Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
969c7921 | 59 | #include <linux/stop_machine.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 PZ |
70 | #include <linux/debugfs.h> |
71 | #include <linux/ctype.h> | |
6cd8a4bb | 72 | #include <linux/ftrace.h> |
5a0e3ad6 | 73 | #include <linux/slab.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
335d7afb | 77 | #include <asm/mutex.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
5091faa4 | 81 | #include "sched_autogroup.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee | 233 | /* |
c4a8849a | 234 | * sched_domains_mutex serializes calls to init_sched_domains, |
712555ee HC |
235 | * detach_destroy_domains and partition_sched_domains. |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
5091faa4 MG |
274 | |
275 | #ifdef CONFIG_SCHED_AUTOGROUP | |
276 | struct autogroup *autogroup; | |
277 | #endif | |
29f59db3 SV |
278 | }; |
279 | ||
3d4b47b4 | 280 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 281 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 282 | |
e9036b36 CG |
283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
284 | ||
07e06b01 | 285 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 286 | |
cb4ad1ff | 287 | /* |
2e084786 LJ |
288 | * A weight of 0 or 1 can cause arithmetics problems. |
289 | * A weight of a cfs_rq is the sum of weights of which entities | |
290 | * are queued on this cfs_rq, so a weight of a entity should not be | |
291 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
292 | * (The default weight is 1024 - so there's no practical |
293 | * limitation from this.) | |
294 | */ | |
18d95a28 | 295 | #define MIN_SHARES 2 |
2e084786 | 296 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 297 | |
07e06b01 | 298 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
299 | #endif |
300 | ||
29f59db3 | 301 | /* Default task group. |
3a252015 | 302 | * Every task in system belong to this group at bootup. |
29f59db3 | 303 | */ |
07e06b01 | 304 | struct task_group root_task_group; |
29f59db3 | 305 | |
7c941438 | 306 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 307 | |
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
3fe1698b PZ |
315 | #ifndef CONFIG_64BIT |
316 | u64 min_vruntime_copy; | |
317 | #endif | |
6aa645ea IM |
318 | |
319 | struct rb_root tasks_timeline; | |
320 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
321 | |
322 | struct list_head tasks; | |
323 | struct list_head *balance_iterator; | |
324 | ||
325 | /* | |
326 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
327 | * It is set to NULL otherwise (i.e when none are currently running). |
328 | */ | |
ac53db59 | 329 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 330 | |
4934a4d3 | 331 | #ifdef CONFIG_SCHED_DEBUG |
5ac5c4d6 | 332 | unsigned int nr_spread_over; |
4934a4d3 | 333 | #endif |
ddc97297 | 334 | |
62160e3f | 335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
336 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
337 | ||
41a2d6cf IM |
338 | /* |
339 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
340 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
341 | * (like users, containers etc.) | |
342 | * | |
343 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
344 | * list is used during load balance. | |
345 | */ | |
3d4b47b4 | 346 | int on_list; |
41a2d6cf IM |
347 | struct list_head leaf_cfs_rq_list; |
348 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
349 | |
350 | #ifdef CONFIG_SMP | |
c09595f6 | 351 | /* |
c8cba857 | 352 | * the part of load.weight contributed by tasks |
c09595f6 | 353 | */ |
c8cba857 | 354 | unsigned long task_weight; |
c09595f6 | 355 | |
c8cba857 PZ |
356 | /* |
357 | * h_load = weight * f(tg) | |
358 | * | |
359 | * Where f(tg) is the recursive weight fraction assigned to | |
360 | * this group. | |
361 | */ | |
362 | unsigned long h_load; | |
c09595f6 | 363 | |
c8cba857 | 364 | /* |
3b3d190e PT |
365 | * Maintaining per-cpu shares distribution for group scheduling |
366 | * | |
367 | * load_stamp is the last time we updated the load average | |
368 | * load_last is the last time we updated the load average and saw load | |
369 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 370 | */ |
2069dd75 PZ |
371 | u64 load_avg; |
372 | u64 load_period; | |
3b3d190e | 373 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 374 | |
2069dd75 | 375 | unsigned long load_contribution; |
c09595f6 | 376 | #endif |
6aa645ea IM |
377 | #endif |
378 | }; | |
1da177e4 | 379 | |
6aa645ea IM |
380 | /* Real-Time classes' related field in a runqueue: */ |
381 | struct rt_rq { | |
382 | struct rt_prio_array active; | |
63489e45 | 383 | unsigned long rt_nr_running; |
052f1dc7 | 384 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
385 | struct { |
386 | int curr; /* highest queued rt task prio */ | |
398a153b | 387 | #ifdef CONFIG_SMP |
e864c499 | 388 | int next; /* next highest */ |
398a153b | 389 | #endif |
e864c499 | 390 | } highest_prio; |
6f505b16 | 391 | #endif |
fa85ae24 | 392 | #ifdef CONFIG_SMP |
73fe6aae | 393 | unsigned long rt_nr_migratory; |
a1ba4d8b | 394 | unsigned long rt_nr_total; |
a22d7fc1 | 395 | int overloaded; |
917b627d | 396 | struct plist_head pushable_tasks; |
fa85ae24 | 397 | #endif |
6f505b16 | 398 | int rt_throttled; |
fa85ae24 | 399 | u64 rt_time; |
ac086bc2 | 400 | u64 rt_runtime; |
ea736ed5 | 401 | /* Nests inside the rq lock: */ |
0986b11b | 402 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 403 | |
052f1dc7 | 404 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
405 | unsigned long rt_nr_boosted; |
406 | ||
6f505b16 PZ |
407 | struct rq *rq; |
408 | struct list_head leaf_rt_rq_list; | |
409 | struct task_group *tg; | |
6f505b16 | 410 | #endif |
6aa645ea IM |
411 | }; |
412 | ||
57d885fe GH |
413 | #ifdef CONFIG_SMP |
414 | ||
415 | /* | |
416 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
417 | * variables. Each exclusive cpuset essentially defines an island domain by |
418 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
419 | * exclusive cpuset is created, we also create and attach a new root-domain |
420 | * object. | |
421 | * | |
57d885fe GH |
422 | */ |
423 | struct root_domain { | |
424 | atomic_t refcount; | |
dce840a0 | 425 | struct rcu_head rcu; |
c6c4927b RR |
426 | cpumask_var_t span; |
427 | cpumask_var_t online; | |
637f5085 | 428 | |
0eab9146 | 429 | /* |
637f5085 GH |
430 | * The "RT overload" flag: it gets set if a CPU has more than |
431 | * one runnable RT task. | |
432 | */ | |
c6c4927b | 433 | cpumask_var_t rto_mask; |
0eab9146 | 434 | atomic_t rto_count; |
6e0534f2 | 435 | struct cpupri cpupri; |
57d885fe GH |
436 | }; |
437 | ||
dc938520 GH |
438 | /* |
439 | * By default the system creates a single root-domain with all cpus as | |
440 | * members (mimicking the global state we have today). | |
441 | */ | |
57d885fe GH |
442 | static struct root_domain def_root_domain; |
443 | ||
ed2d372c | 444 | #endif /* CONFIG_SMP */ |
57d885fe | 445 | |
1da177e4 LT |
446 | /* |
447 | * This is the main, per-CPU runqueue data structure. | |
448 | * | |
449 | * Locking rule: those places that want to lock multiple runqueues | |
450 | * (such as the load balancing or the thread migration code), lock | |
451 | * acquire operations must be ordered by ascending &runqueue. | |
452 | */ | |
70b97a7f | 453 | struct rq { |
d8016491 | 454 | /* runqueue lock: */ |
05fa785c | 455 | raw_spinlock_t lock; |
1da177e4 LT |
456 | |
457 | /* | |
458 | * nr_running and cpu_load should be in the same cacheline because | |
459 | * remote CPUs use both these fields when doing load calculation. | |
460 | */ | |
461 | unsigned long nr_running; | |
6aa645ea IM |
462 | #define CPU_LOAD_IDX_MAX 5 |
463 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 464 | unsigned long last_load_update_tick; |
46cb4b7c | 465 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 466 | u64 nohz_stamp; |
83cd4fe2 | 467 | unsigned char nohz_balance_kick; |
46cb4b7c | 468 | #endif |
61eadef6 | 469 | int skip_clock_update; |
a64692a3 | 470 | |
d8016491 IM |
471 | /* capture load from *all* tasks on this cpu: */ |
472 | struct load_weight load; | |
6aa645ea IM |
473 | unsigned long nr_load_updates; |
474 | u64 nr_switches; | |
475 | ||
476 | struct cfs_rq cfs; | |
6f505b16 | 477 | struct rt_rq rt; |
6f505b16 | 478 | |
6aa645ea | 479 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
480 | /* list of leaf cfs_rq on this cpu: */ |
481 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
482 | #endif |
483 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 484 | struct list_head leaf_rt_rq_list; |
1da177e4 | 485 | #endif |
1da177e4 LT |
486 | |
487 | /* | |
488 | * This is part of a global counter where only the total sum | |
489 | * over all CPUs matters. A task can increase this counter on | |
490 | * one CPU and if it got migrated afterwards it may decrease | |
491 | * it on another CPU. Always updated under the runqueue lock: | |
492 | */ | |
493 | unsigned long nr_uninterruptible; | |
494 | ||
34f971f6 | 495 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 496 | unsigned long next_balance; |
1da177e4 | 497 | struct mm_struct *prev_mm; |
6aa645ea | 498 | |
3e51f33f | 499 | u64 clock; |
305e6835 | 500 | u64 clock_task; |
6aa645ea | 501 | |
1da177e4 LT |
502 | atomic_t nr_iowait; |
503 | ||
504 | #ifdef CONFIG_SMP | |
0eab9146 | 505 | struct root_domain *rd; |
1da177e4 LT |
506 | struct sched_domain *sd; |
507 | ||
e51fd5e2 PZ |
508 | unsigned long cpu_power; |
509 | ||
a0a522ce | 510 | unsigned char idle_at_tick; |
1da177e4 | 511 | /* For active balancing */ |
3f029d3c | 512 | int post_schedule; |
1da177e4 LT |
513 | int active_balance; |
514 | int push_cpu; | |
969c7921 | 515 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
516 | /* cpu of this runqueue: */ |
517 | int cpu; | |
1f11eb6a | 518 | int online; |
1da177e4 | 519 | |
a8a51d5e | 520 | unsigned long avg_load_per_task; |
1da177e4 | 521 | |
e9e9250b PZ |
522 | u64 rt_avg; |
523 | u64 age_stamp; | |
1b9508f6 MG |
524 | u64 idle_stamp; |
525 | u64 avg_idle; | |
1da177e4 LT |
526 | #endif |
527 | ||
aa483808 VP |
528 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
529 | u64 prev_irq_time; | |
530 | #endif | |
531 | ||
dce48a84 TG |
532 | /* calc_load related fields */ |
533 | unsigned long calc_load_update; | |
534 | long calc_load_active; | |
535 | ||
8f4d37ec | 536 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
537 | #ifdef CONFIG_SMP |
538 | int hrtick_csd_pending; | |
539 | struct call_single_data hrtick_csd; | |
540 | #endif | |
8f4d37ec PZ |
541 | struct hrtimer hrtick_timer; |
542 | #endif | |
543 | ||
1da177e4 LT |
544 | #ifdef CONFIG_SCHEDSTATS |
545 | /* latency stats */ | |
546 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
547 | unsigned long long rq_cpu_time; |
548 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
549 | |
550 | /* sys_sched_yield() stats */ | |
480b9434 | 551 | unsigned int yld_count; |
1da177e4 LT |
552 | |
553 | /* schedule() stats */ | |
480b9434 KC |
554 | unsigned int sched_switch; |
555 | unsigned int sched_count; | |
556 | unsigned int sched_goidle; | |
1da177e4 LT |
557 | |
558 | /* try_to_wake_up() stats */ | |
480b9434 KC |
559 | unsigned int ttwu_count; |
560 | unsigned int ttwu_local; | |
1da177e4 | 561 | #endif |
317f3941 PZ |
562 | |
563 | #ifdef CONFIG_SMP | |
564 | struct task_struct *wake_list; | |
565 | #endif | |
1da177e4 LT |
566 | }; |
567 | ||
f34e3b61 | 568 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 569 | |
a64692a3 | 570 | |
1e5a7405 | 571 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 572 | |
0a2966b4 CL |
573 | static inline int cpu_of(struct rq *rq) |
574 | { | |
575 | #ifdef CONFIG_SMP | |
576 | return rq->cpu; | |
577 | #else | |
578 | return 0; | |
579 | #endif | |
580 | } | |
581 | ||
497f0ab3 | 582 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d | 583 | rcu_dereference_check((p), \ |
dce840a0 | 584 | rcu_read_lock_held() || \ |
d11c563d PM |
585 | lockdep_is_held(&sched_domains_mutex)) |
586 | ||
674311d5 NP |
587 | /* |
588 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 589 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
590 | * |
591 | * The domain tree of any CPU may only be accessed from within | |
592 | * preempt-disabled sections. | |
593 | */ | |
48f24c4d | 594 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 595 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
596 | |
597 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
598 | #define this_rq() (&__get_cpu_var(runqueues)) | |
599 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
600 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 601 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 602 | |
dc61b1d6 PZ |
603 | #ifdef CONFIG_CGROUP_SCHED |
604 | ||
605 | /* | |
606 | * Return the group to which this tasks belongs. | |
607 | * | |
608 | * We use task_subsys_state_check() and extend the RCU verification | |
0122ec5b | 609 | * with lockdep_is_held(&p->pi_lock) because cpu_cgroup_attach() |
dc61b1d6 PZ |
610 | * holds that lock for each task it moves into the cgroup. Therefore |
611 | * by holding that lock, we pin the task to the current cgroup. | |
612 | */ | |
613 | static inline struct task_group *task_group(struct task_struct *p) | |
614 | { | |
5091faa4 | 615 | struct task_group *tg; |
dc61b1d6 PZ |
616 | struct cgroup_subsys_state *css; |
617 | ||
618 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
0122ec5b | 619 | lockdep_is_held(&p->pi_lock)); |
5091faa4 MG |
620 | tg = container_of(css, struct task_group, css); |
621 | ||
622 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
623 | } |
624 | ||
625 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
626 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
627 | { | |
628 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
629 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
630 | p->se.parent = task_group(p)->se[cpu]; | |
631 | #endif | |
632 | ||
633 | #ifdef CONFIG_RT_GROUP_SCHED | |
634 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
635 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
636 | #endif | |
637 | } | |
638 | ||
639 | #else /* CONFIG_CGROUP_SCHED */ | |
640 | ||
641 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
642 | static inline struct task_group *task_group(struct task_struct *p) | |
643 | { | |
644 | return NULL; | |
645 | } | |
646 | ||
647 | #endif /* CONFIG_CGROUP_SCHED */ | |
648 | ||
fe44d621 | 649 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 650 | |
fe44d621 | 651 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 652 | { |
fe44d621 | 653 | s64 delta; |
305e6835 | 654 | |
61eadef6 | 655 | if (rq->skip_clock_update > 0) |
f26f9aff | 656 | return; |
aa483808 | 657 | |
fe44d621 PZ |
658 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
659 | rq->clock += delta; | |
660 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
661 | } |
662 | ||
bf5c91ba IM |
663 | /* |
664 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
665 | */ | |
666 | #ifdef CONFIG_SCHED_DEBUG | |
667 | # define const_debug __read_mostly | |
668 | #else | |
669 | # define const_debug static const | |
670 | #endif | |
671 | ||
017730c1 | 672 | /** |
1fd06bb1 | 673 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 674 | * @cpu: the processor in question. |
017730c1 | 675 | * |
017730c1 IM |
676 | * This interface allows printk to be called with the runqueue lock |
677 | * held and know whether or not it is OK to wake up the klogd. | |
678 | */ | |
89f19f04 | 679 | int runqueue_is_locked(int cpu) |
017730c1 | 680 | { |
05fa785c | 681 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
682 | } |
683 | ||
bf5c91ba IM |
684 | /* |
685 | * Debugging: various feature bits | |
686 | */ | |
f00b45c1 PZ |
687 | |
688 | #define SCHED_FEAT(name, enabled) \ | |
689 | __SCHED_FEAT_##name , | |
690 | ||
bf5c91ba | 691 | enum { |
f00b45c1 | 692 | #include "sched_features.h" |
bf5c91ba IM |
693 | }; |
694 | ||
f00b45c1 PZ |
695 | #undef SCHED_FEAT |
696 | ||
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | (1UL << __SCHED_FEAT_##name) * enabled | | |
699 | ||
bf5c91ba | 700 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
701 | #include "sched_features.h" |
702 | 0; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
706 | #ifdef CONFIG_SCHED_DEBUG | |
707 | #define SCHED_FEAT(name, enabled) \ | |
708 | #name , | |
709 | ||
983ed7a6 | 710 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
711 | #include "sched_features.h" |
712 | NULL | |
713 | }; | |
714 | ||
715 | #undef SCHED_FEAT | |
716 | ||
34f3a814 | 717 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 718 | { |
f00b45c1 PZ |
719 | int i; |
720 | ||
721 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
722 | if (!(sysctl_sched_features & (1UL << i))) |
723 | seq_puts(m, "NO_"); | |
724 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 725 | } |
34f3a814 | 726 | seq_puts(m, "\n"); |
f00b45c1 | 727 | |
34f3a814 | 728 | return 0; |
f00b45c1 PZ |
729 | } |
730 | ||
731 | static ssize_t | |
732 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
733 | size_t cnt, loff_t *ppos) | |
734 | { | |
735 | char buf[64]; | |
7740191c | 736 | char *cmp; |
f00b45c1 PZ |
737 | int neg = 0; |
738 | int i; | |
739 | ||
740 | if (cnt > 63) | |
741 | cnt = 63; | |
742 | ||
743 | if (copy_from_user(&buf, ubuf, cnt)) | |
744 | return -EFAULT; | |
745 | ||
746 | buf[cnt] = 0; | |
7740191c | 747 | cmp = strstrip(buf); |
f00b45c1 | 748 | |
524429c3 | 749 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
750 | neg = 1; |
751 | cmp += 3; | |
752 | } | |
753 | ||
754 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 755 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
756 | if (neg) |
757 | sysctl_sched_features &= ~(1UL << i); | |
758 | else | |
759 | sysctl_sched_features |= (1UL << i); | |
760 | break; | |
761 | } | |
762 | } | |
763 | ||
764 | if (!sched_feat_names[i]) | |
765 | return -EINVAL; | |
766 | ||
42994724 | 767 | *ppos += cnt; |
f00b45c1 PZ |
768 | |
769 | return cnt; | |
770 | } | |
771 | ||
34f3a814 LZ |
772 | static int sched_feat_open(struct inode *inode, struct file *filp) |
773 | { | |
774 | return single_open(filp, sched_feat_show, NULL); | |
775 | } | |
776 | ||
828c0950 | 777 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
778 | .open = sched_feat_open, |
779 | .write = sched_feat_write, | |
780 | .read = seq_read, | |
781 | .llseek = seq_lseek, | |
782 | .release = single_release, | |
f00b45c1 PZ |
783 | }; |
784 | ||
785 | static __init int sched_init_debug(void) | |
786 | { | |
f00b45c1 PZ |
787 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
788 | &sched_feat_fops); | |
789 | ||
790 | return 0; | |
791 | } | |
792 | late_initcall(sched_init_debug); | |
793 | ||
794 | #endif | |
795 | ||
796 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 797 | |
b82d9fdd PZ |
798 | /* |
799 | * Number of tasks to iterate in a single balance run. | |
800 | * Limited because this is done with IRQs disabled. | |
801 | */ | |
802 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
803 | ||
e9e9250b PZ |
804 | /* |
805 | * period over which we average the RT time consumption, measured | |
806 | * in ms. | |
807 | * | |
808 | * default: 1s | |
809 | */ | |
810 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
811 | ||
fa85ae24 | 812 | /* |
9f0c1e56 | 813 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
814 | * default: 1s |
815 | */ | |
9f0c1e56 | 816 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 817 | |
6892b75e IM |
818 | static __read_mostly int scheduler_running; |
819 | ||
9f0c1e56 PZ |
820 | /* |
821 | * part of the period that we allow rt tasks to run in us. | |
822 | * default: 0.95s | |
823 | */ | |
824 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 825 | |
d0b27fa7 PZ |
826 | static inline u64 global_rt_period(void) |
827 | { | |
828 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
829 | } | |
830 | ||
831 | static inline u64 global_rt_runtime(void) | |
832 | { | |
e26873bb | 833 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
834 | return RUNTIME_INF; |
835 | ||
836 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
837 | } | |
fa85ae24 | 838 | |
1da177e4 | 839 | #ifndef prepare_arch_switch |
4866cde0 NP |
840 | # define prepare_arch_switch(next) do { } while (0) |
841 | #endif | |
842 | #ifndef finish_arch_switch | |
843 | # define finish_arch_switch(prev) do { } while (0) | |
844 | #endif | |
845 | ||
051a1d1a DA |
846 | static inline int task_current(struct rq *rq, struct task_struct *p) |
847 | { | |
848 | return rq->curr == p; | |
849 | } | |
850 | ||
70b97a7f | 851 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 852 | { |
3ca7a440 PZ |
853 | #ifdef CONFIG_SMP |
854 | return p->on_cpu; | |
855 | #else | |
051a1d1a | 856 | return task_current(rq, p); |
3ca7a440 | 857 | #endif |
4866cde0 NP |
858 | } |
859 | ||
3ca7a440 | 860 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 862 | { |
3ca7a440 PZ |
863 | #ifdef CONFIG_SMP |
864 | /* | |
865 | * We can optimise this out completely for !SMP, because the | |
866 | * SMP rebalancing from interrupt is the only thing that cares | |
867 | * here. | |
868 | */ | |
869 | next->on_cpu = 1; | |
870 | #endif | |
4866cde0 NP |
871 | } |
872 | ||
70b97a7f | 873 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 874 | { |
3ca7a440 PZ |
875 | #ifdef CONFIG_SMP |
876 | /* | |
877 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
878 | * We must ensure this doesn't happen until the switch is completely | |
879 | * finished. | |
880 | */ | |
881 | smp_wmb(); | |
882 | prev->on_cpu = 0; | |
883 | #endif | |
da04c035 IM |
884 | #ifdef CONFIG_DEBUG_SPINLOCK |
885 | /* this is a valid case when another task releases the spinlock */ | |
886 | rq->lock.owner = current; | |
887 | #endif | |
8a25d5de IM |
888 | /* |
889 | * If we are tracking spinlock dependencies then we have to | |
890 | * fix up the runqueue lock - which gets 'carried over' from | |
891 | * prev into current: | |
892 | */ | |
893 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
894 | ||
05fa785c | 895 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
896 | } |
897 | ||
898 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 899 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * We can optimise this out completely for !SMP, because the | |
904 | * SMP rebalancing from interrupt is the only thing that cares | |
905 | * here. | |
906 | */ | |
3ca7a440 | 907 | next->on_cpu = 1; |
4866cde0 NP |
908 | #endif |
909 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 910 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 911 | #else |
05fa785c | 912 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
913 | #endif |
914 | } | |
915 | ||
70b97a7f | 916 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
917 | { |
918 | #ifdef CONFIG_SMP | |
919 | /* | |
3ca7a440 | 920 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
921 | * We must ensure this doesn't happen until the switch is completely |
922 | * finished. | |
923 | */ | |
924 | smp_wmb(); | |
3ca7a440 | 925 | prev->on_cpu = 0; |
4866cde0 NP |
926 | #endif |
927 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
928 | local_irq_enable(); | |
1da177e4 | 929 | #endif |
4866cde0 NP |
930 | } |
931 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 932 | |
0970d299 | 933 | /* |
0122ec5b | 934 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 935 | */ |
70b97a7f | 936 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
937 | __acquires(rq->lock) |
938 | { | |
0970d299 PZ |
939 | struct rq *rq; |
940 | ||
0122ec5b PZ |
941 | lockdep_assert_held(&p->pi_lock); |
942 | ||
3a5c359a | 943 | for (;;) { |
0970d299 | 944 | rq = task_rq(p); |
05fa785c | 945 | raw_spin_lock(&rq->lock); |
65cc8e48 | 946 | if (likely(rq == task_rq(p))) |
3a5c359a | 947 | return rq; |
05fa785c | 948 | raw_spin_unlock(&rq->lock); |
b29739f9 | 949 | } |
b29739f9 IM |
950 | } |
951 | ||
1da177e4 | 952 | /* |
0122ec5b | 953 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 954 | */ |
70b97a7f | 955 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 956 | __acquires(p->pi_lock) |
1da177e4 LT |
957 | __acquires(rq->lock) |
958 | { | |
70b97a7f | 959 | struct rq *rq; |
1da177e4 | 960 | |
3a5c359a | 961 | for (;;) { |
0122ec5b | 962 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 963 | rq = task_rq(p); |
05fa785c | 964 | raw_spin_lock(&rq->lock); |
65cc8e48 | 965 | if (likely(rq == task_rq(p))) |
3a5c359a | 966 | return rq; |
0122ec5b PZ |
967 | raw_spin_unlock(&rq->lock); |
968 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 969 | } |
1da177e4 LT |
970 | } |
971 | ||
a9957449 | 972 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
973 | __releases(rq->lock) |
974 | { | |
05fa785c | 975 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
976 | } |
977 | ||
0122ec5b PZ |
978 | static inline void |
979 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 980 | __releases(rq->lock) |
0122ec5b | 981 | __releases(p->pi_lock) |
1da177e4 | 982 | { |
0122ec5b PZ |
983 | raw_spin_unlock(&rq->lock); |
984 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
985 | } |
986 | ||
1da177e4 | 987 | /* |
cc2a73b5 | 988 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 989 | */ |
a9957449 | 990 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
991 | __acquires(rq->lock) |
992 | { | |
70b97a7f | 993 | struct rq *rq; |
1da177e4 LT |
994 | |
995 | local_irq_disable(); | |
996 | rq = this_rq(); | |
05fa785c | 997 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
998 | |
999 | return rq; | |
1000 | } | |
1001 | ||
8f4d37ec PZ |
1002 | #ifdef CONFIG_SCHED_HRTICK |
1003 | /* | |
1004 | * Use HR-timers to deliver accurate preemption points. | |
1005 | * | |
1006 | * Its all a bit involved since we cannot program an hrt while holding the | |
1007 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1008 | * reschedule event. | |
1009 | * | |
1010 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1011 | * rq->lock. | |
1012 | */ | |
8f4d37ec PZ |
1013 | |
1014 | /* | |
1015 | * Use hrtick when: | |
1016 | * - enabled by features | |
1017 | * - hrtimer is actually high res | |
1018 | */ | |
1019 | static inline int hrtick_enabled(struct rq *rq) | |
1020 | { | |
1021 | if (!sched_feat(HRTICK)) | |
1022 | return 0; | |
ba42059f | 1023 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1024 | return 0; |
8f4d37ec PZ |
1025 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1026 | } | |
1027 | ||
8f4d37ec PZ |
1028 | static void hrtick_clear(struct rq *rq) |
1029 | { | |
1030 | if (hrtimer_active(&rq->hrtick_timer)) | |
1031 | hrtimer_cancel(&rq->hrtick_timer); | |
1032 | } | |
1033 | ||
8f4d37ec PZ |
1034 | /* |
1035 | * High-resolution timer tick. | |
1036 | * Runs from hardirq context with interrupts disabled. | |
1037 | */ | |
1038 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1039 | { | |
1040 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1041 | ||
1042 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1043 | ||
05fa785c | 1044 | raw_spin_lock(&rq->lock); |
3e51f33f | 1045 | update_rq_clock(rq); |
8f4d37ec | 1046 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1047 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1048 | |
1049 | return HRTIMER_NORESTART; | |
1050 | } | |
1051 | ||
95e904c7 | 1052 | #ifdef CONFIG_SMP |
31656519 PZ |
1053 | /* |
1054 | * called from hardirq (IPI) context | |
1055 | */ | |
1056 | static void __hrtick_start(void *arg) | |
b328ca18 | 1057 | { |
31656519 | 1058 | struct rq *rq = arg; |
b328ca18 | 1059 | |
05fa785c | 1060 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1061 | hrtimer_restart(&rq->hrtick_timer); |
1062 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1063 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1064 | } |
1065 | ||
31656519 PZ |
1066 | /* |
1067 | * Called to set the hrtick timer state. | |
1068 | * | |
1069 | * called with rq->lock held and irqs disabled | |
1070 | */ | |
1071 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1072 | { |
31656519 PZ |
1073 | struct hrtimer *timer = &rq->hrtick_timer; |
1074 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1075 | |
cc584b21 | 1076 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1077 | |
1078 | if (rq == this_rq()) { | |
1079 | hrtimer_restart(timer); | |
1080 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1081 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1082 | rq->hrtick_csd_pending = 1; |
1083 | } | |
b328ca18 PZ |
1084 | } |
1085 | ||
1086 | static int | |
1087 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1088 | { | |
1089 | int cpu = (int)(long)hcpu; | |
1090 | ||
1091 | switch (action) { | |
1092 | case CPU_UP_CANCELED: | |
1093 | case CPU_UP_CANCELED_FROZEN: | |
1094 | case CPU_DOWN_PREPARE: | |
1095 | case CPU_DOWN_PREPARE_FROZEN: | |
1096 | case CPU_DEAD: | |
1097 | case CPU_DEAD_FROZEN: | |
31656519 | 1098 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1099 | return NOTIFY_OK; |
1100 | } | |
1101 | ||
1102 | return NOTIFY_DONE; | |
1103 | } | |
1104 | ||
fa748203 | 1105 | static __init void init_hrtick(void) |
b328ca18 PZ |
1106 | { |
1107 | hotcpu_notifier(hotplug_hrtick, 0); | |
1108 | } | |
31656519 PZ |
1109 | #else |
1110 | /* | |
1111 | * Called to set the hrtick timer state. | |
1112 | * | |
1113 | * called with rq->lock held and irqs disabled | |
1114 | */ | |
1115 | static void hrtick_start(struct rq *rq, u64 delay) | |
1116 | { | |
7f1e2ca9 | 1117 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1118 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1119 | } |
b328ca18 | 1120 | |
006c75f1 | 1121 | static inline void init_hrtick(void) |
8f4d37ec | 1122 | { |
8f4d37ec | 1123 | } |
31656519 | 1124 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1125 | |
31656519 | 1126 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1127 | { |
31656519 PZ |
1128 | #ifdef CONFIG_SMP |
1129 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1130 | |
31656519 PZ |
1131 | rq->hrtick_csd.flags = 0; |
1132 | rq->hrtick_csd.func = __hrtick_start; | |
1133 | rq->hrtick_csd.info = rq; | |
1134 | #endif | |
8f4d37ec | 1135 | |
31656519 PZ |
1136 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1137 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1138 | } |
006c75f1 | 1139 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1140 | static inline void hrtick_clear(struct rq *rq) |
1141 | { | |
1142 | } | |
1143 | ||
8f4d37ec PZ |
1144 | static inline void init_rq_hrtick(struct rq *rq) |
1145 | { | |
1146 | } | |
1147 | ||
b328ca18 PZ |
1148 | static inline void init_hrtick(void) |
1149 | { | |
1150 | } | |
006c75f1 | 1151 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1152 | |
c24d20db IM |
1153 | /* |
1154 | * resched_task - mark a task 'to be rescheduled now'. | |
1155 | * | |
1156 | * On UP this means the setting of the need_resched flag, on SMP it | |
1157 | * might also involve a cross-CPU call to trigger the scheduler on | |
1158 | * the target CPU. | |
1159 | */ | |
1160 | #ifdef CONFIG_SMP | |
1161 | ||
1162 | #ifndef tsk_is_polling | |
1163 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1164 | #endif | |
1165 | ||
31656519 | 1166 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1167 | { |
1168 | int cpu; | |
1169 | ||
05fa785c | 1170 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1171 | |
5ed0cec0 | 1172 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1173 | return; |
1174 | ||
5ed0cec0 | 1175 | set_tsk_need_resched(p); |
c24d20db IM |
1176 | |
1177 | cpu = task_cpu(p); | |
1178 | if (cpu == smp_processor_id()) | |
1179 | return; | |
1180 | ||
1181 | /* NEED_RESCHED must be visible before we test polling */ | |
1182 | smp_mb(); | |
1183 | if (!tsk_is_polling(p)) | |
1184 | smp_send_reschedule(cpu); | |
1185 | } | |
1186 | ||
1187 | static void resched_cpu(int cpu) | |
1188 | { | |
1189 | struct rq *rq = cpu_rq(cpu); | |
1190 | unsigned long flags; | |
1191 | ||
05fa785c | 1192 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1193 | return; |
1194 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1195 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1196 | } |
06d8308c TG |
1197 | |
1198 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1199 | /* |
1200 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1201 | * from an idle cpu. This is good for power-savings. | |
1202 | * | |
1203 | * We don't do similar optimization for completely idle system, as | |
1204 | * selecting an idle cpu will add more delays to the timers than intended | |
1205 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1206 | */ | |
1207 | int get_nohz_timer_target(void) | |
1208 | { | |
1209 | int cpu = smp_processor_id(); | |
1210 | int i; | |
1211 | struct sched_domain *sd; | |
1212 | ||
057f3fad | 1213 | rcu_read_lock(); |
83cd4fe2 | 1214 | for_each_domain(cpu, sd) { |
057f3fad PZ |
1215 | for_each_cpu(i, sched_domain_span(sd)) { |
1216 | if (!idle_cpu(i)) { | |
1217 | cpu = i; | |
1218 | goto unlock; | |
1219 | } | |
1220 | } | |
83cd4fe2 | 1221 | } |
057f3fad PZ |
1222 | unlock: |
1223 | rcu_read_unlock(); | |
83cd4fe2 VP |
1224 | return cpu; |
1225 | } | |
06d8308c TG |
1226 | /* |
1227 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1228 | * idle CPU then this timer might expire before the next timer event | |
1229 | * which is scheduled to wake up that CPU. In case of a completely | |
1230 | * idle system the next event might even be infinite time into the | |
1231 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1232 | * leaves the inner idle loop so the newly added timer is taken into | |
1233 | * account when the CPU goes back to idle and evaluates the timer | |
1234 | * wheel for the next timer event. | |
1235 | */ | |
1236 | void wake_up_idle_cpu(int cpu) | |
1237 | { | |
1238 | struct rq *rq = cpu_rq(cpu); | |
1239 | ||
1240 | if (cpu == smp_processor_id()) | |
1241 | return; | |
1242 | ||
1243 | /* | |
1244 | * This is safe, as this function is called with the timer | |
1245 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1246 | * to idle and has not yet set rq->curr to idle then it will | |
1247 | * be serialized on the timer wheel base lock and take the new | |
1248 | * timer into account automatically. | |
1249 | */ | |
1250 | if (rq->curr != rq->idle) | |
1251 | return; | |
1252 | ||
1253 | /* | |
1254 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1255 | * lockless. The worst case is that the other CPU runs the | |
1256 | * idle task through an additional NOOP schedule() | |
1257 | */ | |
5ed0cec0 | 1258 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1259 | |
1260 | /* NEED_RESCHED must be visible before we test polling */ | |
1261 | smp_mb(); | |
1262 | if (!tsk_is_polling(rq->idle)) | |
1263 | smp_send_reschedule(cpu); | |
1264 | } | |
39c0cbe2 | 1265 | |
6d6bc0ad | 1266 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1267 | |
e9e9250b PZ |
1268 | static u64 sched_avg_period(void) |
1269 | { | |
1270 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1271 | } | |
1272 | ||
1273 | static void sched_avg_update(struct rq *rq) | |
1274 | { | |
1275 | s64 period = sched_avg_period(); | |
1276 | ||
1277 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1278 | /* |
1279 | * Inline assembly required to prevent the compiler | |
1280 | * optimising this loop into a divmod call. | |
1281 | * See __iter_div_u64_rem() for another example of this. | |
1282 | */ | |
1283 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1284 | rq->age_stamp += period; |
1285 | rq->rt_avg /= 2; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1290 | { | |
1291 | rq->rt_avg += rt_delta; | |
1292 | sched_avg_update(rq); | |
1293 | } | |
1294 | ||
6d6bc0ad | 1295 | #else /* !CONFIG_SMP */ |
31656519 | 1296 | static void resched_task(struct task_struct *p) |
c24d20db | 1297 | { |
05fa785c | 1298 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1299 | set_tsk_need_resched(p); |
c24d20db | 1300 | } |
e9e9250b PZ |
1301 | |
1302 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1303 | { | |
1304 | } | |
da2b71ed SS |
1305 | |
1306 | static void sched_avg_update(struct rq *rq) | |
1307 | { | |
1308 | } | |
6d6bc0ad | 1309 | #endif /* CONFIG_SMP */ |
c24d20db | 1310 | |
45bf76df IM |
1311 | #if BITS_PER_LONG == 32 |
1312 | # define WMULT_CONST (~0UL) | |
1313 | #else | |
1314 | # define WMULT_CONST (1UL << 32) | |
1315 | #endif | |
1316 | ||
1317 | #define WMULT_SHIFT 32 | |
1318 | ||
194081eb IM |
1319 | /* |
1320 | * Shift right and round: | |
1321 | */ | |
cf2ab469 | 1322 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1323 | |
a7be37ac PZ |
1324 | /* |
1325 | * delta *= weight / lw | |
1326 | */ | |
cb1c4fc9 | 1327 | static unsigned long |
45bf76df IM |
1328 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1329 | struct load_weight *lw) | |
1330 | { | |
1331 | u64 tmp; | |
1332 | ||
db670dac SB |
1333 | tmp = (u64)delta_exec * weight; |
1334 | ||
7a232e03 LJ |
1335 | if (!lw->inv_weight) { |
1336 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1337 | lw->inv_weight = 1; | |
1338 | else | |
db670dac | 1339 | lw->inv_weight = WMULT_CONST / lw->weight; |
7a232e03 | 1340 | } |
45bf76df | 1341 | |
45bf76df IM |
1342 | /* |
1343 | * Check whether we'd overflow the 64-bit multiplication: | |
1344 | */ | |
194081eb | 1345 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1346 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1347 | WMULT_SHIFT/2); |
1348 | else | |
cf2ab469 | 1349 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1350 | |
ecf691da | 1351 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1352 | } |
1353 | ||
1091985b | 1354 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1355 | { |
1356 | lw->weight += inc; | |
e89996ae | 1357 | lw->inv_weight = 0; |
45bf76df IM |
1358 | } |
1359 | ||
1091985b | 1360 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1361 | { |
1362 | lw->weight -= dec; | |
e89996ae | 1363 | lw->inv_weight = 0; |
45bf76df IM |
1364 | } |
1365 | ||
2069dd75 PZ |
1366 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1367 | { | |
1368 | lw->weight = w; | |
1369 | lw->inv_weight = 0; | |
1370 | } | |
1371 | ||
2dd73a4f PW |
1372 | /* |
1373 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1374 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1375 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1376 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1377 | * scaled version of the new time slice allocation that they receive on time |
1378 | * slice expiry etc. | |
1379 | */ | |
1380 | ||
cce7ade8 PZ |
1381 | #define WEIGHT_IDLEPRIO 3 |
1382 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1383 | |
1384 | /* | |
1385 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1386 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1387 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1388 | * that remained on nice 0. | |
1389 | * | |
1390 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1391 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1392 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1393 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1394 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1395 | */ |
1396 | static const int prio_to_weight[40] = { | |
254753dc IM |
1397 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1398 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1399 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1400 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1401 | /* 0 */ 1024, 820, 655, 526, 423, | |
1402 | /* 5 */ 335, 272, 215, 172, 137, | |
1403 | /* 10 */ 110, 87, 70, 56, 45, | |
1404 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1405 | }; |
1406 | ||
5714d2de IM |
1407 | /* |
1408 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1409 | * | |
1410 | * In cases where the weight does not change often, we can use the | |
1411 | * precalculated inverse to speed up arithmetics by turning divisions | |
1412 | * into multiplications: | |
1413 | */ | |
dd41f596 | 1414 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1415 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1416 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1417 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1418 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1419 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1420 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1421 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1422 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1423 | }; |
2dd73a4f | 1424 | |
ef12fefa BR |
1425 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1426 | enum cpuacct_stat_index { | |
1427 | CPUACCT_STAT_USER, /* ... user mode */ | |
1428 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1429 | ||
1430 | CPUACCT_STAT_NSTATS, | |
1431 | }; | |
1432 | ||
d842de87 SV |
1433 | #ifdef CONFIG_CGROUP_CPUACCT |
1434 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1435 | static void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1437 | #else |
1438 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1439 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1440 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1441 | #endif |
1442 | ||
18d95a28 PZ |
1443 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1444 | { | |
1445 | update_load_add(&rq->load, load); | |
1446 | } | |
1447 | ||
1448 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1449 | { | |
1450 | update_load_sub(&rq->load, load); | |
1451 | } | |
1452 | ||
7940ca36 | 1453 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1454 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1455 | |
1456 | /* | |
1457 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1458 | * leaving it for the final time. | |
1459 | */ | |
eb755805 | 1460 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1461 | { |
1462 | struct task_group *parent, *child; | |
eb755805 | 1463 | int ret; |
c09595f6 PZ |
1464 | |
1465 | rcu_read_lock(); | |
1466 | parent = &root_task_group; | |
1467 | down: | |
eb755805 PZ |
1468 | ret = (*down)(parent, data); |
1469 | if (ret) | |
1470 | goto out_unlock; | |
c09595f6 PZ |
1471 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1472 | parent = child; | |
1473 | goto down; | |
1474 | ||
1475 | up: | |
1476 | continue; | |
1477 | } | |
eb755805 PZ |
1478 | ret = (*up)(parent, data); |
1479 | if (ret) | |
1480 | goto out_unlock; | |
c09595f6 PZ |
1481 | |
1482 | child = parent; | |
1483 | parent = parent->parent; | |
1484 | if (parent) | |
1485 | goto up; | |
eb755805 | 1486 | out_unlock: |
c09595f6 | 1487 | rcu_read_unlock(); |
eb755805 PZ |
1488 | |
1489 | return ret; | |
c09595f6 PZ |
1490 | } |
1491 | ||
eb755805 PZ |
1492 | static int tg_nop(struct task_group *tg, void *data) |
1493 | { | |
1494 | return 0; | |
c09595f6 | 1495 | } |
eb755805 PZ |
1496 | #endif |
1497 | ||
1498 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1499 | /* Used instead of source_load when we know the type == 0 */ |
1500 | static unsigned long weighted_cpuload(const int cpu) | |
1501 | { | |
1502 | return cpu_rq(cpu)->load.weight; | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Return a low guess at the load of a migration-source cpu weighted | |
1507 | * according to the scheduling class and "nice" value. | |
1508 | * | |
1509 | * We want to under-estimate the load of migration sources, to | |
1510 | * balance conservatively. | |
1511 | */ | |
1512 | static unsigned long source_load(int cpu, int type) | |
1513 | { | |
1514 | struct rq *rq = cpu_rq(cpu); | |
1515 | unsigned long total = weighted_cpuload(cpu); | |
1516 | ||
1517 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1518 | return total; | |
1519 | ||
1520 | return min(rq->cpu_load[type-1], total); | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Return a high guess at the load of a migration-target cpu weighted | |
1525 | * according to the scheduling class and "nice" value. | |
1526 | */ | |
1527 | static unsigned long target_load(int cpu, int type) | |
1528 | { | |
1529 | struct rq *rq = cpu_rq(cpu); | |
1530 | unsigned long total = weighted_cpuload(cpu); | |
1531 | ||
1532 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1533 | return total; | |
1534 | ||
1535 | return max(rq->cpu_load[type-1], total); | |
1536 | } | |
1537 | ||
ae154be1 PZ |
1538 | static unsigned long power_of(int cpu) |
1539 | { | |
e51fd5e2 | 1540 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1541 | } |
1542 | ||
eb755805 PZ |
1543 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1544 | ||
1545 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1546 | { | |
1547 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1548 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1549 | |
4cd42620 SR |
1550 | if (nr_running) |
1551 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1552 | else |
1553 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1554 | |
1555 | return rq->avg_load_per_task; | |
1556 | } | |
1557 | ||
1558 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1559 | |
c09595f6 | 1560 | /* |
c8cba857 PZ |
1561 | * Compute the cpu's hierarchical load factor for each task group. |
1562 | * This needs to be done in a top-down fashion because the load of a child | |
1563 | * group is a fraction of its parents load. | |
c09595f6 | 1564 | */ |
eb755805 | 1565 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1566 | { |
c8cba857 | 1567 | unsigned long load; |
eb755805 | 1568 | long cpu = (long)data; |
c09595f6 | 1569 | |
c8cba857 PZ |
1570 | if (!tg->parent) { |
1571 | load = cpu_rq(cpu)->load.weight; | |
1572 | } else { | |
1573 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1574 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1575 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1576 | } | |
c09595f6 | 1577 | |
c8cba857 | 1578 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1579 | |
eb755805 | 1580 | return 0; |
c09595f6 PZ |
1581 | } |
1582 | ||
eb755805 | 1583 | static void update_h_load(long cpu) |
c09595f6 | 1584 | { |
eb755805 | 1585 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1586 | } |
1587 | ||
18d95a28 PZ |
1588 | #endif |
1589 | ||
8f45e2b5 GH |
1590 | #ifdef CONFIG_PREEMPT |
1591 | ||
b78bb868 PZ |
1592 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1593 | ||
70574a99 | 1594 | /* |
8f45e2b5 GH |
1595 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1596 | * way at the expense of forcing extra atomic operations in all | |
1597 | * invocations. This assures that the double_lock is acquired using the | |
1598 | * same underlying policy as the spinlock_t on this architecture, which | |
1599 | * reduces latency compared to the unfair variant below. However, it | |
1600 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1601 | */ |
8f45e2b5 GH |
1602 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1603 | __releases(this_rq->lock) | |
1604 | __acquires(busiest->lock) | |
1605 | __acquires(this_rq->lock) | |
1606 | { | |
05fa785c | 1607 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1608 | double_rq_lock(this_rq, busiest); |
1609 | ||
1610 | return 1; | |
1611 | } | |
1612 | ||
1613 | #else | |
1614 | /* | |
1615 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1616 | * latency by eliminating extra atomic operations when the locks are | |
1617 | * already in proper order on entry. This favors lower cpu-ids and will | |
1618 | * grant the double lock to lower cpus over higher ids under contention, | |
1619 | * regardless of entry order into the function. | |
1620 | */ | |
1621 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1622 | __releases(this_rq->lock) |
1623 | __acquires(busiest->lock) | |
1624 | __acquires(this_rq->lock) | |
1625 | { | |
1626 | int ret = 0; | |
1627 | ||
05fa785c | 1628 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1629 | if (busiest < this_rq) { |
05fa785c TG |
1630 | raw_spin_unlock(&this_rq->lock); |
1631 | raw_spin_lock(&busiest->lock); | |
1632 | raw_spin_lock_nested(&this_rq->lock, | |
1633 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1634 | ret = 1; |
1635 | } else | |
05fa785c TG |
1636 | raw_spin_lock_nested(&busiest->lock, |
1637 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1638 | } |
1639 | return ret; | |
1640 | } | |
1641 | ||
8f45e2b5 GH |
1642 | #endif /* CONFIG_PREEMPT */ |
1643 | ||
1644 | /* | |
1645 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1646 | */ | |
1647 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1648 | { | |
1649 | if (unlikely(!irqs_disabled())) { | |
1650 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1651 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1652 | BUG_ON(1); |
1653 | } | |
1654 | ||
1655 | return _double_lock_balance(this_rq, busiest); | |
1656 | } | |
1657 | ||
70574a99 AD |
1658 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1659 | __releases(busiest->lock) | |
1660 | { | |
05fa785c | 1661 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1662 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1663 | } | |
1e3c88bd PZ |
1664 | |
1665 | /* | |
1666 | * double_rq_lock - safely lock two runqueues | |
1667 | * | |
1668 | * Note this does not disable interrupts like task_rq_lock, | |
1669 | * you need to do so manually before calling. | |
1670 | */ | |
1671 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1672 | __acquires(rq1->lock) | |
1673 | __acquires(rq2->lock) | |
1674 | { | |
1675 | BUG_ON(!irqs_disabled()); | |
1676 | if (rq1 == rq2) { | |
1677 | raw_spin_lock(&rq1->lock); | |
1678 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1679 | } else { | |
1680 | if (rq1 < rq2) { | |
1681 | raw_spin_lock(&rq1->lock); | |
1682 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1683 | } else { | |
1684 | raw_spin_lock(&rq2->lock); | |
1685 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1686 | } | |
1687 | } | |
1e3c88bd PZ |
1688 | } |
1689 | ||
1690 | /* | |
1691 | * double_rq_unlock - safely unlock two runqueues | |
1692 | * | |
1693 | * Note this does not restore interrupts like task_rq_unlock, | |
1694 | * you need to do so manually after calling. | |
1695 | */ | |
1696 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1697 | __releases(rq1->lock) | |
1698 | __releases(rq2->lock) | |
1699 | { | |
1700 | raw_spin_unlock(&rq1->lock); | |
1701 | if (rq1 != rq2) | |
1702 | raw_spin_unlock(&rq2->lock); | |
1703 | else | |
1704 | __release(rq2->lock); | |
1705 | } | |
1706 | ||
d95f4122 MG |
1707 | #else /* CONFIG_SMP */ |
1708 | ||
1709 | /* | |
1710 | * double_rq_lock - safely lock two runqueues | |
1711 | * | |
1712 | * Note this does not disable interrupts like task_rq_lock, | |
1713 | * you need to do so manually before calling. | |
1714 | */ | |
1715 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1716 | __acquires(rq1->lock) | |
1717 | __acquires(rq2->lock) | |
1718 | { | |
1719 | BUG_ON(!irqs_disabled()); | |
1720 | BUG_ON(rq1 != rq2); | |
1721 | raw_spin_lock(&rq1->lock); | |
1722 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * double_rq_unlock - safely unlock two runqueues | |
1727 | * | |
1728 | * Note this does not restore interrupts like task_rq_unlock, | |
1729 | * you need to do so manually after calling. | |
1730 | */ | |
1731 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1732 | __releases(rq1->lock) | |
1733 | __releases(rq2->lock) | |
1734 | { | |
1735 | BUG_ON(rq1 != rq2); | |
1736 | raw_spin_unlock(&rq1->lock); | |
1737 | __release(rq2->lock); | |
1738 | } | |
1739 | ||
18d95a28 PZ |
1740 | #endif |
1741 | ||
74f5187a | 1742 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1743 | static void update_sysctl(void); |
acb4a848 | 1744 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1745 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1746 | |
cd29fe6f PZ |
1747 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1748 | { | |
1749 | set_task_rq(p, cpu); | |
1750 | #ifdef CONFIG_SMP | |
1751 | /* | |
1752 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1753 | * successfuly executed on another CPU. We must ensure that updates of | |
1754 | * per-task data have been completed by this moment. | |
1755 | */ | |
1756 | smp_wmb(); | |
1757 | task_thread_info(p)->cpu = cpu; | |
1758 | #endif | |
1759 | } | |
dce48a84 | 1760 | |
1e3c88bd | 1761 | static const struct sched_class rt_sched_class; |
dd41f596 | 1762 | |
34f971f6 | 1763 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1764 | #define for_each_class(class) \ |
1765 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1766 | |
1e3c88bd PZ |
1767 | #include "sched_stats.h" |
1768 | ||
c09595f6 | 1769 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1770 | { |
1771 | rq->nr_running++; | |
9c217245 IM |
1772 | } |
1773 | ||
c09595f6 | 1774 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1775 | { |
1776 | rq->nr_running--; | |
9c217245 IM |
1777 | } |
1778 | ||
45bf76df IM |
1779 | static void set_load_weight(struct task_struct *p) |
1780 | { | |
dd41f596 IM |
1781 | /* |
1782 | * SCHED_IDLE tasks get minimal weight: | |
1783 | */ | |
1784 | if (p->policy == SCHED_IDLE) { | |
1785 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1786 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1787 | return; | |
1788 | } | |
71f8bd46 | 1789 | |
dd41f596 IM |
1790 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1791 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1792 | } |
1793 | ||
371fd7e7 | 1794 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1795 | { |
a64692a3 | 1796 | update_rq_clock(rq); |
dd41f596 | 1797 | sched_info_queued(p); |
371fd7e7 | 1798 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1799 | } |
1800 | ||
371fd7e7 | 1801 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1802 | { |
a64692a3 | 1803 | update_rq_clock(rq); |
46ac22ba | 1804 | sched_info_dequeued(p); |
371fd7e7 | 1805 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1806 | } |
1807 | ||
1e3c88bd PZ |
1808 | /* |
1809 | * activate_task - move a task to the runqueue. | |
1810 | */ | |
371fd7e7 | 1811 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1812 | { |
1813 | if (task_contributes_to_load(p)) | |
1814 | rq->nr_uninterruptible--; | |
1815 | ||
371fd7e7 | 1816 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1817 | inc_nr_running(rq); |
1818 | } | |
1819 | ||
1820 | /* | |
1821 | * deactivate_task - remove a task from the runqueue. | |
1822 | */ | |
371fd7e7 | 1823 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1824 | { |
1825 | if (task_contributes_to_load(p)) | |
1826 | rq->nr_uninterruptible++; | |
1827 | ||
371fd7e7 | 1828 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1829 | dec_nr_running(rq); |
1830 | } | |
1831 | ||
b52bfee4 VP |
1832 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1833 | ||
305e6835 VP |
1834 | /* |
1835 | * There are no locks covering percpu hardirq/softirq time. | |
1836 | * They are only modified in account_system_vtime, on corresponding CPU | |
1837 | * with interrupts disabled. So, writes are safe. | |
1838 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1839 | * This may result in other CPU reading this CPU's irq time and can | |
1840 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1841 | * or new value with a side effect of accounting a slice of irq time to wrong |
1842 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1843 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1844 | */ |
b52bfee4 VP |
1845 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1846 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1847 | ||
1848 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1849 | static int sched_clock_irqtime; | |
1850 | ||
1851 | void enable_sched_clock_irqtime(void) | |
1852 | { | |
1853 | sched_clock_irqtime = 1; | |
1854 | } | |
1855 | ||
1856 | void disable_sched_clock_irqtime(void) | |
1857 | { | |
1858 | sched_clock_irqtime = 0; | |
1859 | } | |
1860 | ||
8e92c201 PZ |
1861 | #ifndef CONFIG_64BIT |
1862 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1863 | ||
1864 | static inline void irq_time_write_begin(void) | |
1865 | { | |
1866 | __this_cpu_inc(irq_time_seq.sequence); | |
1867 | smp_wmb(); | |
1868 | } | |
1869 | ||
1870 | static inline void irq_time_write_end(void) | |
1871 | { | |
1872 | smp_wmb(); | |
1873 | __this_cpu_inc(irq_time_seq.sequence); | |
1874 | } | |
1875 | ||
1876 | static inline u64 irq_time_read(int cpu) | |
1877 | { | |
1878 | u64 irq_time; | |
1879 | unsigned seq; | |
1880 | ||
1881 | do { | |
1882 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1883 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1884 | per_cpu(cpu_hardirq_time, cpu); | |
1885 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1886 | ||
1887 | return irq_time; | |
1888 | } | |
1889 | #else /* CONFIG_64BIT */ | |
1890 | static inline void irq_time_write_begin(void) | |
1891 | { | |
1892 | } | |
1893 | ||
1894 | static inline void irq_time_write_end(void) | |
1895 | { | |
1896 | } | |
1897 | ||
1898 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1899 | { |
305e6835 VP |
1900 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1901 | } | |
8e92c201 | 1902 | #endif /* CONFIG_64BIT */ |
305e6835 | 1903 | |
fe44d621 PZ |
1904 | /* |
1905 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1906 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1907 | */ | |
b52bfee4 VP |
1908 | void account_system_vtime(struct task_struct *curr) |
1909 | { | |
1910 | unsigned long flags; | |
fe44d621 | 1911 | s64 delta; |
b52bfee4 | 1912 | int cpu; |
b52bfee4 VP |
1913 | |
1914 | if (!sched_clock_irqtime) | |
1915 | return; | |
1916 | ||
1917 | local_irq_save(flags); | |
1918 | ||
b52bfee4 | 1919 | cpu = smp_processor_id(); |
fe44d621 PZ |
1920 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1921 | __this_cpu_add(irq_start_time, delta); | |
1922 | ||
8e92c201 | 1923 | irq_time_write_begin(); |
b52bfee4 VP |
1924 | /* |
1925 | * We do not account for softirq time from ksoftirqd here. | |
1926 | * We want to continue accounting softirq time to ksoftirqd thread | |
1927 | * in that case, so as not to confuse scheduler with a special task | |
1928 | * that do not consume any time, but still wants to run. | |
1929 | */ | |
1930 | if (hardirq_count()) | |
fe44d621 | 1931 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1932 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1933 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1934 | |
8e92c201 | 1935 | irq_time_write_end(); |
b52bfee4 VP |
1936 | local_irq_restore(flags); |
1937 | } | |
b7dadc38 | 1938 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1939 | |
fe44d621 | 1940 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1941 | { |
fe44d621 PZ |
1942 | s64 irq_delta; |
1943 | ||
8e92c201 | 1944 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1945 | |
1946 | /* | |
1947 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1948 | * this case when a previous update_rq_clock() happened inside a | |
1949 | * {soft,}irq region. | |
1950 | * | |
1951 | * When this happens, we stop ->clock_task and only update the | |
1952 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1953 | * update will consume the rest. This ensures ->clock_task is | |
1954 | * monotonic. | |
1955 | * | |
1956 | * It does however cause some slight miss-attribution of {soft,}irq | |
1957 | * time, a more accurate solution would be to update the irq_time using | |
1958 | * the current rq->clock timestamp, except that would require using | |
1959 | * atomic ops. | |
1960 | */ | |
1961 | if (irq_delta > delta) | |
1962 | irq_delta = delta; | |
1963 | ||
1964 | rq->prev_irq_time += irq_delta; | |
1965 | delta -= irq_delta; | |
1966 | rq->clock_task += delta; | |
1967 | ||
1968 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1969 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1970 | } |
1971 | ||
abb74cef VP |
1972 | static int irqtime_account_hi_update(void) |
1973 | { | |
1974 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1975 | unsigned long flags; | |
1976 | u64 latest_ns; | |
1977 | int ret = 0; | |
1978 | ||
1979 | local_irq_save(flags); | |
1980 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1981 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1982 | ret = 1; | |
1983 | local_irq_restore(flags); | |
1984 | return ret; | |
1985 | } | |
1986 | ||
1987 | static int irqtime_account_si_update(void) | |
1988 | { | |
1989 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1990 | unsigned long flags; | |
1991 | u64 latest_ns; | |
1992 | int ret = 0; | |
1993 | ||
1994 | local_irq_save(flags); | |
1995 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1996 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1997 | ret = 1; | |
1998 | local_irq_restore(flags); | |
1999 | return ret; | |
2000 | } | |
2001 | ||
fe44d621 | 2002 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 2003 | |
abb74cef VP |
2004 | #define sched_clock_irqtime (0) |
2005 | ||
fe44d621 | 2006 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 2007 | { |
fe44d621 | 2008 | rq->clock_task += delta; |
305e6835 VP |
2009 | } |
2010 | ||
fe44d621 | 2011 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 2012 | |
1e3c88bd PZ |
2013 | #include "sched_idletask.c" |
2014 | #include "sched_fair.c" | |
2015 | #include "sched_rt.c" | |
5091faa4 | 2016 | #include "sched_autogroup.c" |
34f971f6 | 2017 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2018 | #ifdef CONFIG_SCHED_DEBUG |
2019 | # include "sched_debug.c" | |
2020 | #endif | |
2021 | ||
34f971f6 PZ |
2022 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2023 | { | |
2024 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2025 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2026 | ||
2027 | if (stop) { | |
2028 | /* | |
2029 | * Make it appear like a SCHED_FIFO task, its something | |
2030 | * userspace knows about and won't get confused about. | |
2031 | * | |
2032 | * Also, it will make PI more or less work without too | |
2033 | * much confusion -- but then, stop work should not | |
2034 | * rely on PI working anyway. | |
2035 | */ | |
2036 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2037 | ||
2038 | stop->sched_class = &stop_sched_class; | |
2039 | } | |
2040 | ||
2041 | cpu_rq(cpu)->stop = stop; | |
2042 | ||
2043 | if (old_stop) { | |
2044 | /* | |
2045 | * Reset it back to a normal scheduling class so that | |
2046 | * it can die in pieces. | |
2047 | */ | |
2048 | old_stop->sched_class = &rt_sched_class; | |
2049 | } | |
2050 | } | |
2051 | ||
14531189 | 2052 | /* |
dd41f596 | 2053 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2054 | */ |
14531189 IM |
2055 | static inline int __normal_prio(struct task_struct *p) |
2056 | { | |
dd41f596 | 2057 | return p->static_prio; |
14531189 IM |
2058 | } |
2059 | ||
b29739f9 IM |
2060 | /* |
2061 | * Calculate the expected normal priority: i.e. priority | |
2062 | * without taking RT-inheritance into account. Might be | |
2063 | * boosted by interactivity modifiers. Changes upon fork, | |
2064 | * setprio syscalls, and whenever the interactivity | |
2065 | * estimator recalculates. | |
2066 | */ | |
36c8b586 | 2067 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2068 | { |
2069 | int prio; | |
2070 | ||
e05606d3 | 2071 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2072 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2073 | else | |
2074 | prio = __normal_prio(p); | |
2075 | return prio; | |
2076 | } | |
2077 | ||
2078 | /* | |
2079 | * Calculate the current priority, i.e. the priority | |
2080 | * taken into account by the scheduler. This value might | |
2081 | * be boosted by RT tasks, or might be boosted by | |
2082 | * interactivity modifiers. Will be RT if the task got | |
2083 | * RT-boosted. If not then it returns p->normal_prio. | |
2084 | */ | |
36c8b586 | 2085 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2086 | { |
2087 | p->normal_prio = normal_prio(p); | |
2088 | /* | |
2089 | * If we are RT tasks or we were boosted to RT priority, | |
2090 | * keep the priority unchanged. Otherwise, update priority | |
2091 | * to the normal priority: | |
2092 | */ | |
2093 | if (!rt_prio(p->prio)) | |
2094 | return p->normal_prio; | |
2095 | return p->prio; | |
2096 | } | |
2097 | ||
1da177e4 LT |
2098 | /** |
2099 | * task_curr - is this task currently executing on a CPU? | |
2100 | * @p: the task in question. | |
2101 | */ | |
36c8b586 | 2102 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2103 | { |
2104 | return cpu_curr(task_cpu(p)) == p; | |
2105 | } | |
2106 | ||
cb469845 SR |
2107 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2108 | const struct sched_class *prev_class, | |
da7a735e | 2109 | int oldprio) |
cb469845 SR |
2110 | { |
2111 | if (prev_class != p->sched_class) { | |
2112 | if (prev_class->switched_from) | |
da7a735e PZ |
2113 | prev_class->switched_from(rq, p); |
2114 | p->sched_class->switched_to(rq, p); | |
2115 | } else if (oldprio != p->prio) | |
2116 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2117 | } |
2118 | ||
1e5a7405 PZ |
2119 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2120 | { | |
2121 | const struct sched_class *class; | |
2122 | ||
2123 | if (p->sched_class == rq->curr->sched_class) { | |
2124 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2125 | } else { | |
2126 | for_each_class(class) { | |
2127 | if (class == rq->curr->sched_class) | |
2128 | break; | |
2129 | if (class == p->sched_class) { | |
2130 | resched_task(rq->curr); | |
2131 | break; | |
2132 | } | |
2133 | } | |
2134 | } | |
2135 | ||
2136 | /* | |
2137 | * A queue event has occurred, and we're going to schedule. In | |
2138 | * this case, we can save a useless back to back clock update. | |
2139 | */ | |
fd2f4419 | 2140 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2141 | rq->skip_clock_update = 1; |
2142 | } | |
2143 | ||
1da177e4 | 2144 | #ifdef CONFIG_SMP |
cc367732 IM |
2145 | /* |
2146 | * Is this task likely cache-hot: | |
2147 | */ | |
e7693a36 | 2148 | static int |
cc367732 IM |
2149 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2150 | { | |
2151 | s64 delta; | |
2152 | ||
e6c8fba7 PZ |
2153 | if (p->sched_class != &fair_sched_class) |
2154 | return 0; | |
2155 | ||
ef8002f6 NR |
2156 | if (unlikely(p->policy == SCHED_IDLE)) |
2157 | return 0; | |
2158 | ||
f540a608 IM |
2159 | /* |
2160 | * Buddy candidates are cache hot: | |
2161 | */ | |
f685ceac | 2162 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2163 | (&p->se == cfs_rq_of(&p->se)->next || |
2164 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2165 | return 1; |
2166 | ||
6bc1665b IM |
2167 | if (sysctl_sched_migration_cost == -1) |
2168 | return 1; | |
2169 | if (sysctl_sched_migration_cost == 0) | |
2170 | return 0; | |
2171 | ||
cc367732 IM |
2172 | delta = now - p->se.exec_start; |
2173 | ||
2174 | return delta < (s64)sysctl_sched_migration_cost; | |
2175 | } | |
2176 | ||
dd41f596 | 2177 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2178 | { |
e2912009 PZ |
2179 | #ifdef CONFIG_SCHED_DEBUG |
2180 | /* | |
2181 | * We should never call set_task_cpu() on a blocked task, | |
2182 | * ttwu() will sort out the placement. | |
2183 | */ | |
077614ee PZ |
2184 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2185 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
2186 | |
2187 | #ifdef CONFIG_LOCKDEP | |
2188 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || | |
2189 | lockdep_is_held(&task_rq(p)->lock))); | |
2190 | #endif | |
e2912009 PZ |
2191 | #endif |
2192 | ||
de1d7286 | 2193 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2194 | |
0c69774e PZ |
2195 | if (task_cpu(p) != new_cpu) { |
2196 | p->se.nr_migrations++; | |
2197 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2198 | } | |
dd41f596 IM |
2199 | |
2200 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2201 | } |
2202 | ||
969c7921 | 2203 | struct migration_arg { |
36c8b586 | 2204 | struct task_struct *task; |
1da177e4 | 2205 | int dest_cpu; |
70b97a7f | 2206 | }; |
1da177e4 | 2207 | |
969c7921 TH |
2208 | static int migration_cpu_stop(void *data); |
2209 | ||
1da177e4 LT |
2210 | /* |
2211 | * wait_task_inactive - wait for a thread to unschedule. | |
2212 | * | |
85ba2d86 RM |
2213 | * If @match_state is nonzero, it's the @p->state value just checked and |
2214 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2215 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2216 | * we return a positive number (its total switch count). If a second call | |
2217 | * a short while later returns the same number, the caller can be sure that | |
2218 | * @p has remained unscheduled the whole time. | |
2219 | * | |
1da177e4 LT |
2220 | * The caller must ensure that the task *will* unschedule sometime soon, |
2221 | * else this function might spin for a *long* time. This function can't | |
2222 | * be called with interrupts off, or it may introduce deadlock with | |
2223 | * smp_call_function() if an IPI is sent by the same process we are | |
2224 | * waiting to become inactive. | |
2225 | */ | |
85ba2d86 | 2226 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2227 | { |
2228 | unsigned long flags; | |
dd41f596 | 2229 | int running, on_rq; |
85ba2d86 | 2230 | unsigned long ncsw; |
70b97a7f | 2231 | struct rq *rq; |
1da177e4 | 2232 | |
3a5c359a AK |
2233 | for (;;) { |
2234 | /* | |
2235 | * We do the initial early heuristics without holding | |
2236 | * any task-queue locks at all. We'll only try to get | |
2237 | * the runqueue lock when things look like they will | |
2238 | * work out! | |
2239 | */ | |
2240 | rq = task_rq(p); | |
fa490cfd | 2241 | |
3a5c359a AK |
2242 | /* |
2243 | * If the task is actively running on another CPU | |
2244 | * still, just relax and busy-wait without holding | |
2245 | * any locks. | |
2246 | * | |
2247 | * NOTE! Since we don't hold any locks, it's not | |
2248 | * even sure that "rq" stays as the right runqueue! | |
2249 | * But we don't care, since "task_running()" will | |
2250 | * return false if the runqueue has changed and p | |
2251 | * is actually now running somewhere else! | |
2252 | */ | |
85ba2d86 RM |
2253 | while (task_running(rq, p)) { |
2254 | if (match_state && unlikely(p->state != match_state)) | |
2255 | return 0; | |
3a5c359a | 2256 | cpu_relax(); |
85ba2d86 | 2257 | } |
fa490cfd | 2258 | |
3a5c359a AK |
2259 | /* |
2260 | * Ok, time to look more closely! We need the rq | |
2261 | * lock now, to be *sure*. If we're wrong, we'll | |
2262 | * just go back and repeat. | |
2263 | */ | |
2264 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2265 | trace_sched_wait_task(p); |
3a5c359a | 2266 | running = task_running(rq, p); |
fd2f4419 | 2267 | on_rq = p->on_rq; |
85ba2d86 | 2268 | ncsw = 0; |
f31e11d8 | 2269 | if (!match_state || p->state == match_state) |
93dcf55f | 2270 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 2271 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 2272 | |
85ba2d86 RM |
2273 | /* |
2274 | * If it changed from the expected state, bail out now. | |
2275 | */ | |
2276 | if (unlikely(!ncsw)) | |
2277 | break; | |
2278 | ||
3a5c359a AK |
2279 | /* |
2280 | * Was it really running after all now that we | |
2281 | * checked with the proper locks actually held? | |
2282 | * | |
2283 | * Oops. Go back and try again.. | |
2284 | */ | |
2285 | if (unlikely(running)) { | |
2286 | cpu_relax(); | |
2287 | continue; | |
2288 | } | |
fa490cfd | 2289 | |
3a5c359a AK |
2290 | /* |
2291 | * It's not enough that it's not actively running, | |
2292 | * it must be off the runqueue _entirely_, and not | |
2293 | * preempted! | |
2294 | * | |
80dd99b3 | 2295 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2296 | * running right now), it's preempted, and we should |
2297 | * yield - it could be a while. | |
2298 | */ | |
2299 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2300 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2301 | ||
2302 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2303 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2304 | continue; |
2305 | } | |
fa490cfd | 2306 | |
3a5c359a AK |
2307 | /* |
2308 | * Ahh, all good. It wasn't running, and it wasn't | |
2309 | * runnable, which means that it will never become | |
2310 | * running in the future either. We're all done! | |
2311 | */ | |
2312 | break; | |
2313 | } | |
85ba2d86 RM |
2314 | |
2315 | return ncsw; | |
1da177e4 LT |
2316 | } |
2317 | ||
2318 | /*** | |
2319 | * kick_process - kick a running thread to enter/exit the kernel | |
2320 | * @p: the to-be-kicked thread | |
2321 | * | |
2322 | * Cause a process which is running on another CPU to enter | |
2323 | * kernel-mode, without any delay. (to get signals handled.) | |
2324 | * | |
25985edc | 2325 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2326 | * because all it wants to ensure is that the remote task enters |
2327 | * the kernel. If the IPI races and the task has been migrated | |
2328 | * to another CPU then no harm is done and the purpose has been | |
2329 | * achieved as well. | |
2330 | */ | |
36c8b586 | 2331 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2332 | { |
2333 | int cpu; | |
2334 | ||
2335 | preempt_disable(); | |
2336 | cpu = task_cpu(p); | |
2337 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2338 | smp_send_reschedule(cpu); | |
2339 | preempt_enable(); | |
2340 | } | |
b43e3521 | 2341 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2342 | #endif /* CONFIG_SMP */ |
1da177e4 | 2343 | |
970b13ba | 2344 | #ifdef CONFIG_SMP |
30da688e | 2345 | /* |
013fdb80 | 2346 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 2347 | */ |
5da9a0fb PZ |
2348 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2349 | { | |
2350 | int dest_cpu; | |
2351 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2352 | ||
2353 | /* Look for allowed, online CPU in same node. */ | |
2354 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2355 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2356 | return dest_cpu; | |
2357 | ||
2358 | /* Any allowed, online CPU? */ | |
2359 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2360 | if (dest_cpu < nr_cpu_ids) | |
2361 | return dest_cpu; | |
2362 | ||
2363 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2364 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2365 | /* | |
2366 | * Don't tell them about moving exiting tasks or | |
2367 | * kernel threads (both mm NULL), since they never | |
2368 | * leave kernel. | |
2369 | */ | |
2370 | if (p->mm && printk_ratelimit()) { | |
2371 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2372 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2373 | } |
2374 | ||
2375 | return dest_cpu; | |
2376 | } | |
2377 | ||
e2912009 | 2378 | /* |
013fdb80 | 2379 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 2380 | */ |
970b13ba | 2381 | static inline |
7608dec2 | 2382 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2383 | { |
7608dec2 | 2384 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
2385 | |
2386 | /* | |
2387 | * In order not to call set_task_cpu() on a blocking task we need | |
2388 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2389 | * cpu. | |
2390 | * | |
2391 | * Since this is common to all placement strategies, this lives here. | |
2392 | * | |
2393 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2394 | * not worry about this generic constraint ] | |
2395 | */ | |
2396 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2397 | !cpu_online(cpu))) |
5da9a0fb | 2398 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2399 | |
2400 | return cpu; | |
970b13ba | 2401 | } |
09a40af5 MG |
2402 | |
2403 | static void update_avg(u64 *avg, u64 sample) | |
2404 | { | |
2405 | s64 diff = sample - *avg; | |
2406 | *avg += diff >> 3; | |
2407 | } | |
970b13ba PZ |
2408 | #endif |
2409 | ||
d7c01d27 | 2410 | static void |
b84cb5df | 2411 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2412 | { |
d7c01d27 | 2413 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
2414 | struct rq *rq = this_rq(); |
2415 | ||
d7c01d27 PZ |
2416 | #ifdef CONFIG_SMP |
2417 | int this_cpu = smp_processor_id(); | |
2418 | ||
2419 | if (cpu == this_cpu) { | |
2420 | schedstat_inc(rq, ttwu_local); | |
2421 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2422 | } else { | |
2423 | struct sched_domain *sd; | |
2424 | ||
2425 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 2426 | rcu_read_lock(); |
d7c01d27 PZ |
2427 | for_each_domain(this_cpu, sd) { |
2428 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2429 | schedstat_inc(sd, ttwu_wake_remote); | |
2430 | break; | |
2431 | } | |
2432 | } | |
057f3fad | 2433 | rcu_read_unlock(); |
d7c01d27 PZ |
2434 | } |
2435 | #endif /* CONFIG_SMP */ | |
2436 | ||
2437 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2438 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2439 | |
2440 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2441 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2442 | |
2443 | if (cpu != task_cpu(p)) | |
9ed3811a | 2444 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
9ed3811a | 2445 | |
d7c01d27 PZ |
2446 | #endif /* CONFIG_SCHEDSTATS */ |
2447 | } | |
2448 | ||
2449 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2450 | { | |
9ed3811a | 2451 | activate_task(rq, p, en_flags); |
fd2f4419 | 2452 | p->on_rq = 1; |
c2f7115e PZ |
2453 | |
2454 | /* if a worker is waking up, notify workqueue */ | |
2455 | if (p->flags & PF_WQ_WORKER) | |
2456 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2457 | } |
2458 | ||
23f41eeb PZ |
2459 | /* |
2460 | * Mark the task runnable and perform wakeup-preemption. | |
2461 | */ | |
89363381 | 2462 | static void |
23f41eeb | 2463 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 2464 | { |
89363381 | 2465 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2466 | check_preempt_curr(rq, p, wake_flags); |
2467 | ||
2468 | p->state = TASK_RUNNING; | |
2469 | #ifdef CONFIG_SMP | |
2470 | if (p->sched_class->task_woken) | |
2471 | p->sched_class->task_woken(rq, p); | |
2472 | ||
2473 | if (unlikely(rq->idle_stamp)) { | |
2474 | u64 delta = rq->clock - rq->idle_stamp; | |
2475 | u64 max = 2*sysctl_sched_migration_cost; | |
2476 | ||
2477 | if (delta > max) | |
2478 | rq->avg_idle = max; | |
2479 | else | |
2480 | update_avg(&rq->avg_idle, delta); | |
2481 | rq->idle_stamp = 0; | |
2482 | } | |
2483 | #endif | |
2484 | } | |
2485 | ||
c05fbafb PZ |
2486 | static void |
2487 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
2488 | { | |
2489 | #ifdef CONFIG_SMP | |
2490 | if (p->sched_contributes_to_load) | |
2491 | rq->nr_uninterruptible--; | |
2492 | #endif | |
2493 | ||
2494 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
2495 | ttwu_do_wakeup(rq, p, wake_flags); | |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2500 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2501 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2502 | * the task is still ->on_rq. | |
2503 | */ | |
2504 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2505 | { | |
2506 | struct rq *rq; | |
2507 | int ret = 0; | |
2508 | ||
2509 | rq = __task_rq_lock(p); | |
2510 | if (p->on_rq) { | |
2511 | ttwu_do_wakeup(rq, p, wake_flags); | |
2512 | ret = 1; | |
2513 | } | |
2514 | __task_rq_unlock(rq); | |
2515 | ||
2516 | return ret; | |
2517 | } | |
2518 | ||
317f3941 PZ |
2519 | #ifdef CONFIG_SMP |
2520 | static void sched_ttwu_pending(void) | |
2521 | { | |
2522 | struct rq *rq = this_rq(); | |
2523 | struct task_struct *list = xchg(&rq->wake_list, NULL); | |
2524 | ||
2525 | if (!list) | |
2526 | return; | |
2527 | ||
2528 | raw_spin_lock(&rq->lock); | |
2529 | ||
2530 | while (list) { | |
2531 | struct task_struct *p = list; | |
2532 | list = list->wake_entry; | |
2533 | ttwu_do_activate(rq, p, 0); | |
2534 | } | |
2535 | ||
2536 | raw_spin_unlock(&rq->lock); | |
2537 | } | |
2538 | ||
2539 | void scheduler_ipi(void) | |
2540 | { | |
2541 | sched_ttwu_pending(); | |
2542 | } | |
2543 | ||
2544 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
2545 | { | |
2546 | struct rq *rq = cpu_rq(cpu); | |
2547 | struct task_struct *next = rq->wake_list; | |
2548 | ||
2549 | for (;;) { | |
2550 | struct task_struct *old = next; | |
2551 | ||
2552 | p->wake_entry = next; | |
2553 | next = cmpxchg(&rq->wake_list, old, p); | |
2554 | if (next == old) | |
2555 | break; | |
2556 | } | |
2557 | ||
2558 | if (!next) | |
2559 | smp_send_reschedule(cpu); | |
2560 | } | |
2561 | #endif | |
2562 | ||
c05fbafb PZ |
2563 | static void ttwu_queue(struct task_struct *p, int cpu) |
2564 | { | |
2565 | struct rq *rq = cpu_rq(cpu); | |
2566 | ||
17d9f311 | 2567 | #if defined(CONFIG_SMP) |
317f3941 PZ |
2568 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { |
2569 | ttwu_queue_remote(p, cpu); | |
2570 | return; | |
2571 | } | |
2572 | #endif | |
2573 | ||
c05fbafb PZ |
2574 | raw_spin_lock(&rq->lock); |
2575 | ttwu_do_activate(rq, p, 0); | |
2576 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
2577 | } |
2578 | ||
2579 | /** | |
1da177e4 | 2580 | * try_to_wake_up - wake up a thread |
9ed3811a | 2581 | * @p: the thread to be awakened |
1da177e4 | 2582 | * @state: the mask of task states that can be woken |
9ed3811a | 2583 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2584 | * |
2585 | * Put it on the run-queue if it's not already there. The "current" | |
2586 | * thread is always on the run-queue (except when the actual | |
2587 | * re-schedule is in progress), and as such you're allowed to do | |
2588 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2589 | * runnable without the overhead of this. | |
2590 | * | |
9ed3811a TH |
2591 | * Returns %true if @p was woken up, %false if it was already running |
2592 | * or @state didn't match @p's state. | |
1da177e4 | 2593 | */ |
e4a52bcb PZ |
2594 | static int |
2595 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2596 | { |
1da177e4 | 2597 | unsigned long flags; |
c05fbafb | 2598 | int cpu, success = 0; |
2398f2c6 | 2599 | |
04e2f174 | 2600 | smp_wmb(); |
013fdb80 | 2601 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2602 | if (!(p->state & state)) |
1da177e4 LT |
2603 | goto out; |
2604 | ||
c05fbafb | 2605 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2606 | cpu = task_cpu(p); |
1da177e4 | 2607 | |
c05fbafb PZ |
2608 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2609 | goto stat; | |
1da177e4 | 2610 | |
1da177e4 | 2611 | #ifdef CONFIG_SMP |
e9c84311 | 2612 | /* |
c05fbafb PZ |
2613 | * If the owning (remote) cpu is still in the middle of schedule() with |
2614 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 2615 | */ |
e4a52bcb PZ |
2616 | while (p->on_cpu) { |
2617 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2618 | /* | |
2619 | * If called from interrupt context we could have landed in the | |
2620 | * middle of schedule(), in this case we should take care not | |
2621 | * to spin on ->on_cpu if p is current, since that would | |
2622 | * deadlock. | |
2623 | */ | |
c05fbafb PZ |
2624 | if (p == current) { |
2625 | ttwu_queue(p, cpu); | |
2626 | goto stat; | |
2627 | } | |
e4a52bcb PZ |
2628 | #endif |
2629 | cpu_relax(); | |
371fd7e7 | 2630 | } |
0970d299 | 2631 | /* |
e4a52bcb | 2632 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 2633 | */ |
e4a52bcb | 2634 | smp_rmb(); |
1da177e4 | 2635 | |
a8e4f2ea | 2636 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2637 | p->state = TASK_WAKING; |
e7693a36 | 2638 | |
e4a52bcb | 2639 | if (p->sched_class->task_waking) |
74f8e4b2 | 2640 | p->sched_class->task_waking(p); |
efbbd05a | 2641 | |
7608dec2 | 2642 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
c05fbafb | 2643 | if (task_cpu(p) != cpu) |
e4a52bcb | 2644 | set_task_cpu(p, cpu); |
1da177e4 | 2645 | #endif /* CONFIG_SMP */ |
1da177e4 | 2646 | |
c05fbafb PZ |
2647 | ttwu_queue(p, cpu); |
2648 | stat: | |
b84cb5df | 2649 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2650 | out: |
013fdb80 | 2651 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2652 | |
2653 | return success; | |
2654 | } | |
2655 | ||
21aa9af0 TH |
2656 | /** |
2657 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2658 | * @p: the thread to be awakened | |
2659 | * | |
2acca55e | 2660 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2661 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2662 | * the current task. |
21aa9af0 TH |
2663 | */ |
2664 | static void try_to_wake_up_local(struct task_struct *p) | |
2665 | { | |
2666 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2667 | |
2668 | BUG_ON(rq != this_rq()); | |
2669 | BUG_ON(p == current); | |
2670 | lockdep_assert_held(&rq->lock); | |
2671 | ||
2acca55e PZ |
2672 | if (!raw_spin_trylock(&p->pi_lock)) { |
2673 | raw_spin_unlock(&rq->lock); | |
2674 | raw_spin_lock(&p->pi_lock); | |
2675 | raw_spin_lock(&rq->lock); | |
2676 | } | |
2677 | ||
21aa9af0 | 2678 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2679 | goto out; |
21aa9af0 | 2680 | |
fd2f4419 | 2681 | if (!p->on_rq) |
d7c01d27 PZ |
2682 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2683 | ||
23f41eeb | 2684 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2685 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2686 | out: |
2687 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2688 | } |
2689 | ||
50fa610a DH |
2690 | /** |
2691 | * wake_up_process - Wake up a specific process | |
2692 | * @p: The process to be woken up. | |
2693 | * | |
2694 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2695 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2696 | * running. | |
2697 | * | |
2698 | * It may be assumed that this function implies a write memory barrier before | |
2699 | * changing the task state if and only if any tasks are woken up. | |
2700 | */ | |
7ad5b3a5 | 2701 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2702 | { |
d9514f6c | 2703 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2704 | } |
1da177e4 LT |
2705 | EXPORT_SYMBOL(wake_up_process); |
2706 | ||
7ad5b3a5 | 2707 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2708 | { |
2709 | return try_to_wake_up(p, state, 0); | |
2710 | } | |
2711 | ||
1da177e4 LT |
2712 | /* |
2713 | * Perform scheduler related setup for a newly forked process p. | |
2714 | * p is forked by current. | |
dd41f596 IM |
2715 | * |
2716 | * __sched_fork() is basic setup used by init_idle() too: | |
2717 | */ | |
2718 | static void __sched_fork(struct task_struct *p) | |
2719 | { | |
fd2f4419 PZ |
2720 | p->on_rq = 0; |
2721 | ||
2722 | p->se.on_rq = 0; | |
dd41f596 IM |
2723 | p->se.exec_start = 0; |
2724 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2725 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2726 | p->se.nr_migrations = 0; |
da7a735e | 2727 | p->se.vruntime = 0; |
fd2f4419 | 2728 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2729 | |
2730 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2731 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2732 | #endif |
476d139c | 2733 | |
fa717060 | 2734 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2735 | |
e107be36 AK |
2736 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2737 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2738 | #endif | |
dd41f596 IM |
2739 | } |
2740 | ||
2741 | /* | |
2742 | * fork()/clone()-time setup: | |
2743 | */ | |
3e51e3ed | 2744 | void sched_fork(struct task_struct *p) |
dd41f596 | 2745 | { |
0122ec5b | 2746 | unsigned long flags; |
dd41f596 IM |
2747 | int cpu = get_cpu(); |
2748 | ||
2749 | __sched_fork(p); | |
06b83b5f | 2750 | /* |
0017d735 | 2751 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2752 | * nobody will actually run it, and a signal or other external |
2753 | * event cannot wake it up and insert it on the runqueue either. | |
2754 | */ | |
0017d735 | 2755 | p->state = TASK_RUNNING; |
dd41f596 | 2756 | |
b9dc29e7 MG |
2757 | /* |
2758 | * Revert to default priority/policy on fork if requested. | |
2759 | */ | |
2760 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2761 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2762 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2763 | p->normal_prio = p->static_prio; |
2764 | } | |
b9dc29e7 | 2765 | |
6c697bdf MG |
2766 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2767 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2768 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2769 | set_load_weight(p); |
2770 | } | |
2771 | ||
b9dc29e7 MG |
2772 | /* |
2773 | * We don't need the reset flag anymore after the fork. It has | |
2774 | * fulfilled its duty: | |
2775 | */ | |
2776 | p->sched_reset_on_fork = 0; | |
2777 | } | |
ca94c442 | 2778 | |
f83f9ac2 PW |
2779 | /* |
2780 | * Make sure we do not leak PI boosting priority to the child. | |
2781 | */ | |
2782 | p->prio = current->normal_prio; | |
2783 | ||
2ddbf952 HS |
2784 | if (!rt_prio(p->prio)) |
2785 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2786 | |
cd29fe6f PZ |
2787 | if (p->sched_class->task_fork) |
2788 | p->sched_class->task_fork(p); | |
2789 | ||
86951599 PZ |
2790 | /* |
2791 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2792 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2793 | * is ran before sched_fork(). | |
2794 | * | |
2795 | * Silence PROVE_RCU. | |
2796 | */ | |
0122ec5b | 2797 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2798 | set_task_cpu(p, cpu); |
0122ec5b | 2799 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2800 | |
52f17b6c | 2801 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2802 | if (likely(sched_info_on())) |
52f17b6c | 2803 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2804 | #endif |
3ca7a440 PZ |
2805 | #if defined(CONFIG_SMP) |
2806 | p->on_cpu = 0; | |
4866cde0 | 2807 | #endif |
1da177e4 | 2808 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2809 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2810 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2811 | #endif |
806c09a7 | 2812 | #ifdef CONFIG_SMP |
917b627d | 2813 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2814 | #endif |
917b627d | 2815 | |
476d139c | 2816 | put_cpu(); |
1da177e4 LT |
2817 | } |
2818 | ||
2819 | /* | |
2820 | * wake_up_new_task - wake up a newly created task for the first time. | |
2821 | * | |
2822 | * This function will do some initial scheduler statistics housekeeping | |
2823 | * that must be done for every newly created context, then puts the task | |
2824 | * on the runqueue and wakes it. | |
2825 | */ | |
3e51e3ed | 2826 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2827 | { |
2828 | unsigned long flags; | |
dd41f596 | 2829 | struct rq *rq; |
fabf318e | 2830 | |
ab2515c4 | 2831 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2832 | #ifdef CONFIG_SMP |
2833 | /* | |
2834 | * Fork balancing, do it here and not earlier because: | |
2835 | * - cpus_allowed can change in the fork path | |
2836 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2837 | */ |
ab2515c4 | 2838 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2839 | #endif |
2840 | ||
ab2515c4 | 2841 | rq = __task_rq_lock(p); |
cd29fe6f | 2842 | activate_task(rq, p, 0); |
fd2f4419 | 2843 | p->on_rq = 1; |
89363381 | 2844 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2845 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2846 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2847 | if (p->sched_class->task_woken) |
2848 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2849 | #endif |
0122ec5b | 2850 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2851 | } |
2852 | ||
e107be36 AK |
2853 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2854 | ||
2855 | /** | |
80dd99b3 | 2856 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2857 | * @notifier: notifier struct to register |
e107be36 AK |
2858 | */ |
2859 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2860 | { | |
2861 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2862 | } | |
2863 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2864 | ||
2865 | /** | |
2866 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2867 | * @notifier: notifier struct to unregister |
e107be36 AK |
2868 | * |
2869 | * This is safe to call from within a preemption notifier. | |
2870 | */ | |
2871 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2872 | { | |
2873 | hlist_del(¬ifier->link); | |
2874 | } | |
2875 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2876 | ||
2877 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2878 | { | |
2879 | struct preempt_notifier *notifier; | |
2880 | struct hlist_node *node; | |
2881 | ||
2882 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2883 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2884 | } | |
2885 | ||
2886 | static void | |
2887 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2888 | struct task_struct *next) | |
2889 | { | |
2890 | struct preempt_notifier *notifier; | |
2891 | struct hlist_node *node; | |
2892 | ||
2893 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2894 | notifier->ops->sched_out(notifier, next); | |
2895 | } | |
2896 | ||
6d6bc0ad | 2897 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2898 | |
2899 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2900 | { | |
2901 | } | |
2902 | ||
2903 | static void | |
2904 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2905 | struct task_struct *next) | |
2906 | { | |
2907 | } | |
2908 | ||
6d6bc0ad | 2909 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2910 | |
4866cde0 NP |
2911 | /** |
2912 | * prepare_task_switch - prepare to switch tasks | |
2913 | * @rq: the runqueue preparing to switch | |
421cee29 | 2914 | * @prev: the current task that is being switched out |
4866cde0 NP |
2915 | * @next: the task we are going to switch to. |
2916 | * | |
2917 | * This is called with the rq lock held and interrupts off. It must | |
2918 | * be paired with a subsequent finish_task_switch after the context | |
2919 | * switch. | |
2920 | * | |
2921 | * prepare_task_switch sets up locking and calls architecture specific | |
2922 | * hooks. | |
2923 | */ | |
e107be36 AK |
2924 | static inline void |
2925 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2926 | struct task_struct *next) | |
4866cde0 | 2927 | { |
fe4b04fa PZ |
2928 | sched_info_switch(prev, next); |
2929 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2930 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2931 | prepare_lock_switch(rq, next); |
2932 | prepare_arch_switch(next); | |
fe4b04fa | 2933 | trace_sched_switch(prev, next); |
4866cde0 NP |
2934 | } |
2935 | ||
1da177e4 LT |
2936 | /** |
2937 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2938 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2939 | * @prev: the thread we just switched away from. |
2940 | * | |
4866cde0 NP |
2941 | * finish_task_switch must be called after the context switch, paired |
2942 | * with a prepare_task_switch call before the context switch. | |
2943 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2944 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2945 | * |
2946 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2947 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2948 | * with the lock held can cause deadlocks; see schedule() for |
2949 | * details.) | |
2950 | */ | |
a9957449 | 2951 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2952 | __releases(rq->lock) |
2953 | { | |
1da177e4 | 2954 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2955 | long prev_state; |
1da177e4 LT |
2956 | |
2957 | rq->prev_mm = NULL; | |
2958 | ||
2959 | /* | |
2960 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2961 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2962 | * schedule one last time. The schedule call will never return, and |
2963 | * the scheduled task must drop that reference. | |
c394cc9f | 2964 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2965 | * still held, otherwise prev could be scheduled on another cpu, die |
2966 | * there before we look at prev->state, and then the reference would | |
2967 | * be dropped twice. | |
2968 | * Manfred Spraul <manfred@colorfullife.com> | |
2969 | */ | |
55a101f8 | 2970 | prev_state = prev->state; |
4866cde0 | 2971 | finish_arch_switch(prev); |
8381f65d JI |
2972 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2973 | local_irq_disable(); | |
2974 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2975 | perf_event_task_sched_in(current); |
8381f65d JI |
2976 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2977 | local_irq_enable(); | |
2978 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2979 | finish_lock_switch(rq, prev); |
e8fa1362 | 2980 | |
e107be36 | 2981 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2982 | if (mm) |
2983 | mmdrop(mm); | |
c394cc9f | 2984 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2985 | /* |
2986 | * Remove function-return probe instances associated with this | |
2987 | * task and put them back on the free list. | |
9761eea8 | 2988 | */ |
c6fd91f0 | 2989 | kprobe_flush_task(prev); |
1da177e4 | 2990 | put_task_struct(prev); |
c6fd91f0 | 2991 | } |
1da177e4 LT |
2992 | } |
2993 | ||
3f029d3c GH |
2994 | #ifdef CONFIG_SMP |
2995 | ||
2996 | /* assumes rq->lock is held */ | |
2997 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2998 | { | |
2999 | if (prev->sched_class->pre_schedule) | |
3000 | prev->sched_class->pre_schedule(rq, prev); | |
3001 | } | |
3002 | ||
3003 | /* rq->lock is NOT held, but preemption is disabled */ | |
3004 | static inline void post_schedule(struct rq *rq) | |
3005 | { | |
3006 | if (rq->post_schedule) { | |
3007 | unsigned long flags; | |
3008 | ||
05fa785c | 3009 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
3010 | if (rq->curr->sched_class->post_schedule) |
3011 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 3012 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
3013 | |
3014 | rq->post_schedule = 0; | |
3015 | } | |
3016 | } | |
3017 | ||
3018 | #else | |
da19ab51 | 3019 | |
3f029d3c GH |
3020 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3021 | { | |
3022 | } | |
3023 | ||
3024 | static inline void post_schedule(struct rq *rq) | |
3025 | { | |
1da177e4 LT |
3026 | } |
3027 | ||
3f029d3c GH |
3028 | #endif |
3029 | ||
1da177e4 LT |
3030 | /** |
3031 | * schedule_tail - first thing a freshly forked thread must call. | |
3032 | * @prev: the thread we just switched away from. | |
3033 | */ | |
36c8b586 | 3034 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3035 | __releases(rq->lock) |
3036 | { | |
70b97a7f IM |
3037 | struct rq *rq = this_rq(); |
3038 | ||
4866cde0 | 3039 | finish_task_switch(rq, prev); |
da19ab51 | 3040 | |
3f029d3c GH |
3041 | /* |
3042 | * FIXME: do we need to worry about rq being invalidated by the | |
3043 | * task_switch? | |
3044 | */ | |
3045 | post_schedule(rq); | |
70b97a7f | 3046 | |
4866cde0 NP |
3047 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3048 | /* In this case, finish_task_switch does not reenable preemption */ | |
3049 | preempt_enable(); | |
3050 | #endif | |
1da177e4 | 3051 | if (current->set_child_tid) |
b488893a | 3052 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3053 | } |
3054 | ||
3055 | /* | |
3056 | * context_switch - switch to the new MM and the new | |
3057 | * thread's register state. | |
3058 | */ | |
dd41f596 | 3059 | static inline void |
70b97a7f | 3060 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3061 | struct task_struct *next) |
1da177e4 | 3062 | { |
dd41f596 | 3063 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3064 | |
e107be36 | 3065 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3066 | |
dd41f596 IM |
3067 | mm = next->mm; |
3068 | oldmm = prev->active_mm; | |
9226d125 ZA |
3069 | /* |
3070 | * For paravirt, this is coupled with an exit in switch_to to | |
3071 | * combine the page table reload and the switch backend into | |
3072 | * one hypercall. | |
3073 | */ | |
224101ed | 3074 | arch_start_context_switch(prev); |
9226d125 | 3075 | |
31915ab4 | 3076 | if (!mm) { |
1da177e4 LT |
3077 | next->active_mm = oldmm; |
3078 | atomic_inc(&oldmm->mm_count); | |
3079 | enter_lazy_tlb(oldmm, next); | |
3080 | } else | |
3081 | switch_mm(oldmm, mm, next); | |
3082 | ||
31915ab4 | 3083 | if (!prev->mm) { |
1da177e4 | 3084 | prev->active_mm = NULL; |
1da177e4 LT |
3085 | rq->prev_mm = oldmm; |
3086 | } | |
3a5f5e48 IM |
3087 | /* |
3088 | * Since the runqueue lock will be released by the next | |
3089 | * task (which is an invalid locking op but in the case | |
3090 | * of the scheduler it's an obvious special-case), so we | |
3091 | * do an early lockdep release here: | |
3092 | */ | |
3093 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3094 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3095 | #endif |
1da177e4 LT |
3096 | |
3097 | /* Here we just switch the register state and the stack. */ | |
3098 | switch_to(prev, next, prev); | |
3099 | ||
dd41f596 IM |
3100 | barrier(); |
3101 | /* | |
3102 | * this_rq must be evaluated again because prev may have moved | |
3103 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3104 | * frame will be invalid. | |
3105 | */ | |
3106 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3107 | } |
3108 | ||
3109 | /* | |
3110 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3111 | * | |
3112 | * externally visible scheduler statistics: current number of runnable | |
3113 | * threads, current number of uninterruptible-sleeping threads, total | |
3114 | * number of context switches performed since bootup. | |
3115 | */ | |
3116 | unsigned long nr_running(void) | |
3117 | { | |
3118 | unsigned long i, sum = 0; | |
3119 | ||
3120 | for_each_online_cpu(i) | |
3121 | sum += cpu_rq(i)->nr_running; | |
3122 | ||
3123 | return sum; | |
f711f609 | 3124 | } |
1da177e4 LT |
3125 | |
3126 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3127 | { |
1da177e4 | 3128 | unsigned long i, sum = 0; |
f711f609 | 3129 | |
0a945022 | 3130 | for_each_possible_cpu(i) |
1da177e4 | 3131 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3132 | |
3133 | /* | |
1da177e4 LT |
3134 | * Since we read the counters lockless, it might be slightly |
3135 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3136 | */ |
1da177e4 LT |
3137 | if (unlikely((long)sum < 0)) |
3138 | sum = 0; | |
f711f609 | 3139 | |
1da177e4 | 3140 | return sum; |
f711f609 | 3141 | } |
f711f609 | 3142 | |
1da177e4 | 3143 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3144 | { |
cc94abfc SR |
3145 | int i; |
3146 | unsigned long long sum = 0; | |
46cb4b7c | 3147 | |
0a945022 | 3148 | for_each_possible_cpu(i) |
1da177e4 | 3149 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3150 | |
1da177e4 LT |
3151 | return sum; |
3152 | } | |
483b4ee6 | 3153 | |
1da177e4 LT |
3154 | unsigned long nr_iowait(void) |
3155 | { | |
3156 | unsigned long i, sum = 0; | |
483b4ee6 | 3157 | |
0a945022 | 3158 | for_each_possible_cpu(i) |
1da177e4 | 3159 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3160 | |
1da177e4 LT |
3161 | return sum; |
3162 | } | |
483b4ee6 | 3163 | |
8c215bd3 | 3164 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3165 | { |
8c215bd3 | 3166 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3167 | return atomic_read(&this->nr_iowait); |
3168 | } | |
46cb4b7c | 3169 | |
69d25870 AV |
3170 | unsigned long this_cpu_load(void) |
3171 | { | |
3172 | struct rq *this = this_rq(); | |
3173 | return this->cpu_load[0]; | |
3174 | } | |
e790fb0b | 3175 | |
46cb4b7c | 3176 | |
dce48a84 TG |
3177 | /* Variables and functions for calc_load */ |
3178 | static atomic_long_t calc_load_tasks; | |
3179 | static unsigned long calc_load_update; | |
3180 | unsigned long avenrun[3]; | |
3181 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3182 | |
74f5187a PZ |
3183 | static long calc_load_fold_active(struct rq *this_rq) |
3184 | { | |
3185 | long nr_active, delta = 0; | |
3186 | ||
3187 | nr_active = this_rq->nr_running; | |
3188 | nr_active += (long) this_rq->nr_uninterruptible; | |
3189 | ||
3190 | if (nr_active != this_rq->calc_load_active) { | |
3191 | delta = nr_active - this_rq->calc_load_active; | |
3192 | this_rq->calc_load_active = nr_active; | |
3193 | } | |
3194 | ||
3195 | return delta; | |
3196 | } | |
3197 | ||
0f004f5a PZ |
3198 | static unsigned long |
3199 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3200 | { | |
3201 | load *= exp; | |
3202 | load += active * (FIXED_1 - exp); | |
3203 | load += 1UL << (FSHIFT - 1); | |
3204 | return load >> FSHIFT; | |
3205 | } | |
3206 | ||
74f5187a PZ |
3207 | #ifdef CONFIG_NO_HZ |
3208 | /* | |
3209 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3210 | * | |
3211 | * When making the ILB scale, we should try to pull this in as well. | |
3212 | */ | |
3213 | static atomic_long_t calc_load_tasks_idle; | |
3214 | ||
3215 | static void calc_load_account_idle(struct rq *this_rq) | |
3216 | { | |
3217 | long delta; | |
3218 | ||
3219 | delta = calc_load_fold_active(this_rq); | |
3220 | if (delta) | |
3221 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3222 | } | |
3223 | ||
3224 | static long calc_load_fold_idle(void) | |
3225 | { | |
3226 | long delta = 0; | |
3227 | ||
3228 | /* | |
3229 | * Its got a race, we don't care... | |
3230 | */ | |
3231 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3232 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3233 | ||
3234 | return delta; | |
3235 | } | |
0f004f5a PZ |
3236 | |
3237 | /** | |
3238 | * fixed_power_int - compute: x^n, in O(log n) time | |
3239 | * | |
3240 | * @x: base of the power | |
3241 | * @frac_bits: fractional bits of @x | |
3242 | * @n: power to raise @x to. | |
3243 | * | |
3244 | * By exploiting the relation between the definition of the natural power | |
3245 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3246 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3247 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3248 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3249 | * of course trivially computable in O(log_2 n), the length of our binary | |
3250 | * vector. | |
3251 | */ | |
3252 | static unsigned long | |
3253 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3254 | { | |
3255 | unsigned long result = 1UL << frac_bits; | |
3256 | ||
3257 | if (n) for (;;) { | |
3258 | if (n & 1) { | |
3259 | result *= x; | |
3260 | result += 1UL << (frac_bits - 1); | |
3261 | result >>= frac_bits; | |
3262 | } | |
3263 | n >>= 1; | |
3264 | if (!n) | |
3265 | break; | |
3266 | x *= x; | |
3267 | x += 1UL << (frac_bits - 1); | |
3268 | x >>= frac_bits; | |
3269 | } | |
3270 | ||
3271 | return result; | |
3272 | } | |
3273 | ||
3274 | /* | |
3275 | * a1 = a0 * e + a * (1 - e) | |
3276 | * | |
3277 | * a2 = a1 * e + a * (1 - e) | |
3278 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3279 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3280 | * | |
3281 | * a3 = a2 * e + a * (1 - e) | |
3282 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3283 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3284 | * | |
3285 | * ... | |
3286 | * | |
3287 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3288 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3289 | * = a0 * e^n + a * (1 - e^n) | |
3290 | * | |
3291 | * [1] application of the geometric series: | |
3292 | * | |
3293 | * n 1 - x^(n+1) | |
3294 | * S_n := \Sum x^i = ------------- | |
3295 | * i=0 1 - x | |
3296 | */ | |
3297 | static unsigned long | |
3298 | calc_load_n(unsigned long load, unsigned long exp, | |
3299 | unsigned long active, unsigned int n) | |
3300 | { | |
3301 | ||
3302 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3303 | } | |
3304 | ||
3305 | /* | |
3306 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3307 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3308 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3309 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3310 | * | |
3311 | * Once we've updated the global active value, we need to apply the exponential | |
3312 | * weights adjusted to the number of cycles missed. | |
3313 | */ | |
3314 | static void calc_global_nohz(unsigned long ticks) | |
3315 | { | |
3316 | long delta, active, n; | |
3317 | ||
3318 | if (time_before(jiffies, calc_load_update)) | |
3319 | return; | |
3320 | ||
3321 | /* | |
3322 | * If we crossed a calc_load_update boundary, make sure to fold | |
3323 | * any pending idle changes, the respective CPUs might have | |
3324 | * missed the tick driven calc_load_account_active() update | |
3325 | * due to NO_HZ. | |
3326 | */ | |
3327 | delta = calc_load_fold_idle(); | |
3328 | if (delta) | |
3329 | atomic_long_add(delta, &calc_load_tasks); | |
3330 | ||
3331 | /* | |
3332 | * If we were idle for multiple load cycles, apply them. | |
3333 | */ | |
3334 | if (ticks >= LOAD_FREQ) { | |
3335 | n = ticks / LOAD_FREQ; | |
3336 | ||
3337 | active = atomic_long_read(&calc_load_tasks); | |
3338 | active = active > 0 ? active * FIXED_1 : 0; | |
3339 | ||
3340 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3341 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3342 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3343 | ||
3344 | calc_load_update += n * LOAD_FREQ; | |
3345 | } | |
3346 | ||
3347 | /* | |
3348 | * Its possible the remainder of the above division also crosses | |
3349 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3350 | * which comes after this will take care of that. | |
3351 | * | |
3352 | * Consider us being 11 ticks before a cycle completion, and us | |
3353 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3354 | * age us 4 cycles, and the test in calc_global_load() will | |
3355 | * pick up the final one. | |
3356 | */ | |
3357 | } | |
74f5187a PZ |
3358 | #else |
3359 | static void calc_load_account_idle(struct rq *this_rq) | |
3360 | { | |
3361 | } | |
3362 | ||
3363 | static inline long calc_load_fold_idle(void) | |
3364 | { | |
3365 | return 0; | |
3366 | } | |
0f004f5a PZ |
3367 | |
3368 | static void calc_global_nohz(unsigned long ticks) | |
3369 | { | |
3370 | } | |
74f5187a PZ |
3371 | #endif |
3372 | ||
2d02494f TG |
3373 | /** |
3374 | * get_avenrun - get the load average array | |
3375 | * @loads: pointer to dest load array | |
3376 | * @offset: offset to add | |
3377 | * @shift: shift count to shift the result left | |
3378 | * | |
3379 | * These values are estimates at best, so no need for locking. | |
3380 | */ | |
3381 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3382 | { | |
3383 | loads[0] = (avenrun[0] + offset) << shift; | |
3384 | loads[1] = (avenrun[1] + offset) << shift; | |
3385 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3386 | } |
46cb4b7c | 3387 | |
46cb4b7c | 3388 | /* |
dce48a84 TG |
3389 | * calc_load - update the avenrun load estimates 10 ticks after the |
3390 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3391 | */ |
0f004f5a | 3392 | void calc_global_load(unsigned long ticks) |
7835b98b | 3393 | { |
dce48a84 | 3394 | long active; |
1da177e4 | 3395 | |
0f004f5a PZ |
3396 | calc_global_nohz(ticks); |
3397 | ||
3398 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3399 | return; |
1da177e4 | 3400 | |
dce48a84 TG |
3401 | active = atomic_long_read(&calc_load_tasks); |
3402 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3403 | |
dce48a84 TG |
3404 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3405 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3406 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3407 | |
dce48a84 TG |
3408 | calc_load_update += LOAD_FREQ; |
3409 | } | |
1da177e4 | 3410 | |
dce48a84 | 3411 | /* |
74f5187a PZ |
3412 | * Called from update_cpu_load() to periodically update this CPU's |
3413 | * active count. | |
dce48a84 TG |
3414 | */ |
3415 | static void calc_load_account_active(struct rq *this_rq) | |
3416 | { | |
74f5187a | 3417 | long delta; |
08c183f3 | 3418 | |
74f5187a PZ |
3419 | if (time_before(jiffies, this_rq->calc_load_update)) |
3420 | return; | |
783609c6 | 3421 | |
74f5187a PZ |
3422 | delta = calc_load_fold_active(this_rq); |
3423 | delta += calc_load_fold_idle(); | |
3424 | if (delta) | |
dce48a84 | 3425 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3426 | |
3427 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3428 | } |
3429 | ||
fdf3e95d VP |
3430 | /* |
3431 | * The exact cpuload at various idx values, calculated at every tick would be | |
3432 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3433 | * | |
3434 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3435 | * on nth tick when cpu may be busy, then we have: | |
3436 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3437 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3438 | * | |
3439 | * decay_load_missed() below does efficient calculation of | |
3440 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3441 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3442 | * | |
3443 | * The calculation is approximated on a 128 point scale. | |
3444 | * degrade_zero_ticks is the number of ticks after which load at any | |
3445 | * particular idx is approximated to be zero. | |
3446 | * degrade_factor is a precomputed table, a row for each load idx. | |
3447 | * Each column corresponds to degradation factor for a power of two ticks, | |
3448 | * based on 128 point scale. | |
3449 | * Example: | |
3450 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3451 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3452 | * | |
3453 | * With this power of 2 load factors, we can degrade the load n times | |
3454 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3455 | * n mult/shifts needed by the exact degradation. | |
3456 | */ | |
3457 | #define DEGRADE_SHIFT 7 | |
3458 | static const unsigned char | |
3459 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3460 | static const unsigned char | |
3461 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3462 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3463 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3464 | {96, 72, 40, 12, 1, 0, 0}, | |
3465 | {112, 98, 75, 43, 15, 1, 0}, | |
3466 | {120, 112, 98, 76, 45, 16, 2} }; | |
3467 | ||
3468 | /* | |
3469 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3470 | * would be when CPU is idle and so we just decay the old load without | |
3471 | * adding any new load. | |
3472 | */ | |
3473 | static unsigned long | |
3474 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3475 | { | |
3476 | int j = 0; | |
3477 | ||
3478 | if (!missed_updates) | |
3479 | return load; | |
3480 | ||
3481 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3482 | return 0; | |
3483 | ||
3484 | if (idx == 1) | |
3485 | return load >> missed_updates; | |
3486 | ||
3487 | while (missed_updates) { | |
3488 | if (missed_updates % 2) | |
3489 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3490 | ||
3491 | missed_updates >>= 1; | |
3492 | j++; | |
3493 | } | |
3494 | return load; | |
3495 | } | |
3496 | ||
46cb4b7c | 3497 | /* |
dd41f596 | 3498 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3499 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3500 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3501 | */ |
dd41f596 | 3502 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3503 | { |
495eca49 | 3504 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3505 | unsigned long curr_jiffies = jiffies; |
3506 | unsigned long pending_updates; | |
dd41f596 | 3507 | int i, scale; |
46cb4b7c | 3508 | |
dd41f596 | 3509 | this_rq->nr_load_updates++; |
46cb4b7c | 3510 | |
fdf3e95d VP |
3511 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3512 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3513 | return; | |
3514 | ||
3515 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3516 | this_rq->last_load_update_tick = curr_jiffies; | |
3517 | ||
dd41f596 | 3518 | /* Update our load: */ |
fdf3e95d VP |
3519 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3520 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3521 | unsigned long old_load, new_load; |
7d1e6a9b | 3522 | |
dd41f596 | 3523 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3524 | |
dd41f596 | 3525 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3526 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3527 | new_load = this_load; |
a25707f3 IM |
3528 | /* |
3529 | * Round up the averaging division if load is increasing. This | |
3530 | * prevents us from getting stuck on 9 if the load is 10, for | |
3531 | * example. | |
3532 | */ | |
3533 | if (new_load > old_load) | |
fdf3e95d VP |
3534 | new_load += scale - 1; |
3535 | ||
3536 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3537 | } |
da2b71ed SS |
3538 | |
3539 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3540 | } |
3541 | ||
3542 | static void update_cpu_load_active(struct rq *this_rq) | |
3543 | { | |
3544 | update_cpu_load(this_rq); | |
46cb4b7c | 3545 | |
74f5187a | 3546 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3547 | } |
3548 | ||
dd41f596 | 3549 | #ifdef CONFIG_SMP |
8a0be9ef | 3550 | |
46cb4b7c | 3551 | /* |
38022906 PZ |
3552 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3553 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3554 | */ |
38022906 | 3555 | void sched_exec(void) |
46cb4b7c | 3556 | { |
38022906 | 3557 | struct task_struct *p = current; |
1da177e4 | 3558 | unsigned long flags; |
0017d735 | 3559 | int dest_cpu; |
46cb4b7c | 3560 | |
8f42ced9 | 3561 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 3562 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3563 | if (dest_cpu == smp_processor_id()) |
3564 | goto unlock; | |
38022906 | 3565 | |
8f42ced9 | 3566 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3567 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3568 | |
8f42ced9 PZ |
3569 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3570 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3571 | return; |
3572 | } | |
0017d735 | 3573 | unlock: |
8f42ced9 | 3574 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3575 | } |
dd41f596 | 3576 | |
1da177e4 LT |
3577 | #endif |
3578 | ||
1da177e4 LT |
3579 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3580 | ||
3581 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3582 | ||
3583 | /* | |
c5f8d995 | 3584 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3585 | * @p in case that task is currently running. |
c5f8d995 HS |
3586 | * |
3587 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3588 | */ |
c5f8d995 HS |
3589 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3590 | { | |
3591 | u64 ns = 0; | |
3592 | ||
3593 | if (task_current(rq, p)) { | |
3594 | update_rq_clock(rq); | |
305e6835 | 3595 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3596 | if ((s64)ns < 0) |
3597 | ns = 0; | |
3598 | } | |
3599 | ||
3600 | return ns; | |
3601 | } | |
3602 | ||
bb34d92f | 3603 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3604 | { |
1da177e4 | 3605 | unsigned long flags; |
41b86e9c | 3606 | struct rq *rq; |
bb34d92f | 3607 | u64 ns = 0; |
48f24c4d | 3608 | |
41b86e9c | 3609 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 3610 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 3611 | task_rq_unlock(rq, p, &flags); |
1508487e | 3612 | |
c5f8d995 HS |
3613 | return ns; |
3614 | } | |
f06febc9 | 3615 | |
c5f8d995 HS |
3616 | /* |
3617 | * Return accounted runtime for the task. | |
3618 | * In case the task is currently running, return the runtime plus current's | |
3619 | * pending runtime that have not been accounted yet. | |
3620 | */ | |
3621 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3622 | { | |
3623 | unsigned long flags; | |
3624 | struct rq *rq; | |
3625 | u64 ns = 0; | |
3626 | ||
3627 | rq = task_rq_lock(p, &flags); | |
3628 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3629 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
3630 | |
3631 | return ns; | |
3632 | } | |
48f24c4d | 3633 | |
c5f8d995 HS |
3634 | /* |
3635 | * Return sum_exec_runtime for the thread group. | |
3636 | * In case the task is currently running, return the sum plus current's | |
3637 | * pending runtime that have not been accounted yet. | |
3638 | * | |
3639 | * Note that the thread group might have other running tasks as well, | |
3640 | * so the return value not includes other pending runtime that other | |
3641 | * running tasks might have. | |
3642 | */ | |
3643 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3644 | { | |
3645 | struct task_cputime totals; | |
3646 | unsigned long flags; | |
3647 | struct rq *rq; | |
3648 | u64 ns; | |
3649 | ||
3650 | rq = task_rq_lock(p, &flags); | |
3651 | thread_group_cputime(p, &totals); | |
3652 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3653 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 3654 | |
1da177e4 LT |
3655 | return ns; |
3656 | } | |
3657 | ||
1da177e4 LT |
3658 | /* |
3659 | * Account user cpu time to a process. | |
3660 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3661 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3662 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3663 | */ |
457533a7 MS |
3664 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3665 | cputime_t cputime_scaled) | |
1da177e4 LT |
3666 | { |
3667 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3668 | cputime64_t tmp; | |
3669 | ||
457533a7 | 3670 | /* Add user time to process. */ |
1da177e4 | 3671 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3672 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3673 | account_group_user_time(p, cputime); |
1da177e4 LT |
3674 | |
3675 | /* Add user time to cpustat. */ | |
3676 | tmp = cputime_to_cputime64(cputime); | |
3677 | if (TASK_NICE(p) > 0) | |
3678 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3679 | else | |
3680 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3681 | |
3682 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3683 | /* Account for user time used */ |
3684 | acct_update_integrals(p); | |
1da177e4 LT |
3685 | } |
3686 | ||
94886b84 LV |
3687 | /* |
3688 | * Account guest cpu time to a process. | |
3689 | * @p: the process that the cpu time gets accounted to | |
3690 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3691 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3692 | */ |
457533a7 MS |
3693 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3694 | cputime_t cputime_scaled) | |
94886b84 LV |
3695 | { |
3696 | cputime64_t tmp; | |
3697 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3698 | ||
3699 | tmp = cputime_to_cputime64(cputime); | |
3700 | ||
457533a7 | 3701 | /* Add guest time to process. */ |
94886b84 | 3702 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3703 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3704 | account_group_user_time(p, cputime); |
94886b84 LV |
3705 | p->gtime = cputime_add(p->gtime, cputime); |
3706 | ||
457533a7 | 3707 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3708 | if (TASK_NICE(p) > 0) { |
3709 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3710 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3711 | } else { | |
3712 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3713 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3714 | } | |
94886b84 LV |
3715 | } |
3716 | ||
70a89a66 VP |
3717 | /* |
3718 | * Account system cpu time to a process and desired cpustat field | |
3719 | * @p: the process that the cpu time gets accounted to | |
3720 | * @cputime: the cpu time spent in kernel space since the last update | |
3721 | * @cputime_scaled: cputime scaled by cpu frequency | |
3722 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3723 | */ | |
3724 | static inline | |
3725 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3726 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3727 | { | |
3728 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3729 | ||
3730 | /* Add system time to process. */ | |
3731 | p->stime = cputime_add(p->stime, cputime); | |
3732 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3733 | account_group_system_time(p, cputime); | |
3734 | ||
3735 | /* Add system time to cpustat. */ | |
3736 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3737 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3738 | ||
3739 | /* Account for system time used */ | |
3740 | acct_update_integrals(p); | |
3741 | } | |
3742 | ||
1da177e4 LT |
3743 | /* |
3744 | * Account system cpu time to a process. | |
3745 | * @p: the process that the cpu time gets accounted to | |
3746 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3747 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3748 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3749 | */ |
3750 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3751 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3752 | { |
3753 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3754 | cputime64_t *target_cputime64; |
1da177e4 | 3755 | |
983ed7a6 | 3756 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3757 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3758 | return; |
3759 | } | |
94886b84 | 3760 | |
1da177e4 | 3761 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3762 | target_cputime64 = &cpustat->irq; |
75e1056f | 3763 | else if (in_serving_softirq()) |
70a89a66 | 3764 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3765 | else |
70a89a66 | 3766 | target_cputime64 = &cpustat->system; |
ef12fefa | 3767 | |
70a89a66 | 3768 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3769 | } |
3770 | ||
c66f08be | 3771 | /* |
1da177e4 | 3772 | * Account for involuntary wait time. |
544b4a1f | 3773 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3774 | */ |
79741dd3 | 3775 | void account_steal_time(cputime_t cputime) |
c66f08be | 3776 | { |
79741dd3 MS |
3777 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3778 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3779 | ||
3780 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3781 | } |
3782 | ||
1da177e4 | 3783 | /* |
79741dd3 MS |
3784 | * Account for idle time. |
3785 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3786 | */ |
79741dd3 | 3787 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3788 | { |
3789 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3790 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3791 | struct rq *rq = this_rq(); |
1da177e4 | 3792 | |
79741dd3 MS |
3793 | if (atomic_read(&rq->nr_iowait) > 0) |
3794 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3795 | else | |
3796 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3797 | } |
3798 | ||
79741dd3 MS |
3799 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3800 | ||
abb74cef VP |
3801 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3802 | /* | |
3803 | * Account a tick to a process and cpustat | |
3804 | * @p: the process that the cpu time gets accounted to | |
3805 | * @user_tick: is the tick from userspace | |
3806 | * @rq: the pointer to rq | |
3807 | * | |
3808 | * Tick demultiplexing follows the order | |
3809 | * - pending hardirq update | |
3810 | * - pending softirq update | |
3811 | * - user_time | |
3812 | * - idle_time | |
3813 | * - system time | |
3814 | * - check for guest_time | |
3815 | * - else account as system_time | |
3816 | * | |
3817 | * Check for hardirq is done both for system and user time as there is | |
3818 | * no timer going off while we are on hardirq and hence we may never get an | |
3819 | * opportunity to update it solely in system time. | |
3820 | * p->stime and friends are only updated on system time and not on irq | |
3821 | * softirq as those do not count in task exec_runtime any more. | |
3822 | */ | |
3823 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3824 | struct rq *rq) | |
3825 | { | |
3826 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3827 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3828 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3829 | ||
3830 | if (irqtime_account_hi_update()) { | |
3831 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3832 | } else if (irqtime_account_si_update()) { | |
3833 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3834 | } else if (this_cpu_ksoftirqd() == p) { |
3835 | /* | |
3836 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3837 | * So, we have to handle it separately here. | |
3838 | * Also, p->stime needs to be updated for ksoftirqd. | |
3839 | */ | |
3840 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3841 | &cpustat->softirq); | |
abb74cef VP |
3842 | } else if (user_tick) { |
3843 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3844 | } else if (p == rq->idle) { | |
3845 | account_idle_time(cputime_one_jiffy); | |
3846 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3847 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3848 | } else { | |
3849 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3850 | &cpustat->system); | |
3851 | } | |
3852 | } | |
3853 | ||
3854 | static void irqtime_account_idle_ticks(int ticks) | |
3855 | { | |
3856 | int i; | |
3857 | struct rq *rq = this_rq(); | |
3858 | ||
3859 | for (i = 0; i < ticks; i++) | |
3860 | irqtime_account_process_tick(current, 0, rq); | |
3861 | } | |
544b4a1f | 3862 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3863 | static void irqtime_account_idle_ticks(int ticks) {} |
3864 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3865 | struct rq *rq) {} | |
544b4a1f | 3866 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3867 | |
3868 | /* | |
3869 | * Account a single tick of cpu time. | |
3870 | * @p: the process that the cpu time gets accounted to | |
3871 | * @user_tick: indicates if the tick is a user or a system tick | |
3872 | */ | |
3873 | void account_process_tick(struct task_struct *p, int user_tick) | |
3874 | { | |
a42548a1 | 3875 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3876 | struct rq *rq = this_rq(); |
3877 | ||
abb74cef VP |
3878 | if (sched_clock_irqtime) { |
3879 | irqtime_account_process_tick(p, user_tick, rq); | |
3880 | return; | |
3881 | } | |
3882 | ||
79741dd3 | 3883 | if (user_tick) |
a42548a1 | 3884 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3885 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3886 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3887 | one_jiffy_scaled); |
3888 | else | |
a42548a1 | 3889 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3890 | } |
3891 | ||
3892 | /* | |
3893 | * Account multiple ticks of steal time. | |
3894 | * @p: the process from which the cpu time has been stolen | |
3895 | * @ticks: number of stolen ticks | |
3896 | */ | |
3897 | void account_steal_ticks(unsigned long ticks) | |
3898 | { | |
3899 | account_steal_time(jiffies_to_cputime(ticks)); | |
3900 | } | |
3901 | ||
3902 | /* | |
3903 | * Account multiple ticks of idle time. | |
3904 | * @ticks: number of stolen ticks | |
3905 | */ | |
3906 | void account_idle_ticks(unsigned long ticks) | |
3907 | { | |
abb74cef VP |
3908 | |
3909 | if (sched_clock_irqtime) { | |
3910 | irqtime_account_idle_ticks(ticks); | |
3911 | return; | |
3912 | } | |
3913 | ||
79741dd3 | 3914 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3915 | } |
3916 | ||
79741dd3 MS |
3917 | #endif |
3918 | ||
49048622 BS |
3919 | /* |
3920 | * Use precise platform statistics if available: | |
3921 | */ | |
3922 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3923 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3924 | { |
d99ca3b9 HS |
3925 | *ut = p->utime; |
3926 | *st = p->stime; | |
49048622 BS |
3927 | } |
3928 | ||
0cf55e1e | 3929 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3930 | { |
0cf55e1e HS |
3931 | struct task_cputime cputime; |
3932 | ||
3933 | thread_group_cputime(p, &cputime); | |
3934 | ||
3935 | *ut = cputime.utime; | |
3936 | *st = cputime.stime; | |
49048622 BS |
3937 | } |
3938 | #else | |
761b1d26 HS |
3939 | |
3940 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3941 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3942 | #endif |
3943 | ||
d180c5bc | 3944 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3945 | { |
d99ca3b9 | 3946 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3947 | |
3948 | /* | |
3949 | * Use CFS's precise accounting: | |
3950 | */ | |
d180c5bc | 3951 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3952 | |
3953 | if (total) { | |
e75e863d | 3954 | u64 temp = rtime; |
d180c5bc | 3955 | |
e75e863d | 3956 | temp *= utime; |
49048622 | 3957 | do_div(temp, total); |
d180c5bc HS |
3958 | utime = (cputime_t)temp; |
3959 | } else | |
3960 | utime = rtime; | |
49048622 | 3961 | |
d180c5bc HS |
3962 | /* |
3963 | * Compare with previous values, to keep monotonicity: | |
3964 | */ | |
761b1d26 | 3965 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3966 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3967 | |
d99ca3b9 HS |
3968 | *ut = p->prev_utime; |
3969 | *st = p->prev_stime; | |
49048622 BS |
3970 | } |
3971 | ||
0cf55e1e HS |
3972 | /* |
3973 | * Must be called with siglock held. | |
3974 | */ | |
3975 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3976 | { |
0cf55e1e HS |
3977 | struct signal_struct *sig = p->signal; |
3978 | struct task_cputime cputime; | |
3979 | cputime_t rtime, utime, total; | |
49048622 | 3980 | |
0cf55e1e | 3981 | thread_group_cputime(p, &cputime); |
49048622 | 3982 | |
0cf55e1e HS |
3983 | total = cputime_add(cputime.utime, cputime.stime); |
3984 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3985 | |
0cf55e1e | 3986 | if (total) { |
e75e863d | 3987 | u64 temp = rtime; |
49048622 | 3988 | |
e75e863d | 3989 | temp *= cputime.utime; |
0cf55e1e HS |
3990 | do_div(temp, total); |
3991 | utime = (cputime_t)temp; | |
3992 | } else | |
3993 | utime = rtime; | |
3994 | ||
3995 | sig->prev_utime = max(sig->prev_utime, utime); | |
3996 | sig->prev_stime = max(sig->prev_stime, | |
3997 | cputime_sub(rtime, sig->prev_utime)); | |
3998 | ||
3999 | *ut = sig->prev_utime; | |
4000 | *st = sig->prev_stime; | |
49048622 | 4001 | } |
49048622 | 4002 | #endif |
49048622 | 4003 | |
7835b98b CL |
4004 | /* |
4005 | * This function gets called by the timer code, with HZ frequency. | |
4006 | * We call it with interrupts disabled. | |
7835b98b CL |
4007 | */ |
4008 | void scheduler_tick(void) | |
4009 | { | |
7835b98b CL |
4010 | int cpu = smp_processor_id(); |
4011 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4012 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4013 | |
4014 | sched_clock_tick(); | |
dd41f596 | 4015 | |
05fa785c | 4016 | raw_spin_lock(&rq->lock); |
3e51f33f | 4017 | update_rq_clock(rq); |
fdf3e95d | 4018 | update_cpu_load_active(rq); |
fa85ae24 | 4019 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 4020 | raw_spin_unlock(&rq->lock); |
7835b98b | 4021 | |
e9d2b064 | 4022 | perf_event_task_tick(); |
e220d2dc | 4023 | |
e418e1c2 | 4024 | #ifdef CONFIG_SMP |
dd41f596 IM |
4025 | rq->idle_at_tick = idle_cpu(cpu); |
4026 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4027 | #endif |
1da177e4 LT |
4028 | } |
4029 | ||
132380a0 | 4030 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4031 | { |
4032 | if (in_lock_functions(addr)) { | |
4033 | addr = CALLER_ADDR2; | |
4034 | if (in_lock_functions(addr)) | |
4035 | addr = CALLER_ADDR3; | |
4036 | } | |
4037 | return addr; | |
4038 | } | |
1da177e4 | 4039 | |
7e49fcce SR |
4040 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4041 | defined(CONFIG_PREEMPT_TRACER)) | |
4042 | ||
43627582 | 4043 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4044 | { |
6cd8a4bb | 4045 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4046 | /* |
4047 | * Underflow? | |
4048 | */ | |
9a11b49a IM |
4049 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4050 | return; | |
6cd8a4bb | 4051 | #endif |
1da177e4 | 4052 | preempt_count() += val; |
6cd8a4bb | 4053 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4054 | /* |
4055 | * Spinlock count overflowing soon? | |
4056 | */ | |
33859f7f MOS |
4057 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4058 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4059 | #endif |
4060 | if (preempt_count() == val) | |
4061 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4062 | } |
4063 | EXPORT_SYMBOL(add_preempt_count); | |
4064 | ||
43627582 | 4065 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4066 | { |
6cd8a4bb | 4067 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4068 | /* |
4069 | * Underflow? | |
4070 | */ | |
01e3eb82 | 4071 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4072 | return; |
1da177e4 LT |
4073 | /* |
4074 | * Is the spinlock portion underflowing? | |
4075 | */ | |
9a11b49a IM |
4076 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4077 | !(preempt_count() & PREEMPT_MASK))) | |
4078 | return; | |
6cd8a4bb | 4079 | #endif |
9a11b49a | 4080 | |
6cd8a4bb SR |
4081 | if (preempt_count() == val) |
4082 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4083 | preempt_count() -= val; |
4084 | } | |
4085 | EXPORT_SYMBOL(sub_preempt_count); | |
4086 | ||
4087 | #endif | |
4088 | ||
4089 | /* | |
dd41f596 | 4090 | * Print scheduling while atomic bug: |
1da177e4 | 4091 | */ |
dd41f596 | 4092 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4093 | { |
838225b4 SS |
4094 | struct pt_regs *regs = get_irq_regs(); |
4095 | ||
3df0fc5b PZ |
4096 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4097 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4098 | |
dd41f596 | 4099 | debug_show_held_locks(prev); |
e21f5b15 | 4100 | print_modules(); |
dd41f596 IM |
4101 | if (irqs_disabled()) |
4102 | print_irqtrace_events(prev); | |
838225b4 SS |
4103 | |
4104 | if (regs) | |
4105 | show_regs(regs); | |
4106 | else | |
4107 | dump_stack(); | |
dd41f596 | 4108 | } |
1da177e4 | 4109 | |
dd41f596 IM |
4110 | /* |
4111 | * Various schedule()-time debugging checks and statistics: | |
4112 | */ | |
4113 | static inline void schedule_debug(struct task_struct *prev) | |
4114 | { | |
1da177e4 | 4115 | /* |
41a2d6cf | 4116 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4117 | * schedule() atomically, we ignore that path for now. |
4118 | * Otherwise, whine if we are scheduling when we should not be. | |
4119 | */ | |
3f33a7ce | 4120 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4121 | __schedule_bug(prev); |
4122 | ||
1da177e4 LT |
4123 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4124 | ||
2d72376b | 4125 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
4126 | } |
4127 | ||
6cecd084 | 4128 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4129 | { |
61eadef6 | 4130 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 4131 | update_rq_clock(rq); |
6cecd084 | 4132 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4133 | } |
4134 | ||
dd41f596 IM |
4135 | /* |
4136 | * Pick up the highest-prio task: | |
4137 | */ | |
4138 | static inline struct task_struct * | |
b67802ea | 4139 | pick_next_task(struct rq *rq) |
dd41f596 | 4140 | { |
5522d5d5 | 4141 | const struct sched_class *class; |
dd41f596 | 4142 | struct task_struct *p; |
1da177e4 LT |
4143 | |
4144 | /* | |
dd41f596 IM |
4145 | * Optimization: we know that if all tasks are in |
4146 | * the fair class we can call that function directly: | |
1da177e4 | 4147 | */ |
dd41f596 | 4148 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4149 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4150 | if (likely(p)) |
4151 | return p; | |
1da177e4 LT |
4152 | } |
4153 | ||
34f971f6 | 4154 | for_each_class(class) { |
fb8d4724 | 4155 | p = class->pick_next_task(rq); |
dd41f596 IM |
4156 | if (p) |
4157 | return p; | |
dd41f596 | 4158 | } |
34f971f6 PZ |
4159 | |
4160 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4161 | } |
1da177e4 | 4162 | |
dd41f596 IM |
4163 | /* |
4164 | * schedule() is the main scheduler function. | |
4165 | */ | |
ff743345 | 4166 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4167 | { |
4168 | struct task_struct *prev, *next; | |
67ca7bde | 4169 | unsigned long *switch_count; |
dd41f596 | 4170 | struct rq *rq; |
31656519 | 4171 | int cpu; |
dd41f596 | 4172 | |
ff743345 PZ |
4173 | need_resched: |
4174 | preempt_disable(); | |
dd41f596 IM |
4175 | cpu = smp_processor_id(); |
4176 | rq = cpu_rq(cpu); | |
25502a6c | 4177 | rcu_note_context_switch(cpu); |
dd41f596 | 4178 | prev = rq->curr; |
dd41f596 | 4179 | |
dd41f596 | 4180 | schedule_debug(prev); |
1da177e4 | 4181 | |
31656519 | 4182 | if (sched_feat(HRTICK)) |
f333fdc9 | 4183 | hrtick_clear(rq); |
8f4d37ec | 4184 | |
05fa785c | 4185 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4186 | |
246d86b5 | 4187 | switch_count = &prev->nivcsw; |
1da177e4 | 4188 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4189 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4190 | prev->state = TASK_RUNNING; |
21aa9af0 | 4191 | } else { |
2acca55e PZ |
4192 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
4193 | prev->on_rq = 0; | |
4194 | ||
21aa9af0 | 4195 | /* |
2acca55e PZ |
4196 | * If a worker went to sleep, notify and ask workqueue |
4197 | * whether it wants to wake up a task to maintain | |
4198 | * concurrency. | |
21aa9af0 TH |
4199 | */ |
4200 | if (prev->flags & PF_WQ_WORKER) { | |
4201 | struct task_struct *to_wakeup; | |
4202 | ||
4203 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4204 | if (to_wakeup) | |
4205 | try_to_wake_up_local(to_wakeup); | |
4206 | } | |
fd2f4419 | 4207 | |
6631e635 | 4208 | /* |
2acca55e PZ |
4209 | * If we are going to sleep and we have plugged IO |
4210 | * queued, make sure to submit it to avoid deadlocks. | |
6631e635 LT |
4211 | */ |
4212 | if (blk_needs_flush_plug(prev)) { | |
4213 | raw_spin_unlock(&rq->lock); | |
a237c1c5 | 4214 | blk_schedule_flush_plug(prev); |
6631e635 LT |
4215 | raw_spin_lock(&rq->lock); |
4216 | } | |
21aa9af0 | 4217 | } |
dd41f596 | 4218 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4219 | } |
4220 | ||
3f029d3c | 4221 | pre_schedule(rq, prev); |
f65eda4f | 4222 | |
dd41f596 | 4223 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4224 | idle_balance(cpu, rq); |
1da177e4 | 4225 | |
df1c99d4 | 4226 | put_prev_task(rq, prev); |
b67802ea | 4227 | next = pick_next_task(rq); |
f26f9aff MG |
4228 | clear_tsk_need_resched(prev); |
4229 | rq->skip_clock_update = 0; | |
1da177e4 | 4230 | |
1da177e4 | 4231 | if (likely(prev != next)) { |
1da177e4 LT |
4232 | rq->nr_switches++; |
4233 | rq->curr = next; | |
4234 | ++*switch_count; | |
4235 | ||
dd41f596 | 4236 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4237 | /* |
246d86b5 ON |
4238 | * The context switch have flipped the stack from under us |
4239 | * and restored the local variables which were saved when | |
4240 | * this task called schedule() in the past. prev == current | |
4241 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4242 | */ |
4243 | cpu = smp_processor_id(); | |
4244 | rq = cpu_rq(cpu); | |
1da177e4 | 4245 | } else |
05fa785c | 4246 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4247 | |
3f029d3c | 4248 | post_schedule(rq); |
1da177e4 | 4249 | |
1da177e4 | 4250 | preempt_enable_no_resched(); |
ff743345 | 4251 | if (need_resched()) |
1da177e4 LT |
4252 | goto need_resched; |
4253 | } | |
1da177e4 LT |
4254 | EXPORT_SYMBOL(schedule); |
4255 | ||
c08f7829 | 4256 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4257 | |
c6eb3dda PZ |
4258 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4259 | { | |
4260 | bool ret = false; | |
0d66bf6d | 4261 | |
c6eb3dda PZ |
4262 | rcu_read_lock(); |
4263 | if (lock->owner != owner) | |
4264 | goto fail; | |
0d66bf6d PZ |
4265 | |
4266 | /* | |
c6eb3dda PZ |
4267 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4268 | * lock->owner still matches owner, if that fails, owner might | |
4269 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4270 | * ensures the memory stays valid. | |
0d66bf6d | 4271 | */ |
c6eb3dda | 4272 | barrier(); |
0d66bf6d | 4273 | |
c6eb3dda PZ |
4274 | ret = owner->on_cpu; |
4275 | fail: | |
4276 | rcu_read_unlock(); | |
0d66bf6d | 4277 | |
c6eb3dda PZ |
4278 | return ret; |
4279 | } | |
0d66bf6d | 4280 | |
c6eb3dda PZ |
4281 | /* |
4282 | * Look out! "owner" is an entirely speculative pointer | |
4283 | * access and not reliable. | |
4284 | */ | |
4285 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4286 | { | |
4287 | if (!sched_feat(OWNER_SPIN)) | |
4288 | return 0; | |
0d66bf6d | 4289 | |
c6eb3dda PZ |
4290 | while (owner_running(lock, owner)) { |
4291 | if (need_resched()) | |
0d66bf6d PZ |
4292 | return 0; |
4293 | ||
335d7afb | 4294 | arch_mutex_cpu_relax(); |
0d66bf6d | 4295 | } |
4b402210 | 4296 | |
c6eb3dda PZ |
4297 | /* |
4298 | * If the owner changed to another task there is likely | |
4299 | * heavy contention, stop spinning. | |
4300 | */ | |
4301 | if (lock->owner) | |
4302 | return 0; | |
4303 | ||
0d66bf6d PZ |
4304 | return 1; |
4305 | } | |
4306 | #endif | |
4307 | ||
1da177e4 LT |
4308 | #ifdef CONFIG_PREEMPT |
4309 | /* | |
2ed6e34f | 4310 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4311 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4312 | * occur there and call schedule directly. |
4313 | */ | |
d1f74e20 | 4314 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4315 | { |
4316 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4317 | |
1da177e4 LT |
4318 | /* |
4319 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4320 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4321 | */ |
beed33a8 | 4322 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4323 | return; |
4324 | ||
3a5c359a | 4325 | do { |
d1f74e20 | 4326 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4327 | schedule(); |
d1f74e20 | 4328 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4329 | |
3a5c359a AK |
4330 | /* |
4331 | * Check again in case we missed a preemption opportunity | |
4332 | * between schedule and now. | |
4333 | */ | |
4334 | barrier(); | |
5ed0cec0 | 4335 | } while (need_resched()); |
1da177e4 | 4336 | } |
1da177e4 LT |
4337 | EXPORT_SYMBOL(preempt_schedule); |
4338 | ||
4339 | /* | |
2ed6e34f | 4340 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4341 | * off of irq context. |
4342 | * Note, that this is called and return with irqs disabled. This will | |
4343 | * protect us against recursive calling from irq. | |
4344 | */ | |
4345 | asmlinkage void __sched preempt_schedule_irq(void) | |
4346 | { | |
4347 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4348 | |
2ed6e34f | 4349 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4350 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4351 | ||
3a5c359a AK |
4352 | do { |
4353 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4354 | local_irq_enable(); |
4355 | schedule(); | |
4356 | local_irq_disable(); | |
3a5c359a | 4357 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4358 | |
3a5c359a AK |
4359 | /* |
4360 | * Check again in case we missed a preemption opportunity | |
4361 | * between schedule and now. | |
4362 | */ | |
4363 | barrier(); | |
5ed0cec0 | 4364 | } while (need_resched()); |
1da177e4 LT |
4365 | } |
4366 | ||
4367 | #endif /* CONFIG_PREEMPT */ | |
4368 | ||
63859d4f | 4369 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4370 | void *key) |
1da177e4 | 4371 | { |
63859d4f | 4372 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4373 | } |
1da177e4 LT |
4374 | EXPORT_SYMBOL(default_wake_function); |
4375 | ||
4376 | /* | |
41a2d6cf IM |
4377 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4378 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4379 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4380 | * | |
4381 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4382 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4383 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4384 | */ | |
78ddb08f | 4385 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4386 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4387 | { |
2e45874c | 4388 | wait_queue_t *curr, *next; |
1da177e4 | 4389 | |
2e45874c | 4390 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4391 | unsigned flags = curr->flags; |
4392 | ||
63859d4f | 4393 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4394 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4395 | break; |
4396 | } | |
4397 | } | |
4398 | ||
4399 | /** | |
4400 | * __wake_up - wake up threads blocked on a waitqueue. | |
4401 | * @q: the waitqueue | |
4402 | * @mode: which threads | |
4403 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4404 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4405 | * |
4406 | * It may be assumed that this function implies a write memory barrier before | |
4407 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4408 | */ |
7ad5b3a5 | 4409 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4410 | int nr_exclusive, void *key) |
1da177e4 LT |
4411 | { |
4412 | unsigned long flags; | |
4413 | ||
4414 | spin_lock_irqsave(&q->lock, flags); | |
4415 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4416 | spin_unlock_irqrestore(&q->lock, flags); | |
4417 | } | |
1da177e4 LT |
4418 | EXPORT_SYMBOL(__wake_up); |
4419 | ||
4420 | /* | |
4421 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4422 | */ | |
7ad5b3a5 | 4423 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4424 | { |
4425 | __wake_up_common(q, mode, 1, 0, NULL); | |
4426 | } | |
22c43c81 | 4427 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4428 | |
4ede816a DL |
4429 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4430 | { | |
4431 | __wake_up_common(q, mode, 1, 0, key); | |
4432 | } | |
bf294b41 | 4433 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4434 | |
1da177e4 | 4435 | /** |
4ede816a | 4436 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4437 | * @q: the waitqueue |
4438 | * @mode: which threads | |
4439 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4440 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4441 | * |
4442 | * The sync wakeup differs that the waker knows that it will schedule | |
4443 | * away soon, so while the target thread will be woken up, it will not | |
4444 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4445 | * with each other. This can prevent needless bouncing between CPUs. | |
4446 | * | |
4447 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4448 | * |
4449 | * It may be assumed that this function implies a write memory barrier before | |
4450 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4451 | */ |
4ede816a DL |
4452 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4453 | int nr_exclusive, void *key) | |
1da177e4 LT |
4454 | { |
4455 | unsigned long flags; | |
7d478721 | 4456 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4457 | |
4458 | if (unlikely(!q)) | |
4459 | return; | |
4460 | ||
4461 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4462 | wake_flags = 0; |
1da177e4 LT |
4463 | |
4464 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4465 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4466 | spin_unlock_irqrestore(&q->lock, flags); |
4467 | } | |
4ede816a DL |
4468 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4469 | ||
4470 | /* | |
4471 | * __wake_up_sync - see __wake_up_sync_key() | |
4472 | */ | |
4473 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4474 | { | |
4475 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4476 | } | |
1da177e4 LT |
4477 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4478 | ||
65eb3dc6 KD |
4479 | /** |
4480 | * complete: - signals a single thread waiting on this completion | |
4481 | * @x: holds the state of this particular completion | |
4482 | * | |
4483 | * This will wake up a single thread waiting on this completion. Threads will be | |
4484 | * awakened in the same order in which they were queued. | |
4485 | * | |
4486 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4487 | * |
4488 | * It may be assumed that this function implies a write memory barrier before | |
4489 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4490 | */ |
b15136e9 | 4491 | void complete(struct completion *x) |
1da177e4 LT |
4492 | { |
4493 | unsigned long flags; | |
4494 | ||
4495 | spin_lock_irqsave(&x->wait.lock, flags); | |
4496 | x->done++; | |
d9514f6c | 4497 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4498 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4499 | } | |
4500 | EXPORT_SYMBOL(complete); | |
4501 | ||
65eb3dc6 KD |
4502 | /** |
4503 | * complete_all: - signals all threads waiting on this completion | |
4504 | * @x: holds the state of this particular completion | |
4505 | * | |
4506 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4507 | * |
4508 | * It may be assumed that this function implies a write memory barrier before | |
4509 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4510 | */ |
b15136e9 | 4511 | void complete_all(struct completion *x) |
1da177e4 LT |
4512 | { |
4513 | unsigned long flags; | |
4514 | ||
4515 | spin_lock_irqsave(&x->wait.lock, flags); | |
4516 | x->done += UINT_MAX/2; | |
d9514f6c | 4517 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4518 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4519 | } | |
4520 | EXPORT_SYMBOL(complete_all); | |
4521 | ||
8cbbe86d AK |
4522 | static inline long __sched |
4523 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4524 | { |
1da177e4 LT |
4525 | if (!x->done) { |
4526 | DECLARE_WAITQUEUE(wait, current); | |
4527 | ||
a93d2f17 | 4528 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4529 | do { |
94d3d824 | 4530 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4531 | timeout = -ERESTARTSYS; |
4532 | break; | |
8cbbe86d AK |
4533 | } |
4534 | __set_current_state(state); | |
1da177e4 LT |
4535 | spin_unlock_irq(&x->wait.lock); |
4536 | timeout = schedule_timeout(timeout); | |
4537 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4538 | } while (!x->done && timeout); |
1da177e4 | 4539 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4540 | if (!x->done) |
4541 | return timeout; | |
1da177e4 LT |
4542 | } |
4543 | x->done--; | |
ea71a546 | 4544 | return timeout ?: 1; |
1da177e4 | 4545 | } |
1da177e4 | 4546 | |
8cbbe86d AK |
4547 | static long __sched |
4548 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4549 | { |
1da177e4 LT |
4550 | might_sleep(); |
4551 | ||
4552 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4553 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4554 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4555 | return timeout; |
4556 | } | |
1da177e4 | 4557 | |
65eb3dc6 KD |
4558 | /** |
4559 | * wait_for_completion: - waits for completion of a task | |
4560 | * @x: holds the state of this particular completion | |
4561 | * | |
4562 | * This waits to be signaled for completion of a specific task. It is NOT | |
4563 | * interruptible and there is no timeout. | |
4564 | * | |
4565 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4566 | * and interrupt capability. Also see complete(). | |
4567 | */ | |
b15136e9 | 4568 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4569 | { |
4570 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4571 | } |
8cbbe86d | 4572 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4573 | |
65eb3dc6 KD |
4574 | /** |
4575 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4576 | * @x: holds the state of this particular completion | |
4577 | * @timeout: timeout value in jiffies | |
4578 | * | |
4579 | * This waits for either a completion of a specific task to be signaled or for a | |
4580 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4581 | * interruptible. | |
4582 | */ | |
b15136e9 | 4583 | unsigned long __sched |
8cbbe86d | 4584 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4585 | { |
8cbbe86d | 4586 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4587 | } |
8cbbe86d | 4588 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4589 | |
65eb3dc6 KD |
4590 | /** |
4591 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4592 | * @x: holds the state of this particular completion | |
4593 | * | |
4594 | * This waits for completion of a specific task to be signaled. It is | |
4595 | * interruptible. | |
4596 | */ | |
8cbbe86d | 4597 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4598 | { |
51e97990 AK |
4599 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4600 | if (t == -ERESTARTSYS) | |
4601 | return t; | |
4602 | return 0; | |
0fec171c | 4603 | } |
8cbbe86d | 4604 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4605 | |
65eb3dc6 KD |
4606 | /** |
4607 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4608 | * @x: holds the state of this particular completion | |
4609 | * @timeout: timeout value in jiffies | |
4610 | * | |
4611 | * This waits for either a completion of a specific task to be signaled or for a | |
4612 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4613 | */ | |
6bf41237 | 4614 | long __sched |
8cbbe86d AK |
4615 | wait_for_completion_interruptible_timeout(struct completion *x, |
4616 | unsigned long timeout) | |
0fec171c | 4617 | { |
8cbbe86d | 4618 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4619 | } |
8cbbe86d | 4620 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4621 | |
65eb3dc6 KD |
4622 | /** |
4623 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4624 | * @x: holds the state of this particular completion | |
4625 | * | |
4626 | * This waits to be signaled for completion of a specific task. It can be | |
4627 | * interrupted by a kill signal. | |
4628 | */ | |
009e577e MW |
4629 | int __sched wait_for_completion_killable(struct completion *x) |
4630 | { | |
4631 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4632 | if (t == -ERESTARTSYS) | |
4633 | return t; | |
4634 | return 0; | |
4635 | } | |
4636 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4637 | ||
0aa12fb4 SW |
4638 | /** |
4639 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4640 | * @x: holds the state of this particular completion | |
4641 | * @timeout: timeout value in jiffies | |
4642 | * | |
4643 | * This waits for either a completion of a specific task to be | |
4644 | * signaled or for a specified timeout to expire. It can be | |
4645 | * interrupted by a kill signal. The timeout is in jiffies. | |
4646 | */ | |
6bf41237 | 4647 | long __sched |
0aa12fb4 SW |
4648 | wait_for_completion_killable_timeout(struct completion *x, |
4649 | unsigned long timeout) | |
4650 | { | |
4651 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4652 | } | |
4653 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4654 | ||
be4de352 DC |
4655 | /** |
4656 | * try_wait_for_completion - try to decrement a completion without blocking | |
4657 | * @x: completion structure | |
4658 | * | |
4659 | * Returns: 0 if a decrement cannot be done without blocking | |
4660 | * 1 if a decrement succeeded. | |
4661 | * | |
4662 | * If a completion is being used as a counting completion, | |
4663 | * attempt to decrement the counter without blocking. This | |
4664 | * enables us to avoid waiting if the resource the completion | |
4665 | * is protecting is not available. | |
4666 | */ | |
4667 | bool try_wait_for_completion(struct completion *x) | |
4668 | { | |
7539a3b3 | 4669 | unsigned long flags; |
be4de352 DC |
4670 | int ret = 1; |
4671 | ||
7539a3b3 | 4672 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4673 | if (!x->done) |
4674 | ret = 0; | |
4675 | else | |
4676 | x->done--; | |
7539a3b3 | 4677 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4678 | return ret; |
4679 | } | |
4680 | EXPORT_SYMBOL(try_wait_for_completion); | |
4681 | ||
4682 | /** | |
4683 | * completion_done - Test to see if a completion has any waiters | |
4684 | * @x: completion structure | |
4685 | * | |
4686 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4687 | * 1 if there are no waiters. | |
4688 | * | |
4689 | */ | |
4690 | bool completion_done(struct completion *x) | |
4691 | { | |
7539a3b3 | 4692 | unsigned long flags; |
be4de352 DC |
4693 | int ret = 1; |
4694 | ||
7539a3b3 | 4695 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4696 | if (!x->done) |
4697 | ret = 0; | |
7539a3b3 | 4698 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4699 | return ret; |
4700 | } | |
4701 | EXPORT_SYMBOL(completion_done); | |
4702 | ||
8cbbe86d AK |
4703 | static long __sched |
4704 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4705 | { |
0fec171c IM |
4706 | unsigned long flags; |
4707 | wait_queue_t wait; | |
4708 | ||
4709 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4710 | |
8cbbe86d | 4711 | __set_current_state(state); |
1da177e4 | 4712 | |
8cbbe86d AK |
4713 | spin_lock_irqsave(&q->lock, flags); |
4714 | __add_wait_queue(q, &wait); | |
4715 | spin_unlock(&q->lock); | |
4716 | timeout = schedule_timeout(timeout); | |
4717 | spin_lock_irq(&q->lock); | |
4718 | __remove_wait_queue(q, &wait); | |
4719 | spin_unlock_irqrestore(&q->lock, flags); | |
4720 | ||
4721 | return timeout; | |
4722 | } | |
4723 | ||
4724 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4725 | { | |
4726 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4727 | } |
1da177e4 LT |
4728 | EXPORT_SYMBOL(interruptible_sleep_on); |
4729 | ||
0fec171c | 4730 | long __sched |
95cdf3b7 | 4731 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4732 | { |
8cbbe86d | 4733 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4734 | } |
1da177e4 LT |
4735 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4736 | ||
0fec171c | 4737 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4738 | { |
8cbbe86d | 4739 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4740 | } |
1da177e4 LT |
4741 | EXPORT_SYMBOL(sleep_on); |
4742 | ||
0fec171c | 4743 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4744 | { |
8cbbe86d | 4745 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4746 | } |
1da177e4 LT |
4747 | EXPORT_SYMBOL(sleep_on_timeout); |
4748 | ||
b29739f9 IM |
4749 | #ifdef CONFIG_RT_MUTEXES |
4750 | ||
4751 | /* | |
4752 | * rt_mutex_setprio - set the current priority of a task | |
4753 | * @p: task | |
4754 | * @prio: prio value (kernel-internal form) | |
4755 | * | |
4756 | * This function changes the 'effective' priority of a task. It does | |
4757 | * not touch ->normal_prio like __setscheduler(). | |
4758 | * | |
4759 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4760 | */ | |
36c8b586 | 4761 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4762 | { |
83b699ed | 4763 | int oldprio, on_rq, running; |
70b97a7f | 4764 | struct rq *rq; |
83ab0aa0 | 4765 | const struct sched_class *prev_class; |
b29739f9 IM |
4766 | |
4767 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4768 | ||
0122ec5b | 4769 | rq = __task_rq_lock(p); |
b29739f9 | 4770 | |
a8027073 | 4771 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4772 | oldprio = p->prio; |
83ab0aa0 | 4773 | prev_class = p->sched_class; |
fd2f4419 | 4774 | on_rq = p->on_rq; |
051a1d1a | 4775 | running = task_current(rq, p); |
0e1f3483 | 4776 | if (on_rq) |
69be72c1 | 4777 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4778 | if (running) |
4779 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4780 | |
4781 | if (rt_prio(prio)) | |
4782 | p->sched_class = &rt_sched_class; | |
4783 | else | |
4784 | p->sched_class = &fair_sched_class; | |
4785 | ||
b29739f9 IM |
4786 | p->prio = prio; |
4787 | ||
0e1f3483 HS |
4788 | if (running) |
4789 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4790 | if (on_rq) |
371fd7e7 | 4791 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4792 | |
da7a735e | 4793 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4794 | __task_rq_unlock(rq); |
b29739f9 IM |
4795 | } |
4796 | ||
4797 | #endif | |
4798 | ||
36c8b586 | 4799 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4800 | { |
dd41f596 | 4801 | int old_prio, delta, on_rq; |
1da177e4 | 4802 | unsigned long flags; |
70b97a7f | 4803 | struct rq *rq; |
1da177e4 LT |
4804 | |
4805 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4806 | return; | |
4807 | /* | |
4808 | * We have to be careful, if called from sys_setpriority(), | |
4809 | * the task might be in the middle of scheduling on another CPU. | |
4810 | */ | |
4811 | rq = task_rq_lock(p, &flags); | |
4812 | /* | |
4813 | * The RT priorities are set via sched_setscheduler(), but we still | |
4814 | * allow the 'normal' nice value to be set - but as expected | |
4815 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4816 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4817 | */ |
e05606d3 | 4818 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4819 | p->static_prio = NICE_TO_PRIO(nice); |
4820 | goto out_unlock; | |
4821 | } | |
fd2f4419 | 4822 | on_rq = p->on_rq; |
c09595f6 | 4823 | if (on_rq) |
69be72c1 | 4824 | dequeue_task(rq, p, 0); |
1da177e4 | 4825 | |
1da177e4 | 4826 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4827 | set_load_weight(p); |
b29739f9 IM |
4828 | old_prio = p->prio; |
4829 | p->prio = effective_prio(p); | |
4830 | delta = p->prio - old_prio; | |
1da177e4 | 4831 | |
dd41f596 | 4832 | if (on_rq) { |
371fd7e7 | 4833 | enqueue_task(rq, p, 0); |
1da177e4 | 4834 | /* |
d5f9f942 AM |
4835 | * If the task increased its priority or is running and |
4836 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4837 | */ |
d5f9f942 | 4838 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4839 | resched_task(rq->curr); |
4840 | } | |
4841 | out_unlock: | |
0122ec5b | 4842 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 4843 | } |
1da177e4 LT |
4844 | EXPORT_SYMBOL(set_user_nice); |
4845 | ||
e43379f1 MM |
4846 | /* |
4847 | * can_nice - check if a task can reduce its nice value | |
4848 | * @p: task | |
4849 | * @nice: nice value | |
4850 | */ | |
36c8b586 | 4851 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4852 | { |
024f4747 MM |
4853 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4854 | int nice_rlim = 20 - nice; | |
48f24c4d | 4855 | |
78d7d407 | 4856 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4857 | capable(CAP_SYS_NICE)); |
4858 | } | |
4859 | ||
1da177e4 LT |
4860 | #ifdef __ARCH_WANT_SYS_NICE |
4861 | ||
4862 | /* | |
4863 | * sys_nice - change the priority of the current process. | |
4864 | * @increment: priority increment | |
4865 | * | |
4866 | * sys_setpriority is a more generic, but much slower function that | |
4867 | * does similar things. | |
4868 | */ | |
5add95d4 | 4869 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4870 | { |
48f24c4d | 4871 | long nice, retval; |
1da177e4 LT |
4872 | |
4873 | /* | |
4874 | * Setpriority might change our priority at the same moment. | |
4875 | * We don't have to worry. Conceptually one call occurs first | |
4876 | * and we have a single winner. | |
4877 | */ | |
e43379f1 MM |
4878 | if (increment < -40) |
4879 | increment = -40; | |
1da177e4 LT |
4880 | if (increment > 40) |
4881 | increment = 40; | |
4882 | ||
2b8f836f | 4883 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4884 | if (nice < -20) |
4885 | nice = -20; | |
4886 | if (nice > 19) | |
4887 | nice = 19; | |
4888 | ||
e43379f1 MM |
4889 | if (increment < 0 && !can_nice(current, nice)) |
4890 | return -EPERM; | |
4891 | ||
1da177e4 LT |
4892 | retval = security_task_setnice(current, nice); |
4893 | if (retval) | |
4894 | return retval; | |
4895 | ||
4896 | set_user_nice(current, nice); | |
4897 | return 0; | |
4898 | } | |
4899 | ||
4900 | #endif | |
4901 | ||
4902 | /** | |
4903 | * task_prio - return the priority value of a given task. | |
4904 | * @p: the task in question. | |
4905 | * | |
4906 | * This is the priority value as seen by users in /proc. | |
4907 | * RT tasks are offset by -200. Normal tasks are centered | |
4908 | * around 0, value goes from -16 to +15. | |
4909 | */ | |
36c8b586 | 4910 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4911 | { |
4912 | return p->prio - MAX_RT_PRIO; | |
4913 | } | |
4914 | ||
4915 | /** | |
4916 | * task_nice - return the nice value of a given task. | |
4917 | * @p: the task in question. | |
4918 | */ | |
36c8b586 | 4919 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4920 | { |
4921 | return TASK_NICE(p); | |
4922 | } | |
150d8bed | 4923 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4924 | |
4925 | /** | |
4926 | * idle_cpu - is a given cpu idle currently? | |
4927 | * @cpu: the processor in question. | |
4928 | */ | |
4929 | int idle_cpu(int cpu) | |
4930 | { | |
4931 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4932 | } | |
4933 | ||
1da177e4 LT |
4934 | /** |
4935 | * idle_task - return the idle task for a given cpu. | |
4936 | * @cpu: the processor in question. | |
4937 | */ | |
36c8b586 | 4938 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4939 | { |
4940 | return cpu_rq(cpu)->idle; | |
4941 | } | |
4942 | ||
4943 | /** | |
4944 | * find_process_by_pid - find a process with a matching PID value. | |
4945 | * @pid: the pid in question. | |
4946 | */ | |
a9957449 | 4947 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4948 | { |
228ebcbe | 4949 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4950 | } |
4951 | ||
4952 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4953 | static void |
4954 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4955 | { |
1da177e4 LT |
4956 | p->policy = policy; |
4957 | p->rt_priority = prio; | |
b29739f9 IM |
4958 | p->normal_prio = normal_prio(p); |
4959 | /* we are holding p->pi_lock already */ | |
4960 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4961 | if (rt_prio(p->prio)) |
4962 | p->sched_class = &rt_sched_class; | |
4963 | else | |
4964 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4965 | set_load_weight(p); |
1da177e4 LT |
4966 | } |
4967 | ||
c69e8d9c DH |
4968 | /* |
4969 | * check the target process has a UID that matches the current process's | |
4970 | */ | |
4971 | static bool check_same_owner(struct task_struct *p) | |
4972 | { | |
4973 | const struct cred *cred = current_cred(), *pcred; | |
4974 | bool match; | |
4975 | ||
4976 | rcu_read_lock(); | |
4977 | pcred = __task_cred(p); | |
b0e77598 SH |
4978 | if (cred->user->user_ns == pcred->user->user_ns) |
4979 | match = (cred->euid == pcred->euid || | |
4980 | cred->euid == pcred->uid); | |
4981 | else | |
4982 | match = false; | |
c69e8d9c DH |
4983 | rcu_read_unlock(); |
4984 | return match; | |
4985 | } | |
4986 | ||
961ccddd | 4987 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4988 | const struct sched_param *param, bool user) |
1da177e4 | 4989 | { |
83b699ed | 4990 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4991 | unsigned long flags; |
83ab0aa0 | 4992 | const struct sched_class *prev_class; |
70b97a7f | 4993 | struct rq *rq; |
ca94c442 | 4994 | int reset_on_fork; |
1da177e4 | 4995 | |
66e5393a SR |
4996 | /* may grab non-irq protected spin_locks */ |
4997 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4998 | recheck: |
4999 | /* double check policy once rq lock held */ | |
ca94c442 LP |
5000 | if (policy < 0) { |
5001 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5002 | policy = oldpolicy = p->policy; |
ca94c442 LP |
5003 | } else { |
5004 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
5005 | policy &= ~SCHED_RESET_ON_FORK; | |
5006 | ||
5007 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
5008 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
5009 | policy != SCHED_IDLE) | |
5010 | return -EINVAL; | |
5011 | } | |
5012 | ||
1da177e4 LT |
5013 | /* |
5014 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5015 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5016 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5017 | */ |
5018 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5019 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5020 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5021 | return -EINVAL; |
e05606d3 | 5022 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5023 | return -EINVAL; |
5024 | ||
37e4ab3f OC |
5025 | /* |
5026 | * Allow unprivileged RT tasks to decrease priority: | |
5027 | */ | |
961ccddd | 5028 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5029 | if (rt_policy(policy)) { |
a44702e8 ON |
5030 | unsigned long rlim_rtprio = |
5031 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
5032 | |
5033 | /* can't set/change the rt policy */ | |
5034 | if (policy != p->policy && !rlim_rtprio) | |
5035 | return -EPERM; | |
5036 | ||
5037 | /* can't increase priority */ | |
5038 | if (param->sched_priority > p->rt_priority && | |
5039 | param->sched_priority > rlim_rtprio) | |
5040 | return -EPERM; | |
5041 | } | |
c02aa73b | 5042 | |
dd41f596 | 5043 | /* |
c02aa73b DH |
5044 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5045 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5046 | */ |
c02aa73b DH |
5047 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
5048 | if (!can_nice(p, TASK_NICE(p))) | |
5049 | return -EPERM; | |
5050 | } | |
5fe1d75f | 5051 | |
37e4ab3f | 5052 | /* can't change other user's priorities */ |
c69e8d9c | 5053 | if (!check_same_owner(p)) |
37e4ab3f | 5054 | return -EPERM; |
ca94c442 LP |
5055 | |
5056 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
5057 | if (p->sched_reset_on_fork && !reset_on_fork) | |
5058 | return -EPERM; | |
37e4ab3f | 5059 | } |
1da177e4 | 5060 | |
725aad24 | 5061 | if (user) { |
b0ae1981 | 5062 | retval = security_task_setscheduler(p); |
725aad24 JF |
5063 | if (retval) |
5064 | return retval; | |
5065 | } | |
5066 | ||
b29739f9 IM |
5067 | /* |
5068 | * make sure no PI-waiters arrive (or leave) while we are | |
5069 | * changing the priority of the task: | |
0122ec5b | 5070 | * |
25985edc | 5071 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5072 | * runqueue lock must be held. |
5073 | */ | |
0122ec5b | 5074 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 5075 | |
34f971f6 PZ |
5076 | /* |
5077 | * Changing the policy of the stop threads its a very bad idea | |
5078 | */ | |
5079 | if (p == rq->stop) { | |
0122ec5b | 5080 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
5081 | return -EINVAL; |
5082 | } | |
5083 | ||
a51e9198 DF |
5084 | /* |
5085 | * If not changing anything there's no need to proceed further: | |
5086 | */ | |
5087 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5088 | param->sched_priority == p->rt_priority))) { | |
5089 | ||
5090 | __task_rq_unlock(rq); | |
5091 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5092 | return 0; | |
5093 | } | |
5094 | ||
dc61b1d6 PZ |
5095 | #ifdef CONFIG_RT_GROUP_SCHED |
5096 | if (user) { | |
5097 | /* | |
5098 | * Do not allow realtime tasks into groups that have no runtime | |
5099 | * assigned. | |
5100 | */ | |
5101 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5102 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5103 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 5104 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
5105 | return -EPERM; |
5106 | } | |
5107 | } | |
5108 | #endif | |
5109 | ||
1da177e4 LT |
5110 | /* recheck policy now with rq lock held */ |
5111 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5112 | policy = oldpolicy = -1; | |
0122ec5b | 5113 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
5114 | goto recheck; |
5115 | } | |
fd2f4419 | 5116 | on_rq = p->on_rq; |
051a1d1a | 5117 | running = task_current(rq, p); |
0e1f3483 | 5118 | if (on_rq) |
2e1cb74a | 5119 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5120 | if (running) |
5121 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5122 | |
ca94c442 LP |
5123 | p->sched_reset_on_fork = reset_on_fork; |
5124 | ||
1da177e4 | 5125 | oldprio = p->prio; |
83ab0aa0 | 5126 | prev_class = p->sched_class; |
dd41f596 | 5127 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5128 | |
0e1f3483 HS |
5129 | if (running) |
5130 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5131 | if (on_rq) |
dd41f596 | 5132 | activate_task(rq, p, 0); |
cb469845 | 5133 | |
da7a735e | 5134 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5135 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 5136 | |
95e02ca9 TG |
5137 | rt_mutex_adjust_pi(p); |
5138 | ||
1da177e4 LT |
5139 | return 0; |
5140 | } | |
961ccddd RR |
5141 | |
5142 | /** | |
5143 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5144 | * @p: the task in question. | |
5145 | * @policy: new policy. | |
5146 | * @param: structure containing the new RT priority. | |
5147 | * | |
5148 | * NOTE that the task may be already dead. | |
5149 | */ | |
5150 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5151 | const struct sched_param *param) |
961ccddd RR |
5152 | { |
5153 | return __sched_setscheduler(p, policy, param, true); | |
5154 | } | |
1da177e4 LT |
5155 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5156 | ||
961ccddd RR |
5157 | /** |
5158 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5159 | * @p: the task in question. | |
5160 | * @policy: new policy. | |
5161 | * @param: structure containing the new RT priority. | |
5162 | * | |
5163 | * Just like sched_setscheduler, only don't bother checking if the | |
5164 | * current context has permission. For example, this is needed in | |
5165 | * stop_machine(): we create temporary high priority worker threads, | |
5166 | * but our caller might not have that capability. | |
5167 | */ | |
5168 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5169 | const struct sched_param *param) |
961ccddd RR |
5170 | { |
5171 | return __sched_setscheduler(p, policy, param, false); | |
5172 | } | |
5173 | ||
95cdf3b7 IM |
5174 | static int |
5175 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5176 | { |
1da177e4 LT |
5177 | struct sched_param lparam; |
5178 | struct task_struct *p; | |
36c8b586 | 5179 | int retval; |
1da177e4 LT |
5180 | |
5181 | if (!param || pid < 0) | |
5182 | return -EINVAL; | |
5183 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5184 | return -EFAULT; | |
5fe1d75f ON |
5185 | |
5186 | rcu_read_lock(); | |
5187 | retval = -ESRCH; | |
1da177e4 | 5188 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5189 | if (p != NULL) |
5190 | retval = sched_setscheduler(p, policy, &lparam); | |
5191 | rcu_read_unlock(); | |
36c8b586 | 5192 | |
1da177e4 LT |
5193 | return retval; |
5194 | } | |
5195 | ||
5196 | /** | |
5197 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5198 | * @pid: the pid in question. | |
5199 | * @policy: new policy. | |
5200 | * @param: structure containing the new RT priority. | |
5201 | */ | |
5add95d4 HC |
5202 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5203 | struct sched_param __user *, param) | |
1da177e4 | 5204 | { |
c21761f1 JB |
5205 | /* negative values for policy are not valid */ |
5206 | if (policy < 0) | |
5207 | return -EINVAL; | |
5208 | ||
1da177e4 LT |
5209 | return do_sched_setscheduler(pid, policy, param); |
5210 | } | |
5211 | ||
5212 | /** | |
5213 | * sys_sched_setparam - set/change the RT priority of a thread | |
5214 | * @pid: the pid in question. | |
5215 | * @param: structure containing the new RT priority. | |
5216 | */ | |
5add95d4 | 5217 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5218 | { |
5219 | return do_sched_setscheduler(pid, -1, param); | |
5220 | } | |
5221 | ||
5222 | /** | |
5223 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5224 | * @pid: the pid in question. | |
5225 | */ | |
5add95d4 | 5226 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5227 | { |
36c8b586 | 5228 | struct task_struct *p; |
3a5c359a | 5229 | int retval; |
1da177e4 LT |
5230 | |
5231 | if (pid < 0) | |
3a5c359a | 5232 | return -EINVAL; |
1da177e4 LT |
5233 | |
5234 | retval = -ESRCH; | |
5fe85be0 | 5235 | rcu_read_lock(); |
1da177e4 LT |
5236 | p = find_process_by_pid(pid); |
5237 | if (p) { | |
5238 | retval = security_task_getscheduler(p); | |
5239 | if (!retval) | |
ca94c442 LP |
5240 | retval = p->policy |
5241 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5242 | } |
5fe85be0 | 5243 | rcu_read_unlock(); |
1da177e4 LT |
5244 | return retval; |
5245 | } | |
5246 | ||
5247 | /** | |
ca94c442 | 5248 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5249 | * @pid: the pid in question. |
5250 | * @param: structure containing the RT priority. | |
5251 | */ | |
5add95d4 | 5252 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5253 | { |
5254 | struct sched_param lp; | |
36c8b586 | 5255 | struct task_struct *p; |
3a5c359a | 5256 | int retval; |
1da177e4 LT |
5257 | |
5258 | if (!param || pid < 0) | |
3a5c359a | 5259 | return -EINVAL; |
1da177e4 | 5260 | |
5fe85be0 | 5261 | rcu_read_lock(); |
1da177e4 LT |
5262 | p = find_process_by_pid(pid); |
5263 | retval = -ESRCH; | |
5264 | if (!p) | |
5265 | goto out_unlock; | |
5266 | ||
5267 | retval = security_task_getscheduler(p); | |
5268 | if (retval) | |
5269 | goto out_unlock; | |
5270 | ||
5271 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5272 | rcu_read_unlock(); |
1da177e4 LT |
5273 | |
5274 | /* | |
5275 | * This one might sleep, we cannot do it with a spinlock held ... | |
5276 | */ | |
5277 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5278 | ||
1da177e4 LT |
5279 | return retval; |
5280 | ||
5281 | out_unlock: | |
5fe85be0 | 5282 | rcu_read_unlock(); |
1da177e4 LT |
5283 | return retval; |
5284 | } | |
5285 | ||
96f874e2 | 5286 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5287 | { |
5a16f3d3 | 5288 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5289 | struct task_struct *p; |
5290 | int retval; | |
1da177e4 | 5291 | |
95402b38 | 5292 | get_online_cpus(); |
23f5d142 | 5293 | rcu_read_lock(); |
1da177e4 LT |
5294 | |
5295 | p = find_process_by_pid(pid); | |
5296 | if (!p) { | |
23f5d142 | 5297 | rcu_read_unlock(); |
95402b38 | 5298 | put_online_cpus(); |
1da177e4 LT |
5299 | return -ESRCH; |
5300 | } | |
5301 | ||
23f5d142 | 5302 | /* Prevent p going away */ |
1da177e4 | 5303 | get_task_struct(p); |
23f5d142 | 5304 | rcu_read_unlock(); |
1da177e4 | 5305 | |
5a16f3d3 RR |
5306 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5307 | retval = -ENOMEM; | |
5308 | goto out_put_task; | |
5309 | } | |
5310 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5311 | retval = -ENOMEM; | |
5312 | goto out_free_cpus_allowed; | |
5313 | } | |
1da177e4 | 5314 | retval = -EPERM; |
b0e77598 | 5315 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5316 | goto out_unlock; |
5317 | ||
b0ae1981 | 5318 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5319 | if (retval) |
5320 | goto out_unlock; | |
5321 | ||
5a16f3d3 RR |
5322 | cpuset_cpus_allowed(p, cpus_allowed); |
5323 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5324 | again: |
5a16f3d3 | 5325 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5326 | |
8707d8b8 | 5327 | if (!retval) { |
5a16f3d3 RR |
5328 | cpuset_cpus_allowed(p, cpus_allowed); |
5329 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5330 | /* |
5331 | * We must have raced with a concurrent cpuset | |
5332 | * update. Just reset the cpus_allowed to the | |
5333 | * cpuset's cpus_allowed | |
5334 | */ | |
5a16f3d3 | 5335 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5336 | goto again; |
5337 | } | |
5338 | } | |
1da177e4 | 5339 | out_unlock: |
5a16f3d3 RR |
5340 | free_cpumask_var(new_mask); |
5341 | out_free_cpus_allowed: | |
5342 | free_cpumask_var(cpus_allowed); | |
5343 | out_put_task: | |
1da177e4 | 5344 | put_task_struct(p); |
95402b38 | 5345 | put_online_cpus(); |
1da177e4 LT |
5346 | return retval; |
5347 | } | |
5348 | ||
5349 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5350 | struct cpumask *new_mask) |
1da177e4 | 5351 | { |
96f874e2 RR |
5352 | if (len < cpumask_size()) |
5353 | cpumask_clear(new_mask); | |
5354 | else if (len > cpumask_size()) | |
5355 | len = cpumask_size(); | |
5356 | ||
1da177e4 LT |
5357 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5358 | } | |
5359 | ||
5360 | /** | |
5361 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5362 | * @pid: pid of the process | |
5363 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5364 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5365 | */ | |
5add95d4 HC |
5366 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5367 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5368 | { |
5a16f3d3 | 5369 | cpumask_var_t new_mask; |
1da177e4 LT |
5370 | int retval; |
5371 | ||
5a16f3d3 RR |
5372 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5373 | return -ENOMEM; | |
1da177e4 | 5374 | |
5a16f3d3 RR |
5375 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5376 | if (retval == 0) | |
5377 | retval = sched_setaffinity(pid, new_mask); | |
5378 | free_cpumask_var(new_mask); | |
5379 | return retval; | |
1da177e4 LT |
5380 | } |
5381 | ||
96f874e2 | 5382 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5383 | { |
36c8b586 | 5384 | struct task_struct *p; |
31605683 | 5385 | unsigned long flags; |
1da177e4 | 5386 | int retval; |
1da177e4 | 5387 | |
95402b38 | 5388 | get_online_cpus(); |
23f5d142 | 5389 | rcu_read_lock(); |
1da177e4 LT |
5390 | |
5391 | retval = -ESRCH; | |
5392 | p = find_process_by_pid(pid); | |
5393 | if (!p) | |
5394 | goto out_unlock; | |
5395 | ||
e7834f8f DQ |
5396 | retval = security_task_getscheduler(p); |
5397 | if (retval) | |
5398 | goto out_unlock; | |
5399 | ||
013fdb80 | 5400 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 5401 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 5402 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5403 | |
5404 | out_unlock: | |
23f5d142 | 5405 | rcu_read_unlock(); |
95402b38 | 5406 | put_online_cpus(); |
1da177e4 | 5407 | |
9531b62f | 5408 | return retval; |
1da177e4 LT |
5409 | } |
5410 | ||
5411 | /** | |
5412 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5413 | * @pid: pid of the process | |
5414 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5415 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5416 | */ | |
5add95d4 HC |
5417 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5418 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5419 | { |
5420 | int ret; | |
f17c8607 | 5421 | cpumask_var_t mask; |
1da177e4 | 5422 | |
84fba5ec | 5423 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5424 | return -EINVAL; |
5425 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5426 | return -EINVAL; |
5427 | ||
f17c8607 RR |
5428 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5429 | return -ENOMEM; | |
1da177e4 | 5430 | |
f17c8607 RR |
5431 | ret = sched_getaffinity(pid, mask); |
5432 | if (ret == 0) { | |
8bc037fb | 5433 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5434 | |
5435 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5436 | ret = -EFAULT; |
5437 | else | |
cd3d8031 | 5438 | ret = retlen; |
f17c8607 RR |
5439 | } |
5440 | free_cpumask_var(mask); | |
1da177e4 | 5441 | |
f17c8607 | 5442 | return ret; |
1da177e4 LT |
5443 | } |
5444 | ||
5445 | /** | |
5446 | * sys_sched_yield - yield the current processor to other threads. | |
5447 | * | |
dd41f596 IM |
5448 | * This function yields the current CPU to other tasks. If there are no |
5449 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5450 | */ |
5add95d4 | 5451 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5452 | { |
70b97a7f | 5453 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5454 | |
2d72376b | 5455 | schedstat_inc(rq, yld_count); |
4530d7ab | 5456 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5457 | |
5458 | /* | |
5459 | * Since we are going to call schedule() anyway, there's | |
5460 | * no need to preempt or enable interrupts: | |
5461 | */ | |
5462 | __release(rq->lock); | |
8a25d5de | 5463 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5464 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5465 | preempt_enable_no_resched(); |
5466 | ||
5467 | schedule(); | |
5468 | ||
5469 | return 0; | |
5470 | } | |
5471 | ||
d86ee480 PZ |
5472 | static inline int should_resched(void) |
5473 | { | |
5474 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5475 | } | |
5476 | ||
e7b38404 | 5477 | static void __cond_resched(void) |
1da177e4 | 5478 | { |
e7aaaa69 FW |
5479 | add_preempt_count(PREEMPT_ACTIVE); |
5480 | schedule(); | |
5481 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5482 | } |
5483 | ||
02b67cc3 | 5484 | int __sched _cond_resched(void) |
1da177e4 | 5485 | { |
d86ee480 | 5486 | if (should_resched()) { |
1da177e4 LT |
5487 | __cond_resched(); |
5488 | return 1; | |
5489 | } | |
5490 | return 0; | |
5491 | } | |
02b67cc3 | 5492 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5493 | |
5494 | /* | |
613afbf8 | 5495 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5496 | * call schedule, and on return reacquire the lock. |
5497 | * | |
41a2d6cf | 5498 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5499 | * operations here to prevent schedule() from being called twice (once via |
5500 | * spin_unlock(), once by hand). | |
5501 | */ | |
613afbf8 | 5502 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5503 | { |
d86ee480 | 5504 | int resched = should_resched(); |
6df3cecb JK |
5505 | int ret = 0; |
5506 | ||
f607c668 PZ |
5507 | lockdep_assert_held(lock); |
5508 | ||
95c354fe | 5509 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5510 | spin_unlock(lock); |
d86ee480 | 5511 | if (resched) |
95c354fe NP |
5512 | __cond_resched(); |
5513 | else | |
5514 | cpu_relax(); | |
6df3cecb | 5515 | ret = 1; |
1da177e4 | 5516 | spin_lock(lock); |
1da177e4 | 5517 | } |
6df3cecb | 5518 | return ret; |
1da177e4 | 5519 | } |
613afbf8 | 5520 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5521 | |
613afbf8 | 5522 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5523 | { |
5524 | BUG_ON(!in_softirq()); | |
5525 | ||
d86ee480 | 5526 | if (should_resched()) { |
98d82567 | 5527 | local_bh_enable(); |
1da177e4 LT |
5528 | __cond_resched(); |
5529 | local_bh_disable(); | |
5530 | return 1; | |
5531 | } | |
5532 | return 0; | |
5533 | } | |
613afbf8 | 5534 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5535 | |
1da177e4 LT |
5536 | /** |
5537 | * yield - yield the current processor to other threads. | |
5538 | * | |
72fd4a35 | 5539 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5540 | * thread runnable and calls sys_sched_yield(). |
5541 | */ | |
5542 | void __sched yield(void) | |
5543 | { | |
5544 | set_current_state(TASK_RUNNING); | |
5545 | sys_sched_yield(); | |
5546 | } | |
1da177e4 LT |
5547 | EXPORT_SYMBOL(yield); |
5548 | ||
d95f4122 MG |
5549 | /** |
5550 | * yield_to - yield the current processor to another thread in | |
5551 | * your thread group, or accelerate that thread toward the | |
5552 | * processor it's on. | |
16addf95 RD |
5553 | * @p: target task |
5554 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5555 | * |
5556 | * It's the caller's job to ensure that the target task struct | |
5557 | * can't go away on us before we can do any checks. | |
5558 | * | |
5559 | * Returns true if we indeed boosted the target task. | |
5560 | */ | |
5561 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5562 | { | |
5563 | struct task_struct *curr = current; | |
5564 | struct rq *rq, *p_rq; | |
5565 | unsigned long flags; | |
5566 | bool yielded = 0; | |
5567 | ||
5568 | local_irq_save(flags); | |
5569 | rq = this_rq(); | |
5570 | ||
5571 | again: | |
5572 | p_rq = task_rq(p); | |
5573 | double_rq_lock(rq, p_rq); | |
5574 | while (task_rq(p) != p_rq) { | |
5575 | double_rq_unlock(rq, p_rq); | |
5576 | goto again; | |
5577 | } | |
5578 | ||
5579 | if (!curr->sched_class->yield_to_task) | |
5580 | goto out; | |
5581 | ||
5582 | if (curr->sched_class != p->sched_class) | |
5583 | goto out; | |
5584 | ||
5585 | if (task_running(p_rq, p) || p->state) | |
5586 | goto out; | |
5587 | ||
5588 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5589 | if (yielded) { |
d95f4122 | 5590 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5591 | /* |
5592 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5593 | * fairness. | |
5594 | */ | |
5595 | if (preempt && rq != p_rq) | |
5596 | resched_task(p_rq->curr); | |
5597 | } | |
d95f4122 MG |
5598 | |
5599 | out: | |
5600 | double_rq_unlock(rq, p_rq); | |
5601 | local_irq_restore(flags); | |
5602 | ||
5603 | if (yielded) | |
5604 | schedule(); | |
5605 | ||
5606 | return yielded; | |
5607 | } | |
5608 | EXPORT_SYMBOL_GPL(yield_to); | |
5609 | ||
1da177e4 | 5610 | /* |
41a2d6cf | 5611 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5612 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5613 | */ |
5614 | void __sched io_schedule(void) | |
5615 | { | |
54d35f29 | 5616 | struct rq *rq = raw_rq(); |
1da177e4 | 5617 | |
0ff92245 | 5618 | delayacct_blkio_start(); |
1da177e4 | 5619 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5620 | blk_flush_plug(current); |
8f0dfc34 | 5621 | current->in_iowait = 1; |
1da177e4 | 5622 | schedule(); |
8f0dfc34 | 5623 | current->in_iowait = 0; |
1da177e4 | 5624 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5625 | delayacct_blkio_end(); |
1da177e4 | 5626 | } |
1da177e4 LT |
5627 | EXPORT_SYMBOL(io_schedule); |
5628 | ||
5629 | long __sched io_schedule_timeout(long timeout) | |
5630 | { | |
54d35f29 | 5631 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5632 | long ret; |
5633 | ||
0ff92245 | 5634 | delayacct_blkio_start(); |
1da177e4 | 5635 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5636 | blk_flush_plug(current); |
8f0dfc34 | 5637 | current->in_iowait = 1; |
1da177e4 | 5638 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5639 | current->in_iowait = 0; |
1da177e4 | 5640 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5641 | delayacct_blkio_end(); |
1da177e4 LT |
5642 | return ret; |
5643 | } | |
5644 | ||
5645 | /** | |
5646 | * sys_sched_get_priority_max - return maximum RT priority. | |
5647 | * @policy: scheduling class. | |
5648 | * | |
5649 | * this syscall returns the maximum rt_priority that can be used | |
5650 | * by a given scheduling class. | |
5651 | */ | |
5add95d4 | 5652 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5653 | { |
5654 | int ret = -EINVAL; | |
5655 | ||
5656 | switch (policy) { | |
5657 | case SCHED_FIFO: | |
5658 | case SCHED_RR: | |
5659 | ret = MAX_USER_RT_PRIO-1; | |
5660 | break; | |
5661 | case SCHED_NORMAL: | |
b0a9499c | 5662 | case SCHED_BATCH: |
dd41f596 | 5663 | case SCHED_IDLE: |
1da177e4 LT |
5664 | ret = 0; |
5665 | break; | |
5666 | } | |
5667 | return ret; | |
5668 | } | |
5669 | ||
5670 | /** | |
5671 | * sys_sched_get_priority_min - return minimum RT priority. | |
5672 | * @policy: scheduling class. | |
5673 | * | |
5674 | * this syscall returns the minimum rt_priority that can be used | |
5675 | * by a given scheduling class. | |
5676 | */ | |
5add95d4 | 5677 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5678 | { |
5679 | int ret = -EINVAL; | |
5680 | ||
5681 | switch (policy) { | |
5682 | case SCHED_FIFO: | |
5683 | case SCHED_RR: | |
5684 | ret = 1; | |
5685 | break; | |
5686 | case SCHED_NORMAL: | |
b0a9499c | 5687 | case SCHED_BATCH: |
dd41f596 | 5688 | case SCHED_IDLE: |
1da177e4 LT |
5689 | ret = 0; |
5690 | } | |
5691 | return ret; | |
5692 | } | |
5693 | ||
5694 | /** | |
5695 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5696 | * @pid: pid of the process. | |
5697 | * @interval: userspace pointer to the timeslice value. | |
5698 | * | |
5699 | * this syscall writes the default timeslice value of a given process | |
5700 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5701 | */ | |
17da2bd9 | 5702 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5703 | struct timespec __user *, interval) |
1da177e4 | 5704 | { |
36c8b586 | 5705 | struct task_struct *p; |
a4ec24b4 | 5706 | unsigned int time_slice; |
dba091b9 TG |
5707 | unsigned long flags; |
5708 | struct rq *rq; | |
3a5c359a | 5709 | int retval; |
1da177e4 | 5710 | struct timespec t; |
1da177e4 LT |
5711 | |
5712 | if (pid < 0) | |
3a5c359a | 5713 | return -EINVAL; |
1da177e4 LT |
5714 | |
5715 | retval = -ESRCH; | |
1a551ae7 | 5716 | rcu_read_lock(); |
1da177e4 LT |
5717 | p = find_process_by_pid(pid); |
5718 | if (!p) | |
5719 | goto out_unlock; | |
5720 | ||
5721 | retval = security_task_getscheduler(p); | |
5722 | if (retval) | |
5723 | goto out_unlock; | |
5724 | ||
dba091b9 TG |
5725 | rq = task_rq_lock(p, &flags); |
5726 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5727 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5728 | |
1a551ae7 | 5729 | rcu_read_unlock(); |
a4ec24b4 | 5730 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5731 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5732 | return retval; |
3a5c359a | 5733 | |
1da177e4 | 5734 | out_unlock: |
1a551ae7 | 5735 | rcu_read_unlock(); |
1da177e4 LT |
5736 | return retval; |
5737 | } | |
5738 | ||
7c731e0a | 5739 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5740 | |
82a1fcb9 | 5741 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5742 | { |
1da177e4 | 5743 | unsigned long free = 0; |
36c8b586 | 5744 | unsigned state; |
1da177e4 | 5745 | |
1da177e4 | 5746 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5747 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5748 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5749 | #if BITS_PER_LONG == 32 |
1da177e4 | 5750 | if (state == TASK_RUNNING) |
3df0fc5b | 5751 | printk(KERN_CONT " running "); |
1da177e4 | 5752 | else |
3df0fc5b | 5753 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5754 | #else |
5755 | if (state == TASK_RUNNING) | |
3df0fc5b | 5756 | printk(KERN_CONT " running task "); |
1da177e4 | 5757 | else |
3df0fc5b | 5758 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5759 | #endif |
5760 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5761 | free = stack_not_used(p); |
1da177e4 | 5762 | #endif |
3df0fc5b | 5763 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5764 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5765 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5766 | |
5fb5e6de | 5767 | show_stack(p, NULL); |
1da177e4 LT |
5768 | } |
5769 | ||
e59e2ae2 | 5770 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5771 | { |
36c8b586 | 5772 | struct task_struct *g, *p; |
1da177e4 | 5773 | |
4bd77321 | 5774 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5775 | printk(KERN_INFO |
5776 | " task PC stack pid father\n"); | |
1da177e4 | 5777 | #else |
3df0fc5b PZ |
5778 | printk(KERN_INFO |
5779 | " task PC stack pid father\n"); | |
1da177e4 LT |
5780 | #endif |
5781 | read_lock(&tasklist_lock); | |
5782 | do_each_thread(g, p) { | |
5783 | /* | |
5784 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5785 | * console might take a lot of time: |
1da177e4 LT |
5786 | */ |
5787 | touch_nmi_watchdog(); | |
39bc89fd | 5788 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5789 | sched_show_task(p); |
1da177e4 LT |
5790 | } while_each_thread(g, p); |
5791 | ||
04c9167f JF |
5792 | touch_all_softlockup_watchdogs(); |
5793 | ||
dd41f596 IM |
5794 | #ifdef CONFIG_SCHED_DEBUG |
5795 | sysrq_sched_debug_show(); | |
5796 | #endif | |
1da177e4 | 5797 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5798 | /* |
5799 | * Only show locks if all tasks are dumped: | |
5800 | */ | |
93335a21 | 5801 | if (!state_filter) |
e59e2ae2 | 5802 | debug_show_all_locks(); |
1da177e4 LT |
5803 | } |
5804 | ||
1df21055 IM |
5805 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5806 | { | |
dd41f596 | 5807 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5808 | } |
5809 | ||
f340c0d1 IM |
5810 | /** |
5811 | * init_idle - set up an idle thread for a given CPU | |
5812 | * @idle: task in question | |
5813 | * @cpu: cpu the idle task belongs to | |
5814 | * | |
5815 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5816 | * flag, to make booting more robust. | |
5817 | */ | |
5c1e1767 | 5818 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5819 | { |
70b97a7f | 5820 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5821 | unsigned long flags; |
5822 | ||
05fa785c | 5823 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5824 | |
dd41f596 | 5825 | __sched_fork(idle); |
06b83b5f | 5826 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5827 | idle->se.exec_start = sched_clock(); |
5828 | ||
96f874e2 | 5829 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5830 | /* |
5831 | * We're having a chicken and egg problem, even though we are | |
5832 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5833 | * lockdep check in task_group() will fail. | |
5834 | * | |
5835 | * Similar case to sched_fork(). / Alternatively we could | |
5836 | * use task_rq_lock() here and obtain the other rq->lock. | |
5837 | * | |
5838 | * Silence PROVE_RCU | |
5839 | */ | |
5840 | rcu_read_lock(); | |
dd41f596 | 5841 | __set_task_cpu(idle, cpu); |
6506cf6c | 5842 | rcu_read_unlock(); |
1da177e4 | 5843 | |
1da177e4 | 5844 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
5845 | #if defined(CONFIG_SMP) |
5846 | idle->on_cpu = 1; | |
4866cde0 | 5847 | #endif |
05fa785c | 5848 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5849 | |
5850 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5851 | task_thread_info(idle)->preempt_count = 0; |
625f2a37 | 5852 | |
dd41f596 IM |
5853 | /* |
5854 | * The idle tasks have their own, simple scheduling class: | |
5855 | */ | |
5856 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5857 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5858 | } |
5859 | ||
5860 | /* | |
5861 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5862 | * indicates which cpus entered this state. This is used | |
5863 | * in the rcu update to wait only for active cpus. For system | |
5864 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5865 | * always be CPU_BITS_NONE. |
1da177e4 | 5866 | */ |
6a7b3dc3 | 5867 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5868 | |
19978ca6 IM |
5869 | /* |
5870 | * Increase the granularity value when there are more CPUs, | |
5871 | * because with more CPUs the 'effective latency' as visible | |
5872 | * to users decreases. But the relationship is not linear, | |
5873 | * so pick a second-best guess by going with the log2 of the | |
5874 | * number of CPUs. | |
5875 | * | |
5876 | * This idea comes from the SD scheduler of Con Kolivas: | |
5877 | */ | |
acb4a848 | 5878 | static int get_update_sysctl_factor(void) |
19978ca6 | 5879 | { |
4ca3ef71 | 5880 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5881 | unsigned int factor; |
5882 | ||
5883 | switch (sysctl_sched_tunable_scaling) { | |
5884 | case SCHED_TUNABLESCALING_NONE: | |
5885 | factor = 1; | |
5886 | break; | |
5887 | case SCHED_TUNABLESCALING_LINEAR: | |
5888 | factor = cpus; | |
5889 | break; | |
5890 | case SCHED_TUNABLESCALING_LOG: | |
5891 | default: | |
5892 | factor = 1 + ilog2(cpus); | |
5893 | break; | |
5894 | } | |
19978ca6 | 5895 | |
acb4a848 CE |
5896 | return factor; |
5897 | } | |
19978ca6 | 5898 | |
acb4a848 CE |
5899 | static void update_sysctl(void) |
5900 | { | |
5901 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5902 | |
0bcdcf28 CE |
5903 | #define SET_SYSCTL(name) \ |
5904 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5905 | SET_SYSCTL(sched_min_granularity); | |
5906 | SET_SYSCTL(sched_latency); | |
5907 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5908 | #undef SET_SYSCTL |
5909 | } | |
55cd5340 | 5910 | |
0bcdcf28 CE |
5911 | static inline void sched_init_granularity(void) |
5912 | { | |
5913 | update_sysctl(); | |
19978ca6 IM |
5914 | } |
5915 | ||
1da177e4 LT |
5916 | #ifdef CONFIG_SMP |
5917 | /* | |
5918 | * This is how migration works: | |
5919 | * | |
969c7921 TH |
5920 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5921 | * stop_one_cpu(). | |
5922 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5923 | * off the CPU) | |
5924 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5925 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5926 | * it and puts it into the right queue. |
969c7921 TH |
5927 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5928 | * is done. | |
1da177e4 LT |
5929 | */ |
5930 | ||
5931 | /* | |
5932 | * Change a given task's CPU affinity. Migrate the thread to a | |
5933 | * proper CPU and schedule it away if the CPU it's executing on | |
5934 | * is removed from the allowed bitmask. | |
5935 | * | |
5936 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5937 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5938 | * call is not atomic; no spinlocks may be held. |
5939 | */ | |
96f874e2 | 5940 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5941 | { |
5942 | unsigned long flags; | |
70b97a7f | 5943 | struct rq *rq; |
969c7921 | 5944 | unsigned int dest_cpu; |
48f24c4d | 5945 | int ret = 0; |
1da177e4 LT |
5946 | |
5947 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5948 | |
db44fc01 YZ |
5949 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
5950 | goto out; | |
5951 | ||
6ad4c188 | 5952 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5953 | ret = -EINVAL; |
5954 | goto out; | |
5955 | } | |
5956 | ||
db44fc01 | 5957 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
5958 | ret = -EINVAL; |
5959 | goto out; | |
5960 | } | |
5961 | ||
73fe6aae | 5962 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5963 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5964 | else { |
96f874e2 RR |
5965 | cpumask_copy(&p->cpus_allowed, new_mask); |
5966 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5967 | } |
5968 | ||
1da177e4 | 5969 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5970 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5971 | goto out; |
5972 | ||
969c7921 | 5973 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 5974 | if (p->on_rq) { |
969c7921 | 5975 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 5976 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 5977 | task_rq_unlock(rq, p, &flags); |
969c7921 | 5978 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5979 | tlb_migrate_finish(p->mm); |
5980 | return 0; | |
5981 | } | |
5982 | out: | |
0122ec5b | 5983 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 5984 | |
1da177e4 LT |
5985 | return ret; |
5986 | } | |
cd8ba7cd | 5987 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5988 | |
5989 | /* | |
41a2d6cf | 5990 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5991 | * this because either it can't run here any more (set_cpus_allowed() |
5992 | * away from this CPU, or CPU going down), or because we're | |
5993 | * attempting to rebalance this task on exec (sched_exec). | |
5994 | * | |
5995 | * So we race with normal scheduler movements, but that's OK, as long | |
5996 | * as the task is no longer on this CPU. | |
efc30814 KK |
5997 | * |
5998 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5999 | */ |
efc30814 | 6000 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6001 | { |
70b97a7f | 6002 | struct rq *rq_dest, *rq_src; |
e2912009 | 6003 | int ret = 0; |
1da177e4 | 6004 | |
e761b772 | 6005 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6006 | return ret; |
1da177e4 LT |
6007 | |
6008 | rq_src = cpu_rq(src_cpu); | |
6009 | rq_dest = cpu_rq(dest_cpu); | |
6010 | ||
0122ec5b | 6011 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
6012 | double_rq_lock(rq_src, rq_dest); |
6013 | /* Already moved. */ | |
6014 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6015 | goto done; |
1da177e4 | 6016 | /* Affinity changed (again). */ |
96f874e2 | 6017 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6018 | goto fail; |
1da177e4 | 6019 | |
e2912009 PZ |
6020 | /* |
6021 | * If we're not on a rq, the next wake-up will ensure we're | |
6022 | * placed properly. | |
6023 | */ | |
fd2f4419 | 6024 | if (p->on_rq) { |
2e1cb74a | 6025 | deactivate_task(rq_src, p, 0); |
e2912009 | 6026 | set_task_cpu(p, dest_cpu); |
dd41f596 | 6027 | activate_task(rq_dest, p, 0); |
15afe09b | 6028 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6029 | } |
b1e38734 | 6030 | done: |
efc30814 | 6031 | ret = 1; |
b1e38734 | 6032 | fail: |
1da177e4 | 6033 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 6034 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 6035 | return ret; |
1da177e4 LT |
6036 | } |
6037 | ||
6038 | /* | |
969c7921 TH |
6039 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
6040 | * and performs thread migration by bumping thread off CPU then | |
6041 | * 'pushing' onto another runqueue. | |
1da177e4 | 6042 | */ |
969c7921 | 6043 | static int migration_cpu_stop(void *data) |
1da177e4 | 6044 | { |
969c7921 | 6045 | struct migration_arg *arg = data; |
f7b4cddc | 6046 | |
969c7921 TH |
6047 | /* |
6048 | * The original target cpu might have gone down and we might | |
6049 | * be on another cpu but it doesn't matter. | |
6050 | */ | |
f7b4cddc | 6051 | local_irq_disable(); |
969c7921 | 6052 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 6053 | local_irq_enable(); |
1da177e4 | 6054 | return 0; |
f7b4cddc ON |
6055 | } |
6056 | ||
1da177e4 | 6057 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6058 | |
054b9108 | 6059 | /* |
48c5ccae PZ |
6060 | * Ensures that the idle task is using init_mm right before its cpu goes |
6061 | * offline. | |
054b9108 | 6062 | */ |
48c5ccae | 6063 | void idle_task_exit(void) |
1da177e4 | 6064 | { |
48c5ccae | 6065 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6066 | |
48c5ccae | 6067 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6068 | |
48c5ccae PZ |
6069 | if (mm != &init_mm) |
6070 | switch_mm(mm, &init_mm, current); | |
6071 | mmdrop(mm); | |
1da177e4 LT |
6072 | } |
6073 | ||
6074 | /* | |
6075 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6076 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6077 | * for performance reasons the counter is not stricly tracking tasks to | |
6078 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6079 | * to keep the global sum constant after CPU-down: | |
6080 | */ | |
70b97a7f | 6081 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6082 | { |
6ad4c188 | 6083 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6084 | |
1da177e4 LT |
6085 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6086 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6087 | } |
6088 | ||
dd41f596 | 6089 | /* |
48c5ccae | 6090 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6091 | */ |
48c5ccae | 6092 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6093 | { |
48c5ccae PZ |
6094 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6095 | rq->calc_load_active = 0; | |
1da177e4 LT |
6096 | } |
6097 | ||
48f24c4d | 6098 | /* |
48c5ccae PZ |
6099 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6100 | * try_to_wake_up()->select_task_rq(). | |
6101 | * | |
6102 | * Called with rq->lock held even though we'er in stop_machine() and | |
6103 | * there's no concurrency possible, we hold the required locks anyway | |
6104 | * because of lock validation efforts. | |
1da177e4 | 6105 | */ |
48c5ccae | 6106 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6107 | { |
70b97a7f | 6108 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6109 | struct task_struct *next, *stop = rq->stop; |
6110 | int dest_cpu; | |
1da177e4 LT |
6111 | |
6112 | /* | |
48c5ccae PZ |
6113 | * Fudge the rq selection such that the below task selection loop |
6114 | * doesn't get stuck on the currently eligible stop task. | |
6115 | * | |
6116 | * We're currently inside stop_machine() and the rq is either stuck | |
6117 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6118 | * either way we should never end up calling schedule() until we're | |
6119 | * done here. | |
1da177e4 | 6120 | */ |
48c5ccae | 6121 | rq->stop = NULL; |
48f24c4d | 6122 | |
dd41f596 | 6123 | for ( ; ; ) { |
48c5ccae PZ |
6124 | /* |
6125 | * There's this thread running, bail when that's the only | |
6126 | * remaining thread. | |
6127 | */ | |
6128 | if (rq->nr_running == 1) | |
dd41f596 | 6129 | break; |
48c5ccae | 6130 | |
b67802ea | 6131 | next = pick_next_task(rq); |
48c5ccae | 6132 | BUG_ON(!next); |
79c53799 | 6133 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6134 | |
48c5ccae PZ |
6135 | /* Find suitable destination for @next, with force if needed. */ |
6136 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6137 | raw_spin_unlock(&rq->lock); | |
6138 | ||
6139 | __migrate_task(next, dead_cpu, dest_cpu); | |
6140 | ||
6141 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6142 | } |
dce48a84 | 6143 | |
48c5ccae | 6144 | rq->stop = stop; |
dce48a84 | 6145 | } |
48c5ccae | 6146 | |
1da177e4 LT |
6147 | #endif /* CONFIG_HOTPLUG_CPU */ |
6148 | ||
e692ab53 NP |
6149 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6150 | ||
6151 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6152 | { |
6153 | .procname = "sched_domain", | |
c57baf1e | 6154 | .mode = 0555, |
e0361851 | 6155 | }, |
56992309 | 6156 | {} |
e692ab53 NP |
6157 | }; |
6158 | ||
6159 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6160 | { |
6161 | .procname = "kernel", | |
c57baf1e | 6162 | .mode = 0555, |
e0361851 AD |
6163 | .child = sd_ctl_dir, |
6164 | }, | |
56992309 | 6165 | {} |
e692ab53 NP |
6166 | }; |
6167 | ||
6168 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6169 | { | |
6170 | struct ctl_table *entry = | |
5cf9f062 | 6171 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6172 | |
e692ab53 NP |
6173 | return entry; |
6174 | } | |
6175 | ||
6382bc90 MM |
6176 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6177 | { | |
cd790076 | 6178 | struct ctl_table *entry; |
6382bc90 | 6179 | |
cd790076 MM |
6180 | /* |
6181 | * In the intermediate directories, both the child directory and | |
6182 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6183 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6184 | * static strings and all have proc handlers. |
6185 | */ | |
6186 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6187 | if (entry->child) |
6188 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6189 | if (entry->proc_handler == NULL) |
6190 | kfree(entry->procname); | |
6191 | } | |
6382bc90 MM |
6192 | |
6193 | kfree(*tablep); | |
6194 | *tablep = NULL; | |
6195 | } | |
6196 | ||
e692ab53 | 6197 | static void |
e0361851 | 6198 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6199 | const char *procname, void *data, int maxlen, |
6200 | mode_t mode, proc_handler *proc_handler) | |
6201 | { | |
e692ab53 NP |
6202 | entry->procname = procname; |
6203 | entry->data = data; | |
6204 | entry->maxlen = maxlen; | |
6205 | entry->mode = mode; | |
6206 | entry->proc_handler = proc_handler; | |
6207 | } | |
6208 | ||
6209 | static struct ctl_table * | |
6210 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6211 | { | |
a5d8c348 | 6212 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6213 | |
ad1cdc1d MM |
6214 | if (table == NULL) |
6215 | return NULL; | |
6216 | ||
e0361851 | 6217 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6218 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6219 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6220 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6221 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6222 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6223 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6224 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6225 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6226 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6227 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6228 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6229 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6230 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6231 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6232 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6233 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6234 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6235 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6236 | &sd->cache_nice_tries, |
6237 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6238 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6239 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6240 | set_table_entry(&table[11], "name", sd->name, |
6241 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6242 | /* &table[12] is terminator */ | |
e692ab53 NP |
6243 | |
6244 | return table; | |
6245 | } | |
6246 | ||
9a4e7159 | 6247 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6248 | { |
6249 | struct ctl_table *entry, *table; | |
6250 | struct sched_domain *sd; | |
6251 | int domain_num = 0, i; | |
6252 | char buf[32]; | |
6253 | ||
6254 | for_each_domain(cpu, sd) | |
6255 | domain_num++; | |
6256 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6257 | if (table == NULL) |
6258 | return NULL; | |
e692ab53 NP |
6259 | |
6260 | i = 0; | |
6261 | for_each_domain(cpu, sd) { | |
6262 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6263 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6264 | entry->mode = 0555; |
e692ab53 NP |
6265 | entry->child = sd_alloc_ctl_domain_table(sd); |
6266 | entry++; | |
6267 | i++; | |
6268 | } | |
6269 | return table; | |
6270 | } | |
6271 | ||
6272 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6273 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6274 | { |
6ad4c188 | 6275 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6276 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6277 | char buf[32]; | |
6278 | ||
7378547f MM |
6279 | WARN_ON(sd_ctl_dir[0].child); |
6280 | sd_ctl_dir[0].child = entry; | |
6281 | ||
ad1cdc1d MM |
6282 | if (entry == NULL) |
6283 | return; | |
6284 | ||
6ad4c188 | 6285 | for_each_possible_cpu(i) { |
e692ab53 | 6286 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6287 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6288 | entry->mode = 0555; |
e692ab53 | 6289 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6290 | entry++; |
e692ab53 | 6291 | } |
7378547f MM |
6292 | |
6293 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6294 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6295 | } | |
6382bc90 | 6296 | |
7378547f | 6297 | /* may be called multiple times per register */ |
6382bc90 MM |
6298 | static void unregister_sched_domain_sysctl(void) |
6299 | { | |
7378547f MM |
6300 | if (sd_sysctl_header) |
6301 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6302 | sd_sysctl_header = NULL; |
7378547f MM |
6303 | if (sd_ctl_dir[0].child) |
6304 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6305 | } |
e692ab53 | 6306 | #else |
6382bc90 MM |
6307 | static void register_sched_domain_sysctl(void) |
6308 | { | |
6309 | } | |
6310 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6311 | { |
6312 | } | |
6313 | #endif | |
6314 | ||
1f11eb6a GH |
6315 | static void set_rq_online(struct rq *rq) |
6316 | { | |
6317 | if (!rq->online) { | |
6318 | const struct sched_class *class; | |
6319 | ||
c6c4927b | 6320 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6321 | rq->online = 1; |
6322 | ||
6323 | for_each_class(class) { | |
6324 | if (class->rq_online) | |
6325 | class->rq_online(rq); | |
6326 | } | |
6327 | } | |
6328 | } | |
6329 | ||
6330 | static void set_rq_offline(struct rq *rq) | |
6331 | { | |
6332 | if (rq->online) { | |
6333 | const struct sched_class *class; | |
6334 | ||
6335 | for_each_class(class) { | |
6336 | if (class->rq_offline) | |
6337 | class->rq_offline(rq); | |
6338 | } | |
6339 | ||
c6c4927b | 6340 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6341 | rq->online = 0; |
6342 | } | |
6343 | } | |
6344 | ||
1da177e4 LT |
6345 | /* |
6346 | * migration_call - callback that gets triggered when a CPU is added. | |
6347 | * Here we can start up the necessary migration thread for the new CPU. | |
6348 | */ | |
48f24c4d IM |
6349 | static int __cpuinit |
6350 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6351 | { |
48f24c4d | 6352 | int cpu = (long)hcpu; |
1da177e4 | 6353 | unsigned long flags; |
969c7921 | 6354 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6355 | |
48c5ccae | 6356 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6357 | |
1da177e4 | 6358 | case CPU_UP_PREPARE: |
a468d389 | 6359 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6360 | break; |
48f24c4d | 6361 | |
1da177e4 | 6362 | case CPU_ONLINE: |
1f94ef59 | 6363 | /* Update our root-domain */ |
05fa785c | 6364 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6365 | if (rq->rd) { |
c6c4927b | 6366 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6367 | |
6368 | set_rq_online(rq); | |
1f94ef59 | 6369 | } |
05fa785c | 6370 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6371 | break; |
48f24c4d | 6372 | |
1da177e4 | 6373 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6374 | case CPU_DYING: |
317f3941 | 6375 | sched_ttwu_pending(); |
57d885fe | 6376 | /* Update our root-domain */ |
05fa785c | 6377 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6378 | if (rq->rd) { |
c6c4927b | 6379 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6380 | set_rq_offline(rq); |
57d885fe | 6381 | } |
48c5ccae PZ |
6382 | migrate_tasks(cpu); |
6383 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6384 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6385 | |
6386 | migrate_nr_uninterruptible(rq); | |
6387 | calc_global_load_remove(rq); | |
57d885fe | 6388 | break; |
1da177e4 LT |
6389 | #endif |
6390 | } | |
49c022e6 PZ |
6391 | |
6392 | update_max_interval(); | |
6393 | ||
1da177e4 LT |
6394 | return NOTIFY_OK; |
6395 | } | |
6396 | ||
f38b0820 PM |
6397 | /* |
6398 | * Register at high priority so that task migration (migrate_all_tasks) | |
6399 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6400 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6401 | */ |
26c2143b | 6402 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6403 | .notifier_call = migration_call, |
50a323b7 | 6404 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6405 | }; |
6406 | ||
3a101d05 TH |
6407 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6408 | unsigned long action, void *hcpu) | |
6409 | { | |
6410 | switch (action & ~CPU_TASKS_FROZEN) { | |
6411 | case CPU_ONLINE: | |
6412 | case CPU_DOWN_FAILED: | |
6413 | set_cpu_active((long)hcpu, true); | |
6414 | return NOTIFY_OK; | |
6415 | default: | |
6416 | return NOTIFY_DONE; | |
6417 | } | |
6418 | } | |
6419 | ||
6420 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6421 | unsigned long action, void *hcpu) | |
6422 | { | |
6423 | switch (action & ~CPU_TASKS_FROZEN) { | |
6424 | case CPU_DOWN_PREPARE: | |
6425 | set_cpu_active((long)hcpu, false); | |
6426 | return NOTIFY_OK; | |
6427 | default: | |
6428 | return NOTIFY_DONE; | |
6429 | } | |
6430 | } | |
6431 | ||
7babe8db | 6432 | static int __init migration_init(void) |
1da177e4 LT |
6433 | { |
6434 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6435 | int err; |
48f24c4d | 6436 | |
3a101d05 | 6437 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6438 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6439 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6440 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6441 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6442 | |
3a101d05 TH |
6443 | /* Register cpu active notifiers */ |
6444 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6445 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6446 | ||
a004cd42 | 6447 | return 0; |
1da177e4 | 6448 | } |
7babe8db | 6449 | early_initcall(migration_init); |
1da177e4 LT |
6450 | #endif |
6451 | ||
6452 | #ifdef CONFIG_SMP | |
476f3534 | 6453 | |
4cb98839 PZ |
6454 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
6455 | ||
3e9830dc | 6456 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6457 | |
f6630114 MT |
6458 | static __read_mostly int sched_domain_debug_enabled; |
6459 | ||
6460 | static int __init sched_domain_debug_setup(char *str) | |
6461 | { | |
6462 | sched_domain_debug_enabled = 1; | |
6463 | ||
6464 | return 0; | |
6465 | } | |
6466 | early_param("sched_debug", sched_domain_debug_setup); | |
6467 | ||
7c16ec58 | 6468 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6469 | struct cpumask *groupmask) |
1da177e4 | 6470 | { |
4dcf6aff | 6471 | struct sched_group *group = sd->groups; |
434d53b0 | 6472 | char str[256]; |
1da177e4 | 6473 | |
968ea6d8 | 6474 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6475 | cpumask_clear(groupmask); |
4dcf6aff IM |
6476 | |
6477 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6478 | ||
6479 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6480 | printk("does not load-balance\n"); |
4dcf6aff | 6481 | if (sd->parent) |
3df0fc5b PZ |
6482 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6483 | " has parent"); | |
4dcf6aff | 6484 | return -1; |
41c7ce9a NP |
6485 | } |
6486 | ||
3df0fc5b | 6487 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6488 | |
758b2cdc | 6489 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6490 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6491 | "CPU%d\n", cpu); | |
4dcf6aff | 6492 | } |
758b2cdc | 6493 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6494 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6495 | " CPU%d\n", cpu); | |
4dcf6aff | 6496 | } |
1da177e4 | 6497 | |
4dcf6aff | 6498 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6499 | do { |
4dcf6aff | 6500 | if (!group) { |
3df0fc5b PZ |
6501 | printk("\n"); |
6502 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6503 | break; |
6504 | } | |
6505 | ||
18a3885f | 6506 | if (!group->cpu_power) { |
3df0fc5b PZ |
6507 | printk(KERN_CONT "\n"); |
6508 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6509 | "set\n"); | |
4dcf6aff IM |
6510 | break; |
6511 | } | |
1da177e4 | 6512 | |
758b2cdc | 6513 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6514 | printk(KERN_CONT "\n"); |
6515 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6516 | break; |
6517 | } | |
1da177e4 | 6518 | |
758b2cdc | 6519 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6520 | printk(KERN_CONT "\n"); |
6521 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6522 | break; |
6523 | } | |
1da177e4 | 6524 | |
758b2cdc | 6525 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6526 | |
968ea6d8 | 6527 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6528 | |
3df0fc5b | 6529 | printk(KERN_CONT " %s", str); |
18a3885f | 6530 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6531 | printk(KERN_CONT " (cpu_power = %d)", |
6532 | group->cpu_power); | |
381512cf | 6533 | } |
1da177e4 | 6534 | |
4dcf6aff IM |
6535 | group = group->next; |
6536 | } while (group != sd->groups); | |
3df0fc5b | 6537 | printk(KERN_CONT "\n"); |
1da177e4 | 6538 | |
758b2cdc | 6539 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6540 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6541 | |
758b2cdc RR |
6542 | if (sd->parent && |
6543 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6544 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6545 | "of domain->span\n"); | |
4dcf6aff IM |
6546 | return 0; |
6547 | } | |
1da177e4 | 6548 | |
4dcf6aff IM |
6549 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6550 | { | |
6551 | int level = 0; | |
1da177e4 | 6552 | |
f6630114 MT |
6553 | if (!sched_domain_debug_enabled) |
6554 | return; | |
6555 | ||
4dcf6aff IM |
6556 | if (!sd) { |
6557 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6558 | return; | |
6559 | } | |
1da177e4 | 6560 | |
4dcf6aff IM |
6561 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6562 | ||
6563 | for (;;) { | |
4cb98839 | 6564 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 6565 | break; |
1da177e4 LT |
6566 | level++; |
6567 | sd = sd->parent; | |
33859f7f | 6568 | if (!sd) |
4dcf6aff IM |
6569 | break; |
6570 | } | |
1da177e4 | 6571 | } |
6d6bc0ad | 6572 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6573 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6574 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6575 | |
1a20ff27 | 6576 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6577 | { |
758b2cdc | 6578 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6579 | return 1; |
6580 | ||
6581 | /* Following flags need at least 2 groups */ | |
6582 | if (sd->flags & (SD_LOAD_BALANCE | | |
6583 | SD_BALANCE_NEWIDLE | | |
6584 | SD_BALANCE_FORK | | |
89c4710e SS |
6585 | SD_BALANCE_EXEC | |
6586 | SD_SHARE_CPUPOWER | | |
6587 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6588 | if (sd->groups != sd->groups->next) |
6589 | return 0; | |
6590 | } | |
6591 | ||
6592 | /* Following flags don't use groups */ | |
c88d5910 | 6593 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6594 | return 0; |
6595 | ||
6596 | return 1; | |
6597 | } | |
6598 | ||
48f24c4d IM |
6599 | static int |
6600 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6601 | { |
6602 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6603 | ||
6604 | if (sd_degenerate(parent)) | |
6605 | return 1; | |
6606 | ||
758b2cdc | 6607 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6608 | return 0; |
6609 | ||
245af2c7 SS |
6610 | /* Flags needing groups don't count if only 1 group in parent */ |
6611 | if (parent->groups == parent->groups->next) { | |
6612 | pflags &= ~(SD_LOAD_BALANCE | | |
6613 | SD_BALANCE_NEWIDLE | | |
6614 | SD_BALANCE_FORK | | |
89c4710e SS |
6615 | SD_BALANCE_EXEC | |
6616 | SD_SHARE_CPUPOWER | | |
6617 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6618 | if (nr_node_ids == 1) |
6619 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6620 | } |
6621 | if (~cflags & pflags) | |
6622 | return 0; | |
6623 | ||
6624 | return 1; | |
6625 | } | |
6626 | ||
dce840a0 | 6627 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 6628 | { |
dce840a0 | 6629 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 6630 | |
68e74568 | 6631 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
6632 | free_cpumask_var(rd->rto_mask); |
6633 | free_cpumask_var(rd->online); | |
6634 | free_cpumask_var(rd->span); | |
6635 | kfree(rd); | |
6636 | } | |
6637 | ||
57d885fe GH |
6638 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6639 | { | |
a0490fa3 | 6640 | struct root_domain *old_rd = NULL; |
57d885fe | 6641 | unsigned long flags; |
57d885fe | 6642 | |
05fa785c | 6643 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6644 | |
6645 | if (rq->rd) { | |
a0490fa3 | 6646 | old_rd = rq->rd; |
57d885fe | 6647 | |
c6c4927b | 6648 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6649 | set_rq_offline(rq); |
57d885fe | 6650 | |
c6c4927b | 6651 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6652 | |
a0490fa3 IM |
6653 | /* |
6654 | * If we dont want to free the old_rt yet then | |
6655 | * set old_rd to NULL to skip the freeing later | |
6656 | * in this function: | |
6657 | */ | |
6658 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6659 | old_rd = NULL; | |
57d885fe GH |
6660 | } |
6661 | ||
6662 | atomic_inc(&rd->refcount); | |
6663 | rq->rd = rd; | |
6664 | ||
c6c4927b | 6665 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6666 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6667 | set_rq_online(rq); |
57d885fe | 6668 | |
05fa785c | 6669 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6670 | |
6671 | if (old_rd) | |
dce840a0 | 6672 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
6673 | } |
6674 | ||
68c38fc3 | 6675 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6676 | { |
6677 | memset(rd, 0, sizeof(*rd)); | |
6678 | ||
68c38fc3 | 6679 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6680 | goto out; |
68c38fc3 | 6681 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6682 | goto free_span; |
68c38fc3 | 6683 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6684 | goto free_online; |
6e0534f2 | 6685 | |
68c38fc3 | 6686 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6687 | goto free_rto_mask; |
c6c4927b | 6688 | return 0; |
6e0534f2 | 6689 | |
68e74568 RR |
6690 | free_rto_mask: |
6691 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6692 | free_online: |
6693 | free_cpumask_var(rd->online); | |
6694 | free_span: | |
6695 | free_cpumask_var(rd->span); | |
0c910d28 | 6696 | out: |
c6c4927b | 6697 | return -ENOMEM; |
57d885fe GH |
6698 | } |
6699 | ||
6700 | static void init_defrootdomain(void) | |
6701 | { | |
68c38fc3 | 6702 | init_rootdomain(&def_root_domain); |
c6c4927b | 6703 | |
57d885fe GH |
6704 | atomic_set(&def_root_domain.refcount, 1); |
6705 | } | |
6706 | ||
dc938520 | 6707 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6708 | { |
6709 | struct root_domain *rd; | |
6710 | ||
6711 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6712 | if (!rd) | |
6713 | return NULL; | |
6714 | ||
68c38fc3 | 6715 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6716 | kfree(rd); |
6717 | return NULL; | |
6718 | } | |
57d885fe GH |
6719 | |
6720 | return rd; | |
6721 | } | |
6722 | ||
dce840a0 PZ |
6723 | static void free_sched_domain(struct rcu_head *rcu) |
6724 | { | |
6725 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
6726 | if (atomic_dec_and_test(&sd->groups->ref)) | |
6727 | kfree(sd->groups); | |
6728 | kfree(sd); | |
6729 | } | |
6730 | ||
6731 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
6732 | { | |
6733 | call_rcu(&sd->rcu, free_sched_domain); | |
6734 | } | |
6735 | ||
6736 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
6737 | { | |
6738 | for (; sd; sd = sd->parent) | |
6739 | destroy_sched_domain(sd, cpu); | |
6740 | } | |
6741 | ||
1da177e4 | 6742 | /* |
0eab9146 | 6743 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6744 | * hold the hotplug lock. |
6745 | */ | |
0eab9146 IM |
6746 | static void |
6747 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6748 | { |
70b97a7f | 6749 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6750 | struct sched_domain *tmp; |
6751 | ||
6752 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6753 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6754 | struct sched_domain *parent = tmp->parent; |
6755 | if (!parent) | |
6756 | break; | |
f29c9b1c | 6757 | |
1a848870 | 6758 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6759 | tmp->parent = parent->parent; |
1a848870 SS |
6760 | if (parent->parent) |
6761 | parent->parent->child = tmp; | |
dce840a0 | 6762 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6763 | } else |
6764 | tmp = tmp->parent; | |
245af2c7 SS |
6765 | } |
6766 | ||
1a848870 | 6767 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6768 | tmp = sd; |
245af2c7 | 6769 | sd = sd->parent; |
dce840a0 | 6770 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6771 | if (sd) |
6772 | sd->child = NULL; | |
6773 | } | |
1da177e4 | 6774 | |
4cb98839 | 6775 | sched_domain_debug(sd, cpu); |
1da177e4 | 6776 | |
57d885fe | 6777 | rq_attach_root(rq, rd); |
dce840a0 | 6778 | tmp = rq->sd; |
674311d5 | 6779 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6780 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
6781 | } |
6782 | ||
6783 | /* cpus with isolated domains */ | |
dcc30a35 | 6784 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6785 | |
6786 | /* Setup the mask of cpus configured for isolated domains */ | |
6787 | static int __init isolated_cpu_setup(char *str) | |
6788 | { | |
bdddd296 | 6789 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6790 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6791 | return 1; |
6792 | } | |
6793 | ||
8927f494 | 6794 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6795 | |
9c1cfda2 | 6796 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6797 | |
9c1cfda2 | 6798 | #ifdef CONFIG_NUMA |
198e2f18 | 6799 | |
9c1cfda2 JH |
6800 | /** |
6801 | * find_next_best_node - find the next node to include in a sched_domain | |
6802 | * @node: node whose sched_domain we're building | |
6803 | * @used_nodes: nodes already in the sched_domain | |
6804 | * | |
41a2d6cf | 6805 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6806 | * finds the closest node not already in the @used_nodes map. |
6807 | * | |
6808 | * Should use nodemask_t. | |
6809 | */ | |
c5f59f08 | 6810 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 6811 | { |
7142d17e | 6812 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
6813 | |
6814 | min_val = INT_MAX; | |
6815 | ||
076ac2af | 6816 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6817 | /* Start at @node */ |
076ac2af | 6818 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6819 | |
6820 | if (!nr_cpus_node(n)) | |
6821 | continue; | |
6822 | ||
6823 | /* Skip already used nodes */ | |
c5f59f08 | 6824 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6825 | continue; |
6826 | ||
6827 | /* Simple min distance search */ | |
6828 | val = node_distance(node, n); | |
6829 | ||
6830 | if (val < min_val) { | |
6831 | min_val = val; | |
6832 | best_node = n; | |
6833 | } | |
6834 | } | |
6835 | ||
7142d17e HD |
6836 | if (best_node != -1) |
6837 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
6838 | return best_node; |
6839 | } | |
6840 | ||
6841 | /** | |
6842 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6843 | * @node: node whose cpumask we're constructing | |
73486722 | 6844 | * @span: resulting cpumask |
9c1cfda2 | 6845 | * |
41a2d6cf | 6846 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6847 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6848 | * out optimally. | |
6849 | */ | |
96f874e2 | 6850 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6851 | { |
c5f59f08 | 6852 | nodemask_t used_nodes; |
48f24c4d | 6853 | int i; |
9c1cfda2 | 6854 | |
6ca09dfc | 6855 | cpumask_clear(span); |
c5f59f08 | 6856 | nodes_clear(used_nodes); |
9c1cfda2 | 6857 | |
6ca09dfc | 6858 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6859 | node_set(node, used_nodes); |
9c1cfda2 JH |
6860 | |
6861 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6862 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
6863 | if (next_node < 0) |
6864 | break; | |
6ca09dfc | 6865 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6866 | } |
9c1cfda2 | 6867 | } |
d3081f52 PZ |
6868 | |
6869 | static const struct cpumask *cpu_node_mask(int cpu) | |
6870 | { | |
6871 | lockdep_assert_held(&sched_domains_mutex); | |
6872 | ||
6873 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
6874 | ||
6875 | return sched_domains_tmpmask; | |
6876 | } | |
2c402dc3 PZ |
6877 | |
6878 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
6879 | { | |
6880 | return cpu_possible_mask; | |
6881 | } | |
6d6bc0ad | 6882 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6883 | |
d3081f52 PZ |
6884 | static const struct cpumask *cpu_cpu_mask(int cpu) |
6885 | { | |
6886 | return cpumask_of_node(cpu_to_node(cpu)); | |
6887 | } | |
6888 | ||
5c45bf27 | 6889 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6890 | |
dce840a0 PZ |
6891 | struct sd_data { |
6892 | struct sched_domain **__percpu sd; | |
6893 | struct sched_group **__percpu sg; | |
6894 | }; | |
6895 | ||
49a02c51 | 6896 | struct s_data { |
21d42ccf | 6897 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6898 | struct root_domain *rd; |
6899 | }; | |
6900 | ||
2109b99e | 6901 | enum s_alloc { |
2109b99e | 6902 | sa_rootdomain, |
21d42ccf | 6903 | sa_sd, |
dce840a0 | 6904 | sa_sd_storage, |
2109b99e AH |
6905 | sa_none, |
6906 | }; | |
6907 | ||
54ab4ff4 PZ |
6908 | struct sched_domain_topology_level; |
6909 | ||
6910 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
6911 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
6912 | ||
6913 | struct sched_domain_topology_level { | |
2c402dc3 PZ |
6914 | sched_domain_init_f init; |
6915 | sched_domain_mask_f mask; | |
54ab4ff4 | 6916 | struct sd_data data; |
eb7a74e6 PZ |
6917 | }; |
6918 | ||
9c1cfda2 | 6919 | /* |
dce840a0 | 6920 | * Assumes the sched_domain tree is fully constructed |
9c1cfda2 | 6921 | */ |
dce840a0 | 6922 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6923 | { |
dce840a0 PZ |
6924 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6925 | struct sched_domain *child = sd->child; | |
1da177e4 | 6926 | |
dce840a0 PZ |
6927 | if (child) |
6928 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6929 | |
6711cab4 | 6930 | if (sg) |
dce840a0 PZ |
6931 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
6932 | ||
6933 | return cpu; | |
1e9f28fa | 6934 | } |
1e9f28fa | 6935 | |
01a08546 | 6936 | /* |
dce840a0 PZ |
6937 | * build_sched_groups takes the cpumask we wish to span, and a pointer |
6938 | * to a function which identifies what group(along with sched group) a CPU | |
6939 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids | |
6940 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
6941 | * | |
6942 | * build_sched_groups will build a circular linked list of the groups | |
6943 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6944 | * and ->cpu_power to 0. | |
01a08546 | 6945 | */ |
dce840a0 | 6946 | static void |
f96225fd | 6947 | build_sched_groups(struct sched_domain *sd) |
1da177e4 | 6948 | { |
dce840a0 PZ |
6949 | struct sched_group *first = NULL, *last = NULL; |
6950 | struct sd_data *sdd = sd->private; | |
6951 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6952 | struct cpumask *covered; |
dce840a0 | 6953 | int i; |
9c1cfda2 | 6954 | |
f96225fd PZ |
6955 | lockdep_assert_held(&sched_domains_mutex); |
6956 | covered = sched_domains_tmpmask; | |
6957 | ||
dce840a0 | 6958 | cpumask_clear(covered); |
6711cab4 | 6959 | |
dce840a0 PZ |
6960 | for_each_cpu(i, span) { |
6961 | struct sched_group *sg; | |
6962 | int group = get_group(i, sdd, &sg); | |
6963 | int j; | |
6711cab4 | 6964 | |
dce840a0 PZ |
6965 | if (cpumask_test_cpu(i, covered)) |
6966 | continue; | |
6711cab4 | 6967 | |
dce840a0 PZ |
6968 | cpumask_clear(sched_group_cpus(sg)); |
6969 | sg->cpu_power = 0; | |
0601a88d | 6970 | |
dce840a0 PZ |
6971 | for_each_cpu(j, span) { |
6972 | if (get_group(j, sdd, NULL) != group) | |
6973 | continue; | |
0601a88d | 6974 | |
dce840a0 PZ |
6975 | cpumask_set_cpu(j, covered); |
6976 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6977 | } | |
0601a88d | 6978 | |
dce840a0 PZ |
6979 | if (!first) |
6980 | first = sg; | |
6981 | if (last) | |
6982 | last->next = sg; | |
6983 | last = sg; | |
6984 | } | |
6985 | last->next = first; | |
0601a88d | 6986 | } |
51888ca2 | 6987 | |
89c4710e SS |
6988 | /* |
6989 | * Initialize sched groups cpu_power. | |
6990 | * | |
6991 | * cpu_power indicates the capacity of sched group, which is used while | |
6992 | * distributing the load between different sched groups in a sched domain. | |
6993 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6994 | * there are asymmetries in the topology. If there are asymmetries, group | |
6995 | * having more cpu_power will pickup more load compared to the group having | |
6996 | * less cpu_power. | |
89c4710e SS |
6997 | */ |
6998 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6999 | { | |
89c4710e SS |
7000 | WARN_ON(!sd || !sd->groups); |
7001 | ||
13318a71 | 7002 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7003 | return; |
7004 | ||
aae6d3dd SS |
7005 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7006 | ||
d274cb30 | 7007 | update_group_power(sd, cpu); |
89c4710e SS |
7008 | } |
7009 | ||
7c16ec58 MT |
7010 | /* |
7011 | * Initializers for schedule domains | |
7012 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7013 | */ | |
7014 | ||
a5d8c348 IM |
7015 | #ifdef CONFIG_SCHED_DEBUG |
7016 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7017 | #else | |
7018 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7019 | #endif | |
7020 | ||
54ab4ff4 PZ |
7021 | #define SD_INIT_FUNC(type) \ |
7022 | static noinline struct sched_domain * \ | |
7023 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
7024 | { \ | |
7025 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
7026 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
7027 | SD_INIT_NAME(sd, type); \ |
7028 | sd->private = &tl->data; \ | |
7029 | return sd; \ | |
7c16ec58 MT |
7030 | } |
7031 | ||
7032 | SD_INIT_FUNC(CPU) | |
7033 | #ifdef CONFIG_NUMA | |
7034 | SD_INIT_FUNC(ALLNODES) | |
7035 | SD_INIT_FUNC(NODE) | |
7036 | #endif | |
7037 | #ifdef CONFIG_SCHED_SMT | |
7038 | SD_INIT_FUNC(SIBLING) | |
7039 | #endif | |
7040 | #ifdef CONFIG_SCHED_MC | |
7041 | SD_INIT_FUNC(MC) | |
7042 | #endif | |
01a08546 HC |
7043 | #ifdef CONFIG_SCHED_BOOK |
7044 | SD_INIT_FUNC(BOOK) | |
7045 | #endif | |
7c16ec58 | 7046 | |
1d3504fc | 7047 | static int default_relax_domain_level = -1; |
60495e77 | 7048 | int sched_domain_level_max; |
1d3504fc HS |
7049 | |
7050 | static int __init setup_relax_domain_level(char *str) | |
7051 | { | |
30e0e178 LZ |
7052 | unsigned long val; |
7053 | ||
7054 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 7055 | if (val < sched_domain_level_max) |
30e0e178 LZ |
7056 | default_relax_domain_level = val; |
7057 | ||
1d3504fc HS |
7058 | return 1; |
7059 | } | |
7060 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7061 | ||
7062 | static void set_domain_attribute(struct sched_domain *sd, | |
7063 | struct sched_domain_attr *attr) | |
7064 | { | |
7065 | int request; | |
7066 | ||
7067 | if (!attr || attr->relax_domain_level < 0) { | |
7068 | if (default_relax_domain_level < 0) | |
7069 | return; | |
7070 | else | |
7071 | request = default_relax_domain_level; | |
7072 | } else | |
7073 | request = attr->relax_domain_level; | |
7074 | if (request < sd->level) { | |
7075 | /* turn off idle balance on this domain */ | |
c88d5910 | 7076 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7077 | } else { |
7078 | /* turn on idle balance on this domain */ | |
c88d5910 | 7079 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7080 | } |
7081 | } | |
7082 | ||
54ab4ff4 PZ |
7083 | static void __sdt_free(const struct cpumask *cpu_map); |
7084 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
7085 | ||
2109b99e AH |
7086 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7087 | const struct cpumask *cpu_map) | |
7088 | { | |
7089 | switch (what) { | |
2109b99e | 7090 | case sa_rootdomain: |
822ff793 PZ |
7091 | if (!atomic_read(&d->rd->refcount)) |
7092 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
7093 | case sa_sd: |
7094 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 7095 | case sa_sd_storage: |
54ab4ff4 | 7096 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
7097 | case sa_none: |
7098 | break; | |
7099 | } | |
7100 | } | |
3404c8d9 | 7101 | |
2109b99e AH |
7102 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7103 | const struct cpumask *cpu_map) | |
7104 | { | |
dce840a0 PZ |
7105 | memset(d, 0, sizeof(*d)); |
7106 | ||
54ab4ff4 PZ |
7107 | if (__sdt_alloc(cpu_map)) |
7108 | return sa_sd_storage; | |
dce840a0 PZ |
7109 | d->sd = alloc_percpu(struct sched_domain *); |
7110 | if (!d->sd) | |
7111 | return sa_sd_storage; | |
2109b99e | 7112 | d->rd = alloc_rootdomain(); |
dce840a0 | 7113 | if (!d->rd) |
21d42ccf | 7114 | return sa_sd; |
2109b99e AH |
7115 | return sa_rootdomain; |
7116 | } | |
57d885fe | 7117 | |
dce840a0 PZ |
7118 | /* |
7119 | * NULL the sd_data elements we've used to build the sched_domain and | |
7120 | * sched_group structure so that the subsequent __free_domain_allocs() | |
7121 | * will not free the data we're using. | |
7122 | */ | |
7123 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
7124 | { | |
7125 | struct sd_data *sdd = sd->private; | |
7126 | struct sched_group *sg = sd->groups; | |
7127 | ||
7128 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
7129 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
7130 | ||
7131 | if (cpu == cpumask_first(sched_group_cpus(sg))) { | |
7132 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sg, cpu) != sg); | |
7133 | *per_cpu_ptr(sdd->sg, cpu) = NULL; | |
7134 | } | |
7135 | } | |
7136 | ||
2c402dc3 PZ |
7137 | #ifdef CONFIG_SCHED_SMT |
7138 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 7139 | { |
2c402dc3 | 7140 | return topology_thread_cpumask(cpu); |
3bd65a80 | 7141 | } |
2c402dc3 | 7142 | #endif |
7f4588f3 | 7143 | |
d069b916 PZ |
7144 | /* |
7145 | * Topology list, bottom-up. | |
7146 | */ | |
2c402dc3 | 7147 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
7148 | #ifdef CONFIG_SCHED_SMT |
7149 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 7150 | #endif |
1e9f28fa | 7151 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 7152 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 7153 | #endif |
d069b916 PZ |
7154 | #ifdef CONFIG_SCHED_BOOK |
7155 | { sd_init_BOOK, cpu_book_mask, }, | |
7156 | #endif | |
7157 | { sd_init_CPU, cpu_cpu_mask, }, | |
7158 | #ifdef CONFIG_NUMA | |
7159 | { sd_init_NODE, cpu_node_mask, }, | |
7160 | { sd_init_ALLNODES, cpu_allnodes_mask, }, | |
1da177e4 | 7161 | #endif |
eb7a74e6 PZ |
7162 | { NULL, }, |
7163 | }; | |
7164 | ||
7165 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
7166 | ||
54ab4ff4 PZ |
7167 | static int __sdt_alloc(const struct cpumask *cpu_map) |
7168 | { | |
7169 | struct sched_domain_topology_level *tl; | |
7170 | int j; | |
7171 | ||
7172 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7173 | struct sd_data *sdd = &tl->data; | |
7174 | ||
7175 | sdd->sd = alloc_percpu(struct sched_domain *); | |
7176 | if (!sdd->sd) | |
7177 | return -ENOMEM; | |
7178 | ||
7179 | sdd->sg = alloc_percpu(struct sched_group *); | |
7180 | if (!sdd->sg) | |
7181 | return -ENOMEM; | |
7182 | ||
7183 | for_each_cpu(j, cpu_map) { | |
7184 | struct sched_domain *sd; | |
7185 | struct sched_group *sg; | |
7186 | ||
7187 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
7188 | GFP_KERNEL, cpu_to_node(j)); | |
7189 | if (!sd) | |
7190 | return -ENOMEM; | |
7191 | ||
7192 | *per_cpu_ptr(sdd->sd, j) = sd; | |
7193 | ||
7194 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7195 | GFP_KERNEL, cpu_to_node(j)); | |
7196 | if (!sg) | |
7197 | return -ENOMEM; | |
7198 | ||
7199 | *per_cpu_ptr(sdd->sg, j) = sg; | |
7200 | } | |
7201 | } | |
7202 | ||
7203 | return 0; | |
7204 | } | |
7205 | ||
7206 | static void __sdt_free(const struct cpumask *cpu_map) | |
7207 | { | |
7208 | struct sched_domain_topology_level *tl; | |
7209 | int j; | |
7210 | ||
7211 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7212 | struct sd_data *sdd = &tl->data; | |
7213 | ||
7214 | for_each_cpu(j, cpu_map) { | |
7215 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
7216 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
7217 | } | |
7218 | free_percpu(sdd->sd); | |
7219 | free_percpu(sdd->sg); | |
7220 | } | |
7221 | } | |
7222 | ||
2c402dc3 PZ |
7223 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
7224 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 7225 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
7226 | int cpu) |
7227 | { | |
54ab4ff4 | 7228 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 7229 | if (!sd) |
d069b916 | 7230 | return child; |
2c402dc3 PZ |
7231 | |
7232 | set_domain_attribute(sd, attr); | |
7233 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
7234 | if (child) { |
7235 | sd->level = child->level + 1; | |
7236 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 7237 | child->parent = sd; |
60495e77 | 7238 | } |
d069b916 | 7239 | sd->child = child; |
2c402dc3 PZ |
7240 | |
7241 | return sd; | |
7242 | } | |
7243 | ||
2109b99e AH |
7244 | /* |
7245 | * Build sched domains for a given set of cpus and attach the sched domains | |
7246 | * to the individual cpus | |
7247 | */ | |
dce840a0 PZ |
7248 | static int build_sched_domains(const struct cpumask *cpu_map, |
7249 | struct sched_domain_attr *attr) | |
2109b99e AH |
7250 | { |
7251 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 7252 | struct sched_domain *sd; |
2109b99e | 7253 | struct s_data d; |
822ff793 | 7254 | int i, ret = -ENOMEM; |
9c1cfda2 | 7255 | |
2109b99e AH |
7256 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7257 | if (alloc_state != sa_rootdomain) | |
7258 | goto error; | |
9c1cfda2 | 7259 | |
dce840a0 | 7260 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7261 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7262 | struct sched_domain_topology_level *tl; |
7263 | ||
3bd65a80 | 7264 | sd = NULL; |
2c402dc3 PZ |
7265 | for (tl = sched_domain_topology; tl->init; tl++) |
7266 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); | |
d274cb30 | 7267 | |
d069b916 PZ |
7268 | while (sd->child) |
7269 | sd = sd->child; | |
7270 | ||
21d42ccf | 7271 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
7272 | } |
7273 | ||
7274 | /* Build the groups for the domains */ | |
7275 | for_each_cpu(i, cpu_map) { | |
7276 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7277 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
7278 | get_group(i, sd->private, &sd->groups); | |
7279 | atomic_inc(&sd->groups->ref); | |
21d42ccf | 7280 | |
dce840a0 PZ |
7281 | if (i != cpumask_first(sched_domain_span(sd))) |
7282 | continue; | |
7283 | ||
f96225fd | 7284 | build_sched_groups(sd); |
1cf51902 | 7285 | } |
a06dadbe | 7286 | } |
9c1cfda2 | 7287 | |
1da177e4 | 7288 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
7289 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7290 | if (!cpumask_test_cpu(i, cpu_map)) | |
7291 | continue; | |
9c1cfda2 | 7292 | |
dce840a0 PZ |
7293 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7294 | claim_allocations(i, sd); | |
cd4ea6ae | 7295 | init_sched_groups_power(i, sd); |
dce840a0 | 7296 | } |
f712c0c7 | 7297 | } |
9c1cfda2 | 7298 | |
1da177e4 | 7299 | /* Attach the domains */ |
dce840a0 | 7300 | rcu_read_lock(); |
abcd083a | 7301 | for_each_cpu(i, cpu_map) { |
21d42ccf | 7302 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 7303 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7304 | } |
dce840a0 | 7305 | rcu_read_unlock(); |
51888ca2 | 7306 | |
822ff793 | 7307 | ret = 0; |
51888ca2 | 7308 | error: |
2109b99e | 7309 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7310 | return ret; |
1da177e4 | 7311 | } |
029190c5 | 7312 | |
acc3f5d7 | 7313 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7314 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7315 | static struct sched_domain_attr *dattr_cur; |
7316 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7317 | |
7318 | /* | |
7319 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7320 | * cpumask) fails, then fallback to a single sched domain, |
7321 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7322 | */ |
4212823f | 7323 | static cpumask_var_t fallback_doms; |
029190c5 | 7324 | |
ee79d1bd HC |
7325 | /* |
7326 | * arch_update_cpu_topology lets virtualized architectures update the | |
7327 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7328 | * or 0 if it stayed the same. | |
7329 | */ | |
7330 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7331 | { |
ee79d1bd | 7332 | return 0; |
22e52b07 HC |
7333 | } |
7334 | ||
acc3f5d7 RR |
7335 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7336 | { | |
7337 | int i; | |
7338 | cpumask_var_t *doms; | |
7339 | ||
7340 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7341 | if (!doms) | |
7342 | return NULL; | |
7343 | for (i = 0; i < ndoms; i++) { | |
7344 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7345 | free_sched_domains(doms, i); | |
7346 | return NULL; | |
7347 | } | |
7348 | } | |
7349 | return doms; | |
7350 | } | |
7351 | ||
7352 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7353 | { | |
7354 | unsigned int i; | |
7355 | for (i = 0; i < ndoms; i++) | |
7356 | free_cpumask_var(doms[i]); | |
7357 | kfree(doms); | |
7358 | } | |
7359 | ||
1a20ff27 | 7360 | /* |
41a2d6cf | 7361 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7362 | * For now this just excludes isolated cpus, but could be used to |
7363 | * exclude other special cases in the future. | |
1a20ff27 | 7364 | */ |
c4a8849a | 7365 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7366 | { |
7378547f MM |
7367 | int err; |
7368 | ||
22e52b07 | 7369 | arch_update_cpu_topology(); |
029190c5 | 7370 | ndoms_cur = 1; |
acc3f5d7 | 7371 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7372 | if (!doms_cur) |
acc3f5d7 RR |
7373 | doms_cur = &fallback_doms; |
7374 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7375 | dattr_cur = NULL; |
dce840a0 | 7376 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7377 | register_sched_domain_sysctl(); |
7378547f MM |
7378 | |
7379 | return err; | |
1a20ff27 DG |
7380 | } |
7381 | ||
1a20ff27 DG |
7382 | /* |
7383 | * Detach sched domains from a group of cpus specified in cpu_map | |
7384 | * These cpus will now be attached to the NULL domain | |
7385 | */ | |
96f874e2 | 7386 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7387 | { |
7388 | int i; | |
7389 | ||
dce840a0 | 7390 | rcu_read_lock(); |
abcd083a | 7391 | for_each_cpu(i, cpu_map) |
57d885fe | 7392 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7393 | rcu_read_unlock(); |
1a20ff27 DG |
7394 | } |
7395 | ||
1d3504fc HS |
7396 | /* handle null as "default" */ |
7397 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7398 | struct sched_domain_attr *new, int idx_new) | |
7399 | { | |
7400 | struct sched_domain_attr tmp; | |
7401 | ||
7402 | /* fast path */ | |
7403 | if (!new && !cur) | |
7404 | return 1; | |
7405 | ||
7406 | tmp = SD_ATTR_INIT; | |
7407 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7408 | new ? (new + idx_new) : &tmp, | |
7409 | sizeof(struct sched_domain_attr)); | |
7410 | } | |
7411 | ||
029190c5 PJ |
7412 | /* |
7413 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7414 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7415 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7416 | * It destroys each deleted domain and builds each new domain. | |
7417 | * | |
acc3f5d7 | 7418 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7419 | * The masks don't intersect (don't overlap.) We should setup one |
7420 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7421 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7422 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7423 | * it as it is. | |
7424 | * | |
acc3f5d7 RR |
7425 | * The passed in 'doms_new' should be allocated using |
7426 | * alloc_sched_domains. This routine takes ownership of it and will | |
7427 | * free_sched_domains it when done with it. If the caller failed the | |
7428 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7429 | * and partition_sched_domains() will fallback to the single partition | |
7430 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7431 | * |
96f874e2 | 7432 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7433 | * ndoms_new == 0 is a special case for destroying existing domains, |
7434 | * and it will not create the default domain. | |
dfb512ec | 7435 | * |
029190c5 PJ |
7436 | * Call with hotplug lock held |
7437 | */ | |
acc3f5d7 | 7438 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7439 | struct sched_domain_attr *dattr_new) |
029190c5 | 7440 | { |
dfb512ec | 7441 | int i, j, n; |
d65bd5ec | 7442 | int new_topology; |
029190c5 | 7443 | |
712555ee | 7444 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7445 | |
7378547f MM |
7446 | /* always unregister in case we don't destroy any domains */ |
7447 | unregister_sched_domain_sysctl(); | |
7448 | ||
d65bd5ec HC |
7449 | /* Let architecture update cpu core mappings. */ |
7450 | new_topology = arch_update_cpu_topology(); | |
7451 | ||
dfb512ec | 7452 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7453 | |
7454 | /* Destroy deleted domains */ | |
7455 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7456 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7457 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7458 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7459 | goto match1; |
7460 | } | |
7461 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7462 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7463 | match1: |
7464 | ; | |
7465 | } | |
7466 | ||
e761b772 MK |
7467 | if (doms_new == NULL) { |
7468 | ndoms_cur = 0; | |
acc3f5d7 | 7469 | doms_new = &fallback_doms; |
6ad4c188 | 7470 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7471 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7472 | } |
7473 | ||
029190c5 PJ |
7474 | /* Build new domains */ |
7475 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7476 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7477 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7478 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7479 | goto match2; |
7480 | } | |
7481 | /* no match - add a new doms_new */ | |
dce840a0 | 7482 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7483 | match2: |
7484 | ; | |
7485 | } | |
7486 | ||
7487 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7488 | if (doms_cur != &fallback_doms) |
7489 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7490 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7491 | doms_cur = doms_new; |
1d3504fc | 7492 | dattr_cur = dattr_new; |
029190c5 | 7493 | ndoms_cur = ndoms_new; |
7378547f MM |
7494 | |
7495 | register_sched_domain_sysctl(); | |
a1835615 | 7496 | |
712555ee | 7497 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7498 | } |
7499 | ||
5c45bf27 | 7500 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 7501 | static void reinit_sched_domains(void) |
5c45bf27 | 7502 | { |
95402b38 | 7503 | get_online_cpus(); |
dfb512ec MK |
7504 | |
7505 | /* Destroy domains first to force the rebuild */ | |
7506 | partition_sched_domains(0, NULL, NULL); | |
7507 | ||
e761b772 | 7508 | rebuild_sched_domains(); |
95402b38 | 7509 | put_online_cpus(); |
5c45bf27 SS |
7510 | } |
7511 | ||
7512 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7513 | { | |
afb8a9b7 | 7514 | unsigned int level = 0; |
5c45bf27 | 7515 | |
afb8a9b7 GS |
7516 | if (sscanf(buf, "%u", &level) != 1) |
7517 | return -EINVAL; | |
7518 | ||
7519 | /* | |
7520 | * level is always be positive so don't check for | |
7521 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7522 | * What happens on 0 or 1 byte write, | |
7523 | * need to check for count as well? | |
7524 | */ | |
7525 | ||
7526 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7527 | return -EINVAL; |
7528 | ||
7529 | if (smt) | |
afb8a9b7 | 7530 | sched_smt_power_savings = level; |
5c45bf27 | 7531 | else |
afb8a9b7 | 7532 | sched_mc_power_savings = level; |
5c45bf27 | 7533 | |
c4a8849a | 7534 | reinit_sched_domains(); |
5c45bf27 | 7535 | |
c70f22d2 | 7536 | return count; |
5c45bf27 SS |
7537 | } |
7538 | ||
5c45bf27 | 7539 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7540 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7541 | struct sysdev_class_attribute *attr, |
f718cd4a | 7542 | char *page) |
5c45bf27 SS |
7543 | { |
7544 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7545 | } | |
f718cd4a | 7546 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7547 | struct sysdev_class_attribute *attr, |
48f24c4d | 7548 | const char *buf, size_t count) |
5c45bf27 SS |
7549 | { |
7550 | return sched_power_savings_store(buf, count, 0); | |
7551 | } | |
f718cd4a AK |
7552 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7553 | sched_mc_power_savings_show, | |
7554 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7555 | #endif |
7556 | ||
7557 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7558 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7559 | struct sysdev_class_attribute *attr, |
f718cd4a | 7560 | char *page) |
5c45bf27 SS |
7561 | { |
7562 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7563 | } | |
f718cd4a | 7564 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7565 | struct sysdev_class_attribute *attr, |
48f24c4d | 7566 | const char *buf, size_t count) |
5c45bf27 SS |
7567 | { |
7568 | return sched_power_savings_store(buf, count, 1); | |
7569 | } | |
f718cd4a AK |
7570 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7571 | sched_smt_power_savings_show, | |
6707de00 AB |
7572 | sched_smt_power_savings_store); |
7573 | #endif | |
7574 | ||
39aac648 | 7575 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7576 | { |
7577 | int err = 0; | |
7578 | ||
7579 | #ifdef CONFIG_SCHED_SMT | |
7580 | if (smt_capable()) | |
7581 | err = sysfs_create_file(&cls->kset.kobj, | |
7582 | &attr_sched_smt_power_savings.attr); | |
7583 | #endif | |
7584 | #ifdef CONFIG_SCHED_MC | |
7585 | if (!err && mc_capable()) | |
7586 | err = sysfs_create_file(&cls->kset.kobj, | |
7587 | &attr_sched_mc_power_savings.attr); | |
7588 | #endif | |
7589 | return err; | |
7590 | } | |
6d6bc0ad | 7591 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7592 | |
1da177e4 | 7593 | /* |
3a101d05 TH |
7594 | * Update cpusets according to cpu_active mask. If cpusets are |
7595 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7596 | * around partition_sched_domains(). | |
1da177e4 | 7597 | */ |
0b2e918a TH |
7598 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7599 | void *hcpu) | |
e761b772 | 7600 | { |
3a101d05 | 7601 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7602 | case CPU_ONLINE: |
6ad4c188 | 7603 | case CPU_DOWN_FAILED: |
3a101d05 | 7604 | cpuset_update_active_cpus(); |
e761b772 | 7605 | return NOTIFY_OK; |
3a101d05 TH |
7606 | default: |
7607 | return NOTIFY_DONE; | |
7608 | } | |
7609 | } | |
e761b772 | 7610 | |
0b2e918a TH |
7611 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7612 | void *hcpu) | |
3a101d05 TH |
7613 | { |
7614 | switch (action & ~CPU_TASKS_FROZEN) { | |
7615 | case CPU_DOWN_PREPARE: | |
7616 | cpuset_update_active_cpus(); | |
7617 | return NOTIFY_OK; | |
e761b772 MK |
7618 | default: |
7619 | return NOTIFY_DONE; | |
7620 | } | |
7621 | } | |
e761b772 MK |
7622 | |
7623 | static int update_runtime(struct notifier_block *nfb, | |
7624 | unsigned long action, void *hcpu) | |
1da177e4 | 7625 | { |
7def2be1 PZ |
7626 | int cpu = (int)(long)hcpu; |
7627 | ||
1da177e4 | 7628 | switch (action) { |
1da177e4 | 7629 | case CPU_DOWN_PREPARE: |
8bb78442 | 7630 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7631 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7632 | return NOTIFY_OK; |
7633 | ||
1da177e4 | 7634 | case CPU_DOWN_FAILED: |
8bb78442 | 7635 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7636 | case CPU_ONLINE: |
8bb78442 | 7637 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7638 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7639 | return NOTIFY_OK; |
7640 | ||
1da177e4 LT |
7641 | default: |
7642 | return NOTIFY_DONE; | |
7643 | } | |
1da177e4 | 7644 | } |
1da177e4 LT |
7645 | |
7646 | void __init sched_init_smp(void) | |
7647 | { | |
dcc30a35 RR |
7648 | cpumask_var_t non_isolated_cpus; |
7649 | ||
7650 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7651 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7652 | |
95402b38 | 7653 | get_online_cpus(); |
712555ee | 7654 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7655 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7656 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7657 | if (cpumask_empty(non_isolated_cpus)) | |
7658 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7659 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7660 | put_online_cpus(); |
e761b772 | 7661 | |
3a101d05 TH |
7662 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7663 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7664 | |
7665 | /* RT runtime code needs to handle some hotplug events */ | |
7666 | hotcpu_notifier(update_runtime, 0); | |
7667 | ||
b328ca18 | 7668 | init_hrtick(); |
5c1e1767 NP |
7669 | |
7670 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7671 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7672 | BUG(); |
19978ca6 | 7673 | sched_init_granularity(); |
dcc30a35 | 7674 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7675 | |
0e3900e6 | 7676 | init_sched_rt_class(); |
1da177e4 LT |
7677 | } |
7678 | #else | |
7679 | void __init sched_init_smp(void) | |
7680 | { | |
19978ca6 | 7681 | sched_init_granularity(); |
1da177e4 LT |
7682 | } |
7683 | #endif /* CONFIG_SMP */ | |
7684 | ||
cd1bb94b AB |
7685 | const_debug unsigned int sysctl_timer_migration = 1; |
7686 | ||
1da177e4 LT |
7687 | int in_sched_functions(unsigned long addr) |
7688 | { | |
1da177e4 LT |
7689 | return in_lock_functions(addr) || |
7690 | (addr >= (unsigned long)__sched_text_start | |
7691 | && addr < (unsigned long)__sched_text_end); | |
7692 | } | |
7693 | ||
a9957449 | 7694 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7695 | { |
7696 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7697 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7698 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7699 | cfs_rq->rq = rq; | |
f07333bf | 7700 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 7701 | #ifdef CONFIG_SMP |
f07333bf | 7702 | cfs_rq->load_stamp = 1; |
6ea72f12 | 7703 | #endif |
dd41f596 | 7704 | #endif |
67e9fb2a | 7705 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7706 | } |
7707 | ||
fa85ae24 PZ |
7708 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7709 | { | |
7710 | struct rt_prio_array *array; | |
7711 | int i; | |
7712 | ||
7713 | array = &rt_rq->active; | |
7714 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7715 | INIT_LIST_HEAD(array->queue + i); | |
7716 | __clear_bit(i, array->bitmap); | |
7717 | } | |
7718 | /* delimiter for bitsearch: */ | |
7719 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7720 | ||
052f1dc7 | 7721 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7722 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7723 | #ifdef CONFIG_SMP |
e864c499 | 7724 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7725 | #endif |
48d5e258 | 7726 | #endif |
fa85ae24 PZ |
7727 | #ifdef CONFIG_SMP |
7728 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7729 | rt_rq->overloaded = 0; |
05fa785c | 7730 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7731 | #endif |
7732 | ||
7733 | rt_rq->rt_time = 0; | |
7734 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7735 | rt_rq->rt_runtime = 0; |
0986b11b | 7736 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7737 | |
052f1dc7 | 7738 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7739 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7740 | rt_rq->rq = rq; |
7741 | #endif | |
fa85ae24 PZ |
7742 | } |
7743 | ||
6f505b16 | 7744 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7745 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7746 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7747 | struct sched_entity *parent) |
6f505b16 | 7748 | { |
ec7dc8ac | 7749 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7750 | tg->cfs_rq[cpu] = cfs_rq; |
7751 | init_cfs_rq(cfs_rq, rq); | |
7752 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7753 | |
7754 | tg->se[cpu] = se; | |
07e06b01 | 7755 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
7756 | if (!se) |
7757 | return; | |
7758 | ||
ec7dc8ac DG |
7759 | if (!parent) |
7760 | se->cfs_rq = &rq->cfs; | |
7761 | else | |
7762 | se->cfs_rq = parent->my_q; | |
7763 | ||
6f505b16 | 7764 | se->my_q = cfs_rq; |
9437178f | 7765 | update_load_set(&se->load, 0); |
ec7dc8ac | 7766 | se->parent = parent; |
6f505b16 | 7767 | } |
052f1dc7 | 7768 | #endif |
6f505b16 | 7769 | |
052f1dc7 | 7770 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7771 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7772 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7773 | struct sched_rt_entity *parent) |
6f505b16 | 7774 | { |
ec7dc8ac DG |
7775 | struct rq *rq = cpu_rq(cpu); |
7776 | ||
6f505b16 PZ |
7777 | tg->rt_rq[cpu] = rt_rq; |
7778 | init_rt_rq(rt_rq, rq); | |
7779 | rt_rq->tg = tg; | |
ac086bc2 | 7780 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7781 | |
7782 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7783 | if (!rt_se) |
7784 | return; | |
7785 | ||
ec7dc8ac DG |
7786 | if (!parent) |
7787 | rt_se->rt_rq = &rq->rt; | |
7788 | else | |
7789 | rt_se->rt_rq = parent->my_q; | |
7790 | ||
6f505b16 | 7791 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7792 | rt_se->parent = parent; |
6f505b16 PZ |
7793 | INIT_LIST_HEAD(&rt_se->run_list); |
7794 | } | |
7795 | #endif | |
7796 | ||
1da177e4 LT |
7797 | void __init sched_init(void) |
7798 | { | |
dd41f596 | 7799 | int i, j; |
434d53b0 MT |
7800 | unsigned long alloc_size = 0, ptr; |
7801 | ||
7802 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7803 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7804 | #endif | |
7805 | #ifdef CONFIG_RT_GROUP_SCHED | |
7806 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7807 | #endif |
df7c8e84 | 7808 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7809 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7810 | #endif |
434d53b0 | 7811 | if (alloc_size) { |
36b7b6d4 | 7812 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7813 | |
7814 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7815 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7816 | ptr += nr_cpu_ids * sizeof(void **); |
7817 | ||
07e06b01 | 7818 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7819 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7820 | |
6d6bc0ad | 7821 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7822 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7823 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7824 | ptr += nr_cpu_ids * sizeof(void **); |
7825 | ||
07e06b01 | 7826 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7827 | ptr += nr_cpu_ids * sizeof(void **); |
7828 | ||
6d6bc0ad | 7829 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7830 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7831 | for_each_possible_cpu(i) { | |
7832 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7833 | ptr += cpumask_size(); | |
7834 | } | |
7835 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7836 | } |
dd41f596 | 7837 | |
57d885fe GH |
7838 | #ifdef CONFIG_SMP |
7839 | init_defrootdomain(); | |
7840 | #endif | |
7841 | ||
d0b27fa7 PZ |
7842 | init_rt_bandwidth(&def_rt_bandwidth, |
7843 | global_rt_period(), global_rt_runtime()); | |
7844 | ||
7845 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7846 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7847 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7848 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7849 | |
7c941438 | 7850 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7851 | list_add(&root_task_group.list, &task_groups); |
7852 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 7853 | autogroup_init(&init_task); |
7c941438 | 7854 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7855 | |
0a945022 | 7856 | for_each_possible_cpu(i) { |
70b97a7f | 7857 | struct rq *rq; |
1da177e4 LT |
7858 | |
7859 | rq = cpu_rq(i); | |
05fa785c | 7860 | raw_spin_lock_init(&rq->lock); |
7897986b | 7861 | rq->nr_running = 0; |
dce48a84 TG |
7862 | rq->calc_load_active = 0; |
7863 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7864 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7865 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7866 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 7867 | root_task_group.shares = root_task_group_load; |
6f505b16 | 7868 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7869 | /* |
07e06b01 | 7870 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7871 | * |
7872 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7873 | * gets 100% of the cpu resources in the system. This overall | |
7874 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7875 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7876 | * based on each entity's (task or task-group's) weight |
7877 | * (se->load.weight). | |
7878 | * | |
07e06b01 | 7879 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7880 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7881 | * then A0's share of the cpu resource is: | |
7882 | * | |
0d905bca | 7883 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7884 | * |
07e06b01 YZ |
7885 | * We achieve this by letting root_task_group's tasks sit |
7886 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7887 | */ |
07e06b01 | 7888 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7889 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7890 | ||
7891 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7892 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7893 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7894 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7895 | #endif |
1da177e4 | 7896 | |
dd41f596 IM |
7897 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7898 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7899 | |
7900 | rq->last_load_update_tick = jiffies; | |
7901 | ||
1da177e4 | 7902 | #ifdef CONFIG_SMP |
41c7ce9a | 7903 | rq->sd = NULL; |
57d885fe | 7904 | rq->rd = NULL; |
e51fd5e2 | 7905 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7906 | rq->post_schedule = 0; |
1da177e4 | 7907 | rq->active_balance = 0; |
dd41f596 | 7908 | rq->next_balance = jiffies; |
1da177e4 | 7909 | rq->push_cpu = 0; |
0a2966b4 | 7910 | rq->cpu = i; |
1f11eb6a | 7911 | rq->online = 0; |
eae0c9df MG |
7912 | rq->idle_stamp = 0; |
7913 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7914 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7915 | #ifdef CONFIG_NO_HZ |
7916 | rq->nohz_balance_kick = 0; | |
7917 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7918 | #endif | |
1da177e4 | 7919 | #endif |
8f4d37ec | 7920 | init_rq_hrtick(rq); |
1da177e4 | 7921 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7922 | } |
7923 | ||
2dd73a4f | 7924 | set_load_weight(&init_task); |
b50f60ce | 7925 | |
e107be36 AK |
7926 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7927 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7928 | #endif | |
7929 | ||
c9819f45 | 7930 | #ifdef CONFIG_SMP |
962cf36c | 7931 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7932 | #endif |
7933 | ||
b50f60ce | 7934 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7935 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7936 | #endif |
7937 | ||
1da177e4 LT |
7938 | /* |
7939 | * The boot idle thread does lazy MMU switching as well: | |
7940 | */ | |
7941 | atomic_inc(&init_mm.mm_count); | |
7942 | enter_lazy_tlb(&init_mm, current); | |
7943 | ||
7944 | /* | |
7945 | * Make us the idle thread. Technically, schedule() should not be | |
7946 | * called from this thread, however somewhere below it might be, | |
7947 | * but because we are the idle thread, we just pick up running again | |
7948 | * when this runqueue becomes "idle". | |
7949 | */ | |
7950 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7951 | |
7952 | calc_load_update = jiffies + LOAD_FREQ; | |
7953 | ||
dd41f596 IM |
7954 | /* |
7955 | * During early bootup we pretend to be a normal task: | |
7956 | */ | |
7957 | current->sched_class = &fair_sched_class; | |
6892b75e | 7958 | |
6a7b3dc3 | 7959 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7960 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7961 | #ifdef CONFIG_SMP |
4cb98839 | 7962 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
7d1e6a9b | 7963 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
7964 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
7965 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
7966 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
7967 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
7968 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 7969 | #endif |
bdddd296 RR |
7970 | /* May be allocated at isolcpus cmdline parse time */ |
7971 | if (cpu_isolated_map == NULL) | |
7972 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7973 | #endif /* SMP */ |
6a7b3dc3 | 7974 | |
6892b75e | 7975 | scheduler_running = 1; |
1da177e4 LT |
7976 | } |
7977 | ||
7978 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7979 | static inline int preempt_count_equals(int preempt_offset) |
7980 | { | |
234da7bc | 7981 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7982 | |
4ba8216c | 7983 | return (nested == preempt_offset); |
e4aafea2 FW |
7984 | } |
7985 | ||
d894837f | 7986 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7987 | { |
48f24c4d | 7988 | #ifdef in_atomic |
1da177e4 LT |
7989 | static unsigned long prev_jiffy; /* ratelimiting */ |
7990 | ||
e4aafea2 FW |
7991 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7992 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7993 | return; |
7994 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7995 | return; | |
7996 | prev_jiffy = jiffies; | |
7997 | ||
3df0fc5b PZ |
7998 | printk(KERN_ERR |
7999 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8000 | file, line); | |
8001 | printk(KERN_ERR | |
8002 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8003 | in_atomic(), irqs_disabled(), | |
8004 | current->pid, current->comm); | |
aef745fc IM |
8005 | |
8006 | debug_show_held_locks(current); | |
8007 | if (irqs_disabled()) | |
8008 | print_irqtrace_events(current); | |
8009 | dump_stack(); | |
1da177e4 LT |
8010 | #endif |
8011 | } | |
8012 | EXPORT_SYMBOL(__might_sleep); | |
8013 | #endif | |
8014 | ||
8015 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8016 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8017 | { | |
da7a735e PZ |
8018 | const struct sched_class *prev_class = p->sched_class; |
8019 | int old_prio = p->prio; | |
3a5e4dc1 | 8020 | int on_rq; |
3e51f33f | 8021 | |
fd2f4419 | 8022 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8023 | if (on_rq) |
8024 | deactivate_task(rq, p, 0); | |
8025 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8026 | if (on_rq) { | |
8027 | activate_task(rq, p, 0); | |
8028 | resched_task(rq->curr); | |
8029 | } | |
da7a735e PZ |
8030 | |
8031 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8032 | } |
8033 | ||
1da177e4 LT |
8034 | void normalize_rt_tasks(void) |
8035 | { | |
a0f98a1c | 8036 | struct task_struct *g, *p; |
1da177e4 | 8037 | unsigned long flags; |
70b97a7f | 8038 | struct rq *rq; |
1da177e4 | 8039 | |
4cf5d77a | 8040 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8041 | do_each_thread(g, p) { |
178be793 IM |
8042 | /* |
8043 | * Only normalize user tasks: | |
8044 | */ | |
8045 | if (!p->mm) | |
8046 | continue; | |
8047 | ||
6cfb0d5d | 8048 | p->se.exec_start = 0; |
6cfb0d5d | 8049 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8050 | p->se.statistics.wait_start = 0; |
8051 | p->se.statistics.sleep_start = 0; | |
8052 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8053 | #endif |
dd41f596 IM |
8054 | |
8055 | if (!rt_task(p)) { | |
8056 | /* | |
8057 | * Renice negative nice level userspace | |
8058 | * tasks back to 0: | |
8059 | */ | |
8060 | if (TASK_NICE(p) < 0 && p->mm) | |
8061 | set_user_nice(p, 0); | |
1da177e4 | 8062 | continue; |
dd41f596 | 8063 | } |
1da177e4 | 8064 | |
1d615482 | 8065 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8066 | rq = __task_rq_lock(p); |
1da177e4 | 8067 | |
178be793 | 8068 | normalize_task(rq, p); |
3a5e4dc1 | 8069 | |
b29739f9 | 8070 | __task_rq_unlock(rq); |
1d615482 | 8071 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8072 | } while_each_thread(g, p); |
8073 | ||
4cf5d77a | 8074 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8075 | } |
8076 | ||
8077 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8078 | |
67fc4e0c | 8079 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8080 | /* |
67fc4e0c | 8081 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8082 | * |
8083 | * They can only be called when the whole system has been | |
8084 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8085 | * activity can take place. Using them for anything else would | |
8086 | * be a serious bug, and as a result, they aren't even visible | |
8087 | * under any other configuration. | |
8088 | */ | |
8089 | ||
8090 | /** | |
8091 | * curr_task - return the current task for a given cpu. | |
8092 | * @cpu: the processor in question. | |
8093 | * | |
8094 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8095 | */ | |
36c8b586 | 8096 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8097 | { |
8098 | return cpu_curr(cpu); | |
8099 | } | |
8100 | ||
67fc4e0c JW |
8101 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8102 | ||
8103 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8104 | /** |
8105 | * set_curr_task - set the current task for a given cpu. | |
8106 | * @cpu: the processor in question. | |
8107 | * @p: the task pointer to set. | |
8108 | * | |
8109 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8110 | * are serviced on a separate stack. It allows the architecture to switch the |
8111 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8112 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8113 | * and caller must save the original value of the current task (see | |
8114 | * curr_task() above) and restore that value before reenabling interrupts and | |
8115 | * re-starting the system. | |
8116 | * | |
8117 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8118 | */ | |
36c8b586 | 8119 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8120 | { |
8121 | cpu_curr(cpu) = p; | |
8122 | } | |
8123 | ||
8124 | #endif | |
29f59db3 | 8125 | |
bccbe08a PZ |
8126 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8127 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8128 | { |
8129 | int i; | |
8130 | ||
8131 | for_each_possible_cpu(i) { | |
8132 | if (tg->cfs_rq) | |
8133 | kfree(tg->cfs_rq[i]); | |
8134 | if (tg->se) | |
8135 | kfree(tg->se[i]); | |
6f505b16 PZ |
8136 | } |
8137 | ||
8138 | kfree(tg->cfs_rq); | |
8139 | kfree(tg->se); | |
6f505b16 PZ |
8140 | } |
8141 | ||
ec7dc8ac DG |
8142 | static |
8143 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8144 | { |
29f59db3 | 8145 | struct cfs_rq *cfs_rq; |
eab17229 | 8146 | struct sched_entity *se; |
29f59db3 SV |
8147 | int i; |
8148 | ||
434d53b0 | 8149 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8150 | if (!tg->cfs_rq) |
8151 | goto err; | |
434d53b0 | 8152 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8153 | if (!tg->se) |
8154 | goto err; | |
052f1dc7 PZ |
8155 | |
8156 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8157 | |
8158 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8159 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8160 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8161 | if (!cfs_rq) |
8162 | goto err; | |
8163 | ||
eab17229 LZ |
8164 | se = kzalloc_node(sizeof(struct sched_entity), |
8165 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8166 | if (!se) |
dfc12eb2 | 8167 | goto err_free_rq; |
29f59db3 | 8168 | |
3d4b47b4 | 8169 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8170 | } |
8171 | ||
8172 | return 1; | |
8173 | ||
49246274 | 8174 | err_free_rq: |
dfc12eb2 | 8175 | kfree(cfs_rq); |
49246274 | 8176 | err: |
bccbe08a PZ |
8177 | return 0; |
8178 | } | |
8179 | ||
bccbe08a PZ |
8180 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8181 | { | |
3d4b47b4 PZ |
8182 | struct rq *rq = cpu_rq(cpu); |
8183 | unsigned long flags; | |
3d4b47b4 PZ |
8184 | |
8185 | /* | |
8186 | * Only empty task groups can be destroyed; so we can speculatively | |
8187 | * check on_list without danger of it being re-added. | |
8188 | */ | |
8189 | if (!tg->cfs_rq[cpu]->on_list) | |
8190 | return; | |
8191 | ||
8192 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8193 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8194 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8195 | } |
6d6bc0ad | 8196 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8197 | static inline void free_fair_sched_group(struct task_group *tg) |
8198 | { | |
8199 | } | |
8200 | ||
ec7dc8ac DG |
8201 | static inline |
8202 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8203 | { |
8204 | return 1; | |
8205 | } | |
8206 | ||
bccbe08a PZ |
8207 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8208 | { | |
8209 | } | |
6d6bc0ad | 8210 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8211 | |
8212 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8213 | static void free_rt_sched_group(struct task_group *tg) |
8214 | { | |
8215 | int i; | |
8216 | ||
d0b27fa7 PZ |
8217 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8218 | ||
bccbe08a PZ |
8219 | for_each_possible_cpu(i) { |
8220 | if (tg->rt_rq) | |
8221 | kfree(tg->rt_rq[i]); | |
8222 | if (tg->rt_se) | |
8223 | kfree(tg->rt_se[i]); | |
8224 | } | |
8225 | ||
8226 | kfree(tg->rt_rq); | |
8227 | kfree(tg->rt_se); | |
8228 | } | |
8229 | ||
ec7dc8ac DG |
8230 | static |
8231 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8232 | { |
8233 | struct rt_rq *rt_rq; | |
eab17229 | 8234 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8235 | int i; |
8236 | ||
434d53b0 | 8237 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8238 | if (!tg->rt_rq) |
8239 | goto err; | |
434d53b0 | 8240 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8241 | if (!tg->rt_se) |
8242 | goto err; | |
8243 | ||
d0b27fa7 PZ |
8244 | init_rt_bandwidth(&tg->rt_bandwidth, |
8245 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8246 | |
8247 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8248 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8249 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8250 | if (!rt_rq) |
8251 | goto err; | |
29f59db3 | 8252 | |
eab17229 LZ |
8253 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8254 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8255 | if (!rt_se) |
dfc12eb2 | 8256 | goto err_free_rq; |
29f59db3 | 8257 | |
3d4b47b4 | 8258 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8259 | } |
8260 | ||
bccbe08a PZ |
8261 | return 1; |
8262 | ||
49246274 | 8263 | err_free_rq: |
dfc12eb2 | 8264 | kfree(rt_rq); |
49246274 | 8265 | err: |
bccbe08a PZ |
8266 | return 0; |
8267 | } | |
6d6bc0ad | 8268 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8269 | static inline void free_rt_sched_group(struct task_group *tg) |
8270 | { | |
8271 | } | |
8272 | ||
ec7dc8ac DG |
8273 | static inline |
8274 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8275 | { |
8276 | return 1; | |
8277 | } | |
6d6bc0ad | 8278 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8279 | |
7c941438 | 8280 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8281 | static void free_sched_group(struct task_group *tg) |
8282 | { | |
8283 | free_fair_sched_group(tg); | |
8284 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8285 | autogroup_free(tg); |
bccbe08a PZ |
8286 | kfree(tg); |
8287 | } | |
8288 | ||
8289 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8290 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8291 | { |
8292 | struct task_group *tg; | |
8293 | unsigned long flags; | |
bccbe08a PZ |
8294 | |
8295 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8296 | if (!tg) | |
8297 | return ERR_PTR(-ENOMEM); | |
8298 | ||
ec7dc8ac | 8299 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8300 | goto err; |
8301 | ||
ec7dc8ac | 8302 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8303 | goto err; |
8304 | ||
8ed36996 | 8305 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8306 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8307 | |
8308 | WARN_ON(!parent); /* root should already exist */ | |
8309 | ||
8310 | tg->parent = parent; | |
f473aa5e | 8311 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8312 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8313 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8314 | |
9b5b7751 | 8315 | return tg; |
29f59db3 SV |
8316 | |
8317 | err: | |
6f505b16 | 8318 | free_sched_group(tg); |
29f59db3 SV |
8319 | return ERR_PTR(-ENOMEM); |
8320 | } | |
8321 | ||
9b5b7751 | 8322 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8323 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8324 | { |
29f59db3 | 8325 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8326 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8327 | } |
8328 | ||
9b5b7751 | 8329 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8330 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8331 | { |
8ed36996 | 8332 | unsigned long flags; |
9b5b7751 | 8333 | int i; |
29f59db3 | 8334 | |
3d4b47b4 PZ |
8335 | /* end participation in shares distribution */ |
8336 | for_each_possible_cpu(i) | |
bccbe08a | 8337 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8338 | |
8339 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8340 | list_del_rcu(&tg->list); |
f473aa5e | 8341 | list_del_rcu(&tg->siblings); |
8ed36996 | 8342 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8343 | |
9b5b7751 | 8344 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8345 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8346 | } |
8347 | ||
9b5b7751 | 8348 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8349 | * The caller of this function should have put the task in its new group |
8350 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8351 | * reflect its new group. | |
9b5b7751 SV |
8352 | */ |
8353 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8354 | { |
8355 | int on_rq, running; | |
8356 | unsigned long flags; | |
8357 | struct rq *rq; | |
8358 | ||
8359 | rq = task_rq_lock(tsk, &flags); | |
8360 | ||
051a1d1a | 8361 | running = task_current(rq, tsk); |
fd2f4419 | 8362 | on_rq = tsk->on_rq; |
29f59db3 | 8363 | |
0e1f3483 | 8364 | if (on_rq) |
29f59db3 | 8365 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8366 | if (unlikely(running)) |
8367 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8368 | |
810b3817 | 8369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8370 | if (tsk->sched_class->task_move_group) |
8371 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8372 | else | |
810b3817 | 8373 | #endif |
b2b5ce02 | 8374 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8375 | |
0e1f3483 HS |
8376 | if (unlikely(running)) |
8377 | tsk->sched_class->set_curr_task(rq); | |
8378 | if (on_rq) | |
371fd7e7 | 8379 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8380 | |
0122ec5b | 8381 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 8382 | } |
7c941438 | 8383 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8384 | |
052f1dc7 | 8385 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8386 | static DEFINE_MUTEX(shares_mutex); |
8387 | ||
4cf86d77 | 8388 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8389 | { |
8390 | int i; | |
8ed36996 | 8391 | unsigned long flags; |
c61935fd | 8392 | |
ec7dc8ac DG |
8393 | /* |
8394 | * We can't change the weight of the root cgroup. | |
8395 | */ | |
8396 | if (!tg->se[0]) | |
8397 | return -EINVAL; | |
8398 | ||
18d95a28 PZ |
8399 | if (shares < MIN_SHARES) |
8400 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8401 | else if (shares > MAX_SHARES) |
8402 | shares = MAX_SHARES; | |
62fb1851 | 8403 | |
8ed36996 | 8404 | mutex_lock(&shares_mutex); |
9b5b7751 | 8405 | if (tg->shares == shares) |
5cb350ba | 8406 | goto done; |
29f59db3 | 8407 | |
9b5b7751 | 8408 | tg->shares = shares; |
c09595f6 | 8409 | for_each_possible_cpu(i) { |
9437178f PT |
8410 | struct rq *rq = cpu_rq(i); |
8411 | struct sched_entity *se; | |
8412 | ||
8413 | se = tg->se[i]; | |
8414 | /* Propagate contribution to hierarchy */ | |
8415 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8416 | for_each_sched_entity(se) | |
6d5ab293 | 8417 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8418 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8419 | } |
29f59db3 | 8420 | |
5cb350ba | 8421 | done: |
8ed36996 | 8422 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8423 | return 0; |
29f59db3 SV |
8424 | } |
8425 | ||
5cb350ba DG |
8426 | unsigned long sched_group_shares(struct task_group *tg) |
8427 | { | |
8428 | return tg->shares; | |
8429 | } | |
052f1dc7 | 8430 | #endif |
5cb350ba | 8431 | |
052f1dc7 | 8432 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8433 | /* |
9f0c1e56 | 8434 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8435 | */ |
9f0c1e56 PZ |
8436 | static DEFINE_MUTEX(rt_constraints_mutex); |
8437 | ||
8438 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8439 | { | |
8440 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8441 | return 1ULL << 20; |
9f0c1e56 | 8442 | |
9a7e0b18 | 8443 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8444 | } |
8445 | ||
9a7e0b18 PZ |
8446 | /* Must be called with tasklist_lock held */ |
8447 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8448 | { |
9a7e0b18 | 8449 | struct task_struct *g, *p; |
b40b2e8e | 8450 | |
9a7e0b18 PZ |
8451 | do_each_thread(g, p) { |
8452 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8453 | return 1; | |
8454 | } while_each_thread(g, p); | |
b40b2e8e | 8455 | |
9a7e0b18 PZ |
8456 | return 0; |
8457 | } | |
b40b2e8e | 8458 | |
9a7e0b18 PZ |
8459 | struct rt_schedulable_data { |
8460 | struct task_group *tg; | |
8461 | u64 rt_period; | |
8462 | u64 rt_runtime; | |
8463 | }; | |
b40b2e8e | 8464 | |
9a7e0b18 PZ |
8465 | static int tg_schedulable(struct task_group *tg, void *data) |
8466 | { | |
8467 | struct rt_schedulable_data *d = data; | |
8468 | struct task_group *child; | |
8469 | unsigned long total, sum = 0; | |
8470 | u64 period, runtime; | |
b40b2e8e | 8471 | |
9a7e0b18 PZ |
8472 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8473 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8474 | |
9a7e0b18 PZ |
8475 | if (tg == d->tg) { |
8476 | period = d->rt_period; | |
8477 | runtime = d->rt_runtime; | |
b40b2e8e | 8478 | } |
b40b2e8e | 8479 | |
4653f803 PZ |
8480 | /* |
8481 | * Cannot have more runtime than the period. | |
8482 | */ | |
8483 | if (runtime > period && runtime != RUNTIME_INF) | |
8484 | return -EINVAL; | |
6f505b16 | 8485 | |
4653f803 PZ |
8486 | /* |
8487 | * Ensure we don't starve existing RT tasks. | |
8488 | */ | |
9a7e0b18 PZ |
8489 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8490 | return -EBUSY; | |
6f505b16 | 8491 | |
9a7e0b18 | 8492 | total = to_ratio(period, runtime); |
6f505b16 | 8493 | |
4653f803 PZ |
8494 | /* |
8495 | * Nobody can have more than the global setting allows. | |
8496 | */ | |
8497 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8498 | return -EINVAL; | |
6f505b16 | 8499 | |
4653f803 PZ |
8500 | /* |
8501 | * The sum of our children's runtime should not exceed our own. | |
8502 | */ | |
9a7e0b18 PZ |
8503 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8504 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8505 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8506 | |
9a7e0b18 PZ |
8507 | if (child == d->tg) { |
8508 | period = d->rt_period; | |
8509 | runtime = d->rt_runtime; | |
8510 | } | |
6f505b16 | 8511 | |
9a7e0b18 | 8512 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8513 | } |
6f505b16 | 8514 | |
9a7e0b18 PZ |
8515 | if (sum > total) |
8516 | return -EINVAL; | |
8517 | ||
8518 | return 0; | |
6f505b16 PZ |
8519 | } |
8520 | ||
9a7e0b18 | 8521 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8522 | { |
9a7e0b18 PZ |
8523 | struct rt_schedulable_data data = { |
8524 | .tg = tg, | |
8525 | .rt_period = period, | |
8526 | .rt_runtime = runtime, | |
8527 | }; | |
8528 | ||
8529 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8530 | } |
8531 | ||
d0b27fa7 PZ |
8532 | static int tg_set_bandwidth(struct task_group *tg, |
8533 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8534 | { |
ac086bc2 | 8535 | int i, err = 0; |
9f0c1e56 | 8536 | |
9f0c1e56 | 8537 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8538 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8539 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8540 | if (err) | |
9f0c1e56 | 8541 | goto unlock; |
ac086bc2 | 8542 | |
0986b11b | 8543 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8544 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8545 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8546 | |
8547 | for_each_possible_cpu(i) { | |
8548 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8549 | ||
0986b11b | 8550 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8551 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8552 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8553 | } |
0986b11b | 8554 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8555 | unlock: |
521f1a24 | 8556 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8557 | mutex_unlock(&rt_constraints_mutex); |
8558 | ||
8559 | return err; | |
6f505b16 PZ |
8560 | } |
8561 | ||
d0b27fa7 PZ |
8562 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8563 | { | |
8564 | u64 rt_runtime, rt_period; | |
8565 | ||
8566 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8567 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8568 | if (rt_runtime_us < 0) | |
8569 | rt_runtime = RUNTIME_INF; | |
8570 | ||
8571 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8572 | } | |
8573 | ||
9f0c1e56 PZ |
8574 | long sched_group_rt_runtime(struct task_group *tg) |
8575 | { | |
8576 | u64 rt_runtime_us; | |
8577 | ||
d0b27fa7 | 8578 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8579 | return -1; |
8580 | ||
d0b27fa7 | 8581 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8582 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8583 | return rt_runtime_us; | |
8584 | } | |
d0b27fa7 PZ |
8585 | |
8586 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8587 | { | |
8588 | u64 rt_runtime, rt_period; | |
8589 | ||
8590 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8591 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8592 | ||
619b0488 R |
8593 | if (rt_period == 0) |
8594 | return -EINVAL; | |
8595 | ||
d0b27fa7 PZ |
8596 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8597 | } | |
8598 | ||
8599 | long sched_group_rt_period(struct task_group *tg) | |
8600 | { | |
8601 | u64 rt_period_us; | |
8602 | ||
8603 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8604 | do_div(rt_period_us, NSEC_PER_USEC); | |
8605 | return rt_period_us; | |
8606 | } | |
8607 | ||
8608 | static int sched_rt_global_constraints(void) | |
8609 | { | |
4653f803 | 8610 | u64 runtime, period; |
d0b27fa7 PZ |
8611 | int ret = 0; |
8612 | ||
ec5d4989 HS |
8613 | if (sysctl_sched_rt_period <= 0) |
8614 | return -EINVAL; | |
8615 | ||
4653f803 PZ |
8616 | runtime = global_rt_runtime(); |
8617 | period = global_rt_period(); | |
8618 | ||
8619 | /* | |
8620 | * Sanity check on the sysctl variables. | |
8621 | */ | |
8622 | if (runtime > period && runtime != RUNTIME_INF) | |
8623 | return -EINVAL; | |
10b612f4 | 8624 | |
d0b27fa7 | 8625 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8626 | read_lock(&tasklist_lock); |
4653f803 | 8627 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8628 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8629 | mutex_unlock(&rt_constraints_mutex); |
8630 | ||
8631 | return ret; | |
8632 | } | |
54e99124 DG |
8633 | |
8634 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8635 | { | |
8636 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8637 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8638 | return 0; | |
8639 | ||
8640 | return 1; | |
8641 | } | |
8642 | ||
6d6bc0ad | 8643 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8644 | static int sched_rt_global_constraints(void) |
8645 | { | |
ac086bc2 PZ |
8646 | unsigned long flags; |
8647 | int i; | |
8648 | ||
ec5d4989 HS |
8649 | if (sysctl_sched_rt_period <= 0) |
8650 | return -EINVAL; | |
8651 | ||
60aa605d PZ |
8652 | /* |
8653 | * There's always some RT tasks in the root group | |
8654 | * -- migration, kstopmachine etc.. | |
8655 | */ | |
8656 | if (sysctl_sched_rt_runtime == 0) | |
8657 | return -EBUSY; | |
8658 | ||
0986b11b | 8659 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8660 | for_each_possible_cpu(i) { |
8661 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8662 | ||
0986b11b | 8663 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8664 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8665 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8666 | } |
0986b11b | 8667 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8668 | |
d0b27fa7 PZ |
8669 | return 0; |
8670 | } | |
6d6bc0ad | 8671 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8672 | |
8673 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8674 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8675 | loff_t *ppos) |
8676 | { | |
8677 | int ret; | |
8678 | int old_period, old_runtime; | |
8679 | static DEFINE_MUTEX(mutex); | |
8680 | ||
8681 | mutex_lock(&mutex); | |
8682 | old_period = sysctl_sched_rt_period; | |
8683 | old_runtime = sysctl_sched_rt_runtime; | |
8684 | ||
8d65af78 | 8685 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8686 | |
8687 | if (!ret && write) { | |
8688 | ret = sched_rt_global_constraints(); | |
8689 | if (ret) { | |
8690 | sysctl_sched_rt_period = old_period; | |
8691 | sysctl_sched_rt_runtime = old_runtime; | |
8692 | } else { | |
8693 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8694 | def_rt_bandwidth.rt_period = | |
8695 | ns_to_ktime(global_rt_period()); | |
8696 | } | |
8697 | } | |
8698 | mutex_unlock(&mutex); | |
8699 | ||
8700 | return ret; | |
8701 | } | |
68318b8e | 8702 | |
052f1dc7 | 8703 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8704 | |
8705 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8706 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8707 | { |
2b01dfe3 PM |
8708 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8709 | struct task_group, css); | |
68318b8e SV |
8710 | } |
8711 | ||
8712 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8713 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8714 | { |
ec7dc8ac | 8715 | struct task_group *tg, *parent; |
68318b8e | 8716 | |
2b01dfe3 | 8717 | if (!cgrp->parent) { |
68318b8e | 8718 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8719 | return &root_task_group.css; |
68318b8e SV |
8720 | } |
8721 | ||
ec7dc8ac DG |
8722 | parent = cgroup_tg(cgrp->parent); |
8723 | tg = sched_create_group(parent); | |
68318b8e SV |
8724 | if (IS_ERR(tg)) |
8725 | return ERR_PTR(-ENOMEM); | |
8726 | ||
68318b8e SV |
8727 | return &tg->css; |
8728 | } | |
8729 | ||
41a2d6cf IM |
8730 | static void |
8731 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8732 | { |
2b01dfe3 | 8733 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8734 | |
8735 | sched_destroy_group(tg); | |
8736 | } | |
8737 | ||
41a2d6cf | 8738 | static int |
be367d09 | 8739 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8740 | { |
b68aa230 | 8741 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8742 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8743 | return -EINVAL; |
8744 | #else | |
68318b8e SV |
8745 | /* We don't support RT-tasks being in separate groups */ |
8746 | if (tsk->sched_class != &fair_sched_class) | |
8747 | return -EINVAL; | |
b68aa230 | 8748 | #endif |
be367d09 BB |
8749 | return 0; |
8750 | } | |
68318b8e | 8751 | |
be367d09 BB |
8752 | static int |
8753 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8754 | struct task_struct *tsk, bool threadgroup) | |
8755 | { | |
8756 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8757 | if (retval) | |
8758 | return retval; | |
8759 | if (threadgroup) { | |
8760 | struct task_struct *c; | |
8761 | rcu_read_lock(); | |
8762 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8763 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8764 | if (retval) { | |
8765 | rcu_read_unlock(); | |
8766 | return retval; | |
8767 | } | |
8768 | } | |
8769 | rcu_read_unlock(); | |
8770 | } | |
68318b8e SV |
8771 | return 0; |
8772 | } | |
8773 | ||
8774 | static void | |
2b01dfe3 | 8775 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8776 | struct cgroup *old_cont, struct task_struct *tsk, |
8777 | bool threadgroup) | |
68318b8e SV |
8778 | { |
8779 | sched_move_task(tsk); | |
be367d09 BB |
8780 | if (threadgroup) { |
8781 | struct task_struct *c; | |
8782 | rcu_read_lock(); | |
8783 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8784 | sched_move_task(c); | |
8785 | } | |
8786 | rcu_read_unlock(); | |
8787 | } | |
68318b8e SV |
8788 | } |
8789 | ||
068c5cc5 | 8790 | static void |
d41d5a01 PZ |
8791 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
8792 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
8793 | { |
8794 | /* | |
8795 | * cgroup_exit() is called in the copy_process() failure path. | |
8796 | * Ignore this case since the task hasn't ran yet, this avoids | |
8797 | * trying to poke a half freed task state from generic code. | |
8798 | */ | |
8799 | if (!(task->flags & PF_EXITING)) | |
8800 | return; | |
8801 | ||
8802 | sched_move_task(task); | |
8803 | } | |
8804 | ||
052f1dc7 | 8805 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8806 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8807 | u64 shareval) |
68318b8e | 8808 | { |
2b01dfe3 | 8809 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8810 | } |
8811 | ||
f4c753b7 | 8812 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8813 | { |
2b01dfe3 | 8814 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8815 | |
8816 | return (u64) tg->shares; | |
8817 | } | |
6d6bc0ad | 8818 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8819 | |
052f1dc7 | 8820 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8821 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8822 | s64 val) |
6f505b16 | 8823 | { |
06ecb27c | 8824 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8825 | } |
8826 | ||
06ecb27c | 8827 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8828 | { |
06ecb27c | 8829 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8830 | } |
d0b27fa7 PZ |
8831 | |
8832 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8833 | u64 rt_period_us) | |
8834 | { | |
8835 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8836 | } | |
8837 | ||
8838 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8839 | { | |
8840 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8841 | } | |
6d6bc0ad | 8842 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8843 | |
fe5c7cc2 | 8844 | static struct cftype cpu_files[] = { |
052f1dc7 | 8845 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8846 | { |
8847 | .name = "shares", | |
f4c753b7 PM |
8848 | .read_u64 = cpu_shares_read_u64, |
8849 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8850 | }, |
052f1dc7 PZ |
8851 | #endif |
8852 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8853 | { |
9f0c1e56 | 8854 | .name = "rt_runtime_us", |
06ecb27c PM |
8855 | .read_s64 = cpu_rt_runtime_read, |
8856 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8857 | }, |
d0b27fa7 PZ |
8858 | { |
8859 | .name = "rt_period_us", | |
f4c753b7 PM |
8860 | .read_u64 = cpu_rt_period_read_uint, |
8861 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8862 | }, |
052f1dc7 | 8863 | #endif |
68318b8e SV |
8864 | }; |
8865 | ||
8866 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8867 | { | |
fe5c7cc2 | 8868 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8869 | } |
8870 | ||
8871 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8872 | .name = "cpu", |
8873 | .create = cpu_cgroup_create, | |
8874 | .destroy = cpu_cgroup_destroy, | |
8875 | .can_attach = cpu_cgroup_can_attach, | |
8876 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8877 | .exit = cpu_cgroup_exit, |
38605cae IM |
8878 | .populate = cpu_cgroup_populate, |
8879 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8880 | .early_init = 1, |
8881 | }; | |
8882 | ||
052f1dc7 | 8883 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8884 | |
8885 | #ifdef CONFIG_CGROUP_CPUACCT | |
8886 | ||
8887 | /* | |
8888 | * CPU accounting code for task groups. | |
8889 | * | |
8890 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8891 | * (balbir@in.ibm.com). | |
8892 | */ | |
8893 | ||
934352f2 | 8894 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8895 | struct cpuacct { |
8896 | struct cgroup_subsys_state css; | |
8897 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8898 | u64 __percpu *cpuusage; |
ef12fefa | 8899 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8900 | struct cpuacct *parent; |
d842de87 SV |
8901 | }; |
8902 | ||
8903 | struct cgroup_subsys cpuacct_subsys; | |
8904 | ||
8905 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8906 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8907 | { |
32cd756a | 8908 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8909 | struct cpuacct, css); |
8910 | } | |
8911 | ||
8912 | /* return cpu accounting group to which this task belongs */ | |
8913 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8914 | { | |
8915 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8916 | struct cpuacct, css); | |
8917 | } | |
8918 | ||
8919 | /* create a new cpu accounting group */ | |
8920 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8921 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8922 | { |
8923 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8924 | int i; |
d842de87 SV |
8925 | |
8926 | if (!ca) | |
ef12fefa | 8927 | goto out; |
d842de87 SV |
8928 | |
8929 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8930 | if (!ca->cpuusage) |
8931 | goto out_free_ca; | |
8932 | ||
8933 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8934 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8935 | goto out_free_counters; | |
d842de87 | 8936 | |
934352f2 BR |
8937 | if (cgrp->parent) |
8938 | ca->parent = cgroup_ca(cgrp->parent); | |
8939 | ||
d842de87 | 8940 | return &ca->css; |
ef12fefa BR |
8941 | |
8942 | out_free_counters: | |
8943 | while (--i >= 0) | |
8944 | percpu_counter_destroy(&ca->cpustat[i]); | |
8945 | free_percpu(ca->cpuusage); | |
8946 | out_free_ca: | |
8947 | kfree(ca); | |
8948 | out: | |
8949 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8950 | } |
8951 | ||
8952 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8953 | static void |
32cd756a | 8954 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8955 | { |
32cd756a | 8956 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8957 | int i; |
d842de87 | 8958 | |
ef12fefa BR |
8959 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8960 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8961 | free_percpu(ca->cpuusage); |
8962 | kfree(ca); | |
8963 | } | |
8964 | ||
720f5498 KC |
8965 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8966 | { | |
b36128c8 | 8967 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8968 | u64 data; |
8969 | ||
8970 | #ifndef CONFIG_64BIT | |
8971 | /* | |
8972 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8973 | */ | |
05fa785c | 8974 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8975 | data = *cpuusage; |
05fa785c | 8976 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8977 | #else |
8978 | data = *cpuusage; | |
8979 | #endif | |
8980 | ||
8981 | return data; | |
8982 | } | |
8983 | ||
8984 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8985 | { | |
b36128c8 | 8986 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8987 | |
8988 | #ifndef CONFIG_64BIT | |
8989 | /* | |
8990 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8991 | */ | |
05fa785c | 8992 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8993 | *cpuusage = val; |
05fa785c | 8994 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8995 | #else |
8996 | *cpuusage = val; | |
8997 | #endif | |
8998 | } | |
8999 | ||
d842de87 | 9000 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9001 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9002 | { |
32cd756a | 9003 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9004 | u64 totalcpuusage = 0; |
9005 | int i; | |
9006 | ||
720f5498 KC |
9007 | for_each_present_cpu(i) |
9008 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9009 | |
9010 | return totalcpuusage; | |
9011 | } | |
9012 | ||
0297b803 DG |
9013 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9014 | u64 reset) | |
9015 | { | |
9016 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9017 | int err = 0; | |
9018 | int i; | |
9019 | ||
9020 | if (reset) { | |
9021 | err = -EINVAL; | |
9022 | goto out; | |
9023 | } | |
9024 | ||
720f5498 KC |
9025 | for_each_present_cpu(i) |
9026 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9027 | |
0297b803 DG |
9028 | out: |
9029 | return err; | |
9030 | } | |
9031 | ||
e9515c3c KC |
9032 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9033 | struct seq_file *m) | |
9034 | { | |
9035 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9036 | u64 percpu; | |
9037 | int i; | |
9038 | ||
9039 | for_each_present_cpu(i) { | |
9040 | percpu = cpuacct_cpuusage_read(ca, i); | |
9041 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9042 | } | |
9043 | seq_printf(m, "\n"); | |
9044 | return 0; | |
9045 | } | |
9046 | ||
ef12fefa BR |
9047 | static const char *cpuacct_stat_desc[] = { |
9048 | [CPUACCT_STAT_USER] = "user", | |
9049 | [CPUACCT_STAT_SYSTEM] = "system", | |
9050 | }; | |
9051 | ||
9052 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9053 | struct cgroup_map_cb *cb) | |
9054 | { | |
9055 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9056 | int i; | |
9057 | ||
9058 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9059 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9060 | val = cputime64_to_clock_t(val); | |
9061 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9062 | } | |
9063 | return 0; | |
9064 | } | |
9065 | ||
d842de87 SV |
9066 | static struct cftype files[] = { |
9067 | { | |
9068 | .name = "usage", | |
f4c753b7 PM |
9069 | .read_u64 = cpuusage_read, |
9070 | .write_u64 = cpuusage_write, | |
d842de87 | 9071 | }, |
e9515c3c KC |
9072 | { |
9073 | .name = "usage_percpu", | |
9074 | .read_seq_string = cpuacct_percpu_seq_read, | |
9075 | }, | |
ef12fefa BR |
9076 | { |
9077 | .name = "stat", | |
9078 | .read_map = cpuacct_stats_show, | |
9079 | }, | |
d842de87 SV |
9080 | }; |
9081 | ||
32cd756a | 9082 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9083 | { |
32cd756a | 9084 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9085 | } |
9086 | ||
9087 | /* | |
9088 | * charge this task's execution time to its accounting group. | |
9089 | * | |
9090 | * called with rq->lock held. | |
9091 | */ | |
9092 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9093 | { | |
9094 | struct cpuacct *ca; | |
934352f2 | 9095 | int cpu; |
d842de87 | 9096 | |
c40c6f85 | 9097 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9098 | return; |
9099 | ||
934352f2 | 9100 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9101 | |
9102 | rcu_read_lock(); | |
9103 | ||
d842de87 | 9104 | ca = task_ca(tsk); |
d842de87 | 9105 | |
934352f2 | 9106 | for (; ca; ca = ca->parent) { |
b36128c8 | 9107 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9108 | *cpuusage += cputime; |
9109 | } | |
a18b83b7 BR |
9110 | |
9111 | rcu_read_unlock(); | |
d842de87 SV |
9112 | } |
9113 | ||
fa535a77 AB |
9114 | /* |
9115 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9116 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9117 | * percpu_counter_add with values large enough to always overflow the | |
9118 | * per cpu batch limit causing bad SMP scalability. | |
9119 | * | |
9120 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9121 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9122 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9123 | */ | |
9124 | #ifdef CONFIG_SMP | |
9125 | #define CPUACCT_BATCH \ | |
9126 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9127 | #else | |
9128 | #define CPUACCT_BATCH 0 | |
9129 | #endif | |
9130 | ||
ef12fefa BR |
9131 | /* |
9132 | * Charge the system/user time to the task's accounting group. | |
9133 | */ | |
9134 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9135 | enum cpuacct_stat_index idx, cputime_t val) | |
9136 | { | |
9137 | struct cpuacct *ca; | |
fa535a77 | 9138 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9139 | |
9140 | if (unlikely(!cpuacct_subsys.active)) | |
9141 | return; | |
9142 | ||
9143 | rcu_read_lock(); | |
9144 | ca = task_ca(tsk); | |
9145 | ||
9146 | do { | |
fa535a77 | 9147 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9148 | ca = ca->parent; |
9149 | } while (ca); | |
9150 | rcu_read_unlock(); | |
9151 | } | |
9152 | ||
d842de87 SV |
9153 | struct cgroup_subsys cpuacct_subsys = { |
9154 | .name = "cpuacct", | |
9155 | .create = cpuacct_create, | |
9156 | .destroy = cpuacct_destroy, | |
9157 | .populate = cpuacct_populate, | |
9158 | .subsys_id = cpuacct_subsys_id, | |
9159 | }; | |
9160 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9161 |