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; |
6aa645ea IM |
315 | |
316 | struct rb_root tasks_timeline; | |
317 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
318 | |
319 | struct list_head tasks; | |
320 | struct list_head *balance_iterator; | |
321 | ||
322 | /* | |
323 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
324 | * It is set to NULL otherwise (i.e when none are currently running). |
325 | */ | |
ac53db59 | 326 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 327 | |
5ac5c4d6 | 328 | unsigned int nr_spread_over; |
ddc97297 | 329 | |
62160e3f | 330 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
331 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
332 | ||
41a2d6cf IM |
333 | /* |
334 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
335 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
336 | * (like users, containers etc.) | |
337 | * | |
338 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
339 | * list is used during load balance. | |
340 | */ | |
3d4b47b4 | 341 | int on_list; |
41a2d6cf IM |
342 | struct list_head leaf_cfs_rq_list; |
343 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
344 | |
345 | #ifdef CONFIG_SMP | |
c09595f6 | 346 | /* |
c8cba857 | 347 | * the part of load.weight contributed by tasks |
c09595f6 | 348 | */ |
c8cba857 | 349 | unsigned long task_weight; |
c09595f6 | 350 | |
c8cba857 PZ |
351 | /* |
352 | * h_load = weight * f(tg) | |
353 | * | |
354 | * Where f(tg) is the recursive weight fraction assigned to | |
355 | * this group. | |
356 | */ | |
357 | unsigned long h_load; | |
c09595f6 | 358 | |
c8cba857 | 359 | /* |
3b3d190e PT |
360 | * Maintaining per-cpu shares distribution for group scheduling |
361 | * | |
362 | * load_stamp is the last time we updated the load average | |
363 | * load_last is the last time we updated the load average and saw load | |
364 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 365 | */ |
2069dd75 PZ |
366 | u64 load_avg; |
367 | u64 load_period; | |
3b3d190e | 368 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 369 | |
2069dd75 | 370 | unsigned long load_contribution; |
c09595f6 | 371 | #endif |
6aa645ea IM |
372 | #endif |
373 | }; | |
1da177e4 | 374 | |
6aa645ea IM |
375 | /* Real-Time classes' related field in a runqueue: */ |
376 | struct rt_rq { | |
377 | struct rt_prio_array active; | |
63489e45 | 378 | unsigned long rt_nr_running; |
052f1dc7 | 379 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
380 | struct { |
381 | int curr; /* highest queued rt task prio */ | |
398a153b | 382 | #ifdef CONFIG_SMP |
e864c499 | 383 | int next; /* next highest */ |
398a153b | 384 | #endif |
e864c499 | 385 | } highest_prio; |
6f505b16 | 386 | #endif |
fa85ae24 | 387 | #ifdef CONFIG_SMP |
73fe6aae | 388 | unsigned long rt_nr_migratory; |
a1ba4d8b | 389 | unsigned long rt_nr_total; |
a22d7fc1 | 390 | int overloaded; |
917b627d | 391 | struct plist_head pushable_tasks; |
fa85ae24 | 392 | #endif |
6f505b16 | 393 | int rt_throttled; |
fa85ae24 | 394 | u64 rt_time; |
ac086bc2 | 395 | u64 rt_runtime; |
ea736ed5 | 396 | /* Nests inside the rq lock: */ |
0986b11b | 397 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 398 | |
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
400 | unsigned long rt_nr_boosted; |
401 | ||
6f505b16 PZ |
402 | struct rq *rq; |
403 | struct list_head leaf_rt_rq_list; | |
404 | struct task_group *tg; | |
6f505b16 | 405 | #endif |
6aa645ea IM |
406 | }; |
407 | ||
57d885fe GH |
408 | #ifdef CONFIG_SMP |
409 | ||
410 | /* | |
411 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
412 | * variables. Each exclusive cpuset essentially defines an island domain by |
413 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
414 | * exclusive cpuset is created, we also create and attach a new root-domain |
415 | * object. | |
416 | * | |
57d885fe GH |
417 | */ |
418 | struct root_domain { | |
419 | atomic_t refcount; | |
dce840a0 | 420 | struct rcu_head rcu; |
c6c4927b RR |
421 | cpumask_var_t span; |
422 | cpumask_var_t online; | |
637f5085 | 423 | |
0eab9146 | 424 | /* |
637f5085 GH |
425 | * The "RT overload" flag: it gets set if a CPU has more than |
426 | * one runnable RT task. | |
427 | */ | |
c6c4927b | 428 | cpumask_var_t rto_mask; |
0eab9146 | 429 | atomic_t rto_count; |
6e0534f2 | 430 | struct cpupri cpupri; |
57d885fe GH |
431 | }; |
432 | ||
dc938520 GH |
433 | /* |
434 | * By default the system creates a single root-domain with all cpus as | |
435 | * members (mimicking the global state we have today). | |
436 | */ | |
57d885fe GH |
437 | static struct root_domain def_root_domain; |
438 | ||
ed2d372c | 439 | #endif /* CONFIG_SMP */ |
57d885fe | 440 | |
1da177e4 LT |
441 | /* |
442 | * This is the main, per-CPU runqueue data structure. | |
443 | * | |
444 | * Locking rule: those places that want to lock multiple runqueues | |
445 | * (such as the load balancing or the thread migration code), lock | |
446 | * acquire operations must be ordered by ascending &runqueue. | |
447 | */ | |
70b97a7f | 448 | struct rq { |
d8016491 | 449 | /* runqueue lock: */ |
05fa785c | 450 | raw_spinlock_t lock; |
1da177e4 LT |
451 | |
452 | /* | |
453 | * nr_running and cpu_load should be in the same cacheline because | |
454 | * remote CPUs use both these fields when doing load calculation. | |
455 | */ | |
456 | unsigned long nr_running; | |
6aa645ea IM |
457 | #define CPU_LOAD_IDX_MAX 5 |
458 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 459 | unsigned long last_load_update_tick; |
46cb4b7c | 460 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 461 | u64 nohz_stamp; |
83cd4fe2 | 462 | unsigned char nohz_balance_kick; |
46cb4b7c | 463 | #endif |
a64692a3 MG |
464 | unsigned int skip_clock_update; |
465 | ||
d8016491 IM |
466 | /* capture load from *all* tasks on this cpu: */ |
467 | struct load_weight load; | |
6aa645ea IM |
468 | unsigned long nr_load_updates; |
469 | u64 nr_switches; | |
470 | ||
471 | struct cfs_rq cfs; | |
6f505b16 | 472 | struct rt_rq rt; |
6f505b16 | 473 | |
6aa645ea | 474 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
475 | /* list of leaf cfs_rq on this cpu: */ |
476 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
477 | #endif |
478 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 479 | struct list_head leaf_rt_rq_list; |
1da177e4 | 480 | #endif |
1da177e4 LT |
481 | |
482 | /* | |
483 | * This is part of a global counter where only the total sum | |
484 | * over all CPUs matters. A task can increase this counter on | |
485 | * one CPU and if it got migrated afterwards it may decrease | |
486 | * it on another CPU. Always updated under the runqueue lock: | |
487 | */ | |
488 | unsigned long nr_uninterruptible; | |
489 | ||
34f971f6 | 490 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 491 | unsigned long next_balance; |
1da177e4 | 492 | struct mm_struct *prev_mm; |
6aa645ea | 493 | |
3e51f33f | 494 | u64 clock; |
305e6835 | 495 | u64 clock_task; |
6aa645ea | 496 | |
1da177e4 LT |
497 | atomic_t nr_iowait; |
498 | ||
499 | #ifdef CONFIG_SMP | |
0eab9146 | 500 | struct root_domain *rd; |
1da177e4 LT |
501 | struct sched_domain *sd; |
502 | ||
e51fd5e2 PZ |
503 | unsigned long cpu_power; |
504 | ||
a0a522ce | 505 | unsigned char idle_at_tick; |
1da177e4 | 506 | /* For active balancing */ |
3f029d3c | 507 | int post_schedule; |
1da177e4 LT |
508 | int active_balance; |
509 | int push_cpu; | |
969c7921 | 510 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
511 | /* cpu of this runqueue: */ |
512 | int cpu; | |
1f11eb6a | 513 | int online; |
1da177e4 | 514 | |
a8a51d5e | 515 | unsigned long avg_load_per_task; |
1da177e4 | 516 | |
e9e9250b PZ |
517 | u64 rt_avg; |
518 | u64 age_stamp; | |
1b9508f6 MG |
519 | u64 idle_stamp; |
520 | u64 avg_idle; | |
1da177e4 LT |
521 | #endif |
522 | ||
aa483808 VP |
523 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
524 | u64 prev_irq_time; | |
525 | #endif | |
526 | ||
dce48a84 TG |
527 | /* calc_load related fields */ |
528 | unsigned long calc_load_update; | |
529 | long calc_load_active; | |
530 | ||
8f4d37ec | 531 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
532 | #ifdef CONFIG_SMP |
533 | int hrtick_csd_pending; | |
534 | struct call_single_data hrtick_csd; | |
535 | #endif | |
8f4d37ec PZ |
536 | struct hrtimer hrtick_timer; |
537 | #endif | |
538 | ||
1da177e4 LT |
539 | #ifdef CONFIG_SCHEDSTATS |
540 | /* latency stats */ | |
541 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
542 | unsigned long long rq_cpu_time; |
543 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
544 | |
545 | /* sys_sched_yield() stats */ | |
480b9434 | 546 | unsigned int yld_count; |
1da177e4 LT |
547 | |
548 | /* schedule() stats */ | |
480b9434 KC |
549 | unsigned int sched_switch; |
550 | unsigned int sched_count; | |
551 | unsigned int sched_goidle; | |
1da177e4 LT |
552 | |
553 | /* try_to_wake_up() stats */ | |
480b9434 KC |
554 | unsigned int ttwu_count; |
555 | unsigned int ttwu_local; | |
1da177e4 LT |
556 | #endif |
557 | }; | |
558 | ||
f34e3b61 | 559 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 560 | |
a64692a3 | 561 | |
1e5a7405 | 562 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 563 | |
0a2966b4 CL |
564 | static inline int cpu_of(struct rq *rq) |
565 | { | |
566 | #ifdef CONFIG_SMP | |
567 | return rq->cpu; | |
568 | #else | |
569 | return 0; | |
570 | #endif | |
571 | } | |
572 | ||
497f0ab3 | 573 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d | 574 | rcu_dereference_check((p), \ |
dce840a0 | 575 | rcu_read_lock_held() || \ |
d11c563d PM |
576 | lockdep_is_held(&sched_domains_mutex)) |
577 | ||
674311d5 NP |
578 | /* |
579 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 580 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
581 | * |
582 | * The domain tree of any CPU may only be accessed from within | |
583 | * preempt-disabled sections. | |
584 | */ | |
48f24c4d | 585 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 586 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
587 | |
588 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
589 | #define this_rq() (&__get_cpu_var(runqueues)) | |
590 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
591 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 592 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 593 | |
dc61b1d6 PZ |
594 | #ifdef CONFIG_CGROUP_SCHED |
595 | ||
596 | /* | |
597 | * Return the group to which this tasks belongs. | |
598 | * | |
599 | * We use task_subsys_state_check() and extend the RCU verification | |
600 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
601 | * holds that lock for each task it moves into the cgroup. Therefore | |
602 | * by holding that lock, we pin the task to the current cgroup. | |
603 | */ | |
604 | static inline struct task_group *task_group(struct task_struct *p) | |
605 | { | |
5091faa4 | 606 | struct task_group *tg; |
dc61b1d6 PZ |
607 | struct cgroup_subsys_state *css; |
608 | ||
609 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
610 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
611 | tg = container_of(css, struct task_group, css); |
612 | ||
613 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
614 | } |
615 | ||
616 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
617 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
618 | { | |
619 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
620 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
621 | p->se.parent = task_group(p)->se[cpu]; | |
622 | #endif | |
623 | ||
624 | #ifdef CONFIG_RT_GROUP_SCHED | |
625 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
626 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
627 | #endif | |
628 | } | |
629 | ||
630 | #else /* CONFIG_CGROUP_SCHED */ | |
631 | ||
632 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
633 | static inline struct task_group *task_group(struct task_struct *p) | |
634 | { | |
635 | return NULL; | |
636 | } | |
637 | ||
638 | #endif /* CONFIG_CGROUP_SCHED */ | |
639 | ||
fe44d621 | 640 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 641 | |
fe44d621 | 642 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 643 | { |
fe44d621 | 644 | s64 delta; |
305e6835 | 645 | |
f26f9aff MG |
646 | if (rq->skip_clock_update) |
647 | return; | |
aa483808 | 648 | |
fe44d621 PZ |
649 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
650 | rq->clock += delta; | |
651 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
652 | } |
653 | ||
bf5c91ba IM |
654 | /* |
655 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
656 | */ | |
657 | #ifdef CONFIG_SCHED_DEBUG | |
658 | # define const_debug __read_mostly | |
659 | #else | |
660 | # define const_debug static const | |
661 | #endif | |
662 | ||
017730c1 | 663 | /** |
1fd06bb1 | 664 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 665 | * @cpu: the processor in question. |
017730c1 | 666 | * |
017730c1 IM |
667 | * This interface allows printk to be called with the runqueue lock |
668 | * held and know whether or not it is OK to wake up the klogd. | |
669 | */ | |
89f19f04 | 670 | int runqueue_is_locked(int cpu) |
017730c1 | 671 | { |
05fa785c | 672 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
673 | } |
674 | ||
bf5c91ba IM |
675 | /* |
676 | * Debugging: various feature bits | |
677 | */ | |
f00b45c1 PZ |
678 | |
679 | #define SCHED_FEAT(name, enabled) \ | |
680 | __SCHED_FEAT_##name , | |
681 | ||
bf5c91ba | 682 | enum { |
f00b45c1 | 683 | #include "sched_features.h" |
bf5c91ba IM |
684 | }; |
685 | ||
f00b45c1 PZ |
686 | #undef SCHED_FEAT |
687 | ||
688 | #define SCHED_FEAT(name, enabled) \ | |
689 | (1UL << __SCHED_FEAT_##name) * enabled | | |
690 | ||
bf5c91ba | 691 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
692 | #include "sched_features.h" |
693 | 0; | |
694 | ||
695 | #undef SCHED_FEAT | |
696 | ||
697 | #ifdef CONFIG_SCHED_DEBUG | |
698 | #define SCHED_FEAT(name, enabled) \ | |
699 | #name , | |
700 | ||
983ed7a6 | 701 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
702 | #include "sched_features.h" |
703 | NULL | |
704 | }; | |
705 | ||
706 | #undef SCHED_FEAT | |
707 | ||
34f3a814 | 708 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 709 | { |
f00b45c1 PZ |
710 | int i; |
711 | ||
712 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
713 | if (!(sysctl_sched_features & (1UL << i))) |
714 | seq_puts(m, "NO_"); | |
715 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 716 | } |
34f3a814 | 717 | seq_puts(m, "\n"); |
f00b45c1 | 718 | |
34f3a814 | 719 | return 0; |
f00b45c1 PZ |
720 | } |
721 | ||
722 | static ssize_t | |
723 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
724 | size_t cnt, loff_t *ppos) | |
725 | { | |
726 | char buf[64]; | |
7740191c | 727 | char *cmp; |
f00b45c1 PZ |
728 | int neg = 0; |
729 | int i; | |
730 | ||
731 | if (cnt > 63) | |
732 | cnt = 63; | |
733 | ||
734 | if (copy_from_user(&buf, ubuf, cnt)) | |
735 | return -EFAULT; | |
736 | ||
737 | buf[cnt] = 0; | |
7740191c | 738 | cmp = strstrip(buf); |
f00b45c1 | 739 | |
524429c3 | 740 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
741 | neg = 1; |
742 | cmp += 3; | |
743 | } | |
744 | ||
745 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 746 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
747 | if (neg) |
748 | sysctl_sched_features &= ~(1UL << i); | |
749 | else | |
750 | sysctl_sched_features |= (1UL << i); | |
751 | break; | |
752 | } | |
753 | } | |
754 | ||
755 | if (!sched_feat_names[i]) | |
756 | return -EINVAL; | |
757 | ||
42994724 | 758 | *ppos += cnt; |
f00b45c1 PZ |
759 | |
760 | return cnt; | |
761 | } | |
762 | ||
34f3a814 LZ |
763 | static int sched_feat_open(struct inode *inode, struct file *filp) |
764 | { | |
765 | return single_open(filp, sched_feat_show, NULL); | |
766 | } | |
767 | ||
828c0950 | 768 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
769 | .open = sched_feat_open, |
770 | .write = sched_feat_write, | |
771 | .read = seq_read, | |
772 | .llseek = seq_lseek, | |
773 | .release = single_release, | |
f00b45c1 PZ |
774 | }; |
775 | ||
776 | static __init int sched_init_debug(void) | |
777 | { | |
f00b45c1 PZ |
778 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
779 | &sched_feat_fops); | |
780 | ||
781 | return 0; | |
782 | } | |
783 | late_initcall(sched_init_debug); | |
784 | ||
785 | #endif | |
786 | ||
787 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 788 | |
b82d9fdd PZ |
789 | /* |
790 | * Number of tasks to iterate in a single balance run. | |
791 | * Limited because this is done with IRQs disabled. | |
792 | */ | |
793 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
794 | ||
e9e9250b PZ |
795 | /* |
796 | * period over which we average the RT time consumption, measured | |
797 | * in ms. | |
798 | * | |
799 | * default: 1s | |
800 | */ | |
801 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
802 | ||
fa85ae24 | 803 | /* |
9f0c1e56 | 804 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
805 | * default: 1s |
806 | */ | |
9f0c1e56 | 807 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 808 | |
6892b75e IM |
809 | static __read_mostly int scheduler_running; |
810 | ||
9f0c1e56 PZ |
811 | /* |
812 | * part of the period that we allow rt tasks to run in us. | |
813 | * default: 0.95s | |
814 | */ | |
815 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 816 | |
d0b27fa7 PZ |
817 | static inline u64 global_rt_period(void) |
818 | { | |
819 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
820 | } | |
821 | ||
822 | static inline u64 global_rt_runtime(void) | |
823 | { | |
e26873bb | 824 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
825 | return RUNTIME_INF; |
826 | ||
827 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
828 | } | |
fa85ae24 | 829 | |
1da177e4 | 830 | #ifndef prepare_arch_switch |
4866cde0 NP |
831 | # define prepare_arch_switch(next) do { } while (0) |
832 | #endif | |
833 | #ifndef finish_arch_switch | |
834 | # define finish_arch_switch(prev) do { } while (0) | |
835 | #endif | |
836 | ||
051a1d1a DA |
837 | static inline int task_current(struct rq *rq, struct task_struct *p) |
838 | { | |
839 | return rq->curr == p; | |
840 | } | |
841 | ||
4866cde0 | 842 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 843 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 844 | { |
051a1d1a | 845 | return task_current(rq, p); |
4866cde0 NP |
846 | } |
847 | ||
70b97a7f | 848 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
849 | { |
850 | } | |
851 | ||
70b97a7f | 852 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 853 | { |
da04c035 IM |
854 | #ifdef CONFIG_DEBUG_SPINLOCK |
855 | /* this is a valid case when another task releases the spinlock */ | |
856 | rq->lock.owner = current; | |
857 | #endif | |
8a25d5de IM |
858 | /* |
859 | * If we are tracking spinlock dependencies then we have to | |
860 | * fix up the runqueue lock - which gets 'carried over' from | |
861 | * prev into current: | |
862 | */ | |
863 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
864 | ||
05fa785c | 865 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
866 | } |
867 | ||
868 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 869 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
870 | { |
871 | #ifdef CONFIG_SMP | |
872 | return p->oncpu; | |
873 | #else | |
051a1d1a | 874 | return task_current(rq, p); |
4866cde0 NP |
875 | #endif |
876 | } | |
877 | ||
70b97a7f | 878 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
879 | { |
880 | #ifdef CONFIG_SMP | |
881 | /* | |
882 | * We can optimise this out completely for !SMP, because the | |
883 | * SMP rebalancing from interrupt is the only thing that cares | |
884 | * here. | |
885 | */ | |
886 | next->oncpu = 1; | |
887 | #endif | |
888 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 889 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 890 | #else |
05fa785c | 891 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
892 | #endif |
893 | } | |
894 | ||
70b97a7f | 895 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
896 | { |
897 | #ifdef CONFIG_SMP | |
898 | /* | |
899 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
900 | * We must ensure this doesn't happen until the switch is completely | |
901 | * finished. | |
902 | */ | |
903 | smp_wmb(); | |
904 | prev->oncpu = 0; | |
905 | #endif | |
906 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
907 | local_irq_enable(); | |
1da177e4 | 908 | #endif |
4866cde0 NP |
909 | } |
910 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 911 | |
0970d299 | 912 | /* |
65cc8e48 PZ |
913 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
914 | * against ttwu(). | |
0970d299 PZ |
915 | */ |
916 | static inline int task_is_waking(struct task_struct *p) | |
917 | { | |
0017d735 | 918 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
919 | } |
920 | ||
b29739f9 IM |
921 | /* |
922 | * __task_rq_lock - lock the runqueue a given task resides on. | |
923 | * Must be called interrupts disabled. | |
924 | */ | |
70b97a7f | 925 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
926 | __acquires(rq->lock) |
927 | { | |
0970d299 PZ |
928 | struct rq *rq; |
929 | ||
3a5c359a | 930 | for (;;) { |
0970d299 | 931 | rq = task_rq(p); |
05fa785c | 932 | raw_spin_lock(&rq->lock); |
65cc8e48 | 933 | if (likely(rq == task_rq(p))) |
3a5c359a | 934 | return rq; |
05fa785c | 935 | raw_spin_unlock(&rq->lock); |
b29739f9 | 936 | } |
b29739f9 IM |
937 | } |
938 | ||
1da177e4 LT |
939 | /* |
940 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 941 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
942 | * explicitly disabling preemption. |
943 | */ | |
70b97a7f | 944 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
945 | __acquires(rq->lock) |
946 | { | |
70b97a7f | 947 | struct rq *rq; |
1da177e4 | 948 | |
3a5c359a AK |
949 | for (;;) { |
950 | local_irq_save(*flags); | |
951 | rq = task_rq(p); | |
05fa785c | 952 | raw_spin_lock(&rq->lock); |
65cc8e48 | 953 | if (likely(rq == task_rq(p))) |
3a5c359a | 954 | return rq; |
05fa785c | 955 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 956 | } |
1da177e4 LT |
957 | } |
958 | ||
a9957449 | 959 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
960 | __releases(rq->lock) |
961 | { | |
05fa785c | 962 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
963 | } |
964 | ||
70b97a7f | 965 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
966 | __releases(rq->lock) |
967 | { | |
05fa785c | 968 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
969 | } |
970 | ||
1da177e4 | 971 | /* |
cc2a73b5 | 972 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 973 | */ |
a9957449 | 974 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
975 | __acquires(rq->lock) |
976 | { | |
70b97a7f | 977 | struct rq *rq; |
1da177e4 LT |
978 | |
979 | local_irq_disable(); | |
980 | rq = this_rq(); | |
05fa785c | 981 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
982 | |
983 | return rq; | |
984 | } | |
985 | ||
8f4d37ec PZ |
986 | #ifdef CONFIG_SCHED_HRTICK |
987 | /* | |
988 | * Use HR-timers to deliver accurate preemption points. | |
989 | * | |
990 | * Its all a bit involved since we cannot program an hrt while holding the | |
991 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
992 | * reschedule event. | |
993 | * | |
994 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
995 | * rq->lock. | |
996 | */ | |
8f4d37ec PZ |
997 | |
998 | /* | |
999 | * Use hrtick when: | |
1000 | * - enabled by features | |
1001 | * - hrtimer is actually high res | |
1002 | */ | |
1003 | static inline int hrtick_enabled(struct rq *rq) | |
1004 | { | |
1005 | if (!sched_feat(HRTICK)) | |
1006 | return 0; | |
ba42059f | 1007 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1008 | return 0; |
8f4d37ec PZ |
1009 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1010 | } | |
1011 | ||
8f4d37ec PZ |
1012 | static void hrtick_clear(struct rq *rq) |
1013 | { | |
1014 | if (hrtimer_active(&rq->hrtick_timer)) | |
1015 | hrtimer_cancel(&rq->hrtick_timer); | |
1016 | } | |
1017 | ||
8f4d37ec PZ |
1018 | /* |
1019 | * High-resolution timer tick. | |
1020 | * Runs from hardirq context with interrupts disabled. | |
1021 | */ | |
1022 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1023 | { | |
1024 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1025 | ||
1026 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1027 | ||
05fa785c | 1028 | raw_spin_lock(&rq->lock); |
3e51f33f | 1029 | update_rq_clock(rq); |
8f4d37ec | 1030 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1031 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1032 | |
1033 | return HRTIMER_NORESTART; | |
1034 | } | |
1035 | ||
95e904c7 | 1036 | #ifdef CONFIG_SMP |
31656519 PZ |
1037 | /* |
1038 | * called from hardirq (IPI) context | |
1039 | */ | |
1040 | static void __hrtick_start(void *arg) | |
b328ca18 | 1041 | { |
31656519 | 1042 | struct rq *rq = arg; |
b328ca18 | 1043 | |
05fa785c | 1044 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1045 | hrtimer_restart(&rq->hrtick_timer); |
1046 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1047 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1048 | } |
1049 | ||
31656519 PZ |
1050 | /* |
1051 | * Called to set the hrtick timer state. | |
1052 | * | |
1053 | * called with rq->lock held and irqs disabled | |
1054 | */ | |
1055 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1056 | { |
31656519 PZ |
1057 | struct hrtimer *timer = &rq->hrtick_timer; |
1058 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1059 | |
cc584b21 | 1060 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1061 | |
1062 | if (rq == this_rq()) { | |
1063 | hrtimer_restart(timer); | |
1064 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1065 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1066 | rq->hrtick_csd_pending = 1; |
1067 | } | |
b328ca18 PZ |
1068 | } |
1069 | ||
1070 | static int | |
1071 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1072 | { | |
1073 | int cpu = (int)(long)hcpu; | |
1074 | ||
1075 | switch (action) { | |
1076 | case CPU_UP_CANCELED: | |
1077 | case CPU_UP_CANCELED_FROZEN: | |
1078 | case CPU_DOWN_PREPARE: | |
1079 | case CPU_DOWN_PREPARE_FROZEN: | |
1080 | case CPU_DEAD: | |
1081 | case CPU_DEAD_FROZEN: | |
31656519 | 1082 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1083 | return NOTIFY_OK; |
1084 | } | |
1085 | ||
1086 | return NOTIFY_DONE; | |
1087 | } | |
1088 | ||
fa748203 | 1089 | static __init void init_hrtick(void) |
b328ca18 PZ |
1090 | { |
1091 | hotcpu_notifier(hotplug_hrtick, 0); | |
1092 | } | |
31656519 PZ |
1093 | #else |
1094 | /* | |
1095 | * Called to set the hrtick timer state. | |
1096 | * | |
1097 | * called with rq->lock held and irqs disabled | |
1098 | */ | |
1099 | static void hrtick_start(struct rq *rq, u64 delay) | |
1100 | { | |
7f1e2ca9 | 1101 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1102 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1103 | } |
b328ca18 | 1104 | |
006c75f1 | 1105 | static inline void init_hrtick(void) |
8f4d37ec | 1106 | { |
8f4d37ec | 1107 | } |
31656519 | 1108 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1109 | |
31656519 | 1110 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1111 | { |
31656519 PZ |
1112 | #ifdef CONFIG_SMP |
1113 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1114 | |
31656519 PZ |
1115 | rq->hrtick_csd.flags = 0; |
1116 | rq->hrtick_csd.func = __hrtick_start; | |
1117 | rq->hrtick_csd.info = rq; | |
1118 | #endif | |
8f4d37ec | 1119 | |
31656519 PZ |
1120 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1121 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1122 | } |
006c75f1 | 1123 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1124 | static inline void hrtick_clear(struct rq *rq) |
1125 | { | |
1126 | } | |
1127 | ||
8f4d37ec PZ |
1128 | static inline void init_rq_hrtick(struct rq *rq) |
1129 | { | |
1130 | } | |
1131 | ||
b328ca18 PZ |
1132 | static inline void init_hrtick(void) |
1133 | { | |
1134 | } | |
006c75f1 | 1135 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1136 | |
c24d20db IM |
1137 | /* |
1138 | * resched_task - mark a task 'to be rescheduled now'. | |
1139 | * | |
1140 | * On UP this means the setting of the need_resched flag, on SMP it | |
1141 | * might also involve a cross-CPU call to trigger the scheduler on | |
1142 | * the target CPU. | |
1143 | */ | |
1144 | #ifdef CONFIG_SMP | |
1145 | ||
1146 | #ifndef tsk_is_polling | |
1147 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1148 | #endif | |
1149 | ||
31656519 | 1150 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1151 | { |
1152 | int cpu; | |
1153 | ||
05fa785c | 1154 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1155 | |
5ed0cec0 | 1156 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1157 | return; |
1158 | ||
5ed0cec0 | 1159 | set_tsk_need_resched(p); |
c24d20db IM |
1160 | |
1161 | cpu = task_cpu(p); | |
1162 | if (cpu == smp_processor_id()) | |
1163 | return; | |
1164 | ||
1165 | /* NEED_RESCHED must be visible before we test polling */ | |
1166 | smp_mb(); | |
1167 | if (!tsk_is_polling(p)) | |
1168 | smp_send_reschedule(cpu); | |
1169 | } | |
1170 | ||
1171 | static void resched_cpu(int cpu) | |
1172 | { | |
1173 | struct rq *rq = cpu_rq(cpu); | |
1174 | unsigned long flags; | |
1175 | ||
05fa785c | 1176 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1177 | return; |
1178 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1179 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1180 | } |
06d8308c TG |
1181 | |
1182 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1183 | /* |
1184 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1185 | * from an idle cpu. This is good for power-savings. | |
1186 | * | |
1187 | * We don't do similar optimization for completely idle system, as | |
1188 | * selecting an idle cpu will add more delays to the timers than intended | |
1189 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1190 | */ | |
1191 | int get_nohz_timer_target(void) | |
1192 | { | |
1193 | int cpu = smp_processor_id(); | |
1194 | int i; | |
1195 | struct sched_domain *sd; | |
1196 | ||
1197 | for_each_domain(cpu, sd) { | |
1198 | for_each_cpu(i, sched_domain_span(sd)) | |
1199 | if (!idle_cpu(i)) | |
1200 | return i; | |
1201 | } | |
1202 | return cpu; | |
1203 | } | |
06d8308c TG |
1204 | /* |
1205 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1206 | * idle CPU then this timer might expire before the next timer event | |
1207 | * which is scheduled to wake up that CPU. In case of a completely | |
1208 | * idle system the next event might even be infinite time into the | |
1209 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1210 | * leaves the inner idle loop so the newly added timer is taken into | |
1211 | * account when the CPU goes back to idle and evaluates the timer | |
1212 | * wheel for the next timer event. | |
1213 | */ | |
1214 | void wake_up_idle_cpu(int cpu) | |
1215 | { | |
1216 | struct rq *rq = cpu_rq(cpu); | |
1217 | ||
1218 | if (cpu == smp_processor_id()) | |
1219 | return; | |
1220 | ||
1221 | /* | |
1222 | * This is safe, as this function is called with the timer | |
1223 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1224 | * to idle and has not yet set rq->curr to idle then it will | |
1225 | * be serialized on the timer wheel base lock and take the new | |
1226 | * timer into account automatically. | |
1227 | */ | |
1228 | if (rq->curr != rq->idle) | |
1229 | return; | |
1230 | ||
1231 | /* | |
1232 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1233 | * lockless. The worst case is that the other CPU runs the | |
1234 | * idle task through an additional NOOP schedule() | |
1235 | */ | |
5ed0cec0 | 1236 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1237 | |
1238 | /* NEED_RESCHED must be visible before we test polling */ | |
1239 | smp_mb(); | |
1240 | if (!tsk_is_polling(rq->idle)) | |
1241 | smp_send_reschedule(cpu); | |
1242 | } | |
39c0cbe2 | 1243 | |
6d6bc0ad | 1244 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1245 | |
e9e9250b PZ |
1246 | static u64 sched_avg_period(void) |
1247 | { | |
1248 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1249 | } | |
1250 | ||
1251 | static void sched_avg_update(struct rq *rq) | |
1252 | { | |
1253 | s64 period = sched_avg_period(); | |
1254 | ||
1255 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1256 | /* |
1257 | * Inline assembly required to prevent the compiler | |
1258 | * optimising this loop into a divmod call. | |
1259 | * See __iter_div_u64_rem() for another example of this. | |
1260 | */ | |
1261 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1262 | rq->age_stamp += period; |
1263 | rq->rt_avg /= 2; | |
1264 | } | |
1265 | } | |
1266 | ||
1267 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1268 | { | |
1269 | rq->rt_avg += rt_delta; | |
1270 | sched_avg_update(rq); | |
1271 | } | |
1272 | ||
6d6bc0ad | 1273 | #else /* !CONFIG_SMP */ |
31656519 | 1274 | static void resched_task(struct task_struct *p) |
c24d20db | 1275 | { |
05fa785c | 1276 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1277 | set_tsk_need_resched(p); |
c24d20db | 1278 | } |
e9e9250b PZ |
1279 | |
1280 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1281 | { | |
1282 | } | |
da2b71ed SS |
1283 | |
1284 | static void sched_avg_update(struct rq *rq) | |
1285 | { | |
1286 | } | |
6d6bc0ad | 1287 | #endif /* CONFIG_SMP */ |
c24d20db | 1288 | |
45bf76df IM |
1289 | #if BITS_PER_LONG == 32 |
1290 | # define WMULT_CONST (~0UL) | |
1291 | #else | |
1292 | # define WMULT_CONST (1UL << 32) | |
1293 | #endif | |
1294 | ||
1295 | #define WMULT_SHIFT 32 | |
1296 | ||
194081eb IM |
1297 | /* |
1298 | * Shift right and round: | |
1299 | */ | |
cf2ab469 | 1300 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1301 | |
a7be37ac PZ |
1302 | /* |
1303 | * delta *= weight / lw | |
1304 | */ | |
cb1c4fc9 | 1305 | static unsigned long |
45bf76df IM |
1306 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1307 | struct load_weight *lw) | |
1308 | { | |
1309 | u64 tmp; | |
1310 | ||
7a232e03 LJ |
1311 | if (!lw->inv_weight) { |
1312 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1313 | lw->inv_weight = 1; | |
1314 | else | |
1315 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1316 | / (lw->weight+1); | |
1317 | } | |
45bf76df IM |
1318 | |
1319 | tmp = (u64)delta_exec * weight; | |
1320 | /* | |
1321 | * Check whether we'd overflow the 64-bit multiplication: | |
1322 | */ | |
194081eb | 1323 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1324 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1325 | WMULT_SHIFT/2); |
1326 | else | |
cf2ab469 | 1327 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1328 | |
ecf691da | 1329 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1333 | { |
1334 | lw->weight += inc; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
1091985b | 1338 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1339 | { |
1340 | lw->weight -= dec; | |
e89996ae | 1341 | lw->inv_weight = 0; |
45bf76df IM |
1342 | } |
1343 | ||
2069dd75 PZ |
1344 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1345 | { | |
1346 | lw->weight = w; | |
1347 | lw->inv_weight = 0; | |
1348 | } | |
1349 | ||
2dd73a4f PW |
1350 | /* |
1351 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1352 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1353 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1354 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1355 | * scaled version of the new time slice allocation that they receive on time |
1356 | * slice expiry etc. | |
1357 | */ | |
1358 | ||
cce7ade8 PZ |
1359 | #define WEIGHT_IDLEPRIO 3 |
1360 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1361 | |
1362 | /* | |
1363 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1364 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1365 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1366 | * that remained on nice 0. | |
1367 | * | |
1368 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1369 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1370 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1371 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1372 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1373 | */ |
1374 | static const int prio_to_weight[40] = { | |
254753dc IM |
1375 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1376 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1377 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1378 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1379 | /* 0 */ 1024, 820, 655, 526, 423, | |
1380 | /* 5 */ 335, 272, 215, 172, 137, | |
1381 | /* 10 */ 110, 87, 70, 56, 45, | |
1382 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1383 | }; |
1384 | ||
5714d2de IM |
1385 | /* |
1386 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1387 | * | |
1388 | * In cases where the weight does not change often, we can use the | |
1389 | * precalculated inverse to speed up arithmetics by turning divisions | |
1390 | * into multiplications: | |
1391 | */ | |
dd41f596 | 1392 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1393 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1394 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1395 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1396 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1397 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1398 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1399 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1400 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1401 | }; |
2dd73a4f | 1402 | |
ef12fefa BR |
1403 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1404 | enum cpuacct_stat_index { | |
1405 | CPUACCT_STAT_USER, /* ... user mode */ | |
1406 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1407 | ||
1408 | CPUACCT_STAT_NSTATS, | |
1409 | }; | |
1410 | ||
d842de87 SV |
1411 | #ifdef CONFIG_CGROUP_CPUACCT |
1412 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1413 | static void cpuacct_update_stats(struct task_struct *tsk, |
1414 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1415 | #else |
1416 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1417 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1418 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1419 | #endif |
1420 | ||
18d95a28 PZ |
1421 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1422 | { | |
1423 | update_load_add(&rq->load, load); | |
1424 | } | |
1425 | ||
1426 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1427 | { | |
1428 | update_load_sub(&rq->load, load); | |
1429 | } | |
1430 | ||
7940ca36 | 1431 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1432 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1433 | |
1434 | /* | |
1435 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1436 | * leaving it for the final time. | |
1437 | */ | |
eb755805 | 1438 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1439 | { |
1440 | struct task_group *parent, *child; | |
eb755805 | 1441 | int ret; |
c09595f6 PZ |
1442 | |
1443 | rcu_read_lock(); | |
1444 | parent = &root_task_group; | |
1445 | down: | |
eb755805 PZ |
1446 | ret = (*down)(parent, data); |
1447 | if (ret) | |
1448 | goto out_unlock; | |
c09595f6 PZ |
1449 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1450 | parent = child; | |
1451 | goto down; | |
1452 | ||
1453 | up: | |
1454 | continue; | |
1455 | } | |
eb755805 PZ |
1456 | ret = (*up)(parent, data); |
1457 | if (ret) | |
1458 | goto out_unlock; | |
c09595f6 PZ |
1459 | |
1460 | child = parent; | |
1461 | parent = parent->parent; | |
1462 | if (parent) | |
1463 | goto up; | |
eb755805 | 1464 | out_unlock: |
c09595f6 | 1465 | rcu_read_unlock(); |
eb755805 PZ |
1466 | |
1467 | return ret; | |
c09595f6 PZ |
1468 | } |
1469 | ||
eb755805 PZ |
1470 | static int tg_nop(struct task_group *tg, void *data) |
1471 | { | |
1472 | return 0; | |
c09595f6 | 1473 | } |
eb755805 PZ |
1474 | #endif |
1475 | ||
1476 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1477 | /* Used instead of source_load when we know the type == 0 */ |
1478 | static unsigned long weighted_cpuload(const int cpu) | |
1479 | { | |
1480 | return cpu_rq(cpu)->load.weight; | |
1481 | } | |
1482 | ||
1483 | /* | |
1484 | * Return a low guess at the load of a migration-source cpu weighted | |
1485 | * according to the scheduling class and "nice" value. | |
1486 | * | |
1487 | * We want to under-estimate the load of migration sources, to | |
1488 | * balance conservatively. | |
1489 | */ | |
1490 | static unsigned long source_load(int cpu, int type) | |
1491 | { | |
1492 | struct rq *rq = cpu_rq(cpu); | |
1493 | unsigned long total = weighted_cpuload(cpu); | |
1494 | ||
1495 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1496 | return total; | |
1497 | ||
1498 | return min(rq->cpu_load[type-1], total); | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * Return a high guess at the load of a migration-target cpu weighted | |
1503 | * according to the scheduling class and "nice" value. | |
1504 | */ | |
1505 | static unsigned long target_load(int cpu, int type) | |
1506 | { | |
1507 | struct rq *rq = cpu_rq(cpu); | |
1508 | unsigned long total = weighted_cpuload(cpu); | |
1509 | ||
1510 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1511 | return total; | |
1512 | ||
1513 | return max(rq->cpu_load[type-1], total); | |
1514 | } | |
1515 | ||
ae154be1 PZ |
1516 | static unsigned long power_of(int cpu) |
1517 | { | |
e51fd5e2 | 1518 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1519 | } |
1520 | ||
eb755805 PZ |
1521 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1522 | ||
1523 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1524 | { | |
1525 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1526 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1527 | |
4cd42620 SR |
1528 | if (nr_running) |
1529 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1530 | else |
1531 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1532 | |
1533 | return rq->avg_load_per_task; | |
1534 | } | |
1535 | ||
1536 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1537 | |
c09595f6 | 1538 | /* |
c8cba857 PZ |
1539 | * Compute the cpu's hierarchical load factor for each task group. |
1540 | * This needs to be done in a top-down fashion because the load of a child | |
1541 | * group is a fraction of its parents load. | |
c09595f6 | 1542 | */ |
eb755805 | 1543 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1544 | { |
c8cba857 | 1545 | unsigned long load; |
eb755805 | 1546 | long cpu = (long)data; |
c09595f6 | 1547 | |
c8cba857 PZ |
1548 | if (!tg->parent) { |
1549 | load = cpu_rq(cpu)->load.weight; | |
1550 | } else { | |
1551 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1552 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1553 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1554 | } | |
c09595f6 | 1555 | |
c8cba857 | 1556 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1557 | |
eb755805 | 1558 | return 0; |
c09595f6 PZ |
1559 | } |
1560 | ||
eb755805 | 1561 | static void update_h_load(long cpu) |
c09595f6 | 1562 | { |
eb755805 | 1563 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1564 | } |
1565 | ||
18d95a28 PZ |
1566 | #endif |
1567 | ||
8f45e2b5 GH |
1568 | #ifdef CONFIG_PREEMPT |
1569 | ||
b78bb868 PZ |
1570 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1571 | ||
70574a99 | 1572 | /* |
8f45e2b5 GH |
1573 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1574 | * way at the expense of forcing extra atomic operations in all | |
1575 | * invocations. This assures that the double_lock is acquired using the | |
1576 | * same underlying policy as the spinlock_t on this architecture, which | |
1577 | * reduces latency compared to the unfair variant below. However, it | |
1578 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1579 | */ |
8f45e2b5 GH |
1580 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1581 | __releases(this_rq->lock) | |
1582 | __acquires(busiest->lock) | |
1583 | __acquires(this_rq->lock) | |
1584 | { | |
05fa785c | 1585 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1586 | double_rq_lock(this_rq, busiest); |
1587 | ||
1588 | return 1; | |
1589 | } | |
1590 | ||
1591 | #else | |
1592 | /* | |
1593 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1594 | * latency by eliminating extra atomic operations when the locks are | |
1595 | * already in proper order on entry. This favors lower cpu-ids and will | |
1596 | * grant the double lock to lower cpus over higher ids under contention, | |
1597 | * regardless of entry order into the function. | |
1598 | */ | |
1599 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1600 | __releases(this_rq->lock) |
1601 | __acquires(busiest->lock) | |
1602 | __acquires(this_rq->lock) | |
1603 | { | |
1604 | int ret = 0; | |
1605 | ||
05fa785c | 1606 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1607 | if (busiest < this_rq) { |
05fa785c TG |
1608 | raw_spin_unlock(&this_rq->lock); |
1609 | raw_spin_lock(&busiest->lock); | |
1610 | raw_spin_lock_nested(&this_rq->lock, | |
1611 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1612 | ret = 1; |
1613 | } else | |
05fa785c TG |
1614 | raw_spin_lock_nested(&busiest->lock, |
1615 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1616 | } |
1617 | return ret; | |
1618 | } | |
1619 | ||
8f45e2b5 GH |
1620 | #endif /* CONFIG_PREEMPT */ |
1621 | ||
1622 | /* | |
1623 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1624 | */ | |
1625 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1626 | { | |
1627 | if (unlikely(!irqs_disabled())) { | |
1628 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1629 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1630 | BUG_ON(1); |
1631 | } | |
1632 | ||
1633 | return _double_lock_balance(this_rq, busiest); | |
1634 | } | |
1635 | ||
70574a99 AD |
1636 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1637 | __releases(busiest->lock) | |
1638 | { | |
05fa785c | 1639 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1640 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1641 | } | |
1e3c88bd PZ |
1642 | |
1643 | /* | |
1644 | * double_rq_lock - safely lock two runqueues | |
1645 | * | |
1646 | * Note this does not disable interrupts like task_rq_lock, | |
1647 | * you need to do so manually before calling. | |
1648 | */ | |
1649 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1650 | __acquires(rq1->lock) | |
1651 | __acquires(rq2->lock) | |
1652 | { | |
1653 | BUG_ON(!irqs_disabled()); | |
1654 | if (rq1 == rq2) { | |
1655 | raw_spin_lock(&rq1->lock); | |
1656 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1657 | } else { | |
1658 | if (rq1 < rq2) { | |
1659 | raw_spin_lock(&rq1->lock); | |
1660 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1661 | } else { | |
1662 | raw_spin_lock(&rq2->lock); | |
1663 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1664 | } | |
1665 | } | |
1e3c88bd PZ |
1666 | } |
1667 | ||
1668 | /* | |
1669 | * double_rq_unlock - safely unlock two runqueues | |
1670 | * | |
1671 | * Note this does not restore interrupts like task_rq_unlock, | |
1672 | * you need to do so manually after calling. | |
1673 | */ | |
1674 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1675 | __releases(rq1->lock) | |
1676 | __releases(rq2->lock) | |
1677 | { | |
1678 | raw_spin_unlock(&rq1->lock); | |
1679 | if (rq1 != rq2) | |
1680 | raw_spin_unlock(&rq2->lock); | |
1681 | else | |
1682 | __release(rq2->lock); | |
1683 | } | |
1684 | ||
d95f4122 MG |
1685 | #else /* CONFIG_SMP */ |
1686 | ||
1687 | /* | |
1688 | * double_rq_lock - safely lock two runqueues | |
1689 | * | |
1690 | * Note this does not disable interrupts like task_rq_lock, | |
1691 | * you need to do so manually before calling. | |
1692 | */ | |
1693 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1694 | __acquires(rq1->lock) | |
1695 | __acquires(rq2->lock) | |
1696 | { | |
1697 | BUG_ON(!irqs_disabled()); | |
1698 | BUG_ON(rq1 != rq2); | |
1699 | raw_spin_lock(&rq1->lock); | |
1700 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1701 | } | |
1702 | ||
1703 | /* | |
1704 | * double_rq_unlock - safely unlock two runqueues | |
1705 | * | |
1706 | * Note this does not restore interrupts like task_rq_unlock, | |
1707 | * you need to do so manually after calling. | |
1708 | */ | |
1709 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1710 | __releases(rq1->lock) | |
1711 | __releases(rq2->lock) | |
1712 | { | |
1713 | BUG_ON(rq1 != rq2); | |
1714 | raw_spin_unlock(&rq1->lock); | |
1715 | __release(rq2->lock); | |
1716 | } | |
1717 | ||
18d95a28 PZ |
1718 | #endif |
1719 | ||
74f5187a | 1720 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1721 | static void update_sysctl(void); |
acb4a848 | 1722 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1723 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1724 | |
cd29fe6f PZ |
1725 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1726 | { | |
1727 | set_task_rq(p, cpu); | |
1728 | #ifdef CONFIG_SMP | |
1729 | /* | |
1730 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1731 | * successfuly executed on another CPU. We must ensure that updates of | |
1732 | * per-task data have been completed by this moment. | |
1733 | */ | |
1734 | smp_wmb(); | |
1735 | task_thread_info(p)->cpu = cpu; | |
1736 | #endif | |
1737 | } | |
dce48a84 | 1738 | |
1e3c88bd | 1739 | static const struct sched_class rt_sched_class; |
dd41f596 | 1740 | |
34f971f6 | 1741 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1742 | #define for_each_class(class) \ |
1743 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1744 | |
1e3c88bd PZ |
1745 | #include "sched_stats.h" |
1746 | ||
c09595f6 | 1747 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1748 | { |
1749 | rq->nr_running++; | |
9c217245 IM |
1750 | } |
1751 | ||
c09595f6 | 1752 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1753 | { |
1754 | rq->nr_running--; | |
9c217245 IM |
1755 | } |
1756 | ||
45bf76df IM |
1757 | static void set_load_weight(struct task_struct *p) |
1758 | { | |
dd41f596 IM |
1759 | /* |
1760 | * SCHED_IDLE tasks get minimal weight: | |
1761 | */ | |
1762 | if (p->policy == SCHED_IDLE) { | |
1763 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1764 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1765 | return; | |
1766 | } | |
71f8bd46 | 1767 | |
dd41f596 IM |
1768 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1769 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1770 | } |
1771 | ||
371fd7e7 | 1772 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1773 | { |
a64692a3 | 1774 | update_rq_clock(rq); |
dd41f596 | 1775 | sched_info_queued(p); |
371fd7e7 | 1776 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1777 | p->se.on_rq = 1; |
71f8bd46 IM |
1778 | } |
1779 | ||
371fd7e7 | 1780 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1781 | { |
a64692a3 | 1782 | update_rq_clock(rq); |
46ac22ba | 1783 | sched_info_dequeued(p); |
371fd7e7 | 1784 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1785 | p->se.on_rq = 0; |
71f8bd46 IM |
1786 | } |
1787 | ||
1e3c88bd PZ |
1788 | /* |
1789 | * activate_task - move a task to the runqueue. | |
1790 | */ | |
371fd7e7 | 1791 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1792 | { |
1793 | if (task_contributes_to_load(p)) | |
1794 | rq->nr_uninterruptible--; | |
1795 | ||
371fd7e7 | 1796 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1797 | inc_nr_running(rq); |
1798 | } | |
1799 | ||
1800 | /* | |
1801 | * deactivate_task - remove a task from the runqueue. | |
1802 | */ | |
371fd7e7 | 1803 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1804 | { |
1805 | if (task_contributes_to_load(p)) | |
1806 | rq->nr_uninterruptible++; | |
1807 | ||
371fd7e7 | 1808 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1809 | dec_nr_running(rq); |
1810 | } | |
1811 | ||
b52bfee4 VP |
1812 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1813 | ||
305e6835 VP |
1814 | /* |
1815 | * There are no locks covering percpu hardirq/softirq time. | |
1816 | * They are only modified in account_system_vtime, on corresponding CPU | |
1817 | * with interrupts disabled. So, writes are safe. | |
1818 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1819 | * This may result in other CPU reading this CPU's irq time and can | |
1820 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1821 | * or new value with a side effect of accounting a slice of irq time to wrong |
1822 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1823 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1824 | */ |
b52bfee4 VP |
1825 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1826 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1827 | ||
1828 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1829 | static int sched_clock_irqtime; | |
1830 | ||
1831 | void enable_sched_clock_irqtime(void) | |
1832 | { | |
1833 | sched_clock_irqtime = 1; | |
1834 | } | |
1835 | ||
1836 | void disable_sched_clock_irqtime(void) | |
1837 | { | |
1838 | sched_clock_irqtime = 0; | |
1839 | } | |
1840 | ||
8e92c201 PZ |
1841 | #ifndef CONFIG_64BIT |
1842 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1843 | ||
1844 | static inline void irq_time_write_begin(void) | |
1845 | { | |
1846 | __this_cpu_inc(irq_time_seq.sequence); | |
1847 | smp_wmb(); | |
1848 | } | |
1849 | ||
1850 | static inline void irq_time_write_end(void) | |
1851 | { | |
1852 | smp_wmb(); | |
1853 | __this_cpu_inc(irq_time_seq.sequence); | |
1854 | } | |
1855 | ||
1856 | static inline u64 irq_time_read(int cpu) | |
1857 | { | |
1858 | u64 irq_time; | |
1859 | unsigned seq; | |
1860 | ||
1861 | do { | |
1862 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1863 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1864 | per_cpu(cpu_hardirq_time, cpu); | |
1865 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1866 | ||
1867 | return irq_time; | |
1868 | } | |
1869 | #else /* CONFIG_64BIT */ | |
1870 | static inline void irq_time_write_begin(void) | |
1871 | { | |
1872 | } | |
1873 | ||
1874 | static inline void irq_time_write_end(void) | |
1875 | { | |
1876 | } | |
1877 | ||
1878 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1879 | { |
305e6835 VP |
1880 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1881 | } | |
8e92c201 | 1882 | #endif /* CONFIG_64BIT */ |
305e6835 | 1883 | |
fe44d621 PZ |
1884 | /* |
1885 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1886 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1887 | */ | |
b52bfee4 VP |
1888 | void account_system_vtime(struct task_struct *curr) |
1889 | { | |
1890 | unsigned long flags; | |
fe44d621 | 1891 | s64 delta; |
b52bfee4 | 1892 | int cpu; |
b52bfee4 VP |
1893 | |
1894 | if (!sched_clock_irqtime) | |
1895 | return; | |
1896 | ||
1897 | local_irq_save(flags); | |
1898 | ||
b52bfee4 | 1899 | cpu = smp_processor_id(); |
fe44d621 PZ |
1900 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1901 | __this_cpu_add(irq_start_time, delta); | |
1902 | ||
8e92c201 | 1903 | irq_time_write_begin(); |
b52bfee4 VP |
1904 | /* |
1905 | * We do not account for softirq time from ksoftirqd here. | |
1906 | * We want to continue accounting softirq time to ksoftirqd thread | |
1907 | * in that case, so as not to confuse scheduler with a special task | |
1908 | * that do not consume any time, but still wants to run. | |
1909 | */ | |
1910 | if (hardirq_count()) | |
fe44d621 | 1911 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1912 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1913 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1914 | |
8e92c201 | 1915 | irq_time_write_end(); |
b52bfee4 VP |
1916 | local_irq_restore(flags); |
1917 | } | |
b7dadc38 | 1918 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1919 | |
fe44d621 | 1920 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1921 | { |
fe44d621 PZ |
1922 | s64 irq_delta; |
1923 | ||
8e92c201 | 1924 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1925 | |
1926 | /* | |
1927 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1928 | * this case when a previous update_rq_clock() happened inside a | |
1929 | * {soft,}irq region. | |
1930 | * | |
1931 | * When this happens, we stop ->clock_task and only update the | |
1932 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1933 | * update will consume the rest. This ensures ->clock_task is | |
1934 | * monotonic. | |
1935 | * | |
1936 | * It does however cause some slight miss-attribution of {soft,}irq | |
1937 | * time, a more accurate solution would be to update the irq_time using | |
1938 | * the current rq->clock timestamp, except that would require using | |
1939 | * atomic ops. | |
1940 | */ | |
1941 | if (irq_delta > delta) | |
1942 | irq_delta = delta; | |
1943 | ||
1944 | rq->prev_irq_time += irq_delta; | |
1945 | delta -= irq_delta; | |
1946 | rq->clock_task += delta; | |
1947 | ||
1948 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1949 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1950 | } |
1951 | ||
abb74cef VP |
1952 | static int irqtime_account_hi_update(void) |
1953 | { | |
1954 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1955 | unsigned long flags; | |
1956 | u64 latest_ns; | |
1957 | int ret = 0; | |
1958 | ||
1959 | local_irq_save(flags); | |
1960 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1961 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1962 | ret = 1; | |
1963 | local_irq_restore(flags); | |
1964 | return ret; | |
1965 | } | |
1966 | ||
1967 | static int irqtime_account_si_update(void) | |
1968 | { | |
1969 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1970 | unsigned long flags; | |
1971 | u64 latest_ns; | |
1972 | int ret = 0; | |
1973 | ||
1974 | local_irq_save(flags); | |
1975 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1976 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1977 | ret = 1; | |
1978 | local_irq_restore(flags); | |
1979 | return ret; | |
1980 | } | |
1981 | ||
fe44d621 | 1982 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1983 | |
abb74cef VP |
1984 | #define sched_clock_irqtime (0) |
1985 | ||
fe44d621 | 1986 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1987 | { |
fe44d621 | 1988 | rq->clock_task += delta; |
305e6835 VP |
1989 | } |
1990 | ||
fe44d621 | 1991 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 1992 | |
1e3c88bd PZ |
1993 | #include "sched_idletask.c" |
1994 | #include "sched_fair.c" | |
1995 | #include "sched_rt.c" | |
5091faa4 | 1996 | #include "sched_autogroup.c" |
34f971f6 | 1997 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1998 | #ifdef CONFIG_SCHED_DEBUG |
1999 | # include "sched_debug.c" | |
2000 | #endif | |
2001 | ||
34f971f6 PZ |
2002 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2003 | { | |
2004 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2005 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2006 | ||
2007 | if (stop) { | |
2008 | /* | |
2009 | * Make it appear like a SCHED_FIFO task, its something | |
2010 | * userspace knows about and won't get confused about. | |
2011 | * | |
2012 | * Also, it will make PI more or less work without too | |
2013 | * much confusion -- but then, stop work should not | |
2014 | * rely on PI working anyway. | |
2015 | */ | |
2016 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2017 | ||
2018 | stop->sched_class = &stop_sched_class; | |
2019 | } | |
2020 | ||
2021 | cpu_rq(cpu)->stop = stop; | |
2022 | ||
2023 | if (old_stop) { | |
2024 | /* | |
2025 | * Reset it back to a normal scheduling class so that | |
2026 | * it can die in pieces. | |
2027 | */ | |
2028 | old_stop->sched_class = &rt_sched_class; | |
2029 | } | |
2030 | } | |
2031 | ||
14531189 | 2032 | /* |
dd41f596 | 2033 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2034 | */ |
14531189 IM |
2035 | static inline int __normal_prio(struct task_struct *p) |
2036 | { | |
dd41f596 | 2037 | return p->static_prio; |
14531189 IM |
2038 | } |
2039 | ||
b29739f9 IM |
2040 | /* |
2041 | * Calculate the expected normal priority: i.e. priority | |
2042 | * without taking RT-inheritance into account. Might be | |
2043 | * boosted by interactivity modifiers. Changes upon fork, | |
2044 | * setprio syscalls, and whenever the interactivity | |
2045 | * estimator recalculates. | |
2046 | */ | |
36c8b586 | 2047 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2048 | { |
2049 | int prio; | |
2050 | ||
e05606d3 | 2051 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2052 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2053 | else | |
2054 | prio = __normal_prio(p); | |
2055 | return prio; | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Calculate the current priority, i.e. the priority | |
2060 | * taken into account by the scheduler. This value might | |
2061 | * be boosted by RT tasks, or might be boosted by | |
2062 | * interactivity modifiers. Will be RT if the task got | |
2063 | * RT-boosted. If not then it returns p->normal_prio. | |
2064 | */ | |
36c8b586 | 2065 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2066 | { |
2067 | p->normal_prio = normal_prio(p); | |
2068 | /* | |
2069 | * If we are RT tasks or we were boosted to RT priority, | |
2070 | * keep the priority unchanged. Otherwise, update priority | |
2071 | * to the normal priority: | |
2072 | */ | |
2073 | if (!rt_prio(p->prio)) | |
2074 | return p->normal_prio; | |
2075 | return p->prio; | |
2076 | } | |
2077 | ||
1da177e4 LT |
2078 | /** |
2079 | * task_curr - is this task currently executing on a CPU? | |
2080 | * @p: the task in question. | |
2081 | */ | |
36c8b586 | 2082 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2083 | { |
2084 | return cpu_curr(task_cpu(p)) == p; | |
2085 | } | |
2086 | ||
cb469845 SR |
2087 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2088 | const struct sched_class *prev_class, | |
da7a735e | 2089 | int oldprio) |
cb469845 SR |
2090 | { |
2091 | if (prev_class != p->sched_class) { | |
2092 | if (prev_class->switched_from) | |
da7a735e PZ |
2093 | prev_class->switched_from(rq, p); |
2094 | p->sched_class->switched_to(rq, p); | |
2095 | } else if (oldprio != p->prio) | |
2096 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2097 | } |
2098 | ||
1e5a7405 PZ |
2099 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2100 | { | |
2101 | const struct sched_class *class; | |
2102 | ||
2103 | if (p->sched_class == rq->curr->sched_class) { | |
2104 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2105 | } else { | |
2106 | for_each_class(class) { | |
2107 | if (class == rq->curr->sched_class) | |
2108 | break; | |
2109 | if (class == p->sched_class) { | |
2110 | resched_task(rq->curr); | |
2111 | break; | |
2112 | } | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * A queue event has occurred, and we're going to schedule. In | |
2118 | * this case, we can save a useless back to back clock update. | |
2119 | */ | |
f26f9aff | 2120 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2121 | rq->skip_clock_update = 1; |
2122 | } | |
2123 | ||
1da177e4 | 2124 | #ifdef CONFIG_SMP |
cc367732 IM |
2125 | /* |
2126 | * Is this task likely cache-hot: | |
2127 | */ | |
e7693a36 | 2128 | static int |
cc367732 IM |
2129 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2130 | { | |
2131 | s64 delta; | |
2132 | ||
e6c8fba7 PZ |
2133 | if (p->sched_class != &fair_sched_class) |
2134 | return 0; | |
2135 | ||
ef8002f6 NR |
2136 | if (unlikely(p->policy == SCHED_IDLE)) |
2137 | return 0; | |
2138 | ||
f540a608 IM |
2139 | /* |
2140 | * Buddy candidates are cache hot: | |
2141 | */ | |
f685ceac | 2142 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2143 | (&p->se == cfs_rq_of(&p->se)->next || |
2144 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2145 | return 1; |
2146 | ||
6bc1665b IM |
2147 | if (sysctl_sched_migration_cost == -1) |
2148 | return 1; | |
2149 | if (sysctl_sched_migration_cost == 0) | |
2150 | return 0; | |
2151 | ||
cc367732 IM |
2152 | delta = now - p->se.exec_start; |
2153 | ||
2154 | return delta < (s64)sysctl_sched_migration_cost; | |
2155 | } | |
2156 | ||
dd41f596 | 2157 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2158 | { |
e2912009 PZ |
2159 | #ifdef CONFIG_SCHED_DEBUG |
2160 | /* | |
2161 | * We should never call set_task_cpu() on a blocked task, | |
2162 | * ttwu() will sort out the placement. | |
2163 | */ | |
077614ee PZ |
2164 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2165 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2166 | #endif |
2167 | ||
de1d7286 | 2168 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2169 | |
0c69774e PZ |
2170 | if (task_cpu(p) != new_cpu) { |
2171 | p->se.nr_migrations++; | |
2172 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2173 | } | |
dd41f596 IM |
2174 | |
2175 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2176 | } |
2177 | ||
969c7921 | 2178 | struct migration_arg { |
36c8b586 | 2179 | struct task_struct *task; |
1da177e4 | 2180 | int dest_cpu; |
70b97a7f | 2181 | }; |
1da177e4 | 2182 | |
969c7921 TH |
2183 | static int migration_cpu_stop(void *data); |
2184 | ||
1da177e4 LT |
2185 | /* |
2186 | * The task's runqueue lock must be held. | |
2187 | * Returns true if you have to wait for migration thread. | |
2188 | */ | |
b7a2b39d | 2189 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2190 | { |
1da177e4 LT |
2191 | /* |
2192 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2193 | * the next wake-up will properly place the task. |
1da177e4 | 2194 | */ |
969c7921 | 2195 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2196 | } |
2197 | ||
2198 | /* | |
2199 | * wait_task_inactive - wait for a thread to unschedule. | |
2200 | * | |
85ba2d86 RM |
2201 | * If @match_state is nonzero, it's the @p->state value just checked and |
2202 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2203 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2204 | * we return a positive number (its total switch count). If a second call | |
2205 | * a short while later returns the same number, the caller can be sure that | |
2206 | * @p has remained unscheduled the whole time. | |
2207 | * | |
1da177e4 LT |
2208 | * The caller must ensure that the task *will* unschedule sometime soon, |
2209 | * else this function might spin for a *long* time. This function can't | |
2210 | * be called with interrupts off, or it may introduce deadlock with | |
2211 | * smp_call_function() if an IPI is sent by the same process we are | |
2212 | * waiting to become inactive. | |
2213 | */ | |
85ba2d86 | 2214 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2215 | { |
2216 | unsigned long flags; | |
dd41f596 | 2217 | int running, on_rq; |
85ba2d86 | 2218 | unsigned long ncsw; |
70b97a7f | 2219 | struct rq *rq; |
1da177e4 | 2220 | |
3a5c359a AK |
2221 | for (;;) { |
2222 | /* | |
2223 | * We do the initial early heuristics without holding | |
2224 | * any task-queue locks at all. We'll only try to get | |
2225 | * the runqueue lock when things look like they will | |
2226 | * work out! | |
2227 | */ | |
2228 | rq = task_rq(p); | |
fa490cfd | 2229 | |
3a5c359a AK |
2230 | /* |
2231 | * If the task is actively running on another CPU | |
2232 | * still, just relax and busy-wait without holding | |
2233 | * any locks. | |
2234 | * | |
2235 | * NOTE! Since we don't hold any locks, it's not | |
2236 | * even sure that "rq" stays as the right runqueue! | |
2237 | * But we don't care, since "task_running()" will | |
2238 | * return false if the runqueue has changed and p | |
2239 | * is actually now running somewhere else! | |
2240 | */ | |
85ba2d86 RM |
2241 | while (task_running(rq, p)) { |
2242 | if (match_state && unlikely(p->state != match_state)) | |
2243 | return 0; | |
3a5c359a | 2244 | cpu_relax(); |
85ba2d86 | 2245 | } |
fa490cfd | 2246 | |
3a5c359a AK |
2247 | /* |
2248 | * Ok, time to look more closely! We need the rq | |
2249 | * lock now, to be *sure*. If we're wrong, we'll | |
2250 | * just go back and repeat. | |
2251 | */ | |
2252 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2253 | trace_sched_wait_task(p); |
3a5c359a AK |
2254 | running = task_running(rq, p); |
2255 | on_rq = p->se.on_rq; | |
85ba2d86 | 2256 | ncsw = 0; |
f31e11d8 | 2257 | if (!match_state || p->state == match_state) |
93dcf55f | 2258 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2259 | task_rq_unlock(rq, &flags); |
fa490cfd | 2260 | |
85ba2d86 RM |
2261 | /* |
2262 | * If it changed from the expected state, bail out now. | |
2263 | */ | |
2264 | if (unlikely(!ncsw)) | |
2265 | break; | |
2266 | ||
3a5c359a AK |
2267 | /* |
2268 | * Was it really running after all now that we | |
2269 | * checked with the proper locks actually held? | |
2270 | * | |
2271 | * Oops. Go back and try again.. | |
2272 | */ | |
2273 | if (unlikely(running)) { | |
2274 | cpu_relax(); | |
2275 | continue; | |
2276 | } | |
fa490cfd | 2277 | |
3a5c359a AK |
2278 | /* |
2279 | * It's not enough that it's not actively running, | |
2280 | * it must be off the runqueue _entirely_, and not | |
2281 | * preempted! | |
2282 | * | |
80dd99b3 | 2283 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2284 | * running right now), it's preempted, and we should |
2285 | * yield - it could be a while. | |
2286 | */ | |
2287 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2288 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2289 | ||
2290 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2291 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2292 | continue; |
2293 | } | |
fa490cfd | 2294 | |
3a5c359a AK |
2295 | /* |
2296 | * Ahh, all good. It wasn't running, and it wasn't | |
2297 | * runnable, which means that it will never become | |
2298 | * running in the future either. We're all done! | |
2299 | */ | |
2300 | break; | |
2301 | } | |
85ba2d86 RM |
2302 | |
2303 | return ncsw; | |
1da177e4 LT |
2304 | } |
2305 | ||
2306 | /*** | |
2307 | * kick_process - kick a running thread to enter/exit the kernel | |
2308 | * @p: the to-be-kicked thread | |
2309 | * | |
2310 | * Cause a process which is running on another CPU to enter | |
2311 | * kernel-mode, without any delay. (to get signals handled.) | |
2312 | * | |
25985edc | 2313 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2314 | * because all it wants to ensure is that the remote task enters |
2315 | * the kernel. If the IPI races and the task has been migrated | |
2316 | * to another CPU then no harm is done and the purpose has been | |
2317 | * achieved as well. | |
2318 | */ | |
36c8b586 | 2319 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2320 | { |
2321 | int cpu; | |
2322 | ||
2323 | preempt_disable(); | |
2324 | cpu = task_cpu(p); | |
2325 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2326 | smp_send_reschedule(cpu); | |
2327 | preempt_enable(); | |
2328 | } | |
b43e3521 | 2329 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2330 | #endif /* CONFIG_SMP */ |
1da177e4 | 2331 | |
970b13ba | 2332 | #ifdef CONFIG_SMP |
30da688e ON |
2333 | /* |
2334 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2335 | */ | |
5da9a0fb PZ |
2336 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2337 | { | |
2338 | int dest_cpu; | |
2339 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2340 | ||
2341 | /* Look for allowed, online CPU in same node. */ | |
2342 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2343 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2344 | return dest_cpu; | |
2345 | ||
2346 | /* Any allowed, online CPU? */ | |
2347 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2348 | if (dest_cpu < nr_cpu_ids) | |
2349 | return dest_cpu; | |
2350 | ||
2351 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2352 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2353 | /* | |
2354 | * Don't tell them about moving exiting tasks or | |
2355 | * kernel threads (both mm NULL), since they never | |
2356 | * leave kernel. | |
2357 | */ | |
2358 | if (p->mm && printk_ratelimit()) { | |
2359 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2360 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2361 | } |
2362 | ||
2363 | return dest_cpu; | |
2364 | } | |
2365 | ||
e2912009 | 2366 | /* |
30da688e | 2367 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2368 | */ |
970b13ba | 2369 | static inline |
0017d735 | 2370 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2371 | { |
0017d735 | 2372 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2373 | |
2374 | /* | |
2375 | * In order not to call set_task_cpu() on a blocking task we need | |
2376 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2377 | * cpu. | |
2378 | * | |
2379 | * Since this is common to all placement strategies, this lives here. | |
2380 | * | |
2381 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2382 | * not worry about this generic constraint ] | |
2383 | */ | |
2384 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2385 | !cpu_online(cpu))) |
5da9a0fb | 2386 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2387 | |
2388 | return cpu; | |
970b13ba | 2389 | } |
09a40af5 MG |
2390 | |
2391 | static void update_avg(u64 *avg, u64 sample) | |
2392 | { | |
2393 | s64 diff = sample - *avg; | |
2394 | *avg += diff >> 3; | |
2395 | } | |
970b13ba PZ |
2396 | #endif |
2397 | ||
9ed3811a TH |
2398 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2399 | bool is_sync, bool is_migrate, bool is_local, | |
2400 | unsigned long en_flags) | |
2401 | { | |
2402 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2403 | if (is_sync) | |
2404 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2405 | if (is_migrate) | |
2406 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2407 | if (is_local) | |
2408 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2409 | else | |
2410 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2411 | ||
2412 | activate_task(rq, p, en_flags); | |
2413 | } | |
2414 | ||
2415 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2416 | int wake_flags, bool success) | |
2417 | { | |
2418 | trace_sched_wakeup(p, success); | |
2419 | check_preempt_curr(rq, p, wake_flags); | |
2420 | ||
2421 | p->state = TASK_RUNNING; | |
2422 | #ifdef CONFIG_SMP | |
2423 | if (p->sched_class->task_woken) | |
2424 | p->sched_class->task_woken(rq, p); | |
2425 | ||
2426 | if (unlikely(rq->idle_stamp)) { | |
2427 | u64 delta = rq->clock - rq->idle_stamp; | |
2428 | u64 max = 2*sysctl_sched_migration_cost; | |
2429 | ||
2430 | if (delta > max) | |
2431 | rq->avg_idle = max; | |
2432 | else | |
2433 | update_avg(&rq->avg_idle, delta); | |
2434 | rq->idle_stamp = 0; | |
2435 | } | |
2436 | #endif | |
21aa9af0 TH |
2437 | /* if a worker is waking up, notify workqueue */ |
2438 | if ((p->flags & PF_WQ_WORKER) && success) | |
2439 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2440 | } |
2441 | ||
2442 | /** | |
1da177e4 | 2443 | * try_to_wake_up - wake up a thread |
9ed3811a | 2444 | * @p: the thread to be awakened |
1da177e4 | 2445 | * @state: the mask of task states that can be woken |
9ed3811a | 2446 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2447 | * |
2448 | * Put it on the run-queue if it's not already there. The "current" | |
2449 | * thread is always on the run-queue (except when the actual | |
2450 | * re-schedule is in progress), and as such you're allowed to do | |
2451 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2452 | * runnable without the overhead of this. | |
2453 | * | |
9ed3811a TH |
2454 | * Returns %true if @p was woken up, %false if it was already running |
2455 | * or @state didn't match @p's state. | |
1da177e4 | 2456 | */ |
7d478721 PZ |
2457 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2458 | int wake_flags) | |
1da177e4 | 2459 | { |
cc367732 | 2460 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2461 | unsigned long flags; |
371fd7e7 | 2462 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2463 | struct rq *rq; |
1da177e4 | 2464 | |
e9c84311 | 2465 | this_cpu = get_cpu(); |
2398f2c6 | 2466 | |
04e2f174 | 2467 | smp_wmb(); |
ab3b3aa5 | 2468 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2469 | if (!(p->state & state)) |
1da177e4 LT |
2470 | goto out; |
2471 | ||
dd41f596 | 2472 | if (p->se.on_rq) |
1da177e4 LT |
2473 | goto out_running; |
2474 | ||
2475 | cpu = task_cpu(p); | |
cc367732 | 2476 | orig_cpu = cpu; |
1da177e4 LT |
2477 | |
2478 | #ifdef CONFIG_SMP | |
2479 | if (unlikely(task_running(rq, p))) | |
2480 | goto out_activate; | |
2481 | ||
e9c84311 PZ |
2482 | /* |
2483 | * In order to handle concurrent wakeups and release the rq->lock | |
2484 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2485 | * |
2486 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2487 | */ |
cc87f76a PZ |
2488 | if (task_contributes_to_load(p)) { |
2489 | if (likely(cpu_online(orig_cpu))) | |
2490 | rq->nr_uninterruptible--; | |
2491 | else | |
2492 | this_rq()->nr_uninterruptible--; | |
2493 | } | |
e9c84311 | 2494 | p->state = TASK_WAKING; |
efbbd05a | 2495 | |
371fd7e7 | 2496 | if (p->sched_class->task_waking) { |
efbbd05a | 2497 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2498 | en_flags |= ENQUEUE_WAKING; |
2499 | } | |
efbbd05a | 2500 | |
0017d735 PZ |
2501 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2502 | if (cpu != orig_cpu) | |
5d2f5a61 | 2503 | set_task_cpu(p, cpu); |
0017d735 | 2504 | __task_rq_unlock(rq); |
ab19cb23 | 2505 | |
0970d299 PZ |
2506 | rq = cpu_rq(cpu); |
2507 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2508 | |
0970d299 PZ |
2509 | /* |
2510 | * We migrated the task without holding either rq->lock, however | |
2511 | * since the task is not on the task list itself, nobody else | |
2512 | * will try and migrate the task, hence the rq should match the | |
2513 | * cpu we just moved it to. | |
2514 | */ | |
2515 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2516 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2517 | |
e7693a36 GH |
2518 | #ifdef CONFIG_SCHEDSTATS |
2519 | schedstat_inc(rq, ttwu_count); | |
2520 | if (cpu == this_cpu) | |
2521 | schedstat_inc(rq, ttwu_local); | |
2522 | else { | |
2523 | struct sched_domain *sd; | |
2524 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2525 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2526 | schedstat_inc(sd, ttwu_wake_remote); |
2527 | break; | |
2528 | } | |
2529 | } | |
2530 | } | |
6d6bc0ad | 2531 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2532 | |
1da177e4 LT |
2533 | out_activate: |
2534 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2535 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2536 | cpu == this_cpu, en_flags); | |
1da177e4 | 2537 | success = 1; |
1da177e4 | 2538 | out_running: |
9ed3811a | 2539 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2540 | out: |
2541 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2542 | put_cpu(); |
1da177e4 LT |
2543 | |
2544 | return success; | |
2545 | } | |
2546 | ||
21aa9af0 TH |
2547 | /** |
2548 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2549 | * @p: the thread to be awakened | |
2550 | * | |
b595076a | 2551 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 TH |
2552 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2553 | * the current task. this_rq() stays locked over invocation. | |
2554 | */ | |
2555 | static void try_to_wake_up_local(struct task_struct *p) | |
2556 | { | |
2557 | struct rq *rq = task_rq(p); | |
2558 | bool success = false; | |
2559 | ||
2560 | BUG_ON(rq != this_rq()); | |
2561 | BUG_ON(p == current); | |
2562 | lockdep_assert_held(&rq->lock); | |
2563 | ||
2564 | if (!(p->state & TASK_NORMAL)) | |
2565 | return; | |
2566 | ||
2567 | if (!p->se.on_rq) { | |
2568 | if (likely(!task_running(rq, p))) { | |
2569 | schedstat_inc(rq, ttwu_count); | |
2570 | schedstat_inc(rq, ttwu_local); | |
2571 | } | |
2572 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2573 | success = true; | |
2574 | } | |
2575 | ttwu_post_activation(p, rq, 0, success); | |
2576 | } | |
2577 | ||
50fa610a DH |
2578 | /** |
2579 | * wake_up_process - Wake up a specific process | |
2580 | * @p: The process to be woken up. | |
2581 | * | |
2582 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2583 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2584 | * running. | |
2585 | * | |
2586 | * It may be assumed that this function implies a write memory barrier before | |
2587 | * changing the task state if and only if any tasks are woken up. | |
2588 | */ | |
7ad5b3a5 | 2589 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2590 | { |
d9514f6c | 2591 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2592 | } |
1da177e4 LT |
2593 | EXPORT_SYMBOL(wake_up_process); |
2594 | ||
7ad5b3a5 | 2595 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2596 | { |
2597 | return try_to_wake_up(p, state, 0); | |
2598 | } | |
2599 | ||
1da177e4 LT |
2600 | /* |
2601 | * Perform scheduler related setup for a newly forked process p. | |
2602 | * p is forked by current. | |
dd41f596 IM |
2603 | * |
2604 | * __sched_fork() is basic setup used by init_idle() too: | |
2605 | */ | |
2606 | static void __sched_fork(struct task_struct *p) | |
2607 | { | |
dd41f596 IM |
2608 | p->se.exec_start = 0; |
2609 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2610 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2611 | p->se.nr_migrations = 0; |
da7a735e | 2612 | p->se.vruntime = 0; |
6cfb0d5d IM |
2613 | |
2614 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2615 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2616 | #endif |
476d139c | 2617 | |
fa717060 | 2618 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2619 | p->se.on_rq = 0; |
4a55bd5e | 2620 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2621 | |
e107be36 AK |
2622 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2623 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2624 | #endif | |
dd41f596 IM |
2625 | } |
2626 | ||
2627 | /* | |
2628 | * fork()/clone()-time setup: | |
2629 | */ | |
2630 | void sched_fork(struct task_struct *p, int clone_flags) | |
2631 | { | |
2632 | int cpu = get_cpu(); | |
2633 | ||
2634 | __sched_fork(p); | |
06b83b5f | 2635 | /* |
0017d735 | 2636 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2637 | * nobody will actually run it, and a signal or other external |
2638 | * event cannot wake it up and insert it on the runqueue either. | |
2639 | */ | |
0017d735 | 2640 | p->state = TASK_RUNNING; |
dd41f596 | 2641 | |
b9dc29e7 MG |
2642 | /* |
2643 | * Revert to default priority/policy on fork if requested. | |
2644 | */ | |
2645 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2646 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2647 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2648 | p->normal_prio = p->static_prio; |
2649 | } | |
b9dc29e7 | 2650 | |
6c697bdf MG |
2651 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2652 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2653 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2654 | set_load_weight(p); |
2655 | } | |
2656 | ||
b9dc29e7 MG |
2657 | /* |
2658 | * We don't need the reset flag anymore after the fork. It has | |
2659 | * fulfilled its duty: | |
2660 | */ | |
2661 | p->sched_reset_on_fork = 0; | |
2662 | } | |
ca94c442 | 2663 | |
f83f9ac2 PW |
2664 | /* |
2665 | * Make sure we do not leak PI boosting priority to the child. | |
2666 | */ | |
2667 | p->prio = current->normal_prio; | |
2668 | ||
2ddbf952 HS |
2669 | if (!rt_prio(p->prio)) |
2670 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2671 | |
cd29fe6f PZ |
2672 | if (p->sched_class->task_fork) |
2673 | p->sched_class->task_fork(p); | |
2674 | ||
86951599 PZ |
2675 | /* |
2676 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2677 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2678 | * is ran before sched_fork(). | |
2679 | * | |
2680 | * Silence PROVE_RCU. | |
2681 | */ | |
2682 | rcu_read_lock(); | |
5f3edc1b | 2683 | set_task_cpu(p, cpu); |
86951599 | 2684 | rcu_read_unlock(); |
5f3edc1b | 2685 | |
52f17b6c | 2686 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2687 | if (likely(sched_info_on())) |
52f17b6c | 2688 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2689 | #endif |
d6077cb8 | 2690 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2691 | p->oncpu = 0; |
2692 | #endif | |
1da177e4 | 2693 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2694 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2695 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2696 | #endif |
806c09a7 | 2697 | #ifdef CONFIG_SMP |
917b627d | 2698 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2699 | #endif |
917b627d | 2700 | |
476d139c | 2701 | put_cpu(); |
1da177e4 LT |
2702 | } |
2703 | ||
2704 | /* | |
2705 | * wake_up_new_task - wake up a newly created task for the first time. | |
2706 | * | |
2707 | * This function will do some initial scheduler statistics housekeeping | |
2708 | * that must be done for every newly created context, then puts the task | |
2709 | * on the runqueue and wakes it. | |
2710 | */ | |
7ad5b3a5 | 2711 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2712 | { |
2713 | unsigned long flags; | |
dd41f596 | 2714 | struct rq *rq; |
c890692b | 2715 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2716 | |
2717 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2718 | rq = task_rq_lock(p, &flags); |
2719 | p->state = TASK_WAKING; | |
2720 | ||
fabf318e PZ |
2721 | /* |
2722 | * Fork balancing, do it here and not earlier because: | |
2723 | * - cpus_allowed can change in the fork path | |
2724 | * - any previously selected cpu might disappear through hotplug | |
2725 | * | |
0017d735 PZ |
2726 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2727 | * without people poking at ->cpus_allowed. | |
fabf318e | 2728 | */ |
0017d735 | 2729 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2730 | set_task_cpu(p, cpu); |
1da177e4 | 2731 | |
06b83b5f | 2732 | p->state = TASK_RUNNING; |
0017d735 PZ |
2733 | task_rq_unlock(rq, &flags); |
2734 | #endif | |
2735 | ||
2736 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2737 | activate_task(rq, p, 0); |
27a9da65 | 2738 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2739 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2740 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2741 | if (p->sched_class->task_woken) |
2742 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2743 | #endif |
dd41f596 | 2744 | task_rq_unlock(rq, &flags); |
fabf318e | 2745 | put_cpu(); |
1da177e4 LT |
2746 | } |
2747 | ||
e107be36 AK |
2748 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2749 | ||
2750 | /** | |
80dd99b3 | 2751 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2752 | * @notifier: notifier struct to register |
e107be36 AK |
2753 | */ |
2754 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2755 | { | |
2756 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2757 | } | |
2758 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2759 | ||
2760 | /** | |
2761 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2762 | * @notifier: notifier struct to unregister |
e107be36 AK |
2763 | * |
2764 | * This is safe to call from within a preemption notifier. | |
2765 | */ | |
2766 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2767 | { | |
2768 | hlist_del(¬ifier->link); | |
2769 | } | |
2770 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2771 | ||
2772 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2773 | { | |
2774 | struct preempt_notifier *notifier; | |
2775 | struct hlist_node *node; | |
2776 | ||
2777 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2778 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2779 | } | |
2780 | ||
2781 | static void | |
2782 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2783 | struct task_struct *next) | |
2784 | { | |
2785 | struct preempt_notifier *notifier; | |
2786 | struct hlist_node *node; | |
2787 | ||
2788 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2789 | notifier->ops->sched_out(notifier, next); | |
2790 | } | |
2791 | ||
6d6bc0ad | 2792 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2793 | |
2794 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2795 | { | |
2796 | } | |
2797 | ||
2798 | static void | |
2799 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2800 | struct task_struct *next) | |
2801 | { | |
2802 | } | |
2803 | ||
6d6bc0ad | 2804 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2805 | |
4866cde0 NP |
2806 | /** |
2807 | * prepare_task_switch - prepare to switch tasks | |
2808 | * @rq: the runqueue preparing to switch | |
421cee29 | 2809 | * @prev: the current task that is being switched out |
4866cde0 NP |
2810 | * @next: the task we are going to switch to. |
2811 | * | |
2812 | * This is called with the rq lock held and interrupts off. It must | |
2813 | * be paired with a subsequent finish_task_switch after the context | |
2814 | * switch. | |
2815 | * | |
2816 | * prepare_task_switch sets up locking and calls architecture specific | |
2817 | * hooks. | |
2818 | */ | |
e107be36 AK |
2819 | static inline void |
2820 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2821 | struct task_struct *next) | |
4866cde0 | 2822 | { |
fe4b04fa PZ |
2823 | sched_info_switch(prev, next); |
2824 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2825 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2826 | prepare_lock_switch(rq, next); |
2827 | prepare_arch_switch(next); | |
fe4b04fa | 2828 | trace_sched_switch(prev, next); |
4866cde0 NP |
2829 | } |
2830 | ||
1da177e4 LT |
2831 | /** |
2832 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2833 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2834 | * @prev: the thread we just switched away from. |
2835 | * | |
4866cde0 NP |
2836 | * finish_task_switch must be called after the context switch, paired |
2837 | * with a prepare_task_switch call before the context switch. | |
2838 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2839 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2840 | * |
2841 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2842 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2843 | * with the lock held can cause deadlocks; see schedule() for |
2844 | * details.) | |
2845 | */ | |
a9957449 | 2846 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2847 | __releases(rq->lock) |
2848 | { | |
1da177e4 | 2849 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2850 | long prev_state; |
1da177e4 LT |
2851 | |
2852 | rq->prev_mm = NULL; | |
2853 | ||
2854 | /* | |
2855 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2856 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2857 | * schedule one last time. The schedule call will never return, and |
2858 | * the scheduled task must drop that reference. | |
c394cc9f | 2859 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2860 | * still held, otherwise prev could be scheduled on another cpu, die |
2861 | * there before we look at prev->state, and then the reference would | |
2862 | * be dropped twice. | |
2863 | * Manfred Spraul <manfred@colorfullife.com> | |
2864 | */ | |
55a101f8 | 2865 | prev_state = prev->state; |
4866cde0 | 2866 | finish_arch_switch(prev); |
8381f65d JI |
2867 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2868 | local_irq_disable(); | |
2869 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2870 | perf_event_task_sched_in(current); |
8381f65d JI |
2871 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2872 | local_irq_enable(); | |
2873 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2874 | finish_lock_switch(rq, prev); |
e8fa1362 | 2875 | |
e107be36 | 2876 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2877 | if (mm) |
2878 | mmdrop(mm); | |
c394cc9f | 2879 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2880 | /* |
2881 | * Remove function-return probe instances associated with this | |
2882 | * task and put them back on the free list. | |
9761eea8 | 2883 | */ |
c6fd91f0 | 2884 | kprobe_flush_task(prev); |
1da177e4 | 2885 | put_task_struct(prev); |
c6fd91f0 | 2886 | } |
1da177e4 LT |
2887 | } |
2888 | ||
3f029d3c GH |
2889 | #ifdef CONFIG_SMP |
2890 | ||
2891 | /* assumes rq->lock is held */ | |
2892 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2893 | { | |
2894 | if (prev->sched_class->pre_schedule) | |
2895 | prev->sched_class->pre_schedule(rq, prev); | |
2896 | } | |
2897 | ||
2898 | /* rq->lock is NOT held, but preemption is disabled */ | |
2899 | static inline void post_schedule(struct rq *rq) | |
2900 | { | |
2901 | if (rq->post_schedule) { | |
2902 | unsigned long flags; | |
2903 | ||
05fa785c | 2904 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2905 | if (rq->curr->sched_class->post_schedule) |
2906 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2907 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2908 | |
2909 | rq->post_schedule = 0; | |
2910 | } | |
2911 | } | |
2912 | ||
2913 | #else | |
da19ab51 | 2914 | |
3f029d3c GH |
2915 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2916 | { | |
2917 | } | |
2918 | ||
2919 | static inline void post_schedule(struct rq *rq) | |
2920 | { | |
1da177e4 LT |
2921 | } |
2922 | ||
3f029d3c GH |
2923 | #endif |
2924 | ||
1da177e4 LT |
2925 | /** |
2926 | * schedule_tail - first thing a freshly forked thread must call. | |
2927 | * @prev: the thread we just switched away from. | |
2928 | */ | |
36c8b586 | 2929 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2930 | __releases(rq->lock) |
2931 | { | |
70b97a7f IM |
2932 | struct rq *rq = this_rq(); |
2933 | ||
4866cde0 | 2934 | finish_task_switch(rq, prev); |
da19ab51 | 2935 | |
3f029d3c GH |
2936 | /* |
2937 | * FIXME: do we need to worry about rq being invalidated by the | |
2938 | * task_switch? | |
2939 | */ | |
2940 | post_schedule(rq); | |
70b97a7f | 2941 | |
4866cde0 NP |
2942 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2943 | /* In this case, finish_task_switch does not reenable preemption */ | |
2944 | preempt_enable(); | |
2945 | #endif | |
1da177e4 | 2946 | if (current->set_child_tid) |
b488893a | 2947 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2948 | } |
2949 | ||
2950 | /* | |
2951 | * context_switch - switch to the new MM and the new | |
2952 | * thread's register state. | |
2953 | */ | |
dd41f596 | 2954 | static inline void |
70b97a7f | 2955 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2956 | struct task_struct *next) |
1da177e4 | 2957 | { |
dd41f596 | 2958 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2959 | |
e107be36 | 2960 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2961 | |
dd41f596 IM |
2962 | mm = next->mm; |
2963 | oldmm = prev->active_mm; | |
9226d125 ZA |
2964 | /* |
2965 | * For paravirt, this is coupled with an exit in switch_to to | |
2966 | * combine the page table reload and the switch backend into | |
2967 | * one hypercall. | |
2968 | */ | |
224101ed | 2969 | arch_start_context_switch(prev); |
9226d125 | 2970 | |
31915ab4 | 2971 | if (!mm) { |
1da177e4 LT |
2972 | next->active_mm = oldmm; |
2973 | atomic_inc(&oldmm->mm_count); | |
2974 | enter_lazy_tlb(oldmm, next); | |
2975 | } else | |
2976 | switch_mm(oldmm, mm, next); | |
2977 | ||
31915ab4 | 2978 | if (!prev->mm) { |
1da177e4 | 2979 | prev->active_mm = NULL; |
1da177e4 LT |
2980 | rq->prev_mm = oldmm; |
2981 | } | |
3a5f5e48 IM |
2982 | /* |
2983 | * Since the runqueue lock will be released by the next | |
2984 | * task (which is an invalid locking op but in the case | |
2985 | * of the scheduler it's an obvious special-case), so we | |
2986 | * do an early lockdep release here: | |
2987 | */ | |
2988 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2989 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2990 | #endif |
1da177e4 LT |
2991 | |
2992 | /* Here we just switch the register state and the stack. */ | |
2993 | switch_to(prev, next, prev); | |
2994 | ||
dd41f596 IM |
2995 | barrier(); |
2996 | /* | |
2997 | * this_rq must be evaluated again because prev may have moved | |
2998 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2999 | * frame will be invalid. | |
3000 | */ | |
3001 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3002 | } |
3003 | ||
3004 | /* | |
3005 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3006 | * | |
3007 | * externally visible scheduler statistics: current number of runnable | |
3008 | * threads, current number of uninterruptible-sleeping threads, total | |
3009 | * number of context switches performed since bootup. | |
3010 | */ | |
3011 | unsigned long nr_running(void) | |
3012 | { | |
3013 | unsigned long i, sum = 0; | |
3014 | ||
3015 | for_each_online_cpu(i) | |
3016 | sum += cpu_rq(i)->nr_running; | |
3017 | ||
3018 | return sum; | |
f711f609 | 3019 | } |
1da177e4 LT |
3020 | |
3021 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3022 | { |
1da177e4 | 3023 | unsigned long i, sum = 0; |
f711f609 | 3024 | |
0a945022 | 3025 | for_each_possible_cpu(i) |
1da177e4 | 3026 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3027 | |
3028 | /* | |
1da177e4 LT |
3029 | * Since we read the counters lockless, it might be slightly |
3030 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3031 | */ |
1da177e4 LT |
3032 | if (unlikely((long)sum < 0)) |
3033 | sum = 0; | |
f711f609 | 3034 | |
1da177e4 | 3035 | return sum; |
f711f609 | 3036 | } |
f711f609 | 3037 | |
1da177e4 | 3038 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3039 | { |
cc94abfc SR |
3040 | int i; |
3041 | unsigned long long sum = 0; | |
46cb4b7c | 3042 | |
0a945022 | 3043 | for_each_possible_cpu(i) |
1da177e4 | 3044 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3045 | |
1da177e4 LT |
3046 | return sum; |
3047 | } | |
483b4ee6 | 3048 | |
1da177e4 LT |
3049 | unsigned long nr_iowait(void) |
3050 | { | |
3051 | unsigned long i, sum = 0; | |
483b4ee6 | 3052 | |
0a945022 | 3053 | for_each_possible_cpu(i) |
1da177e4 | 3054 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3055 | |
1da177e4 LT |
3056 | return sum; |
3057 | } | |
483b4ee6 | 3058 | |
8c215bd3 | 3059 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3060 | { |
8c215bd3 | 3061 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3062 | return atomic_read(&this->nr_iowait); |
3063 | } | |
46cb4b7c | 3064 | |
69d25870 AV |
3065 | unsigned long this_cpu_load(void) |
3066 | { | |
3067 | struct rq *this = this_rq(); | |
3068 | return this->cpu_load[0]; | |
3069 | } | |
e790fb0b | 3070 | |
46cb4b7c | 3071 | |
dce48a84 TG |
3072 | /* Variables and functions for calc_load */ |
3073 | static atomic_long_t calc_load_tasks; | |
3074 | static unsigned long calc_load_update; | |
3075 | unsigned long avenrun[3]; | |
3076 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3077 | |
74f5187a PZ |
3078 | static long calc_load_fold_active(struct rq *this_rq) |
3079 | { | |
3080 | long nr_active, delta = 0; | |
3081 | ||
3082 | nr_active = this_rq->nr_running; | |
3083 | nr_active += (long) this_rq->nr_uninterruptible; | |
3084 | ||
3085 | if (nr_active != this_rq->calc_load_active) { | |
3086 | delta = nr_active - this_rq->calc_load_active; | |
3087 | this_rq->calc_load_active = nr_active; | |
3088 | } | |
3089 | ||
3090 | return delta; | |
3091 | } | |
3092 | ||
0f004f5a PZ |
3093 | static unsigned long |
3094 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3095 | { | |
3096 | load *= exp; | |
3097 | load += active * (FIXED_1 - exp); | |
3098 | load += 1UL << (FSHIFT - 1); | |
3099 | return load >> FSHIFT; | |
3100 | } | |
3101 | ||
74f5187a PZ |
3102 | #ifdef CONFIG_NO_HZ |
3103 | /* | |
3104 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3105 | * | |
3106 | * When making the ILB scale, we should try to pull this in as well. | |
3107 | */ | |
3108 | static atomic_long_t calc_load_tasks_idle; | |
3109 | ||
3110 | static void calc_load_account_idle(struct rq *this_rq) | |
3111 | { | |
3112 | long delta; | |
3113 | ||
3114 | delta = calc_load_fold_active(this_rq); | |
3115 | if (delta) | |
3116 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3117 | } | |
3118 | ||
3119 | static long calc_load_fold_idle(void) | |
3120 | { | |
3121 | long delta = 0; | |
3122 | ||
3123 | /* | |
3124 | * Its got a race, we don't care... | |
3125 | */ | |
3126 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3127 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3128 | ||
3129 | return delta; | |
3130 | } | |
0f004f5a PZ |
3131 | |
3132 | /** | |
3133 | * fixed_power_int - compute: x^n, in O(log n) time | |
3134 | * | |
3135 | * @x: base of the power | |
3136 | * @frac_bits: fractional bits of @x | |
3137 | * @n: power to raise @x to. | |
3138 | * | |
3139 | * By exploiting the relation between the definition of the natural power | |
3140 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3141 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3142 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3143 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3144 | * of course trivially computable in O(log_2 n), the length of our binary | |
3145 | * vector. | |
3146 | */ | |
3147 | static unsigned long | |
3148 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3149 | { | |
3150 | unsigned long result = 1UL << frac_bits; | |
3151 | ||
3152 | if (n) for (;;) { | |
3153 | if (n & 1) { | |
3154 | result *= x; | |
3155 | result += 1UL << (frac_bits - 1); | |
3156 | result >>= frac_bits; | |
3157 | } | |
3158 | n >>= 1; | |
3159 | if (!n) | |
3160 | break; | |
3161 | x *= x; | |
3162 | x += 1UL << (frac_bits - 1); | |
3163 | x >>= frac_bits; | |
3164 | } | |
3165 | ||
3166 | return result; | |
3167 | } | |
3168 | ||
3169 | /* | |
3170 | * a1 = a0 * e + a * (1 - e) | |
3171 | * | |
3172 | * a2 = a1 * e + a * (1 - e) | |
3173 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3174 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3175 | * | |
3176 | * a3 = a2 * e + a * (1 - e) | |
3177 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3178 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3179 | * | |
3180 | * ... | |
3181 | * | |
3182 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3183 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3184 | * = a0 * e^n + a * (1 - e^n) | |
3185 | * | |
3186 | * [1] application of the geometric series: | |
3187 | * | |
3188 | * n 1 - x^(n+1) | |
3189 | * S_n := \Sum x^i = ------------- | |
3190 | * i=0 1 - x | |
3191 | */ | |
3192 | static unsigned long | |
3193 | calc_load_n(unsigned long load, unsigned long exp, | |
3194 | unsigned long active, unsigned int n) | |
3195 | { | |
3196 | ||
3197 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3198 | } | |
3199 | ||
3200 | /* | |
3201 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3202 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3203 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3204 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3205 | * | |
3206 | * Once we've updated the global active value, we need to apply the exponential | |
3207 | * weights adjusted to the number of cycles missed. | |
3208 | */ | |
3209 | static void calc_global_nohz(unsigned long ticks) | |
3210 | { | |
3211 | long delta, active, n; | |
3212 | ||
3213 | if (time_before(jiffies, calc_load_update)) | |
3214 | return; | |
3215 | ||
3216 | /* | |
3217 | * If we crossed a calc_load_update boundary, make sure to fold | |
3218 | * any pending idle changes, the respective CPUs might have | |
3219 | * missed the tick driven calc_load_account_active() update | |
3220 | * due to NO_HZ. | |
3221 | */ | |
3222 | delta = calc_load_fold_idle(); | |
3223 | if (delta) | |
3224 | atomic_long_add(delta, &calc_load_tasks); | |
3225 | ||
3226 | /* | |
3227 | * If we were idle for multiple load cycles, apply them. | |
3228 | */ | |
3229 | if (ticks >= LOAD_FREQ) { | |
3230 | n = ticks / LOAD_FREQ; | |
3231 | ||
3232 | active = atomic_long_read(&calc_load_tasks); | |
3233 | active = active > 0 ? active * FIXED_1 : 0; | |
3234 | ||
3235 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3236 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3237 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3238 | ||
3239 | calc_load_update += n * LOAD_FREQ; | |
3240 | } | |
3241 | ||
3242 | /* | |
3243 | * Its possible the remainder of the above division also crosses | |
3244 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3245 | * which comes after this will take care of that. | |
3246 | * | |
3247 | * Consider us being 11 ticks before a cycle completion, and us | |
3248 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3249 | * age us 4 cycles, and the test in calc_global_load() will | |
3250 | * pick up the final one. | |
3251 | */ | |
3252 | } | |
74f5187a PZ |
3253 | #else |
3254 | static void calc_load_account_idle(struct rq *this_rq) | |
3255 | { | |
3256 | } | |
3257 | ||
3258 | static inline long calc_load_fold_idle(void) | |
3259 | { | |
3260 | return 0; | |
3261 | } | |
0f004f5a PZ |
3262 | |
3263 | static void calc_global_nohz(unsigned long ticks) | |
3264 | { | |
3265 | } | |
74f5187a PZ |
3266 | #endif |
3267 | ||
2d02494f TG |
3268 | /** |
3269 | * get_avenrun - get the load average array | |
3270 | * @loads: pointer to dest load array | |
3271 | * @offset: offset to add | |
3272 | * @shift: shift count to shift the result left | |
3273 | * | |
3274 | * These values are estimates at best, so no need for locking. | |
3275 | */ | |
3276 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3277 | { | |
3278 | loads[0] = (avenrun[0] + offset) << shift; | |
3279 | loads[1] = (avenrun[1] + offset) << shift; | |
3280 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3281 | } |
46cb4b7c | 3282 | |
46cb4b7c | 3283 | /* |
dce48a84 TG |
3284 | * calc_load - update the avenrun load estimates 10 ticks after the |
3285 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3286 | */ |
0f004f5a | 3287 | void calc_global_load(unsigned long ticks) |
7835b98b | 3288 | { |
dce48a84 | 3289 | long active; |
1da177e4 | 3290 | |
0f004f5a PZ |
3291 | calc_global_nohz(ticks); |
3292 | ||
3293 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3294 | return; |
1da177e4 | 3295 | |
dce48a84 TG |
3296 | active = atomic_long_read(&calc_load_tasks); |
3297 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3298 | |
dce48a84 TG |
3299 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3300 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3301 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3302 | |
dce48a84 TG |
3303 | calc_load_update += LOAD_FREQ; |
3304 | } | |
1da177e4 | 3305 | |
dce48a84 | 3306 | /* |
74f5187a PZ |
3307 | * Called from update_cpu_load() to periodically update this CPU's |
3308 | * active count. | |
dce48a84 TG |
3309 | */ |
3310 | static void calc_load_account_active(struct rq *this_rq) | |
3311 | { | |
74f5187a | 3312 | long delta; |
08c183f3 | 3313 | |
74f5187a PZ |
3314 | if (time_before(jiffies, this_rq->calc_load_update)) |
3315 | return; | |
783609c6 | 3316 | |
74f5187a PZ |
3317 | delta = calc_load_fold_active(this_rq); |
3318 | delta += calc_load_fold_idle(); | |
3319 | if (delta) | |
dce48a84 | 3320 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3321 | |
3322 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3323 | } |
3324 | ||
fdf3e95d VP |
3325 | /* |
3326 | * The exact cpuload at various idx values, calculated at every tick would be | |
3327 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3328 | * | |
3329 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3330 | * on nth tick when cpu may be busy, then we have: | |
3331 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3332 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3333 | * | |
3334 | * decay_load_missed() below does efficient calculation of | |
3335 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3336 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3337 | * | |
3338 | * The calculation is approximated on a 128 point scale. | |
3339 | * degrade_zero_ticks is the number of ticks after which load at any | |
3340 | * particular idx is approximated to be zero. | |
3341 | * degrade_factor is a precomputed table, a row for each load idx. | |
3342 | * Each column corresponds to degradation factor for a power of two ticks, | |
3343 | * based on 128 point scale. | |
3344 | * Example: | |
3345 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3346 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3347 | * | |
3348 | * With this power of 2 load factors, we can degrade the load n times | |
3349 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3350 | * n mult/shifts needed by the exact degradation. | |
3351 | */ | |
3352 | #define DEGRADE_SHIFT 7 | |
3353 | static const unsigned char | |
3354 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3355 | static const unsigned char | |
3356 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3357 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3358 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3359 | {96, 72, 40, 12, 1, 0, 0}, | |
3360 | {112, 98, 75, 43, 15, 1, 0}, | |
3361 | {120, 112, 98, 76, 45, 16, 2} }; | |
3362 | ||
3363 | /* | |
3364 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3365 | * would be when CPU is idle and so we just decay the old load without | |
3366 | * adding any new load. | |
3367 | */ | |
3368 | static unsigned long | |
3369 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3370 | { | |
3371 | int j = 0; | |
3372 | ||
3373 | if (!missed_updates) | |
3374 | return load; | |
3375 | ||
3376 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3377 | return 0; | |
3378 | ||
3379 | if (idx == 1) | |
3380 | return load >> missed_updates; | |
3381 | ||
3382 | while (missed_updates) { | |
3383 | if (missed_updates % 2) | |
3384 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3385 | ||
3386 | missed_updates >>= 1; | |
3387 | j++; | |
3388 | } | |
3389 | return load; | |
3390 | } | |
3391 | ||
46cb4b7c | 3392 | /* |
dd41f596 | 3393 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3394 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3395 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3396 | */ |
dd41f596 | 3397 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3398 | { |
495eca49 | 3399 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3400 | unsigned long curr_jiffies = jiffies; |
3401 | unsigned long pending_updates; | |
dd41f596 | 3402 | int i, scale; |
46cb4b7c | 3403 | |
dd41f596 | 3404 | this_rq->nr_load_updates++; |
46cb4b7c | 3405 | |
fdf3e95d VP |
3406 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3407 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3408 | return; | |
3409 | ||
3410 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3411 | this_rq->last_load_update_tick = curr_jiffies; | |
3412 | ||
dd41f596 | 3413 | /* Update our load: */ |
fdf3e95d VP |
3414 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3415 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3416 | unsigned long old_load, new_load; |
7d1e6a9b | 3417 | |
dd41f596 | 3418 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3419 | |
dd41f596 | 3420 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3421 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3422 | new_load = this_load; |
a25707f3 IM |
3423 | /* |
3424 | * Round up the averaging division if load is increasing. This | |
3425 | * prevents us from getting stuck on 9 if the load is 10, for | |
3426 | * example. | |
3427 | */ | |
3428 | if (new_load > old_load) | |
fdf3e95d VP |
3429 | new_load += scale - 1; |
3430 | ||
3431 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3432 | } |
da2b71ed SS |
3433 | |
3434 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3435 | } |
3436 | ||
3437 | static void update_cpu_load_active(struct rq *this_rq) | |
3438 | { | |
3439 | update_cpu_load(this_rq); | |
46cb4b7c | 3440 | |
74f5187a | 3441 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3442 | } |
3443 | ||
dd41f596 | 3444 | #ifdef CONFIG_SMP |
8a0be9ef | 3445 | |
46cb4b7c | 3446 | /* |
38022906 PZ |
3447 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3448 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3449 | */ |
38022906 | 3450 | void sched_exec(void) |
46cb4b7c | 3451 | { |
38022906 | 3452 | struct task_struct *p = current; |
1da177e4 | 3453 | unsigned long flags; |
70b97a7f | 3454 | struct rq *rq; |
0017d735 | 3455 | int dest_cpu; |
46cb4b7c | 3456 | |
1da177e4 | 3457 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3458 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3459 | if (dest_cpu == smp_processor_id()) | |
3460 | goto unlock; | |
38022906 | 3461 | |
46cb4b7c | 3462 | /* |
38022906 | 3463 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3464 | */ |
30da688e | 3465 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3466 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3467 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3468 | |
1da177e4 | 3469 | task_rq_unlock(rq, &flags); |
969c7921 | 3470 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3471 | return; |
3472 | } | |
0017d735 | 3473 | unlock: |
1da177e4 | 3474 | task_rq_unlock(rq, &flags); |
1da177e4 | 3475 | } |
dd41f596 | 3476 | |
1da177e4 LT |
3477 | #endif |
3478 | ||
1da177e4 LT |
3479 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3480 | ||
3481 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3482 | ||
3483 | /* | |
c5f8d995 | 3484 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3485 | * @p in case that task is currently running. |
c5f8d995 HS |
3486 | * |
3487 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3488 | */ |
c5f8d995 HS |
3489 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3490 | { | |
3491 | u64 ns = 0; | |
3492 | ||
3493 | if (task_current(rq, p)) { | |
3494 | update_rq_clock(rq); | |
305e6835 | 3495 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3496 | if ((s64)ns < 0) |
3497 | ns = 0; | |
3498 | } | |
3499 | ||
3500 | return ns; | |
3501 | } | |
3502 | ||
bb34d92f | 3503 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3504 | { |
1da177e4 | 3505 | unsigned long flags; |
41b86e9c | 3506 | struct rq *rq; |
bb34d92f | 3507 | u64 ns = 0; |
48f24c4d | 3508 | |
41b86e9c | 3509 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3510 | ns = do_task_delta_exec(p, rq); |
3511 | task_rq_unlock(rq, &flags); | |
1508487e | 3512 | |
c5f8d995 HS |
3513 | return ns; |
3514 | } | |
f06febc9 | 3515 | |
c5f8d995 HS |
3516 | /* |
3517 | * Return accounted runtime for the task. | |
3518 | * In case the task is currently running, return the runtime plus current's | |
3519 | * pending runtime that have not been accounted yet. | |
3520 | */ | |
3521 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3522 | { | |
3523 | unsigned long flags; | |
3524 | struct rq *rq; | |
3525 | u64 ns = 0; | |
3526 | ||
3527 | rq = task_rq_lock(p, &flags); | |
3528 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3529 | task_rq_unlock(rq, &flags); | |
3530 | ||
3531 | return ns; | |
3532 | } | |
48f24c4d | 3533 | |
c5f8d995 HS |
3534 | /* |
3535 | * Return sum_exec_runtime for the thread group. | |
3536 | * In case the task is currently running, return the sum plus current's | |
3537 | * pending runtime that have not been accounted yet. | |
3538 | * | |
3539 | * Note that the thread group might have other running tasks as well, | |
3540 | * so the return value not includes other pending runtime that other | |
3541 | * running tasks might have. | |
3542 | */ | |
3543 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3544 | { | |
3545 | struct task_cputime totals; | |
3546 | unsigned long flags; | |
3547 | struct rq *rq; | |
3548 | u64 ns; | |
3549 | ||
3550 | rq = task_rq_lock(p, &flags); | |
3551 | thread_group_cputime(p, &totals); | |
3552 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3553 | task_rq_unlock(rq, &flags); |
48f24c4d | 3554 | |
1da177e4 LT |
3555 | return ns; |
3556 | } | |
3557 | ||
1da177e4 LT |
3558 | /* |
3559 | * Account user cpu time to a process. | |
3560 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3561 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3562 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3563 | */ |
457533a7 MS |
3564 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3565 | cputime_t cputime_scaled) | |
1da177e4 LT |
3566 | { |
3567 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3568 | cputime64_t tmp; | |
3569 | ||
457533a7 | 3570 | /* Add user time to process. */ |
1da177e4 | 3571 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3572 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3573 | account_group_user_time(p, cputime); |
1da177e4 LT |
3574 | |
3575 | /* Add user time to cpustat. */ | |
3576 | tmp = cputime_to_cputime64(cputime); | |
3577 | if (TASK_NICE(p) > 0) | |
3578 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3579 | else | |
3580 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3581 | |
3582 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3583 | /* Account for user time used */ |
3584 | acct_update_integrals(p); | |
1da177e4 LT |
3585 | } |
3586 | ||
94886b84 LV |
3587 | /* |
3588 | * Account guest cpu time to a process. | |
3589 | * @p: the process that the cpu time gets accounted to | |
3590 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3591 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3592 | */ |
457533a7 MS |
3593 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3594 | cputime_t cputime_scaled) | |
94886b84 LV |
3595 | { |
3596 | cputime64_t tmp; | |
3597 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3598 | ||
3599 | tmp = cputime_to_cputime64(cputime); | |
3600 | ||
457533a7 | 3601 | /* Add guest time to process. */ |
94886b84 | 3602 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3603 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3604 | account_group_user_time(p, cputime); |
94886b84 LV |
3605 | p->gtime = cputime_add(p->gtime, cputime); |
3606 | ||
457533a7 | 3607 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3608 | if (TASK_NICE(p) > 0) { |
3609 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3610 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3611 | } else { | |
3612 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3613 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3614 | } | |
94886b84 LV |
3615 | } |
3616 | ||
70a89a66 VP |
3617 | /* |
3618 | * Account system cpu time to a process and desired cpustat field | |
3619 | * @p: the process that the cpu time gets accounted to | |
3620 | * @cputime: the cpu time spent in kernel space since the last update | |
3621 | * @cputime_scaled: cputime scaled by cpu frequency | |
3622 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3623 | */ | |
3624 | static inline | |
3625 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3626 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3627 | { | |
3628 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3629 | ||
3630 | /* Add system time to process. */ | |
3631 | p->stime = cputime_add(p->stime, cputime); | |
3632 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3633 | account_group_system_time(p, cputime); | |
3634 | ||
3635 | /* Add system time to cpustat. */ | |
3636 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3637 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3638 | ||
3639 | /* Account for system time used */ | |
3640 | acct_update_integrals(p); | |
3641 | } | |
3642 | ||
1da177e4 LT |
3643 | /* |
3644 | * Account system cpu time to a process. | |
3645 | * @p: the process that the cpu time gets accounted to | |
3646 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3647 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3648 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3649 | */ |
3650 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3651 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3652 | { |
3653 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3654 | cputime64_t *target_cputime64; |
1da177e4 | 3655 | |
983ed7a6 | 3656 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3657 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3658 | return; |
3659 | } | |
94886b84 | 3660 | |
1da177e4 | 3661 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3662 | target_cputime64 = &cpustat->irq; |
75e1056f | 3663 | else if (in_serving_softirq()) |
70a89a66 | 3664 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3665 | else |
70a89a66 | 3666 | target_cputime64 = &cpustat->system; |
ef12fefa | 3667 | |
70a89a66 | 3668 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3669 | } |
3670 | ||
c66f08be | 3671 | /* |
1da177e4 | 3672 | * Account for involuntary wait time. |
544b4a1f | 3673 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3674 | */ |
79741dd3 | 3675 | void account_steal_time(cputime_t cputime) |
c66f08be | 3676 | { |
79741dd3 MS |
3677 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3678 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3679 | ||
3680 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3681 | } |
3682 | ||
1da177e4 | 3683 | /* |
79741dd3 MS |
3684 | * Account for idle time. |
3685 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3686 | */ |
79741dd3 | 3687 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3688 | { |
3689 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3690 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3691 | struct rq *rq = this_rq(); |
1da177e4 | 3692 | |
79741dd3 MS |
3693 | if (atomic_read(&rq->nr_iowait) > 0) |
3694 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3695 | else | |
3696 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3697 | } |
3698 | ||
79741dd3 MS |
3699 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3700 | ||
abb74cef VP |
3701 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3702 | /* | |
3703 | * Account a tick to a process and cpustat | |
3704 | * @p: the process that the cpu time gets accounted to | |
3705 | * @user_tick: is the tick from userspace | |
3706 | * @rq: the pointer to rq | |
3707 | * | |
3708 | * Tick demultiplexing follows the order | |
3709 | * - pending hardirq update | |
3710 | * - pending softirq update | |
3711 | * - user_time | |
3712 | * - idle_time | |
3713 | * - system time | |
3714 | * - check for guest_time | |
3715 | * - else account as system_time | |
3716 | * | |
3717 | * Check for hardirq is done both for system and user time as there is | |
3718 | * no timer going off while we are on hardirq and hence we may never get an | |
3719 | * opportunity to update it solely in system time. | |
3720 | * p->stime and friends are only updated on system time and not on irq | |
3721 | * softirq as those do not count in task exec_runtime any more. | |
3722 | */ | |
3723 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3724 | struct rq *rq) | |
3725 | { | |
3726 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3727 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3728 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3729 | ||
3730 | if (irqtime_account_hi_update()) { | |
3731 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3732 | } else if (irqtime_account_si_update()) { | |
3733 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3734 | } else if (this_cpu_ksoftirqd() == p) { |
3735 | /* | |
3736 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3737 | * So, we have to handle it separately here. | |
3738 | * Also, p->stime needs to be updated for ksoftirqd. | |
3739 | */ | |
3740 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3741 | &cpustat->softirq); | |
abb74cef VP |
3742 | } else if (user_tick) { |
3743 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3744 | } else if (p == rq->idle) { | |
3745 | account_idle_time(cputime_one_jiffy); | |
3746 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3747 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3748 | } else { | |
3749 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3750 | &cpustat->system); | |
3751 | } | |
3752 | } | |
3753 | ||
3754 | static void irqtime_account_idle_ticks(int ticks) | |
3755 | { | |
3756 | int i; | |
3757 | struct rq *rq = this_rq(); | |
3758 | ||
3759 | for (i = 0; i < ticks; i++) | |
3760 | irqtime_account_process_tick(current, 0, rq); | |
3761 | } | |
544b4a1f | 3762 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3763 | static void irqtime_account_idle_ticks(int ticks) {} |
3764 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3765 | struct rq *rq) {} | |
544b4a1f | 3766 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3767 | |
3768 | /* | |
3769 | * Account a single tick of cpu time. | |
3770 | * @p: the process that the cpu time gets accounted to | |
3771 | * @user_tick: indicates if the tick is a user or a system tick | |
3772 | */ | |
3773 | void account_process_tick(struct task_struct *p, int user_tick) | |
3774 | { | |
a42548a1 | 3775 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3776 | struct rq *rq = this_rq(); |
3777 | ||
abb74cef VP |
3778 | if (sched_clock_irqtime) { |
3779 | irqtime_account_process_tick(p, user_tick, rq); | |
3780 | return; | |
3781 | } | |
3782 | ||
79741dd3 | 3783 | if (user_tick) |
a42548a1 | 3784 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3785 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3786 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3787 | one_jiffy_scaled); |
3788 | else | |
a42548a1 | 3789 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3790 | } |
3791 | ||
3792 | /* | |
3793 | * Account multiple ticks of steal time. | |
3794 | * @p: the process from which the cpu time has been stolen | |
3795 | * @ticks: number of stolen ticks | |
3796 | */ | |
3797 | void account_steal_ticks(unsigned long ticks) | |
3798 | { | |
3799 | account_steal_time(jiffies_to_cputime(ticks)); | |
3800 | } | |
3801 | ||
3802 | /* | |
3803 | * Account multiple ticks of idle time. | |
3804 | * @ticks: number of stolen ticks | |
3805 | */ | |
3806 | void account_idle_ticks(unsigned long ticks) | |
3807 | { | |
abb74cef VP |
3808 | |
3809 | if (sched_clock_irqtime) { | |
3810 | irqtime_account_idle_ticks(ticks); | |
3811 | return; | |
3812 | } | |
3813 | ||
79741dd3 | 3814 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3815 | } |
3816 | ||
79741dd3 MS |
3817 | #endif |
3818 | ||
49048622 BS |
3819 | /* |
3820 | * Use precise platform statistics if available: | |
3821 | */ | |
3822 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3823 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3824 | { |
d99ca3b9 HS |
3825 | *ut = p->utime; |
3826 | *st = p->stime; | |
49048622 BS |
3827 | } |
3828 | ||
0cf55e1e | 3829 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3830 | { |
0cf55e1e HS |
3831 | struct task_cputime cputime; |
3832 | ||
3833 | thread_group_cputime(p, &cputime); | |
3834 | ||
3835 | *ut = cputime.utime; | |
3836 | *st = cputime.stime; | |
49048622 BS |
3837 | } |
3838 | #else | |
761b1d26 HS |
3839 | |
3840 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3841 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3842 | #endif |
3843 | ||
d180c5bc | 3844 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3845 | { |
d99ca3b9 | 3846 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3847 | |
3848 | /* | |
3849 | * Use CFS's precise accounting: | |
3850 | */ | |
d180c5bc | 3851 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3852 | |
3853 | if (total) { | |
e75e863d | 3854 | u64 temp = rtime; |
d180c5bc | 3855 | |
e75e863d | 3856 | temp *= utime; |
49048622 | 3857 | do_div(temp, total); |
d180c5bc HS |
3858 | utime = (cputime_t)temp; |
3859 | } else | |
3860 | utime = rtime; | |
49048622 | 3861 | |
d180c5bc HS |
3862 | /* |
3863 | * Compare with previous values, to keep monotonicity: | |
3864 | */ | |
761b1d26 | 3865 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3866 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3867 | |
d99ca3b9 HS |
3868 | *ut = p->prev_utime; |
3869 | *st = p->prev_stime; | |
49048622 BS |
3870 | } |
3871 | ||
0cf55e1e HS |
3872 | /* |
3873 | * Must be called with siglock held. | |
3874 | */ | |
3875 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3876 | { |
0cf55e1e HS |
3877 | struct signal_struct *sig = p->signal; |
3878 | struct task_cputime cputime; | |
3879 | cputime_t rtime, utime, total; | |
49048622 | 3880 | |
0cf55e1e | 3881 | thread_group_cputime(p, &cputime); |
49048622 | 3882 | |
0cf55e1e HS |
3883 | total = cputime_add(cputime.utime, cputime.stime); |
3884 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3885 | |
0cf55e1e | 3886 | if (total) { |
e75e863d | 3887 | u64 temp = rtime; |
49048622 | 3888 | |
e75e863d | 3889 | temp *= cputime.utime; |
0cf55e1e HS |
3890 | do_div(temp, total); |
3891 | utime = (cputime_t)temp; | |
3892 | } else | |
3893 | utime = rtime; | |
3894 | ||
3895 | sig->prev_utime = max(sig->prev_utime, utime); | |
3896 | sig->prev_stime = max(sig->prev_stime, | |
3897 | cputime_sub(rtime, sig->prev_utime)); | |
3898 | ||
3899 | *ut = sig->prev_utime; | |
3900 | *st = sig->prev_stime; | |
49048622 | 3901 | } |
49048622 | 3902 | #endif |
49048622 | 3903 | |
7835b98b CL |
3904 | /* |
3905 | * This function gets called by the timer code, with HZ frequency. | |
3906 | * We call it with interrupts disabled. | |
3907 | * | |
3908 | * It also gets called by the fork code, when changing the parent's | |
3909 | * timeslices. | |
3910 | */ | |
3911 | void scheduler_tick(void) | |
3912 | { | |
7835b98b CL |
3913 | int cpu = smp_processor_id(); |
3914 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3915 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3916 | |
3917 | sched_clock_tick(); | |
dd41f596 | 3918 | |
05fa785c | 3919 | raw_spin_lock(&rq->lock); |
3e51f33f | 3920 | update_rq_clock(rq); |
fdf3e95d | 3921 | update_cpu_load_active(rq); |
fa85ae24 | 3922 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3923 | raw_spin_unlock(&rq->lock); |
7835b98b | 3924 | |
e9d2b064 | 3925 | perf_event_task_tick(); |
e220d2dc | 3926 | |
e418e1c2 | 3927 | #ifdef CONFIG_SMP |
dd41f596 IM |
3928 | rq->idle_at_tick = idle_cpu(cpu); |
3929 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3930 | #endif |
1da177e4 LT |
3931 | } |
3932 | ||
132380a0 | 3933 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3934 | { |
3935 | if (in_lock_functions(addr)) { | |
3936 | addr = CALLER_ADDR2; | |
3937 | if (in_lock_functions(addr)) | |
3938 | addr = CALLER_ADDR3; | |
3939 | } | |
3940 | return addr; | |
3941 | } | |
1da177e4 | 3942 | |
7e49fcce SR |
3943 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3944 | defined(CONFIG_PREEMPT_TRACER)) | |
3945 | ||
43627582 | 3946 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3947 | { |
6cd8a4bb | 3948 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3949 | /* |
3950 | * Underflow? | |
3951 | */ | |
9a11b49a IM |
3952 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3953 | return; | |
6cd8a4bb | 3954 | #endif |
1da177e4 | 3955 | preempt_count() += val; |
6cd8a4bb | 3956 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3957 | /* |
3958 | * Spinlock count overflowing soon? | |
3959 | */ | |
33859f7f MOS |
3960 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3961 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3962 | #endif |
3963 | if (preempt_count() == val) | |
3964 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3965 | } |
3966 | EXPORT_SYMBOL(add_preempt_count); | |
3967 | ||
43627582 | 3968 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3969 | { |
6cd8a4bb | 3970 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3971 | /* |
3972 | * Underflow? | |
3973 | */ | |
01e3eb82 | 3974 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3975 | return; |
1da177e4 LT |
3976 | /* |
3977 | * Is the spinlock portion underflowing? | |
3978 | */ | |
9a11b49a IM |
3979 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3980 | !(preempt_count() & PREEMPT_MASK))) | |
3981 | return; | |
6cd8a4bb | 3982 | #endif |
9a11b49a | 3983 | |
6cd8a4bb SR |
3984 | if (preempt_count() == val) |
3985 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3986 | preempt_count() -= val; |
3987 | } | |
3988 | EXPORT_SYMBOL(sub_preempt_count); | |
3989 | ||
3990 | #endif | |
3991 | ||
3992 | /* | |
dd41f596 | 3993 | * Print scheduling while atomic bug: |
1da177e4 | 3994 | */ |
dd41f596 | 3995 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3996 | { |
838225b4 SS |
3997 | struct pt_regs *regs = get_irq_regs(); |
3998 | ||
3df0fc5b PZ |
3999 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4000 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4001 | |
dd41f596 | 4002 | debug_show_held_locks(prev); |
e21f5b15 | 4003 | print_modules(); |
dd41f596 IM |
4004 | if (irqs_disabled()) |
4005 | print_irqtrace_events(prev); | |
838225b4 SS |
4006 | |
4007 | if (regs) | |
4008 | show_regs(regs); | |
4009 | else | |
4010 | dump_stack(); | |
dd41f596 | 4011 | } |
1da177e4 | 4012 | |
dd41f596 IM |
4013 | /* |
4014 | * Various schedule()-time debugging checks and statistics: | |
4015 | */ | |
4016 | static inline void schedule_debug(struct task_struct *prev) | |
4017 | { | |
1da177e4 | 4018 | /* |
41a2d6cf | 4019 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4020 | * schedule() atomically, we ignore that path for now. |
4021 | * Otherwise, whine if we are scheduling when we should not be. | |
4022 | */ | |
3f33a7ce | 4023 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4024 | __schedule_bug(prev); |
4025 | ||
1da177e4 LT |
4026 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4027 | ||
2d72376b | 4028 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4029 | #ifdef CONFIG_SCHEDSTATS |
4030 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 4031 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 4032 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
4033 | } |
4034 | #endif | |
dd41f596 IM |
4035 | } |
4036 | ||
6cecd084 | 4037 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4038 | { |
a64692a3 MG |
4039 | if (prev->se.on_rq) |
4040 | update_rq_clock(rq); | |
6cecd084 | 4041 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4042 | } |
4043 | ||
dd41f596 IM |
4044 | /* |
4045 | * Pick up the highest-prio task: | |
4046 | */ | |
4047 | static inline struct task_struct * | |
b67802ea | 4048 | pick_next_task(struct rq *rq) |
dd41f596 | 4049 | { |
5522d5d5 | 4050 | const struct sched_class *class; |
dd41f596 | 4051 | struct task_struct *p; |
1da177e4 LT |
4052 | |
4053 | /* | |
dd41f596 IM |
4054 | * Optimization: we know that if all tasks are in |
4055 | * the fair class we can call that function directly: | |
1da177e4 | 4056 | */ |
dd41f596 | 4057 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4058 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4059 | if (likely(p)) |
4060 | return p; | |
1da177e4 LT |
4061 | } |
4062 | ||
34f971f6 | 4063 | for_each_class(class) { |
fb8d4724 | 4064 | p = class->pick_next_task(rq); |
dd41f596 IM |
4065 | if (p) |
4066 | return p; | |
dd41f596 | 4067 | } |
34f971f6 PZ |
4068 | |
4069 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4070 | } |
1da177e4 | 4071 | |
dd41f596 IM |
4072 | /* |
4073 | * schedule() is the main scheduler function. | |
4074 | */ | |
ff743345 | 4075 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4076 | { |
4077 | struct task_struct *prev, *next; | |
67ca7bde | 4078 | unsigned long *switch_count; |
dd41f596 | 4079 | struct rq *rq; |
31656519 | 4080 | int cpu; |
dd41f596 | 4081 | |
ff743345 PZ |
4082 | need_resched: |
4083 | preempt_disable(); | |
dd41f596 IM |
4084 | cpu = smp_processor_id(); |
4085 | rq = cpu_rq(cpu); | |
25502a6c | 4086 | rcu_note_context_switch(cpu); |
dd41f596 | 4087 | prev = rq->curr; |
dd41f596 | 4088 | |
dd41f596 | 4089 | schedule_debug(prev); |
1da177e4 | 4090 | |
31656519 | 4091 | if (sched_feat(HRTICK)) |
f333fdc9 | 4092 | hrtick_clear(rq); |
8f4d37ec | 4093 | |
05fa785c | 4094 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4095 | |
246d86b5 | 4096 | switch_count = &prev->nivcsw; |
1da177e4 | 4097 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4098 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4099 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4100 | } else { |
4101 | /* | |
4102 | * If a worker is going to sleep, notify and | |
4103 | * ask workqueue whether it wants to wake up a | |
4104 | * task to maintain concurrency. If so, wake | |
4105 | * up the task. | |
4106 | */ | |
4107 | if (prev->flags & PF_WQ_WORKER) { | |
4108 | struct task_struct *to_wakeup; | |
4109 | ||
4110 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4111 | if (to_wakeup) | |
4112 | try_to_wake_up_local(to_wakeup); | |
4113 | } | |
371fd7e7 | 4114 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 4115 | } |
dd41f596 | 4116 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4117 | } |
4118 | ||
73c10101 JA |
4119 | /* |
4120 | * If we are going to sleep and we have plugged IO queued, make | |
4121 | * sure to submit it to avoid deadlocks. | |
4122 | */ | |
4123 | if (prev->state != TASK_RUNNING && blk_needs_flush_plug(prev)) { | |
4124 | raw_spin_unlock(&rq->lock); | |
4125 | blk_flush_plug(prev); | |
4126 | raw_spin_lock(&rq->lock); | |
4127 | } | |
4128 | ||
3f029d3c | 4129 | pre_schedule(rq, prev); |
f65eda4f | 4130 | |
dd41f596 | 4131 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4132 | idle_balance(cpu, rq); |
1da177e4 | 4133 | |
df1c99d4 | 4134 | put_prev_task(rq, prev); |
b67802ea | 4135 | next = pick_next_task(rq); |
f26f9aff MG |
4136 | clear_tsk_need_resched(prev); |
4137 | rq->skip_clock_update = 0; | |
1da177e4 | 4138 | |
1da177e4 | 4139 | if (likely(prev != next)) { |
1da177e4 LT |
4140 | rq->nr_switches++; |
4141 | rq->curr = next; | |
4142 | ++*switch_count; | |
4143 | ||
dd41f596 | 4144 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4145 | /* |
246d86b5 ON |
4146 | * The context switch have flipped the stack from under us |
4147 | * and restored the local variables which were saved when | |
4148 | * this task called schedule() in the past. prev == current | |
4149 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4150 | */ |
4151 | cpu = smp_processor_id(); | |
4152 | rq = cpu_rq(cpu); | |
1da177e4 | 4153 | } else |
05fa785c | 4154 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4155 | |
3f029d3c | 4156 | post_schedule(rq); |
1da177e4 | 4157 | |
1da177e4 | 4158 | preempt_enable_no_resched(); |
ff743345 | 4159 | if (need_resched()) |
1da177e4 LT |
4160 | goto need_resched; |
4161 | } | |
1da177e4 LT |
4162 | EXPORT_SYMBOL(schedule); |
4163 | ||
c08f7829 | 4164 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4165 | /* |
4166 | * Look out! "owner" is an entirely speculative pointer | |
4167 | * access and not reliable. | |
4168 | */ | |
4169 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4170 | { | |
4171 | unsigned int cpu; | |
4172 | struct rq *rq; | |
4173 | ||
4174 | if (!sched_feat(OWNER_SPIN)) | |
4175 | return 0; | |
4176 | ||
4177 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4178 | /* | |
4179 | * Need to access the cpu field knowing that | |
4180 | * DEBUG_PAGEALLOC could have unmapped it if | |
4181 | * the mutex owner just released it and exited. | |
4182 | */ | |
4183 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4184 | return 0; |
0d66bf6d PZ |
4185 | #else |
4186 | cpu = owner->cpu; | |
4187 | #endif | |
4188 | ||
4189 | /* | |
4190 | * Even if the access succeeded (likely case), | |
4191 | * the cpu field may no longer be valid. | |
4192 | */ | |
4193 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4194 | return 0; |
0d66bf6d PZ |
4195 | |
4196 | /* | |
4197 | * We need to validate that we can do a | |
4198 | * get_cpu() and that we have the percpu area. | |
4199 | */ | |
4200 | if (!cpu_online(cpu)) | |
4b402210 | 4201 | return 0; |
0d66bf6d PZ |
4202 | |
4203 | rq = cpu_rq(cpu); | |
4204 | ||
4205 | for (;;) { | |
4206 | /* | |
4207 | * Owner changed, break to re-assess state. | |
4208 | */ | |
9d0f4dcc TC |
4209 | if (lock->owner != owner) { |
4210 | /* | |
4211 | * If the lock has switched to a different owner, | |
4212 | * we likely have heavy contention. Return 0 to quit | |
4213 | * optimistic spinning and not contend further: | |
4214 | */ | |
4215 | if (lock->owner) | |
4216 | return 0; | |
0d66bf6d | 4217 | break; |
9d0f4dcc | 4218 | } |
0d66bf6d PZ |
4219 | |
4220 | /* | |
4221 | * Is that owner really running on that cpu? | |
4222 | */ | |
4223 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4224 | return 0; | |
4225 | ||
335d7afb | 4226 | arch_mutex_cpu_relax(); |
0d66bf6d | 4227 | } |
4b402210 | 4228 | |
0d66bf6d PZ |
4229 | return 1; |
4230 | } | |
4231 | #endif | |
4232 | ||
1da177e4 LT |
4233 | #ifdef CONFIG_PREEMPT |
4234 | /* | |
2ed6e34f | 4235 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4236 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4237 | * occur there and call schedule directly. |
4238 | */ | |
d1f74e20 | 4239 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4240 | { |
4241 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4242 | |
1da177e4 LT |
4243 | /* |
4244 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4245 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4246 | */ |
beed33a8 | 4247 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4248 | return; |
4249 | ||
3a5c359a | 4250 | do { |
d1f74e20 | 4251 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4252 | schedule(); |
d1f74e20 | 4253 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4254 | |
3a5c359a AK |
4255 | /* |
4256 | * Check again in case we missed a preemption opportunity | |
4257 | * between schedule and now. | |
4258 | */ | |
4259 | barrier(); | |
5ed0cec0 | 4260 | } while (need_resched()); |
1da177e4 | 4261 | } |
1da177e4 LT |
4262 | EXPORT_SYMBOL(preempt_schedule); |
4263 | ||
4264 | /* | |
2ed6e34f | 4265 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4266 | * off of irq context. |
4267 | * Note, that this is called and return with irqs disabled. This will | |
4268 | * protect us against recursive calling from irq. | |
4269 | */ | |
4270 | asmlinkage void __sched preempt_schedule_irq(void) | |
4271 | { | |
4272 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4273 | |
2ed6e34f | 4274 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4275 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4276 | ||
3a5c359a AK |
4277 | do { |
4278 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4279 | local_irq_enable(); |
4280 | schedule(); | |
4281 | local_irq_disable(); | |
3a5c359a | 4282 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4283 | |
3a5c359a AK |
4284 | /* |
4285 | * Check again in case we missed a preemption opportunity | |
4286 | * between schedule and now. | |
4287 | */ | |
4288 | barrier(); | |
5ed0cec0 | 4289 | } while (need_resched()); |
1da177e4 LT |
4290 | } |
4291 | ||
4292 | #endif /* CONFIG_PREEMPT */ | |
4293 | ||
63859d4f | 4294 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4295 | void *key) |
1da177e4 | 4296 | { |
63859d4f | 4297 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4298 | } |
1da177e4 LT |
4299 | EXPORT_SYMBOL(default_wake_function); |
4300 | ||
4301 | /* | |
41a2d6cf IM |
4302 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4303 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4304 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4305 | * | |
4306 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4307 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4308 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4309 | */ | |
78ddb08f | 4310 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4311 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4312 | { |
2e45874c | 4313 | wait_queue_t *curr, *next; |
1da177e4 | 4314 | |
2e45874c | 4315 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4316 | unsigned flags = curr->flags; |
4317 | ||
63859d4f | 4318 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4319 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4320 | break; |
4321 | } | |
4322 | } | |
4323 | ||
4324 | /** | |
4325 | * __wake_up - wake up threads blocked on a waitqueue. | |
4326 | * @q: the waitqueue | |
4327 | * @mode: which threads | |
4328 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4329 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4330 | * |
4331 | * It may be assumed that this function implies a write memory barrier before | |
4332 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4333 | */ |
7ad5b3a5 | 4334 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4335 | int nr_exclusive, void *key) |
1da177e4 LT |
4336 | { |
4337 | unsigned long flags; | |
4338 | ||
4339 | spin_lock_irqsave(&q->lock, flags); | |
4340 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4341 | spin_unlock_irqrestore(&q->lock, flags); | |
4342 | } | |
1da177e4 LT |
4343 | EXPORT_SYMBOL(__wake_up); |
4344 | ||
4345 | /* | |
4346 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4347 | */ | |
7ad5b3a5 | 4348 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4349 | { |
4350 | __wake_up_common(q, mode, 1, 0, NULL); | |
4351 | } | |
22c43c81 | 4352 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4353 | |
4ede816a DL |
4354 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4355 | { | |
4356 | __wake_up_common(q, mode, 1, 0, key); | |
4357 | } | |
bf294b41 | 4358 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4359 | |
1da177e4 | 4360 | /** |
4ede816a | 4361 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4362 | * @q: the waitqueue |
4363 | * @mode: which threads | |
4364 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4365 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4366 | * |
4367 | * The sync wakeup differs that the waker knows that it will schedule | |
4368 | * away soon, so while the target thread will be woken up, it will not | |
4369 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4370 | * with each other. This can prevent needless bouncing between CPUs. | |
4371 | * | |
4372 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4373 | * |
4374 | * It may be assumed that this function implies a write memory barrier before | |
4375 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4376 | */ |
4ede816a DL |
4377 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4378 | int nr_exclusive, void *key) | |
1da177e4 LT |
4379 | { |
4380 | unsigned long flags; | |
7d478721 | 4381 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4382 | |
4383 | if (unlikely(!q)) | |
4384 | return; | |
4385 | ||
4386 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4387 | wake_flags = 0; |
1da177e4 LT |
4388 | |
4389 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4390 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4391 | spin_unlock_irqrestore(&q->lock, flags); |
4392 | } | |
4ede816a DL |
4393 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4394 | ||
4395 | /* | |
4396 | * __wake_up_sync - see __wake_up_sync_key() | |
4397 | */ | |
4398 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4399 | { | |
4400 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4401 | } | |
1da177e4 LT |
4402 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4403 | ||
65eb3dc6 KD |
4404 | /** |
4405 | * complete: - signals a single thread waiting on this completion | |
4406 | * @x: holds the state of this particular completion | |
4407 | * | |
4408 | * This will wake up a single thread waiting on this completion. Threads will be | |
4409 | * awakened in the same order in which they were queued. | |
4410 | * | |
4411 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4412 | * |
4413 | * It may be assumed that this function implies a write memory barrier before | |
4414 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4415 | */ |
b15136e9 | 4416 | void complete(struct completion *x) |
1da177e4 LT |
4417 | { |
4418 | unsigned long flags; | |
4419 | ||
4420 | spin_lock_irqsave(&x->wait.lock, flags); | |
4421 | x->done++; | |
d9514f6c | 4422 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4423 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4424 | } | |
4425 | EXPORT_SYMBOL(complete); | |
4426 | ||
65eb3dc6 KD |
4427 | /** |
4428 | * complete_all: - signals all threads waiting on this completion | |
4429 | * @x: holds the state of this particular completion | |
4430 | * | |
4431 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4432 | * |
4433 | * It may be assumed that this function implies a write memory barrier before | |
4434 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4435 | */ |
b15136e9 | 4436 | void complete_all(struct completion *x) |
1da177e4 LT |
4437 | { |
4438 | unsigned long flags; | |
4439 | ||
4440 | spin_lock_irqsave(&x->wait.lock, flags); | |
4441 | x->done += UINT_MAX/2; | |
d9514f6c | 4442 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4443 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4444 | } | |
4445 | EXPORT_SYMBOL(complete_all); | |
4446 | ||
8cbbe86d AK |
4447 | static inline long __sched |
4448 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4449 | { |
1da177e4 LT |
4450 | if (!x->done) { |
4451 | DECLARE_WAITQUEUE(wait, current); | |
4452 | ||
a93d2f17 | 4453 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4454 | do { |
94d3d824 | 4455 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4456 | timeout = -ERESTARTSYS; |
4457 | break; | |
8cbbe86d AK |
4458 | } |
4459 | __set_current_state(state); | |
1da177e4 LT |
4460 | spin_unlock_irq(&x->wait.lock); |
4461 | timeout = schedule_timeout(timeout); | |
4462 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4463 | } while (!x->done && timeout); |
1da177e4 | 4464 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4465 | if (!x->done) |
4466 | return timeout; | |
1da177e4 LT |
4467 | } |
4468 | x->done--; | |
ea71a546 | 4469 | return timeout ?: 1; |
1da177e4 | 4470 | } |
1da177e4 | 4471 | |
8cbbe86d AK |
4472 | static long __sched |
4473 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4474 | { |
1da177e4 LT |
4475 | might_sleep(); |
4476 | ||
4477 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4478 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4479 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4480 | return timeout; |
4481 | } | |
1da177e4 | 4482 | |
65eb3dc6 KD |
4483 | /** |
4484 | * wait_for_completion: - waits for completion of a task | |
4485 | * @x: holds the state of this particular completion | |
4486 | * | |
4487 | * This waits to be signaled for completion of a specific task. It is NOT | |
4488 | * interruptible and there is no timeout. | |
4489 | * | |
4490 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4491 | * and interrupt capability. Also see complete(). | |
4492 | */ | |
b15136e9 | 4493 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4494 | { |
4495 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4496 | } |
8cbbe86d | 4497 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4498 | |
65eb3dc6 KD |
4499 | /** |
4500 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4501 | * @x: holds the state of this particular completion | |
4502 | * @timeout: timeout value in jiffies | |
4503 | * | |
4504 | * This waits for either a completion of a specific task to be signaled or for a | |
4505 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4506 | * interruptible. | |
4507 | */ | |
b15136e9 | 4508 | unsigned long __sched |
8cbbe86d | 4509 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4510 | { |
8cbbe86d | 4511 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4512 | } |
8cbbe86d | 4513 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4514 | |
65eb3dc6 KD |
4515 | /** |
4516 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4517 | * @x: holds the state of this particular completion | |
4518 | * | |
4519 | * This waits for completion of a specific task to be signaled. It is | |
4520 | * interruptible. | |
4521 | */ | |
8cbbe86d | 4522 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4523 | { |
51e97990 AK |
4524 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4525 | if (t == -ERESTARTSYS) | |
4526 | return t; | |
4527 | return 0; | |
0fec171c | 4528 | } |
8cbbe86d | 4529 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4530 | |
65eb3dc6 KD |
4531 | /** |
4532 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4533 | * @x: holds the state of this particular completion | |
4534 | * @timeout: timeout value in jiffies | |
4535 | * | |
4536 | * This waits for either a completion of a specific task to be signaled or for a | |
4537 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4538 | */ | |
6bf41237 | 4539 | long __sched |
8cbbe86d AK |
4540 | wait_for_completion_interruptible_timeout(struct completion *x, |
4541 | unsigned long timeout) | |
0fec171c | 4542 | { |
8cbbe86d | 4543 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4544 | } |
8cbbe86d | 4545 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4546 | |
65eb3dc6 KD |
4547 | /** |
4548 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4549 | * @x: holds the state of this particular completion | |
4550 | * | |
4551 | * This waits to be signaled for completion of a specific task. It can be | |
4552 | * interrupted by a kill signal. | |
4553 | */ | |
009e577e MW |
4554 | int __sched wait_for_completion_killable(struct completion *x) |
4555 | { | |
4556 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4557 | if (t == -ERESTARTSYS) | |
4558 | return t; | |
4559 | return 0; | |
4560 | } | |
4561 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4562 | ||
0aa12fb4 SW |
4563 | /** |
4564 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4565 | * @x: holds the state of this particular completion | |
4566 | * @timeout: timeout value in jiffies | |
4567 | * | |
4568 | * This waits for either a completion of a specific task to be | |
4569 | * signaled or for a specified timeout to expire. It can be | |
4570 | * interrupted by a kill signal. The timeout is in jiffies. | |
4571 | */ | |
6bf41237 | 4572 | long __sched |
0aa12fb4 SW |
4573 | wait_for_completion_killable_timeout(struct completion *x, |
4574 | unsigned long timeout) | |
4575 | { | |
4576 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4577 | } | |
4578 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4579 | ||
be4de352 DC |
4580 | /** |
4581 | * try_wait_for_completion - try to decrement a completion without blocking | |
4582 | * @x: completion structure | |
4583 | * | |
4584 | * Returns: 0 if a decrement cannot be done without blocking | |
4585 | * 1 if a decrement succeeded. | |
4586 | * | |
4587 | * If a completion is being used as a counting completion, | |
4588 | * attempt to decrement the counter without blocking. This | |
4589 | * enables us to avoid waiting if the resource the completion | |
4590 | * is protecting is not available. | |
4591 | */ | |
4592 | bool try_wait_for_completion(struct completion *x) | |
4593 | { | |
7539a3b3 | 4594 | unsigned long flags; |
be4de352 DC |
4595 | int ret = 1; |
4596 | ||
7539a3b3 | 4597 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4598 | if (!x->done) |
4599 | ret = 0; | |
4600 | else | |
4601 | x->done--; | |
7539a3b3 | 4602 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4603 | return ret; |
4604 | } | |
4605 | EXPORT_SYMBOL(try_wait_for_completion); | |
4606 | ||
4607 | /** | |
4608 | * completion_done - Test to see if a completion has any waiters | |
4609 | * @x: completion structure | |
4610 | * | |
4611 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4612 | * 1 if there are no waiters. | |
4613 | * | |
4614 | */ | |
4615 | bool completion_done(struct completion *x) | |
4616 | { | |
7539a3b3 | 4617 | unsigned long flags; |
be4de352 DC |
4618 | int ret = 1; |
4619 | ||
7539a3b3 | 4620 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4621 | if (!x->done) |
4622 | ret = 0; | |
7539a3b3 | 4623 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4624 | return ret; |
4625 | } | |
4626 | EXPORT_SYMBOL(completion_done); | |
4627 | ||
8cbbe86d AK |
4628 | static long __sched |
4629 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4630 | { |
0fec171c IM |
4631 | unsigned long flags; |
4632 | wait_queue_t wait; | |
4633 | ||
4634 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4635 | |
8cbbe86d | 4636 | __set_current_state(state); |
1da177e4 | 4637 | |
8cbbe86d AK |
4638 | spin_lock_irqsave(&q->lock, flags); |
4639 | __add_wait_queue(q, &wait); | |
4640 | spin_unlock(&q->lock); | |
4641 | timeout = schedule_timeout(timeout); | |
4642 | spin_lock_irq(&q->lock); | |
4643 | __remove_wait_queue(q, &wait); | |
4644 | spin_unlock_irqrestore(&q->lock, flags); | |
4645 | ||
4646 | return timeout; | |
4647 | } | |
4648 | ||
4649 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4650 | { | |
4651 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4652 | } |
1da177e4 LT |
4653 | EXPORT_SYMBOL(interruptible_sleep_on); |
4654 | ||
0fec171c | 4655 | long __sched |
95cdf3b7 | 4656 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4657 | { |
8cbbe86d | 4658 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4659 | } |
1da177e4 LT |
4660 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4661 | ||
0fec171c | 4662 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4663 | { |
8cbbe86d | 4664 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4665 | } |
1da177e4 LT |
4666 | EXPORT_SYMBOL(sleep_on); |
4667 | ||
0fec171c | 4668 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4669 | { |
8cbbe86d | 4670 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4671 | } |
1da177e4 LT |
4672 | EXPORT_SYMBOL(sleep_on_timeout); |
4673 | ||
b29739f9 IM |
4674 | #ifdef CONFIG_RT_MUTEXES |
4675 | ||
4676 | /* | |
4677 | * rt_mutex_setprio - set the current priority of a task | |
4678 | * @p: task | |
4679 | * @prio: prio value (kernel-internal form) | |
4680 | * | |
4681 | * This function changes the 'effective' priority of a task. It does | |
4682 | * not touch ->normal_prio like __setscheduler(). | |
4683 | * | |
4684 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4685 | */ | |
36c8b586 | 4686 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4687 | { |
4688 | unsigned long flags; | |
83b699ed | 4689 | int oldprio, on_rq, running; |
70b97a7f | 4690 | struct rq *rq; |
83ab0aa0 | 4691 | const struct sched_class *prev_class; |
b29739f9 IM |
4692 | |
4693 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4694 | ||
4695 | rq = task_rq_lock(p, &flags); | |
4696 | ||
a8027073 | 4697 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4698 | oldprio = p->prio; |
83ab0aa0 | 4699 | prev_class = p->sched_class; |
dd41f596 | 4700 | on_rq = p->se.on_rq; |
051a1d1a | 4701 | running = task_current(rq, p); |
0e1f3483 | 4702 | if (on_rq) |
69be72c1 | 4703 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4704 | if (running) |
4705 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4706 | |
4707 | if (rt_prio(prio)) | |
4708 | p->sched_class = &rt_sched_class; | |
4709 | else | |
4710 | p->sched_class = &fair_sched_class; | |
4711 | ||
b29739f9 IM |
4712 | p->prio = prio; |
4713 | ||
0e1f3483 HS |
4714 | if (running) |
4715 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4716 | if (on_rq) |
371fd7e7 | 4717 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4718 | |
da7a735e | 4719 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 IM |
4720 | task_rq_unlock(rq, &flags); |
4721 | } | |
4722 | ||
4723 | #endif | |
4724 | ||
36c8b586 | 4725 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4726 | { |
dd41f596 | 4727 | int old_prio, delta, on_rq; |
1da177e4 | 4728 | unsigned long flags; |
70b97a7f | 4729 | struct rq *rq; |
1da177e4 LT |
4730 | |
4731 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4732 | return; | |
4733 | /* | |
4734 | * We have to be careful, if called from sys_setpriority(), | |
4735 | * the task might be in the middle of scheduling on another CPU. | |
4736 | */ | |
4737 | rq = task_rq_lock(p, &flags); | |
4738 | /* | |
4739 | * The RT priorities are set via sched_setscheduler(), but we still | |
4740 | * allow the 'normal' nice value to be set - but as expected | |
4741 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4742 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4743 | */ |
e05606d3 | 4744 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4745 | p->static_prio = NICE_TO_PRIO(nice); |
4746 | goto out_unlock; | |
4747 | } | |
dd41f596 | 4748 | on_rq = p->se.on_rq; |
c09595f6 | 4749 | if (on_rq) |
69be72c1 | 4750 | dequeue_task(rq, p, 0); |
1da177e4 | 4751 | |
1da177e4 | 4752 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4753 | set_load_weight(p); |
b29739f9 IM |
4754 | old_prio = p->prio; |
4755 | p->prio = effective_prio(p); | |
4756 | delta = p->prio - old_prio; | |
1da177e4 | 4757 | |
dd41f596 | 4758 | if (on_rq) { |
371fd7e7 | 4759 | enqueue_task(rq, p, 0); |
1da177e4 | 4760 | /* |
d5f9f942 AM |
4761 | * If the task increased its priority or is running and |
4762 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4763 | */ |
d5f9f942 | 4764 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4765 | resched_task(rq->curr); |
4766 | } | |
4767 | out_unlock: | |
4768 | task_rq_unlock(rq, &flags); | |
4769 | } | |
1da177e4 LT |
4770 | EXPORT_SYMBOL(set_user_nice); |
4771 | ||
e43379f1 MM |
4772 | /* |
4773 | * can_nice - check if a task can reduce its nice value | |
4774 | * @p: task | |
4775 | * @nice: nice value | |
4776 | */ | |
36c8b586 | 4777 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4778 | { |
024f4747 MM |
4779 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4780 | int nice_rlim = 20 - nice; | |
48f24c4d | 4781 | |
78d7d407 | 4782 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4783 | capable(CAP_SYS_NICE)); |
4784 | } | |
4785 | ||
1da177e4 LT |
4786 | #ifdef __ARCH_WANT_SYS_NICE |
4787 | ||
4788 | /* | |
4789 | * sys_nice - change the priority of the current process. | |
4790 | * @increment: priority increment | |
4791 | * | |
4792 | * sys_setpriority is a more generic, but much slower function that | |
4793 | * does similar things. | |
4794 | */ | |
5add95d4 | 4795 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4796 | { |
48f24c4d | 4797 | long nice, retval; |
1da177e4 LT |
4798 | |
4799 | /* | |
4800 | * Setpriority might change our priority at the same moment. | |
4801 | * We don't have to worry. Conceptually one call occurs first | |
4802 | * and we have a single winner. | |
4803 | */ | |
e43379f1 MM |
4804 | if (increment < -40) |
4805 | increment = -40; | |
1da177e4 LT |
4806 | if (increment > 40) |
4807 | increment = 40; | |
4808 | ||
2b8f836f | 4809 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4810 | if (nice < -20) |
4811 | nice = -20; | |
4812 | if (nice > 19) | |
4813 | nice = 19; | |
4814 | ||
e43379f1 MM |
4815 | if (increment < 0 && !can_nice(current, nice)) |
4816 | return -EPERM; | |
4817 | ||
1da177e4 LT |
4818 | retval = security_task_setnice(current, nice); |
4819 | if (retval) | |
4820 | return retval; | |
4821 | ||
4822 | set_user_nice(current, nice); | |
4823 | return 0; | |
4824 | } | |
4825 | ||
4826 | #endif | |
4827 | ||
4828 | /** | |
4829 | * task_prio - return the priority value of a given task. | |
4830 | * @p: the task in question. | |
4831 | * | |
4832 | * This is the priority value as seen by users in /proc. | |
4833 | * RT tasks are offset by -200. Normal tasks are centered | |
4834 | * around 0, value goes from -16 to +15. | |
4835 | */ | |
36c8b586 | 4836 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4837 | { |
4838 | return p->prio - MAX_RT_PRIO; | |
4839 | } | |
4840 | ||
4841 | /** | |
4842 | * task_nice - return the nice value of a given task. | |
4843 | * @p: the task in question. | |
4844 | */ | |
36c8b586 | 4845 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4846 | { |
4847 | return TASK_NICE(p); | |
4848 | } | |
150d8bed | 4849 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4850 | |
4851 | /** | |
4852 | * idle_cpu - is a given cpu idle currently? | |
4853 | * @cpu: the processor in question. | |
4854 | */ | |
4855 | int idle_cpu(int cpu) | |
4856 | { | |
4857 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4858 | } | |
4859 | ||
1da177e4 LT |
4860 | /** |
4861 | * idle_task - return the idle task for a given cpu. | |
4862 | * @cpu: the processor in question. | |
4863 | */ | |
36c8b586 | 4864 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4865 | { |
4866 | return cpu_rq(cpu)->idle; | |
4867 | } | |
4868 | ||
4869 | /** | |
4870 | * find_process_by_pid - find a process with a matching PID value. | |
4871 | * @pid: the pid in question. | |
4872 | */ | |
a9957449 | 4873 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4874 | { |
228ebcbe | 4875 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4876 | } |
4877 | ||
4878 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4879 | static void |
4880 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4881 | { |
dd41f596 | 4882 | BUG_ON(p->se.on_rq); |
48f24c4d | 4883 | |
1da177e4 LT |
4884 | p->policy = policy; |
4885 | p->rt_priority = prio; | |
b29739f9 IM |
4886 | p->normal_prio = normal_prio(p); |
4887 | /* we are holding p->pi_lock already */ | |
4888 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4889 | if (rt_prio(p->prio)) |
4890 | p->sched_class = &rt_sched_class; | |
4891 | else | |
4892 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4893 | set_load_weight(p); |
1da177e4 LT |
4894 | } |
4895 | ||
c69e8d9c DH |
4896 | /* |
4897 | * check the target process has a UID that matches the current process's | |
4898 | */ | |
4899 | static bool check_same_owner(struct task_struct *p) | |
4900 | { | |
4901 | const struct cred *cred = current_cred(), *pcred; | |
4902 | bool match; | |
4903 | ||
4904 | rcu_read_lock(); | |
4905 | pcred = __task_cred(p); | |
b0e77598 SH |
4906 | if (cred->user->user_ns == pcred->user->user_ns) |
4907 | match = (cred->euid == pcred->euid || | |
4908 | cred->euid == pcred->uid); | |
4909 | else | |
4910 | match = false; | |
c69e8d9c DH |
4911 | rcu_read_unlock(); |
4912 | return match; | |
4913 | } | |
4914 | ||
961ccddd | 4915 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4916 | const struct sched_param *param, bool user) |
1da177e4 | 4917 | { |
83b699ed | 4918 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4919 | unsigned long flags; |
83ab0aa0 | 4920 | const struct sched_class *prev_class; |
70b97a7f | 4921 | struct rq *rq; |
ca94c442 | 4922 | int reset_on_fork; |
1da177e4 | 4923 | |
66e5393a SR |
4924 | /* may grab non-irq protected spin_locks */ |
4925 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4926 | recheck: |
4927 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4928 | if (policy < 0) { |
4929 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4930 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4931 | } else { |
4932 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4933 | policy &= ~SCHED_RESET_ON_FORK; | |
4934 | ||
4935 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4936 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4937 | policy != SCHED_IDLE) | |
4938 | return -EINVAL; | |
4939 | } | |
4940 | ||
1da177e4 LT |
4941 | /* |
4942 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4943 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4944 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4945 | */ |
4946 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4947 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4948 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4949 | return -EINVAL; |
e05606d3 | 4950 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4951 | return -EINVAL; |
4952 | ||
37e4ab3f OC |
4953 | /* |
4954 | * Allow unprivileged RT tasks to decrease priority: | |
4955 | */ | |
961ccddd | 4956 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4957 | if (rt_policy(policy)) { |
a44702e8 ON |
4958 | unsigned long rlim_rtprio = |
4959 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4960 | |
4961 | /* can't set/change the rt policy */ | |
4962 | if (policy != p->policy && !rlim_rtprio) | |
4963 | return -EPERM; | |
4964 | ||
4965 | /* can't increase priority */ | |
4966 | if (param->sched_priority > p->rt_priority && | |
4967 | param->sched_priority > rlim_rtprio) | |
4968 | return -EPERM; | |
4969 | } | |
c02aa73b | 4970 | |
dd41f596 | 4971 | /* |
c02aa73b DH |
4972 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4973 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4974 | */ |
c02aa73b DH |
4975 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4976 | if (!can_nice(p, TASK_NICE(p))) | |
4977 | return -EPERM; | |
4978 | } | |
5fe1d75f | 4979 | |
37e4ab3f | 4980 | /* can't change other user's priorities */ |
c69e8d9c | 4981 | if (!check_same_owner(p)) |
37e4ab3f | 4982 | return -EPERM; |
ca94c442 LP |
4983 | |
4984 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4985 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4986 | return -EPERM; | |
37e4ab3f | 4987 | } |
1da177e4 | 4988 | |
725aad24 | 4989 | if (user) { |
b0ae1981 | 4990 | retval = security_task_setscheduler(p); |
725aad24 JF |
4991 | if (retval) |
4992 | return retval; | |
4993 | } | |
4994 | ||
b29739f9 IM |
4995 | /* |
4996 | * make sure no PI-waiters arrive (or leave) while we are | |
4997 | * changing the priority of the task: | |
4998 | */ | |
1d615482 | 4999 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 | 5000 | /* |
25985edc | 5001 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5002 | * runqueue lock must be held. |
5003 | */ | |
b29739f9 | 5004 | rq = __task_rq_lock(p); |
dc61b1d6 | 5005 | |
34f971f6 PZ |
5006 | /* |
5007 | * Changing the policy of the stop threads its a very bad idea | |
5008 | */ | |
5009 | if (p == rq->stop) { | |
5010 | __task_rq_unlock(rq); | |
5011 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5012 | return -EINVAL; | |
5013 | } | |
5014 | ||
a51e9198 DF |
5015 | /* |
5016 | * If not changing anything there's no need to proceed further: | |
5017 | */ | |
5018 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5019 | param->sched_priority == p->rt_priority))) { | |
5020 | ||
5021 | __task_rq_unlock(rq); | |
5022 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5023 | return 0; | |
5024 | } | |
5025 | ||
dc61b1d6 PZ |
5026 | #ifdef CONFIG_RT_GROUP_SCHED |
5027 | if (user) { | |
5028 | /* | |
5029 | * Do not allow realtime tasks into groups that have no runtime | |
5030 | * assigned. | |
5031 | */ | |
5032 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5033 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5034 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
5035 | __task_rq_unlock(rq); |
5036 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5037 | return -EPERM; | |
5038 | } | |
5039 | } | |
5040 | #endif | |
5041 | ||
1da177e4 LT |
5042 | /* recheck policy now with rq lock held */ |
5043 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5044 | policy = oldpolicy = -1; | |
b29739f9 | 5045 | __task_rq_unlock(rq); |
1d615482 | 5046 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5047 | goto recheck; |
5048 | } | |
dd41f596 | 5049 | on_rq = p->se.on_rq; |
051a1d1a | 5050 | running = task_current(rq, p); |
0e1f3483 | 5051 | if (on_rq) |
2e1cb74a | 5052 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5053 | if (running) |
5054 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5055 | |
ca94c442 LP |
5056 | p->sched_reset_on_fork = reset_on_fork; |
5057 | ||
1da177e4 | 5058 | oldprio = p->prio; |
83ab0aa0 | 5059 | prev_class = p->sched_class; |
dd41f596 | 5060 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5061 | |
0e1f3483 HS |
5062 | if (running) |
5063 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5064 | if (on_rq) |
dd41f596 | 5065 | activate_task(rq, p, 0); |
cb469845 | 5066 | |
da7a735e | 5067 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 | 5068 | __task_rq_unlock(rq); |
1d615482 | 5069 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 5070 | |
95e02ca9 TG |
5071 | rt_mutex_adjust_pi(p); |
5072 | ||
1da177e4 LT |
5073 | return 0; |
5074 | } | |
961ccddd RR |
5075 | |
5076 | /** | |
5077 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5078 | * @p: the task in question. | |
5079 | * @policy: new policy. | |
5080 | * @param: structure containing the new RT priority. | |
5081 | * | |
5082 | * NOTE that the task may be already dead. | |
5083 | */ | |
5084 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5085 | const struct sched_param *param) |
961ccddd RR |
5086 | { |
5087 | return __sched_setscheduler(p, policy, param, true); | |
5088 | } | |
1da177e4 LT |
5089 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5090 | ||
961ccddd RR |
5091 | /** |
5092 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5093 | * @p: the task in question. | |
5094 | * @policy: new policy. | |
5095 | * @param: structure containing the new RT priority. | |
5096 | * | |
5097 | * Just like sched_setscheduler, only don't bother checking if the | |
5098 | * current context has permission. For example, this is needed in | |
5099 | * stop_machine(): we create temporary high priority worker threads, | |
5100 | * but our caller might not have that capability. | |
5101 | */ | |
5102 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5103 | const struct sched_param *param) |
961ccddd RR |
5104 | { |
5105 | return __sched_setscheduler(p, policy, param, false); | |
5106 | } | |
5107 | ||
95cdf3b7 IM |
5108 | static int |
5109 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5110 | { |
1da177e4 LT |
5111 | struct sched_param lparam; |
5112 | struct task_struct *p; | |
36c8b586 | 5113 | int retval; |
1da177e4 LT |
5114 | |
5115 | if (!param || pid < 0) | |
5116 | return -EINVAL; | |
5117 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5118 | return -EFAULT; | |
5fe1d75f ON |
5119 | |
5120 | rcu_read_lock(); | |
5121 | retval = -ESRCH; | |
1da177e4 | 5122 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5123 | if (p != NULL) |
5124 | retval = sched_setscheduler(p, policy, &lparam); | |
5125 | rcu_read_unlock(); | |
36c8b586 | 5126 | |
1da177e4 LT |
5127 | return retval; |
5128 | } | |
5129 | ||
5130 | /** | |
5131 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5132 | * @pid: the pid in question. | |
5133 | * @policy: new policy. | |
5134 | * @param: structure containing the new RT priority. | |
5135 | */ | |
5add95d4 HC |
5136 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5137 | struct sched_param __user *, param) | |
1da177e4 | 5138 | { |
c21761f1 JB |
5139 | /* negative values for policy are not valid */ |
5140 | if (policy < 0) | |
5141 | return -EINVAL; | |
5142 | ||
1da177e4 LT |
5143 | return do_sched_setscheduler(pid, policy, param); |
5144 | } | |
5145 | ||
5146 | /** | |
5147 | * sys_sched_setparam - set/change the RT priority of a thread | |
5148 | * @pid: the pid in question. | |
5149 | * @param: structure containing the new RT priority. | |
5150 | */ | |
5add95d4 | 5151 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5152 | { |
5153 | return do_sched_setscheduler(pid, -1, param); | |
5154 | } | |
5155 | ||
5156 | /** | |
5157 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5158 | * @pid: the pid in question. | |
5159 | */ | |
5add95d4 | 5160 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5161 | { |
36c8b586 | 5162 | struct task_struct *p; |
3a5c359a | 5163 | int retval; |
1da177e4 LT |
5164 | |
5165 | if (pid < 0) | |
3a5c359a | 5166 | return -EINVAL; |
1da177e4 LT |
5167 | |
5168 | retval = -ESRCH; | |
5fe85be0 | 5169 | rcu_read_lock(); |
1da177e4 LT |
5170 | p = find_process_by_pid(pid); |
5171 | if (p) { | |
5172 | retval = security_task_getscheduler(p); | |
5173 | if (!retval) | |
ca94c442 LP |
5174 | retval = p->policy |
5175 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5176 | } |
5fe85be0 | 5177 | rcu_read_unlock(); |
1da177e4 LT |
5178 | return retval; |
5179 | } | |
5180 | ||
5181 | /** | |
ca94c442 | 5182 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5183 | * @pid: the pid in question. |
5184 | * @param: structure containing the RT priority. | |
5185 | */ | |
5add95d4 | 5186 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5187 | { |
5188 | struct sched_param lp; | |
36c8b586 | 5189 | struct task_struct *p; |
3a5c359a | 5190 | int retval; |
1da177e4 LT |
5191 | |
5192 | if (!param || pid < 0) | |
3a5c359a | 5193 | return -EINVAL; |
1da177e4 | 5194 | |
5fe85be0 | 5195 | rcu_read_lock(); |
1da177e4 LT |
5196 | p = find_process_by_pid(pid); |
5197 | retval = -ESRCH; | |
5198 | if (!p) | |
5199 | goto out_unlock; | |
5200 | ||
5201 | retval = security_task_getscheduler(p); | |
5202 | if (retval) | |
5203 | goto out_unlock; | |
5204 | ||
5205 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5206 | rcu_read_unlock(); |
1da177e4 LT |
5207 | |
5208 | /* | |
5209 | * This one might sleep, we cannot do it with a spinlock held ... | |
5210 | */ | |
5211 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5212 | ||
1da177e4 LT |
5213 | return retval; |
5214 | ||
5215 | out_unlock: | |
5fe85be0 | 5216 | rcu_read_unlock(); |
1da177e4 LT |
5217 | return retval; |
5218 | } | |
5219 | ||
96f874e2 | 5220 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5221 | { |
5a16f3d3 | 5222 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5223 | struct task_struct *p; |
5224 | int retval; | |
1da177e4 | 5225 | |
95402b38 | 5226 | get_online_cpus(); |
23f5d142 | 5227 | rcu_read_lock(); |
1da177e4 LT |
5228 | |
5229 | p = find_process_by_pid(pid); | |
5230 | if (!p) { | |
23f5d142 | 5231 | rcu_read_unlock(); |
95402b38 | 5232 | put_online_cpus(); |
1da177e4 LT |
5233 | return -ESRCH; |
5234 | } | |
5235 | ||
23f5d142 | 5236 | /* Prevent p going away */ |
1da177e4 | 5237 | get_task_struct(p); |
23f5d142 | 5238 | rcu_read_unlock(); |
1da177e4 | 5239 | |
5a16f3d3 RR |
5240 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5241 | retval = -ENOMEM; | |
5242 | goto out_put_task; | |
5243 | } | |
5244 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5245 | retval = -ENOMEM; | |
5246 | goto out_free_cpus_allowed; | |
5247 | } | |
1da177e4 | 5248 | retval = -EPERM; |
b0e77598 | 5249 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5250 | goto out_unlock; |
5251 | ||
b0ae1981 | 5252 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5253 | if (retval) |
5254 | goto out_unlock; | |
5255 | ||
5a16f3d3 RR |
5256 | cpuset_cpus_allowed(p, cpus_allowed); |
5257 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5258 | again: |
5a16f3d3 | 5259 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5260 | |
8707d8b8 | 5261 | if (!retval) { |
5a16f3d3 RR |
5262 | cpuset_cpus_allowed(p, cpus_allowed); |
5263 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5264 | /* |
5265 | * We must have raced with a concurrent cpuset | |
5266 | * update. Just reset the cpus_allowed to the | |
5267 | * cpuset's cpus_allowed | |
5268 | */ | |
5a16f3d3 | 5269 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5270 | goto again; |
5271 | } | |
5272 | } | |
1da177e4 | 5273 | out_unlock: |
5a16f3d3 RR |
5274 | free_cpumask_var(new_mask); |
5275 | out_free_cpus_allowed: | |
5276 | free_cpumask_var(cpus_allowed); | |
5277 | out_put_task: | |
1da177e4 | 5278 | put_task_struct(p); |
95402b38 | 5279 | put_online_cpus(); |
1da177e4 LT |
5280 | return retval; |
5281 | } | |
5282 | ||
5283 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5284 | struct cpumask *new_mask) |
1da177e4 | 5285 | { |
96f874e2 RR |
5286 | if (len < cpumask_size()) |
5287 | cpumask_clear(new_mask); | |
5288 | else if (len > cpumask_size()) | |
5289 | len = cpumask_size(); | |
5290 | ||
1da177e4 LT |
5291 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5292 | } | |
5293 | ||
5294 | /** | |
5295 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5296 | * @pid: pid of the process | |
5297 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5298 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5299 | */ | |
5add95d4 HC |
5300 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5301 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5302 | { |
5a16f3d3 | 5303 | cpumask_var_t new_mask; |
1da177e4 LT |
5304 | int retval; |
5305 | ||
5a16f3d3 RR |
5306 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5307 | return -ENOMEM; | |
1da177e4 | 5308 | |
5a16f3d3 RR |
5309 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5310 | if (retval == 0) | |
5311 | retval = sched_setaffinity(pid, new_mask); | |
5312 | free_cpumask_var(new_mask); | |
5313 | return retval; | |
1da177e4 LT |
5314 | } |
5315 | ||
96f874e2 | 5316 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5317 | { |
36c8b586 | 5318 | struct task_struct *p; |
31605683 TG |
5319 | unsigned long flags; |
5320 | struct rq *rq; | |
1da177e4 | 5321 | int retval; |
1da177e4 | 5322 | |
95402b38 | 5323 | get_online_cpus(); |
23f5d142 | 5324 | rcu_read_lock(); |
1da177e4 LT |
5325 | |
5326 | retval = -ESRCH; | |
5327 | p = find_process_by_pid(pid); | |
5328 | if (!p) | |
5329 | goto out_unlock; | |
5330 | ||
e7834f8f DQ |
5331 | retval = security_task_getscheduler(p); |
5332 | if (retval) | |
5333 | goto out_unlock; | |
5334 | ||
31605683 | 5335 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5336 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5337 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5338 | |
5339 | out_unlock: | |
23f5d142 | 5340 | rcu_read_unlock(); |
95402b38 | 5341 | put_online_cpus(); |
1da177e4 | 5342 | |
9531b62f | 5343 | return retval; |
1da177e4 LT |
5344 | } |
5345 | ||
5346 | /** | |
5347 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5348 | * @pid: pid of the process | |
5349 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5350 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5351 | */ | |
5add95d4 HC |
5352 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5353 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5354 | { |
5355 | int ret; | |
f17c8607 | 5356 | cpumask_var_t mask; |
1da177e4 | 5357 | |
84fba5ec | 5358 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5359 | return -EINVAL; |
5360 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5361 | return -EINVAL; |
5362 | ||
f17c8607 RR |
5363 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5364 | return -ENOMEM; | |
1da177e4 | 5365 | |
f17c8607 RR |
5366 | ret = sched_getaffinity(pid, mask); |
5367 | if (ret == 0) { | |
8bc037fb | 5368 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5369 | |
5370 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5371 | ret = -EFAULT; |
5372 | else | |
cd3d8031 | 5373 | ret = retlen; |
f17c8607 RR |
5374 | } |
5375 | free_cpumask_var(mask); | |
1da177e4 | 5376 | |
f17c8607 | 5377 | return ret; |
1da177e4 LT |
5378 | } |
5379 | ||
5380 | /** | |
5381 | * sys_sched_yield - yield the current processor to other threads. | |
5382 | * | |
dd41f596 IM |
5383 | * This function yields the current CPU to other tasks. If there are no |
5384 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5385 | */ |
5add95d4 | 5386 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5387 | { |
70b97a7f | 5388 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5389 | |
2d72376b | 5390 | schedstat_inc(rq, yld_count); |
4530d7ab | 5391 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5392 | |
5393 | /* | |
5394 | * Since we are going to call schedule() anyway, there's | |
5395 | * no need to preempt or enable interrupts: | |
5396 | */ | |
5397 | __release(rq->lock); | |
8a25d5de | 5398 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5399 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5400 | preempt_enable_no_resched(); |
5401 | ||
5402 | schedule(); | |
5403 | ||
5404 | return 0; | |
5405 | } | |
5406 | ||
d86ee480 PZ |
5407 | static inline int should_resched(void) |
5408 | { | |
5409 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5410 | } | |
5411 | ||
e7b38404 | 5412 | static void __cond_resched(void) |
1da177e4 | 5413 | { |
e7aaaa69 FW |
5414 | add_preempt_count(PREEMPT_ACTIVE); |
5415 | schedule(); | |
5416 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5417 | } |
5418 | ||
02b67cc3 | 5419 | int __sched _cond_resched(void) |
1da177e4 | 5420 | { |
d86ee480 | 5421 | if (should_resched()) { |
1da177e4 LT |
5422 | __cond_resched(); |
5423 | return 1; | |
5424 | } | |
5425 | return 0; | |
5426 | } | |
02b67cc3 | 5427 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5428 | |
5429 | /* | |
613afbf8 | 5430 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5431 | * call schedule, and on return reacquire the lock. |
5432 | * | |
41a2d6cf | 5433 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5434 | * operations here to prevent schedule() from being called twice (once via |
5435 | * spin_unlock(), once by hand). | |
5436 | */ | |
613afbf8 | 5437 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5438 | { |
d86ee480 | 5439 | int resched = should_resched(); |
6df3cecb JK |
5440 | int ret = 0; |
5441 | ||
f607c668 PZ |
5442 | lockdep_assert_held(lock); |
5443 | ||
95c354fe | 5444 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5445 | spin_unlock(lock); |
d86ee480 | 5446 | if (resched) |
95c354fe NP |
5447 | __cond_resched(); |
5448 | else | |
5449 | cpu_relax(); | |
6df3cecb | 5450 | ret = 1; |
1da177e4 | 5451 | spin_lock(lock); |
1da177e4 | 5452 | } |
6df3cecb | 5453 | return ret; |
1da177e4 | 5454 | } |
613afbf8 | 5455 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5456 | |
613afbf8 | 5457 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5458 | { |
5459 | BUG_ON(!in_softirq()); | |
5460 | ||
d86ee480 | 5461 | if (should_resched()) { |
98d82567 | 5462 | local_bh_enable(); |
1da177e4 LT |
5463 | __cond_resched(); |
5464 | local_bh_disable(); | |
5465 | return 1; | |
5466 | } | |
5467 | return 0; | |
5468 | } | |
613afbf8 | 5469 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5470 | |
1da177e4 LT |
5471 | /** |
5472 | * yield - yield the current processor to other threads. | |
5473 | * | |
72fd4a35 | 5474 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5475 | * thread runnable and calls sys_sched_yield(). |
5476 | */ | |
5477 | void __sched yield(void) | |
5478 | { | |
5479 | set_current_state(TASK_RUNNING); | |
5480 | sys_sched_yield(); | |
5481 | } | |
1da177e4 LT |
5482 | EXPORT_SYMBOL(yield); |
5483 | ||
d95f4122 MG |
5484 | /** |
5485 | * yield_to - yield the current processor to another thread in | |
5486 | * your thread group, or accelerate that thread toward the | |
5487 | * processor it's on. | |
16addf95 RD |
5488 | * @p: target task |
5489 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5490 | * |
5491 | * It's the caller's job to ensure that the target task struct | |
5492 | * can't go away on us before we can do any checks. | |
5493 | * | |
5494 | * Returns true if we indeed boosted the target task. | |
5495 | */ | |
5496 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5497 | { | |
5498 | struct task_struct *curr = current; | |
5499 | struct rq *rq, *p_rq; | |
5500 | unsigned long flags; | |
5501 | bool yielded = 0; | |
5502 | ||
5503 | local_irq_save(flags); | |
5504 | rq = this_rq(); | |
5505 | ||
5506 | again: | |
5507 | p_rq = task_rq(p); | |
5508 | double_rq_lock(rq, p_rq); | |
5509 | while (task_rq(p) != p_rq) { | |
5510 | double_rq_unlock(rq, p_rq); | |
5511 | goto again; | |
5512 | } | |
5513 | ||
5514 | if (!curr->sched_class->yield_to_task) | |
5515 | goto out; | |
5516 | ||
5517 | if (curr->sched_class != p->sched_class) | |
5518 | goto out; | |
5519 | ||
5520 | if (task_running(p_rq, p) || p->state) | |
5521 | goto out; | |
5522 | ||
5523 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5524 | if (yielded) { |
d95f4122 | 5525 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5526 | /* |
5527 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5528 | * fairness. | |
5529 | */ | |
5530 | if (preempt && rq != p_rq) | |
5531 | resched_task(p_rq->curr); | |
5532 | } | |
d95f4122 MG |
5533 | |
5534 | out: | |
5535 | double_rq_unlock(rq, p_rq); | |
5536 | local_irq_restore(flags); | |
5537 | ||
5538 | if (yielded) | |
5539 | schedule(); | |
5540 | ||
5541 | return yielded; | |
5542 | } | |
5543 | EXPORT_SYMBOL_GPL(yield_to); | |
5544 | ||
1da177e4 | 5545 | /* |
41a2d6cf | 5546 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5547 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5548 | */ |
5549 | void __sched io_schedule(void) | |
5550 | { | |
54d35f29 | 5551 | struct rq *rq = raw_rq(); |
1da177e4 | 5552 | |
0ff92245 | 5553 | delayacct_blkio_start(); |
1da177e4 | 5554 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5555 | blk_flush_plug(current); |
8f0dfc34 | 5556 | current->in_iowait = 1; |
1da177e4 | 5557 | schedule(); |
8f0dfc34 | 5558 | current->in_iowait = 0; |
1da177e4 | 5559 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5560 | delayacct_blkio_end(); |
1da177e4 | 5561 | } |
1da177e4 LT |
5562 | EXPORT_SYMBOL(io_schedule); |
5563 | ||
5564 | long __sched io_schedule_timeout(long timeout) | |
5565 | { | |
54d35f29 | 5566 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5567 | long ret; |
5568 | ||
0ff92245 | 5569 | delayacct_blkio_start(); |
1da177e4 | 5570 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5571 | blk_flush_plug(current); |
8f0dfc34 | 5572 | current->in_iowait = 1; |
1da177e4 | 5573 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5574 | current->in_iowait = 0; |
1da177e4 | 5575 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5576 | delayacct_blkio_end(); |
1da177e4 LT |
5577 | return ret; |
5578 | } | |
5579 | ||
5580 | /** | |
5581 | * sys_sched_get_priority_max - return maximum RT priority. | |
5582 | * @policy: scheduling class. | |
5583 | * | |
5584 | * this syscall returns the maximum rt_priority that can be used | |
5585 | * by a given scheduling class. | |
5586 | */ | |
5add95d4 | 5587 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5588 | { |
5589 | int ret = -EINVAL; | |
5590 | ||
5591 | switch (policy) { | |
5592 | case SCHED_FIFO: | |
5593 | case SCHED_RR: | |
5594 | ret = MAX_USER_RT_PRIO-1; | |
5595 | break; | |
5596 | case SCHED_NORMAL: | |
b0a9499c | 5597 | case SCHED_BATCH: |
dd41f596 | 5598 | case SCHED_IDLE: |
1da177e4 LT |
5599 | ret = 0; |
5600 | break; | |
5601 | } | |
5602 | return ret; | |
5603 | } | |
5604 | ||
5605 | /** | |
5606 | * sys_sched_get_priority_min - return minimum RT priority. | |
5607 | * @policy: scheduling class. | |
5608 | * | |
5609 | * this syscall returns the minimum rt_priority that can be used | |
5610 | * by a given scheduling class. | |
5611 | */ | |
5add95d4 | 5612 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5613 | { |
5614 | int ret = -EINVAL; | |
5615 | ||
5616 | switch (policy) { | |
5617 | case SCHED_FIFO: | |
5618 | case SCHED_RR: | |
5619 | ret = 1; | |
5620 | break; | |
5621 | case SCHED_NORMAL: | |
b0a9499c | 5622 | case SCHED_BATCH: |
dd41f596 | 5623 | case SCHED_IDLE: |
1da177e4 LT |
5624 | ret = 0; |
5625 | } | |
5626 | return ret; | |
5627 | } | |
5628 | ||
5629 | /** | |
5630 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5631 | * @pid: pid of the process. | |
5632 | * @interval: userspace pointer to the timeslice value. | |
5633 | * | |
5634 | * this syscall writes the default timeslice value of a given process | |
5635 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5636 | */ | |
17da2bd9 | 5637 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5638 | struct timespec __user *, interval) |
1da177e4 | 5639 | { |
36c8b586 | 5640 | struct task_struct *p; |
a4ec24b4 | 5641 | unsigned int time_slice; |
dba091b9 TG |
5642 | unsigned long flags; |
5643 | struct rq *rq; | |
3a5c359a | 5644 | int retval; |
1da177e4 | 5645 | struct timespec t; |
1da177e4 LT |
5646 | |
5647 | if (pid < 0) | |
3a5c359a | 5648 | return -EINVAL; |
1da177e4 LT |
5649 | |
5650 | retval = -ESRCH; | |
1a551ae7 | 5651 | rcu_read_lock(); |
1da177e4 LT |
5652 | p = find_process_by_pid(pid); |
5653 | if (!p) | |
5654 | goto out_unlock; | |
5655 | ||
5656 | retval = security_task_getscheduler(p); | |
5657 | if (retval) | |
5658 | goto out_unlock; | |
5659 | ||
dba091b9 TG |
5660 | rq = task_rq_lock(p, &flags); |
5661 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5662 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5663 | |
1a551ae7 | 5664 | rcu_read_unlock(); |
a4ec24b4 | 5665 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5666 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5667 | return retval; |
3a5c359a | 5668 | |
1da177e4 | 5669 | out_unlock: |
1a551ae7 | 5670 | rcu_read_unlock(); |
1da177e4 LT |
5671 | return retval; |
5672 | } | |
5673 | ||
7c731e0a | 5674 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5675 | |
82a1fcb9 | 5676 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5677 | { |
1da177e4 | 5678 | unsigned long free = 0; |
36c8b586 | 5679 | unsigned state; |
1da177e4 | 5680 | |
1da177e4 | 5681 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5682 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5683 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5684 | #if BITS_PER_LONG == 32 |
1da177e4 | 5685 | if (state == TASK_RUNNING) |
3df0fc5b | 5686 | printk(KERN_CONT " running "); |
1da177e4 | 5687 | else |
3df0fc5b | 5688 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5689 | #else |
5690 | if (state == TASK_RUNNING) | |
3df0fc5b | 5691 | printk(KERN_CONT " running task "); |
1da177e4 | 5692 | else |
3df0fc5b | 5693 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5694 | #endif |
5695 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5696 | free = stack_not_used(p); |
1da177e4 | 5697 | #endif |
3df0fc5b | 5698 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5699 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5700 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5701 | |
5fb5e6de | 5702 | show_stack(p, NULL); |
1da177e4 LT |
5703 | } |
5704 | ||
e59e2ae2 | 5705 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5706 | { |
36c8b586 | 5707 | struct task_struct *g, *p; |
1da177e4 | 5708 | |
4bd77321 | 5709 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5710 | printk(KERN_INFO |
5711 | " task PC stack pid father\n"); | |
1da177e4 | 5712 | #else |
3df0fc5b PZ |
5713 | printk(KERN_INFO |
5714 | " task PC stack pid father\n"); | |
1da177e4 LT |
5715 | #endif |
5716 | read_lock(&tasklist_lock); | |
5717 | do_each_thread(g, p) { | |
5718 | /* | |
5719 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5720 | * console might take a lot of time: |
1da177e4 LT |
5721 | */ |
5722 | touch_nmi_watchdog(); | |
39bc89fd | 5723 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5724 | sched_show_task(p); |
1da177e4 LT |
5725 | } while_each_thread(g, p); |
5726 | ||
04c9167f JF |
5727 | touch_all_softlockup_watchdogs(); |
5728 | ||
dd41f596 IM |
5729 | #ifdef CONFIG_SCHED_DEBUG |
5730 | sysrq_sched_debug_show(); | |
5731 | #endif | |
1da177e4 | 5732 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5733 | /* |
5734 | * Only show locks if all tasks are dumped: | |
5735 | */ | |
93335a21 | 5736 | if (!state_filter) |
e59e2ae2 | 5737 | debug_show_all_locks(); |
1da177e4 LT |
5738 | } |
5739 | ||
1df21055 IM |
5740 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5741 | { | |
dd41f596 | 5742 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5743 | } |
5744 | ||
f340c0d1 IM |
5745 | /** |
5746 | * init_idle - set up an idle thread for a given CPU | |
5747 | * @idle: task in question | |
5748 | * @cpu: cpu the idle task belongs to | |
5749 | * | |
5750 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5751 | * flag, to make booting more robust. | |
5752 | */ | |
5c1e1767 | 5753 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5754 | { |
70b97a7f | 5755 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5756 | unsigned long flags; |
5757 | ||
05fa785c | 5758 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5759 | |
dd41f596 | 5760 | __sched_fork(idle); |
06b83b5f | 5761 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5762 | idle->se.exec_start = sched_clock(); |
5763 | ||
96f874e2 | 5764 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5765 | /* |
5766 | * We're having a chicken and egg problem, even though we are | |
5767 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5768 | * lockdep check in task_group() will fail. | |
5769 | * | |
5770 | * Similar case to sched_fork(). / Alternatively we could | |
5771 | * use task_rq_lock() here and obtain the other rq->lock. | |
5772 | * | |
5773 | * Silence PROVE_RCU | |
5774 | */ | |
5775 | rcu_read_lock(); | |
dd41f596 | 5776 | __set_task_cpu(idle, cpu); |
6506cf6c | 5777 | rcu_read_unlock(); |
1da177e4 | 5778 | |
1da177e4 | 5779 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5780 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5781 | idle->oncpu = 1; | |
5782 | #endif | |
05fa785c | 5783 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5784 | |
5785 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5786 | #if defined(CONFIG_PREEMPT) |
5787 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5788 | #else | |
a1261f54 | 5789 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5790 | #endif |
dd41f596 IM |
5791 | /* |
5792 | * The idle tasks have their own, simple scheduling class: | |
5793 | */ | |
5794 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5795 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5796 | } |
5797 | ||
5798 | /* | |
5799 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5800 | * indicates which cpus entered this state. This is used | |
5801 | * in the rcu update to wait only for active cpus. For system | |
5802 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5803 | * always be CPU_BITS_NONE. |
1da177e4 | 5804 | */ |
6a7b3dc3 | 5805 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5806 | |
19978ca6 IM |
5807 | /* |
5808 | * Increase the granularity value when there are more CPUs, | |
5809 | * because with more CPUs the 'effective latency' as visible | |
5810 | * to users decreases. But the relationship is not linear, | |
5811 | * so pick a second-best guess by going with the log2 of the | |
5812 | * number of CPUs. | |
5813 | * | |
5814 | * This idea comes from the SD scheduler of Con Kolivas: | |
5815 | */ | |
acb4a848 | 5816 | static int get_update_sysctl_factor(void) |
19978ca6 | 5817 | { |
4ca3ef71 | 5818 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5819 | unsigned int factor; |
5820 | ||
5821 | switch (sysctl_sched_tunable_scaling) { | |
5822 | case SCHED_TUNABLESCALING_NONE: | |
5823 | factor = 1; | |
5824 | break; | |
5825 | case SCHED_TUNABLESCALING_LINEAR: | |
5826 | factor = cpus; | |
5827 | break; | |
5828 | case SCHED_TUNABLESCALING_LOG: | |
5829 | default: | |
5830 | factor = 1 + ilog2(cpus); | |
5831 | break; | |
5832 | } | |
19978ca6 | 5833 | |
acb4a848 CE |
5834 | return factor; |
5835 | } | |
19978ca6 | 5836 | |
acb4a848 CE |
5837 | static void update_sysctl(void) |
5838 | { | |
5839 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5840 | |
0bcdcf28 CE |
5841 | #define SET_SYSCTL(name) \ |
5842 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5843 | SET_SYSCTL(sched_min_granularity); | |
5844 | SET_SYSCTL(sched_latency); | |
5845 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5846 | #undef SET_SYSCTL |
5847 | } | |
55cd5340 | 5848 | |
0bcdcf28 CE |
5849 | static inline void sched_init_granularity(void) |
5850 | { | |
5851 | update_sysctl(); | |
19978ca6 IM |
5852 | } |
5853 | ||
1da177e4 LT |
5854 | #ifdef CONFIG_SMP |
5855 | /* | |
5856 | * This is how migration works: | |
5857 | * | |
969c7921 TH |
5858 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5859 | * stop_one_cpu(). | |
5860 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5861 | * off the CPU) | |
5862 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5863 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5864 | * it and puts it into the right queue. |
969c7921 TH |
5865 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5866 | * is done. | |
1da177e4 LT |
5867 | */ |
5868 | ||
5869 | /* | |
5870 | * Change a given task's CPU affinity. Migrate the thread to a | |
5871 | * proper CPU and schedule it away if the CPU it's executing on | |
5872 | * is removed from the allowed bitmask. | |
5873 | * | |
5874 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5875 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5876 | * call is not atomic; no spinlocks may be held. |
5877 | */ | |
96f874e2 | 5878 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5879 | { |
5880 | unsigned long flags; | |
70b97a7f | 5881 | struct rq *rq; |
969c7921 | 5882 | unsigned int dest_cpu; |
48f24c4d | 5883 | int ret = 0; |
1da177e4 | 5884 | |
65cc8e48 PZ |
5885 | /* |
5886 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5887 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5888 | */ | |
5889 | again: | |
5890 | while (task_is_waking(p)) | |
5891 | cpu_relax(); | |
1da177e4 | 5892 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5893 | if (task_is_waking(p)) { |
5894 | task_rq_unlock(rq, &flags); | |
5895 | goto again; | |
5896 | } | |
e2912009 | 5897 | |
6ad4c188 | 5898 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5899 | ret = -EINVAL; |
5900 | goto out; | |
5901 | } | |
5902 | ||
9985b0ba | 5903 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5904 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5905 | ret = -EINVAL; |
5906 | goto out; | |
5907 | } | |
5908 | ||
73fe6aae | 5909 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5910 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5911 | else { |
96f874e2 RR |
5912 | cpumask_copy(&p->cpus_allowed, new_mask); |
5913 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5914 | } |
5915 | ||
1da177e4 | 5916 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5917 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5918 | goto out; |
5919 | ||
969c7921 | 5920 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5921 | if (migrate_task(p, rq)) { |
969c7921 | 5922 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5923 | /* Need help from migration thread: drop lock and wait. */ |
5924 | task_rq_unlock(rq, &flags); | |
969c7921 | 5925 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5926 | tlb_migrate_finish(p->mm); |
5927 | return 0; | |
5928 | } | |
5929 | out: | |
5930 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5931 | |
1da177e4 LT |
5932 | return ret; |
5933 | } | |
cd8ba7cd | 5934 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5935 | |
5936 | /* | |
41a2d6cf | 5937 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5938 | * this because either it can't run here any more (set_cpus_allowed() |
5939 | * away from this CPU, or CPU going down), or because we're | |
5940 | * attempting to rebalance this task on exec (sched_exec). | |
5941 | * | |
5942 | * So we race with normal scheduler movements, but that's OK, as long | |
5943 | * as the task is no longer on this CPU. | |
efc30814 KK |
5944 | * |
5945 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5946 | */ |
efc30814 | 5947 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5948 | { |
70b97a7f | 5949 | struct rq *rq_dest, *rq_src; |
e2912009 | 5950 | int ret = 0; |
1da177e4 | 5951 | |
e761b772 | 5952 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5953 | return ret; |
1da177e4 LT |
5954 | |
5955 | rq_src = cpu_rq(src_cpu); | |
5956 | rq_dest = cpu_rq(dest_cpu); | |
5957 | ||
5958 | double_rq_lock(rq_src, rq_dest); | |
5959 | /* Already moved. */ | |
5960 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5961 | goto done; |
1da177e4 | 5962 | /* Affinity changed (again). */ |
96f874e2 | 5963 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5964 | goto fail; |
1da177e4 | 5965 | |
e2912009 PZ |
5966 | /* |
5967 | * If we're not on a rq, the next wake-up will ensure we're | |
5968 | * placed properly. | |
5969 | */ | |
5970 | if (p->se.on_rq) { | |
2e1cb74a | 5971 | deactivate_task(rq_src, p, 0); |
e2912009 | 5972 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5973 | activate_task(rq_dest, p, 0); |
15afe09b | 5974 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5975 | } |
b1e38734 | 5976 | done: |
efc30814 | 5977 | ret = 1; |
b1e38734 | 5978 | fail: |
1da177e4 | 5979 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5980 | return ret; |
1da177e4 LT |
5981 | } |
5982 | ||
5983 | /* | |
969c7921 TH |
5984 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5985 | * and performs thread migration by bumping thread off CPU then | |
5986 | * 'pushing' onto another runqueue. | |
1da177e4 | 5987 | */ |
969c7921 | 5988 | static int migration_cpu_stop(void *data) |
1da177e4 | 5989 | { |
969c7921 | 5990 | struct migration_arg *arg = data; |
f7b4cddc | 5991 | |
969c7921 TH |
5992 | /* |
5993 | * The original target cpu might have gone down and we might | |
5994 | * be on another cpu but it doesn't matter. | |
5995 | */ | |
f7b4cddc | 5996 | local_irq_disable(); |
969c7921 | 5997 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5998 | local_irq_enable(); |
1da177e4 | 5999 | return 0; |
f7b4cddc ON |
6000 | } |
6001 | ||
1da177e4 | 6002 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6003 | |
054b9108 | 6004 | /* |
48c5ccae PZ |
6005 | * Ensures that the idle task is using init_mm right before its cpu goes |
6006 | * offline. | |
054b9108 | 6007 | */ |
48c5ccae | 6008 | void idle_task_exit(void) |
1da177e4 | 6009 | { |
48c5ccae | 6010 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6011 | |
48c5ccae | 6012 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6013 | |
48c5ccae PZ |
6014 | if (mm != &init_mm) |
6015 | switch_mm(mm, &init_mm, current); | |
6016 | mmdrop(mm); | |
1da177e4 LT |
6017 | } |
6018 | ||
6019 | /* | |
6020 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6021 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6022 | * for performance reasons the counter is not stricly tracking tasks to | |
6023 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6024 | * to keep the global sum constant after CPU-down: | |
6025 | */ | |
70b97a7f | 6026 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6027 | { |
6ad4c188 | 6028 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6029 | |
1da177e4 LT |
6030 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6031 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6032 | } |
6033 | ||
dd41f596 | 6034 | /* |
48c5ccae | 6035 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6036 | */ |
48c5ccae | 6037 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6038 | { |
48c5ccae PZ |
6039 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6040 | rq->calc_load_active = 0; | |
1da177e4 LT |
6041 | } |
6042 | ||
48f24c4d | 6043 | /* |
48c5ccae PZ |
6044 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6045 | * try_to_wake_up()->select_task_rq(). | |
6046 | * | |
6047 | * Called with rq->lock held even though we'er in stop_machine() and | |
6048 | * there's no concurrency possible, we hold the required locks anyway | |
6049 | * because of lock validation efforts. | |
1da177e4 | 6050 | */ |
48c5ccae | 6051 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6052 | { |
70b97a7f | 6053 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6054 | struct task_struct *next, *stop = rq->stop; |
6055 | int dest_cpu; | |
1da177e4 LT |
6056 | |
6057 | /* | |
48c5ccae PZ |
6058 | * Fudge the rq selection such that the below task selection loop |
6059 | * doesn't get stuck on the currently eligible stop task. | |
6060 | * | |
6061 | * We're currently inside stop_machine() and the rq is either stuck | |
6062 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6063 | * either way we should never end up calling schedule() until we're | |
6064 | * done here. | |
1da177e4 | 6065 | */ |
48c5ccae | 6066 | rq->stop = NULL; |
48f24c4d | 6067 | |
dd41f596 | 6068 | for ( ; ; ) { |
48c5ccae PZ |
6069 | /* |
6070 | * There's this thread running, bail when that's the only | |
6071 | * remaining thread. | |
6072 | */ | |
6073 | if (rq->nr_running == 1) | |
dd41f596 | 6074 | break; |
48c5ccae | 6075 | |
b67802ea | 6076 | next = pick_next_task(rq); |
48c5ccae | 6077 | BUG_ON(!next); |
79c53799 | 6078 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6079 | |
48c5ccae PZ |
6080 | /* Find suitable destination for @next, with force if needed. */ |
6081 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6082 | raw_spin_unlock(&rq->lock); | |
6083 | ||
6084 | __migrate_task(next, dead_cpu, dest_cpu); | |
6085 | ||
6086 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6087 | } |
dce48a84 | 6088 | |
48c5ccae | 6089 | rq->stop = stop; |
dce48a84 | 6090 | } |
48c5ccae | 6091 | |
1da177e4 LT |
6092 | #endif /* CONFIG_HOTPLUG_CPU */ |
6093 | ||
e692ab53 NP |
6094 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6095 | ||
6096 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6097 | { |
6098 | .procname = "sched_domain", | |
c57baf1e | 6099 | .mode = 0555, |
e0361851 | 6100 | }, |
56992309 | 6101 | {} |
e692ab53 NP |
6102 | }; |
6103 | ||
6104 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6105 | { |
6106 | .procname = "kernel", | |
c57baf1e | 6107 | .mode = 0555, |
e0361851 AD |
6108 | .child = sd_ctl_dir, |
6109 | }, | |
56992309 | 6110 | {} |
e692ab53 NP |
6111 | }; |
6112 | ||
6113 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6114 | { | |
6115 | struct ctl_table *entry = | |
5cf9f062 | 6116 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6117 | |
e692ab53 NP |
6118 | return entry; |
6119 | } | |
6120 | ||
6382bc90 MM |
6121 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6122 | { | |
cd790076 | 6123 | struct ctl_table *entry; |
6382bc90 | 6124 | |
cd790076 MM |
6125 | /* |
6126 | * In the intermediate directories, both the child directory and | |
6127 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6128 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6129 | * static strings and all have proc handlers. |
6130 | */ | |
6131 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6132 | if (entry->child) |
6133 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6134 | if (entry->proc_handler == NULL) |
6135 | kfree(entry->procname); | |
6136 | } | |
6382bc90 MM |
6137 | |
6138 | kfree(*tablep); | |
6139 | *tablep = NULL; | |
6140 | } | |
6141 | ||
e692ab53 | 6142 | static void |
e0361851 | 6143 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6144 | const char *procname, void *data, int maxlen, |
6145 | mode_t mode, proc_handler *proc_handler) | |
6146 | { | |
e692ab53 NP |
6147 | entry->procname = procname; |
6148 | entry->data = data; | |
6149 | entry->maxlen = maxlen; | |
6150 | entry->mode = mode; | |
6151 | entry->proc_handler = proc_handler; | |
6152 | } | |
6153 | ||
6154 | static struct ctl_table * | |
6155 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6156 | { | |
a5d8c348 | 6157 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6158 | |
ad1cdc1d MM |
6159 | if (table == NULL) |
6160 | return NULL; | |
6161 | ||
e0361851 | 6162 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6163 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6164 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6165 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6166 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6167 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6168 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6169 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6170 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6171 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6172 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6173 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6174 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6175 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6176 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6177 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6178 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6179 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6180 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6181 | &sd->cache_nice_tries, |
6182 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6183 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6184 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6185 | set_table_entry(&table[11], "name", sd->name, |
6186 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6187 | /* &table[12] is terminator */ | |
e692ab53 NP |
6188 | |
6189 | return table; | |
6190 | } | |
6191 | ||
9a4e7159 | 6192 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6193 | { |
6194 | struct ctl_table *entry, *table; | |
6195 | struct sched_domain *sd; | |
6196 | int domain_num = 0, i; | |
6197 | char buf[32]; | |
6198 | ||
6199 | for_each_domain(cpu, sd) | |
6200 | domain_num++; | |
6201 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6202 | if (table == NULL) |
6203 | return NULL; | |
e692ab53 NP |
6204 | |
6205 | i = 0; | |
6206 | for_each_domain(cpu, sd) { | |
6207 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6208 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6209 | entry->mode = 0555; |
e692ab53 NP |
6210 | entry->child = sd_alloc_ctl_domain_table(sd); |
6211 | entry++; | |
6212 | i++; | |
6213 | } | |
6214 | return table; | |
6215 | } | |
6216 | ||
6217 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6218 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6219 | { |
6ad4c188 | 6220 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6221 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6222 | char buf[32]; | |
6223 | ||
7378547f MM |
6224 | WARN_ON(sd_ctl_dir[0].child); |
6225 | sd_ctl_dir[0].child = entry; | |
6226 | ||
ad1cdc1d MM |
6227 | if (entry == NULL) |
6228 | return; | |
6229 | ||
6ad4c188 | 6230 | for_each_possible_cpu(i) { |
e692ab53 | 6231 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6232 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6233 | entry->mode = 0555; |
e692ab53 | 6234 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6235 | entry++; |
e692ab53 | 6236 | } |
7378547f MM |
6237 | |
6238 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6239 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6240 | } | |
6382bc90 | 6241 | |
7378547f | 6242 | /* may be called multiple times per register */ |
6382bc90 MM |
6243 | static void unregister_sched_domain_sysctl(void) |
6244 | { | |
7378547f MM |
6245 | if (sd_sysctl_header) |
6246 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6247 | sd_sysctl_header = NULL; |
7378547f MM |
6248 | if (sd_ctl_dir[0].child) |
6249 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6250 | } |
e692ab53 | 6251 | #else |
6382bc90 MM |
6252 | static void register_sched_domain_sysctl(void) |
6253 | { | |
6254 | } | |
6255 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6256 | { |
6257 | } | |
6258 | #endif | |
6259 | ||
1f11eb6a GH |
6260 | static void set_rq_online(struct rq *rq) |
6261 | { | |
6262 | if (!rq->online) { | |
6263 | const struct sched_class *class; | |
6264 | ||
c6c4927b | 6265 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6266 | rq->online = 1; |
6267 | ||
6268 | for_each_class(class) { | |
6269 | if (class->rq_online) | |
6270 | class->rq_online(rq); | |
6271 | } | |
6272 | } | |
6273 | } | |
6274 | ||
6275 | static void set_rq_offline(struct rq *rq) | |
6276 | { | |
6277 | if (rq->online) { | |
6278 | const struct sched_class *class; | |
6279 | ||
6280 | for_each_class(class) { | |
6281 | if (class->rq_offline) | |
6282 | class->rq_offline(rq); | |
6283 | } | |
6284 | ||
c6c4927b | 6285 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6286 | rq->online = 0; |
6287 | } | |
6288 | } | |
6289 | ||
1da177e4 LT |
6290 | /* |
6291 | * migration_call - callback that gets triggered when a CPU is added. | |
6292 | * Here we can start up the necessary migration thread for the new CPU. | |
6293 | */ | |
48f24c4d IM |
6294 | static int __cpuinit |
6295 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6296 | { |
48f24c4d | 6297 | int cpu = (long)hcpu; |
1da177e4 | 6298 | unsigned long flags; |
969c7921 | 6299 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6300 | |
48c5ccae | 6301 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6302 | |
1da177e4 | 6303 | case CPU_UP_PREPARE: |
a468d389 | 6304 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6305 | break; |
48f24c4d | 6306 | |
1da177e4 | 6307 | case CPU_ONLINE: |
1f94ef59 | 6308 | /* Update our root-domain */ |
05fa785c | 6309 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6310 | if (rq->rd) { |
c6c4927b | 6311 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6312 | |
6313 | set_rq_online(rq); | |
1f94ef59 | 6314 | } |
05fa785c | 6315 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6316 | break; |
48f24c4d | 6317 | |
1da177e4 | 6318 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6319 | case CPU_DYING: |
57d885fe | 6320 | /* Update our root-domain */ |
05fa785c | 6321 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6322 | if (rq->rd) { |
c6c4927b | 6323 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6324 | set_rq_offline(rq); |
57d885fe | 6325 | } |
48c5ccae PZ |
6326 | migrate_tasks(cpu); |
6327 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6328 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6329 | |
6330 | migrate_nr_uninterruptible(rq); | |
6331 | calc_global_load_remove(rq); | |
57d885fe | 6332 | break; |
1da177e4 LT |
6333 | #endif |
6334 | } | |
49c022e6 PZ |
6335 | |
6336 | update_max_interval(); | |
6337 | ||
1da177e4 LT |
6338 | return NOTIFY_OK; |
6339 | } | |
6340 | ||
f38b0820 PM |
6341 | /* |
6342 | * Register at high priority so that task migration (migrate_all_tasks) | |
6343 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6344 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6345 | */ |
26c2143b | 6346 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6347 | .notifier_call = migration_call, |
50a323b7 | 6348 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6349 | }; |
6350 | ||
3a101d05 TH |
6351 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6352 | unsigned long action, void *hcpu) | |
6353 | { | |
6354 | switch (action & ~CPU_TASKS_FROZEN) { | |
6355 | case CPU_ONLINE: | |
6356 | case CPU_DOWN_FAILED: | |
6357 | set_cpu_active((long)hcpu, true); | |
6358 | return NOTIFY_OK; | |
6359 | default: | |
6360 | return NOTIFY_DONE; | |
6361 | } | |
6362 | } | |
6363 | ||
6364 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6365 | unsigned long action, void *hcpu) | |
6366 | { | |
6367 | switch (action & ~CPU_TASKS_FROZEN) { | |
6368 | case CPU_DOWN_PREPARE: | |
6369 | set_cpu_active((long)hcpu, false); | |
6370 | return NOTIFY_OK; | |
6371 | default: | |
6372 | return NOTIFY_DONE; | |
6373 | } | |
6374 | } | |
6375 | ||
7babe8db | 6376 | static int __init migration_init(void) |
1da177e4 LT |
6377 | { |
6378 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6379 | int err; |
48f24c4d | 6380 | |
3a101d05 | 6381 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6382 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6383 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6384 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6385 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6386 | |
3a101d05 TH |
6387 | /* Register cpu active notifiers */ |
6388 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6389 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6390 | ||
a004cd42 | 6391 | return 0; |
1da177e4 | 6392 | } |
7babe8db | 6393 | early_initcall(migration_init); |
1da177e4 LT |
6394 | #endif |
6395 | ||
6396 | #ifdef CONFIG_SMP | |
476f3534 | 6397 | |
4cb98839 PZ |
6398 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
6399 | ||
3e9830dc | 6400 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6401 | |
f6630114 MT |
6402 | static __read_mostly int sched_domain_debug_enabled; |
6403 | ||
6404 | static int __init sched_domain_debug_setup(char *str) | |
6405 | { | |
6406 | sched_domain_debug_enabled = 1; | |
6407 | ||
6408 | return 0; | |
6409 | } | |
6410 | early_param("sched_debug", sched_domain_debug_setup); | |
6411 | ||
7c16ec58 | 6412 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6413 | struct cpumask *groupmask) |
1da177e4 | 6414 | { |
4dcf6aff | 6415 | struct sched_group *group = sd->groups; |
434d53b0 | 6416 | char str[256]; |
1da177e4 | 6417 | |
968ea6d8 | 6418 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6419 | cpumask_clear(groupmask); |
4dcf6aff IM |
6420 | |
6421 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6422 | ||
6423 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6424 | printk("does not load-balance\n"); |
4dcf6aff | 6425 | if (sd->parent) |
3df0fc5b PZ |
6426 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6427 | " has parent"); | |
4dcf6aff | 6428 | return -1; |
41c7ce9a NP |
6429 | } |
6430 | ||
3df0fc5b | 6431 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6432 | |
758b2cdc | 6433 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6434 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6435 | "CPU%d\n", cpu); | |
4dcf6aff | 6436 | } |
758b2cdc | 6437 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6438 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6439 | " CPU%d\n", cpu); | |
4dcf6aff | 6440 | } |
1da177e4 | 6441 | |
4dcf6aff | 6442 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6443 | do { |
4dcf6aff | 6444 | if (!group) { |
3df0fc5b PZ |
6445 | printk("\n"); |
6446 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6447 | break; |
6448 | } | |
6449 | ||
18a3885f | 6450 | if (!group->cpu_power) { |
3df0fc5b PZ |
6451 | printk(KERN_CONT "\n"); |
6452 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6453 | "set\n"); | |
4dcf6aff IM |
6454 | break; |
6455 | } | |
1da177e4 | 6456 | |
758b2cdc | 6457 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6458 | printk(KERN_CONT "\n"); |
6459 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6460 | break; |
6461 | } | |
1da177e4 | 6462 | |
758b2cdc | 6463 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6464 | printk(KERN_CONT "\n"); |
6465 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6466 | break; |
6467 | } | |
1da177e4 | 6468 | |
758b2cdc | 6469 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6470 | |
968ea6d8 | 6471 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6472 | |
3df0fc5b | 6473 | printk(KERN_CONT " %s", str); |
18a3885f | 6474 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6475 | printk(KERN_CONT " (cpu_power = %d)", |
6476 | group->cpu_power); | |
381512cf | 6477 | } |
1da177e4 | 6478 | |
4dcf6aff IM |
6479 | group = group->next; |
6480 | } while (group != sd->groups); | |
3df0fc5b | 6481 | printk(KERN_CONT "\n"); |
1da177e4 | 6482 | |
758b2cdc | 6483 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6484 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6485 | |
758b2cdc RR |
6486 | if (sd->parent && |
6487 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6488 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6489 | "of domain->span\n"); | |
4dcf6aff IM |
6490 | return 0; |
6491 | } | |
1da177e4 | 6492 | |
4dcf6aff IM |
6493 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6494 | { | |
6495 | int level = 0; | |
1da177e4 | 6496 | |
f6630114 MT |
6497 | if (!sched_domain_debug_enabled) |
6498 | return; | |
6499 | ||
4dcf6aff IM |
6500 | if (!sd) { |
6501 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6502 | return; | |
6503 | } | |
1da177e4 | 6504 | |
4dcf6aff IM |
6505 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6506 | ||
6507 | for (;;) { | |
4cb98839 | 6508 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 6509 | break; |
1da177e4 LT |
6510 | level++; |
6511 | sd = sd->parent; | |
33859f7f | 6512 | if (!sd) |
4dcf6aff IM |
6513 | break; |
6514 | } | |
1da177e4 | 6515 | } |
6d6bc0ad | 6516 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6517 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6518 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6519 | |
1a20ff27 | 6520 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6521 | { |
758b2cdc | 6522 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6523 | return 1; |
6524 | ||
6525 | /* Following flags need at least 2 groups */ | |
6526 | if (sd->flags & (SD_LOAD_BALANCE | | |
6527 | SD_BALANCE_NEWIDLE | | |
6528 | SD_BALANCE_FORK | | |
89c4710e SS |
6529 | SD_BALANCE_EXEC | |
6530 | SD_SHARE_CPUPOWER | | |
6531 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6532 | if (sd->groups != sd->groups->next) |
6533 | return 0; | |
6534 | } | |
6535 | ||
6536 | /* Following flags don't use groups */ | |
c88d5910 | 6537 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6538 | return 0; |
6539 | ||
6540 | return 1; | |
6541 | } | |
6542 | ||
48f24c4d IM |
6543 | static int |
6544 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6545 | { |
6546 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6547 | ||
6548 | if (sd_degenerate(parent)) | |
6549 | return 1; | |
6550 | ||
758b2cdc | 6551 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6552 | return 0; |
6553 | ||
245af2c7 SS |
6554 | /* Flags needing groups don't count if only 1 group in parent */ |
6555 | if (parent->groups == parent->groups->next) { | |
6556 | pflags &= ~(SD_LOAD_BALANCE | | |
6557 | SD_BALANCE_NEWIDLE | | |
6558 | SD_BALANCE_FORK | | |
89c4710e SS |
6559 | SD_BALANCE_EXEC | |
6560 | SD_SHARE_CPUPOWER | | |
6561 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6562 | if (nr_node_ids == 1) |
6563 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6564 | } |
6565 | if (~cflags & pflags) | |
6566 | return 0; | |
6567 | ||
6568 | return 1; | |
6569 | } | |
6570 | ||
dce840a0 | 6571 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 6572 | { |
dce840a0 | 6573 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 6574 | |
68e74568 | 6575 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
6576 | free_cpumask_var(rd->rto_mask); |
6577 | free_cpumask_var(rd->online); | |
6578 | free_cpumask_var(rd->span); | |
6579 | kfree(rd); | |
6580 | } | |
6581 | ||
57d885fe GH |
6582 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6583 | { | |
a0490fa3 | 6584 | struct root_domain *old_rd = NULL; |
57d885fe | 6585 | unsigned long flags; |
57d885fe | 6586 | |
05fa785c | 6587 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6588 | |
6589 | if (rq->rd) { | |
a0490fa3 | 6590 | old_rd = rq->rd; |
57d885fe | 6591 | |
c6c4927b | 6592 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6593 | set_rq_offline(rq); |
57d885fe | 6594 | |
c6c4927b | 6595 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6596 | |
a0490fa3 IM |
6597 | /* |
6598 | * If we dont want to free the old_rt yet then | |
6599 | * set old_rd to NULL to skip the freeing later | |
6600 | * in this function: | |
6601 | */ | |
6602 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6603 | old_rd = NULL; | |
57d885fe GH |
6604 | } |
6605 | ||
6606 | atomic_inc(&rd->refcount); | |
6607 | rq->rd = rd; | |
6608 | ||
c6c4927b | 6609 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6610 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6611 | set_rq_online(rq); |
57d885fe | 6612 | |
05fa785c | 6613 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6614 | |
6615 | if (old_rd) | |
dce840a0 | 6616 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
6617 | } |
6618 | ||
68c38fc3 | 6619 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6620 | { |
6621 | memset(rd, 0, sizeof(*rd)); | |
6622 | ||
68c38fc3 | 6623 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6624 | goto out; |
68c38fc3 | 6625 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6626 | goto free_span; |
68c38fc3 | 6627 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6628 | goto free_online; |
6e0534f2 | 6629 | |
68c38fc3 | 6630 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6631 | goto free_rto_mask; |
c6c4927b | 6632 | return 0; |
6e0534f2 | 6633 | |
68e74568 RR |
6634 | free_rto_mask: |
6635 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6636 | free_online: |
6637 | free_cpumask_var(rd->online); | |
6638 | free_span: | |
6639 | free_cpumask_var(rd->span); | |
0c910d28 | 6640 | out: |
c6c4927b | 6641 | return -ENOMEM; |
57d885fe GH |
6642 | } |
6643 | ||
6644 | static void init_defrootdomain(void) | |
6645 | { | |
68c38fc3 | 6646 | init_rootdomain(&def_root_domain); |
c6c4927b | 6647 | |
57d885fe GH |
6648 | atomic_set(&def_root_domain.refcount, 1); |
6649 | } | |
6650 | ||
dc938520 | 6651 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6652 | { |
6653 | struct root_domain *rd; | |
6654 | ||
6655 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6656 | if (!rd) | |
6657 | return NULL; | |
6658 | ||
68c38fc3 | 6659 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6660 | kfree(rd); |
6661 | return NULL; | |
6662 | } | |
57d885fe GH |
6663 | |
6664 | return rd; | |
6665 | } | |
6666 | ||
dce840a0 PZ |
6667 | static void free_sched_domain(struct rcu_head *rcu) |
6668 | { | |
6669 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
6670 | if (atomic_dec_and_test(&sd->groups->ref)) | |
6671 | kfree(sd->groups); | |
6672 | kfree(sd); | |
6673 | } | |
6674 | ||
6675 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
6676 | { | |
6677 | call_rcu(&sd->rcu, free_sched_domain); | |
6678 | } | |
6679 | ||
6680 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
6681 | { | |
6682 | for (; sd; sd = sd->parent) | |
6683 | destroy_sched_domain(sd, cpu); | |
6684 | } | |
6685 | ||
1da177e4 | 6686 | /* |
0eab9146 | 6687 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6688 | * hold the hotplug lock. |
6689 | */ | |
0eab9146 IM |
6690 | static void |
6691 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6692 | { |
70b97a7f | 6693 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6694 | struct sched_domain *tmp; |
6695 | ||
6696 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6697 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6698 | struct sched_domain *parent = tmp->parent; |
6699 | if (!parent) | |
6700 | break; | |
f29c9b1c | 6701 | |
1a848870 | 6702 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6703 | tmp->parent = parent->parent; |
1a848870 SS |
6704 | if (parent->parent) |
6705 | parent->parent->child = tmp; | |
dce840a0 | 6706 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6707 | } else |
6708 | tmp = tmp->parent; | |
245af2c7 SS |
6709 | } |
6710 | ||
1a848870 | 6711 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6712 | tmp = sd; |
245af2c7 | 6713 | sd = sd->parent; |
dce840a0 | 6714 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6715 | if (sd) |
6716 | sd->child = NULL; | |
6717 | } | |
1da177e4 | 6718 | |
4cb98839 | 6719 | sched_domain_debug(sd, cpu); |
1da177e4 | 6720 | |
57d885fe | 6721 | rq_attach_root(rq, rd); |
dce840a0 | 6722 | tmp = rq->sd; |
674311d5 | 6723 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6724 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
6725 | } |
6726 | ||
6727 | /* cpus with isolated domains */ | |
dcc30a35 | 6728 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6729 | |
6730 | /* Setup the mask of cpus configured for isolated domains */ | |
6731 | static int __init isolated_cpu_setup(char *str) | |
6732 | { | |
bdddd296 | 6733 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6734 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6735 | return 1; |
6736 | } | |
6737 | ||
8927f494 | 6738 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6739 | |
9c1cfda2 | 6740 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6741 | |
9c1cfda2 | 6742 | #ifdef CONFIG_NUMA |
198e2f18 | 6743 | |
9c1cfda2 JH |
6744 | /** |
6745 | * find_next_best_node - find the next node to include in a sched_domain | |
6746 | * @node: node whose sched_domain we're building | |
6747 | * @used_nodes: nodes already in the sched_domain | |
6748 | * | |
41a2d6cf | 6749 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6750 | * finds the closest node not already in the @used_nodes map. |
6751 | * | |
6752 | * Should use nodemask_t. | |
6753 | */ | |
c5f59f08 | 6754 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6755 | { |
6756 | int i, n, val, min_val, best_node = 0; | |
6757 | ||
6758 | min_val = INT_MAX; | |
6759 | ||
076ac2af | 6760 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6761 | /* Start at @node */ |
076ac2af | 6762 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6763 | |
6764 | if (!nr_cpus_node(n)) | |
6765 | continue; | |
6766 | ||
6767 | /* Skip already used nodes */ | |
c5f59f08 | 6768 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6769 | continue; |
6770 | ||
6771 | /* Simple min distance search */ | |
6772 | val = node_distance(node, n); | |
6773 | ||
6774 | if (val < min_val) { | |
6775 | min_val = val; | |
6776 | best_node = n; | |
6777 | } | |
6778 | } | |
6779 | ||
c5f59f08 | 6780 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6781 | return best_node; |
6782 | } | |
6783 | ||
6784 | /** | |
6785 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6786 | * @node: node whose cpumask we're constructing | |
73486722 | 6787 | * @span: resulting cpumask |
9c1cfda2 | 6788 | * |
41a2d6cf | 6789 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6790 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6791 | * out optimally. | |
6792 | */ | |
96f874e2 | 6793 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6794 | { |
c5f59f08 | 6795 | nodemask_t used_nodes; |
48f24c4d | 6796 | int i; |
9c1cfda2 | 6797 | |
6ca09dfc | 6798 | cpumask_clear(span); |
c5f59f08 | 6799 | nodes_clear(used_nodes); |
9c1cfda2 | 6800 | |
6ca09dfc | 6801 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6802 | node_set(node, used_nodes); |
9c1cfda2 JH |
6803 | |
6804 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6805 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6806 | |
6ca09dfc | 6807 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6808 | } |
9c1cfda2 | 6809 | } |
d3081f52 PZ |
6810 | |
6811 | static const struct cpumask *cpu_node_mask(int cpu) | |
6812 | { | |
6813 | lockdep_assert_held(&sched_domains_mutex); | |
6814 | ||
6815 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
6816 | ||
6817 | return sched_domains_tmpmask; | |
6818 | } | |
6d6bc0ad | 6819 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6820 | |
d3081f52 PZ |
6821 | static const struct cpumask *cpu_cpu_mask(int cpu) |
6822 | { | |
6823 | return cpumask_of_node(cpu_to_node(cpu)); | |
6824 | } | |
6825 | ||
5c45bf27 | 6826 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6827 | |
dce840a0 PZ |
6828 | struct sd_data { |
6829 | struct sched_domain **__percpu sd; | |
6830 | struct sched_group **__percpu sg; | |
6831 | }; | |
6832 | ||
49a02c51 | 6833 | struct s_data { |
21d42ccf | 6834 | struct sched_domain ** __percpu sd; |
dce840a0 | 6835 | struct sd_data sdd[SD_LV_MAX]; |
49a02c51 AH |
6836 | struct root_domain *rd; |
6837 | }; | |
6838 | ||
2109b99e | 6839 | enum s_alloc { |
2109b99e | 6840 | sa_rootdomain, |
21d42ccf | 6841 | sa_sd, |
dce840a0 | 6842 | sa_sd_storage, |
2109b99e AH |
6843 | sa_none, |
6844 | }; | |
6845 | ||
9c1cfda2 | 6846 | /* |
dce840a0 | 6847 | * Assumes the sched_domain tree is fully constructed |
9c1cfda2 | 6848 | */ |
dce840a0 | 6849 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6850 | { |
dce840a0 PZ |
6851 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6852 | struct sched_domain *child = sd->child; | |
1da177e4 | 6853 | |
dce840a0 PZ |
6854 | if (child) |
6855 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6856 | |
6711cab4 | 6857 | if (sg) |
dce840a0 PZ |
6858 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
6859 | ||
6860 | return cpu; | |
1e9f28fa | 6861 | } |
1e9f28fa | 6862 | |
01a08546 | 6863 | /* |
dce840a0 PZ |
6864 | * build_sched_groups takes the cpumask we wish to span, and a pointer |
6865 | * to a function which identifies what group(along with sched group) a CPU | |
6866 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids | |
6867 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
6868 | * | |
6869 | * build_sched_groups will build a circular linked list of the groups | |
6870 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6871 | * and ->cpu_power to 0. | |
01a08546 | 6872 | */ |
dce840a0 | 6873 | static void |
f96225fd | 6874 | build_sched_groups(struct sched_domain *sd) |
1da177e4 | 6875 | { |
dce840a0 PZ |
6876 | struct sched_group *first = NULL, *last = NULL; |
6877 | struct sd_data *sdd = sd->private; | |
6878 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6879 | struct cpumask *covered; |
dce840a0 | 6880 | int i; |
9c1cfda2 | 6881 | |
f96225fd PZ |
6882 | lockdep_assert_held(&sched_domains_mutex); |
6883 | covered = sched_domains_tmpmask; | |
6884 | ||
dce840a0 | 6885 | cpumask_clear(covered); |
6711cab4 | 6886 | |
dce840a0 PZ |
6887 | for_each_cpu(i, span) { |
6888 | struct sched_group *sg; | |
6889 | int group = get_group(i, sdd, &sg); | |
6890 | int j; | |
6711cab4 | 6891 | |
dce840a0 PZ |
6892 | if (cpumask_test_cpu(i, covered)) |
6893 | continue; | |
6711cab4 | 6894 | |
dce840a0 PZ |
6895 | cpumask_clear(sched_group_cpus(sg)); |
6896 | sg->cpu_power = 0; | |
0601a88d | 6897 | |
dce840a0 PZ |
6898 | for_each_cpu(j, span) { |
6899 | if (get_group(j, sdd, NULL) != group) | |
6900 | continue; | |
0601a88d | 6901 | |
dce840a0 PZ |
6902 | cpumask_set_cpu(j, covered); |
6903 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6904 | } | |
0601a88d | 6905 | |
dce840a0 PZ |
6906 | if (!first) |
6907 | first = sg; | |
6908 | if (last) | |
6909 | last->next = sg; | |
6910 | last = sg; | |
6911 | } | |
6912 | last->next = first; | |
0601a88d | 6913 | } |
51888ca2 | 6914 | |
89c4710e SS |
6915 | /* |
6916 | * Initialize sched groups cpu_power. | |
6917 | * | |
6918 | * cpu_power indicates the capacity of sched group, which is used while | |
6919 | * distributing the load between different sched groups in a sched domain. | |
6920 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6921 | * there are asymmetries in the topology. If there are asymmetries, group | |
6922 | * having more cpu_power will pickup more load compared to the group having | |
6923 | * less cpu_power. | |
89c4710e SS |
6924 | */ |
6925 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6926 | { | |
89c4710e SS |
6927 | WARN_ON(!sd || !sd->groups); |
6928 | ||
13318a71 | 6929 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6930 | return; |
6931 | ||
aae6d3dd SS |
6932 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
6933 | ||
d274cb30 | 6934 | update_group_power(sd, cpu); |
89c4710e SS |
6935 | } |
6936 | ||
7c16ec58 MT |
6937 | /* |
6938 | * Initializers for schedule domains | |
6939 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6940 | */ | |
6941 | ||
a5d8c348 IM |
6942 | #ifdef CONFIG_SCHED_DEBUG |
6943 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6944 | #else | |
6945 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6946 | #endif | |
6947 | ||
dce840a0 PZ |
6948 | #define SD_INIT_FUNC(type) \ |
6949 | static noinline struct sched_domain *sd_init_##type(struct s_data *d, int cpu) \ | |
6950 | { \ | |
6951 | struct sched_domain *sd = *per_cpu_ptr(d->sdd[SD_LV_##type].sd, cpu); \ | |
6952 | *sd = SD_##type##_INIT; \ | |
6953 | sd->level = SD_LV_##type; \ | |
6954 | SD_INIT_NAME(sd, type); \ | |
6955 | sd->private = &d->sdd[SD_LV_##type]; \ | |
6956 | return sd; \ | |
7c16ec58 MT |
6957 | } |
6958 | ||
6959 | SD_INIT_FUNC(CPU) | |
6960 | #ifdef CONFIG_NUMA | |
6961 | SD_INIT_FUNC(ALLNODES) | |
6962 | SD_INIT_FUNC(NODE) | |
6963 | #endif | |
6964 | #ifdef CONFIG_SCHED_SMT | |
6965 | SD_INIT_FUNC(SIBLING) | |
6966 | #endif | |
6967 | #ifdef CONFIG_SCHED_MC | |
6968 | SD_INIT_FUNC(MC) | |
6969 | #endif | |
01a08546 HC |
6970 | #ifdef CONFIG_SCHED_BOOK |
6971 | SD_INIT_FUNC(BOOK) | |
6972 | #endif | |
7c16ec58 | 6973 | |
1d3504fc HS |
6974 | static int default_relax_domain_level = -1; |
6975 | ||
6976 | static int __init setup_relax_domain_level(char *str) | |
6977 | { | |
30e0e178 LZ |
6978 | unsigned long val; |
6979 | ||
6980 | val = simple_strtoul(str, NULL, 0); | |
6981 | if (val < SD_LV_MAX) | |
6982 | default_relax_domain_level = val; | |
6983 | ||
1d3504fc HS |
6984 | return 1; |
6985 | } | |
6986 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6987 | ||
6988 | static void set_domain_attribute(struct sched_domain *sd, | |
6989 | struct sched_domain_attr *attr) | |
6990 | { | |
6991 | int request; | |
6992 | ||
6993 | if (!attr || attr->relax_domain_level < 0) { | |
6994 | if (default_relax_domain_level < 0) | |
6995 | return; | |
6996 | else | |
6997 | request = default_relax_domain_level; | |
6998 | } else | |
6999 | request = attr->relax_domain_level; | |
7000 | if (request < sd->level) { | |
7001 | /* turn off idle balance on this domain */ | |
c88d5910 | 7002 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7003 | } else { |
7004 | /* turn on idle balance on this domain */ | |
c88d5910 | 7005 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7006 | } |
7007 | } | |
7008 | ||
2109b99e AH |
7009 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7010 | const struct cpumask *cpu_map) | |
7011 | { | |
dce840a0 PZ |
7012 | int i, j; |
7013 | ||
2109b99e | 7014 | switch (what) { |
2109b99e | 7015 | case sa_rootdomain: |
822ff793 PZ |
7016 | if (!atomic_read(&d->rd->refcount)) |
7017 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
7018 | case sa_sd: |
7019 | free_percpu(d->sd); /* fall through */ | |
dce840a0 PZ |
7020 | case sa_sd_storage: |
7021 | for (i = 0; i < SD_LV_MAX; i++) { | |
7022 | for_each_cpu(j, cpu_map) { | |
7023 | kfree(*per_cpu_ptr(d->sdd[i].sd, j)); | |
7024 | kfree(*per_cpu_ptr(d->sdd[i].sg, j)); | |
7025 | } | |
7026 | free_percpu(d->sdd[i].sd); | |
7027 | free_percpu(d->sdd[i].sg); | |
7028 | } /* fall through */ | |
2109b99e AH |
7029 | case sa_none: |
7030 | break; | |
7031 | } | |
7032 | } | |
3404c8d9 | 7033 | |
2109b99e AH |
7034 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7035 | const struct cpumask *cpu_map) | |
7036 | { | |
dce840a0 PZ |
7037 | int i, j; |
7038 | ||
7039 | memset(d, 0, sizeof(*d)); | |
7040 | ||
dce840a0 PZ |
7041 | for (i = 0; i < SD_LV_MAX; i++) { |
7042 | d->sdd[i].sd = alloc_percpu(struct sched_domain *); | |
7043 | if (!d->sdd[i].sd) | |
7044 | return sa_sd_storage; | |
7045 | ||
7046 | d->sdd[i].sg = alloc_percpu(struct sched_group *); | |
7047 | if (!d->sdd[i].sg) | |
7048 | return sa_sd_storage; | |
7049 | ||
7050 | for_each_cpu(j, cpu_map) { | |
7051 | struct sched_domain *sd; | |
7052 | struct sched_group *sg; | |
7053 | ||
7054 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
7055 | GFP_KERNEL, cpu_to_node(j)); | |
7056 | if (!sd) | |
7057 | return sa_sd_storage; | |
7058 | ||
7059 | *per_cpu_ptr(d->sdd[i].sd, j) = sd; | |
7060 | ||
7061 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7062 | GFP_KERNEL, cpu_to_node(j)); | |
7063 | if (!sg) | |
7064 | return sa_sd_storage; | |
7065 | ||
7066 | *per_cpu_ptr(d->sdd[i].sg, j) = sg; | |
7067 | } | |
21d42ccf | 7068 | } |
dce840a0 PZ |
7069 | d->sd = alloc_percpu(struct sched_domain *); |
7070 | if (!d->sd) | |
7071 | return sa_sd_storage; | |
2109b99e | 7072 | d->rd = alloc_rootdomain(); |
dce840a0 | 7073 | if (!d->rd) |
21d42ccf | 7074 | return sa_sd; |
2109b99e AH |
7075 | return sa_rootdomain; |
7076 | } | |
57d885fe | 7077 | |
dce840a0 PZ |
7078 | /* |
7079 | * NULL the sd_data elements we've used to build the sched_domain and | |
7080 | * sched_group structure so that the subsequent __free_domain_allocs() | |
7081 | * will not free the data we're using. | |
7082 | */ | |
7083 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
7084 | { | |
7085 | struct sd_data *sdd = sd->private; | |
7086 | struct sched_group *sg = sd->groups; | |
7087 | ||
7088 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
7089 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
7090 | ||
7091 | if (cpu == cpumask_first(sched_group_cpus(sg))) { | |
7092 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sg, cpu) != sg); | |
7093 | *per_cpu_ptr(sdd->sg, cpu) = NULL; | |
7094 | } | |
7095 | } | |
7096 | ||
3bd65a80 PZ |
7097 | static struct sched_domain *__build_allnodes_sched_domain(struct s_data *d, |
7098 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7099 | struct sched_domain *parent, int i) | |
7f4588f3 AH |
7100 | { |
7101 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7102 | #ifdef CONFIG_NUMA |
3bd65a80 PZ |
7103 | sd = sd_init_ALLNODES(d, i); |
7104 | set_domain_attribute(sd, attr); | |
d3081f52 | 7105 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_possible_mask); |
3bd65a80 PZ |
7106 | sd->parent = parent; |
7107 | if (parent) | |
7108 | parent->child = sd; | |
7109 | #endif | |
7110 | return sd; | |
7111 | } | |
7f4588f3 | 7112 | |
3bd65a80 PZ |
7113 | static struct sched_domain *__build_node_sched_domain(struct s_data *d, |
7114 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7115 | struct sched_domain *parent, int i) | |
7116 | { | |
7117 | struct sched_domain *sd = NULL; | |
7118 | #ifdef CONFIG_NUMA | |
dce840a0 | 7119 | sd = sd_init_NODE(d, i); |
7f4588f3 | 7120 | set_domain_attribute(sd, attr); |
d3081f52 | 7121 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_node_mask(i)); |
7f4588f3 AH |
7122 | sd->parent = parent; |
7123 | if (parent) | |
7124 | parent->child = sd; | |
1da177e4 | 7125 | #endif |
7f4588f3 AH |
7126 | return sd; |
7127 | } | |
1da177e4 | 7128 | |
87cce662 AH |
7129 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7130 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7131 | struct sched_domain *parent, int i) | |
7132 | { | |
7133 | struct sched_domain *sd; | |
dce840a0 | 7134 | sd = sd_init_CPU(d, i); |
87cce662 | 7135 | set_domain_attribute(sd, attr); |
d3081f52 | 7136 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_cpu_mask(i)); |
87cce662 AH |
7137 | sd->parent = parent; |
7138 | if (parent) | |
7139 | parent->child = sd; | |
87cce662 AH |
7140 | return sd; |
7141 | } | |
1da177e4 | 7142 | |
01a08546 HC |
7143 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7144 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7145 | struct sched_domain *parent, int i) | |
7146 | { | |
7147 | struct sched_domain *sd = parent; | |
7148 | #ifdef CONFIG_SCHED_BOOK | |
dce840a0 | 7149 | sd = sd_init_BOOK(d, i); |
01a08546 HC |
7150 | set_domain_attribute(sd, attr); |
7151 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7152 | sd->parent = parent; | |
7153 | parent->child = sd; | |
01a08546 HC |
7154 | #endif |
7155 | return sd; | |
7156 | } | |
7157 | ||
410c4081 AH |
7158 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7159 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7160 | struct sched_domain *parent, int i) | |
7161 | { | |
7162 | struct sched_domain *sd = parent; | |
1e9f28fa | 7163 | #ifdef CONFIG_SCHED_MC |
dce840a0 | 7164 | sd = sd_init_MC(d, i); |
410c4081 AH |
7165 | set_domain_attribute(sd, attr); |
7166 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7167 | sd->parent = parent; | |
7168 | parent->child = sd; | |
1e9f28fa | 7169 | #endif |
410c4081 AH |
7170 | return sd; |
7171 | } | |
1e9f28fa | 7172 | |
d8173535 AH |
7173 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7174 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7175 | struct sched_domain *parent, int i) | |
7176 | { | |
7177 | struct sched_domain *sd = parent; | |
1da177e4 | 7178 | #ifdef CONFIG_SCHED_SMT |
dce840a0 | 7179 | sd = sd_init_SIBLING(d, i); |
d8173535 AH |
7180 | set_domain_attribute(sd, attr); |
7181 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7182 | sd->parent = parent; | |
7183 | parent->child = sd; | |
1da177e4 | 7184 | #endif |
d8173535 AH |
7185 | return sd; |
7186 | } | |
1da177e4 | 7187 | |
2109b99e AH |
7188 | /* |
7189 | * Build sched domains for a given set of cpus and attach the sched domains | |
7190 | * to the individual cpus | |
7191 | */ | |
dce840a0 PZ |
7192 | static int build_sched_domains(const struct cpumask *cpu_map, |
7193 | struct sched_domain_attr *attr) | |
2109b99e AH |
7194 | { |
7195 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 7196 | struct sched_domain *sd; |
2109b99e | 7197 | struct s_data d; |
822ff793 | 7198 | int i, ret = -ENOMEM; |
9c1cfda2 | 7199 | |
2109b99e AH |
7200 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7201 | if (alloc_state != sa_rootdomain) | |
7202 | goto error; | |
9c1cfda2 | 7203 | |
dce840a0 | 7204 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7205 | for_each_cpu(i, cpu_map) { |
3bd65a80 PZ |
7206 | sd = NULL; |
7207 | sd = __build_allnodes_sched_domain(&d, cpu_map, attr, sd, i); | |
7208 | sd = __build_node_sched_domain(&d, cpu_map, attr, sd, i); | |
87cce662 | 7209 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7210 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7211 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7212 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
d274cb30 | 7213 | |
21d42ccf | 7214 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
7215 | } |
7216 | ||
7217 | /* Build the groups for the domains */ | |
7218 | for_each_cpu(i, cpu_map) { | |
7219 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7220 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
7221 | get_group(i, sd->private, &sd->groups); | |
7222 | atomic_inc(&sd->groups->ref); | |
21d42ccf | 7223 | |
dce840a0 PZ |
7224 | if (i != cpumask_first(sched_domain_span(sd))) |
7225 | continue; | |
7226 | ||
f96225fd | 7227 | build_sched_groups(sd); |
1cf51902 | 7228 | } |
a06dadbe | 7229 | } |
9c1cfda2 | 7230 | |
1da177e4 | 7231 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
7232 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7233 | if (!cpumask_test_cpu(i, cpu_map)) | |
7234 | continue; | |
9c1cfda2 | 7235 | |
dce840a0 PZ |
7236 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7237 | claim_allocations(i, sd); | |
cd4ea6ae | 7238 | init_sched_groups_power(i, sd); |
dce840a0 | 7239 | } |
f712c0c7 | 7240 | } |
9c1cfda2 | 7241 | |
1da177e4 | 7242 | /* Attach the domains */ |
dce840a0 | 7243 | rcu_read_lock(); |
abcd083a | 7244 | for_each_cpu(i, cpu_map) { |
21d42ccf | 7245 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 7246 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7247 | } |
dce840a0 | 7248 | rcu_read_unlock(); |
51888ca2 | 7249 | |
822ff793 | 7250 | ret = 0; |
51888ca2 | 7251 | error: |
2109b99e | 7252 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7253 | return ret; |
1da177e4 | 7254 | } |
029190c5 | 7255 | |
acc3f5d7 | 7256 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7257 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7258 | static struct sched_domain_attr *dattr_cur; |
7259 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7260 | |
7261 | /* | |
7262 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7263 | * cpumask) fails, then fallback to a single sched domain, |
7264 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7265 | */ |
4212823f | 7266 | static cpumask_var_t fallback_doms; |
029190c5 | 7267 | |
ee79d1bd HC |
7268 | /* |
7269 | * arch_update_cpu_topology lets virtualized architectures update the | |
7270 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7271 | * or 0 if it stayed the same. | |
7272 | */ | |
7273 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7274 | { |
ee79d1bd | 7275 | return 0; |
22e52b07 HC |
7276 | } |
7277 | ||
acc3f5d7 RR |
7278 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7279 | { | |
7280 | int i; | |
7281 | cpumask_var_t *doms; | |
7282 | ||
7283 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7284 | if (!doms) | |
7285 | return NULL; | |
7286 | for (i = 0; i < ndoms; i++) { | |
7287 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7288 | free_sched_domains(doms, i); | |
7289 | return NULL; | |
7290 | } | |
7291 | } | |
7292 | return doms; | |
7293 | } | |
7294 | ||
7295 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7296 | { | |
7297 | unsigned int i; | |
7298 | for (i = 0; i < ndoms; i++) | |
7299 | free_cpumask_var(doms[i]); | |
7300 | kfree(doms); | |
7301 | } | |
7302 | ||
1a20ff27 | 7303 | /* |
41a2d6cf | 7304 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7305 | * For now this just excludes isolated cpus, but could be used to |
7306 | * exclude other special cases in the future. | |
1a20ff27 | 7307 | */ |
c4a8849a | 7308 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7309 | { |
7378547f MM |
7310 | int err; |
7311 | ||
22e52b07 | 7312 | arch_update_cpu_topology(); |
029190c5 | 7313 | ndoms_cur = 1; |
acc3f5d7 | 7314 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7315 | if (!doms_cur) |
acc3f5d7 RR |
7316 | doms_cur = &fallback_doms; |
7317 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7318 | dattr_cur = NULL; |
dce840a0 | 7319 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7320 | register_sched_domain_sysctl(); |
7378547f MM |
7321 | |
7322 | return err; | |
1a20ff27 DG |
7323 | } |
7324 | ||
1a20ff27 DG |
7325 | /* |
7326 | * Detach sched domains from a group of cpus specified in cpu_map | |
7327 | * These cpus will now be attached to the NULL domain | |
7328 | */ | |
96f874e2 | 7329 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7330 | { |
7331 | int i; | |
7332 | ||
dce840a0 | 7333 | rcu_read_lock(); |
abcd083a | 7334 | for_each_cpu(i, cpu_map) |
57d885fe | 7335 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7336 | rcu_read_unlock(); |
1a20ff27 DG |
7337 | } |
7338 | ||
1d3504fc HS |
7339 | /* handle null as "default" */ |
7340 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7341 | struct sched_domain_attr *new, int idx_new) | |
7342 | { | |
7343 | struct sched_domain_attr tmp; | |
7344 | ||
7345 | /* fast path */ | |
7346 | if (!new && !cur) | |
7347 | return 1; | |
7348 | ||
7349 | tmp = SD_ATTR_INIT; | |
7350 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7351 | new ? (new + idx_new) : &tmp, | |
7352 | sizeof(struct sched_domain_attr)); | |
7353 | } | |
7354 | ||
029190c5 PJ |
7355 | /* |
7356 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7357 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7358 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7359 | * It destroys each deleted domain and builds each new domain. | |
7360 | * | |
acc3f5d7 | 7361 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7362 | * The masks don't intersect (don't overlap.) We should setup one |
7363 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7364 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7365 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7366 | * it as it is. | |
7367 | * | |
acc3f5d7 RR |
7368 | * The passed in 'doms_new' should be allocated using |
7369 | * alloc_sched_domains. This routine takes ownership of it and will | |
7370 | * free_sched_domains it when done with it. If the caller failed the | |
7371 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7372 | * and partition_sched_domains() will fallback to the single partition | |
7373 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7374 | * |
96f874e2 | 7375 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7376 | * ndoms_new == 0 is a special case for destroying existing domains, |
7377 | * and it will not create the default domain. | |
dfb512ec | 7378 | * |
029190c5 PJ |
7379 | * Call with hotplug lock held |
7380 | */ | |
acc3f5d7 | 7381 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7382 | struct sched_domain_attr *dattr_new) |
029190c5 | 7383 | { |
dfb512ec | 7384 | int i, j, n; |
d65bd5ec | 7385 | int new_topology; |
029190c5 | 7386 | |
712555ee | 7387 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7388 | |
7378547f MM |
7389 | /* always unregister in case we don't destroy any domains */ |
7390 | unregister_sched_domain_sysctl(); | |
7391 | ||
d65bd5ec HC |
7392 | /* Let architecture update cpu core mappings. */ |
7393 | new_topology = arch_update_cpu_topology(); | |
7394 | ||
dfb512ec | 7395 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7396 | |
7397 | /* Destroy deleted domains */ | |
7398 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7399 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7400 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7401 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7402 | goto match1; |
7403 | } | |
7404 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7405 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7406 | match1: |
7407 | ; | |
7408 | } | |
7409 | ||
e761b772 MK |
7410 | if (doms_new == NULL) { |
7411 | ndoms_cur = 0; | |
acc3f5d7 | 7412 | doms_new = &fallback_doms; |
6ad4c188 | 7413 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7414 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7415 | } |
7416 | ||
029190c5 PJ |
7417 | /* Build new domains */ |
7418 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7419 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7420 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7421 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7422 | goto match2; |
7423 | } | |
7424 | /* no match - add a new doms_new */ | |
dce840a0 | 7425 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7426 | match2: |
7427 | ; | |
7428 | } | |
7429 | ||
7430 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7431 | if (doms_cur != &fallback_doms) |
7432 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7433 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7434 | doms_cur = doms_new; |
1d3504fc | 7435 | dattr_cur = dattr_new; |
029190c5 | 7436 | ndoms_cur = ndoms_new; |
7378547f MM |
7437 | |
7438 | register_sched_domain_sysctl(); | |
a1835615 | 7439 | |
712555ee | 7440 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7441 | } |
7442 | ||
5c45bf27 | 7443 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 7444 | static void reinit_sched_domains(void) |
5c45bf27 | 7445 | { |
95402b38 | 7446 | get_online_cpus(); |
dfb512ec MK |
7447 | |
7448 | /* Destroy domains first to force the rebuild */ | |
7449 | partition_sched_domains(0, NULL, NULL); | |
7450 | ||
e761b772 | 7451 | rebuild_sched_domains(); |
95402b38 | 7452 | put_online_cpus(); |
5c45bf27 SS |
7453 | } |
7454 | ||
7455 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7456 | { | |
afb8a9b7 | 7457 | unsigned int level = 0; |
5c45bf27 | 7458 | |
afb8a9b7 GS |
7459 | if (sscanf(buf, "%u", &level) != 1) |
7460 | return -EINVAL; | |
7461 | ||
7462 | /* | |
7463 | * level is always be positive so don't check for | |
7464 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7465 | * What happens on 0 or 1 byte write, | |
7466 | * need to check for count as well? | |
7467 | */ | |
7468 | ||
7469 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7470 | return -EINVAL; |
7471 | ||
7472 | if (smt) | |
afb8a9b7 | 7473 | sched_smt_power_savings = level; |
5c45bf27 | 7474 | else |
afb8a9b7 | 7475 | sched_mc_power_savings = level; |
5c45bf27 | 7476 | |
c4a8849a | 7477 | reinit_sched_domains(); |
5c45bf27 | 7478 | |
c70f22d2 | 7479 | return count; |
5c45bf27 SS |
7480 | } |
7481 | ||
5c45bf27 | 7482 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7483 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7484 | struct sysdev_class_attribute *attr, |
f718cd4a | 7485 | char *page) |
5c45bf27 SS |
7486 | { |
7487 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7488 | } | |
f718cd4a | 7489 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7490 | struct sysdev_class_attribute *attr, |
48f24c4d | 7491 | const char *buf, size_t count) |
5c45bf27 SS |
7492 | { |
7493 | return sched_power_savings_store(buf, count, 0); | |
7494 | } | |
f718cd4a AK |
7495 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7496 | sched_mc_power_savings_show, | |
7497 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7498 | #endif |
7499 | ||
7500 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7501 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7502 | struct sysdev_class_attribute *attr, |
f718cd4a | 7503 | char *page) |
5c45bf27 SS |
7504 | { |
7505 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7506 | } | |
f718cd4a | 7507 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7508 | struct sysdev_class_attribute *attr, |
48f24c4d | 7509 | const char *buf, size_t count) |
5c45bf27 SS |
7510 | { |
7511 | return sched_power_savings_store(buf, count, 1); | |
7512 | } | |
f718cd4a AK |
7513 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7514 | sched_smt_power_savings_show, | |
6707de00 AB |
7515 | sched_smt_power_savings_store); |
7516 | #endif | |
7517 | ||
39aac648 | 7518 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7519 | { |
7520 | int err = 0; | |
7521 | ||
7522 | #ifdef CONFIG_SCHED_SMT | |
7523 | if (smt_capable()) | |
7524 | err = sysfs_create_file(&cls->kset.kobj, | |
7525 | &attr_sched_smt_power_savings.attr); | |
7526 | #endif | |
7527 | #ifdef CONFIG_SCHED_MC | |
7528 | if (!err && mc_capable()) | |
7529 | err = sysfs_create_file(&cls->kset.kobj, | |
7530 | &attr_sched_mc_power_savings.attr); | |
7531 | #endif | |
7532 | return err; | |
7533 | } | |
6d6bc0ad | 7534 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7535 | |
1da177e4 | 7536 | /* |
3a101d05 TH |
7537 | * Update cpusets according to cpu_active mask. If cpusets are |
7538 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7539 | * around partition_sched_domains(). | |
1da177e4 | 7540 | */ |
0b2e918a TH |
7541 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7542 | void *hcpu) | |
e761b772 | 7543 | { |
3a101d05 | 7544 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7545 | case CPU_ONLINE: |
6ad4c188 | 7546 | case CPU_DOWN_FAILED: |
3a101d05 | 7547 | cpuset_update_active_cpus(); |
e761b772 | 7548 | return NOTIFY_OK; |
3a101d05 TH |
7549 | default: |
7550 | return NOTIFY_DONE; | |
7551 | } | |
7552 | } | |
e761b772 | 7553 | |
0b2e918a TH |
7554 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7555 | void *hcpu) | |
3a101d05 TH |
7556 | { |
7557 | switch (action & ~CPU_TASKS_FROZEN) { | |
7558 | case CPU_DOWN_PREPARE: | |
7559 | cpuset_update_active_cpus(); | |
7560 | return NOTIFY_OK; | |
e761b772 MK |
7561 | default: |
7562 | return NOTIFY_DONE; | |
7563 | } | |
7564 | } | |
e761b772 MK |
7565 | |
7566 | static int update_runtime(struct notifier_block *nfb, | |
7567 | unsigned long action, void *hcpu) | |
1da177e4 | 7568 | { |
7def2be1 PZ |
7569 | int cpu = (int)(long)hcpu; |
7570 | ||
1da177e4 | 7571 | switch (action) { |
1da177e4 | 7572 | case CPU_DOWN_PREPARE: |
8bb78442 | 7573 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7574 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7575 | return NOTIFY_OK; |
7576 | ||
1da177e4 | 7577 | case CPU_DOWN_FAILED: |
8bb78442 | 7578 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7579 | case CPU_ONLINE: |
8bb78442 | 7580 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7581 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7582 | return NOTIFY_OK; |
7583 | ||
1da177e4 LT |
7584 | default: |
7585 | return NOTIFY_DONE; | |
7586 | } | |
1da177e4 | 7587 | } |
1da177e4 LT |
7588 | |
7589 | void __init sched_init_smp(void) | |
7590 | { | |
dcc30a35 RR |
7591 | cpumask_var_t non_isolated_cpus; |
7592 | ||
7593 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7594 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7595 | |
95402b38 | 7596 | get_online_cpus(); |
712555ee | 7597 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7598 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7599 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7600 | if (cpumask_empty(non_isolated_cpus)) | |
7601 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7602 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7603 | put_online_cpus(); |
e761b772 | 7604 | |
3a101d05 TH |
7605 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7606 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7607 | |
7608 | /* RT runtime code needs to handle some hotplug events */ | |
7609 | hotcpu_notifier(update_runtime, 0); | |
7610 | ||
b328ca18 | 7611 | init_hrtick(); |
5c1e1767 NP |
7612 | |
7613 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7614 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7615 | BUG(); |
19978ca6 | 7616 | sched_init_granularity(); |
dcc30a35 | 7617 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7618 | |
0e3900e6 | 7619 | init_sched_rt_class(); |
1da177e4 LT |
7620 | } |
7621 | #else | |
7622 | void __init sched_init_smp(void) | |
7623 | { | |
19978ca6 | 7624 | sched_init_granularity(); |
1da177e4 LT |
7625 | } |
7626 | #endif /* CONFIG_SMP */ | |
7627 | ||
cd1bb94b AB |
7628 | const_debug unsigned int sysctl_timer_migration = 1; |
7629 | ||
1da177e4 LT |
7630 | int in_sched_functions(unsigned long addr) |
7631 | { | |
1da177e4 LT |
7632 | return in_lock_functions(addr) || |
7633 | (addr >= (unsigned long)__sched_text_start | |
7634 | && addr < (unsigned long)__sched_text_end); | |
7635 | } | |
7636 | ||
a9957449 | 7637 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7638 | { |
7639 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7640 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7641 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7642 | cfs_rq->rq = rq; | |
f07333bf | 7643 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 7644 | #ifdef CONFIG_SMP |
f07333bf | 7645 | cfs_rq->load_stamp = 1; |
6ea72f12 | 7646 | #endif |
dd41f596 | 7647 | #endif |
67e9fb2a | 7648 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7649 | } |
7650 | ||
fa85ae24 PZ |
7651 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7652 | { | |
7653 | struct rt_prio_array *array; | |
7654 | int i; | |
7655 | ||
7656 | array = &rt_rq->active; | |
7657 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7658 | INIT_LIST_HEAD(array->queue + i); | |
7659 | __clear_bit(i, array->bitmap); | |
7660 | } | |
7661 | /* delimiter for bitsearch: */ | |
7662 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7663 | ||
052f1dc7 | 7664 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7665 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7666 | #ifdef CONFIG_SMP |
e864c499 | 7667 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7668 | #endif |
48d5e258 | 7669 | #endif |
fa85ae24 PZ |
7670 | #ifdef CONFIG_SMP |
7671 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7672 | rt_rq->overloaded = 0; |
05fa785c | 7673 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7674 | #endif |
7675 | ||
7676 | rt_rq->rt_time = 0; | |
7677 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7678 | rt_rq->rt_runtime = 0; |
0986b11b | 7679 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7680 | |
052f1dc7 | 7681 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7682 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7683 | rt_rq->rq = rq; |
7684 | #endif | |
fa85ae24 PZ |
7685 | } |
7686 | ||
6f505b16 | 7687 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7688 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7689 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7690 | struct sched_entity *parent) |
6f505b16 | 7691 | { |
ec7dc8ac | 7692 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7693 | tg->cfs_rq[cpu] = cfs_rq; |
7694 | init_cfs_rq(cfs_rq, rq); | |
7695 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7696 | |
7697 | tg->se[cpu] = se; | |
07e06b01 | 7698 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
7699 | if (!se) |
7700 | return; | |
7701 | ||
ec7dc8ac DG |
7702 | if (!parent) |
7703 | se->cfs_rq = &rq->cfs; | |
7704 | else | |
7705 | se->cfs_rq = parent->my_q; | |
7706 | ||
6f505b16 | 7707 | se->my_q = cfs_rq; |
9437178f | 7708 | update_load_set(&se->load, 0); |
ec7dc8ac | 7709 | se->parent = parent; |
6f505b16 | 7710 | } |
052f1dc7 | 7711 | #endif |
6f505b16 | 7712 | |
052f1dc7 | 7713 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7714 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7715 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7716 | struct sched_rt_entity *parent) |
6f505b16 | 7717 | { |
ec7dc8ac DG |
7718 | struct rq *rq = cpu_rq(cpu); |
7719 | ||
6f505b16 PZ |
7720 | tg->rt_rq[cpu] = rt_rq; |
7721 | init_rt_rq(rt_rq, rq); | |
7722 | rt_rq->tg = tg; | |
ac086bc2 | 7723 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7724 | |
7725 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7726 | if (!rt_se) |
7727 | return; | |
7728 | ||
ec7dc8ac DG |
7729 | if (!parent) |
7730 | rt_se->rt_rq = &rq->rt; | |
7731 | else | |
7732 | rt_se->rt_rq = parent->my_q; | |
7733 | ||
6f505b16 | 7734 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7735 | rt_se->parent = parent; |
6f505b16 PZ |
7736 | INIT_LIST_HEAD(&rt_se->run_list); |
7737 | } | |
7738 | #endif | |
7739 | ||
1da177e4 LT |
7740 | void __init sched_init(void) |
7741 | { | |
dd41f596 | 7742 | int i, j; |
434d53b0 MT |
7743 | unsigned long alloc_size = 0, ptr; |
7744 | ||
7745 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7746 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7747 | #endif | |
7748 | #ifdef CONFIG_RT_GROUP_SCHED | |
7749 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7750 | #endif |
df7c8e84 | 7751 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7752 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7753 | #endif |
434d53b0 | 7754 | if (alloc_size) { |
36b7b6d4 | 7755 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7756 | |
7757 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7758 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7759 | ptr += nr_cpu_ids * sizeof(void **); |
7760 | ||
07e06b01 | 7761 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7762 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7763 | |
6d6bc0ad | 7764 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7765 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7766 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7767 | ptr += nr_cpu_ids * sizeof(void **); |
7768 | ||
07e06b01 | 7769 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7770 | ptr += nr_cpu_ids * sizeof(void **); |
7771 | ||
6d6bc0ad | 7772 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7773 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7774 | for_each_possible_cpu(i) { | |
7775 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7776 | ptr += cpumask_size(); | |
7777 | } | |
7778 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7779 | } |
dd41f596 | 7780 | |
57d885fe GH |
7781 | #ifdef CONFIG_SMP |
7782 | init_defrootdomain(); | |
7783 | #endif | |
7784 | ||
d0b27fa7 PZ |
7785 | init_rt_bandwidth(&def_rt_bandwidth, |
7786 | global_rt_period(), global_rt_runtime()); | |
7787 | ||
7788 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7789 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7790 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7791 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7792 | |
7c941438 | 7793 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7794 | list_add(&root_task_group.list, &task_groups); |
7795 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 7796 | autogroup_init(&init_task); |
7c941438 | 7797 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7798 | |
0a945022 | 7799 | for_each_possible_cpu(i) { |
70b97a7f | 7800 | struct rq *rq; |
1da177e4 LT |
7801 | |
7802 | rq = cpu_rq(i); | |
05fa785c | 7803 | raw_spin_lock_init(&rq->lock); |
7897986b | 7804 | rq->nr_running = 0; |
dce48a84 TG |
7805 | rq->calc_load_active = 0; |
7806 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7807 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7808 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7809 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 7810 | root_task_group.shares = root_task_group_load; |
6f505b16 | 7811 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7812 | /* |
07e06b01 | 7813 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7814 | * |
7815 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7816 | * gets 100% of the cpu resources in the system. This overall | |
7817 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7818 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7819 | * based on each entity's (task or task-group's) weight |
7820 | * (se->load.weight). | |
7821 | * | |
07e06b01 | 7822 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7823 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7824 | * then A0's share of the cpu resource is: | |
7825 | * | |
0d905bca | 7826 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7827 | * |
07e06b01 YZ |
7828 | * We achieve this by letting root_task_group's tasks sit |
7829 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7830 | */ |
07e06b01 | 7831 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7832 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7833 | ||
7834 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7835 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7836 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7837 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7838 | #endif |
1da177e4 | 7839 | |
dd41f596 IM |
7840 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7841 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7842 | |
7843 | rq->last_load_update_tick = jiffies; | |
7844 | ||
1da177e4 | 7845 | #ifdef CONFIG_SMP |
41c7ce9a | 7846 | rq->sd = NULL; |
57d885fe | 7847 | rq->rd = NULL; |
e51fd5e2 | 7848 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7849 | rq->post_schedule = 0; |
1da177e4 | 7850 | rq->active_balance = 0; |
dd41f596 | 7851 | rq->next_balance = jiffies; |
1da177e4 | 7852 | rq->push_cpu = 0; |
0a2966b4 | 7853 | rq->cpu = i; |
1f11eb6a | 7854 | rq->online = 0; |
eae0c9df MG |
7855 | rq->idle_stamp = 0; |
7856 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7857 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7858 | #ifdef CONFIG_NO_HZ |
7859 | rq->nohz_balance_kick = 0; | |
7860 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7861 | #endif | |
1da177e4 | 7862 | #endif |
8f4d37ec | 7863 | init_rq_hrtick(rq); |
1da177e4 | 7864 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7865 | } |
7866 | ||
2dd73a4f | 7867 | set_load_weight(&init_task); |
b50f60ce | 7868 | |
e107be36 AK |
7869 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7870 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7871 | #endif | |
7872 | ||
c9819f45 | 7873 | #ifdef CONFIG_SMP |
962cf36c | 7874 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7875 | #endif |
7876 | ||
b50f60ce | 7877 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7878 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7879 | #endif |
7880 | ||
1da177e4 LT |
7881 | /* |
7882 | * The boot idle thread does lazy MMU switching as well: | |
7883 | */ | |
7884 | atomic_inc(&init_mm.mm_count); | |
7885 | enter_lazy_tlb(&init_mm, current); | |
7886 | ||
7887 | /* | |
7888 | * Make us the idle thread. Technically, schedule() should not be | |
7889 | * called from this thread, however somewhere below it might be, | |
7890 | * but because we are the idle thread, we just pick up running again | |
7891 | * when this runqueue becomes "idle". | |
7892 | */ | |
7893 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7894 | |
7895 | calc_load_update = jiffies + LOAD_FREQ; | |
7896 | ||
dd41f596 IM |
7897 | /* |
7898 | * During early bootup we pretend to be a normal task: | |
7899 | */ | |
7900 | current->sched_class = &fair_sched_class; | |
6892b75e | 7901 | |
6a7b3dc3 | 7902 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7903 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7904 | #ifdef CONFIG_SMP |
4cb98839 | 7905 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
7d1e6a9b | 7906 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
7907 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
7908 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
7909 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
7910 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
7911 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 7912 | #endif |
bdddd296 RR |
7913 | /* May be allocated at isolcpus cmdline parse time */ |
7914 | if (cpu_isolated_map == NULL) | |
7915 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7916 | #endif /* SMP */ |
6a7b3dc3 | 7917 | |
6892b75e | 7918 | scheduler_running = 1; |
1da177e4 LT |
7919 | } |
7920 | ||
7921 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7922 | static inline int preempt_count_equals(int preempt_offset) |
7923 | { | |
234da7bc | 7924 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7925 | |
4ba8216c | 7926 | return (nested == preempt_offset); |
e4aafea2 FW |
7927 | } |
7928 | ||
d894837f | 7929 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7930 | { |
48f24c4d | 7931 | #ifdef in_atomic |
1da177e4 LT |
7932 | static unsigned long prev_jiffy; /* ratelimiting */ |
7933 | ||
e4aafea2 FW |
7934 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7935 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7936 | return; |
7937 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7938 | return; | |
7939 | prev_jiffy = jiffies; | |
7940 | ||
3df0fc5b PZ |
7941 | printk(KERN_ERR |
7942 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7943 | file, line); | |
7944 | printk(KERN_ERR | |
7945 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7946 | in_atomic(), irqs_disabled(), | |
7947 | current->pid, current->comm); | |
aef745fc IM |
7948 | |
7949 | debug_show_held_locks(current); | |
7950 | if (irqs_disabled()) | |
7951 | print_irqtrace_events(current); | |
7952 | dump_stack(); | |
1da177e4 LT |
7953 | #endif |
7954 | } | |
7955 | EXPORT_SYMBOL(__might_sleep); | |
7956 | #endif | |
7957 | ||
7958 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7959 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7960 | { | |
da7a735e PZ |
7961 | const struct sched_class *prev_class = p->sched_class; |
7962 | int old_prio = p->prio; | |
3a5e4dc1 | 7963 | int on_rq; |
3e51f33f | 7964 | |
3a5e4dc1 AK |
7965 | on_rq = p->se.on_rq; |
7966 | if (on_rq) | |
7967 | deactivate_task(rq, p, 0); | |
7968 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7969 | if (on_rq) { | |
7970 | activate_task(rq, p, 0); | |
7971 | resched_task(rq->curr); | |
7972 | } | |
da7a735e PZ |
7973 | |
7974 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7975 | } |
7976 | ||
1da177e4 LT |
7977 | void normalize_rt_tasks(void) |
7978 | { | |
a0f98a1c | 7979 | struct task_struct *g, *p; |
1da177e4 | 7980 | unsigned long flags; |
70b97a7f | 7981 | struct rq *rq; |
1da177e4 | 7982 | |
4cf5d77a | 7983 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7984 | do_each_thread(g, p) { |
178be793 IM |
7985 | /* |
7986 | * Only normalize user tasks: | |
7987 | */ | |
7988 | if (!p->mm) | |
7989 | continue; | |
7990 | ||
6cfb0d5d | 7991 | p->se.exec_start = 0; |
6cfb0d5d | 7992 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7993 | p->se.statistics.wait_start = 0; |
7994 | p->se.statistics.sleep_start = 0; | |
7995 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7996 | #endif |
dd41f596 IM |
7997 | |
7998 | if (!rt_task(p)) { | |
7999 | /* | |
8000 | * Renice negative nice level userspace | |
8001 | * tasks back to 0: | |
8002 | */ | |
8003 | if (TASK_NICE(p) < 0 && p->mm) | |
8004 | set_user_nice(p, 0); | |
1da177e4 | 8005 | continue; |
dd41f596 | 8006 | } |
1da177e4 | 8007 | |
1d615482 | 8008 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8009 | rq = __task_rq_lock(p); |
1da177e4 | 8010 | |
178be793 | 8011 | normalize_task(rq, p); |
3a5e4dc1 | 8012 | |
b29739f9 | 8013 | __task_rq_unlock(rq); |
1d615482 | 8014 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8015 | } while_each_thread(g, p); |
8016 | ||
4cf5d77a | 8017 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8018 | } |
8019 | ||
8020 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8021 | |
67fc4e0c | 8022 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8023 | /* |
67fc4e0c | 8024 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8025 | * |
8026 | * They can only be called when the whole system has been | |
8027 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8028 | * activity can take place. Using them for anything else would | |
8029 | * be a serious bug, and as a result, they aren't even visible | |
8030 | * under any other configuration. | |
8031 | */ | |
8032 | ||
8033 | /** | |
8034 | * curr_task - return the current task for a given cpu. | |
8035 | * @cpu: the processor in question. | |
8036 | * | |
8037 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8038 | */ | |
36c8b586 | 8039 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8040 | { |
8041 | return cpu_curr(cpu); | |
8042 | } | |
8043 | ||
67fc4e0c JW |
8044 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8045 | ||
8046 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8047 | /** |
8048 | * set_curr_task - set the current task for a given cpu. | |
8049 | * @cpu: the processor in question. | |
8050 | * @p: the task pointer to set. | |
8051 | * | |
8052 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8053 | * are serviced on a separate stack. It allows the architecture to switch the |
8054 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8055 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8056 | * and caller must save the original value of the current task (see | |
8057 | * curr_task() above) and restore that value before reenabling interrupts and | |
8058 | * re-starting the system. | |
8059 | * | |
8060 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8061 | */ | |
36c8b586 | 8062 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8063 | { |
8064 | cpu_curr(cpu) = p; | |
8065 | } | |
8066 | ||
8067 | #endif | |
29f59db3 | 8068 | |
bccbe08a PZ |
8069 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8070 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8071 | { |
8072 | int i; | |
8073 | ||
8074 | for_each_possible_cpu(i) { | |
8075 | if (tg->cfs_rq) | |
8076 | kfree(tg->cfs_rq[i]); | |
8077 | if (tg->se) | |
8078 | kfree(tg->se[i]); | |
6f505b16 PZ |
8079 | } |
8080 | ||
8081 | kfree(tg->cfs_rq); | |
8082 | kfree(tg->se); | |
6f505b16 PZ |
8083 | } |
8084 | ||
ec7dc8ac DG |
8085 | static |
8086 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8087 | { |
29f59db3 | 8088 | struct cfs_rq *cfs_rq; |
eab17229 | 8089 | struct sched_entity *se; |
29f59db3 SV |
8090 | int i; |
8091 | ||
434d53b0 | 8092 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8093 | if (!tg->cfs_rq) |
8094 | goto err; | |
434d53b0 | 8095 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8096 | if (!tg->se) |
8097 | goto err; | |
052f1dc7 PZ |
8098 | |
8099 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8100 | |
8101 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8102 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8103 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8104 | if (!cfs_rq) |
8105 | goto err; | |
8106 | ||
eab17229 LZ |
8107 | se = kzalloc_node(sizeof(struct sched_entity), |
8108 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8109 | if (!se) |
dfc12eb2 | 8110 | goto err_free_rq; |
29f59db3 | 8111 | |
3d4b47b4 | 8112 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8113 | } |
8114 | ||
8115 | return 1; | |
8116 | ||
49246274 | 8117 | err_free_rq: |
dfc12eb2 | 8118 | kfree(cfs_rq); |
49246274 | 8119 | err: |
bccbe08a PZ |
8120 | return 0; |
8121 | } | |
8122 | ||
bccbe08a PZ |
8123 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8124 | { | |
3d4b47b4 PZ |
8125 | struct rq *rq = cpu_rq(cpu); |
8126 | unsigned long flags; | |
3d4b47b4 PZ |
8127 | |
8128 | /* | |
8129 | * Only empty task groups can be destroyed; so we can speculatively | |
8130 | * check on_list without danger of it being re-added. | |
8131 | */ | |
8132 | if (!tg->cfs_rq[cpu]->on_list) | |
8133 | return; | |
8134 | ||
8135 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8136 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8137 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8138 | } |
6d6bc0ad | 8139 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8140 | static inline void free_fair_sched_group(struct task_group *tg) |
8141 | { | |
8142 | } | |
8143 | ||
ec7dc8ac DG |
8144 | static inline |
8145 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8146 | { |
8147 | return 1; | |
8148 | } | |
8149 | ||
bccbe08a PZ |
8150 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8151 | { | |
8152 | } | |
6d6bc0ad | 8153 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8154 | |
8155 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8156 | static void free_rt_sched_group(struct task_group *tg) |
8157 | { | |
8158 | int i; | |
8159 | ||
d0b27fa7 PZ |
8160 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8161 | ||
bccbe08a PZ |
8162 | for_each_possible_cpu(i) { |
8163 | if (tg->rt_rq) | |
8164 | kfree(tg->rt_rq[i]); | |
8165 | if (tg->rt_se) | |
8166 | kfree(tg->rt_se[i]); | |
8167 | } | |
8168 | ||
8169 | kfree(tg->rt_rq); | |
8170 | kfree(tg->rt_se); | |
8171 | } | |
8172 | ||
ec7dc8ac DG |
8173 | static |
8174 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8175 | { |
8176 | struct rt_rq *rt_rq; | |
eab17229 | 8177 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8178 | struct rq *rq; |
8179 | int i; | |
8180 | ||
434d53b0 | 8181 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8182 | if (!tg->rt_rq) |
8183 | goto err; | |
434d53b0 | 8184 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8185 | if (!tg->rt_se) |
8186 | goto err; | |
8187 | ||
d0b27fa7 PZ |
8188 | init_rt_bandwidth(&tg->rt_bandwidth, |
8189 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8190 | |
8191 | for_each_possible_cpu(i) { | |
8192 | rq = cpu_rq(i); | |
8193 | ||
eab17229 LZ |
8194 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8195 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8196 | if (!rt_rq) |
8197 | goto err; | |
29f59db3 | 8198 | |
eab17229 LZ |
8199 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8200 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8201 | if (!rt_se) |
dfc12eb2 | 8202 | goto err_free_rq; |
29f59db3 | 8203 | |
3d4b47b4 | 8204 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8205 | } |
8206 | ||
bccbe08a PZ |
8207 | return 1; |
8208 | ||
49246274 | 8209 | err_free_rq: |
dfc12eb2 | 8210 | kfree(rt_rq); |
49246274 | 8211 | err: |
bccbe08a PZ |
8212 | return 0; |
8213 | } | |
6d6bc0ad | 8214 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8215 | static inline void free_rt_sched_group(struct task_group *tg) |
8216 | { | |
8217 | } | |
8218 | ||
ec7dc8ac DG |
8219 | static inline |
8220 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8221 | { |
8222 | return 1; | |
8223 | } | |
6d6bc0ad | 8224 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8225 | |
7c941438 | 8226 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8227 | static void free_sched_group(struct task_group *tg) |
8228 | { | |
8229 | free_fair_sched_group(tg); | |
8230 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8231 | autogroup_free(tg); |
bccbe08a PZ |
8232 | kfree(tg); |
8233 | } | |
8234 | ||
8235 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8236 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8237 | { |
8238 | struct task_group *tg; | |
8239 | unsigned long flags; | |
bccbe08a PZ |
8240 | |
8241 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8242 | if (!tg) | |
8243 | return ERR_PTR(-ENOMEM); | |
8244 | ||
ec7dc8ac | 8245 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8246 | goto err; |
8247 | ||
ec7dc8ac | 8248 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8249 | goto err; |
8250 | ||
8ed36996 | 8251 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8252 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8253 | |
8254 | WARN_ON(!parent); /* root should already exist */ | |
8255 | ||
8256 | tg->parent = parent; | |
f473aa5e | 8257 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8258 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8259 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8260 | |
9b5b7751 | 8261 | return tg; |
29f59db3 SV |
8262 | |
8263 | err: | |
6f505b16 | 8264 | free_sched_group(tg); |
29f59db3 SV |
8265 | return ERR_PTR(-ENOMEM); |
8266 | } | |
8267 | ||
9b5b7751 | 8268 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8269 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8270 | { |
29f59db3 | 8271 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8272 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8273 | } |
8274 | ||
9b5b7751 | 8275 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8276 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8277 | { |
8ed36996 | 8278 | unsigned long flags; |
9b5b7751 | 8279 | int i; |
29f59db3 | 8280 | |
3d4b47b4 PZ |
8281 | /* end participation in shares distribution */ |
8282 | for_each_possible_cpu(i) | |
bccbe08a | 8283 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8284 | |
8285 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8286 | list_del_rcu(&tg->list); |
f473aa5e | 8287 | list_del_rcu(&tg->siblings); |
8ed36996 | 8288 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8289 | |
9b5b7751 | 8290 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8291 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8292 | } |
8293 | ||
9b5b7751 | 8294 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8295 | * The caller of this function should have put the task in its new group |
8296 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8297 | * reflect its new group. | |
9b5b7751 SV |
8298 | */ |
8299 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8300 | { |
8301 | int on_rq, running; | |
8302 | unsigned long flags; | |
8303 | struct rq *rq; | |
8304 | ||
8305 | rq = task_rq_lock(tsk, &flags); | |
8306 | ||
051a1d1a | 8307 | running = task_current(rq, tsk); |
29f59db3 SV |
8308 | on_rq = tsk->se.on_rq; |
8309 | ||
0e1f3483 | 8310 | if (on_rq) |
29f59db3 | 8311 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8312 | if (unlikely(running)) |
8313 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8314 | |
810b3817 | 8315 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8316 | if (tsk->sched_class->task_move_group) |
8317 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8318 | else | |
810b3817 | 8319 | #endif |
b2b5ce02 | 8320 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8321 | |
0e1f3483 HS |
8322 | if (unlikely(running)) |
8323 | tsk->sched_class->set_curr_task(rq); | |
8324 | if (on_rq) | |
371fd7e7 | 8325 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8326 | |
29f59db3 SV |
8327 | task_rq_unlock(rq, &flags); |
8328 | } | |
7c941438 | 8329 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8330 | |
052f1dc7 | 8331 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8332 | static DEFINE_MUTEX(shares_mutex); |
8333 | ||
4cf86d77 | 8334 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8335 | { |
8336 | int i; | |
8ed36996 | 8337 | unsigned long flags; |
c61935fd | 8338 | |
ec7dc8ac DG |
8339 | /* |
8340 | * We can't change the weight of the root cgroup. | |
8341 | */ | |
8342 | if (!tg->se[0]) | |
8343 | return -EINVAL; | |
8344 | ||
18d95a28 PZ |
8345 | if (shares < MIN_SHARES) |
8346 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8347 | else if (shares > MAX_SHARES) |
8348 | shares = MAX_SHARES; | |
62fb1851 | 8349 | |
8ed36996 | 8350 | mutex_lock(&shares_mutex); |
9b5b7751 | 8351 | if (tg->shares == shares) |
5cb350ba | 8352 | goto done; |
29f59db3 | 8353 | |
9b5b7751 | 8354 | tg->shares = shares; |
c09595f6 | 8355 | for_each_possible_cpu(i) { |
9437178f PT |
8356 | struct rq *rq = cpu_rq(i); |
8357 | struct sched_entity *se; | |
8358 | ||
8359 | se = tg->se[i]; | |
8360 | /* Propagate contribution to hierarchy */ | |
8361 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8362 | for_each_sched_entity(se) | |
6d5ab293 | 8363 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8364 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8365 | } |
29f59db3 | 8366 | |
5cb350ba | 8367 | done: |
8ed36996 | 8368 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8369 | return 0; |
29f59db3 SV |
8370 | } |
8371 | ||
5cb350ba DG |
8372 | unsigned long sched_group_shares(struct task_group *tg) |
8373 | { | |
8374 | return tg->shares; | |
8375 | } | |
052f1dc7 | 8376 | #endif |
5cb350ba | 8377 | |
052f1dc7 | 8378 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8379 | /* |
9f0c1e56 | 8380 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8381 | */ |
9f0c1e56 PZ |
8382 | static DEFINE_MUTEX(rt_constraints_mutex); |
8383 | ||
8384 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8385 | { | |
8386 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8387 | return 1ULL << 20; |
9f0c1e56 | 8388 | |
9a7e0b18 | 8389 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8390 | } |
8391 | ||
9a7e0b18 PZ |
8392 | /* Must be called with tasklist_lock held */ |
8393 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8394 | { |
9a7e0b18 | 8395 | struct task_struct *g, *p; |
b40b2e8e | 8396 | |
9a7e0b18 PZ |
8397 | do_each_thread(g, p) { |
8398 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8399 | return 1; | |
8400 | } while_each_thread(g, p); | |
b40b2e8e | 8401 | |
9a7e0b18 PZ |
8402 | return 0; |
8403 | } | |
b40b2e8e | 8404 | |
9a7e0b18 PZ |
8405 | struct rt_schedulable_data { |
8406 | struct task_group *tg; | |
8407 | u64 rt_period; | |
8408 | u64 rt_runtime; | |
8409 | }; | |
b40b2e8e | 8410 | |
9a7e0b18 PZ |
8411 | static int tg_schedulable(struct task_group *tg, void *data) |
8412 | { | |
8413 | struct rt_schedulable_data *d = data; | |
8414 | struct task_group *child; | |
8415 | unsigned long total, sum = 0; | |
8416 | u64 period, runtime; | |
b40b2e8e | 8417 | |
9a7e0b18 PZ |
8418 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8419 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8420 | |
9a7e0b18 PZ |
8421 | if (tg == d->tg) { |
8422 | period = d->rt_period; | |
8423 | runtime = d->rt_runtime; | |
b40b2e8e | 8424 | } |
b40b2e8e | 8425 | |
4653f803 PZ |
8426 | /* |
8427 | * Cannot have more runtime than the period. | |
8428 | */ | |
8429 | if (runtime > period && runtime != RUNTIME_INF) | |
8430 | return -EINVAL; | |
6f505b16 | 8431 | |
4653f803 PZ |
8432 | /* |
8433 | * Ensure we don't starve existing RT tasks. | |
8434 | */ | |
9a7e0b18 PZ |
8435 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8436 | return -EBUSY; | |
6f505b16 | 8437 | |
9a7e0b18 | 8438 | total = to_ratio(period, runtime); |
6f505b16 | 8439 | |
4653f803 PZ |
8440 | /* |
8441 | * Nobody can have more than the global setting allows. | |
8442 | */ | |
8443 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8444 | return -EINVAL; | |
6f505b16 | 8445 | |
4653f803 PZ |
8446 | /* |
8447 | * The sum of our children's runtime should not exceed our own. | |
8448 | */ | |
9a7e0b18 PZ |
8449 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8450 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8451 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8452 | |
9a7e0b18 PZ |
8453 | if (child == d->tg) { |
8454 | period = d->rt_period; | |
8455 | runtime = d->rt_runtime; | |
8456 | } | |
6f505b16 | 8457 | |
9a7e0b18 | 8458 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8459 | } |
6f505b16 | 8460 | |
9a7e0b18 PZ |
8461 | if (sum > total) |
8462 | return -EINVAL; | |
8463 | ||
8464 | return 0; | |
6f505b16 PZ |
8465 | } |
8466 | ||
9a7e0b18 | 8467 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8468 | { |
9a7e0b18 PZ |
8469 | struct rt_schedulable_data data = { |
8470 | .tg = tg, | |
8471 | .rt_period = period, | |
8472 | .rt_runtime = runtime, | |
8473 | }; | |
8474 | ||
8475 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8476 | } |
8477 | ||
d0b27fa7 PZ |
8478 | static int tg_set_bandwidth(struct task_group *tg, |
8479 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8480 | { |
ac086bc2 | 8481 | int i, err = 0; |
9f0c1e56 | 8482 | |
9f0c1e56 | 8483 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8484 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8485 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8486 | if (err) | |
9f0c1e56 | 8487 | goto unlock; |
ac086bc2 | 8488 | |
0986b11b | 8489 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8490 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8491 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8492 | |
8493 | for_each_possible_cpu(i) { | |
8494 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8495 | ||
0986b11b | 8496 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8497 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8498 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8499 | } |
0986b11b | 8500 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8501 | unlock: |
521f1a24 | 8502 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8503 | mutex_unlock(&rt_constraints_mutex); |
8504 | ||
8505 | return err; | |
6f505b16 PZ |
8506 | } |
8507 | ||
d0b27fa7 PZ |
8508 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8509 | { | |
8510 | u64 rt_runtime, rt_period; | |
8511 | ||
8512 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8513 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8514 | if (rt_runtime_us < 0) | |
8515 | rt_runtime = RUNTIME_INF; | |
8516 | ||
8517 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8518 | } | |
8519 | ||
9f0c1e56 PZ |
8520 | long sched_group_rt_runtime(struct task_group *tg) |
8521 | { | |
8522 | u64 rt_runtime_us; | |
8523 | ||
d0b27fa7 | 8524 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8525 | return -1; |
8526 | ||
d0b27fa7 | 8527 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8528 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8529 | return rt_runtime_us; | |
8530 | } | |
d0b27fa7 PZ |
8531 | |
8532 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8533 | { | |
8534 | u64 rt_runtime, rt_period; | |
8535 | ||
8536 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8537 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8538 | ||
619b0488 R |
8539 | if (rt_period == 0) |
8540 | return -EINVAL; | |
8541 | ||
d0b27fa7 PZ |
8542 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8543 | } | |
8544 | ||
8545 | long sched_group_rt_period(struct task_group *tg) | |
8546 | { | |
8547 | u64 rt_period_us; | |
8548 | ||
8549 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8550 | do_div(rt_period_us, NSEC_PER_USEC); | |
8551 | return rt_period_us; | |
8552 | } | |
8553 | ||
8554 | static int sched_rt_global_constraints(void) | |
8555 | { | |
4653f803 | 8556 | u64 runtime, period; |
d0b27fa7 PZ |
8557 | int ret = 0; |
8558 | ||
ec5d4989 HS |
8559 | if (sysctl_sched_rt_period <= 0) |
8560 | return -EINVAL; | |
8561 | ||
4653f803 PZ |
8562 | runtime = global_rt_runtime(); |
8563 | period = global_rt_period(); | |
8564 | ||
8565 | /* | |
8566 | * Sanity check on the sysctl variables. | |
8567 | */ | |
8568 | if (runtime > period && runtime != RUNTIME_INF) | |
8569 | return -EINVAL; | |
10b612f4 | 8570 | |
d0b27fa7 | 8571 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8572 | read_lock(&tasklist_lock); |
4653f803 | 8573 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8574 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8575 | mutex_unlock(&rt_constraints_mutex); |
8576 | ||
8577 | return ret; | |
8578 | } | |
54e99124 DG |
8579 | |
8580 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8581 | { | |
8582 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8583 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8584 | return 0; | |
8585 | ||
8586 | return 1; | |
8587 | } | |
8588 | ||
6d6bc0ad | 8589 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8590 | static int sched_rt_global_constraints(void) |
8591 | { | |
ac086bc2 PZ |
8592 | unsigned long flags; |
8593 | int i; | |
8594 | ||
ec5d4989 HS |
8595 | if (sysctl_sched_rt_period <= 0) |
8596 | return -EINVAL; | |
8597 | ||
60aa605d PZ |
8598 | /* |
8599 | * There's always some RT tasks in the root group | |
8600 | * -- migration, kstopmachine etc.. | |
8601 | */ | |
8602 | if (sysctl_sched_rt_runtime == 0) | |
8603 | return -EBUSY; | |
8604 | ||
0986b11b | 8605 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8606 | for_each_possible_cpu(i) { |
8607 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8608 | ||
0986b11b | 8609 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8610 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8611 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8612 | } |
0986b11b | 8613 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8614 | |
d0b27fa7 PZ |
8615 | return 0; |
8616 | } | |
6d6bc0ad | 8617 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8618 | |
8619 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8620 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8621 | loff_t *ppos) |
8622 | { | |
8623 | int ret; | |
8624 | int old_period, old_runtime; | |
8625 | static DEFINE_MUTEX(mutex); | |
8626 | ||
8627 | mutex_lock(&mutex); | |
8628 | old_period = sysctl_sched_rt_period; | |
8629 | old_runtime = sysctl_sched_rt_runtime; | |
8630 | ||
8d65af78 | 8631 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8632 | |
8633 | if (!ret && write) { | |
8634 | ret = sched_rt_global_constraints(); | |
8635 | if (ret) { | |
8636 | sysctl_sched_rt_period = old_period; | |
8637 | sysctl_sched_rt_runtime = old_runtime; | |
8638 | } else { | |
8639 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8640 | def_rt_bandwidth.rt_period = | |
8641 | ns_to_ktime(global_rt_period()); | |
8642 | } | |
8643 | } | |
8644 | mutex_unlock(&mutex); | |
8645 | ||
8646 | return ret; | |
8647 | } | |
68318b8e | 8648 | |
052f1dc7 | 8649 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8650 | |
8651 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8652 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8653 | { |
2b01dfe3 PM |
8654 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8655 | struct task_group, css); | |
68318b8e SV |
8656 | } |
8657 | ||
8658 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8659 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8660 | { |
ec7dc8ac | 8661 | struct task_group *tg, *parent; |
68318b8e | 8662 | |
2b01dfe3 | 8663 | if (!cgrp->parent) { |
68318b8e | 8664 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8665 | return &root_task_group.css; |
68318b8e SV |
8666 | } |
8667 | ||
ec7dc8ac DG |
8668 | parent = cgroup_tg(cgrp->parent); |
8669 | tg = sched_create_group(parent); | |
68318b8e SV |
8670 | if (IS_ERR(tg)) |
8671 | return ERR_PTR(-ENOMEM); | |
8672 | ||
68318b8e SV |
8673 | return &tg->css; |
8674 | } | |
8675 | ||
41a2d6cf IM |
8676 | static void |
8677 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8678 | { |
2b01dfe3 | 8679 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8680 | |
8681 | sched_destroy_group(tg); | |
8682 | } | |
8683 | ||
41a2d6cf | 8684 | static int |
be367d09 | 8685 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8686 | { |
b68aa230 | 8687 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8688 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8689 | return -EINVAL; |
8690 | #else | |
68318b8e SV |
8691 | /* We don't support RT-tasks being in separate groups */ |
8692 | if (tsk->sched_class != &fair_sched_class) | |
8693 | return -EINVAL; | |
b68aa230 | 8694 | #endif |
be367d09 BB |
8695 | return 0; |
8696 | } | |
68318b8e | 8697 | |
be367d09 BB |
8698 | static int |
8699 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8700 | struct task_struct *tsk, bool threadgroup) | |
8701 | { | |
8702 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8703 | if (retval) | |
8704 | return retval; | |
8705 | if (threadgroup) { | |
8706 | struct task_struct *c; | |
8707 | rcu_read_lock(); | |
8708 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8709 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8710 | if (retval) { | |
8711 | rcu_read_unlock(); | |
8712 | return retval; | |
8713 | } | |
8714 | } | |
8715 | rcu_read_unlock(); | |
8716 | } | |
68318b8e SV |
8717 | return 0; |
8718 | } | |
8719 | ||
8720 | static void | |
2b01dfe3 | 8721 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8722 | struct cgroup *old_cont, struct task_struct *tsk, |
8723 | bool threadgroup) | |
68318b8e SV |
8724 | { |
8725 | sched_move_task(tsk); | |
be367d09 BB |
8726 | if (threadgroup) { |
8727 | struct task_struct *c; | |
8728 | rcu_read_lock(); | |
8729 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8730 | sched_move_task(c); | |
8731 | } | |
8732 | rcu_read_unlock(); | |
8733 | } | |
68318b8e SV |
8734 | } |
8735 | ||
068c5cc5 | 8736 | static void |
d41d5a01 PZ |
8737 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
8738 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
8739 | { |
8740 | /* | |
8741 | * cgroup_exit() is called in the copy_process() failure path. | |
8742 | * Ignore this case since the task hasn't ran yet, this avoids | |
8743 | * trying to poke a half freed task state from generic code. | |
8744 | */ | |
8745 | if (!(task->flags & PF_EXITING)) | |
8746 | return; | |
8747 | ||
8748 | sched_move_task(task); | |
8749 | } | |
8750 | ||
052f1dc7 | 8751 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8752 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8753 | u64 shareval) |
68318b8e | 8754 | { |
2b01dfe3 | 8755 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8756 | } |
8757 | ||
f4c753b7 | 8758 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8759 | { |
2b01dfe3 | 8760 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8761 | |
8762 | return (u64) tg->shares; | |
8763 | } | |
6d6bc0ad | 8764 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8765 | |
052f1dc7 | 8766 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8767 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8768 | s64 val) |
6f505b16 | 8769 | { |
06ecb27c | 8770 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8771 | } |
8772 | ||
06ecb27c | 8773 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8774 | { |
06ecb27c | 8775 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8776 | } |
d0b27fa7 PZ |
8777 | |
8778 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8779 | u64 rt_period_us) | |
8780 | { | |
8781 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8782 | } | |
8783 | ||
8784 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8785 | { | |
8786 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8787 | } | |
6d6bc0ad | 8788 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8789 | |
fe5c7cc2 | 8790 | static struct cftype cpu_files[] = { |
052f1dc7 | 8791 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8792 | { |
8793 | .name = "shares", | |
f4c753b7 PM |
8794 | .read_u64 = cpu_shares_read_u64, |
8795 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8796 | }, |
052f1dc7 PZ |
8797 | #endif |
8798 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8799 | { |
9f0c1e56 | 8800 | .name = "rt_runtime_us", |
06ecb27c PM |
8801 | .read_s64 = cpu_rt_runtime_read, |
8802 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8803 | }, |
d0b27fa7 PZ |
8804 | { |
8805 | .name = "rt_period_us", | |
f4c753b7 PM |
8806 | .read_u64 = cpu_rt_period_read_uint, |
8807 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8808 | }, |
052f1dc7 | 8809 | #endif |
68318b8e SV |
8810 | }; |
8811 | ||
8812 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8813 | { | |
fe5c7cc2 | 8814 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8815 | } |
8816 | ||
8817 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8818 | .name = "cpu", |
8819 | .create = cpu_cgroup_create, | |
8820 | .destroy = cpu_cgroup_destroy, | |
8821 | .can_attach = cpu_cgroup_can_attach, | |
8822 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8823 | .exit = cpu_cgroup_exit, |
38605cae IM |
8824 | .populate = cpu_cgroup_populate, |
8825 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8826 | .early_init = 1, |
8827 | }; | |
8828 | ||
052f1dc7 | 8829 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8830 | |
8831 | #ifdef CONFIG_CGROUP_CPUACCT | |
8832 | ||
8833 | /* | |
8834 | * CPU accounting code for task groups. | |
8835 | * | |
8836 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8837 | * (balbir@in.ibm.com). | |
8838 | */ | |
8839 | ||
934352f2 | 8840 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8841 | struct cpuacct { |
8842 | struct cgroup_subsys_state css; | |
8843 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8844 | u64 __percpu *cpuusage; |
ef12fefa | 8845 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8846 | struct cpuacct *parent; |
d842de87 SV |
8847 | }; |
8848 | ||
8849 | struct cgroup_subsys cpuacct_subsys; | |
8850 | ||
8851 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8852 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8853 | { |
32cd756a | 8854 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8855 | struct cpuacct, css); |
8856 | } | |
8857 | ||
8858 | /* return cpu accounting group to which this task belongs */ | |
8859 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8860 | { | |
8861 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8862 | struct cpuacct, css); | |
8863 | } | |
8864 | ||
8865 | /* create a new cpu accounting group */ | |
8866 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8867 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8868 | { |
8869 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8870 | int i; |
d842de87 SV |
8871 | |
8872 | if (!ca) | |
ef12fefa | 8873 | goto out; |
d842de87 SV |
8874 | |
8875 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8876 | if (!ca->cpuusage) |
8877 | goto out_free_ca; | |
8878 | ||
8879 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8880 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8881 | goto out_free_counters; | |
d842de87 | 8882 | |
934352f2 BR |
8883 | if (cgrp->parent) |
8884 | ca->parent = cgroup_ca(cgrp->parent); | |
8885 | ||
d842de87 | 8886 | return &ca->css; |
ef12fefa BR |
8887 | |
8888 | out_free_counters: | |
8889 | while (--i >= 0) | |
8890 | percpu_counter_destroy(&ca->cpustat[i]); | |
8891 | free_percpu(ca->cpuusage); | |
8892 | out_free_ca: | |
8893 | kfree(ca); | |
8894 | out: | |
8895 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8896 | } |
8897 | ||
8898 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8899 | static void |
32cd756a | 8900 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8901 | { |
32cd756a | 8902 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8903 | int i; |
d842de87 | 8904 | |
ef12fefa BR |
8905 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8906 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8907 | free_percpu(ca->cpuusage); |
8908 | kfree(ca); | |
8909 | } | |
8910 | ||
720f5498 KC |
8911 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8912 | { | |
b36128c8 | 8913 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8914 | u64 data; |
8915 | ||
8916 | #ifndef CONFIG_64BIT | |
8917 | /* | |
8918 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8919 | */ | |
05fa785c | 8920 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8921 | data = *cpuusage; |
05fa785c | 8922 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8923 | #else |
8924 | data = *cpuusage; | |
8925 | #endif | |
8926 | ||
8927 | return data; | |
8928 | } | |
8929 | ||
8930 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8931 | { | |
b36128c8 | 8932 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8933 | |
8934 | #ifndef CONFIG_64BIT | |
8935 | /* | |
8936 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8937 | */ | |
05fa785c | 8938 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8939 | *cpuusage = val; |
05fa785c | 8940 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8941 | #else |
8942 | *cpuusage = val; | |
8943 | #endif | |
8944 | } | |
8945 | ||
d842de87 | 8946 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8947 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8948 | { |
32cd756a | 8949 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8950 | u64 totalcpuusage = 0; |
8951 | int i; | |
8952 | ||
720f5498 KC |
8953 | for_each_present_cpu(i) |
8954 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8955 | |
8956 | return totalcpuusage; | |
8957 | } | |
8958 | ||
0297b803 DG |
8959 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8960 | u64 reset) | |
8961 | { | |
8962 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8963 | int err = 0; | |
8964 | int i; | |
8965 | ||
8966 | if (reset) { | |
8967 | err = -EINVAL; | |
8968 | goto out; | |
8969 | } | |
8970 | ||
720f5498 KC |
8971 | for_each_present_cpu(i) |
8972 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8973 | |
0297b803 DG |
8974 | out: |
8975 | return err; | |
8976 | } | |
8977 | ||
e9515c3c KC |
8978 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8979 | struct seq_file *m) | |
8980 | { | |
8981 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8982 | u64 percpu; | |
8983 | int i; | |
8984 | ||
8985 | for_each_present_cpu(i) { | |
8986 | percpu = cpuacct_cpuusage_read(ca, i); | |
8987 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8988 | } | |
8989 | seq_printf(m, "\n"); | |
8990 | return 0; | |
8991 | } | |
8992 | ||
ef12fefa BR |
8993 | static const char *cpuacct_stat_desc[] = { |
8994 | [CPUACCT_STAT_USER] = "user", | |
8995 | [CPUACCT_STAT_SYSTEM] = "system", | |
8996 | }; | |
8997 | ||
8998 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8999 | struct cgroup_map_cb *cb) | |
9000 | { | |
9001 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9002 | int i; | |
9003 | ||
9004 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9005 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9006 | val = cputime64_to_clock_t(val); | |
9007 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9008 | } | |
9009 | return 0; | |
9010 | } | |
9011 | ||
d842de87 SV |
9012 | static struct cftype files[] = { |
9013 | { | |
9014 | .name = "usage", | |
f4c753b7 PM |
9015 | .read_u64 = cpuusage_read, |
9016 | .write_u64 = cpuusage_write, | |
d842de87 | 9017 | }, |
e9515c3c KC |
9018 | { |
9019 | .name = "usage_percpu", | |
9020 | .read_seq_string = cpuacct_percpu_seq_read, | |
9021 | }, | |
ef12fefa BR |
9022 | { |
9023 | .name = "stat", | |
9024 | .read_map = cpuacct_stats_show, | |
9025 | }, | |
d842de87 SV |
9026 | }; |
9027 | ||
32cd756a | 9028 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9029 | { |
32cd756a | 9030 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9031 | } |
9032 | ||
9033 | /* | |
9034 | * charge this task's execution time to its accounting group. | |
9035 | * | |
9036 | * called with rq->lock held. | |
9037 | */ | |
9038 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9039 | { | |
9040 | struct cpuacct *ca; | |
934352f2 | 9041 | int cpu; |
d842de87 | 9042 | |
c40c6f85 | 9043 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9044 | return; |
9045 | ||
934352f2 | 9046 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9047 | |
9048 | rcu_read_lock(); | |
9049 | ||
d842de87 | 9050 | ca = task_ca(tsk); |
d842de87 | 9051 | |
934352f2 | 9052 | for (; ca; ca = ca->parent) { |
b36128c8 | 9053 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9054 | *cpuusage += cputime; |
9055 | } | |
a18b83b7 BR |
9056 | |
9057 | rcu_read_unlock(); | |
d842de87 SV |
9058 | } |
9059 | ||
fa535a77 AB |
9060 | /* |
9061 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9062 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9063 | * percpu_counter_add with values large enough to always overflow the | |
9064 | * per cpu batch limit causing bad SMP scalability. | |
9065 | * | |
9066 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9067 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9068 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9069 | */ | |
9070 | #ifdef CONFIG_SMP | |
9071 | #define CPUACCT_BATCH \ | |
9072 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9073 | #else | |
9074 | #define CPUACCT_BATCH 0 | |
9075 | #endif | |
9076 | ||
ef12fefa BR |
9077 | /* |
9078 | * Charge the system/user time to the task's accounting group. | |
9079 | */ | |
9080 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9081 | enum cpuacct_stat_index idx, cputime_t val) | |
9082 | { | |
9083 | struct cpuacct *ca; | |
fa535a77 | 9084 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9085 | |
9086 | if (unlikely(!cpuacct_subsys.active)) | |
9087 | return; | |
9088 | ||
9089 | rcu_read_lock(); | |
9090 | ca = task_ca(tsk); | |
9091 | ||
9092 | do { | |
fa535a77 | 9093 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9094 | ca = ca->parent; |
9095 | } while (ca); | |
9096 | rcu_read_unlock(); | |
9097 | } | |
9098 | ||
d842de87 SV |
9099 | struct cgroup_subsys cpuacct_subsys = { |
9100 | .name = "cpuacct", | |
9101 | .create = cpuacct_create, | |
9102 | .destroy = cpuacct_destroy, | |
9103 | .populate = cpuacct_populate, | |
9104 | .subsys_id = cpuacct_subsys_id, | |
9105 | }; | |
9106 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9107 |