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 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
969c7921 | 59 | #include <linux/stop_machine.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 PZ |
70 | #include <linux/debugfs.h> |
71 | #include <linux/ctype.h> | |
6cd8a4bb | 72 | #include <linux/ftrace.h> |
5a0e3ad6 | 73 | #include <linux/slab.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
335d7afb | 77 | #include <asm/mutex.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
5091faa4 | 81 | #include "sched_autogroup.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee HC |
233 | /* |
234 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
235 | * detach_destroy_domains and partition_sched_domains. | |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
5091faa4 MG |
274 | |
275 | #ifdef CONFIG_SCHED_AUTOGROUP | |
276 | struct autogroup *autogroup; | |
277 | #endif | |
29f59db3 SV |
278 | }; |
279 | ||
3d4b47b4 | 280 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 281 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 282 | |
e9036b36 CG |
283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
284 | ||
07e06b01 | 285 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 286 | |
cb4ad1ff | 287 | /* |
2e084786 LJ |
288 | * A weight of 0 or 1 can cause arithmetics problems. |
289 | * A weight of a cfs_rq is the sum of weights of which entities | |
290 | * are queued on this cfs_rq, so a weight of a entity should not be | |
291 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
292 | * (The default weight is 1024 - so there's no practical |
293 | * limitation from this.) | |
294 | */ | |
18d95a28 | 295 | #define MIN_SHARES 2 |
2e084786 | 296 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 297 | |
07e06b01 | 298 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
299 | #endif |
300 | ||
29f59db3 | 301 | /* Default task group. |
3a252015 | 302 | * Every task in system belong to this group at bootup. |
29f59db3 | 303 | */ |
07e06b01 | 304 | struct task_group root_task_group; |
29f59db3 | 305 | |
7c941438 | 306 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 307 | |
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
3fe1698b PZ |
315 | #ifndef CONFIG_64BIT |
316 | u64 min_vruntime_copy; | |
317 | #endif | |
6aa645ea IM |
318 | |
319 | struct rb_root tasks_timeline; | |
320 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
321 | |
322 | struct list_head tasks; | |
323 | struct list_head *balance_iterator; | |
324 | ||
325 | /* | |
326 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
327 | * It is set to NULL otherwise (i.e when none are currently running). |
328 | */ | |
ac53db59 | 329 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 330 | |
5ac5c4d6 | 331 | unsigned int nr_spread_over; |
ddc97297 | 332 | |
62160e3f | 333 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
334 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
335 | ||
41a2d6cf IM |
336 | /* |
337 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
338 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
339 | * (like users, containers etc.) | |
340 | * | |
341 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
342 | * list is used during load balance. | |
343 | */ | |
3d4b47b4 | 344 | int on_list; |
41a2d6cf IM |
345 | struct list_head leaf_cfs_rq_list; |
346 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
347 | |
348 | #ifdef CONFIG_SMP | |
c09595f6 | 349 | /* |
c8cba857 | 350 | * the part of load.weight contributed by tasks |
c09595f6 | 351 | */ |
c8cba857 | 352 | unsigned long task_weight; |
c09595f6 | 353 | |
c8cba857 PZ |
354 | /* |
355 | * h_load = weight * f(tg) | |
356 | * | |
357 | * Where f(tg) is the recursive weight fraction assigned to | |
358 | * this group. | |
359 | */ | |
360 | unsigned long h_load; | |
c09595f6 | 361 | |
c8cba857 | 362 | /* |
3b3d190e PT |
363 | * Maintaining per-cpu shares distribution for group scheduling |
364 | * | |
365 | * load_stamp is the last time we updated the load average | |
366 | * load_last is the last time we updated the load average and saw load | |
367 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 368 | */ |
2069dd75 PZ |
369 | u64 load_avg; |
370 | u64 load_period; | |
3b3d190e | 371 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 372 | |
2069dd75 | 373 | unsigned long load_contribution; |
c09595f6 | 374 | #endif |
6aa645ea IM |
375 | #endif |
376 | }; | |
1da177e4 | 377 | |
6aa645ea IM |
378 | /* Real-Time classes' related field in a runqueue: */ |
379 | struct rt_rq { | |
380 | struct rt_prio_array active; | |
63489e45 | 381 | unsigned long rt_nr_running; |
052f1dc7 | 382 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
383 | struct { |
384 | int curr; /* highest queued rt task prio */ | |
398a153b | 385 | #ifdef CONFIG_SMP |
e864c499 | 386 | int next; /* next highest */ |
398a153b | 387 | #endif |
e864c499 | 388 | } highest_prio; |
6f505b16 | 389 | #endif |
fa85ae24 | 390 | #ifdef CONFIG_SMP |
73fe6aae | 391 | unsigned long rt_nr_migratory; |
a1ba4d8b | 392 | unsigned long rt_nr_total; |
a22d7fc1 | 393 | int overloaded; |
917b627d | 394 | struct plist_head pushable_tasks; |
fa85ae24 | 395 | #endif |
6f505b16 | 396 | int rt_throttled; |
fa85ae24 | 397 | u64 rt_time; |
ac086bc2 | 398 | u64 rt_runtime; |
ea736ed5 | 399 | /* Nests inside the rq lock: */ |
0986b11b | 400 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 401 | |
052f1dc7 | 402 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
403 | unsigned long rt_nr_boosted; |
404 | ||
6f505b16 PZ |
405 | struct rq *rq; |
406 | struct list_head leaf_rt_rq_list; | |
407 | struct task_group *tg; | |
6f505b16 | 408 | #endif |
6aa645ea IM |
409 | }; |
410 | ||
57d885fe GH |
411 | #ifdef CONFIG_SMP |
412 | ||
413 | /* | |
414 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
415 | * variables. Each exclusive cpuset essentially defines an island domain by |
416 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
417 | * exclusive cpuset is created, we also create and attach a new root-domain |
418 | * object. | |
419 | * | |
57d885fe GH |
420 | */ |
421 | struct root_domain { | |
422 | atomic_t refcount; | |
c6c4927b RR |
423 | cpumask_var_t span; |
424 | cpumask_var_t online; | |
637f5085 | 425 | |
0eab9146 | 426 | /* |
637f5085 GH |
427 | * The "RT overload" flag: it gets set if a CPU has more than |
428 | * one runnable RT task. | |
429 | */ | |
c6c4927b | 430 | cpumask_var_t rto_mask; |
0eab9146 | 431 | atomic_t rto_count; |
6e0534f2 | 432 | struct cpupri cpupri; |
57d885fe GH |
433 | }; |
434 | ||
dc938520 GH |
435 | /* |
436 | * By default the system creates a single root-domain with all cpus as | |
437 | * members (mimicking the global state we have today). | |
438 | */ | |
57d885fe GH |
439 | static struct root_domain def_root_domain; |
440 | ||
ed2d372c | 441 | #endif /* CONFIG_SMP */ |
57d885fe | 442 | |
1da177e4 LT |
443 | /* |
444 | * This is the main, per-CPU runqueue data structure. | |
445 | * | |
446 | * Locking rule: those places that want to lock multiple runqueues | |
447 | * (such as the load balancing or the thread migration code), lock | |
448 | * acquire operations must be ordered by ascending &runqueue. | |
449 | */ | |
70b97a7f | 450 | struct rq { |
d8016491 | 451 | /* runqueue lock: */ |
05fa785c | 452 | raw_spinlock_t lock; |
1da177e4 LT |
453 | |
454 | /* | |
455 | * nr_running and cpu_load should be in the same cacheline because | |
456 | * remote CPUs use both these fields when doing load calculation. | |
457 | */ | |
458 | unsigned long nr_running; | |
6aa645ea IM |
459 | #define CPU_LOAD_IDX_MAX 5 |
460 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 461 | unsigned long last_load_update_tick; |
46cb4b7c | 462 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 463 | u64 nohz_stamp; |
83cd4fe2 | 464 | unsigned char nohz_balance_kick; |
46cb4b7c | 465 | #endif |
a64692a3 MG |
466 | unsigned int skip_clock_update; |
467 | ||
d8016491 IM |
468 | /* capture load from *all* tasks on this cpu: */ |
469 | struct load_weight load; | |
6aa645ea IM |
470 | unsigned long nr_load_updates; |
471 | u64 nr_switches; | |
472 | ||
473 | struct cfs_rq cfs; | |
6f505b16 | 474 | struct rt_rq rt; |
6f505b16 | 475 | |
6aa645ea | 476 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
477 | /* list of leaf cfs_rq on this cpu: */ |
478 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
479 | #endif |
480 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 481 | struct list_head leaf_rt_rq_list; |
1da177e4 | 482 | #endif |
1da177e4 LT |
483 | |
484 | /* | |
485 | * This is part of a global counter where only the total sum | |
486 | * over all CPUs matters. A task can increase this counter on | |
487 | * one CPU and if it got migrated afterwards it may decrease | |
488 | * it on another CPU. Always updated under the runqueue lock: | |
489 | */ | |
490 | unsigned long nr_uninterruptible; | |
491 | ||
34f971f6 | 492 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 493 | unsigned long next_balance; |
1da177e4 | 494 | struct mm_struct *prev_mm; |
6aa645ea | 495 | |
3e51f33f | 496 | u64 clock; |
305e6835 | 497 | u64 clock_task; |
6aa645ea | 498 | |
1da177e4 LT |
499 | atomic_t nr_iowait; |
500 | ||
501 | #ifdef CONFIG_SMP | |
0eab9146 | 502 | struct root_domain *rd; |
1da177e4 LT |
503 | struct sched_domain *sd; |
504 | ||
e51fd5e2 PZ |
505 | unsigned long cpu_power; |
506 | ||
a0a522ce | 507 | unsigned char idle_at_tick; |
1da177e4 | 508 | /* For active balancing */ |
3f029d3c | 509 | int post_schedule; |
1da177e4 LT |
510 | int active_balance; |
511 | int push_cpu; | |
969c7921 | 512 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
513 | /* cpu of this runqueue: */ |
514 | int cpu; | |
1f11eb6a | 515 | int online; |
1da177e4 | 516 | |
a8a51d5e | 517 | unsigned long avg_load_per_task; |
1da177e4 | 518 | |
e9e9250b PZ |
519 | u64 rt_avg; |
520 | u64 age_stamp; | |
1b9508f6 MG |
521 | u64 idle_stamp; |
522 | u64 avg_idle; | |
1da177e4 LT |
523 | #endif |
524 | ||
aa483808 VP |
525 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
526 | u64 prev_irq_time; | |
527 | #endif | |
528 | ||
dce48a84 TG |
529 | /* calc_load related fields */ |
530 | unsigned long calc_load_update; | |
531 | long calc_load_active; | |
532 | ||
8f4d37ec | 533 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
534 | #ifdef CONFIG_SMP |
535 | int hrtick_csd_pending; | |
536 | struct call_single_data hrtick_csd; | |
537 | #endif | |
8f4d37ec PZ |
538 | struct hrtimer hrtick_timer; |
539 | #endif | |
540 | ||
1da177e4 LT |
541 | #ifdef CONFIG_SCHEDSTATS |
542 | /* latency stats */ | |
543 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
544 | unsigned long long rq_cpu_time; |
545 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
546 | |
547 | /* sys_sched_yield() stats */ | |
480b9434 | 548 | unsigned int yld_count; |
1da177e4 LT |
549 | |
550 | /* schedule() stats */ | |
480b9434 KC |
551 | unsigned int sched_switch; |
552 | unsigned int sched_count; | |
553 | unsigned int sched_goidle; | |
1da177e4 LT |
554 | |
555 | /* try_to_wake_up() stats */ | |
480b9434 KC |
556 | unsigned int ttwu_count; |
557 | unsigned int ttwu_local; | |
1da177e4 | 558 | #endif |
317f3941 PZ |
559 | |
560 | #ifdef CONFIG_SMP | |
561 | struct task_struct *wake_list; | |
562 | #endif | |
1da177e4 LT |
563 | }; |
564 | ||
f34e3b61 | 565 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 566 | |
a64692a3 | 567 | |
1e5a7405 | 568 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 569 | |
0a2966b4 CL |
570 | static inline int cpu_of(struct rq *rq) |
571 | { | |
572 | #ifdef CONFIG_SMP | |
573 | return rq->cpu; | |
574 | #else | |
575 | return 0; | |
576 | #endif | |
577 | } | |
578 | ||
497f0ab3 | 579 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
580 | rcu_dereference_check((p), \ |
581 | rcu_read_lock_sched_held() || \ | |
582 | lockdep_is_held(&sched_domains_mutex)) | |
583 | ||
674311d5 NP |
584 | /* |
585 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 586 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
587 | * |
588 | * The domain tree of any CPU may only be accessed from within | |
589 | * preempt-disabled sections. | |
590 | */ | |
48f24c4d | 591 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 592 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
593 | |
594 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
595 | #define this_rq() (&__get_cpu_var(runqueues)) | |
596 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
597 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 598 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 599 | |
dc61b1d6 PZ |
600 | #ifdef CONFIG_CGROUP_SCHED |
601 | ||
602 | /* | |
603 | * Return the group to which this tasks belongs. | |
604 | * | |
605 | * We use task_subsys_state_check() and extend the RCU verification | |
0122ec5b | 606 | * with lockdep_is_held(&p->pi_lock) because cpu_cgroup_attach() |
dc61b1d6 PZ |
607 | * holds that lock for each task it moves into the cgroup. Therefore |
608 | * by holding that lock, we pin the task to the current cgroup. | |
609 | */ | |
610 | static inline struct task_group *task_group(struct task_struct *p) | |
611 | { | |
5091faa4 | 612 | struct task_group *tg; |
dc61b1d6 PZ |
613 | struct cgroup_subsys_state *css; |
614 | ||
615 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
0122ec5b | 616 | lockdep_is_held(&p->pi_lock)); |
5091faa4 MG |
617 | tg = container_of(css, struct task_group, css); |
618 | ||
619 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
620 | } |
621 | ||
622 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
623 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
624 | { | |
625 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
626 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
627 | p->se.parent = task_group(p)->se[cpu]; | |
628 | #endif | |
629 | ||
630 | #ifdef CONFIG_RT_GROUP_SCHED | |
631 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
632 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
633 | #endif | |
634 | } | |
635 | ||
636 | #else /* CONFIG_CGROUP_SCHED */ | |
637 | ||
638 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
639 | static inline struct task_group *task_group(struct task_struct *p) | |
640 | { | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | #endif /* CONFIG_CGROUP_SCHED */ | |
645 | ||
fe44d621 | 646 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 647 | |
fe44d621 | 648 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 649 | { |
fe44d621 | 650 | s64 delta; |
305e6835 | 651 | |
f26f9aff MG |
652 | if (rq->skip_clock_update) |
653 | return; | |
aa483808 | 654 | |
fe44d621 PZ |
655 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
656 | rq->clock += delta; | |
657 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
658 | } |
659 | ||
bf5c91ba IM |
660 | /* |
661 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
662 | */ | |
663 | #ifdef CONFIG_SCHED_DEBUG | |
664 | # define const_debug __read_mostly | |
665 | #else | |
666 | # define const_debug static const | |
667 | #endif | |
668 | ||
017730c1 | 669 | /** |
1fd06bb1 | 670 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 671 | * @cpu: the processor in question. |
017730c1 | 672 | * |
017730c1 IM |
673 | * This interface allows printk to be called with the runqueue lock |
674 | * held and know whether or not it is OK to wake up the klogd. | |
675 | */ | |
89f19f04 | 676 | int runqueue_is_locked(int cpu) |
017730c1 | 677 | { |
05fa785c | 678 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
679 | } |
680 | ||
bf5c91ba IM |
681 | /* |
682 | * Debugging: various feature bits | |
683 | */ | |
f00b45c1 PZ |
684 | |
685 | #define SCHED_FEAT(name, enabled) \ | |
686 | __SCHED_FEAT_##name , | |
687 | ||
bf5c91ba | 688 | enum { |
f00b45c1 | 689 | #include "sched_features.h" |
bf5c91ba IM |
690 | }; |
691 | ||
f00b45c1 PZ |
692 | #undef SCHED_FEAT |
693 | ||
694 | #define SCHED_FEAT(name, enabled) \ | |
695 | (1UL << __SCHED_FEAT_##name) * enabled | | |
696 | ||
bf5c91ba | 697 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
698 | #include "sched_features.h" |
699 | 0; | |
700 | ||
701 | #undef SCHED_FEAT | |
702 | ||
703 | #ifdef CONFIG_SCHED_DEBUG | |
704 | #define SCHED_FEAT(name, enabled) \ | |
705 | #name , | |
706 | ||
983ed7a6 | 707 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
708 | #include "sched_features.h" |
709 | NULL | |
710 | }; | |
711 | ||
712 | #undef SCHED_FEAT | |
713 | ||
34f3a814 | 714 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 715 | { |
f00b45c1 PZ |
716 | int i; |
717 | ||
718 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
719 | if (!(sysctl_sched_features & (1UL << i))) |
720 | seq_puts(m, "NO_"); | |
721 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 722 | } |
34f3a814 | 723 | seq_puts(m, "\n"); |
f00b45c1 | 724 | |
34f3a814 | 725 | return 0; |
f00b45c1 PZ |
726 | } |
727 | ||
728 | static ssize_t | |
729 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
730 | size_t cnt, loff_t *ppos) | |
731 | { | |
732 | char buf[64]; | |
7740191c | 733 | char *cmp; |
f00b45c1 PZ |
734 | int neg = 0; |
735 | int i; | |
736 | ||
737 | if (cnt > 63) | |
738 | cnt = 63; | |
739 | ||
740 | if (copy_from_user(&buf, ubuf, cnt)) | |
741 | return -EFAULT; | |
742 | ||
743 | buf[cnt] = 0; | |
7740191c | 744 | cmp = strstrip(buf); |
f00b45c1 | 745 | |
524429c3 | 746 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
747 | neg = 1; |
748 | cmp += 3; | |
749 | } | |
750 | ||
751 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 752 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
753 | if (neg) |
754 | sysctl_sched_features &= ~(1UL << i); | |
755 | else | |
756 | sysctl_sched_features |= (1UL << i); | |
757 | break; | |
758 | } | |
759 | } | |
760 | ||
761 | if (!sched_feat_names[i]) | |
762 | return -EINVAL; | |
763 | ||
42994724 | 764 | *ppos += cnt; |
f00b45c1 PZ |
765 | |
766 | return cnt; | |
767 | } | |
768 | ||
34f3a814 LZ |
769 | static int sched_feat_open(struct inode *inode, struct file *filp) |
770 | { | |
771 | return single_open(filp, sched_feat_show, NULL); | |
772 | } | |
773 | ||
828c0950 | 774 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
775 | .open = sched_feat_open, |
776 | .write = sched_feat_write, | |
777 | .read = seq_read, | |
778 | .llseek = seq_lseek, | |
779 | .release = single_release, | |
f00b45c1 PZ |
780 | }; |
781 | ||
782 | static __init int sched_init_debug(void) | |
783 | { | |
f00b45c1 PZ |
784 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
785 | &sched_feat_fops); | |
786 | ||
787 | return 0; | |
788 | } | |
789 | late_initcall(sched_init_debug); | |
790 | ||
791 | #endif | |
792 | ||
793 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 794 | |
b82d9fdd PZ |
795 | /* |
796 | * Number of tasks to iterate in a single balance run. | |
797 | * Limited because this is done with IRQs disabled. | |
798 | */ | |
799 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
800 | ||
e9e9250b PZ |
801 | /* |
802 | * period over which we average the RT time consumption, measured | |
803 | * in ms. | |
804 | * | |
805 | * default: 1s | |
806 | */ | |
807 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
808 | ||
fa85ae24 | 809 | /* |
9f0c1e56 | 810 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
811 | * default: 1s |
812 | */ | |
9f0c1e56 | 813 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 814 | |
6892b75e IM |
815 | static __read_mostly int scheduler_running; |
816 | ||
9f0c1e56 PZ |
817 | /* |
818 | * part of the period that we allow rt tasks to run in us. | |
819 | * default: 0.95s | |
820 | */ | |
821 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 822 | |
d0b27fa7 PZ |
823 | static inline u64 global_rt_period(void) |
824 | { | |
825 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
826 | } | |
827 | ||
828 | static inline u64 global_rt_runtime(void) | |
829 | { | |
e26873bb | 830 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
831 | return RUNTIME_INF; |
832 | ||
833 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
834 | } | |
fa85ae24 | 835 | |
1da177e4 | 836 | #ifndef prepare_arch_switch |
4866cde0 NP |
837 | # define prepare_arch_switch(next) do { } while (0) |
838 | #endif | |
839 | #ifndef finish_arch_switch | |
840 | # define finish_arch_switch(prev) do { } while (0) | |
841 | #endif | |
842 | ||
051a1d1a DA |
843 | static inline int task_current(struct rq *rq, struct task_struct *p) |
844 | { | |
845 | return rq->curr == p; | |
846 | } | |
847 | ||
70b97a7f | 848 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 849 | { |
3ca7a440 PZ |
850 | #ifdef CONFIG_SMP |
851 | return p->on_cpu; | |
852 | #else | |
051a1d1a | 853 | return task_current(rq, p); |
3ca7a440 | 854 | #endif |
4866cde0 NP |
855 | } |
856 | ||
3ca7a440 | 857 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 858 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 859 | { |
3ca7a440 PZ |
860 | #ifdef CONFIG_SMP |
861 | /* | |
862 | * We can optimise this out completely for !SMP, because the | |
863 | * SMP rebalancing from interrupt is the only thing that cares | |
864 | * here. | |
865 | */ | |
866 | next->on_cpu = 1; | |
867 | #endif | |
4866cde0 NP |
868 | } |
869 | ||
70b97a7f | 870 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 871 | { |
3ca7a440 PZ |
872 | #ifdef CONFIG_SMP |
873 | /* | |
874 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
875 | * We must ensure this doesn't happen until the switch is completely | |
876 | * finished. | |
877 | */ | |
878 | smp_wmb(); | |
879 | prev->on_cpu = 0; | |
880 | #endif | |
da04c035 IM |
881 | #ifdef CONFIG_DEBUG_SPINLOCK |
882 | /* this is a valid case when another task releases the spinlock */ | |
883 | rq->lock.owner = current; | |
884 | #endif | |
8a25d5de IM |
885 | /* |
886 | * If we are tracking spinlock dependencies then we have to | |
887 | * fix up the runqueue lock - which gets 'carried over' from | |
888 | * prev into current: | |
889 | */ | |
890 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
891 | ||
05fa785c | 892 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
893 | } |
894 | ||
895 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 896 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
897 | { |
898 | #ifdef CONFIG_SMP | |
899 | /* | |
900 | * We can optimise this out completely for !SMP, because the | |
901 | * SMP rebalancing from interrupt is the only thing that cares | |
902 | * here. | |
903 | */ | |
3ca7a440 | 904 | next->on_cpu = 1; |
4866cde0 NP |
905 | #endif |
906 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 907 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 908 | #else |
05fa785c | 909 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
910 | #endif |
911 | } | |
912 | ||
70b97a7f | 913 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
914 | { |
915 | #ifdef CONFIG_SMP | |
916 | /* | |
3ca7a440 | 917 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
918 | * We must ensure this doesn't happen until the switch is completely |
919 | * finished. | |
920 | */ | |
921 | smp_wmb(); | |
3ca7a440 | 922 | prev->on_cpu = 0; |
4866cde0 NP |
923 | #endif |
924 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
925 | local_irq_enable(); | |
1da177e4 | 926 | #endif |
4866cde0 NP |
927 | } |
928 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 929 | |
0970d299 | 930 | /* |
0122ec5b | 931 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 932 | */ |
70b97a7f | 933 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
934 | __acquires(rq->lock) |
935 | { | |
0970d299 PZ |
936 | struct rq *rq; |
937 | ||
0122ec5b PZ |
938 | lockdep_assert_held(&p->pi_lock); |
939 | ||
3a5c359a | 940 | for (;;) { |
0970d299 | 941 | rq = task_rq(p); |
05fa785c | 942 | raw_spin_lock(&rq->lock); |
65cc8e48 | 943 | if (likely(rq == task_rq(p))) |
3a5c359a | 944 | return rq; |
05fa785c | 945 | raw_spin_unlock(&rq->lock); |
b29739f9 | 946 | } |
b29739f9 IM |
947 | } |
948 | ||
1da177e4 | 949 | /* |
0122ec5b | 950 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 951 | */ |
70b97a7f | 952 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 953 | __acquires(p->pi_lock) |
1da177e4 LT |
954 | __acquires(rq->lock) |
955 | { | |
70b97a7f | 956 | struct rq *rq; |
1da177e4 | 957 | |
3a5c359a | 958 | for (;;) { |
0122ec5b | 959 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 960 | rq = task_rq(p); |
05fa785c | 961 | raw_spin_lock(&rq->lock); |
65cc8e48 | 962 | if (likely(rq == task_rq(p))) |
3a5c359a | 963 | return rq; |
0122ec5b PZ |
964 | raw_spin_unlock(&rq->lock); |
965 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 966 | } |
1da177e4 LT |
967 | } |
968 | ||
a9957449 | 969 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
970 | __releases(rq->lock) |
971 | { | |
05fa785c | 972 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
973 | } |
974 | ||
0122ec5b PZ |
975 | static inline void |
976 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 977 | __releases(rq->lock) |
0122ec5b | 978 | __releases(p->pi_lock) |
1da177e4 | 979 | { |
0122ec5b PZ |
980 | raw_spin_unlock(&rq->lock); |
981 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
982 | } |
983 | ||
1da177e4 | 984 | /* |
cc2a73b5 | 985 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 986 | */ |
a9957449 | 987 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
988 | __acquires(rq->lock) |
989 | { | |
70b97a7f | 990 | struct rq *rq; |
1da177e4 LT |
991 | |
992 | local_irq_disable(); | |
993 | rq = this_rq(); | |
05fa785c | 994 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
995 | |
996 | return rq; | |
997 | } | |
998 | ||
8f4d37ec PZ |
999 | #ifdef CONFIG_SCHED_HRTICK |
1000 | /* | |
1001 | * Use HR-timers to deliver accurate preemption points. | |
1002 | * | |
1003 | * Its all a bit involved since we cannot program an hrt while holding the | |
1004 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1005 | * reschedule event. | |
1006 | * | |
1007 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1008 | * rq->lock. | |
1009 | */ | |
8f4d37ec PZ |
1010 | |
1011 | /* | |
1012 | * Use hrtick when: | |
1013 | * - enabled by features | |
1014 | * - hrtimer is actually high res | |
1015 | */ | |
1016 | static inline int hrtick_enabled(struct rq *rq) | |
1017 | { | |
1018 | if (!sched_feat(HRTICK)) | |
1019 | return 0; | |
ba42059f | 1020 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1021 | return 0; |
8f4d37ec PZ |
1022 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1023 | } | |
1024 | ||
8f4d37ec PZ |
1025 | static void hrtick_clear(struct rq *rq) |
1026 | { | |
1027 | if (hrtimer_active(&rq->hrtick_timer)) | |
1028 | hrtimer_cancel(&rq->hrtick_timer); | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | /* |
1032 | * High-resolution timer tick. | |
1033 | * Runs from hardirq context with interrupts disabled. | |
1034 | */ | |
1035 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1036 | { | |
1037 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1038 | ||
1039 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1040 | ||
05fa785c | 1041 | raw_spin_lock(&rq->lock); |
3e51f33f | 1042 | update_rq_clock(rq); |
8f4d37ec | 1043 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1044 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1045 | |
1046 | return HRTIMER_NORESTART; | |
1047 | } | |
1048 | ||
95e904c7 | 1049 | #ifdef CONFIG_SMP |
31656519 PZ |
1050 | /* |
1051 | * called from hardirq (IPI) context | |
1052 | */ | |
1053 | static void __hrtick_start(void *arg) | |
b328ca18 | 1054 | { |
31656519 | 1055 | struct rq *rq = arg; |
b328ca18 | 1056 | |
05fa785c | 1057 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1058 | hrtimer_restart(&rq->hrtick_timer); |
1059 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1060 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1061 | } |
1062 | ||
31656519 PZ |
1063 | /* |
1064 | * Called to set the hrtick timer state. | |
1065 | * | |
1066 | * called with rq->lock held and irqs disabled | |
1067 | */ | |
1068 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1069 | { |
31656519 PZ |
1070 | struct hrtimer *timer = &rq->hrtick_timer; |
1071 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1072 | |
cc584b21 | 1073 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1074 | |
1075 | if (rq == this_rq()) { | |
1076 | hrtimer_restart(timer); | |
1077 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1078 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1079 | rq->hrtick_csd_pending = 1; |
1080 | } | |
b328ca18 PZ |
1081 | } |
1082 | ||
1083 | static int | |
1084 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1085 | { | |
1086 | int cpu = (int)(long)hcpu; | |
1087 | ||
1088 | switch (action) { | |
1089 | case CPU_UP_CANCELED: | |
1090 | case CPU_UP_CANCELED_FROZEN: | |
1091 | case CPU_DOWN_PREPARE: | |
1092 | case CPU_DOWN_PREPARE_FROZEN: | |
1093 | case CPU_DEAD: | |
1094 | case CPU_DEAD_FROZEN: | |
31656519 | 1095 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1096 | return NOTIFY_OK; |
1097 | } | |
1098 | ||
1099 | return NOTIFY_DONE; | |
1100 | } | |
1101 | ||
fa748203 | 1102 | static __init void init_hrtick(void) |
b328ca18 PZ |
1103 | { |
1104 | hotcpu_notifier(hotplug_hrtick, 0); | |
1105 | } | |
31656519 PZ |
1106 | #else |
1107 | /* | |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
1113 | { | |
7f1e2ca9 | 1114 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1115 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1116 | } |
b328ca18 | 1117 | |
006c75f1 | 1118 | static inline void init_hrtick(void) |
8f4d37ec | 1119 | { |
8f4d37ec | 1120 | } |
31656519 | 1121 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1122 | |
31656519 | 1123 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1124 | { |
31656519 PZ |
1125 | #ifdef CONFIG_SMP |
1126 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1127 | |
31656519 PZ |
1128 | rq->hrtick_csd.flags = 0; |
1129 | rq->hrtick_csd.func = __hrtick_start; | |
1130 | rq->hrtick_csd.info = rq; | |
1131 | #endif | |
8f4d37ec | 1132 | |
31656519 PZ |
1133 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1134 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1135 | } |
006c75f1 | 1136 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1137 | static inline void hrtick_clear(struct rq *rq) |
1138 | { | |
1139 | } | |
1140 | ||
8f4d37ec PZ |
1141 | static inline void init_rq_hrtick(struct rq *rq) |
1142 | { | |
1143 | } | |
1144 | ||
b328ca18 PZ |
1145 | static inline void init_hrtick(void) |
1146 | { | |
1147 | } | |
006c75f1 | 1148 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1149 | |
c24d20db IM |
1150 | /* |
1151 | * resched_task - mark a task 'to be rescheduled now'. | |
1152 | * | |
1153 | * On UP this means the setting of the need_resched flag, on SMP it | |
1154 | * might also involve a cross-CPU call to trigger the scheduler on | |
1155 | * the target CPU. | |
1156 | */ | |
1157 | #ifdef CONFIG_SMP | |
1158 | ||
1159 | #ifndef tsk_is_polling | |
1160 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1161 | #endif | |
1162 | ||
31656519 | 1163 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1164 | { |
1165 | int cpu; | |
1166 | ||
05fa785c | 1167 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1168 | |
5ed0cec0 | 1169 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1170 | return; |
1171 | ||
5ed0cec0 | 1172 | set_tsk_need_resched(p); |
c24d20db IM |
1173 | |
1174 | cpu = task_cpu(p); | |
1175 | if (cpu == smp_processor_id()) | |
1176 | return; | |
1177 | ||
1178 | /* NEED_RESCHED must be visible before we test polling */ | |
1179 | smp_mb(); | |
1180 | if (!tsk_is_polling(p)) | |
1181 | smp_send_reschedule(cpu); | |
1182 | } | |
1183 | ||
1184 | static void resched_cpu(int cpu) | |
1185 | { | |
1186 | struct rq *rq = cpu_rq(cpu); | |
1187 | unsigned long flags; | |
1188 | ||
05fa785c | 1189 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1190 | return; |
1191 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1192 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1193 | } |
06d8308c TG |
1194 | |
1195 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1196 | /* |
1197 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1198 | * from an idle cpu. This is good for power-savings. | |
1199 | * | |
1200 | * We don't do similar optimization for completely idle system, as | |
1201 | * selecting an idle cpu will add more delays to the timers than intended | |
1202 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1203 | */ | |
1204 | int get_nohz_timer_target(void) | |
1205 | { | |
1206 | int cpu = smp_processor_id(); | |
1207 | int i; | |
1208 | struct sched_domain *sd; | |
1209 | ||
1210 | for_each_domain(cpu, sd) { | |
1211 | for_each_cpu(i, sched_domain_span(sd)) | |
1212 | if (!idle_cpu(i)) | |
1213 | return i; | |
1214 | } | |
1215 | return cpu; | |
1216 | } | |
06d8308c TG |
1217 | /* |
1218 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1219 | * idle CPU then this timer might expire before the next timer event | |
1220 | * which is scheduled to wake up that CPU. In case of a completely | |
1221 | * idle system the next event might even be infinite time into the | |
1222 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1223 | * leaves the inner idle loop so the newly added timer is taken into | |
1224 | * account when the CPU goes back to idle and evaluates the timer | |
1225 | * wheel for the next timer event. | |
1226 | */ | |
1227 | void wake_up_idle_cpu(int cpu) | |
1228 | { | |
1229 | struct rq *rq = cpu_rq(cpu); | |
1230 | ||
1231 | if (cpu == smp_processor_id()) | |
1232 | return; | |
1233 | ||
1234 | /* | |
1235 | * This is safe, as this function is called with the timer | |
1236 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1237 | * to idle and has not yet set rq->curr to idle then it will | |
1238 | * be serialized on the timer wheel base lock and take the new | |
1239 | * timer into account automatically. | |
1240 | */ | |
1241 | if (rq->curr != rq->idle) | |
1242 | return; | |
1243 | ||
1244 | /* | |
1245 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1246 | * lockless. The worst case is that the other CPU runs the | |
1247 | * idle task through an additional NOOP schedule() | |
1248 | */ | |
5ed0cec0 | 1249 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1250 | |
1251 | /* NEED_RESCHED must be visible before we test polling */ | |
1252 | smp_mb(); | |
1253 | if (!tsk_is_polling(rq->idle)) | |
1254 | smp_send_reschedule(cpu); | |
1255 | } | |
39c0cbe2 | 1256 | |
6d6bc0ad | 1257 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1258 | |
e9e9250b PZ |
1259 | static u64 sched_avg_period(void) |
1260 | { | |
1261 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1262 | } | |
1263 | ||
1264 | static void sched_avg_update(struct rq *rq) | |
1265 | { | |
1266 | s64 period = sched_avg_period(); | |
1267 | ||
1268 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1269 | /* |
1270 | * Inline assembly required to prevent the compiler | |
1271 | * optimising this loop into a divmod call. | |
1272 | * See __iter_div_u64_rem() for another example of this. | |
1273 | */ | |
1274 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1275 | rq->age_stamp += period; |
1276 | rq->rt_avg /= 2; | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1281 | { | |
1282 | rq->rt_avg += rt_delta; | |
1283 | sched_avg_update(rq); | |
1284 | } | |
1285 | ||
6d6bc0ad | 1286 | #else /* !CONFIG_SMP */ |
31656519 | 1287 | static void resched_task(struct task_struct *p) |
c24d20db | 1288 | { |
05fa785c | 1289 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1290 | set_tsk_need_resched(p); |
c24d20db | 1291 | } |
e9e9250b PZ |
1292 | |
1293 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1294 | { | |
1295 | } | |
da2b71ed SS |
1296 | |
1297 | static void sched_avg_update(struct rq *rq) | |
1298 | { | |
1299 | } | |
6d6bc0ad | 1300 | #endif /* CONFIG_SMP */ |
c24d20db | 1301 | |
45bf76df IM |
1302 | #if BITS_PER_LONG == 32 |
1303 | # define WMULT_CONST (~0UL) | |
1304 | #else | |
1305 | # define WMULT_CONST (1UL << 32) | |
1306 | #endif | |
1307 | ||
1308 | #define WMULT_SHIFT 32 | |
1309 | ||
194081eb IM |
1310 | /* |
1311 | * Shift right and round: | |
1312 | */ | |
cf2ab469 | 1313 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1314 | |
a7be37ac PZ |
1315 | /* |
1316 | * delta *= weight / lw | |
1317 | */ | |
cb1c4fc9 | 1318 | static unsigned long |
45bf76df IM |
1319 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1320 | struct load_weight *lw) | |
1321 | { | |
1322 | u64 tmp; | |
1323 | ||
7a232e03 LJ |
1324 | if (!lw->inv_weight) { |
1325 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1326 | lw->inv_weight = 1; | |
1327 | else | |
1328 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1329 | / (lw->weight+1); | |
1330 | } | |
45bf76df IM |
1331 | |
1332 | tmp = (u64)delta_exec * weight; | |
1333 | /* | |
1334 | * Check whether we'd overflow the 64-bit multiplication: | |
1335 | */ | |
194081eb | 1336 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1337 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1338 | WMULT_SHIFT/2); |
1339 | else | |
cf2ab469 | 1340 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1341 | |
ecf691da | 1342 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1343 | } |
1344 | ||
1091985b | 1345 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1346 | { |
1347 | lw->weight += inc; | |
e89996ae | 1348 | lw->inv_weight = 0; |
45bf76df IM |
1349 | } |
1350 | ||
1091985b | 1351 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1352 | { |
1353 | lw->weight -= dec; | |
e89996ae | 1354 | lw->inv_weight = 0; |
45bf76df IM |
1355 | } |
1356 | ||
2069dd75 PZ |
1357 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1358 | { | |
1359 | lw->weight = w; | |
1360 | lw->inv_weight = 0; | |
1361 | } | |
1362 | ||
2dd73a4f PW |
1363 | /* |
1364 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1365 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1366 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1367 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1368 | * scaled version of the new time slice allocation that they receive on time |
1369 | * slice expiry etc. | |
1370 | */ | |
1371 | ||
cce7ade8 PZ |
1372 | #define WEIGHT_IDLEPRIO 3 |
1373 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1374 | |
1375 | /* | |
1376 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1377 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1378 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1379 | * that remained on nice 0. | |
1380 | * | |
1381 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1382 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1383 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1384 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1385 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1386 | */ |
1387 | static const int prio_to_weight[40] = { | |
254753dc IM |
1388 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1389 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1390 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1391 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1392 | /* 0 */ 1024, 820, 655, 526, 423, | |
1393 | /* 5 */ 335, 272, 215, 172, 137, | |
1394 | /* 10 */ 110, 87, 70, 56, 45, | |
1395 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1396 | }; |
1397 | ||
5714d2de IM |
1398 | /* |
1399 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1400 | * | |
1401 | * In cases where the weight does not change often, we can use the | |
1402 | * precalculated inverse to speed up arithmetics by turning divisions | |
1403 | * into multiplications: | |
1404 | */ | |
dd41f596 | 1405 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1406 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1407 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1408 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1409 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1410 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1411 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1412 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1413 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1414 | }; |
2dd73a4f | 1415 | |
ef12fefa BR |
1416 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1417 | enum cpuacct_stat_index { | |
1418 | CPUACCT_STAT_USER, /* ... user mode */ | |
1419 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1420 | ||
1421 | CPUACCT_STAT_NSTATS, | |
1422 | }; | |
1423 | ||
d842de87 SV |
1424 | #ifdef CONFIG_CGROUP_CPUACCT |
1425 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1426 | static void cpuacct_update_stats(struct task_struct *tsk, |
1427 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1428 | #else |
1429 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1430 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1431 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1432 | #endif |
1433 | ||
18d95a28 PZ |
1434 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1435 | { | |
1436 | update_load_add(&rq->load, load); | |
1437 | } | |
1438 | ||
1439 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1440 | { | |
1441 | update_load_sub(&rq->load, load); | |
1442 | } | |
1443 | ||
7940ca36 | 1444 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1445 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1446 | |
1447 | /* | |
1448 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1449 | * leaving it for the final time. | |
1450 | */ | |
eb755805 | 1451 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1452 | { |
1453 | struct task_group *parent, *child; | |
eb755805 | 1454 | int ret; |
c09595f6 PZ |
1455 | |
1456 | rcu_read_lock(); | |
1457 | parent = &root_task_group; | |
1458 | down: | |
eb755805 PZ |
1459 | ret = (*down)(parent, data); |
1460 | if (ret) | |
1461 | goto out_unlock; | |
c09595f6 PZ |
1462 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1463 | parent = child; | |
1464 | goto down; | |
1465 | ||
1466 | up: | |
1467 | continue; | |
1468 | } | |
eb755805 PZ |
1469 | ret = (*up)(parent, data); |
1470 | if (ret) | |
1471 | goto out_unlock; | |
c09595f6 PZ |
1472 | |
1473 | child = parent; | |
1474 | parent = parent->parent; | |
1475 | if (parent) | |
1476 | goto up; | |
eb755805 | 1477 | out_unlock: |
c09595f6 | 1478 | rcu_read_unlock(); |
eb755805 PZ |
1479 | |
1480 | return ret; | |
c09595f6 PZ |
1481 | } |
1482 | ||
eb755805 PZ |
1483 | static int tg_nop(struct task_group *tg, void *data) |
1484 | { | |
1485 | return 0; | |
c09595f6 | 1486 | } |
eb755805 PZ |
1487 | #endif |
1488 | ||
1489 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1490 | /* Used instead of source_load when we know the type == 0 */ |
1491 | static unsigned long weighted_cpuload(const int cpu) | |
1492 | { | |
1493 | return cpu_rq(cpu)->load.weight; | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * Return a low guess at the load of a migration-source cpu weighted | |
1498 | * according to the scheduling class and "nice" value. | |
1499 | * | |
1500 | * We want to under-estimate the load of migration sources, to | |
1501 | * balance conservatively. | |
1502 | */ | |
1503 | static unsigned long source_load(int cpu, int type) | |
1504 | { | |
1505 | struct rq *rq = cpu_rq(cpu); | |
1506 | unsigned long total = weighted_cpuload(cpu); | |
1507 | ||
1508 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1509 | return total; | |
1510 | ||
1511 | return min(rq->cpu_load[type-1], total); | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * Return a high guess at the load of a migration-target cpu weighted | |
1516 | * according to the scheduling class and "nice" value. | |
1517 | */ | |
1518 | static unsigned long target_load(int cpu, int type) | |
1519 | { | |
1520 | struct rq *rq = cpu_rq(cpu); | |
1521 | unsigned long total = weighted_cpuload(cpu); | |
1522 | ||
1523 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1524 | return total; | |
1525 | ||
1526 | return max(rq->cpu_load[type-1], total); | |
1527 | } | |
1528 | ||
ae154be1 PZ |
1529 | static unsigned long power_of(int cpu) |
1530 | { | |
e51fd5e2 | 1531 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1532 | } |
1533 | ||
eb755805 PZ |
1534 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1535 | ||
1536 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1537 | { | |
1538 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1539 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1540 | |
4cd42620 SR |
1541 | if (nr_running) |
1542 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1543 | else |
1544 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1545 | |
1546 | return rq->avg_load_per_task; | |
1547 | } | |
1548 | ||
1549 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1550 | |
c09595f6 | 1551 | /* |
c8cba857 PZ |
1552 | * Compute the cpu's hierarchical load factor for each task group. |
1553 | * This needs to be done in a top-down fashion because the load of a child | |
1554 | * group is a fraction of its parents load. | |
c09595f6 | 1555 | */ |
eb755805 | 1556 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1557 | { |
c8cba857 | 1558 | unsigned long load; |
eb755805 | 1559 | long cpu = (long)data; |
c09595f6 | 1560 | |
c8cba857 PZ |
1561 | if (!tg->parent) { |
1562 | load = cpu_rq(cpu)->load.weight; | |
1563 | } else { | |
1564 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1565 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1566 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1567 | } | |
c09595f6 | 1568 | |
c8cba857 | 1569 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1570 | |
eb755805 | 1571 | return 0; |
c09595f6 PZ |
1572 | } |
1573 | ||
eb755805 | 1574 | static void update_h_load(long cpu) |
c09595f6 | 1575 | { |
eb755805 | 1576 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1577 | } |
1578 | ||
18d95a28 PZ |
1579 | #endif |
1580 | ||
8f45e2b5 GH |
1581 | #ifdef CONFIG_PREEMPT |
1582 | ||
b78bb868 PZ |
1583 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1584 | ||
70574a99 | 1585 | /* |
8f45e2b5 GH |
1586 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1587 | * way at the expense of forcing extra atomic operations in all | |
1588 | * invocations. This assures that the double_lock is acquired using the | |
1589 | * same underlying policy as the spinlock_t on this architecture, which | |
1590 | * reduces latency compared to the unfair variant below. However, it | |
1591 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1592 | */ |
8f45e2b5 GH |
1593 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1594 | __releases(this_rq->lock) | |
1595 | __acquires(busiest->lock) | |
1596 | __acquires(this_rq->lock) | |
1597 | { | |
05fa785c | 1598 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1599 | double_rq_lock(this_rq, busiest); |
1600 | ||
1601 | return 1; | |
1602 | } | |
1603 | ||
1604 | #else | |
1605 | /* | |
1606 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1607 | * latency by eliminating extra atomic operations when the locks are | |
1608 | * already in proper order on entry. This favors lower cpu-ids and will | |
1609 | * grant the double lock to lower cpus over higher ids under contention, | |
1610 | * regardless of entry order into the function. | |
1611 | */ | |
1612 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1613 | __releases(this_rq->lock) |
1614 | __acquires(busiest->lock) | |
1615 | __acquires(this_rq->lock) | |
1616 | { | |
1617 | int ret = 0; | |
1618 | ||
05fa785c | 1619 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1620 | if (busiest < this_rq) { |
05fa785c TG |
1621 | raw_spin_unlock(&this_rq->lock); |
1622 | raw_spin_lock(&busiest->lock); | |
1623 | raw_spin_lock_nested(&this_rq->lock, | |
1624 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1625 | ret = 1; |
1626 | } else | |
05fa785c TG |
1627 | raw_spin_lock_nested(&busiest->lock, |
1628 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1629 | } |
1630 | return ret; | |
1631 | } | |
1632 | ||
8f45e2b5 GH |
1633 | #endif /* CONFIG_PREEMPT */ |
1634 | ||
1635 | /* | |
1636 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1637 | */ | |
1638 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1639 | { | |
1640 | if (unlikely(!irqs_disabled())) { | |
1641 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1642 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1643 | BUG_ON(1); |
1644 | } | |
1645 | ||
1646 | return _double_lock_balance(this_rq, busiest); | |
1647 | } | |
1648 | ||
70574a99 AD |
1649 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1650 | __releases(busiest->lock) | |
1651 | { | |
05fa785c | 1652 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1653 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1654 | } | |
1e3c88bd PZ |
1655 | |
1656 | /* | |
1657 | * double_rq_lock - safely lock two runqueues | |
1658 | * | |
1659 | * Note this does not disable interrupts like task_rq_lock, | |
1660 | * you need to do so manually before calling. | |
1661 | */ | |
1662 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1663 | __acquires(rq1->lock) | |
1664 | __acquires(rq2->lock) | |
1665 | { | |
1666 | BUG_ON(!irqs_disabled()); | |
1667 | if (rq1 == rq2) { | |
1668 | raw_spin_lock(&rq1->lock); | |
1669 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1670 | } else { | |
1671 | if (rq1 < rq2) { | |
1672 | raw_spin_lock(&rq1->lock); | |
1673 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1674 | } else { | |
1675 | raw_spin_lock(&rq2->lock); | |
1676 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1677 | } | |
1678 | } | |
1e3c88bd PZ |
1679 | } |
1680 | ||
1681 | /* | |
1682 | * double_rq_unlock - safely unlock two runqueues | |
1683 | * | |
1684 | * Note this does not restore interrupts like task_rq_unlock, | |
1685 | * you need to do so manually after calling. | |
1686 | */ | |
1687 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1688 | __releases(rq1->lock) | |
1689 | __releases(rq2->lock) | |
1690 | { | |
1691 | raw_spin_unlock(&rq1->lock); | |
1692 | if (rq1 != rq2) | |
1693 | raw_spin_unlock(&rq2->lock); | |
1694 | else | |
1695 | __release(rq2->lock); | |
1696 | } | |
1697 | ||
d95f4122 MG |
1698 | #else /* CONFIG_SMP */ |
1699 | ||
1700 | /* | |
1701 | * double_rq_lock - safely lock two runqueues | |
1702 | * | |
1703 | * Note this does not disable interrupts like task_rq_lock, | |
1704 | * you need to do so manually before calling. | |
1705 | */ | |
1706 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1707 | __acquires(rq1->lock) | |
1708 | __acquires(rq2->lock) | |
1709 | { | |
1710 | BUG_ON(!irqs_disabled()); | |
1711 | BUG_ON(rq1 != rq2); | |
1712 | raw_spin_lock(&rq1->lock); | |
1713 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1714 | } | |
1715 | ||
1716 | /* | |
1717 | * double_rq_unlock - safely unlock two runqueues | |
1718 | * | |
1719 | * Note this does not restore interrupts like task_rq_unlock, | |
1720 | * you need to do so manually after calling. | |
1721 | */ | |
1722 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1723 | __releases(rq1->lock) | |
1724 | __releases(rq2->lock) | |
1725 | { | |
1726 | BUG_ON(rq1 != rq2); | |
1727 | raw_spin_unlock(&rq1->lock); | |
1728 | __release(rq2->lock); | |
1729 | } | |
1730 | ||
18d95a28 PZ |
1731 | #endif |
1732 | ||
74f5187a | 1733 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1734 | static void update_sysctl(void); |
acb4a848 | 1735 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1736 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1737 | |
cd29fe6f PZ |
1738 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1739 | { | |
1740 | set_task_rq(p, cpu); | |
1741 | #ifdef CONFIG_SMP | |
1742 | /* | |
1743 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1744 | * successfuly executed on another CPU. We must ensure that updates of | |
1745 | * per-task data have been completed by this moment. | |
1746 | */ | |
1747 | smp_wmb(); | |
1748 | task_thread_info(p)->cpu = cpu; | |
1749 | #endif | |
1750 | } | |
dce48a84 | 1751 | |
1e3c88bd | 1752 | static const struct sched_class rt_sched_class; |
dd41f596 | 1753 | |
34f971f6 | 1754 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1755 | #define for_each_class(class) \ |
1756 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1757 | |
1e3c88bd PZ |
1758 | #include "sched_stats.h" |
1759 | ||
c09595f6 | 1760 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1761 | { |
1762 | rq->nr_running++; | |
9c217245 IM |
1763 | } |
1764 | ||
c09595f6 | 1765 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1766 | { |
1767 | rq->nr_running--; | |
9c217245 IM |
1768 | } |
1769 | ||
45bf76df IM |
1770 | static void set_load_weight(struct task_struct *p) |
1771 | { | |
dd41f596 IM |
1772 | /* |
1773 | * SCHED_IDLE tasks get minimal weight: | |
1774 | */ | |
1775 | if (p->policy == SCHED_IDLE) { | |
1776 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1777 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1778 | return; | |
1779 | } | |
71f8bd46 | 1780 | |
dd41f596 IM |
1781 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1782 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1783 | } |
1784 | ||
371fd7e7 | 1785 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1786 | { |
a64692a3 | 1787 | update_rq_clock(rq); |
dd41f596 | 1788 | sched_info_queued(p); |
371fd7e7 | 1789 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1790 | } |
1791 | ||
371fd7e7 | 1792 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1793 | { |
a64692a3 | 1794 | update_rq_clock(rq); |
46ac22ba | 1795 | sched_info_dequeued(p); |
371fd7e7 | 1796 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1797 | } |
1798 | ||
1e3c88bd PZ |
1799 | /* |
1800 | * activate_task - move a task to the runqueue. | |
1801 | */ | |
371fd7e7 | 1802 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1803 | { |
1804 | if (task_contributes_to_load(p)) | |
1805 | rq->nr_uninterruptible--; | |
1806 | ||
371fd7e7 | 1807 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1808 | inc_nr_running(rq); |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * deactivate_task - remove a task from the runqueue. | |
1813 | */ | |
371fd7e7 | 1814 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1815 | { |
1816 | if (task_contributes_to_load(p)) | |
1817 | rq->nr_uninterruptible++; | |
1818 | ||
371fd7e7 | 1819 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1820 | dec_nr_running(rq); |
1821 | } | |
1822 | ||
b52bfee4 VP |
1823 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1824 | ||
305e6835 VP |
1825 | /* |
1826 | * There are no locks covering percpu hardirq/softirq time. | |
1827 | * They are only modified in account_system_vtime, on corresponding CPU | |
1828 | * with interrupts disabled. So, writes are safe. | |
1829 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1830 | * This may result in other CPU reading this CPU's irq time and can | |
1831 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1832 | * or new value with a side effect of accounting a slice of irq time to wrong |
1833 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1834 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1835 | */ |
b52bfee4 VP |
1836 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1837 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1838 | ||
1839 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1840 | static int sched_clock_irqtime; | |
1841 | ||
1842 | void enable_sched_clock_irqtime(void) | |
1843 | { | |
1844 | sched_clock_irqtime = 1; | |
1845 | } | |
1846 | ||
1847 | void disable_sched_clock_irqtime(void) | |
1848 | { | |
1849 | sched_clock_irqtime = 0; | |
1850 | } | |
1851 | ||
8e92c201 PZ |
1852 | #ifndef CONFIG_64BIT |
1853 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1854 | ||
1855 | static inline void irq_time_write_begin(void) | |
1856 | { | |
1857 | __this_cpu_inc(irq_time_seq.sequence); | |
1858 | smp_wmb(); | |
1859 | } | |
1860 | ||
1861 | static inline void irq_time_write_end(void) | |
1862 | { | |
1863 | smp_wmb(); | |
1864 | __this_cpu_inc(irq_time_seq.sequence); | |
1865 | } | |
1866 | ||
1867 | static inline u64 irq_time_read(int cpu) | |
1868 | { | |
1869 | u64 irq_time; | |
1870 | unsigned seq; | |
1871 | ||
1872 | do { | |
1873 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1874 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1875 | per_cpu(cpu_hardirq_time, cpu); | |
1876 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1877 | ||
1878 | return irq_time; | |
1879 | } | |
1880 | #else /* CONFIG_64BIT */ | |
1881 | static inline void irq_time_write_begin(void) | |
1882 | { | |
1883 | } | |
1884 | ||
1885 | static inline void irq_time_write_end(void) | |
1886 | { | |
1887 | } | |
1888 | ||
1889 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1890 | { |
305e6835 VP |
1891 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1892 | } | |
8e92c201 | 1893 | #endif /* CONFIG_64BIT */ |
305e6835 | 1894 | |
fe44d621 PZ |
1895 | /* |
1896 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1897 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1898 | */ | |
b52bfee4 VP |
1899 | void account_system_vtime(struct task_struct *curr) |
1900 | { | |
1901 | unsigned long flags; | |
fe44d621 | 1902 | s64 delta; |
b52bfee4 | 1903 | int cpu; |
b52bfee4 VP |
1904 | |
1905 | if (!sched_clock_irqtime) | |
1906 | return; | |
1907 | ||
1908 | local_irq_save(flags); | |
1909 | ||
b52bfee4 | 1910 | cpu = smp_processor_id(); |
fe44d621 PZ |
1911 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1912 | __this_cpu_add(irq_start_time, delta); | |
1913 | ||
8e92c201 | 1914 | irq_time_write_begin(); |
b52bfee4 VP |
1915 | /* |
1916 | * We do not account for softirq time from ksoftirqd here. | |
1917 | * We want to continue accounting softirq time to ksoftirqd thread | |
1918 | * in that case, so as not to confuse scheduler with a special task | |
1919 | * that do not consume any time, but still wants to run. | |
1920 | */ | |
1921 | if (hardirq_count()) | |
fe44d621 | 1922 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1923 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1924 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1925 | |
8e92c201 | 1926 | irq_time_write_end(); |
b52bfee4 VP |
1927 | local_irq_restore(flags); |
1928 | } | |
b7dadc38 | 1929 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1930 | |
fe44d621 | 1931 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1932 | { |
fe44d621 PZ |
1933 | s64 irq_delta; |
1934 | ||
8e92c201 | 1935 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1936 | |
1937 | /* | |
1938 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1939 | * this case when a previous update_rq_clock() happened inside a | |
1940 | * {soft,}irq region. | |
1941 | * | |
1942 | * When this happens, we stop ->clock_task and only update the | |
1943 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1944 | * update will consume the rest. This ensures ->clock_task is | |
1945 | * monotonic. | |
1946 | * | |
1947 | * It does however cause some slight miss-attribution of {soft,}irq | |
1948 | * time, a more accurate solution would be to update the irq_time using | |
1949 | * the current rq->clock timestamp, except that would require using | |
1950 | * atomic ops. | |
1951 | */ | |
1952 | if (irq_delta > delta) | |
1953 | irq_delta = delta; | |
1954 | ||
1955 | rq->prev_irq_time += irq_delta; | |
1956 | delta -= irq_delta; | |
1957 | rq->clock_task += delta; | |
1958 | ||
1959 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1960 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1961 | } |
1962 | ||
abb74cef VP |
1963 | static int irqtime_account_hi_update(void) |
1964 | { | |
1965 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1966 | unsigned long flags; | |
1967 | u64 latest_ns; | |
1968 | int ret = 0; | |
1969 | ||
1970 | local_irq_save(flags); | |
1971 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1972 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1973 | ret = 1; | |
1974 | local_irq_restore(flags); | |
1975 | return ret; | |
1976 | } | |
1977 | ||
1978 | static int irqtime_account_si_update(void) | |
1979 | { | |
1980 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1981 | unsigned long flags; | |
1982 | u64 latest_ns; | |
1983 | int ret = 0; | |
1984 | ||
1985 | local_irq_save(flags); | |
1986 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1987 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1988 | ret = 1; | |
1989 | local_irq_restore(flags); | |
1990 | return ret; | |
1991 | } | |
1992 | ||
fe44d621 | 1993 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1994 | |
abb74cef VP |
1995 | #define sched_clock_irqtime (0) |
1996 | ||
fe44d621 | 1997 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1998 | { |
fe44d621 | 1999 | rq->clock_task += delta; |
305e6835 VP |
2000 | } |
2001 | ||
fe44d621 | 2002 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 2003 | |
1e3c88bd PZ |
2004 | #include "sched_idletask.c" |
2005 | #include "sched_fair.c" | |
2006 | #include "sched_rt.c" | |
5091faa4 | 2007 | #include "sched_autogroup.c" |
34f971f6 | 2008 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2009 | #ifdef CONFIG_SCHED_DEBUG |
2010 | # include "sched_debug.c" | |
2011 | #endif | |
2012 | ||
34f971f6 PZ |
2013 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2014 | { | |
2015 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2016 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2017 | ||
2018 | if (stop) { | |
2019 | /* | |
2020 | * Make it appear like a SCHED_FIFO task, its something | |
2021 | * userspace knows about and won't get confused about. | |
2022 | * | |
2023 | * Also, it will make PI more or less work without too | |
2024 | * much confusion -- but then, stop work should not | |
2025 | * rely on PI working anyway. | |
2026 | */ | |
2027 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2028 | ||
2029 | stop->sched_class = &stop_sched_class; | |
2030 | } | |
2031 | ||
2032 | cpu_rq(cpu)->stop = stop; | |
2033 | ||
2034 | if (old_stop) { | |
2035 | /* | |
2036 | * Reset it back to a normal scheduling class so that | |
2037 | * it can die in pieces. | |
2038 | */ | |
2039 | old_stop->sched_class = &rt_sched_class; | |
2040 | } | |
2041 | } | |
2042 | ||
14531189 | 2043 | /* |
dd41f596 | 2044 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2045 | */ |
14531189 IM |
2046 | static inline int __normal_prio(struct task_struct *p) |
2047 | { | |
dd41f596 | 2048 | return p->static_prio; |
14531189 IM |
2049 | } |
2050 | ||
b29739f9 IM |
2051 | /* |
2052 | * Calculate the expected normal priority: i.e. priority | |
2053 | * without taking RT-inheritance into account. Might be | |
2054 | * boosted by interactivity modifiers. Changes upon fork, | |
2055 | * setprio syscalls, and whenever the interactivity | |
2056 | * estimator recalculates. | |
2057 | */ | |
36c8b586 | 2058 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2059 | { |
2060 | int prio; | |
2061 | ||
e05606d3 | 2062 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2063 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2064 | else | |
2065 | prio = __normal_prio(p); | |
2066 | return prio; | |
2067 | } | |
2068 | ||
2069 | /* | |
2070 | * Calculate the current priority, i.e. the priority | |
2071 | * taken into account by the scheduler. This value might | |
2072 | * be boosted by RT tasks, or might be boosted by | |
2073 | * interactivity modifiers. Will be RT if the task got | |
2074 | * RT-boosted. If not then it returns p->normal_prio. | |
2075 | */ | |
36c8b586 | 2076 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2077 | { |
2078 | p->normal_prio = normal_prio(p); | |
2079 | /* | |
2080 | * If we are RT tasks or we were boosted to RT priority, | |
2081 | * keep the priority unchanged. Otherwise, update priority | |
2082 | * to the normal priority: | |
2083 | */ | |
2084 | if (!rt_prio(p->prio)) | |
2085 | return p->normal_prio; | |
2086 | return p->prio; | |
2087 | } | |
2088 | ||
1da177e4 LT |
2089 | /** |
2090 | * task_curr - is this task currently executing on a CPU? | |
2091 | * @p: the task in question. | |
2092 | */ | |
36c8b586 | 2093 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2094 | { |
2095 | return cpu_curr(task_cpu(p)) == p; | |
2096 | } | |
2097 | ||
cb469845 SR |
2098 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2099 | const struct sched_class *prev_class, | |
da7a735e | 2100 | int oldprio) |
cb469845 SR |
2101 | { |
2102 | if (prev_class != p->sched_class) { | |
2103 | if (prev_class->switched_from) | |
da7a735e PZ |
2104 | prev_class->switched_from(rq, p); |
2105 | p->sched_class->switched_to(rq, p); | |
2106 | } else if (oldprio != p->prio) | |
2107 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2108 | } |
2109 | ||
1e5a7405 PZ |
2110 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2111 | { | |
2112 | const struct sched_class *class; | |
2113 | ||
2114 | if (p->sched_class == rq->curr->sched_class) { | |
2115 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2116 | } else { | |
2117 | for_each_class(class) { | |
2118 | if (class == rq->curr->sched_class) | |
2119 | break; | |
2120 | if (class == p->sched_class) { | |
2121 | resched_task(rq->curr); | |
2122 | break; | |
2123 | } | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * A queue event has occurred, and we're going to schedule. In | |
2129 | * this case, we can save a useless back to back clock update. | |
2130 | */ | |
fd2f4419 | 2131 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2132 | rq->skip_clock_update = 1; |
2133 | } | |
2134 | ||
1da177e4 | 2135 | #ifdef CONFIG_SMP |
cc367732 IM |
2136 | /* |
2137 | * Is this task likely cache-hot: | |
2138 | */ | |
e7693a36 | 2139 | static int |
cc367732 IM |
2140 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2141 | { | |
2142 | s64 delta; | |
2143 | ||
e6c8fba7 PZ |
2144 | if (p->sched_class != &fair_sched_class) |
2145 | return 0; | |
2146 | ||
ef8002f6 NR |
2147 | if (unlikely(p->policy == SCHED_IDLE)) |
2148 | return 0; | |
2149 | ||
f540a608 IM |
2150 | /* |
2151 | * Buddy candidates are cache hot: | |
2152 | */ | |
f685ceac | 2153 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2154 | (&p->se == cfs_rq_of(&p->se)->next || |
2155 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2156 | return 1; |
2157 | ||
6bc1665b IM |
2158 | if (sysctl_sched_migration_cost == -1) |
2159 | return 1; | |
2160 | if (sysctl_sched_migration_cost == 0) | |
2161 | return 0; | |
2162 | ||
cc367732 IM |
2163 | delta = now - p->se.exec_start; |
2164 | ||
2165 | return delta < (s64)sysctl_sched_migration_cost; | |
2166 | } | |
2167 | ||
dd41f596 | 2168 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2169 | { |
e2912009 PZ |
2170 | #ifdef CONFIG_SCHED_DEBUG |
2171 | /* | |
2172 | * We should never call set_task_cpu() on a blocked task, | |
2173 | * ttwu() will sort out the placement. | |
2174 | */ | |
077614ee PZ |
2175 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2176 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
2177 | |
2178 | #ifdef CONFIG_LOCKDEP | |
2179 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || | |
2180 | lockdep_is_held(&task_rq(p)->lock))); | |
2181 | #endif | |
e2912009 PZ |
2182 | #endif |
2183 | ||
de1d7286 | 2184 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2185 | |
0c69774e PZ |
2186 | if (task_cpu(p) != new_cpu) { |
2187 | p->se.nr_migrations++; | |
2188 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2189 | } | |
dd41f596 IM |
2190 | |
2191 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2192 | } |
2193 | ||
969c7921 | 2194 | struct migration_arg { |
36c8b586 | 2195 | struct task_struct *task; |
1da177e4 | 2196 | int dest_cpu; |
70b97a7f | 2197 | }; |
1da177e4 | 2198 | |
969c7921 TH |
2199 | static int migration_cpu_stop(void *data); |
2200 | ||
1da177e4 LT |
2201 | /* |
2202 | * The task's runqueue lock must be held. | |
2203 | * Returns true if you have to wait for migration thread. | |
2204 | */ | |
7608dec2 | 2205 | static bool need_migrate_task(struct task_struct *p) |
1da177e4 | 2206 | { |
1da177e4 LT |
2207 | /* |
2208 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2209 | * the next wake-up will properly place the task. |
1da177e4 | 2210 | */ |
7608dec2 PZ |
2211 | bool running = p->on_rq || p->on_cpu; |
2212 | smp_rmb(); /* finish_lock_switch() */ | |
2213 | return running; | |
1da177e4 LT |
2214 | } |
2215 | ||
2216 | /* | |
2217 | * wait_task_inactive - wait for a thread to unschedule. | |
2218 | * | |
85ba2d86 RM |
2219 | * If @match_state is nonzero, it's the @p->state value just checked and |
2220 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2221 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2222 | * we return a positive number (its total switch count). If a second call | |
2223 | * a short while later returns the same number, the caller can be sure that | |
2224 | * @p has remained unscheduled the whole time. | |
2225 | * | |
1da177e4 LT |
2226 | * The caller must ensure that the task *will* unschedule sometime soon, |
2227 | * else this function might spin for a *long* time. This function can't | |
2228 | * be called with interrupts off, or it may introduce deadlock with | |
2229 | * smp_call_function() if an IPI is sent by the same process we are | |
2230 | * waiting to become inactive. | |
2231 | */ | |
85ba2d86 | 2232 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2233 | { |
2234 | unsigned long flags; | |
dd41f596 | 2235 | int running, on_rq; |
85ba2d86 | 2236 | unsigned long ncsw; |
70b97a7f | 2237 | struct rq *rq; |
1da177e4 | 2238 | |
3a5c359a AK |
2239 | for (;;) { |
2240 | /* | |
2241 | * We do the initial early heuristics without holding | |
2242 | * any task-queue locks at all. We'll only try to get | |
2243 | * the runqueue lock when things look like they will | |
2244 | * work out! | |
2245 | */ | |
2246 | rq = task_rq(p); | |
fa490cfd | 2247 | |
3a5c359a AK |
2248 | /* |
2249 | * If the task is actively running on another CPU | |
2250 | * still, just relax and busy-wait without holding | |
2251 | * any locks. | |
2252 | * | |
2253 | * NOTE! Since we don't hold any locks, it's not | |
2254 | * even sure that "rq" stays as the right runqueue! | |
2255 | * But we don't care, since "task_running()" will | |
2256 | * return false if the runqueue has changed and p | |
2257 | * is actually now running somewhere else! | |
2258 | */ | |
85ba2d86 RM |
2259 | while (task_running(rq, p)) { |
2260 | if (match_state && unlikely(p->state != match_state)) | |
2261 | return 0; | |
3a5c359a | 2262 | cpu_relax(); |
85ba2d86 | 2263 | } |
fa490cfd | 2264 | |
3a5c359a AK |
2265 | /* |
2266 | * Ok, time to look more closely! We need the rq | |
2267 | * lock now, to be *sure*. If we're wrong, we'll | |
2268 | * just go back and repeat. | |
2269 | */ | |
2270 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2271 | trace_sched_wait_task(p); |
3a5c359a | 2272 | running = task_running(rq, p); |
fd2f4419 | 2273 | on_rq = p->on_rq; |
85ba2d86 | 2274 | ncsw = 0; |
f31e11d8 | 2275 | if (!match_state || p->state == match_state) |
93dcf55f | 2276 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 2277 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 2278 | |
85ba2d86 RM |
2279 | /* |
2280 | * If it changed from the expected state, bail out now. | |
2281 | */ | |
2282 | if (unlikely(!ncsw)) | |
2283 | break; | |
2284 | ||
3a5c359a AK |
2285 | /* |
2286 | * Was it really running after all now that we | |
2287 | * checked with the proper locks actually held? | |
2288 | * | |
2289 | * Oops. Go back and try again.. | |
2290 | */ | |
2291 | if (unlikely(running)) { | |
2292 | cpu_relax(); | |
2293 | continue; | |
2294 | } | |
fa490cfd | 2295 | |
3a5c359a AK |
2296 | /* |
2297 | * It's not enough that it's not actively running, | |
2298 | * it must be off the runqueue _entirely_, and not | |
2299 | * preempted! | |
2300 | * | |
80dd99b3 | 2301 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2302 | * running right now), it's preempted, and we should |
2303 | * yield - it could be a while. | |
2304 | */ | |
2305 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2306 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2307 | ||
2308 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2309 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2310 | continue; |
2311 | } | |
fa490cfd | 2312 | |
3a5c359a AK |
2313 | /* |
2314 | * Ahh, all good. It wasn't running, and it wasn't | |
2315 | * runnable, which means that it will never become | |
2316 | * running in the future either. We're all done! | |
2317 | */ | |
2318 | break; | |
2319 | } | |
85ba2d86 RM |
2320 | |
2321 | return ncsw; | |
1da177e4 LT |
2322 | } |
2323 | ||
2324 | /*** | |
2325 | * kick_process - kick a running thread to enter/exit the kernel | |
2326 | * @p: the to-be-kicked thread | |
2327 | * | |
2328 | * Cause a process which is running on another CPU to enter | |
2329 | * kernel-mode, without any delay. (to get signals handled.) | |
2330 | * | |
25985edc | 2331 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2332 | * because all it wants to ensure is that the remote task enters |
2333 | * the kernel. If the IPI races and the task has been migrated | |
2334 | * to another CPU then no harm is done and the purpose has been | |
2335 | * achieved as well. | |
2336 | */ | |
36c8b586 | 2337 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2338 | { |
2339 | int cpu; | |
2340 | ||
2341 | preempt_disable(); | |
2342 | cpu = task_cpu(p); | |
2343 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2344 | smp_send_reschedule(cpu); | |
2345 | preempt_enable(); | |
2346 | } | |
b43e3521 | 2347 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2348 | #endif /* CONFIG_SMP */ |
1da177e4 | 2349 | |
970b13ba | 2350 | #ifdef CONFIG_SMP |
30da688e | 2351 | /* |
013fdb80 | 2352 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 2353 | */ |
5da9a0fb PZ |
2354 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2355 | { | |
2356 | int dest_cpu; | |
2357 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2358 | ||
2359 | /* Look for allowed, online CPU in same node. */ | |
2360 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2361 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2362 | return dest_cpu; | |
2363 | ||
2364 | /* Any allowed, online CPU? */ | |
2365 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2366 | if (dest_cpu < nr_cpu_ids) | |
2367 | return dest_cpu; | |
2368 | ||
2369 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2370 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2371 | /* | |
2372 | * Don't tell them about moving exiting tasks or | |
2373 | * kernel threads (both mm NULL), since they never | |
2374 | * leave kernel. | |
2375 | */ | |
2376 | if (p->mm && printk_ratelimit()) { | |
2377 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2378 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2379 | } |
2380 | ||
2381 | return dest_cpu; | |
2382 | } | |
2383 | ||
e2912009 | 2384 | /* |
013fdb80 | 2385 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 2386 | */ |
970b13ba | 2387 | static inline |
7608dec2 | 2388 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2389 | { |
7608dec2 | 2390 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
2391 | |
2392 | /* | |
2393 | * In order not to call set_task_cpu() on a blocking task we need | |
2394 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2395 | * cpu. | |
2396 | * | |
2397 | * Since this is common to all placement strategies, this lives here. | |
2398 | * | |
2399 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2400 | * not worry about this generic constraint ] | |
2401 | */ | |
2402 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2403 | !cpu_online(cpu))) |
5da9a0fb | 2404 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2405 | |
2406 | return cpu; | |
970b13ba | 2407 | } |
09a40af5 MG |
2408 | |
2409 | static void update_avg(u64 *avg, u64 sample) | |
2410 | { | |
2411 | s64 diff = sample - *avg; | |
2412 | *avg += diff >> 3; | |
2413 | } | |
970b13ba PZ |
2414 | #endif |
2415 | ||
d7c01d27 | 2416 | static void |
b84cb5df | 2417 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2418 | { |
d7c01d27 | 2419 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
2420 | struct rq *rq = this_rq(); |
2421 | ||
d7c01d27 PZ |
2422 | #ifdef CONFIG_SMP |
2423 | int this_cpu = smp_processor_id(); | |
2424 | ||
2425 | if (cpu == this_cpu) { | |
2426 | schedstat_inc(rq, ttwu_local); | |
2427 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2428 | } else { | |
2429 | struct sched_domain *sd; | |
2430 | ||
2431 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2432 | for_each_domain(this_cpu, sd) { | |
2433 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2434 | schedstat_inc(sd, ttwu_wake_remote); | |
2435 | break; | |
2436 | } | |
2437 | } | |
2438 | } | |
2439 | #endif /* CONFIG_SMP */ | |
2440 | ||
2441 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2442 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2443 | |
2444 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2445 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2446 | |
2447 | if (cpu != task_cpu(p)) | |
9ed3811a | 2448 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
9ed3811a | 2449 | |
d7c01d27 PZ |
2450 | #endif /* CONFIG_SCHEDSTATS */ |
2451 | } | |
2452 | ||
2453 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2454 | { | |
9ed3811a | 2455 | activate_task(rq, p, en_flags); |
fd2f4419 | 2456 | p->on_rq = 1; |
c2f7115e PZ |
2457 | |
2458 | /* if a worker is waking up, notify workqueue */ | |
2459 | if (p->flags & PF_WQ_WORKER) | |
2460 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2461 | } |
2462 | ||
23f41eeb PZ |
2463 | /* |
2464 | * Mark the task runnable and perform wakeup-preemption. | |
2465 | */ | |
89363381 | 2466 | static void |
23f41eeb | 2467 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 2468 | { |
89363381 | 2469 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2470 | check_preempt_curr(rq, p, wake_flags); |
2471 | ||
2472 | p->state = TASK_RUNNING; | |
2473 | #ifdef CONFIG_SMP | |
2474 | if (p->sched_class->task_woken) | |
2475 | p->sched_class->task_woken(rq, p); | |
2476 | ||
2477 | if (unlikely(rq->idle_stamp)) { | |
2478 | u64 delta = rq->clock - rq->idle_stamp; | |
2479 | u64 max = 2*sysctl_sched_migration_cost; | |
2480 | ||
2481 | if (delta > max) | |
2482 | rq->avg_idle = max; | |
2483 | else | |
2484 | update_avg(&rq->avg_idle, delta); | |
2485 | rq->idle_stamp = 0; | |
2486 | } | |
2487 | #endif | |
2488 | } | |
2489 | ||
c05fbafb PZ |
2490 | static void |
2491 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
2492 | { | |
2493 | #ifdef CONFIG_SMP | |
2494 | if (p->sched_contributes_to_load) | |
2495 | rq->nr_uninterruptible--; | |
2496 | #endif | |
2497 | ||
2498 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
2499 | ttwu_do_wakeup(rq, p, wake_flags); | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2504 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2505 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2506 | * the task is still ->on_rq. | |
2507 | */ | |
2508 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2509 | { | |
2510 | struct rq *rq; | |
2511 | int ret = 0; | |
2512 | ||
2513 | rq = __task_rq_lock(p); | |
2514 | if (p->on_rq) { | |
2515 | ttwu_do_wakeup(rq, p, wake_flags); | |
2516 | ret = 1; | |
2517 | } | |
2518 | __task_rq_unlock(rq); | |
2519 | ||
2520 | return ret; | |
2521 | } | |
2522 | ||
317f3941 PZ |
2523 | #ifdef CONFIG_SMP |
2524 | static void sched_ttwu_pending(void) | |
2525 | { | |
2526 | struct rq *rq = this_rq(); | |
2527 | struct task_struct *list = xchg(&rq->wake_list, NULL); | |
2528 | ||
2529 | if (!list) | |
2530 | return; | |
2531 | ||
2532 | raw_spin_lock(&rq->lock); | |
2533 | ||
2534 | while (list) { | |
2535 | struct task_struct *p = list; | |
2536 | list = list->wake_entry; | |
2537 | ttwu_do_activate(rq, p, 0); | |
2538 | } | |
2539 | ||
2540 | raw_spin_unlock(&rq->lock); | |
2541 | } | |
2542 | ||
2543 | void scheduler_ipi(void) | |
2544 | { | |
2545 | sched_ttwu_pending(); | |
2546 | } | |
2547 | ||
2548 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
2549 | { | |
2550 | struct rq *rq = cpu_rq(cpu); | |
2551 | struct task_struct *next = rq->wake_list; | |
2552 | ||
2553 | for (;;) { | |
2554 | struct task_struct *old = next; | |
2555 | ||
2556 | p->wake_entry = next; | |
2557 | next = cmpxchg(&rq->wake_list, old, p); | |
2558 | if (next == old) | |
2559 | break; | |
2560 | } | |
2561 | ||
2562 | if (!next) | |
2563 | smp_send_reschedule(cpu); | |
2564 | } | |
2565 | #endif | |
2566 | ||
c05fbafb PZ |
2567 | static void ttwu_queue(struct task_struct *p, int cpu) |
2568 | { | |
2569 | struct rq *rq = cpu_rq(cpu); | |
2570 | ||
317f3941 PZ |
2571 | #if defined(CONFIG_SMP) && defined(CONFIG_SCHED_TTWU_QUEUE) |
2572 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { | |
2573 | ttwu_queue_remote(p, cpu); | |
2574 | return; | |
2575 | } | |
2576 | #endif | |
2577 | ||
c05fbafb PZ |
2578 | raw_spin_lock(&rq->lock); |
2579 | ttwu_do_activate(rq, p, 0); | |
2580 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
2581 | } |
2582 | ||
2583 | /** | |
1da177e4 | 2584 | * try_to_wake_up - wake up a thread |
9ed3811a | 2585 | * @p: the thread to be awakened |
1da177e4 | 2586 | * @state: the mask of task states that can be woken |
9ed3811a | 2587 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2588 | * |
2589 | * Put it on the run-queue if it's not already there. The "current" | |
2590 | * thread is always on the run-queue (except when the actual | |
2591 | * re-schedule is in progress), and as such you're allowed to do | |
2592 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2593 | * runnable without the overhead of this. | |
2594 | * | |
9ed3811a TH |
2595 | * Returns %true if @p was woken up, %false if it was already running |
2596 | * or @state didn't match @p's state. | |
1da177e4 | 2597 | */ |
e4a52bcb PZ |
2598 | static int |
2599 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2600 | { |
1da177e4 | 2601 | unsigned long flags; |
c05fbafb | 2602 | int cpu, success = 0; |
2398f2c6 | 2603 | |
04e2f174 | 2604 | smp_wmb(); |
013fdb80 | 2605 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2606 | if (!(p->state & state)) |
1da177e4 LT |
2607 | goto out; |
2608 | ||
c05fbafb | 2609 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2610 | cpu = task_cpu(p); |
1da177e4 | 2611 | |
c05fbafb PZ |
2612 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2613 | goto stat; | |
1da177e4 | 2614 | |
1da177e4 | 2615 | #ifdef CONFIG_SMP |
e9c84311 | 2616 | /* |
c05fbafb PZ |
2617 | * If the owning (remote) cpu is still in the middle of schedule() with |
2618 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 2619 | */ |
e4a52bcb PZ |
2620 | while (p->on_cpu) { |
2621 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2622 | /* | |
2623 | * If called from interrupt context we could have landed in the | |
2624 | * middle of schedule(), in this case we should take care not | |
2625 | * to spin on ->on_cpu if p is current, since that would | |
2626 | * deadlock. | |
2627 | */ | |
c05fbafb PZ |
2628 | if (p == current) { |
2629 | ttwu_queue(p, cpu); | |
2630 | goto stat; | |
2631 | } | |
e4a52bcb PZ |
2632 | #endif |
2633 | cpu_relax(); | |
371fd7e7 | 2634 | } |
0970d299 | 2635 | /* |
e4a52bcb | 2636 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 2637 | */ |
e4a52bcb | 2638 | smp_rmb(); |
1da177e4 | 2639 | |
a8e4f2ea | 2640 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2641 | p->state = TASK_WAKING; |
efbbd05a | 2642 | |
e4a52bcb | 2643 | if (p->sched_class->task_waking) |
74f8e4b2 | 2644 | p->sched_class->task_waking(p); |
e7693a36 | 2645 | |
7608dec2 | 2646 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
c05fbafb | 2647 | if (task_cpu(p) != cpu) |
e4a52bcb | 2648 | set_task_cpu(p, cpu); |
1da177e4 | 2649 | #endif /* CONFIG_SMP */ |
1da177e4 | 2650 | |
c05fbafb PZ |
2651 | ttwu_queue(p, cpu); |
2652 | stat: | |
b84cb5df | 2653 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2654 | out: |
013fdb80 | 2655 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2656 | |
2657 | return success; | |
2658 | } | |
2659 | ||
21aa9af0 TH |
2660 | /** |
2661 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2662 | * @p: the thread to be awakened | |
2663 | * | |
2acca55e | 2664 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2665 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2666 | * the current task. |
21aa9af0 TH |
2667 | */ |
2668 | static void try_to_wake_up_local(struct task_struct *p) | |
2669 | { | |
2670 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2671 | |
2672 | BUG_ON(rq != this_rq()); | |
2673 | BUG_ON(p == current); | |
2674 | lockdep_assert_held(&rq->lock); | |
2675 | ||
2acca55e PZ |
2676 | if (!raw_spin_trylock(&p->pi_lock)) { |
2677 | raw_spin_unlock(&rq->lock); | |
2678 | raw_spin_lock(&p->pi_lock); | |
2679 | raw_spin_lock(&rq->lock); | |
2680 | } | |
2681 | ||
21aa9af0 | 2682 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2683 | goto out; |
21aa9af0 | 2684 | |
fd2f4419 | 2685 | if (!p->on_rq) |
d7c01d27 PZ |
2686 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2687 | ||
23f41eeb | 2688 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2689 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2690 | out: |
2691 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2692 | } |
2693 | ||
50fa610a DH |
2694 | /** |
2695 | * wake_up_process - Wake up a specific process | |
2696 | * @p: The process to be woken up. | |
2697 | * | |
2698 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2699 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2700 | * running. | |
2701 | * | |
2702 | * It may be assumed that this function implies a write memory barrier before | |
2703 | * changing the task state if and only if any tasks are woken up. | |
2704 | */ | |
7ad5b3a5 | 2705 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2706 | { |
d9514f6c | 2707 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2708 | } |
1da177e4 LT |
2709 | EXPORT_SYMBOL(wake_up_process); |
2710 | ||
7ad5b3a5 | 2711 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2712 | { |
2713 | return try_to_wake_up(p, state, 0); | |
2714 | } | |
2715 | ||
1da177e4 LT |
2716 | /* |
2717 | * Perform scheduler related setup for a newly forked process p. | |
2718 | * p is forked by current. | |
dd41f596 IM |
2719 | * |
2720 | * __sched_fork() is basic setup used by init_idle() too: | |
2721 | */ | |
2722 | static void __sched_fork(struct task_struct *p) | |
2723 | { | |
fd2f4419 PZ |
2724 | p->on_rq = 0; |
2725 | ||
2726 | p->se.on_rq = 0; | |
dd41f596 IM |
2727 | p->se.exec_start = 0; |
2728 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2729 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2730 | p->se.nr_migrations = 0; |
da7a735e | 2731 | p->se.vruntime = 0; |
fd2f4419 | 2732 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2733 | |
2734 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2735 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2736 | #endif |
476d139c | 2737 | |
fa717060 | 2738 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2739 | |
e107be36 AK |
2740 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2741 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2742 | #endif | |
dd41f596 IM |
2743 | } |
2744 | ||
2745 | /* | |
2746 | * fork()/clone()-time setup: | |
2747 | */ | |
2748 | void sched_fork(struct task_struct *p, int clone_flags) | |
2749 | { | |
0122ec5b | 2750 | unsigned long flags; |
dd41f596 IM |
2751 | int cpu = get_cpu(); |
2752 | ||
2753 | __sched_fork(p); | |
06b83b5f | 2754 | /* |
0017d735 | 2755 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2756 | * nobody will actually run it, and a signal or other external |
2757 | * event cannot wake it up and insert it on the runqueue either. | |
2758 | */ | |
0017d735 | 2759 | p->state = TASK_RUNNING; |
dd41f596 | 2760 | |
b9dc29e7 MG |
2761 | /* |
2762 | * Revert to default priority/policy on fork if requested. | |
2763 | */ | |
2764 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2765 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2766 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2767 | p->normal_prio = p->static_prio; |
2768 | } | |
b9dc29e7 | 2769 | |
6c697bdf MG |
2770 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2771 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2772 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2773 | set_load_weight(p); |
2774 | } | |
2775 | ||
b9dc29e7 MG |
2776 | /* |
2777 | * We don't need the reset flag anymore after the fork. It has | |
2778 | * fulfilled its duty: | |
2779 | */ | |
2780 | p->sched_reset_on_fork = 0; | |
2781 | } | |
ca94c442 | 2782 | |
f83f9ac2 PW |
2783 | /* |
2784 | * Make sure we do not leak PI boosting priority to the child. | |
2785 | */ | |
2786 | p->prio = current->normal_prio; | |
2787 | ||
2ddbf952 HS |
2788 | if (!rt_prio(p->prio)) |
2789 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2790 | |
cd29fe6f PZ |
2791 | if (p->sched_class->task_fork) |
2792 | p->sched_class->task_fork(p); | |
2793 | ||
86951599 PZ |
2794 | /* |
2795 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2796 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2797 | * is ran before sched_fork(). | |
2798 | * | |
2799 | * Silence PROVE_RCU. | |
2800 | */ | |
0122ec5b | 2801 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2802 | set_task_cpu(p, cpu); |
0122ec5b | 2803 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2804 | |
52f17b6c | 2805 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2806 | if (likely(sched_info_on())) |
52f17b6c | 2807 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2808 | #endif |
3ca7a440 PZ |
2809 | #if defined(CONFIG_SMP) |
2810 | p->on_cpu = 0; | |
4866cde0 | 2811 | #endif |
1da177e4 | 2812 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2813 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2814 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2815 | #endif |
806c09a7 | 2816 | #ifdef CONFIG_SMP |
917b627d | 2817 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2818 | #endif |
917b627d | 2819 | |
476d139c | 2820 | put_cpu(); |
1da177e4 LT |
2821 | } |
2822 | ||
2823 | /* | |
2824 | * wake_up_new_task - wake up a newly created task for the first time. | |
2825 | * | |
2826 | * This function will do some initial scheduler statistics housekeeping | |
2827 | * that must be done for every newly created context, then puts the task | |
2828 | * on the runqueue and wakes it. | |
2829 | */ | |
7ad5b3a5 | 2830 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2831 | { |
2832 | unsigned long flags; | |
dd41f596 | 2833 | struct rq *rq; |
fabf318e | 2834 | |
ab2515c4 | 2835 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2836 | #ifdef CONFIG_SMP |
2837 | /* | |
2838 | * Fork balancing, do it here and not earlier because: | |
2839 | * - cpus_allowed can change in the fork path | |
2840 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2841 | */ |
ab2515c4 | 2842 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2843 | #endif |
2844 | ||
ab2515c4 | 2845 | rq = __task_rq_lock(p); |
cd29fe6f | 2846 | activate_task(rq, p, 0); |
fd2f4419 | 2847 | p->on_rq = 1; |
89363381 | 2848 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2849 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2850 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2851 | if (p->sched_class->task_woken) |
2852 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2853 | #endif |
0122ec5b | 2854 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2855 | } |
2856 | ||
e107be36 AK |
2857 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2858 | ||
2859 | /** | |
80dd99b3 | 2860 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2861 | * @notifier: notifier struct to register |
e107be36 AK |
2862 | */ |
2863 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2864 | { | |
2865 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2866 | } | |
2867 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2868 | ||
2869 | /** | |
2870 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2871 | * @notifier: notifier struct to unregister |
e107be36 AK |
2872 | * |
2873 | * This is safe to call from within a preemption notifier. | |
2874 | */ | |
2875 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2876 | { | |
2877 | hlist_del(¬ifier->link); | |
2878 | } | |
2879 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2880 | ||
2881 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2882 | { | |
2883 | struct preempt_notifier *notifier; | |
2884 | struct hlist_node *node; | |
2885 | ||
2886 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2887 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2888 | } | |
2889 | ||
2890 | static void | |
2891 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2892 | struct task_struct *next) | |
2893 | { | |
2894 | struct preempt_notifier *notifier; | |
2895 | struct hlist_node *node; | |
2896 | ||
2897 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2898 | notifier->ops->sched_out(notifier, next); | |
2899 | } | |
2900 | ||
6d6bc0ad | 2901 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2902 | |
2903 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2904 | { | |
2905 | } | |
2906 | ||
2907 | static void | |
2908 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2909 | struct task_struct *next) | |
2910 | { | |
2911 | } | |
2912 | ||
6d6bc0ad | 2913 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2914 | |
4866cde0 NP |
2915 | /** |
2916 | * prepare_task_switch - prepare to switch tasks | |
2917 | * @rq: the runqueue preparing to switch | |
421cee29 | 2918 | * @prev: the current task that is being switched out |
4866cde0 NP |
2919 | * @next: the task we are going to switch to. |
2920 | * | |
2921 | * This is called with the rq lock held and interrupts off. It must | |
2922 | * be paired with a subsequent finish_task_switch after the context | |
2923 | * switch. | |
2924 | * | |
2925 | * prepare_task_switch sets up locking and calls architecture specific | |
2926 | * hooks. | |
2927 | */ | |
e107be36 AK |
2928 | static inline void |
2929 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2930 | struct task_struct *next) | |
4866cde0 | 2931 | { |
fe4b04fa PZ |
2932 | sched_info_switch(prev, next); |
2933 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2934 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2935 | prepare_lock_switch(rq, next); |
2936 | prepare_arch_switch(next); | |
fe4b04fa | 2937 | trace_sched_switch(prev, next); |
4866cde0 NP |
2938 | } |
2939 | ||
1da177e4 LT |
2940 | /** |
2941 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2942 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2943 | * @prev: the thread we just switched away from. |
2944 | * | |
4866cde0 NP |
2945 | * finish_task_switch must be called after the context switch, paired |
2946 | * with a prepare_task_switch call before the context switch. | |
2947 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2948 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2949 | * |
2950 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2951 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2952 | * with the lock held can cause deadlocks; see schedule() for |
2953 | * details.) | |
2954 | */ | |
a9957449 | 2955 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2956 | __releases(rq->lock) |
2957 | { | |
1da177e4 | 2958 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2959 | long prev_state; |
1da177e4 LT |
2960 | |
2961 | rq->prev_mm = NULL; | |
2962 | ||
2963 | /* | |
2964 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2965 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2966 | * schedule one last time. The schedule call will never return, and |
2967 | * the scheduled task must drop that reference. | |
c394cc9f | 2968 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2969 | * still held, otherwise prev could be scheduled on another cpu, die |
2970 | * there before we look at prev->state, and then the reference would | |
2971 | * be dropped twice. | |
2972 | * Manfred Spraul <manfred@colorfullife.com> | |
2973 | */ | |
55a101f8 | 2974 | prev_state = prev->state; |
4866cde0 | 2975 | finish_arch_switch(prev); |
8381f65d JI |
2976 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2977 | local_irq_disable(); | |
2978 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2979 | perf_event_task_sched_in(current); |
8381f65d JI |
2980 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2981 | local_irq_enable(); | |
2982 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2983 | finish_lock_switch(rq, prev); |
e8fa1362 | 2984 | |
e107be36 | 2985 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2986 | if (mm) |
2987 | mmdrop(mm); | |
c394cc9f | 2988 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2989 | /* |
2990 | * Remove function-return probe instances associated with this | |
2991 | * task and put them back on the free list. | |
9761eea8 | 2992 | */ |
c6fd91f0 | 2993 | kprobe_flush_task(prev); |
1da177e4 | 2994 | put_task_struct(prev); |
c6fd91f0 | 2995 | } |
1da177e4 LT |
2996 | } |
2997 | ||
3f029d3c GH |
2998 | #ifdef CONFIG_SMP |
2999 | ||
3000 | /* assumes rq->lock is held */ | |
3001 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
3002 | { | |
3003 | if (prev->sched_class->pre_schedule) | |
3004 | prev->sched_class->pre_schedule(rq, prev); | |
3005 | } | |
3006 | ||
3007 | /* rq->lock is NOT held, but preemption is disabled */ | |
3008 | static inline void post_schedule(struct rq *rq) | |
3009 | { | |
3010 | if (rq->post_schedule) { | |
3011 | unsigned long flags; | |
3012 | ||
05fa785c | 3013 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
3014 | if (rq->curr->sched_class->post_schedule) |
3015 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 3016 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
3017 | |
3018 | rq->post_schedule = 0; | |
3019 | } | |
3020 | } | |
3021 | ||
3022 | #else | |
da19ab51 | 3023 | |
3f029d3c GH |
3024 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3025 | { | |
3026 | } | |
3027 | ||
3028 | static inline void post_schedule(struct rq *rq) | |
3029 | { | |
1da177e4 LT |
3030 | } |
3031 | ||
3f029d3c GH |
3032 | #endif |
3033 | ||
1da177e4 LT |
3034 | /** |
3035 | * schedule_tail - first thing a freshly forked thread must call. | |
3036 | * @prev: the thread we just switched away from. | |
3037 | */ | |
36c8b586 | 3038 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3039 | __releases(rq->lock) |
3040 | { | |
70b97a7f IM |
3041 | struct rq *rq = this_rq(); |
3042 | ||
4866cde0 | 3043 | finish_task_switch(rq, prev); |
da19ab51 | 3044 | |
3f029d3c GH |
3045 | /* |
3046 | * FIXME: do we need to worry about rq being invalidated by the | |
3047 | * task_switch? | |
3048 | */ | |
3049 | post_schedule(rq); | |
70b97a7f | 3050 | |
4866cde0 NP |
3051 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3052 | /* In this case, finish_task_switch does not reenable preemption */ | |
3053 | preempt_enable(); | |
3054 | #endif | |
1da177e4 | 3055 | if (current->set_child_tid) |
b488893a | 3056 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3057 | } |
3058 | ||
3059 | /* | |
3060 | * context_switch - switch to the new MM and the new | |
3061 | * thread's register state. | |
3062 | */ | |
dd41f596 | 3063 | static inline void |
70b97a7f | 3064 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3065 | struct task_struct *next) |
1da177e4 | 3066 | { |
dd41f596 | 3067 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3068 | |
e107be36 | 3069 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3070 | |
dd41f596 IM |
3071 | mm = next->mm; |
3072 | oldmm = prev->active_mm; | |
9226d125 ZA |
3073 | /* |
3074 | * For paravirt, this is coupled with an exit in switch_to to | |
3075 | * combine the page table reload and the switch backend into | |
3076 | * one hypercall. | |
3077 | */ | |
224101ed | 3078 | arch_start_context_switch(prev); |
9226d125 | 3079 | |
31915ab4 | 3080 | if (!mm) { |
1da177e4 LT |
3081 | next->active_mm = oldmm; |
3082 | atomic_inc(&oldmm->mm_count); | |
3083 | enter_lazy_tlb(oldmm, next); | |
3084 | } else | |
3085 | switch_mm(oldmm, mm, next); | |
3086 | ||
31915ab4 | 3087 | if (!prev->mm) { |
1da177e4 | 3088 | prev->active_mm = NULL; |
1da177e4 LT |
3089 | rq->prev_mm = oldmm; |
3090 | } | |
3a5f5e48 IM |
3091 | /* |
3092 | * Since the runqueue lock will be released by the next | |
3093 | * task (which is an invalid locking op but in the case | |
3094 | * of the scheduler it's an obvious special-case), so we | |
3095 | * do an early lockdep release here: | |
3096 | */ | |
3097 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3098 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3099 | #endif |
1da177e4 LT |
3100 | |
3101 | /* Here we just switch the register state and the stack. */ | |
3102 | switch_to(prev, next, prev); | |
3103 | ||
dd41f596 IM |
3104 | barrier(); |
3105 | /* | |
3106 | * this_rq must be evaluated again because prev may have moved | |
3107 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3108 | * frame will be invalid. | |
3109 | */ | |
3110 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3111 | } |
3112 | ||
3113 | /* | |
3114 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3115 | * | |
3116 | * externally visible scheduler statistics: current number of runnable | |
3117 | * threads, current number of uninterruptible-sleeping threads, total | |
3118 | * number of context switches performed since bootup. | |
3119 | */ | |
3120 | unsigned long nr_running(void) | |
3121 | { | |
3122 | unsigned long i, sum = 0; | |
3123 | ||
3124 | for_each_online_cpu(i) | |
3125 | sum += cpu_rq(i)->nr_running; | |
3126 | ||
3127 | return sum; | |
f711f609 | 3128 | } |
1da177e4 LT |
3129 | |
3130 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3131 | { |
1da177e4 | 3132 | unsigned long i, sum = 0; |
f711f609 | 3133 | |
0a945022 | 3134 | for_each_possible_cpu(i) |
1da177e4 | 3135 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3136 | |
3137 | /* | |
1da177e4 LT |
3138 | * Since we read the counters lockless, it might be slightly |
3139 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3140 | */ |
1da177e4 LT |
3141 | if (unlikely((long)sum < 0)) |
3142 | sum = 0; | |
f711f609 | 3143 | |
1da177e4 | 3144 | return sum; |
f711f609 | 3145 | } |
f711f609 | 3146 | |
1da177e4 | 3147 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3148 | { |
cc94abfc SR |
3149 | int i; |
3150 | unsigned long long sum = 0; | |
46cb4b7c | 3151 | |
0a945022 | 3152 | for_each_possible_cpu(i) |
1da177e4 | 3153 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3154 | |
1da177e4 LT |
3155 | return sum; |
3156 | } | |
483b4ee6 | 3157 | |
1da177e4 LT |
3158 | unsigned long nr_iowait(void) |
3159 | { | |
3160 | unsigned long i, sum = 0; | |
483b4ee6 | 3161 | |
0a945022 | 3162 | for_each_possible_cpu(i) |
1da177e4 | 3163 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3164 | |
1da177e4 LT |
3165 | return sum; |
3166 | } | |
483b4ee6 | 3167 | |
8c215bd3 | 3168 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3169 | { |
8c215bd3 | 3170 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3171 | return atomic_read(&this->nr_iowait); |
3172 | } | |
46cb4b7c | 3173 | |
69d25870 AV |
3174 | unsigned long this_cpu_load(void) |
3175 | { | |
3176 | struct rq *this = this_rq(); | |
3177 | return this->cpu_load[0]; | |
3178 | } | |
e790fb0b | 3179 | |
46cb4b7c | 3180 | |
dce48a84 TG |
3181 | /* Variables and functions for calc_load */ |
3182 | static atomic_long_t calc_load_tasks; | |
3183 | static unsigned long calc_load_update; | |
3184 | unsigned long avenrun[3]; | |
3185 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3186 | |
74f5187a PZ |
3187 | static long calc_load_fold_active(struct rq *this_rq) |
3188 | { | |
3189 | long nr_active, delta = 0; | |
3190 | ||
3191 | nr_active = this_rq->nr_running; | |
3192 | nr_active += (long) this_rq->nr_uninterruptible; | |
3193 | ||
3194 | if (nr_active != this_rq->calc_load_active) { | |
3195 | delta = nr_active - this_rq->calc_load_active; | |
3196 | this_rq->calc_load_active = nr_active; | |
3197 | } | |
3198 | ||
3199 | return delta; | |
3200 | } | |
3201 | ||
0f004f5a PZ |
3202 | static unsigned long |
3203 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3204 | { | |
3205 | load *= exp; | |
3206 | load += active * (FIXED_1 - exp); | |
3207 | load += 1UL << (FSHIFT - 1); | |
3208 | return load >> FSHIFT; | |
3209 | } | |
3210 | ||
74f5187a PZ |
3211 | #ifdef CONFIG_NO_HZ |
3212 | /* | |
3213 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3214 | * | |
3215 | * When making the ILB scale, we should try to pull this in as well. | |
3216 | */ | |
3217 | static atomic_long_t calc_load_tasks_idle; | |
3218 | ||
3219 | static void calc_load_account_idle(struct rq *this_rq) | |
3220 | { | |
3221 | long delta; | |
3222 | ||
3223 | delta = calc_load_fold_active(this_rq); | |
3224 | if (delta) | |
3225 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3226 | } | |
3227 | ||
3228 | static long calc_load_fold_idle(void) | |
3229 | { | |
3230 | long delta = 0; | |
3231 | ||
3232 | /* | |
3233 | * Its got a race, we don't care... | |
3234 | */ | |
3235 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3236 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3237 | ||
3238 | return delta; | |
3239 | } | |
0f004f5a PZ |
3240 | |
3241 | /** | |
3242 | * fixed_power_int - compute: x^n, in O(log n) time | |
3243 | * | |
3244 | * @x: base of the power | |
3245 | * @frac_bits: fractional bits of @x | |
3246 | * @n: power to raise @x to. | |
3247 | * | |
3248 | * By exploiting the relation between the definition of the natural power | |
3249 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3250 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3251 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3252 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3253 | * of course trivially computable in O(log_2 n), the length of our binary | |
3254 | * vector. | |
3255 | */ | |
3256 | static unsigned long | |
3257 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3258 | { | |
3259 | unsigned long result = 1UL << frac_bits; | |
3260 | ||
3261 | if (n) for (;;) { | |
3262 | if (n & 1) { | |
3263 | result *= x; | |
3264 | result += 1UL << (frac_bits - 1); | |
3265 | result >>= frac_bits; | |
3266 | } | |
3267 | n >>= 1; | |
3268 | if (!n) | |
3269 | break; | |
3270 | x *= x; | |
3271 | x += 1UL << (frac_bits - 1); | |
3272 | x >>= frac_bits; | |
3273 | } | |
3274 | ||
3275 | return result; | |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * a1 = a0 * e + a * (1 - e) | |
3280 | * | |
3281 | * a2 = a1 * e + a * (1 - e) | |
3282 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3283 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3284 | * | |
3285 | * a3 = a2 * e + a * (1 - e) | |
3286 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3287 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3288 | * | |
3289 | * ... | |
3290 | * | |
3291 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3292 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3293 | * = a0 * e^n + a * (1 - e^n) | |
3294 | * | |
3295 | * [1] application of the geometric series: | |
3296 | * | |
3297 | * n 1 - x^(n+1) | |
3298 | * S_n := \Sum x^i = ------------- | |
3299 | * i=0 1 - x | |
3300 | */ | |
3301 | static unsigned long | |
3302 | calc_load_n(unsigned long load, unsigned long exp, | |
3303 | unsigned long active, unsigned int n) | |
3304 | { | |
3305 | ||
3306 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3307 | } | |
3308 | ||
3309 | /* | |
3310 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3311 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3312 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3313 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3314 | * | |
3315 | * Once we've updated the global active value, we need to apply the exponential | |
3316 | * weights adjusted to the number of cycles missed. | |
3317 | */ | |
3318 | static void calc_global_nohz(unsigned long ticks) | |
3319 | { | |
3320 | long delta, active, n; | |
3321 | ||
3322 | if (time_before(jiffies, calc_load_update)) | |
3323 | return; | |
3324 | ||
3325 | /* | |
3326 | * If we crossed a calc_load_update boundary, make sure to fold | |
3327 | * any pending idle changes, the respective CPUs might have | |
3328 | * missed the tick driven calc_load_account_active() update | |
3329 | * due to NO_HZ. | |
3330 | */ | |
3331 | delta = calc_load_fold_idle(); | |
3332 | if (delta) | |
3333 | atomic_long_add(delta, &calc_load_tasks); | |
3334 | ||
3335 | /* | |
3336 | * If we were idle for multiple load cycles, apply them. | |
3337 | */ | |
3338 | if (ticks >= LOAD_FREQ) { | |
3339 | n = ticks / LOAD_FREQ; | |
3340 | ||
3341 | active = atomic_long_read(&calc_load_tasks); | |
3342 | active = active > 0 ? active * FIXED_1 : 0; | |
3343 | ||
3344 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3345 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3346 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3347 | ||
3348 | calc_load_update += n * LOAD_FREQ; | |
3349 | } | |
3350 | ||
3351 | /* | |
3352 | * Its possible the remainder of the above division also crosses | |
3353 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3354 | * which comes after this will take care of that. | |
3355 | * | |
3356 | * Consider us being 11 ticks before a cycle completion, and us | |
3357 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3358 | * age us 4 cycles, and the test in calc_global_load() will | |
3359 | * pick up the final one. | |
3360 | */ | |
3361 | } | |
74f5187a PZ |
3362 | #else |
3363 | static void calc_load_account_idle(struct rq *this_rq) | |
3364 | { | |
3365 | } | |
3366 | ||
3367 | static inline long calc_load_fold_idle(void) | |
3368 | { | |
3369 | return 0; | |
3370 | } | |
0f004f5a PZ |
3371 | |
3372 | static void calc_global_nohz(unsigned long ticks) | |
3373 | { | |
3374 | } | |
74f5187a PZ |
3375 | #endif |
3376 | ||
2d02494f TG |
3377 | /** |
3378 | * get_avenrun - get the load average array | |
3379 | * @loads: pointer to dest load array | |
3380 | * @offset: offset to add | |
3381 | * @shift: shift count to shift the result left | |
3382 | * | |
3383 | * These values are estimates at best, so no need for locking. | |
3384 | */ | |
3385 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3386 | { | |
3387 | loads[0] = (avenrun[0] + offset) << shift; | |
3388 | loads[1] = (avenrun[1] + offset) << shift; | |
3389 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3390 | } |
46cb4b7c | 3391 | |
46cb4b7c | 3392 | /* |
dce48a84 TG |
3393 | * calc_load - update the avenrun load estimates 10 ticks after the |
3394 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3395 | */ |
0f004f5a | 3396 | void calc_global_load(unsigned long ticks) |
7835b98b | 3397 | { |
dce48a84 | 3398 | long active; |
1da177e4 | 3399 | |
0f004f5a PZ |
3400 | calc_global_nohz(ticks); |
3401 | ||
3402 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3403 | return; |
1da177e4 | 3404 | |
dce48a84 TG |
3405 | active = atomic_long_read(&calc_load_tasks); |
3406 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3407 | |
dce48a84 TG |
3408 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3409 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3410 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3411 | |
dce48a84 TG |
3412 | calc_load_update += LOAD_FREQ; |
3413 | } | |
1da177e4 | 3414 | |
dce48a84 | 3415 | /* |
74f5187a PZ |
3416 | * Called from update_cpu_load() to periodically update this CPU's |
3417 | * active count. | |
dce48a84 TG |
3418 | */ |
3419 | static void calc_load_account_active(struct rq *this_rq) | |
3420 | { | |
74f5187a | 3421 | long delta; |
08c183f3 | 3422 | |
74f5187a PZ |
3423 | if (time_before(jiffies, this_rq->calc_load_update)) |
3424 | return; | |
783609c6 | 3425 | |
74f5187a PZ |
3426 | delta = calc_load_fold_active(this_rq); |
3427 | delta += calc_load_fold_idle(); | |
3428 | if (delta) | |
dce48a84 | 3429 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3430 | |
3431 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3432 | } |
3433 | ||
fdf3e95d VP |
3434 | /* |
3435 | * The exact cpuload at various idx values, calculated at every tick would be | |
3436 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3437 | * | |
3438 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3439 | * on nth tick when cpu may be busy, then we have: | |
3440 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3441 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3442 | * | |
3443 | * decay_load_missed() below does efficient calculation of | |
3444 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3445 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3446 | * | |
3447 | * The calculation is approximated on a 128 point scale. | |
3448 | * degrade_zero_ticks is the number of ticks after which load at any | |
3449 | * particular idx is approximated to be zero. | |
3450 | * degrade_factor is a precomputed table, a row for each load idx. | |
3451 | * Each column corresponds to degradation factor for a power of two ticks, | |
3452 | * based on 128 point scale. | |
3453 | * Example: | |
3454 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3455 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3456 | * | |
3457 | * With this power of 2 load factors, we can degrade the load n times | |
3458 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3459 | * n mult/shifts needed by the exact degradation. | |
3460 | */ | |
3461 | #define DEGRADE_SHIFT 7 | |
3462 | static const unsigned char | |
3463 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3464 | static const unsigned char | |
3465 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3466 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3467 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3468 | {96, 72, 40, 12, 1, 0, 0}, | |
3469 | {112, 98, 75, 43, 15, 1, 0}, | |
3470 | {120, 112, 98, 76, 45, 16, 2} }; | |
3471 | ||
3472 | /* | |
3473 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3474 | * would be when CPU is idle and so we just decay the old load without | |
3475 | * adding any new load. | |
3476 | */ | |
3477 | static unsigned long | |
3478 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3479 | { | |
3480 | int j = 0; | |
3481 | ||
3482 | if (!missed_updates) | |
3483 | return load; | |
3484 | ||
3485 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3486 | return 0; | |
3487 | ||
3488 | if (idx == 1) | |
3489 | return load >> missed_updates; | |
3490 | ||
3491 | while (missed_updates) { | |
3492 | if (missed_updates % 2) | |
3493 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3494 | ||
3495 | missed_updates >>= 1; | |
3496 | j++; | |
3497 | } | |
3498 | return load; | |
3499 | } | |
3500 | ||
46cb4b7c | 3501 | /* |
dd41f596 | 3502 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3503 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3504 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3505 | */ |
dd41f596 | 3506 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3507 | { |
495eca49 | 3508 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3509 | unsigned long curr_jiffies = jiffies; |
3510 | unsigned long pending_updates; | |
dd41f596 | 3511 | int i, scale; |
46cb4b7c | 3512 | |
dd41f596 | 3513 | this_rq->nr_load_updates++; |
46cb4b7c | 3514 | |
fdf3e95d VP |
3515 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3516 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3517 | return; | |
3518 | ||
3519 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3520 | this_rq->last_load_update_tick = curr_jiffies; | |
3521 | ||
dd41f596 | 3522 | /* Update our load: */ |
fdf3e95d VP |
3523 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3524 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3525 | unsigned long old_load, new_load; |
7d1e6a9b | 3526 | |
dd41f596 | 3527 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3528 | |
dd41f596 | 3529 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3530 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3531 | new_load = this_load; |
a25707f3 IM |
3532 | /* |
3533 | * Round up the averaging division if load is increasing. This | |
3534 | * prevents us from getting stuck on 9 if the load is 10, for | |
3535 | * example. | |
3536 | */ | |
3537 | if (new_load > old_load) | |
fdf3e95d VP |
3538 | new_load += scale - 1; |
3539 | ||
3540 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3541 | } |
da2b71ed SS |
3542 | |
3543 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3544 | } |
3545 | ||
3546 | static void update_cpu_load_active(struct rq *this_rq) | |
3547 | { | |
3548 | update_cpu_load(this_rq); | |
46cb4b7c | 3549 | |
74f5187a | 3550 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3551 | } |
3552 | ||
dd41f596 | 3553 | #ifdef CONFIG_SMP |
8a0be9ef | 3554 | |
46cb4b7c | 3555 | /* |
38022906 PZ |
3556 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3557 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3558 | */ |
38022906 | 3559 | void sched_exec(void) |
46cb4b7c | 3560 | { |
38022906 | 3561 | struct task_struct *p = current; |
1da177e4 | 3562 | unsigned long flags; |
0017d735 | 3563 | int dest_cpu; |
46cb4b7c | 3564 | |
8f42ced9 | 3565 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 3566 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3567 | if (dest_cpu == smp_processor_id()) |
3568 | goto unlock; | |
38022906 | 3569 | |
8f42ced9 | 3570 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3571 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3572 | |
8f42ced9 PZ |
3573 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3574 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3575 | return; |
3576 | } | |
0017d735 | 3577 | unlock: |
8f42ced9 | 3578 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3579 | } |
dd41f596 | 3580 | |
1da177e4 LT |
3581 | #endif |
3582 | ||
1da177e4 LT |
3583 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3584 | ||
3585 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3586 | ||
3587 | /* | |
c5f8d995 | 3588 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3589 | * @p in case that task is currently running. |
c5f8d995 HS |
3590 | * |
3591 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3592 | */ |
c5f8d995 HS |
3593 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3594 | { | |
3595 | u64 ns = 0; | |
3596 | ||
3597 | if (task_current(rq, p)) { | |
3598 | update_rq_clock(rq); | |
305e6835 | 3599 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3600 | if ((s64)ns < 0) |
3601 | ns = 0; | |
3602 | } | |
3603 | ||
3604 | return ns; | |
3605 | } | |
3606 | ||
bb34d92f | 3607 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3608 | { |
1da177e4 | 3609 | unsigned long flags; |
41b86e9c | 3610 | struct rq *rq; |
bb34d92f | 3611 | u64 ns = 0; |
48f24c4d | 3612 | |
41b86e9c | 3613 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 3614 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 3615 | task_rq_unlock(rq, p, &flags); |
1508487e | 3616 | |
c5f8d995 HS |
3617 | return ns; |
3618 | } | |
f06febc9 | 3619 | |
c5f8d995 HS |
3620 | /* |
3621 | * Return accounted runtime for the task. | |
3622 | * In case the task is currently running, return the runtime plus current's | |
3623 | * pending runtime that have not been accounted yet. | |
3624 | */ | |
3625 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3626 | { | |
3627 | unsigned long flags; | |
3628 | struct rq *rq; | |
3629 | u64 ns = 0; | |
3630 | ||
3631 | rq = task_rq_lock(p, &flags); | |
3632 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3633 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
3634 | |
3635 | return ns; | |
3636 | } | |
48f24c4d | 3637 | |
c5f8d995 HS |
3638 | /* |
3639 | * Return sum_exec_runtime for the thread group. | |
3640 | * In case the task is currently running, return the sum plus current's | |
3641 | * pending runtime that have not been accounted yet. | |
3642 | * | |
3643 | * Note that the thread group might have other running tasks as well, | |
3644 | * so the return value not includes other pending runtime that other | |
3645 | * running tasks might have. | |
3646 | */ | |
3647 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3648 | { | |
3649 | struct task_cputime totals; | |
3650 | unsigned long flags; | |
3651 | struct rq *rq; | |
3652 | u64 ns; | |
3653 | ||
3654 | rq = task_rq_lock(p, &flags); | |
3655 | thread_group_cputime(p, &totals); | |
3656 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3657 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 3658 | |
1da177e4 LT |
3659 | return ns; |
3660 | } | |
3661 | ||
1da177e4 LT |
3662 | /* |
3663 | * Account user cpu time to a process. | |
3664 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3665 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3666 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3667 | */ |
457533a7 MS |
3668 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3669 | cputime_t cputime_scaled) | |
1da177e4 LT |
3670 | { |
3671 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3672 | cputime64_t tmp; | |
3673 | ||
457533a7 | 3674 | /* Add user time to process. */ |
1da177e4 | 3675 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3676 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3677 | account_group_user_time(p, cputime); |
1da177e4 LT |
3678 | |
3679 | /* Add user time to cpustat. */ | |
3680 | tmp = cputime_to_cputime64(cputime); | |
3681 | if (TASK_NICE(p) > 0) | |
3682 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3683 | else | |
3684 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3685 | |
3686 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3687 | /* Account for user time used */ |
3688 | acct_update_integrals(p); | |
1da177e4 LT |
3689 | } |
3690 | ||
94886b84 LV |
3691 | /* |
3692 | * Account guest cpu time to a process. | |
3693 | * @p: the process that the cpu time gets accounted to | |
3694 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3695 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3696 | */ |
457533a7 MS |
3697 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3698 | cputime_t cputime_scaled) | |
94886b84 LV |
3699 | { |
3700 | cputime64_t tmp; | |
3701 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3702 | ||
3703 | tmp = cputime_to_cputime64(cputime); | |
3704 | ||
457533a7 | 3705 | /* Add guest time to process. */ |
94886b84 | 3706 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3707 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3708 | account_group_user_time(p, cputime); |
94886b84 LV |
3709 | p->gtime = cputime_add(p->gtime, cputime); |
3710 | ||
457533a7 | 3711 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3712 | if (TASK_NICE(p) > 0) { |
3713 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3714 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3715 | } else { | |
3716 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3717 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3718 | } | |
94886b84 LV |
3719 | } |
3720 | ||
70a89a66 VP |
3721 | /* |
3722 | * Account system cpu time to a process and desired cpustat field | |
3723 | * @p: the process that the cpu time gets accounted to | |
3724 | * @cputime: the cpu time spent in kernel space since the last update | |
3725 | * @cputime_scaled: cputime scaled by cpu frequency | |
3726 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3727 | */ | |
3728 | static inline | |
3729 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3730 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3731 | { | |
3732 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3733 | ||
3734 | /* Add system time to process. */ | |
3735 | p->stime = cputime_add(p->stime, cputime); | |
3736 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3737 | account_group_system_time(p, cputime); | |
3738 | ||
3739 | /* Add system time to cpustat. */ | |
3740 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3741 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3742 | ||
3743 | /* Account for system time used */ | |
3744 | acct_update_integrals(p); | |
3745 | } | |
3746 | ||
1da177e4 LT |
3747 | /* |
3748 | * Account system cpu time to a process. | |
3749 | * @p: the process that the cpu time gets accounted to | |
3750 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3751 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3752 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3753 | */ |
3754 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3755 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3756 | { |
3757 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3758 | cputime64_t *target_cputime64; |
1da177e4 | 3759 | |
983ed7a6 | 3760 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3761 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3762 | return; |
3763 | } | |
94886b84 | 3764 | |
1da177e4 | 3765 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3766 | target_cputime64 = &cpustat->irq; |
75e1056f | 3767 | else if (in_serving_softirq()) |
70a89a66 | 3768 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3769 | else |
70a89a66 | 3770 | target_cputime64 = &cpustat->system; |
ef12fefa | 3771 | |
70a89a66 | 3772 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3773 | } |
3774 | ||
c66f08be | 3775 | /* |
1da177e4 | 3776 | * Account for involuntary wait time. |
544b4a1f | 3777 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3778 | */ |
79741dd3 | 3779 | void account_steal_time(cputime_t cputime) |
c66f08be | 3780 | { |
79741dd3 MS |
3781 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3782 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3783 | ||
3784 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3785 | } |
3786 | ||
1da177e4 | 3787 | /* |
79741dd3 MS |
3788 | * Account for idle time. |
3789 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3790 | */ |
79741dd3 | 3791 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3792 | { |
3793 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3794 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3795 | struct rq *rq = this_rq(); |
1da177e4 | 3796 | |
79741dd3 MS |
3797 | if (atomic_read(&rq->nr_iowait) > 0) |
3798 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3799 | else | |
3800 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3801 | } |
3802 | ||
79741dd3 MS |
3803 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3804 | ||
abb74cef VP |
3805 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3806 | /* | |
3807 | * Account a tick to a process and cpustat | |
3808 | * @p: the process that the cpu time gets accounted to | |
3809 | * @user_tick: is the tick from userspace | |
3810 | * @rq: the pointer to rq | |
3811 | * | |
3812 | * Tick demultiplexing follows the order | |
3813 | * - pending hardirq update | |
3814 | * - pending softirq update | |
3815 | * - user_time | |
3816 | * - idle_time | |
3817 | * - system time | |
3818 | * - check for guest_time | |
3819 | * - else account as system_time | |
3820 | * | |
3821 | * Check for hardirq is done both for system and user time as there is | |
3822 | * no timer going off while we are on hardirq and hence we may never get an | |
3823 | * opportunity to update it solely in system time. | |
3824 | * p->stime and friends are only updated on system time and not on irq | |
3825 | * softirq as those do not count in task exec_runtime any more. | |
3826 | */ | |
3827 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3828 | struct rq *rq) | |
3829 | { | |
3830 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3831 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3832 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3833 | ||
3834 | if (irqtime_account_hi_update()) { | |
3835 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3836 | } else if (irqtime_account_si_update()) { | |
3837 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3838 | } else if (this_cpu_ksoftirqd() == p) { |
3839 | /* | |
3840 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3841 | * So, we have to handle it separately here. | |
3842 | * Also, p->stime needs to be updated for ksoftirqd. | |
3843 | */ | |
3844 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3845 | &cpustat->softirq); | |
abb74cef VP |
3846 | } else if (user_tick) { |
3847 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3848 | } else if (p == rq->idle) { | |
3849 | account_idle_time(cputime_one_jiffy); | |
3850 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3851 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3852 | } else { | |
3853 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3854 | &cpustat->system); | |
3855 | } | |
3856 | } | |
3857 | ||
3858 | static void irqtime_account_idle_ticks(int ticks) | |
3859 | { | |
3860 | int i; | |
3861 | struct rq *rq = this_rq(); | |
3862 | ||
3863 | for (i = 0; i < ticks; i++) | |
3864 | irqtime_account_process_tick(current, 0, rq); | |
3865 | } | |
544b4a1f | 3866 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3867 | static void irqtime_account_idle_ticks(int ticks) {} |
3868 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3869 | struct rq *rq) {} | |
544b4a1f | 3870 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3871 | |
3872 | /* | |
3873 | * Account a single tick of cpu time. | |
3874 | * @p: the process that the cpu time gets accounted to | |
3875 | * @user_tick: indicates if the tick is a user or a system tick | |
3876 | */ | |
3877 | void account_process_tick(struct task_struct *p, int user_tick) | |
3878 | { | |
a42548a1 | 3879 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3880 | struct rq *rq = this_rq(); |
3881 | ||
abb74cef VP |
3882 | if (sched_clock_irqtime) { |
3883 | irqtime_account_process_tick(p, user_tick, rq); | |
3884 | return; | |
3885 | } | |
3886 | ||
79741dd3 | 3887 | if (user_tick) |
a42548a1 | 3888 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3889 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3890 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3891 | one_jiffy_scaled); |
3892 | else | |
a42548a1 | 3893 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3894 | } |
3895 | ||
3896 | /* | |
3897 | * Account multiple ticks of steal time. | |
3898 | * @p: the process from which the cpu time has been stolen | |
3899 | * @ticks: number of stolen ticks | |
3900 | */ | |
3901 | void account_steal_ticks(unsigned long ticks) | |
3902 | { | |
3903 | account_steal_time(jiffies_to_cputime(ticks)); | |
3904 | } | |
3905 | ||
3906 | /* | |
3907 | * Account multiple ticks of idle time. | |
3908 | * @ticks: number of stolen ticks | |
3909 | */ | |
3910 | void account_idle_ticks(unsigned long ticks) | |
3911 | { | |
abb74cef VP |
3912 | |
3913 | if (sched_clock_irqtime) { | |
3914 | irqtime_account_idle_ticks(ticks); | |
3915 | return; | |
3916 | } | |
3917 | ||
79741dd3 | 3918 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3919 | } |
3920 | ||
79741dd3 MS |
3921 | #endif |
3922 | ||
49048622 BS |
3923 | /* |
3924 | * Use precise platform statistics if available: | |
3925 | */ | |
3926 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3927 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3928 | { |
d99ca3b9 HS |
3929 | *ut = p->utime; |
3930 | *st = p->stime; | |
49048622 BS |
3931 | } |
3932 | ||
0cf55e1e | 3933 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3934 | { |
0cf55e1e HS |
3935 | struct task_cputime cputime; |
3936 | ||
3937 | thread_group_cputime(p, &cputime); | |
3938 | ||
3939 | *ut = cputime.utime; | |
3940 | *st = cputime.stime; | |
49048622 BS |
3941 | } |
3942 | #else | |
761b1d26 HS |
3943 | |
3944 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3945 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3946 | #endif |
3947 | ||
d180c5bc | 3948 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3949 | { |
d99ca3b9 | 3950 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3951 | |
3952 | /* | |
3953 | * Use CFS's precise accounting: | |
3954 | */ | |
d180c5bc | 3955 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3956 | |
3957 | if (total) { | |
e75e863d | 3958 | u64 temp = rtime; |
d180c5bc | 3959 | |
e75e863d | 3960 | temp *= utime; |
49048622 | 3961 | do_div(temp, total); |
d180c5bc HS |
3962 | utime = (cputime_t)temp; |
3963 | } else | |
3964 | utime = rtime; | |
49048622 | 3965 | |
d180c5bc HS |
3966 | /* |
3967 | * Compare with previous values, to keep monotonicity: | |
3968 | */ | |
761b1d26 | 3969 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3970 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3971 | |
d99ca3b9 HS |
3972 | *ut = p->prev_utime; |
3973 | *st = p->prev_stime; | |
49048622 BS |
3974 | } |
3975 | ||
0cf55e1e HS |
3976 | /* |
3977 | * Must be called with siglock held. | |
3978 | */ | |
3979 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3980 | { |
0cf55e1e HS |
3981 | struct signal_struct *sig = p->signal; |
3982 | struct task_cputime cputime; | |
3983 | cputime_t rtime, utime, total; | |
49048622 | 3984 | |
0cf55e1e | 3985 | thread_group_cputime(p, &cputime); |
49048622 | 3986 | |
0cf55e1e HS |
3987 | total = cputime_add(cputime.utime, cputime.stime); |
3988 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3989 | |
0cf55e1e | 3990 | if (total) { |
e75e863d | 3991 | u64 temp = rtime; |
49048622 | 3992 | |
e75e863d | 3993 | temp *= cputime.utime; |
0cf55e1e HS |
3994 | do_div(temp, total); |
3995 | utime = (cputime_t)temp; | |
3996 | } else | |
3997 | utime = rtime; | |
3998 | ||
3999 | sig->prev_utime = max(sig->prev_utime, utime); | |
4000 | sig->prev_stime = max(sig->prev_stime, | |
4001 | cputime_sub(rtime, sig->prev_utime)); | |
4002 | ||
4003 | *ut = sig->prev_utime; | |
4004 | *st = sig->prev_stime; | |
49048622 | 4005 | } |
49048622 | 4006 | #endif |
49048622 | 4007 | |
7835b98b CL |
4008 | /* |
4009 | * This function gets called by the timer code, with HZ frequency. | |
4010 | * We call it with interrupts disabled. | |
4011 | * | |
4012 | * It also gets called by the fork code, when changing the parent's | |
4013 | * timeslices. | |
4014 | */ | |
4015 | void scheduler_tick(void) | |
4016 | { | |
7835b98b CL |
4017 | int cpu = smp_processor_id(); |
4018 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4019 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4020 | |
4021 | sched_clock_tick(); | |
dd41f596 | 4022 | |
05fa785c | 4023 | raw_spin_lock(&rq->lock); |
3e51f33f | 4024 | update_rq_clock(rq); |
fdf3e95d | 4025 | update_cpu_load_active(rq); |
fa85ae24 | 4026 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 4027 | raw_spin_unlock(&rq->lock); |
7835b98b | 4028 | |
e9d2b064 | 4029 | perf_event_task_tick(); |
e220d2dc | 4030 | |
e418e1c2 | 4031 | #ifdef CONFIG_SMP |
dd41f596 IM |
4032 | rq->idle_at_tick = idle_cpu(cpu); |
4033 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4034 | #endif |
1da177e4 LT |
4035 | } |
4036 | ||
132380a0 | 4037 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4038 | { |
4039 | if (in_lock_functions(addr)) { | |
4040 | addr = CALLER_ADDR2; | |
4041 | if (in_lock_functions(addr)) | |
4042 | addr = CALLER_ADDR3; | |
4043 | } | |
4044 | return addr; | |
4045 | } | |
1da177e4 | 4046 | |
7e49fcce SR |
4047 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4048 | defined(CONFIG_PREEMPT_TRACER)) | |
4049 | ||
43627582 | 4050 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4051 | { |
6cd8a4bb | 4052 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4053 | /* |
4054 | * Underflow? | |
4055 | */ | |
9a11b49a IM |
4056 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4057 | return; | |
6cd8a4bb | 4058 | #endif |
1da177e4 | 4059 | preempt_count() += val; |
6cd8a4bb | 4060 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4061 | /* |
4062 | * Spinlock count overflowing soon? | |
4063 | */ | |
33859f7f MOS |
4064 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4065 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4066 | #endif |
4067 | if (preempt_count() == val) | |
4068 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4069 | } |
4070 | EXPORT_SYMBOL(add_preempt_count); | |
4071 | ||
43627582 | 4072 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4073 | { |
6cd8a4bb | 4074 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4075 | /* |
4076 | * Underflow? | |
4077 | */ | |
01e3eb82 | 4078 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4079 | return; |
1da177e4 LT |
4080 | /* |
4081 | * Is the spinlock portion underflowing? | |
4082 | */ | |
9a11b49a IM |
4083 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4084 | !(preempt_count() & PREEMPT_MASK))) | |
4085 | return; | |
6cd8a4bb | 4086 | #endif |
9a11b49a | 4087 | |
6cd8a4bb SR |
4088 | if (preempt_count() == val) |
4089 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4090 | preempt_count() -= val; |
4091 | } | |
4092 | EXPORT_SYMBOL(sub_preempt_count); | |
4093 | ||
4094 | #endif | |
4095 | ||
4096 | /* | |
dd41f596 | 4097 | * Print scheduling while atomic bug: |
1da177e4 | 4098 | */ |
dd41f596 | 4099 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4100 | { |
838225b4 SS |
4101 | struct pt_regs *regs = get_irq_regs(); |
4102 | ||
3df0fc5b PZ |
4103 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4104 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4105 | |
dd41f596 | 4106 | debug_show_held_locks(prev); |
e21f5b15 | 4107 | print_modules(); |
dd41f596 IM |
4108 | if (irqs_disabled()) |
4109 | print_irqtrace_events(prev); | |
838225b4 SS |
4110 | |
4111 | if (regs) | |
4112 | show_regs(regs); | |
4113 | else | |
4114 | dump_stack(); | |
dd41f596 | 4115 | } |
1da177e4 | 4116 | |
dd41f596 IM |
4117 | /* |
4118 | * Various schedule()-time debugging checks and statistics: | |
4119 | */ | |
4120 | static inline void schedule_debug(struct task_struct *prev) | |
4121 | { | |
1da177e4 | 4122 | /* |
41a2d6cf | 4123 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4124 | * schedule() atomically, we ignore that path for now. |
4125 | * Otherwise, whine if we are scheduling when we should not be. | |
4126 | */ | |
3f33a7ce | 4127 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4128 | __schedule_bug(prev); |
4129 | ||
1da177e4 LT |
4130 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4131 | ||
2d72376b | 4132 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4133 | #ifdef CONFIG_SCHEDSTATS |
4134 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 4135 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 4136 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
4137 | } |
4138 | #endif | |
dd41f596 IM |
4139 | } |
4140 | ||
6cecd084 | 4141 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4142 | { |
fd2f4419 | 4143 | if (prev->on_rq) |
a64692a3 | 4144 | update_rq_clock(rq); |
6cecd084 | 4145 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4146 | } |
4147 | ||
dd41f596 IM |
4148 | /* |
4149 | * Pick up the highest-prio task: | |
4150 | */ | |
4151 | static inline struct task_struct * | |
b67802ea | 4152 | pick_next_task(struct rq *rq) |
dd41f596 | 4153 | { |
5522d5d5 | 4154 | const struct sched_class *class; |
dd41f596 | 4155 | struct task_struct *p; |
1da177e4 LT |
4156 | |
4157 | /* | |
dd41f596 IM |
4158 | * Optimization: we know that if all tasks are in |
4159 | * the fair class we can call that function directly: | |
1da177e4 | 4160 | */ |
dd41f596 | 4161 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4162 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4163 | if (likely(p)) |
4164 | return p; | |
1da177e4 LT |
4165 | } |
4166 | ||
34f971f6 | 4167 | for_each_class(class) { |
fb8d4724 | 4168 | p = class->pick_next_task(rq); |
dd41f596 IM |
4169 | if (p) |
4170 | return p; | |
dd41f596 | 4171 | } |
34f971f6 PZ |
4172 | |
4173 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4174 | } |
1da177e4 | 4175 | |
dd41f596 IM |
4176 | /* |
4177 | * schedule() is the main scheduler function. | |
4178 | */ | |
ff743345 | 4179 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4180 | { |
4181 | struct task_struct *prev, *next; | |
67ca7bde | 4182 | unsigned long *switch_count; |
dd41f596 | 4183 | struct rq *rq; |
31656519 | 4184 | int cpu; |
dd41f596 | 4185 | |
ff743345 PZ |
4186 | need_resched: |
4187 | preempt_disable(); | |
dd41f596 IM |
4188 | cpu = smp_processor_id(); |
4189 | rq = cpu_rq(cpu); | |
25502a6c | 4190 | rcu_note_context_switch(cpu); |
dd41f596 | 4191 | prev = rq->curr; |
dd41f596 | 4192 | |
dd41f596 | 4193 | schedule_debug(prev); |
1da177e4 | 4194 | |
31656519 | 4195 | if (sched_feat(HRTICK)) |
f333fdc9 | 4196 | hrtick_clear(rq); |
8f4d37ec | 4197 | |
05fa785c | 4198 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4199 | |
246d86b5 | 4200 | switch_count = &prev->nivcsw; |
1da177e4 | 4201 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4202 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4203 | prev->state = TASK_RUNNING; |
21aa9af0 | 4204 | } else { |
2acca55e PZ |
4205 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
4206 | prev->on_rq = 0; | |
4207 | ||
21aa9af0 | 4208 | /* |
2acca55e PZ |
4209 | * If a worker went to sleep, notify and ask workqueue |
4210 | * whether it wants to wake up a task to maintain | |
4211 | * concurrency. | |
21aa9af0 TH |
4212 | */ |
4213 | if (prev->flags & PF_WQ_WORKER) { | |
4214 | struct task_struct *to_wakeup; | |
4215 | ||
4216 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4217 | if (to_wakeup) | |
4218 | try_to_wake_up_local(to_wakeup); | |
4219 | } | |
6631e635 LT |
4220 | |
4221 | /* | |
2acca55e PZ |
4222 | * If we are going to sleep and we have plugged IO |
4223 | * queued, make sure to submit it to avoid deadlocks. | |
6631e635 LT |
4224 | */ |
4225 | if (blk_needs_flush_plug(prev)) { | |
4226 | raw_spin_unlock(&rq->lock); | |
a237c1c5 | 4227 | blk_schedule_flush_plug(prev); |
6631e635 LT |
4228 | raw_spin_lock(&rq->lock); |
4229 | } | |
21aa9af0 | 4230 | } |
dd41f596 | 4231 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4232 | } |
4233 | ||
3f029d3c | 4234 | pre_schedule(rq, prev); |
f65eda4f | 4235 | |
dd41f596 | 4236 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4237 | idle_balance(cpu, rq); |
1da177e4 | 4238 | |
df1c99d4 | 4239 | put_prev_task(rq, prev); |
b67802ea | 4240 | next = pick_next_task(rq); |
f26f9aff MG |
4241 | clear_tsk_need_resched(prev); |
4242 | rq->skip_clock_update = 0; | |
1da177e4 | 4243 | |
1da177e4 | 4244 | if (likely(prev != next)) { |
1da177e4 LT |
4245 | rq->nr_switches++; |
4246 | rq->curr = next; | |
4247 | ++*switch_count; | |
4248 | ||
dd41f596 | 4249 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4250 | /* |
246d86b5 ON |
4251 | * The context switch have flipped the stack from under us |
4252 | * and restored the local variables which were saved when | |
4253 | * this task called schedule() in the past. prev == current | |
4254 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4255 | */ |
4256 | cpu = smp_processor_id(); | |
4257 | rq = cpu_rq(cpu); | |
1da177e4 | 4258 | } else |
05fa785c | 4259 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4260 | |
3f029d3c | 4261 | post_schedule(rq); |
1da177e4 | 4262 | |
1da177e4 | 4263 | preempt_enable_no_resched(); |
ff743345 | 4264 | if (need_resched()) |
1da177e4 LT |
4265 | goto need_resched; |
4266 | } | |
1da177e4 LT |
4267 | EXPORT_SYMBOL(schedule); |
4268 | ||
c08f7829 | 4269 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4270 | |
c6eb3dda PZ |
4271 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4272 | { | |
4273 | bool ret = false; | |
0d66bf6d | 4274 | |
c6eb3dda PZ |
4275 | rcu_read_lock(); |
4276 | if (lock->owner != owner) | |
4277 | goto fail; | |
0d66bf6d PZ |
4278 | |
4279 | /* | |
c6eb3dda PZ |
4280 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4281 | * lock->owner still matches owner, if that fails, owner might | |
4282 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4283 | * ensures the memory stays valid. | |
0d66bf6d | 4284 | */ |
c6eb3dda | 4285 | barrier(); |
0d66bf6d | 4286 | |
c6eb3dda PZ |
4287 | ret = owner->on_cpu; |
4288 | fail: | |
4289 | rcu_read_unlock(); | |
0d66bf6d | 4290 | |
c6eb3dda PZ |
4291 | return ret; |
4292 | } | |
0d66bf6d | 4293 | |
c6eb3dda PZ |
4294 | /* |
4295 | * Look out! "owner" is an entirely speculative pointer | |
4296 | * access and not reliable. | |
4297 | */ | |
4298 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4299 | { | |
4300 | if (!sched_feat(OWNER_SPIN)) | |
4301 | return 0; | |
0d66bf6d | 4302 | |
c6eb3dda PZ |
4303 | while (owner_running(lock, owner)) { |
4304 | if (need_resched()) | |
0d66bf6d PZ |
4305 | return 0; |
4306 | ||
335d7afb | 4307 | arch_mutex_cpu_relax(); |
0d66bf6d | 4308 | } |
4b402210 | 4309 | |
c6eb3dda PZ |
4310 | /* |
4311 | * If the owner changed to another task there is likely | |
4312 | * heavy contention, stop spinning. | |
4313 | */ | |
4314 | if (lock->owner) | |
4315 | return 0; | |
4316 | ||
0d66bf6d PZ |
4317 | return 1; |
4318 | } | |
4319 | #endif | |
4320 | ||
1da177e4 LT |
4321 | #ifdef CONFIG_PREEMPT |
4322 | /* | |
2ed6e34f | 4323 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4324 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4325 | * occur there and call schedule directly. |
4326 | */ | |
d1f74e20 | 4327 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4328 | { |
4329 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4330 | |
1da177e4 LT |
4331 | /* |
4332 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4333 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4334 | */ |
beed33a8 | 4335 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4336 | return; |
4337 | ||
3a5c359a | 4338 | do { |
d1f74e20 | 4339 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4340 | schedule(); |
d1f74e20 | 4341 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4342 | |
3a5c359a AK |
4343 | /* |
4344 | * Check again in case we missed a preemption opportunity | |
4345 | * between schedule and now. | |
4346 | */ | |
4347 | barrier(); | |
5ed0cec0 | 4348 | } while (need_resched()); |
1da177e4 | 4349 | } |
1da177e4 LT |
4350 | EXPORT_SYMBOL(preempt_schedule); |
4351 | ||
4352 | /* | |
2ed6e34f | 4353 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4354 | * off of irq context. |
4355 | * Note, that this is called and return with irqs disabled. This will | |
4356 | * protect us against recursive calling from irq. | |
4357 | */ | |
4358 | asmlinkage void __sched preempt_schedule_irq(void) | |
4359 | { | |
4360 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4361 | |
2ed6e34f | 4362 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4363 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4364 | ||
3a5c359a AK |
4365 | do { |
4366 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4367 | local_irq_enable(); |
4368 | schedule(); | |
4369 | local_irq_disable(); | |
3a5c359a | 4370 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4371 | |
3a5c359a AK |
4372 | /* |
4373 | * Check again in case we missed a preemption opportunity | |
4374 | * between schedule and now. | |
4375 | */ | |
4376 | barrier(); | |
5ed0cec0 | 4377 | } while (need_resched()); |
1da177e4 LT |
4378 | } |
4379 | ||
4380 | #endif /* CONFIG_PREEMPT */ | |
4381 | ||
63859d4f | 4382 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4383 | void *key) |
1da177e4 | 4384 | { |
63859d4f | 4385 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4386 | } |
1da177e4 LT |
4387 | EXPORT_SYMBOL(default_wake_function); |
4388 | ||
4389 | /* | |
41a2d6cf IM |
4390 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4391 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4392 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4393 | * | |
4394 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4395 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4396 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4397 | */ | |
78ddb08f | 4398 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4399 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4400 | { |
2e45874c | 4401 | wait_queue_t *curr, *next; |
1da177e4 | 4402 | |
2e45874c | 4403 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4404 | unsigned flags = curr->flags; |
4405 | ||
63859d4f | 4406 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4407 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4408 | break; |
4409 | } | |
4410 | } | |
4411 | ||
4412 | /** | |
4413 | * __wake_up - wake up threads blocked on a waitqueue. | |
4414 | * @q: the waitqueue | |
4415 | * @mode: which threads | |
4416 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4417 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4418 | * |
4419 | * It may be assumed that this function implies a write memory barrier before | |
4420 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4421 | */ |
7ad5b3a5 | 4422 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4423 | int nr_exclusive, void *key) |
1da177e4 LT |
4424 | { |
4425 | unsigned long flags; | |
4426 | ||
4427 | spin_lock_irqsave(&q->lock, flags); | |
4428 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4429 | spin_unlock_irqrestore(&q->lock, flags); | |
4430 | } | |
1da177e4 LT |
4431 | EXPORT_SYMBOL(__wake_up); |
4432 | ||
4433 | /* | |
4434 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4435 | */ | |
7ad5b3a5 | 4436 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4437 | { |
4438 | __wake_up_common(q, mode, 1, 0, NULL); | |
4439 | } | |
22c43c81 | 4440 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4441 | |
4ede816a DL |
4442 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4443 | { | |
4444 | __wake_up_common(q, mode, 1, 0, key); | |
4445 | } | |
bf294b41 | 4446 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4447 | |
1da177e4 | 4448 | /** |
4ede816a | 4449 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4450 | * @q: the waitqueue |
4451 | * @mode: which threads | |
4452 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4453 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4454 | * |
4455 | * The sync wakeup differs that the waker knows that it will schedule | |
4456 | * away soon, so while the target thread will be woken up, it will not | |
4457 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4458 | * with each other. This can prevent needless bouncing between CPUs. | |
4459 | * | |
4460 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4461 | * |
4462 | * It may be assumed that this function implies a write memory barrier before | |
4463 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4464 | */ |
4ede816a DL |
4465 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4466 | int nr_exclusive, void *key) | |
1da177e4 LT |
4467 | { |
4468 | unsigned long flags; | |
7d478721 | 4469 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4470 | |
4471 | if (unlikely(!q)) | |
4472 | return; | |
4473 | ||
4474 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4475 | wake_flags = 0; |
1da177e4 LT |
4476 | |
4477 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4478 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4479 | spin_unlock_irqrestore(&q->lock, flags); |
4480 | } | |
4ede816a DL |
4481 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4482 | ||
4483 | /* | |
4484 | * __wake_up_sync - see __wake_up_sync_key() | |
4485 | */ | |
4486 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4487 | { | |
4488 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4489 | } | |
1da177e4 LT |
4490 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4491 | ||
65eb3dc6 KD |
4492 | /** |
4493 | * complete: - signals a single thread waiting on this completion | |
4494 | * @x: holds the state of this particular completion | |
4495 | * | |
4496 | * This will wake up a single thread waiting on this completion. Threads will be | |
4497 | * awakened in the same order in which they were queued. | |
4498 | * | |
4499 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4500 | * |
4501 | * It may be assumed that this function implies a write memory barrier before | |
4502 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4503 | */ |
b15136e9 | 4504 | void complete(struct completion *x) |
1da177e4 LT |
4505 | { |
4506 | unsigned long flags; | |
4507 | ||
4508 | spin_lock_irqsave(&x->wait.lock, flags); | |
4509 | x->done++; | |
d9514f6c | 4510 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4511 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4512 | } | |
4513 | EXPORT_SYMBOL(complete); | |
4514 | ||
65eb3dc6 KD |
4515 | /** |
4516 | * complete_all: - signals all threads waiting on this completion | |
4517 | * @x: holds the state of this particular completion | |
4518 | * | |
4519 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4520 | * |
4521 | * It may be assumed that this function implies a write memory barrier before | |
4522 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4523 | */ |
b15136e9 | 4524 | void complete_all(struct completion *x) |
1da177e4 LT |
4525 | { |
4526 | unsigned long flags; | |
4527 | ||
4528 | spin_lock_irqsave(&x->wait.lock, flags); | |
4529 | x->done += UINT_MAX/2; | |
d9514f6c | 4530 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4531 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4532 | } | |
4533 | EXPORT_SYMBOL(complete_all); | |
4534 | ||
8cbbe86d AK |
4535 | static inline long __sched |
4536 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4537 | { |
1da177e4 LT |
4538 | if (!x->done) { |
4539 | DECLARE_WAITQUEUE(wait, current); | |
4540 | ||
a93d2f17 | 4541 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4542 | do { |
94d3d824 | 4543 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4544 | timeout = -ERESTARTSYS; |
4545 | break; | |
8cbbe86d AK |
4546 | } |
4547 | __set_current_state(state); | |
1da177e4 LT |
4548 | spin_unlock_irq(&x->wait.lock); |
4549 | timeout = schedule_timeout(timeout); | |
4550 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4551 | } while (!x->done && timeout); |
1da177e4 | 4552 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4553 | if (!x->done) |
4554 | return timeout; | |
1da177e4 LT |
4555 | } |
4556 | x->done--; | |
ea71a546 | 4557 | return timeout ?: 1; |
1da177e4 | 4558 | } |
1da177e4 | 4559 | |
8cbbe86d AK |
4560 | static long __sched |
4561 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4562 | { |
1da177e4 LT |
4563 | might_sleep(); |
4564 | ||
4565 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4566 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4567 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4568 | return timeout; |
4569 | } | |
1da177e4 | 4570 | |
65eb3dc6 KD |
4571 | /** |
4572 | * wait_for_completion: - waits for completion of a task | |
4573 | * @x: holds the state of this particular completion | |
4574 | * | |
4575 | * This waits to be signaled for completion of a specific task. It is NOT | |
4576 | * interruptible and there is no timeout. | |
4577 | * | |
4578 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4579 | * and interrupt capability. Also see complete(). | |
4580 | */ | |
b15136e9 | 4581 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4582 | { |
4583 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4584 | } |
8cbbe86d | 4585 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4586 | |
65eb3dc6 KD |
4587 | /** |
4588 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4589 | * @x: holds the state of this particular completion | |
4590 | * @timeout: timeout value in jiffies | |
4591 | * | |
4592 | * This waits for either a completion of a specific task to be signaled or for a | |
4593 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4594 | * interruptible. | |
4595 | */ | |
b15136e9 | 4596 | unsigned long __sched |
8cbbe86d | 4597 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4598 | { |
8cbbe86d | 4599 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4600 | } |
8cbbe86d | 4601 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4602 | |
65eb3dc6 KD |
4603 | /** |
4604 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4605 | * @x: holds the state of this particular completion | |
4606 | * | |
4607 | * This waits for completion of a specific task to be signaled. It is | |
4608 | * interruptible. | |
4609 | */ | |
8cbbe86d | 4610 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4611 | { |
51e97990 AK |
4612 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4613 | if (t == -ERESTARTSYS) | |
4614 | return t; | |
4615 | return 0; | |
0fec171c | 4616 | } |
8cbbe86d | 4617 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4618 | |
65eb3dc6 KD |
4619 | /** |
4620 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4621 | * @x: holds the state of this particular completion | |
4622 | * @timeout: timeout value in jiffies | |
4623 | * | |
4624 | * This waits for either a completion of a specific task to be signaled or for a | |
4625 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4626 | */ | |
6bf41237 | 4627 | long __sched |
8cbbe86d AK |
4628 | wait_for_completion_interruptible_timeout(struct completion *x, |
4629 | unsigned long timeout) | |
0fec171c | 4630 | { |
8cbbe86d | 4631 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4632 | } |
8cbbe86d | 4633 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4634 | |
65eb3dc6 KD |
4635 | /** |
4636 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4637 | * @x: holds the state of this particular completion | |
4638 | * | |
4639 | * This waits to be signaled for completion of a specific task. It can be | |
4640 | * interrupted by a kill signal. | |
4641 | */ | |
009e577e MW |
4642 | int __sched wait_for_completion_killable(struct completion *x) |
4643 | { | |
4644 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4645 | if (t == -ERESTARTSYS) | |
4646 | return t; | |
4647 | return 0; | |
4648 | } | |
4649 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4650 | ||
0aa12fb4 SW |
4651 | /** |
4652 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4653 | * @x: holds the state of this particular completion | |
4654 | * @timeout: timeout value in jiffies | |
4655 | * | |
4656 | * This waits for either a completion of a specific task to be | |
4657 | * signaled or for a specified timeout to expire. It can be | |
4658 | * interrupted by a kill signal. The timeout is in jiffies. | |
4659 | */ | |
6bf41237 | 4660 | long __sched |
0aa12fb4 SW |
4661 | wait_for_completion_killable_timeout(struct completion *x, |
4662 | unsigned long timeout) | |
4663 | { | |
4664 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4665 | } | |
4666 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4667 | ||
be4de352 DC |
4668 | /** |
4669 | * try_wait_for_completion - try to decrement a completion without blocking | |
4670 | * @x: completion structure | |
4671 | * | |
4672 | * Returns: 0 if a decrement cannot be done without blocking | |
4673 | * 1 if a decrement succeeded. | |
4674 | * | |
4675 | * If a completion is being used as a counting completion, | |
4676 | * attempt to decrement the counter without blocking. This | |
4677 | * enables us to avoid waiting if the resource the completion | |
4678 | * is protecting is not available. | |
4679 | */ | |
4680 | bool try_wait_for_completion(struct completion *x) | |
4681 | { | |
7539a3b3 | 4682 | unsigned long flags; |
be4de352 DC |
4683 | int ret = 1; |
4684 | ||
7539a3b3 | 4685 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4686 | if (!x->done) |
4687 | ret = 0; | |
4688 | else | |
4689 | x->done--; | |
7539a3b3 | 4690 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4691 | return ret; |
4692 | } | |
4693 | EXPORT_SYMBOL(try_wait_for_completion); | |
4694 | ||
4695 | /** | |
4696 | * completion_done - Test to see if a completion has any waiters | |
4697 | * @x: completion structure | |
4698 | * | |
4699 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4700 | * 1 if there are no waiters. | |
4701 | * | |
4702 | */ | |
4703 | bool completion_done(struct completion *x) | |
4704 | { | |
7539a3b3 | 4705 | unsigned long flags; |
be4de352 DC |
4706 | int ret = 1; |
4707 | ||
7539a3b3 | 4708 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4709 | if (!x->done) |
4710 | ret = 0; | |
7539a3b3 | 4711 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4712 | return ret; |
4713 | } | |
4714 | EXPORT_SYMBOL(completion_done); | |
4715 | ||
8cbbe86d AK |
4716 | static long __sched |
4717 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4718 | { |
0fec171c IM |
4719 | unsigned long flags; |
4720 | wait_queue_t wait; | |
4721 | ||
4722 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4723 | |
8cbbe86d | 4724 | __set_current_state(state); |
1da177e4 | 4725 | |
8cbbe86d AK |
4726 | spin_lock_irqsave(&q->lock, flags); |
4727 | __add_wait_queue(q, &wait); | |
4728 | spin_unlock(&q->lock); | |
4729 | timeout = schedule_timeout(timeout); | |
4730 | spin_lock_irq(&q->lock); | |
4731 | __remove_wait_queue(q, &wait); | |
4732 | spin_unlock_irqrestore(&q->lock, flags); | |
4733 | ||
4734 | return timeout; | |
4735 | } | |
4736 | ||
4737 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4738 | { | |
4739 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4740 | } |
1da177e4 LT |
4741 | EXPORT_SYMBOL(interruptible_sleep_on); |
4742 | ||
0fec171c | 4743 | long __sched |
95cdf3b7 | 4744 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4745 | { |
8cbbe86d | 4746 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4747 | } |
1da177e4 LT |
4748 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4749 | ||
0fec171c | 4750 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4751 | { |
8cbbe86d | 4752 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4753 | } |
1da177e4 LT |
4754 | EXPORT_SYMBOL(sleep_on); |
4755 | ||
0fec171c | 4756 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4757 | { |
8cbbe86d | 4758 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4759 | } |
1da177e4 LT |
4760 | EXPORT_SYMBOL(sleep_on_timeout); |
4761 | ||
b29739f9 IM |
4762 | #ifdef CONFIG_RT_MUTEXES |
4763 | ||
4764 | /* | |
4765 | * rt_mutex_setprio - set the current priority of a task | |
4766 | * @p: task | |
4767 | * @prio: prio value (kernel-internal form) | |
4768 | * | |
4769 | * This function changes the 'effective' priority of a task. It does | |
4770 | * not touch ->normal_prio like __setscheduler(). | |
4771 | * | |
4772 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4773 | */ | |
36c8b586 | 4774 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4775 | { |
83b699ed | 4776 | int oldprio, on_rq, running; |
70b97a7f | 4777 | struct rq *rq; |
83ab0aa0 | 4778 | const struct sched_class *prev_class; |
b29739f9 IM |
4779 | |
4780 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4781 | ||
0122ec5b | 4782 | rq = __task_rq_lock(p); |
b29739f9 | 4783 | |
a8027073 | 4784 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4785 | oldprio = p->prio; |
83ab0aa0 | 4786 | prev_class = p->sched_class; |
fd2f4419 | 4787 | on_rq = p->on_rq; |
051a1d1a | 4788 | running = task_current(rq, p); |
0e1f3483 | 4789 | if (on_rq) |
69be72c1 | 4790 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4791 | if (running) |
4792 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4793 | |
4794 | if (rt_prio(prio)) | |
4795 | p->sched_class = &rt_sched_class; | |
4796 | else | |
4797 | p->sched_class = &fair_sched_class; | |
4798 | ||
b29739f9 IM |
4799 | p->prio = prio; |
4800 | ||
0e1f3483 HS |
4801 | if (running) |
4802 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4803 | if (on_rq) |
371fd7e7 | 4804 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4805 | |
da7a735e | 4806 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4807 | __task_rq_unlock(rq); |
b29739f9 IM |
4808 | } |
4809 | ||
4810 | #endif | |
4811 | ||
36c8b586 | 4812 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4813 | { |
dd41f596 | 4814 | int old_prio, delta, on_rq; |
1da177e4 | 4815 | unsigned long flags; |
70b97a7f | 4816 | struct rq *rq; |
1da177e4 LT |
4817 | |
4818 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4819 | return; | |
4820 | /* | |
4821 | * We have to be careful, if called from sys_setpriority(), | |
4822 | * the task might be in the middle of scheduling on another CPU. | |
4823 | */ | |
4824 | rq = task_rq_lock(p, &flags); | |
4825 | /* | |
4826 | * The RT priorities are set via sched_setscheduler(), but we still | |
4827 | * allow the 'normal' nice value to be set - but as expected | |
4828 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4829 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4830 | */ |
e05606d3 | 4831 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4832 | p->static_prio = NICE_TO_PRIO(nice); |
4833 | goto out_unlock; | |
4834 | } | |
fd2f4419 | 4835 | on_rq = p->on_rq; |
c09595f6 | 4836 | if (on_rq) |
69be72c1 | 4837 | dequeue_task(rq, p, 0); |
1da177e4 | 4838 | |
1da177e4 | 4839 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4840 | set_load_weight(p); |
b29739f9 IM |
4841 | old_prio = p->prio; |
4842 | p->prio = effective_prio(p); | |
4843 | delta = p->prio - old_prio; | |
1da177e4 | 4844 | |
dd41f596 | 4845 | if (on_rq) { |
371fd7e7 | 4846 | enqueue_task(rq, p, 0); |
1da177e4 | 4847 | /* |
d5f9f942 AM |
4848 | * If the task increased its priority or is running and |
4849 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4850 | */ |
d5f9f942 | 4851 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4852 | resched_task(rq->curr); |
4853 | } | |
4854 | out_unlock: | |
0122ec5b | 4855 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 4856 | } |
1da177e4 LT |
4857 | EXPORT_SYMBOL(set_user_nice); |
4858 | ||
e43379f1 MM |
4859 | /* |
4860 | * can_nice - check if a task can reduce its nice value | |
4861 | * @p: task | |
4862 | * @nice: nice value | |
4863 | */ | |
36c8b586 | 4864 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4865 | { |
024f4747 MM |
4866 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4867 | int nice_rlim = 20 - nice; | |
48f24c4d | 4868 | |
78d7d407 | 4869 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4870 | capable(CAP_SYS_NICE)); |
4871 | } | |
4872 | ||
1da177e4 LT |
4873 | #ifdef __ARCH_WANT_SYS_NICE |
4874 | ||
4875 | /* | |
4876 | * sys_nice - change the priority of the current process. | |
4877 | * @increment: priority increment | |
4878 | * | |
4879 | * sys_setpriority is a more generic, but much slower function that | |
4880 | * does similar things. | |
4881 | */ | |
5add95d4 | 4882 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4883 | { |
48f24c4d | 4884 | long nice, retval; |
1da177e4 LT |
4885 | |
4886 | /* | |
4887 | * Setpriority might change our priority at the same moment. | |
4888 | * We don't have to worry. Conceptually one call occurs first | |
4889 | * and we have a single winner. | |
4890 | */ | |
e43379f1 MM |
4891 | if (increment < -40) |
4892 | increment = -40; | |
1da177e4 LT |
4893 | if (increment > 40) |
4894 | increment = 40; | |
4895 | ||
2b8f836f | 4896 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4897 | if (nice < -20) |
4898 | nice = -20; | |
4899 | if (nice > 19) | |
4900 | nice = 19; | |
4901 | ||
e43379f1 MM |
4902 | if (increment < 0 && !can_nice(current, nice)) |
4903 | return -EPERM; | |
4904 | ||
1da177e4 LT |
4905 | retval = security_task_setnice(current, nice); |
4906 | if (retval) | |
4907 | return retval; | |
4908 | ||
4909 | set_user_nice(current, nice); | |
4910 | return 0; | |
4911 | } | |
4912 | ||
4913 | #endif | |
4914 | ||
4915 | /** | |
4916 | * task_prio - return the priority value of a given task. | |
4917 | * @p: the task in question. | |
4918 | * | |
4919 | * This is the priority value as seen by users in /proc. | |
4920 | * RT tasks are offset by -200. Normal tasks are centered | |
4921 | * around 0, value goes from -16 to +15. | |
4922 | */ | |
36c8b586 | 4923 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4924 | { |
4925 | return p->prio - MAX_RT_PRIO; | |
4926 | } | |
4927 | ||
4928 | /** | |
4929 | * task_nice - return the nice value of a given task. | |
4930 | * @p: the task in question. | |
4931 | */ | |
36c8b586 | 4932 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4933 | { |
4934 | return TASK_NICE(p); | |
4935 | } | |
150d8bed | 4936 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4937 | |
4938 | /** | |
4939 | * idle_cpu - is a given cpu idle currently? | |
4940 | * @cpu: the processor in question. | |
4941 | */ | |
4942 | int idle_cpu(int cpu) | |
4943 | { | |
4944 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4945 | } | |
4946 | ||
1da177e4 LT |
4947 | /** |
4948 | * idle_task - return the idle task for a given cpu. | |
4949 | * @cpu: the processor in question. | |
4950 | */ | |
36c8b586 | 4951 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4952 | { |
4953 | return cpu_rq(cpu)->idle; | |
4954 | } | |
4955 | ||
4956 | /** | |
4957 | * find_process_by_pid - find a process with a matching PID value. | |
4958 | * @pid: the pid in question. | |
4959 | */ | |
a9957449 | 4960 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4961 | { |
228ebcbe | 4962 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4963 | } |
4964 | ||
4965 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4966 | static void |
4967 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4968 | { |
1da177e4 LT |
4969 | p->policy = policy; |
4970 | p->rt_priority = prio; | |
b29739f9 IM |
4971 | p->normal_prio = normal_prio(p); |
4972 | /* we are holding p->pi_lock already */ | |
4973 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4974 | if (rt_prio(p->prio)) |
4975 | p->sched_class = &rt_sched_class; | |
4976 | else | |
4977 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4978 | set_load_weight(p); |
1da177e4 LT |
4979 | } |
4980 | ||
c69e8d9c DH |
4981 | /* |
4982 | * check the target process has a UID that matches the current process's | |
4983 | */ | |
4984 | static bool check_same_owner(struct task_struct *p) | |
4985 | { | |
4986 | const struct cred *cred = current_cred(), *pcred; | |
4987 | bool match; | |
4988 | ||
4989 | rcu_read_lock(); | |
4990 | pcred = __task_cred(p); | |
b0e77598 SH |
4991 | if (cred->user->user_ns == pcred->user->user_ns) |
4992 | match = (cred->euid == pcred->euid || | |
4993 | cred->euid == pcred->uid); | |
4994 | else | |
4995 | match = false; | |
c69e8d9c DH |
4996 | rcu_read_unlock(); |
4997 | return match; | |
4998 | } | |
4999 | ||
961ccddd | 5000 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 5001 | const struct sched_param *param, bool user) |
1da177e4 | 5002 | { |
83b699ed | 5003 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5004 | unsigned long flags; |
83ab0aa0 | 5005 | const struct sched_class *prev_class; |
70b97a7f | 5006 | struct rq *rq; |
ca94c442 | 5007 | int reset_on_fork; |
1da177e4 | 5008 | |
66e5393a SR |
5009 | /* may grab non-irq protected spin_locks */ |
5010 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5011 | recheck: |
5012 | /* double check policy once rq lock held */ | |
ca94c442 LP |
5013 | if (policy < 0) { |
5014 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5015 | policy = oldpolicy = p->policy; |
ca94c442 LP |
5016 | } else { |
5017 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
5018 | policy &= ~SCHED_RESET_ON_FORK; | |
5019 | ||
5020 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
5021 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
5022 | policy != SCHED_IDLE) | |
5023 | return -EINVAL; | |
5024 | } | |
5025 | ||
1da177e4 LT |
5026 | /* |
5027 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5028 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5029 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5030 | */ |
5031 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5032 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5033 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5034 | return -EINVAL; |
e05606d3 | 5035 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5036 | return -EINVAL; |
5037 | ||
37e4ab3f OC |
5038 | /* |
5039 | * Allow unprivileged RT tasks to decrease priority: | |
5040 | */ | |
961ccddd | 5041 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5042 | if (rt_policy(policy)) { |
a44702e8 ON |
5043 | unsigned long rlim_rtprio = |
5044 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
5045 | |
5046 | /* can't set/change the rt policy */ | |
5047 | if (policy != p->policy && !rlim_rtprio) | |
5048 | return -EPERM; | |
5049 | ||
5050 | /* can't increase priority */ | |
5051 | if (param->sched_priority > p->rt_priority && | |
5052 | param->sched_priority > rlim_rtprio) | |
5053 | return -EPERM; | |
5054 | } | |
c02aa73b | 5055 | |
dd41f596 | 5056 | /* |
c02aa73b DH |
5057 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5058 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5059 | */ |
c02aa73b DH |
5060 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
5061 | if (!can_nice(p, TASK_NICE(p))) | |
5062 | return -EPERM; | |
5063 | } | |
5fe1d75f | 5064 | |
37e4ab3f | 5065 | /* can't change other user's priorities */ |
c69e8d9c | 5066 | if (!check_same_owner(p)) |
37e4ab3f | 5067 | return -EPERM; |
ca94c442 LP |
5068 | |
5069 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
5070 | if (p->sched_reset_on_fork && !reset_on_fork) | |
5071 | return -EPERM; | |
37e4ab3f | 5072 | } |
1da177e4 | 5073 | |
725aad24 | 5074 | if (user) { |
b0ae1981 | 5075 | retval = security_task_setscheduler(p); |
725aad24 JF |
5076 | if (retval) |
5077 | return retval; | |
5078 | } | |
5079 | ||
b29739f9 IM |
5080 | /* |
5081 | * make sure no PI-waiters arrive (or leave) while we are | |
5082 | * changing the priority of the task: | |
0122ec5b | 5083 | * |
25985edc | 5084 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5085 | * runqueue lock must be held. |
5086 | */ | |
0122ec5b | 5087 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 5088 | |
34f971f6 PZ |
5089 | /* |
5090 | * Changing the policy of the stop threads its a very bad idea | |
5091 | */ | |
5092 | if (p == rq->stop) { | |
0122ec5b | 5093 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
5094 | return -EINVAL; |
5095 | } | |
5096 | ||
a51e9198 DF |
5097 | /* |
5098 | * If not changing anything there's no need to proceed further: | |
5099 | */ | |
5100 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5101 | param->sched_priority == p->rt_priority))) { | |
5102 | ||
5103 | __task_rq_unlock(rq); | |
5104 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5105 | return 0; | |
5106 | } | |
5107 | ||
dc61b1d6 PZ |
5108 | #ifdef CONFIG_RT_GROUP_SCHED |
5109 | if (user) { | |
5110 | /* | |
5111 | * Do not allow realtime tasks into groups that have no runtime | |
5112 | * assigned. | |
5113 | */ | |
5114 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5115 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5116 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 5117 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
5118 | return -EPERM; |
5119 | } | |
5120 | } | |
5121 | #endif | |
5122 | ||
1da177e4 LT |
5123 | /* recheck policy now with rq lock held */ |
5124 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5125 | policy = oldpolicy = -1; | |
0122ec5b | 5126 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
5127 | goto recheck; |
5128 | } | |
fd2f4419 | 5129 | on_rq = p->on_rq; |
051a1d1a | 5130 | running = task_current(rq, p); |
0e1f3483 | 5131 | if (on_rq) |
2e1cb74a | 5132 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5133 | if (running) |
5134 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5135 | |
ca94c442 LP |
5136 | p->sched_reset_on_fork = reset_on_fork; |
5137 | ||
1da177e4 | 5138 | oldprio = p->prio; |
83ab0aa0 | 5139 | prev_class = p->sched_class; |
dd41f596 | 5140 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5141 | |
0e1f3483 HS |
5142 | if (running) |
5143 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5144 | if (on_rq) |
dd41f596 | 5145 | activate_task(rq, p, 0); |
cb469845 | 5146 | |
da7a735e | 5147 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5148 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 5149 | |
95e02ca9 TG |
5150 | rt_mutex_adjust_pi(p); |
5151 | ||
1da177e4 LT |
5152 | return 0; |
5153 | } | |
961ccddd RR |
5154 | |
5155 | /** | |
5156 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5157 | * @p: the task in question. | |
5158 | * @policy: new policy. | |
5159 | * @param: structure containing the new RT priority. | |
5160 | * | |
5161 | * NOTE that the task may be already dead. | |
5162 | */ | |
5163 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5164 | const struct sched_param *param) |
961ccddd RR |
5165 | { |
5166 | return __sched_setscheduler(p, policy, param, true); | |
5167 | } | |
1da177e4 LT |
5168 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5169 | ||
961ccddd RR |
5170 | /** |
5171 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5172 | * @p: the task in question. | |
5173 | * @policy: new policy. | |
5174 | * @param: structure containing the new RT priority. | |
5175 | * | |
5176 | * Just like sched_setscheduler, only don't bother checking if the | |
5177 | * current context has permission. For example, this is needed in | |
5178 | * stop_machine(): we create temporary high priority worker threads, | |
5179 | * but our caller might not have that capability. | |
5180 | */ | |
5181 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5182 | const struct sched_param *param) |
961ccddd RR |
5183 | { |
5184 | return __sched_setscheduler(p, policy, param, false); | |
5185 | } | |
5186 | ||
95cdf3b7 IM |
5187 | static int |
5188 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5189 | { |
1da177e4 LT |
5190 | struct sched_param lparam; |
5191 | struct task_struct *p; | |
36c8b586 | 5192 | int retval; |
1da177e4 LT |
5193 | |
5194 | if (!param || pid < 0) | |
5195 | return -EINVAL; | |
5196 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5197 | return -EFAULT; | |
5fe1d75f ON |
5198 | |
5199 | rcu_read_lock(); | |
5200 | retval = -ESRCH; | |
1da177e4 | 5201 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5202 | if (p != NULL) |
5203 | retval = sched_setscheduler(p, policy, &lparam); | |
5204 | rcu_read_unlock(); | |
36c8b586 | 5205 | |
1da177e4 LT |
5206 | return retval; |
5207 | } | |
5208 | ||
5209 | /** | |
5210 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5211 | * @pid: the pid in question. | |
5212 | * @policy: new policy. | |
5213 | * @param: structure containing the new RT priority. | |
5214 | */ | |
5add95d4 HC |
5215 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5216 | struct sched_param __user *, param) | |
1da177e4 | 5217 | { |
c21761f1 JB |
5218 | /* negative values for policy are not valid */ |
5219 | if (policy < 0) | |
5220 | return -EINVAL; | |
5221 | ||
1da177e4 LT |
5222 | return do_sched_setscheduler(pid, policy, param); |
5223 | } | |
5224 | ||
5225 | /** | |
5226 | * sys_sched_setparam - set/change the RT priority of a thread | |
5227 | * @pid: the pid in question. | |
5228 | * @param: structure containing the new RT priority. | |
5229 | */ | |
5add95d4 | 5230 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5231 | { |
5232 | return do_sched_setscheduler(pid, -1, param); | |
5233 | } | |
5234 | ||
5235 | /** | |
5236 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5237 | * @pid: the pid in question. | |
5238 | */ | |
5add95d4 | 5239 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5240 | { |
36c8b586 | 5241 | struct task_struct *p; |
3a5c359a | 5242 | int retval; |
1da177e4 LT |
5243 | |
5244 | if (pid < 0) | |
3a5c359a | 5245 | return -EINVAL; |
1da177e4 LT |
5246 | |
5247 | retval = -ESRCH; | |
5fe85be0 | 5248 | rcu_read_lock(); |
1da177e4 LT |
5249 | p = find_process_by_pid(pid); |
5250 | if (p) { | |
5251 | retval = security_task_getscheduler(p); | |
5252 | if (!retval) | |
ca94c442 LP |
5253 | retval = p->policy |
5254 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5255 | } |
5fe85be0 | 5256 | rcu_read_unlock(); |
1da177e4 LT |
5257 | return retval; |
5258 | } | |
5259 | ||
5260 | /** | |
ca94c442 | 5261 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5262 | * @pid: the pid in question. |
5263 | * @param: structure containing the RT priority. | |
5264 | */ | |
5add95d4 | 5265 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5266 | { |
5267 | struct sched_param lp; | |
36c8b586 | 5268 | struct task_struct *p; |
3a5c359a | 5269 | int retval; |
1da177e4 LT |
5270 | |
5271 | if (!param || pid < 0) | |
3a5c359a | 5272 | return -EINVAL; |
1da177e4 | 5273 | |
5fe85be0 | 5274 | rcu_read_lock(); |
1da177e4 LT |
5275 | p = find_process_by_pid(pid); |
5276 | retval = -ESRCH; | |
5277 | if (!p) | |
5278 | goto out_unlock; | |
5279 | ||
5280 | retval = security_task_getscheduler(p); | |
5281 | if (retval) | |
5282 | goto out_unlock; | |
5283 | ||
5284 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5285 | rcu_read_unlock(); |
1da177e4 LT |
5286 | |
5287 | /* | |
5288 | * This one might sleep, we cannot do it with a spinlock held ... | |
5289 | */ | |
5290 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5291 | ||
1da177e4 LT |
5292 | return retval; |
5293 | ||
5294 | out_unlock: | |
5fe85be0 | 5295 | rcu_read_unlock(); |
1da177e4 LT |
5296 | return retval; |
5297 | } | |
5298 | ||
96f874e2 | 5299 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5300 | { |
5a16f3d3 | 5301 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5302 | struct task_struct *p; |
5303 | int retval; | |
1da177e4 | 5304 | |
95402b38 | 5305 | get_online_cpus(); |
23f5d142 | 5306 | rcu_read_lock(); |
1da177e4 LT |
5307 | |
5308 | p = find_process_by_pid(pid); | |
5309 | if (!p) { | |
23f5d142 | 5310 | rcu_read_unlock(); |
95402b38 | 5311 | put_online_cpus(); |
1da177e4 LT |
5312 | return -ESRCH; |
5313 | } | |
5314 | ||
23f5d142 | 5315 | /* Prevent p going away */ |
1da177e4 | 5316 | get_task_struct(p); |
23f5d142 | 5317 | rcu_read_unlock(); |
1da177e4 | 5318 | |
5a16f3d3 RR |
5319 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5320 | retval = -ENOMEM; | |
5321 | goto out_put_task; | |
5322 | } | |
5323 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5324 | retval = -ENOMEM; | |
5325 | goto out_free_cpus_allowed; | |
5326 | } | |
1da177e4 | 5327 | retval = -EPERM; |
b0e77598 | 5328 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5329 | goto out_unlock; |
5330 | ||
b0ae1981 | 5331 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5332 | if (retval) |
5333 | goto out_unlock; | |
5334 | ||
5a16f3d3 RR |
5335 | cpuset_cpus_allowed(p, cpus_allowed); |
5336 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5337 | again: |
5a16f3d3 | 5338 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5339 | |
8707d8b8 | 5340 | if (!retval) { |
5a16f3d3 RR |
5341 | cpuset_cpus_allowed(p, cpus_allowed); |
5342 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5343 | /* |
5344 | * We must have raced with a concurrent cpuset | |
5345 | * update. Just reset the cpus_allowed to the | |
5346 | * cpuset's cpus_allowed | |
5347 | */ | |
5a16f3d3 | 5348 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5349 | goto again; |
5350 | } | |
5351 | } | |
1da177e4 | 5352 | out_unlock: |
5a16f3d3 RR |
5353 | free_cpumask_var(new_mask); |
5354 | out_free_cpus_allowed: | |
5355 | free_cpumask_var(cpus_allowed); | |
5356 | out_put_task: | |
1da177e4 | 5357 | put_task_struct(p); |
95402b38 | 5358 | put_online_cpus(); |
1da177e4 LT |
5359 | return retval; |
5360 | } | |
5361 | ||
5362 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5363 | struct cpumask *new_mask) |
1da177e4 | 5364 | { |
96f874e2 RR |
5365 | if (len < cpumask_size()) |
5366 | cpumask_clear(new_mask); | |
5367 | else if (len > cpumask_size()) | |
5368 | len = cpumask_size(); | |
5369 | ||
1da177e4 LT |
5370 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5371 | } | |
5372 | ||
5373 | /** | |
5374 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5375 | * @pid: pid of the process | |
5376 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5377 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5378 | */ | |
5add95d4 HC |
5379 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5380 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5381 | { |
5a16f3d3 | 5382 | cpumask_var_t new_mask; |
1da177e4 LT |
5383 | int retval; |
5384 | ||
5a16f3d3 RR |
5385 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5386 | return -ENOMEM; | |
1da177e4 | 5387 | |
5a16f3d3 RR |
5388 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5389 | if (retval == 0) | |
5390 | retval = sched_setaffinity(pid, new_mask); | |
5391 | free_cpumask_var(new_mask); | |
5392 | return retval; | |
1da177e4 LT |
5393 | } |
5394 | ||
96f874e2 | 5395 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5396 | { |
36c8b586 | 5397 | struct task_struct *p; |
31605683 | 5398 | unsigned long flags; |
1da177e4 | 5399 | int retval; |
1da177e4 | 5400 | |
95402b38 | 5401 | get_online_cpus(); |
23f5d142 | 5402 | rcu_read_lock(); |
1da177e4 LT |
5403 | |
5404 | retval = -ESRCH; | |
5405 | p = find_process_by_pid(pid); | |
5406 | if (!p) | |
5407 | goto out_unlock; | |
5408 | ||
e7834f8f DQ |
5409 | retval = security_task_getscheduler(p); |
5410 | if (retval) | |
5411 | goto out_unlock; | |
5412 | ||
013fdb80 | 5413 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 5414 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 5415 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5416 | |
5417 | out_unlock: | |
23f5d142 | 5418 | rcu_read_unlock(); |
95402b38 | 5419 | put_online_cpus(); |
1da177e4 | 5420 | |
9531b62f | 5421 | return retval; |
1da177e4 LT |
5422 | } |
5423 | ||
5424 | /** | |
5425 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5426 | * @pid: pid of the process | |
5427 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5428 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5429 | */ | |
5add95d4 HC |
5430 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5431 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5432 | { |
5433 | int ret; | |
f17c8607 | 5434 | cpumask_var_t mask; |
1da177e4 | 5435 | |
84fba5ec | 5436 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5437 | return -EINVAL; |
5438 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5439 | return -EINVAL; |
5440 | ||
f17c8607 RR |
5441 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5442 | return -ENOMEM; | |
1da177e4 | 5443 | |
f17c8607 RR |
5444 | ret = sched_getaffinity(pid, mask); |
5445 | if (ret == 0) { | |
8bc037fb | 5446 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5447 | |
5448 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5449 | ret = -EFAULT; |
5450 | else | |
cd3d8031 | 5451 | ret = retlen; |
f17c8607 RR |
5452 | } |
5453 | free_cpumask_var(mask); | |
1da177e4 | 5454 | |
f17c8607 | 5455 | return ret; |
1da177e4 LT |
5456 | } |
5457 | ||
5458 | /** | |
5459 | * sys_sched_yield - yield the current processor to other threads. | |
5460 | * | |
dd41f596 IM |
5461 | * This function yields the current CPU to other tasks. If there are no |
5462 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5463 | */ |
5add95d4 | 5464 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5465 | { |
70b97a7f | 5466 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5467 | |
2d72376b | 5468 | schedstat_inc(rq, yld_count); |
4530d7ab | 5469 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5470 | |
5471 | /* | |
5472 | * Since we are going to call schedule() anyway, there's | |
5473 | * no need to preempt or enable interrupts: | |
5474 | */ | |
5475 | __release(rq->lock); | |
8a25d5de | 5476 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5477 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5478 | preempt_enable_no_resched(); |
5479 | ||
5480 | schedule(); | |
5481 | ||
5482 | return 0; | |
5483 | } | |
5484 | ||
d86ee480 PZ |
5485 | static inline int should_resched(void) |
5486 | { | |
5487 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5488 | } | |
5489 | ||
e7b38404 | 5490 | static void __cond_resched(void) |
1da177e4 | 5491 | { |
e7aaaa69 FW |
5492 | add_preempt_count(PREEMPT_ACTIVE); |
5493 | schedule(); | |
5494 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5495 | } |
5496 | ||
02b67cc3 | 5497 | int __sched _cond_resched(void) |
1da177e4 | 5498 | { |
d86ee480 | 5499 | if (should_resched()) { |
1da177e4 LT |
5500 | __cond_resched(); |
5501 | return 1; | |
5502 | } | |
5503 | return 0; | |
5504 | } | |
02b67cc3 | 5505 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5506 | |
5507 | /* | |
613afbf8 | 5508 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5509 | * call schedule, and on return reacquire the lock. |
5510 | * | |
41a2d6cf | 5511 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5512 | * operations here to prevent schedule() from being called twice (once via |
5513 | * spin_unlock(), once by hand). | |
5514 | */ | |
613afbf8 | 5515 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5516 | { |
d86ee480 | 5517 | int resched = should_resched(); |
6df3cecb JK |
5518 | int ret = 0; |
5519 | ||
f607c668 PZ |
5520 | lockdep_assert_held(lock); |
5521 | ||
95c354fe | 5522 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5523 | spin_unlock(lock); |
d86ee480 | 5524 | if (resched) |
95c354fe NP |
5525 | __cond_resched(); |
5526 | else | |
5527 | cpu_relax(); | |
6df3cecb | 5528 | ret = 1; |
1da177e4 | 5529 | spin_lock(lock); |
1da177e4 | 5530 | } |
6df3cecb | 5531 | return ret; |
1da177e4 | 5532 | } |
613afbf8 | 5533 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5534 | |
613afbf8 | 5535 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5536 | { |
5537 | BUG_ON(!in_softirq()); | |
5538 | ||
d86ee480 | 5539 | if (should_resched()) { |
98d82567 | 5540 | local_bh_enable(); |
1da177e4 LT |
5541 | __cond_resched(); |
5542 | local_bh_disable(); | |
5543 | return 1; | |
5544 | } | |
5545 | return 0; | |
5546 | } | |
613afbf8 | 5547 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5548 | |
1da177e4 LT |
5549 | /** |
5550 | * yield - yield the current processor to other threads. | |
5551 | * | |
72fd4a35 | 5552 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5553 | * thread runnable and calls sys_sched_yield(). |
5554 | */ | |
5555 | void __sched yield(void) | |
5556 | { | |
5557 | set_current_state(TASK_RUNNING); | |
5558 | sys_sched_yield(); | |
5559 | } | |
1da177e4 LT |
5560 | EXPORT_SYMBOL(yield); |
5561 | ||
d95f4122 MG |
5562 | /** |
5563 | * yield_to - yield the current processor to another thread in | |
5564 | * your thread group, or accelerate that thread toward the | |
5565 | * processor it's on. | |
16addf95 RD |
5566 | * @p: target task |
5567 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5568 | * |
5569 | * It's the caller's job to ensure that the target task struct | |
5570 | * can't go away on us before we can do any checks. | |
5571 | * | |
5572 | * Returns true if we indeed boosted the target task. | |
5573 | */ | |
5574 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5575 | { | |
5576 | struct task_struct *curr = current; | |
5577 | struct rq *rq, *p_rq; | |
5578 | unsigned long flags; | |
5579 | bool yielded = 0; | |
5580 | ||
5581 | local_irq_save(flags); | |
5582 | rq = this_rq(); | |
5583 | ||
5584 | again: | |
5585 | p_rq = task_rq(p); | |
5586 | double_rq_lock(rq, p_rq); | |
5587 | while (task_rq(p) != p_rq) { | |
5588 | double_rq_unlock(rq, p_rq); | |
5589 | goto again; | |
5590 | } | |
5591 | ||
5592 | if (!curr->sched_class->yield_to_task) | |
5593 | goto out; | |
5594 | ||
5595 | if (curr->sched_class != p->sched_class) | |
5596 | goto out; | |
5597 | ||
5598 | if (task_running(p_rq, p) || p->state) | |
5599 | goto out; | |
5600 | ||
5601 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5602 | if (yielded) { |
d95f4122 | 5603 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5604 | /* |
5605 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5606 | * fairness. | |
5607 | */ | |
5608 | if (preempt && rq != p_rq) | |
5609 | resched_task(p_rq->curr); | |
5610 | } | |
d95f4122 MG |
5611 | |
5612 | out: | |
5613 | double_rq_unlock(rq, p_rq); | |
5614 | local_irq_restore(flags); | |
5615 | ||
5616 | if (yielded) | |
5617 | schedule(); | |
5618 | ||
5619 | return yielded; | |
5620 | } | |
5621 | EXPORT_SYMBOL_GPL(yield_to); | |
5622 | ||
1da177e4 | 5623 | /* |
41a2d6cf | 5624 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5625 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5626 | */ |
5627 | void __sched io_schedule(void) | |
5628 | { | |
54d35f29 | 5629 | struct rq *rq = raw_rq(); |
1da177e4 | 5630 | |
0ff92245 | 5631 | delayacct_blkio_start(); |
1da177e4 | 5632 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5633 | blk_flush_plug(current); |
8f0dfc34 | 5634 | current->in_iowait = 1; |
1da177e4 | 5635 | schedule(); |
8f0dfc34 | 5636 | current->in_iowait = 0; |
1da177e4 | 5637 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5638 | delayacct_blkio_end(); |
1da177e4 | 5639 | } |
1da177e4 LT |
5640 | EXPORT_SYMBOL(io_schedule); |
5641 | ||
5642 | long __sched io_schedule_timeout(long timeout) | |
5643 | { | |
54d35f29 | 5644 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5645 | long ret; |
5646 | ||
0ff92245 | 5647 | delayacct_blkio_start(); |
1da177e4 | 5648 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5649 | blk_flush_plug(current); |
8f0dfc34 | 5650 | current->in_iowait = 1; |
1da177e4 | 5651 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5652 | current->in_iowait = 0; |
1da177e4 | 5653 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5654 | delayacct_blkio_end(); |
1da177e4 LT |
5655 | return ret; |
5656 | } | |
5657 | ||
5658 | /** | |
5659 | * sys_sched_get_priority_max - return maximum RT priority. | |
5660 | * @policy: scheduling class. | |
5661 | * | |
5662 | * this syscall returns the maximum rt_priority that can be used | |
5663 | * by a given scheduling class. | |
5664 | */ | |
5add95d4 | 5665 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5666 | { |
5667 | int ret = -EINVAL; | |
5668 | ||
5669 | switch (policy) { | |
5670 | case SCHED_FIFO: | |
5671 | case SCHED_RR: | |
5672 | ret = MAX_USER_RT_PRIO-1; | |
5673 | break; | |
5674 | case SCHED_NORMAL: | |
b0a9499c | 5675 | case SCHED_BATCH: |
dd41f596 | 5676 | case SCHED_IDLE: |
1da177e4 LT |
5677 | ret = 0; |
5678 | break; | |
5679 | } | |
5680 | return ret; | |
5681 | } | |
5682 | ||
5683 | /** | |
5684 | * sys_sched_get_priority_min - return minimum RT priority. | |
5685 | * @policy: scheduling class. | |
5686 | * | |
5687 | * this syscall returns the minimum rt_priority that can be used | |
5688 | * by a given scheduling class. | |
5689 | */ | |
5add95d4 | 5690 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5691 | { |
5692 | int ret = -EINVAL; | |
5693 | ||
5694 | switch (policy) { | |
5695 | case SCHED_FIFO: | |
5696 | case SCHED_RR: | |
5697 | ret = 1; | |
5698 | break; | |
5699 | case SCHED_NORMAL: | |
b0a9499c | 5700 | case SCHED_BATCH: |
dd41f596 | 5701 | case SCHED_IDLE: |
1da177e4 LT |
5702 | ret = 0; |
5703 | } | |
5704 | return ret; | |
5705 | } | |
5706 | ||
5707 | /** | |
5708 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5709 | * @pid: pid of the process. | |
5710 | * @interval: userspace pointer to the timeslice value. | |
5711 | * | |
5712 | * this syscall writes the default timeslice value of a given process | |
5713 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5714 | */ | |
17da2bd9 | 5715 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5716 | struct timespec __user *, interval) |
1da177e4 | 5717 | { |
36c8b586 | 5718 | struct task_struct *p; |
a4ec24b4 | 5719 | unsigned int time_slice; |
dba091b9 TG |
5720 | unsigned long flags; |
5721 | struct rq *rq; | |
3a5c359a | 5722 | int retval; |
1da177e4 | 5723 | struct timespec t; |
1da177e4 LT |
5724 | |
5725 | if (pid < 0) | |
3a5c359a | 5726 | return -EINVAL; |
1da177e4 LT |
5727 | |
5728 | retval = -ESRCH; | |
1a551ae7 | 5729 | rcu_read_lock(); |
1da177e4 LT |
5730 | p = find_process_by_pid(pid); |
5731 | if (!p) | |
5732 | goto out_unlock; | |
5733 | ||
5734 | retval = security_task_getscheduler(p); | |
5735 | if (retval) | |
5736 | goto out_unlock; | |
5737 | ||
dba091b9 TG |
5738 | rq = task_rq_lock(p, &flags); |
5739 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5740 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5741 | |
1a551ae7 | 5742 | rcu_read_unlock(); |
a4ec24b4 | 5743 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5744 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5745 | return retval; |
3a5c359a | 5746 | |
1da177e4 | 5747 | out_unlock: |
1a551ae7 | 5748 | rcu_read_unlock(); |
1da177e4 LT |
5749 | return retval; |
5750 | } | |
5751 | ||
7c731e0a | 5752 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5753 | |
82a1fcb9 | 5754 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5755 | { |
1da177e4 | 5756 | unsigned long free = 0; |
36c8b586 | 5757 | unsigned state; |
1da177e4 | 5758 | |
1da177e4 | 5759 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5760 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5761 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5762 | #if BITS_PER_LONG == 32 |
1da177e4 | 5763 | if (state == TASK_RUNNING) |
3df0fc5b | 5764 | printk(KERN_CONT " running "); |
1da177e4 | 5765 | else |
3df0fc5b | 5766 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5767 | #else |
5768 | if (state == TASK_RUNNING) | |
3df0fc5b | 5769 | printk(KERN_CONT " running task "); |
1da177e4 | 5770 | else |
3df0fc5b | 5771 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5772 | #endif |
5773 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5774 | free = stack_not_used(p); |
1da177e4 | 5775 | #endif |
3df0fc5b | 5776 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5777 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5778 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5779 | |
5fb5e6de | 5780 | show_stack(p, NULL); |
1da177e4 LT |
5781 | } |
5782 | ||
e59e2ae2 | 5783 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5784 | { |
36c8b586 | 5785 | struct task_struct *g, *p; |
1da177e4 | 5786 | |
4bd77321 | 5787 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5788 | printk(KERN_INFO |
5789 | " task PC stack pid father\n"); | |
1da177e4 | 5790 | #else |
3df0fc5b PZ |
5791 | printk(KERN_INFO |
5792 | " task PC stack pid father\n"); | |
1da177e4 LT |
5793 | #endif |
5794 | read_lock(&tasklist_lock); | |
5795 | do_each_thread(g, p) { | |
5796 | /* | |
5797 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5798 | * console might take a lot of time: |
1da177e4 LT |
5799 | */ |
5800 | touch_nmi_watchdog(); | |
39bc89fd | 5801 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5802 | sched_show_task(p); |
1da177e4 LT |
5803 | } while_each_thread(g, p); |
5804 | ||
04c9167f JF |
5805 | touch_all_softlockup_watchdogs(); |
5806 | ||
dd41f596 IM |
5807 | #ifdef CONFIG_SCHED_DEBUG |
5808 | sysrq_sched_debug_show(); | |
5809 | #endif | |
1da177e4 | 5810 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5811 | /* |
5812 | * Only show locks if all tasks are dumped: | |
5813 | */ | |
93335a21 | 5814 | if (!state_filter) |
e59e2ae2 | 5815 | debug_show_all_locks(); |
1da177e4 LT |
5816 | } |
5817 | ||
1df21055 IM |
5818 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5819 | { | |
dd41f596 | 5820 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5821 | } |
5822 | ||
f340c0d1 IM |
5823 | /** |
5824 | * init_idle - set up an idle thread for a given CPU | |
5825 | * @idle: task in question | |
5826 | * @cpu: cpu the idle task belongs to | |
5827 | * | |
5828 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5829 | * flag, to make booting more robust. | |
5830 | */ | |
5c1e1767 | 5831 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5832 | { |
70b97a7f | 5833 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5834 | unsigned long flags; |
5835 | ||
05fa785c | 5836 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5837 | |
dd41f596 | 5838 | __sched_fork(idle); |
06b83b5f | 5839 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5840 | idle->se.exec_start = sched_clock(); |
5841 | ||
96f874e2 | 5842 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5843 | /* |
5844 | * We're having a chicken and egg problem, even though we are | |
5845 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5846 | * lockdep check in task_group() will fail. | |
5847 | * | |
5848 | * Similar case to sched_fork(). / Alternatively we could | |
5849 | * use task_rq_lock() here and obtain the other rq->lock. | |
5850 | * | |
5851 | * Silence PROVE_RCU | |
5852 | */ | |
5853 | rcu_read_lock(); | |
dd41f596 | 5854 | __set_task_cpu(idle, cpu); |
6506cf6c | 5855 | rcu_read_unlock(); |
1da177e4 | 5856 | |
1da177e4 | 5857 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
5858 | #if defined(CONFIG_SMP) |
5859 | idle->on_cpu = 1; | |
4866cde0 | 5860 | #endif |
05fa785c | 5861 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5862 | |
5863 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5864 | #if defined(CONFIG_PREEMPT) |
5865 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5866 | #else | |
a1261f54 | 5867 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5868 | #endif |
dd41f596 IM |
5869 | /* |
5870 | * The idle tasks have their own, simple scheduling class: | |
5871 | */ | |
5872 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5873 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5874 | } |
5875 | ||
5876 | /* | |
5877 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5878 | * indicates which cpus entered this state. This is used | |
5879 | * in the rcu update to wait only for active cpus. For system | |
5880 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5881 | * always be CPU_BITS_NONE. |
1da177e4 | 5882 | */ |
6a7b3dc3 | 5883 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5884 | |
19978ca6 IM |
5885 | /* |
5886 | * Increase the granularity value when there are more CPUs, | |
5887 | * because with more CPUs the 'effective latency' as visible | |
5888 | * to users decreases. But the relationship is not linear, | |
5889 | * so pick a second-best guess by going with the log2 of the | |
5890 | * number of CPUs. | |
5891 | * | |
5892 | * This idea comes from the SD scheduler of Con Kolivas: | |
5893 | */ | |
acb4a848 | 5894 | static int get_update_sysctl_factor(void) |
19978ca6 | 5895 | { |
4ca3ef71 | 5896 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5897 | unsigned int factor; |
5898 | ||
5899 | switch (sysctl_sched_tunable_scaling) { | |
5900 | case SCHED_TUNABLESCALING_NONE: | |
5901 | factor = 1; | |
5902 | break; | |
5903 | case SCHED_TUNABLESCALING_LINEAR: | |
5904 | factor = cpus; | |
5905 | break; | |
5906 | case SCHED_TUNABLESCALING_LOG: | |
5907 | default: | |
5908 | factor = 1 + ilog2(cpus); | |
5909 | break; | |
5910 | } | |
19978ca6 | 5911 | |
acb4a848 CE |
5912 | return factor; |
5913 | } | |
19978ca6 | 5914 | |
acb4a848 CE |
5915 | static void update_sysctl(void) |
5916 | { | |
5917 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5918 | |
0bcdcf28 CE |
5919 | #define SET_SYSCTL(name) \ |
5920 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5921 | SET_SYSCTL(sched_min_granularity); | |
5922 | SET_SYSCTL(sched_latency); | |
5923 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5924 | #undef SET_SYSCTL |
5925 | } | |
55cd5340 | 5926 | |
0bcdcf28 CE |
5927 | static inline void sched_init_granularity(void) |
5928 | { | |
5929 | update_sysctl(); | |
19978ca6 IM |
5930 | } |
5931 | ||
1da177e4 LT |
5932 | #ifdef CONFIG_SMP |
5933 | /* | |
5934 | * This is how migration works: | |
5935 | * | |
969c7921 TH |
5936 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5937 | * stop_one_cpu(). | |
5938 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5939 | * off the CPU) | |
5940 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5941 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5942 | * it and puts it into the right queue. |
969c7921 TH |
5943 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5944 | * is done. | |
1da177e4 LT |
5945 | */ |
5946 | ||
5947 | /* | |
5948 | * Change a given task's CPU affinity. Migrate the thread to a | |
5949 | * proper CPU and schedule it away if the CPU it's executing on | |
5950 | * is removed from the allowed bitmask. | |
5951 | * | |
5952 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5953 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5954 | * call is not atomic; no spinlocks may be held. |
5955 | */ | |
96f874e2 | 5956 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5957 | { |
5958 | unsigned long flags; | |
70b97a7f | 5959 | struct rq *rq; |
969c7921 | 5960 | unsigned int dest_cpu; |
48f24c4d | 5961 | int ret = 0; |
1da177e4 LT |
5962 | |
5963 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5964 | |
6ad4c188 | 5965 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5966 | ret = -EINVAL; |
5967 | goto out; | |
5968 | } | |
5969 | ||
9985b0ba | 5970 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5971 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5972 | ret = -EINVAL; |
5973 | goto out; | |
5974 | } | |
5975 | ||
73fe6aae | 5976 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5977 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5978 | else { |
96f874e2 RR |
5979 | cpumask_copy(&p->cpus_allowed, new_mask); |
5980 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5981 | } |
5982 | ||
1da177e4 | 5983 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5984 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5985 | goto out; |
5986 | ||
969c7921 | 5987 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
7608dec2 | 5988 | if (need_migrate_task(p)) { |
969c7921 | 5989 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 5990 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 5991 | task_rq_unlock(rq, p, &flags); |
969c7921 | 5992 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5993 | tlb_migrate_finish(p->mm); |
5994 | return 0; | |
5995 | } | |
5996 | out: | |
0122ec5b | 5997 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 5998 | |
1da177e4 LT |
5999 | return ret; |
6000 | } | |
cd8ba7cd | 6001 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6002 | |
6003 | /* | |
41a2d6cf | 6004 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6005 | * this because either it can't run here any more (set_cpus_allowed() |
6006 | * away from this CPU, or CPU going down), or because we're | |
6007 | * attempting to rebalance this task on exec (sched_exec). | |
6008 | * | |
6009 | * So we race with normal scheduler movements, but that's OK, as long | |
6010 | * as the task is no longer on this CPU. | |
efc30814 KK |
6011 | * |
6012 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6013 | */ |
efc30814 | 6014 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6015 | { |
70b97a7f | 6016 | struct rq *rq_dest, *rq_src; |
e2912009 | 6017 | int ret = 0; |
1da177e4 | 6018 | |
e761b772 | 6019 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6020 | return ret; |
1da177e4 LT |
6021 | |
6022 | rq_src = cpu_rq(src_cpu); | |
6023 | rq_dest = cpu_rq(dest_cpu); | |
6024 | ||
0122ec5b | 6025 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
6026 | double_rq_lock(rq_src, rq_dest); |
6027 | /* Already moved. */ | |
6028 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6029 | goto done; |
1da177e4 | 6030 | /* Affinity changed (again). */ |
96f874e2 | 6031 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6032 | goto fail; |
1da177e4 | 6033 | |
e2912009 PZ |
6034 | /* |
6035 | * If we're not on a rq, the next wake-up will ensure we're | |
6036 | * placed properly. | |
6037 | */ | |
fd2f4419 | 6038 | if (p->on_rq) { |
2e1cb74a | 6039 | deactivate_task(rq_src, p, 0); |
e2912009 | 6040 | set_task_cpu(p, dest_cpu); |
dd41f596 | 6041 | activate_task(rq_dest, p, 0); |
15afe09b | 6042 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6043 | } |
b1e38734 | 6044 | done: |
efc30814 | 6045 | ret = 1; |
b1e38734 | 6046 | fail: |
1da177e4 | 6047 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 6048 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 6049 | return ret; |
1da177e4 LT |
6050 | } |
6051 | ||
6052 | /* | |
969c7921 TH |
6053 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
6054 | * and performs thread migration by bumping thread off CPU then | |
6055 | * 'pushing' onto another runqueue. | |
1da177e4 | 6056 | */ |
969c7921 | 6057 | static int migration_cpu_stop(void *data) |
1da177e4 | 6058 | { |
969c7921 | 6059 | struct migration_arg *arg = data; |
f7b4cddc | 6060 | |
969c7921 TH |
6061 | /* |
6062 | * The original target cpu might have gone down and we might | |
6063 | * be on another cpu but it doesn't matter. | |
6064 | */ | |
f7b4cddc | 6065 | local_irq_disable(); |
969c7921 | 6066 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 6067 | local_irq_enable(); |
1da177e4 | 6068 | return 0; |
f7b4cddc ON |
6069 | } |
6070 | ||
1da177e4 | 6071 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6072 | |
054b9108 | 6073 | /* |
48c5ccae PZ |
6074 | * Ensures that the idle task is using init_mm right before its cpu goes |
6075 | * offline. | |
054b9108 | 6076 | */ |
48c5ccae | 6077 | void idle_task_exit(void) |
1da177e4 | 6078 | { |
48c5ccae | 6079 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6080 | |
48c5ccae | 6081 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6082 | |
48c5ccae PZ |
6083 | if (mm != &init_mm) |
6084 | switch_mm(mm, &init_mm, current); | |
6085 | mmdrop(mm); | |
1da177e4 LT |
6086 | } |
6087 | ||
6088 | /* | |
6089 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6090 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6091 | * for performance reasons the counter is not stricly tracking tasks to | |
6092 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6093 | * to keep the global sum constant after CPU-down: | |
6094 | */ | |
70b97a7f | 6095 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6096 | { |
6ad4c188 | 6097 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6098 | |
1da177e4 LT |
6099 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6100 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6101 | } |
6102 | ||
dd41f596 | 6103 | /* |
48c5ccae | 6104 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6105 | */ |
48c5ccae | 6106 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6107 | { |
48c5ccae PZ |
6108 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6109 | rq->calc_load_active = 0; | |
1da177e4 LT |
6110 | } |
6111 | ||
48f24c4d | 6112 | /* |
48c5ccae PZ |
6113 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6114 | * try_to_wake_up()->select_task_rq(). | |
6115 | * | |
6116 | * Called with rq->lock held even though we'er in stop_machine() and | |
6117 | * there's no concurrency possible, we hold the required locks anyway | |
6118 | * because of lock validation efforts. | |
1da177e4 | 6119 | */ |
48c5ccae | 6120 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6121 | { |
70b97a7f | 6122 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6123 | struct task_struct *next, *stop = rq->stop; |
6124 | int dest_cpu; | |
1da177e4 LT |
6125 | |
6126 | /* | |
48c5ccae PZ |
6127 | * Fudge the rq selection such that the below task selection loop |
6128 | * doesn't get stuck on the currently eligible stop task. | |
6129 | * | |
6130 | * We're currently inside stop_machine() and the rq is either stuck | |
6131 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6132 | * either way we should never end up calling schedule() until we're | |
6133 | * done here. | |
1da177e4 | 6134 | */ |
48c5ccae | 6135 | rq->stop = NULL; |
48f24c4d | 6136 | |
dd41f596 | 6137 | for ( ; ; ) { |
48c5ccae PZ |
6138 | /* |
6139 | * There's this thread running, bail when that's the only | |
6140 | * remaining thread. | |
6141 | */ | |
6142 | if (rq->nr_running == 1) | |
dd41f596 | 6143 | break; |
48c5ccae | 6144 | |
b67802ea | 6145 | next = pick_next_task(rq); |
48c5ccae | 6146 | BUG_ON(!next); |
79c53799 | 6147 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6148 | |
48c5ccae PZ |
6149 | /* Find suitable destination for @next, with force if needed. */ |
6150 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6151 | raw_spin_unlock(&rq->lock); | |
6152 | ||
6153 | __migrate_task(next, dead_cpu, dest_cpu); | |
6154 | ||
6155 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6156 | } |
dce48a84 | 6157 | |
48c5ccae | 6158 | rq->stop = stop; |
dce48a84 | 6159 | } |
48c5ccae | 6160 | |
1da177e4 LT |
6161 | #endif /* CONFIG_HOTPLUG_CPU */ |
6162 | ||
e692ab53 NP |
6163 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6164 | ||
6165 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6166 | { |
6167 | .procname = "sched_domain", | |
c57baf1e | 6168 | .mode = 0555, |
e0361851 | 6169 | }, |
56992309 | 6170 | {} |
e692ab53 NP |
6171 | }; |
6172 | ||
6173 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6174 | { |
6175 | .procname = "kernel", | |
c57baf1e | 6176 | .mode = 0555, |
e0361851 AD |
6177 | .child = sd_ctl_dir, |
6178 | }, | |
56992309 | 6179 | {} |
e692ab53 NP |
6180 | }; |
6181 | ||
6182 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6183 | { | |
6184 | struct ctl_table *entry = | |
5cf9f062 | 6185 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6186 | |
e692ab53 NP |
6187 | return entry; |
6188 | } | |
6189 | ||
6382bc90 MM |
6190 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6191 | { | |
cd790076 | 6192 | struct ctl_table *entry; |
6382bc90 | 6193 | |
cd790076 MM |
6194 | /* |
6195 | * In the intermediate directories, both the child directory and | |
6196 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6197 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6198 | * static strings and all have proc handlers. |
6199 | */ | |
6200 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6201 | if (entry->child) |
6202 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6203 | if (entry->proc_handler == NULL) |
6204 | kfree(entry->procname); | |
6205 | } | |
6382bc90 MM |
6206 | |
6207 | kfree(*tablep); | |
6208 | *tablep = NULL; | |
6209 | } | |
6210 | ||
e692ab53 | 6211 | static void |
e0361851 | 6212 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6213 | const char *procname, void *data, int maxlen, |
6214 | mode_t mode, proc_handler *proc_handler) | |
6215 | { | |
e692ab53 NP |
6216 | entry->procname = procname; |
6217 | entry->data = data; | |
6218 | entry->maxlen = maxlen; | |
6219 | entry->mode = mode; | |
6220 | entry->proc_handler = proc_handler; | |
6221 | } | |
6222 | ||
6223 | static struct ctl_table * | |
6224 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6225 | { | |
a5d8c348 | 6226 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6227 | |
ad1cdc1d MM |
6228 | if (table == NULL) |
6229 | return NULL; | |
6230 | ||
e0361851 | 6231 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6232 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6233 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6234 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6235 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6236 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6237 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6238 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6239 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6240 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6241 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6242 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6243 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6244 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6245 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6246 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6247 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6248 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6249 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6250 | &sd->cache_nice_tries, |
6251 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6252 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6253 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6254 | set_table_entry(&table[11], "name", sd->name, |
6255 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6256 | /* &table[12] is terminator */ | |
e692ab53 NP |
6257 | |
6258 | return table; | |
6259 | } | |
6260 | ||
9a4e7159 | 6261 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6262 | { |
6263 | struct ctl_table *entry, *table; | |
6264 | struct sched_domain *sd; | |
6265 | int domain_num = 0, i; | |
6266 | char buf[32]; | |
6267 | ||
6268 | for_each_domain(cpu, sd) | |
6269 | domain_num++; | |
6270 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6271 | if (table == NULL) |
6272 | return NULL; | |
e692ab53 NP |
6273 | |
6274 | i = 0; | |
6275 | for_each_domain(cpu, sd) { | |
6276 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6277 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6278 | entry->mode = 0555; |
e692ab53 NP |
6279 | entry->child = sd_alloc_ctl_domain_table(sd); |
6280 | entry++; | |
6281 | i++; | |
6282 | } | |
6283 | return table; | |
6284 | } | |
6285 | ||
6286 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6287 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6288 | { |
6ad4c188 | 6289 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6290 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6291 | char buf[32]; | |
6292 | ||
7378547f MM |
6293 | WARN_ON(sd_ctl_dir[0].child); |
6294 | sd_ctl_dir[0].child = entry; | |
6295 | ||
ad1cdc1d MM |
6296 | if (entry == NULL) |
6297 | return; | |
6298 | ||
6ad4c188 | 6299 | for_each_possible_cpu(i) { |
e692ab53 | 6300 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6301 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6302 | entry->mode = 0555; |
e692ab53 | 6303 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6304 | entry++; |
e692ab53 | 6305 | } |
7378547f MM |
6306 | |
6307 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6308 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6309 | } | |
6382bc90 | 6310 | |
7378547f | 6311 | /* may be called multiple times per register */ |
6382bc90 MM |
6312 | static void unregister_sched_domain_sysctl(void) |
6313 | { | |
7378547f MM |
6314 | if (sd_sysctl_header) |
6315 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6316 | sd_sysctl_header = NULL; |
7378547f MM |
6317 | if (sd_ctl_dir[0].child) |
6318 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6319 | } |
e692ab53 | 6320 | #else |
6382bc90 MM |
6321 | static void register_sched_domain_sysctl(void) |
6322 | { | |
6323 | } | |
6324 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6325 | { |
6326 | } | |
6327 | #endif | |
6328 | ||
1f11eb6a GH |
6329 | static void set_rq_online(struct rq *rq) |
6330 | { | |
6331 | if (!rq->online) { | |
6332 | const struct sched_class *class; | |
6333 | ||
c6c4927b | 6334 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6335 | rq->online = 1; |
6336 | ||
6337 | for_each_class(class) { | |
6338 | if (class->rq_online) | |
6339 | class->rq_online(rq); | |
6340 | } | |
6341 | } | |
6342 | } | |
6343 | ||
6344 | static void set_rq_offline(struct rq *rq) | |
6345 | { | |
6346 | if (rq->online) { | |
6347 | const struct sched_class *class; | |
6348 | ||
6349 | for_each_class(class) { | |
6350 | if (class->rq_offline) | |
6351 | class->rq_offline(rq); | |
6352 | } | |
6353 | ||
c6c4927b | 6354 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6355 | rq->online = 0; |
6356 | } | |
6357 | } | |
6358 | ||
1da177e4 LT |
6359 | /* |
6360 | * migration_call - callback that gets triggered when a CPU is added. | |
6361 | * Here we can start up the necessary migration thread for the new CPU. | |
6362 | */ | |
48f24c4d IM |
6363 | static int __cpuinit |
6364 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6365 | { |
48f24c4d | 6366 | int cpu = (long)hcpu; |
1da177e4 | 6367 | unsigned long flags; |
969c7921 | 6368 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6369 | |
48c5ccae | 6370 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6371 | |
1da177e4 | 6372 | case CPU_UP_PREPARE: |
a468d389 | 6373 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6374 | break; |
48f24c4d | 6375 | |
1da177e4 | 6376 | case CPU_ONLINE: |
1f94ef59 | 6377 | /* Update our root-domain */ |
05fa785c | 6378 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6379 | if (rq->rd) { |
c6c4927b | 6380 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6381 | |
6382 | set_rq_online(rq); | |
1f94ef59 | 6383 | } |
05fa785c | 6384 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6385 | break; |
48f24c4d | 6386 | |
1da177e4 | 6387 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6388 | case CPU_DYING: |
317f3941 | 6389 | sched_ttwu_pending(); |
57d885fe | 6390 | /* Update our root-domain */ |
05fa785c | 6391 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6392 | if (rq->rd) { |
c6c4927b | 6393 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6394 | set_rq_offline(rq); |
57d885fe | 6395 | } |
48c5ccae PZ |
6396 | migrate_tasks(cpu); |
6397 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6398 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6399 | |
6400 | migrate_nr_uninterruptible(rq); | |
6401 | calc_global_load_remove(rq); | |
57d885fe | 6402 | break; |
1da177e4 LT |
6403 | #endif |
6404 | } | |
49c022e6 PZ |
6405 | |
6406 | update_max_interval(); | |
6407 | ||
1da177e4 LT |
6408 | return NOTIFY_OK; |
6409 | } | |
6410 | ||
f38b0820 PM |
6411 | /* |
6412 | * Register at high priority so that task migration (migrate_all_tasks) | |
6413 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6414 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6415 | */ |
26c2143b | 6416 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6417 | .notifier_call = migration_call, |
50a323b7 | 6418 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6419 | }; |
6420 | ||
3a101d05 TH |
6421 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6422 | unsigned long action, void *hcpu) | |
6423 | { | |
6424 | switch (action & ~CPU_TASKS_FROZEN) { | |
6425 | case CPU_ONLINE: | |
6426 | case CPU_DOWN_FAILED: | |
6427 | set_cpu_active((long)hcpu, true); | |
6428 | return NOTIFY_OK; | |
6429 | default: | |
6430 | return NOTIFY_DONE; | |
6431 | } | |
6432 | } | |
6433 | ||
6434 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6435 | unsigned long action, void *hcpu) | |
6436 | { | |
6437 | switch (action & ~CPU_TASKS_FROZEN) { | |
6438 | case CPU_DOWN_PREPARE: | |
6439 | set_cpu_active((long)hcpu, false); | |
6440 | return NOTIFY_OK; | |
6441 | default: | |
6442 | return NOTIFY_DONE; | |
6443 | } | |
6444 | } | |
6445 | ||
7babe8db | 6446 | static int __init migration_init(void) |
1da177e4 LT |
6447 | { |
6448 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6449 | int err; |
48f24c4d | 6450 | |
3a101d05 | 6451 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6452 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6453 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6454 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6455 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6456 | |
3a101d05 TH |
6457 | /* Register cpu active notifiers */ |
6458 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6459 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6460 | ||
a004cd42 | 6461 | return 0; |
1da177e4 | 6462 | } |
7babe8db | 6463 | early_initcall(migration_init); |
1da177e4 LT |
6464 | #endif |
6465 | ||
6466 | #ifdef CONFIG_SMP | |
476f3534 | 6467 | |
3e9830dc | 6468 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6469 | |
f6630114 MT |
6470 | static __read_mostly int sched_domain_debug_enabled; |
6471 | ||
6472 | static int __init sched_domain_debug_setup(char *str) | |
6473 | { | |
6474 | sched_domain_debug_enabled = 1; | |
6475 | ||
6476 | return 0; | |
6477 | } | |
6478 | early_param("sched_debug", sched_domain_debug_setup); | |
6479 | ||
7c16ec58 | 6480 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6481 | struct cpumask *groupmask) |
1da177e4 | 6482 | { |
4dcf6aff | 6483 | struct sched_group *group = sd->groups; |
434d53b0 | 6484 | char str[256]; |
1da177e4 | 6485 | |
968ea6d8 | 6486 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6487 | cpumask_clear(groupmask); |
4dcf6aff IM |
6488 | |
6489 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6490 | ||
6491 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6492 | printk("does not load-balance\n"); |
4dcf6aff | 6493 | if (sd->parent) |
3df0fc5b PZ |
6494 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6495 | " has parent"); | |
4dcf6aff | 6496 | return -1; |
41c7ce9a NP |
6497 | } |
6498 | ||
3df0fc5b | 6499 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6500 | |
758b2cdc | 6501 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6502 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6503 | "CPU%d\n", cpu); | |
4dcf6aff | 6504 | } |
758b2cdc | 6505 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6506 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6507 | " CPU%d\n", cpu); | |
4dcf6aff | 6508 | } |
1da177e4 | 6509 | |
4dcf6aff | 6510 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6511 | do { |
4dcf6aff | 6512 | if (!group) { |
3df0fc5b PZ |
6513 | printk("\n"); |
6514 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6515 | break; |
6516 | } | |
6517 | ||
18a3885f | 6518 | if (!group->cpu_power) { |
3df0fc5b PZ |
6519 | printk(KERN_CONT "\n"); |
6520 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6521 | "set\n"); | |
4dcf6aff IM |
6522 | break; |
6523 | } | |
1da177e4 | 6524 | |
758b2cdc | 6525 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6526 | printk(KERN_CONT "\n"); |
6527 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6528 | break; |
6529 | } | |
1da177e4 | 6530 | |
758b2cdc | 6531 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6532 | printk(KERN_CONT "\n"); |
6533 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6534 | break; |
6535 | } | |
1da177e4 | 6536 | |
758b2cdc | 6537 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6538 | |
968ea6d8 | 6539 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6540 | |
3df0fc5b | 6541 | printk(KERN_CONT " %s", str); |
18a3885f | 6542 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6543 | printk(KERN_CONT " (cpu_power = %d)", |
6544 | group->cpu_power); | |
381512cf | 6545 | } |
1da177e4 | 6546 | |
4dcf6aff IM |
6547 | group = group->next; |
6548 | } while (group != sd->groups); | |
3df0fc5b | 6549 | printk(KERN_CONT "\n"); |
1da177e4 | 6550 | |
758b2cdc | 6551 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6552 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6553 | |
758b2cdc RR |
6554 | if (sd->parent && |
6555 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6556 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6557 | "of domain->span\n"); | |
4dcf6aff IM |
6558 | return 0; |
6559 | } | |
1da177e4 | 6560 | |
4dcf6aff IM |
6561 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6562 | { | |
d5dd3db1 | 6563 | cpumask_var_t groupmask; |
4dcf6aff | 6564 | int level = 0; |
1da177e4 | 6565 | |
f6630114 MT |
6566 | if (!sched_domain_debug_enabled) |
6567 | return; | |
6568 | ||
4dcf6aff IM |
6569 | if (!sd) { |
6570 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6571 | return; | |
6572 | } | |
1da177e4 | 6573 | |
4dcf6aff IM |
6574 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6575 | ||
d5dd3db1 | 6576 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6577 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6578 | return; | |
6579 | } | |
6580 | ||
4dcf6aff | 6581 | for (;;) { |
7c16ec58 | 6582 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6583 | break; |
1da177e4 LT |
6584 | level++; |
6585 | sd = sd->parent; | |
33859f7f | 6586 | if (!sd) |
4dcf6aff IM |
6587 | break; |
6588 | } | |
d5dd3db1 | 6589 | free_cpumask_var(groupmask); |
1da177e4 | 6590 | } |
6d6bc0ad | 6591 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6592 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6593 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6594 | |
1a20ff27 | 6595 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6596 | { |
758b2cdc | 6597 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6598 | return 1; |
6599 | ||
6600 | /* Following flags need at least 2 groups */ | |
6601 | if (sd->flags & (SD_LOAD_BALANCE | | |
6602 | SD_BALANCE_NEWIDLE | | |
6603 | SD_BALANCE_FORK | | |
89c4710e SS |
6604 | SD_BALANCE_EXEC | |
6605 | SD_SHARE_CPUPOWER | | |
6606 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6607 | if (sd->groups != sd->groups->next) |
6608 | return 0; | |
6609 | } | |
6610 | ||
6611 | /* Following flags don't use groups */ | |
c88d5910 | 6612 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6613 | return 0; |
6614 | ||
6615 | return 1; | |
6616 | } | |
6617 | ||
48f24c4d IM |
6618 | static int |
6619 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6620 | { |
6621 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6622 | ||
6623 | if (sd_degenerate(parent)) | |
6624 | return 1; | |
6625 | ||
758b2cdc | 6626 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6627 | return 0; |
6628 | ||
245af2c7 SS |
6629 | /* Flags needing groups don't count if only 1 group in parent */ |
6630 | if (parent->groups == parent->groups->next) { | |
6631 | pflags &= ~(SD_LOAD_BALANCE | | |
6632 | SD_BALANCE_NEWIDLE | | |
6633 | SD_BALANCE_FORK | | |
89c4710e SS |
6634 | SD_BALANCE_EXEC | |
6635 | SD_SHARE_CPUPOWER | | |
6636 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6637 | if (nr_node_ids == 1) |
6638 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6639 | } |
6640 | if (~cflags & pflags) | |
6641 | return 0; | |
6642 | ||
6643 | return 1; | |
6644 | } | |
6645 | ||
c6c4927b RR |
6646 | static void free_rootdomain(struct root_domain *rd) |
6647 | { | |
047106ad PZ |
6648 | synchronize_sched(); |
6649 | ||
68e74568 RR |
6650 | cpupri_cleanup(&rd->cpupri); |
6651 | ||
c6c4927b RR |
6652 | free_cpumask_var(rd->rto_mask); |
6653 | free_cpumask_var(rd->online); | |
6654 | free_cpumask_var(rd->span); | |
6655 | kfree(rd); | |
6656 | } | |
6657 | ||
57d885fe GH |
6658 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6659 | { | |
a0490fa3 | 6660 | struct root_domain *old_rd = NULL; |
57d885fe | 6661 | unsigned long flags; |
57d885fe | 6662 | |
05fa785c | 6663 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6664 | |
6665 | if (rq->rd) { | |
a0490fa3 | 6666 | old_rd = rq->rd; |
57d885fe | 6667 | |
c6c4927b | 6668 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6669 | set_rq_offline(rq); |
57d885fe | 6670 | |
c6c4927b | 6671 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6672 | |
a0490fa3 IM |
6673 | /* |
6674 | * If we dont want to free the old_rt yet then | |
6675 | * set old_rd to NULL to skip the freeing later | |
6676 | * in this function: | |
6677 | */ | |
6678 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6679 | old_rd = NULL; | |
57d885fe GH |
6680 | } |
6681 | ||
6682 | atomic_inc(&rd->refcount); | |
6683 | rq->rd = rd; | |
6684 | ||
c6c4927b | 6685 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6686 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6687 | set_rq_online(rq); |
57d885fe | 6688 | |
05fa785c | 6689 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6690 | |
6691 | if (old_rd) | |
6692 | free_rootdomain(old_rd); | |
57d885fe GH |
6693 | } |
6694 | ||
68c38fc3 | 6695 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6696 | { |
6697 | memset(rd, 0, sizeof(*rd)); | |
6698 | ||
68c38fc3 | 6699 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6700 | goto out; |
68c38fc3 | 6701 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6702 | goto free_span; |
68c38fc3 | 6703 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6704 | goto free_online; |
6e0534f2 | 6705 | |
68c38fc3 | 6706 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6707 | goto free_rto_mask; |
c6c4927b | 6708 | return 0; |
6e0534f2 | 6709 | |
68e74568 RR |
6710 | free_rto_mask: |
6711 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6712 | free_online: |
6713 | free_cpumask_var(rd->online); | |
6714 | free_span: | |
6715 | free_cpumask_var(rd->span); | |
0c910d28 | 6716 | out: |
c6c4927b | 6717 | return -ENOMEM; |
57d885fe GH |
6718 | } |
6719 | ||
6720 | static void init_defrootdomain(void) | |
6721 | { | |
68c38fc3 | 6722 | init_rootdomain(&def_root_domain); |
c6c4927b | 6723 | |
57d885fe GH |
6724 | atomic_set(&def_root_domain.refcount, 1); |
6725 | } | |
6726 | ||
dc938520 | 6727 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6728 | { |
6729 | struct root_domain *rd; | |
6730 | ||
6731 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6732 | if (!rd) | |
6733 | return NULL; | |
6734 | ||
68c38fc3 | 6735 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6736 | kfree(rd); |
6737 | return NULL; | |
6738 | } | |
57d885fe GH |
6739 | |
6740 | return rd; | |
6741 | } | |
6742 | ||
1da177e4 | 6743 | /* |
0eab9146 | 6744 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6745 | * hold the hotplug lock. |
6746 | */ | |
0eab9146 IM |
6747 | static void |
6748 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6749 | { |
70b97a7f | 6750 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6751 | struct sched_domain *tmp; |
6752 | ||
669c55e9 PZ |
6753 | for (tmp = sd; tmp; tmp = tmp->parent) |
6754 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6755 | ||
245af2c7 | 6756 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6757 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6758 | struct sched_domain *parent = tmp->parent; |
6759 | if (!parent) | |
6760 | break; | |
f29c9b1c | 6761 | |
1a848870 | 6762 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6763 | tmp->parent = parent->parent; |
1a848870 SS |
6764 | if (parent->parent) |
6765 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6766 | } else |
6767 | tmp = tmp->parent; | |
245af2c7 SS |
6768 | } |
6769 | ||
1a848870 | 6770 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6771 | sd = sd->parent; |
1a848870 SS |
6772 | if (sd) |
6773 | sd->child = NULL; | |
6774 | } | |
1da177e4 LT |
6775 | |
6776 | sched_domain_debug(sd, cpu); | |
6777 | ||
57d885fe | 6778 | rq_attach_root(rq, rd); |
674311d5 | 6779 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6780 | } |
6781 | ||
6782 | /* cpus with isolated domains */ | |
dcc30a35 | 6783 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6784 | |
6785 | /* Setup the mask of cpus configured for isolated domains */ | |
6786 | static int __init isolated_cpu_setup(char *str) | |
6787 | { | |
bdddd296 | 6788 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6789 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6790 | return 1; |
6791 | } | |
6792 | ||
8927f494 | 6793 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6794 | |
6795 | /* | |
6711cab4 SS |
6796 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6797 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6798 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6799 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6800 | * |
6801 | * init_sched_build_groups will build a circular linked list of the groups | |
6802 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6803 | * and ->cpu_power to 0. | |
6804 | */ | |
a616058b | 6805 | static void |
96f874e2 RR |
6806 | init_sched_build_groups(const struct cpumask *span, |
6807 | const struct cpumask *cpu_map, | |
6808 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6809 | struct sched_group **sg, |
96f874e2 RR |
6810 | struct cpumask *tmpmask), |
6811 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6812 | { |
6813 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6814 | int i; |
6815 | ||
96f874e2 | 6816 | cpumask_clear(covered); |
7c16ec58 | 6817 | |
abcd083a | 6818 | for_each_cpu(i, span) { |
6711cab4 | 6819 | struct sched_group *sg; |
7c16ec58 | 6820 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6821 | int j; |
6822 | ||
758b2cdc | 6823 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6824 | continue; |
6825 | ||
758b2cdc | 6826 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6827 | sg->cpu_power = 0; |
1da177e4 | 6828 | |
abcd083a | 6829 | for_each_cpu(j, span) { |
7c16ec58 | 6830 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6831 | continue; |
6832 | ||
96f874e2 | 6833 | cpumask_set_cpu(j, covered); |
758b2cdc | 6834 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6835 | } |
6836 | if (!first) | |
6837 | first = sg; | |
6838 | if (last) | |
6839 | last->next = sg; | |
6840 | last = sg; | |
6841 | } | |
6842 | last->next = first; | |
6843 | } | |
6844 | ||
9c1cfda2 | 6845 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6846 | |
9c1cfda2 | 6847 | #ifdef CONFIG_NUMA |
198e2f18 | 6848 | |
9c1cfda2 JH |
6849 | /** |
6850 | * find_next_best_node - find the next node to include in a sched_domain | |
6851 | * @node: node whose sched_domain we're building | |
6852 | * @used_nodes: nodes already in the sched_domain | |
6853 | * | |
41a2d6cf | 6854 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6855 | * finds the closest node not already in the @used_nodes map. |
6856 | * | |
6857 | * Should use nodemask_t. | |
6858 | */ | |
c5f59f08 | 6859 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6860 | { |
6861 | int i, n, val, min_val, best_node = 0; | |
6862 | ||
6863 | min_val = INT_MAX; | |
6864 | ||
076ac2af | 6865 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6866 | /* Start at @node */ |
076ac2af | 6867 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6868 | |
6869 | if (!nr_cpus_node(n)) | |
6870 | continue; | |
6871 | ||
6872 | /* Skip already used nodes */ | |
c5f59f08 | 6873 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6874 | continue; |
6875 | ||
6876 | /* Simple min distance search */ | |
6877 | val = node_distance(node, n); | |
6878 | ||
6879 | if (val < min_val) { | |
6880 | min_val = val; | |
6881 | best_node = n; | |
6882 | } | |
6883 | } | |
6884 | ||
c5f59f08 | 6885 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6886 | return best_node; |
6887 | } | |
6888 | ||
6889 | /** | |
6890 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6891 | * @node: node whose cpumask we're constructing | |
73486722 | 6892 | * @span: resulting cpumask |
9c1cfda2 | 6893 | * |
41a2d6cf | 6894 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6895 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6896 | * out optimally. | |
6897 | */ | |
96f874e2 | 6898 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6899 | { |
c5f59f08 | 6900 | nodemask_t used_nodes; |
48f24c4d | 6901 | int i; |
9c1cfda2 | 6902 | |
6ca09dfc | 6903 | cpumask_clear(span); |
c5f59f08 | 6904 | nodes_clear(used_nodes); |
9c1cfda2 | 6905 | |
6ca09dfc | 6906 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6907 | node_set(node, used_nodes); |
9c1cfda2 JH |
6908 | |
6909 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6910 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6911 | |
6ca09dfc | 6912 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6913 | } |
9c1cfda2 | 6914 | } |
6d6bc0ad | 6915 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6916 | |
5c45bf27 | 6917 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6918 | |
6c99e9ad RR |
6919 | /* |
6920 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6921 | * |
6922 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6923 | * and struct sched_domain. ) | |
6c99e9ad RR |
6924 | */ |
6925 | struct static_sched_group { | |
6926 | struct sched_group sg; | |
6927 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6928 | }; | |
6929 | ||
6930 | struct static_sched_domain { | |
6931 | struct sched_domain sd; | |
6932 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6933 | }; | |
6934 | ||
49a02c51 AH |
6935 | struct s_data { |
6936 | #ifdef CONFIG_NUMA | |
6937 | int sd_allnodes; | |
6938 | cpumask_var_t domainspan; | |
6939 | cpumask_var_t covered; | |
6940 | cpumask_var_t notcovered; | |
6941 | #endif | |
6942 | cpumask_var_t nodemask; | |
6943 | cpumask_var_t this_sibling_map; | |
6944 | cpumask_var_t this_core_map; | |
01a08546 | 6945 | cpumask_var_t this_book_map; |
49a02c51 AH |
6946 | cpumask_var_t send_covered; |
6947 | cpumask_var_t tmpmask; | |
6948 | struct sched_group **sched_group_nodes; | |
6949 | struct root_domain *rd; | |
6950 | }; | |
6951 | ||
2109b99e AH |
6952 | enum s_alloc { |
6953 | sa_sched_groups = 0, | |
6954 | sa_rootdomain, | |
6955 | sa_tmpmask, | |
6956 | sa_send_covered, | |
01a08546 | 6957 | sa_this_book_map, |
2109b99e AH |
6958 | sa_this_core_map, |
6959 | sa_this_sibling_map, | |
6960 | sa_nodemask, | |
6961 | sa_sched_group_nodes, | |
6962 | #ifdef CONFIG_NUMA | |
6963 | sa_notcovered, | |
6964 | sa_covered, | |
6965 | sa_domainspan, | |
6966 | #endif | |
6967 | sa_none, | |
6968 | }; | |
6969 | ||
9c1cfda2 | 6970 | /* |
48f24c4d | 6971 | * SMT sched-domains: |
9c1cfda2 | 6972 | */ |
1da177e4 | 6973 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6974 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6975 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6976 | |
41a2d6cf | 6977 | static int |
96f874e2 RR |
6978 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6979 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6980 | { |
6711cab4 | 6981 | if (sg) |
1871e52c | 6982 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6983 | return cpu; |
6984 | } | |
6d6bc0ad | 6985 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6986 | |
48f24c4d IM |
6987 | /* |
6988 | * multi-core sched-domains: | |
6989 | */ | |
1e9f28fa | 6990 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6991 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6992 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6993 | |
41a2d6cf | 6994 | static int |
96f874e2 RR |
6995 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6996 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6997 | { |
6711cab4 | 6998 | int group; |
f269893c | 6999 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 7000 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7001 | group = cpumask_first(mask); |
f269893c HC |
7002 | #else |
7003 | group = cpu; | |
7004 | #endif | |
6711cab4 | 7005 | if (sg) |
6c99e9ad | 7006 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7007 | return group; |
1e9f28fa | 7008 | } |
f269893c | 7009 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 7010 | |
01a08546 HC |
7011 | /* |
7012 | * book sched-domains: | |
7013 | */ | |
7014 | #ifdef CONFIG_SCHED_BOOK | |
7015 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
7016 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
7017 | ||
41a2d6cf | 7018 | static int |
01a08546 HC |
7019 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
7020 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7021 | { |
01a08546 HC |
7022 | int group = cpu; |
7023 | #ifdef CONFIG_SCHED_MC | |
7024 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
7025 | group = cpumask_first(mask); | |
7026 | #elif defined(CONFIG_SCHED_SMT) | |
7027 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
7028 | group = cpumask_first(mask); | |
7029 | #endif | |
6711cab4 | 7030 | if (sg) |
01a08546 HC |
7031 | *sg = &per_cpu(sched_group_book, group).sg; |
7032 | return group; | |
1e9f28fa | 7033 | } |
01a08546 | 7034 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 7035 | |
6c99e9ad RR |
7036 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7037 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7038 | |
41a2d6cf | 7039 | static int |
96f874e2 RR |
7040 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7041 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7042 | { |
6711cab4 | 7043 | int group; |
01a08546 HC |
7044 | #ifdef CONFIG_SCHED_BOOK |
7045 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
7046 | group = cpumask_first(mask); | |
7047 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 7048 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7049 | group = cpumask_first(mask); |
1e9f28fa | 7050 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 7051 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7052 | group = cpumask_first(mask); |
1da177e4 | 7053 | #else |
6711cab4 | 7054 | group = cpu; |
1da177e4 | 7055 | #endif |
6711cab4 | 7056 | if (sg) |
6c99e9ad | 7057 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7058 | return group; |
1da177e4 LT |
7059 | } |
7060 | ||
7061 | #ifdef CONFIG_NUMA | |
1da177e4 | 7062 | /* |
9c1cfda2 JH |
7063 | * The init_sched_build_groups can't handle what we want to do with node |
7064 | * groups, so roll our own. Now each node has its own list of groups which | |
7065 | * gets dynamically allocated. | |
1da177e4 | 7066 | */ |
62ea9ceb | 7067 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7068 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7069 | |
62ea9ceb | 7070 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7071 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7072 | |
96f874e2 RR |
7073 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7074 | struct sched_group **sg, | |
7075 | struct cpumask *nodemask) | |
9c1cfda2 | 7076 | { |
6711cab4 SS |
7077 | int group; |
7078 | ||
6ca09dfc | 7079 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7080 | group = cpumask_first(nodemask); |
6711cab4 SS |
7081 | |
7082 | if (sg) | |
6c99e9ad | 7083 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7084 | return group; |
1da177e4 | 7085 | } |
6711cab4 | 7086 | |
08069033 SS |
7087 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7088 | { | |
7089 | struct sched_group *sg = group_head; | |
7090 | int j; | |
7091 | ||
7092 | if (!sg) | |
7093 | return; | |
3a5c359a | 7094 | do { |
758b2cdc | 7095 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7096 | struct sched_domain *sd; |
08069033 | 7097 | |
6c99e9ad | 7098 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 7099 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
7100 | /* |
7101 | * Only add "power" once for each | |
7102 | * physical package. | |
7103 | */ | |
7104 | continue; | |
7105 | } | |
08069033 | 7106 | |
18a3885f | 7107 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
7108 | } |
7109 | sg = sg->next; | |
7110 | } while (sg != group_head); | |
08069033 | 7111 | } |
0601a88d AH |
7112 | |
7113 | static int build_numa_sched_groups(struct s_data *d, | |
7114 | const struct cpumask *cpu_map, int num) | |
7115 | { | |
7116 | struct sched_domain *sd; | |
7117 | struct sched_group *sg, *prev; | |
7118 | int n, j; | |
7119 | ||
7120 | cpumask_clear(d->covered); | |
7121 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
7122 | if (cpumask_empty(d->nodemask)) { | |
7123 | d->sched_group_nodes[num] = NULL; | |
7124 | goto out; | |
7125 | } | |
7126 | ||
7127 | sched_domain_node_span(num, d->domainspan); | |
7128 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
7129 | ||
7130 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7131 | GFP_KERNEL, num); | |
7132 | if (!sg) { | |
3df0fc5b PZ |
7133 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
7134 | num); | |
0601a88d AH |
7135 | return -ENOMEM; |
7136 | } | |
7137 | d->sched_group_nodes[num] = sg; | |
7138 | ||
7139 | for_each_cpu(j, d->nodemask) { | |
7140 | sd = &per_cpu(node_domains, j).sd; | |
7141 | sd->groups = sg; | |
7142 | } | |
7143 | ||
18a3885f | 7144 | sg->cpu_power = 0; |
0601a88d AH |
7145 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7146 | sg->next = sg; | |
7147 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7148 | ||
7149 | prev = sg; | |
7150 | for (j = 0; j < nr_node_ids; j++) { | |
7151 | n = (num + j) % nr_node_ids; | |
7152 | cpumask_complement(d->notcovered, d->covered); | |
7153 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7154 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7155 | if (cpumask_empty(d->tmpmask)) | |
7156 | break; | |
7157 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7158 | if (cpumask_empty(d->tmpmask)) | |
7159 | continue; | |
7160 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7161 | GFP_KERNEL, num); | |
7162 | if (!sg) { | |
3df0fc5b PZ |
7163 | printk(KERN_WARNING |
7164 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7165 | return -ENOMEM; |
7166 | } | |
18a3885f | 7167 | sg->cpu_power = 0; |
0601a88d AH |
7168 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7169 | sg->next = prev->next; | |
7170 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7171 | prev->next = sg; | |
7172 | prev = sg; | |
7173 | } | |
7174 | out: | |
7175 | return 0; | |
7176 | } | |
6d6bc0ad | 7177 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7178 | |
a616058b | 7179 | #ifdef CONFIG_NUMA |
51888ca2 | 7180 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7181 | static void free_sched_groups(const struct cpumask *cpu_map, |
7182 | struct cpumask *nodemask) | |
51888ca2 | 7183 | { |
a616058b | 7184 | int cpu, i; |
51888ca2 | 7185 | |
abcd083a | 7186 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7187 | struct sched_group **sched_group_nodes |
7188 | = sched_group_nodes_bycpu[cpu]; | |
7189 | ||
51888ca2 SV |
7190 | if (!sched_group_nodes) |
7191 | continue; | |
7192 | ||
076ac2af | 7193 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7194 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7195 | ||
6ca09dfc | 7196 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7197 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7198 | continue; |
7199 | ||
7200 | if (sg == NULL) | |
7201 | continue; | |
7202 | sg = sg->next; | |
7203 | next_sg: | |
7204 | oldsg = sg; | |
7205 | sg = sg->next; | |
7206 | kfree(oldsg); | |
7207 | if (oldsg != sched_group_nodes[i]) | |
7208 | goto next_sg; | |
7209 | } | |
7210 | kfree(sched_group_nodes); | |
7211 | sched_group_nodes_bycpu[cpu] = NULL; | |
7212 | } | |
51888ca2 | 7213 | } |
6d6bc0ad | 7214 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7215 | static void free_sched_groups(const struct cpumask *cpu_map, |
7216 | struct cpumask *nodemask) | |
a616058b SS |
7217 | { |
7218 | } | |
6d6bc0ad | 7219 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7220 | |
89c4710e SS |
7221 | /* |
7222 | * Initialize sched groups cpu_power. | |
7223 | * | |
7224 | * cpu_power indicates the capacity of sched group, which is used while | |
7225 | * distributing the load between different sched groups in a sched domain. | |
7226 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7227 | * there are asymmetries in the topology. If there are asymmetries, group | |
7228 | * having more cpu_power will pickup more load compared to the group having | |
7229 | * less cpu_power. | |
89c4710e SS |
7230 | */ |
7231 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7232 | { | |
7233 | struct sched_domain *child; | |
7234 | struct sched_group *group; | |
f93e65c1 PZ |
7235 | long power; |
7236 | int weight; | |
89c4710e SS |
7237 | |
7238 | WARN_ON(!sd || !sd->groups); | |
7239 | ||
13318a71 | 7240 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7241 | return; |
7242 | ||
aae6d3dd SS |
7243 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7244 | ||
89c4710e SS |
7245 | child = sd->child; |
7246 | ||
18a3885f | 7247 | sd->groups->cpu_power = 0; |
5517d86b | 7248 | |
f93e65c1 PZ |
7249 | if (!child) { |
7250 | power = SCHED_LOAD_SCALE; | |
7251 | weight = cpumask_weight(sched_domain_span(sd)); | |
7252 | /* | |
7253 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7254 | * Usually multiple threads get a better yield out of |
7255 | * that one core than a single thread would have, | |
7256 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7257 | */ |
a52bfd73 PZ |
7258 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7259 | power *= sd->smt_gain; | |
f93e65c1 | 7260 | power /= weight; |
a52bfd73 PZ |
7261 | power >>= SCHED_LOAD_SHIFT; |
7262 | } | |
18a3885f | 7263 | sd->groups->cpu_power += power; |
89c4710e SS |
7264 | return; |
7265 | } | |
7266 | ||
89c4710e | 7267 | /* |
f93e65c1 | 7268 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7269 | */ |
7270 | group = child->groups; | |
7271 | do { | |
18a3885f | 7272 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7273 | group = group->next; |
7274 | } while (group != child->groups); | |
7275 | } | |
7276 | ||
7c16ec58 MT |
7277 | /* |
7278 | * Initializers for schedule domains | |
7279 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7280 | */ | |
7281 | ||
a5d8c348 IM |
7282 | #ifdef CONFIG_SCHED_DEBUG |
7283 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7284 | #else | |
7285 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7286 | #endif | |
7287 | ||
7c16ec58 | 7288 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7289 | |
7c16ec58 MT |
7290 | #define SD_INIT_FUNC(type) \ |
7291 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7292 | { \ | |
7293 | memset(sd, 0, sizeof(*sd)); \ | |
7294 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7295 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7296 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7297 | } |
7298 | ||
7299 | SD_INIT_FUNC(CPU) | |
7300 | #ifdef CONFIG_NUMA | |
7301 | SD_INIT_FUNC(ALLNODES) | |
7302 | SD_INIT_FUNC(NODE) | |
7303 | #endif | |
7304 | #ifdef CONFIG_SCHED_SMT | |
7305 | SD_INIT_FUNC(SIBLING) | |
7306 | #endif | |
7307 | #ifdef CONFIG_SCHED_MC | |
7308 | SD_INIT_FUNC(MC) | |
7309 | #endif | |
01a08546 HC |
7310 | #ifdef CONFIG_SCHED_BOOK |
7311 | SD_INIT_FUNC(BOOK) | |
7312 | #endif | |
7c16ec58 | 7313 | |
1d3504fc HS |
7314 | static int default_relax_domain_level = -1; |
7315 | ||
7316 | static int __init setup_relax_domain_level(char *str) | |
7317 | { | |
30e0e178 LZ |
7318 | unsigned long val; |
7319 | ||
7320 | val = simple_strtoul(str, NULL, 0); | |
7321 | if (val < SD_LV_MAX) | |
7322 | default_relax_domain_level = val; | |
7323 | ||
1d3504fc HS |
7324 | return 1; |
7325 | } | |
7326 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7327 | ||
7328 | static void set_domain_attribute(struct sched_domain *sd, | |
7329 | struct sched_domain_attr *attr) | |
7330 | { | |
7331 | int request; | |
7332 | ||
7333 | if (!attr || attr->relax_domain_level < 0) { | |
7334 | if (default_relax_domain_level < 0) | |
7335 | return; | |
7336 | else | |
7337 | request = default_relax_domain_level; | |
7338 | } else | |
7339 | request = attr->relax_domain_level; | |
7340 | if (request < sd->level) { | |
7341 | /* turn off idle balance on this domain */ | |
c88d5910 | 7342 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7343 | } else { |
7344 | /* turn on idle balance on this domain */ | |
c88d5910 | 7345 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7346 | } |
7347 | } | |
7348 | ||
2109b99e AH |
7349 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7350 | const struct cpumask *cpu_map) | |
7351 | { | |
7352 | switch (what) { | |
7353 | case sa_sched_groups: | |
7354 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7355 | d->sched_group_nodes = NULL; | |
7356 | case sa_rootdomain: | |
7357 | free_rootdomain(d->rd); /* fall through */ | |
7358 | case sa_tmpmask: | |
7359 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7360 | case sa_send_covered: | |
7361 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7362 | case sa_this_book_map: |
7363 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7364 | case sa_this_core_map: |
7365 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7366 | case sa_this_sibling_map: | |
7367 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7368 | case sa_nodemask: | |
7369 | free_cpumask_var(d->nodemask); /* fall through */ | |
7370 | case sa_sched_group_nodes: | |
d1b55138 | 7371 | #ifdef CONFIG_NUMA |
2109b99e AH |
7372 | kfree(d->sched_group_nodes); /* fall through */ |
7373 | case sa_notcovered: | |
7374 | free_cpumask_var(d->notcovered); /* fall through */ | |
7375 | case sa_covered: | |
7376 | free_cpumask_var(d->covered); /* fall through */ | |
7377 | case sa_domainspan: | |
7378 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7379 | #endif |
2109b99e AH |
7380 | case sa_none: |
7381 | break; | |
7382 | } | |
7383 | } | |
3404c8d9 | 7384 | |
2109b99e AH |
7385 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7386 | const struct cpumask *cpu_map) | |
7387 | { | |
3404c8d9 | 7388 | #ifdef CONFIG_NUMA |
2109b99e AH |
7389 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7390 | return sa_none; | |
7391 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7392 | return sa_domainspan; | |
7393 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7394 | return sa_covered; | |
7395 | /* Allocate the per-node list of sched groups */ | |
7396 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7397 | sizeof(struct sched_group *), GFP_KERNEL); | |
7398 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7399 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7400 | return sa_notcovered; |
d1b55138 | 7401 | } |
2109b99e | 7402 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7403 | #endif |
2109b99e AH |
7404 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7405 | return sa_sched_group_nodes; | |
7406 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7407 | return sa_nodemask; | |
7408 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7409 | return sa_this_sibling_map; | |
01a08546 | 7410 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7411 | return sa_this_core_map; |
01a08546 HC |
7412 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7413 | return sa_this_book_map; | |
2109b99e AH |
7414 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7415 | return sa_send_covered; | |
7416 | d->rd = alloc_rootdomain(); | |
7417 | if (!d->rd) { | |
3df0fc5b | 7418 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7419 | return sa_tmpmask; |
57d885fe | 7420 | } |
2109b99e AH |
7421 | return sa_rootdomain; |
7422 | } | |
57d885fe | 7423 | |
7f4588f3 AH |
7424 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7425 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7426 | { | |
7427 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7428 | #ifdef CONFIG_NUMA |
7f4588f3 | 7429 | struct sched_domain *parent; |
1da177e4 | 7430 | |
7f4588f3 AH |
7431 | d->sd_allnodes = 0; |
7432 | if (cpumask_weight(cpu_map) > | |
7433 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7434 | sd = &per_cpu(allnodes_domains, i).sd; | |
7435 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7436 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7437 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7438 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7439 | d->sd_allnodes = 1; | |
7440 | } | |
7441 | parent = sd; | |
7442 | ||
7443 | sd = &per_cpu(node_domains, i).sd; | |
7444 | SD_INIT(sd, NODE); | |
7445 | set_domain_attribute(sd, attr); | |
7446 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7447 | sd->parent = parent; | |
7448 | if (parent) | |
7449 | parent->child = sd; | |
7450 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7451 | #endif |
7f4588f3 AH |
7452 | return sd; |
7453 | } | |
1da177e4 | 7454 | |
87cce662 AH |
7455 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7456 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7457 | struct sched_domain *parent, int i) | |
7458 | { | |
7459 | struct sched_domain *sd; | |
7460 | sd = &per_cpu(phys_domains, i).sd; | |
7461 | SD_INIT(sd, CPU); | |
7462 | set_domain_attribute(sd, attr); | |
7463 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7464 | sd->parent = parent; | |
7465 | if (parent) | |
7466 | parent->child = sd; | |
7467 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7468 | return sd; | |
7469 | } | |
1da177e4 | 7470 | |
01a08546 HC |
7471 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7472 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7473 | struct sched_domain *parent, int i) | |
7474 | { | |
7475 | struct sched_domain *sd = parent; | |
7476 | #ifdef CONFIG_SCHED_BOOK | |
7477 | sd = &per_cpu(book_domains, i).sd; | |
7478 | SD_INIT(sd, BOOK); | |
7479 | set_domain_attribute(sd, attr); | |
7480 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7481 | sd->parent = parent; | |
7482 | parent->child = sd; | |
7483 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7484 | #endif | |
7485 | return sd; | |
7486 | } | |
7487 | ||
410c4081 AH |
7488 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7489 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7490 | struct sched_domain *parent, int i) | |
7491 | { | |
7492 | struct sched_domain *sd = parent; | |
1e9f28fa | 7493 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7494 | sd = &per_cpu(core_domains, i).sd; |
7495 | SD_INIT(sd, MC); | |
7496 | set_domain_attribute(sd, attr); | |
7497 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7498 | sd->parent = parent; | |
7499 | parent->child = sd; | |
7500 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7501 | #endif |
410c4081 AH |
7502 | return sd; |
7503 | } | |
1e9f28fa | 7504 | |
d8173535 AH |
7505 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7506 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7507 | struct sched_domain *parent, int i) | |
7508 | { | |
7509 | struct sched_domain *sd = parent; | |
1da177e4 | 7510 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7511 | sd = &per_cpu(cpu_domains, i).sd; |
7512 | SD_INIT(sd, SIBLING); | |
7513 | set_domain_attribute(sd, attr); | |
7514 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7515 | sd->parent = parent; | |
7516 | parent->child = sd; | |
7517 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7518 | #endif |
d8173535 AH |
7519 | return sd; |
7520 | } | |
1da177e4 | 7521 | |
0e8e85c9 AH |
7522 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7523 | const struct cpumask *cpu_map, int cpu) | |
7524 | { | |
7525 | switch (l) { | |
1da177e4 | 7526 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7527 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7528 | cpumask_and(d->this_sibling_map, cpu_map, | |
7529 | topology_thread_cpumask(cpu)); | |
7530 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7531 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7532 | &cpu_to_cpu_group, | |
7533 | d->send_covered, d->tmpmask); | |
7534 | break; | |
1da177e4 | 7535 | #endif |
1e9f28fa | 7536 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7537 | case SD_LV_MC: /* set up multi-core groups */ |
7538 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7539 | if (cpu == cpumask_first(d->this_core_map)) | |
7540 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7541 | &cpu_to_core_group, | |
7542 | d->send_covered, d->tmpmask); | |
7543 | break; | |
01a08546 HC |
7544 | #endif |
7545 | #ifdef CONFIG_SCHED_BOOK | |
7546 | case SD_LV_BOOK: /* set up book groups */ | |
7547 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7548 | if (cpu == cpumask_first(d->this_book_map)) | |
7549 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7550 | &cpu_to_book_group, | |
7551 | d->send_covered, d->tmpmask); | |
7552 | break; | |
1e9f28fa | 7553 | #endif |
86548096 AH |
7554 | case SD_LV_CPU: /* set up physical groups */ |
7555 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7556 | if (!cpumask_empty(d->nodemask)) | |
7557 | init_sched_build_groups(d->nodemask, cpu_map, | |
7558 | &cpu_to_phys_group, | |
7559 | d->send_covered, d->tmpmask); | |
7560 | break; | |
1da177e4 | 7561 | #ifdef CONFIG_NUMA |
de616e36 AH |
7562 | case SD_LV_ALLNODES: |
7563 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7564 | d->send_covered, d->tmpmask); | |
7565 | break; | |
7566 | #endif | |
0e8e85c9 AH |
7567 | default: |
7568 | break; | |
7c16ec58 | 7569 | } |
0e8e85c9 | 7570 | } |
9c1cfda2 | 7571 | |
2109b99e AH |
7572 | /* |
7573 | * Build sched domains for a given set of cpus and attach the sched domains | |
7574 | * to the individual cpus | |
7575 | */ | |
7576 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7577 | struct sched_domain_attr *attr) | |
7578 | { | |
7579 | enum s_alloc alloc_state = sa_none; | |
7580 | struct s_data d; | |
294b0c96 | 7581 | struct sched_domain *sd; |
2109b99e | 7582 | int i; |
7c16ec58 | 7583 | #ifdef CONFIG_NUMA |
2109b99e | 7584 | d.sd_allnodes = 0; |
7c16ec58 | 7585 | #endif |
9c1cfda2 | 7586 | |
2109b99e AH |
7587 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7588 | if (alloc_state != sa_rootdomain) | |
7589 | goto error; | |
7590 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7591 | |
1da177e4 | 7592 | /* |
1a20ff27 | 7593 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7594 | */ |
abcd083a | 7595 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7596 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7597 | cpu_map); | |
9761eea8 | 7598 | |
7f4588f3 | 7599 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7600 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7601 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7602 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7603 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7604 | } |
9c1cfda2 | 7605 | |
abcd083a | 7606 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7607 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7608 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7609 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7610 | } |
9c1cfda2 | 7611 | |
1da177e4 | 7612 | /* Set up physical groups */ |
86548096 AH |
7613 | for (i = 0; i < nr_node_ids; i++) |
7614 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7615 | |
1da177e4 LT |
7616 | #ifdef CONFIG_NUMA |
7617 | /* Set up node groups */ | |
de616e36 AH |
7618 | if (d.sd_allnodes) |
7619 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7620 | |
0601a88d AH |
7621 | for (i = 0; i < nr_node_ids; i++) |
7622 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7623 | goto error; |
1da177e4 LT |
7624 | #endif |
7625 | ||
7626 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7627 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7628 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7629 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7630 | init_sched_groups_power(i, sd); |
5c45bf27 | 7631 | } |
1da177e4 | 7632 | #endif |
1e9f28fa | 7633 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7634 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7635 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7636 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7637 | } |
7638 | #endif | |
01a08546 HC |
7639 | #ifdef CONFIG_SCHED_BOOK |
7640 | for_each_cpu(i, cpu_map) { | |
7641 | sd = &per_cpu(book_domains, i).sd; | |
7642 | init_sched_groups_power(i, sd); | |
7643 | } | |
7644 | #endif | |
1e9f28fa | 7645 | |
abcd083a | 7646 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7647 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7648 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7649 | } |
7650 | ||
9c1cfda2 | 7651 | #ifdef CONFIG_NUMA |
076ac2af | 7652 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7653 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7654 | |
49a02c51 | 7655 | if (d.sd_allnodes) { |
6711cab4 | 7656 | struct sched_group *sg; |
f712c0c7 | 7657 | |
96f874e2 | 7658 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7659 | d.tmpmask); |
f712c0c7 SS |
7660 | init_numa_sched_groups_power(sg); |
7661 | } | |
9c1cfda2 JH |
7662 | #endif |
7663 | ||
1da177e4 | 7664 | /* Attach the domains */ |
abcd083a | 7665 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7666 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7667 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7668 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7669 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7670 | #elif defined(CONFIG_SCHED_BOOK) |
7671 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7672 | #else |
6c99e9ad | 7673 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7674 | #endif |
49a02c51 | 7675 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7676 | } |
51888ca2 | 7677 | |
2109b99e AH |
7678 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7679 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7680 | return 0; | |
51888ca2 | 7681 | |
51888ca2 | 7682 | error: |
2109b99e AH |
7683 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7684 | return -ENOMEM; | |
1da177e4 | 7685 | } |
029190c5 | 7686 | |
96f874e2 | 7687 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7688 | { |
7689 | return __build_sched_domains(cpu_map, NULL); | |
7690 | } | |
7691 | ||
acc3f5d7 | 7692 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7693 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7694 | static struct sched_domain_attr *dattr_cur; |
7695 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7696 | |
7697 | /* | |
7698 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7699 | * cpumask) fails, then fallback to a single sched domain, |
7700 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7701 | */ |
4212823f | 7702 | static cpumask_var_t fallback_doms; |
029190c5 | 7703 | |
ee79d1bd HC |
7704 | /* |
7705 | * arch_update_cpu_topology lets virtualized architectures update the | |
7706 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7707 | * or 0 if it stayed the same. | |
7708 | */ | |
7709 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7710 | { |
ee79d1bd | 7711 | return 0; |
22e52b07 HC |
7712 | } |
7713 | ||
acc3f5d7 RR |
7714 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7715 | { | |
7716 | int i; | |
7717 | cpumask_var_t *doms; | |
7718 | ||
7719 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7720 | if (!doms) | |
7721 | return NULL; | |
7722 | for (i = 0; i < ndoms; i++) { | |
7723 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7724 | free_sched_domains(doms, i); | |
7725 | return NULL; | |
7726 | } | |
7727 | } | |
7728 | return doms; | |
7729 | } | |
7730 | ||
7731 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7732 | { | |
7733 | unsigned int i; | |
7734 | for (i = 0; i < ndoms; i++) | |
7735 | free_cpumask_var(doms[i]); | |
7736 | kfree(doms); | |
7737 | } | |
7738 | ||
1a20ff27 | 7739 | /* |
41a2d6cf | 7740 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7741 | * For now this just excludes isolated cpus, but could be used to |
7742 | * exclude other special cases in the future. | |
1a20ff27 | 7743 | */ |
96f874e2 | 7744 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7745 | { |
7378547f MM |
7746 | int err; |
7747 | ||
22e52b07 | 7748 | arch_update_cpu_topology(); |
029190c5 | 7749 | ndoms_cur = 1; |
acc3f5d7 | 7750 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7751 | if (!doms_cur) |
acc3f5d7 RR |
7752 | doms_cur = &fallback_doms; |
7753 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7754 | dattr_cur = NULL; |
acc3f5d7 | 7755 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7756 | register_sched_domain_sysctl(); |
7378547f MM |
7757 | |
7758 | return err; | |
1a20ff27 DG |
7759 | } |
7760 | ||
96f874e2 RR |
7761 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7762 | struct cpumask *tmpmask) | |
1da177e4 | 7763 | { |
7c16ec58 | 7764 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7765 | } |
1da177e4 | 7766 | |
1a20ff27 DG |
7767 | /* |
7768 | * Detach sched domains from a group of cpus specified in cpu_map | |
7769 | * These cpus will now be attached to the NULL domain | |
7770 | */ | |
96f874e2 | 7771 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7772 | { |
96f874e2 RR |
7773 | /* Save because hotplug lock held. */ |
7774 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7775 | int i; |
7776 | ||
abcd083a | 7777 | for_each_cpu(i, cpu_map) |
57d885fe | 7778 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7779 | synchronize_sched(); |
96f874e2 | 7780 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7781 | } |
7782 | ||
1d3504fc HS |
7783 | /* handle null as "default" */ |
7784 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7785 | struct sched_domain_attr *new, int idx_new) | |
7786 | { | |
7787 | struct sched_domain_attr tmp; | |
7788 | ||
7789 | /* fast path */ | |
7790 | if (!new && !cur) | |
7791 | return 1; | |
7792 | ||
7793 | tmp = SD_ATTR_INIT; | |
7794 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7795 | new ? (new + idx_new) : &tmp, | |
7796 | sizeof(struct sched_domain_attr)); | |
7797 | } | |
7798 | ||
029190c5 PJ |
7799 | /* |
7800 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7801 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7802 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7803 | * It destroys each deleted domain and builds each new domain. | |
7804 | * | |
acc3f5d7 | 7805 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7806 | * The masks don't intersect (don't overlap.) We should setup one |
7807 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7808 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7809 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7810 | * it as it is. | |
7811 | * | |
acc3f5d7 RR |
7812 | * The passed in 'doms_new' should be allocated using |
7813 | * alloc_sched_domains. This routine takes ownership of it and will | |
7814 | * free_sched_domains it when done with it. If the caller failed the | |
7815 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7816 | * and partition_sched_domains() will fallback to the single partition | |
7817 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7818 | * |
96f874e2 | 7819 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7820 | * ndoms_new == 0 is a special case for destroying existing domains, |
7821 | * and it will not create the default domain. | |
dfb512ec | 7822 | * |
029190c5 PJ |
7823 | * Call with hotplug lock held |
7824 | */ | |
acc3f5d7 | 7825 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7826 | struct sched_domain_attr *dattr_new) |
029190c5 | 7827 | { |
dfb512ec | 7828 | int i, j, n; |
d65bd5ec | 7829 | int new_topology; |
029190c5 | 7830 | |
712555ee | 7831 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7832 | |
7378547f MM |
7833 | /* always unregister in case we don't destroy any domains */ |
7834 | unregister_sched_domain_sysctl(); | |
7835 | ||
d65bd5ec HC |
7836 | /* Let architecture update cpu core mappings. */ |
7837 | new_topology = arch_update_cpu_topology(); | |
7838 | ||
dfb512ec | 7839 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7840 | |
7841 | /* Destroy deleted domains */ | |
7842 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7843 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7844 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7845 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7846 | goto match1; |
7847 | } | |
7848 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7849 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7850 | match1: |
7851 | ; | |
7852 | } | |
7853 | ||
e761b772 MK |
7854 | if (doms_new == NULL) { |
7855 | ndoms_cur = 0; | |
acc3f5d7 | 7856 | doms_new = &fallback_doms; |
6ad4c188 | 7857 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7858 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7859 | } |
7860 | ||
029190c5 PJ |
7861 | /* Build new domains */ |
7862 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7863 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7864 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7865 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7866 | goto match2; |
7867 | } | |
7868 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7869 | __build_sched_domains(doms_new[i], |
1d3504fc | 7870 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7871 | match2: |
7872 | ; | |
7873 | } | |
7874 | ||
7875 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7876 | if (doms_cur != &fallback_doms) |
7877 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7878 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7879 | doms_cur = doms_new; |
1d3504fc | 7880 | dattr_cur = dattr_new; |
029190c5 | 7881 | ndoms_cur = ndoms_new; |
7378547f MM |
7882 | |
7883 | register_sched_domain_sysctl(); | |
a1835615 | 7884 | |
712555ee | 7885 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7886 | } |
7887 | ||
5c45bf27 | 7888 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7889 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7890 | { |
95402b38 | 7891 | get_online_cpus(); |
dfb512ec MK |
7892 | |
7893 | /* Destroy domains first to force the rebuild */ | |
7894 | partition_sched_domains(0, NULL, NULL); | |
7895 | ||
e761b772 | 7896 | rebuild_sched_domains(); |
95402b38 | 7897 | put_online_cpus(); |
5c45bf27 SS |
7898 | } |
7899 | ||
7900 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7901 | { | |
afb8a9b7 | 7902 | unsigned int level = 0; |
5c45bf27 | 7903 | |
afb8a9b7 GS |
7904 | if (sscanf(buf, "%u", &level) != 1) |
7905 | return -EINVAL; | |
7906 | ||
7907 | /* | |
7908 | * level is always be positive so don't check for | |
7909 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7910 | * What happens on 0 or 1 byte write, | |
7911 | * need to check for count as well? | |
7912 | */ | |
7913 | ||
7914 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7915 | return -EINVAL; |
7916 | ||
7917 | if (smt) | |
afb8a9b7 | 7918 | sched_smt_power_savings = level; |
5c45bf27 | 7919 | else |
afb8a9b7 | 7920 | sched_mc_power_savings = level; |
5c45bf27 | 7921 | |
c70f22d2 | 7922 | arch_reinit_sched_domains(); |
5c45bf27 | 7923 | |
c70f22d2 | 7924 | return count; |
5c45bf27 SS |
7925 | } |
7926 | ||
5c45bf27 | 7927 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7928 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7929 | struct sysdev_class_attribute *attr, |
f718cd4a | 7930 | char *page) |
5c45bf27 SS |
7931 | { |
7932 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7933 | } | |
f718cd4a | 7934 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7935 | struct sysdev_class_attribute *attr, |
48f24c4d | 7936 | const char *buf, size_t count) |
5c45bf27 SS |
7937 | { |
7938 | return sched_power_savings_store(buf, count, 0); | |
7939 | } | |
f718cd4a AK |
7940 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7941 | sched_mc_power_savings_show, | |
7942 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7943 | #endif |
7944 | ||
7945 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7946 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7947 | struct sysdev_class_attribute *attr, |
f718cd4a | 7948 | char *page) |
5c45bf27 SS |
7949 | { |
7950 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7951 | } | |
f718cd4a | 7952 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7953 | struct sysdev_class_attribute *attr, |
48f24c4d | 7954 | const char *buf, size_t count) |
5c45bf27 SS |
7955 | { |
7956 | return sched_power_savings_store(buf, count, 1); | |
7957 | } | |
f718cd4a AK |
7958 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7959 | sched_smt_power_savings_show, | |
6707de00 AB |
7960 | sched_smt_power_savings_store); |
7961 | #endif | |
7962 | ||
39aac648 | 7963 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7964 | { |
7965 | int err = 0; | |
7966 | ||
7967 | #ifdef CONFIG_SCHED_SMT | |
7968 | if (smt_capable()) | |
7969 | err = sysfs_create_file(&cls->kset.kobj, | |
7970 | &attr_sched_smt_power_savings.attr); | |
7971 | #endif | |
7972 | #ifdef CONFIG_SCHED_MC | |
7973 | if (!err && mc_capable()) | |
7974 | err = sysfs_create_file(&cls->kset.kobj, | |
7975 | &attr_sched_mc_power_savings.attr); | |
7976 | #endif | |
7977 | return err; | |
7978 | } | |
6d6bc0ad | 7979 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7980 | |
1da177e4 | 7981 | /* |
3a101d05 TH |
7982 | * Update cpusets according to cpu_active mask. If cpusets are |
7983 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7984 | * around partition_sched_domains(). | |
1da177e4 | 7985 | */ |
0b2e918a TH |
7986 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7987 | void *hcpu) | |
e761b772 | 7988 | { |
3a101d05 | 7989 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7990 | case CPU_ONLINE: |
6ad4c188 | 7991 | case CPU_DOWN_FAILED: |
3a101d05 | 7992 | cpuset_update_active_cpus(); |
e761b772 | 7993 | return NOTIFY_OK; |
3a101d05 TH |
7994 | default: |
7995 | return NOTIFY_DONE; | |
7996 | } | |
7997 | } | |
e761b772 | 7998 | |
0b2e918a TH |
7999 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
8000 | void *hcpu) | |
3a101d05 TH |
8001 | { |
8002 | switch (action & ~CPU_TASKS_FROZEN) { | |
8003 | case CPU_DOWN_PREPARE: | |
8004 | cpuset_update_active_cpus(); | |
8005 | return NOTIFY_OK; | |
e761b772 MK |
8006 | default: |
8007 | return NOTIFY_DONE; | |
8008 | } | |
8009 | } | |
e761b772 MK |
8010 | |
8011 | static int update_runtime(struct notifier_block *nfb, | |
8012 | unsigned long action, void *hcpu) | |
1da177e4 | 8013 | { |
7def2be1 PZ |
8014 | int cpu = (int)(long)hcpu; |
8015 | ||
1da177e4 | 8016 | switch (action) { |
1da177e4 | 8017 | case CPU_DOWN_PREPARE: |
8bb78442 | 8018 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8019 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8020 | return NOTIFY_OK; |
8021 | ||
1da177e4 | 8022 | case CPU_DOWN_FAILED: |
8bb78442 | 8023 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8024 | case CPU_ONLINE: |
8bb78442 | 8025 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8026 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8027 | return NOTIFY_OK; |
8028 | ||
1da177e4 LT |
8029 | default: |
8030 | return NOTIFY_DONE; | |
8031 | } | |
1da177e4 | 8032 | } |
1da177e4 LT |
8033 | |
8034 | void __init sched_init_smp(void) | |
8035 | { | |
dcc30a35 RR |
8036 | cpumask_var_t non_isolated_cpus; |
8037 | ||
8038 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 8039 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 8040 | |
434d53b0 MT |
8041 | #if defined(CONFIG_NUMA) |
8042 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8043 | GFP_KERNEL); | |
8044 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8045 | #endif | |
95402b38 | 8046 | get_online_cpus(); |
712555ee | 8047 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 8048 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
8049 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
8050 | if (cpumask_empty(non_isolated_cpus)) | |
8051 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8052 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8053 | put_online_cpus(); |
e761b772 | 8054 | |
3a101d05 TH |
8055 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
8056 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
8057 | |
8058 | /* RT runtime code needs to handle some hotplug events */ | |
8059 | hotcpu_notifier(update_runtime, 0); | |
8060 | ||
b328ca18 | 8061 | init_hrtick(); |
5c1e1767 NP |
8062 | |
8063 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8064 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8065 | BUG(); |
19978ca6 | 8066 | sched_init_granularity(); |
dcc30a35 | 8067 | free_cpumask_var(non_isolated_cpus); |
4212823f | 8068 | |
0e3900e6 | 8069 | init_sched_rt_class(); |
1da177e4 LT |
8070 | } |
8071 | #else | |
8072 | void __init sched_init_smp(void) | |
8073 | { | |
19978ca6 | 8074 | sched_init_granularity(); |
1da177e4 LT |
8075 | } |
8076 | #endif /* CONFIG_SMP */ | |
8077 | ||
cd1bb94b AB |
8078 | const_debug unsigned int sysctl_timer_migration = 1; |
8079 | ||
1da177e4 LT |
8080 | int in_sched_functions(unsigned long addr) |
8081 | { | |
1da177e4 LT |
8082 | return in_lock_functions(addr) || |
8083 | (addr >= (unsigned long)__sched_text_start | |
8084 | && addr < (unsigned long)__sched_text_end); | |
8085 | } | |
8086 | ||
a9957449 | 8087 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8088 | { |
8089 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8090 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8091 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8092 | cfs_rq->rq = rq; | |
f07333bf | 8093 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 8094 | #ifdef CONFIG_SMP |
f07333bf | 8095 | cfs_rq->load_stamp = 1; |
6ea72f12 | 8096 | #endif |
dd41f596 | 8097 | #endif |
67e9fb2a | 8098 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8099 | } |
8100 | ||
fa85ae24 PZ |
8101 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8102 | { | |
8103 | struct rt_prio_array *array; | |
8104 | int i; | |
8105 | ||
8106 | array = &rt_rq->active; | |
8107 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8108 | INIT_LIST_HEAD(array->queue + i); | |
8109 | __clear_bit(i, array->bitmap); | |
8110 | } | |
8111 | /* delimiter for bitsearch: */ | |
8112 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8113 | ||
052f1dc7 | 8114 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8115 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8116 | #ifdef CONFIG_SMP |
e864c499 | 8117 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8118 | #endif |
48d5e258 | 8119 | #endif |
fa85ae24 PZ |
8120 | #ifdef CONFIG_SMP |
8121 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8122 | rt_rq->overloaded = 0; |
05fa785c | 8123 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8124 | #endif |
8125 | ||
8126 | rt_rq->rt_time = 0; | |
8127 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8128 | rt_rq->rt_runtime = 0; |
0986b11b | 8129 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 8130 | |
052f1dc7 | 8131 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8132 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8133 | rt_rq->rq = rq; |
8134 | #endif | |
fa85ae24 PZ |
8135 | } |
8136 | ||
6f505b16 | 8137 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8138 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8139 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8140 | struct sched_entity *parent) |
6f505b16 | 8141 | { |
ec7dc8ac | 8142 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8143 | tg->cfs_rq[cpu] = cfs_rq; |
8144 | init_cfs_rq(cfs_rq, rq); | |
8145 | cfs_rq->tg = tg; | |
6f505b16 PZ |
8146 | |
8147 | tg->se[cpu] = se; | |
07e06b01 | 8148 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8149 | if (!se) |
8150 | return; | |
8151 | ||
ec7dc8ac DG |
8152 | if (!parent) |
8153 | se->cfs_rq = &rq->cfs; | |
8154 | else | |
8155 | se->cfs_rq = parent->my_q; | |
8156 | ||
6f505b16 | 8157 | se->my_q = cfs_rq; |
9437178f | 8158 | update_load_set(&se->load, 0); |
ec7dc8ac | 8159 | se->parent = parent; |
6f505b16 | 8160 | } |
052f1dc7 | 8161 | #endif |
6f505b16 | 8162 | |
052f1dc7 | 8163 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8164 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8165 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8166 | struct sched_rt_entity *parent) |
6f505b16 | 8167 | { |
ec7dc8ac DG |
8168 | struct rq *rq = cpu_rq(cpu); |
8169 | ||
6f505b16 PZ |
8170 | tg->rt_rq[cpu] = rt_rq; |
8171 | init_rt_rq(rt_rq, rq); | |
8172 | rt_rq->tg = tg; | |
ac086bc2 | 8173 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8174 | |
8175 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8176 | if (!rt_se) |
8177 | return; | |
8178 | ||
ec7dc8ac DG |
8179 | if (!parent) |
8180 | rt_se->rt_rq = &rq->rt; | |
8181 | else | |
8182 | rt_se->rt_rq = parent->my_q; | |
8183 | ||
6f505b16 | 8184 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8185 | rt_se->parent = parent; |
6f505b16 PZ |
8186 | INIT_LIST_HEAD(&rt_se->run_list); |
8187 | } | |
8188 | #endif | |
8189 | ||
1da177e4 LT |
8190 | void __init sched_init(void) |
8191 | { | |
dd41f596 | 8192 | int i, j; |
434d53b0 MT |
8193 | unsigned long alloc_size = 0, ptr; |
8194 | ||
8195 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8196 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8197 | #endif | |
8198 | #ifdef CONFIG_RT_GROUP_SCHED | |
8199 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8200 | #endif |
df7c8e84 | 8201 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8202 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8203 | #endif |
434d53b0 | 8204 | if (alloc_size) { |
36b7b6d4 | 8205 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8206 | |
8207 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8208 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8209 | ptr += nr_cpu_ids * sizeof(void **); |
8210 | ||
07e06b01 | 8211 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8212 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8213 | |
6d6bc0ad | 8214 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8215 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8216 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8217 | ptr += nr_cpu_ids * sizeof(void **); |
8218 | ||
07e06b01 | 8219 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8220 | ptr += nr_cpu_ids * sizeof(void **); |
8221 | ||
6d6bc0ad | 8222 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8223 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8224 | for_each_possible_cpu(i) { | |
8225 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8226 | ptr += cpumask_size(); | |
8227 | } | |
8228 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8229 | } |
dd41f596 | 8230 | |
57d885fe GH |
8231 | #ifdef CONFIG_SMP |
8232 | init_defrootdomain(); | |
8233 | #endif | |
8234 | ||
d0b27fa7 PZ |
8235 | init_rt_bandwidth(&def_rt_bandwidth, |
8236 | global_rt_period(), global_rt_runtime()); | |
8237 | ||
8238 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8239 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8240 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8241 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8242 | |
7c941438 | 8243 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8244 | list_add(&root_task_group.list, &task_groups); |
8245 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8246 | autogroup_init(&init_task); |
7c941438 | 8247 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8248 | |
0a945022 | 8249 | for_each_possible_cpu(i) { |
70b97a7f | 8250 | struct rq *rq; |
1da177e4 LT |
8251 | |
8252 | rq = cpu_rq(i); | |
05fa785c | 8253 | raw_spin_lock_init(&rq->lock); |
7897986b | 8254 | rq->nr_running = 0; |
dce48a84 TG |
8255 | rq->calc_load_active = 0; |
8256 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8257 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8258 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8259 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8260 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8261 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8262 | /* |
07e06b01 | 8263 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8264 | * |
8265 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8266 | * gets 100% of the cpu resources in the system. This overall | |
8267 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8268 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8269 | * based on each entity's (task or task-group's) weight |
8270 | * (se->load.weight). | |
8271 | * | |
07e06b01 | 8272 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8273 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8274 | * then A0's share of the cpu resource is: | |
8275 | * | |
0d905bca | 8276 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8277 | * |
07e06b01 YZ |
8278 | * We achieve this by letting root_task_group's tasks sit |
8279 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8280 | */ |
07e06b01 | 8281 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8282 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8283 | ||
8284 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8285 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8286 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8287 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8288 | #endif |
1da177e4 | 8289 | |
dd41f596 IM |
8290 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8291 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8292 | |
8293 | rq->last_load_update_tick = jiffies; | |
8294 | ||
1da177e4 | 8295 | #ifdef CONFIG_SMP |
41c7ce9a | 8296 | rq->sd = NULL; |
57d885fe | 8297 | rq->rd = NULL; |
e51fd5e2 | 8298 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8299 | rq->post_schedule = 0; |
1da177e4 | 8300 | rq->active_balance = 0; |
dd41f596 | 8301 | rq->next_balance = jiffies; |
1da177e4 | 8302 | rq->push_cpu = 0; |
0a2966b4 | 8303 | rq->cpu = i; |
1f11eb6a | 8304 | rq->online = 0; |
eae0c9df MG |
8305 | rq->idle_stamp = 0; |
8306 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8307 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8308 | #ifdef CONFIG_NO_HZ |
8309 | rq->nohz_balance_kick = 0; | |
8310 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8311 | #endif | |
1da177e4 | 8312 | #endif |
8f4d37ec | 8313 | init_rq_hrtick(rq); |
1da177e4 | 8314 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8315 | } |
8316 | ||
2dd73a4f | 8317 | set_load_weight(&init_task); |
b50f60ce | 8318 | |
e107be36 AK |
8319 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8320 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8321 | #endif | |
8322 | ||
c9819f45 | 8323 | #ifdef CONFIG_SMP |
962cf36c | 8324 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8325 | #endif |
8326 | ||
b50f60ce | 8327 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8328 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8329 | #endif |
8330 | ||
1da177e4 LT |
8331 | /* |
8332 | * The boot idle thread does lazy MMU switching as well: | |
8333 | */ | |
8334 | atomic_inc(&init_mm.mm_count); | |
8335 | enter_lazy_tlb(&init_mm, current); | |
8336 | ||
8337 | /* | |
8338 | * Make us the idle thread. Technically, schedule() should not be | |
8339 | * called from this thread, however somewhere below it might be, | |
8340 | * but because we are the idle thread, we just pick up running again | |
8341 | * when this runqueue becomes "idle". | |
8342 | */ | |
8343 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8344 | |
8345 | calc_load_update = jiffies + LOAD_FREQ; | |
8346 | ||
dd41f596 IM |
8347 | /* |
8348 | * During early bootup we pretend to be a normal task: | |
8349 | */ | |
8350 | current->sched_class = &fair_sched_class; | |
6892b75e | 8351 | |
6a7b3dc3 | 8352 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8353 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8354 | #ifdef CONFIG_SMP |
7d1e6a9b | 8355 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8356 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8357 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8358 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8359 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8360 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8361 | #endif |
bdddd296 RR |
8362 | /* May be allocated at isolcpus cmdline parse time */ |
8363 | if (cpu_isolated_map == NULL) | |
8364 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8365 | #endif /* SMP */ |
6a7b3dc3 | 8366 | |
6892b75e | 8367 | scheduler_running = 1; |
1da177e4 LT |
8368 | } |
8369 | ||
8370 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8371 | static inline int preempt_count_equals(int preempt_offset) |
8372 | { | |
234da7bc | 8373 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8374 | |
4ba8216c | 8375 | return (nested == preempt_offset); |
e4aafea2 FW |
8376 | } |
8377 | ||
d894837f | 8378 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8379 | { |
48f24c4d | 8380 | #ifdef in_atomic |
1da177e4 LT |
8381 | static unsigned long prev_jiffy; /* ratelimiting */ |
8382 | ||
e4aafea2 FW |
8383 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8384 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8385 | return; |
8386 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8387 | return; | |
8388 | prev_jiffy = jiffies; | |
8389 | ||
3df0fc5b PZ |
8390 | printk(KERN_ERR |
8391 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8392 | file, line); | |
8393 | printk(KERN_ERR | |
8394 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8395 | in_atomic(), irqs_disabled(), | |
8396 | current->pid, current->comm); | |
aef745fc IM |
8397 | |
8398 | debug_show_held_locks(current); | |
8399 | if (irqs_disabled()) | |
8400 | print_irqtrace_events(current); | |
8401 | dump_stack(); | |
1da177e4 LT |
8402 | #endif |
8403 | } | |
8404 | EXPORT_SYMBOL(__might_sleep); | |
8405 | #endif | |
8406 | ||
8407 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8408 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8409 | { | |
da7a735e PZ |
8410 | const struct sched_class *prev_class = p->sched_class; |
8411 | int old_prio = p->prio; | |
3a5e4dc1 | 8412 | int on_rq; |
3e51f33f | 8413 | |
fd2f4419 | 8414 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8415 | if (on_rq) |
8416 | deactivate_task(rq, p, 0); | |
8417 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8418 | if (on_rq) { | |
8419 | activate_task(rq, p, 0); | |
8420 | resched_task(rq->curr); | |
8421 | } | |
da7a735e PZ |
8422 | |
8423 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8424 | } |
8425 | ||
1da177e4 LT |
8426 | void normalize_rt_tasks(void) |
8427 | { | |
a0f98a1c | 8428 | struct task_struct *g, *p; |
1da177e4 | 8429 | unsigned long flags; |
70b97a7f | 8430 | struct rq *rq; |
1da177e4 | 8431 | |
4cf5d77a | 8432 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8433 | do_each_thread(g, p) { |
178be793 IM |
8434 | /* |
8435 | * Only normalize user tasks: | |
8436 | */ | |
8437 | if (!p->mm) | |
8438 | continue; | |
8439 | ||
6cfb0d5d | 8440 | p->se.exec_start = 0; |
6cfb0d5d | 8441 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8442 | p->se.statistics.wait_start = 0; |
8443 | p->se.statistics.sleep_start = 0; | |
8444 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8445 | #endif |
dd41f596 IM |
8446 | |
8447 | if (!rt_task(p)) { | |
8448 | /* | |
8449 | * Renice negative nice level userspace | |
8450 | * tasks back to 0: | |
8451 | */ | |
8452 | if (TASK_NICE(p) < 0 && p->mm) | |
8453 | set_user_nice(p, 0); | |
1da177e4 | 8454 | continue; |
dd41f596 | 8455 | } |
1da177e4 | 8456 | |
1d615482 | 8457 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8458 | rq = __task_rq_lock(p); |
1da177e4 | 8459 | |
178be793 | 8460 | normalize_task(rq, p); |
3a5e4dc1 | 8461 | |
b29739f9 | 8462 | __task_rq_unlock(rq); |
1d615482 | 8463 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8464 | } while_each_thread(g, p); |
8465 | ||
4cf5d77a | 8466 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8467 | } |
8468 | ||
8469 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8470 | |
67fc4e0c | 8471 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8472 | /* |
67fc4e0c | 8473 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8474 | * |
8475 | * They can only be called when the whole system has been | |
8476 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8477 | * activity can take place. Using them for anything else would | |
8478 | * be a serious bug, and as a result, they aren't even visible | |
8479 | * under any other configuration. | |
8480 | */ | |
8481 | ||
8482 | /** | |
8483 | * curr_task - return the current task for a given cpu. | |
8484 | * @cpu: the processor in question. | |
8485 | * | |
8486 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8487 | */ | |
36c8b586 | 8488 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8489 | { |
8490 | return cpu_curr(cpu); | |
8491 | } | |
8492 | ||
67fc4e0c JW |
8493 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8494 | ||
8495 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8496 | /** |
8497 | * set_curr_task - set the current task for a given cpu. | |
8498 | * @cpu: the processor in question. | |
8499 | * @p: the task pointer to set. | |
8500 | * | |
8501 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8502 | * are serviced on a separate stack. It allows the architecture to switch the |
8503 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8504 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8505 | * and caller must save the original value of the current task (see | |
8506 | * curr_task() above) and restore that value before reenabling interrupts and | |
8507 | * re-starting the system. | |
8508 | * | |
8509 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8510 | */ | |
36c8b586 | 8511 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8512 | { |
8513 | cpu_curr(cpu) = p; | |
8514 | } | |
8515 | ||
8516 | #endif | |
29f59db3 | 8517 | |
bccbe08a PZ |
8518 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8519 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8520 | { |
8521 | int i; | |
8522 | ||
8523 | for_each_possible_cpu(i) { | |
8524 | if (tg->cfs_rq) | |
8525 | kfree(tg->cfs_rq[i]); | |
8526 | if (tg->se) | |
8527 | kfree(tg->se[i]); | |
6f505b16 PZ |
8528 | } |
8529 | ||
8530 | kfree(tg->cfs_rq); | |
8531 | kfree(tg->se); | |
6f505b16 PZ |
8532 | } |
8533 | ||
ec7dc8ac DG |
8534 | static |
8535 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8536 | { |
29f59db3 | 8537 | struct cfs_rq *cfs_rq; |
eab17229 | 8538 | struct sched_entity *se; |
29f59db3 SV |
8539 | int i; |
8540 | ||
434d53b0 | 8541 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8542 | if (!tg->cfs_rq) |
8543 | goto err; | |
434d53b0 | 8544 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8545 | if (!tg->se) |
8546 | goto err; | |
052f1dc7 PZ |
8547 | |
8548 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8549 | |
8550 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8551 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8552 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8553 | if (!cfs_rq) |
8554 | goto err; | |
8555 | ||
eab17229 LZ |
8556 | se = kzalloc_node(sizeof(struct sched_entity), |
8557 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8558 | if (!se) |
dfc12eb2 | 8559 | goto err_free_rq; |
29f59db3 | 8560 | |
3d4b47b4 | 8561 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8562 | } |
8563 | ||
8564 | return 1; | |
8565 | ||
49246274 | 8566 | err_free_rq: |
dfc12eb2 | 8567 | kfree(cfs_rq); |
49246274 | 8568 | err: |
bccbe08a PZ |
8569 | return 0; |
8570 | } | |
8571 | ||
bccbe08a PZ |
8572 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8573 | { | |
3d4b47b4 PZ |
8574 | struct rq *rq = cpu_rq(cpu); |
8575 | unsigned long flags; | |
3d4b47b4 PZ |
8576 | |
8577 | /* | |
8578 | * Only empty task groups can be destroyed; so we can speculatively | |
8579 | * check on_list without danger of it being re-added. | |
8580 | */ | |
8581 | if (!tg->cfs_rq[cpu]->on_list) | |
8582 | return; | |
8583 | ||
8584 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8585 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8586 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8587 | } |
6d6bc0ad | 8588 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8589 | static inline void free_fair_sched_group(struct task_group *tg) |
8590 | { | |
8591 | } | |
8592 | ||
ec7dc8ac DG |
8593 | static inline |
8594 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8595 | { |
8596 | return 1; | |
8597 | } | |
8598 | ||
bccbe08a PZ |
8599 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8600 | { | |
8601 | } | |
6d6bc0ad | 8602 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8603 | |
8604 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8605 | static void free_rt_sched_group(struct task_group *tg) |
8606 | { | |
8607 | int i; | |
8608 | ||
d0b27fa7 PZ |
8609 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8610 | ||
bccbe08a PZ |
8611 | for_each_possible_cpu(i) { |
8612 | if (tg->rt_rq) | |
8613 | kfree(tg->rt_rq[i]); | |
8614 | if (tg->rt_se) | |
8615 | kfree(tg->rt_se[i]); | |
8616 | } | |
8617 | ||
8618 | kfree(tg->rt_rq); | |
8619 | kfree(tg->rt_se); | |
8620 | } | |
8621 | ||
ec7dc8ac DG |
8622 | static |
8623 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8624 | { |
8625 | struct rt_rq *rt_rq; | |
eab17229 | 8626 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8627 | struct rq *rq; |
8628 | int i; | |
8629 | ||
434d53b0 | 8630 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8631 | if (!tg->rt_rq) |
8632 | goto err; | |
434d53b0 | 8633 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8634 | if (!tg->rt_se) |
8635 | goto err; | |
8636 | ||
d0b27fa7 PZ |
8637 | init_rt_bandwidth(&tg->rt_bandwidth, |
8638 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8639 | |
8640 | for_each_possible_cpu(i) { | |
8641 | rq = cpu_rq(i); | |
8642 | ||
eab17229 LZ |
8643 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8644 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8645 | if (!rt_rq) |
8646 | goto err; | |
29f59db3 | 8647 | |
eab17229 LZ |
8648 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8649 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8650 | if (!rt_se) |
dfc12eb2 | 8651 | goto err_free_rq; |
29f59db3 | 8652 | |
3d4b47b4 | 8653 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8654 | } |
8655 | ||
bccbe08a PZ |
8656 | return 1; |
8657 | ||
49246274 | 8658 | err_free_rq: |
dfc12eb2 | 8659 | kfree(rt_rq); |
49246274 | 8660 | err: |
bccbe08a PZ |
8661 | return 0; |
8662 | } | |
6d6bc0ad | 8663 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8664 | static inline void free_rt_sched_group(struct task_group *tg) |
8665 | { | |
8666 | } | |
8667 | ||
ec7dc8ac DG |
8668 | static inline |
8669 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8670 | { |
8671 | return 1; | |
8672 | } | |
6d6bc0ad | 8673 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8674 | |
7c941438 | 8675 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8676 | static void free_sched_group(struct task_group *tg) |
8677 | { | |
8678 | free_fair_sched_group(tg); | |
8679 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8680 | autogroup_free(tg); |
bccbe08a PZ |
8681 | kfree(tg); |
8682 | } | |
8683 | ||
8684 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8685 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8686 | { |
8687 | struct task_group *tg; | |
8688 | unsigned long flags; | |
bccbe08a PZ |
8689 | |
8690 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8691 | if (!tg) | |
8692 | return ERR_PTR(-ENOMEM); | |
8693 | ||
ec7dc8ac | 8694 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8695 | goto err; |
8696 | ||
ec7dc8ac | 8697 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8698 | goto err; |
8699 | ||
8ed36996 | 8700 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8701 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8702 | |
8703 | WARN_ON(!parent); /* root should already exist */ | |
8704 | ||
8705 | tg->parent = parent; | |
f473aa5e | 8706 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8707 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8708 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8709 | |
9b5b7751 | 8710 | return tg; |
29f59db3 SV |
8711 | |
8712 | err: | |
6f505b16 | 8713 | free_sched_group(tg); |
29f59db3 SV |
8714 | return ERR_PTR(-ENOMEM); |
8715 | } | |
8716 | ||
9b5b7751 | 8717 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8718 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8719 | { |
29f59db3 | 8720 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8721 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8722 | } |
8723 | ||
9b5b7751 | 8724 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8725 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8726 | { |
8ed36996 | 8727 | unsigned long flags; |
9b5b7751 | 8728 | int i; |
29f59db3 | 8729 | |
3d4b47b4 PZ |
8730 | /* end participation in shares distribution */ |
8731 | for_each_possible_cpu(i) | |
bccbe08a | 8732 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8733 | |
8734 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8735 | list_del_rcu(&tg->list); |
f473aa5e | 8736 | list_del_rcu(&tg->siblings); |
8ed36996 | 8737 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8738 | |
9b5b7751 | 8739 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8740 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8741 | } |
8742 | ||
9b5b7751 | 8743 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8744 | * The caller of this function should have put the task in its new group |
8745 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8746 | * reflect its new group. | |
9b5b7751 SV |
8747 | */ |
8748 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8749 | { |
8750 | int on_rq, running; | |
8751 | unsigned long flags; | |
8752 | struct rq *rq; | |
8753 | ||
8754 | rq = task_rq_lock(tsk, &flags); | |
8755 | ||
051a1d1a | 8756 | running = task_current(rq, tsk); |
fd2f4419 | 8757 | on_rq = tsk->on_rq; |
29f59db3 | 8758 | |
0e1f3483 | 8759 | if (on_rq) |
29f59db3 | 8760 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8761 | if (unlikely(running)) |
8762 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8763 | |
810b3817 | 8764 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8765 | if (tsk->sched_class->task_move_group) |
8766 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8767 | else | |
810b3817 | 8768 | #endif |
b2b5ce02 | 8769 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8770 | |
0e1f3483 HS |
8771 | if (unlikely(running)) |
8772 | tsk->sched_class->set_curr_task(rq); | |
8773 | if (on_rq) | |
371fd7e7 | 8774 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8775 | |
0122ec5b | 8776 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 8777 | } |
7c941438 | 8778 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8779 | |
052f1dc7 | 8780 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8781 | static DEFINE_MUTEX(shares_mutex); |
8782 | ||
4cf86d77 | 8783 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8784 | { |
8785 | int i; | |
8ed36996 | 8786 | unsigned long flags; |
c61935fd | 8787 | |
ec7dc8ac DG |
8788 | /* |
8789 | * We can't change the weight of the root cgroup. | |
8790 | */ | |
8791 | if (!tg->se[0]) | |
8792 | return -EINVAL; | |
8793 | ||
18d95a28 PZ |
8794 | if (shares < MIN_SHARES) |
8795 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8796 | else if (shares > MAX_SHARES) |
8797 | shares = MAX_SHARES; | |
62fb1851 | 8798 | |
8ed36996 | 8799 | mutex_lock(&shares_mutex); |
9b5b7751 | 8800 | if (tg->shares == shares) |
5cb350ba | 8801 | goto done; |
29f59db3 | 8802 | |
9b5b7751 | 8803 | tg->shares = shares; |
c09595f6 | 8804 | for_each_possible_cpu(i) { |
9437178f PT |
8805 | struct rq *rq = cpu_rq(i); |
8806 | struct sched_entity *se; | |
8807 | ||
8808 | se = tg->se[i]; | |
8809 | /* Propagate contribution to hierarchy */ | |
8810 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8811 | for_each_sched_entity(se) | |
6d5ab293 | 8812 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8813 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8814 | } |
29f59db3 | 8815 | |
5cb350ba | 8816 | done: |
8ed36996 | 8817 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8818 | return 0; |
29f59db3 SV |
8819 | } |
8820 | ||
5cb350ba DG |
8821 | unsigned long sched_group_shares(struct task_group *tg) |
8822 | { | |
8823 | return tg->shares; | |
8824 | } | |
052f1dc7 | 8825 | #endif |
5cb350ba | 8826 | |
052f1dc7 | 8827 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8828 | /* |
9f0c1e56 | 8829 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8830 | */ |
9f0c1e56 PZ |
8831 | static DEFINE_MUTEX(rt_constraints_mutex); |
8832 | ||
8833 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8834 | { | |
8835 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8836 | return 1ULL << 20; |
9f0c1e56 | 8837 | |
9a7e0b18 | 8838 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8839 | } |
8840 | ||
9a7e0b18 PZ |
8841 | /* Must be called with tasklist_lock held */ |
8842 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8843 | { |
9a7e0b18 | 8844 | struct task_struct *g, *p; |
b40b2e8e | 8845 | |
9a7e0b18 PZ |
8846 | do_each_thread(g, p) { |
8847 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8848 | return 1; | |
8849 | } while_each_thread(g, p); | |
b40b2e8e | 8850 | |
9a7e0b18 PZ |
8851 | return 0; |
8852 | } | |
b40b2e8e | 8853 | |
9a7e0b18 PZ |
8854 | struct rt_schedulable_data { |
8855 | struct task_group *tg; | |
8856 | u64 rt_period; | |
8857 | u64 rt_runtime; | |
8858 | }; | |
b40b2e8e | 8859 | |
9a7e0b18 PZ |
8860 | static int tg_schedulable(struct task_group *tg, void *data) |
8861 | { | |
8862 | struct rt_schedulable_data *d = data; | |
8863 | struct task_group *child; | |
8864 | unsigned long total, sum = 0; | |
8865 | u64 period, runtime; | |
b40b2e8e | 8866 | |
9a7e0b18 PZ |
8867 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8868 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8869 | |
9a7e0b18 PZ |
8870 | if (tg == d->tg) { |
8871 | period = d->rt_period; | |
8872 | runtime = d->rt_runtime; | |
b40b2e8e | 8873 | } |
b40b2e8e | 8874 | |
4653f803 PZ |
8875 | /* |
8876 | * Cannot have more runtime than the period. | |
8877 | */ | |
8878 | if (runtime > period && runtime != RUNTIME_INF) | |
8879 | return -EINVAL; | |
6f505b16 | 8880 | |
4653f803 PZ |
8881 | /* |
8882 | * Ensure we don't starve existing RT tasks. | |
8883 | */ | |
9a7e0b18 PZ |
8884 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8885 | return -EBUSY; | |
6f505b16 | 8886 | |
9a7e0b18 | 8887 | total = to_ratio(period, runtime); |
6f505b16 | 8888 | |
4653f803 PZ |
8889 | /* |
8890 | * Nobody can have more than the global setting allows. | |
8891 | */ | |
8892 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8893 | return -EINVAL; | |
6f505b16 | 8894 | |
4653f803 PZ |
8895 | /* |
8896 | * The sum of our children's runtime should not exceed our own. | |
8897 | */ | |
9a7e0b18 PZ |
8898 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8899 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8900 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8901 | |
9a7e0b18 PZ |
8902 | if (child == d->tg) { |
8903 | period = d->rt_period; | |
8904 | runtime = d->rt_runtime; | |
8905 | } | |
6f505b16 | 8906 | |
9a7e0b18 | 8907 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8908 | } |
6f505b16 | 8909 | |
9a7e0b18 PZ |
8910 | if (sum > total) |
8911 | return -EINVAL; | |
8912 | ||
8913 | return 0; | |
6f505b16 PZ |
8914 | } |
8915 | ||
9a7e0b18 | 8916 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8917 | { |
9a7e0b18 PZ |
8918 | struct rt_schedulable_data data = { |
8919 | .tg = tg, | |
8920 | .rt_period = period, | |
8921 | .rt_runtime = runtime, | |
8922 | }; | |
8923 | ||
8924 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8925 | } |
8926 | ||
d0b27fa7 PZ |
8927 | static int tg_set_bandwidth(struct task_group *tg, |
8928 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8929 | { |
ac086bc2 | 8930 | int i, err = 0; |
9f0c1e56 | 8931 | |
9f0c1e56 | 8932 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8933 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8934 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8935 | if (err) | |
9f0c1e56 | 8936 | goto unlock; |
ac086bc2 | 8937 | |
0986b11b | 8938 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8939 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8940 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8941 | |
8942 | for_each_possible_cpu(i) { | |
8943 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8944 | ||
0986b11b | 8945 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8946 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8947 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8948 | } |
0986b11b | 8949 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8950 | unlock: |
521f1a24 | 8951 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8952 | mutex_unlock(&rt_constraints_mutex); |
8953 | ||
8954 | return err; | |
6f505b16 PZ |
8955 | } |
8956 | ||
d0b27fa7 PZ |
8957 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8958 | { | |
8959 | u64 rt_runtime, rt_period; | |
8960 | ||
8961 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8962 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8963 | if (rt_runtime_us < 0) | |
8964 | rt_runtime = RUNTIME_INF; | |
8965 | ||
8966 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8967 | } | |
8968 | ||
9f0c1e56 PZ |
8969 | long sched_group_rt_runtime(struct task_group *tg) |
8970 | { | |
8971 | u64 rt_runtime_us; | |
8972 | ||
d0b27fa7 | 8973 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8974 | return -1; |
8975 | ||
d0b27fa7 | 8976 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8977 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8978 | return rt_runtime_us; | |
8979 | } | |
d0b27fa7 PZ |
8980 | |
8981 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8982 | { | |
8983 | u64 rt_runtime, rt_period; | |
8984 | ||
8985 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8986 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8987 | ||
619b0488 R |
8988 | if (rt_period == 0) |
8989 | return -EINVAL; | |
8990 | ||
d0b27fa7 PZ |
8991 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8992 | } | |
8993 | ||
8994 | long sched_group_rt_period(struct task_group *tg) | |
8995 | { | |
8996 | u64 rt_period_us; | |
8997 | ||
8998 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8999 | do_div(rt_period_us, NSEC_PER_USEC); | |
9000 | return rt_period_us; | |
9001 | } | |
9002 | ||
9003 | static int sched_rt_global_constraints(void) | |
9004 | { | |
4653f803 | 9005 | u64 runtime, period; |
d0b27fa7 PZ |
9006 | int ret = 0; |
9007 | ||
ec5d4989 HS |
9008 | if (sysctl_sched_rt_period <= 0) |
9009 | return -EINVAL; | |
9010 | ||
4653f803 PZ |
9011 | runtime = global_rt_runtime(); |
9012 | period = global_rt_period(); | |
9013 | ||
9014 | /* | |
9015 | * Sanity check on the sysctl variables. | |
9016 | */ | |
9017 | if (runtime > period && runtime != RUNTIME_INF) | |
9018 | return -EINVAL; | |
10b612f4 | 9019 | |
d0b27fa7 | 9020 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9021 | read_lock(&tasklist_lock); |
4653f803 | 9022 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9023 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9024 | mutex_unlock(&rt_constraints_mutex); |
9025 | ||
9026 | return ret; | |
9027 | } | |
54e99124 DG |
9028 | |
9029 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9030 | { | |
9031 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9032 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9033 | return 0; | |
9034 | ||
9035 | return 1; | |
9036 | } | |
9037 | ||
6d6bc0ad | 9038 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9039 | static int sched_rt_global_constraints(void) |
9040 | { | |
ac086bc2 PZ |
9041 | unsigned long flags; |
9042 | int i; | |
9043 | ||
ec5d4989 HS |
9044 | if (sysctl_sched_rt_period <= 0) |
9045 | return -EINVAL; | |
9046 | ||
60aa605d PZ |
9047 | /* |
9048 | * There's always some RT tasks in the root group | |
9049 | * -- migration, kstopmachine etc.. | |
9050 | */ | |
9051 | if (sysctl_sched_rt_runtime == 0) | |
9052 | return -EBUSY; | |
9053 | ||
0986b11b | 9054 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
9055 | for_each_possible_cpu(i) { |
9056 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9057 | ||
0986b11b | 9058 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9059 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 9060 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9061 | } |
0986b11b | 9062 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 9063 | |
d0b27fa7 PZ |
9064 | return 0; |
9065 | } | |
6d6bc0ad | 9066 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9067 | |
9068 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 9069 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
9070 | loff_t *ppos) |
9071 | { | |
9072 | int ret; | |
9073 | int old_period, old_runtime; | |
9074 | static DEFINE_MUTEX(mutex); | |
9075 | ||
9076 | mutex_lock(&mutex); | |
9077 | old_period = sysctl_sched_rt_period; | |
9078 | old_runtime = sysctl_sched_rt_runtime; | |
9079 | ||
8d65af78 | 9080 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9081 | |
9082 | if (!ret && write) { | |
9083 | ret = sched_rt_global_constraints(); | |
9084 | if (ret) { | |
9085 | sysctl_sched_rt_period = old_period; | |
9086 | sysctl_sched_rt_runtime = old_runtime; | |
9087 | } else { | |
9088 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9089 | def_rt_bandwidth.rt_period = | |
9090 | ns_to_ktime(global_rt_period()); | |
9091 | } | |
9092 | } | |
9093 | mutex_unlock(&mutex); | |
9094 | ||
9095 | return ret; | |
9096 | } | |
68318b8e | 9097 | |
052f1dc7 | 9098 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9099 | |
9100 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9101 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9102 | { |
2b01dfe3 PM |
9103 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9104 | struct task_group, css); | |
68318b8e SV |
9105 | } |
9106 | ||
9107 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9108 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9109 | { |
ec7dc8ac | 9110 | struct task_group *tg, *parent; |
68318b8e | 9111 | |
2b01dfe3 | 9112 | if (!cgrp->parent) { |
68318b8e | 9113 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9114 | return &root_task_group.css; |
68318b8e SV |
9115 | } |
9116 | ||
ec7dc8ac DG |
9117 | parent = cgroup_tg(cgrp->parent); |
9118 | tg = sched_create_group(parent); | |
68318b8e SV |
9119 | if (IS_ERR(tg)) |
9120 | return ERR_PTR(-ENOMEM); | |
9121 | ||
68318b8e SV |
9122 | return &tg->css; |
9123 | } | |
9124 | ||
41a2d6cf IM |
9125 | static void |
9126 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9127 | { |
2b01dfe3 | 9128 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9129 | |
9130 | sched_destroy_group(tg); | |
9131 | } | |
9132 | ||
41a2d6cf | 9133 | static int |
be367d09 | 9134 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9135 | { |
b68aa230 | 9136 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9137 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9138 | return -EINVAL; |
9139 | #else | |
68318b8e SV |
9140 | /* We don't support RT-tasks being in separate groups */ |
9141 | if (tsk->sched_class != &fair_sched_class) | |
9142 | return -EINVAL; | |
b68aa230 | 9143 | #endif |
be367d09 BB |
9144 | return 0; |
9145 | } | |
68318b8e | 9146 | |
be367d09 BB |
9147 | static int |
9148 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9149 | struct task_struct *tsk, bool threadgroup) | |
9150 | { | |
9151 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9152 | if (retval) | |
9153 | return retval; | |
9154 | if (threadgroup) { | |
9155 | struct task_struct *c; | |
9156 | rcu_read_lock(); | |
9157 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9158 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9159 | if (retval) { | |
9160 | rcu_read_unlock(); | |
9161 | return retval; | |
9162 | } | |
9163 | } | |
9164 | rcu_read_unlock(); | |
9165 | } | |
68318b8e SV |
9166 | return 0; |
9167 | } | |
9168 | ||
9169 | static void | |
2b01dfe3 | 9170 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9171 | struct cgroup *old_cont, struct task_struct *tsk, |
9172 | bool threadgroup) | |
68318b8e SV |
9173 | { |
9174 | sched_move_task(tsk); | |
be367d09 BB |
9175 | if (threadgroup) { |
9176 | struct task_struct *c; | |
9177 | rcu_read_lock(); | |
9178 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9179 | sched_move_task(c); | |
9180 | } | |
9181 | rcu_read_unlock(); | |
9182 | } | |
68318b8e SV |
9183 | } |
9184 | ||
068c5cc5 | 9185 | static void |
d41d5a01 PZ |
9186 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
9187 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
9188 | { |
9189 | /* | |
9190 | * cgroup_exit() is called in the copy_process() failure path. | |
9191 | * Ignore this case since the task hasn't ran yet, this avoids | |
9192 | * trying to poke a half freed task state from generic code. | |
9193 | */ | |
9194 | if (!(task->flags & PF_EXITING)) | |
9195 | return; | |
9196 | ||
9197 | sched_move_task(task); | |
9198 | } | |
9199 | ||
052f1dc7 | 9200 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9201 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9202 | u64 shareval) |
68318b8e | 9203 | { |
2b01dfe3 | 9204 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9205 | } |
9206 | ||
f4c753b7 | 9207 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9208 | { |
2b01dfe3 | 9209 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9210 | |
9211 | return (u64) tg->shares; | |
9212 | } | |
6d6bc0ad | 9213 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9214 | |
052f1dc7 | 9215 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9216 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9217 | s64 val) |
6f505b16 | 9218 | { |
06ecb27c | 9219 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9220 | } |
9221 | ||
06ecb27c | 9222 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9223 | { |
06ecb27c | 9224 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9225 | } |
d0b27fa7 PZ |
9226 | |
9227 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9228 | u64 rt_period_us) | |
9229 | { | |
9230 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9231 | } | |
9232 | ||
9233 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9234 | { | |
9235 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9236 | } | |
6d6bc0ad | 9237 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9238 | |
fe5c7cc2 | 9239 | static struct cftype cpu_files[] = { |
052f1dc7 | 9240 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9241 | { |
9242 | .name = "shares", | |
f4c753b7 PM |
9243 | .read_u64 = cpu_shares_read_u64, |
9244 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9245 | }, |
052f1dc7 PZ |
9246 | #endif |
9247 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9248 | { |
9f0c1e56 | 9249 | .name = "rt_runtime_us", |
06ecb27c PM |
9250 | .read_s64 = cpu_rt_runtime_read, |
9251 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9252 | }, |
d0b27fa7 PZ |
9253 | { |
9254 | .name = "rt_period_us", | |
f4c753b7 PM |
9255 | .read_u64 = cpu_rt_period_read_uint, |
9256 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9257 | }, |
052f1dc7 | 9258 | #endif |
68318b8e SV |
9259 | }; |
9260 | ||
9261 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9262 | { | |
fe5c7cc2 | 9263 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9264 | } |
9265 | ||
9266 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9267 | .name = "cpu", |
9268 | .create = cpu_cgroup_create, | |
9269 | .destroy = cpu_cgroup_destroy, | |
9270 | .can_attach = cpu_cgroup_can_attach, | |
9271 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 9272 | .exit = cpu_cgroup_exit, |
38605cae IM |
9273 | .populate = cpu_cgroup_populate, |
9274 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9275 | .early_init = 1, |
9276 | }; | |
9277 | ||
052f1dc7 | 9278 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9279 | |
9280 | #ifdef CONFIG_CGROUP_CPUACCT | |
9281 | ||
9282 | /* | |
9283 | * CPU accounting code for task groups. | |
9284 | * | |
9285 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9286 | * (balbir@in.ibm.com). | |
9287 | */ | |
9288 | ||
934352f2 | 9289 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9290 | struct cpuacct { |
9291 | struct cgroup_subsys_state css; | |
9292 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9293 | u64 __percpu *cpuusage; |
ef12fefa | 9294 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9295 | struct cpuacct *parent; |
d842de87 SV |
9296 | }; |
9297 | ||
9298 | struct cgroup_subsys cpuacct_subsys; | |
9299 | ||
9300 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9301 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9302 | { |
32cd756a | 9303 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9304 | struct cpuacct, css); |
9305 | } | |
9306 | ||
9307 | /* return cpu accounting group to which this task belongs */ | |
9308 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9309 | { | |
9310 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9311 | struct cpuacct, css); | |
9312 | } | |
9313 | ||
9314 | /* create a new cpu accounting group */ | |
9315 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9316 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9317 | { |
9318 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9319 | int i; |
d842de87 SV |
9320 | |
9321 | if (!ca) | |
ef12fefa | 9322 | goto out; |
d842de87 SV |
9323 | |
9324 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9325 | if (!ca->cpuusage) |
9326 | goto out_free_ca; | |
9327 | ||
9328 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9329 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9330 | goto out_free_counters; | |
d842de87 | 9331 | |
934352f2 BR |
9332 | if (cgrp->parent) |
9333 | ca->parent = cgroup_ca(cgrp->parent); | |
9334 | ||
d842de87 | 9335 | return &ca->css; |
ef12fefa BR |
9336 | |
9337 | out_free_counters: | |
9338 | while (--i >= 0) | |
9339 | percpu_counter_destroy(&ca->cpustat[i]); | |
9340 | free_percpu(ca->cpuusage); | |
9341 | out_free_ca: | |
9342 | kfree(ca); | |
9343 | out: | |
9344 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9345 | } |
9346 | ||
9347 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9348 | static void |
32cd756a | 9349 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9350 | { |
32cd756a | 9351 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9352 | int i; |
d842de87 | 9353 | |
ef12fefa BR |
9354 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9355 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9356 | free_percpu(ca->cpuusage); |
9357 | kfree(ca); | |
9358 | } | |
9359 | ||
720f5498 KC |
9360 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9361 | { | |
b36128c8 | 9362 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9363 | u64 data; |
9364 | ||
9365 | #ifndef CONFIG_64BIT | |
9366 | /* | |
9367 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9368 | */ | |
05fa785c | 9369 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9370 | data = *cpuusage; |
05fa785c | 9371 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9372 | #else |
9373 | data = *cpuusage; | |
9374 | #endif | |
9375 | ||
9376 | return data; | |
9377 | } | |
9378 | ||
9379 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9380 | { | |
b36128c8 | 9381 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9382 | |
9383 | #ifndef CONFIG_64BIT | |
9384 | /* | |
9385 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9386 | */ | |
05fa785c | 9387 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9388 | *cpuusage = val; |
05fa785c | 9389 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9390 | #else |
9391 | *cpuusage = val; | |
9392 | #endif | |
9393 | } | |
9394 | ||
d842de87 | 9395 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9396 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9397 | { |
32cd756a | 9398 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9399 | u64 totalcpuusage = 0; |
9400 | int i; | |
9401 | ||
720f5498 KC |
9402 | for_each_present_cpu(i) |
9403 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9404 | |
9405 | return totalcpuusage; | |
9406 | } | |
9407 | ||
0297b803 DG |
9408 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9409 | u64 reset) | |
9410 | { | |
9411 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9412 | int err = 0; | |
9413 | int i; | |
9414 | ||
9415 | if (reset) { | |
9416 | err = -EINVAL; | |
9417 | goto out; | |
9418 | } | |
9419 | ||
720f5498 KC |
9420 | for_each_present_cpu(i) |
9421 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9422 | |
0297b803 DG |
9423 | out: |
9424 | return err; | |
9425 | } | |
9426 | ||
e9515c3c KC |
9427 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9428 | struct seq_file *m) | |
9429 | { | |
9430 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9431 | u64 percpu; | |
9432 | int i; | |
9433 | ||
9434 | for_each_present_cpu(i) { | |
9435 | percpu = cpuacct_cpuusage_read(ca, i); | |
9436 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9437 | } | |
9438 | seq_printf(m, "\n"); | |
9439 | return 0; | |
9440 | } | |
9441 | ||
ef12fefa BR |
9442 | static const char *cpuacct_stat_desc[] = { |
9443 | [CPUACCT_STAT_USER] = "user", | |
9444 | [CPUACCT_STAT_SYSTEM] = "system", | |
9445 | }; | |
9446 | ||
9447 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9448 | struct cgroup_map_cb *cb) | |
9449 | { | |
9450 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9451 | int i; | |
9452 | ||
9453 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9454 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9455 | val = cputime64_to_clock_t(val); | |
9456 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9457 | } | |
9458 | return 0; | |
9459 | } | |
9460 | ||
d842de87 SV |
9461 | static struct cftype files[] = { |
9462 | { | |
9463 | .name = "usage", | |
f4c753b7 PM |
9464 | .read_u64 = cpuusage_read, |
9465 | .write_u64 = cpuusage_write, | |
d842de87 | 9466 | }, |
e9515c3c KC |
9467 | { |
9468 | .name = "usage_percpu", | |
9469 | .read_seq_string = cpuacct_percpu_seq_read, | |
9470 | }, | |
ef12fefa BR |
9471 | { |
9472 | .name = "stat", | |
9473 | .read_map = cpuacct_stats_show, | |
9474 | }, | |
d842de87 SV |
9475 | }; |
9476 | ||
32cd756a | 9477 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9478 | { |
32cd756a | 9479 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9480 | } |
9481 | ||
9482 | /* | |
9483 | * charge this task's execution time to its accounting group. | |
9484 | * | |
9485 | * called with rq->lock held. | |
9486 | */ | |
9487 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9488 | { | |
9489 | struct cpuacct *ca; | |
934352f2 | 9490 | int cpu; |
d842de87 | 9491 | |
c40c6f85 | 9492 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9493 | return; |
9494 | ||
934352f2 | 9495 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9496 | |
9497 | rcu_read_lock(); | |
9498 | ||
d842de87 | 9499 | ca = task_ca(tsk); |
d842de87 | 9500 | |
934352f2 | 9501 | for (; ca; ca = ca->parent) { |
b36128c8 | 9502 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9503 | *cpuusage += cputime; |
9504 | } | |
a18b83b7 BR |
9505 | |
9506 | rcu_read_unlock(); | |
d842de87 SV |
9507 | } |
9508 | ||
fa535a77 AB |
9509 | /* |
9510 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9511 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9512 | * percpu_counter_add with values large enough to always overflow the | |
9513 | * per cpu batch limit causing bad SMP scalability. | |
9514 | * | |
9515 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9516 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9517 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9518 | */ | |
9519 | #ifdef CONFIG_SMP | |
9520 | #define CPUACCT_BATCH \ | |
9521 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9522 | #else | |
9523 | #define CPUACCT_BATCH 0 | |
9524 | #endif | |
9525 | ||
ef12fefa BR |
9526 | /* |
9527 | * Charge the system/user time to the task's accounting group. | |
9528 | */ | |
9529 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9530 | enum cpuacct_stat_index idx, cputime_t val) | |
9531 | { | |
9532 | struct cpuacct *ca; | |
fa535a77 | 9533 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9534 | |
9535 | if (unlikely(!cpuacct_subsys.active)) | |
9536 | return; | |
9537 | ||
9538 | rcu_read_lock(); | |
9539 | ca = task_ca(tsk); | |
9540 | ||
9541 | do { | |
fa535a77 | 9542 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9543 | ca = ca->parent; |
9544 | } while (ca); | |
9545 | rcu_read_unlock(); | |
9546 | } | |
9547 | ||
d842de87 SV |
9548 | struct cgroup_subsys cpuacct_subsys = { |
9549 | .name = "cpuacct", | |
9550 | .create = cpuacct_create, | |
9551 | .destroy = cpuacct_destroy, | |
9552 | .populate = cpuacct_populate, | |
9553 | .subsys_id = cpuacct_subsys_id, | |
9554 | }; | |
9555 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9556 |