Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
335d7afb | 78 | #include <asm/mutex.h> |
1da177e4 | 79 | |
6e0534f2 | 80 | #include "sched_cpupri.h" |
21aa9af0 | 81 | #include "workqueue_sched.h" |
5091faa4 | 82 | #include "sched_autogroup.h" |
6e0534f2 | 83 | |
a8d154b0 | 84 | #define CREATE_TRACE_POINTS |
ad8d75ff | 85 | #include <trace/events/sched.h> |
a8d154b0 | 86 | |
1da177e4 LT |
87 | /* |
88 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
89 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
90 | * and back. | |
91 | */ | |
92 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
93 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
94 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
95 | ||
96 | /* | |
97 | * 'User priority' is the nice value converted to something we | |
98 | * can work with better when scaling various scheduler parameters, | |
99 | * it's a [ 0 ... 39 ] range. | |
100 | */ | |
101 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
102 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
103 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
104 | ||
105 | /* | |
d7876a08 | 106 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 107 | */ |
d6322faf | 108 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 109 | |
6aa645ea IM |
110 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
111 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
112 | ||
1da177e4 LT |
113 | /* |
114 | * These are the 'tuning knobs' of the scheduler: | |
115 | * | |
a4ec24b4 | 116 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
117 | * Timeslices get refilled after they expire. |
118 | */ | |
1da177e4 | 119 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 120 | |
d0b27fa7 PZ |
121 | /* |
122 | * single value that denotes runtime == period, ie unlimited time. | |
123 | */ | |
124 | #define RUNTIME_INF ((u64)~0ULL) | |
125 | ||
e05606d3 IM |
126 | static inline int rt_policy(int policy) |
127 | { | |
3f33a7ce | 128 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
129 | return 1; |
130 | return 0; | |
131 | } | |
132 | ||
133 | static inline int task_has_rt_policy(struct task_struct *p) | |
134 | { | |
135 | return rt_policy(p->policy); | |
136 | } | |
137 | ||
1da177e4 | 138 | /* |
6aa645ea | 139 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 140 | */ |
6aa645ea IM |
141 | struct rt_prio_array { |
142 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
143 | struct list_head queue[MAX_RT_PRIO]; | |
144 | }; | |
145 | ||
d0b27fa7 | 146 | struct rt_bandwidth { |
ea736ed5 | 147 | /* nests inside the rq lock: */ |
0986b11b | 148 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
149 | ktime_t rt_period; |
150 | u64 rt_runtime; | |
151 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
152 | }; |
153 | ||
154 | static struct rt_bandwidth def_rt_bandwidth; | |
155 | ||
156 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
157 | ||
158 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
159 | { | |
160 | struct rt_bandwidth *rt_b = | |
161 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
162 | ktime_t now; | |
163 | int overrun; | |
164 | int idle = 0; | |
165 | ||
166 | for (;;) { | |
167 | now = hrtimer_cb_get_time(timer); | |
168 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
169 | ||
170 | if (!overrun) | |
171 | break; | |
172 | ||
173 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
174 | } | |
175 | ||
176 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
177 | } | |
178 | ||
179 | static | |
180 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
181 | { | |
182 | rt_b->rt_period = ns_to_ktime(period); | |
183 | rt_b->rt_runtime = runtime; | |
184 | ||
0986b11b | 185 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 186 | |
d0b27fa7 PZ |
187 | hrtimer_init(&rt_b->rt_period_timer, |
188 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
189 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
190 | } |
191 | ||
c8bfff6d KH |
192 | static inline int rt_bandwidth_enabled(void) |
193 | { | |
194 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
195 | } |
196 | ||
197 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
198 | { | |
199 | ktime_t now; | |
200 | ||
cac64d00 | 201 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
202 | return; |
203 | ||
204 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
205 | return; | |
206 | ||
0986b11b | 207 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 208 | for (;;) { |
7f1e2ca9 PZ |
209 | unsigned long delta; |
210 | ktime_t soft, hard; | |
211 | ||
d0b27fa7 PZ |
212 | if (hrtimer_active(&rt_b->rt_period_timer)) |
213 | break; | |
214 | ||
215 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
216 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
217 | |
218 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
219 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
220 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
221 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 222 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 223 | } |
0986b11b | 224 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
225 | } |
226 | ||
227 | #ifdef CONFIG_RT_GROUP_SCHED | |
228 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
229 | { | |
230 | hrtimer_cancel(&rt_b->rt_period_timer); | |
231 | } | |
232 | #endif | |
233 | ||
712555ee HC |
234 | /* |
235 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
236 | * detach_destroy_domains and partition_sched_domains. | |
237 | */ | |
238 | static DEFINE_MUTEX(sched_domains_mutex); | |
239 | ||
7c941438 | 240 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 241 | |
68318b8e SV |
242 | #include <linux/cgroup.h> |
243 | ||
29f59db3 SV |
244 | struct cfs_rq; |
245 | ||
6f505b16 PZ |
246 | static LIST_HEAD(task_groups); |
247 | ||
29f59db3 | 248 | /* task group related information */ |
4cf86d77 | 249 | struct task_group { |
68318b8e | 250 | struct cgroup_subsys_state css; |
6c415b92 | 251 | |
052f1dc7 | 252 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
253 | /* schedulable entities of this group on each cpu */ |
254 | struct sched_entity **se; | |
255 | /* runqueue "owned" by this group on each cpu */ | |
256 | struct cfs_rq **cfs_rq; | |
257 | unsigned long shares; | |
2069dd75 PZ |
258 | |
259 | atomic_t load_weight; | |
052f1dc7 PZ |
260 | #endif |
261 | ||
262 | #ifdef CONFIG_RT_GROUP_SCHED | |
263 | struct sched_rt_entity **rt_se; | |
264 | struct rt_rq **rt_rq; | |
265 | ||
d0b27fa7 | 266 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 267 | #endif |
6b2d7700 | 268 | |
ae8393e5 | 269 | struct rcu_head rcu; |
6f505b16 | 270 | struct list_head list; |
f473aa5e PZ |
271 | |
272 | struct task_group *parent; | |
273 | struct list_head siblings; | |
274 | struct list_head children; | |
5091faa4 MG |
275 | |
276 | #ifdef CONFIG_SCHED_AUTOGROUP | |
277 | struct autogroup *autogroup; | |
278 | #endif | |
29f59db3 SV |
279 | }; |
280 | ||
3d4b47b4 | 281 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 282 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 283 | |
e9036b36 CG |
284 | #ifdef CONFIG_FAIR_GROUP_SCHED |
285 | ||
07e06b01 | 286 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 287 | |
cb4ad1ff | 288 | /* |
2e084786 LJ |
289 | * A weight of 0 or 1 can cause arithmetics problems. |
290 | * A weight of a cfs_rq is the sum of weights of which entities | |
291 | * are queued on this cfs_rq, so a weight of a entity should not be | |
292 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
293 | * (The default weight is 1024 - so there's no practical |
294 | * limitation from this.) | |
295 | */ | |
18d95a28 | 296 | #define MIN_SHARES 2 |
2e084786 | 297 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 298 | |
07e06b01 | 299 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
300 | #endif |
301 | ||
29f59db3 | 302 | /* Default task group. |
3a252015 | 303 | * Every task in system belong to this group at bootup. |
29f59db3 | 304 | */ |
07e06b01 | 305 | struct task_group root_task_group; |
29f59db3 | 306 | |
7c941438 | 307 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 308 | |
6aa645ea IM |
309 | /* CFS-related fields in a runqueue */ |
310 | struct cfs_rq { | |
311 | struct load_weight load; | |
312 | unsigned long nr_running; | |
313 | ||
6aa645ea | 314 | u64 exec_clock; |
e9acbff6 | 315 | u64 min_vruntime; |
6aa645ea IM |
316 | |
317 | struct rb_root tasks_timeline; | |
318 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
319 | |
320 | struct list_head tasks; | |
321 | struct list_head *balance_iterator; | |
322 | ||
323 | /* | |
324 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
325 | * It is set to NULL otherwise (i.e when none are currently running). |
326 | */ | |
ac53db59 | 327 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 328 | |
5ac5c4d6 | 329 | unsigned int nr_spread_over; |
ddc97297 | 330 | |
62160e3f | 331 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
332 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
333 | ||
41a2d6cf IM |
334 | /* |
335 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
336 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
337 | * (like users, containers etc.) | |
338 | * | |
339 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
340 | * list is used during load balance. | |
341 | */ | |
3d4b47b4 | 342 | int on_list; |
41a2d6cf IM |
343 | struct list_head leaf_cfs_rq_list; |
344 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
345 | |
346 | #ifdef CONFIG_SMP | |
c09595f6 | 347 | /* |
c8cba857 | 348 | * the part of load.weight contributed by tasks |
c09595f6 | 349 | */ |
c8cba857 | 350 | unsigned long task_weight; |
c09595f6 | 351 | |
c8cba857 PZ |
352 | /* |
353 | * h_load = weight * f(tg) | |
354 | * | |
355 | * Where f(tg) is the recursive weight fraction assigned to | |
356 | * this group. | |
357 | */ | |
358 | unsigned long h_load; | |
c09595f6 | 359 | |
c8cba857 | 360 | /* |
3b3d190e PT |
361 | * Maintaining per-cpu shares distribution for group scheduling |
362 | * | |
363 | * load_stamp is the last time we updated the load average | |
364 | * load_last is the last time we updated the load average and saw load | |
365 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 366 | */ |
2069dd75 PZ |
367 | u64 load_avg; |
368 | u64 load_period; | |
3b3d190e | 369 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 370 | |
2069dd75 | 371 | unsigned long load_contribution; |
c09595f6 | 372 | #endif |
6aa645ea IM |
373 | #endif |
374 | }; | |
1da177e4 | 375 | |
6aa645ea IM |
376 | /* Real-Time classes' related field in a runqueue: */ |
377 | struct rt_rq { | |
378 | struct rt_prio_array active; | |
63489e45 | 379 | unsigned long rt_nr_running; |
052f1dc7 | 380 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
381 | struct { |
382 | int curr; /* highest queued rt task prio */ | |
398a153b | 383 | #ifdef CONFIG_SMP |
e864c499 | 384 | int next; /* next highest */ |
398a153b | 385 | #endif |
e864c499 | 386 | } highest_prio; |
6f505b16 | 387 | #endif |
fa85ae24 | 388 | #ifdef CONFIG_SMP |
73fe6aae | 389 | unsigned long rt_nr_migratory; |
a1ba4d8b | 390 | unsigned long rt_nr_total; |
a22d7fc1 | 391 | int overloaded; |
917b627d | 392 | struct plist_head pushable_tasks; |
fa85ae24 | 393 | #endif |
6f505b16 | 394 | int rt_throttled; |
fa85ae24 | 395 | u64 rt_time; |
ac086bc2 | 396 | u64 rt_runtime; |
ea736ed5 | 397 | /* Nests inside the rq lock: */ |
0986b11b | 398 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 399 | |
052f1dc7 | 400 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
401 | unsigned long rt_nr_boosted; |
402 | ||
6f505b16 PZ |
403 | struct rq *rq; |
404 | struct list_head leaf_rt_rq_list; | |
405 | struct task_group *tg; | |
6f505b16 | 406 | #endif |
6aa645ea IM |
407 | }; |
408 | ||
57d885fe GH |
409 | #ifdef CONFIG_SMP |
410 | ||
411 | /* | |
412 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
413 | * variables. Each exclusive cpuset essentially defines an island domain by |
414 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
415 | * exclusive cpuset is created, we also create and attach a new root-domain |
416 | * object. | |
417 | * | |
57d885fe GH |
418 | */ |
419 | struct root_domain { | |
420 | atomic_t refcount; | |
c6c4927b RR |
421 | cpumask_var_t span; |
422 | cpumask_var_t online; | |
637f5085 | 423 | |
0eab9146 | 424 | /* |
637f5085 GH |
425 | * The "RT overload" flag: it gets set if a CPU has more than |
426 | * one runnable RT task. | |
427 | */ | |
c6c4927b | 428 | cpumask_var_t rto_mask; |
0eab9146 | 429 | atomic_t rto_count; |
6e0534f2 | 430 | struct cpupri cpupri; |
57d885fe GH |
431 | }; |
432 | ||
dc938520 GH |
433 | /* |
434 | * By default the system creates a single root-domain with all cpus as | |
435 | * members (mimicking the global state we have today). | |
436 | */ | |
57d885fe GH |
437 | static struct root_domain def_root_domain; |
438 | ||
ed2d372c | 439 | #endif /* CONFIG_SMP */ |
57d885fe | 440 | |
1da177e4 LT |
441 | /* |
442 | * This is the main, per-CPU runqueue data structure. | |
443 | * | |
444 | * Locking rule: those places that want to lock multiple runqueues | |
445 | * (such as the load balancing or the thread migration code), lock | |
446 | * acquire operations must be ordered by ascending &runqueue. | |
447 | */ | |
70b97a7f | 448 | struct rq { |
d8016491 | 449 | /* runqueue lock: */ |
05fa785c | 450 | raw_spinlock_t lock; |
1da177e4 LT |
451 | |
452 | /* | |
453 | * nr_running and cpu_load should be in the same cacheline because | |
454 | * remote CPUs use both these fields when doing load calculation. | |
455 | */ | |
456 | unsigned long nr_running; | |
6aa645ea IM |
457 | #define CPU_LOAD_IDX_MAX 5 |
458 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 459 | unsigned long last_load_update_tick; |
46cb4b7c | 460 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 461 | u64 nohz_stamp; |
83cd4fe2 | 462 | unsigned char nohz_balance_kick; |
46cb4b7c | 463 | #endif |
a64692a3 MG |
464 | unsigned int skip_clock_update; |
465 | ||
d8016491 IM |
466 | /* capture load from *all* tasks on this cpu: */ |
467 | struct load_weight load; | |
6aa645ea IM |
468 | unsigned long nr_load_updates; |
469 | u64 nr_switches; | |
470 | ||
471 | struct cfs_rq cfs; | |
6f505b16 | 472 | struct rt_rq rt; |
6f505b16 | 473 | |
6aa645ea | 474 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
475 | /* list of leaf cfs_rq on this cpu: */ |
476 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
477 | #endif |
478 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 479 | struct list_head leaf_rt_rq_list; |
1da177e4 | 480 | #endif |
1da177e4 LT |
481 | |
482 | /* | |
483 | * This is part of a global counter where only the total sum | |
484 | * over all CPUs matters. A task can increase this counter on | |
485 | * one CPU and if it got migrated afterwards it may decrease | |
486 | * it on another CPU. Always updated under the runqueue lock: | |
487 | */ | |
488 | unsigned long nr_uninterruptible; | |
489 | ||
34f971f6 | 490 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 491 | unsigned long next_balance; |
1da177e4 | 492 | struct mm_struct *prev_mm; |
6aa645ea | 493 | |
3e51f33f | 494 | u64 clock; |
305e6835 | 495 | u64 clock_task; |
6aa645ea | 496 | |
1da177e4 LT |
497 | atomic_t nr_iowait; |
498 | ||
499 | #ifdef CONFIG_SMP | |
0eab9146 | 500 | struct root_domain *rd; |
1da177e4 LT |
501 | struct sched_domain *sd; |
502 | ||
e51fd5e2 PZ |
503 | unsigned long cpu_power; |
504 | ||
a0a522ce | 505 | unsigned char idle_at_tick; |
1da177e4 | 506 | /* For active balancing */ |
3f029d3c | 507 | int post_schedule; |
1da177e4 LT |
508 | int active_balance; |
509 | int push_cpu; | |
969c7921 | 510 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
511 | /* cpu of this runqueue: */ |
512 | int cpu; | |
1f11eb6a | 513 | int online; |
1da177e4 | 514 | |
a8a51d5e | 515 | unsigned long avg_load_per_task; |
1da177e4 | 516 | |
e9e9250b PZ |
517 | u64 rt_avg; |
518 | u64 age_stamp; | |
1b9508f6 MG |
519 | u64 idle_stamp; |
520 | u64 avg_idle; | |
1da177e4 LT |
521 | #endif |
522 | ||
aa483808 VP |
523 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
524 | u64 prev_irq_time; | |
525 | #endif | |
526 | ||
dce48a84 TG |
527 | /* calc_load related fields */ |
528 | unsigned long calc_load_update; | |
529 | long calc_load_active; | |
530 | ||
8f4d37ec | 531 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
532 | #ifdef CONFIG_SMP |
533 | int hrtick_csd_pending; | |
534 | struct call_single_data hrtick_csd; | |
535 | #endif | |
8f4d37ec PZ |
536 | struct hrtimer hrtick_timer; |
537 | #endif | |
538 | ||
1da177e4 LT |
539 | #ifdef CONFIG_SCHEDSTATS |
540 | /* latency stats */ | |
541 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
542 | unsigned long long rq_cpu_time; |
543 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
544 | |
545 | /* sys_sched_yield() stats */ | |
480b9434 | 546 | unsigned int yld_count; |
1da177e4 LT |
547 | |
548 | /* schedule() stats */ | |
480b9434 KC |
549 | unsigned int sched_switch; |
550 | unsigned int sched_count; | |
551 | unsigned int sched_goidle; | |
1da177e4 LT |
552 | |
553 | /* try_to_wake_up() stats */ | |
480b9434 KC |
554 | unsigned int ttwu_count; |
555 | unsigned int ttwu_local; | |
1da177e4 LT |
556 | #endif |
557 | }; | |
558 | ||
f34e3b61 | 559 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 560 | |
a64692a3 | 561 | |
1e5a7405 | 562 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 563 | |
0a2966b4 CL |
564 | static inline int cpu_of(struct rq *rq) |
565 | { | |
566 | #ifdef CONFIG_SMP | |
567 | return rq->cpu; | |
568 | #else | |
569 | return 0; | |
570 | #endif | |
571 | } | |
572 | ||
497f0ab3 | 573 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
574 | rcu_dereference_check((p), \ |
575 | rcu_read_lock_sched_held() || \ | |
576 | lockdep_is_held(&sched_domains_mutex)) | |
577 | ||
674311d5 NP |
578 | /* |
579 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 580 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
581 | * |
582 | * The domain tree of any CPU may only be accessed from within | |
583 | * preempt-disabled sections. | |
584 | */ | |
48f24c4d | 585 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 586 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
587 | |
588 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
589 | #define this_rq() (&__get_cpu_var(runqueues)) | |
590 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
591 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 592 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 593 | |
dc61b1d6 PZ |
594 | #ifdef CONFIG_CGROUP_SCHED |
595 | ||
596 | /* | |
597 | * Return the group to which this tasks belongs. | |
598 | * | |
599 | * We use task_subsys_state_check() and extend the RCU verification | |
600 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
601 | * holds that lock for each task it moves into the cgroup. Therefore | |
602 | * by holding that lock, we pin the task to the current cgroup. | |
603 | */ | |
604 | static inline struct task_group *task_group(struct task_struct *p) | |
605 | { | |
5091faa4 | 606 | struct task_group *tg; |
dc61b1d6 PZ |
607 | struct cgroup_subsys_state *css; |
608 | ||
068c5cc5 PZ |
609 | if (p->flags & PF_EXITING) |
610 | return &root_task_group; | |
611 | ||
dc61b1d6 PZ |
612 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, |
613 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
614 | tg = container_of(css, struct task_group, css); |
615 | ||
616 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
617 | } |
618 | ||
619 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
620 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
621 | { | |
622 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
623 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
624 | p->se.parent = task_group(p)->se[cpu]; | |
625 | #endif | |
626 | ||
627 | #ifdef CONFIG_RT_GROUP_SCHED | |
628 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
629 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
630 | #endif | |
631 | } | |
632 | ||
633 | #else /* CONFIG_CGROUP_SCHED */ | |
634 | ||
635 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
636 | static inline struct task_group *task_group(struct task_struct *p) | |
637 | { | |
638 | return NULL; | |
639 | } | |
640 | ||
641 | #endif /* CONFIG_CGROUP_SCHED */ | |
642 | ||
fe44d621 | 643 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 644 | |
fe44d621 | 645 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 646 | { |
fe44d621 | 647 | s64 delta; |
305e6835 | 648 | |
f26f9aff MG |
649 | if (rq->skip_clock_update) |
650 | return; | |
aa483808 | 651 | |
fe44d621 PZ |
652 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
653 | rq->clock += delta; | |
654 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
655 | } |
656 | ||
bf5c91ba IM |
657 | /* |
658 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
659 | */ | |
660 | #ifdef CONFIG_SCHED_DEBUG | |
661 | # define const_debug __read_mostly | |
662 | #else | |
663 | # define const_debug static const | |
664 | #endif | |
665 | ||
017730c1 IM |
666 | /** |
667 | * runqueue_is_locked | |
e17b38bf | 668 | * @cpu: the processor in question. |
017730c1 IM |
669 | * |
670 | * Returns true if the current cpu runqueue is locked. | |
671 | * This interface allows printk to be called with the runqueue lock | |
672 | * held and know whether or not it is OK to wake up the klogd. | |
673 | */ | |
89f19f04 | 674 | int runqueue_is_locked(int cpu) |
017730c1 | 675 | { |
05fa785c | 676 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
677 | } |
678 | ||
bf5c91ba IM |
679 | /* |
680 | * Debugging: various feature bits | |
681 | */ | |
f00b45c1 PZ |
682 | |
683 | #define SCHED_FEAT(name, enabled) \ | |
684 | __SCHED_FEAT_##name , | |
685 | ||
bf5c91ba | 686 | enum { |
f00b45c1 | 687 | #include "sched_features.h" |
bf5c91ba IM |
688 | }; |
689 | ||
f00b45c1 PZ |
690 | #undef SCHED_FEAT |
691 | ||
692 | #define SCHED_FEAT(name, enabled) \ | |
693 | (1UL << __SCHED_FEAT_##name) * enabled | | |
694 | ||
bf5c91ba | 695 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
696 | #include "sched_features.h" |
697 | 0; | |
698 | ||
699 | #undef SCHED_FEAT | |
700 | ||
701 | #ifdef CONFIG_SCHED_DEBUG | |
702 | #define SCHED_FEAT(name, enabled) \ | |
703 | #name , | |
704 | ||
983ed7a6 | 705 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
706 | #include "sched_features.h" |
707 | NULL | |
708 | }; | |
709 | ||
710 | #undef SCHED_FEAT | |
711 | ||
34f3a814 | 712 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 713 | { |
f00b45c1 PZ |
714 | int i; |
715 | ||
716 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
717 | if (!(sysctl_sched_features & (1UL << i))) |
718 | seq_puts(m, "NO_"); | |
719 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 720 | } |
34f3a814 | 721 | seq_puts(m, "\n"); |
f00b45c1 | 722 | |
34f3a814 | 723 | return 0; |
f00b45c1 PZ |
724 | } |
725 | ||
726 | static ssize_t | |
727 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
728 | size_t cnt, loff_t *ppos) | |
729 | { | |
730 | char buf[64]; | |
7740191c | 731 | char *cmp; |
f00b45c1 PZ |
732 | int neg = 0; |
733 | int i; | |
734 | ||
735 | if (cnt > 63) | |
736 | cnt = 63; | |
737 | ||
738 | if (copy_from_user(&buf, ubuf, cnt)) | |
739 | return -EFAULT; | |
740 | ||
741 | buf[cnt] = 0; | |
7740191c | 742 | cmp = strstrip(buf); |
f00b45c1 | 743 | |
524429c3 | 744 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
745 | neg = 1; |
746 | cmp += 3; | |
747 | } | |
748 | ||
749 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 750 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
751 | if (neg) |
752 | sysctl_sched_features &= ~(1UL << i); | |
753 | else | |
754 | sysctl_sched_features |= (1UL << i); | |
755 | break; | |
756 | } | |
757 | } | |
758 | ||
759 | if (!sched_feat_names[i]) | |
760 | return -EINVAL; | |
761 | ||
42994724 | 762 | *ppos += cnt; |
f00b45c1 PZ |
763 | |
764 | return cnt; | |
765 | } | |
766 | ||
34f3a814 LZ |
767 | static int sched_feat_open(struct inode *inode, struct file *filp) |
768 | { | |
769 | return single_open(filp, sched_feat_show, NULL); | |
770 | } | |
771 | ||
828c0950 | 772 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
773 | .open = sched_feat_open, |
774 | .write = sched_feat_write, | |
775 | .read = seq_read, | |
776 | .llseek = seq_lseek, | |
777 | .release = single_release, | |
f00b45c1 PZ |
778 | }; |
779 | ||
780 | static __init int sched_init_debug(void) | |
781 | { | |
f00b45c1 PZ |
782 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
783 | &sched_feat_fops); | |
784 | ||
785 | return 0; | |
786 | } | |
787 | late_initcall(sched_init_debug); | |
788 | ||
789 | #endif | |
790 | ||
791 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 792 | |
b82d9fdd PZ |
793 | /* |
794 | * Number of tasks to iterate in a single balance run. | |
795 | * Limited because this is done with IRQs disabled. | |
796 | */ | |
797 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
798 | ||
e9e9250b PZ |
799 | /* |
800 | * period over which we average the RT time consumption, measured | |
801 | * in ms. | |
802 | * | |
803 | * default: 1s | |
804 | */ | |
805 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
806 | ||
fa85ae24 | 807 | /* |
9f0c1e56 | 808 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
809 | * default: 1s |
810 | */ | |
9f0c1e56 | 811 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 812 | |
6892b75e IM |
813 | static __read_mostly int scheduler_running; |
814 | ||
9f0c1e56 PZ |
815 | /* |
816 | * part of the period that we allow rt tasks to run in us. | |
817 | * default: 0.95s | |
818 | */ | |
819 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 820 | |
d0b27fa7 PZ |
821 | static inline u64 global_rt_period(void) |
822 | { | |
823 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
824 | } | |
825 | ||
826 | static inline u64 global_rt_runtime(void) | |
827 | { | |
e26873bb | 828 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
829 | return RUNTIME_INF; |
830 | ||
831 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
832 | } | |
fa85ae24 | 833 | |
1da177e4 | 834 | #ifndef prepare_arch_switch |
4866cde0 NP |
835 | # define prepare_arch_switch(next) do { } while (0) |
836 | #endif | |
837 | #ifndef finish_arch_switch | |
838 | # define finish_arch_switch(prev) do { } while (0) | |
839 | #endif | |
840 | ||
051a1d1a DA |
841 | static inline int task_current(struct rq *rq, struct task_struct *p) |
842 | { | |
843 | return rq->curr == p; | |
844 | } | |
845 | ||
4866cde0 | 846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 847 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 848 | { |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | } |
851 | ||
70b97a7f | 852 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
853 | { |
854 | } | |
855 | ||
70b97a7f | 856 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 857 | { |
da04c035 IM |
858 | #ifdef CONFIG_DEBUG_SPINLOCK |
859 | /* this is a valid case when another task releases the spinlock */ | |
860 | rq->lock.owner = current; | |
861 | #endif | |
8a25d5de IM |
862 | /* |
863 | * If we are tracking spinlock dependencies then we have to | |
864 | * fix up the runqueue lock - which gets 'carried over' from | |
865 | * prev into current: | |
866 | */ | |
867 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
868 | ||
05fa785c | 869 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
870 | } |
871 | ||
872 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
874 | { |
875 | #ifdef CONFIG_SMP | |
876 | return p->oncpu; | |
877 | #else | |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | #endif |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * We can optimise this out completely for !SMP, because the | |
887 | * SMP rebalancing from interrupt is the only thing that cares | |
888 | * here. | |
889 | */ | |
890 | next->oncpu = 1; | |
891 | #endif | |
892 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 893 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 894 | #else |
05fa785c | 895 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
896 | #endif |
897 | } | |
898 | ||
70b97a7f | 899 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
904 | * We must ensure this doesn't happen until the switch is completely | |
905 | * finished. | |
906 | */ | |
907 | smp_wmb(); | |
908 | prev->oncpu = 0; | |
909 | #endif | |
910 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
911 | local_irq_enable(); | |
1da177e4 | 912 | #endif |
4866cde0 NP |
913 | } |
914 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 915 | |
0970d299 | 916 | /* |
65cc8e48 PZ |
917 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
918 | * against ttwu(). | |
0970d299 PZ |
919 | */ |
920 | static inline int task_is_waking(struct task_struct *p) | |
921 | { | |
0017d735 | 922 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
923 | } |
924 | ||
b29739f9 IM |
925 | /* |
926 | * __task_rq_lock - lock the runqueue a given task resides on. | |
927 | * Must be called interrupts disabled. | |
928 | */ | |
70b97a7f | 929 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
930 | __acquires(rq->lock) |
931 | { | |
0970d299 PZ |
932 | struct rq *rq; |
933 | ||
3a5c359a | 934 | for (;;) { |
0970d299 | 935 | rq = task_rq(p); |
05fa785c | 936 | raw_spin_lock(&rq->lock); |
65cc8e48 | 937 | if (likely(rq == task_rq(p))) |
3a5c359a | 938 | return rq; |
05fa785c | 939 | raw_spin_unlock(&rq->lock); |
b29739f9 | 940 | } |
b29739f9 IM |
941 | } |
942 | ||
1da177e4 LT |
943 | /* |
944 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 945 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
946 | * explicitly disabling preemption. |
947 | */ | |
70b97a7f | 948 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
949 | __acquires(rq->lock) |
950 | { | |
70b97a7f | 951 | struct rq *rq; |
1da177e4 | 952 | |
3a5c359a AK |
953 | for (;;) { |
954 | local_irq_save(*flags); | |
955 | rq = task_rq(p); | |
05fa785c | 956 | raw_spin_lock(&rq->lock); |
65cc8e48 | 957 | if (likely(rq == task_rq(p))) |
3a5c359a | 958 | return rq; |
05fa785c | 959 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 960 | } |
1da177e4 LT |
961 | } |
962 | ||
a9957449 | 963 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
964 | __releases(rq->lock) |
965 | { | |
05fa785c | 966 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
967 | } |
968 | ||
70b97a7f | 969 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
970 | __releases(rq->lock) |
971 | { | |
05fa785c | 972 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
973 | } |
974 | ||
1da177e4 | 975 | /* |
cc2a73b5 | 976 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 977 | */ |
a9957449 | 978 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
979 | __acquires(rq->lock) |
980 | { | |
70b97a7f | 981 | struct rq *rq; |
1da177e4 LT |
982 | |
983 | local_irq_disable(); | |
984 | rq = this_rq(); | |
05fa785c | 985 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
986 | |
987 | return rq; | |
988 | } | |
989 | ||
8f4d37ec PZ |
990 | #ifdef CONFIG_SCHED_HRTICK |
991 | /* | |
992 | * Use HR-timers to deliver accurate preemption points. | |
993 | * | |
994 | * Its all a bit involved since we cannot program an hrt while holding the | |
995 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
996 | * reschedule event. | |
997 | * | |
998 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
999 | * rq->lock. | |
1000 | */ | |
8f4d37ec PZ |
1001 | |
1002 | /* | |
1003 | * Use hrtick when: | |
1004 | * - enabled by features | |
1005 | * - hrtimer is actually high res | |
1006 | */ | |
1007 | static inline int hrtick_enabled(struct rq *rq) | |
1008 | { | |
1009 | if (!sched_feat(HRTICK)) | |
1010 | return 0; | |
ba42059f | 1011 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1012 | return 0; |
8f4d37ec PZ |
1013 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1014 | } | |
1015 | ||
8f4d37ec PZ |
1016 | static void hrtick_clear(struct rq *rq) |
1017 | { | |
1018 | if (hrtimer_active(&rq->hrtick_timer)) | |
1019 | hrtimer_cancel(&rq->hrtick_timer); | |
1020 | } | |
1021 | ||
8f4d37ec PZ |
1022 | /* |
1023 | * High-resolution timer tick. | |
1024 | * Runs from hardirq context with interrupts disabled. | |
1025 | */ | |
1026 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1027 | { | |
1028 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1029 | ||
1030 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1031 | ||
05fa785c | 1032 | raw_spin_lock(&rq->lock); |
3e51f33f | 1033 | update_rq_clock(rq); |
8f4d37ec | 1034 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1035 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1036 | |
1037 | return HRTIMER_NORESTART; | |
1038 | } | |
1039 | ||
95e904c7 | 1040 | #ifdef CONFIG_SMP |
31656519 PZ |
1041 | /* |
1042 | * called from hardirq (IPI) context | |
1043 | */ | |
1044 | static void __hrtick_start(void *arg) | |
b328ca18 | 1045 | { |
31656519 | 1046 | struct rq *rq = arg; |
b328ca18 | 1047 | |
05fa785c | 1048 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1049 | hrtimer_restart(&rq->hrtick_timer); |
1050 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1051 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1052 | } |
1053 | ||
31656519 PZ |
1054 | /* |
1055 | * Called to set the hrtick timer state. | |
1056 | * | |
1057 | * called with rq->lock held and irqs disabled | |
1058 | */ | |
1059 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1060 | { |
31656519 PZ |
1061 | struct hrtimer *timer = &rq->hrtick_timer; |
1062 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1063 | |
cc584b21 | 1064 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1065 | |
1066 | if (rq == this_rq()) { | |
1067 | hrtimer_restart(timer); | |
1068 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1069 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1070 | rq->hrtick_csd_pending = 1; |
1071 | } | |
b328ca18 PZ |
1072 | } |
1073 | ||
1074 | static int | |
1075 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1076 | { | |
1077 | int cpu = (int)(long)hcpu; | |
1078 | ||
1079 | switch (action) { | |
1080 | case CPU_UP_CANCELED: | |
1081 | case CPU_UP_CANCELED_FROZEN: | |
1082 | case CPU_DOWN_PREPARE: | |
1083 | case CPU_DOWN_PREPARE_FROZEN: | |
1084 | case CPU_DEAD: | |
1085 | case CPU_DEAD_FROZEN: | |
31656519 | 1086 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1087 | return NOTIFY_OK; |
1088 | } | |
1089 | ||
1090 | return NOTIFY_DONE; | |
1091 | } | |
1092 | ||
fa748203 | 1093 | static __init void init_hrtick(void) |
b328ca18 PZ |
1094 | { |
1095 | hotcpu_notifier(hotplug_hrtick, 0); | |
1096 | } | |
31656519 PZ |
1097 | #else |
1098 | /* | |
1099 | * Called to set the hrtick timer state. | |
1100 | * | |
1101 | * called with rq->lock held and irqs disabled | |
1102 | */ | |
1103 | static void hrtick_start(struct rq *rq, u64 delay) | |
1104 | { | |
7f1e2ca9 | 1105 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1106 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1107 | } |
b328ca18 | 1108 | |
006c75f1 | 1109 | static inline void init_hrtick(void) |
8f4d37ec | 1110 | { |
8f4d37ec | 1111 | } |
31656519 | 1112 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1113 | |
31656519 | 1114 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1115 | { |
31656519 PZ |
1116 | #ifdef CONFIG_SMP |
1117 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1118 | |
31656519 PZ |
1119 | rq->hrtick_csd.flags = 0; |
1120 | rq->hrtick_csd.func = __hrtick_start; | |
1121 | rq->hrtick_csd.info = rq; | |
1122 | #endif | |
8f4d37ec | 1123 | |
31656519 PZ |
1124 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1125 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1126 | } |
006c75f1 | 1127 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1128 | static inline void hrtick_clear(struct rq *rq) |
1129 | { | |
1130 | } | |
1131 | ||
8f4d37ec PZ |
1132 | static inline void init_rq_hrtick(struct rq *rq) |
1133 | { | |
1134 | } | |
1135 | ||
b328ca18 PZ |
1136 | static inline void init_hrtick(void) |
1137 | { | |
1138 | } | |
006c75f1 | 1139 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1140 | |
c24d20db IM |
1141 | /* |
1142 | * resched_task - mark a task 'to be rescheduled now'. | |
1143 | * | |
1144 | * On UP this means the setting of the need_resched flag, on SMP it | |
1145 | * might also involve a cross-CPU call to trigger the scheduler on | |
1146 | * the target CPU. | |
1147 | */ | |
1148 | #ifdef CONFIG_SMP | |
1149 | ||
1150 | #ifndef tsk_is_polling | |
1151 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1152 | #endif | |
1153 | ||
31656519 | 1154 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1155 | { |
1156 | int cpu; | |
1157 | ||
05fa785c | 1158 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1159 | |
5ed0cec0 | 1160 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1161 | return; |
1162 | ||
5ed0cec0 | 1163 | set_tsk_need_resched(p); |
c24d20db IM |
1164 | |
1165 | cpu = task_cpu(p); | |
1166 | if (cpu == smp_processor_id()) | |
1167 | return; | |
1168 | ||
1169 | /* NEED_RESCHED must be visible before we test polling */ | |
1170 | smp_mb(); | |
1171 | if (!tsk_is_polling(p)) | |
1172 | smp_send_reschedule(cpu); | |
1173 | } | |
1174 | ||
1175 | static void resched_cpu(int cpu) | |
1176 | { | |
1177 | struct rq *rq = cpu_rq(cpu); | |
1178 | unsigned long flags; | |
1179 | ||
05fa785c | 1180 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1181 | return; |
1182 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1183 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1184 | } |
06d8308c TG |
1185 | |
1186 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1187 | /* |
1188 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1189 | * from an idle cpu. This is good for power-savings. | |
1190 | * | |
1191 | * We don't do similar optimization for completely idle system, as | |
1192 | * selecting an idle cpu will add more delays to the timers than intended | |
1193 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1194 | */ | |
1195 | int get_nohz_timer_target(void) | |
1196 | { | |
1197 | int cpu = smp_processor_id(); | |
1198 | int i; | |
1199 | struct sched_domain *sd; | |
1200 | ||
1201 | for_each_domain(cpu, sd) { | |
1202 | for_each_cpu(i, sched_domain_span(sd)) | |
1203 | if (!idle_cpu(i)) | |
1204 | return i; | |
1205 | } | |
1206 | return cpu; | |
1207 | } | |
06d8308c TG |
1208 | /* |
1209 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1210 | * idle CPU then this timer might expire before the next timer event | |
1211 | * which is scheduled to wake up that CPU. In case of a completely | |
1212 | * idle system the next event might even be infinite time into the | |
1213 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1214 | * leaves the inner idle loop so the newly added timer is taken into | |
1215 | * account when the CPU goes back to idle and evaluates the timer | |
1216 | * wheel for the next timer event. | |
1217 | */ | |
1218 | void wake_up_idle_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | ||
1222 | if (cpu == smp_processor_id()) | |
1223 | return; | |
1224 | ||
1225 | /* | |
1226 | * This is safe, as this function is called with the timer | |
1227 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1228 | * to idle and has not yet set rq->curr to idle then it will | |
1229 | * be serialized on the timer wheel base lock and take the new | |
1230 | * timer into account automatically. | |
1231 | */ | |
1232 | if (rq->curr != rq->idle) | |
1233 | return; | |
1234 | ||
1235 | /* | |
1236 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1237 | * lockless. The worst case is that the other CPU runs the | |
1238 | * idle task through an additional NOOP schedule() | |
1239 | */ | |
5ed0cec0 | 1240 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1241 | |
1242 | /* NEED_RESCHED must be visible before we test polling */ | |
1243 | smp_mb(); | |
1244 | if (!tsk_is_polling(rq->idle)) | |
1245 | smp_send_reschedule(cpu); | |
1246 | } | |
39c0cbe2 | 1247 | |
6d6bc0ad | 1248 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1249 | |
e9e9250b PZ |
1250 | static u64 sched_avg_period(void) |
1251 | { | |
1252 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1253 | } | |
1254 | ||
1255 | static void sched_avg_update(struct rq *rq) | |
1256 | { | |
1257 | s64 period = sched_avg_period(); | |
1258 | ||
1259 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1260 | /* |
1261 | * Inline assembly required to prevent the compiler | |
1262 | * optimising this loop into a divmod call. | |
1263 | * See __iter_div_u64_rem() for another example of this. | |
1264 | */ | |
1265 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1266 | rq->age_stamp += period; |
1267 | rq->rt_avg /= 2; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1272 | { | |
1273 | rq->rt_avg += rt_delta; | |
1274 | sched_avg_update(rq); | |
1275 | } | |
1276 | ||
6d6bc0ad | 1277 | #else /* !CONFIG_SMP */ |
31656519 | 1278 | static void resched_task(struct task_struct *p) |
c24d20db | 1279 | { |
05fa785c | 1280 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1281 | set_tsk_need_resched(p); |
c24d20db | 1282 | } |
e9e9250b PZ |
1283 | |
1284 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1285 | { | |
1286 | } | |
da2b71ed SS |
1287 | |
1288 | static void sched_avg_update(struct rq *rq) | |
1289 | { | |
1290 | } | |
6d6bc0ad | 1291 | #endif /* CONFIG_SMP */ |
c24d20db | 1292 | |
45bf76df IM |
1293 | #if BITS_PER_LONG == 32 |
1294 | # define WMULT_CONST (~0UL) | |
1295 | #else | |
1296 | # define WMULT_CONST (1UL << 32) | |
1297 | #endif | |
1298 | ||
1299 | #define WMULT_SHIFT 32 | |
1300 | ||
194081eb IM |
1301 | /* |
1302 | * Shift right and round: | |
1303 | */ | |
cf2ab469 | 1304 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1305 | |
a7be37ac PZ |
1306 | /* |
1307 | * delta *= weight / lw | |
1308 | */ | |
cb1c4fc9 | 1309 | static unsigned long |
45bf76df IM |
1310 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1311 | struct load_weight *lw) | |
1312 | { | |
1313 | u64 tmp; | |
1314 | ||
7a232e03 LJ |
1315 | if (!lw->inv_weight) { |
1316 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1317 | lw->inv_weight = 1; | |
1318 | else | |
1319 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1320 | / (lw->weight+1); | |
1321 | } | |
45bf76df IM |
1322 | |
1323 | tmp = (u64)delta_exec * weight; | |
1324 | /* | |
1325 | * Check whether we'd overflow the 64-bit multiplication: | |
1326 | */ | |
194081eb | 1327 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1328 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1329 | WMULT_SHIFT/2); |
1330 | else | |
cf2ab469 | 1331 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1332 | |
ecf691da | 1333 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1337 | { |
1338 | lw->weight += inc; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
1091985b | 1342 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1343 | { |
1344 | lw->weight -= dec; | |
e89996ae | 1345 | lw->inv_weight = 0; |
45bf76df IM |
1346 | } |
1347 | ||
2069dd75 PZ |
1348 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1349 | { | |
1350 | lw->weight = w; | |
1351 | lw->inv_weight = 0; | |
1352 | } | |
1353 | ||
2dd73a4f PW |
1354 | /* |
1355 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1356 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1357 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1358 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1359 | * scaled version of the new time slice allocation that they receive on time |
1360 | * slice expiry etc. | |
1361 | */ | |
1362 | ||
cce7ade8 PZ |
1363 | #define WEIGHT_IDLEPRIO 3 |
1364 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1365 | |
1366 | /* | |
1367 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1368 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1369 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1370 | * that remained on nice 0. | |
1371 | * | |
1372 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1373 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1374 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1375 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1376 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1377 | */ |
1378 | static const int prio_to_weight[40] = { | |
254753dc IM |
1379 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1380 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1381 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1382 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1383 | /* 0 */ 1024, 820, 655, 526, 423, | |
1384 | /* 5 */ 335, 272, 215, 172, 137, | |
1385 | /* 10 */ 110, 87, 70, 56, 45, | |
1386 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1387 | }; |
1388 | ||
5714d2de IM |
1389 | /* |
1390 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1391 | * | |
1392 | * In cases where the weight does not change often, we can use the | |
1393 | * precalculated inverse to speed up arithmetics by turning divisions | |
1394 | * into multiplications: | |
1395 | */ | |
dd41f596 | 1396 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1397 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1398 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1399 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1400 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1401 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1402 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1403 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1404 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1405 | }; |
2dd73a4f | 1406 | |
ef12fefa BR |
1407 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1408 | enum cpuacct_stat_index { | |
1409 | CPUACCT_STAT_USER, /* ... user mode */ | |
1410 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1411 | ||
1412 | CPUACCT_STAT_NSTATS, | |
1413 | }; | |
1414 | ||
d842de87 SV |
1415 | #ifdef CONFIG_CGROUP_CPUACCT |
1416 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1417 | static void cpuacct_update_stats(struct task_struct *tsk, |
1418 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1419 | #else |
1420 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1421 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1422 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1423 | #endif |
1424 | ||
18d95a28 PZ |
1425 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1426 | { | |
1427 | update_load_add(&rq->load, load); | |
1428 | } | |
1429 | ||
1430 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1431 | { | |
1432 | update_load_sub(&rq->load, load); | |
1433 | } | |
1434 | ||
7940ca36 | 1435 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1436 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1437 | |
1438 | /* | |
1439 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1440 | * leaving it for the final time. | |
1441 | */ | |
eb755805 | 1442 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1443 | { |
1444 | struct task_group *parent, *child; | |
eb755805 | 1445 | int ret; |
c09595f6 PZ |
1446 | |
1447 | rcu_read_lock(); | |
1448 | parent = &root_task_group; | |
1449 | down: | |
eb755805 PZ |
1450 | ret = (*down)(parent, data); |
1451 | if (ret) | |
1452 | goto out_unlock; | |
c09595f6 PZ |
1453 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1454 | parent = child; | |
1455 | goto down; | |
1456 | ||
1457 | up: | |
1458 | continue; | |
1459 | } | |
eb755805 PZ |
1460 | ret = (*up)(parent, data); |
1461 | if (ret) | |
1462 | goto out_unlock; | |
c09595f6 PZ |
1463 | |
1464 | child = parent; | |
1465 | parent = parent->parent; | |
1466 | if (parent) | |
1467 | goto up; | |
eb755805 | 1468 | out_unlock: |
c09595f6 | 1469 | rcu_read_unlock(); |
eb755805 PZ |
1470 | |
1471 | return ret; | |
c09595f6 PZ |
1472 | } |
1473 | ||
eb755805 PZ |
1474 | static int tg_nop(struct task_group *tg, void *data) |
1475 | { | |
1476 | return 0; | |
c09595f6 | 1477 | } |
eb755805 PZ |
1478 | #endif |
1479 | ||
1480 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1481 | /* Used instead of source_load when we know the type == 0 */ |
1482 | static unsigned long weighted_cpuload(const int cpu) | |
1483 | { | |
1484 | return cpu_rq(cpu)->load.weight; | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Return a low guess at the load of a migration-source cpu weighted | |
1489 | * according to the scheduling class and "nice" value. | |
1490 | * | |
1491 | * We want to under-estimate the load of migration sources, to | |
1492 | * balance conservatively. | |
1493 | */ | |
1494 | static unsigned long source_load(int cpu, int type) | |
1495 | { | |
1496 | struct rq *rq = cpu_rq(cpu); | |
1497 | unsigned long total = weighted_cpuload(cpu); | |
1498 | ||
1499 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1500 | return total; | |
1501 | ||
1502 | return min(rq->cpu_load[type-1], total); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Return a high guess at the load of a migration-target cpu weighted | |
1507 | * according to the scheduling class and "nice" value. | |
1508 | */ | |
1509 | static unsigned long target_load(int cpu, int type) | |
1510 | { | |
1511 | struct rq *rq = cpu_rq(cpu); | |
1512 | unsigned long total = weighted_cpuload(cpu); | |
1513 | ||
1514 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1515 | return total; | |
1516 | ||
1517 | return max(rq->cpu_load[type-1], total); | |
1518 | } | |
1519 | ||
ae154be1 PZ |
1520 | static unsigned long power_of(int cpu) |
1521 | { | |
e51fd5e2 | 1522 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1523 | } |
1524 | ||
eb755805 PZ |
1525 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1526 | ||
1527 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1528 | { | |
1529 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1530 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1531 | |
4cd42620 SR |
1532 | if (nr_running) |
1533 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1534 | else |
1535 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1536 | |
1537 | return rq->avg_load_per_task; | |
1538 | } | |
1539 | ||
1540 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1541 | |
c09595f6 | 1542 | /* |
c8cba857 PZ |
1543 | * Compute the cpu's hierarchical load factor for each task group. |
1544 | * This needs to be done in a top-down fashion because the load of a child | |
1545 | * group is a fraction of its parents load. | |
c09595f6 | 1546 | */ |
eb755805 | 1547 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1548 | { |
c8cba857 | 1549 | unsigned long load; |
eb755805 | 1550 | long cpu = (long)data; |
c09595f6 | 1551 | |
c8cba857 PZ |
1552 | if (!tg->parent) { |
1553 | load = cpu_rq(cpu)->load.weight; | |
1554 | } else { | |
1555 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1556 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1557 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1558 | } | |
c09595f6 | 1559 | |
c8cba857 | 1560 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1561 | |
eb755805 | 1562 | return 0; |
c09595f6 PZ |
1563 | } |
1564 | ||
eb755805 | 1565 | static void update_h_load(long cpu) |
c09595f6 | 1566 | { |
eb755805 | 1567 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1568 | } |
1569 | ||
18d95a28 PZ |
1570 | #endif |
1571 | ||
8f45e2b5 GH |
1572 | #ifdef CONFIG_PREEMPT |
1573 | ||
b78bb868 PZ |
1574 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1575 | ||
70574a99 | 1576 | /* |
8f45e2b5 GH |
1577 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1578 | * way at the expense of forcing extra atomic operations in all | |
1579 | * invocations. This assures that the double_lock is acquired using the | |
1580 | * same underlying policy as the spinlock_t on this architecture, which | |
1581 | * reduces latency compared to the unfair variant below. However, it | |
1582 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1583 | */ |
8f45e2b5 GH |
1584 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1585 | __releases(this_rq->lock) | |
1586 | __acquires(busiest->lock) | |
1587 | __acquires(this_rq->lock) | |
1588 | { | |
05fa785c | 1589 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1590 | double_rq_lock(this_rq, busiest); |
1591 | ||
1592 | return 1; | |
1593 | } | |
1594 | ||
1595 | #else | |
1596 | /* | |
1597 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1598 | * latency by eliminating extra atomic operations when the locks are | |
1599 | * already in proper order on entry. This favors lower cpu-ids and will | |
1600 | * grant the double lock to lower cpus over higher ids under contention, | |
1601 | * regardless of entry order into the function. | |
1602 | */ | |
1603 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1604 | __releases(this_rq->lock) |
1605 | __acquires(busiest->lock) | |
1606 | __acquires(this_rq->lock) | |
1607 | { | |
1608 | int ret = 0; | |
1609 | ||
05fa785c | 1610 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1611 | if (busiest < this_rq) { |
05fa785c TG |
1612 | raw_spin_unlock(&this_rq->lock); |
1613 | raw_spin_lock(&busiest->lock); | |
1614 | raw_spin_lock_nested(&this_rq->lock, | |
1615 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1616 | ret = 1; |
1617 | } else | |
05fa785c TG |
1618 | raw_spin_lock_nested(&busiest->lock, |
1619 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1620 | } |
1621 | return ret; | |
1622 | } | |
1623 | ||
8f45e2b5 GH |
1624 | #endif /* CONFIG_PREEMPT */ |
1625 | ||
1626 | /* | |
1627 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1628 | */ | |
1629 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1630 | { | |
1631 | if (unlikely(!irqs_disabled())) { | |
1632 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1633 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1634 | BUG_ON(1); |
1635 | } | |
1636 | ||
1637 | return _double_lock_balance(this_rq, busiest); | |
1638 | } | |
1639 | ||
70574a99 AD |
1640 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1641 | __releases(busiest->lock) | |
1642 | { | |
05fa785c | 1643 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1644 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1645 | } | |
1e3c88bd PZ |
1646 | |
1647 | /* | |
1648 | * double_rq_lock - safely lock two runqueues | |
1649 | * | |
1650 | * Note this does not disable interrupts like task_rq_lock, | |
1651 | * you need to do so manually before calling. | |
1652 | */ | |
1653 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1654 | __acquires(rq1->lock) | |
1655 | __acquires(rq2->lock) | |
1656 | { | |
1657 | BUG_ON(!irqs_disabled()); | |
1658 | if (rq1 == rq2) { | |
1659 | raw_spin_lock(&rq1->lock); | |
1660 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1661 | } else { | |
1662 | if (rq1 < rq2) { | |
1663 | raw_spin_lock(&rq1->lock); | |
1664 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1665 | } else { | |
1666 | raw_spin_lock(&rq2->lock); | |
1667 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1668 | } | |
1669 | } | |
1e3c88bd PZ |
1670 | } |
1671 | ||
1672 | /* | |
1673 | * double_rq_unlock - safely unlock two runqueues | |
1674 | * | |
1675 | * Note this does not restore interrupts like task_rq_unlock, | |
1676 | * you need to do so manually after calling. | |
1677 | */ | |
1678 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1679 | __releases(rq1->lock) | |
1680 | __releases(rq2->lock) | |
1681 | { | |
1682 | raw_spin_unlock(&rq1->lock); | |
1683 | if (rq1 != rq2) | |
1684 | raw_spin_unlock(&rq2->lock); | |
1685 | else | |
1686 | __release(rq2->lock); | |
1687 | } | |
1688 | ||
d95f4122 MG |
1689 | #else /* CONFIG_SMP */ |
1690 | ||
1691 | /* | |
1692 | * double_rq_lock - safely lock two runqueues | |
1693 | * | |
1694 | * Note this does not disable interrupts like task_rq_lock, | |
1695 | * you need to do so manually before calling. | |
1696 | */ | |
1697 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1698 | __acquires(rq1->lock) | |
1699 | __acquires(rq2->lock) | |
1700 | { | |
1701 | BUG_ON(!irqs_disabled()); | |
1702 | BUG_ON(rq1 != rq2); | |
1703 | raw_spin_lock(&rq1->lock); | |
1704 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1705 | } | |
1706 | ||
1707 | /* | |
1708 | * double_rq_unlock - safely unlock two runqueues | |
1709 | * | |
1710 | * Note this does not restore interrupts like task_rq_unlock, | |
1711 | * you need to do so manually after calling. | |
1712 | */ | |
1713 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1714 | __releases(rq1->lock) | |
1715 | __releases(rq2->lock) | |
1716 | { | |
1717 | BUG_ON(rq1 != rq2); | |
1718 | raw_spin_unlock(&rq1->lock); | |
1719 | __release(rq2->lock); | |
1720 | } | |
1721 | ||
18d95a28 PZ |
1722 | #endif |
1723 | ||
74f5187a | 1724 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1725 | static void update_sysctl(void); |
acb4a848 | 1726 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1727 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1728 | |
cd29fe6f PZ |
1729 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1730 | { | |
1731 | set_task_rq(p, cpu); | |
1732 | #ifdef CONFIG_SMP | |
1733 | /* | |
1734 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1735 | * successfuly executed on another CPU. We must ensure that updates of | |
1736 | * per-task data have been completed by this moment. | |
1737 | */ | |
1738 | smp_wmb(); | |
1739 | task_thread_info(p)->cpu = cpu; | |
1740 | #endif | |
1741 | } | |
dce48a84 | 1742 | |
1e3c88bd | 1743 | static const struct sched_class rt_sched_class; |
dd41f596 | 1744 | |
34f971f6 | 1745 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1746 | #define for_each_class(class) \ |
1747 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1748 | |
1e3c88bd PZ |
1749 | #include "sched_stats.h" |
1750 | ||
c09595f6 | 1751 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1752 | { |
1753 | rq->nr_running++; | |
9c217245 IM |
1754 | } |
1755 | ||
c09595f6 | 1756 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1757 | { |
1758 | rq->nr_running--; | |
9c217245 IM |
1759 | } |
1760 | ||
45bf76df IM |
1761 | static void set_load_weight(struct task_struct *p) |
1762 | { | |
dd41f596 IM |
1763 | /* |
1764 | * SCHED_IDLE tasks get minimal weight: | |
1765 | */ | |
1766 | if (p->policy == SCHED_IDLE) { | |
1767 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1768 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1769 | return; | |
1770 | } | |
71f8bd46 | 1771 | |
dd41f596 IM |
1772 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1773 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1774 | } |
1775 | ||
371fd7e7 | 1776 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1777 | { |
a64692a3 | 1778 | update_rq_clock(rq); |
dd41f596 | 1779 | sched_info_queued(p); |
371fd7e7 | 1780 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1781 | p->se.on_rq = 1; |
71f8bd46 IM |
1782 | } |
1783 | ||
371fd7e7 | 1784 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1785 | { |
a64692a3 | 1786 | update_rq_clock(rq); |
46ac22ba | 1787 | sched_info_dequeued(p); |
371fd7e7 | 1788 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1789 | p->se.on_rq = 0; |
71f8bd46 IM |
1790 | } |
1791 | ||
1e3c88bd PZ |
1792 | /* |
1793 | * activate_task - move a task to the runqueue. | |
1794 | */ | |
371fd7e7 | 1795 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1796 | { |
1797 | if (task_contributes_to_load(p)) | |
1798 | rq->nr_uninterruptible--; | |
1799 | ||
371fd7e7 | 1800 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1801 | inc_nr_running(rq); |
1802 | } | |
1803 | ||
1804 | /* | |
1805 | * deactivate_task - remove a task from the runqueue. | |
1806 | */ | |
371fd7e7 | 1807 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1808 | { |
1809 | if (task_contributes_to_load(p)) | |
1810 | rq->nr_uninterruptible++; | |
1811 | ||
371fd7e7 | 1812 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1813 | dec_nr_running(rq); |
1814 | } | |
1815 | ||
b52bfee4 VP |
1816 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1817 | ||
305e6835 VP |
1818 | /* |
1819 | * There are no locks covering percpu hardirq/softirq time. | |
1820 | * They are only modified in account_system_vtime, on corresponding CPU | |
1821 | * with interrupts disabled. So, writes are safe. | |
1822 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1823 | * This may result in other CPU reading this CPU's irq time and can | |
1824 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1825 | * or new value with a side effect of accounting a slice of irq time to wrong |
1826 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1827 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1828 | */ |
b52bfee4 VP |
1829 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1830 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1831 | ||
1832 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1833 | static int sched_clock_irqtime; | |
1834 | ||
1835 | void enable_sched_clock_irqtime(void) | |
1836 | { | |
1837 | sched_clock_irqtime = 1; | |
1838 | } | |
1839 | ||
1840 | void disable_sched_clock_irqtime(void) | |
1841 | { | |
1842 | sched_clock_irqtime = 0; | |
1843 | } | |
1844 | ||
8e92c201 PZ |
1845 | #ifndef CONFIG_64BIT |
1846 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1847 | ||
1848 | static inline void irq_time_write_begin(void) | |
1849 | { | |
1850 | __this_cpu_inc(irq_time_seq.sequence); | |
1851 | smp_wmb(); | |
1852 | } | |
1853 | ||
1854 | static inline void irq_time_write_end(void) | |
1855 | { | |
1856 | smp_wmb(); | |
1857 | __this_cpu_inc(irq_time_seq.sequence); | |
1858 | } | |
1859 | ||
1860 | static inline u64 irq_time_read(int cpu) | |
1861 | { | |
1862 | u64 irq_time; | |
1863 | unsigned seq; | |
1864 | ||
1865 | do { | |
1866 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1867 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1868 | per_cpu(cpu_hardirq_time, cpu); | |
1869 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1870 | ||
1871 | return irq_time; | |
1872 | } | |
1873 | #else /* CONFIG_64BIT */ | |
1874 | static inline void irq_time_write_begin(void) | |
1875 | { | |
1876 | } | |
1877 | ||
1878 | static inline void irq_time_write_end(void) | |
1879 | { | |
1880 | } | |
1881 | ||
1882 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1883 | { |
305e6835 VP |
1884 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1885 | } | |
8e92c201 | 1886 | #endif /* CONFIG_64BIT */ |
305e6835 | 1887 | |
fe44d621 PZ |
1888 | /* |
1889 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1890 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1891 | */ | |
b52bfee4 VP |
1892 | void account_system_vtime(struct task_struct *curr) |
1893 | { | |
1894 | unsigned long flags; | |
fe44d621 | 1895 | s64 delta; |
b52bfee4 | 1896 | int cpu; |
b52bfee4 VP |
1897 | |
1898 | if (!sched_clock_irqtime) | |
1899 | return; | |
1900 | ||
1901 | local_irq_save(flags); | |
1902 | ||
b52bfee4 | 1903 | cpu = smp_processor_id(); |
fe44d621 PZ |
1904 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1905 | __this_cpu_add(irq_start_time, delta); | |
1906 | ||
8e92c201 | 1907 | irq_time_write_begin(); |
b52bfee4 VP |
1908 | /* |
1909 | * We do not account for softirq time from ksoftirqd here. | |
1910 | * We want to continue accounting softirq time to ksoftirqd thread | |
1911 | * in that case, so as not to confuse scheduler with a special task | |
1912 | * that do not consume any time, but still wants to run. | |
1913 | */ | |
1914 | if (hardirq_count()) | |
fe44d621 | 1915 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1916 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1917 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1918 | |
8e92c201 | 1919 | irq_time_write_end(); |
b52bfee4 VP |
1920 | local_irq_restore(flags); |
1921 | } | |
b7dadc38 | 1922 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1923 | |
fe44d621 | 1924 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1925 | { |
fe44d621 PZ |
1926 | s64 irq_delta; |
1927 | ||
8e92c201 | 1928 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1929 | |
1930 | /* | |
1931 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1932 | * this case when a previous update_rq_clock() happened inside a | |
1933 | * {soft,}irq region. | |
1934 | * | |
1935 | * When this happens, we stop ->clock_task and only update the | |
1936 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1937 | * update will consume the rest. This ensures ->clock_task is | |
1938 | * monotonic. | |
1939 | * | |
1940 | * It does however cause some slight miss-attribution of {soft,}irq | |
1941 | * time, a more accurate solution would be to update the irq_time using | |
1942 | * the current rq->clock timestamp, except that would require using | |
1943 | * atomic ops. | |
1944 | */ | |
1945 | if (irq_delta > delta) | |
1946 | irq_delta = delta; | |
1947 | ||
1948 | rq->prev_irq_time += irq_delta; | |
1949 | delta -= irq_delta; | |
1950 | rq->clock_task += delta; | |
1951 | ||
1952 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1953 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1954 | } |
1955 | ||
abb74cef VP |
1956 | static int irqtime_account_hi_update(void) |
1957 | { | |
1958 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1959 | unsigned long flags; | |
1960 | u64 latest_ns; | |
1961 | int ret = 0; | |
1962 | ||
1963 | local_irq_save(flags); | |
1964 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1965 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1966 | ret = 1; | |
1967 | local_irq_restore(flags); | |
1968 | return ret; | |
1969 | } | |
1970 | ||
1971 | static int irqtime_account_si_update(void) | |
1972 | { | |
1973 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1974 | unsigned long flags; | |
1975 | u64 latest_ns; | |
1976 | int ret = 0; | |
1977 | ||
1978 | local_irq_save(flags); | |
1979 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1980 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1981 | ret = 1; | |
1982 | local_irq_restore(flags); | |
1983 | return ret; | |
1984 | } | |
1985 | ||
fe44d621 | 1986 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1987 | |
abb74cef VP |
1988 | #define sched_clock_irqtime (0) |
1989 | ||
fe44d621 | 1990 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1991 | { |
fe44d621 | 1992 | rq->clock_task += delta; |
305e6835 VP |
1993 | } |
1994 | ||
fe44d621 | 1995 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 1996 | |
1e3c88bd PZ |
1997 | #include "sched_idletask.c" |
1998 | #include "sched_fair.c" | |
1999 | #include "sched_rt.c" | |
5091faa4 | 2000 | #include "sched_autogroup.c" |
34f971f6 | 2001 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2002 | #ifdef CONFIG_SCHED_DEBUG |
2003 | # include "sched_debug.c" | |
2004 | #endif | |
2005 | ||
34f971f6 PZ |
2006 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2007 | { | |
2008 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2009 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2010 | ||
2011 | if (stop) { | |
2012 | /* | |
2013 | * Make it appear like a SCHED_FIFO task, its something | |
2014 | * userspace knows about and won't get confused about. | |
2015 | * | |
2016 | * Also, it will make PI more or less work without too | |
2017 | * much confusion -- but then, stop work should not | |
2018 | * rely on PI working anyway. | |
2019 | */ | |
2020 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2021 | ||
2022 | stop->sched_class = &stop_sched_class; | |
2023 | } | |
2024 | ||
2025 | cpu_rq(cpu)->stop = stop; | |
2026 | ||
2027 | if (old_stop) { | |
2028 | /* | |
2029 | * Reset it back to a normal scheduling class so that | |
2030 | * it can die in pieces. | |
2031 | */ | |
2032 | old_stop->sched_class = &rt_sched_class; | |
2033 | } | |
2034 | } | |
2035 | ||
14531189 | 2036 | /* |
dd41f596 | 2037 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2038 | */ |
14531189 IM |
2039 | static inline int __normal_prio(struct task_struct *p) |
2040 | { | |
dd41f596 | 2041 | return p->static_prio; |
14531189 IM |
2042 | } |
2043 | ||
b29739f9 IM |
2044 | /* |
2045 | * Calculate the expected normal priority: i.e. priority | |
2046 | * without taking RT-inheritance into account. Might be | |
2047 | * boosted by interactivity modifiers. Changes upon fork, | |
2048 | * setprio syscalls, and whenever the interactivity | |
2049 | * estimator recalculates. | |
2050 | */ | |
36c8b586 | 2051 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2052 | { |
2053 | int prio; | |
2054 | ||
e05606d3 | 2055 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2056 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2057 | else | |
2058 | prio = __normal_prio(p); | |
2059 | return prio; | |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * Calculate the current priority, i.e. the priority | |
2064 | * taken into account by the scheduler. This value might | |
2065 | * be boosted by RT tasks, or might be boosted by | |
2066 | * interactivity modifiers. Will be RT if the task got | |
2067 | * RT-boosted. If not then it returns p->normal_prio. | |
2068 | */ | |
36c8b586 | 2069 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2070 | { |
2071 | p->normal_prio = normal_prio(p); | |
2072 | /* | |
2073 | * If we are RT tasks or we were boosted to RT priority, | |
2074 | * keep the priority unchanged. Otherwise, update priority | |
2075 | * to the normal priority: | |
2076 | */ | |
2077 | if (!rt_prio(p->prio)) | |
2078 | return p->normal_prio; | |
2079 | return p->prio; | |
2080 | } | |
2081 | ||
1da177e4 LT |
2082 | /** |
2083 | * task_curr - is this task currently executing on a CPU? | |
2084 | * @p: the task in question. | |
2085 | */ | |
36c8b586 | 2086 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2087 | { |
2088 | return cpu_curr(task_cpu(p)) == p; | |
2089 | } | |
2090 | ||
cb469845 SR |
2091 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2092 | const struct sched_class *prev_class, | |
da7a735e | 2093 | int oldprio) |
cb469845 SR |
2094 | { |
2095 | if (prev_class != p->sched_class) { | |
2096 | if (prev_class->switched_from) | |
da7a735e PZ |
2097 | prev_class->switched_from(rq, p); |
2098 | p->sched_class->switched_to(rq, p); | |
2099 | } else if (oldprio != p->prio) | |
2100 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2101 | } |
2102 | ||
1e5a7405 PZ |
2103 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2104 | { | |
2105 | const struct sched_class *class; | |
2106 | ||
2107 | if (p->sched_class == rq->curr->sched_class) { | |
2108 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2109 | } else { | |
2110 | for_each_class(class) { | |
2111 | if (class == rq->curr->sched_class) | |
2112 | break; | |
2113 | if (class == p->sched_class) { | |
2114 | resched_task(rq->curr); | |
2115 | break; | |
2116 | } | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * A queue event has occurred, and we're going to schedule. In | |
2122 | * this case, we can save a useless back to back clock update. | |
2123 | */ | |
f26f9aff | 2124 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2125 | rq->skip_clock_update = 1; |
2126 | } | |
2127 | ||
1da177e4 | 2128 | #ifdef CONFIG_SMP |
cc367732 IM |
2129 | /* |
2130 | * Is this task likely cache-hot: | |
2131 | */ | |
e7693a36 | 2132 | static int |
cc367732 IM |
2133 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2134 | { | |
2135 | s64 delta; | |
2136 | ||
e6c8fba7 PZ |
2137 | if (p->sched_class != &fair_sched_class) |
2138 | return 0; | |
2139 | ||
ef8002f6 NR |
2140 | if (unlikely(p->policy == SCHED_IDLE)) |
2141 | return 0; | |
2142 | ||
f540a608 IM |
2143 | /* |
2144 | * Buddy candidates are cache hot: | |
2145 | */ | |
f685ceac | 2146 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2147 | (&p->se == cfs_rq_of(&p->se)->next || |
2148 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2149 | return 1; |
2150 | ||
6bc1665b IM |
2151 | if (sysctl_sched_migration_cost == -1) |
2152 | return 1; | |
2153 | if (sysctl_sched_migration_cost == 0) | |
2154 | return 0; | |
2155 | ||
cc367732 IM |
2156 | delta = now - p->se.exec_start; |
2157 | ||
2158 | return delta < (s64)sysctl_sched_migration_cost; | |
2159 | } | |
2160 | ||
dd41f596 | 2161 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2162 | { |
e2912009 PZ |
2163 | #ifdef CONFIG_SCHED_DEBUG |
2164 | /* | |
2165 | * We should never call set_task_cpu() on a blocked task, | |
2166 | * ttwu() will sort out the placement. | |
2167 | */ | |
077614ee PZ |
2168 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2169 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2170 | #endif |
2171 | ||
de1d7286 | 2172 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2173 | |
0c69774e PZ |
2174 | if (task_cpu(p) != new_cpu) { |
2175 | p->se.nr_migrations++; | |
2176 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2177 | } | |
dd41f596 IM |
2178 | |
2179 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2180 | } |
2181 | ||
969c7921 | 2182 | struct migration_arg { |
36c8b586 | 2183 | struct task_struct *task; |
1da177e4 | 2184 | int dest_cpu; |
70b97a7f | 2185 | }; |
1da177e4 | 2186 | |
969c7921 TH |
2187 | static int migration_cpu_stop(void *data); |
2188 | ||
1da177e4 LT |
2189 | /* |
2190 | * The task's runqueue lock must be held. | |
2191 | * Returns true if you have to wait for migration thread. | |
2192 | */ | |
b7a2b39d | 2193 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2194 | { |
1da177e4 LT |
2195 | /* |
2196 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2197 | * the next wake-up will properly place the task. |
1da177e4 | 2198 | */ |
969c7921 | 2199 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2200 | } |
2201 | ||
2202 | /* | |
2203 | * wait_task_inactive - wait for a thread to unschedule. | |
2204 | * | |
85ba2d86 RM |
2205 | * If @match_state is nonzero, it's the @p->state value just checked and |
2206 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2207 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2208 | * we return a positive number (its total switch count). If a second call | |
2209 | * a short while later returns the same number, the caller can be sure that | |
2210 | * @p has remained unscheduled the whole time. | |
2211 | * | |
1da177e4 LT |
2212 | * The caller must ensure that the task *will* unschedule sometime soon, |
2213 | * else this function might spin for a *long* time. This function can't | |
2214 | * be called with interrupts off, or it may introduce deadlock with | |
2215 | * smp_call_function() if an IPI is sent by the same process we are | |
2216 | * waiting to become inactive. | |
2217 | */ | |
85ba2d86 | 2218 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2219 | { |
2220 | unsigned long flags; | |
dd41f596 | 2221 | int running, on_rq; |
85ba2d86 | 2222 | unsigned long ncsw; |
70b97a7f | 2223 | struct rq *rq; |
1da177e4 | 2224 | |
3a5c359a AK |
2225 | for (;;) { |
2226 | /* | |
2227 | * We do the initial early heuristics without holding | |
2228 | * any task-queue locks at all. We'll only try to get | |
2229 | * the runqueue lock when things look like they will | |
2230 | * work out! | |
2231 | */ | |
2232 | rq = task_rq(p); | |
fa490cfd | 2233 | |
3a5c359a AK |
2234 | /* |
2235 | * If the task is actively running on another CPU | |
2236 | * still, just relax and busy-wait without holding | |
2237 | * any locks. | |
2238 | * | |
2239 | * NOTE! Since we don't hold any locks, it's not | |
2240 | * even sure that "rq" stays as the right runqueue! | |
2241 | * But we don't care, since "task_running()" will | |
2242 | * return false if the runqueue has changed and p | |
2243 | * is actually now running somewhere else! | |
2244 | */ | |
85ba2d86 RM |
2245 | while (task_running(rq, p)) { |
2246 | if (match_state && unlikely(p->state != match_state)) | |
2247 | return 0; | |
3a5c359a | 2248 | cpu_relax(); |
85ba2d86 | 2249 | } |
fa490cfd | 2250 | |
3a5c359a AK |
2251 | /* |
2252 | * Ok, time to look more closely! We need the rq | |
2253 | * lock now, to be *sure*. If we're wrong, we'll | |
2254 | * just go back and repeat. | |
2255 | */ | |
2256 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2257 | trace_sched_wait_task(p); |
3a5c359a AK |
2258 | running = task_running(rq, p); |
2259 | on_rq = p->se.on_rq; | |
85ba2d86 | 2260 | ncsw = 0; |
f31e11d8 | 2261 | if (!match_state || p->state == match_state) |
93dcf55f | 2262 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2263 | task_rq_unlock(rq, &flags); |
fa490cfd | 2264 | |
85ba2d86 RM |
2265 | /* |
2266 | * If it changed from the expected state, bail out now. | |
2267 | */ | |
2268 | if (unlikely(!ncsw)) | |
2269 | break; | |
2270 | ||
3a5c359a AK |
2271 | /* |
2272 | * Was it really running after all now that we | |
2273 | * checked with the proper locks actually held? | |
2274 | * | |
2275 | * Oops. Go back and try again.. | |
2276 | */ | |
2277 | if (unlikely(running)) { | |
2278 | cpu_relax(); | |
2279 | continue; | |
2280 | } | |
fa490cfd | 2281 | |
3a5c359a AK |
2282 | /* |
2283 | * It's not enough that it's not actively running, | |
2284 | * it must be off the runqueue _entirely_, and not | |
2285 | * preempted! | |
2286 | * | |
80dd99b3 | 2287 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2288 | * running right now), it's preempted, and we should |
2289 | * yield - it could be a while. | |
2290 | */ | |
2291 | if (unlikely(on_rq)) { | |
2292 | schedule_timeout_uninterruptible(1); | |
2293 | continue; | |
2294 | } | |
fa490cfd | 2295 | |
3a5c359a AK |
2296 | /* |
2297 | * Ahh, all good. It wasn't running, and it wasn't | |
2298 | * runnable, which means that it will never become | |
2299 | * running in the future either. We're all done! | |
2300 | */ | |
2301 | break; | |
2302 | } | |
85ba2d86 RM |
2303 | |
2304 | return ncsw; | |
1da177e4 LT |
2305 | } |
2306 | ||
2307 | /*** | |
2308 | * kick_process - kick a running thread to enter/exit the kernel | |
2309 | * @p: the to-be-kicked thread | |
2310 | * | |
2311 | * Cause a process which is running on another CPU to enter | |
2312 | * kernel-mode, without any delay. (to get signals handled.) | |
2313 | * | |
2314 | * NOTE: this function doesnt have to take the runqueue lock, | |
2315 | * because all it wants to ensure is that the remote task enters | |
2316 | * the kernel. If the IPI races and the task has been migrated | |
2317 | * to another CPU then no harm is done and the purpose has been | |
2318 | * achieved as well. | |
2319 | */ | |
36c8b586 | 2320 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2321 | { |
2322 | int cpu; | |
2323 | ||
2324 | preempt_disable(); | |
2325 | cpu = task_cpu(p); | |
2326 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2327 | smp_send_reschedule(cpu); | |
2328 | preempt_enable(); | |
2329 | } | |
b43e3521 | 2330 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2331 | #endif /* CONFIG_SMP */ |
1da177e4 | 2332 | |
0793a61d TG |
2333 | /** |
2334 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2335 | * @p: the task to evaluate | |
2336 | * @func: the function to be called | |
2337 | * @info: the function call argument | |
2338 | * | |
2339 | * Calls the function @func when the task is currently running. This might | |
2340 | * be on the current CPU, which just calls the function directly | |
2341 | */ | |
2342 | void task_oncpu_function_call(struct task_struct *p, | |
2343 | void (*func) (void *info), void *info) | |
2344 | { | |
2345 | int cpu; | |
2346 | ||
2347 | preempt_disable(); | |
2348 | cpu = task_cpu(p); | |
2349 | if (task_curr(p)) | |
2350 | smp_call_function_single(cpu, func, info, 1); | |
2351 | preempt_enable(); | |
2352 | } | |
2353 | ||
970b13ba | 2354 | #ifdef CONFIG_SMP |
30da688e ON |
2355 | /* |
2356 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2357 | */ | |
5da9a0fb PZ |
2358 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2359 | { | |
2360 | int dest_cpu; | |
2361 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2362 | ||
2363 | /* Look for allowed, online CPU in same node. */ | |
2364 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2365 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2366 | return dest_cpu; | |
2367 | ||
2368 | /* Any allowed, online CPU? */ | |
2369 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2370 | if (dest_cpu < nr_cpu_ids) | |
2371 | return dest_cpu; | |
2372 | ||
2373 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2374 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2375 | /* | |
2376 | * Don't tell them about moving exiting tasks or | |
2377 | * kernel threads (both mm NULL), since they never | |
2378 | * leave kernel. | |
2379 | */ | |
2380 | if (p->mm && printk_ratelimit()) { | |
2381 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2382 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2383 | } |
2384 | ||
2385 | return dest_cpu; | |
2386 | } | |
2387 | ||
e2912009 | 2388 | /* |
30da688e | 2389 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2390 | */ |
970b13ba | 2391 | static inline |
0017d735 | 2392 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2393 | { |
0017d735 | 2394 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2395 | |
2396 | /* | |
2397 | * In order not to call set_task_cpu() on a blocking task we need | |
2398 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2399 | * cpu. | |
2400 | * | |
2401 | * Since this is common to all placement strategies, this lives here. | |
2402 | * | |
2403 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2404 | * not worry about this generic constraint ] | |
2405 | */ | |
2406 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2407 | !cpu_online(cpu))) |
5da9a0fb | 2408 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2409 | |
2410 | return cpu; | |
970b13ba | 2411 | } |
09a40af5 MG |
2412 | |
2413 | static void update_avg(u64 *avg, u64 sample) | |
2414 | { | |
2415 | s64 diff = sample - *avg; | |
2416 | *avg += diff >> 3; | |
2417 | } | |
970b13ba PZ |
2418 | #endif |
2419 | ||
9ed3811a TH |
2420 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2421 | bool is_sync, bool is_migrate, bool is_local, | |
2422 | unsigned long en_flags) | |
2423 | { | |
2424 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2425 | if (is_sync) | |
2426 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2427 | if (is_migrate) | |
2428 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2429 | if (is_local) | |
2430 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2431 | else | |
2432 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2433 | ||
2434 | activate_task(rq, p, en_flags); | |
2435 | } | |
2436 | ||
2437 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2438 | int wake_flags, bool success) | |
2439 | { | |
2440 | trace_sched_wakeup(p, success); | |
2441 | check_preempt_curr(rq, p, wake_flags); | |
2442 | ||
2443 | p->state = TASK_RUNNING; | |
2444 | #ifdef CONFIG_SMP | |
2445 | if (p->sched_class->task_woken) | |
2446 | p->sched_class->task_woken(rq, p); | |
2447 | ||
2448 | if (unlikely(rq->idle_stamp)) { | |
2449 | u64 delta = rq->clock - rq->idle_stamp; | |
2450 | u64 max = 2*sysctl_sched_migration_cost; | |
2451 | ||
2452 | if (delta > max) | |
2453 | rq->avg_idle = max; | |
2454 | else | |
2455 | update_avg(&rq->avg_idle, delta); | |
2456 | rq->idle_stamp = 0; | |
2457 | } | |
2458 | #endif | |
21aa9af0 TH |
2459 | /* if a worker is waking up, notify workqueue */ |
2460 | if ((p->flags & PF_WQ_WORKER) && success) | |
2461 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2462 | } |
2463 | ||
2464 | /** | |
1da177e4 | 2465 | * try_to_wake_up - wake up a thread |
9ed3811a | 2466 | * @p: the thread to be awakened |
1da177e4 | 2467 | * @state: the mask of task states that can be woken |
9ed3811a | 2468 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2469 | * |
2470 | * Put it on the run-queue if it's not already there. The "current" | |
2471 | * thread is always on the run-queue (except when the actual | |
2472 | * re-schedule is in progress), and as such you're allowed to do | |
2473 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2474 | * runnable without the overhead of this. | |
2475 | * | |
9ed3811a TH |
2476 | * Returns %true if @p was woken up, %false if it was already running |
2477 | * or @state didn't match @p's state. | |
1da177e4 | 2478 | */ |
7d478721 PZ |
2479 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2480 | int wake_flags) | |
1da177e4 | 2481 | { |
cc367732 | 2482 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2483 | unsigned long flags; |
371fd7e7 | 2484 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2485 | struct rq *rq; |
1da177e4 | 2486 | |
e9c84311 | 2487 | this_cpu = get_cpu(); |
2398f2c6 | 2488 | |
04e2f174 | 2489 | smp_wmb(); |
ab3b3aa5 | 2490 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2491 | if (!(p->state & state)) |
1da177e4 LT |
2492 | goto out; |
2493 | ||
dd41f596 | 2494 | if (p->se.on_rq) |
1da177e4 LT |
2495 | goto out_running; |
2496 | ||
2497 | cpu = task_cpu(p); | |
cc367732 | 2498 | orig_cpu = cpu; |
1da177e4 LT |
2499 | |
2500 | #ifdef CONFIG_SMP | |
2501 | if (unlikely(task_running(rq, p))) | |
2502 | goto out_activate; | |
2503 | ||
e9c84311 PZ |
2504 | /* |
2505 | * In order to handle concurrent wakeups and release the rq->lock | |
2506 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2507 | * |
2508 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2509 | */ |
cc87f76a PZ |
2510 | if (task_contributes_to_load(p)) { |
2511 | if (likely(cpu_online(orig_cpu))) | |
2512 | rq->nr_uninterruptible--; | |
2513 | else | |
2514 | this_rq()->nr_uninterruptible--; | |
2515 | } | |
e9c84311 | 2516 | p->state = TASK_WAKING; |
efbbd05a | 2517 | |
371fd7e7 | 2518 | if (p->sched_class->task_waking) { |
efbbd05a | 2519 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2520 | en_flags |= ENQUEUE_WAKING; |
2521 | } | |
efbbd05a | 2522 | |
0017d735 PZ |
2523 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2524 | if (cpu != orig_cpu) | |
5d2f5a61 | 2525 | set_task_cpu(p, cpu); |
0017d735 | 2526 | __task_rq_unlock(rq); |
ab19cb23 | 2527 | |
0970d299 PZ |
2528 | rq = cpu_rq(cpu); |
2529 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2530 | |
0970d299 PZ |
2531 | /* |
2532 | * We migrated the task without holding either rq->lock, however | |
2533 | * since the task is not on the task list itself, nobody else | |
2534 | * will try and migrate the task, hence the rq should match the | |
2535 | * cpu we just moved it to. | |
2536 | */ | |
2537 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2538 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2539 | |
e7693a36 GH |
2540 | #ifdef CONFIG_SCHEDSTATS |
2541 | schedstat_inc(rq, ttwu_count); | |
2542 | if (cpu == this_cpu) | |
2543 | schedstat_inc(rq, ttwu_local); | |
2544 | else { | |
2545 | struct sched_domain *sd; | |
2546 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2547 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2548 | schedstat_inc(sd, ttwu_wake_remote); |
2549 | break; | |
2550 | } | |
2551 | } | |
2552 | } | |
6d6bc0ad | 2553 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2554 | |
1da177e4 LT |
2555 | out_activate: |
2556 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2557 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2558 | cpu == this_cpu, en_flags); | |
1da177e4 | 2559 | success = 1; |
1da177e4 | 2560 | out_running: |
9ed3811a | 2561 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2562 | out: |
2563 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2564 | put_cpu(); |
1da177e4 LT |
2565 | |
2566 | return success; | |
2567 | } | |
2568 | ||
21aa9af0 TH |
2569 | /** |
2570 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2571 | * @p: the thread to be awakened | |
2572 | * | |
b595076a | 2573 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 TH |
2574 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2575 | * the current task. this_rq() stays locked over invocation. | |
2576 | */ | |
2577 | static void try_to_wake_up_local(struct task_struct *p) | |
2578 | { | |
2579 | struct rq *rq = task_rq(p); | |
2580 | bool success = false; | |
2581 | ||
2582 | BUG_ON(rq != this_rq()); | |
2583 | BUG_ON(p == current); | |
2584 | lockdep_assert_held(&rq->lock); | |
2585 | ||
2586 | if (!(p->state & TASK_NORMAL)) | |
2587 | return; | |
2588 | ||
2589 | if (!p->se.on_rq) { | |
2590 | if (likely(!task_running(rq, p))) { | |
2591 | schedstat_inc(rq, ttwu_count); | |
2592 | schedstat_inc(rq, ttwu_local); | |
2593 | } | |
2594 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2595 | success = true; | |
2596 | } | |
2597 | ttwu_post_activation(p, rq, 0, success); | |
2598 | } | |
2599 | ||
50fa610a DH |
2600 | /** |
2601 | * wake_up_process - Wake up a specific process | |
2602 | * @p: The process to be woken up. | |
2603 | * | |
2604 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2605 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2606 | * running. | |
2607 | * | |
2608 | * It may be assumed that this function implies a write memory barrier before | |
2609 | * changing the task state if and only if any tasks are woken up. | |
2610 | */ | |
7ad5b3a5 | 2611 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2612 | { |
d9514f6c | 2613 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2614 | } |
1da177e4 LT |
2615 | EXPORT_SYMBOL(wake_up_process); |
2616 | ||
7ad5b3a5 | 2617 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2618 | { |
2619 | return try_to_wake_up(p, state, 0); | |
2620 | } | |
2621 | ||
1da177e4 LT |
2622 | /* |
2623 | * Perform scheduler related setup for a newly forked process p. | |
2624 | * p is forked by current. | |
dd41f596 IM |
2625 | * |
2626 | * __sched_fork() is basic setup used by init_idle() too: | |
2627 | */ | |
2628 | static void __sched_fork(struct task_struct *p) | |
2629 | { | |
dd41f596 IM |
2630 | p->se.exec_start = 0; |
2631 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2632 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2633 | p->se.nr_migrations = 0; |
da7a735e | 2634 | p->se.vruntime = 0; |
6cfb0d5d IM |
2635 | |
2636 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2637 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2638 | #endif |
476d139c | 2639 | |
fa717060 | 2640 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2641 | p->se.on_rq = 0; |
4a55bd5e | 2642 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2643 | |
e107be36 AK |
2644 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2645 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2646 | #endif | |
dd41f596 IM |
2647 | } |
2648 | ||
2649 | /* | |
2650 | * fork()/clone()-time setup: | |
2651 | */ | |
2652 | void sched_fork(struct task_struct *p, int clone_flags) | |
2653 | { | |
2654 | int cpu = get_cpu(); | |
2655 | ||
2656 | __sched_fork(p); | |
06b83b5f | 2657 | /* |
0017d735 | 2658 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2659 | * nobody will actually run it, and a signal or other external |
2660 | * event cannot wake it up and insert it on the runqueue either. | |
2661 | */ | |
0017d735 | 2662 | p->state = TASK_RUNNING; |
dd41f596 | 2663 | |
b9dc29e7 MG |
2664 | /* |
2665 | * Revert to default priority/policy on fork if requested. | |
2666 | */ | |
2667 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2668 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2669 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2670 | p->normal_prio = p->static_prio; |
2671 | } | |
b9dc29e7 | 2672 | |
6c697bdf MG |
2673 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2674 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2675 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2676 | set_load_weight(p); |
2677 | } | |
2678 | ||
b9dc29e7 MG |
2679 | /* |
2680 | * We don't need the reset flag anymore after the fork. It has | |
2681 | * fulfilled its duty: | |
2682 | */ | |
2683 | p->sched_reset_on_fork = 0; | |
2684 | } | |
ca94c442 | 2685 | |
f83f9ac2 PW |
2686 | /* |
2687 | * Make sure we do not leak PI boosting priority to the child. | |
2688 | */ | |
2689 | p->prio = current->normal_prio; | |
2690 | ||
2ddbf952 HS |
2691 | if (!rt_prio(p->prio)) |
2692 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2693 | |
cd29fe6f PZ |
2694 | if (p->sched_class->task_fork) |
2695 | p->sched_class->task_fork(p); | |
2696 | ||
86951599 PZ |
2697 | /* |
2698 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2699 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2700 | * is ran before sched_fork(). | |
2701 | * | |
2702 | * Silence PROVE_RCU. | |
2703 | */ | |
2704 | rcu_read_lock(); | |
5f3edc1b | 2705 | set_task_cpu(p, cpu); |
86951599 | 2706 | rcu_read_unlock(); |
5f3edc1b | 2707 | |
52f17b6c | 2708 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2709 | if (likely(sched_info_on())) |
52f17b6c | 2710 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2711 | #endif |
d6077cb8 | 2712 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2713 | p->oncpu = 0; |
2714 | #endif | |
1da177e4 | 2715 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2716 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2717 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2718 | #endif |
806c09a7 | 2719 | #ifdef CONFIG_SMP |
917b627d | 2720 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2721 | #endif |
917b627d | 2722 | |
476d139c | 2723 | put_cpu(); |
1da177e4 LT |
2724 | } |
2725 | ||
2726 | /* | |
2727 | * wake_up_new_task - wake up a newly created task for the first time. | |
2728 | * | |
2729 | * This function will do some initial scheduler statistics housekeeping | |
2730 | * that must be done for every newly created context, then puts the task | |
2731 | * on the runqueue and wakes it. | |
2732 | */ | |
7ad5b3a5 | 2733 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2734 | { |
2735 | unsigned long flags; | |
dd41f596 | 2736 | struct rq *rq; |
c890692b | 2737 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2738 | |
2739 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2740 | rq = task_rq_lock(p, &flags); |
2741 | p->state = TASK_WAKING; | |
2742 | ||
fabf318e PZ |
2743 | /* |
2744 | * Fork balancing, do it here and not earlier because: | |
2745 | * - cpus_allowed can change in the fork path | |
2746 | * - any previously selected cpu might disappear through hotplug | |
2747 | * | |
0017d735 PZ |
2748 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2749 | * without people poking at ->cpus_allowed. | |
fabf318e | 2750 | */ |
0017d735 | 2751 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2752 | set_task_cpu(p, cpu); |
1da177e4 | 2753 | |
06b83b5f | 2754 | p->state = TASK_RUNNING; |
0017d735 PZ |
2755 | task_rq_unlock(rq, &flags); |
2756 | #endif | |
2757 | ||
2758 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2759 | activate_task(rq, p, 0); |
27a9da65 | 2760 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2761 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2762 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2763 | if (p->sched_class->task_woken) |
2764 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2765 | #endif |
dd41f596 | 2766 | task_rq_unlock(rq, &flags); |
fabf318e | 2767 | put_cpu(); |
1da177e4 LT |
2768 | } |
2769 | ||
e107be36 AK |
2770 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2771 | ||
2772 | /** | |
80dd99b3 | 2773 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2774 | * @notifier: notifier struct to register |
e107be36 AK |
2775 | */ |
2776 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2777 | { | |
2778 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2779 | } | |
2780 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2781 | ||
2782 | /** | |
2783 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2784 | * @notifier: notifier struct to unregister |
e107be36 AK |
2785 | * |
2786 | * This is safe to call from within a preemption notifier. | |
2787 | */ | |
2788 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2789 | { | |
2790 | hlist_del(¬ifier->link); | |
2791 | } | |
2792 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2793 | ||
2794 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2795 | { | |
2796 | struct preempt_notifier *notifier; | |
2797 | struct hlist_node *node; | |
2798 | ||
2799 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2800 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2801 | } | |
2802 | ||
2803 | static void | |
2804 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2805 | struct task_struct *next) | |
2806 | { | |
2807 | struct preempt_notifier *notifier; | |
2808 | struct hlist_node *node; | |
2809 | ||
2810 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2811 | notifier->ops->sched_out(notifier, next); | |
2812 | } | |
2813 | ||
6d6bc0ad | 2814 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2815 | |
2816 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2817 | { | |
2818 | } | |
2819 | ||
2820 | static void | |
2821 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2822 | struct task_struct *next) | |
2823 | { | |
2824 | } | |
2825 | ||
6d6bc0ad | 2826 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2827 | |
4866cde0 NP |
2828 | /** |
2829 | * prepare_task_switch - prepare to switch tasks | |
2830 | * @rq: the runqueue preparing to switch | |
421cee29 | 2831 | * @prev: the current task that is being switched out |
4866cde0 NP |
2832 | * @next: the task we are going to switch to. |
2833 | * | |
2834 | * This is called with the rq lock held and interrupts off. It must | |
2835 | * be paired with a subsequent finish_task_switch after the context | |
2836 | * switch. | |
2837 | * | |
2838 | * prepare_task_switch sets up locking and calls architecture specific | |
2839 | * hooks. | |
2840 | */ | |
e107be36 AK |
2841 | static inline void |
2842 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2843 | struct task_struct *next) | |
4866cde0 | 2844 | { |
e107be36 | 2845 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2846 | prepare_lock_switch(rq, next); |
2847 | prepare_arch_switch(next); | |
2848 | } | |
2849 | ||
1da177e4 LT |
2850 | /** |
2851 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2852 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2853 | * @prev: the thread we just switched away from. |
2854 | * | |
4866cde0 NP |
2855 | * finish_task_switch must be called after the context switch, paired |
2856 | * with a prepare_task_switch call before the context switch. | |
2857 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2858 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2859 | * |
2860 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2861 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2862 | * with the lock held can cause deadlocks; see schedule() for |
2863 | * details.) | |
2864 | */ | |
a9957449 | 2865 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2866 | __releases(rq->lock) |
2867 | { | |
1da177e4 | 2868 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2869 | long prev_state; |
1da177e4 LT |
2870 | |
2871 | rq->prev_mm = NULL; | |
2872 | ||
2873 | /* | |
2874 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2875 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2876 | * schedule one last time. The schedule call will never return, and |
2877 | * the scheduled task must drop that reference. | |
c394cc9f | 2878 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2879 | * still held, otherwise prev could be scheduled on another cpu, die |
2880 | * there before we look at prev->state, and then the reference would | |
2881 | * be dropped twice. | |
2882 | * Manfred Spraul <manfred@colorfullife.com> | |
2883 | */ | |
55a101f8 | 2884 | prev_state = prev->state; |
4866cde0 | 2885 | finish_arch_switch(prev); |
8381f65d JI |
2886 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2887 | local_irq_disable(); | |
2888 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2889 | perf_event_task_sched_in(current); |
8381f65d JI |
2890 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2891 | local_irq_enable(); | |
2892 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2893 | finish_lock_switch(rq, prev); |
e8fa1362 | 2894 | |
e107be36 | 2895 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2896 | if (mm) |
2897 | mmdrop(mm); | |
c394cc9f | 2898 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2899 | /* |
2900 | * Remove function-return probe instances associated with this | |
2901 | * task and put them back on the free list. | |
9761eea8 | 2902 | */ |
c6fd91f0 | 2903 | kprobe_flush_task(prev); |
1da177e4 | 2904 | put_task_struct(prev); |
c6fd91f0 | 2905 | } |
1da177e4 LT |
2906 | } |
2907 | ||
3f029d3c GH |
2908 | #ifdef CONFIG_SMP |
2909 | ||
2910 | /* assumes rq->lock is held */ | |
2911 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2912 | { | |
2913 | if (prev->sched_class->pre_schedule) | |
2914 | prev->sched_class->pre_schedule(rq, prev); | |
2915 | } | |
2916 | ||
2917 | /* rq->lock is NOT held, but preemption is disabled */ | |
2918 | static inline void post_schedule(struct rq *rq) | |
2919 | { | |
2920 | if (rq->post_schedule) { | |
2921 | unsigned long flags; | |
2922 | ||
05fa785c | 2923 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2924 | if (rq->curr->sched_class->post_schedule) |
2925 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2926 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2927 | |
2928 | rq->post_schedule = 0; | |
2929 | } | |
2930 | } | |
2931 | ||
2932 | #else | |
da19ab51 | 2933 | |
3f029d3c GH |
2934 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2935 | { | |
2936 | } | |
2937 | ||
2938 | static inline void post_schedule(struct rq *rq) | |
2939 | { | |
1da177e4 LT |
2940 | } |
2941 | ||
3f029d3c GH |
2942 | #endif |
2943 | ||
1da177e4 LT |
2944 | /** |
2945 | * schedule_tail - first thing a freshly forked thread must call. | |
2946 | * @prev: the thread we just switched away from. | |
2947 | */ | |
36c8b586 | 2948 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2949 | __releases(rq->lock) |
2950 | { | |
70b97a7f IM |
2951 | struct rq *rq = this_rq(); |
2952 | ||
4866cde0 | 2953 | finish_task_switch(rq, prev); |
da19ab51 | 2954 | |
3f029d3c GH |
2955 | /* |
2956 | * FIXME: do we need to worry about rq being invalidated by the | |
2957 | * task_switch? | |
2958 | */ | |
2959 | post_schedule(rq); | |
70b97a7f | 2960 | |
4866cde0 NP |
2961 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2962 | /* In this case, finish_task_switch does not reenable preemption */ | |
2963 | preempt_enable(); | |
2964 | #endif | |
1da177e4 | 2965 | if (current->set_child_tid) |
b488893a | 2966 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2967 | } |
2968 | ||
2969 | /* | |
2970 | * context_switch - switch to the new MM and the new | |
2971 | * thread's register state. | |
2972 | */ | |
dd41f596 | 2973 | static inline void |
70b97a7f | 2974 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2975 | struct task_struct *next) |
1da177e4 | 2976 | { |
dd41f596 | 2977 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2978 | |
e107be36 | 2979 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2980 | trace_sched_switch(prev, next); |
dd41f596 IM |
2981 | mm = next->mm; |
2982 | oldmm = prev->active_mm; | |
9226d125 ZA |
2983 | /* |
2984 | * For paravirt, this is coupled with an exit in switch_to to | |
2985 | * combine the page table reload and the switch backend into | |
2986 | * one hypercall. | |
2987 | */ | |
224101ed | 2988 | arch_start_context_switch(prev); |
9226d125 | 2989 | |
31915ab4 | 2990 | if (!mm) { |
1da177e4 LT |
2991 | next->active_mm = oldmm; |
2992 | atomic_inc(&oldmm->mm_count); | |
2993 | enter_lazy_tlb(oldmm, next); | |
2994 | } else | |
2995 | switch_mm(oldmm, mm, next); | |
2996 | ||
31915ab4 | 2997 | if (!prev->mm) { |
1da177e4 | 2998 | prev->active_mm = NULL; |
1da177e4 LT |
2999 | rq->prev_mm = oldmm; |
3000 | } | |
3a5f5e48 IM |
3001 | /* |
3002 | * Since the runqueue lock will be released by the next | |
3003 | * task (which is an invalid locking op but in the case | |
3004 | * of the scheduler it's an obvious special-case), so we | |
3005 | * do an early lockdep release here: | |
3006 | */ | |
3007 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3008 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3009 | #endif |
1da177e4 LT |
3010 | |
3011 | /* Here we just switch the register state and the stack. */ | |
3012 | switch_to(prev, next, prev); | |
3013 | ||
dd41f596 IM |
3014 | barrier(); |
3015 | /* | |
3016 | * this_rq must be evaluated again because prev may have moved | |
3017 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3018 | * frame will be invalid. | |
3019 | */ | |
3020 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3021 | } |
3022 | ||
3023 | /* | |
3024 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3025 | * | |
3026 | * externally visible scheduler statistics: current number of runnable | |
3027 | * threads, current number of uninterruptible-sleeping threads, total | |
3028 | * number of context switches performed since bootup. | |
3029 | */ | |
3030 | unsigned long nr_running(void) | |
3031 | { | |
3032 | unsigned long i, sum = 0; | |
3033 | ||
3034 | for_each_online_cpu(i) | |
3035 | sum += cpu_rq(i)->nr_running; | |
3036 | ||
3037 | return sum; | |
f711f609 | 3038 | } |
1da177e4 LT |
3039 | |
3040 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3041 | { |
1da177e4 | 3042 | unsigned long i, sum = 0; |
f711f609 | 3043 | |
0a945022 | 3044 | for_each_possible_cpu(i) |
1da177e4 | 3045 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3046 | |
3047 | /* | |
1da177e4 LT |
3048 | * Since we read the counters lockless, it might be slightly |
3049 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3050 | */ |
1da177e4 LT |
3051 | if (unlikely((long)sum < 0)) |
3052 | sum = 0; | |
f711f609 | 3053 | |
1da177e4 | 3054 | return sum; |
f711f609 | 3055 | } |
f711f609 | 3056 | |
1da177e4 | 3057 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3058 | { |
cc94abfc SR |
3059 | int i; |
3060 | unsigned long long sum = 0; | |
46cb4b7c | 3061 | |
0a945022 | 3062 | for_each_possible_cpu(i) |
1da177e4 | 3063 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3064 | |
1da177e4 LT |
3065 | return sum; |
3066 | } | |
483b4ee6 | 3067 | |
1da177e4 LT |
3068 | unsigned long nr_iowait(void) |
3069 | { | |
3070 | unsigned long i, sum = 0; | |
483b4ee6 | 3071 | |
0a945022 | 3072 | for_each_possible_cpu(i) |
1da177e4 | 3073 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3074 | |
1da177e4 LT |
3075 | return sum; |
3076 | } | |
483b4ee6 | 3077 | |
8c215bd3 | 3078 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3079 | { |
8c215bd3 | 3080 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3081 | return atomic_read(&this->nr_iowait); |
3082 | } | |
46cb4b7c | 3083 | |
69d25870 AV |
3084 | unsigned long this_cpu_load(void) |
3085 | { | |
3086 | struct rq *this = this_rq(); | |
3087 | return this->cpu_load[0]; | |
3088 | } | |
e790fb0b | 3089 | |
46cb4b7c | 3090 | |
dce48a84 TG |
3091 | /* Variables and functions for calc_load */ |
3092 | static atomic_long_t calc_load_tasks; | |
3093 | static unsigned long calc_load_update; | |
3094 | unsigned long avenrun[3]; | |
3095 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3096 | |
74f5187a PZ |
3097 | static long calc_load_fold_active(struct rq *this_rq) |
3098 | { | |
3099 | long nr_active, delta = 0; | |
3100 | ||
3101 | nr_active = this_rq->nr_running; | |
3102 | nr_active += (long) this_rq->nr_uninterruptible; | |
3103 | ||
3104 | if (nr_active != this_rq->calc_load_active) { | |
3105 | delta = nr_active - this_rq->calc_load_active; | |
3106 | this_rq->calc_load_active = nr_active; | |
3107 | } | |
3108 | ||
3109 | return delta; | |
3110 | } | |
3111 | ||
0f004f5a PZ |
3112 | static unsigned long |
3113 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3114 | { | |
3115 | load *= exp; | |
3116 | load += active * (FIXED_1 - exp); | |
3117 | load += 1UL << (FSHIFT - 1); | |
3118 | return load >> FSHIFT; | |
3119 | } | |
3120 | ||
74f5187a PZ |
3121 | #ifdef CONFIG_NO_HZ |
3122 | /* | |
3123 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3124 | * | |
3125 | * When making the ILB scale, we should try to pull this in as well. | |
3126 | */ | |
3127 | static atomic_long_t calc_load_tasks_idle; | |
3128 | ||
3129 | static void calc_load_account_idle(struct rq *this_rq) | |
3130 | { | |
3131 | long delta; | |
3132 | ||
3133 | delta = calc_load_fold_active(this_rq); | |
3134 | if (delta) | |
3135 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3136 | } | |
3137 | ||
3138 | static long calc_load_fold_idle(void) | |
3139 | { | |
3140 | long delta = 0; | |
3141 | ||
3142 | /* | |
3143 | * Its got a race, we don't care... | |
3144 | */ | |
3145 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3146 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3147 | ||
3148 | return delta; | |
3149 | } | |
0f004f5a PZ |
3150 | |
3151 | /** | |
3152 | * fixed_power_int - compute: x^n, in O(log n) time | |
3153 | * | |
3154 | * @x: base of the power | |
3155 | * @frac_bits: fractional bits of @x | |
3156 | * @n: power to raise @x to. | |
3157 | * | |
3158 | * By exploiting the relation between the definition of the natural power | |
3159 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3160 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3161 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3162 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3163 | * of course trivially computable in O(log_2 n), the length of our binary | |
3164 | * vector. | |
3165 | */ | |
3166 | static unsigned long | |
3167 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3168 | { | |
3169 | unsigned long result = 1UL << frac_bits; | |
3170 | ||
3171 | if (n) for (;;) { | |
3172 | if (n & 1) { | |
3173 | result *= x; | |
3174 | result += 1UL << (frac_bits - 1); | |
3175 | result >>= frac_bits; | |
3176 | } | |
3177 | n >>= 1; | |
3178 | if (!n) | |
3179 | break; | |
3180 | x *= x; | |
3181 | x += 1UL << (frac_bits - 1); | |
3182 | x >>= frac_bits; | |
3183 | } | |
3184 | ||
3185 | return result; | |
3186 | } | |
3187 | ||
3188 | /* | |
3189 | * a1 = a0 * e + a * (1 - e) | |
3190 | * | |
3191 | * a2 = a1 * e + a * (1 - e) | |
3192 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3193 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3194 | * | |
3195 | * a3 = a2 * e + a * (1 - e) | |
3196 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3197 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3198 | * | |
3199 | * ... | |
3200 | * | |
3201 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3202 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3203 | * = a0 * e^n + a * (1 - e^n) | |
3204 | * | |
3205 | * [1] application of the geometric series: | |
3206 | * | |
3207 | * n 1 - x^(n+1) | |
3208 | * S_n := \Sum x^i = ------------- | |
3209 | * i=0 1 - x | |
3210 | */ | |
3211 | static unsigned long | |
3212 | calc_load_n(unsigned long load, unsigned long exp, | |
3213 | unsigned long active, unsigned int n) | |
3214 | { | |
3215 | ||
3216 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3217 | } | |
3218 | ||
3219 | /* | |
3220 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3221 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3222 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3223 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3224 | * | |
3225 | * Once we've updated the global active value, we need to apply the exponential | |
3226 | * weights adjusted to the number of cycles missed. | |
3227 | */ | |
3228 | static void calc_global_nohz(unsigned long ticks) | |
3229 | { | |
3230 | long delta, active, n; | |
3231 | ||
3232 | if (time_before(jiffies, calc_load_update)) | |
3233 | return; | |
3234 | ||
3235 | /* | |
3236 | * If we crossed a calc_load_update boundary, make sure to fold | |
3237 | * any pending idle changes, the respective CPUs might have | |
3238 | * missed the tick driven calc_load_account_active() update | |
3239 | * due to NO_HZ. | |
3240 | */ | |
3241 | delta = calc_load_fold_idle(); | |
3242 | if (delta) | |
3243 | atomic_long_add(delta, &calc_load_tasks); | |
3244 | ||
3245 | /* | |
3246 | * If we were idle for multiple load cycles, apply them. | |
3247 | */ | |
3248 | if (ticks >= LOAD_FREQ) { | |
3249 | n = ticks / LOAD_FREQ; | |
3250 | ||
3251 | active = atomic_long_read(&calc_load_tasks); | |
3252 | active = active > 0 ? active * FIXED_1 : 0; | |
3253 | ||
3254 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3255 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3256 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3257 | ||
3258 | calc_load_update += n * LOAD_FREQ; | |
3259 | } | |
3260 | ||
3261 | /* | |
3262 | * Its possible the remainder of the above division also crosses | |
3263 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3264 | * which comes after this will take care of that. | |
3265 | * | |
3266 | * Consider us being 11 ticks before a cycle completion, and us | |
3267 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3268 | * age us 4 cycles, and the test in calc_global_load() will | |
3269 | * pick up the final one. | |
3270 | */ | |
3271 | } | |
74f5187a PZ |
3272 | #else |
3273 | static void calc_load_account_idle(struct rq *this_rq) | |
3274 | { | |
3275 | } | |
3276 | ||
3277 | static inline long calc_load_fold_idle(void) | |
3278 | { | |
3279 | return 0; | |
3280 | } | |
0f004f5a PZ |
3281 | |
3282 | static void calc_global_nohz(unsigned long ticks) | |
3283 | { | |
3284 | } | |
74f5187a PZ |
3285 | #endif |
3286 | ||
2d02494f TG |
3287 | /** |
3288 | * get_avenrun - get the load average array | |
3289 | * @loads: pointer to dest load array | |
3290 | * @offset: offset to add | |
3291 | * @shift: shift count to shift the result left | |
3292 | * | |
3293 | * These values are estimates at best, so no need for locking. | |
3294 | */ | |
3295 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3296 | { | |
3297 | loads[0] = (avenrun[0] + offset) << shift; | |
3298 | loads[1] = (avenrun[1] + offset) << shift; | |
3299 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3300 | } |
46cb4b7c | 3301 | |
46cb4b7c | 3302 | /* |
dce48a84 TG |
3303 | * calc_load - update the avenrun load estimates 10 ticks after the |
3304 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3305 | */ |
0f004f5a | 3306 | void calc_global_load(unsigned long ticks) |
7835b98b | 3307 | { |
dce48a84 | 3308 | long active; |
1da177e4 | 3309 | |
0f004f5a PZ |
3310 | calc_global_nohz(ticks); |
3311 | ||
3312 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3313 | return; |
1da177e4 | 3314 | |
dce48a84 TG |
3315 | active = atomic_long_read(&calc_load_tasks); |
3316 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3317 | |
dce48a84 TG |
3318 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3319 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3320 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3321 | |
dce48a84 TG |
3322 | calc_load_update += LOAD_FREQ; |
3323 | } | |
1da177e4 | 3324 | |
dce48a84 | 3325 | /* |
74f5187a PZ |
3326 | * Called from update_cpu_load() to periodically update this CPU's |
3327 | * active count. | |
dce48a84 TG |
3328 | */ |
3329 | static void calc_load_account_active(struct rq *this_rq) | |
3330 | { | |
74f5187a | 3331 | long delta; |
08c183f3 | 3332 | |
74f5187a PZ |
3333 | if (time_before(jiffies, this_rq->calc_load_update)) |
3334 | return; | |
783609c6 | 3335 | |
74f5187a PZ |
3336 | delta = calc_load_fold_active(this_rq); |
3337 | delta += calc_load_fold_idle(); | |
3338 | if (delta) | |
dce48a84 | 3339 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3340 | |
3341 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3342 | } |
3343 | ||
fdf3e95d VP |
3344 | /* |
3345 | * The exact cpuload at various idx values, calculated at every tick would be | |
3346 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3347 | * | |
3348 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3349 | * on nth tick when cpu may be busy, then we have: | |
3350 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3351 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3352 | * | |
3353 | * decay_load_missed() below does efficient calculation of | |
3354 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3355 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3356 | * | |
3357 | * The calculation is approximated on a 128 point scale. | |
3358 | * degrade_zero_ticks is the number of ticks after which load at any | |
3359 | * particular idx is approximated to be zero. | |
3360 | * degrade_factor is a precomputed table, a row for each load idx. | |
3361 | * Each column corresponds to degradation factor for a power of two ticks, | |
3362 | * based on 128 point scale. | |
3363 | * Example: | |
3364 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3365 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3366 | * | |
3367 | * With this power of 2 load factors, we can degrade the load n times | |
3368 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3369 | * n mult/shifts needed by the exact degradation. | |
3370 | */ | |
3371 | #define DEGRADE_SHIFT 7 | |
3372 | static const unsigned char | |
3373 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3374 | static const unsigned char | |
3375 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3376 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3377 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3378 | {96, 72, 40, 12, 1, 0, 0}, | |
3379 | {112, 98, 75, 43, 15, 1, 0}, | |
3380 | {120, 112, 98, 76, 45, 16, 2} }; | |
3381 | ||
3382 | /* | |
3383 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3384 | * would be when CPU is idle and so we just decay the old load without | |
3385 | * adding any new load. | |
3386 | */ | |
3387 | static unsigned long | |
3388 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3389 | { | |
3390 | int j = 0; | |
3391 | ||
3392 | if (!missed_updates) | |
3393 | return load; | |
3394 | ||
3395 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3396 | return 0; | |
3397 | ||
3398 | if (idx == 1) | |
3399 | return load >> missed_updates; | |
3400 | ||
3401 | while (missed_updates) { | |
3402 | if (missed_updates % 2) | |
3403 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3404 | ||
3405 | missed_updates >>= 1; | |
3406 | j++; | |
3407 | } | |
3408 | return load; | |
3409 | } | |
3410 | ||
46cb4b7c | 3411 | /* |
dd41f596 | 3412 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3413 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3414 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3415 | */ |
dd41f596 | 3416 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3417 | { |
495eca49 | 3418 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3419 | unsigned long curr_jiffies = jiffies; |
3420 | unsigned long pending_updates; | |
dd41f596 | 3421 | int i, scale; |
46cb4b7c | 3422 | |
dd41f596 | 3423 | this_rq->nr_load_updates++; |
46cb4b7c | 3424 | |
fdf3e95d VP |
3425 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3426 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3427 | return; | |
3428 | ||
3429 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3430 | this_rq->last_load_update_tick = curr_jiffies; | |
3431 | ||
dd41f596 | 3432 | /* Update our load: */ |
fdf3e95d VP |
3433 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3434 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3435 | unsigned long old_load, new_load; |
7d1e6a9b | 3436 | |
dd41f596 | 3437 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3438 | |
dd41f596 | 3439 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3440 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3441 | new_load = this_load; |
a25707f3 IM |
3442 | /* |
3443 | * Round up the averaging division if load is increasing. This | |
3444 | * prevents us from getting stuck on 9 if the load is 10, for | |
3445 | * example. | |
3446 | */ | |
3447 | if (new_load > old_load) | |
fdf3e95d VP |
3448 | new_load += scale - 1; |
3449 | ||
3450 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3451 | } |
da2b71ed SS |
3452 | |
3453 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3454 | } |
3455 | ||
3456 | static void update_cpu_load_active(struct rq *this_rq) | |
3457 | { | |
3458 | update_cpu_load(this_rq); | |
46cb4b7c | 3459 | |
74f5187a | 3460 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3461 | } |
3462 | ||
dd41f596 | 3463 | #ifdef CONFIG_SMP |
8a0be9ef | 3464 | |
46cb4b7c | 3465 | /* |
38022906 PZ |
3466 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3467 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3468 | */ |
38022906 | 3469 | void sched_exec(void) |
46cb4b7c | 3470 | { |
38022906 | 3471 | struct task_struct *p = current; |
1da177e4 | 3472 | unsigned long flags; |
70b97a7f | 3473 | struct rq *rq; |
0017d735 | 3474 | int dest_cpu; |
46cb4b7c | 3475 | |
1da177e4 | 3476 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3477 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3478 | if (dest_cpu == smp_processor_id()) | |
3479 | goto unlock; | |
38022906 | 3480 | |
46cb4b7c | 3481 | /* |
38022906 | 3482 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3483 | */ |
30da688e | 3484 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3485 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3486 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3487 | |
1da177e4 | 3488 | task_rq_unlock(rq, &flags); |
969c7921 | 3489 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3490 | return; |
3491 | } | |
0017d735 | 3492 | unlock: |
1da177e4 | 3493 | task_rq_unlock(rq, &flags); |
1da177e4 | 3494 | } |
dd41f596 | 3495 | |
1da177e4 LT |
3496 | #endif |
3497 | ||
1da177e4 LT |
3498 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3499 | ||
3500 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3501 | ||
3502 | /* | |
c5f8d995 | 3503 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3504 | * @p in case that task is currently running. |
c5f8d995 HS |
3505 | * |
3506 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3507 | */ |
c5f8d995 HS |
3508 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3509 | { | |
3510 | u64 ns = 0; | |
3511 | ||
3512 | if (task_current(rq, p)) { | |
3513 | update_rq_clock(rq); | |
305e6835 | 3514 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3515 | if ((s64)ns < 0) |
3516 | ns = 0; | |
3517 | } | |
3518 | ||
3519 | return ns; | |
3520 | } | |
3521 | ||
bb34d92f | 3522 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3523 | { |
1da177e4 | 3524 | unsigned long flags; |
41b86e9c | 3525 | struct rq *rq; |
bb34d92f | 3526 | u64 ns = 0; |
48f24c4d | 3527 | |
41b86e9c | 3528 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3529 | ns = do_task_delta_exec(p, rq); |
3530 | task_rq_unlock(rq, &flags); | |
1508487e | 3531 | |
c5f8d995 HS |
3532 | return ns; |
3533 | } | |
f06febc9 | 3534 | |
c5f8d995 HS |
3535 | /* |
3536 | * Return accounted runtime for the task. | |
3537 | * In case the task is currently running, return the runtime plus current's | |
3538 | * pending runtime that have not been accounted yet. | |
3539 | */ | |
3540 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3541 | { | |
3542 | unsigned long flags; | |
3543 | struct rq *rq; | |
3544 | u64 ns = 0; | |
3545 | ||
3546 | rq = task_rq_lock(p, &flags); | |
3547 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3548 | task_rq_unlock(rq, &flags); | |
3549 | ||
3550 | return ns; | |
3551 | } | |
48f24c4d | 3552 | |
c5f8d995 HS |
3553 | /* |
3554 | * Return sum_exec_runtime for the thread group. | |
3555 | * In case the task is currently running, return the sum plus current's | |
3556 | * pending runtime that have not been accounted yet. | |
3557 | * | |
3558 | * Note that the thread group might have other running tasks as well, | |
3559 | * so the return value not includes other pending runtime that other | |
3560 | * running tasks might have. | |
3561 | */ | |
3562 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3563 | { | |
3564 | struct task_cputime totals; | |
3565 | unsigned long flags; | |
3566 | struct rq *rq; | |
3567 | u64 ns; | |
3568 | ||
3569 | rq = task_rq_lock(p, &flags); | |
3570 | thread_group_cputime(p, &totals); | |
3571 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3572 | task_rq_unlock(rq, &flags); |
48f24c4d | 3573 | |
1da177e4 LT |
3574 | return ns; |
3575 | } | |
3576 | ||
1da177e4 LT |
3577 | /* |
3578 | * Account user cpu time to a process. | |
3579 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3580 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3581 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3582 | */ |
457533a7 MS |
3583 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3584 | cputime_t cputime_scaled) | |
1da177e4 LT |
3585 | { |
3586 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3587 | cputime64_t tmp; | |
3588 | ||
457533a7 | 3589 | /* Add user time to process. */ |
1da177e4 | 3590 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3591 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3592 | account_group_user_time(p, cputime); |
1da177e4 LT |
3593 | |
3594 | /* Add user time to cpustat. */ | |
3595 | tmp = cputime_to_cputime64(cputime); | |
3596 | if (TASK_NICE(p) > 0) | |
3597 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3598 | else | |
3599 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3600 | |
3601 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3602 | /* Account for user time used */ |
3603 | acct_update_integrals(p); | |
1da177e4 LT |
3604 | } |
3605 | ||
94886b84 LV |
3606 | /* |
3607 | * Account guest cpu time to a process. | |
3608 | * @p: the process that the cpu time gets accounted to | |
3609 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3610 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3611 | */ |
457533a7 MS |
3612 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3613 | cputime_t cputime_scaled) | |
94886b84 LV |
3614 | { |
3615 | cputime64_t tmp; | |
3616 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3617 | ||
3618 | tmp = cputime_to_cputime64(cputime); | |
3619 | ||
457533a7 | 3620 | /* Add guest time to process. */ |
94886b84 | 3621 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3622 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3623 | account_group_user_time(p, cputime); |
94886b84 LV |
3624 | p->gtime = cputime_add(p->gtime, cputime); |
3625 | ||
457533a7 | 3626 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3627 | if (TASK_NICE(p) > 0) { |
3628 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3629 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3630 | } else { | |
3631 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3632 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3633 | } | |
94886b84 LV |
3634 | } |
3635 | ||
70a89a66 VP |
3636 | /* |
3637 | * Account system cpu time to a process and desired cpustat field | |
3638 | * @p: the process that the cpu time gets accounted to | |
3639 | * @cputime: the cpu time spent in kernel space since the last update | |
3640 | * @cputime_scaled: cputime scaled by cpu frequency | |
3641 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3642 | */ | |
3643 | static inline | |
3644 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3645 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3646 | { | |
3647 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3648 | ||
3649 | /* Add system time to process. */ | |
3650 | p->stime = cputime_add(p->stime, cputime); | |
3651 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3652 | account_group_system_time(p, cputime); | |
3653 | ||
3654 | /* Add system time to cpustat. */ | |
3655 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3656 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3657 | ||
3658 | /* Account for system time used */ | |
3659 | acct_update_integrals(p); | |
3660 | } | |
3661 | ||
1da177e4 LT |
3662 | /* |
3663 | * Account system cpu time to a process. | |
3664 | * @p: the process that the cpu time gets accounted to | |
3665 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3666 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3667 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3668 | */ |
3669 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3670 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3671 | { |
3672 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3673 | cputime64_t *target_cputime64; |
1da177e4 | 3674 | |
983ed7a6 | 3675 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3676 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3677 | return; |
3678 | } | |
94886b84 | 3679 | |
1da177e4 | 3680 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3681 | target_cputime64 = &cpustat->irq; |
75e1056f | 3682 | else if (in_serving_softirq()) |
70a89a66 | 3683 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3684 | else |
70a89a66 | 3685 | target_cputime64 = &cpustat->system; |
79741dd3 | 3686 | |
70a89a66 | 3687 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3688 | } |
3689 | ||
7e949870 | 3690 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
abb74cef VP |
3691 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3692 | /* | |
3693 | * Account a tick to a process and cpustat | |
3694 | * @p: the process that the cpu time gets accounted to | |
3695 | * @user_tick: is the tick from userspace | |
3696 | * @rq: the pointer to rq | |
3697 | * | |
3698 | * Tick demultiplexing follows the order | |
3699 | * - pending hardirq update | |
3700 | * - pending softirq update | |
3701 | * - user_time | |
3702 | * - idle_time | |
3703 | * - system time | |
3704 | * - check for guest_time | |
3705 | * - else account as system_time | |
3706 | * | |
3707 | * Check for hardirq is done both for system and user time as there is | |
3708 | * no timer going off while we are on hardirq and hence we may never get an | |
3709 | * opportunity to update it solely in system time. | |
3710 | * p->stime and friends are only updated on system time and not on irq | |
3711 | * softirq as those do not count in task exec_runtime any more. | |
3712 | */ | |
3713 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3714 | struct rq *rq) | |
3715 | { | |
3716 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3717 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3718 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3719 | ||
3720 | if (irqtime_account_hi_update()) { | |
3721 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3722 | } else if (irqtime_account_si_update()) { | |
3723 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3724 | } else if (this_cpu_ksoftirqd() == p) { |
3725 | /* | |
3726 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3727 | * So, we have to handle it separately here. | |
3728 | * Also, p->stime needs to be updated for ksoftirqd. | |
3729 | */ | |
3730 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3731 | &cpustat->softirq); | |
abb74cef VP |
3732 | } else if (user_tick) { |
3733 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3734 | } else if (p == rq->idle) { | |
3735 | account_idle_time(cputime_one_jiffy); | |
3736 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3737 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3738 | } else { | |
3739 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3740 | &cpustat->system); | |
3741 | } | |
3742 | } | |
3743 | ||
3744 | static void irqtime_account_idle_ticks(int ticks) | |
3745 | { | |
3746 | int i; | |
3747 | struct rq *rq = this_rq(); | |
3748 | ||
3749 | for (i = 0; i < ticks; i++) | |
3750 | irqtime_account_process_tick(current, 0, rq); | |
3751 | } | |
3752 | #else | |
3753 | static void irqtime_account_idle_ticks(int ticks) {} | |
3754 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3755 | struct rq *rq) {} | |
3756 | #endif | |
7e949870 | 3757 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING */ |
abb74cef | 3758 | |
c66f08be | 3759 | /* |
1da177e4 | 3760 | * Account for involuntary wait time. |
1da177e4 | 3761 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3762 | */ |
79741dd3 | 3763 | void account_steal_time(cputime_t cputime) |
c66f08be | 3764 | { |
79741dd3 MS |
3765 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3766 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3767 | ||
3768 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3769 | } |
3770 | ||
1da177e4 | 3771 | /* |
79741dd3 MS |
3772 | * Account for idle time. |
3773 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3774 | */ |
79741dd3 | 3775 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3776 | { |
3777 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3778 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3779 | struct rq *rq = this_rq(); |
1da177e4 | 3780 | |
79741dd3 MS |
3781 | if (atomic_read(&rq->nr_iowait) > 0) |
3782 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3783 | else | |
3784 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3785 | } |
3786 | ||
79741dd3 MS |
3787 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3788 | ||
3789 | /* | |
3790 | * Account a single tick of cpu time. | |
3791 | * @p: the process that the cpu time gets accounted to | |
3792 | * @user_tick: indicates if the tick is a user or a system tick | |
3793 | */ | |
3794 | void account_process_tick(struct task_struct *p, int user_tick) | |
3795 | { | |
a42548a1 | 3796 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3797 | struct rq *rq = this_rq(); |
3798 | ||
abb74cef VP |
3799 | if (sched_clock_irqtime) { |
3800 | irqtime_account_process_tick(p, user_tick, rq); | |
3801 | return; | |
3802 | } | |
3803 | ||
79741dd3 | 3804 | if (user_tick) |
a42548a1 | 3805 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3806 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3807 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3808 | one_jiffy_scaled); |
3809 | else | |
a42548a1 | 3810 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3811 | } |
3812 | ||
3813 | /* | |
3814 | * Account multiple ticks of steal time. | |
3815 | * @p: the process from which the cpu time has been stolen | |
3816 | * @ticks: number of stolen ticks | |
3817 | */ | |
3818 | void account_steal_ticks(unsigned long ticks) | |
3819 | { | |
3820 | account_steal_time(jiffies_to_cputime(ticks)); | |
3821 | } | |
3822 | ||
3823 | /* | |
3824 | * Account multiple ticks of idle time. | |
3825 | * @ticks: number of stolen ticks | |
3826 | */ | |
3827 | void account_idle_ticks(unsigned long ticks) | |
3828 | { | |
abb74cef VP |
3829 | |
3830 | if (sched_clock_irqtime) { | |
3831 | irqtime_account_idle_ticks(ticks); | |
3832 | return; | |
3833 | } | |
3834 | ||
79741dd3 | 3835 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3836 | } |
3837 | ||
79741dd3 MS |
3838 | #endif |
3839 | ||
49048622 BS |
3840 | /* |
3841 | * Use precise platform statistics if available: | |
3842 | */ | |
3843 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3844 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3845 | { |
d99ca3b9 HS |
3846 | *ut = p->utime; |
3847 | *st = p->stime; | |
49048622 BS |
3848 | } |
3849 | ||
0cf55e1e | 3850 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3851 | { |
0cf55e1e HS |
3852 | struct task_cputime cputime; |
3853 | ||
3854 | thread_group_cputime(p, &cputime); | |
3855 | ||
3856 | *ut = cputime.utime; | |
3857 | *st = cputime.stime; | |
49048622 BS |
3858 | } |
3859 | #else | |
761b1d26 HS |
3860 | |
3861 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3862 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3863 | #endif |
3864 | ||
d180c5bc | 3865 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3866 | { |
d99ca3b9 | 3867 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3868 | |
3869 | /* | |
3870 | * Use CFS's precise accounting: | |
3871 | */ | |
d180c5bc | 3872 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3873 | |
3874 | if (total) { | |
e75e863d | 3875 | u64 temp = rtime; |
d180c5bc | 3876 | |
e75e863d | 3877 | temp *= utime; |
49048622 | 3878 | do_div(temp, total); |
d180c5bc HS |
3879 | utime = (cputime_t)temp; |
3880 | } else | |
3881 | utime = rtime; | |
49048622 | 3882 | |
d180c5bc HS |
3883 | /* |
3884 | * Compare with previous values, to keep monotonicity: | |
3885 | */ | |
761b1d26 | 3886 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3887 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3888 | |
d99ca3b9 HS |
3889 | *ut = p->prev_utime; |
3890 | *st = p->prev_stime; | |
49048622 BS |
3891 | } |
3892 | ||
0cf55e1e HS |
3893 | /* |
3894 | * Must be called with siglock held. | |
3895 | */ | |
3896 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3897 | { |
0cf55e1e HS |
3898 | struct signal_struct *sig = p->signal; |
3899 | struct task_cputime cputime; | |
3900 | cputime_t rtime, utime, total; | |
49048622 | 3901 | |
0cf55e1e | 3902 | thread_group_cputime(p, &cputime); |
49048622 | 3903 | |
0cf55e1e HS |
3904 | total = cputime_add(cputime.utime, cputime.stime); |
3905 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3906 | |
0cf55e1e | 3907 | if (total) { |
e75e863d | 3908 | u64 temp = rtime; |
49048622 | 3909 | |
e75e863d | 3910 | temp *= cputime.utime; |
0cf55e1e HS |
3911 | do_div(temp, total); |
3912 | utime = (cputime_t)temp; | |
3913 | } else | |
3914 | utime = rtime; | |
3915 | ||
3916 | sig->prev_utime = max(sig->prev_utime, utime); | |
3917 | sig->prev_stime = max(sig->prev_stime, | |
3918 | cputime_sub(rtime, sig->prev_utime)); | |
3919 | ||
3920 | *ut = sig->prev_utime; | |
3921 | *st = sig->prev_stime; | |
49048622 | 3922 | } |
49048622 | 3923 | #endif |
49048622 | 3924 | |
7835b98b CL |
3925 | /* |
3926 | * This function gets called by the timer code, with HZ frequency. | |
3927 | * We call it with interrupts disabled. | |
3928 | * | |
3929 | * It also gets called by the fork code, when changing the parent's | |
3930 | * timeslices. | |
3931 | */ | |
3932 | void scheduler_tick(void) | |
3933 | { | |
7835b98b CL |
3934 | int cpu = smp_processor_id(); |
3935 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3936 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3937 | |
3938 | sched_clock_tick(); | |
dd41f596 | 3939 | |
05fa785c | 3940 | raw_spin_lock(&rq->lock); |
3e51f33f | 3941 | update_rq_clock(rq); |
fdf3e95d | 3942 | update_cpu_load_active(rq); |
fa85ae24 | 3943 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3944 | raw_spin_unlock(&rq->lock); |
7835b98b | 3945 | |
e9d2b064 | 3946 | perf_event_task_tick(); |
e220d2dc | 3947 | |
e418e1c2 | 3948 | #ifdef CONFIG_SMP |
dd41f596 IM |
3949 | rq->idle_at_tick = idle_cpu(cpu); |
3950 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3951 | #endif |
1da177e4 LT |
3952 | } |
3953 | ||
132380a0 | 3954 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3955 | { |
3956 | if (in_lock_functions(addr)) { | |
3957 | addr = CALLER_ADDR2; | |
3958 | if (in_lock_functions(addr)) | |
3959 | addr = CALLER_ADDR3; | |
3960 | } | |
3961 | return addr; | |
3962 | } | |
1da177e4 | 3963 | |
7e49fcce SR |
3964 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3965 | defined(CONFIG_PREEMPT_TRACER)) | |
3966 | ||
43627582 | 3967 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3968 | { |
6cd8a4bb | 3969 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3970 | /* |
3971 | * Underflow? | |
3972 | */ | |
9a11b49a IM |
3973 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3974 | return; | |
6cd8a4bb | 3975 | #endif |
1da177e4 | 3976 | preempt_count() += val; |
6cd8a4bb | 3977 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3978 | /* |
3979 | * Spinlock count overflowing soon? | |
3980 | */ | |
33859f7f MOS |
3981 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3982 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3983 | #endif |
3984 | if (preempt_count() == val) | |
3985 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3986 | } |
3987 | EXPORT_SYMBOL(add_preempt_count); | |
3988 | ||
43627582 | 3989 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3990 | { |
6cd8a4bb | 3991 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3992 | /* |
3993 | * Underflow? | |
3994 | */ | |
01e3eb82 | 3995 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3996 | return; |
1da177e4 LT |
3997 | /* |
3998 | * Is the spinlock portion underflowing? | |
3999 | */ | |
9a11b49a IM |
4000 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4001 | !(preempt_count() & PREEMPT_MASK))) | |
4002 | return; | |
6cd8a4bb | 4003 | #endif |
9a11b49a | 4004 | |
6cd8a4bb SR |
4005 | if (preempt_count() == val) |
4006 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4007 | preempt_count() -= val; |
4008 | } | |
4009 | EXPORT_SYMBOL(sub_preempt_count); | |
4010 | ||
4011 | #endif | |
4012 | ||
4013 | /* | |
dd41f596 | 4014 | * Print scheduling while atomic bug: |
1da177e4 | 4015 | */ |
dd41f596 | 4016 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4017 | { |
838225b4 SS |
4018 | struct pt_regs *regs = get_irq_regs(); |
4019 | ||
3df0fc5b PZ |
4020 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4021 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4022 | |
dd41f596 | 4023 | debug_show_held_locks(prev); |
e21f5b15 | 4024 | print_modules(); |
dd41f596 IM |
4025 | if (irqs_disabled()) |
4026 | print_irqtrace_events(prev); | |
838225b4 SS |
4027 | |
4028 | if (regs) | |
4029 | show_regs(regs); | |
4030 | else | |
4031 | dump_stack(); | |
dd41f596 | 4032 | } |
1da177e4 | 4033 | |
dd41f596 IM |
4034 | /* |
4035 | * Various schedule()-time debugging checks and statistics: | |
4036 | */ | |
4037 | static inline void schedule_debug(struct task_struct *prev) | |
4038 | { | |
1da177e4 | 4039 | /* |
41a2d6cf | 4040 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4041 | * schedule() atomically, we ignore that path for now. |
4042 | * Otherwise, whine if we are scheduling when we should not be. | |
4043 | */ | |
3f33a7ce | 4044 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4045 | __schedule_bug(prev); |
4046 | ||
1da177e4 LT |
4047 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4048 | ||
2d72376b | 4049 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4050 | #ifdef CONFIG_SCHEDSTATS |
4051 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 4052 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 4053 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
4054 | } |
4055 | #endif | |
dd41f596 IM |
4056 | } |
4057 | ||
6cecd084 | 4058 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4059 | { |
a64692a3 MG |
4060 | if (prev->se.on_rq) |
4061 | update_rq_clock(rq); | |
6cecd084 | 4062 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4063 | } |
4064 | ||
dd41f596 IM |
4065 | /* |
4066 | * Pick up the highest-prio task: | |
4067 | */ | |
4068 | static inline struct task_struct * | |
b67802ea | 4069 | pick_next_task(struct rq *rq) |
dd41f596 | 4070 | { |
5522d5d5 | 4071 | const struct sched_class *class; |
dd41f596 | 4072 | struct task_struct *p; |
1da177e4 LT |
4073 | |
4074 | /* | |
dd41f596 IM |
4075 | * Optimization: we know that if all tasks are in |
4076 | * the fair class we can call that function directly: | |
1da177e4 | 4077 | */ |
dd41f596 | 4078 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4079 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4080 | if (likely(p)) |
4081 | return p; | |
1da177e4 LT |
4082 | } |
4083 | ||
34f971f6 | 4084 | for_each_class(class) { |
fb8d4724 | 4085 | p = class->pick_next_task(rq); |
dd41f596 IM |
4086 | if (p) |
4087 | return p; | |
dd41f596 | 4088 | } |
34f971f6 PZ |
4089 | |
4090 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4091 | } |
1da177e4 | 4092 | |
dd41f596 IM |
4093 | /* |
4094 | * schedule() is the main scheduler function. | |
4095 | */ | |
ff743345 | 4096 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4097 | { |
4098 | struct task_struct *prev, *next; | |
67ca7bde | 4099 | unsigned long *switch_count; |
dd41f596 | 4100 | struct rq *rq; |
31656519 | 4101 | int cpu; |
dd41f596 | 4102 | |
ff743345 PZ |
4103 | need_resched: |
4104 | preempt_disable(); | |
dd41f596 IM |
4105 | cpu = smp_processor_id(); |
4106 | rq = cpu_rq(cpu); | |
25502a6c | 4107 | rcu_note_context_switch(cpu); |
dd41f596 | 4108 | prev = rq->curr; |
dd41f596 IM |
4109 | |
4110 | release_kernel_lock(prev); | |
4111 | need_resched_nonpreemptible: | |
4112 | ||
4113 | schedule_debug(prev); | |
1da177e4 | 4114 | |
31656519 | 4115 | if (sched_feat(HRTICK)) |
f333fdc9 | 4116 | hrtick_clear(rq); |
8f4d37ec | 4117 | |
05fa785c | 4118 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4119 | |
246d86b5 | 4120 | switch_count = &prev->nivcsw; |
1da177e4 | 4121 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4122 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4123 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4124 | } else { |
4125 | /* | |
4126 | * If a worker is going to sleep, notify and | |
4127 | * ask workqueue whether it wants to wake up a | |
4128 | * task to maintain concurrency. If so, wake | |
4129 | * up the task. | |
4130 | */ | |
4131 | if (prev->flags & PF_WQ_WORKER) { | |
4132 | struct task_struct *to_wakeup; | |
4133 | ||
4134 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4135 | if (to_wakeup) | |
4136 | try_to_wake_up_local(to_wakeup); | |
4137 | } | |
371fd7e7 | 4138 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 4139 | } |
dd41f596 | 4140 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4141 | } |
4142 | ||
3f029d3c | 4143 | pre_schedule(rq, prev); |
f65eda4f | 4144 | |
dd41f596 | 4145 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4146 | idle_balance(cpu, rq); |
1da177e4 | 4147 | |
df1c99d4 | 4148 | put_prev_task(rq, prev); |
b67802ea | 4149 | next = pick_next_task(rq); |
f26f9aff MG |
4150 | clear_tsk_need_resched(prev); |
4151 | rq->skip_clock_update = 0; | |
1da177e4 | 4152 | |
1da177e4 | 4153 | if (likely(prev != next)) { |
673a90a1 | 4154 | sched_info_switch(prev, next); |
49f47433 | 4155 | perf_event_task_sched_out(prev, next); |
673a90a1 | 4156 | |
1da177e4 LT |
4157 | rq->nr_switches++; |
4158 | rq->curr = next; | |
4159 | ++*switch_count; | |
4160 | ||
dd41f596 | 4161 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4162 | /* |
246d86b5 ON |
4163 | * The context switch have flipped the stack from under us |
4164 | * and restored the local variables which were saved when | |
4165 | * this task called schedule() in the past. prev == current | |
4166 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4167 | */ |
4168 | cpu = smp_processor_id(); | |
4169 | rq = cpu_rq(cpu); | |
1da177e4 | 4170 | } else |
05fa785c | 4171 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4172 | |
3f029d3c | 4173 | post_schedule(rq); |
1da177e4 | 4174 | |
246d86b5 | 4175 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 4176 | goto need_resched_nonpreemptible; |
8f4d37ec | 4177 | |
1da177e4 | 4178 | preempt_enable_no_resched(); |
ff743345 | 4179 | if (need_resched()) |
1da177e4 LT |
4180 | goto need_resched; |
4181 | } | |
1da177e4 LT |
4182 | EXPORT_SYMBOL(schedule); |
4183 | ||
c08f7829 | 4184 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4185 | /* |
4186 | * Look out! "owner" is an entirely speculative pointer | |
4187 | * access and not reliable. | |
4188 | */ | |
4189 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4190 | { | |
4191 | unsigned int cpu; | |
4192 | struct rq *rq; | |
4193 | ||
4194 | if (!sched_feat(OWNER_SPIN)) | |
4195 | return 0; | |
4196 | ||
4197 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4198 | /* | |
4199 | * Need to access the cpu field knowing that | |
4200 | * DEBUG_PAGEALLOC could have unmapped it if | |
4201 | * the mutex owner just released it and exited. | |
4202 | */ | |
4203 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4204 | return 0; |
0d66bf6d PZ |
4205 | #else |
4206 | cpu = owner->cpu; | |
4207 | #endif | |
4208 | ||
4209 | /* | |
4210 | * Even if the access succeeded (likely case), | |
4211 | * the cpu field may no longer be valid. | |
4212 | */ | |
4213 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4214 | return 0; |
0d66bf6d PZ |
4215 | |
4216 | /* | |
4217 | * We need to validate that we can do a | |
4218 | * get_cpu() and that we have the percpu area. | |
4219 | */ | |
4220 | if (!cpu_online(cpu)) | |
4b402210 | 4221 | return 0; |
0d66bf6d PZ |
4222 | |
4223 | rq = cpu_rq(cpu); | |
4224 | ||
4225 | for (;;) { | |
4226 | /* | |
4227 | * Owner changed, break to re-assess state. | |
4228 | */ | |
9d0f4dcc TC |
4229 | if (lock->owner != owner) { |
4230 | /* | |
4231 | * If the lock has switched to a different owner, | |
4232 | * we likely have heavy contention. Return 0 to quit | |
4233 | * optimistic spinning and not contend further: | |
4234 | */ | |
4235 | if (lock->owner) | |
4236 | return 0; | |
0d66bf6d | 4237 | break; |
9d0f4dcc | 4238 | } |
0d66bf6d PZ |
4239 | |
4240 | /* | |
4241 | * Is that owner really running on that cpu? | |
4242 | */ | |
4243 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4244 | return 0; | |
4245 | ||
335d7afb | 4246 | arch_mutex_cpu_relax(); |
0d66bf6d | 4247 | } |
4b402210 | 4248 | |
0d66bf6d PZ |
4249 | return 1; |
4250 | } | |
4251 | #endif | |
4252 | ||
1da177e4 LT |
4253 | #ifdef CONFIG_PREEMPT |
4254 | /* | |
2ed6e34f | 4255 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4256 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4257 | * occur there and call schedule directly. |
4258 | */ | |
d1f74e20 | 4259 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4260 | { |
4261 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4262 | |
1da177e4 LT |
4263 | /* |
4264 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4265 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4266 | */ |
beed33a8 | 4267 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4268 | return; |
4269 | ||
3a5c359a | 4270 | do { |
d1f74e20 | 4271 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4272 | schedule(); |
d1f74e20 | 4273 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4274 | |
3a5c359a AK |
4275 | /* |
4276 | * Check again in case we missed a preemption opportunity | |
4277 | * between schedule and now. | |
4278 | */ | |
4279 | barrier(); | |
5ed0cec0 | 4280 | } while (need_resched()); |
1da177e4 | 4281 | } |
1da177e4 LT |
4282 | EXPORT_SYMBOL(preempt_schedule); |
4283 | ||
4284 | /* | |
2ed6e34f | 4285 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4286 | * off of irq context. |
4287 | * Note, that this is called and return with irqs disabled. This will | |
4288 | * protect us against recursive calling from irq. | |
4289 | */ | |
4290 | asmlinkage void __sched preempt_schedule_irq(void) | |
4291 | { | |
4292 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4293 | |
2ed6e34f | 4294 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4295 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4296 | ||
3a5c359a AK |
4297 | do { |
4298 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4299 | local_irq_enable(); |
4300 | schedule(); | |
4301 | local_irq_disable(); | |
3a5c359a | 4302 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4303 | |
3a5c359a AK |
4304 | /* |
4305 | * Check again in case we missed a preemption opportunity | |
4306 | * between schedule and now. | |
4307 | */ | |
4308 | barrier(); | |
5ed0cec0 | 4309 | } while (need_resched()); |
1da177e4 LT |
4310 | } |
4311 | ||
4312 | #endif /* CONFIG_PREEMPT */ | |
4313 | ||
63859d4f | 4314 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4315 | void *key) |
1da177e4 | 4316 | { |
63859d4f | 4317 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4318 | } |
1da177e4 LT |
4319 | EXPORT_SYMBOL(default_wake_function); |
4320 | ||
4321 | /* | |
41a2d6cf IM |
4322 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4323 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4324 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4325 | * | |
4326 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4327 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4328 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4329 | */ | |
78ddb08f | 4330 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4331 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4332 | { |
2e45874c | 4333 | wait_queue_t *curr, *next; |
1da177e4 | 4334 | |
2e45874c | 4335 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4336 | unsigned flags = curr->flags; |
4337 | ||
63859d4f | 4338 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4339 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4340 | break; |
4341 | } | |
4342 | } | |
4343 | ||
4344 | /** | |
4345 | * __wake_up - wake up threads blocked on a waitqueue. | |
4346 | * @q: the waitqueue | |
4347 | * @mode: which threads | |
4348 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4349 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4350 | * |
4351 | * It may be assumed that this function implies a write memory barrier before | |
4352 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4353 | */ |
7ad5b3a5 | 4354 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4355 | int nr_exclusive, void *key) |
1da177e4 LT |
4356 | { |
4357 | unsigned long flags; | |
4358 | ||
4359 | spin_lock_irqsave(&q->lock, flags); | |
4360 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4361 | spin_unlock_irqrestore(&q->lock, flags); | |
4362 | } | |
1da177e4 LT |
4363 | EXPORT_SYMBOL(__wake_up); |
4364 | ||
4365 | /* | |
4366 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4367 | */ | |
7ad5b3a5 | 4368 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4369 | { |
4370 | __wake_up_common(q, mode, 1, 0, NULL); | |
4371 | } | |
22c43c81 | 4372 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4373 | |
4ede816a DL |
4374 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4375 | { | |
4376 | __wake_up_common(q, mode, 1, 0, key); | |
4377 | } | |
4378 | ||
1da177e4 | 4379 | /** |
4ede816a | 4380 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4381 | * @q: the waitqueue |
4382 | * @mode: which threads | |
4383 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4384 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4385 | * |
4386 | * The sync wakeup differs that the waker knows that it will schedule | |
4387 | * away soon, so while the target thread will be woken up, it will not | |
4388 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4389 | * with each other. This can prevent needless bouncing between CPUs. | |
4390 | * | |
4391 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4392 | * |
4393 | * It may be assumed that this function implies a write memory barrier before | |
4394 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4395 | */ |
4ede816a DL |
4396 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4397 | int nr_exclusive, void *key) | |
1da177e4 LT |
4398 | { |
4399 | unsigned long flags; | |
7d478721 | 4400 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4401 | |
4402 | if (unlikely(!q)) | |
4403 | return; | |
4404 | ||
4405 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4406 | wake_flags = 0; |
1da177e4 LT |
4407 | |
4408 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4409 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4410 | spin_unlock_irqrestore(&q->lock, flags); |
4411 | } | |
4ede816a DL |
4412 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4413 | ||
4414 | /* | |
4415 | * __wake_up_sync - see __wake_up_sync_key() | |
4416 | */ | |
4417 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4418 | { | |
4419 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4420 | } | |
1da177e4 LT |
4421 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4422 | ||
65eb3dc6 KD |
4423 | /** |
4424 | * complete: - signals a single thread waiting on this completion | |
4425 | * @x: holds the state of this particular completion | |
4426 | * | |
4427 | * This will wake up a single thread waiting on this completion. Threads will be | |
4428 | * awakened in the same order in which they were queued. | |
4429 | * | |
4430 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4431 | * |
4432 | * It may be assumed that this function implies a write memory barrier before | |
4433 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4434 | */ |
b15136e9 | 4435 | void complete(struct completion *x) |
1da177e4 LT |
4436 | { |
4437 | unsigned long flags; | |
4438 | ||
4439 | spin_lock_irqsave(&x->wait.lock, flags); | |
4440 | x->done++; | |
d9514f6c | 4441 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4442 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4443 | } | |
4444 | EXPORT_SYMBOL(complete); | |
4445 | ||
65eb3dc6 KD |
4446 | /** |
4447 | * complete_all: - signals all threads waiting on this completion | |
4448 | * @x: holds the state of this particular completion | |
4449 | * | |
4450 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4451 | * |
4452 | * It may be assumed that this function implies a write memory barrier before | |
4453 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4454 | */ |
b15136e9 | 4455 | void complete_all(struct completion *x) |
1da177e4 LT |
4456 | { |
4457 | unsigned long flags; | |
4458 | ||
4459 | spin_lock_irqsave(&x->wait.lock, flags); | |
4460 | x->done += UINT_MAX/2; | |
d9514f6c | 4461 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4462 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4463 | } | |
4464 | EXPORT_SYMBOL(complete_all); | |
4465 | ||
8cbbe86d AK |
4466 | static inline long __sched |
4467 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4468 | { |
1da177e4 LT |
4469 | if (!x->done) { |
4470 | DECLARE_WAITQUEUE(wait, current); | |
4471 | ||
a93d2f17 | 4472 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4473 | do { |
94d3d824 | 4474 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4475 | timeout = -ERESTARTSYS; |
4476 | break; | |
8cbbe86d AK |
4477 | } |
4478 | __set_current_state(state); | |
1da177e4 LT |
4479 | spin_unlock_irq(&x->wait.lock); |
4480 | timeout = schedule_timeout(timeout); | |
4481 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4482 | } while (!x->done && timeout); |
1da177e4 | 4483 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4484 | if (!x->done) |
4485 | return timeout; | |
1da177e4 LT |
4486 | } |
4487 | x->done--; | |
ea71a546 | 4488 | return timeout ?: 1; |
1da177e4 | 4489 | } |
1da177e4 | 4490 | |
8cbbe86d AK |
4491 | static long __sched |
4492 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4493 | { |
1da177e4 LT |
4494 | might_sleep(); |
4495 | ||
4496 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4497 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4498 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4499 | return timeout; |
4500 | } | |
1da177e4 | 4501 | |
65eb3dc6 KD |
4502 | /** |
4503 | * wait_for_completion: - waits for completion of a task | |
4504 | * @x: holds the state of this particular completion | |
4505 | * | |
4506 | * This waits to be signaled for completion of a specific task. It is NOT | |
4507 | * interruptible and there is no timeout. | |
4508 | * | |
4509 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4510 | * and interrupt capability. Also see complete(). | |
4511 | */ | |
b15136e9 | 4512 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4513 | { |
4514 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4515 | } |
8cbbe86d | 4516 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4517 | |
65eb3dc6 KD |
4518 | /** |
4519 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4520 | * @x: holds the state of this particular completion | |
4521 | * @timeout: timeout value in jiffies | |
4522 | * | |
4523 | * This waits for either a completion of a specific task to be signaled or for a | |
4524 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4525 | * interruptible. | |
4526 | */ | |
b15136e9 | 4527 | unsigned long __sched |
8cbbe86d | 4528 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4529 | { |
8cbbe86d | 4530 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4531 | } |
8cbbe86d | 4532 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4533 | |
65eb3dc6 KD |
4534 | /** |
4535 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4536 | * @x: holds the state of this particular completion | |
4537 | * | |
4538 | * This waits for completion of a specific task to be signaled. It is | |
4539 | * interruptible. | |
4540 | */ | |
8cbbe86d | 4541 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4542 | { |
51e97990 AK |
4543 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4544 | if (t == -ERESTARTSYS) | |
4545 | return t; | |
4546 | return 0; | |
0fec171c | 4547 | } |
8cbbe86d | 4548 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4549 | |
65eb3dc6 KD |
4550 | /** |
4551 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4552 | * @x: holds the state of this particular completion | |
4553 | * @timeout: timeout value in jiffies | |
4554 | * | |
4555 | * This waits for either a completion of a specific task to be signaled or for a | |
4556 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4557 | */ | |
6bf41237 | 4558 | long __sched |
8cbbe86d AK |
4559 | wait_for_completion_interruptible_timeout(struct completion *x, |
4560 | unsigned long timeout) | |
0fec171c | 4561 | { |
8cbbe86d | 4562 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4563 | } |
8cbbe86d | 4564 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4565 | |
65eb3dc6 KD |
4566 | /** |
4567 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4568 | * @x: holds the state of this particular completion | |
4569 | * | |
4570 | * This waits to be signaled for completion of a specific task. It can be | |
4571 | * interrupted by a kill signal. | |
4572 | */ | |
009e577e MW |
4573 | int __sched wait_for_completion_killable(struct completion *x) |
4574 | { | |
4575 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4576 | if (t == -ERESTARTSYS) | |
4577 | return t; | |
4578 | return 0; | |
4579 | } | |
4580 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4581 | ||
0aa12fb4 SW |
4582 | /** |
4583 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4584 | * @x: holds the state of this particular completion | |
4585 | * @timeout: timeout value in jiffies | |
4586 | * | |
4587 | * This waits for either a completion of a specific task to be | |
4588 | * signaled or for a specified timeout to expire. It can be | |
4589 | * interrupted by a kill signal. The timeout is in jiffies. | |
4590 | */ | |
6bf41237 | 4591 | long __sched |
0aa12fb4 SW |
4592 | wait_for_completion_killable_timeout(struct completion *x, |
4593 | unsigned long timeout) | |
4594 | { | |
4595 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4596 | } | |
4597 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4598 | ||
be4de352 DC |
4599 | /** |
4600 | * try_wait_for_completion - try to decrement a completion without blocking | |
4601 | * @x: completion structure | |
4602 | * | |
4603 | * Returns: 0 if a decrement cannot be done without blocking | |
4604 | * 1 if a decrement succeeded. | |
4605 | * | |
4606 | * If a completion is being used as a counting completion, | |
4607 | * attempt to decrement the counter without blocking. This | |
4608 | * enables us to avoid waiting if the resource the completion | |
4609 | * is protecting is not available. | |
4610 | */ | |
4611 | bool try_wait_for_completion(struct completion *x) | |
4612 | { | |
7539a3b3 | 4613 | unsigned long flags; |
be4de352 DC |
4614 | int ret = 1; |
4615 | ||
7539a3b3 | 4616 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4617 | if (!x->done) |
4618 | ret = 0; | |
4619 | else | |
4620 | x->done--; | |
7539a3b3 | 4621 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4622 | return ret; |
4623 | } | |
4624 | EXPORT_SYMBOL(try_wait_for_completion); | |
4625 | ||
4626 | /** | |
4627 | * completion_done - Test to see if a completion has any waiters | |
4628 | * @x: completion structure | |
4629 | * | |
4630 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4631 | * 1 if there are no waiters. | |
4632 | * | |
4633 | */ | |
4634 | bool completion_done(struct completion *x) | |
4635 | { | |
7539a3b3 | 4636 | unsigned long flags; |
be4de352 DC |
4637 | int ret = 1; |
4638 | ||
7539a3b3 | 4639 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4640 | if (!x->done) |
4641 | ret = 0; | |
7539a3b3 | 4642 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4643 | return ret; |
4644 | } | |
4645 | EXPORT_SYMBOL(completion_done); | |
4646 | ||
8cbbe86d AK |
4647 | static long __sched |
4648 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4649 | { |
0fec171c IM |
4650 | unsigned long flags; |
4651 | wait_queue_t wait; | |
4652 | ||
4653 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4654 | |
8cbbe86d | 4655 | __set_current_state(state); |
1da177e4 | 4656 | |
8cbbe86d AK |
4657 | spin_lock_irqsave(&q->lock, flags); |
4658 | __add_wait_queue(q, &wait); | |
4659 | spin_unlock(&q->lock); | |
4660 | timeout = schedule_timeout(timeout); | |
4661 | spin_lock_irq(&q->lock); | |
4662 | __remove_wait_queue(q, &wait); | |
4663 | spin_unlock_irqrestore(&q->lock, flags); | |
4664 | ||
4665 | return timeout; | |
4666 | } | |
4667 | ||
4668 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4669 | { | |
4670 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4671 | } |
1da177e4 LT |
4672 | EXPORT_SYMBOL(interruptible_sleep_on); |
4673 | ||
0fec171c | 4674 | long __sched |
95cdf3b7 | 4675 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4676 | { |
8cbbe86d | 4677 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4678 | } |
1da177e4 LT |
4679 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4680 | ||
0fec171c | 4681 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4682 | { |
8cbbe86d | 4683 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4684 | } |
1da177e4 LT |
4685 | EXPORT_SYMBOL(sleep_on); |
4686 | ||
0fec171c | 4687 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4688 | { |
8cbbe86d | 4689 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4690 | } |
1da177e4 LT |
4691 | EXPORT_SYMBOL(sleep_on_timeout); |
4692 | ||
b29739f9 IM |
4693 | #ifdef CONFIG_RT_MUTEXES |
4694 | ||
4695 | /* | |
4696 | * rt_mutex_setprio - set the current priority of a task | |
4697 | * @p: task | |
4698 | * @prio: prio value (kernel-internal form) | |
4699 | * | |
4700 | * This function changes the 'effective' priority of a task. It does | |
4701 | * not touch ->normal_prio like __setscheduler(). | |
4702 | * | |
4703 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4704 | */ | |
36c8b586 | 4705 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4706 | { |
4707 | unsigned long flags; | |
83b699ed | 4708 | int oldprio, on_rq, running; |
70b97a7f | 4709 | struct rq *rq; |
83ab0aa0 | 4710 | const struct sched_class *prev_class; |
b29739f9 IM |
4711 | |
4712 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4713 | ||
4714 | rq = task_rq_lock(p, &flags); | |
4715 | ||
a8027073 | 4716 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4717 | oldprio = p->prio; |
83ab0aa0 | 4718 | prev_class = p->sched_class; |
dd41f596 | 4719 | on_rq = p->se.on_rq; |
051a1d1a | 4720 | running = task_current(rq, p); |
0e1f3483 | 4721 | if (on_rq) |
69be72c1 | 4722 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4723 | if (running) |
4724 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4725 | |
4726 | if (rt_prio(prio)) | |
4727 | p->sched_class = &rt_sched_class; | |
4728 | else | |
4729 | p->sched_class = &fair_sched_class; | |
4730 | ||
b29739f9 IM |
4731 | p->prio = prio; |
4732 | ||
0e1f3483 HS |
4733 | if (running) |
4734 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4735 | if (on_rq) |
371fd7e7 | 4736 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4737 | |
da7a735e | 4738 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 IM |
4739 | task_rq_unlock(rq, &flags); |
4740 | } | |
4741 | ||
4742 | #endif | |
4743 | ||
36c8b586 | 4744 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4745 | { |
dd41f596 | 4746 | int old_prio, delta, on_rq; |
1da177e4 | 4747 | unsigned long flags; |
70b97a7f | 4748 | struct rq *rq; |
1da177e4 LT |
4749 | |
4750 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4751 | return; | |
4752 | /* | |
4753 | * We have to be careful, if called from sys_setpriority(), | |
4754 | * the task might be in the middle of scheduling on another CPU. | |
4755 | */ | |
4756 | rq = task_rq_lock(p, &flags); | |
4757 | /* | |
4758 | * The RT priorities are set via sched_setscheduler(), but we still | |
4759 | * allow the 'normal' nice value to be set - but as expected | |
4760 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4761 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4762 | */ |
e05606d3 | 4763 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4764 | p->static_prio = NICE_TO_PRIO(nice); |
4765 | goto out_unlock; | |
4766 | } | |
dd41f596 | 4767 | on_rq = p->se.on_rq; |
c09595f6 | 4768 | if (on_rq) |
69be72c1 | 4769 | dequeue_task(rq, p, 0); |
1da177e4 | 4770 | |
1da177e4 | 4771 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4772 | set_load_weight(p); |
b29739f9 IM |
4773 | old_prio = p->prio; |
4774 | p->prio = effective_prio(p); | |
4775 | delta = p->prio - old_prio; | |
1da177e4 | 4776 | |
dd41f596 | 4777 | if (on_rq) { |
371fd7e7 | 4778 | enqueue_task(rq, p, 0); |
1da177e4 | 4779 | /* |
d5f9f942 AM |
4780 | * If the task increased its priority or is running and |
4781 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4782 | */ |
d5f9f942 | 4783 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4784 | resched_task(rq->curr); |
4785 | } | |
4786 | out_unlock: | |
4787 | task_rq_unlock(rq, &flags); | |
4788 | } | |
1da177e4 LT |
4789 | EXPORT_SYMBOL(set_user_nice); |
4790 | ||
e43379f1 MM |
4791 | /* |
4792 | * can_nice - check if a task can reduce its nice value | |
4793 | * @p: task | |
4794 | * @nice: nice value | |
4795 | */ | |
36c8b586 | 4796 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4797 | { |
024f4747 MM |
4798 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4799 | int nice_rlim = 20 - nice; | |
48f24c4d | 4800 | |
78d7d407 | 4801 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4802 | capable(CAP_SYS_NICE)); |
4803 | } | |
4804 | ||
1da177e4 LT |
4805 | #ifdef __ARCH_WANT_SYS_NICE |
4806 | ||
4807 | /* | |
4808 | * sys_nice - change the priority of the current process. | |
4809 | * @increment: priority increment | |
4810 | * | |
4811 | * sys_setpriority is a more generic, but much slower function that | |
4812 | * does similar things. | |
4813 | */ | |
5add95d4 | 4814 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4815 | { |
48f24c4d | 4816 | long nice, retval; |
1da177e4 LT |
4817 | |
4818 | /* | |
4819 | * Setpriority might change our priority at the same moment. | |
4820 | * We don't have to worry. Conceptually one call occurs first | |
4821 | * and we have a single winner. | |
4822 | */ | |
e43379f1 MM |
4823 | if (increment < -40) |
4824 | increment = -40; | |
1da177e4 LT |
4825 | if (increment > 40) |
4826 | increment = 40; | |
4827 | ||
2b8f836f | 4828 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4829 | if (nice < -20) |
4830 | nice = -20; | |
4831 | if (nice > 19) | |
4832 | nice = 19; | |
4833 | ||
e43379f1 MM |
4834 | if (increment < 0 && !can_nice(current, nice)) |
4835 | return -EPERM; | |
4836 | ||
1da177e4 LT |
4837 | retval = security_task_setnice(current, nice); |
4838 | if (retval) | |
4839 | return retval; | |
4840 | ||
4841 | set_user_nice(current, nice); | |
4842 | return 0; | |
4843 | } | |
4844 | ||
4845 | #endif | |
4846 | ||
4847 | /** | |
4848 | * task_prio - return the priority value of a given task. | |
4849 | * @p: the task in question. | |
4850 | * | |
4851 | * This is the priority value as seen by users in /proc. | |
4852 | * RT tasks are offset by -200. Normal tasks are centered | |
4853 | * around 0, value goes from -16 to +15. | |
4854 | */ | |
36c8b586 | 4855 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4856 | { |
4857 | return p->prio - MAX_RT_PRIO; | |
4858 | } | |
4859 | ||
4860 | /** | |
4861 | * task_nice - return the nice value of a given task. | |
4862 | * @p: the task in question. | |
4863 | */ | |
36c8b586 | 4864 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4865 | { |
4866 | return TASK_NICE(p); | |
4867 | } | |
150d8bed | 4868 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4869 | |
4870 | /** | |
4871 | * idle_cpu - is a given cpu idle currently? | |
4872 | * @cpu: the processor in question. | |
4873 | */ | |
4874 | int idle_cpu(int cpu) | |
4875 | { | |
4876 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4877 | } | |
4878 | ||
1da177e4 LT |
4879 | /** |
4880 | * idle_task - return the idle task for a given cpu. | |
4881 | * @cpu: the processor in question. | |
4882 | */ | |
36c8b586 | 4883 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4884 | { |
4885 | return cpu_rq(cpu)->idle; | |
4886 | } | |
4887 | ||
4888 | /** | |
4889 | * find_process_by_pid - find a process with a matching PID value. | |
4890 | * @pid: the pid in question. | |
4891 | */ | |
a9957449 | 4892 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4893 | { |
228ebcbe | 4894 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4895 | } |
4896 | ||
4897 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4898 | static void |
4899 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4900 | { |
dd41f596 | 4901 | BUG_ON(p->se.on_rq); |
48f24c4d | 4902 | |
1da177e4 LT |
4903 | p->policy = policy; |
4904 | p->rt_priority = prio; | |
b29739f9 IM |
4905 | p->normal_prio = normal_prio(p); |
4906 | /* we are holding p->pi_lock already */ | |
4907 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4908 | if (rt_prio(p->prio)) |
4909 | p->sched_class = &rt_sched_class; | |
4910 | else | |
4911 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4912 | set_load_weight(p); |
1da177e4 LT |
4913 | } |
4914 | ||
c69e8d9c DH |
4915 | /* |
4916 | * check the target process has a UID that matches the current process's | |
4917 | */ | |
4918 | static bool check_same_owner(struct task_struct *p) | |
4919 | { | |
4920 | const struct cred *cred = current_cred(), *pcred; | |
4921 | bool match; | |
4922 | ||
4923 | rcu_read_lock(); | |
4924 | pcred = __task_cred(p); | |
4925 | match = (cred->euid == pcred->euid || | |
4926 | cred->euid == pcred->uid); | |
4927 | rcu_read_unlock(); | |
4928 | return match; | |
4929 | } | |
4930 | ||
961ccddd | 4931 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4932 | const struct sched_param *param, bool user) |
1da177e4 | 4933 | { |
83b699ed | 4934 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4935 | unsigned long flags; |
83ab0aa0 | 4936 | const struct sched_class *prev_class; |
70b97a7f | 4937 | struct rq *rq; |
ca94c442 | 4938 | int reset_on_fork; |
1da177e4 | 4939 | |
66e5393a SR |
4940 | /* may grab non-irq protected spin_locks */ |
4941 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4942 | recheck: |
4943 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4944 | if (policy < 0) { |
4945 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4946 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4947 | } else { |
4948 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4949 | policy &= ~SCHED_RESET_ON_FORK; | |
4950 | ||
4951 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4952 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4953 | policy != SCHED_IDLE) | |
4954 | return -EINVAL; | |
4955 | } | |
4956 | ||
1da177e4 LT |
4957 | /* |
4958 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4959 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4960 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4961 | */ |
4962 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4963 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4964 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4965 | return -EINVAL; |
e05606d3 | 4966 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4967 | return -EINVAL; |
4968 | ||
37e4ab3f OC |
4969 | /* |
4970 | * Allow unprivileged RT tasks to decrease priority: | |
4971 | */ | |
961ccddd | 4972 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4973 | if (rt_policy(policy)) { |
a44702e8 ON |
4974 | unsigned long rlim_rtprio = |
4975 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4976 | |
4977 | /* can't set/change the rt policy */ | |
4978 | if (policy != p->policy && !rlim_rtprio) | |
4979 | return -EPERM; | |
4980 | ||
4981 | /* can't increase priority */ | |
4982 | if (param->sched_priority > p->rt_priority && | |
4983 | param->sched_priority > rlim_rtprio) | |
4984 | return -EPERM; | |
4985 | } | |
dd41f596 IM |
4986 | /* |
4987 | * Like positive nice levels, dont allow tasks to | |
4988 | * move out of SCHED_IDLE either: | |
4989 | */ | |
4990 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4991 | return -EPERM; | |
5fe1d75f | 4992 | |
37e4ab3f | 4993 | /* can't change other user's priorities */ |
c69e8d9c | 4994 | if (!check_same_owner(p)) |
37e4ab3f | 4995 | return -EPERM; |
ca94c442 LP |
4996 | |
4997 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4998 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4999 | return -EPERM; | |
37e4ab3f | 5000 | } |
1da177e4 | 5001 | |
725aad24 | 5002 | if (user) { |
b0ae1981 | 5003 | retval = security_task_setscheduler(p); |
725aad24 JF |
5004 | if (retval) |
5005 | return retval; | |
5006 | } | |
5007 | ||
b29739f9 IM |
5008 | /* |
5009 | * make sure no PI-waiters arrive (or leave) while we are | |
5010 | * changing the priority of the task: | |
5011 | */ | |
1d615482 | 5012 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
5013 | /* |
5014 | * To be able to change p->policy safely, the apropriate | |
5015 | * runqueue lock must be held. | |
5016 | */ | |
b29739f9 | 5017 | rq = __task_rq_lock(p); |
dc61b1d6 | 5018 | |
34f971f6 PZ |
5019 | /* |
5020 | * Changing the policy of the stop threads its a very bad idea | |
5021 | */ | |
5022 | if (p == rq->stop) { | |
5023 | __task_rq_unlock(rq); | |
5024 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5025 | return -EINVAL; | |
5026 | } | |
5027 | ||
dc61b1d6 PZ |
5028 | #ifdef CONFIG_RT_GROUP_SCHED |
5029 | if (user) { | |
5030 | /* | |
5031 | * Do not allow realtime tasks into groups that have no runtime | |
5032 | * assigned. | |
5033 | */ | |
5034 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5035 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5036 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
5037 | __task_rq_unlock(rq); |
5038 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5039 | return -EPERM; | |
5040 | } | |
5041 | } | |
5042 | #endif | |
5043 | ||
1da177e4 LT |
5044 | /* recheck policy now with rq lock held */ |
5045 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5046 | policy = oldpolicy = -1; | |
b29739f9 | 5047 | __task_rq_unlock(rq); |
1d615482 | 5048 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5049 | goto recheck; |
5050 | } | |
dd41f596 | 5051 | on_rq = p->se.on_rq; |
051a1d1a | 5052 | running = task_current(rq, p); |
0e1f3483 | 5053 | if (on_rq) |
2e1cb74a | 5054 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5055 | if (running) |
5056 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5057 | |
ca94c442 LP |
5058 | p->sched_reset_on_fork = reset_on_fork; |
5059 | ||
1da177e4 | 5060 | oldprio = p->prio; |
83ab0aa0 | 5061 | prev_class = p->sched_class; |
dd41f596 | 5062 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5063 | |
0e1f3483 HS |
5064 | if (running) |
5065 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5066 | if (on_rq) |
dd41f596 | 5067 | activate_task(rq, p, 0); |
cb469845 | 5068 | |
da7a735e | 5069 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 | 5070 | __task_rq_unlock(rq); |
1d615482 | 5071 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 5072 | |
95e02ca9 TG |
5073 | rt_mutex_adjust_pi(p); |
5074 | ||
1da177e4 LT |
5075 | return 0; |
5076 | } | |
961ccddd RR |
5077 | |
5078 | /** | |
5079 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5080 | * @p: the task in question. | |
5081 | * @policy: new policy. | |
5082 | * @param: structure containing the new RT priority. | |
5083 | * | |
5084 | * NOTE that the task may be already dead. | |
5085 | */ | |
5086 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5087 | const struct sched_param *param) |
961ccddd RR |
5088 | { |
5089 | return __sched_setscheduler(p, policy, param, true); | |
5090 | } | |
1da177e4 LT |
5091 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5092 | ||
961ccddd RR |
5093 | /** |
5094 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5095 | * @p: the task in question. | |
5096 | * @policy: new policy. | |
5097 | * @param: structure containing the new RT priority. | |
5098 | * | |
5099 | * Just like sched_setscheduler, only don't bother checking if the | |
5100 | * current context has permission. For example, this is needed in | |
5101 | * stop_machine(): we create temporary high priority worker threads, | |
5102 | * but our caller might not have that capability. | |
5103 | */ | |
5104 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5105 | const struct sched_param *param) |
961ccddd RR |
5106 | { |
5107 | return __sched_setscheduler(p, policy, param, false); | |
5108 | } | |
5109 | ||
95cdf3b7 IM |
5110 | static int |
5111 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5112 | { |
1da177e4 LT |
5113 | struct sched_param lparam; |
5114 | struct task_struct *p; | |
36c8b586 | 5115 | int retval; |
1da177e4 LT |
5116 | |
5117 | if (!param || pid < 0) | |
5118 | return -EINVAL; | |
5119 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5120 | return -EFAULT; | |
5fe1d75f ON |
5121 | |
5122 | rcu_read_lock(); | |
5123 | retval = -ESRCH; | |
1da177e4 | 5124 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5125 | if (p != NULL) |
5126 | retval = sched_setscheduler(p, policy, &lparam); | |
5127 | rcu_read_unlock(); | |
36c8b586 | 5128 | |
1da177e4 LT |
5129 | return retval; |
5130 | } | |
5131 | ||
5132 | /** | |
5133 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5134 | * @pid: the pid in question. | |
5135 | * @policy: new policy. | |
5136 | * @param: structure containing the new RT priority. | |
5137 | */ | |
5add95d4 HC |
5138 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5139 | struct sched_param __user *, param) | |
1da177e4 | 5140 | { |
c21761f1 JB |
5141 | /* negative values for policy are not valid */ |
5142 | if (policy < 0) | |
5143 | return -EINVAL; | |
5144 | ||
1da177e4 LT |
5145 | return do_sched_setscheduler(pid, policy, param); |
5146 | } | |
5147 | ||
5148 | /** | |
5149 | * sys_sched_setparam - set/change the RT priority of a thread | |
5150 | * @pid: the pid in question. | |
5151 | * @param: structure containing the new RT priority. | |
5152 | */ | |
5add95d4 | 5153 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5154 | { |
5155 | return do_sched_setscheduler(pid, -1, param); | |
5156 | } | |
5157 | ||
5158 | /** | |
5159 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5160 | * @pid: the pid in question. | |
5161 | */ | |
5add95d4 | 5162 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5163 | { |
36c8b586 | 5164 | struct task_struct *p; |
3a5c359a | 5165 | int retval; |
1da177e4 LT |
5166 | |
5167 | if (pid < 0) | |
3a5c359a | 5168 | return -EINVAL; |
1da177e4 LT |
5169 | |
5170 | retval = -ESRCH; | |
5fe85be0 | 5171 | rcu_read_lock(); |
1da177e4 LT |
5172 | p = find_process_by_pid(pid); |
5173 | if (p) { | |
5174 | retval = security_task_getscheduler(p); | |
5175 | if (!retval) | |
ca94c442 LP |
5176 | retval = p->policy |
5177 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5178 | } |
5fe85be0 | 5179 | rcu_read_unlock(); |
1da177e4 LT |
5180 | return retval; |
5181 | } | |
5182 | ||
5183 | /** | |
ca94c442 | 5184 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5185 | * @pid: the pid in question. |
5186 | * @param: structure containing the RT priority. | |
5187 | */ | |
5add95d4 | 5188 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5189 | { |
5190 | struct sched_param lp; | |
36c8b586 | 5191 | struct task_struct *p; |
3a5c359a | 5192 | int retval; |
1da177e4 LT |
5193 | |
5194 | if (!param || pid < 0) | |
3a5c359a | 5195 | return -EINVAL; |
1da177e4 | 5196 | |
5fe85be0 | 5197 | rcu_read_lock(); |
1da177e4 LT |
5198 | p = find_process_by_pid(pid); |
5199 | retval = -ESRCH; | |
5200 | if (!p) | |
5201 | goto out_unlock; | |
5202 | ||
5203 | retval = security_task_getscheduler(p); | |
5204 | if (retval) | |
5205 | goto out_unlock; | |
5206 | ||
5207 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5208 | rcu_read_unlock(); |
1da177e4 LT |
5209 | |
5210 | /* | |
5211 | * This one might sleep, we cannot do it with a spinlock held ... | |
5212 | */ | |
5213 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5214 | ||
1da177e4 LT |
5215 | return retval; |
5216 | ||
5217 | out_unlock: | |
5fe85be0 | 5218 | rcu_read_unlock(); |
1da177e4 LT |
5219 | return retval; |
5220 | } | |
5221 | ||
96f874e2 | 5222 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5223 | { |
5a16f3d3 | 5224 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5225 | struct task_struct *p; |
5226 | int retval; | |
1da177e4 | 5227 | |
95402b38 | 5228 | get_online_cpus(); |
23f5d142 | 5229 | rcu_read_lock(); |
1da177e4 LT |
5230 | |
5231 | p = find_process_by_pid(pid); | |
5232 | if (!p) { | |
23f5d142 | 5233 | rcu_read_unlock(); |
95402b38 | 5234 | put_online_cpus(); |
1da177e4 LT |
5235 | return -ESRCH; |
5236 | } | |
5237 | ||
23f5d142 | 5238 | /* Prevent p going away */ |
1da177e4 | 5239 | get_task_struct(p); |
23f5d142 | 5240 | rcu_read_unlock(); |
1da177e4 | 5241 | |
5a16f3d3 RR |
5242 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5243 | retval = -ENOMEM; | |
5244 | goto out_put_task; | |
5245 | } | |
5246 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5247 | retval = -ENOMEM; | |
5248 | goto out_free_cpus_allowed; | |
5249 | } | |
1da177e4 | 5250 | retval = -EPERM; |
c69e8d9c | 5251 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5252 | goto out_unlock; |
5253 | ||
b0ae1981 | 5254 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5255 | if (retval) |
5256 | goto out_unlock; | |
5257 | ||
5a16f3d3 RR |
5258 | cpuset_cpus_allowed(p, cpus_allowed); |
5259 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5260 | again: |
5a16f3d3 | 5261 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5262 | |
8707d8b8 | 5263 | if (!retval) { |
5a16f3d3 RR |
5264 | cpuset_cpus_allowed(p, cpus_allowed); |
5265 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5266 | /* |
5267 | * We must have raced with a concurrent cpuset | |
5268 | * update. Just reset the cpus_allowed to the | |
5269 | * cpuset's cpus_allowed | |
5270 | */ | |
5a16f3d3 | 5271 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5272 | goto again; |
5273 | } | |
5274 | } | |
1da177e4 | 5275 | out_unlock: |
5a16f3d3 RR |
5276 | free_cpumask_var(new_mask); |
5277 | out_free_cpus_allowed: | |
5278 | free_cpumask_var(cpus_allowed); | |
5279 | out_put_task: | |
1da177e4 | 5280 | put_task_struct(p); |
95402b38 | 5281 | put_online_cpus(); |
1da177e4 LT |
5282 | return retval; |
5283 | } | |
5284 | ||
5285 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5286 | struct cpumask *new_mask) |
1da177e4 | 5287 | { |
96f874e2 RR |
5288 | if (len < cpumask_size()) |
5289 | cpumask_clear(new_mask); | |
5290 | else if (len > cpumask_size()) | |
5291 | len = cpumask_size(); | |
5292 | ||
1da177e4 LT |
5293 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5294 | } | |
5295 | ||
5296 | /** | |
5297 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5298 | * @pid: pid of the process | |
5299 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5300 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5301 | */ | |
5add95d4 HC |
5302 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5303 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5304 | { |
5a16f3d3 | 5305 | cpumask_var_t new_mask; |
1da177e4 LT |
5306 | int retval; |
5307 | ||
5a16f3d3 RR |
5308 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5309 | return -ENOMEM; | |
1da177e4 | 5310 | |
5a16f3d3 RR |
5311 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5312 | if (retval == 0) | |
5313 | retval = sched_setaffinity(pid, new_mask); | |
5314 | free_cpumask_var(new_mask); | |
5315 | return retval; | |
1da177e4 LT |
5316 | } |
5317 | ||
96f874e2 | 5318 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5319 | { |
36c8b586 | 5320 | struct task_struct *p; |
31605683 TG |
5321 | unsigned long flags; |
5322 | struct rq *rq; | |
1da177e4 | 5323 | int retval; |
1da177e4 | 5324 | |
95402b38 | 5325 | get_online_cpus(); |
23f5d142 | 5326 | rcu_read_lock(); |
1da177e4 LT |
5327 | |
5328 | retval = -ESRCH; | |
5329 | p = find_process_by_pid(pid); | |
5330 | if (!p) | |
5331 | goto out_unlock; | |
5332 | ||
e7834f8f DQ |
5333 | retval = security_task_getscheduler(p); |
5334 | if (retval) | |
5335 | goto out_unlock; | |
5336 | ||
31605683 | 5337 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5338 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5339 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5340 | |
5341 | out_unlock: | |
23f5d142 | 5342 | rcu_read_unlock(); |
95402b38 | 5343 | put_online_cpus(); |
1da177e4 | 5344 | |
9531b62f | 5345 | return retval; |
1da177e4 LT |
5346 | } |
5347 | ||
5348 | /** | |
5349 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5350 | * @pid: pid of the process | |
5351 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5352 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5353 | */ | |
5add95d4 HC |
5354 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5355 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5356 | { |
5357 | int ret; | |
f17c8607 | 5358 | cpumask_var_t mask; |
1da177e4 | 5359 | |
84fba5ec | 5360 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5361 | return -EINVAL; |
5362 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5363 | return -EINVAL; |
5364 | ||
f17c8607 RR |
5365 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5366 | return -ENOMEM; | |
1da177e4 | 5367 | |
f17c8607 RR |
5368 | ret = sched_getaffinity(pid, mask); |
5369 | if (ret == 0) { | |
8bc037fb | 5370 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5371 | |
5372 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5373 | ret = -EFAULT; |
5374 | else | |
cd3d8031 | 5375 | ret = retlen; |
f17c8607 RR |
5376 | } |
5377 | free_cpumask_var(mask); | |
1da177e4 | 5378 | |
f17c8607 | 5379 | return ret; |
1da177e4 LT |
5380 | } |
5381 | ||
5382 | /** | |
5383 | * sys_sched_yield - yield the current processor to other threads. | |
5384 | * | |
dd41f596 IM |
5385 | * This function yields the current CPU to other tasks. If there are no |
5386 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5387 | */ |
5add95d4 | 5388 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5389 | { |
70b97a7f | 5390 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5391 | |
2d72376b | 5392 | schedstat_inc(rq, yld_count); |
4530d7ab | 5393 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5394 | |
5395 | /* | |
5396 | * Since we are going to call schedule() anyway, there's | |
5397 | * no need to preempt or enable interrupts: | |
5398 | */ | |
5399 | __release(rq->lock); | |
8a25d5de | 5400 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5401 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5402 | preempt_enable_no_resched(); |
5403 | ||
5404 | schedule(); | |
5405 | ||
5406 | return 0; | |
5407 | } | |
5408 | ||
d86ee480 PZ |
5409 | static inline int should_resched(void) |
5410 | { | |
5411 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5412 | } | |
5413 | ||
e7b38404 | 5414 | static void __cond_resched(void) |
1da177e4 | 5415 | { |
e7aaaa69 FW |
5416 | add_preempt_count(PREEMPT_ACTIVE); |
5417 | schedule(); | |
5418 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5419 | } |
5420 | ||
02b67cc3 | 5421 | int __sched _cond_resched(void) |
1da177e4 | 5422 | { |
d86ee480 | 5423 | if (should_resched()) { |
1da177e4 LT |
5424 | __cond_resched(); |
5425 | return 1; | |
5426 | } | |
5427 | return 0; | |
5428 | } | |
02b67cc3 | 5429 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5430 | |
5431 | /* | |
613afbf8 | 5432 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5433 | * call schedule, and on return reacquire the lock. |
5434 | * | |
41a2d6cf | 5435 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5436 | * operations here to prevent schedule() from being called twice (once via |
5437 | * spin_unlock(), once by hand). | |
5438 | */ | |
613afbf8 | 5439 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5440 | { |
d86ee480 | 5441 | int resched = should_resched(); |
6df3cecb JK |
5442 | int ret = 0; |
5443 | ||
f607c668 PZ |
5444 | lockdep_assert_held(lock); |
5445 | ||
95c354fe | 5446 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5447 | spin_unlock(lock); |
d86ee480 | 5448 | if (resched) |
95c354fe NP |
5449 | __cond_resched(); |
5450 | else | |
5451 | cpu_relax(); | |
6df3cecb | 5452 | ret = 1; |
1da177e4 | 5453 | spin_lock(lock); |
1da177e4 | 5454 | } |
6df3cecb | 5455 | return ret; |
1da177e4 | 5456 | } |
613afbf8 | 5457 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5458 | |
613afbf8 | 5459 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5460 | { |
5461 | BUG_ON(!in_softirq()); | |
5462 | ||
d86ee480 | 5463 | if (should_resched()) { |
98d82567 | 5464 | local_bh_enable(); |
1da177e4 LT |
5465 | __cond_resched(); |
5466 | local_bh_disable(); | |
5467 | return 1; | |
5468 | } | |
5469 | return 0; | |
5470 | } | |
613afbf8 | 5471 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5472 | |
1da177e4 LT |
5473 | /** |
5474 | * yield - yield the current processor to other threads. | |
5475 | * | |
72fd4a35 | 5476 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5477 | * thread runnable and calls sys_sched_yield(). |
5478 | */ | |
5479 | void __sched yield(void) | |
5480 | { | |
5481 | set_current_state(TASK_RUNNING); | |
5482 | sys_sched_yield(); | |
5483 | } | |
1da177e4 LT |
5484 | EXPORT_SYMBOL(yield); |
5485 | ||
d95f4122 MG |
5486 | /** |
5487 | * yield_to - yield the current processor to another thread in | |
5488 | * your thread group, or accelerate that thread toward the | |
5489 | * processor it's on. | |
5490 | * | |
5491 | * It's the caller's job to ensure that the target task struct | |
5492 | * can't go away on us before we can do any checks. | |
5493 | * | |
5494 | * Returns true if we indeed boosted the target task. | |
5495 | */ | |
5496 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5497 | { | |
5498 | struct task_struct *curr = current; | |
5499 | struct rq *rq, *p_rq; | |
5500 | unsigned long flags; | |
5501 | bool yielded = 0; | |
5502 | ||
5503 | local_irq_save(flags); | |
5504 | rq = this_rq(); | |
5505 | ||
5506 | again: | |
5507 | p_rq = task_rq(p); | |
5508 | double_rq_lock(rq, p_rq); | |
5509 | while (task_rq(p) != p_rq) { | |
5510 | double_rq_unlock(rq, p_rq); | |
5511 | goto again; | |
5512 | } | |
5513 | ||
5514 | if (!curr->sched_class->yield_to_task) | |
5515 | goto out; | |
5516 | ||
5517 | if (curr->sched_class != p->sched_class) | |
5518 | goto out; | |
5519 | ||
5520 | if (task_running(p_rq, p) || p->state) | |
5521 | goto out; | |
5522 | ||
5523 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
5524 | if (yielded) | |
5525 | schedstat_inc(rq, yld_count); | |
5526 | ||
5527 | out: | |
5528 | double_rq_unlock(rq, p_rq); | |
5529 | local_irq_restore(flags); | |
5530 | ||
5531 | if (yielded) | |
5532 | schedule(); | |
5533 | ||
5534 | return yielded; | |
5535 | } | |
5536 | EXPORT_SYMBOL_GPL(yield_to); | |
5537 | ||
1da177e4 | 5538 | /* |
41a2d6cf | 5539 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5540 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5541 | */ |
5542 | void __sched io_schedule(void) | |
5543 | { | |
54d35f29 | 5544 | struct rq *rq = raw_rq(); |
1da177e4 | 5545 | |
0ff92245 | 5546 | delayacct_blkio_start(); |
1da177e4 | 5547 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5548 | current->in_iowait = 1; |
1da177e4 | 5549 | schedule(); |
8f0dfc34 | 5550 | current->in_iowait = 0; |
1da177e4 | 5551 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5552 | delayacct_blkio_end(); |
1da177e4 | 5553 | } |
1da177e4 LT |
5554 | EXPORT_SYMBOL(io_schedule); |
5555 | ||
5556 | long __sched io_schedule_timeout(long timeout) | |
5557 | { | |
54d35f29 | 5558 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5559 | long ret; |
5560 | ||
0ff92245 | 5561 | delayacct_blkio_start(); |
1da177e4 | 5562 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5563 | current->in_iowait = 1; |
1da177e4 | 5564 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5565 | current->in_iowait = 0; |
1da177e4 | 5566 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5567 | delayacct_blkio_end(); |
1da177e4 LT |
5568 | return ret; |
5569 | } | |
5570 | ||
5571 | /** | |
5572 | * sys_sched_get_priority_max - return maximum RT priority. | |
5573 | * @policy: scheduling class. | |
5574 | * | |
5575 | * this syscall returns the maximum rt_priority that can be used | |
5576 | * by a given scheduling class. | |
5577 | */ | |
5add95d4 | 5578 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5579 | { |
5580 | int ret = -EINVAL; | |
5581 | ||
5582 | switch (policy) { | |
5583 | case SCHED_FIFO: | |
5584 | case SCHED_RR: | |
5585 | ret = MAX_USER_RT_PRIO-1; | |
5586 | break; | |
5587 | case SCHED_NORMAL: | |
b0a9499c | 5588 | case SCHED_BATCH: |
dd41f596 | 5589 | case SCHED_IDLE: |
1da177e4 LT |
5590 | ret = 0; |
5591 | break; | |
5592 | } | |
5593 | return ret; | |
5594 | } | |
5595 | ||
5596 | /** | |
5597 | * sys_sched_get_priority_min - return minimum RT priority. | |
5598 | * @policy: scheduling class. | |
5599 | * | |
5600 | * this syscall returns the minimum rt_priority that can be used | |
5601 | * by a given scheduling class. | |
5602 | */ | |
5add95d4 | 5603 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5604 | { |
5605 | int ret = -EINVAL; | |
5606 | ||
5607 | switch (policy) { | |
5608 | case SCHED_FIFO: | |
5609 | case SCHED_RR: | |
5610 | ret = 1; | |
5611 | break; | |
5612 | case SCHED_NORMAL: | |
b0a9499c | 5613 | case SCHED_BATCH: |
dd41f596 | 5614 | case SCHED_IDLE: |
1da177e4 LT |
5615 | ret = 0; |
5616 | } | |
5617 | return ret; | |
5618 | } | |
5619 | ||
5620 | /** | |
5621 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5622 | * @pid: pid of the process. | |
5623 | * @interval: userspace pointer to the timeslice value. | |
5624 | * | |
5625 | * this syscall writes the default timeslice value of a given process | |
5626 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5627 | */ | |
17da2bd9 | 5628 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5629 | struct timespec __user *, interval) |
1da177e4 | 5630 | { |
36c8b586 | 5631 | struct task_struct *p; |
a4ec24b4 | 5632 | unsigned int time_slice; |
dba091b9 TG |
5633 | unsigned long flags; |
5634 | struct rq *rq; | |
3a5c359a | 5635 | int retval; |
1da177e4 | 5636 | struct timespec t; |
1da177e4 LT |
5637 | |
5638 | if (pid < 0) | |
3a5c359a | 5639 | return -EINVAL; |
1da177e4 LT |
5640 | |
5641 | retval = -ESRCH; | |
1a551ae7 | 5642 | rcu_read_lock(); |
1da177e4 LT |
5643 | p = find_process_by_pid(pid); |
5644 | if (!p) | |
5645 | goto out_unlock; | |
5646 | ||
5647 | retval = security_task_getscheduler(p); | |
5648 | if (retval) | |
5649 | goto out_unlock; | |
5650 | ||
dba091b9 TG |
5651 | rq = task_rq_lock(p, &flags); |
5652 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5653 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5654 | |
1a551ae7 | 5655 | rcu_read_unlock(); |
a4ec24b4 | 5656 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5657 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5658 | return retval; |
3a5c359a | 5659 | |
1da177e4 | 5660 | out_unlock: |
1a551ae7 | 5661 | rcu_read_unlock(); |
1da177e4 LT |
5662 | return retval; |
5663 | } | |
5664 | ||
7c731e0a | 5665 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5666 | |
82a1fcb9 | 5667 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5668 | { |
1da177e4 | 5669 | unsigned long free = 0; |
36c8b586 | 5670 | unsigned state; |
1da177e4 | 5671 | |
1da177e4 | 5672 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5673 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5674 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5675 | #if BITS_PER_LONG == 32 |
1da177e4 | 5676 | if (state == TASK_RUNNING) |
3df0fc5b | 5677 | printk(KERN_CONT " running "); |
1da177e4 | 5678 | else |
3df0fc5b | 5679 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5680 | #else |
5681 | if (state == TASK_RUNNING) | |
3df0fc5b | 5682 | printk(KERN_CONT " running task "); |
1da177e4 | 5683 | else |
3df0fc5b | 5684 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5685 | #endif |
5686 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5687 | free = stack_not_used(p); |
1da177e4 | 5688 | #endif |
3df0fc5b | 5689 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5690 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5691 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5692 | |
5fb5e6de | 5693 | show_stack(p, NULL); |
1da177e4 LT |
5694 | } |
5695 | ||
e59e2ae2 | 5696 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5697 | { |
36c8b586 | 5698 | struct task_struct *g, *p; |
1da177e4 | 5699 | |
4bd77321 | 5700 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5701 | printk(KERN_INFO |
5702 | " task PC stack pid father\n"); | |
1da177e4 | 5703 | #else |
3df0fc5b PZ |
5704 | printk(KERN_INFO |
5705 | " task PC stack pid father\n"); | |
1da177e4 LT |
5706 | #endif |
5707 | read_lock(&tasklist_lock); | |
5708 | do_each_thread(g, p) { | |
5709 | /* | |
5710 | * reset the NMI-timeout, listing all files on a slow | |
5711 | * console might take alot of time: | |
5712 | */ | |
5713 | touch_nmi_watchdog(); | |
39bc89fd | 5714 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5715 | sched_show_task(p); |
1da177e4 LT |
5716 | } while_each_thread(g, p); |
5717 | ||
04c9167f JF |
5718 | touch_all_softlockup_watchdogs(); |
5719 | ||
dd41f596 IM |
5720 | #ifdef CONFIG_SCHED_DEBUG |
5721 | sysrq_sched_debug_show(); | |
5722 | #endif | |
1da177e4 | 5723 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5724 | /* |
5725 | * Only show locks if all tasks are dumped: | |
5726 | */ | |
93335a21 | 5727 | if (!state_filter) |
e59e2ae2 | 5728 | debug_show_all_locks(); |
1da177e4 LT |
5729 | } |
5730 | ||
1df21055 IM |
5731 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5732 | { | |
dd41f596 | 5733 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5734 | } |
5735 | ||
f340c0d1 IM |
5736 | /** |
5737 | * init_idle - set up an idle thread for a given CPU | |
5738 | * @idle: task in question | |
5739 | * @cpu: cpu the idle task belongs to | |
5740 | * | |
5741 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5742 | * flag, to make booting more robust. | |
5743 | */ | |
5c1e1767 | 5744 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5745 | { |
70b97a7f | 5746 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5747 | unsigned long flags; |
5748 | ||
05fa785c | 5749 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5750 | |
dd41f596 | 5751 | __sched_fork(idle); |
06b83b5f | 5752 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5753 | idle->se.exec_start = sched_clock(); |
5754 | ||
96f874e2 | 5755 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5756 | /* |
5757 | * We're having a chicken and egg problem, even though we are | |
5758 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5759 | * lockdep check in task_group() will fail. | |
5760 | * | |
5761 | * Similar case to sched_fork(). / Alternatively we could | |
5762 | * use task_rq_lock() here and obtain the other rq->lock. | |
5763 | * | |
5764 | * Silence PROVE_RCU | |
5765 | */ | |
5766 | rcu_read_lock(); | |
dd41f596 | 5767 | __set_task_cpu(idle, cpu); |
6506cf6c | 5768 | rcu_read_unlock(); |
1da177e4 | 5769 | |
1da177e4 | 5770 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5771 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5772 | idle->oncpu = 1; | |
5773 | #endif | |
05fa785c | 5774 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5775 | |
5776 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5777 | #if defined(CONFIG_PREEMPT) |
5778 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5779 | #else | |
a1261f54 | 5780 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5781 | #endif |
dd41f596 IM |
5782 | /* |
5783 | * The idle tasks have their own, simple scheduling class: | |
5784 | */ | |
5785 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5786 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5787 | } |
5788 | ||
5789 | /* | |
5790 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5791 | * indicates which cpus entered this state. This is used | |
5792 | * in the rcu update to wait only for active cpus. For system | |
5793 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5794 | * always be CPU_BITS_NONE. |
1da177e4 | 5795 | */ |
6a7b3dc3 | 5796 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5797 | |
19978ca6 IM |
5798 | /* |
5799 | * Increase the granularity value when there are more CPUs, | |
5800 | * because with more CPUs the 'effective latency' as visible | |
5801 | * to users decreases. But the relationship is not linear, | |
5802 | * so pick a second-best guess by going with the log2 of the | |
5803 | * number of CPUs. | |
5804 | * | |
5805 | * This idea comes from the SD scheduler of Con Kolivas: | |
5806 | */ | |
acb4a848 | 5807 | static int get_update_sysctl_factor(void) |
19978ca6 | 5808 | { |
4ca3ef71 | 5809 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5810 | unsigned int factor; |
5811 | ||
5812 | switch (sysctl_sched_tunable_scaling) { | |
5813 | case SCHED_TUNABLESCALING_NONE: | |
5814 | factor = 1; | |
5815 | break; | |
5816 | case SCHED_TUNABLESCALING_LINEAR: | |
5817 | factor = cpus; | |
5818 | break; | |
5819 | case SCHED_TUNABLESCALING_LOG: | |
5820 | default: | |
5821 | factor = 1 + ilog2(cpus); | |
5822 | break; | |
5823 | } | |
19978ca6 | 5824 | |
acb4a848 CE |
5825 | return factor; |
5826 | } | |
19978ca6 | 5827 | |
acb4a848 CE |
5828 | static void update_sysctl(void) |
5829 | { | |
5830 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5831 | |
0bcdcf28 CE |
5832 | #define SET_SYSCTL(name) \ |
5833 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5834 | SET_SYSCTL(sched_min_granularity); | |
5835 | SET_SYSCTL(sched_latency); | |
5836 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5837 | #undef SET_SYSCTL |
5838 | } | |
55cd5340 | 5839 | |
0bcdcf28 CE |
5840 | static inline void sched_init_granularity(void) |
5841 | { | |
5842 | update_sysctl(); | |
19978ca6 IM |
5843 | } |
5844 | ||
1da177e4 LT |
5845 | #ifdef CONFIG_SMP |
5846 | /* | |
5847 | * This is how migration works: | |
5848 | * | |
969c7921 TH |
5849 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5850 | * stop_one_cpu(). | |
5851 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5852 | * off the CPU) | |
5853 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5854 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5855 | * it and puts it into the right queue. |
969c7921 TH |
5856 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5857 | * is done. | |
1da177e4 LT |
5858 | */ |
5859 | ||
5860 | /* | |
5861 | * Change a given task's CPU affinity. Migrate the thread to a | |
5862 | * proper CPU and schedule it away if the CPU it's executing on | |
5863 | * is removed from the allowed bitmask. | |
5864 | * | |
5865 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5866 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5867 | * call is not atomic; no spinlocks may be held. |
5868 | */ | |
96f874e2 | 5869 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5870 | { |
5871 | unsigned long flags; | |
70b97a7f | 5872 | struct rq *rq; |
969c7921 | 5873 | unsigned int dest_cpu; |
48f24c4d | 5874 | int ret = 0; |
1da177e4 | 5875 | |
65cc8e48 PZ |
5876 | /* |
5877 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5878 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5879 | */ | |
5880 | again: | |
5881 | while (task_is_waking(p)) | |
5882 | cpu_relax(); | |
1da177e4 | 5883 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5884 | if (task_is_waking(p)) { |
5885 | task_rq_unlock(rq, &flags); | |
5886 | goto again; | |
5887 | } | |
e2912009 | 5888 | |
6ad4c188 | 5889 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5890 | ret = -EINVAL; |
5891 | goto out; | |
5892 | } | |
5893 | ||
9985b0ba | 5894 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5895 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5896 | ret = -EINVAL; |
5897 | goto out; | |
5898 | } | |
5899 | ||
73fe6aae | 5900 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5901 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5902 | else { |
96f874e2 RR |
5903 | cpumask_copy(&p->cpus_allowed, new_mask); |
5904 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5905 | } |
5906 | ||
1da177e4 | 5907 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5908 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5909 | goto out; |
5910 | ||
969c7921 | 5911 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5912 | if (migrate_task(p, rq)) { |
969c7921 | 5913 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5914 | /* Need help from migration thread: drop lock and wait. */ |
5915 | task_rq_unlock(rq, &flags); | |
969c7921 | 5916 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5917 | tlb_migrate_finish(p->mm); |
5918 | return 0; | |
5919 | } | |
5920 | out: | |
5921 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5922 | |
1da177e4 LT |
5923 | return ret; |
5924 | } | |
cd8ba7cd | 5925 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5926 | |
5927 | /* | |
41a2d6cf | 5928 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5929 | * this because either it can't run here any more (set_cpus_allowed() |
5930 | * away from this CPU, or CPU going down), or because we're | |
5931 | * attempting to rebalance this task on exec (sched_exec). | |
5932 | * | |
5933 | * So we race with normal scheduler movements, but that's OK, as long | |
5934 | * as the task is no longer on this CPU. | |
efc30814 KK |
5935 | * |
5936 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5937 | */ |
efc30814 | 5938 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5939 | { |
70b97a7f | 5940 | struct rq *rq_dest, *rq_src; |
e2912009 | 5941 | int ret = 0; |
1da177e4 | 5942 | |
e761b772 | 5943 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5944 | return ret; |
1da177e4 LT |
5945 | |
5946 | rq_src = cpu_rq(src_cpu); | |
5947 | rq_dest = cpu_rq(dest_cpu); | |
5948 | ||
5949 | double_rq_lock(rq_src, rq_dest); | |
5950 | /* Already moved. */ | |
5951 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5952 | goto done; |
1da177e4 | 5953 | /* Affinity changed (again). */ |
96f874e2 | 5954 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5955 | goto fail; |
1da177e4 | 5956 | |
e2912009 PZ |
5957 | /* |
5958 | * If we're not on a rq, the next wake-up will ensure we're | |
5959 | * placed properly. | |
5960 | */ | |
5961 | if (p->se.on_rq) { | |
2e1cb74a | 5962 | deactivate_task(rq_src, p, 0); |
e2912009 | 5963 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5964 | activate_task(rq_dest, p, 0); |
15afe09b | 5965 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5966 | } |
b1e38734 | 5967 | done: |
efc30814 | 5968 | ret = 1; |
b1e38734 | 5969 | fail: |
1da177e4 | 5970 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5971 | return ret; |
1da177e4 LT |
5972 | } |
5973 | ||
5974 | /* | |
969c7921 TH |
5975 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5976 | * and performs thread migration by bumping thread off CPU then | |
5977 | * 'pushing' onto another runqueue. | |
1da177e4 | 5978 | */ |
969c7921 | 5979 | static int migration_cpu_stop(void *data) |
1da177e4 | 5980 | { |
969c7921 | 5981 | struct migration_arg *arg = data; |
f7b4cddc | 5982 | |
969c7921 TH |
5983 | /* |
5984 | * The original target cpu might have gone down and we might | |
5985 | * be on another cpu but it doesn't matter. | |
5986 | */ | |
f7b4cddc | 5987 | local_irq_disable(); |
969c7921 | 5988 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5989 | local_irq_enable(); |
1da177e4 | 5990 | return 0; |
f7b4cddc ON |
5991 | } |
5992 | ||
1da177e4 | 5993 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5994 | |
054b9108 | 5995 | /* |
48c5ccae PZ |
5996 | * Ensures that the idle task is using init_mm right before its cpu goes |
5997 | * offline. | |
054b9108 | 5998 | */ |
48c5ccae | 5999 | void idle_task_exit(void) |
1da177e4 | 6000 | { |
48c5ccae | 6001 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6002 | |
48c5ccae | 6003 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6004 | |
48c5ccae PZ |
6005 | if (mm != &init_mm) |
6006 | switch_mm(mm, &init_mm, current); | |
6007 | mmdrop(mm); | |
1da177e4 LT |
6008 | } |
6009 | ||
6010 | /* | |
6011 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6012 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6013 | * for performance reasons the counter is not stricly tracking tasks to | |
6014 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6015 | * to keep the global sum constant after CPU-down: | |
6016 | */ | |
70b97a7f | 6017 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6018 | { |
6ad4c188 | 6019 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6020 | |
1da177e4 LT |
6021 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6022 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6023 | } |
6024 | ||
dd41f596 | 6025 | /* |
48c5ccae | 6026 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6027 | */ |
48c5ccae | 6028 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6029 | { |
48c5ccae PZ |
6030 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6031 | rq->calc_load_active = 0; | |
1da177e4 LT |
6032 | } |
6033 | ||
48f24c4d | 6034 | /* |
48c5ccae PZ |
6035 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6036 | * try_to_wake_up()->select_task_rq(). | |
6037 | * | |
6038 | * Called with rq->lock held even though we'er in stop_machine() and | |
6039 | * there's no concurrency possible, we hold the required locks anyway | |
6040 | * because of lock validation efforts. | |
1da177e4 | 6041 | */ |
48c5ccae | 6042 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6043 | { |
70b97a7f | 6044 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6045 | struct task_struct *next, *stop = rq->stop; |
6046 | int dest_cpu; | |
1da177e4 LT |
6047 | |
6048 | /* | |
48c5ccae PZ |
6049 | * Fudge the rq selection such that the below task selection loop |
6050 | * doesn't get stuck on the currently eligible stop task. | |
6051 | * | |
6052 | * We're currently inside stop_machine() and the rq is either stuck | |
6053 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6054 | * either way we should never end up calling schedule() until we're | |
6055 | * done here. | |
1da177e4 | 6056 | */ |
48c5ccae | 6057 | rq->stop = NULL; |
48f24c4d | 6058 | |
dd41f596 | 6059 | for ( ; ; ) { |
48c5ccae PZ |
6060 | /* |
6061 | * There's this thread running, bail when that's the only | |
6062 | * remaining thread. | |
6063 | */ | |
6064 | if (rq->nr_running == 1) | |
dd41f596 | 6065 | break; |
48c5ccae | 6066 | |
b67802ea | 6067 | next = pick_next_task(rq); |
48c5ccae | 6068 | BUG_ON(!next); |
79c53799 | 6069 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6070 | |
48c5ccae PZ |
6071 | /* Find suitable destination for @next, with force if needed. */ |
6072 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6073 | raw_spin_unlock(&rq->lock); | |
6074 | ||
6075 | __migrate_task(next, dead_cpu, dest_cpu); | |
6076 | ||
6077 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6078 | } |
dce48a84 | 6079 | |
48c5ccae | 6080 | rq->stop = stop; |
dce48a84 | 6081 | } |
48c5ccae | 6082 | |
1da177e4 LT |
6083 | #endif /* CONFIG_HOTPLUG_CPU */ |
6084 | ||
e692ab53 NP |
6085 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6086 | ||
6087 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6088 | { |
6089 | .procname = "sched_domain", | |
c57baf1e | 6090 | .mode = 0555, |
e0361851 | 6091 | }, |
56992309 | 6092 | {} |
e692ab53 NP |
6093 | }; |
6094 | ||
6095 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6096 | { |
6097 | .procname = "kernel", | |
c57baf1e | 6098 | .mode = 0555, |
e0361851 AD |
6099 | .child = sd_ctl_dir, |
6100 | }, | |
56992309 | 6101 | {} |
e692ab53 NP |
6102 | }; |
6103 | ||
6104 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6105 | { | |
6106 | struct ctl_table *entry = | |
5cf9f062 | 6107 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6108 | |
e692ab53 NP |
6109 | return entry; |
6110 | } | |
6111 | ||
6382bc90 MM |
6112 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6113 | { | |
cd790076 | 6114 | struct ctl_table *entry; |
6382bc90 | 6115 | |
cd790076 MM |
6116 | /* |
6117 | * In the intermediate directories, both the child directory and | |
6118 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6119 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6120 | * static strings and all have proc handlers. |
6121 | */ | |
6122 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6123 | if (entry->child) |
6124 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6125 | if (entry->proc_handler == NULL) |
6126 | kfree(entry->procname); | |
6127 | } | |
6382bc90 MM |
6128 | |
6129 | kfree(*tablep); | |
6130 | *tablep = NULL; | |
6131 | } | |
6132 | ||
e692ab53 | 6133 | static void |
e0361851 | 6134 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6135 | const char *procname, void *data, int maxlen, |
6136 | mode_t mode, proc_handler *proc_handler) | |
6137 | { | |
e692ab53 NP |
6138 | entry->procname = procname; |
6139 | entry->data = data; | |
6140 | entry->maxlen = maxlen; | |
6141 | entry->mode = mode; | |
6142 | entry->proc_handler = proc_handler; | |
6143 | } | |
6144 | ||
6145 | static struct ctl_table * | |
6146 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6147 | { | |
a5d8c348 | 6148 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6149 | |
ad1cdc1d MM |
6150 | if (table == NULL) |
6151 | return NULL; | |
6152 | ||
e0361851 | 6153 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6154 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6155 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6156 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6157 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6158 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6159 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6160 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6161 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6162 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6163 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6164 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6165 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6166 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6167 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6168 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6169 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6170 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6171 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6172 | &sd->cache_nice_tries, |
6173 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6174 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6175 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6176 | set_table_entry(&table[11], "name", sd->name, |
6177 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6178 | /* &table[12] is terminator */ | |
e692ab53 NP |
6179 | |
6180 | return table; | |
6181 | } | |
6182 | ||
9a4e7159 | 6183 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6184 | { |
6185 | struct ctl_table *entry, *table; | |
6186 | struct sched_domain *sd; | |
6187 | int domain_num = 0, i; | |
6188 | char buf[32]; | |
6189 | ||
6190 | for_each_domain(cpu, sd) | |
6191 | domain_num++; | |
6192 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6193 | if (table == NULL) |
6194 | return NULL; | |
e692ab53 NP |
6195 | |
6196 | i = 0; | |
6197 | for_each_domain(cpu, sd) { | |
6198 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6199 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6200 | entry->mode = 0555; |
e692ab53 NP |
6201 | entry->child = sd_alloc_ctl_domain_table(sd); |
6202 | entry++; | |
6203 | i++; | |
6204 | } | |
6205 | return table; | |
6206 | } | |
6207 | ||
6208 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6209 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6210 | { |
6ad4c188 | 6211 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6212 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6213 | char buf[32]; | |
6214 | ||
7378547f MM |
6215 | WARN_ON(sd_ctl_dir[0].child); |
6216 | sd_ctl_dir[0].child = entry; | |
6217 | ||
ad1cdc1d MM |
6218 | if (entry == NULL) |
6219 | return; | |
6220 | ||
6ad4c188 | 6221 | for_each_possible_cpu(i) { |
e692ab53 | 6222 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6223 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6224 | entry->mode = 0555; |
e692ab53 | 6225 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6226 | entry++; |
e692ab53 | 6227 | } |
7378547f MM |
6228 | |
6229 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6230 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6231 | } | |
6382bc90 | 6232 | |
7378547f | 6233 | /* may be called multiple times per register */ |
6382bc90 MM |
6234 | static void unregister_sched_domain_sysctl(void) |
6235 | { | |
7378547f MM |
6236 | if (sd_sysctl_header) |
6237 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6238 | sd_sysctl_header = NULL; |
7378547f MM |
6239 | if (sd_ctl_dir[0].child) |
6240 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6241 | } |
e692ab53 | 6242 | #else |
6382bc90 MM |
6243 | static void register_sched_domain_sysctl(void) |
6244 | { | |
6245 | } | |
6246 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6247 | { |
6248 | } | |
6249 | #endif | |
6250 | ||
1f11eb6a GH |
6251 | static void set_rq_online(struct rq *rq) |
6252 | { | |
6253 | if (!rq->online) { | |
6254 | const struct sched_class *class; | |
6255 | ||
c6c4927b | 6256 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6257 | rq->online = 1; |
6258 | ||
6259 | for_each_class(class) { | |
6260 | if (class->rq_online) | |
6261 | class->rq_online(rq); | |
6262 | } | |
6263 | } | |
6264 | } | |
6265 | ||
6266 | static void set_rq_offline(struct rq *rq) | |
6267 | { | |
6268 | if (rq->online) { | |
6269 | const struct sched_class *class; | |
6270 | ||
6271 | for_each_class(class) { | |
6272 | if (class->rq_offline) | |
6273 | class->rq_offline(rq); | |
6274 | } | |
6275 | ||
c6c4927b | 6276 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6277 | rq->online = 0; |
6278 | } | |
6279 | } | |
6280 | ||
1da177e4 LT |
6281 | /* |
6282 | * migration_call - callback that gets triggered when a CPU is added. | |
6283 | * Here we can start up the necessary migration thread for the new CPU. | |
6284 | */ | |
48f24c4d IM |
6285 | static int __cpuinit |
6286 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6287 | { |
48f24c4d | 6288 | int cpu = (long)hcpu; |
1da177e4 | 6289 | unsigned long flags; |
969c7921 | 6290 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6291 | |
48c5ccae | 6292 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6293 | |
1da177e4 | 6294 | case CPU_UP_PREPARE: |
a468d389 | 6295 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6296 | break; |
48f24c4d | 6297 | |
1da177e4 | 6298 | case CPU_ONLINE: |
1f94ef59 | 6299 | /* Update our root-domain */ |
05fa785c | 6300 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6301 | if (rq->rd) { |
c6c4927b | 6302 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6303 | |
6304 | set_rq_online(rq); | |
1f94ef59 | 6305 | } |
05fa785c | 6306 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6307 | break; |
48f24c4d | 6308 | |
1da177e4 | 6309 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6310 | case CPU_DYING: |
57d885fe | 6311 | /* Update our root-domain */ |
05fa785c | 6312 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6313 | if (rq->rd) { |
c6c4927b | 6314 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6315 | set_rq_offline(rq); |
57d885fe | 6316 | } |
48c5ccae PZ |
6317 | migrate_tasks(cpu); |
6318 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6319 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6320 | |
6321 | migrate_nr_uninterruptible(rq); | |
6322 | calc_global_load_remove(rq); | |
57d885fe | 6323 | break; |
1da177e4 LT |
6324 | #endif |
6325 | } | |
6326 | return NOTIFY_OK; | |
6327 | } | |
6328 | ||
f38b0820 PM |
6329 | /* |
6330 | * Register at high priority so that task migration (migrate_all_tasks) | |
6331 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6332 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6333 | */ |
26c2143b | 6334 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6335 | .notifier_call = migration_call, |
50a323b7 | 6336 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6337 | }; |
6338 | ||
3a101d05 TH |
6339 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6340 | unsigned long action, void *hcpu) | |
6341 | { | |
6342 | switch (action & ~CPU_TASKS_FROZEN) { | |
6343 | case CPU_ONLINE: | |
6344 | case CPU_DOWN_FAILED: | |
6345 | set_cpu_active((long)hcpu, true); | |
6346 | return NOTIFY_OK; | |
6347 | default: | |
6348 | return NOTIFY_DONE; | |
6349 | } | |
6350 | } | |
6351 | ||
6352 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6353 | unsigned long action, void *hcpu) | |
6354 | { | |
6355 | switch (action & ~CPU_TASKS_FROZEN) { | |
6356 | case CPU_DOWN_PREPARE: | |
6357 | set_cpu_active((long)hcpu, false); | |
6358 | return NOTIFY_OK; | |
6359 | default: | |
6360 | return NOTIFY_DONE; | |
6361 | } | |
6362 | } | |
6363 | ||
7babe8db | 6364 | static int __init migration_init(void) |
1da177e4 LT |
6365 | { |
6366 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6367 | int err; |
48f24c4d | 6368 | |
3a101d05 | 6369 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6370 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6371 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6372 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6373 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6374 | |
3a101d05 TH |
6375 | /* Register cpu active notifiers */ |
6376 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6377 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6378 | ||
a004cd42 | 6379 | return 0; |
1da177e4 | 6380 | } |
7babe8db | 6381 | early_initcall(migration_init); |
1da177e4 LT |
6382 | #endif |
6383 | ||
6384 | #ifdef CONFIG_SMP | |
476f3534 | 6385 | |
3e9830dc | 6386 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6387 | |
f6630114 MT |
6388 | static __read_mostly int sched_domain_debug_enabled; |
6389 | ||
6390 | static int __init sched_domain_debug_setup(char *str) | |
6391 | { | |
6392 | sched_domain_debug_enabled = 1; | |
6393 | ||
6394 | return 0; | |
6395 | } | |
6396 | early_param("sched_debug", sched_domain_debug_setup); | |
6397 | ||
7c16ec58 | 6398 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6399 | struct cpumask *groupmask) |
1da177e4 | 6400 | { |
4dcf6aff | 6401 | struct sched_group *group = sd->groups; |
434d53b0 | 6402 | char str[256]; |
1da177e4 | 6403 | |
968ea6d8 | 6404 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6405 | cpumask_clear(groupmask); |
4dcf6aff IM |
6406 | |
6407 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6408 | ||
6409 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6410 | printk("does not load-balance\n"); |
4dcf6aff | 6411 | if (sd->parent) |
3df0fc5b PZ |
6412 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6413 | " has parent"); | |
4dcf6aff | 6414 | return -1; |
41c7ce9a NP |
6415 | } |
6416 | ||
3df0fc5b | 6417 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6418 | |
758b2cdc | 6419 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6420 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6421 | "CPU%d\n", cpu); | |
4dcf6aff | 6422 | } |
758b2cdc | 6423 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6424 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6425 | " CPU%d\n", cpu); | |
4dcf6aff | 6426 | } |
1da177e4 | 6427 | |
4dcf6aff | 6428 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6429 | do { |
4dcf6aff | 6430 | if (!group) { |
3df0fc5b PZ |
6431 | printk("\n"); |
6432 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6433 | break; |
6434 | } | |
6435 | ||
18a3885f | 6436 | if (!group->cpu_power) { |
3df0fc5b PZ |
6437 | printk(KERN_CONT "\n"); |
6438 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6439 | "set\n"); | |
4dcf6aff IM |
6440 | break; |
6441 | } | |
1da177e4 | 6442 | |
758b2cdc | 6443 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6444 | printk(KERN_CONT "\n"); |
6445 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6446 | break; |
6447 | } | |
1da177e4 | 6448 | |
758b2cdc | 6449 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6450 | printk(KERN_CONT "\n"); |
6451 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6452 | break; |
6453 | } | |
1da177e4 | 6454 | |
758b2cdc | 6455 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6456 | |
968ea6d8 | 6457 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6458 | |
3df0fc5b | 6459 | printk(KERN_CONT " %s", str); |
18a3885f | 6460 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6461 | printk(KERN_CONT " (cpu_power = %d)", |
6462 | group->cpu_power); | |
381512cf | 6463 | } |
1da177e4 | 6464 | |
4dcf6aff IM |
6465 | group = group->next; |
6466 | } while (group != sd->groups); | |
3df0fc5b | 6467 | printk(KERN_CONT "\n"); |
1da177e4 | 6468 | |
758b2cdc | 6469 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6470 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6471 | |
758b2cdc RR |
6472 | if (sd->parent && |
6473 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6474 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6475 | "of domain->span\n"); | |
4dcf6aff IM |
6476 | return 0; |
6477 | } | |
1da177e4 | 6478 | |
4dcf6aff IM |
6479 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6480 | { | |
d5dd3db1 | 6481 | cpumask_var_t groupmask; |
4dcf6aff | 6482 | int level = 0; |
1da177e4 | 6483 | |
f6630114 MT |
6484 | if (!sched_domain_debug_enabled) |
6485 | return; | |
6486 | ||
4dcf6aff IM |
6487 | if (!sd) { |
6488 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6489 | return; | |
6490 | } | |
1da177e4 | 6491 | |
4dcf6aff IM |
6492 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6493 | ||
d5dd3db1 | 6494 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6495 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6496 | return; | |
6497 | } | |
6498 | ||
4dcf6aff | 6499 | for (;;) { |
7c16ec58 | 6500 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6501 | break; |
1da177e4 LT |
6502 | level++; |
6503 | sd = sd->parent; | |
33859f7f | 6504 | if (!sd) |
4dcf6aff IM |
6505 | break; |
6506 | } | |
d5dd3db1 | 6507 | free_cpumask_var(groupmask); |
1da177e4 | 6508 | } |
6d6bc0ad | 6509 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6510 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6511 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6512 | |
1a20ff27 | 6513 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6514 | { |
758b2cdc | 6515 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6516 | return 1; |
6517 | ||
6518 | /* Following flags need at least 2 groups */ | |
6519 | if (sd->flags & (SD_LOAD_BALANCE | | |
6520 | SD_BALANCE_NEWIDLE | | |
6521 | SD_BALANCE_FORK | | |
89c4710e SS |
6522 | SD_BALANCE_EXEC | |
6523 | SD_SHARE_CPUPOWER | | |
6524 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6525 | if (sd->groups != sd->groups->next) |
6526 | return 0; | |
6527 | } | |
6528 | ||
6529 | /* Following flags don't use groups */ | |
c88d5910 | 6530 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6531 | return 0; |
6532 | ||
6533 | return 1; | |
6534 | } | |
6535 | ||
48f24c4d IM |
6536 | static int |
6537 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6538 | { |
6539 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6540 | ||
6541 | if (sd_degenerate(parent)) | |
6542 | return 1; | |
6543 | ||
758b2cdc | 6544 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6545 | return 0; |
6546 | ||
245af2c7 SS |
6547 | /* Flags needing groups don't count if only 1 group in parent */ |
6548 | if (parent->groups == parent->groups->next) { | |
6549 | pflags &= ~(SD_LOAD_BALANCE | | |
6550 | SD_BALANCE_NEWIDLE | | |
6551 | SD_BALANCE_FORK | | |
89c4710e SS |
6552 | SD_BALANCE_EXEC | |
6553 | SD_SHARE_CPUPOWER | | |
6554 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6555 | if (nr_node_ids == 1) |
6556 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6557 | } |
6558 | if (~cflags & pflags) | |
6559 | return 0; | |
6560 | ||
6561 | return 1; | |
6562 | } | |
6563 | ||
c6c4927b RR |
6564 | static void free_rootdomain(struct root_domain *rd) |
6565 | { | |
047106ad PZ |
6566 | synchronize_sched(); |
6567 | ||
68e74568 RR |
6568 | cpupri_cleanup(&rd->cpupri); |
6569 | ||
c6c4927b RR |
6570 | free_cpumask_var(rd->rto_mask); |
6571 | free_cpumask_var(rd->online); | |
6572 | free_cpumask_var(rd->span); | |
6573 | kfree(rd); | |
6574 | } | |
6575 | ||
57d885fe GH |
6576 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6577 | { | |
a0490fa3 | 6578 | struct root_domain *old_rd = NULL; |
57d885fe | 6579 | unsigned long flags; |
57d885fe | 6580 | |
05fa785c | 6581 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6582 | |
6583 | if (rq->rd) { | |
a0490fa3 | 6584 | old_rd = rq->rd; |
57d885fe | 6585 | |
c6c4927b | 6586 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6587 | set_rq_offline(rq); |
57d885fe | 6588 | |
c6c4927b | 6589 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6590 | |
a0490fa3 IM |
6591 | /* |
6592 | * If we dont want to free the old_rt yet then | |
6593 | * set old_rd to NULL to skip the freeing later | |
6594 | * in this function: | |
6595 | */ | |
6596 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6597 | old_rd = NULL; | |
57d885fe GH |
6598 | } |
6599 | ||
6600 | atomic_inc(&rd->refcount); | |
6601 | rq->rd = rd; | |
6602 | ||
c6c4927b | 6603 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6604 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6605 | set_rq_online(rq); |
57d885fe | 6606 | |
05fa785c | 6607 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6608 | |
6609 | if (old_rd) | |
6610 | free_rootdomain(old_rd); | |
57d885fe GH |
6611 | } |
6612 | ||
68c38fc3 | 6613 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6614 | { |
6615 | memset(rd, 0, sizeof(*rd)); | |
6616 | ||
68c38fc3 | 6617 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6618 | goto out; |
68c38fc3 | 6619 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6620 | goto free_span; |
68c38fc3 | 6621 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6622 | goto free_online; |
6e0534f2 | 6623 | |
68c38fc3 | 6624 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6625 | goto free_rto_mask; |
c6c4927b | 6626 | return 0; |
6e0534f2 | 6627 | |
68e74568 RR |
6628 | free_rto_mask: |
6629 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6630 | free_online: |
6631 | free_cpumask_var(rd->online); | |
6632 | free_span: | |
6633 | free_cpumask_var(rd->span); | |
0c910d28 | 6634 | out: |
c6c4927b | 6635 | return -ENOMEM; |
57d885fe GH |
6636 | } |
6637 | ||
6638 | static void init_defrootdomain(void) | |
6639 | { | |
68c38fc3 | 6640 | init_rootdomain(&def_root_domain); |
c6c4927b | 6641 | |
57d885fe GH |
6642 | atomic_set(&def_root_domain.refcount, 1); |
6643 | } | |
6644 | ||
dc938520 | 6645 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6646 | { |
6647 | struct root_domain *rd; | |
6648 | ||
6649 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6650 | if (!rd) | |
6651 | return NULL; | |
6652 | ||
68c38fc3 | 6653 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6654 | kfree(rd); |
6655 | return NULL; | |
6656 | } | |
57d885fe GH |
6657 | |
6658 | return rd; | |
6659 | } | |
6660 | ||
1da177e4 | 6661 | /* |
0eab9146 | 6662 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6663 | * hold the hotplug lock. |
6664 | */ | |
0eab9146 IM |
6665 | static void |
6666 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6667 | { |
70b97a7f | 6668 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6669 | struct sched_domain *tmp; |
6670 | ||
669c55e9 PZ |
6671 | for (tmp = sd; tmp; tmp = tmp->parent) |
6672 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6673 | ||
245af2c7 | 6674 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6675 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6676 | struct sched_domain *parent = tmp->parent; |
6677 | if (!parent) | |
6678 | break; | |
f29c9b1c | 6679 | |
1a848870 | 6680 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6681 | tmp->parent = parent->parent; |
1a848870 SS |
6682 | if (parent->parent) |
6683 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6684 | } else |
6685 | tmp = tmp->parent; | |
245af2c7 SS |
6686 | } |
6687 | ||
1a848870 | 6688 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6689 | sd = sd->parent; |
1a848870 SS |
6690 | if (sd) |
6691 | sd->child = NULL; | |
6692 | } | |
1da177e4 LT |
6693 | |
6694 | sched_domain_debug(sd, cpu); | |
6695 | ||
57d885fe | 6696 | rq_attach_root(rq, rd); |
674311d5 | 6697 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6698 | } |
6699 | ||
6700 | /* cpus with isolated domains */ | |
dcc30a35 | 6701 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6702 | |
6703 | /* Setup the mask of cpus configured for isolated domains */ | |
6704 | static int __init isolated_cpu_setup(char *str) | |
6705 | { | |
bdddd296 | 6706 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6707 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6708 | return 1; |
6709 | } | |
6710 | ||
8927f494 | 6711 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6712 | |
6713 | /* | |
6711cab4 SS |
6714 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6715 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6716 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6717 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6718 | * |
6719 | * init_sched_build_groups will build a circular linked list of the groups | |
6720 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6721 | * and ->cpu_power to 0. | |
6722 | */ | |
a616058b | 6723 | static void |
96f874e2 RR |
6724 | init_sched_build_groups(const struct cpumask *span, |
6725 | const struct cpumask *cpu_map, | |
6726 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6727 | struct sched_group **sg, |
96f874e2 RR |
6728 | struct cpumask *tmpmask), |
6729 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6730 | { |
6731 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6732 | int i; |
6733 | ||
96f874e2 | 6734 | cpumask_clear(covered); |
7c16ec58 | 6735 | |
abcd083a | 6736 | for_each_cpu(i, span) { |
6711cab4 | 6737 | struct sched_group *sg; |
7c16ec58 | 6738 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6739 | int j; |
6740 | ||
758b2cdc | 6741 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6742 | continue; |
6743 | ||
758b2cdc | 6744 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6745 | sg->cpu_power = 0; |
1da177e4 | 6746 | |
abcd083a | 6747 | for_each_cpu(j, span) { |
7c16ec58 | 6748 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6749 | continue; |
6750 | ||
96f874e2 | 6751 | cpumask_set_cpu(j, covered); |
758b2cdc | 6752 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6753 | } |
6754 | if (!first) | |
6755 | first = sg; | |
6756 | if (last) | |
6757 | last->next = sg; | |
6758 | last = sg; | |
6759 | } | |
6760 | last->next = first; | |
6761 | } | |
6762 | ||
9c1cfda2 | 6763 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6764 | |
9c1cfda2 | 6765 | #ifdef CONFIG_NUMA |
198e2f18 | 6766 | |
9c1cfda2 JH |
6767 | /** |
6768 | * find_next_best_node - find the next node to include in a sched_domain | |
6769 | * @node: node whose sched_domain we're building | |
6770 | * @used_nodes: nodes already in the sched_domain | |
6771 | * | |
41a2d6cf | 6772 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6773 | * finds the closest node not already in the @used_nodes map. |
6774 | * | |
6775 | * Should use nodemask_t. | |
6776 | */ | |
c5f59f08 | 6777 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6778 | { |
6779 | int i, n, val, min_val, best_node = 0; | |
6780 | ||
6781 | min_val = INT_MAX; | |
6782 | ||
076ac2af | 6783 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6784 | /* Start at @node */ |
076ac2af | 6785 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6786 | |
6787 | if (!nr_cpus_node(n)) | |
6788 | continue; | |
6789 | ||
6790 | /* Skip already used nodes */ | |
c5f59f08 | 6791 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6792 | continue; |
6793 | ||
6794 | /* Simple min distance search */ | |
6795 | val = node_distance(node, n); | |
6796 | ||
6797 | if (val < min_val) { | |
6798 | min_val = val; | |
6799 | best_node = n; | |
6800 | } | |
6801 | } | |
6802 | ||
c5f59f08 | 6803 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6804 | return best_node; |
6805 | } | |
6806 | ||
6807 | /** | |
6808 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6809 | * @node: node whose cpumask we're constructing | |
73486722 | 6810 | * @span: resulting cpumask |
9c1cfda2 | 6811 | * |
41a2d6cf | 6812 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6813 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6814 | * out optimally. | |
6815 | */ | |
96f874e2 | 6816 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6817 | { |
c5f59f08 | 6818 | nodemask_t used_nodes; |
48f24c4d | 6819 | int i; |
9c1cfda2 | 6820 | |
6ca09dfc | 6821 | cpumask_clear(span); |
c5f59f08 | 6822 | nodes_clear(used_nodes); |
9c1cfda2 | 6823 | |
6ca09dfc | 6824 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6825 | node_set(node, used_nodes); |
9c1cfda2 JH |
6826 | |
6827 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6828 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6829 | |
6ca09dfc | 6830 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6831 | } |
9c1cfda2 | 6832 | } |
6d6bc0ad | 6833 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6834 | |
5c45bf27 | 6835 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6836 | |
6c99e9ad RR |
6837 | /* |
6838 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6839 | * |
6840 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6841 | * and struct sched_domain. ) | |
6c99e9ad RR |
6842 | */ |
6843 | struct static_sched_group { | |
6844 | struct sched_group sg; | |
6845 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6846 | }; | |
6847 | ||
6848 | struct static_sched_domain { | |
6849 | struct sched_domain sd; | |
6850 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6851 | }; | |
6852 | ||
49a02c51 AH |
6853 | struct s_data { |
6854 | #ifdef CONFIG_NUMA | |
6855 | int sd_allnodes; | |
6856 | cpumask_var_t domainspan; | |
6857 | cpumask_var_t covered; | |
6858 | cpumask_var_t notcovered; | |
6859 | #endif | |
6860 | cpumask_var_t nodemask; | |
6861 | cpumask_var_t this_sibling_map; | |
6862 | cpumask_var_t this_core_map; | |
01a08546 | 6863 | cpumask_var_t this_book_map; |
49a02c51 AH |
6864 | cpumask_var_t send_covered; |
6865 | cpumask_var_t tmpmask; | |
6866 | struct sched_group **sched_group_nodes; | |
6867 | struct root_domain *rd; | |
6868 | }; | |
6869 | ||
2109b99e AH |
6870 | enum s_alloc { |
6871 | sa_sched_groups = 0, | |
6872 | sa_rootdomain, | |
6873 | sa_tmpmask, | |
6874 | sa_send_covered, | |
01a08546 | 6875 | sa_this_book_map, |
2109b99e AH |
6876 | sa_this_core_map, |
6877 | sa_this_sibling_map, | |
6878 | sa_nodemask, | |
6879 | sa_sched_group_nodes, | |
6880 | #ifdef CONFIG_NUMA | |
6881 | sa_notcovered, | |
6882 | sa_covered, | |
6883 | sa_domainspan, | |
6884 | #endif | |
6885 | sa_none, | |
6886 | }; | |
6887 | ||
9c1cfda2 | 6888 | /* |
48f24c4d | 6889 | * SMT sched-domains: |
9c1cfda2 | 6890 | */ |
1da177e4 | 6891 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6892 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6893 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6894 | |
41a2d6cf | 6895 | static int |
96f874e2 RR |
6896 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6897 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6898 | { |
6711cab4 | 6899 | if (sg) |
1871e52c | 6900 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6901 | return cpu; |
6902 | } | |
6d6bc0ad | 6903 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6904 | |
48f24c4d IM |
6905 | /* |
6906 | * multi-core sched-domains: | |
6907 | */ | |
1e9f28fa | 6908 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6909 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6910 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6911 | |
41a2d6cf | 6912 | static int |
96f874e2 RR |
6913 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6914 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6915 | { |
6711cab4 | 6916 | int group; |
f269893c | 6917 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6918 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6919 | group = cpumask_first(mask); |
f269893c HC |
6920 | #else |
6921 | group = cpu; | |
6922 | #endif | |
6711cab4 | 6923 | if (sg) |
6c99e9ad | 6924 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6925 | return group; |
1e9f28fa | 6926 | } |
f269893c | 6927 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6928 | |
01a08546 HC |
6929 | /* |
6930 | * book sched-domains: | |
6931 | */ | |
6932 | #ifdef CONFIG_SCHED_BOOK | |
6933 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6934 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6935 | ||
41a2d6cf | 6936 | static int |
01a08546 HC |
6937 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6938 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6939 | { |
01a08546 HC |
6940 | int group = cpu; |
6941 | #ifdef CONFIG_SCHED_MC | |
6942 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6943 | group = cpumask_first(mask); | |
6944 | #elif defined(CONFIG_SCHED_SMT) | |
6945 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6946 | group = cpumask_first(mask); | |
6947 | #endif | |
6711cab4 | 6948 | if (sg) |
01a08546 HC |
6949 | *sg = &per_cpu(sched_group_book, group).sg; |
6950 | return group; | |
1e9f28fa | 6951 | } |
01a08546 | 6952 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6953 | |
6c99e9ad RR |
6954 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6955 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6956 | |
41a2d6cf | 6957 | static int |
96f874e2 RR |
6958 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6959 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6960 | { |
6711cab4 | 6961 | int group; |
01a08546 HC |
6962 | #ifdef CONFIG_SCHED_BOOK |
6963 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6964 | group = cpumask_first(mask); | |
6965 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6966 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6967 | group = cpumask_first(mask); |
1e9f28fa | 6968 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6969 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6970 | group = cpumask_first(mask); |
1da177e4 | 6971 | #else |
6711cab4 | 6972 | group = cpu; |
1da177e4 | 6973 | #endif |
6711cab4 | 6974 | if (sg) |
6c99e9ad | 6975 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6976 | return group; |
1da177e4 LT |
6977 | } |
6978 | ||
6979 | #ifdef CONFIG_NUMA | |
1da177e4 | 6980 | /* |
9c1cfda2 JH |
6981 | * The init_sched_build_groups can't handle what we want to do with node |
6982 | * groups, so roll our own. Now each node has its own list of groups which | |
6983 | * gets dynamically allocated. | |
1da177e4 | 6984 | */ |
62ea9ceb | 6985 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6986 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6987 | |
62ea9ceb | 6988 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6989 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6990 | |
96f874e2 RR |
6991 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6992 | struct sched_group **sg, | |
6993 | struct cpumask *nodemask) | |
9c1cfda2 | 6994 | { |
6711cab4 SS |
6995 | int group; |
6996 | ||
6ca09dfc | 6997 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6998 | group = cpumask_first(nodemask); |
6711cab4 SS |
6999 | |
7000 | if (sg) | |
6c99e9ad | 7001 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7002 | return group; |
1da177e4 | 7003 | } |
6711cab4 | 7004 | |
08069033 SS |
7005 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7006 | { | |
7007 | struct sched_group *sg = group_head; | |
7008 | int j; | |
7009 | ||
7010 | if (!sg) | |
7011 | return; | |
3a5c359a | 7012 | do { |
758b2cdc | 7013 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7014 | struct sched_domain *sd; |
08069033 | 7015 | |
6c99e9ad | 7016 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 7017 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
7018 | /* |
7019 | * Only add "power" once for each | |
7020 | * physical package. | |
7021 | */ | |
7022 | continue; | |
7023 | } | |
08069033 | 7024 | |
18a3885f | 7025 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
7026 | } |
7027 | sg = sg->next; | |
7028 | } while (sg != group_head); | |
08069033 | 7029 | } |
0601a88d AH |
7030 | |
7031 | static int build_numa_sched_groups(struct s_data *d, | |
7032 | const struct cpumask *cpu_map, int num) | |
7033 | { | |
7034 | struct sched_domain *sd; | |
7035 | struct sched_group *sg, *prev; | |
7036 | int n, j; | |
7037 | ||
7038 | cpumask_clear(d->covered); | |
7039 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
7040 | if (cpumask_empty(d->nodemask)) { | |
7041 | d->sched_group_nodes[num] = NULL; | |
7042 | goto out; | |
7043 | } | |
7044 | ||
7045 | sched_domain_node_span(num, d->domainspan); | |
7046 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
7047 | ||
7048 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7049 | GFP_KERNEL, num); | |
7050 | if (!sg) { | |
3df0fc5b PZ |
7051 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
7052 | num); | |
0601a88d AH |
7053 | return -ENOMEM; |
7054 | } | |
7055 | d->sched_group_nodes[num] = sg; | |
7056 | ||
7057 | for_each_cpu(j, d->nodemask) { | |
7058 | sd = &per_cpu(node_domains, j).sd; | |
7059 | sd->groups = sg; | |
7060 | } | |
7061 | ||
18a3885f | 7062 | sg->cpu_power = 0; |
0601a88d AH |
7063 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7064 | sg->next = sg; | |
7065 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7066 | ||
7067 | prev = sg; | |
7068 | for (j = 0; j < nr_node_ids; j++) { | |
7069 | n = (num + j) % nr_node_ids; | |
7070 | cpumask_complement(d->notcovered, d->covered); | |
7071 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7072 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7073 | if (cpumask_empty(d->tmpmask)) | |
7074 | break; | |
7075 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7076 | if (cpumask_empty(d->tmpmask)) | |
7077 | continue; | |
7078 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7079 | GFP_KERNEL, num); | |
7080 | if (!sg) { | |
3df0fc5b PZ |
7081 | printk(KERN_WARNING |
7082 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7083 | return -ENOMEM; |
7084 | } | |
18a3885f | 7085 | sg->cpu_power = 0; |
0601a88d AH |
7086 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7087 | sg->next = prev->next; | |
7088 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7089 | prev->next = sg; | |
7090 | prev = sg; | |
7091 | } | |
7092 | out: | |
7093 | return 0; | |
7094 | } | |
6d6bc0ad | 7095 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7096 | |
a616058b | 7097 | #ifdef CONFIG_NUMA |
51888ca2 | 7098 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7099 | static void free_sched_groups(const struct cpumask *cpu_map, |
7100 | struct cpumask *nodemask) | |
51888ca2 | 7101 | { |
a616058b | 7102 | int cpu, i; |
51888ca2 | 7103 | |
abcd083a | 7104 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7105 | struct sched_group **sched_group_nodes |
7106 | = sched_group_nodes_bycpu[cpu]; | |
7107 | ||
51888ca2 SV |
7108 | if (!sched_group_nodes) |
7109 | continue; | |
7110 | ||
076ac2af | 7111 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7112 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7113 | ||
6ca09dfc | 7114 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7115 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7116 | continue; |
7117 | ||
7118 | if (sg == NULL) | |
7119 | continue; | |
7120 | sg = sg->next; | |
7121 | next_sg: | |
7122 | oldsg = sg; | |
7123 | sg = sg->next; | |
7124 | kfree(oldsg); | |
7125 | if (oldsg != sched_group_nodes[i]) | |
7126 | goto next_sg; | |
7127 | } | |
7128 | kfree(sched_group_nodes); | |
7129 | sched_group_nodes_bycpu[cpu] = NULL; | |
7130 | } | |
51888ca2 | 7131 | } |
6d6bc0ad | 7132 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7133 | static void free_sched_groups(const struct cpumask *cpu_map, |
7134 | struct cpumask *nodemask) | |
a616058b SS |
7135 | { |
7136 | } | |
6d6bc0ad | 7137 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7138 | |
89c4710e SS |
7139 | /* |
7140 | * Initialize sched groups cpu_power. | |
7141 | * | |
7142 | * cpu_power indicates the capacity of sched group, which is used while | |
7143 | * distributing the load between different sched groups in a sched domain. | |
7144 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7145 | * there are asymmetries in the topology. If there are asymmetries, group | |
7146 | * having more cpu_power will pickup more load compared to the group having | |
7147 | * less cpu_power. | |
89c4710e SS |
7148 | */ |
7149 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7150 | { | |
7151 | struct sched_domain *child; | |
7152 | struct sched_group *group; | |
f93e65c1 PZ |
7153 | long power; |
7154 | int weight; | |
89c4710e SS |
7155 | |
7156 | WARN_ON(!sd || !sd->groups); | |
7157 | ||
13318a71 | 7158 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7159 | return; |
7160 | ||
aae6d3dd SS |
7161 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7162 | ||
89c4710e SS |
7163 | child = sd->child; |
7164 | ||
18a3885f | 7165 | sd->groups->cpu_power = 0; |
5517d86b | 7166 | |
f93e65c1 PZ |
7167 | if (!child) { |
7168 | power = SCHED_LOAD_SCALE; | |
7169 | weight = cpumask_weight(sched_domain_span(sd)); | |
7170 | /* | |
7171 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7172 | * Usually multiple threads get a better yield out of |
7173 | * that one core than a single thread would have, | |
7174 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7175 | */ |
a52bfd73 PZ |
7176 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7177 | power *= sd->smt_gain; | |
f93e65c1 | 7178 | power /= weight; |
a52bfd73 PZ |
7179 | power >>= SCHED_LOAD_SHIFT; |
7180 | } | |
18a3885f | 7181 | sd->groups->cpu_power += power; |
89c4710e SS |
7182 | return; |
7183 | } | |
7184 | ||
89c4710e | 7185 | /* |
f93e65c1 | 7186 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7187 | */ |
7188 | group = child->groups; | |
7189 | do { | |
18a3885f | 7190 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7191 | group = group->next; |
7192 | } while (group != child->groups); | |
7193 | } | |
7194 | ||
7c16ec58 MT |
7195 | /* |
7196 | * Initializers for schedule domains | |
7197 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7198 | */ | |
7199 | ||
a5d8c348 IM |
7200 | #ifdef CONFIG_SCHED_DEBUG |
7201 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7202 | #else | |
7203 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7204 | #endif | |
7205 | ||
7c16ec58 | 7206 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7207 | |
7c16ec58 MT |
7208 | #define SD_INIT_FUNC(type) \ |
7209 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7210 | { \ | |
7211 | memset(sd, 0, sizeof(*sd)); \ | |
7212 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7213 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7214 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7215 | } |
7216 | ||
7217 | SD_INIT_FUNC(CPU) | |
7218 | #ifdef CONFIG_NUMA | |
7219 | SD_INIT_FUNC(ALLNODES) | |
7220 | SD_INIT_FUNC(NODE) | |
7221 | #endif | |
7222 | #ifdef CONFIG_SCHED_SMT | |
7223 | SD_INIT_FUNC(SIBLING) | |
7224 | #endif | |
7225 | #ifdef CONFIG_SCHED_MC | |
7226 | SD_INIT_FUNC(MC) | |
7227 | #endif | |
01a08546 HC |
7228 | #ifdef CONFIG_SCHED_BOOK |
7229 | SD_INIT_FUNC(BOOK) | |
7230 | #endif | |
7c16ec58 | 7231 | |
1d3504fc HS |
7232 | static int default_relax_domain_level = -1; |
7233 | ||
7234 | static int __init setup_relax_domain_level(char *str) | |
7235 | { | |
30e0e178 LZ |
7236 | unsigned long val; |
7237 | ||
7238 | val = simple_strtoul(str, NULL, 0); | |
7239 | if (val < SD_LV_MAX) | |
7240 | default_relax_domain_level = val; | |
7241 | ||
1d3504fc HS |
7242 | return 1; |
7243 | } | |
7244 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7245 | ||
7246 | static void set_domain_attribute(struct sched_domain *sd, | |
7247 | struct sched_domain_attr *attr) | |
7248 | { | |
7249 | int request; | |
7250 | ||
7251 | if (!attr || attr->relax_domain_level < 0) { | |
7252 | if (default_relax_domain_level < 0) | |
7253 | return; | |
7254 | else | |
7255 | request = default_relax_domain_level; | |
7256 | } else | |
7257 | request = attr->relax_domain_level; | |
7258 | if (request < sd->level) { | |
7259 | /* turn off idle balance on this domain */ | |
c88d5910 | 7260 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7261 | } else { |
7262 | /* turn on idle balance on this domain */ | |
c88d5910 | 7263 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7264 | } |
7265 | } | |
7266 | ||
2109b99e AH |
7267 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7268 | const struct cpumask *cpu_map) | |
7269 | { | |
7270 | switch (what) { | |
7271 | case sa_sched_groups: | |
7272 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7273 | d->sched_group_nodes = NULL; | |
7274 | case sa_rootdomain: | |
7275 | free_rootdomain(d->rd); /* fall through */ | |
7276 | case sa_tmpmask: | |
7277 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7278 | case sa_send_covered: | |
7279 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7280 | case sa_this_book_map: |
7281 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7282 | case sa_this_core_map: |
7283 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7284 | case sa_this_sibling_map: | |
7285 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7286 | case sa_nodemask: | |
7287 | free_cpumask_var(d->nodemask); /* fall through */ | |
7288 | case sa_sched_group_nodes: | |
d1b55138 | 7289 | #ifdef CONFIG_NUMA |
2109b99e AH |
7290 | kfree(d->sched_group_nodes); /* fall through */ |
7291 | case sa_notcovered: | |
7292 | free_cpumask_var(d->notcovered); /* fall through */ | |
7293 | case sa_covered: | |
7294 | free_cpumask_var(d->covered); /* fall through */ | |
7295 | case sa_domainspan: | |
7296 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7297 | #endif |
2109b99e AH |
7298 | case sa_none: |
7299 | break; | |
7300 | } | |
7301 | } | |
3404c8d9 | 7302 | |
2109b99e AH |
7303 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7304 | const struct cpumask *cpu_map) | |
7305 | { | |
3404c8d9 | 7306 | #ifdef CONFIG_NUMA |
2109b99e AH |
7307 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7308 | return sa_none; | |
7309 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7310 | return sa_domainspan; | |
7311 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7312 | return sa_covered; | |
7313 | /* Allocate the per-node list of sched groups */ | |
7314 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7315 | sizeof(struct sched_group *), GFP_KERNEL); | |
7316 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7317 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7318 | return sa_notcovered; |
d1b55138 | 7319 | } |
2109b99e | 7320 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7321 | #endif |
2109b99e AH |
7322 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7323 | return sa_sched_group_nodes; | |
7324 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7325 | return sa_nodemask; | |
7326 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7327 | return sa_this_sibling_map; | |
01a08546 | 7328 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7329 | return sa_this_core_map; |
01a08546 HC |
7330 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7331 | return sa_this_book_map; | |
2109b99e AH |
7332 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7333 | return sa_send_covered; | |
7334 | d->rd = alloc_rootdomain(); | |
7335 | if (!d->rd) { | |
3df0fc5b | 7336 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7337 | return sa_tmpmask; |
57d885fe | 7338 | } |
2109b99e AH |
7339 | return sa_rootdomain; |
7340 | } | |
57d885fe | 7341 | |
7f4588f3 AH |
7342 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7343 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7344 | { | |
7345 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7346 | #ifdef CONFIG_NUMA |
7f4588f3 | 7347 | struct sched_domain *parent; |
1da177e4 | 7348 | |
7f4588f3 AH |
7349 | d->sd_allnodes = 0; |
7350 | if (cpumask_weight(cpu_map) > | |
7351 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7352 | sd = &per_cpu(allnodes_domains, i).sd; | |
7353 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7354 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7355 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7356 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7357 | d->sd_allnodes = 1; | |
7358 | } | |
7359 | parent = sd; | |
7360 | ||
7361 | sd = &per_cpu(node_domains, i).sd; | |
7362 | SD_INIT(sd, NODE); | |
7363 | set_domain_attribute(sd, attr); | |
7364 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7365 | sd->parent = parent; | |
7366 | if (parent) | |
7367 | parent->child = sd; | |
7368 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7369 | #endif |
7f4588f3 AH |
7370 | return sd; |
7371 | } | |
1da177e4 | 7372 | |
87cce662 AH |
7373 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7374 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7375 | struct sched_domain *parent, int i) | |
7376 | { | |
7377 | struct sched_domain *sd; | |
7378 | sd = &per_cpu(phys_domains, i).sd; | |
7379 | SD_INIT(sd, CPU); | |
7380 | set_domain_attribute(sd, attr); | |
7381 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7382 | sd->parent = parent; | |
7383 | if (parent) | |
7384 | parent->child = sd; | |
7385 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7386 | return sd; | |
7387 | } | |
1da177e4 | 7388 | |
01a08546 HC |
7389 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7390 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7391 | struct sched_domain *parent, int i) | |
7392 | { | |
7393 | struct sched_domain *sd = parent; | |
7394 | #ifdef CONFIG_SCHED_BOOK | |
7395 | sd = &per_cpu(book_domains, i).sd; | |
7396 | SD_INIT(sd, BOOK); | |
7397 | set_domain_attribute(sd, attr); | |
7398 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7399 | sd->parent = parent; | |
7400 | parent->child = sd; | |
7401 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7402 | #endif | |
7403 | return sd; | |
7404 | } | |
7405 | ||
410c4081 AH |
7406 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7407 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7408 | struct sched_domain *parent, int i) | |
7409 | { | |
7410 | struct sched_domain *sd = parent; | |
1e9f28fa | 7411 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7412 | sd = &per_cpu(core_domains, i).sd; |
7413 | SD_INIT(sd, MC); | |
7414 | set_domain_attribute(sd, attr); | |
7415 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7416 | sd->parent = parent; | |
7417 | parent->child = sd; | |
7418 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7419 | #endif |
410c4081 AH |
7420 | return sd; |
7421 | } | |
1e9f28fa | 7422 | |
d8173535 AH |
7423 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7424 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7425 | struct sched_domain *parent, int i) | |
7426 | { | |
7427 | struct sched_domain *sd = parent; | |
1da177e4 | 7428 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7429 | sd = &per_cpu(cpu_domains, i).sd; |
7430 | SD_INIT(sd, SIBLING); | |
7431 | set_domain_attribute(sd, attr); | |
7432 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7433 | sd->parent = parent; | |
7434 | parent->child = sd; | |
7435 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7436 | #endif |
d8173535 AH |
7437 | return sd; |
7438 | } | |
1da177e4 | 7439 | |
0e8e85c9 AH |
7440 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7441 | const struct cpumask *cpu_map, int cpu) | |
7442 | { | |
7443 | switch (l) { | |
1da177e4 | 7444 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7445 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7446 | cpumask_and(d->this_sibling_map, cpu_map, | |
7447 | topology_thread_cpumask(cpu)); | |
7448 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7449 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7450 | &cpu_to_cpu_group, | |
7451 | d->send_covered, d->tmpmask); | |
7452 | break; | |
1da177e4 | 7453 | #endif |
1e9f28fa | 7454 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7455 | case SD_LV_MC: /* set up multi-core groups */ |
7456 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7457 | if (cpu == cpumask_first(d->this_core_map)) | |
7458 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7459 | &cpu_to_core_group, | |
7460 | d->send_covered, d->tmpmask); | |
7461 | break; | |
01a08546 HC |
7462 | #endif |
7463 | #ifdef CONFIG_SCHED_BOOK | |
7464 | case SD_LV_BOOK: /* set up book groups */ | |
7465 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7466 | if (cpu == cpumask_first(d->this_book_map)) | |
7467 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7468 | &cpu_to_book_group, | |
7469 | d->send_covered, d->tmpmask); | |
7470 | break; | |
1e9f28fa | 7471 | #endif |
86548096 AH |
7472 | case SD_LV_CPU: /* set up physical groups */ |
7473 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7474 | if (!cpumask_empty(d->nodemask)) | |
7475 | init_sched_build_groups(d->nodemask, cpu_map, | |
7476 | &cpu_to_phys_group, | |
7477 | d->send_covered, d->tmpmask); | |
7478 | break; | |
1da177e4 | 7479 | #ifdef CONFIG_NUMA |
de616e36 AH |
7480 | case SD_LV_ALLNODES: |
7481 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7482 | d->send_covered, d->tmpmask); | |
7483 | break; | |
7484 | #endif | |
0e8e85c9 AH |
7485 | default: |
7486 | break; | |
7c16ec58 | 7487 | } |
0e8e85c9 | 7488 | } |
9c1cfda2 | 7489 | |
2109b99e AH |
7490 | /* |
7491 | * Build sched domains for a given set of cpus and attach the sched domains | |
7492 | * to the individual cpus | |
7493 | */ | |
7494 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7495 | struct sched_domain_attr *attr) | |
7496 | { | |
7497 | enum s_alloc alloc_state = sa_none; | |
7498 | struct s_data d; | |
294b0c96 | 7499 | struct sched_domain *sd; |
2109b99e | 7500 | int i; |
7c16ec58 | 7501 | #ifdef CONFIG_NUMA |
2109b99e | 7502 | d.sd_allnodes = 0; |
7c16ec58 | 7503 | #endif |
9c1cfda2 | 7504 | |
2109b99e AH |
7505 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7506 | if (alloc_state != sa_rootdomain) | |
7507 | goto error; | |
7508 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7509 | |
1da177e4 | 7510 | /* |
1a20ff27 | 7511 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7512 | */ |
abcd083a | 7513 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7514 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7515 | cpu_map); | |
9761eea8 | 7516 | |
7f4588f3 | 7517 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7518 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7519 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7520 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7521 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7522 | } |
9c1cfda2 | 7523 | |
abcd083a | 7524 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7525 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7526 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7527 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7528 | } |
9c1cfda2 | 7529 | |
1da177e4 | 7530 | /* Set up physical groups */ |
86548096 AH |
7531 | for (i = 0; i < nr_node_ids; i++) |
7532 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7533 | |
1da177e4 LT |
7534 | #ifdef CONFIG_NUMA |
7535 | /* Set up node groups */ | |
de616e36 AH |
7536 | if (d.sd_allnodes) |
7537 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7538 | |
0601a88d AH |
7539 | for (i = 0; i < nr_node_ids; i++) |
7540 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7541 | goto error; |
1da177e4 LT |
7542 | #endif |
7543 | ||
7544 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7545 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7546 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7547 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7548 | init_sched_groups_power(i, sd); |
5c45bf27 | 7549 | } |
1da177e4 | 7550 | #endif |
1e9f28fa | 7551 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7552 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7553 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7554 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7555 | } |
7556 | #endif | |
01a08546 HC |
7557 | #ifdef CONFIG_SCHED_BOOK |
7558 | for_each_cpu(i, cpu_map) { | |
7559 | sd = &per_cpu(book_domains, i).sd; | |
7560 | init_sched_groups_power(i, sd); | |
7561 | } | |
7562 | #endif | |
1e9f28fa | 7563 | |
abcd083a | 7564 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7565 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7566 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7567 | } |
7568 | ||
9c1cfda2 | 7569 | #ifdef CONFIG_NUMA |
076ac2af | 7570 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7571 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7572 | |
49a02c51 | 7573 | if (d.sd_allnodes) { |
6711cab4 | 7574 | struct sched_group *sg; |
f712c0c7 | 7575 | |
96f874e2 | 7576 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7577 | d.tmpmask); |
f712c0c7 SS |
7578 | init_numa_sched_groups_power(sg); |
7579 | } | |
9c1cfda2 JH |
7580 | #endif |
7581 | ||
1da177e4 | 7582 | /* Attach the domains */ |
abcd083a | 7583 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7584 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7585 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7586 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7587 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7588 | #elif defined(CONFIG_SCHED_BOOK) |
7589 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7590 | #else |
6c99e9ad | 7591 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7592 | #endif |
49a02c51 | 7593 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7594 | } |
51888ca2 | 7595 | |
2109b99e AH |
7596 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7597 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7598 | return 0; | |
51888ca2 | 7599 | |
51888ca2 | 7600 | error: |
2109b99e AH |
7601 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7602 | return -ENOMEM; | |
1da177e4 | 7603 | } |
029190c5 | 7604 | |
96f874e2 | 7605 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7606 | { |
7607 | return __build_sched_domains(cpu_map, NULL); | |
7608 | } | |
7609 | ||
acc3f5d7 | 7610 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7611 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7612 | static struct sched_domain_attr *dattr_cur; |
7613 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7614 | |
7615 | /* | |
7616 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7617 | * cpumask) fails, then fallback to a single sched domain, |
7618 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7619 | */ |
4212823f | 7620 | static cpumask_var_t fallback_doms; |
029190c5 | 7621 | |
ee79d1bd HC |
7622 | /* |
7623 | * arch_update_cpu_topology lets virtualized architectures update the | |
7624 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7625 | * or 0 if it stayed the same. | |
7626 | */ | |
7627 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7628 | { |
ee79d1bd | 7629 | return 0; |
22e52b07 HC |
7630 | } |
7631 | ||
acc3f5d7 RR |
7632 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7633 | { | |
7634 | int i; | |
7635 | cpumask_var_t *doms; | |
7636 | ||
7637 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7638 | if (!doms) | |
7639 | return NULL; | |
7640 | for (i = 0; i < ndoms; i++) { | |
7641 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7642 | free_sched_domains(doms, i); | |
7643 | return NULL; | |
7644 | } | |
7645 | } | |
7646 | return doms; | |
7647 | } | |
7648 | ||
7649 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7650 | { | |
7651 | unsigned int i; | |
7652 | for (i = 0; i < ndoms; i++) | |
7653 | free_cpumask_var(doms[i]); | |
7654 | kfree(doms); | |
7655 | } | |
7656 | ||
1a20ff27 | 7657 | /* |
41a2d6cf | 7658 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7659 | * For now this just excludes isolated cpus, but could be used to |
7660 | * exclude other special cases in the future. | |
1a20ff27 | 7661 | */ |
96f874e2 | 7662 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7663 | { |
7378547f MM |
7664 | int err; |
7665 | ||
22e52b07 | 7666 | arch_update_cpu_topology(); |
029190c5 | 7667 | ndoms_cur = 1; |
acc3f5d7 | 7668 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7669 | if (!doms_cur) |
acc3f5d7 RR |
7670 | doms_cur = &fallback_doms; |
7671 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7672 | dattr_cur = NULL; |
acc3f5d7 | 7673 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7674 | register_sched_domain_sysctl(); |
7378547f MM |
7675 | |
7676 | return err; | |
1a20ff27 DG |
7677 | } |
7678 | ||
96f874e2 RR |
7679 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7680 | struct cpumask *tmpmask) | |
1da177e4 | 7681 | { |
7c16ec58 | 7682 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7683 | } |
1da177e4 | 7684 | |
1a20ff27 DG |
7685 | /* |
7686 | * Detach sched domains from a group of cpus specified in cpu_map | |
7687 | * These cpus will now be attached to the NULL domain | |
7688 | */ | |
96f874e2 | 7689 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7690 | { |
96f874e2 RR |
7691 | /* Save because hotplug lock held. */ |
7692 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7693 | int i; |
7694 | ||
abcd083a | 7695 | for_each_cpu(i, cpu_map) |
57d885fe | 7696 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7697 | synchronize_sched(); |
96f874e2 | 7698 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7699 | } |
7700 | ||
1d3504fc HS |
7701 | /* handle null as "default" */ |
7702 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7703 | struct sched_domain_attr *new, int idx_new) | |
7704 | { | |
7705 | struct sched_domain_attr tmp; | |
7706 | ||
7707 | /* fast path */ | |
7708 | if (!new && !cur) | |
7709 | return 1; | |
7710 | ||
7711 | tmp = SD_ATTR_INIT; | |
7712 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7713 | new ? (new + idx_new) : &tmp, | |
7714 | sizeof(struct sched_domain_attr)); | |
7715 | } | |
7716 | ||
029190c5 PJ |
7717 | /* |
7718 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7719 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7720 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7721 | * It destroys each deleted domain and builds each new domain. | |
7722 | * | |
acc3f5d7 | 7723 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7724 | * The masks don't intersect (don't overlap.) We should setup one |
7725 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7726 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7727 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7728 | * it as it is. | |
7729 | * | |
acc3f5d7 RR |
7730 | * The passed in 'doms_new' should be allocated using |
7731 | * alloc_sched_domains. This routine takes ownership of it and will | |
7732 | * free_sched_domains it when done with it. If the caller failed the | |
7733 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7734 | * and partition_sched_domains() will fallback to the single partition | |
7735 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7736 | * |
96f874e2 | 7737 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7738 | * ndoms_new == 0 is a special case for destroying existing domains, |
7739 | * and it will not create the default domain. | |
dfb512ec | 7740 | * |
029190c5 PJ |
7741 | * Call with hotplug lock held |
7742 | */ | |
acc3f5d7 | 7743 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7744 | struct sched_domain_attr *dattr_new) |
029190c5 | 7745 | { |
dfb512ec | 7746 | int i, j, n; |
d65bd5ec | 7747 | int new_topology; |
029190c5 | 7748 | |
712555ee | 7749 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7750 | |
7378547f MM |
7751 | /* always unregister in case we don't destroy any domains */ |
7752 | unregister_sched_domain_sysctl(); | |
7753 | ||
d65bd5ec HC |
7754 | /* Let architecture update cpu core mappings. */ |
7755 | new_topology = arch_update_cpu_topology(); | |
7756 | ||
dfb512ec | 7757 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7758 | |
7759 | /* Destroy deleted domains */ | |
7760 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7761 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7762 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7763 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7764 | goto match1; |
7765 | } | |
7766 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7767 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7768 | match1: |
7769 | ; | |
7770 | } | |
7771 | ||
e761b772 MK |
7772 | if (doms_new == NULL) { |
7773 | ndoms_cur = 0; | |
acc3f5d7 | 7774 | doms_new = &fallback_doms; |
6ad4c188 | 7775 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7776 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7777 | } |
7778 | ||
029190c5 PJ |
7779 | /* Build new domains */ |
7780 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7781 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7782 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7783 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7784 | goto match2; |
7785 | } | |
7786 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7787 | __build_sched_domains(doms_new[i], |
1d3504fc | 7788 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7789 | match2: |
7790 | ; | |
7791 | } | |
7792 | ||
7793 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7794 | if (doms_cur != &fallback_doms) |
7795 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7796 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7797 | doms_cur = doms_new; |
1d3504fc | 7798 | dattr_cur = dattr_new; |
029190c5 | 7799 | ndoms_cur = ndoms_new; |
7378547f MM |
7800 | |
7801 | register_sched_domain_sysctl(); | |
a1835615 | 7802 | |
712555ee | 7803 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7804 | } |
7805 | ||
5c45bf27 | 7806 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7807 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7808 | { |
95402b38 | 7809 | get_online_cpus(); |
dfb512ec MK |
7810 | |
7811 | /* Destroy domains first to force the rebuild */ | |
7812 | partition_sched_domains(0, NULL, NULL); | |
7813 | ||
e761b772 | 7814 | rebuild_sched_domains(); |
95402b38 | 7815 | put_online_cpus(); |
5c45bf27 SS |
7816 | } |
7817 | ||
7818 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7819 | { | |
afb8a9b7 | 7820 | unsigned int level = 0; |
5c45bf27 | 7821 | |
afb8a9b7 GS |
7822 | if (sscanf(buf, "%u", &level) != 1) |
7823 | return -EINVAL; | |
7824 | ||
7825 | /* | |
7826 | * level is always be positive so don't check for | |
7827 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7828 | * What happens on 0 or 1 byte write, | |
7829 | * need to check for count as well? | |
7830 | */ | |
7831 | ||
7832 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7833 | return -EINVAL; |
7834 | ||
7835 | if (smt) | |
afb8a9b7 | 7836 | sched_smt_power_savings = level; |
5c45bf27 | 7837 | else |
afb8a9b7 | 7838 | sched_mc_power_savings = level; |
5c45bf27 | 7839 | |
c70f22d2 | 7840 | arch_reinit_sched_domains(); |
5c45bf27 | 7841 | |
c70f22d2 | 7842 | return count; |
5c45bf27 SS |
7843 | } |
7844 | ||
5c45bf27 | 7845 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7846 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7847 | struct sysdev_class_attribute *attr, |
f718cd4a | 7848 | char *page) |
5c45bf27 SS |
7849 | { |
7850 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7851 | } | |
f718cd4a | 7852 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7853 | struct sysdev_class_attribute *attr, |
48f24c4d | 7854 | const char *buf, size_t count) |
5c45bf27 SS |
7855 | { |
7856 | return sched_power_savings_store(buf, count, 0); | |
7857 | } | |
f718cd4a AK |
7858 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7859 | sched_mc_power_savings_show, | |
7860 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7861 | #endif |
7862 | ||
7863 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7864 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7865 | struct sysdev_class_attribute *attr, |
f718cd4a | 7866 | char *page) |
5c45bf27 SS |
7867 | { |
7868 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7869 | } | |
f718cd4a | 7870 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7871 | struct sysdev_class_attribute *attr, |
48f24c4d | 7872 | const char *buf, size_t count) |
5c45bf27 SS |
7873 | { |
7874 | return sched_power_savings_store(buf, count, 1); | |
7875 | } | |
f718cd4a AK |
7876 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7877 | sched_smt_power_savings_show, | |
6707de00 AB |
7878 | sched_smt_power_savings_store); |
7879 | #endif | |
7880 | ||
39aac648 | 7881 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7882 | { |
7883 | int err = 0; | |
7884 | ||
7885 | #ifdef CONFIG_SCHED_SMT | |
7886 | if (smt_capable()) | |
7887 | err = sysfs_create_file(&cls->kset.kobj, | |
7888 | &attr_sched_smt_power_savings.attr); | |
7889 | #endif | |
7890 | #ifdef CONFIG_SCHED_MC | |
7891 | if (!err && mc_capable()) | |
7892 | err = sysfs_create_file(&cls->kset.kobj, | |
7893 | &attr_sched_mc_power_savings.attr); | |
7894 | #endif | |
7895 | return err; | |
7896 | } | |
6d6bc0ad | 7897 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7898 | |
1da177e4 | 7899 | /* |
3a101d05 TH |
7900 | * Update cpusets according to cpu_active mask. If cpusets are |
7901 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7902 | * around partition_sched_domains(). | |
1da177e4 | 7903 | */ |
0b2e918a TH |
7904 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7905 | void *hcpu) | |
e761b772 | 7906 | { |
3a101d05 | 7907 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7908 | case CPU_ONLINE: |
6ad4c188 | 7909 | case CPU_DOWN_FAILED: |
3a101d05 | 7910 | cpuset_update_active_cpus(); |
e761b772 | 7911 | return NOTIFY_OK; |
3a101d05 TH |
7912 | default: |
7913 | return NOTIFY_DONE; | |
7914 | } | |
7915 | } | |
e761b772 | 7916 | |
0b2e918a TH |
7917 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7918 | void *hcpu) | |
3a101d05 TH |
7919 | { |
7920 | switch (action & ~CPU_TASKS_FROZEN) { | |
7921 | case CPU_DOWN_PREPARE: | |
7922 | cpuset_update_active_cpus(); | |
7923 | return NOTIFY_OK; | |
e761b772 MK |
7924 | default: |
7925 | return NOTIFY_DONE; | |
7926 | } | |
7927 | } | |
e761b772 MK |
7928 | |
7929 | static int update_runtime(struct notifier_block *nfb, | |
7930 | unsigned long action, void *hcpu) | |
1da177e4 | 7931 | { |
7def2be1 PZ |
7932 | int cpu = (int)(long)hcpu; |
7933 | ||
1da177e4 | 7934 | switch (action) { |
1da177e4 | 7935 | case CPU_DOWN_PREPARE: |
8bb78442 | 7936 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7937 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7938 | return NOTIFY_OK; |
7939 | ||
1da177e4 | 7940 | case CPU_DOWN_FAILED: |
8bb78442 | 7941 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7942 | case CPU_ONLINE: |
8bb78442 | 7943 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7944 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7945 | return NOTIFY_OK; |
7946 | ||
1da177e4 LT |
7947 | default: |
7948 | return NOTIFY_DONE; | |
7949 | } | |
1da177e4 | 7950 | } |
1da177e4 LT |
7951 | |
7952 | void __init sched_init_smp(void) | |
7953 | { | |
dcc30a35 RR |
7954 | cpumask_var_t non_isolated_cpus; |
7955 | ||
7956 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7957 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7958 | |
434d53b0 MT |
7959 | #if defined(CONFIG_NUMA) |
7960 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7961 | GFP_KERNEL); | |
7962 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7963 | #endif | |
95402b38 | 7964 | get_online_cpus(); |
712555ee | 7965 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7966 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7967 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7968 | if (cpumask_empty(non_isolated_cpus)) | |
7969 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7970 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7971 | put_online_cpus(); |
e761b772 | 7972 | |
3a101d05 TH |
7973 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7974 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7975 | |
7976 | /* RT runtime code needs to handle some hotplug events */ | |
7977 | hotcpu_notifier(update_runtime, 0); | |
7978 | ||
b328ca18 | 7979 | init_hrtick(); |
5c1e1767 NP |
7980 | |
7981 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7982 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7983 | BUG(); |
19978ca6 | 7984 | sched_init_granularity(); |
dcc30a35 | 7985 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7986 | |
0e3900e6 | 7987 | init_sched_rt_class(); |
1da177e4 LT |
7988 | } |
7989 | #else | |
7990 | void __init sched_init_smp(void) | |
7991 | { | |
19978ca6 | 7992 | sched_init_granularity(); |
1da177e4 LT |
7993 | } |
7994 | #endif /* CONFIG_SMP */ | |
7995 | ||
cd1bb94b AB |
7996 | const_debug unsigned int sysctl_timer_migration = 1; |
7997 | ||
1da177e4 LT |
7998 | int in_sched_functions(unsigned long addr) |
7999 | { | |
1da177e4 LT |
8000 | return in_lock_functions(addr) || |
8001 | (addr >= (unsigned long)__sched_text_start | |
8002 | && addr < (unsigned long)__sched_text_end); | |
8003 | } | |
8004 | ||
a9957449 | 8005 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8006 | { |
8007 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8008 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8009 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8010 | cfs_rq->rq = rq; | |
f07333bf | 8011 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 8012 | #ifdef CONFIG_SMP |
f07333bf | 8013 | cfs_rq->load_stamp = 1; |
6ea72f12 | 8014 | #endif |
dd41f596 | 8015 | #endif |
67e9fb2a | 8016 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8017 | } |
8018 | ||
fa85ae24 PZ |
8019 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8020 | { | |
8021 | struct rt_prio_array *array; | |
8022 | int i; | |
8023 | ||
8024 | array = &rt_rq->active; | |
8025 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8026 | INIT_LIST_HEAD(array->queue + i); | |
8027 | __clear_bit(i, array->bitmap); | |
8028 | } | |
8029 | /* delimiter for bitsearch: */ | |
8030 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8031 | ||
052f1dc7 | 8032 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8033 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8034 | #ifdef CONFIG_SMP |
e864c499 | 8035 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8036 | #endif |
48d5e258 | 8037 | #endif |
fa85ae24 PZ |
8038 | #ifdef CONFIG_SMP |
8039 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8040 | rt_rq->overloaded = 0; |
05fa785c | 8041 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8042 | #endif |
8043 | ||
8044 | rt_rq->rt_time = 0; | |
8045 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8046 | rt_rq->rt_runtime = 0; |
0986b11b | 8047 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 8048 | |
052f1dc7 | 8049 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8050 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8051 | rt_rq->rq = rq; |
8052 | #endif | |
fa85ae24 PZ |
8053 | } |
8054 | ||
6f505b16 | 8055 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8056 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8057 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8058 | struct sched_entity *parent) |
6f505b16 | 8059 | { |
ec7dc8ac | 8060 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8061 | tg->cfs_rq[cpu] = cfs_rq; |
8062 | init_cfs_rq(cfs_rq, rq); | |
8063 | cfs_rq->tg = tg; | |
6f505b16 PZ |
8064 | |
8065 | tg->se[cpu] = se; | |
07e06b01 | 8066 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8067 | if (!se) |
8068 | return; | |
8069 | ||
ec7dc8ac DG |
8070 | if (!parent) |
8071 | se->cfs_rq = &rq->cfs; | |
8072 | else | |
8073 | se->cfs_rq = parent->my_q; | |
8074 | ||
6f505b16 | 8075 | se->my_q = cfs_rq; |
9437178f | 8076 | update_load_set(&se->load, 0); |
ec7dc8ac | 8077 | se->parent = parent; |
6f505b16 | 8078 | } |
052f1dc7 | 8079 | #endif |
6f505b16 | 8080 | |
052f1dc7 | 8081 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8082 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8083 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8084 | struct sched_rt_entity *parent) |
6f505b16 | 8085 | { |
ec7dc8ac DG |
8086 | struct rq *rq = cpu_rq(cpu); |
8087 | ||
6f505b16 PZ |
8088 | tg->rt_rq[cpu] = rt_rq; |
8089 | init_rt_rq(rt_rq, rq); | |
8090 | rt_rq->tg = tg; | |
ac086bc2 | 8091 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8092 | |
8093 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8094 | if (!rt_se) |
8095 | return; | |
8096 | ||
ec7dc8ac DG |
8097 | if (!parent) |
8098 | rt_se->rt_rq = &rq->rt; | |
8099 | else | |
8100 | rt_se->rt_rq = parent->my_q; | |
8101 | ||
6f505b16 | 8102 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8103 | rt_se->parent = parent; |
6f505b16 PZ |
8104 | INIT_LIST_HEAD(&rt_se->run_list); |
8105 | } | |
8106 | #endif | |
8107 | ||
1da177e4 LT |
8108 | void __init sched_init(void) |
8109 | { | |
dd41f596 | 8110 | int i, j; |
434d53b0 MT |
8111 | unsigned long alloc_size = 0, ptr; |
8112 | ||
8113 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8114 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8115 | #endif | |
8116 | #ifdef CONFIG_RT_GROUP_SCHED | |
8117 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8118 | #endif |
df7c8e84 | 8119 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8120 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8121 | #endif |
434d53b0 | 8122 | if (alloc_size) { |
36b7b6d4 | 8123 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8124 | |
8125 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8126 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8127 | ptr += nr_cpu_ids * sizeof(void **); |
8128 | ||
07e06b01 | 8129 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8130 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8131 | |
6d6bc0ad | 8132 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8133 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8134 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8135 | ptr += nr_cpu_ids * sizeof(void **); |
8136 | ||
07e06b01 | 8137 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8138 | ptr += nr_cpu_ids * sizeof(void **); |
8139 | ||
6d6bc0ad | 8140 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8141 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8142 | for_each_possible_cpu(i) { | |
8143 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8144 | ptr += cpumask_size(); | |
8145 | } | |
8146 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8147 | } |
dd41f596 | 8148 | |
57d885fe GH |
8149 | #ifdef CONFIG_SMP |
8150 | init_defrootdomain(); | |
8151 | #endif | |
8152 | ||
d0b27fa7 PZ |
8153 | init_rt_bandwidth(&def_rt_bandwidth, |
8154 | global_rt_period(), global_rt_runtime()); | |
8155 | ||
8156 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8157 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8158 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8159 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8160 | |
7c941438 | 8161 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8162 | list_add(&root_task_group.list, &task_groups); |
8163 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8164 | autogroup_init(&init_task); |
7c941438 | 8165 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8166 | |
0a945022 | 8167 | for_each_possible_cpu(i) { |
70b97a7f | 8168 | struct rq *rq; |
1da177e4 LT |
8169 | |
8170 | rq = cpu_rq(i); | |
05fa785c | 8171 | raw_spin_lock_init(&rq->lock); |
7897986b | 8172 | rq->nr_running = 0; |
dce48a84 TG |
8173 | rq->calc_load_active = 0; |
8174 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8175 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8176 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8177 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8178 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8179 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8180 | /* |
07e06b01 | 8181 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8182 | * |
8183 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8184 | * gets 100% of the cpu resources in the system. This overall | |
8185 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8186 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8187 | * based on each entity's (task or task-group's) weight |
8188 | * (se->load.weight). | |
8189 | * | |
07e06b01 | 8190 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8191 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8192 | * then A0's share of the cpu resource is: | |
8193 | * | |
0d905bca | 8194 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8195 | * |
07e06b01 YZ |
8196 | * We achieve this by letting root_task_group's tasks sit |
8197 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8198 | */ |
07e06b01 | 8199 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8200 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8201 | ||
8202 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8203 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8204 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8205 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8206 | #endif |
1da177e4 | 8207 | |
dd41f596 IM |
8208 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8209 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8210 | |
8211 | rq->last_load_update_tick = jiffies; | |
8212 | ||
1da177e4 | 8213 | #ifdef CONFIG_SMP |
41c7ce9a | 8214 | rq->sd = NULL; |
57d885fe | 8215 | rq->rd = NULL; |
e51fd5e2 | 8216 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8217 | rq->post_schedule = 0; |
1da177e4 | 8218 | rq->active_balance = 0; |
dd41f596 | 8219 | rq->next_balance = jiffies; |
1da177e4 | 8220 | rq->push_cpu = 0; |
0a2966b4 | 8221 | rq->cpu = i; |
1f11eb6a | 8222 | rq->online = 0; |
eae0c9df MG |
8223 | rq->idle_stamp = 0; |
8224 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8225 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8226 | #ifdef CONFIG_NO_HZ |
8227 | rq->nohz_balance_kick = 0; | |
8228 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8229 | #endif | |
1da177e4 | 8230 | #endif |
8f4d37ec | 8231 | init_rq_hrtick(rq); |
1da177e4 | 8232 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8233 | } |
8234 | ||
2dd73a4f | 8235 | set_load_weight(&init_task); |
b50f60ce | 8236 | |
e107be36 AK |
8237 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8238 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8239 | #endif | |
8240 | ||
c9819f45 | 8241 | #ifdef CONFIG_SMP |
962cf36c | 8242 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8243 | #endif |
8244 | ||
b50f60ce | 8245 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8246 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8247 | #endif |
8248 | ||
1da177e4 LT |
8249 | /* |
8250 | * The boot idle thread does lazy MMU switching as well: | |
8251 | */ | |
8252 | atomic_inc(&init_mm.mm_count); | |
8253 | enter_lazy_tlb(&init_mm, current); | |
8254 | ||
8255 | /* | |
8256 | * Make us the idle thread. Technically, schedule() should not be | |
8257 | * called from this thread, however somewhere below it might be, | |
8258 | * but because we are the idle thread, we just pick up running again | |
8259 | * when this runqueue becomes "idle". | |
8260 | */ | |
8261 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8262 | |
8263 | calc_load_update = jiffies + LOAD_FREQ; | |
8264 | ||
dd41f596 IM |
8265 | /* |
8266 | * During early bootup we pretend to be a normal task: | |
8267 | */ | |
8268 | current->sched_class = &fair_sched_class; | |
6892b75e | 8269 | |
6a7b3dc3 | 8270 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8271 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8272 | #ifdef CONFIG_SMP |
7d1e6a9b | 8273 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8274 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8275 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8276 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8277 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8278 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8279 | #endif |
bdddd296 RR |
8280 | /* May be allocated at isolcpus cmdline parse time */ |
8281 | if (cpu_isolated_map == NULL) | |
8282 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8283 | #endif /* SMP */ |
6a7b3dc3 | 8284 | |
6892b75e | 8285 | scheduler_running = 1; |
1da177e4 LT |
8286 | } |
8287 | ||
8288 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8289 | static inline int preempt_count_equals(int preempt_offset) |
8290 | { | |
234da7bc | 8291 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
8292 | |
8293 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
8294 | } | |
8295 | ||
d894837f | 8296 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8297 | { |
48f24c4d | 8298 | #ifdef in_atomic |
1da177e4 LT |
8299 | static unsigned long prev_jiffy; /* ratelimiting */ |
8300 | ||
e4aafea2 FW |
8301 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8302 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8303 | return; |
8304 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8305 | return; | |
8306 | prev_jiffy = jiffies; | |
8307 | ||
3df0fc5b PZ |
8308 | printk(KERN_ERR |
8309 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8310 | file, line); | |
8311 | printk(KERN_ERR | |
8312 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8313 | in_atomic(), irqs_disabled(), | |
8314 | current->pid, current->comm); | |
aef745fc IM |
8315 | |
8316 | debug_show_held_locks(current); | |
8317 | if (irqs_disabled()) | |
8318 | print_irqtrace_events(current); | |
8319 | dump_stack(); | |
1da177e4 LT |
8320 | #endif |
8321 | } | |
8322 | EXPORT_SYMBOL(__might_sleep); | |
8323 | #endif | |
8324 | ||
8325 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8326 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8327 | { | |
da7a735e PZ |
8328 | const struct sched_class *prev_class = p->sched_class; |
8329 | int old_prio = p->prio; | |
3a5e4dc1 | 8330 | int on_rq; |
3e51f33f | 8331 | |
3a5e4dc1 AK |
8332 | on_rq = p->se.on_rq; |
8333 | if (on_rq) | |
8334 | deactivate_task(rq, p, 0); | |
8335 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8336 | if (on_rq) { | |
8337 | activate_task(rq, p, 0); | |
8338 | resched_task(rq->curr); | |
8339 | } | |
da7a735e PZ |
8340 | |
8341 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8342 | } |
8343 | ||
1da177e4 LT |
8344 | void normalize_rt_tasks(void) |
8345 | { | |
a0f98a1c | 8346 | struct task_struct *g, *p; |
1da177e4 | 8347 | unsigned long flags; |
70b97a7f | 8348 | struct rq *rq; |
1da177e4 | 8349 | |
4cf5d77a | 8350 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8351 | do_each_thread(g, p) { |
178be793 IM |
8352 | /* |
8353 | * Only normalize user tasks: | |
8354 | */ | |
8355 | if (!p->mm) | |
8356 | continue; | |
8357 | ||
6cfb0d5d | 8358 | p->se.exec_start = 0; |
6cfb0d5d | 8359 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8360 | p->se.statistics.wait_start = 0; |
8361 | p->se.statistics.sleep_start = 0; | |
8362 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8363 | #endif |
dd41f596 IM |
8364 | |
8365 | if (!rt_task(p)) { | |
8366 | /* | |
8367 | * Renice negative nice level userspace | |
8368 | * tasks back to 0: | |
8369 | */ | |
8370 | if (TASK_NICE(p) < 0 && p->mm) | |
8371 | set_user_nice(p, 0); | |
1da177e4 | 8372 | continue; |
dd41f596 | 8373 | } |
1da177e4 | 8374 | |
1d615482 | 8375 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8376 | rq = __task_rq_lock(p); |
1da177e4 | 8377 | |
178be793 | 8378 | normalize_task(rq, p); |
3a5e4dc1 | 8379 | |
b29739f9 | 8380 | __task_rq_unlock(rq); |
1d615482 | 8381 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8382 | } while_each_thread(g, p); |
8383 | ||
4cf5d77a | 8384 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8385 | } |
8386 | ||
8387 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8388 | |
67fc4e0c | 8389 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8390 | /* |
67fc4e0c | 8391 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8392 | * |
8393 | * They can only be called when the whole system has been | |
8394 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8395 | * activity can take place. Using them for anything else would | |
8396 | * be a serious bug, and as a result, they aren't even visible | |
8397 | * under any other configuration. | |
8398 | */ | |
8399 | ||
8400 | /** | |
8401 | * curr_task - return the current task for a given cpu. | |
8402 | * @cpu: the processor in question. | |
8403 | * | |
8404 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8405 | */ | |
36c8b586 | 8406 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8407 | { |
8408 | return cpu_curr(cpu); | |
8409 | } | |
8410 | ||
67fc4e0c JW |
8411 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8412 | ||
8413 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8414 | /** |
8415 | * set_curr_task - set the current task for a given cpu. | |
8416 | * @cpu: the processor in question. | |
8417 | * @p: the task pointer to set. | |
8418 | * | |
8419 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8420 | * are serviced on a separate stack. It allows the architecture to switch the |
8421 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8422 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8423 | * and caller must save the original value of the current task (see | |
8424 | * curr_task() above) and restore that value before reenabling interrupts and | |
8425 | * re-starting the system. | |
8426 | * | |
8427 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8428 | */ | |
36c8b586 | 8429 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8430 | { |
8431 | cpu_curr(cpu) = p; | |
8432 | } | |
8433 | ||
8434 | #endif | |
29f59db3 | 8435 | |
bccbe08a PZ |
8436 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8437 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8438 | { |
8439 | int i; | |
8440 | ||
8441 | for_each_possible_cpu(i) { | |
8442 | if (tg->cfs_rq) | |
8443 | kfree(tg->cfs_rq[i]); | |
8444 | if (tg->se) | |
8445 | kfree(tg->se[i]); | |
6f505b16 PZ |
8446 | } |
8447 | ||
8448 | kfree(tg->cfs_rq); | |
8449 | kfree(tg->se); | |
6f505b16 PZ |
8450 | } |
8451 | ||
ec7dc8ac DG |
8452 | static |
8453 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8454 | { |
29f59db3 | 8455 | struct cfs_rq *cfs_rq; |
eab17229 | 8456 | struct sched_entity *se; |
9b5b7751 | 8457 | struct rq *rq; |
29f59db3 SV |
8458 | int i; |
8459 | ||
434d53b0 | 8460 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8461 | if (!tg->cfs_rq) |
8462 | goto err; | |
434d53b0 | 8463 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8464 | if (!tg->se) |
8465 | goto err; | |
052f1dc7 PZ |
8466 | |
8467 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8468 | |
8469 | for_each_possible_cpu(i) { | |
9b5b7751 | 8470 | rq = cpu_rq(i); |
29f59db3 | 8471 | |
eab17229 LZ |
8472 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8473 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8474 | if (!cfs_rq) |
8475 | goto err; | |
8476 | ||
eab17229 LZ |
8477 | se = kzalloc_node(sizeof(struct sched_entity), |
8478 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8479 | if (!se) |
dfc12eb2 | 8480 | goto err_free_rq; |
29f59db3 | 8481 | |
3d4b47b4 | 8482 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8483 | } |
8484 | ||
8485 | return 1; | |
8486 | ||
49246274 | 8487 | err_free_rq: |
dfc12eb2 | 8488 | kfree(cfs_rq); |
49246274 | 8489 | err: |
bccbe08a PZ |
8490 | return 0; |
8491 | } | |
8492 | ||
bccbe08a PZ |
8493 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8494 | { | |
3d4b47b4 PZ |
8495 | struct rq *rq = cpu_rq(cpu); |
8496 | unsigned long flags; | |
3d4b47b4 PZ |
8497 | |
8498 | /* | |
8499 | * Only empty task groups can be destroyed; so we can speculatively | |
8500 | * check on_list without danger of it being re-added. | |
8501 | */ | |
8502 | if (!tg->cfs_rq[cpu]->on_list) | |
8503 | return; | |
8504 | ||
8505 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8506 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8507 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8508 | } |
6d6bc0ad | 8509 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8510 | static inline void free_fair_sched_group(struct task_group *tg) |
8511 | { | |
8512 | } | |
8513 | ||
ec7dc8ac DG |
8514 | static inline |
8515 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8516 | { |
8517 | return 1; | |
8518 | } | |
8519 | ||
bccbe08a PZ |
8520 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8521 | { | |
8522 | } | |
6d6bc0ad | 8523 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8524 | |
8525 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8526 | static void free_rt_sched_group(struct task_group *tg) |
8527 | { | |
8528 | int i; | |
8529 | ||
d0b27fa7 PZ |
8530 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8531 | ||
bccbe08a PZ |
8532 | for_each_possible_cpu(i) { |
8533 | if (tg->rt_rq) | |
8534 | kfree(tg->rt_rq[i]); | |
8535 | if (tg->rt_se) | |
8536 | kfree(tg->rt_se[i]); | |
8537 | } | |
8538 | ||
8539 | kfree(tg->rt_rq); | |
8540 | kfree(tg->rt_se); | |
8541 | } | |
8542 | ||
ec7dc8ac DG |
8543 | static |
8544 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8545 | { |
8546 | struct rt_rq *rt_rq; | |
eab17229 | 8547 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8548 | struct rq *rq; |
8549 | int i; | |
8550 | ||
434d53b0 | 8551 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8552 | if (!tg->rt_rq) |
8553 | goto err; | |
434d53b0 | 8554 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8555 | if (!tg->rt_se) |
8556 | goto err; | |
8557 | ||
d0b27fa7 PZ |
8558 | init_rt_bandwidth(&tg->rt_bandwidth, |
8559 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8560 | |
8561 | for_each_possible_cpu(i) { | |
8562 | rq = cpu_rq(i); | |
8563 | ||
eab17229 LZ |
8564 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8565 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8566 | if (!rt_rq) |
8567 | goto err; | |
29f59db3 | 8568 | |
eab17229 LZ |
8569 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8570 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8571 | if (!rt_se) |
dfc12eb2 | 8572 | goto err_free_rq; |
29f59db3 | 8573 | |
3d4b47b4 | 8574 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8575 | } |
8576 | ||
bccbe08a PZ |
8577 | return 1; |
8578 | ||
49246274 | 8579 | err_free_rq: |
dfc12eb2 | 8580 | kfree(rt_rq); |
49246274 | 8581 | err: |
bccbe08a PZ |
8582 | return 0; |
8583 | } | |
6d6bc0ad | 8584 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8585 | static inline void free_rt_sched_group(struct task_group *tg) |
8586 | { | |
8587 | } | |
8588 | ||
ec7dc8ac DG |
8589 | static inline |
8590 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8591 | { |
8592 | return 1; | |
8593 | } | |
6d6bc0ad | 8594 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8595 | |
7c941438 | 8596 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8597 | static void free_sched_group(struct task_group *tg) |
8598 | { | |
8599 | free_fair_sched_group(tg); | |
8600 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8601 | autogroup_free(tg); |
bccbe08a PZ |
8602 | kfree(tg); |
8603 | } | |
8604 | ||
8605 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8606 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8607 | { |
8608 | struct task_group *tg; | |
8609 | unsigned long flags; | |
bccbe08a PZ |
8610 | |
8611 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8612 | if (!tg) | |
8613 | return ERR_PTR(-ENOMEM); | |
8614 | ||
ec7dc8ac | 8615 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8616 | goto err; |
8617 | ||
ec7dc8ac | 8618 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8619 | goto err; |
8620 | ||
8ed36996 | 8621 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8622 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8623 | |
8624 | WARN_ON(!parent); /* root should already exist */ | |
8625 | ||
8626 | tg->parent = parent; | |
f473aa5e | 8627 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8628 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8629 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8630 | |
9b5b7751 | 8631 | return tg; |
29f59db3 SV |
8632 | |
8633 | err: | |
6f505b16 | 8634 | free_sched_group(tg); |
29f59db3 SV |
8635 | return ERR_PTR(-ENOMEM); |
8636 | } | |
8637 | ||
9b5b7751 | 8638 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8639 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8640 | { |
29f59db3 | 8641 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8642 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8643 | } |
8644 | ||
9b5b7751 | 8645 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8646 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8647 | { |
8ed36996 | 8648 | unsigned long flags; |
9b5b7751 | 8649 | int i; |
29f59db3 | 8650 | |
3d4b47b4 PZ |
8651 | /* end participation in shares distribution */ |
8652 | for_each_possible_cpu(i) | |
bccbe08a | 8653 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8654 | |
8655 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8656 | list_del_rcu(&tg->list); |
f473aa5e | 8657 | list_del_rcu(&tg->siblings); |
8ed36996 | 8658 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8659 | |
9b5b7751 | 8660 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8661 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8662 | } |
8663 | ||
9b5b7751 | 8664 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8665 | * The caller of this function should have put the task in its new group |
8666 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8667 | * reflect its new group. | |
9b5b7751 SV |
8668 | */ |
8669 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8670 | { |
8671 | int on_rq, running; | |
8672 | unsigned long flags; | |
8673 | struct rq *rq; | |
8674 | ||
8675 | rq = task_rq_lock(tsk, &flags); | |
8676 | ||
051a1d1a | 8677 | running = task_current(rq, tsk); |
29f59db3 SV |
8678 | on_rq = tsk->se.on_rq; |
8679 | ||
0e1f3483 | 8680 | if (on_rq) |
29f59db3 | 8681 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8682 | if (unlikely(running)) |
8683 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8684 | |
810b3817 | 8685 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8686 | if (tsk->sched_class->task_move_group) |
8687 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8688 | else | |
810b3817 | 8689 | #endif |
b2b5ce02 | 8690 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8691 | |
0e1f3483 HS |
8692 | if (unlikely(running)) |
8693 | tsk->sched_class->set_curr_task(rq); | |
8694 | if (on_rq) | |
371fd7e7 | 8695 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8696 | |
29f59db3 SV |
8697 | task_rq_unlock(rq, &flags); |
8698 | } | |
7c941438 | 8699 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8700 | |
052f1dc7 | 8701 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8702 | static DEFINE_MUTEX(shares_mutex); |
8703 | ||
4cf86d77 | 8704 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8705 | { |
8706 | int i; | |
8ed36996 | 8707 | unsigned long flags; |
c61935fd | 8708 | |
ec7dc8ac DG |
8709 | /* |
8710 | * We can't change the weight of the root cgroup. | |
8711 | */ | |
8712 | if (!tg->se[0]) | |
8713 | return -EINVAL; | |
8714 | ||
18d95a28 PZ |
8715 | if (shares < MIN_SHARES) |
8716 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8717 | else if (shares > MAX_SHARES) |
8718 | shares = MAX_SHARES; | |
62fb1851 | 8719 | |
8ed36996 | 8720 | mutex_lock(&shares_mutex); |
9b5b7751 | 8721 | if (tg->shares == shares) |
5cb350ba | 8722 | goto done; |
29f59db3 | 8723 | |
9b5b7751 | 8724 | tg->shares = shares; |
c09595f6 | 8725 | for_each_possible_cpu(i) { |
9437178f PT |
8726 | struct rq *rq = cpu_rq(i); |
8727 | struct sched_entity *se; | |
8728 | ||
8729 | se = tg->se[i]; | |
8730 | /* Propagate contribution to hierarchy */ | |
8731 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8732 | for_each_sched_entity(se) | |
6d5ab293 | 8733 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8734 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8735 | } |
29f59db3 | 8736 | |
5cb350ba | 8737 | done: |
8ed36996 | 8738 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8739 | return 0; |
29f59db3 SV |
8740 | } |
8741 | ||
5cb350ba DG |
8742 | unsigned long sched_group_shares(struct task_group *tg) |
8743 | { | |
8744 | return tg->shares; | |
8745 | } | |
052f1dc7 | 8746 | #endif |
5cb350ba | 8747 | |
052f1dc7 | 8748 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8749 | /* |
9f0c1e56 | 8750 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8751 | */ |
9f0c1e56 PZ |
8752 | static DEFINE_MUTEX(rt_constraints_mutex); |
8753 | ||
8754 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8755 | { | |
8756 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8757 | return 1ULL << 20; |
9f0c1e56 | 8758 | |
9a7e0b18 | 8759 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8760 | } |
8761 | ||
9a7e0b18 PZ |
8762 | /* Must be called with tasklist_lock held */ |
8763 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8764 | { |
9a7e0b18 | 8765 | struct task_struct *g, *p; |
b40b2e8e | 8766 | |
9a7e0b18 PZ |
8767 | do_each_thread(g, p) { |
8768 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8769 | return 1; | |
8770 | } while_each_thread(g, p); | |
b40b2e8e | 8771 | |
9a7e0b18 PZ |
8772 | return 0; |
8773 | } | |
b40b2e8e | 8774 | |
9a7e0b18 PZ |
8775 | struct rt_schedulable_data { |
8776 | struct task_group *tg; | |
8777 | u64 rt_period; | |
8778 | u64 rt_runtime; | |
8779 | }; | |
b40b2e8e | 8780 | |
9a7e0b18 PZ |
8781 | static int tg_schedulable(struct task_group *tg, void *data) |
8782 | { | |
8783 | struct rt_schedulable_data *d = data; | |
8784 | struct task_group *child; | |
8785 | unsigned long total, sum = 0; | |
8786 | u64 period, runtime; | |
b40b2e8e | 8787 | |
9a7e0b18 PZ |
8788 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8789 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8790 | |
9a7e0b18 PZ |
8791 | if (tg == d->tg) { |
8792 | period = d->rt_period; | |
8793 | runtime = d->rt_runtime; | |
b40b2e8e | 8794 | } |
b40b2e8e | 8795 | |
4653f803 PZ |
8796 | /* |
8797 | * Cannot have more runtime than the period. | |
8798 | */ | |
8799 | if (runtime > period && runtime != RUNTIME_INF) | |
8800 | return -EINVAL; | |
6f505b16 | 8801 | |
4653f803 PZ |
8802 | /* |
8803 | * Ensure we don't starve existing RT tasks. | |
8804 | */ | |
9a7e0b18 PZ |
8805 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8806 | return -EBUSY; | |
6f505b16 | 8807 | |
9a7e0b18 | 8808 | total = to_ratio(period, runtime); |
6f505b16 | 8809 | |
4653f803 PZ |
8810 | /* |
8811 | * Nobody can have more than the global setting allows. | |
8812 | */ | |
8813 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8814 | return -EINVAL; | |
6f505b16 | 8815 | |
4653f803 PZ |
8816 | /* |
8817 | * The sum of our children's runtime should not exceed our own. | |
8818 | */ | |
9a7e0b18 PZ |
8819 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8820 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8821 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8822 | |
9a7e0b18 PZ |
8823 | if (child == d->tg) { |
8824 | period = d->rt_period; | |
8825 | runtime = d->rt_runtime; | |
8826 | } | |
6f505b16 | 8827 | |
9a7e0b18 | 8828 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8829 | } |
6f505b16 | 8830 | |
9a7e0b18 PZ |
8831 | if (sum > total) |
8832 | return -EINVAL; | |
8833 | ||
8834 | return 0; | |
6f505b16 PZ |
8835 | } |
8836 | ||
9a7e0b18 | 8837 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8838 | { |
9a7e0b18 PZ |
8839 | struct rt_schedulable_data data = { |
8840 | .tg = tg, | |
8841 | .rt_period = period, | |
8842 | .rt_runtime = runtime, | |
8843 | }; | |
8844 | ||
8845 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8846 | } |
8847 | ||
d0b27fa7 PZ |
8848 | static int tg_set_bandwidth(struct task_group *tg, |
8849 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8850 | { |
ac086bc2 | 8851 | int i, err = 0; |
9f0c1e56 | 8852 | |
9f0c1e56 | 8853 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8854 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8855 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8856 | if (err) | |
9f0c1e56 | 8857 | goto unlock; |
ac086bc2 | 8858 | |
0986b11b | 8859 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8860 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8861 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8862 | |
8863 | for_each_possible_cpu(i) { | |
8864 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8865 | ||
0986b11b | 8866 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8867 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8868 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8869 | } |
0986b11b | 8870 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8871 | unlock: |
521f1a24 | 8872 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8873 | mutex_unlock(&rt_constraints_mutex); |
8874 | ||
8875 | return err; | |
6f505b16 PZ |
8876 | } |
8877 | ||
d0b27fa7 PZ |
8878 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8879 | { | |
8880 | u64 rt_runtime, rt_period; | |
8881 | ||
8882 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8883 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8884 | if (rt_runtime_us < 0) | |
8885 | rt_runtime = RUNTIME_INF; | |
8886 | ||
8887 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8888 | } | |
8889 | ||
9f0c1e56 PZ |
8890 | long sched_group_rt_runtime(struct task_group *tg) |
8891 | { | |
8892 | u64 rt_runtime_us; | |
8893 | ||
d0b27fa7 | 8894 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8895 | return -1; |
8896 | ||
d0b27fa7 | 8897 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8898 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8899 | return rt_runtime_us; | |
8900 | } | |
d0b27fa7 PZ |
8901 | |
8902 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8903 | { | |
8904 | u64 rt_runtime, rt_period; | |
8905 | ||
8906 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8907 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8908 | ||
619b0488 R |
8909 | if (rt_period == 0) |
8910 | return -EINVAL; | |
8911 | ||
d0b27fa7 PZ |
8912 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8913 | } | |
8914 | ||
8915 | long sched_group_rt_period(struct task_group *tg) | |
8916 | { | |
8917 | u64 rt_period_us; | |
8918 | ||
8919 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8920 | do_div(rt_period_us, NSEC_PER_USEC); | |
8921 | return rt_period_us; | |
8922 | } | |
8923 | ||
8924 | static int sched_rt_global_constraints(void) | |
8925 | { | |
4653f803 | 8926 | u64 runtime, period; |
d0b27fa7 PZ |
8927 | int ret = 0; |
8928 | ||
ec5d4989 HS |
8929 | if (sysctl_sched_rt_period <= 0) |
8930 | return -EINVAL; | |
8931 | ||
4653f803 PZ |
8932 | runtime = global_rt_runtime(); |
8933 | period = global_rt_period(); | |
8934 | ||
8935 | /* | |
8936 | * Sanity check on the sysctl variables. | |
8937 | */ | |
8938 | if (runtime > period && runtime != RUNTIME_INF) | |
8939 | return -EINVAL; | |
10b612f4 | 8940 | |
d0b27fa7 | 8941 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8942 | read_lock(&tasklist_lock); |
4653f803 | 8943 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8944 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8945 | mutex_unlock(&rt_constraints_mutex); |
8946 | ||
8947 | return ret; | |
8948 | } | |
54e99124 DG |
8949 | |
8950 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8951 | { | |
8952 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8953 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8954 | return 0; | |
8955 | ||
8956 | return 1; | |
8957 | } | |
8958 | ||
6d6bc0ad | 8959 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8960 | static int sched_rt_global_constraints(void) |
8961 | { | |
ac086bc2 PZ |
8962 | unsigned long flags; |
8963 | int i; | |
8964 | ||
ec5d4989 HS |
8965 | if (sysctl_sched_rt_period <= 0) |
8966 | return -EINVAL; | |
8967 | ||
60aa605d PZ |
8968 | /* |
8969 | * There's always some RT tasks in the root group | |
8970 | * -- migration, kstopmachine etc.. | |
8971 | */ | |
8972 | if (sysctl_sched_rt_runtime == 0) | |
8973 | return -EBUSY; | |
8974 | ||
0986b11b | 8975 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8976 | for_each_possible_cpu(i) { |
8977 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8978 | ||
0986b11b | 8979 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8980 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8981 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8982 | } |
0986b11b | 8983 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8984 | |
d0b27fa7 PZ |
8985 | return 0; |
8986 | } | |
6d6bc0ad | 8987 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8988 | |
8989 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8990 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8991 | loff_t *ppos) |
8992 | { | |
8993 | int ret; | |
8994 | int old_period, old_runtime; | |
8995 | static DEFINE_MUTEX(mutex); | |
8996 | ||
8997 | mutex_lock(&mutex); | |
8998 | old_period = sysctl_sched_rt_period; | |
8999 | old_runtime = sysctl_sched_rt_runtime; | |
9000 | ||
8d65af78 | 9001 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9002 | |
9003 | if (!ret && write) { | |
9004 | ret = sched_rt_global_constraints(); | |
9005 | if (ret) { | |
9006 | sysctl_sched_rt_period = old_period; | |
9007 | sysctl_sched_rt_runtime = old_runtime; | |
9008 | } else { | |
9009 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9010 | def_rt_bandwidth.rt_period = | |
9011 | ns_to_ktime(global_rt_period()); | |
9012 | } | |
9013 | } | |
9014 | mutex_unlock(&mutex); | |
9015 | ||
9016 | return ret; | |
9017 | } | |
68318b8e | 9018 | |
052f1dc7 | 9019 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9020 | |
9021 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9022 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9023 | { |
2b01dfe3 PM |
9024 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9025 | struct task_group, css); | |
68318b8e SV |
9026 | } |
9027 | ||
9028 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9029 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9030 | { |
ec7dc8ac | 9031 | struct task_group *tg, *parent; |
68318b8e | 9032 | |
2b01dfe3 | 9033 | if (!cgrp->parent) { |
68318b8e | 9034 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9035 | return &root_task_group.css; |
68318b8e SV |
9036 | } |
9037 | ||
ec7dc8ac DG |
9038 | parent = cgroup_tg(cgrp->parent); |
9039 | tg = sched_create_group(parent); | |
68318b8e SV |
9040 | if (IS_ERR(tg)) |
9041 | return ERR_PTR(-ENOMEM); | |
9042 | ||
68318b8e SV |
9043 | return &tg->css; |
9044 | } | |
9045 | ||
41a2d6cf IM |
9046 | static void |
9047 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9048 | { |
2b01dfe3 | 9049 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9050 | |
9051 | sched_destroy_group(tg); | |
9052 | } | |
9053 | ||
41a2d6cf | 9054 | static int |
be367d09 | 9055 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9056 | { |
b68aa230 | 9057 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9058 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9059 | return -EINVAL; |
9060 | #else | |
68318b8e SV |
9061 | /* We don't support RT-tasks being in separate groups */ |
9062 | if (tsk->sched_class != &fair_sched_class) | |
9063 | return -EINVAL; | |
b68aa230 | 9064 | #endif |
be367d09 BB |
9065 | return 0; |
9066 | } | |
68318b8e | 9067 | |
be367d09 BB |
9068 | static int |
9069 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9070 | struct task_struct *tsk, bool threadgroup) | |
9071 | { | |
9072 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9073 | if (retval) | |
9074 | return retval; | |
9075 | if (threadgroup) { | |
9076 | struct task_struct *c; | |
9077 | rcu_read_lock(); | |
9078 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9079 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9080 | if (retval) { | |
9081 | rcu_read_unlock(); | |
9082 | return retval; | |
9083 | } | |
9084 | } | |
9085 | rcu_read_unlock(); | |
9086 | } | |
68318b8e SV |
9087 | return 0; |
9088 | } | |
9089 | ||
9090 | static void | |
2b01dfe3 | 9091 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9092 | struct cgroup *old_cont, struct task_struct *tsk, |
9093 | bool threadgroup) | |
68318b8e SV |
9094 | { |
9095 | sched_move_task(tsk); | |
be367d09 BB |
9096 | if (threadgroup) { |
9097 | struct task_struct *c; | |
9098 | rcu_read_lock(); | |
9099 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9100 | sched_move_task(c); | |
9101 | } | |
9102 | rcu_read_unlock(); | |
9103 | } | |
68318b8e SV |
9104 | } |
9105 | ||
068c5cc5 PZ |
9106 | static void |
9107 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct task_struct *task) | |
9108 | { | |
9109 | /* | |
9110 | * cgroup_exit() is called in the copy_process() failure path. | |
9111 | * Ignore this case since the task hasn't ran yet, this avoids | |
9112 | * trying to poke a half freed task state from generic code. | |
9113 | */ | |
9114 | if (!(task->flags & PF_EXITING)) | |
9115 | return; | |
9116 | ||
9117 | sched_move_task(task); | |
9118 | } | |
9119 | ||
052f1dc7 | 9120 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9121 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9122 | u64 shareval) |
68318b8e | 9123 | { |
2b01dfe3 | 9124 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9125 | } |
9126 | ||
f4c753b7 | 9127 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9128 | { |
2b01dfe3 | 9129 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9130 | |
9131 | return (u64) tg->shares; | |
9132 | } | |
6d6bc0ad | 9133 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9134 | |
052f1dc7 | 9135 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9136 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9137 | s64 val) |
6f505b16 | 9138 | { |
06ecb27c | 9139 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9140 | } |
9141 | ||
06ecb27c | 9142 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9143 | { |
06ecb27c | 9144 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9145 | } |
d0b27fa7 PZ |
9146 | |
9147 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9148 | u64 rt_period_us) | |
9149 | { | |
9150 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9151 | } | |
9152 | ||
9153 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9154 | { | |
9155 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9156 | } | |
6d6bc0ad | 9157 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9158 | |
fe5c7cc2 | 9159 | static struct cftype cpu_files[] = { |
052f1dc7 | 9160 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9161 | { |
9162 | .name = "shares", | |
f4c753b7 PM |
9163 | .read_u64 = cpu_shares_read_u64, |
9164 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9165 | }, |
052f1dc7 PZ |
9166 | #endif |
9167 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9168 | { |
9f0c1e56 | 9169 | .name = "rt_runtime_us", |
06ecb27c PM |
9170 | .read_s64 = cpu_rt_runtime_read, |
9171 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9172 | }, |
d0b27fa7 PZ |
9173 | { |
9174 | .name = "rt_period_us", | |
f4c753b7 PM |
9175 | .read_u64 = cpu_rt_period_read_uint, |
9176 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9177 | }, |
052f1dc7 | 9178 | #endif |
68318b8e SV |
9179 | }; |
9180 | ||
9181 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9182 | { | |
fe5c7cc2 | 9183 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9184 | } |
9185 | ||
9186 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9187 | .name = "cpu", |
9188 | .create = cpu_cgroup_create, | |
9189 | .destroy = cpu_cgroup_destroy, | |
9190 | .can_attach = cpu_cgroup_can_attach, | |
9191 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 9192 | .exit = cpu_cgroup_exit, |
38605cae IM |
9193 | .populate = cpu_cgroup_populate, |
9194 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9195 | .early_init = 1, |
9196 | }; | |
9197 | ||
052f1dc7 | 9198 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9199 | |
9200 | #ifdef CONFIG_CGROUP_CPUACCT | |
9201 | ||
9202 | /* | |
9203 | * CPU accounting code for task groups. | |
9204 | * | |
9205 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9206 | * (balbir@in.ibm.com). | |
9207 | */ | |
9208 | ||
934352f2 | 9209 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9210 | struct cpuacct { |
9211 | struct cgroup_subsys_state css; | |
9212 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9213 | u64 __percpu *cpuusage; |
ef12fefa | 9214 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9215 | struct cpuacct *parent; |
d842de87 SV |
9216 | }; |
9217 | ||
9218 | struct cgroup_subsys cpuacct_subsys; | |
9219 | ||
9220 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9221 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9222 | { |
32cd756a | 9223 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9224 | struct cpuacct, css); |
9225 | } | |
9226 | ||
9227 | /* return cpu accounting group to which this task belongs */ | |
9228 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9229 | { | |
9230 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9231 | struct cpuacct, css); | |
9232 | } | |
9233 | ||
9234 | /* create a new cpu accounting group */ | |
9235 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9236 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9237 | { |
9238 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9239 | int i; |
d842de87 SV |
9240 | |
9241 | if (!ca) | |
ef12fefa | 9242 | goto out; |
d842de87 SV |
9243 | |
9244 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9245 | if (!ca->cpuusage) |
9246 | goto out_free_ca; | |
9247 | ||
9248 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9249 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9250 | goto out_free_counters; | |
d842de87 | 9251 | |
934352f2 BR |
9252 | if (cgrp->parent) |
9253 | ca->parent = cgroup_ca(cgrp->parent); | |
9254 | ||
d842de87 | 9255 | return &ca->css; |
ef12fefa BR |
9256 | |
9257 | out_free_counters: | |
9258 | while (--i >= 0) | |
9259 | percpu_counter_destroy(&ca->cpustat[i]); | |
9260 | free_percpu(ca->cpuusage); | |
9261 | out_free_ca: | |
9262 | kfree(ca); | |
9263 | out: | |
9264 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9265 | } |
9266 | ||
9267 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9268 | static void |
32cd756a | 9269 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9270 | { |
32cd756a | 9271 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9272 | int i; |
d842de87 | 9273 | |
ef12fefa BR |
9274 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9275 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9276 | free_percpu(ca->cpuusage); |
9277 | kfree(ca); | |
9278 | } | |
9279 | ||
720f5498 KC |
9280 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9281 | { | |
b36128c8 | 9282 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9283 | u64 data; |
9284 | ||
9285 | #ifndef CONFIG_64BIT | |
9286 | /* | |
9287 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9288 | */ | |
05fa785c | 9289 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9290 | data = *cpuusage; |
05fa785c | 9291 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9292 | #else |
9293 | data = *cpuusage; | |
9294 | #endif | |
9295 | ||
9296 | return data; | |
9297 | } | |
9298 | ||
9299 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9300 | { | |
b36128c8 | 9301 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9302 | |
9303 | #ifndef CONFIG_64BIT | |
9304 | /* | |
9305 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9306 | */ | |
05fa785c | 9307 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9308 | *cpuusage = val; |
05fa785c | 9309 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9310 | #else |
9311 | *cpuusage = val; | |
9312 | #endif | |
9313 | } | |
9314 | ||
d842de87 | 9315 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9316 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9317 | { |
32cd756a | 9318 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9319 | u64 totalcpuusage = 0; |
9320 | int i; | |
9321 | ||
720f5498 KC |
9322 | for_each_present_cpu(i) |
9323 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9324 | |
9325 | return totalcpuusage; | |
9326 | } | |
9327 | ||
0297b803 DG |
9328 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9329 | u64 reset) | |
9330 | { | |
9331 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9332 | int err = 0; | |
9333 | int i; | |
9334 | ||
9335 | if (reset) { | |
9336 | err = -EINVAL; | |
9337 | goto out; | |
9338 | } | |
9339 | ||
720f5498 KC |
9340 | for_each_present_cpu(i) |
9341 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9342 | |
0297b803 DG |
9343 | out: |
9344 | return err; | |
9345 | } | |
9346 | ||
e9515c3c KC |
9347 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9348 | struct seq_file *m) | |
9349 | { | |
9350 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9351 | u64 percpu; | |
9352 | int i; | |
9353 | ||
9354 | for_each_present_cpu(i) { | |
9355 | percpu = cpuacct_cpuusage_read(ca, i); | |
9356 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9357 | } | |
9358 | seq_printf(m, "\n"); | |
9359 | return 0; | |
9360 | } | |
9361 | ||
ef12fefa BR |
9362 | static const char *cpuacct_stat_desc[] = { |
9363 | [CPUACCT_STAT_USER] = "user", | |
9364 | [CPUACCT_STAT_SYSTEM] = "system", | |
9365 | }; | |
9366 | ||
9367 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9368 | struct cgroup_map_cb *cb) | |
9369 | { | |
9370 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9371 | int i; | |
9372 | ||
9373 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9374 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9375 | val = cputime64_to_clock_t(val); | |
9376 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9377 | } | |
9378 | return 0; | |
9379 | } | |
9380 | ||
d842de87 SV |
9381 | static struct cftype files[] = { |
9382 | { | |
9383 | .name = "usage", | |
f4c753b7 PM |
9384 | .read_u64 = cpuusage_read, |
9385 | .write_u64 = cpuusage_write, | |
d842de87 | 9386 | }, |
e9515c3c KC |
9387 | { |
9388 | .name = "usage_percpu", | |
9389 | .read_seq_string = cpuacct_percpu_seq_read, | |
9390 | }, | |
ef12fefa BR |
9391 | { |
9392 | .name = "stat", | |
9393 | .read_map = cpuacct_stats_show, | |
9394 | }, | |
d842de87 SV |
9395 | }; |
9396 | ||
32cd756a | 9397 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9398 | { |
32cd756a | 9399 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9400 | } |
9401 | ||
9402 | /* | |
9403 | * charge this task's execution time to its accounting group. | |
9404 | * | |
9405 | * called with rq->lock held. | |
9406 | */ | |
9407 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9408 | { | |
9409 | struct cpuacct *ca; | |
934352f2 | 9410 | int cpu; |
d842de87 | 9411 | |
c40c6f85 | 9412 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9413 | return; |
9414 | ||
934352f2 | 9415 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9416 | |
9417 | rcu_read_lock(); | |
9418 | ||
d842de87 | 9419 | ca = task_ca(tsk); |
d842de87 | 9420 | |
934352f2 | 9421 | for (; ca; ca = ca->parent) { |
b36128c8 | 9422 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9423 | *cpuusage += cputime; |
9424 | } | |
a18b83b7 BR |
9425 | |
9426 | rcu_read_unlock(); | |
d842de87 SV |
9427 | } |
9428 | ||
fa535a77 AB |
9429 | /* |
9430 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9431 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9432 | * percpu_counter_add with values large enough to always overflow the | |
9433 | * per cpu batch limit causing bad SMP scalability. | |
9434 | * | |
9435 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9436 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9437 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9438 | */ | |
9439 | #ifdef CONFIG_SMP | |
9440 | #define CPUACCT_BATCH \ | |
9441 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9442 | #else | |
9443 | #define CPUACCT_BATCH 0 | |
9444 | #endif | |
9445 | ||
ef12fefa BR |
9446 | /* |
9447 | * Charge the system/user time to the task's accounting group. | |
9448 | */ | |
9449 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9450 | enum cpuacct_stat_index idx, cputime_t val) | |
9451 | { | |
9452 | struct cpuacct *ca; | |
fa535a77 | 9453 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9454 | |
9455 | if (unlikely(!cpuacct_subsys.active)) | |
9456 | return; | |
9457 | ||
9458 | rcu_read_lock(); | |
9459 | ca = task_ca(tsk); | |
9460 | ||
9461 | do { | |
fa535a77 | 9462 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9463 | ca = ca->parent; |
9464 | } while (ca); | |
9465 | rcu_read_unlock(); | |
9466 | } | |
9467 | ||
d842de87 SV |
9468 | struct cgroup_subsys cpuacct_subsys = { |
9469 | .name = "cpuacct", | |
9470 | .create = cpuacct_create, | |
9471 | .destroy = cpuacct_destroy, | |
9472 | .populate = cpuacct_populate, | |
9473 | .subsys_id = cpuacct_subsys_id, | |
9474 | }; | |
9475 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9476 |