Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
969c7921 | 59 | #include <linux/stop_machine.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 PZ |
70 | #include <linux/debugfs.h> |
71 | #include <linux/ctype.h> | |
6cd8a4bb | 72 | #include <linux/ftrace.h> |
5a0e3ad6 | 73 | #include <linux/slab.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
335d7afb | 77 | #include <asm/mutex.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
5091faa4 | 81 | #include "sched_autogroup.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee HC |
233 | /* |
234 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
235 | * detach_destroy_domains and partition_sched_domains. | |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
5091faa4 MG |
274 | |
275 | #ifdef CONFIG_SCHED_AUTOGROUP | |
276 | struct autogroup *autogroup; | |
277 | #endif | |
29f59db3 SV |
278 | }; |
279 | ||
3d4b47b4 | 280 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 281 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 282 | |
e9036b36 CG |
283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
284 | ||
07e06b01 | 285 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 286 | |
cb4ad1ff | 287 | /* |
2e084786 LJ |
288 | * A weight of 0 or 1 can cause arithmetics problems. |
289 | * A weight of a cfs_rq is the sum of weights of which entities | |
290 | * are queued on this cfs_rq, so a weight of a entity should not be | |
291 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
292 | * (The default weight is 1024 - so there's no practical |
293 | * limitation from this.) | |
294 | */ | |
18d95a28 | 295 | #define MIN_SHARES 2 |
2e084786 | 296 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 297 | |
07e06b01 | 298 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
299 | #endif |
300 | ||
29f59db3 | 301 | /* Default task group. |
3a252015 | 302 | * Every task in system belong to this group at bootup. |
29f59db3 | 303 | */ |
07e06b01 | 304 | struct task_group root_task_group; |
29f59db3 | 305 | |
7c941438 | 306 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 307 | |
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
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; | |
c6c4927b RR |
420 | cpumask_var_t span; |
421 | cpumask_var_t online; | |
637f5085 | 422 | |
0eab9146 | 423 | /* |
637f5085 GH |
424 | * The "RT overload" flag: it gets set if a CPU has more than |
425 | * one runnable RT task. | |
426 | */ | |
c6c4927b | 427 | cpumask_var_t rto_mask; |
0eab9146 | 428 | atomic_t rto_count; |
6e0534f2 | 429 | struct cpupri cpupri; |
57d885fe GH |
430 | }; |
431 | ||
dc938520 GH |
432 | /* |
433 | * By default the system creates a single root-domain with all cpus as | |
434 | * members (mimicking the global state we have today). | |
435 | */ | |
57d885fe GH |
436 | static struct root_domain def_root_domain; |
437 | ||
ed2d372c | 438 | #endif /* CONFIG_SMP */ |
57d885fe | 439 | |
1da177e4 LT |
440 | /* |
441 | * This is the main, per-CPU runqueue data structure. | |
442 | * | |
443 | * Locking rule: those places that want to lock multiple runqueues | |
444 | * (such as the load balancing or the thread migration code), lock | |
445 | * acquire operations must be ordered by ascending &runqueue. | |
446 | */ | |
70b97a7f | 447 | struct rq { |
d8016491 | 448 | /* runqueue lock: */ |
05fa785c | 449 | raw_spinlock_t lock; |
1da177e4 LT |
450 | |
451 | /* | |
452 | * nr_running and cpu_load should be in the same cacheline because | |
453 | * remote CPUs use both these fields when doing load calculation. | |
454 | */ | |
455 | unsigned long nr_running; | |
6aa645ea IM |
456 | #define CPU_LOAD_IDX_MAX 5 |
457 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 458 | unsigned long last_load_update_tick; |
46cb4b7c | 459 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 460 | u64 nohz_stamp; |
83cd4fe2 | 461 | unsigned char nohz_balance_kick; |
46cb4b7c | 462 | #endif |
a64692a3 MG |
463 | unsigned int skip_clock_update; |
464 | ||
d8016491 IM |
465 | /* capture load from *all* tasks on this cpu: */ |
466 | struct load_weight load; | |
6aa645ea IM |
467 | unsigned long nr_load_updates; |
468 | u64 nr_switches; | |
469 | ||
470 | struct cfs_rq cfs; | |
6f505b16 | 471 | struct rt_rq rt; |
6f505b16 | 472 | |
6aa645ea | 473 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
474 | /* list of leaf cfs_rq on this cpu: */ |
475 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
476 | #endif |
477 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 478 | struct list_head leaf_rt_rq_list; |
1da177e4 | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * This is part of a global counter where only the total sum | |
483 | * over all CPUs matters. A task can increase this counter on | |
484 | * one CPU and if it got migrated afterwards it may decrease | |
485 | * it on another CPU. Always updated under the runqueue lock: | |
486 | */ | |
487 | unsigned long nr_uninterruptible; | |
488 | ||
34f971f6 | 489 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 490 | unsigned long next_balance; |
1da177e4 | 491 | struct mm_struct *prev_mm; |
6aa645ea | 492 | |
3e51f33f | 493 | u64 clock; |
305e6835 | 494 | u64 clock_task; |
6aa645ea | 495 | |
1da177e4 LT |
496 | atomic_t nr_iowait; |
497 | ||
498 | #ifdef CONFIG_SMP | |
0eab9146 | 499 | struct root_domain *rd; |
1da177e4 LT |
500 | struct sched_domain *sd; |
501 | ||
e51fd5e2 PZ |
502 | unsigned long cpu_power; |
503 | ||
a0a522ce | 504 | unsigned char idle_at_tick; |
1da177e4 | 505 | /* For active balancing */ |
3f029d3c | 506 | int post_schedule; |
1da177e4 LT |
507 | int active_balance; |
508 | int push_cpu; | |
969c7921 | 509 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
510 | /* cpu of this runqueue: */ |
511 | int cpu; | |
1f11eb6a | 512 | int online; |
1da177e4 | 513 | |
a8a51d5e | 514 | unsigned long avg_load_per_task; |
1da177e4 | 515 | |
e9e9250b PZ |
516 | u64 rt_avg; |
517 | u64 age_stamp; | |
1b9508f6 MG |
518 | u64 idle_stamp; |
519 | u64 avg_idle; | |
1da177e4 LT |
520 | #endif |
521 | ||
aa483808 VP |
522 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
523 | u64 prev_irq_time; | |
524 | #endif | |
525 | ||
dce48a84 TG |
526 | /* calc_load related fields */ |
527 | unsigned long calc_load_update; | |
528 | long calc_load_active; | |
529 | ||
8f4d37ec | 530 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
531 | #ifdef CONFIG_SMP |
532 | int hrtick_csd_pending; | |
533 | struct call_single_data hrtick_csd; | |
534 | #endif | |
8f4d37ec PZ |
535 | struct hrtimer hrtick_timer; |
536 | #endif | |
537 | ||
1da177e4 LT |
538 | #ifdef CONFIG_SCHEDSTATS |
539 | /* latency stats */ | |
540 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
541 | unsigned long long rq_cpu_time; |
542 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
543 | |
544 | /* sys_sched_yield() stats */ | |
480b9434 | 545 | unsigned int yld_count; |
1da177e4 LT |
546 | |
547 | /* schedule() stats */ | |
480b9434 KC |
548 | unsigned int sched_switch; |
549 | unsigned int sched_count; | |
550 | unsigned int sched_goidle; | |
1da177e4 LT |
551 | |
552 | /* try_to_wake_up() stats */ | |
480b9434 KC |
553 | unsigned int ttwu_count; |
554 | unsigned int ttwu_local; | |
1da177e4 LT |
555 | #endif |
556 | }; | |
557 | ||
f34e3b61 | 558 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 559 | |
a64692a3 | 560 | |
1e5a7405 | 561 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 562 | |
0a2966b4 CL |
563 | static inline int cpu_of(struct rq *rq) |
564 | { | |
565 | #ifdef CONFIG_SMP | |
566 | return rq->cpu; | |
567 | #else | |
568 | return 0; | |
569 | #endif | |
570 | } | |
571 | ||
497f0ab3 | 572 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
573 | rcu_dereference_check((p), \ |
574 | rcu_read_lock_sched_held() || \ | |
575 | lockdep_is_held(&sched_domains_mutex)) | |
576 | ||
674311d5 NP |
577 | /* |
578 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 579 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
580 | * |
581 | * The domain tree of any CPU may only be accessed from within | |
582 | * preempt-disabled sections. | |
583 | */ | |
48f24c4d | 584 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 585 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
586 | |
587 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
588 | #define this_rq() (&__get_cpu_var(runqueues)) | |
589 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
590 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 591 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 592 | |
dc61b1d6 PZ |
593 | #ifdef CONFIG_CGROUP_SCHED |
594 | ||
595 | /* | |
596 | * Return the group to which this tasks belongs. | |
597 | * | |
598 | * We use task_subsys_state_check() and extend the RCU verification | |
599 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
600 | * holds that lock for each task it moves into the cgroup. Therefore | |
601 | * by holding that lock, we pin the task to the current cgroup. | |
602 | */ | |
603 | static inline struct task_group *task_group(struct task_struct *p) | |
604 | { | |
5091faa4 | 605 | struct task_group *tg; |
dc61b1d6 PZ |
606 | struct cgroup_subsys_state *css; |
607 | ||
608 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
609 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
610 | tg = container_of(css, struct task_group, css); |
611 | ||
612 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
613 | } |
614 | ||
615 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
616 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
617 | { | |
618 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
619 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
620 | p->se.parent = task_group(p)->se[cpu]; | |
621 | #endif | |
622 | ||
623 | #ifdef CONFIG_RT_GROUP_SCHED | |
624 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
625 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
626 | #endif | |
627 | } | |
628 | ||
629 | #else /* CONFIG_CGROUP_SCHED */ | |
630 | ||
631 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
632 | static inline struct task_group *task_group(struct task_struct *p) | |
633 | { | |
634 | return NULL; | |
635 | } | |
636 | ||
637 | #endif /* CONFIG_CGROUP_SCHED */ | |
638 | ||
fe44d621 | 639 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 640 | |
fe44d621 | 641 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 642 | { |
fe44d621 | 643 | s64 delta; |
305e6835 | 644 | |
f26f9aff MG |
645 | if (rq->skip_clock_update) |
646 | return; | |
aa483808 | 647 | |
fe44d621 PZ |
648 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
649 | rq->clock += delta; | |
650 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
651 | } |
652 | ||
bf5c91ba IM |
653 | /* |
654 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
655 | */ | |
656 | #ifdef CONFIG_SCHED_DEBUG | |
657 | # define const_debug __read_mostly | |
658 | #else | |
659 | # define const_debug static const | |
660 | #endif | |
661 | ||
017730c1 | 662 | /** |
1fd06bb1 | 663 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 664 | * @cpu: the processor in question. |
017730c1 | 665 | * |
017730c1 IM |
666 | * This interface allows printk to be called with the runqueue lock |
667 | * held and know whether or not it is OK to wake up the klogd. | |
668 | */ | |
89f19f04 | 669 | int runqueue_is_locked(int cpu) |
017730c1 | 670 | { |
05fa785c | 671 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
672 | } |
673 | ||
bf5c91ba IM |
674 | /* |
675 | * Debugging: various feature bits | |
676 | */ | |
f00b45c1 PZ |
677 | |
678 | #define SCHED_FEAT(name, enabled) \ | |
679 | __SCHED_FEAT_##name , | |
680 | ||
bf5c91ba | 681 | enum { |
f00b45c1 | 682 | #include "sched_features.h" |
bf5c91ba IM |
683 | }; |
684 | ||
f00b45c1 PZ |
685 | #undef SCHED_FEAT |
686 | ||
687 | #define SCHED_FEAT(name, enabled) \ | |
688 | (1UL << __SCHED_FEAT_##name) * enabled | | |
689 | ||
bf5c91ba | 690 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
691 | #include "sched_features.h" |
692 | 0; | |
693 | ||
694 | #undef SCHED_FEAT | |
695 | ||
696 | #ifdef CONFIG_SCHED_DEBUG | |
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | #name , | |
699 | ||
983ed7a6 | 700 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
701 | #include "sched_features.h" |
702 | NULL | |
703 | }; | |
704 | ||
705 | #undef SCHED_FEAT | |
706 | ||
34f3a814 | 707 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 708 | { |
f00b45c1 PZ |
709 | int i; |
710 | ||
711 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
712 | if (!(sysctl_sched_features & (1UL << i))) |
713 | seq_puts(m, "NO_"); | |
714 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 715 | } |
34f3a814 | 716 | seq_puts(m, "\n"); |
f00b45c1 | 717 | |
34f3a814 | 718 | return 0; |
f00b45c1 PZ |
719 | } |
720 | ||
721 | static ssize_t | |
722 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
723 | size_t cnt, loff_t *ppos) | |
724 | { | |
725 | char buf[64]; | |
7740191c | 726 | char *cmp; |
f00b45c1 PZ |
727 | int neg = 0; |
728 | int i; | |
729 | ||
730 | if (cnt > 63) | |
731 | cnt = 63; | |
732 | ||
733 | if (copy_from_user(&buf, ubuf, cnt)) | |
734 | return -EFAULT; | |
735 | ||
736 | buf[cnt] = 0; | |
7740191c | 737 | cmp = strstrip(buf); |
f00b45c1 | 738 | |
524429c3 | 739 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
740 | neg = 1; |
741 | cmp += 3; | |
742 | } | |
743 | ||
744 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 745 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
746 | if (neg) |
747 | sysctl_sched_features &= ~(1UL << i); | |
748 | else | |
749 | sysctl_sched_features |= (1UL << i); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (!sched_feat_names[i]) | |
755 | return -EINVAL; | |
756 | ||
42994724 | 757 | *ppos += cnt; |
f00b45c1 PZ |
758 | |
759 | return cnt; | |
760 | } | |
761 | ||
34f3a814 LZ |
762 | static int sched_feat_open(struct inode *inode, struct file *filp) |
763 | { | |
764 | return single_open(filp, sched_feat_show, NULL); | |
765 | } | |
766 | ||
828c0950 | 767 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
768 | .open = sched_feat_open, |
769 | .write = sched_feat_write, | |
770 | .read = seq_read, | |
771 | .llseek = seq_lseek, | |
772 | .release = single_release, | |
f00b45c1 PZ |
773 | }; |
774 | ||
775 | static __init int sched_init_debug(void) | |
776 | { | |
f00b45c1 PZ |
777 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
778 | &sched_feat_fops); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | late_initcall(sched_init_debug); | |
783 | ||
784 | #endif | |
785 | ||
786 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 787 | |
b82d9fdd PZ |
788 | /* |
789 | * Number of tasks to iterate in a single balance run. | |
790 | * Limited because this is done with IRQs disabled. | |
791 | */ | |
792 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
793 | ||
e9e9250b PZ |
794 | /* |
795 | * period over which we average the RT time consumption, measured | |
796 | * in ms. | |
797 | * | |
798 | * default: 1s | |
799 | */ | |
800 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
801 | ||
fa85ae24 | 802 | /* |
9f0c1e56 | 803 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
804 | * default: 1s |
805 | */ | |
9f0c1e56 | 806 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 807 | |
6892b75e IM |
808 | static __read_mostly int scheduler_running; |
809 | ||
9f0c1e56 PZ |
810 | /* |
811 | * part of the period that we allow rt tasks to run in us. | |
812 | * default: 0.95s | |
813 | */ | |
814 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 815 | |
d0b27fa7 PZ |
816 | static inline u64 global_rt_period(void) |
817 | { | |
818 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
819 | } | |
820 | ||
821 | static inline u64 global_rt_runtime(void) | |
822 | { | |
e26873bb | 823 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
824 | return RUNTIME_INF; |
825 | ||
826 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
827 | } | |
fa85ae24 | 828 | |
1da177e4 | 829 | #ifndef prepare_arch_switch |
4866cde0 NP |
830 | # define prepare_arch_switch(next) do { } while (0) |
831 | #endif | |
832 | #ifndef finish_arch_switch | |
833 | # define finish_arch_switch(prev) do { } while (0) | |
834 | #endif | |
835 | ||
051a1d1a DA |
836 | static inline int task_current(struct rq *rq, struct task_struct *p) |
837 | { | |
838 | return rq->curr == p; | |
839 | } | |
840 | ||
4866cde0 | 841 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 842 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 843 | { |
051a1d1a | 844 | return task_current(rq, p); |
4866cde0 NP |
845 | } |
846 | ||
70b97a7f | 847 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
848 | { |
849 | } | |
850 | ||
70b97a7f | 851 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 852 | { |
da04c035 IM |
853 | #ifdef CONFIG_DEBUG_SPINLOCK |
854 | /* this is a valid case when another task releases the spinlock */ | |
855 | rq->lock.owner = current; | |
856 | #endif | |
8a25d5de IM |
857 | /* |
858 | * If we are tracking spinlock dependencies then we have to | |
859 | * fix up the runqueue lock - which gets 'carried over' from | |
860 | * prev into current: | |
861 | */ | |
862 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
863 | ||
05fa785c | 864 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
865 | } |
866 | ||
867 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 868 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
869 | { |
870 | #ifdef CONFIG_SMP | |
871 | return p->oncpu; | |
872 | #else | |
051a1d1a | 873 | return task_current(rq, p); |
4866cde0 NP |
874 | #endif |
875 | } | |
876 | ||
70b97a7f | 877 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
878 | { |
879 | #ifdef CONFIG_SMP | |
880 | /* | |
881 | * We can optimise this out completely for !SMP, because the | |
882 | * SMP rebalancing from interrupt is the only thing that cares | |
883 | * here. | |
884 | */ | |
885 | next->oncpu = 1; | |
886 | #endif | |
887 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 888 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 889 | #else |
05fa785c | 890 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
891 | #endif |
892 | } | |
893 | ||
70b97a7f | 894 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
895 | { |
896 | #ifdef CONFIG_SMP | |
897 | /* | |
898 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
899 | * We must ensure this doesn't happen until the switch is completely | |
900 | * finished. | |
901 | */ | |
902 | smp_wmb(); | |
903 | prev->oncpu = 0; | |
904 | #endif | |
905 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
906 | local_irq_enable(); | |
1da177e4 | 907 | #endif |
4866cde0 NP |
908 | } |
909 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 910 | |
0970d299 | 911 | /* |
65cc8e48 PZ |
912 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
913 | * against ttwu(). | |
0970d299 PZ |
914 | */ |
915 | static inline int task_is_waking(struct task_struct *p) | |
916 | { | |
0017d735 | 917 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
918 | } |
919 | ||
b29739f9 IM |
920 | /* |
921 | * __task_rq_lock - lock the runqueue a given task resides on. | |
922 | * Must be called interrupts disabled. | |
923 | */ | |
70b97a7f | 924 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
925 | __acquires(rq->lock) |
926 | { | |
0970d299 PZ |
927 | struct rq *rq; |
928 | ||
3a5c359a | 929 | for (;;) { |
0970d299 | 930 | rq = task_rq(p); |
05fa785c | 931 | raw_spin_lock(&rq->lock); |
65cc8e48 | 932 | if (likely(rq == task_rq(p))) |
3a5c359a | 933 | return rq; |
05fa785c | 934 | raw_spin_unlock(&rq->lock); |
b29739f9 | 935 | } |
b29739f9 IM |
936 | } |
937 | ||
1da177e4 LT |
938 | /* |
939 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 940 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
941 | * explicitly disabling preemption. |
942 | */ | |
70b97a7f | 943 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
944 | __acquires(rq->lock) |
945 | { | |
70b97a7f | 946 | struct rq *rq; |
1da177e4 | 947 | |
3a5c359a AK |
948 | for (;;) { |
949 | local_irq_save(*flags); | |
950 | rq = task_rq(p); | |
05fa785c | 951 | raw_spin_lock(&rq->lock); |
65cc8e48 | 952 | if (likely(rq == task_rq(p))) |
3a5c359a | 953 | return rq; |
05fa785c | 954 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 955 | } |
1da177e4 LT |
956 | } |
957 | ||
a9957449 | 958 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
959 | __releases(rq->lock) |
960 | { | |
05fa785c | 961 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
962 | } |
963 | ||
70b97a7f | 964 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
965 | __releases(rq->lock) |
966 | { | |
05fa785c | 967 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
968 | } |
969 | ||
1da177e4 | 970 | /* |
cc2a73b5 | 971 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 972 | */ |
a9957449 | 973 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
974 | __acquires(rq->lock) |
975 | { | |
70b97a7f | 976 | struct rq *rq; |
1da177e4 LT |
977 | |
978 | local_irq_disable(); | |
979 | rq = this_rq(); | |
05fa785c | 980 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
981 | |
982 | return rq; | |
983 | } | |
984 | ||
8f4d37ec PZ |
985 | #ifdef CONFIG_SCHED_HRTICK |
986 | /* | |
987 | * Use HR-timers to deliver accurate preemption points. | |
988 | * | |
989 | * Its all a bit involved since we cannot program an hrt while holding the | |
990 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
991 | * reschedule event. | |
992 | * | |
993 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
994 | * rq->lock. | |
995 | */ | |
8f4d37ec PZ |
996 | |
997 | /* | |
998 | * Use hrtick when: | |
999 | * - enabled by features | |
1000 | * - hrtimer is actually high res | |
1001 | */ | |
1002 | static inline int hrtick_enabled(struct rq *rq) | |
1003 | { | |
1004 | if (!sched_feat(HRTICK)) | |
1005 | return 0; | |
ba42059f | 1006 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1007 | return 0; |
8f4d37ec PZ |
1008 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1009 | } | |
1010 | ||
8f4d37ec PZ |
1011 | static void hrtick_clear(struct rq *rq) |
1012 | { | |
1013 | if (hrtimer_active(&rq->hrtick_timer)) | |
1014 | hrtimer_cancel(&rq->hrtick_timer); | |
1015 | } | |
1016 | ||
8f4d37ec PZ |
1017 | /* |
1018 | * High-resolution timer tick. | |
1019 | * Runs from hardirq context with interrupts disabled. | |
1020 | */ | |
1021 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1022 | { | |
1023 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1024 | ||
1025 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1026 | ||
05fa785c | 1027 | raw_spin_lock(&rq->lock); |
3e51f33f | 1028 | update_rq_clock(rq); |
8f4d37ec | 1029 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1030 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1031 | |
1032 | return HRTIMER_NORESTART; | |
1033 | } | |
1034 | ||
95e904c7 | 1035 | #ifdef CONFIG_SMP |
31656519 PZ |
1036 | /* |
1037 | * called from hardirq (IPI) context | |
1038 | */ | |
1039 | static void __hrtick_start(void *arg) | |
b328ca18 | 1040 | { |
31656519 | 1041 | struct rq *rq = arg; |
b328ca18 | 1042 | |
05fa785c | 1043 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1044 | hrtimer_restart(&rq->hrtick_timer); |
1045 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1046 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1047 | } |
1048 | ||
31656519 PZ |
1049 | /* |
1050 | * Called to set the hrtick timer state. | |
1051 | * | |
1052 | * called with rq->lock held and irqs disabled | |
1053 | */ | |
1054 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1055 | { |
31656519 PZ |
1056 | struct hrtimer *timer = &rq->hrtick_timer; |
1057 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1058 | |
cc584b21 | 1059 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1060 | |
1061 | if (rq == this_rq()) { | |
1062 | hrtimer_restart(timer); | |
1063 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1064 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1065 | rq->hrtick_csd_pending = 1; |
1066 | } | |
b328ca18 PZ |
1067 | } |
1068 | ||
1069 | static int | |
1070 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1071 | { | |
1072 | int cpu = (int)(long)hcpu; | |
1073 | ||
1074 | switch (action) { | |
1075 | case CPU_UP_CANCELED: | |
1076 | case CPU_UP_CANCELED_FROZEN: | |
1077 | case CPU_DOWN_PREPARE: | |
1078 | case CPU_DOWN_PREPARE_FROZEN: | |
1079 | case CPU_DEAD: | |
1080 | case CPU_DEAD_FROZEN: | |
31656519 | 1081 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1082 | return NOTIFY_OK; |
1083 | } | |
1084 | ||
1085 | return NOTIFY_DONE; | |
1086 | } | |
1087 | ||
fa748203 | 1088 | static __init void init_hrtick(void) |
b328ca18 PZ |
1089 | { |
1090 | hotcpu_notifier(hotplug_hrtick, 0); | |
1091 | } | |
31656519 PZ |
1092 | #else |
1093 | /* | |
1094 | * Called to set the hrtick timer state. | |
1095 | * | |
1096 | * called with rq->lock held and irqs disabled | |
1097 | */ | |
1098 | static void hrtick_start(struct rq *rq, u64 delay) | |
1099 | { | |
7f1e2ca9 | 1100 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1101 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1102 | } |
b328ca18 | 1103 | |
006c75f1 | 1104 | static inline void init_hrtick(void) |
8f4d37ec | 1105 | { |
8f4d37ec | 1106 | } |
31656519 | 1107 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1108 | |
31656519 | 1109 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1110 | { |
31656519 PZ |
1111 | #ifdef CONFIG_SMP |
1112 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1113 | |
31656519 PZ |
1114 | rq->hrtick_csd.flags = 0; |
1115 | rq->hrtick_csd.func = __hrtick_start; | |
1116 | rq->hrtick_csd.info = rq; | |
1117 | #endif | |
8f4d37ec | 1118 | |
31656519 PZ |
1119 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1120 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1121 | } |
006c75f1 | 1122 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1123 | static inline void hrtick_clear(struct rq *rq) |
1124 | { | |
1125 | } | |
1126 | ||
8f4d37ec PZ |
1127 | static inline void init_rq_hrtick(struct rq *rq) |
1128 | { | |
1129 | } | |
1130 | ||
b328ca18 PZ |
1131 | static inline void init_hrtick(void) |
1132 | { | |
1133 | } | |
006c75f1 | 1134 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1135 | |
c24d20db IM |
1136 | /* |
1137 | * resched_task - mark a task 'to be rescheduled now'. | |
1138 | * | |
1139 | * On UP this means the setting of the need_resched flag, on SMP it | |
1140 | * might also involve a cross-CPU call to trigger the scheduler on | |
1141 | * the target CPU. | |
1142 | */ | |
1143 | #ifdef CONFIG_SMP | |
1144 | ||
1145 | #ifndef tsk_is_polling | |
1146 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1147 | #endif | |
1148 | ||
31656519 | 1149 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1150 | { |
1151 | int cpu; | |
1152 | ||
05fa785c | 1153 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1154 | |
5ed0cec0 | 1155 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1156 | return; |
1157 | ||
5ed0cec0 | 1158 | set_tsk_need_resched(p); |
c24d20db IM |
1159 | |
1160 | cpu = task_cpu(p); | |
1161 | if (cpu == smp_processor_id()) | |
1162 | return; | |
1163 | ||
1164 | /* NEED_RESCHED must be visible before we test polling */ | |
1165 | smp_mb(); | |
1166 | if (!tsk_is_polling(p)) | |
1167 | smp_send_reschedule(cpu); | |
1168 | } | |
1169 | ||
1170 | static void resched_cpu(int cpu) | |
1171 | { | |
1172 | struct rq *rq = cpu_rq(cpu); | |
1173 | unsigned long flags; | |
1174 | ||
05fa785c | 1175 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1176 | return; |
1177 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1178 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1179 | } |
06d8308c TG |
1180 | |
1181 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1182 | /* |
1183 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1184 | * from an idle cpu. This is good for power-savings. | |
1185 | * | |
1186 | * We don't do similar optimization for completely idle system, as | |
1187 | * selecting an idle cpu will add more delays to the timers than intended | |
1188 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1189 | */ | |
1190 | int get_nohz_timer_target(void) | |
1191 | { | |
1192 | int cpu = smp_processor_id(); | |
1193 | int i; | |
1194 | struct sched_domain *sd; | |
1195 | ||
1196 | for_each_domain(cpu, sd) { | |
1197 | for_each_cpu(i, sched_domain_span(sd)) | |
1198 | if (!idle_cpu(i)) | |
1199 | return i; | |
1200 | } | |
1201 | return cpu; | |
1202 | } | |
06d8308c TG |
1203 | /* |
1204 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1205 | * idle CPU then this timer might expire before the next timer event | |
1206 | * which is scheduled to wake up that CPU. In case of a completely | |
1207 | * idle system the next event might even be infinite time into the | |
1208 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1209 | * leaves the inner idle loop so the newly added timer is taken into | |
1210 | * account when the CPU goes back to idle and evaluates the timer | |
1211 | * wheel for the next timer event. | |
1212 | */ | |
1213 | void wake_up_idle_cpu(int cpu) | |
1214 | { | |
1215 | struct rq *rq = cpu_rq(cpu); | |
1216 | ||
1217 | if (cpu == smp_processor_id()) | |
1218 | return; | |
1219 | ||
1220 | /* | |
1221 | * This is safe, as this function is called with the timer | |
1222 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1223 | * to idle and has not yet set rq->curr to idle then it will | |
1224 | * be serialized on the timer wheel base lock and take the new | |
1225 | * timer into account automatically. | |
1226 | */ | |
1227 | if (rq->curr != rq->idle) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1232 | * lockless. The worst case is that the other CPU runs the | |
1233 | * idle task through an additional NOOP schedule() | |
1234 | */ | |
5ed0cec0 | 1235 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1236 | |
1237 | /* NEED_RESCHED must be visible before we test polling */ | |
1238 | smp_mb(); | |
1239 | if (!tsk_is_polling(rq->idle)) | |
1240 | smp_send_reschedule(cpu); | |
1241 | } | |
39c0cbe2 | 1242 | |
6d6bc0ad | 1243 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1244 | |
e9e9250b PZ |
1245 | static u64 sched_avg_period(void) |
1246 | { | |
1247 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1248 | } | |
1249 | ||
1250 | static void sched_avg_update(struct rq *rq) | |
1251 | { | |
1252 | s64 period = sched_avg_period(); | |
1253 | ||
1254 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1255 | /* |
1256 | * Inline assembly required to prevent the compiler | |
1257 | * optimising this loop into a divmod call. | |
1258 | * See __iter_div_u64_rem() for another example of this. | |
1259 | */ | |
1260 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1261 | rq->age_stamp += period; |
1262 | rq->rt_avg /= 2; | |
1263 | } | |
1264 | } | |
1265 | ||
1266 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1267 | { | |
1268 | rq->rt_avg += rt_delta; | |
1269 | sched_avg_update(rq); | |
1270 | } | |
1271 | ||
6d6bc0ad | 1272 | #else /* !CONFIG_SMP */ |
31656519 | 1273 | static void resched_task(struct task_struct *p) |
c24d20db | 1274 | { |
05fa785c | 1275 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1276 | set_tsk_need_resched(p); |
c24d20db | 1277 | } |
e9e9250b PZ |
1278 | |
1279 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1280 | { | |
1281 | } | |
da2b71ed SS |
1282 | |
1283 | static void sched_avg_update(struct rq *rq) | |
1284 | { | |
1285 | } | |
6d6bc0ad | 1286 | #endif /* CONFIG_SMP */ |
c24d20db | 1287 | |
45bf76df IM |
1288 | #if BITS_PER_LONG == 32 |
1289 | # define WMULT_CONST (~0UL) | |
1290 | #else | |
1291 | # define WMULT_CONST (1UL << 32) | |
1292 | #endif | |
1293 | ||
1294 | #define WMULT_SHIFT 32 | |
1295 | ||
194081eb IM |
1296 | /* |
1297 | * Shift right and round: | |
1298 | */ | |
cf2ab469 | 1299 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1300 | |
a7be37ac PZ |
1301 | /* |
1302 | * delta *= weight / lw | |
1303 | */ | |
cb1c4fc9 | 1304 | static unsigned long |
45bf76df IM |
1305 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1306 | struct load_weight *lw) | |
1307 | { | |
1308 | u64 tmp; | |
1309 | ||
7a232e03 LJ |
1310 | if (!lw->inv_weight) { |
1311 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1312 | lw->inv_weight = 1; | |
1313 | else | |
1314 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1315 | / (lw->weight+1); | |
1316 | } | |
45bf76df IM |
1317 | |
1318 | tmp = (u64)delta_exec * weight; | |
1319 | /* | |
1320 | * Check whether we'd overflow the 64-bit multiplication: | |
1321 | */ | |
194081eb | 1322 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1323 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1324 | WMULT_SHIFT/2); |
1325 | else | |
cf2ab469 | 1326 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1327 | |
ecf691da | 1328 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1329 | } |
1330 | ||
1091985b | 1331 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1332 | { |
1333 | lw->weight += inc; | |
e89996ae | 1334 | lw->inv_weight = 0; |
45bf76df IM |
1335 | } |
1336 | ||
1091985b | 1337 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1338 | { |
1339 | lw->weight -= dec; | |
e89996ae | 1340 | lw->inv_weight = 0; |
45bf76df IM |
1341 | } |
1342 | ||
2069dd75 PZ |
1343 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1344 | { | |
1345 | lw->weight = w; | |
1346 | lw->inv_weight = 0; | |
1347 | } | |
1348 | ||
2dd73a4f PW |
1349 | /* |
1350 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1351 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1352 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1353 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1354 | * scaled version of the new time slice allocation that they receive on time |
1355 | * slice expiry etc. | |
1356 | */ | |
1357 | ||
cce7ade8 PZ |
1358 | #define WEIGHT_IDLEPRIO 3 |
1359 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1360 | |
1361 | /* | |
1362 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1363 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1364 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1365 | * that remained on nice 0. | |
1366 | * | |
1367 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1368 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1369 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1370 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1371 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1372 | */ |
1373 | static const int prio_to_weight[40] = { | |
254753dc IM |
1374 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1375 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1376 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1377 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1378 | /* 0 */ 1024, 820, 655, 526, 423, | |
1379 | /* 5 */ 335, 272, 215, 172, 137, | |
1380 | /* 10 */ 110, 87, 70, 56, 45, | |
1381 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1382 | }; |
1383 | ||
5714d2de IM |
1384 | /* |
1385 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1386 | * | |
1387 | * In cases where the weight does not change often, we can use the | |
1388 | * precalculated inverse to speed up arithmetics by turning divisions | |
1389 | * into multiplications: | |
1390 | */ | |
dd41f596 | 1391 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1392 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1393 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1394 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1395 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1396 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1397 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1398 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1399 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1400 | }; |
2dd73a4f | 1401 | |
ef12fefa BR |
1402 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1403 | enum cpuacct_stat_index { | |
1404 | CPUACCT_STAT_USER, /* ... user mode */ | |
1405 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1406 | ||
1407 | CPUACCT_STAT_NSTATS, | |
1408 | }; | |
1409 | ||
d842de87 SV |
1410 | #ifdef CONFIG_CGROUP_CPUACCT |
1411 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1412 | static void cpuacct_update_stats(struct task_struct *tsk, |
1413 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1414 | #else |
1415 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1416 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1417 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1418 | #endif |
1419 | ||
18d95a28 PZ |
1420 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1421 | { | |
1422 | update_load_add(&rq->load, load); | |
1423 | } | |
1424 | ||
1425 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1426 | { | |
1427 | update_load_sub(&rq->load, load); | |
1428 | } | |
1429 | ||
7940ca36 | 1430 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1431 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1432 | |
1433 | /* | |
1434 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1435 | * leaving it for the final time. | |
1436 | */ | |
eb755805 | 1437 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1438 | { |
1439 | struct task_group *parent, *child; | |
eb755805 | 1440 | int ret; |
c09595f6 PZ |
1441 | |
1442 | rcu_read_lock(); | |
1443 | parent = &root_task_group; | |
1444 | down: | |
eb755805 PZ |
1445 | ret = (*down)(parent, data); |
1446 | if (ret) | |
1447 | goto out_unlock; | |
c09595f6 PZ |
1448 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1449 | parent = child; | |
1450 | goto down; | |
1451 | ||
1452 | up: | |
1453 | continue; | |
1454 | } | |
eb755805 PZ |
1455 | ret = (*up)(parent, data); |
1456 | if (ret) | |
1457 | goto out_unlock; | |
c09595f6 PZ |
1458 | |
1459 | child = parent; | |
1460 | parent = parent->parent; | |
1461 | if (parent) | |
1462 | goto up; | |
eb755805 | 1463 | out_unlock: |
c09595f6 | 1464 | rcu_read_unlock(); |
eb755805 PZ |
1465 | |
1466 | return ret; | |
c09595f6 PZ |
1467 | } |
1468 | ||
eb755805 PZ |
1469 | static int tg_nop(struct task_group *tg, void *data) |
1470 | { | |
1471 | return 0; | |
c09595f6 | 1472 | } |
eb755805 PZ |
1473 | #endif |
1474 | ||
1475 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1476 | /* Used instead of source_load when we know the type == 0 */ |
1477 | static unsigned long weighted_cpuload(const int cpu) | |
1478 | { | |
1479 | return cpu_rq(cpu)->load.weight; | |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * Return a low guess at the load of a migration-source cpu weighted | |
1484 | * according to the scheduling class and "nice" value. | |
1485 | * | |
1486 | * We want to under-estimate the load of migration sources, to | |
1487 | * balance conservatively. | |
1488 | */ | |
1489 | static unsigned long source_load(int cpu, int type) | |
1490 | { | |
1491 | struct rq *rq = cpu_rq(cpu); | |
1492 | unsigned long total = weighted_cpuload(cpu); | |
1493 | ||
1494 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1495 | return total; | |
1496 | ||
1497 | return min(rq->cpu_load[type-1], total); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Return a high guess at the load of a migration-target cpu weighted | |
1502 | * according to the scheduling class and "nice" value. | |
1503 | */ | |
1504 | static unsigned long target_load(int cpu, int type) | |
1505 | { | |
1506 | struct rq *rq = cpu_rq(cpu); | |
1507 | unsigned long total = weighted_cpuload(cpu); | |
1508 | ||
1509 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1510 | return total; | |
1511 | ||
1512 | return max(rq->cpu_load[type-1], total); | |
1513 | } | |
1514 | ||
ae154be1 PZ |
1515 | static unsigned long power_of(int cpu) |
1516 | { | |
e51fd5e2 | 1517 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1518 | } |
1519 | ||
eb755805 PZ |
1520 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1521 | ||
1522 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1523 | { | |
1524 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1525 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1526 | |
4cd42620 SR |
1527 | if (nr_running) |
1528 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1529 | else |
1530 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1531 | |
1532 | return rq->avg_load_per_task; | |
1533 | } | |
1534 | ||
1535 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1536 | |
c09595f6 | 1537 | /* |
c8cba857 PZ |
1538 | * Compute the cpu's hierarchical load factor for each task group. |
1539 | * This needs to be done in a top-down fashion because the load of a child | |
1540 | * group is a fraction of its parents load. | |
c09595f6 | 1541 | */ |
eb755805 | 1542 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1543 | { |
c8cba857 | 1544 | unsigned long load; |
eb755805 | 1545 | long cpu = (long)data; |
c09595f6 | 1546 | |
c8cba857 PZ |
1547 | if (!tg->parent) { |
1548 | load = cpu_rq(cpu)->load.weight; | |
1549 | } else { | |
1550 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1551 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1552 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1553 | } | |
c09595f6 | 1554 | |
c8cba857 | 1555 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1556 | |
eb755805 | 1557 | return 0; |
c09595f6 PZ |
1558 | } |
1559 | ||
eb755805 | 1560 | static void update_h_load(long cpu) |
c09595f6 | 1561 | { |
eb755805 | 1562 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1563 | } |
1564 | ||
18d95a28 PZ |
1565 | #endif |
1566 | ||
8f45e2b5 GH |
1567 | #ifdef CONFIG_PREEMPT |
1568 | ||
b78bb868 PZ |
1569 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1570 | ||
70574a99 | 1571 | /* |
8f45e2b5 GH |
1572 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1573 | * way at the expense of forcing extra atomic operations in all | |
1574 | * invocations. This assures that the double_lock is acquired using the | |
1575 | * same underlying policy as the spinlock_t on this architecture, which | |
1576 | * reduces latency compared to the unfair variant below. However, it | |
1577 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1578 | */ |
8f45e2b5 GH |
1579 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1580 | __releases(this_rq->lock) | |
1581 | __acquires(busiest->lock) | |
1582 | __acquires(this_rq->lock) | |
1583 | { | |
05fa785c | 1584 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1585 | double_rq_lock(this_rq, busiest); |
1586 | ||
1587 | return 1; | |
1588 | } | |
1589 | ||
1590 | #else | |
1591 | /* | |
1592 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1593 | * latency by eliminating extra atomic operations when the locks are | |
1594 | * already in proper order on entry. This favors lower cpu-ids and will | |
1595 | * grant the double lock to lower cpus over higher ids under contention, | |
1596 | * regardless of entry order into the function. | |
1597 | */ | |
1598 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1599 | __releases(this_rq->lock) |
1600 | __acquires(busiest->lock) | |
1601 | __acquires(this_rq->lock) | |
1602 | { | |
1603 | int ret = 0; | |
1604 | ||
05fa785c | 1605 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1606 | if (busiest < this_rq) { |
05fa785c TG |
1607 | raw_spin_unlock(&this_rq->lock); |
1608 | raw_spin_lock(&busiest->lock); | |
1609 | raw_spin_lock_nested(&this_rq->lock, | |
1610 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1611 | ret = 1; |
1612 | } else | |
05fa785c TG |
1613 | raw_spin_lock_nested(&busiest->lock, |
1614 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1615 | } |
1616 | return ret; | |
1617 | } | |
1618 | ||
8f45e2b5 GH |
1619 | #endif /* CONFIG_PREEMPT */ |
1620 | ||
1621 | /* | |
1622 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1623 | */ | |
1624 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1625 | { | |
1626 | if (unlikely(!irqs_disabled())) { | |
1627 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1628 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1629 | BUG_ON(1); |
1630 | } | |
1631 | ||
1632 | return _double_lock_balance(this_rq, busiest); | |
1633 | } | |
1634 | ||
70574a99 AD |
1635 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1636 | __releases(busiest->lock) | |
1637 | { | |
05fa785c | 1638 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1639 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1640 | } | |
1e3c88bd PZ |
1641 | |
1642 | /* | |
1643 | * double_rq_lock - safely lock two runqueues | |
1644 | * | |
1645 | * Note this does not disable interrupts like task_rq_lock, | |
1646 | * you need to do so manually before calling. | |
1647 | */ | |
1648 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1649 | __acquires(rq1->lock) | |
1650 | __acquires(rq2->lock) | |
1651 | { | |
1652 | BUG_ON(!irqs_disabled()); | |
1653 | if (rq1 == rq2) { | |
1654 | raw_spin_lock(&rq1->lock); | |
1655 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1656 | } else { | |
1657 | if (rq1 < rq2) { | |
1658 | raw_spin_lock(&rq1->lock); | |
1659 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1660 | } else { | |
1661 | raw_spin_lock(&rq2->lock); | |
1662 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1663 | } | |
1664 | } | |
1e3c88bd PZ |
1665 | } |
1666 | ||
1667 | /* | |
1668 | * double_rq_unlock - safely unlock two runqueues | |
1669 | * | |
1670 | * Note this does not restore interrupts like task_rq_unlock, | |
1671 | * you need to do so manually after calling. | |
1672 | */ | |
1673 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1674 | __releases(rq1->lock) | |
1675 | __releases(rq2->lock) | |
1676 | { | |
1677 | raw_spin_unlock(&rq1->lock); | |
1678 | if (rq1 != rq2) | |
1679 | raw_spin_unlock(&rq2->lock); | |
1680 | else | |
1681 | __release(rq2->lock); | |
1682 | } | |
1683 | ||
d95f4122 MG |
1684 | #else /* CONFIG_SMP */ |
1685 | ||
1686 | /* | |
1687 | * double_rq_lock - safely lock two runqueues | |
1688 | * | |
1689 | * Note this does not disable interrupts like task_rq_lock, | |
1690 | * you need to do so manually before calling. | |
1691 | */ | |
1692 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1693 | __acquires(rq1->lock) | |
1694 | __acquires(rq2->lock) | |
1695 | { | |
1696 | BUG_ON(!irqs_disabled()); | |
1697 | BUG_ON(rq1 != rq2); | |
1698 | raw_spin_lock(&rq1->lock); | |
1699 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * double_rq_unlock - safely unlock two runqueues | |
1704 | * | |
1705 | * Note this does not restore interrupts like task_rq_unlock, | |
1706 | * you need to do so manually after calling. | |
1707 | */ | |
1708 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1709 | __releases(rq1->lock) | |
1710 | __releases(rq2->lock) | |
1711 | { | |
1712 | BUG_ON(rq1 != rq2); | |
1713 | raw_spin_unlock(&rq1->lock); | |
1714 | __release(rq2->lock); | |
1715 | } | |
1716 | ||
18d95a28 PZ |
1717 | #endif |
1718 | ||
74f5187a | 1719 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1720 | static void update_sysctl(void); |
acb4a848 | 1721 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1722 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1723 | |
cd29fe6f PZ |
1724 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1725 | { | |
1726 | set_task_rq(p, cpu); | |
1727 | #ifdef CONFIG_SMP | |
1728 | /* | |
1729 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1730 | * successfuly executed on another CPU. We must ensure that updates of | |
1731 | * per-task data have been completed by this moment. | |
1732 | */ | |
1733 | smp_wmb(); | |
1734 | task_thread_info(p)->cpu = cpu; | |
1735 | #endif | |
1736 | } | |
dce48a84 | 1737 | |
1e3c88bd | 1738 | static const struct sched_class rt_sched_class; |
dd41f596 | 1739 | |
34f971f6 | 1740 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1741 | #define for_each_class(class) \ |
1742 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1743 | |
1e3c88bd PZ |
1744 | #include "sched_stats.h" |
1745 | ||
c09595f6 | 1746 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1747 | { |
1748 | rq->nr_running++; | |
9c217245 IM |
1749 | } |
1750 | ||
c09595f6 | 1751 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1752 | { |
1753 | rq->nr_running--; | |
9c217245 IM |
1754 | } |
1755 | ||
45bf76df IM |
1756 | static void set_load_weight(struct task_struct *p) |
1757 | { | |
dd41f596 IM |
1758 | /* |
1759 | * SCHED_IDLE tasks get minimal weight: | |
1760 | */ | |
1761 | if (p->policy == SCHED_IDLE) { | |
1762 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1763 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1764 | return; | |
1765 | } | |
71f8bd46 | 1766 | |
dd41f596 IM |
1767 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1768 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1769 | } |
1770 | ||
371fd7e7 | 1771 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1772 | { |
a64692a3 | 1773 | update_rq_clock(rq); |
dd41f596 | 1774 | sched_info_queued(p); |
371fd7e7 | 1775 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1776 | p->se.on_rq = 1; |
71f8bd46 IM |
1777 | } |
1778 | ||
371fd7e7 | 1779 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1780 | { |
a64692a3 | 1781 | update_rq_clock(rq); |
46ac22ba | 1782 | sched_info_dequeued(p); |
371fd7e7 | 1783 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1784 | p->se.on_rq = 0; |
71f8bd46 IM |
1785 | } |
1786 | ||
1e3c88bd PZ |
1787 | /* |
1788 | * activate_task - move a task to the runqueue. | |
1789 | */ | |
371fd7e7 | 1790 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1791 | { |
1792 | if (task_contributes_to_load(p)) | |
1793 | rq->nr_uninterruptible--; | |
1794 | ||
371fd7e7 | 1795 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1796 | inc_nr_running(rq); |
1797 | } | |
1798 | ||
1799 | /* | |
1800 | * deactivate_task - remove a task from the runqueue. | |
1801 | */ | |
371fd7e7 | 1802 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1803 | { |
1804 | if (task_contributes_to_load(p)) | |
1805 | rq->nr_uninterruptible++; | |
1806 | ||
371fd7e7 | 1807 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1808 | dec_nr_running(rq); |
1809 | } | |
1810 | ||
b52bfee4 VP |
1811 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1812 | ||
305e6835 VP |
1813 | /* |
1814 | * There are no locks covering percpu hardirq/softirq time. | |
1815 | * They are only modified in account_system_vtime, on corresponding CPU | |
1816 | * with interrupts disabled. So, writes are safe. | |
1817 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1818 | * This may result in other CPU reading this CPU's irq time and can | |
1819 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1820 | * or new value with a side effect of accounting a slice of irq time to wrong |
1821 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1822 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1823 | */ |
b52bfee4 VP |
1824 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1825 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1826 | ||
1827 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1828 | static int sched_clock_irqtime; | |
1829 | ||
1830 | void enable_sched_clock_irqtime(void) | |
1831 | { | |
1832 | sched_clock_irqtime = 1; | |
1833 | } | |
1834 | ||
1835 | void disable_sched_clock_irqtime(void) | |
1836 | { | |
1837 | sched_clock_irqtime = 0; | |
1838 | } | |
1839 | ||
8e92c201 PZ |
1840 | #ifndef CONFIG_64BIT |
1841 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1842 | ||
1843 | static inline void irq_time_write_begin(void) | |
1844 | { | |
1845 | __this_cpu_inc(irq_time_seq.sequence); | |
1846 | smp_wmb(); | |
1847 | } | |
1848 | ||
1849 | static inline void irq_time_write_end(void) | |
1850 | { | |
1851 | smp_wmb(); | |
1852 | __this_cpu_inc(irq_time_seq.sequence); | |
1853 | } | |
1854 | ||
1855 | static inline u64 irq_time_read(int cpu) | |
1856 | { | |
1857 | u64 irq_time; | |
1858 | unsigned seq; | |
1859 | ||
1860 | do { | |
1861 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1862 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1863 | per_cpu(cpu_hardirq_time, cpu); | |
1864 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1865 | ||
1866 | return irq_time; | |
1867 | } | |
1868 | #else /* CONFIG_64BIT */ | |
1869 | static inline void irq_time_write_begin(void) | |
1870 | { | |
1871 | } | |
1872 | ||
1873 | static inline void irq_time_write_end(void) | |
1874 | { | |
1875 | } | |
1876 | ||
1877 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1878 | { |
305e6835 VP |
1879 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1880 | } | |
8e92c201 | 1881 | #endif /* CONFIG_64BIT */ |
305e6835 | 1882 | |
fe44d621 PZ |
1883 | /* |
1884 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1885 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1886 | */ | |
b52bfee4 VP |
1887 | void account_system_vtime(struct task_struct *curr) |
1888 | { | |
1889 | unsigned long flags; | |
fe44d621 | 1890 | s64 delta; |
b52bfee4 | 1891 | int cpu; |
b52bfee4 VP |
1892 | |
1893 | if (!sched_clock_irqtime) | |
1894 | return; | |
1895 | ||
1896 | local_irq_save(flags); | |
1897 | ||
b52bfee4 | 1898 | cpu = smp_processor_id(); |
fe44d621 PZ |
1899 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1900 | __this_cpu_add(irq_start_time, delta); | |
1901 | ||
8e92c201 | 1902 | irq_time_write_begin(); |
b52bfee4 VP |
1903 | /* |
1904 | * We do not account for softirq time from ksoftirqd here. | |
1905 | * We want to continue accounting softirq time to ksoftirqd thread | |
1906 | * in that case, so as not to confuse scheduler with a special task | |
1907 | * that do not consume any time, but still wants to run. | |
1908 | */ | |
1909 | if (hardirq_count()) | |
fe44d621 | 1910 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1911 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1912 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1913 | |
8e92c201 | 1914 | irq_time_write_end(); |
b52bfee4 VP |
1915 | local_irq_restore(flags); |
1916 | } | |
b7dadc38 | 1917 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1918 | |
fe44d621 | 1919 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1920 | { |
fe44d621 PZ |
1921 | s64 irq_delta; |
1922 | ||
8e92c201 | 1923 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1924 | |
1925 | /* | |
1926 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1927 | * this case when a previous update_rq_clock() happened inside a | |
1928 | * {soft,}irq region. | |
1929 | * | |
1930 | * When this happens, we stop ->clock_task and only update the | |
1931 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1932 | * update will consume the rest. This ensures ->clock_task is | |
1933 | * monotonic. | |
1934 | * | |
1935 | * It does however cause some slight miss-attribution of {soft,}irq | |
1936 | * time, a more accurate solution would be to update the irq_time using | |
1937 | * the current rq->clock timestamp, except that would require using | |
1938 | * atomic ops. | |
1939 | */ | |
1940 | if (irq_delta > delta) | |
1941 | irq_delta = delta; | |
1942 | ||
1943 | rq->prev_irq_time += irq_delta; | |
1944 | delta -= irq_delta; | |
1945 | rq->clock_task += delta; | |
1946 | ||
1947 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1948 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1949 | } |
1950 | ||
abb74cef VP |
1951 | static int irqtime_account_hi_update(void) |
1952 | { | |
1953 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1954 | unsigned long flags; | |
1955 | u64 latest_ns; | |
1956 | int ret = 0; | |
1957 | ||
1958 | local_irq_save(flags); | |
1959 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1960 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1961 | ret = 1; | |
1962 | local_irq_restore(flags); | |
1963 | return ret; | |
1964 | } | |
1965 | ||
1966 | static int irqtime_account_si_update(void) | |
1967 | { | |
1968 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1969 | unsigned long flags; | |
1970 | u64 latest_ns; | |
1971 | int ret = 0; | |
1972 | ||
1973 | local_irq_save(flags); | |
1974 | latest_ns = this_cpu_read(cpu_softirq_time); | |
1975 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
1976 | ret = 1; | |
1977 | local_irq_restore(flags); | |
1978 | return ret; | |
1979 | } | |
1980 | ||
fe44d621 | 1981 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1982 | |
abb74cef VP |
1983 | #define sched_clock_irqtime (0) |
1984 | ||
fe44d621 | 1985 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1986 | { |
fe44d621 | 1987 | rq->clock_task += delta; |
305e6835 VP |
1988 | } |
1989 | ||
fe44d621 | 1990 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 1991 | |
1e3c88bd PZ |
1992 | #include "sched_idletask.c" |
1993 | #include "sched_fair.c" | |
1994 | #include "sched_rt.c" | |
5091faa4 | 1995 | #include "sched_autogroup.c" |
34f971f6 | 1996 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1997 | #ifdef CONFIG_SCHED_DEBUG |
1998 | # include "sched_debug.c" | |
1999 | #endif | |
2000 | ||
34f971f6 PZ |
2001 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2002 | { | |
2003 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2004 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2005 | ||
2006 | if (stop) { | |
2007 | /* | |
2008 | * Make it appear like a SCHED_FIFO task, its something | |
2009 | * userspace knows about and won't get confused about. | |
2010 | * | |
2011 | * Also, it will make PI more or less work without too | |
2012 | * much confusion -- but then, stop work should not | |
2013 | * rely on PI working anyway. | |
2014 | */ | |
2015 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2016 | ||
2017 | stop->sched_class = &stop_sched_class; | |
2018 | } | |
2019 | ||
2020 | cpu_rq(cpu)->stop = stop; | |
2021 | ||
2022 | if (old_stop) { | |
2023 | /* | |
2024 | * Reset it back to a normal scheduling class so that | |
2025 | * it can die in pieces. | |
2026 | */ | |
2027 | old_stop->sched_class = &rt_sched_class; | |
2028 | } | |
2029 | } | |
2030 | ||
14531189 | 2031 | /* |
dd41f596 | 2032 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2033 | */ |
14531189 IM |
2034 | static inline int __normal_prio(struct task_struct *p) |
2035 | { | |
dd41f596 | 2036 | return p->static_prio; |
14531189 IM |
2037 | } |
2038 | ||
b29739f9 IM |
2039 | /* |
2040 | * Calculate the expected normal priority: i.e. priority | |
2041 | * without taking RT-inheritance into account. Might be | |
2042 | * boosted by interactivity modifiers. Changes upon fork, | |
2043 | * setprio syscalls, and whenever the interactivity | |
2044 | * estimator recalculates. | |
2045 | */ | |
36c8b586 | 2046 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2047 | { |
2048 | int prio; | |
2049 | ||
e05606d3 | 2050 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2051 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2052 | else | |
2053 | prio = __normal_prio(p); | |
2054 | return prio; | |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Calculate the current priority, i.e. the priority | |
2059 | * taken into account by the scheduler. This value might | |
2060 | * be boosted by RT tasks, or might be boosted by | |
2061 | * interactivity modifiers. Will be RT if the task got | |
2062 | * RT-boosted. If not then it returns p->normal_prio. | |
2063 | */ | |
36c8b586 | 2064 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2065 | { |
2066 | p->normal_prio = normal_prio(p); | |
2067 | /* | |
2068 | * If we are RT tasks or we were boosted to RT priority, | |
2069 | * keep the priority unchanged. Otherwise, update priority | |
2070 | * to the normal priority: | |
2071 | */ | |
2072 | if (!rt_prio(p->prio)) | |
2073 | return p->normal_prio; | |
2074 | return p->prio; | |
2075 | } | |
2076 | ||
1da177e4 LT |
2077 | /** |
2078 | * task_curr - is this task currently executing on a CPU? | |
2079 | * @p: the task in question. | |
2080 | */ | |
36c8b586 | 2081 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2082 | { |
2083 | return cpu_curr(task_cpu(p)) == p; | |
2084 | } | |
2085 | ||
cb469845 SR |
2086 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2087 | const struct sched_class *prev_class, | |
da7a735e | 2088 | int oldprio) |
cb469845 SR |
2089 | { |
2090 | if (prev_class != p->sched_class) { | |
2091 | if (prev_class->switched_from) | |
da7a735e PZ |
2092 | prev_class->switched_from(rq, p); |
2093 | p->sched_class->switched_to(rq, p); | |
2094 | } else if (oldprio != p->prio) | |
2095 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2096 | } |
2097 | ||
1e5a7405 PZ |
2098 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2099 | { | |
2100 | const struct sched_class *class; | |
2101 | ||
2102 | if (p->sched_class == rq->curr->sched_class) { | |
2103 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2104 | } else { | |
2105 | for_each_class(class) { | |
2106 | if (class == rq->curr->sched_class) | |
2107 | break; | |
2108 | if (class == p->sched_class) { | |
2109 | resched_task(rq->curr); | |
2110 | break; | |
2111 | } | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | /* | |
2116 | * A queue event has occurred, and we're going to schedule. In | |
2117 | * this case, we can save a useless back to back clock update. | |
2118 | */ | |
f26f9aff | 2119 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2120 | rq->skip_clock_update = 1; |
2121 | } | |
2122 | ||
1da177e4 | 2123 | #ifdef CONFIG_SMP |
cc367732 IM |
2124 | /* |
2125 | * Is this task likely cache-hot: | |
2126 | */ | |
e7693a36 | 2127 | static int |
cc367732 IM |
2128 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2129 | { | |
2130 | s64 delta; | |
2131 | ||
e6c8fba7 PZ |
2132 | if (p->sched_class != &fair_sched_class) |
2133 | return 0; | |
2134 | ||
ef8002f6 NR |
2135 | if (unlikely(p->policy == SCHED_IDLE)) |
2136 | return 0; | |
2137 | ||
f540a608 IM |
2138 | /* |
2139 | * Buddy candidates are cache hot: | |
2140 | */ | |
f685ceac | 2141 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2142 | (&p->se == cfs_rq_of(&p->se)->next || |
2143 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2144 | return 1; |
2145 | ||
6bc1665b IM |
2146 | if (sysctl_sched_migration_cost == -1) |
2147 | return 1; | |
2148 | if (sysctl_sched_migration_cost == 0) | |
2149 | return 0; | |
2150 | ||
cc367732 IM |
2151 | delta = now - p->se.exec_start; |
2152 | ||
2153 | return delta < (s64)sysctl_sched_migration_cost; | |
2154 | } | |
2155 | ||
dd41f596 | 2156 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2157 | { |
e2912009 PZ |
2158 | #ifdef CONFIG_SCHED_DEBUG |
2159 | /* | |
2160 | * We should never call set_task_cpu() on a blocked task, | |
2161 | * ttwu() will sort out the placement. | |
2162 | */ | |
077614ee PZ |
2163 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2164 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2165 | #endif |
2166 | ||
de1d7286 | 2167 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2168 | |
0c69774e PZ |
2169 | if (task_cpu(p) != new_cpu) { |
2170 | p->se.nr_migrations++; | |
2171 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2172 | } | |
dd41f596 IM |
2173 | |
2174 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2175 | } |
2176 | ||
969c7921 | 2177 | struct migration_arg { |
36c8b586 | 2178 | struct task_struct *task; |
1da177e4 | 2179 | int dest_cpu; |
70b97a7f | 2180 | }; |
1da177e4 | 2181 | |
969c7921 TH |
2182 | static int migration_cpu_stop(void *data); |
2183 | ||
1da177e4 LT |
2184 | /* |
2185 | * The task's runqueue lock must be held. | |
2186 | * Returns true if you have to wait for migration thread. | |
2187 | */ | |
b7a2b39d | 2188 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2189 | { |
1da177e4 LT |
2190 | /* |
2191 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2192 | * the next wake-up will properly place the task. |
1da177e4 | 2193 | */ |
969c7921 | 2194 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2195 | } |
2196 | ||
2197 | /* | |
2198 | * wait_task_inactive - wait for a thread to unschedule. | |
2199 | * | |
85ba2d86 RM |
2200 | * If @match_state is nonzero, it's the @p->state value just checked and |
2201 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2202 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2203 | * we return a positive number (its total switch count). If a second call | |
2204 | * a short while later returns the same number, the caller can be sure that | |
2205 | * @p has remained unscheduled the whole time. | |
2206 | * | |
1da177e4 LT |
2207 | * The caller must ensure that the task *will* unschedule sometime soon, |
2208 | * else this function might spin for a *long* time. This function can't | |
2209 | * be called with interrupts off, or it may introduce deadlock with | |
2210 | * smp_call_function() if an IPI is sent by the same process we are | |
2211 | * waiting to become inactive. | |
2212 | */ | |
85ba2d86 | 2213 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2214 | { |
2215 | unsigned long flags; | |
dd41f596 | 2216 | int running, on_rq; |
85ba2d86 | 2217 | unsigned long ncsw; |
70b97a7f | 2218 | struct rq *rq; |
1da177e4 | 2219 | |
3a5c359a AK |
2220 | for (;;) { |
2221 | /* | |
2222 | * We do the initial early heuristics without holding | |
2223 | * any task-queue locks at all. We'll only try to get | |
2224 | * the runqueue lock when things look like they will | |
2225 | * work out! | |
2226 | */ | |
2227 | rq = task_rq(p); | |
fa490cfd | 2228 | |
3a5c359a AK |
2229 | /* |
2230 | * If the task is actively running on another CPU | |
2231 | * still, just relax and busy-wait without holding | |
2232 | * any locks. | |
2233 | * | |
2234 | * NOTE! Since we don't hold any locks, it's not | |
2235 | * even sure that "rq" stays as the right runqueue! | |
2236 | * But we don't care, since "task_running()" will | |
2237 | * return false if the runqueue has changed and p | |
2238 | * is actually now running somewhere else! | |
2239 | */ | |
85ba2d86 RM |
2240 | while (task_running(rq, p)) { |
2241 | if (match_state && unlikely(p->state != match_state)) | |
2242 | return 0; | |
3a5c359a | 2243 | cpu_relax(); |
85ba2d86 | 2244 | } |
fa490cfd | 2245 | |
3a5c359a AK |
2246 | /* |
2247 | * Ok, time to look more closely! We need the rq | |
2248 | * lock now, to be *sure*. If we're wrong, we'll | |
2249 | * just go back and repeat. | |
2250 | */ | |
2251 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2252 | trace_sched_wait_task(p); |
3a5c359a AK |
2253 | running = task_running(rq, p); |
2254 | on_rq = p->se.on_rq; | |
85ba2d86 | 2255 | ncsw = 0; |
f31e11d8 | 2256 | if (!match_state || p->state == match_state) |
93dcf55f | 2257 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2258 | task_rq_unlock(rq, &flags); |
fa490cfd | 2259 | |
85ba2d86 RM |
2260 | /* |
2261 | * If it changed from the expected state, bail out now. | |
2262 | */ | |
2263 | if (unlikely(!ncsw)) | |
2264 | break; | |
2265 | ||
3a5c359a AK |
2266 | /* |
2267 | * Was it really running after all now that we | |
2268 | * checked with the proper locks actually held? | |
2269 | * | |
2270 | * Oops. Go back and try again.. | |
2271 | */ | |
2272 | if (unlikely(running)) { | |
2273 | cpu_relax(); | |
2274 | continue; | |
2275 | } | |
fa490cfd | 2276 | |
3a5c359a AK |
2277 | /* |
2278 | * It's not enough that it's not actively running, | |
2279 | * it must be off the runqueue _entirely_, and not | |
2280 | * preempted! | |
2281 | * | |
80dd99b3 | 2282 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2283 | * running right now), it's preempted, and we should |
2284 | * yield - it could be a while. | |
2285 | */ | |
2286 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2287 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2288 | ||
2289 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2290 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2291 | continue; |
2292 | } | |
fa490cfd | 2293 | |
3a5c359a AK |
2294 | /* |
2295 | * Ahh, all good. It wasn't running, and it wasn't | |
2296 | * runnable, which means that it will never become | |
2297 | * running in the future either. We're all done! | |
2298 | */ | |
2299 | break; | |
2300 | } | |
85ba2d86 RM |
2301 | |
2302 | return ncsw; | |
1da177e4 LT |
2303 | } |
2304 | ||
2305 | /*** | |
2306 | * kick_process - kick a running thread to enter/exit the kernel | |
2307 | * @p: the to-be-kicked thread | |
2308 | * | |
2309 | * Cause a process which is running on another CPU to enter | |
2310 | * kernel-mode, without any delay. (to get signals handled.) | |
2311 | * | |
2312 | * NOTE: this function doesnt have to take the runqueue lock, | |
2313 | * because all it wants to ensure is that the remote task enters | |
2314 | * the kernel. If the IPI races and the task has been migrated | |
2315 | * to another CPU then no harm is done and the purpose has been | |
2316 | * achieved as well. | |
2317 | */ | |
36c8b586 | 2318 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2319 | { |
2320 | int cpu; | |
2321 | ||
2322 | preempt_disable(); | |
2323 | cpu = task_cpu(p); | |
2324 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2325 | smp_send_reschedule(cpu); | |
2326 | preempt_enable(); | |
2327 | } | |
b43e3521 | 2328 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2329 | #endif /* CONFIG_SMP */ |
1da177e4 | 2330 | |
970b13ba | 2331 | #ifdef CONFIG_SMP |
30da688e ON |
2332 | /* |
2333 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2334 | */ | |
5da9a0fb PZ |
2335 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2336 | { | |
2337 | int dest_cpu; | |
2338 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2339 | ||
2340 | /* Look for allowed, online CPU in same node. */ | |
2341 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2342 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2343 | return dest_cpu; | |
2344 | ||
2345 | /* Any allowed, online CPU? */ | |
2346 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2347 | if (dest_cpu < nr_cpu_ids) | |
2348 | return dest_cpu; | |
2349 | ||
2350 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2351 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2352 | /* | |
2353 | * Don't tell them about moving exiting tasks or | |
2354 | * kernel threads (both mm NULL), since they never | |
2355 | * leave kernel. | |
2356 | */ | |
2357 | if (p->mm && printk_ratelimit()) { | |
2358 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2359 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2360 | } |
2361 | ||
2362 | return dest_cpu; | |
2363 | } | |
2364 | ||
e2912009 | 2365 | /* |
30da688e | 2366 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2367 | */ |
970b13ba | 2368 | static inline |
0017d735 | 2369 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2370 | { |
0017d735 | 2371 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2372 | |
2373 | /* | |
2374 | * In order not to call set_task_cpu() on a blocking task we need | |
2375 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2376 | * cpu. | |
2377 | * | |
2378 | * Since this is common to all placement strategies, this lives here. | |
2379 | * | |
2380 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2381 | * not worry about this generic constraint ] | |
2382 | */ | |
2383 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2384 | !cpu_online(cpu))) |
5da9a0fb | 2385 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2386 | |
2387 | return cpu; | |
970b13ba | 2388 | } |
09a40af5 MG |
2389 | |
2390 | static void update_avg(u64 *avg, u64 sample) | |
2391 | { | |
2392 | s64 diff = sample - *avg; | |
2393 | *avg += diff >> 3; | |
2394 | } | |
970b13ba PZ |
2395 | #endif |
2396 | ||
9ed3811a TH |
2397 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2398 | bool is_sync, bool is_migrate, bool is_local, | |
2399 | unsigned long en_flags) | |
2400 | { | |
2401 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2402 | if (is_sync) | |
2403 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2404 | if (is_migrate) | |
2405 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2406 | if (is_local) | |
2407 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2408 | else | |
2409 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2410 | ||
2411 | activate_task(rq, p, en_flags); | |
2412 | } | |
2413 | ||
2414 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2415 | int wake_flags, bool success) | |
2416 | { | |
2417 | trace_sched_wakeup(p, success); | |
2418 | check_preempt_curr(rq, p, wake_flags); | |
2419 | ||
2420 | p->state = TASK_RUNNING; | |
2421 | #ifdef CONFIG_SMP | |
2422 | if (p->sched_class->task_woken) | |
2423 | p->sched_class->task_woken(rq, p); | |
2424 | ||
2425 | if (unlikely(rq->idle_stamp)) { | |
2426 | u64 delta = rq->clock - rq->idle_stamp; | |
2427 | u64 max = 2*sysctl_sched_migration_cost; | |
2428 | ||
2429 | if (delta > max) | |
2430 | rq->avg_idle = max; | |
2431 | else | |
2432 | update_avg(&rq->avg_idle, delta); | |
2433 | rq->idle_stamp = 0; | |
2434 | } | |
2435 | #endif | |
21aa9af0 TH |
2436 | /* if a worker is waking up, notify workqueue */ |
2437 | if ((p->flags & PF_WQ_WORKER) && success) | |
2438 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2439 | } |
2440 | ||
2441 | /** | |
1da177e4 | 2442 | * try_to_wake_up - wake up a thread |
9ed3811a | 2443 | * @p: the thread to be awakened |
1da177e4 | 2444 | * @state: the mask of task states that can be woken |
9ed3811a | 2445 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2446 | * |
2447 | * Put it on the run-queue if it's not already there. The "current" | |
2448 | * thread is always on the run-queue (except when the actual | |
2449 | * re-schedule is in progress), and as such you're allowed to do | |
2450 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2451 | * runnable without the overhead of this. | |
2452 | * | |
9ed3811a TH |
2453 | * Returns %true if @p was woken up, %false if it was already running |
2454 | * or @state didn't match @p's state. | |
1da177e4 | 2455 | */ |
7d478721 PZ |
2456 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2457 | int wake_flags) | |
1da177e4 | 2458 | { |
cc367732 | 2459 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2460 | unsigned long flags; |
371fd7e7 | 2461 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2462 | struct rq *rq; |
1da177e4 | 2463 | |
e9c84311 | 2464 | this_cpu = get_cpu(); |
2398f2c6 | 2465 | |
04e2f174 | 2466 | smp_wmb(); |
ab3b3aa5 | 2467 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2468 | if (!(p->state & state)) |
1da177e4 LT |
2469 | goto out; |
2470 | ||
dd41f596 | 2471 | if (p->se.on_rq) |
1da177e4 LT |
2472 | goto out_running; |
2473 | ||
2474 | cpu = task_cpu(p); | |
cc367732 | 2475 | orig_cpu = cpu; |
1da177e4 LT |
2476 | |
2477 | #ifdef CONFIG_SMP | |
2478 | if (unlikely(task_running(rq, p))) | |
2479 | goto out_activate; | |
2480 | ||
e9c84311 PZ |
2481 | /* |
2482 | * In order to handle concurrent wakeups and release the rq->lock | |
2483 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2484 | * |
2485 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2486 | */ |
cc87f76a PZ |
2487 | if (task_contributes_to_load(p)) { |
2488 | if (likely(cpu_online(orig_cpu))) | |
2489 | rq->nr_uninterruptible--; | |
2490 | else | |
2491 | this_rq()->nr_uninterruptible--; | |
2492 | } | |
e9c84311 | 2493 | p->state = TASK_WAKING; |
efbbd05a | 2494 | |
371fd7e7 | 2495 | if (p->sched_class->task_waking) { |
efbbd05a | 2496 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2497 | en_flags |= ENQUEUE_WAKING; |
2498 | } | |
efbbd05a | 2499 | |
0017d735 PZ |
2500 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2501 | if (cpu != orig_cpu) | |
5d2f5a61 | 2502 | set_task_cpu(p, cpu); |
0017d735 | 2503 | __task_rq_unlock(rq); |
ab19cb23 | 2504 | |
0970d299 PZ |
2505 | rq = cpu_rq(cpu); |
2506 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2507 | |
0970d299 PZ |
2508 | /* |
2509 | * We migrated the task without holding either rq->lock, however | |
2510 | * since the task is not on the task list itself, nobody else | |
2511 | * will try and migrate the task, hence the rq should match the | |
2512 | * cpu we just moved it to. | |
2513 | */ | |
2514 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2515 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2516 | |
e7693a36 GH |
2517 | #ifdef CONFIG_SCHEDSTATS |
2518 | schedstat_inc(rq, ttwu_count); | |
2519 | if (cpu == this_cpu) | |
2520 | schedstat_inc(rq, ttwu_local); | |
2521 | else { | |
2522 | struct sched_domain *sd; | |
2523 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2524 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2525 | schedstat_inc(sd, ttwu_wake_remote); |
2526 | break; | |
2527 | } | |
2528 | } | |
2529 | } | |
6d6bc0ad | 2530 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2531 | |
1da177e4 LT |
2532 | out_activate: |
2533 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2534 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2535 | cpu == this_cpu, en_flags); | |
1da177e4 | 2536 | success = 1; |
1da177e4 | 2537 | out_running: |
9ed3811a | 2538 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2539 | out: |
2540 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2541 | put_cpu(); |
1da177e4 LT |
2542 | |
2543 | return success; | |
2544 | } | |
2545 | ||
21aa9af0 TH |
2546 | /** |
2547 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2548 | * @p: the thread to be awakened | |
2549 | * | |
b595076a | 2550 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 TH |
2551 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2552 | * the current task. this_rq() stays locked over invocation. | |
2553 | */ | |
2554 | static void try_to_wake_up_local(struct task_struct *p) | |
2555 | { | |
2556 | struct rq *rq = task_rq(p); | |
2557 | bool success = false; | |
2558 | ||
2559 | BUG_ON(rq != this_rq()); | |
2560 | BUG_ON(p == current); | |
2561 | lockdep_assert_held(&rq->lock); | |
2562 | ||
2563 | if (!(p->state & TASK_NORMAL)) | |
2564 | return; | |
2565 | ||
2566 | if (!p->se.on_rq) { | |
2567 | if (likely(!task_running(rq, p))) { | |
2568 | schedstat_inc(rq, ttwu_count); | |
2569 | schedstat_inc(rq, ttwu_local); | |
2570 | } | |
2571 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2572 | success = true; | |
2573 | } | |
2574 | ttwu_post_activation(p, rq, 0, success); | |
2575 | } | |
2576 | ||
50fa610a DH |
2577 | /** |
2578 | * wake_up_process - Wake up a specific process | |
2579 | * @p: The process to be woken up. | |
2580 | * | |
2581 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2582 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2583 | * running. | |
2584 | * | |
2585 | * It may be assumed that this function implies a write memory barrier before | |
2586 | * changing the task state if and only if any tasks are woken up. | |
2587 | */ | |
7ad5b3a5 | 2588 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2589 | { |
d9514f6c | 2590 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2591 | } |
1da177e4 LT |
2592 | EXPORT_SYMBOL(wake_up_process); |
2593 | ||
7ad5b3a5 | 2594 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2595 | { |
2596 | return try_to_wake_up(p, state, 0); | |
2597 | } | |
2598 | ||
1da177e4 LT |
2599 | /* |
2600 | * Perform scheduler related setup for a newly forked process p. | |
2601 | * p is forked by current. | |
dd41f596 IM |
2602 | * |
2603 | * __sched_fork() is basic setup used by init_idle() too: | |
2604 | */ | |
2605 | static void __sched_fork(struct task_struct *p) | |
2606 | { | |
dd41f596 IM |
2607 | p->se.exec_start = 0; |
2608 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2609 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2610 | p->se.nr_migrations = 0; |
da7a735e | 2611 | p->se.vruntime = 0; |
6cfb0d5d IM |
2612 | |
2613 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2614 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2615 | #endif |
476d139c | 2616 | |
fa717060 | 2617 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2618 | p->se.on_rq = 0; |
4a55bd5e | 2619 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2620 | |
e107be36 AK |
2621 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2622 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2623 | #endif | |
dd41f596 IM |
2624 | } |
2625 | ||
2626 | /* | |
2627 | * fork()/clone()-time setup: | |
2628 | */ | |
2629 | void sched_fork(struct task_struct *p, int clone_flags) | |
2630 | { | |
2631 | int cpu = get_cpu(); | |
2632 | ||
2633 | __sched_fork(p); | |
06b83b5f | 2634 | /* |
0017d735 | 2635 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2636 | * nobody will actually run it, and a signal or other external |
2637 | * event cannot wake it up and insert it on the runqueue either. | |
2638 | */ | |
0017d735 | 2639 | p->state = TASK_RUNNING; |
dd41f596 | 2640 | |
b9dc29e7 MG |
2641 | /* |
2642 | * Revert to default priority/policy on fork if requested. | |
2643 | */ | |
2644 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2645 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2646 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2647 | p->normal_prio = p->static_prio; |
2648 | } | |
b9dc29e7 | 2649 | |
6c697bdf MG |
2650 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2651 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2652 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2653 | set_load_weight(p); |
2654 | } | |
2655 | ||
b9dc29e7 MG |
2656 | /* |
2657 | * We don't need the reset flag anymore after the fork. It has | |
2658 | * fulfilled its duty: | |
2659 | */ | |
2660 | p->sched_reset_on_fork = 0; | |
2661 | } | |
ca94c442 | 2662 | |
f83f9ac2 PW |
2663 | /* |
2664 | * Make sure we do not leak PI boosting priority to the child. | |
2665 | */ | |
2666 | p->prio = current->normal_prio; | |
2667 | ||
2ddbf952 HS |
2668 | if (!rt_prio(p->prio)) |
2669 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2670 | |
cd29fe6f PZ |
2671 | if (p->sched_class->task_fork) |
2672 | p->sched_class->task_fork(p); | |
2673 | ||
86951599 PZ |
2674 | /* |
2675 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2676 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2677 | * is ran before sched_fork(). | |
2678 | * | |
2679 | * Silence PROVE_RCU. | |
2680 | */ | |
2681 | rcu_read_lock(); | |
5f3edc1b | 2682 | set_task_cpu(p, cpu); |
86951599 | 2683 | rcu_read_unlock(); |
5f3edc1b | 2684 | |
52f17b6c | 2685 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2686 | if (likely(sched_info_on())) |
52f17b6c | 2687 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2688 | #endif |
d6077cb8 | 2689 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2690 | p->oncpu = 0; |
2691 | #endif | |
1da177e4 | 2692 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2693 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2694 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2695 | #endif |
806c09a7 | 2696 | #ifdef CONFIG_SMP |
917b627d | 2697 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2698 | #endif |
917b627d | 2699 | |
476d139c | 2700 | put_cpu(); |
1da177e4 LT |
2701 | } |
2702 | ||
2703 | /* | |
2704 | * wake_up_new_task - wake up a newly created task for the first time. | |
2705 | * | |
2706 | * This function will do some initial scheduler statistics housekeeping | |
2707 | * that must be done for every newly created context, then puts the task | |
2708 | * on the runqueue and wakes it. | |
2709 | */ | |
7ad5b3a5 | 2710 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2711 | { |
2712 | unsigned long flags; | |
dd41f596 | 2713 | struct rq *rq; |
c890692b | 2714 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2715 | |
2716 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2717 | rq = task_rq_lock(p, &flags); |
2718 | p->state = TASK_WAKING; | |
2719 | ||
fabf318e PZ |
2720 | /* |
2721 | * Fork balancing, do it here and not earlier because: | |
2722 | * - cpus_allowed can change in the fork path | |
2723 | * - any previously selected cpu might disappear through hotplug | |
2724 | * | |
0017d735 PZ |
2725 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2726 | * without people poking at ->cpus_allowed. | |
fabf318e | 2727 | */ |
0017d735 | 2728 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2729 | set_task_cpu(p, cpu); |
1da177e4 | 2730 | |
06b83b5f | 2731 | p->state = TASK_RUNNING; |
0017d735 PZ |
2732 | task_rq_unlock(rq, &flags); |
2733 | #endif | |
2734 | ||
2735 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2736 | activate_task(rq, p, 0); |
27a9da65 | 2737 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2738 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2739 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2740 | if (p->sched_class->task_woken) |
2741 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2742 | #endif |
dd41f596 | 2743 | task_rq_unlock(rq, &flags); |
fabf318e | 2744 | put_cpu(); |
1da177e4 LT |
2745 | } |
2746 | ||
e107be36 AK |
2747 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2748 | ||
2749 | /** | |
80dd99b3 | 2750 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2751 | * @notifier: notifier struct to register |
e107be36 AK |
2752 | */ |
2753 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2754 | { | |
2755 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2756 | } | |
2757 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2758 | ||
2759 | /** | |
2760 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2761 | * @notifier: notifier struct to unregister |
e107be36 AK |
2762 | * |
2763 | * This is safe to call from within a preemption notifier. | |
2764 | */ | |
2765 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2766 | { | |
2767 | hlist_del(¬ifier->link); | |
2768 | } | |
2769 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2770 | ||
2771 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2772 | { | |
2773 | struct preempt_notifier *notifier; | |
2774 | struct hlist_node *node; | |
2775 | ||
2776 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2777 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2778 | } | |
2779 | ||
2780 | static void | |
2781 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2782 | struct task_struct *next) | |
2783 | { | |
2784 | struct preempt_notifier *notifier; | |
2785 | struct hlist_node *node; | |
2786 | ||
2787 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2788 | notifier->ops->sched_out(notifier, next); | |
2789 | } | |
2790 | ||
6d6bc0ad | 2791 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2792 | |
2793 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2794 | { | |
2795 | } | |
2796 | ||
2797 | static void | |
2798 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2799 | struct task_struct *next) | |
2800 | { | |
2801 | } | |
2802 | ||
6d6bc0ad | 2803 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2804 | |
4866cde0 NP |
2805 | /** |
2806 | * prepare_task_switch - prepare to switch tasks | |
2807 | * @rq: the runqueue preparing to switch | |
421cee29 | 2808 | * @prev: the current task that is being switched out |
4866cde0 NP |
2809 | * @next: the task we are going to switch to. |
2810 | * | |
2811 | * This is called with the rq lock held and interrupts off. It must | |
2812 | * be paired with a subsequent finish_task_switch after the context | |
2813 | * switch. | |
2814 | * | |
2815 | * prepare_task_switch sets up locking and calls architecture specific | |
2816 | * hooks. | |
2817 | */ | |
e107be36 AK |
2818 | static inline void |
2819 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2820 | struct task_struct *next) | |
4866cde0 | 2821 | { |
fe4b04fa PZ |
2822 | sched_info_switch(prev, next); |
2823 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2824 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2825 | prepare_lock_switch(rq, next); |
2826 | prepare_arch_switch(next); | |
fe4b04fa | 2827 | trace_sched_switch(prev, next); |
4866cde0 NP |
2828 | } |
2829 | ||
1da177e4 LT |
2830 | /** |
2831 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2832 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2833 | * @prev: the thread we just switched away from. |
2834 | * | |
4866cde0 NP |
2835 | * finish_task_switch must be called after the context switch, paired |
2836 | * with a prepare_task_switch call before the context switch. | |
2837 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2838 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2839 | * |
2840 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2841 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2842 | * with the lock held can cause deadlocks; see schedule() for |
2843 | * details.) | |
2844 | */ | |
a9957449 | 2845 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2846 | __releases(rq->lock) |
2847 | { | |
1da177e4 | 2848 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2849 | long prev_state; |
1da177e4 LT |
2850 | |
2851 | rq->prev_mm = NULL; | |
2852 | ||
2853 | /* | |
2854 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2855 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2856 | * schedule one last time. The schedule call will never return, and |
2857 | * the scheduled task must drop that reference. | |
c394cc9f | 2858 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2859 | * still held, otherwise prev could be scheduled on another cpu, die |
2860 | * there before we look at prev->state, and then the reference would | |
2861 | * be dropped twice. | |
2862 | * Manfred Spraul <manfred@colorfullife.com> | |
2863 | */ | |
55a101f8 | 2864 | prev_state = prev->state; |
4866cde0 | 2865 | finish_arch_switch(prev); |
8381f65d JI |
2866 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2867 | local_irq_disable(); | |
2868 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2869 | perf_event_task_sched_in(current); |
8381f65d JI |
2870 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2871 | local_irq_enable(); | |
2872 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2873 | finish_lock_switch(rq, prev); |
e8fa1362 | 2874 | |
e107be36 | 2875 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2876 | if (mm) |
2877 | mmdrop(mm); | |
c394cc9f | 2878 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2879 | /* |
2880 | * Remove function-return probe instances associated with this | |
2881 | * task and put them back on the free list. | |
9761eea8 | 2882 | */ |
c6fd91f0 | 2883 | kprobe_flush_task(prev); |
1da177e4 | 2884 | put_task_struct(prev); |
c6fd91f0 | 2885 | } |
1da177e4 LT |
2886 | } |
2887 | ||
3f029d3c GH |
2888 | #ifdef CONFIG_SMP |
2889 | ||
2890 | /* assumes rq->lock is held */ | |
2891 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2892 | { | |
2893 | if (prev->sched_class->pre_schedule) | |
2894 | prev->sched_class->pre_schedule(rq, prev); | |
2895 | } | |
2896 | ||
2897 | /* rq->lock is NOT held, but preemption is disabled */ | |
2898 | static inline void post_schedule(struct rq *rq) | |
2899 | { | |
2900 | if (rq->post_schedule) { | |
2901 | unsigned long flags; | |
2902 | ||
05fa785c | 2903 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2904 | if (rq->curr->sched_class->post_schedule) |
2905 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2906 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2907 | |
2908 | rq->post_schedule = 0; | |
2909 | } | |
2910 | } | |
2911 | ||
2912 | #else | |
da19ab51 | 2913 | |
3f029d3c GH |
2914 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2915 | { | |
2916 | } | |
2917 | ||
2918 | static inline void post_schedule(struct rq *rq) | |
2919 | { | |
1da177e4 LT |
2920 | } |
2921 | ||
3f029d3c GH |
2922 | #endif |
2923 | ||
1da177e4 LT |
2924 | /** |
2925 | * schedule_tail - first thing a freshly forked thread must call. | |
2926 | * @prev: the thread we just switched away from. | |
2927 | */ | |
36c8b586 | 2928 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2929 | __releases(rq->lock) |
2930 | { | |
70b97a7f IM |
2931 | struct rq *rq = this_rq(); |
2932 | ||
4866cde0 | 2933 | finish_task_switch(rq, prev); |
da19ab51 | 2934 | |
3f029d3c GH |
2935 | /* |
2936 | * FIXME: do we need to worry about rq being invalidated by the | |
2937 | * task_switch? | |
2938 | */ | |
2939 | post_schedule(rq); | |
70b97a7f | 2940 | |
4866cde0 NP |
2941 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2942 | /* In this case, finish_task_switch does not reenable preemption */ | |
2943 | preempt_enable(); | |
2944 | #endif | |
1da177e4 | 2945 | if (current->set_child_tid) |
b488893a | 2946 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2947 | } |
2948 | ||
2949 | /* | |
2950 | * context_switch - switch to the new MM and the new | |
2951 | * thread's register state. | |
2952 | */ | |
dd41f596 | 2953 | static inline void |
70b97a7f | 2954 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2955 | struct task_struct *next) |
1da177e4 | 2956 | { |
dd41f596 | 2957 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2958 | |
e107be36 | 2959 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2960 | |
dd41f596 IM |
2961 | mm = next->mm; |
2962 | oldmm = prev->active_mm; | |
9226d125 ZA |
2963 | /* |
2964 | * For paravirt, this is coupled with an exit in switch_to to | |
2965 | * combine the page table reload and the switch backend into | |
2966 | * one hypercall. | |
2967 | */ | |
224101ed | 2968 | arch_start_context_switch(prev); |
9226d125 | 2969 | |
31915ab4 | 2970 | if (!mm) { |
1da177e4 LT |
2971 | next->active_mm = oldmm; |
2972 | atomic_inc(&oldmm->mm_count); | |
2973 | enter_lazy_tlb(oldmm, next); | |
2974 | } else | |
2975 | switch_mm(oldmm, mm, next); | |
2976 | ||
31915ab4 | 2977 | if (!prev->mm) { |
1da177e4 | 2978 | prev->active_mm = NULL; |
1da177e4 LT |
2979 | rq->prev_mm = oldmm; |
2980 | } | |
3a5f5e48 IM |
2981 | /* |
2982 | * Since the runqueue lock will be released by the next | |
2983 | * task (which is an invalid locking op but in the case | |
2984 | * of the scheduler it's an obvious special-case), so we | |
2985 | * do an early lockdep release here: | |
2986 | */ | |
2987 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2988 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2989 | #endif |
1da177e4 LT |
2990 | |
2991 | /* Here we just switch the register state and the stack. */ | |
2992 | switch_to(prev, next, prev); | |
2993 | ||
dd41f596 IM |
2994 | barrier(); |
2995 | /* | |
2996 | * this_rq must be evaluated again because prev may have moved | |
2997 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2998 | * frame will be invalid. | |
2999 | */ | |
3000 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3001 | } |
3002 | ||
3003 | /* | |
3004 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3005 | * | |
3006 | * externally visible scheduler statistics: current number of runnable | |
3007 | * threads, current number of uninterruptible-sleeping threads, total | |
3008 | * number of context switches performed since bootup. | |
3009 | */ | |
3010 | unsigned long nr_running(void) | |
3011 | { | |
3012 | unsigned long i, sum = 0; | |
3013 | ||
3014 | for_each_online_cpu(i) | |
3015 | sum += cpu_rq(i)->nr_running; | |
3016 | ||
3017 | return sum; | |
f711f609 | 3018 | } |
1da177e4 LT |
3019 | |
3020 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3021 | { |
1da177e4 | 3022 | unsigned long i, sum = 0; |
f711f609 | 3023 | |
0a945022 | 3024 | for_each_possible_cpu(i) |
1da177e4 | 3025 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3026 | |
3027 | /* | |
1da177e4 LT |
3028 | * Since we read the counters lockless, it might be slightly |
3029 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3030 | */ |
1da177e4 LT |
3031 | if (unlikely((long)sum < 0)) |
3032 | sum = 0; | |
f711f609 | 3033 | |
1da177e4 | 3034 | return sum; |
f711f609 | 3035 | } |
f711f609 | 3036 | |
1da177e4 | 3037 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3038 | { |
cc94abfc SR |
3039 | int i; |
3040 | unsigned long long sum = 0; | |
46cb4b7c | 3041 | |
0a945022 | 3042 | for_each_possible_cpu(i) |
1da177e4 | 3043 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3044 | |
1da177e4 LT |
3045 | return sum; |
3046 | } | |
483b4ee6 | 3047 | |
1da177e4 LT |
3048 | unsigned long nr_iowait(void) |
3049 | { | |
3050 | unsigned long i, sum = 0; | |
483b4ee6 | 3051 | |
0a945022 | 3052 | for_each_possible_cpu(i) |
1da177e4 | 3053 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3054 | |
1da177e4 LT |
3055 | return sum; |
3056 | } | |
483b4ee6 | 3057 | |
8c215bd3 | 3058 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3059 | { |
8c215bd3 | 3060 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3061 | return atomic_read(&this->nr_iowait); |
3062 | } | |
46cb4b7c | 3063 | |
69d25870 AV |
3064 | unsigned long this_cpu_load(void) |
3065 | { | |
3066 | struct rq *this = this_rq(); | |
3067 | return this->cpu_load[0]; | |
3068 | } | |
e790fb0b | 3069 | |
46cb4b7c | 3070 | |
dce48a84 TG |
3071 | /* Variables and functions for calc_load */ |
3072 | static atomic_long_t calc_load_tasks; | |
3073 | static unsigned long calc_load_update; | |
3074 | unsigned long avenrun[3]; | |
3075 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3076 | |
74f5187a PZ |
3077 | static long calc_load_fold_active(struct rq *this_rq) |
3078 | { | |
3079 | long nr_active, delta = 0; | |
3080 | ||
3081 | nr_active = this_rq->nr_running; | |
3082 | nr_active += (long) this_rq->nr_uninterruptible; | |
3083 | ||
3084 | if (nr_active != this_rq->calc_load_active) { | |
3085 | delta = nr_active - this_rq->calc_load_active; | |
3086 | this_rq->calc_load_active = nr_active; | |
3087 | } | |
3088 | ||
3089 | return delta; | |
3090 | } | |
3091 | ||
0f004f5a PZ |
3092 | static unsigned long |
3093 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3094 | { | |
3095 | load *= exp; | |
3096 | load += active * (FIXED_1 - exp); | |
3097 | load += 1UL << (FSHIFT - 1); | |
3098 | return load >> FSHIFT; | |
3099 | } | |
3100 | ||
74f5187a PZ |
3101 | #ifdef CONFIG_NO_HZ |
3102 | /* | |
3103 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3104 | * | |
3105 | * When making the ILB scale, we should try to pull this in as well. | |
3106 | */ | |
3107 | static atomic_long_t calc_load_tasks_idle; | |
3108 | ||
3109 | static void calc_load_account_idle(struct rq *this_rq) | |
3110 | { | |
3111 | long delta; | |
3112 | ||
3113 | delta = calc_load_fold_active(this_rq); | |
3114 | if (delta) | |
3115 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3116 | } | |
3117 | ||
3118 | static long calc_load_fold_idle(void) | |
3119 | { | |
3120 | long delta = 0; | |
3121 | ||
3122 | /* | |
3123 | * Its got a race, we don't care... | |
3124 | */ | |
3125 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3126 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3127 | ||
3128 | return delta; | |
3129 | } | |
0f004f5a PZ |
3130 | |
3131 | /** | |
3132 | * fixed_power_int - compute: x^n, in O(log n) time | |
3133 | * | |
3134 | * @x: base of the power | |
3135 | * @frac_bits: fractional bits of @x | |
3136 | * @n: power to raise @x to. | |
3137 | * | |
3138 | * By exploiting the relation between the definition of the natural power | |
3139 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3140 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3141 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3142 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3143 | * of course trivially computable in O(log_2 n), the length of our binary | |
3144 | * vector. | |
3145 | */ | |
3146 | static unsigned long | |
3147 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3148 | { | |
3149 | unsigned long result = 1UL << frac_bits; | |
3150 | ||
3151 | if (n) for (;;) { | |
3152 | if (n & 1) { | |
3153 | result *= x; | |
3154 | result += 1UL << (frac_bits - 1); | |
3155 | result >>= frac_bits; | |
3156 | } | |
3157 | n >>= 1; | |
3158 | if (!n) | |
3159 | break; | |
3160 | x *= x; | |
3161 | x += 1UL << (frac_bits - 1); | |
3162 | x >>= frac_bits; | |
3163 | } | |
3164 | ||
3165 | return result; | |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * a1 = a0 * e + a * (1 - e) | |
3170 | * | |
3171 | * a2 = a1 * e + a * (1 - e) | |
3172 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3173 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3174 | * | |
3175 | * a3 = a2 * e + a * (1 - e) | |
3176 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3177 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3178 | * | |
3179 | * ... | |
3180 | * | |
3181 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3182 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3183 | * = a0 * e^n + a * (1 - e^n) | |
3184 | * | |
3185 | * [1] application of the geometric series: | |
3186 | * | |
3187 | * n 1 - x^(n+1) | |
3188 | * S_n := \Sum x^i = ------------- | |
3189 | * i=0 1 - x | |
3190 | */ | |
3191 | static unsigned long | |
3192 | calc_load_n(unsigned long load, unsigned long exp, | |
3193 | unsigned long active, unsigned int n) | |
3194 | { | |
3195 | ||
3196 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3197 | } | |
3198 | ||
3199 | /* | |
3200 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3201 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3202 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3203 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3204 | * | |
3205 | * Once we've updated the global active value, we need to apply the exponential | |
3206 | * weights adjusted to the number of cycles missed. | |
3207 | */ | |
3208 | static void calc_global_nohz(unsigned long ticks) | |
3209 | { | |
3210 | long delta, active, n; | |
3211 | ||
3212 | if (time_before(jiffies, calc_load_update)) | |
3213 | return; | |
3214 | ||
3215 | /* | |
3216 | * If we crossed a calc_load_update boundary, make sure to fold | |
3217 | * any pending idle changes, the respective CPUs might have | |
3218 | * missed the tick driven calc_load_account_active() update | |
3219 | * due to NO_HZ. | |
3220 | */ | |
3221 | delta = calc_load_fold_idle(); | |
3222 | if (delta) | |
3223 | atomic_long_add(delta, &calc_load_tasks); | |
3224 | ||
3225 | /* | |
3226 | * If we were idle for multiple load cycles, apply them. | |
3227 | */ | |
3228 | if (ticks >= LOAD_FREQ) { | |
3229 | n = ticks / LOAD_FREQ; | |
3230 | ||
3231 | active = atomic_long_read(&calc_load_tasks); | |
3232 | active = active > 0 ? active * FIXED_1 : 0; | |
3233 | ||
3234 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3235 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3236 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3237 | ||
3238 | calc_load_update += n * LOAD_FREQ; | |
3239 | } | |
3240 | ||
3241 | /* | |
3242 | * Its possible the remainder of the above division also crosses | |
3243 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3244 | * which comes after this will take care of that. | |
3245 | * | |
3246 | * Consider us being 11 ticks before a cycle completion, and us | |
3247 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3248 | * age us 4 cycles, and the test in calc_global_load() will | |
3249 | * pick up the final one. | |
3250 | */ | |
3251 | } | |
74f5187a PZ |
3252 | #else |
3253 | static void calc_load_account_idle(struct rq *this_rq) | |
3254 | { | |
3255 | } | |
3256 | ||
3257 | static inline long calc_load_fold_idle(void) | |
3258 | { | |
3259 | return 0; | |
3260 | } | |
0f004f5a PZ |
3261 | |
3262 | static void calc_global_nohz(unsigned long ticks) | |
3263 | { | |
3264 | } | |
74f5187a PZ |
3265 | #endif |
3266 | ||
2d02494f TG |
3267 | /** |
3268 | * get_avenrun - get the load average array | |
3269 | * @loads: pointer to dest load array | |
3270 | * @offset: offset to add | |
3271 | * @shift: shift count to shift the result left | |
3272 | * | |
3273 | * These values are estimates at best, so no need for locking. | |
3274 | */ | |
3275 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3276 | { | |
3277 | loads[0] = (avenrun[0] + offset) << shift; | |
3278 | loads[1] = (avenrun[1] + offset) << shift; | |
3279 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3280 | } |
46cb4b7c | 3281 | |
46cb4b7c | 3282 | /* |
dce48a84 TG |
3283 | * calc_load - update the avenrun load estimates 10 ticks after the |
3284 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3285 | */ |
0f004f5a | 3286 | void calc_global_load(unsigned long ticks) |
7835b98b | 3287 | { |
dce48a84 | 3288 | long active; |
1da177e4 | 3289 | |
0f004f5a PZ |
3290 | calc_global_nohz(ticks); |
3291 | ||
3292 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3293 | return; |
1da177e4 | 3294 | |
dce48a84 TG |
3295 | active = atomic_long_read(&calc_load_tasks); |
3296 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3297 | |
dce48a84 TG |
3298 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3299 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3300 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3301 | |
dce48a84 TG |
3302 | calc_load_update += LOAD_FREQ; |
3303 | } | |
1da177e4 | 3304 | |
dce48a84 | 3305 | /* |
74f5187a PZ |
3306 | * Called from update_cpu_load() to periodically update this CPU's |
3307 | * active count. | |
dce48a84 TG |
3308 | */ |
3309 | static void calc_load_account_active(struct rq *this_rq) | |
3310 | { | |
74f5187a | 3311 | long delta; |
08c183f3 | 3312 | |
74f5187a PZ |
3313 | if (time_before(jiffies, this_rq->calc_load_update)) |
3314 | return; | |
783609c6 | 3315 | |
74f5187a PZ |
3316 | delta = calc_load_fold_active(this_rq); |
3317 | delta += calc_load_fold_idle(); | |
3318 | if (delta) | |
dce48a84 | 3319 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3320 | |
3321 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3322 | } |
3323 | ||
fdf3e95d VP |
3324 | /* |
3325 | * The exact cpuload at various idx values, calculated at every tick would be | |
3326 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3327 | * | |
3328 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3329 | * on nth tick when cpu may be busy, then we have: | |
3330 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3331 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3332 | * | |
3333 | * decay_load_missed() below does efficient calculation of | |
3334 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3335 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3336 | * | |
3337 | * The calculation is approximated on a 128 point scale. | |
3338 | * degrade_zero_ticks is the number of ticks after which load at any | |
3339 | * particular idx is approximated to be zero. | |
3340 | * degrade_factor is a precomputed table, a row for each load idx. | |
3341 | * Each column corresponds to degradation factor for a power of two ticks, | |
3342 | * based on 128 point scale. | |
3343 | * Example: | |
3344 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3345 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3346 | * | |
3347 | * With this power of 2 load factors, we can degrade the load n times | |
3348 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3349 | * n mult/shifts needed by the exact degradation. | |
3350 | */ | |
3351 | #define DEGRADE_SHIFT 7 | |
3352 | static const unsigned char | |
3353 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3354 | static const unsigned char | |
3355 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3356 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3357 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3358 | {96, 72, 40, 12, 1, 0, 0}, | |
3359 | {112, 98, 75, 43, 15, 1, 0}, | |
3360 | {120, 112, 98, 76, 45, 16, 2} }; | |
3361 | ||
3362 | /* | |
3363 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3364 | * would be when CPU is idle and so we just decay the old load without | |
3365 | * adding any new load. | |
3366 | */ | |
3367 | static unsigned long | |
3368 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3369 | { | |
3370 | int j = 0; | |
3371 | ||
3372 | if (!missed_updates) | |
3373 | return load; | |
3374 | ||
3375 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3376 | return 0; | |
3377 | ||
3378 | if (idx == 1) | |
3379 | return load >> missed_updates; | |
3380 | ||
3381 | while (missed_updates) { | |
3382 | if (missed_updates % 2) | |
3383 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3384 | ||
3385 | missed_updates >>= 1; | |
3386 | j++; | |
3387 | } | |
3388 | return load; | |
3389 | } | |
3390 | ||
46cb4b7c | 3391 | /* |
dd41f596 | 3392 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3393 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3394 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3395 | */ |
dd41f596 | 3396 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3397 | { |
495eca49 | 3398 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3399 | unsigned long curr_jiffies = jiffies; |
3400 | unsigned long pending_updates; | |
dd41f596 | 3401 | int i, scale; |
46cb4b7c | 3402 | |
dd41f596 | 3403 | this_rq->nr_load_updates++; |
46cb4b7c | 3404 | |
fdf3e95d VP |
3405 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3406 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3407 | return; | |
3408 | ||
3409 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3410 | this_rq->last_load_update_tick = curr_jiffies; | |
3411 | ||
dd41f596 | 3412 | /* Update our load: */ |
fdf3e95d VP |
3413 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3414 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3415 | unsigned long old_load, new_load; |
7d1e6a9b | 3416 | |
dd41f596 | 3417 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3418 | |
dd41f596 | 3419 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3420 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3421 | new_load = this_load; |
a25707f3 IM |
3422 | /* |
3423 | * Round up the averaging division if load is increasing. This | |
3424 | * prevents us from getting stuck on 9 if the load is 10, for | |
3425 | * example. | |
3426 | */ | |
3427 | if (new_load > old_load) | |
fdf3e95d VP |
3428 | new_load += scale - 1; |
3429 | ||
3430 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3431 | } |
da2b71ed SS |
3432 | |
3433 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3434 | } |
3435 | ||
3436 | static void update_cpu_load_active(struct rq *this_rq) | |
3437 | { | |
3438 | update_cpu_load(this_rq); | |
46cb4b7c | 3439 | |
74f5187a | 3440 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3441 | } |
3442 | ||
dd41f596 | 3443 | #ifdef CONFIG_SMP |
8a0be9ef | 3444 | |
46cb4b7c | 3445 | /* |
38022906 PZ |
3446 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3447 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3448 | */ |
38022906 | 3449 | void sched_exec(void) |
46cb4b7c | 3450 | { |
38022906 | 3451 | struct task_struct *p = current; |
1da177e4 | 3452 | unsigned long flags; |
70b97a7f | 3453 | struct rq *rq; |
0017d735 | 3454 | int dest_cpu; |
46cb4b7c | 3455 | |
1da177e4 | 3456 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3457 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3458 | if (dest_cpu == smp_processor_id()) | |
3459 | goto unlock; | |
38022906 | 3460 | |
46cb4b7c | 3461 | /* |
38022906 | 3462 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3463 | */ |
30da688e | 3464 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3465 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3466 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3467 | |
1da177e4 | 3468 | task_rq_unlock(rq, &flags); |
969c7921 | 3469 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3470 | return; |
3471 | } | |
0017d735 | 3472 | unlock: |
1da177e4 | 3473 | task_rq_unlock(rq, &flags); |
1da177e4 | 3474 | } |
dd41f596 | 3475 | |
1da177e4 LT |
3476 | #endif |
3477 | ||
1da177e4 LT |
3478 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3479 | ||
3480 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3481 | ||
3482 | /* | |
c5f8d995 | 3483 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3484 | * @p in case that task is currently running. |
c5f8d995 HS |
3485 | * |
3486 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3487 | */ |
c5f8d995 HS |
3488 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3489 | { | |
3490 | u64 ns = 0; | |
3491 | ||
3492 | if (task_current(rq, p)) { | |
3493 | update_rq_clock(rq); | |
305e6835 | 3494 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3495 | if ((s64)ns < 0) |
3496 | ns = 0; | |
3497 | } | |
3498 | ||
3499 | return ns; | |
3500 | } | |
3501 | ||
bb34d92f | 3502 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3503 | { |
1da177e4 | 3504 | unsigned long flags; |
41b86e9c | 3505 | struct rq *rq; |
bb34d92f | 3506 | u64 ns = 0; |
48f24c4d | 3507 | |
41b86e9c | 3508 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3509 | ns = do_task_delta_exec(p, rq); |
3510 | task_rq_unlock(rq, &flags); | |
1508487e | 3511 | |
c5f8d995 HS |
3512 | return ns; |
3513 | } | |
f06febc9 | 3514 | |
c5f8d995 HS |
3515 | /* |
3516 | * Return accounted runtime for the task. | |
3517 | * In case the task is currently running, return the runtime plus current's | |
3518 | * pending runtime that have not been accounted yet. | |
3519 | */ | |
3520 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3521 | { | |
3522 | unsigned long flags; | |
3523 | struct rq *rq; | |
3524 | u64 ns = 0; | |
3525 | ||
3526 | rq = task_rq_lock(p, &flags); | |
3527 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3528 | task_rq_unlock(rq, &flags); | |
3529 | ||
3530 | return ns; | |
3531 | } | |
48f24c4d | 3532 | |
c5f8d995 HS |
3533 | /* |
3534 | * Return sum_exec_runtime for the thread group. | |
3535 | * In case the task is currently running, return the sum plus current's | |
3536 | * pending runtime that have not been accounted yet. | |
3537 | * | |
3538 | * Note that the thread group might have other running tasks as well, | |
3539 | * so the return value not includes other pending runtime that other | |
3540 | * running tasks might have. | |
3541 | */ | |
3542 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3543 | { | |
3544 | struct task_cputime totals; | |
3545 | unsigned long flags; | |
3546 | struct rq *rq; | |
3547 | u64 ns; | |
3548 | ||
3549 | rq = task_rq_lock(p, &flags); | |
3550 | thread_group_cputime(p, &totals); | |
3551 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3552 | task_rq_unlock(rq, &flags); |
48f24c4d | 3553 | |
1da177e4 LT |
3554 | return ns; |
3555 | } | |
3556 | ||
1da177e4 LT |
3557 | /* |
3558 | * Account user cpu time to a process. | |
3559 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3560 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3561 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3562 | */ |
457533a7 MS |
3563 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3564 | cputime_t cputime_scaled) | |
1da177e4 LT |
3565 | { |
3566 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3567 | cputime64_t tmp; | |
3568 | ||
457533a7 | 3569 | /* Add user time to process. */ |
1da177e4 | 3570 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3571 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3572 | account_group_user_time(p, cputime); |
1da177e4 LT |
3573 | |
3574 | /* Add user time to cpustat. */ | |
3575 | tmp = cputime_to_cputime64(cputime); | |
3576 | if (TASK_NICE(p) > 0) | |
3577 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3578 | else | |
3579 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3580 | |
3581 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3582 | /* Account for user time used */ |
3583 | acct_update_integrals(p); | |
1da177e4 LT |
3584 | } |
3585 | ||
94886b84 LV |
3586 | /* |
3587 | * Account guest cpu time to a process. | |
3588 | * @p: the process that the cpu time gets accounted to | |
3589 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3590 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3591 | */ |
457533a7 MS |
3592 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3593 | cputime_t cputime_scaled) | |
94886b84 LV |
3594 | { |
3595 | cputime64_t tmp; | |
3596 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3597 | ||
3598 | tmp = cputime_to_cputime64(cputime); | |
3599 | ||
457533a7 | 3600 | /* Add guest time to process. */ |
94886b84 | 3601 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3602 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3603 | account_group_user_time(p, cputime); |
94886b84 LV |
3604 | p->gtime = cputime_add(p->gtime, cputime); |
3605 | ||
457533a7 | 3606 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3607 | if (TASK_NICE(p) > 0) { |
3608 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3609 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3610 | } else { | |
3611 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3612 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3613 | } | |
94886b84 LV |
3614 | } |
3615 | ||
70a89a66 VP |
3616 | /* |
3617 | * Account system cpu time to a process and desired cpustat field | |
3618 | * @p: the process that the cpu time gets accounted to | |
3619 | * @cputime: the cpu time spent in kernel space since the last update | |
3620 | * @cputime_scaled: cputime scaled by cpu frequency | |
3621 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3622 | */ | |
3623 | static inline | |
3624 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3625 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3626 | { | |
3627 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3628 | ||
3629 | /* Add system time to process. */ | |
3630 | p->stime = cputime_add(p->stime, cputime); | |
3631 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3632 | account_group_system_time(p, cputime); | |
3633 | ||
3634 | /* Add system time to cpustat. */ | |
3635 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3636 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3637 | ||
3638 | /* Account for system time used */ | |
3639 | acct_update_integrals(p); | |
3640 | } | |
3641 | ||
1da177e4 LT |
3642 | /* |
3643 | * Account system cpu time to a process. | |
3644 | * @p: the process that the cpu time gets accounted to | |
3645 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3646 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3647 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3648 | */ |
3649 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3650 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3651 | { |
3652 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3653 | cputime64_t *target_cputime64; |
1da177e4 | 3654 | |
983ed7a6 | 3655 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3656 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3657 | return; |
3658 | } | |
94886b84 | 3659 | |
1da177e4 | 3660 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3661 | target_cputime64 = &cpustat->irq; |
75e1056f | 3662 | else if (in_serving_softirq()) |
70a89a66 | 3663 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3664 | else |
70a89a66 | 3665 | target_cputime64 = &cpustat->system; |
ef12fefa | 3666 | |
70a89a66 | 3667 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3668 | } |
3669 | ||
c66f08be | 3670 | /* |
1da177e4 | 3671 | * Account for involuntary wait time. |
544b4a1f | 3672 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3673 | */ |
79741dd3 | 3674 | void account_steal_time(cputime_t cputime) |
c66f08be | 3675 | { |
79741dd3 MS |
3676 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3677 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3678 | ||
3679 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3680 | } |
3681 | ||
1da177e4 | 3682 | /* |
79741dd3 MS |
3683 | * Account for idle time. |
3684 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3685 | */ |
79741dd3 | 3686 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3687 | { |
3688 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3689 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3690 | struct rq *rq = this_rq(); |
1da177e4 | 3691 | |
79741dd3 MS |
3692 | if (atomic_read(&rq->nr_iowait) > 0) |
3693 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3694 | else | |
3695 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3696 | } |
3697 | ||
79741dd3 MS |
3698 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3699 | ||
abb74cef VP |
3700 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3701 | /* | |
3702 | * Account a tick to a process and cpustat | |
3703 | * @p: the process that the cpu time gets accounted to | |
3704 | * @user_tick: is the tick from userspace | |
3705 | * @rq: the pointer to rq | |
3706 | * | |
3707 | * Tick demultiplexing follows the order | |
3708 | * - pending hardirq update | |
3709 | * - pending softirq update | |
3710 | * - user_time | |
3711 | * - idle_time | |
3712 | * - system time | |
3713 | * - check for guest_time | |
3714 | * - else account as system_time | |
3715 | * | |
3716 | * Check for hardirq is done both for system and user time as there is | |
3717 | * no timer going off while we are on hardirq and hence we may never get an | |
3718 | * opportunity to update it solely in system time. | |
3719 | * p->stime and friends are only updated on system time and not on irq | |
3720 | * softirq as those do not count in task exec_runtime any more. | |
3721 | */ | |
3722 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3723 | struct rq *rq) | |
3724 | { | |
3725 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3726 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3727 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3728 | ||
3729 | if (irqtime_account_hi_update()) { | |
3730 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3731 | } else if (irqtime_account_si_update()) { | |
3732 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3733 | } else if (this_cpu_ksoftirqd() == p) { |
3734 | /* | |
3735 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3736 | * So, we have to handle it separately here. | |
3737 | * Also, p->stime needs to be updated for ksoftirqd. | |
3738 | */ | |
3739 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3740 | &cpustat->softirq); | |
abb74cef VP |
3741 | } else if (user_tick) { |
3742 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3743 | } else if (p == rq->idle) { | |
3744 | account_idle_time(cputime_one_jiffy); | |
3745 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3746 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3747 | } else { | |
3748 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3749 | &cpustat->system); | |
3750 | } | |
3751 | } | |
3752 | ||
3753 | static void irqtime_account_idle_ticks(int ticks) | |
3754 | { | |
3755 | int i; | |
3756 | struct rq *rq = this_rq(); | |
3757 | ||
3758 | for (i = 0; i < ticks; i++) | |
3759 | irqtime_account_process_tick(current, 0, rq); | |
3760 | } | |
544b4a1f | 3761 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3762 | static void irqtime_account_idle_ticks(int ticks) {} |
3763 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3764 | struct rq *rq) {} | |
544b4a1f | 3765 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3766 | |
3767 | /* | |
3768 | * Account a single tick of cpu time. | |
3769 | * @p: the process that the cpu time gets accounted to | |
3770 | * @user_tick: indicates if the tick is a user or a system tick | |
3771 | */ | |
3772 | void account_process_tick(struct task_struct *p, int user_tick) | |
3773 | { | |
a42548a1 | 3774 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3775 | struct rq *rq = this_rq(); |
3776 | ||
abb74cef VP |
3777 | if (sched_clock_irqtime) { |
3778 | irqtime_account_process_tick(p, user_tick, rq); | |
3779 | return; | |
3780 | } | |
3781 | ||
79741dd3 | 3782 | if (user_tick) |
a42548a1 | 3783 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3784 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3785 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3786 | one_jiffy_scaled); |
3787 | else | |
a42548a1 | 3788 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3789 | } |
3790 | ||
3791 | /* | |
3792 | * Account multiple ticks of steal time. | |
3793 | * @p: the process from which the cpu time has been stolen | |
3794 | * @ticks: number of stolen ticks | |
3795 | */ | |
3796 | void account_steal_ticks(unsigned long ticks) | |
3797 | { | |
3798 | account_steal_time(jiffies_to_cputime(ticks)); | |
3799 | } | |
3800 | ||
3801 | /* | |
3802 | * Account multiple ticks of idle time. | |
3803 | * @ticks: number of stolen ticks | |
3804 | */ | |
3805 | void account_idle_ticks(unsigned long ticks) | |
3806 | { | |
abb74cef VP |
3807 | |
3808 | if (sched_clock_irqtime) { | |
3809 | irqtime_account_idle_ticks(ticks); | |
3810 | return; | |
3811 | } | |
3812 | ||
79741dd3 | 3813 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3814 | } |
3815 | ||
79741dd3 MS |
3816 | #endif |
3817 | ||
49048622 BS |
3818 | /* |
3819 | * Use precise platform statistics if available: | |
3820 | */ | |
3821 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3822 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3823 | { |
d99ca3b9 HS |
3824 | *ut = p->utime; |
3825 | *st = p->stime; | |
49048622 BS |
3826 | } |
3827 | ||
0cf55e1e | 3828 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3829 | { |
0cf55e1e HS |
3830 | struct task_cputime cputime; |
3831 | ||
3832 | thread_group_cputime(p, &cputime); | |
3833 | ||
3834 | *ut = cputime.utime; | |
3835 | *st = cputime.stime; | |
49048622 BS |
3836 | } |
3837 | #else | |
761b1d26 HS |
3838 | |
3839 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3840 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3841 | #endif |
3842 | ||
d180c5bc | 3843 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3844 | { |
d99ca3b9 | 3845 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3846 | |
3847 | /* | |
3848 | * Use CFS's precise accounting: | |
3849 | */ | |
d180c5bc | 3850 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3851 | |
3852 | if (total) { | |
e75e863d | 3853 | u64 temp = rtime; |
d180c5bc | 3854 | |
e75e863d | 3855 | temp *= utime; |
49048622 | 3856 | do_div(temp, total); |
d180c5bc HS |
3857 | utime = (cputime_t)temp; |
3858 | } else | |
3859 | utime = rtime; | |
49048622 | 3860 | |
d180c5bc HS |
3861 | /* |
3862 | * Compare with previous values, to keep monotonicity: | |
3863 | */ | |
761b1d26 | 3864 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3865 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3866 | |
d99ca3b9 HS |
3867 | *ut = p->prev_utime; |
3868 | *st = p->prev_stime; | |
49048622 BS |
3869 | } |
3870 | ||
0cf55e1e HS |
3871 | /* |
3872 | * Must be called with siglock held. | |
3873 | */ | |
3874 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3875 | { |
0cf55e1e HS |
3876 | struct signal_struct *sig = p->signal; |
3877 | struct task_cputime cputime; | |
3878 | cputime_t rtime, utime, total; | |
49048622 | 3879 | |
0cf55e1e | 3880 | thread_group_cputime(p, &cputime); |
49048622 | 3881 | |
0cf55e1e HS |
3882 | total = cputime_add(cputime.utime, cputime.stime); |
3883 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3884 | |
0cf55e1e | 3885 | if (total) { |
e75e863d | 3886 | u64 temp = rtime; |
49048622 | 3887 | |
e75e863d | 3888 | temp *= cputime.utime; |
0cf55e1e HS |
3889 | do_div(temp, total); |
3890 | utime = (cputime_t)temp; | |
3891 | } else | |
3892 | utime = rtime; | |
3893 | ||
3894 | sig->prev_utime = max(sig->prev_utime, utime); | |
3895 | sig->prev_stime = max(sig->prev_stime, | |
3896 | cputime_sub(rtime, sig->prev_utime)); | |
3897 | ||
3898 | *ut = sig->prev_utime; | |
3899 | *st = sig->prev_stime; | |
49048622 | 3900 | } |
49048622 | 3901 | #endif |
49048622 | 3902 | |
7835b98b CL |
3903 | /* |
3904 | * This function gets called by the timer code, with HZ frequency. | |
3905 | * We call it with interrupts disabled. | |
3906 | * | |
3907 | * It also gets called by the fork code, when changing the parent's | |
3908 | * timeslices. | |
3909 | */ | |
3910 | void scheduler_tick(void) | |
3911 | { | |
7835b98b CL |
3912 | int cpu = smp_processor_id(); |
3913 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3914 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3915 | |
3916 | sched_clock_tick(); | |
dd41f596 | 3917 | |
05fa785c | 3918 | raw_spin_lock(&rq->lock); |
3e51f33f | 3919 | update_rq_clock(rq); |
fdf3e95d | 3920 | update_cpu_load_active(rq); |
fa85ae24 | 3921 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3922 | raw_spin_unlock(&rq->lock); |
7835b98b | 3923 | |
e9d2b064 | 3924 | perf_event_task_tick(); |
e220d2dc | 3925 | |
e418e1c2 | 3926 | #ifdef CONFIG_SMP |
dd41f596 IM |
3927 | rq->idle_at_tick = idle_cpu(cpu); |
3928 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3929 | #endif |
1da177e4 LT |
3930 | } |
3931 | ||
132380a0 | 3932 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3933 | { |
3934 | if (in_lock_functions(addr)) { | |
3935 | addr = CALLER_ADDR2; | |
3936 | if (in_lock_functions(addr)) | |
3937 | addr = CALLER_ADDR3; | |
3938 | } | |
3939 | return addr; | |
3940 | } | |
1da177e4 | 3941 | |
7e49fcce SR |
3942 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3943 | defined(CONFIG_PREEMPT_TRACER)) | |
3944 | ||
43627582 | 3945 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3946 | { |
6cd8a4bb | 3947 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3948 | /* |
3949 | * Underflow? | |
3950 | */ | |
9a11b49a IM |
3951 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3952 | return; | |
6cd8a4bb | 3953 | #endif |
1da177e4 | 3954 | preempt_count() += val; |
6cd8a4bb | 3955 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3956 | /* |
3957 | * Spinlock count overflowing soon? | |
3958 | */ | |
33859f7f MOS |
3959 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3960 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3961 | #endif |
3962 | if (preempt_count() == val) | |
3963 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3964 | } |
3965 | EXPORT_SYMBOL(add_preempt_count); | |
3966 | ||
43627582 | 3967 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3968 | { |
6cd8a4bb | 3969 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3970 | /* |
3971 | * Underflow? | |
3972 | */ | |
01e3eb82 | 3973 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3974 | return; |
1da177e4 LT |
3975 | /* |
3976 | * Is the spinlock portion underflowing? | |
3977 | */ | |
9a11b49a IM |
3978 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3979 | !(preempt_count() & PREEMPT_MASK))) | |
3980 | return; | |
6cd8a4bb | 3981 | #endif |
9a11b49a | 3982 | |
6cd8a4bb SR |
3983 | if (preempt_count() == val) |
3984 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3985 | preempt_count() -= val; |
3986 | } | |
3987 | EXPORT_SYMBOL(sub_preempt_count); | |
3988 | ||
3989 | #endif | |
3990 | ||
3991 | /* | |
dd41f596 | 3992 | * Print scheduling while atomic bug: |
1da177e4 | 3993 | */ |
dd41f596 | 3994 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3995 | { |
838225b4 SS |
3996 | struct pt_regs *regs = get_irq_regs(); |
3997 | ||
3df0fc5b PZ |
3998 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3999 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4000 | |
dd41f596 | 4001 | debug_show_held_locks(prev); |
e21f5b15 | 4002 | print_modules(); |
dd41f596 IM |
4003 | if (irqs_disabled()) |
4004 | print_irqtrace_events(prev); | |
838225b4 SS |
4005 | |
4006 | if (regs) | |
4007 | show_regs(regs); | |
4008 | else | |
4009 | dump_stack(); | |
dd41f596 | 4010 | } |
1da177e4 | 4011 | |
dd41f596 IM |
4012 | /* |
4013 | * Various schedule()-time debugging checks and statistics: | |
4014 | */ | |
4015 | static inline void schedule_debug(struct task_struct *prev) | |
4016 | { | |
1da177e4 | 4017 | /* |
41a2d6cf | 4018 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4019 | * schedule() atomically, we ignore that path for now. |
4020 | * Otherwise, whine if we are scheduling when we should not be. | |
4021 | */ | |
3f33a7ce | 4022 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4023 | __schedule_bug(prev); |
4024 | ||
1da177e4 LT |
4025 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4026 | ||
2d72376b | 4027 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4028 | #ifdef CONFIG_SCHEDSTATS |
4029 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 4030 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 4031 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
4032 | } |
4033 | #endif | |
dd41f596 IM |
4034 | } |
4035 | ||
6cecd084 | 4036 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4037 | { |
a64692a3 MG |
4038 | if (prev->se.on_rq) |
4039 | update_rq_clock(rq); | |
6cecd084 | 4040 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4041 | } |
4042 | ||
dd41f596 IM |
4043 | /* |
4044 | * Pick up the highest-prio task: | |
4045 | */ | |
4046 | static inline struct task_struct * | |
b67802ea | 4047 | pick_next_task(struct rq *rq) |
dd41f596 | 4048 | { |
5522d5d5 | 4049 | const struct sched_class *class; |
dd41f596 | 4050 | struct task_struct *p; |
1da177e4 LT |
4051 | |
4052 | /* | |
dd41f596 IM |
4053 | * Optimization: we know that if all tasks are in |
4054 | * the fair class we can call that function directly: | |
1da177e4 | 4055 | */ |
dd41f596 | 4056 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4057 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4058 | if (likely(p)) |
4059 | return p; | |
1da177e4 LT |
4060 | } |
4061 | ||
34f971f6 | 4062 | for_each_class(class) { |
fb8d4724 | 4063 | p = class->pick_next_task(rq); |
dd41f596 IM |
4064 | if (p) |
4065 | return p; | |
dd41f596 | 4066 | } |
34f971f6 PZ |
4067 | |
4068 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4069 | } |
1da177e4 | 4070 | |
dd41f596 IM |
4071 | /* |
4072 | * schedule() is the main scheduler function. | |
4073 | */ | |
ff743345 | 4074 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4075 | { |
4076 | struct task_struct *prev, *next; | |
67ca7bde | 4077 | unsigned long *switch_count; |
dd41f596 | 4078 | struct rq *rq; |
31656519 | 4079 | int cpu; |
dd41f596 | 4080 | |
ff743345 PZ |
4081 | need_resched: |
4082 | preempt_disable(); | |
dd41f596 IM |
4083 | cpu = smp_processor_id(); |
4084 | rq = cpu_rq(cpu); | |
25502a6c | 4085 | rcu_note_context_switch(cpu); |
dd41f596 | 4086 | prev = rq->curr; |
dd41f596 | 4087 | |
dd41f596 | 4088 | schedule_debug(prev); |
1da177e4 | 4089 | |
31656519 | 4090 | if (sched_feat(HRTICK)) |
f333fdc9 | 4091 | hrtick_clear(rq); |
8f4d37ec | 4092 | |
05fa785c | 4093 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4094 | |
246d86b5 | 4095 | switch_count = &prev->nivcsw; |
1da177e4 | 4096 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4097 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4098 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4099 | } else { |
4100 | /* | |
4101 | * If a worker is going to sleep, notify and | |
4102 | * ask workqueue whether it wants to wake up a | |
4103 | * task to maintain concurrency. If so, wake | |
4104 | * up the task. | |
4105 | */ | |
4106 | if (prev->flags & PF_WQ_WORKER) { | |
4107 | struct task_struct *to_wakeup; | |
4108 | ||
4109 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4110 | if (to_wakeup) | |
4111 | try_to_wake_up_local(to_wakeup); | |
4112 | } | |
371fd7e7 | 4113 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 4114 | } |
dd41f596 | 4115 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4116 | } |
4117 | ||
73c10101 JA |
4118 | /* |
4119 | * If we are going to sleep and we have plugged IO queued, make | |
4120 | * sure to submit it to avoid deadlocks. | |
4121 | */ | |
4122 | if (prev->state != TASK_RUNNING && blk_needs_flush_plug(prev)) { | |
4123 | raw_spin_unlock(&rq->lock); | |
4124 | blk_flush_plug(prev); | |
4125 | raw_spin_lock(&rq->lock); | |
4126 | } | |
4127 | ||
3f029d3c | 4128 | pre_schedule(rq, prev); |
f65eda4f | 4129 | |
dd41f596 | 4130 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4131 | idle_balance(cpu, rq); |
1da177e4 | 4132 | |
df1c99d4 | 4133 | put_prev_task(rq, prev); |
b67802ea | 4134 | next = pick_next_task(rq); |
f26f9aff MG |
4135 | clear_tsk_need_resched(prev); |
4136 | rq->skip_clock_update = 0; | |
1da177e4 | 4137 | |
1da177e4 | 4138 | if (likely(prev != next)) { |
1da177e4 LT |
4139 | rq->nr_switches++; |
4140 | rq->curr = next; | |
4141 | ++*switch_count; | |
4142 | ||
dd41f596 | 4143 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4144 | /* |
246d86b5 ON |
4145 | * The context switch have flipped the stack from under us |
4146 | * and restored the local variables which were saved when | |
4147 | * this task called schedule() in the past. prev == current | |
4148 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4149 | */ |
4150 | cpu = smp_processor_id(); | |
4151 | rq = cpu_rq(cpu); | |
1da177e4 | 4152 | } else |
05fa785c | 4153 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4154 | |
3f029d3c | 4155 | post_schedule(rq); |
1da177e4 | 4156 | |
1da177e4 | 4157 | preempt_enable_no_resched(); |
ff743345 | 4158 | if (need_resched()) |
1da177e4 LT |
4159 | goto need_resched; |
4160 | } | |
1da177e4 LT |
4161 | EXPORT_SYMBOL(schedule); |
4162 | ||
c08f7829 | 4163 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4164 | /* |
4165 | * Look out! "owner" is an entirely speculative pointer | |
4166 | * access and not reliable. | |
4167 | */ | |
4168 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4169 | { | |
4170 | unsigned int cpu; | |
4171 | struct rq *rq; | |
4172 | ||
4173 | if (!sched_feat(OWNER_SPIN)) | |
4174 | return 0; | |
4175 | ||
4176 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4177 | /* | |
4178 | * Need to access the cpu field knowing that | |
4179 | * DEBUG_PAGEALLOC could have unmapped it if | |
4180 | * the mutex owner just released it and exited. | |
4181 | */ | |
4182 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4183 | return 0; |
0d66bf6d PZ |
4184 | #else |
4185 | cpu = owner->cpu; | |
4186 | #endif | |
4187 | ||
4188 | /* | |
4189 | * Even if the access succeeded (likely case), | |
4190 | * the cpu field may no longer be valid. | |
4191 | */ | |
4192 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4193 | return 0; |
0d66bf6d PZ |
4194 | |
4195 | /* | |
4196 | * We need to validate that we can do a | |
4197 | * get_cpu() and that we have the percpu area. | |
4198 | */ | |
4199 | if (!cpu_online(cpu)) | |
4b402210 | 4200 | return 0; |
0d66bf6d PZ |
4201 | |
4202 | rq = cpu_rq(cpu); | |
4203 | ||
4204 | for (;;) { | |
4205 | /* | |
4206 | * Owner changed, break to re-assess state. | |
4207 | */ | |
9d0f4dcc TC |
4208 | if (lock->owner != owner) { |
4209 | /* | |
4210 | * If the lock has switched to a different owner, | |
4211 | * we likely have heavy contention. Return 0 to quit | |
4212 | * optimistic spinning and not contend further: | |
4213 | */ | |
4214 | if (lock->owner) | |
4215 | return 0; | |
0d66bf6d | 4216 | break; |
9d0f4dcc | 4217 | } |
0d66bf6d PZ |
4218 | |
4219 | /* | |
4220 | * Is that owner really running on that cpu? | |
4221 | */ | |
4222 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4223 | return 0; | |
4224 | ||
335d7afb | 4225 | arch_mutex_cpu_relax(); |
0d66bf6d | 4226 | } |
4b402210 | 4227 | |
0d66bf6d PZ |
4228 | return 1; |
4229 | } | |
4230 | #endif | |
4231 | ||
1da177e4 LT |
4232 | #ifdef CONFIG_PREEMPT |
4233 | /* | |
2ed6e34f | 4234 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4235 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4236 | * occur there and call schedule directly. |
4237 | */ | |
d1f74e20 | 4238 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4239 | { |
4240 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4241 | |
1da177e4 LT |
4242 | /* |
4243 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4244 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4245 | */ |
beed33a8 | 4246 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4247 | return; |
4248 | ||
3a5c359a | 4249 | do { |
d1f74e20 | 4250 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4251 | schedule(); |
d1f74e20 | 4252 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4253 | |
3a5c359a AK |
4254 | /* |
4255 | * Check again in case we missed a preemption opportunity | |
4256 | * between schedule and now. | |
4257 | */ | |
4258 | barrier(); | |
5ed0cec0 | 4259 | } while (need_resched()); |
1da177e4 | 4260 | } |
1da177e4 LT |
4261 | EXPORT_SYMBOL(preempt_schedule); |
4262 | ||
4263 | /* | |
2ed6e34f | 4264 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4265 | * off of irq context. |
4266 | * Note, that this is called and return with irqs disabled. This will | |
4267 | * protect us against recursive calling from irq. | |
4268 | */ | |
4269 | asmlinkage void __sched preempt_schedule_irq(void) | |
4270 | { | |
4271 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4272 | |
2ed6e34f | 4273 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4274 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4275 | ||
3a5c359a AK |
4276 | do { |
4277 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4278 | local_irq_enable(); |
4279 | schedule(); | |
4280 | local_irq_disable(); | |
3a5c359a | 4281 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4282 | |
3a5c359a AK |
4283 | /* |
4284 | * Check again in case we missed a preemption opportunity | |
4285 | * between schedule and now. | |
4286 | */ | |
4287 | barrier(); | |
5ed0cec0 | 4288 | } while (need_resched()); |
1da177e4 LT |
4289 | } |
4290 | ||
4291 | #endif /* CONFIG_PREEMPT */ | |
4292 | ||
63859d4f | 4293 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4294 | void *key) |
1da177e4 | 4295 | { |
63859d4f | 4296 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4297 | } |
1da177e4 LT |
4298 | EXPORT_SYMBOL(default_wake_function); |
4299 | ||
4300 | /* | |
41a2d6cf IM |
4301 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4302 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4303 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4304 | * | |
4305 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4306 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4307 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4308 | */ | |
78ddb08f | 4309 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4310 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4311 | { |
2e45874c | 4312 | wait_queue_t *curr, *next; |
1da177e4 | 4313 | |
2e45874c | 4314 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4315 | unsigned flags = curr->flags; |
4316 | ||
63859d4f | 4317 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4318 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4319 | break; |
4320 | } | |
4321 | } | |
4322 | ||
4323 | /** | |
4324 | * __wake_up - wake up threads blocked on a waitqueue. | |
4325 | * @q: the waitqueue | |
4326 | * @mode: which threads | |
4327 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4328 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4329 | * |
4330 | * It may be assumed that this function implies a write memory barrier before | |
4331 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4332 | */ |
7ad5b3a5 | 4333 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4334 | int nr_exclusive, void *key) |
1da177e4 LT |
4335 | { |
4336 | unsigned long flags; | |
4337 | ||
4338 | spin_lock_irqsave(&q->lock, flags); | |
4339 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4340 | spin_unlock_irqrestore(&q->lock, flags); | |
4341 | } | |
1da177e4 LT |
4342 | EXPORT_SYMBOL(__wake_up); |
4343 | ||
4344 | /* | |
4345 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4346 | */ | |
7ad5b3a5 | 4347 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4348 | { |
4349 | __wake_up_common(q, mode, 1, 0, NULL); | |
4350 | } | |
22c43c81 | 4351 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4352 | |
4ede816a DL |
4353 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4354 | { | |
4355 | __wake_up_common(q, mode, 1, 0, key); | |
4356 | } | |
bf294b41 | 4357 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4358 | |
1da177e4 | 4359 | /** |
4ede816a | 4360 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4361 | * @q: the waitqueue |
4362 | * @mode: which threads | |
4363 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4364 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4365 | * |
4366 | * The sync wakeup differs that the waker knows that it will schedule | |
4367 | * away soon, so while the target thread will be woken up, it will not | |
4368 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4369 | * with each other. This can prevent needless bouncing between CPUs. | |
4370 | * | |
4371 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4372 | * |
4373 | * It may be assumed that this function implies a write memory barrier before | |
4374 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4375 | */ |
4ede816a DL |
4376 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4377 | int nr_exclusive, void *key) | |
1da177e4 LT |
4378 | { |
4379 | unsigned long flags; | |
7d478721 | 4380 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4381 | |
4382 | if (unlikely(!q)) | |
4383 | return; | |
4384 | ||
4385 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4386 | wake_flags = 0; |
1da177e4 LT |
4387 | |
4388 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4389 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4390 | spin_unlock_irqrestore(&q->lock, flags); |
4391 | } | |
4ede816a DL |
4392 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4393 | ||
4394 | /* | |
4395 | * __wake_up_sync - see __wake_up_sync_key() | |
4396 | */ | |
4397 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4398 | { | |
4399 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4400 | } | |
1da177e4 LT |
4401 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4402 | ||
65eb3dc6 KD |
4403 | /** |
4404 | * complete: - signals a single thread waiting on this completion | |
4405 | * @x: holds the state of this particular completion | |
4406 | * | |
4407 | * This will wake up a single thread waiting on this completion. Threads will be | |
4408 | * awakened in the same order in which they were queued. | |
4409 | * | |
4410 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4411 | * |
4412 | * It may be assumed that this function implies a write memory barrier before | |
4413 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4414 | */ |
b15136e9 | 4415 | void complete(struct completion *x) |
1da177e4 LT |
4416 | { |
4417 | unsigned long flags; | |
4418 | ||
4419 | spin_lock_irqsave(&x->wait.lock, flags); | |
4420 | x->done++; | |
d9514f6c | 4421 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4422 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4423 | } | |
4424 | EXPORT_SYMBOL(complete); | |
4425 | ||
65eb3dc6 KD |
4426 | /** |
4427 | * complete_all: - signals all threads waiting on this completion | |
4428 | * @x: holds the state of this particular completion | |
4429 | * | |
4430 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4431 | * |
4432 | * It may be assumed that this function implies a write memory barrier before | |
4433 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4434 | */ |
b15136e9 | 4435 | void complete_all(struct completion *x) |
1da177e4 LT |
4436 | { |
4437 | unsigned long flags; | |
4438 | ||
4439 | spin_lock_irqsave(&x->wait.lock, flags); | |
4440 | x->done += UINT_MAX/2; | |
d9514f6c | 4441 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4442 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4443 | } | |
4444 | EXPORT_SYMBOL(complete_all); | |
4445 | ||
8cbbe86d AK |
4446 | static inline long __sched |
4447 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4448 | { |
1da177e4 LT |
4449 | if (!x->done) { |
4450 | DECLARE_WAITQUEUE(wait, current); | |
4451 | ||
a93d2f17 | 4452 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4453 | do { |
94d3d824 | 4454 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4455 | timeout = -ERESTARTSYS; |
4456 | break; | |
8cbbe86d AK |
4457 | } |
4458 | __set_current_state(state); | |
1da177e4 LT |
4459 | spin_unlock_irq(&x->wait.lock); |
4460 | timeout = schedule_timeout(timeout); | |
4461 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4462 | } while (!x->done && timeout); |
1da177e4 | 4463 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4464 | if (!x->done) |
4465 | return timeout; | |
1da177e4 LT |
4466 | } |
4467 | x->done--; | |
ea71a546 | 4468 | return timeout ?: 1; |
1da177e4 | 4469 | } |
1da177e4 | 4470 | |
8cbbe86d AK |
4471 | static long __sched |
4472 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4473 | { |
1da177e4 LT |
4474 | might_sleep(); |
4475 | ||
4476 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4477 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4478 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4479 | return timeout; |
4480 | } | |
1da177e4 | 4481 | |
65eb3dc6 KD |
4482 | /** |
4483 | * wait_for_completion: - waits for completion of a task | |
4484 | * @x: holds the state of this particular completion | |
4485 | * | |
4486 | * This waits to be signaled for completion of a specific task. It is NOT | |
4487 | * interruptible and there is no timeout. | |
4488 | * | |
4489 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4490 | * and interrupt capability. Also see complete(). | |
4491 | */ | |
b15136e9 | 4492 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4493 | { |
4494 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4495 | } |
8cbbe86d | 4496 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4497 | |
65eb3dc6 KD |
4498 | /** |
4499 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4500 | * @x: holds the state of this particular completion | |
4501 | * @timeout: timeout value in jiffies | |
4502 | * | |
4503 | * This waits for either a completion of a specific task to be signaled or for a | |
4504 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4505 | * interruptible. | |
4506 | */ | |
b15136e9 | 4507 | unsigned long __sched |
8cbbe86d | 4508 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4509 | { |
8cbbe86d | 4510 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4511 | } |
8cbbe86d | 4512 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4513 | |
65eb3dc6 KD |
4514 | /** |
4515 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4516 | * @x: holds the state of this particular completion | |
4517 | * | |
4518 | * This waits for completion of a specific task to be signaled. It is | |
4519 | * interruptible. | |
4520 | */ | |
8cbbe86d | 4521 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4522 | { |
51e97990 AK |
4523 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4524 | if (t == -ERESTARTSYS) | |
4525 | return t; | |
4526 | return 0; | |
0fec171c | 4527 | } |
8cbbe86d | 4528 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4529 | |
65eb3dc6 KD |
4530 | /** |
4531 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4532 | * @x: holds the state of this particular completion | |
4533 | * @timeout: timeout value in jiffies | |
4534 | * | |
4535 | * This waits for either a completion of a specific task to be signaled or for a | |
4536 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4537 | */ | |
6bf41237 | 4538 | long __sched |
8cbbe86d AK |
4539 | wait_for_completion_interruptible_timeout(struct completion *x, |
4540 | unsigned long timeout) | |
0fec171c | 4541 | { |
8cbbe86d | 4542 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4543 | } |
8cbbe86d | 4544 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4545 | |
65eb3dc6 KD |
4546 | /** |
4547 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4548 | * @x: holds the state of this particular completion | |
4549 | * | |
4550 | * This waits to be signaled for completion of a specific task. It can be | |
4551 | * interrupted by a kill signal. | |
4552 | */ | |
009e577e MW |
4553 | int __sched wait_for_completion_killable(struct completion *x) |
4554 | { | |
4555 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4556 | if (t == -ERESTARTSYS) | |
4557 | return t; | |
4558 | return 0; | |
4559 | } | |
4560 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4561 | ||
0aa12fb4 SW |
4562 | /** |
4563 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4564 | * @x: holds the state of this particular completion | |
4565 | * @timeout: timeout value in jiffies | |
4566 | * | |
4567 | * This waits for either a completion of a specific task to be | |
4568 | * signaled or for a specified timeout to expire. It can be | |
4569 | * interrupted by a kill signal. The timeout is in jiffies. | |
4570 | */ | |
6bf41237 | 4571 | long __sched |
0aa12fb4 SW |
4572 | wait_for_completion_killable_timeout(struct completion *x, |
4573 | unsigned long timeout) | |
4574 | { | |
4575 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4576 | } | |
4577 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4578 | ||
be4de352 DC |
4579 | /** |
4580 | * try_wait_for_completion - try to decrement a completion without blocking | |
4581 | * @x: completion structure | |
4582 | * | |
4583 | * Returns: 0 if a decrement cannot be done without blocking | |
4584 | * 1 if a decrement succeeded. | |
4585 | * | |
4586 | * If a completion is being used as a counting completion, | |
4587 | * attempt to decrement the counter without blocking. This | |
4588 | * enables us to avoid waiting if the resource the completion | |
4589 | * is protecting is not available. | |
4590 | */ | |
4591 | bool try_wait_for_completion(struct completion *x) | |
4592 | { | |
7539a3b3 | 4593 | unsigned long flags; |
be4de352 DC |
4594 | int ret = 1; |
4595 | ||
7539a3b3 | 4596 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4597 | if (!x->done) |
4598 | ret = 0; | |
4599 | else | |
4600 | x->done--; | |
7539a3b3 | 4601 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4602 | return ret; |
4603 | } | |
4604 | EXPORT_SYMBOL(try_wait_for_completion); | |
4605 | ||
4606 | /** | |
4607 | * completion_done - Test to see if a completion has any waiters | |
4608 | * @x: completion structure | |
4609 | * | |
4610 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4611 | * 1 if there are no waiters. | |
4612 | * | |
4613 | */ | |
4614 | bool completion_done(struct completion *x) | |
4615 | { | |
7539a3b3 | 4616 | unsigned long flags; |
be4de352 DC |
4617 | int ret = 1; |
4618 | ||
7539a3b3 | 4619 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4620 | if (!x->done) |
4621 | ret = 0; | |
7539a3b3 | 4622 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4623 | return ret; |
4624 | } | |
4625 | EXPORT_SYMBOL(completion_done); | |
4626 | ||
8cbbe86d AK |
4627 | static long __sched |
4628 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4629 | { |
0fec171c IM |
4630 | unsigned long flags; |
4631 | wait_queue_t wait; | |
4632 | ||
4633 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4634 | |
8cbbe86d | 4635 | __set_current_state(state); |
1da177e4 | 4636 | |
8cbbe86d AK |
4637 | spin_lock_irqsave(&q->lock, flags); |
4638 | __add_wait_queue(q, &wait); | |
4639 | spin_unlock(&q->lock); | |
4640 | timeout = schedule_timeout(timeout); | |
4641 | spin_lock_irq(&q->lock); | |
4642 | __remove_wait_queue(q, &wait); | |
4643 | spin_unlock_irqrestore(&q->lock, flags); | |
4644 | ||
4645 | return timeout; | |
4646 | } | |
4647 | ||
4648 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4649 | { | |
4650 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4651 | } |
1da177e4 LT |
4652 | EXPORT_SYMBOL(interruptible_sleep_on); |
4653 | ||
0fec171c | 4654 | long __sched |
95cdf3b7 | 4655 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4656 | { |
8cbbe86d | 4657 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4658 | } |
1da177e4 LT |
4659 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4660 | ||
0fec171c | 4661 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4662 | { |
8cbbe86d | 4663 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4664 | } |
1da177e4 LT |
4665 | EXPORT_SYMBOL(sleep_on); |
4666 | ||
0fec171c | 4667 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4668 | { |
8cbbe86d | 4669 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4670 | } |
1da177e4 LT |
4671 | EXPORT_SYMBOL(sleep_on_timeout); |
4672 | ||
b29739f9 IM |
4673 | #ifdef CONFIG_RT_MUTEXES |
4674 | ||
4675 | /* | |
4676 | * rt_mutex_setprio - set the current priority of a task | |
4677 | * @p: task | |
4678 | * @prio: prio value (kernel-internal form) | |
4679 | * | |
4680 | * This function changes the 'effective' priority of a task. It does | |
4681 | * not touch ->normal_prio like __setscheduler(). | |
4682 | * | |
4683 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4684 | */ | |
36c8b586 | 4685 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4686 | { |
4687 | unsigned long flags; | |
83b699ed | 4688 | int oldprio, on_rq, running; |
70b97a7f | 4689 | struct rq *rq; |
83ab0aa0 | 4690 | const struct sched_class *prev_class; |
b29739f9 IM |
4691 | |
4692 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4693 | ||
4694 | rq = task_rq_lock(p, &flags); | |
4695 | ||
a8027073 | 4696 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4697 | oldprio = p->prio; |
83ab0aa0 | 4698 | prev_class = p->sched_class; |
dd41f596 | 4699 | on_rq = p->se.on_rq; |
051a1d1a | 4700 | running = task_current(rq, p); |
0e1f3483 | 4701 | if (on_rq) |
69be72c1 | 4702 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4703 | if (running) |
4704 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4705 | |
4706 | if (rt_prio(prio)) | |
4707 | p->sched_class = &rt_sched_class; | |
4708 | else | |
4709 | p->sched_class = &fair_sched_class; | |
4710 | ||
b29739f9 IM |
4711 | p->prio = prio; |
4712 | ||
0e1f3483 HS |
4713 | if (running) |
4714 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4715 | if (on_rq) |
371fd7e7 | 4716 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4717 | |
da7a735e | 4718 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 IM |
4719 | task_rq_unlock(rq, &flags); |
4720 | } | |
4721 | ||
4722 | #endif | |
4723 | ||
36c8b586 | 4724 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4725 | { |
dd41f596 | 4726 | int old_prio, delta, on_rq; |
1da177e4 | 4727 | unsigned long flags; |
70b97a7f | 4728 | struct rq *rq; |
1da177e4 LT |
4729 | |
4730 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4731 | return; | |
4732 | /* | |
4733 | * We have to be careful, if called from sys_setpriority(), | |
4734 | * the task might be in the middle of scheduling on another CPU. | |
4735 | */ | |
4736 | rq = task_rq_lock(p, &flags); | |
4737 | /* | |
4738 | * The RT priorities are set via sched_setscheduler(), but we still | |
4739 | * allow the 'normal' nice value to be set - but as expected | |
4740 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4741 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4742 | */ |
e05606d3 | 4743 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4744 | p->static_prio = NICE_TO_PRIO(nice); |
4745 | goto out_unlock; | |
4746 | } | |
dd41f596 | 4747 | on_rq = p->se.on_rq; |
c09595f6 | 4748 | if (on_rq) |
69be72c1 | 4749 | dequeue_task(rq, p, 0); |
1da177e4 | 4750 | |
1da177e4 | 4751 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4752 | set_load_weight(p); |
b29739f9 IM |
4753 | old_prio = p->prio; |
4754 | p->prio = effective_prio(p); | |
4755 | delta = p->prio - old_prio; | |
1da177e4 | 4756 | |
dd41f596 | 4757 | if (on_rq) { |
371fd7e7 | 4758 | enqueue_task(rq, p, 0); |
1da177e4 | 4759 | /* |
d5f9f942 AM |
4760 | * If the task increased its priority or is running and |
4761 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4762 | */ |
d5f9f942 | 4763 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4764 | resched_task(rq->curr); |
4765 | } | |
4766 | out_unlock: | |
4767 | task_rq_unlock(rq, &flags); | |
4768 | } | |
1da177e4 LT |
4769 | EXPORT_SYMBOL(set_user_nice); |
4770 | ||
e43379f1 MM |
4771 | /* |
4772 | * can_nice - check if a task can reduce its nice value | |
4773 | * @p: task | |
4774 | * @nice: nice value | |
4775 | */ | |
36c8b586 | 4776 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4777 | { |
024f4747 MM |
4778 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4779 | int nice_rlim = 20 - nice; | |
48f24c4d | 4780 | |
78d7d407 | 4781 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4782 | capable(CAP_SYS_NICE)); |
4783 | } | |
4784 | ||
1da177e4 LT |
4785 | #ifdef __ARCH_WANT_SYS_NICE |
4786 | ||
4787 | /* | |
4788 | * sys_nice - change the priority of the current process. | |
4789 | * @increment: priority increment | |
4790 | * | |
4791 | * sys_setpriority is a more generic, but much slower function that | |
4792 | * does similar things. | |
4793 | */ | |
5add95d4 | 4794 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4795 | { |
48f24c4d | 4796 | long nice, retval; |
1da177e4 LT |
4797 | |
4798 | /* | |
4799 | * Setpriority might change our priority at the same moment. | |
4800 | * We don't have to worry. Conceptually one call occurs first | |
4801 | * and we have a single winner. | |
4802 | */ | |
e43379f1 MM |
4803 | if (increment < -40) |
4804 | increment = -40; | |
1da177e4 LT |
4805 | if (increment > 40) |
4806 | increment = 40; | |
4807 | ||
2b8f836f | 4808 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4809 | if (nice < -20) |
4810 | nice = -20; | |
4811 | if (nice > 19) | |
4812 | nice = 19; | |
4813 | ||
e43379f1 MM |
4814 | if (increment < 0 && !can_nice(current, nice)) |
4815 | return -EPERM; | |
4816 | ||
1da177e4 LT |
4817 | retval = security_task_setnice(current, nice); |
4818 | if (retval) | |
4819 | return retval; | |
4820 | ||
4821 | set_user_nice(current, nice); | |
4822 | return 0; | |
4823 | } | |
4824 | ||
4825 | #endif | |
4826 | ||
4827 | /** | |
4828 | * task_prio - return the priority value of a given task. | |
4829 | * @p: the task in question. | |
4830 | * | |
4831 | * This is the priority value as seen by users in /proc. | |
4832 | * RT tasks are offset by -200. Normal tasks are centered | |
4833 | * around 0, value goes from -16 to +15. | |
4834 | */ | |
36c8b586 | 4835 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4836 | { |
4837 | return p->prio - MAX_RT_PRIO; | |
4838 | } | |
4839 | ||
4840 | /** | |
4841 | * task_nice - return the nice value of a given task. | |
4842 | * @p: the task in question. | |
4843 | */ | |
36c8b586 | 4844 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4845 | { |
4846 | return TASK_NICE(p); | |
4847 | } | |
150d8bed | 4848 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4849 | |
4850 | /** | |
4851 | * idle_cpu - is a given cpu idle currently? | |
4852 | * @cpu: the processor in question. | |
4853 | */ | |
4854 | int idle_cpu(int cpu) | |
4855 | { | |
4856 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4857 | } | |
4858 | ||
1da177e4 LT |
4859 | /** |
4860 | * idle_task - return the idle task for a given cpu. | |
4861 | * @cpu: the processor in question. | |
4862 | */ | |
36c8b586 | 4863 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4864 | { |
4865 | return cpu_rq(cpu)->idle; | |
4866 | } | |
4867 | ||
4868 | /** | |
4869 | * find_process_by_pid - find a process with a matching PID value. | |
4870 | * @pid: the pid in question. | |
4871 | */ | |
a9957449 | 4872 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4873 | { |
228ebcbe | 4874 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4875 | } |
4876 | ||
4877 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4878 | static void |
4879 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4880 | { |
dd41f596 | 4881 | BUG_ON(p->se.on_rq); |
48f24c4d | 4882 | |
1da177e4 LT |
4883 | p->policy = policy; |
4884 | p->rt_priority = prio; | |
b29739f9 IM |
4885 | p->normal_prio = normal_prio(p); |
4886 | /* we are holding p->pi_lock already */ | |
4887 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4888 | if (rt_prio(p->prio)) |
4889 | p->sched_class = &rt_sched_class; | |
4890 | else | |
4891 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4892 | set_load_weight(p); |
1da177e4 LT |
4893 | } |
4894 | ||
c69e8d9c DH |
4895 | /* |
4896 | * check the target process has a UID that matches the current process's | |
4897 | */ | |
4898 | static bool check_same_owner(struct task_struct *p) | |
4899 | { | |
4900 | const struct cred *cred = current_cred(), *pcred; | |
4901 | bool match; | |
4902 | ||
4903 | rcu_read_lock(); | |
4904 | pcred = __task_cred(p); | |
b0e77598 SH |
4905 | if (cred->user->user_ns == pcred->user->user_ns) |
4906 | match = (cred->euid == pcred->euid || | |
4907 | cred->euid == pcred->uid); | |
4908 | else | |
4909 | match = false; | |
c69e8d9c DH |
4910 | rcu_read_unlock(); |
4911 | return match; | |
4912 | } | |
4913 | ||
961ccddd | 4914 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4915 | const struct sched_param *param, bool user) |
1da177e4 | 4916 | { |
83b699ed | 4917 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4918 | unsigned long flags; |
83ab0aa0 | 4919 | const struct sched_class *prev_class; |
70b97a7f | 4920 | struct rq *rq; |
ca94c442 | 4921 | int reset_on_fork; |
1da177e4 | 4922 | |
66e5393a SR |
4923 | /* may grab non-irq protected spin_locks */ |
4924 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4925 | recheck: |
4926 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4927 | if (policy < 0) { |
4928 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4929 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4930 | } else { |
4931 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4932 | policy &= ~SCHED_RESET_ON_FORK; | |
4933 | ||
4934 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4935 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4936 | policy != SCHED_IDLE) | |
4937 | return -EINVAL; | |
4938 | } | |
4939 | ||
1da177e4 LT |
4940 | /* |
4941 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4942 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4943 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4944 | */ |
4945 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4946 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4947 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4948 | return -EINVAL; |
e05606d3 | 4949 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4950 | return -EINVAL; |
4951 | ||
37e4ab3f OC |
4952 | /* |
4953 | * Allow unprivileged RT tasks to decrease priority: | |
4954 | */ | |
961ccddd | 4955 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4956 | if (rt_policy(policy)) { |
a44702e8 ON |
4957 | unsigned long rlim_rtprio = |
4958 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4959 | |
4960 | /* can't set/change the rt policy */ | |
4961 | if (policy != p->policy && !rlim_rtprio) | |
4962 | return -EPERM; | |
4963 | ||
4964 | /* can't increase priority */ | |
4965 | if (param->sched_priority > p->rt_priority && | |
4966 | param->sched_priority > rlim_rtprio) | |
4967 | return -EPERM; | |
4968 | } | |
c02aa73b | 4969 | |
dd41f596 | 4970 | /* |
c02aa73b DH |
4971 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4972 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4973 | */ |
c02aa73b DH |
4974 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4975 | if (!can_nice(p, TASK_NICE(p))) | |
4976 | return -EPERM; | |
4977 | } | |
5fe1d75f | 4978 | |
37e4ab3f | 4979 | /* can't change other user's priorities */ |
c69e8d9c | 4980 | if (!check_same_owner(p)) |
37e4ab3f | 4981 | return -EPERM; |
ca94c442 LP |
4982 | |
4983 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4984 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4985 | return -EPERM; | |
37e4ab3f | 4986 | } |
1da177e4 | 4987 | |
725aad24 | 4988 | if (user) { |
b0ae1981 | 4989 | retval = security_task_setscheduler(p); |
725aad24 JF |
4990 | if (retval) |
4991 | return retval; | |
4992 | } | |
4993 | ||
b29739f9 IM |
4994 | /* |
4995 | * make sure no PI-waiters arrive (or leave) while we are | |
4996 | * changing the priority of the task: | |
4997 | */ | |
1d615482 | 4998 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4999 | /* |
5000 | * To be able to change p->policy safely, the apropriate | |
5001 | * runqueue lock must be held. | |
5002 | */ | |
b29739f9 | 5003 | rq = __task_rq_lock(p); |
dc61b1d6 | 5004 | |
34f971f6 PZ |
5005 | /* |
5006 | * Changing the policy of the stop threads its a very bad idea | |
5007 | */ | |
5008 | if (p == rq->stop) { | |
5009 | __task_rq_unlock(rq); | |
5010 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5011 | return -EINVAL; | |
5012 | } | |
5013 | ||
dc61b1d6 PZ |
5014 | #ifdef CONFIG_RT_GROUP_SCHED |
5015 | if (user) { | |
5016 | /* | |
5017 | * Do not allow realtime tasks into groups that have no runtime | |
5018 | * assigned. | |
5019 | */ | |
5020 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5021 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5022 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
5023 | __task_rq_unlock(rq); |
5024 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5025 | return -EPERM; | |
5026 | } | |
5027 | } | |
5028 | #endif | |
5029 | ||
1da177e4 LT |
5030 | /* recheck policy now with rq lock held */ |
5031 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5032 | policy = oldpolicy = -1; | |
b29739f9 | 5033 | __task_rq_unlock(rq); |
1d615482 | 5034 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5035 | goto recheck; |
5036 | } | |
dd41f596 | 5037 | on_rq = p->se.on_rq; |
051a1d1a | 5038 | running = task_current(rq, p); |
0e1f3483 | 5039 | if (on_rq) |
2e1cb74a | 5040 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5041 | if (running) |
5042 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5043 | |
ca94c442 LP |
5044 | p->sched_reset_on_fork = reset_on_fork; |
5045 | ||
1da177e4 | 5046 | oldprio = p->prio; |
83ab0aa0 | 5047 | prev_class = p->sched_class; |
dd41f596 | 5048 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5049 | |
0e1f3483 HS |
5050 | if (running) |
5051 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5052 | if (on_rq) |
dd41f596 | 5053 | activate_task(rq, p, 0); |
cb469845 | 5054 | |
da7a735e | 5055 | check_class_changed(rq, p, prev_class, oldprio); |
b29739f9 | 5056 | __task_rq_unlock(rq); |
1d615482 | 5057 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 5058 | |
95e02ca9 TG |
5059 | rt_mutex_adjust_pi(p); |
5060 | ||
1da177e4 LT |
5061 | return 0; |
5062 | } | |
961ccddd RR |
5063 | |
5064 | /** | |
5065 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5066 | * @p: the task in question. | |
5067 | * @policy: new policy. | |
5068 | * @param: structure containing the new RT priority. | |
5069 | * | |
5070 | * NOTE that the task may be already dead. | |
5071 | */ | |
5072 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5073 | const struct sched_param *param) |
961ccddd RR |
5074 | { |
5075 | return __sched_setscheduler(p, policy, param, true); | |
5076 | } | |
1da177e4 LT |
5077 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5078 | ||
961ccddd RR |
5079 | /** |
5080 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5081 | * @p: the task in question. | |
5082 | * @policy: new policy. | |
5083 | * @param: structure containing the new RT priority. | |
5084 | * | |
5085 | * Just like sched_setscheduler, only don't bother checking if the | |
5086 | * current context has permission. For example, this is needed in | |
5087 | * stop_machine(): we create temporary high priority worker threads, | |
5088 | * but our caller might not have that capability. | |
5089 | */ | |
5090 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5091 | const struct sched_param *param) |
961ccddd RR |
5092 | { |
5093 | return __sched_setscheduler(p, policy, param, false); | |
5094 | } | |
5095 | ||
95cdf3b7 IM |
5096 | static int |
5097 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5098 | { |
1da177e4 LT |
5099 | struct sched_param lparam; |
5100 | struct task_struct *p; | |
36c8b586 | 5101 | int retval; |
1da177e4 LT |
5102 | |
5103 | if (!param || pid < 0) | |
5104 | return -EINVAL; | |
5105 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5106 | return -EFAULT; | |
5fe1d75f ON |
5107 | |
5108 | rcu_read_lock(); | |
5109 | retval = -ESRCH; | |
1da177e4 | 5110 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5111 | if (p != NULL) |
5112 | retval = sched_setscheduler(p, policy, &lparam); | |
5113 | rcu_read_unlock(); | |
36c8b586 | 5114 | |
1da177e4 LT |
5115 | return retval; |
5116 | } | |
5117 | ||
5118 | /** | |
5119 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5120 | * @pid: the pid in question. | |
5121 | * @policy: new policy. | |
5122 | * @param: structure containing the new RT priority. | |
5123 | */ | |
5add95d4 HC |
5124 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5125 | struct sched_param __user *, param) | |
1da177e4 | 5126 | { |
c21761f1 JB |
5127 | /* negative values for policy are not valid */ |
5128 | if (policy < 0) | |
5129 | return -EINVAL; | |
5130 | ||
1da177e4 LT |
5131 | return do_sched_setscheduler(pid, policy, param); |
5132 | } | |
5133 | ||
5134 | /** | |
5135 | * sys_sched_setparam - set/change the RT priority of a thread | |
5136 | * @pid: the pid in question. | |
5137 | * @param: structure containing the new RT priority. | |
5138 | */ | |
5add95d4 | 5139 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5140 | { |
5141 | return do_sched_setscheduler(pid, -1, param); | |
5142 | } | |
5143 | ||
5144 | /** | |
5145 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5146 | * @pid: the pid in question. | |
5147 | */ | |
5add95d4 | 5148 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5149 | { |
36c8b586 | 5150 | struct task_struct *p; |
3a5c359a | 5151 | int retval; |
1da177e4 LT |
5152 | |
5153 | if (pid < 0) | |
3a5c359a | 5154 | return -EINVAL; |
1da177e4 LT |
5155 | |
5156 | retval = -ESRCH; | |
5fe85be0 | 5157 | rcu_read_lock(); |
1da177e4 LT |
5158 | p = find_process_by_pid(pid); |
5159 | if (p) { | |
5160 | retval = security_task_getscheduler(p); | |
5161 | if (!retval) | |
ca94c442 LP |
5162 | retval = p->policy |
5163 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5164 | } |
5fe85be0 | 5165 | rcu_read_unlock(); |
1da177e4 LT |
5166 | return retval; |
5167 | } | |
5168 | ||
5169 | /** | |
ca94c442 | 5170 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5171 | * @pid: the pid in question. |
5172 | * @param: structure containing the RT priority. | |
5173 | */ | |
5add95d4 | 5174 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5175 | { |
5176 | struct sched_param lp; | |
36c8b586 | 5177 | struct task_struct *p; |
3a5c359a | 5178 | int retval; |
1da177e4 LT |
5179 | |
5180 | if (!param || pid < 0) | |
3a5c359a | 5181 | return -EINVAL; |
1da177e4 | 5182 | |
5fe85be0 | 5183 | rcu_read_lock(); |
1da177e4 LT |
5184 | p = find_process_by_pid(pid); |
5185 | retval = -ESRCH; | |
5186 | if (!p) | |
5187 | goto out_unlock; | |
5188 | ||
5189 | retval = security_task_getscheduler(p); | |
5190 | if (retval) | |
5191 | goto out_unlock; | |
5192 | ||
5193 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5194 | rcu_read_unlock(); |
1da177e4 LT |
5195 | |
5196 | /* | |
5197 | * This one might sleep, we cannot do it with a spinlock held ... | |
5198 | */ | |
5199 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5200 | ||
1da177e4 LT |
5201 | return retval; |
5202 | ||
5203 | out_unlock: | |
5fe85be0 | 5204 | rcu_read_unlock(); |
1da177e4 LT |
5205 | return retval; |
5206 | } | |
5207 | ||
96f874e2 | 5208 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5209 | { |
5a16f3d3 | 5210 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5211 | struct task_struct *p; |
5212 | int retval; | |
1da177e4 | 5213 | |
95402b38 | 5214 | get_online_cpus(); |
23f5d142 | 5215 | rcu_read_lock(); |
1da177e4 LT |
5216 | |
5217 | p = find_process_by_pid(pid); | |
5218 | if (!p) { | |
23f5d142 | 5219 | rcu_read_unlock(); |
95402b38 | 5220 | put_online_cpus(); |
1da177e4 LT |
5221 | return -ESRCH; |
5222 | } | |
5223 | ||
23f5d142 | 5224 | /* Prevent p going away */ |
1da177e4 | 5225 | get_task_struct(p); |
23f5d142 | 5226 | rcu_read_unlock(); |
1da177e4 | 5227 | |
5a16f3d3 RR |
5228 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5229 | retval = -ENOMEM; | |
5230 | goto out_put_task; | |
5231 | } | |
5232 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5233 | retval = -ENOMEM; | |
5234 | goto out_free_cpus_allowed; | |
5235 | } | |
1da177e4 | 5236 | retval = -EPERM; |
b0e77598 | 5237 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5238 | goto out_unlock; |
5239 | ||
b0ae1981 | 5240 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5241 | if (retval) |
5242 | goto out_unlock; | |
5243 | ||
5a16f3d3 RR |
5244 | cpuset_cpus_allowed(p, cpus_allowed); |
5245 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5246 | again: |
5a16f3d3 | 5247 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5248 | |
8707d8b8 | 5249 | if (!retval) { |
5a16f3d3 RR |
5250 | cpuset_cpus_allowed(p, cpus_allowed); |
5251 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5252 | /* |
5253 | * We must have raced with a concurrent cpuset | |
5254 | * update. Just reset the cpus_allowed to the | |
5255 | * cpuset's cpus_allowed | |
5256 | */ | |
5a16f3d3 | 5257 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5258 | goto again; |
5259 | } | |
5260 | } | |
1da177e4 | 5261 | out_unlock: |
5a16f3d3 RR |
5262 | free_cpumask_var(new_mask); |
5263 | out_free_cpus_allowed: | |
5264 | free_cpumask_var(cpus_allowed); | |
5265 | out_put_task: | |
1da177e4 | 5266 | put_task_struct(p); |
95402b38 | 5267 | put_online_cpus(); |
1da177e4 LT |
5268 | return retval; |
5269 | } | |
5270 | ||
5271 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5272 | struct cpumask *new_mask) |
1da177e4 | 5273 | { |
96f874e2 RR |
5274 | if (len < cpumask_size()) |
5275 | cpumask_clear(new_mask); | |
5276 | else if (len > cpumask_size()) | |
5277 | len = cpumask_size(); | |
5278 | ||
1da177e4 LT |
5279 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5280 | } | |
5281 | ||
5282 | /** | |
5283 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5284 | * @pid: pid of the process | |
5285 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5286 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5287 | */ | |
5add95d4 HC |
5288 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5289 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5290 | { |
5a16f3d3 | 5291 | cpumask_var_t new_mask; |
1da177e4 LT |
5292 | int retval; |
5293 | ||
5a16f3d3 RR |
5294 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5295 | return -ENOMEM; | |
1da177e4 | 5296 | |
5a16f3d3 RR |
5297 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5298 | if (retval == 0) | |
5299 | retval = sched_setaffinity(pid, new_mask); | |
5300 | free_cpumask_var(new_mask); | |
5301 | return retval; | |
1da177e4 LT |
5302 | } |
5303 | ||
96f874e2 | 5304 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5305 | { |
36c8b586 | 5306 | struct task_struct *p; |
31605683 TG |
5307 | unsigned long flags; |
5308 | struct rq *rq; | |
1da177e4 | 5309 | int retval; |
1da177e4 | 5310 | |
95402b38 | 5311 | get_online_cpus(); |
23f5d142 | 5312 | rcu_read_lock(); |
1da177e4 LT |
5313 | |
5314 | retval = -ESRCH; | |
5315 | p = find_process_by_pid(pid); | |
5316 | if (!p) | |
5317 | goto out_unlock; | |
5318 | ||
e7834f8f DQ |
5319 | retval = security_task_getscheduler(p); |
5320 | if (retval) | |
5321 | goto out_unlock; | |
5322 | ||
31605683 | 5323 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5324 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5325 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5326 | |
5327 | out_unlock: | |
23f5d142 | 5328 | rcu_read_unlock(); |
95402b38 | 5329 | put_online_cpus(); |
1da177e4 | 5330 | |
9531b62f | 5331 | return retval; |
1da177e4 LT |
5332 | } |
5333 | ||
5334 | /** | |
5335 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5336 | * @pid: pid of the process | |
5337 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5338 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5339 | */ | |
5add95d4 HC |
5340 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5341 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5342 | { |
5343 | int ret; | |
f17c8607 | 5344 | cpumask_var_t mask; |
1da177e4 | 5345 | |
84fba5ec | 5346 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5347 | return -EINVAL; |
5348 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5349 | return -EINVAL; |
5350 | ||
f17c8607 RR |
5351 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5352 | return -ENOMEM; | |
1da177e4 | 5353 | |
f17c8607 RR |
5354 | ret = sched_getaffinity(pid, mask); |
5355 | if (ret == 0) { | |
8bc037fb | 5356 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5357 | |
5358 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5359 | ret = -EFAULT; |
5360 | else | |
cd3d8031 | 5361 | ret = retlen; |
f17c8607 RR |
5362 | } |
5363 | free_cpumask_var(mask); | |
1da177e4 | 5364 | |
f17c8607 | 5365 | return ret; |
1da177e4 LT |
5366 | } |
5367 | ||
5368 | /** | |
5369 | * sys_sched_yield - yield the current processor to other threads. | |
5370 | * | |
dd41f596 IM |
5371 | * This function yields the current CPU to other tasks. If there are no |
5372 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5373 | */ |
5add95d4 | 5374 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5375 | { |
70b97a7f | 5376 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5377 | |
2d72376b | 5378 | schedstat_inc(rq, yld_count); |
4530d7ab | 5379 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5380 | |
5381 | /* | |
5382 | * Since we are going to call schedule() anyway, there's | |
5383 | * no need to preempt or enable interrupts: | |
5384 | */ | |
5385 | __release(rq->lock); | |
8a25d5de | 5386 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5387 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5388 | preempt_enable_no_resched(); |
5389 | ||
5390 | schedule(); | |
5391 | ||
5392 | return 0; | |
5393 | } | |
5394 | ||
d86ee480 PZ |
5395 | static inline int should_resched(void) |
5396 | { | |
5397 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5398 | } | |
5399 | ||
e7b38404 | 5400 | static void __cond_resched(void) |
1da177e4 | 5401 | { |
e7aaaa69 FW |
5402 | add_preempt_count(PREEMPT_ACTIVE); |
5403 | schedule(); | |
5404 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5405 | } |
5406 | ||
02b67cc3 | 5407 | int __sched _cond_resched(void) |
1da177e4 | 5408 | { |
d86ee480 | 5409 | if (should_resched()) { |
1da177e4 LT |
5410 | __cond_resched(); |
5411 | return 1; | |
5412 | } | |
5413 | return 0; | |
5414 | } | |
02b67cc3 | 5415 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5416 | |
5417 | /* | |
613afbf8 | 5418 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5419 | * call schedule, and on return reacquire the lock. |
5420 | * | |
41a2d6cf | 5421 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5422 | * operations here to prevent schedule() from being called twice (once via |
5423 | * spin_unlock(), once by hand). | |
5424 | */ | |
613afbf8 | 5425 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5426 | { |
d86ee480 | 5427 | int resched = should_resched(); |
6df3cecb JK |
5428 | int ret = 0; |
5429 | ||
f607c668 PZ |
5430 | lockdep_assert_held(lock); |
5431 | ||
95c354fe | 5432 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5433 | spin_unlock(lock); |
d86ee480 | 5434 | if (resched) |
95c354fe NP |
5435 | __cond_resched(); |
5436 | else | |
5437 | cpu_relax(); | |
6df3cecb | 5438 | ret = 1; |
1da177e4 | 5439 | spin_lock(lock); |
1da177e4 | 5440 | } |
6df3cecb | 5441 | return ret; |
1da177e4 | 5442 | } |
613afbf8 | 5443 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5444 | |
613afbf8 | 5445 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5446 | { |
5447 | BUG_ON(!in_softirq()); | |
5448 | ||
d86ee480 | 5449 | if (should_resched()) { |
98d82567 | 5450 | local_bh_enable(); |
1da177e4 LT |
5451 | __cond_resched(); |
5452 | local_bh_disable(); | |
5453 | return 1; | |
5454 | } | |
5455 | return 0; | |
5456 | } | |
613afbf8 | 5457 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5458 | |
1da177e4 LT |
5459 | /** |
5460 | * yield - yield the current processor to other threads. | |
5461 | * | |
72fd4a35 | 5462 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5463 | * thread runnable and calls sys_sched_yield(). |
5464 | */ | |
5465 | void __sched yield(void) | |
5466 | { | |
5467 | set_current_state(TASK_RUNNING); | |
5468 | sys_sched_yield(); | |
5469 | } | |
1da177e4 LT |
5470 | EXPORT_SYMBOL(yield); |
5471 | ||
d95f4122 MG |
5472 | /** |
5473 | * yield_to - yield the current processor to another thread in | |
5474 | * your thread group, or accelerate that thread toward the | |
5475 | * processor it's on. | |
5476 | * | |
5477 | * It's the caller's job to ensure that the target task struct | |
5478 | * can't go away on us before we can do any checks. | |
5479 | * | |
5480 | * Returns true if we indeed boosted the target task. | |
5481 | */ | |
5482 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5483 | { | |
5484 | struct task_struct *curr = current; | |
5485 | struct rq *rq, *p_rq; | |
5486 | unsigned long flags; | |
5487 | bool yielded = 0; | |
5488 | ||
5489 | local_irq_save(flags); | |
5490 | rq = this_rq(); | |
5491 | ||
5492 | again: | |
5493 | p_rq = task_rq(p); | |
5494 | double_rq_lock(rq, p_rq); | |
5495 | while (task_rq(p) != p_rq) { | |
5496 | double_rq_unlock(rq, p_rq); | |
5497 | goto again; | |
5498 | } | |
5499 | ||
5500 | if (!curr->sched_class->yield_to_task) | |
5501 | goto out; | |
5502 | ||
5503 | if (curr->sched_class != p->sched_class) | |
5504 | goto out; | |
5505 | ||
5506 | if (task_running(p_rq, p) || p->state) | |
5507 | goto out; | |
5508 | ||
5509 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5510 | if (yielded) { |
d95f4122 | 5511 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5512 | /* |
5513 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5514 | * fairness. | |
5515 | */ | |
5516 | if (preempt && rq != p_rq) | |
5517 | resched_task(p_rq->curr); | |
5518 | } | |
d95f4122 MG |
5519 | |
5520 | out: | |
5521 | double_rq_unlock(rq, p_rq); | |
5522 | local_irq_restore(flags); | |
5523 | ||
5524 | if (yielded) | |
5525 | schedule(); | |
5526 | ||
5527 | return yielded; | |
5528 | } | |
5529 | EXPORT_SYMBOL_GPL(yield_to); | |
5530 | ||
1da177e4 | 5531 | /* |
41a2d6cf | 5532 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5533 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5534 | */ |
5535 | void __sched io_schedule(void) | |
5536 | { | |
54d35f29 | 5537 | struct rq *rq = raw_rq(); |
1da177e4 | 5538 | |
0ff92245 | 5539 | delayacct_blkio_start(); |
1da177e4 | 5540 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5541 | blk_flush_plug(current); |
8f0dfc34 | 5542 | current->in_iowait = 1; |
1da177e4 | 5543 | schedule(); |
8f0dfc34 | 5544 | current->in_iowait = 0; |
1da177e4 | 5545 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5546 | delayacct_blkio_end(); |
1da177e4 | 5547 | } |
1da177e4 LT |
5548 | EXPORT_SYMBOL(io_schedule); |
5549 | ||
5550 | long __sched io_schedule_timeout(long timeout) | |
5551 | { | |
54d35f29 | 5552 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5553 | long ret; |
5554 | ||
0ff92245 | 5555 | delayacct_blkio_start(); |
1da177e4 | 5556 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5557 | blk_flush_plug(current); |
8f0dfc34 | 5558 | current->in_iowait = 1; |
1da177e4 | 5559 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5560 | current->in_iowait = 0; |
1da177e4 | 5561 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5562 | delayacct_blkio_end(); |
1da177e4 LT |
5563 | return ret; |
5564 | } | |
5565 | ||
5566 | /** | |
5567 | * sys_sched_get_priority_max - return maximum RT priority. | |
5568 | * @policy: scheduling class. | |
5569 | * | |
5570 | * this syscall returns the maximum rt_priority that can be used | |
5571 | * by a given scheduling class. | |
5572 | */ | |
5add95d4 | 5573 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5574 | { |
5575 | int ret = -EINVAL; | |
5576 | ||
5577 | switch (policy) { | |
5578 | case SCHED_FIFO: | |
5579 | case SCHED_RR: | |
5580 | ret = MAX_USER_RT_PRIO-1; | |
5581 | break; | |
5582 | case SCHED_NORMAL: | |
b0a9499c | 5583 | case SCHED_BATCH: |
dd41f596 | 5584 | case SCHED_IDLE: |
1da177e4 LT |
5585 | ret = 0; |
5586 | break; | |
5587 | } | |
5588 | return ret; | |
5589 | } | |
5590 | ||
5591 | /** | |
5592 | * sys_sched_get_priority_min - return minimum RT priority. | |
5593 | * @policy: scheduling class. | |
5594 | * | |
5595 | * this syscall returns the minimum rt_priority that can be used | |
5596 | * by a given scheduling class. | |
5597 | */ | |
5add95d4 | 5598 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5599 | { |
5600 | int ret = -EINVAL; | |
5601 | ||
5602 | switch (policy) { | |
5603 | case SCHED_FIFO: | |
5604 | case SCHED_RR: | |
5605 | ret = 1; | |
5606 | break; | |
5607 | case SCHED_NORMAL: | |
b0a9499c | 5608 | case SCHED_BATCH: |
dd41f596 | 5609 | case SCHED_IDLE: |
1da177e4 LT |
5610 | ret = 0; |
5611 | } | |
5612 | return ret; | |
5613 | } | |
5614 | ||
5615 | /** | |
5616 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5617 | * @pid: pid of the process. | |
5618 | * @interval: userspace pointer to the timeslice value. | |
5619 | * | |
5620 | * this syscall writes the default timeslice value of a given process | |
5621 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5622 | */ | |
17da2bd9 | 5623 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5624 | struct timespec __user *, interval) |
1da177e4 | 5625 | { |
36c8b586 | 5626 | struct task_struct *p; |
a4ec24b4 | 5627 | unsigned int time_slice; |
dba091b9 TG |
5628 | unsigned long flags; |
5629 | struct rq *rq; | |
3a5c359a | 5630 | int retval; |
1da177e4 | 5631 | struct timespec t; |
1da177e4 LT |
5632 | |
5633 | if (pid < 0) | |
3a5c359a | 5634 | return -EINVAL; |
1da177e4 LT |
5635 | |
5636 | retval = -ESRCH; | |
1a551ae7 | 5637 | rcu_read_lock(); |
1da177e4 LT |
5638 | p = find_process_by_pid(pid); |
5639 | if (!p) | |
5640 | goto out_unlock; | |
5641 | ||
5642 | retval = security_task_getscheduler(p); | |
5643 | if (retval) | |
5644 | goto out_unlock; | |
5645 | ||
dba091b9 TG |
5646 | rq = task_rq_lock(p, &flags); |
5647 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5648 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5649 | |
1a551ae7 | 5650 | rcu_read_unlock(); |
a4ec24b4 | 5651 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5652 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5653 | return retval; |
3a5c359a | 5654 | |
1da177e4 | 5655 | out_unlock: |
1a551ae7 | 5656 | rcu_read_unlock(); |
1da177e4 LT |
5657 | return retval; |
5658 | } | |
5659 | ||
7c731e0a | 5660 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5661 | |
82a1fcb9 | 5662 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5663 | { |
1da177e4 | 5664 | unsigned long free = 0; |
36c8b586 | 5665 | unsigned state; |
1da177e4 | 5666 | |
1da177e4 | 5667 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5668 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5669 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5670 | #if BITS_PER_LONG == 32 |
1da177e4 | 5671 | if (state == TASK_RUNNING) |
3df0fc5b | 5672 | printk(KERN_CONT " running "); |
1da177e4 | 5673 | else |
3df0fc5b | 5674 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5675 | #else |
5676 | if (state == TASK_RUNNING) | |
3df0fc5b | 5677 | printk(KERN_CONT " running task "); |
1da177e4 | 5678 | else |
3df0fc5b | 5679 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5680 | #endif |
5681 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5682 | free = stack_not_used(p); |
1da177e4 | 5683 | #endif |
3df0fc5b | 5684 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5685 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5686 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5687 | |
5fb5e6de | 5688 | show_stack(p, NULL); |
1da177e4 LT |
5689 | } |
5690 | ||
e59e2ae2 | 5691 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5692 | { |
36c8b586 | 5693 | struct task_struct *g, *p; |
1da177e4 | 5694 | |
4bd77321 | 5695 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5696 | printk(KERN_INFO |
5697 | " task PC stack pid father\n"); | |
1da177e4 | 5698 | #else |
3df0fc5b PZ |
5699 | printk(KERN_INFO |
5700 | " task PC stack pid father\n"); | |
1da177e4 LT |
5701 | #endif |
5702 | read_lock(&tasklist_lock); | |
5703 | do_each_thread(g, p) { | |
5704 | /* | |
5705 | * reset the NMI-timeout, listing all files on a slow | |
5706 | * console might take alot of time: | |
5707 | */ | |
5708 | touch_nmi_watchdog(); | |
39bc89fd | 5709 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5710 | sched_show_task(p); |
1da177e4 LT |
5711 | } while_each_thread(g, p); |
5712 | ||
04c9167f JF |
5713 | touch_all_softlockup_watchdogs(); |
5714 | ||
dd41f596 IM |
5715 | #ifdef CONFIG_SCHED_DEBUG |
5716 | sysrq_sched_debug_show(); | |
5717 | #endif | |
1da177e4 | 5718 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5719 | /* |
5720 | * Only show locks if all tasks are dumped: | |
5721 | */ | |
93335a21 | 5722 | if (!state_filter) |
e59e2ae2 | 5723 | debug_show_all_locks(); |
1da177e4 LT |
5724 | } |
5725 | ||
1df21055 IM |
5726 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5727 | { | |
dd41f596 | 5728 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5729 | } |
5730 | ||
f340c0d1 IM |
5731 | /** |
5732 | * init_idle - set up an idle thread for a given CPU | |
5733 | * @idle: task in question | |
5734 | * @cpu: cpu the idle task belongs to | |
5735 | * | |
5736 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5737 | * flag, to make booting more robust. | |
5738 | */ | |
5c1e1767 | 5739 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5740 | { |
70b97a7f | 5741 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5742 | unsigned long flags; |
5743 | ||
05fa785c | 5744 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5745 | |
dd41f596 | 5746 | __sched_fork(idle); |
06b83b5f | 5747 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5748 | idle->se.exec_start = sched_clock(); |
5749 | ||
96f874e2 | 5750 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5751 | /* |
5752 | * We're having a chicken and egg problem, even though we are | |
5753 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5754 | * lockdep check in task_group() will fail. | |
5755 | * | |
5756 | * Similar case to sched_fork(). / Alternatively we could | |
5757 | * use task_rq_lock() here and obtain the other rq->lock. | |
5758 | * | |
5759 | * Silence PROVE_RCU | |
5760 | */ | |
5761 | rcu_read_lock(); | |
dd41f596 | 5762 | __set_task_cpu(idle, cpu); |
6506cf6c | 5763 | rcu_read_unlock(); |
1da177e4 | 5764 | |
1da177e4 | 5765 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5766 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5767 | idle->oncpu = 1; | |
5768 | #endif | |
05fa785c | 5769 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5770 | |
5771 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5772 | #if defined(CONFIG_PREEMPT) |
5773 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5774 | #else | |
a1261f54 | 5775 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5776 | #endif |
dd41f596 IM |
5777 | /* |
5778 | * The idle tasks have their own, simple scheduling class: | |
5779 | */ | |
5780 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5781 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5782 | } |
5783 | ||
5784 | /* | |
5785 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5786 | * indicates which cpus entered this state. This is used | |
5787 | * in the rcu update to wait only for active cpus. For system | |
5788 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5789 | * always be CPU_BITS_NONE. |
1da177e4 | 5790 | */ |
6a7b3dc3 | 5791 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5792 | |
19978ca6 IM |
5793 | /* |
5794 | * Increase the granularity value when there are more CPUs, | |
5795 | * because with more CPUs the 'effective latency' as visible | |
5796 | * to users decreases. But the relationship is not linear, | |
5797 | * so pick a second-best guess by going with the log2 of the | |
5798 | * number of CPUs. | |
5799 | * | |
5800 | * This idea comes from the SD scheduler of Con Kolivas: | |
5801 | */ | |
acb4a848 | 5802 | static int get_update_sysctl_factor(void) |
19978ca6 | 5803 | { |
4ca3ef71 | 5804 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5805 | unsigned int factor; |
5806 | ||
5807 | switch (sysctl_sched_tunable_scaling) { | |
5808 | case SCHED_TUNABLESCALING_NONE: | |
5809 | factor = 1; | |
5810 | break; | |
5811 | case SCHED_TUNABLESCALING_LINEAR: | |
5812 | factor = cpus; | |
5813 | break; | |
5814 | case SCHED_TUNABLESCALING_LOG: | |
5815 | default: | |
5816 | factor = 1 + ilog2(cpus); | |
5817 | break; | |
5818 | } | |
19978ca6 | 5819 | |
acb4a848 CE |
5820 | return factor; |
5821 | } | |
19978ca6 | 5822 | |
acb4a848 CE |
5823 | static void update_sysctl(void) |
5824 | { | |
5825 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5826 | |
0bcdcf28 CE |
5827 | #define SET_SYSCTL(name) \ |
5828 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5829 | SET_SYSCTL(sched_min_granularity); | |
5830 | SET_SYSCTL(sched_latency); | |
5831 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5832 | #undef SET_SYSCTL |
5833 | } | |
55cd5340 | 5834 | |
0bcdcf28 CE |
5835 | static inline void sched_init_granularity(void) |
5836 | { | |
5837 | update_sysctl(); | |
19978ca6 IM |
5838 | } |
5839 | ||
1da177e4 LT |
5840 | #ifdef CONFIG_SMP |
5841 | /* | |
5842 | * This is how migration works: | |
5843 | * | |
969c7921 TH |
5844 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5845 | * stop_one_cpu(). | |
5846 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5847 | * off the CPU) | |
5848 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5849 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5850 | * it and puts it into the right queue. |
969c7921 TH |
5851 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5852 | * is done. | |
1da177e4 LT |
5853 | */ |
5854 | ||
5855 | /* | |
5856 | * Change a given task's CPU affinity. Migrate the thread to a | |
5857 | * proper CPU and schedule it away if the CPU it's executing on | |
5858 | * is removed from the allowed bitmask. | |
5859 | * | |
5860 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5861 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5862 | * call is not atomic; no spinlocks may be held. |
5863 | */ | |
96f874e2 | 5864 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5865 | { |
5866 | unsigned long flags; | |
70b97a7f | 5867 | struct rq *rq; |
969c7921 | 5868 | unsigned int dest_cpu; |
48f24c4d | 5869 | int ret = 0; |
1da177e4 | 5870 | |
65cc8e48 PZ |
5871 | /* |
5872 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5873 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5874 | */ | |
5875 | again: | |
5876 | while (task_is_waking(p)) | |
5877 | cpu_relax(); | |
1da177e4 | 5878 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5879 | if (task_is_waking(p)) { |
5880 | task_rq_unlock(rq, &flags); | |
5881 | goto again; | |
5882 | } | |
e2912009 | 5883 | |
6ad4c188 | 5884 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5885 | ret = -EINVAL; |
5886 | goto out; | |
5887 | } | |
5888 | ||
9985b0ba | 5889 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5890 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5891 | ret = -EINVAL; |
5892 | goto out; | |
5893 | } | |
5894 | ||
73fe6aae | 5895 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5896 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5897 | else { |
96f874e2 RR |
5898 | cpumask_copy(&p->cpus_allowed, new_mask); |
5899 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5900 | } |
5901 | ||
1da177e4 | 5902 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5903 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5904 | goto out; |
5905 | ||
969c7921 | 5906 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5907 | if (migrate_task(p, rq)) { |
969c7921 | 5908 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5909 | /* Need help from migration thread: drop lock and wait. */ |
5910 | task_rq_unlock(rq, &flags); | |
969c7921 | 5911 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5912 | tlb_migrate_finish(p->mm); |
5913 | return 0; | |
5914 | } | |
5915 | out: | |
5916 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5917 | |
1da177e4 LT |
5918 | return ret; |
5919 | } | |
cd8ba7cd | 5920 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5921 | |
5922 | /* | |
41a2d6cf | 5923 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5924 | * this because either it can't run here any more (set_cpus_allowed() |
5925 | * away from this CPU, or CPU going down), or because we're | |
5926 | * attempting to rebalance this task on exec (sched_exec). | |
5927 | * | |
5928 | * So we race with normal scheduler movements, but that's OK, as long | |
5929 | * as the task is no longer on this CPU. | |
efc30814 KK |
5930 | * |
5931 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5932 | */ |
efc30814 | 5933 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5934 | { |
70b97a7f | 5935 | struct rq *rq_dest, *rq_src; |
e2912009 | 5936 | int ret = 0; |
1da177e4 | 5937 | |
e761b772 | 5938 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5939 | return ret; |
1da177e4 LT |
5940 | |
5941 | rq_src = cpu_rq(src_cpu); | |
5942 | rq_dest = cpu_rq(dest_cpu); | |
5943 | ||
5944 | double_rq_lock(rq_src, rq_dest); | |
5945 | /* Already moved. */ | |
5946 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5947 | goto done; |
1da177e4 | 5948 | /* Affinity changed (again). */ |
96f874e2 | 5949 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5950 | goto fail; |
1da177e4 | 5951 | |
e2912009 PZ |
5952 | /* |
5953 | * If we're not on a rq, the next wake-up will ensure we're | |
5954 | * placed properly. | |
5955 | */ | |
5956 | if (p->se.on_rq) { | |
2e1cb74a | 5957 | deactivate_task(rq_src, p, 0); |
e2912009 | 5958 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5959 | activate_task(rq_dest, p, 0); |
15afe09b | 5960 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5961 | } |
b1e38734 | 5962 | done: |
efc30814 | 5963 | ret = 1; |
b1e38734 | 5964 | fail: |
1da177e4 | 5965 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5966 | return ret; |
1da177e4 LT |
5967 | } |
5968 | ||
5969 | /* | |
969c7921 TH |
5970 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5971 | * and performs thread migration by bumping thread off CPU then | |
5972 | * 'pushing' onto another runqueue. | |
1da177e4 | 5973 | */ |
969c7921 | 5974 | static int migration_cpu_stop(void *data) |
1da177e4 | 5975 | { |
969c7921 | 5976 | struct migration_arg *arg = data; |
f7b4cddc | 5977 | |
969c7921 TH |
5978 | /* |
5979 | * The original target cpu might have gone down and we might | |
5980 | * be on another cpu but it doesn't matter. | |
5981 | */ | |
f7b4cddc | 5982 | local_irq_disable(); |
969c7921 | 5983 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5984 | local_irq_enable(); |
1da177e4 | 5985 | return 0; |
f7b4cddc ON |
5986 | } |
5987 | ||
1da177e4 | 5988 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5989 | |
054b9108 | 5990 | /* |
48c5ccae PZ |
5991 | * Ensures that the idle task is using init_mm right before its cpu goes |
5992 | * offline. | |
054b9108 | 5993 | */ |
48c5ccae | 5994 | void idle_task_exit(void) |
1da177e4 | 5995 | { |
48c5ccae | 5996 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5997 | |
48c5ccae | 5998 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5999 | |
48c5ccae PZ |
6000 | if (mm != &init_mm) |
6001 | switch_mm(mm, &init_mm, current); | |
6002 | mmdrop(mm); | |
1da177e4 LT |
6003 | } |
6004 | ||
6005 | /* | |
6006 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6007 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6008 | * for performance reasons the counter is not stricly tracking tasks to | |
6009 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6010 | * to keep the global sum constant after CPU-down: | |
6011 | */ | |
70b97a7f | 6012 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6013 | { |
6ad4c188 | 6014 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6015 | |
1da177e4 LT |
6016 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6017 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6018 | } |
6019 | ||
dd41f596 | 6020 | /* |
48c5ccae | 6021 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6022 | */ |
48c5ccae | 6023 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6024 | { |
48c5ccae PZ |
6025 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6026 | rq->calc_load_active = 0; | |
1da177e4 LT |
6027 | } |
6028 | ||
48f24c4d | 6029 | /* |
48c5ccae PZ |
6030 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6031 | * try_to_wake_up()->select_task_rq(). | |
6032 | * | |
6033 | * Called with rq->lock held even though we'er in stop_machine() and | |
6034 | * there's no concurrency possible, we hold the required locks anyway | |
6035 | * because of lock validation efforts. | |
1da177e4 | 6036 | */ |
48c5ccae | 6037 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6038 | { |
70b97a7f | 6039 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6040 | struct task_struct *next, *stop = rq->stop; |
6041 | int dest_cpu; | |
1da177e4 LT |
6042 | |
6043 | /* | |
48c5ccae PZ |
6044 | * Fudge the rq selection such that the below task selection loop |
6045 | * doesn't get stuck on the currently eligible stop task. | |
6046 | * | |
6047 | * We're currently inside stop_machine() and the rq is either stuck | |
6048 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6049 | * either way we should never end up calling schedule() until we're | |
6050 | * done here. | |
1da177e4 | 6051 | */ |
48c5ccae | 6052 | rq->stop = NULL; |
48f24c4d | 6053 | |
dd41f596 | 6054 | for ( ; ; ) { |
48c5ccae PZ |
6055 | /* |
6056 | * There's this thread running, bail when that's the only | |
6057 | * remaining thread. | |
6058 | */ | |
6059 | if (rq->nr_running == 1) | |
dd41f596 | 6060 | break; |
48c5ccae | 6061 | |
b67802ea | 6062 | next = pick_next_task(rq); |
48c5ccae | 6063 | BUG_ON(!next); |
79c53799 | 6064 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6065 | |
48c5ccae PZ |
6066 | /* Find suitable destination for @next, with force if needed. */ |
6067 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6068 | raw_spin_unlock(&rq->lock); | |
6069 | ||
6070 | __migrate_task(next, dead_cpu, dest_cpu); | |
6071 | ||
6072 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6073 | } |
dce48a84 | 6074 | |
48c5ccae | 6075 | rq->stop = stop; |
dce48a84 | 6076 | } |
48c5ccae | 6077 | |
1da177e4 LT |
6078 | #endif /* CONFIG_HOTPLUG_CPU */ |
6079 | ||
e692ab53 NP |
6080 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6081 | ||
6082 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6083 | { |
6084 | .procname = "sched_domain", | |
c57baf1e | 6085 | .mode = 0555, |
e0361851 | 6086 | }, |
56992309 | 6087 | {} |
e692ab53 NP |
6088 | }; |
6089 | ||
6090 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6091 | { |
6092 | .procname = "kernel", | |
c57baf1e | 6093 | .mode = 0555, |
e0361851 AD |
6094 | .child = sd_ctl_dir, |
6095 | }, | |
56992309 | 6096 | {} |
e692ab53 NP |
6097 | }; |
6098 | ||
6099 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6100 | { | |
6101 | struct ctl_table *entry = | |
5cf9f062 | 6102 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6103 | |
e692ab53 NP |
6104 | return entry; |
6105 | } | |
6106 | ||
6382bc90 MM |
6107 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6108 | { | |
cd790076 | 6109 | struct ctl_table *entry; |
6382bc90 | 6110 | |
cd790076 MM |
6111 | /* |
6112 | * In the intermediate directories, both the child directory and | |
6113 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6114 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6115 | * static strings and all have proc handlers. |
6116 | */ | |
6117 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6118 | if (entry->child) |
6119 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6120 | if (entry->proc_handler == NULL) |
6121 | kfree(entry->procname); | |
6122 | } | |
6382bc90 MM |
6123 | |
6124 | kfree(*tablep); | |
6125 | *tablep = NULL; | |
6126 | } | |
6127 | ||
e692ab53 | 6128 | static void |
e0361851 | 6129 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6130 | const char *procname, void *data, int maxlen, |
6131 | mode_t mode, proc_handler *proc_handler) | |
6132 | { | |
e692ab53 NP |
6133 | entry->procname = procname; |
6134 | entry->data = data; | |
6135 | entry->maxlen = maxlen; | |
6136 | entry->mode = mode; | |
6137 | entry->proc_handler = proc_handler; | |
6138 | } | |
6139 | ||
6140 | static struct ctl_table * | |
6141 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6142 | { | |
a5d8c348 | 6143 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6144 | |
ad1cdc1d MM |
6145 | if (table == NULL) |
6146 | return NULL; | |
6147 | ||
e0361851 | 6148 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6149 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6150 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6151 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6152 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6153 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6154 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6155 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6156 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6157 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6158 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6159 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6160 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6161 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6162 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6163 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6164 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6165 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6166 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6167 | &sd->cache_nice_tries, |
6168 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6169 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6170 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6171 | set_table_entry(&table[11], "name", sd->name, |
6172 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6173 | /* &table[12] is terminator */ | |
e692ab53 NP |
6174 | |
6175 | return table; | |
6176 | } | |
6177 | ||
9a4e7159 | 6178 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6179 | { |
6180 | struct ctl_table *entry, *table; | |
6181 | struct sched_domain *sd; | |
6182 | int domain_num = 0, i; | |
6183 | char buf[32]; | |
6184 | ||
6185 | for_each_domain(cpu, sd) | |
6186 | domain_num++; | |
6187 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6188 | if (table == NULL) |
6189 | return NULL; | |
e692ab53 NP |
6190 | |
6191 | i = 0; | |
6192 | for_each_domain(cpu, sd) { | |
6193 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6194 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6195 | entry->mode = 0555; |
e692ab53 NP |
6196 | entry->child = sd_alloc_ctl_domain_table(sd); |
6197 | entry++; | |
6198 | i++; | |
6199 | } | |
6200 | return table; | |
6201 | } | |
6202 | ||
6203 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6204 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6205 | { |
6ad4c188 | 6206 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6207 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6208 | char buf[32]; | |
6209 | ||
7378547f MM |
6210 | WARN_ON(sd_ctl_dir[0].child); |
6211 | sd_ctl_dir[0].child = entry; | |
6212 | ||
ad1cdc1d MM |
6213 | if (entry == NULL) |
6214 | return; | |
6215 | ||
6ad4c188 | 6216 | for_each_possible_cpu(i) { |
e692ab53 | 6217 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6218 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6219 | entry->mode = 0555; |
e692ab53 | 6220 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6221 | entry++; |
e692ab53 | 6222 | } |
7378547f MM |
6223 | |
6224 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6225 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6226 | } | |
6382bc90 | 6227 | |
7378547f | 6228 | /* may be called multiple times per register */ |
6382bc90 MM |
6229 | static void unregister_sched_domain_sysctl(void) |
6230 | { | |
7378547f MM |
6231 | if (sd_sysctl_header) |
6232 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6233 | sd_sysctl_header = NULL; |
7378547f MM |
6234 | if (sd_ctl_dir[0].child) |
6235 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6236 | } |
e692ab53 | 6237 | #else |
6382bc90 MM |
6238 | static void register_sched_domain_sysctl(void) |
6239 | { | |
6240 | } | |
6241 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6242 | { |
6243 | } | |
6244 | #endif | |
6245 | ||
1f11eb6a GH |
6246 | static void set_rq_online(struct rq *rq) |
6247 | { | |
6248 | if (!rq->online) { | |
6249 | const struct sched_class *class; | |
6250 | ||
c6c4927b | 6251 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6252 | rq->online = 1; |
6253 | ||
6254 | for_each_class(class) { | |
6255 | if (class->rq_online) | |
6256 | class->rq_online(rq); | |
6257 | } | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | static void set_rq_offline(struct rq *rq) | |
6262 | { | |
6263 | if (rq->online) { | |
6264 | const struct sched_class *class; | |
6265 | ||
6266 | for_each_class(class) { | |
6267 | if (class->rq_offline) | |
6268 | class->rq_offline(rq); | |
6269 | } | |
6270 | ||
c6c4927b | 6271 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6272 | rq->online = 0; |
6273 | } | |
6274 | } | |
6275 | ||
1da177e4 LT |
6276 | /* |
6277 | * migration_call - callback that gets triggered when a CPU is added. | |
6278 | * Here we can start up the necessary migration thread for the new CPU. | |
6279 | */ | |
48f24c4d IM |
6280 | static int __cpuinit |
6281 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6282 | { |
48f24c4d | 6283 | int cpu = (long)hcpu; |
1da177e4 | 6284 | unsigned long flags; |
969c7921 | 6285 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6286 | |
48c5ccae | 6287 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6288 | |
1da177e4 | 6289 | case CPU_UP_PREPARE: |
a468d389 | 6290 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6291 | break; |
48f24c4d | 6292 | |
1da177e4 | 6293 | case CPU_ONLINE: |
1f94ef59 | 6294 | /* Update our root-domain */ |
05fa785c | 6295 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6296 | if (rq->rd) { |
c6c4927b | 6297 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6298 | |
6299 | set_rq_online(rq); | |
1f94ef59 | 6300 | } |
05fa785c | 6301 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6302 | break; |
48f24c4d | 6303 | |
1da177e4 | 6304 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6305 | case CPU_DYING: |
57d885fe | 6306 | /* Update our root-domain */ |
05fa785c | 6307 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6308 | if (rq->rd) { |
c6c4927b | 6309 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6310 | set_rq_offline(rq); |
57d885fe | 6311 | } |
48c5ccae PZ |
6312 | migrate_tasks(cpu); |
6313 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6314 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6315 | |
6316 | migrate_nr_uninterruptible(rq); | |
6317 | calc_global_load_remove(rq); | |
57d885fe | 6318 | break; |
1da177e4 LT |
6319 | #endif |
6320 | } | |
6321 | return NOTIFY_OK; | |
6322 | } | |
6323 | ||
f38b0820 PM |
6324 | /* |
6325 | * Register at high priority so that task migration (migrate_all_tasks) | |
6326 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6327 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6328 | */ |
26c2143b | 6329 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6330 | .notifier_call = migration_call, |
50a323b7 | 6331 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6332 | }; |
6333 | ||
3a101d05 TH |
6334 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6335 | unsigned long action, void *hcpu) | |
6336 | { | |
6337 | switch (action & ~CPU_TASKS_FROZEN) { | |
6338 | case CPU_ONLINE: | |
6339 | case CPU_DOWN_FAILED: | |
6340 | set_cpu_active((long)hcpu, true); | |
6341 | return NOTIFY_OK; | |
6342 | default: | |
6343 | return NOTIFY_DONE; | |
6344 | } | |
6345 | } | |
6346 | ||
6347 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6348 | unsigned long action, void *hcpu) | |
6349 | { | |
6350 | switch (action & ~CPU_TASKS_FROZEN) { | |
6351 | case CPU_DOWN_PREPARE: | |
6352 | set_cpu_active((long)hcpu, false); | |
6353 | return NOTIFY_OK; | |
6354 | default: | |
6355 | return NOTIFY_DONE; | |
6356 | } | |
6357 | } | |
6358 | ||
7babe8db | 6359 | static int __init migration_init(void) |
1da177e4 LT |
6360 | { |
6361 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6362 | int err; |
48f24c4d | 6363 | |
3a101d05 | 6364 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6365 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6366 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6367 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6368 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6369 | |
3a101d05 TH |
6370 | /* Register cpu active notifiers */ |
6371 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6372 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6373 | ||
a004cd42 | 6374 | return 0; |
1da177e4 | 6375 | } |
7babe8db | 6376 | early_initcall(migration_init); |
1da177e4 LT |
6377 | #endif |
6378 | ||
6379 | #ifdef CONFIG_SMP | |
476f3534 | 6380 | |
3e9830dc | 6381 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6382 | |
f6630114 MT |
6383 | static __read_mostly int sched_domain_debug_enabled; |
6384 | ||
6385 | static int __init sched_domain_debug_setup(char *str) | |
6386 | { | |
6387 | sched_domain_debug_enabled = 1; | |
6388 | ||
6389 | return 0; | |
6390 | } | |
6391 | early_param("sched_debug", sched_domain_debug_setup); | |
6392 | ||
7c16ec58 | 6393 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6394 | struct cpumask *groupmask) |
1da177e4 | 6395 | { |
4dcf6aff | 6396 | struct sched_group *group = sd->groups; |
434d53b0 | 6397 | char str[256]; |
1da177e4 | 6398 | |
968ea6d8 | 6399 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6400 | cpumask_clear(groupmask); |
4dcf6aff IM |
6401 | |
6402 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6403 | ||
6404 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6405 | printk("does not load-balance\n"); |
4dcf6aff | 6406 | if (sd->parent) |
3df0fc5b PZ |
6407 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6408 | " has parent"); | |
4dcf6aff | 6409 | return -1; |
41c7ce9a NP |
6410 | } |
6411 | ||
3df0fc5b | 6412 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6413 | |
758b2cdc | 6414 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6415 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6416 | "CPU%d\n", cpu); | |
4dcf6aff | 6417 | } |
758b2cdc | 6418 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6419 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6420 | " CPU%d\n", cpu); | |
4dcf6aff | 6421 | } |
1da177e4 | 6422 | |
4dcf6aff | 6423 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6424 | do { |
4dcf6aff | 6425 | if (!group) { |
3df0fc5b PZ |
6426 | printk("\n"); |
6427 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6428 | break; |
6429 | } | |
6430 | ||
18a3885f | 6431 | if (!group->cpu_power) { |
3df0fc5b PZ |
6432 | printk(KERN_CONT "\n"); |
6433 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6434 | "set\n"); | |
4dcf6aff IM |
6435 | break; |
6436 | } | |
1da177e4 | 6437 | |
758b2cdc | 6438 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6439 | printk(KERN_CONT "\n"); |
6440 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6441 | break; |
6442 | } | |
1da177e4 | 6443 | |
758b2cdc | 6444 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6445 | printk(KERN_CONT "\n"); |
6446 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6447 | break; |
6448 | } | |
1da177e4 | 6449 | |
758b2cdc | 6450 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6451 | |
968ea6d8 | 6452 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6453 | |
3df0fc5b | 6454 | printk(KERN_CONT " %s", str); |
18a3885f | 6455 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6456 | printk(KERN_CONT " (cpu_power = %d)", |
6457 | group->cpu_power); | |
381512cf | 6458 | } |
1da177e4 | 6459 | |
4dcf6aff IM |
6460 | group = group->next; |
6461 | } while (group != sd->groups); | |
3df0fc5b | 6462 | printk(KERN_CONT "\n"); |
1da177e4 | 6463 | |
758b2cdc | 6464 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6465 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6466 | |
758b2cdc RR |
6467 | if (sd->parent && |
6468 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6469 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6470 | "of domain->span\n"); | |
4dcf6aff IM |
6471 | return 0; |
6472 | } | |
1da177e4 | 6473 | |
4dcf6aff IM |
6474 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6475 | { | |
d5dd3db1 | 6476 | cpumask_var_t groupmask; |
4dcf6aff | 6477 | int level = 0; |
1da177e4 | 6478 | |
f6630114 MT |
6479 | if (!sched_domain_debug_enabled) |
6480 | return; | |
6481 | ||
4dcf6aff IM |
6482 | if (!sd) { |
6483 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6484 | return; | |
6485 | } | |
1da177e4 | 6486 | |
4dcf6aff IM |
6487 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6488 | ||
d5dd3db1 | 6489 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6490 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6491 | return; | |
6492 | } | |
6493 | ||
4dcf6aff | 6494 | for (;;) { |
7c16ec58 | 6495 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6496 | break; |
1da177e4 LT |
6497 | level++; |
6498 | sd = sd->parent; | |
33859f7f | 6499 | if (!sd) |
4dcf6aff IM |
6500 | break; |
6501 | } | |
d5dd3db1 | 6502 | free_cpumask_var(groupmask); |
1da177e4 | 6503 | } |
6d6bc0ad | 6504 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6505 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6506 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6507 | |
1a20ff27 | 6508 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6509 | { |
758b2cdc | 6510 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6511 | return 1; |
6512 | ||
6513 | /* Following flags need at least 2 groups */ | |
6514 | if (sd->flags & (SD_LOAD_BALANCE | | |
6515 | SD_BALANCE_NEWIDLE | | |
6516 | SD_BALANCE_FORK | | |
89c4710e SS |
6517 | SD_BALANCE_EXEC | |
6518 | SD_SHARE_CPUPOWER | | |
6519 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6520 | if (sd->groups != sd->groups->next) |
6521 | return 0; | |
6522 | } | |
6523 | ||
6524 | /* Following flags don't use groups */ | |
c88d5910 | 6525 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6526 | return 0; |
6527 | ||
6528 | return 1; | |
6529 | } | |
6530 | ||
48f24c4d IM |
6531 | static int |
6532 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6533 | { |
6534 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6535 | ||
6536 | if (sd_degenerate(parent)) | |
6537 | return 1; | |
6538 | ||
758b2cdc | 6539 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6540 | return 0; |
6541 | ||
245af2c7 SS |
6542 | /* Flags needing groups don't count if only 1 group in parent */ |
6543 | if (parent->groups == parent->groups->next) { | |
6544 | pflags &= ~(SD_LOAD_BALANCE | | |
6545 | SD_BALANCE_NEWIDLE | | |
6546 | SD_BALANCE_FORK | | |
89c4710e SS |
6547 | SD_BALANCE_EXEC | |
6548 | SD_SHARE_CPUPOWER | | |
6549 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6550 | if (nr_node_ids == 1) |
6551 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6552 | } |
6553 | if (~cflags & pflags) | |
6554 | return 0; | |
6555 | ||
6556 | return 1; | |
6557 | } | |
6558 | ||
c6c4927b RR |
6559 | static void free_rootdomain(struct root_domain *rd) |
6560 | { | |
047106ad PZ |
6561 | synchronize_sched(); |
6562 | ||
68e74568 RR |
6563 | cpupri_cleanup(&rd->cpupri); |
6564 | ||
c6c4927b RR |
6565 | free_cpumask_var(rd->rto_mask); |
6566 | free_cpumask_var(rd->online); | |
6567 | free_cpumask_var(rd->span); | |
6568 | kfree(rd); | |
6569 | } | |
6570 | ||
57d885fe GH |
6571 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6572 | { | |
a0490fa3 | 6573 | struct root_domain *old_rd = NULL; |
57d885fe | 6574 | unsigned long flags; |
57d885fe | 6575 | |
05fa785c | 6576 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6577 | |
6578 | if (rq->rd) { | |
a0490fa3 | 6579 | old_rd = rq->rd; |
57d885fe | 6580 | |
c6c4927b | 6581 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6582 | set_rq_offline(rq); |
57d885fe | 6583 | |
c6c4927b | 6584 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6585 | |
a0490fa3 IM |
6586 | /* |
6587 | * If we dont want to free the old_rt yet then | |
6588 | * set old_rd to NULL to skip the freeing later | |
6589 | * in this function: | |
6590 | */ | |
6591 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6592 | old_rd = NULL; | |
57d885fe GH |
6593 | } |
6594 | ||
6595 | atomic_inc(&rd->refcount); | |
6596 | rq->rd = rd; | |
6597 | ||
c6c4927b | 6598 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6599 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6600 | set_rq_online(rq); |
57d885fe | 6601 | |
05fa785c | 6602 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6603 | |
6604 | if (old_rd) | |
6605 | free_rootdomain(old_rd); | |
57d885fe GH |
6606 | } |
6607 | ||
68c38fc3 | 6608 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6609 | { |
6610 | memset(rd, 0, sizeof(*rd)); | |
6611 | ||
68c38fc3 | 6612 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6613 | goto out; |
68c38fc3 | 6614 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6615 | goto free_span; |
68c38fc3 | 6616 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6617 | goto free_online; |
6e0534f2 | 6618 | |
68c38fc3 | 6619 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6620 | goto free_rto_mask; |
c6c4927b | 6621 | return 0; |
6e0534f2 | 6622 | |
68e74568 RR |
6623 | free_rto_mask: |
6624 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6625 | free_online: |
6626 | free_cpumask_var(rd->online); | |
6627 | free_span: | |
6628 | free_cpumask_var(rd->span); | |
0c910d28 | 6629 | out: |
c6c4927b | 6630 | return -ENOMEM; |
57d885fe GH |
6631 | } |
6632 | ||
6633 | static void init_defrootdomain(void) | |
6634 | { | |
68c38fc3 | 6635 | init_rootdomain(&def_root_domain); |
c6c4927b | 6636 | |
57d885fe GH |
6637 | atomic_set(&def_root_domain.refcount, 1); |
6638 | } | |
6639 | ||
dc938520 | 6640 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6641 | { |
6642 | struct root_domain *rd; | |
6643 | ||
6644 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6645 | if (!rd) | |
6646 | return NULL; | |
6647 | ||
68c38fc3 | 6648 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6649 | kfree(rd); |
6650 | return NULL; | |
6651 | } | |
57d885fe GH |
6652 | |
6653 | return rd; | |
6654 | } | |
6655 | ||
1da177e4 | 6656 | /* |
0eab9146 | 6657 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6658 | * hold the hotplug lock. |
6659 | */ | |
0eab9146 IM |
6660 | static void |
6661 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6662 | { |
70b97a7f | 6663 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6664 | struct sched_domain *tmp; |
6665 | ||
669c55e9 PZ |
6666 | for (tmp = sd; tmp; tmp = tmp->parent) |
6667 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6668 | ||
245af2c7 | 6669 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6670 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6671 | struct sched_domain *parent = tmp->parent; |
6672 | if (!parent) | |
6673 | break; | |
f29c9b1c | 6674 | |
1a848870 | 6675 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6676 | tmp->parent = parent->parent; |
1a848870 SS |
6677 | if (parent->parent) |
6678 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6679 | } else |
6680 | tmp = tmp->parent; | |
245af2c7 SS |
6681 | } |
6682 | ||
1a848870 | 6683 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6684 | sd = sd->parent; |
1a848870 SS |
6685 | if (sd) |
6686 | sd->child = NULL; | |
6687 | } | |
1da177e4 LT |
6688 | |
6689 | sched_domain_debug(sd, cpu); | |
6690 | ||
57d885fe | 6691 | rq_attach_root(rq, rd); |
674311d5 | 6692 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6693 | } |
6694 | ||
6695 | /* cpus with isolated domains */ | |
dcc30a35 | 6696 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6697 | |
6698 | /* Setup the mask of cpus configured for isolated domains */ | |
6699 | static int __init isolated_cpu_setup(char *str) | |
6700 | { | |
bdddd296 | 6701 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6702 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6703 | return 1; |
6704 | } | |
6705 | ||
8927f494 | 6706 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6707 | |
6708 | /* | |
6711cab4 SS |
6709 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6710 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6711 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6712 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6713 | * |
6714 | * init_sched_build_groups will build a circular linked list of the groups | |
6715 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6716 | * and ->cpu_power to 0. | |
6717 | */ | |
a616058b | 6718 | static void |
96f874e2 RR |
6719 | init_sched_build_groups(const struct cpumask *span, |
6720 | const struct cpumask *cpu_map, | |
6721 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6722 | struct sched_group **sg, |
96f874e2 RR |
6723 | struct cpumask *tmpmask), |
6724 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6725 | { |
6726 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6727 | int i; |
6728 | ||
96f874e2 | 6729 | cpumask_clear(covered); |
7c16ec58 | 6730 | |
abcd083a | 6731 | for_each_cpu(i, span) { |
6711cab4 | 6732 | struct sched_group *sg; |
7c16ec58 | 6733 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6734 | int j; |
6735 | ||
758b2cdc | 6736 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6737 | continue; |
6738 | ||
758b2cdc | 6739 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6740 | sg->cpu_power = 0; |
1da177e4 | 6741 | |
abcd083a | 6742 | for_each_cpu(j, span) { |
7c16ec58 | 6743 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6744 | continue; |
6745 | ||
96f874e2 | 6746 | cpumask_set_cpu(j, covered); |
758b2cdc | 6747 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6748 | } |
6749 | if (!first) | |
6750 | first = sg; | |
6751 | if (last) | |
6752 | last->next = sg; | |
6753 | last = sg; | |
6754 | } | |
6755 | last->next = first; | |
6756 | } | |
6757 | ||
9c1cfda2 | 6758 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6759 | |
9c1cfda2 | 6760 | #ifdef CONFIG_NUMA |
198e2f18 | 6761 | |
9c1cfda2 JH |
6762 | /** |
6763 | * find_next_best_node - find the next node to include in a sched_domain | |
6764 | * @node: node whose sched_domain we're building | |
6765 | * @used_nodes: nodes already in the sched_domain | |
6766 | * | |
41a2d6cf | 6767 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6768 | * finds the closest node not already in the @used_nodes map. |
6769 | * | |
6770 | * Should use nodemask_t. | |
6771 | */ | |
c5f59f08 | 6772 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6773 | { |
6774 | int i, n, val, min_val, best_node = 0; | |
6775 | ||
6776 | min_val = INT_MAX; | |
6777 | ||
076ac2af | 6778 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6779 | /* Start at @node */ |
076ac2af | 6780 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6781 | |
6782 | if (!nr_cpus_node(n)) | |
6783 | continue; | |
6784 | ||
6785 | /* Skip already used nodes */ | |
c5f59f08 | 6786 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6787 | continue; |
6788 | ||
6789 | /* Simple min distance search */ | |
6790 | val = node_distance(node, n); | |
6791 | ||
6792 | if (val < min_val) { | |
6793 | min_val = val; | |
6794 | best_node = n; | |
6795 | } | |
6796 | } | |
6797 | ||
c5f59f08 | 6798 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6799 | return best_node; |
6800 | } | |
6801 | ||
6802 | /** | |
6803 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6804 | * @node: node whose cpumask we're constructing | |
73486722 | 6805 | * @span: resulting cpumask |
9c1cfda2 | 6806 | * |
41a2d6cf | 6807 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6808 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6809 | * out optimally. | |
6810 | */ | |
96f874e2 | 6811 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6812 | { |
c5f59f08 | 6813 | nodemask_t used_nodes; |
48f24c4d | 6814 | int i; |
9c1cfda2 | 6815 | |
6ca09dfc | 6816 | cpumask_clear(span); |
c5f59f08 | 6817 | nodes_clear(used_nodes); |
9c1cfda2 | 6818 | |
6ca09dfc | 6819 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6820 | node_set(node, used_nodes); |
9c1cfda2 JH |
6821 | |
6822 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6823 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6824 | |
6ca09dfc | 6825 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6826 | } |
9c1cfda2 | 6827 | } |
6d6bc0ad | 6828 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6829 | |
5c45bf27 | 6830 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6831 | |
6c99e9ad RR |
6832 | /* |
6833 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6834 | * |
6835 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6836 | * and struct sched_domain. ) | |
6c99e9ad RR |
6837 | */ |
6838 | struct static_sched_group { | |
6839 | struct sched_group sg; | |
6840 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6841 | }; | |
6842 | ||
6843 | struct static_sched_domain { | |
6844 | struct sched_domain sd; | |
6845 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6846 | }; | |
6847 | ||
49a02c51 AH |
6848 | struct s_data { |
6849 | #ifdef CONFIG_NUMA | |
6850 | int sd_allnodes; | |
6851 | cpumask_var_t domainspan; | |
6852 | cpumask_var_t covered; | |
6853 | cpumask_var_t notcovered; | |
6854 | #endif | |
6855 | cpumask_var_t nodemask; | |
6856 | cpumask_var_t this_sibling_map; | |
6857 | cpumask_var_t this_core_map; | |
01a08546 | 6858 | cpumask_var_t this_book_map; |
49a02c51 AH |
6859 | cpumask_var_t send_covered; |
6860 | cpumask_var_t tmpmask; | |
6861 | struct sched_group **sched_group_nodes; | |
6862 | struct root_domain *rd; | |
6863 | }; | |
6864 | ||
2109b99e AH |
6865 | enum s_alloc { |
6866 | sa_sched_groups = 0, | |
6867 | sa_rootdomain, | |
6868 | sa_tmpmask, | |
6869 | sa_send_covered, | |
01a08546 | 6870 | sa_this_book_map, |
2109b99e AH |
6871 | sa_this_core_map, |
6872 | sa_this_sibling_map, | |
6873 | sa_nodemask, | |
6874 | sa_sched_group_nodes, | |
6875 | #ifdef CONFIG_NUMA | |
6876 | sa_notcovered, | |
6877 | sa_covered, | |
6878 | sa_domainspan, | |
6879 | #endif | |
6880 | sa_none, | |
6881 | }; | |
6882 | ||
9c1cfda2 | 6883 | /* |
48f24c4d | 6884 | * SMT sched-domains: |
9c1cfda2 | 6885 | */ |
1da177e4 | 6886 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6887 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6888 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6889 | |
41a2d6cf | 6890 | static int |
96f874e2 RR |
6891 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6892 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6893 | { |
6711cab4 | 6894 | if (sg) |
1871e52c | 6895 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6896 | return cpu; |
6897 | } | |
6d6bc0ad | 6898 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6899 | |
48f24c4d IM |
6900 | /* |
6901 | * multi-core sched-domains: | |
6902 | */ | |
1e9f28fa | 6903 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6904 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6905 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6906 | |
41a2d6cf | 6907 | static int |
96f874e2 RR |
6908 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6909 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6910 | { |
6711cab4 | 6911 | int group; |
f269893c | 6912 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6913 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6914 | group = cpumask_first(mask); |
f269893c HC |
6915 | #else |
6916 | group = cpu; | |
6917 | #endif | |
6711cab4 | 6918 | if (sg) |
6c99e9ad | 6919 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6920 | return group; |
1e9f28fa | 6921 | } |
f269893c | 6922 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6923 | |
01a08546 HC |
6924 | /* |
6925 | * book sched-domains: | |
6926 | */ | |
6927 | #ifdef CONFIG_SCHED_BOOK | |
6928 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6929 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6930 | ||
41a2d6cf | 6931 | static int |
01a08546 HC |
6932 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6933 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6934 | { |
01a08546 HC |
6935 | int group = cpu; |
6936 | #ifdef CONFIG_SCHED_MC | |
6937 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6938 | group = cpumask_first(mask); | |
6939 | #elif defined(CONFIG_SCHED_SMT) | |
6940 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6941 | group = cpumask_first(mask); | |
6942 | #endif | |
6711cab4 | 6943 | if (sg) |
01a08546 HC |
6944 | *sg = &per_cpu(sched_group_book, group).sg; |
6945 | return group; | |
1e9f28fa | 6946 | } |
01a08546 | 6947 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6948 | |
6c99e9ad RR |
6949 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6950 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6951 | |
41a2d6cf | 6952 | static int |
96f874e2 RR |
6953 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6954 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6955 | { |
6711cab4 | 6956 | int group; |
01a08546 HC |
6957 | #ifdef CONFIG_SCHED_BOOK |
6958 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6959 | group = cpumask_first(mask); | |
6960 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6961 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6962 | group = cpumask_first(mask); |
1e9f28fa | 6963 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6964 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6965 | group = cpumask_first(mask); |
1da177e4 | 6966 | #else |
6711cab4 | 6967 | group = cpu; |
1da177e4 | 6968 | #endif |
6711cab4 | 6969 | if (sg) |
6c99e9ad | 6970 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6971 | return group; |
1da177e4 LT |
6972 | } |
6973 | ||
6974 | #ifdef CONFIG_NUMA | |
1da177e4 | 6975 | /* |
9c1cfda2 JH |
6976 | * The init_sched_build_groups can't handle what we want to do with node |
6977 | * groups, so roll our own. Now each node has its own list of groups which | |
6978 | * gets dynamically allocated. | |
1da177e4 | 6979 | */ |
62ea9ceb | 6980 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6981 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6982 | |
62ea9ceb | 6983 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6984 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6985 | |
96f874e2 RR |
6986 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6987 | struct sched_group **sg, | |
6988 | struct cpumask *nodemask) | |
9c1cfda2 | 6989 | { |
6711cab4 SS |
6990 | int group; |
6991 | ||
6ca09dfc | 6992 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6993 | group = cpumask_first(nodemask); |
6711cab4 SS |
6994 | |
6995 | if (sg) | |
6c99e9ad | 6996 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6997 | return group; |
1da177e4 | 6998 | } |
6711cab4 | 6999 | |
08069033 SS |
7000 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7001 | { | |
7002 | struct sched_group *sg = group_head; | |
7003 | int j; | |
7004 | ||
7005 | if (!sg) | |
7006 | return; | |
3a5c359a | 7007 | do { |
758b2cdc | 7008 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7009 | struct sched_domain *sd; |
08069033 | 7010 | |
6c99e9ad | 7011 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 7012 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
7013 | /* |
7014 | * Only add "power" once for each | |
7015 | * physical package. | |
7016 | */ | |
7017 | continue; | |
7018 | } | |
08069033 | 7019 | |
18a3885f | 7020 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
7021 | } |
7022 | sg = sg->next; | |
7023 | } while (sg != group_head); | |
08069033 | 7024 | } |
0601a88d AH |
7025 | |
7026 | static int build_numa_sched_groups(struct s_data *d, | |
7027 | const struct cpumask *cpu_map, int num) | |
7028 | { | |
7029 | struct sched_domain *sd; | |
7030 | struct sched_group *sg, *prev; | |
7031 | int n, j; | |
7032 | ||
7033 | cpumask_clear(d->covered); | |
7034 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
7035 | if (cpumask_empty(d->nodemask)) { | |
7036 | d->sched_group_nodes[num] = NULL; | |
7037 | goto out; | |
7038 | } | |
7039 | ||
7040 | sched_domain_node_span(num, d->domainspan); | |
7041 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
7042 | ||
7043 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7044 | GFP_KERNEL, num); | |
7045 | if (!sg) { | |
3df0fc5b PZ |
7046 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
7047 | num); | |
0601a88d AH |
7048 | return -ENOMEM; |
7049 | } | |
7050 | d->sched_group_nodes[num] = sg; | |
7051 | ||
7052 | for_each_cpu(j, d->nodemask) { | |
7053 | sd = &per_cpu(node_domains, j).sd; | |
7054 | sd->groups = sg; | |
7055 | } | |
7056 | ||
18a3885f | 7057 | sg->cpu_power = 0; |
0601a88d AH |
7058 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7059 | sg->next = sg; | |
7060 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7061 | ||
7062 | prev = sg; | |
7063 | for (j = 0; j < nr_node_ids; j++) { | |
7064 | n = (num + j) % nr_node_ids; | |
7065 | cpumask_complement(d->notcovered, d->covered); | |
7066 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7067 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7068 | if (cpumask_empty(d->tmpmask)) | |
7069 | break; | |
7070 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7071 | if (cpumask_empty(d->tmpmask)) | |
7072 | continue; | |
7073 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7074 | GFP_KERNEL, num); | |
7075 | if (!sg) { | |
3df0fc5b PZ |
7076 | printk(KERN_WARNING |
7077 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7078 | return -ENOMEM; |
7079 | } | |
18a3885f | 7080 | sg->cpu_power = 0; |
0601a88d AH |
7081 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7082 | sg->next = prev->next; | |
7083 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7084 | prev->next = sg; | |
7085 | prev = sg; | |
7086 | } | |
7087 | out: | |
7088 | return 0; | |
7089 | } | |
6d6bc0ad | 7090 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7091 | |
a616058b | 7092 | #ifdef CONFIG_NUMA |
51888ca2 | 7093 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7094 | static void free_sched_groups(const struct cpumask *cpu_map, |
7095 | struct cpumask *nodemask) | |
51888ca2 | 7096 | { |
a616058b | 7097 | int cpu, i; |
51888ca2 | 7098 | |
abcd083a | 7099 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7100 | struct sched_group **sched_group_nodes |
7101 | = sched_group_nodes_bycpu[cpu]; | |
7102 | ||
51888ca2 SV |
7103 | if (!sched_group_nodes) |
7104 | continue; | |
7105 | ||
076ac2af | 7106 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7107 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7108 | ||
6ca09dfc | 7109 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7110 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7111 | continue; |
7112 | ||
7113 | if (sg == NULL) | |
7114 | continue; | |
7115 | sg = sg->next; | |
7116 | next_sg: | |
7117 | oldsg = sg; | |
7118 | sg = sg->next; | |
7119 | kfree(oldsg); | |
7120 | if (oldsg != sched_group_nodes[i]) | |
7121 | goto next_sg; | |
7122 | } | |
7123 | kfree(sched_group_nodes); | |
7124 | sched_group_nodes_bycpu[cpu] = NULL; | |
7125 | } | |
51888ca2 | 7126 | } |
6d6bc0ad | 7127 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7128 | static void free_sched_groups(const struct cpumask *cpu_map, |
7129 | struct cpumask *nodemask) | |
a616058b SS |
7130 | { |
7131 | } | |
6d6bc0ad | 7132 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7133 | |
89c4710e SS |
7134 | /* |
7135 | * Initialize sched groups cpu_power. | |
7136 | * | |
7137 | * cpu_power indicates the capacity of sched group, which is used while | |
7138 | * distributing the load between different sched groups in a sched domain. | |
7139 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7140 | * there are asymmetries in the topology. If there are asymmetries, group | |
7141 | * having more cpu_power will pickup more load compared to the group having | |
7142 | * less cpu_power. | |
89c4710e SS |
7143 | */ |
7144 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7145 | { | |
7146 | struct sched_domain *child; | |
7147 | struct sched_group *group; | |
f93e65c1 PZ |
7148 | long power; |
7149 | int weight; | |
89c4710e SS |
7150 | |
7151 | WARN_ON(!sd || !sd->groups); | |
7152 | ||
13318a71 | 7153 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7154 | return; |
7155 | ||
aae6d3dd SS |
7156 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7157 | ||
89c4710e SS |
7158 | child = sd->child; |
7159 | ||
18a3885f | 7160 | sd->groups->cpu_power = 0; |
5517d86b | 7161 | |
f93e65c1 PZ |
7162 | if (!child) { |
7163 | power = SCHED_LOAD_SCALE; | |
7164 | weight = cpumask_weight(sched_domain_span(sd)); | |
7165 | /* | |
7166 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7167 | * Usually multiple threads get a better yield out of |
7168 | * that one core than a single thread would have, | |
7169 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7170 | */ |
a52bfd73 PZ |
7171 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7172 | power *= sd->smt_gain; | |
f93e65c1 | 7173 | power /= weight; |
a52bfd73 PZ |
7174 | power >>= SCHED_LOAD_SHIFT; |
7175 | } | |
18a3885f | 7176 | sd->groups->cpu_power += power; |
89c4710e SS |
7177 | return; |
7178 | } | |
7179 | ||
89c4710e | 7180 | /* |
f93e65c1 | 7181 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7182 | */ |
7183 | group = child->groups; | |
7184 | do { | |
18a3885f | 7185 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7186 | group = group->next; |
7187 | } while (group != child->groups); | |
7188 | } | |
7189 | ||
7c16ec58 MT |
7190 | /* |
7191 | * Initializers for schedule domains | |
7192 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7193 | */ | |
7194 | ||
a5d8c348 IM |
7195 | #ifdef CONFIG_SCHED_DEBUG |
7196 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7197 | #else | |
7198 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7199 | #endif | |
7200 | ||
7c16ec58 | 7201 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7202 | |
7c16ec58 MT |
7203 | #define SD_INIT_FUNC(type) \ |
7204 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7205 | { \ | |
7206 | memset(sd, 0, sizeof(*sd)); \ | |
7207 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7208 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7209 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7210 | } |
7211 | ||
7212 | SD_INIT_FUNC(CPU) | |
7213 | #ifdef CONFIG_NUMA | |
7214 | SD_INIT_FUNC(ALLNODES) | |
7215 | SD_INIT_FUNC(NODE) | |
7216 | #endif | |
7217 | #ifdef CONFIG_SCHED_SMT | |
7218 | SD_INIT_FUNC(SIBLING) | |
7219 | #endif | |
7220 | #ifdef CONFIG_SCHED_MC | |
7221 | SD_INIT_FUNC(MC) | |
7222 | #endif | |
01a08546 HC |
7223 | #ifdef CONFIG_SCHED_BOOK |
7224 | SD_INIT_FUNC(BOOK) | |
7225 | #endif | |
7c16ec58 | 7226 | |
1d3504fc HS |
7227 | static int default_relax_domain_level = -1; |
7228 | ||
7229 | static int __init setup_relax_domain_level(char *str) | |
7230 | { | |
30e0e178 LZ |
7231 | unsigned long val; |
7232 | ||
7233 | val = simple_strtoul(str, NULL, 0); | |
7234 | if (val < SD_LV_MAX) | |
7235 | default_relax_domain_level = val; | |
7236 | ||
1d3504fc HS |
7237 | return 1; |
7238 | } | |
7239 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7240 | ||
7241 | static void set_domain_attribute(struct sched_domain *sd, | |
7242 | struct sched_domain_attr *attr) | |
7243 | { | |
7244 | int request; | |
7245 | ||
7246 | if (!attr || attr->relax_domain_level < 0) { | |
7247 | if (default_relax_domain_level < 0) | |
7248 | return; | |
7249 | else | |
7250 | request = default_relax_domain_level; | |
7251 | } else | |
7252 | request = attr->relax_domain_level; | |
7253 | if (request < sd->level) { | |
7254 | /* turn off idle balance on this domain */ | |
c88d5910 | 7255 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7256 | } else { |
7257 | /* turn on idle balance on this domain */ | |
c88d5910 | 7258 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7259 | } |
7260 | } | |
7261 | ||
2109b99e AH |
7262 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7263 | const struct cpumask *cpu_map) | |
7264 | { | |
7265 | switch (what) { | |
7266 | case sa_sched_groups: | |
7267 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7268 | d->sched_group_nodes = NULL; | |
7269 | case sa_rootdomain: | |
7270 | free_rootdomain(d->rd); /* fall through */ | |
7271 | case sa_tmpmask: | |
7272 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7273 | case sa_send_covered: | |
7274 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7275 | case sa_this_book_map: |
7276 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7277 | case sa_this_core_map: |
7278 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7279 | case sa_this_sibling_map: | |
7280 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7281 | case sa_nodemask: | |
7282 | free_cpumask_var(d->nodemask); /* fall through */ | |
7283 | case sa_sched_group_nodes: | |
d1b55138 | 7284 | #ifdef CONFIG_NUMA |
2109b99e AH |
7285 | kfree(d->sched_group_nodes); /* fall through */ |
7286 | case sa_notcovered: | |
7287 | free_cpumask_var(d->notcovered); /* fall through */ | |
7288 | case sa_covered: | |
7289 | free_cpumask_var(d->covered); /* fall through */ | |
7290 | case sa_domainspan: | |
7291 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7292 | #endif |
2109b99e AH |
7293 | case sa_none: |
7294 | break; | |
7295 | } | |
7296 | } | |
3404c8d9 | 7297 | |
2109b99e AH |
7298 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7299 | const struct cpumask *cpu_map) | |
7300 | { | |
3404c8d9 | 7301 | #ifdef CONFIG_NUMA |
2109b99e AH |
7302 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7303 | return sa_none; | |
7304 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7305 | return sa_domainspan; | |
7306 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7307 | return sa_covered; | |
7308 | /* Allocate the per-node list of sched groups */ | |
7309 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7310 | sizeof(struct sched_group *), GFP_KERNEL); | |
7311 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7312 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7313 | return sa_notcovered; |
d1b55138 | 7314 | } |
2109b99e | 7315 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7316 | #endif |
2109b99e AH |
7317 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7318 | return sa_sched_group_nodes; | |
7319 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7320 | return sa_nodemask; | |
7321 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7322 | return sa_this_sibling_map; | |
01a08546 | 7323 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7324 | return sa_this_core_map; |
01a08546 HC |
7325 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7326 | return sa_this_book_map; | |
2109b99e AH |
7327 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7328 | return sa_send_covered; | |
7329 | d->rd = alloc_rootdomain(); | |
7330 | if (!d->rd) { | |
3df0fc5b | 7331 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7332 | return sa_tmpmask; |
57d885fe | 7333 | } |
2109b99e AH |
7334 | return sa_rootdomain; |
7335 | } | |
57d885fe | 7336 | |
7f4588f3 AH |
7337 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7338 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7339 | { | |
7340 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7341 | #ifdef CONFIG_NUMA |
7f4588f3 | 7342 | struct sched_domain *parent; |
1da177e4 | 7343 | |
7f4588f3 AH |
7344 | d->sd_allnodes = 0; |
7345 | if (cpumask_weight(cpu_map) > | |
7346 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7347 | sd = &per_cpu(allnodes_domains, i).sd; | |
7348 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7349 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7350 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7351 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7352 | d->sd_allnodes = 1; | |
7353 | } | |
7354 | parent = sd; | |
7355 | ||
7356 | sd = &per_cpu(node_domains, i).sd; | |
7357 | SD_INIT(sd, NODE); | |
7358 | set_domain_attribute(sd, attr); | |
7359 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7360 | sd->parent = parent; | |
7361 | if (parent) | |
7362 | parent->child = sd; | |
7363 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7364 | #endif |
7f4588f3 AH |
7365 | return sd; |
7366 | } | |
1da177e4 | 7367 | |
87cce662 AH |
7368 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7369 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7370 | struct sched_domain *parent, int i) | |
7371 | { | |
7372 | struct sched_domain *sd; | |
7373 | sd = &per_cpu(phys_domains, i).sd; | |
7374 | SD_INIT(sd, CPU); | |
7375 | set_domain_attribute(sd, attr); | |
7376 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7377 | sd->parent = parent; | |
7378 | if (parent) | |
7379 | parent->child = sd; | |
7380 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7381 | return sd; | |
7382 | } | |
1da177e4 | 7383 | |
01a08546 HC |
7384 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7385 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7386 | struct sched_domain *parent, int i) | |
7387 | { | |
7388 | struct sched_domain *sd = parent; | |
7389 | #ifdef CONFIG_SCHED_BOOK | |
7390 | sd = &per_cpu(book_domains, i).sd; | |
7391 | SD_INIT(sd, BOOK); | |
7392 | set_domain_attribute(sd, attr); | |
7393 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7394 | sd->parent = parent; | |
7395 | parent->child = sd; | |
7396 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7397 | #endif | |
7398 | return sd; | |
7399 | } | |
7400 | ||
410c4081 AH |
7401 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7402 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7403 | struct sched_domain *parent, int i) | |
7404 | { | |
7405 | struct sched_domain *sd = parent; | |
1e9f28fa | 7406 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7407 | sd = &per_cpu(core_domains, i).sd; |
7408 | SD_INIT(sd, MC); | |
7409 | set_domain_attribute(sd, attr); | |
7410 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7411 | sd->parent = parent; | |
7412 | parent->child = sd; | |
7413 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7414 | #endif |
410c4081 AH |
7415 | return sd; |
7416 | } | |
1e9f28fa | 7417 | |
d8173535 AH |
7418 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7419 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7420 | struct sched_domain *parent, int i) | |
7421 | { | |
7422 | struct sched_domain *sd = parent; | |
1da177e4 | 7423 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7424 | sd = &per_cpu(cpu_domains, i).sd; |
7425 | SD_INIT(sd, SIBLING); | |
7426 | set_domain_attribute(sd, attr); | |
7427 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7428 | sd->parent = parent; | |
7429 | parent->child = sd; | |
7430 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7431 | #endif |
d8173535 AH |
7432 | return sd; |
7433 | } | |
1da177e4 | 7434 | |
0e8e85c9 AH |
7435 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7436 | const struct cpumask *cpu_map, int cpu) | |
7437 | { | |
7438 | switch (l) { | |
1da177e4 | 7439 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7440 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7441 | cpumask_and(d->this_sibling_map, cpu_map, | |
7442 | topology_thread_cpumask(cpu)); | |
7443 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7444 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7445 | &cpu_to_cpu_group, | |
7446 | d->send_covered, d->tmpmask); | |
7447 | break; | |
1da177e4 | 7448 | #endif |
1e9f28fa | 7449 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7450 | case SD_LV_MC: /* set up multi-core groups */ |
7451 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7452 | if (cpu == cpumask_first(d->this_core_map)) | |
7453 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7454 | &cpu_to_core_group, | |
7455 | d->send_covered, d->tmpmask); | |
7456 | break; | |
01a08546 HC |
7457 | #endif |
7458 | #ifdef CONFIG_SCHED_BOOK | |
7459 | case SD_LV_BOOK: /* set up book groups */ | |
7460 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7461 | if (cpu == cpumask_first(d->this_book_map)) | |
7462 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7463 | &cpu_to_book_group, | |
7464 | d->send_covered, d->tmpmask); | |
7465 | break; | |
1e9f28fa | 7466 | #endif |
86548096 AH |
7467 | case SD_LV_CPU: /* set up physical groups */ |
7468 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7469 | if (!cpumask_empty(d->nodemask)) | |
7470 | init_sched_build_groups(d->nodemask, cpu_map, | |
7471 | &cpu_to_phys_group, | |
7472 | d->send_covered, d->tmpmask); | |
7473 | break; | |
1da177e4 | 7474 | #ifdef CONFIG_NUMA |
de616e36 AH |
7475 | case SD_LV_ALLNODES: |
7476 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7477 | d->send_covered, d->tmpmask); | |
7478 | break; | |
7479 | #endif | |
0e8e85c9 AH |
7480 | default: |
7481 | break; | |
7c16ec58 | 7482 | } |
0e8e85c9 | 7483 | } |
9c1cfda2 | 7484 | |
2109b99e AH |
7485 | /* |
7486 | * Build sched domains for a given set of cpus and attach the sched domains | |
7487 | * to the individual cpus | |
7488 | */ | |
7489 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7490 | struct sched_domain_attr *attr) | |
7491 | { | |
7492 | enum s_alloc alloc_state = sa_none; | |
7493 | struct s_data d; | |
294b0c96 | 7494 | struct sched_domain *sd; |
2109b99e | 7495 | int i; |
7c16ec58 | 7496 | #ifdef CONFIG_NUMA |
2109b99e | 7497 | d.sd_allnodes = 0; |
7c16ec58 | 7498 | #endif |
9c1cfda2 | 7499 | |
2109b99e AH |
7500 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7501 | if (alloc_state != sa_rootdomain) | |
7502 | goto error; | |
7503 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7504 | |
1da177e4 | 7505 | /* |
1a20ff27 | 7506 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7507 | */ |
abcd083a | 7508 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7509 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7510 | cpu_map); | |
9761eea8 | 7511 | |
7f4588f3 | 7512 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7513 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7514 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7515 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7516 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7517 | } |
9c1cfda2 | 7518 | |
abcd083a | 7519 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7520 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7521 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7522 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7523 | } |
9c1cfda2 | 7524 | |
1da177e4 | 7525 | /* Set up physical groups */ |
86548096 AH |
7526 | for (i = 0; i < nr_node_ids; i++) |
7527 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7528 | |
1da177e4 LT |
7529 | #ifdef CONFIG_NUMA |
7530 | /* Set up node groups */ | |
de616e36 AH |
7531 | if (d.sd_allnodes) |
7532 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7533 | |
0601a88d AH |
7534 | for (i = 0; i < nr_node_ids; i++) |
7535 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7536 | goto error; |
1da177e4 LT |
7537 | #endif |
7538 | ||
7539 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7540 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7541 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7542 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7543 | init_sched_groups_power(i, sd); |
5c45bf27 | 7544 | } |
1da177e4 | 7545 | #endif |
1e9f28fa | 7546 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7547 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7548 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7549 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7550 | } |
7551 | #endif | |
01a08546 HC |
7552 | #ifdef CONFIG_SCHED_BOOK |
7553 | for_each_cpu(i, cpu_map) { | |
7554 | sd = &per_cpu(book_domains, i).sd; | |
7555 | init_sched_groups_power(i, sd); | |
7556 | } | |
7557 | #endif | |
1e9f28fa | 7558 | |
abcd083a | 7559 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7560 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7561 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7562 | } |
7563 | ||
9c1cfda2 | 7564 | #ifdef CONFIG_NUMA |
076ac2af | 7565 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7566 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7567 | |
49a02c51 | 7568 | if (d.sd_allnodes) { |
6711cab4 | 7569 | struct sched_group *sg; |
f712c0c7 | 7570 | |
96f874e2 | 7571 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7572 | d.tmpmask); |
f712c0c7 SS |
7573 | init_numa_sched_groups_power(sg); |
7574 | } | |
9c1cfda2 JH |
7575 | #endif |
7576 | ||
1da177e4 | 7577 | /* Attach the domains */ |
abcd083a | 7578 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7579 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7580 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7581 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7582 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7583 | #elif defined(CONFIG_SCHED_BOOK) |
7584 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7585 | #else |
6c99e9ad | 7586 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7587 | #endif |
49a02c51 | 7588 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7589 | } |
51888ca2 | 7590 | |
2109b99e AH |
7591 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7592 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7593 | return 0; | |
51888ca2 | 7594 | |
51888ca2 | 7595 | error: |
2109b99e AH |
7596 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7597 | return -ENOMEM; | |
1da177e4 | 7598 | } |
029190c5 | 7599 | |
96f874e2 | 7600 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7601 | { |
7602 | return __build_sched_domains(cpu_map, NULL); | |
7603 | } | |
7604 | ||
acc3f5d7 | 7605 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7606 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7607 | static struct sched_domain_attr *dattr_cur; |
7608 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7609 | |
7610 | /* | |
7611 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7612 | * cpumask) fails, then fallback to a single sched domain, |
7613 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7614 | */ |
4212823f | 7615 | static cpumask_var_t fallback_doms; |
029190c5 | 7616 | |
ee79d1bd HC |
7617 | /* |
7618 | * arch_update_cpu_topology lets virtualized architectures update the | |
7619 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7620 | * or 0 if it stayed the same. | |
7621 | */ | |
7622 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7623 | { |
ee79d1bd | 7624 | return 0; |
22e52b07 HC |
7625 | } |
7626 | ||
acc3f5d7 RR |
7627 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7628 | { | |
7629 | int i; | |
7630 | cpumask_var_t *doms; | |
7631 | ||
7632 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7633 | if (!doms) | |
7634 | return NULL; | |
7635 | for (i = 0; i < ndoms; i++) { | |
7636 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7637 | free_sched_domains(doms, i); | |
7638 | return NULL; | |
7639 | } | |
7640 | } | |
7641 | return doms; | |
7642 | } | |
7643 | ||
7644 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7645 | { | |
7646 | unsigned int i; | |
7647 | for (i = 0; i < ndoms; i++) | |
7648 | free_cpumask_var(doms[i]); | |
7649 | kfree(doms); | |
7650 | } | |
7651 | ||
1a20ff27 | 7652 | /* |
41a2d6cf | 7653 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7654 | * For now this just excludes isolated cpus, but could be used to |
7655 | * exclude other special cases in the future. | |
1a20ff27 | 7656 | */ |
96f874e2 | 7657 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7658 | { |
7378547f MM |
7659 | int err; |
7660 | ||
22e52b07 | 7661 | arch_update_cpu_topology(); |
029190c5 | 7662 | ndoms_cur = 1; |
acc3f5d7 | 7663 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7664 | if (!doms_cur) |
acc3f5d7 RR |
7665 | doms_cur = &fallback_doms; |
7666 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7667 | dattr_cur = NULL; |
acc3f5d7 | 7668 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7669 | register_sched_domain_sysctl(); |
7378547f MM |
7670 | |
7671 | return err; | |
1a20ff27 DG |
7672 | } |
7673 | ||
96f874e2 RR |
7674 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7675 | struct cpumask *tmpmask) | |
1da177e4 | 7676 | { |
7c16ec58 | 7677 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7678 | } |
1da177e4 | 7679 | |
1a20ff27 DG |
7680 | /* |
7681 | * Detach sched domains from a group of cpus specified in cpu_map | |
7682 | * These cpus will now be attached to the NULL domain | |
7683 | */ | |
96f874e2 | 7684 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7685 | { |
96f874e2 RR |
7686 | /* Save because hotplug lock held. */ |
7687 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7688 | int i; |
7689 | ||
abcd083a | 7690 | for_each_cpu(i, cpu_map) |
57d885fe | 7691 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7692 | synchronize_sched(); |
96f874e2 | 7693 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7694 | } |
7695 | ||
1d3504fc HS |
7696 | /* handle null as "default" */ |
7697 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7698 | struct sched_domain_attr *new, int idx_new) | |
7699 | { | |
7700 | struct sched_domain_attr tmp; | |
7701 | ||
7702 | /* fast path */ | |
7703 | if (!new && !cur) | |
7704 | return 1; | |
7705 | ||
7706 | tmp = SD_ATTR_INIT; | |
7707 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7708 | new ? (new + idx_new) : &tmp, | |
7709 | sizeof(struct sched_domain_attr)); | |
7710 | } | |
7711 | ||
029190c5 PJ |
7712 | /* |
7713 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7714 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7715 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7716 | * It destroys each deleted domain and builds each new domain. | |
7717 | * | |
acc3f5d7 | 7718 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7719 | * The masks don't intersect (don't overlap.) We should setup one |
7720 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7721 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7722 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7723 | * it as it is. | |
7724 | * | |
acc3f5d7 RR |
7725 | * The passed in 'doms_new' should be allocated using |
7726 | * alloc_sched_domains. This routine takes ownership of it and will | |
7727 | * free_sched_domains it when done with it. If the caller failed the | |
7728 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7729 | * and partition_sched_domains() will fallback to the single partition | |
7730 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7731 | * |
96f874e2 | 7732 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7733 | * ndoms_new == 0 is a special case for destroying existing domains, |
7734 | * and it will not create the default domain. | |
dfb512ec | 7735 | * |
029190c5 PJ |
7736 | * Call with hotplug lock held |
7737 | */ | |
acc3f5d7 | 7738 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7739 | struct sched_domain_attr *dattr_new) |
029190c5 | 7740 | { |
dfb512ec | 7741 | int i, j, n; |
d65bd5ec | 7742 | int new_topology; |
029190c5 | 7743 | |
712555ee | 7744 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7745 | |
7378547f MM |
7746 | /* always unregister in case we don't destroy any domains */ |
7747 | unregister_sched_domain_sysctl(); | |
7748 | ||
d65bd5ec HC |
7749 | /* Let architecture update cpu core mappings. */ |
7750 | new_topology = arch_update_cpu_topology(); | |
7751 | ||
dfb512ec | 7752 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7753 | |
7754 | /* Destroy deleted domains */ | |
7755 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7756 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7757 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7758 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7759 | goto match1; |
7760 | } | |
7761 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7762 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7763 | match1: |
7764 | ; | |
7765 | } | |
7766 | ||
e761b772 MK |
7767 | if (doms_new == NULL) { |
7768 | ndoms_cur = 0; | |
acc3f5d7 | 7769 | doms_new = &fallback_doms; |
6ad4c188 | 7770 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7771 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7772 | } |
7773 | ||
029190c5 PJ |
7774 | /* Build new domains */ |
7775 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7776 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7777 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7778 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7779 | goto match2; |
7780 | } | |
7781 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7782 | __build_sched_domains(doms_new[i], |
1d3504fc | 7783 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7784 | match2: |
7785 | ; | |
7786 | } | |
7787 | ||
7788 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7789 | if (doms_cur != &fallback_doms) |
7790 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7791 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7792 | doms_cur = doms_new; |
1d3504fc | 7793 | dattr_cur = dattr_new; |
029190c5 | 7794 | ndoms_cur = ndoms_new; |
7378547f MM |
7795 | |
7796 | register_sched_domain_sysctl(); | |
a1835615 | 7797 | |
712555ee | 7798 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7799 | } |
7800 | ||
5c45bf27 | 7801 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7802 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7803 | { |
95402b38 | 7804 | get_online_cpus(); |
dfb512ec MK |
7805 | |
7806 | /* Destroy domains first to force the rebuild */ | |
7807 | partition_sched_domains(0, NULL, NULL); | |
7808 | ||
e761b772 | 7809 | rebuild_sched_domains(); |
95402b38 | 7810 | put_online_cpus(); |
5c45bf27 SS |
7811 | } |
7812 | ||
7813 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7814 | { | |
afb8a9b7 | 7815 | unsigned int level = 0; |
5c45bf27 | 7816 | |
afb8a9b7 GS |
7817 | if (sscanf(buf, "%u", &level) != 1) |
7818 | return -EINVAL; | |
7819 | ||
7820 | /* | |
7821 | * level is always be positive so don't check for | |
7822 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7823 | * What happens on 0 or 1 byte write, | |
7824 | * need to check for count as well? | |
7825 | */ | |
7826 | ||
7827 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7828 | return -EINVAL; |
7829 | ||
7830 | if (smt) | |
afb8a9b7 | 7831 | sched_smt_power_savings = level; |
5c45bf27 | 7832 | else |
afb8a9b7 | 7833 | sched_mc_power_savings = level; |
5c45bf27 | 7834 | |
c70f22d2 | 7835 | arch_reinit_sched_domains(); |
5c45bf27 | 7836 | |
c70f22d2 | 7837 | return count; |
5c45bf27 SS |
7838 | } |
7839 | ||
5c45bf27 | 7840 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7841 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7842 | struct sysdev_class_attribute *attr, |
f718cd4a | 7843 | char *page) |
5c45bf27 SS |
7844 | { |
7845 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7846 | } | |
f718cd4a | 7847 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7848 | struct sysdev_class_attribute *attr, |
48f24c4d | 7849 | const char *buf, size_t count) |
5c45bf27 SS |
7850 | { |
7851 | return sched_power_savings_store(buf, count, 0); | |
7852 | } | |
f718cd4a AK |
7853 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7854 | sched_mc_power_savings_show, | |
7855 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7856 | #endif |
7857 | ||
7858 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7859 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7860 | struct sysdev_class_attribute *attr, |
f718cd4a | 7861 | char *page) |
5c45bf27 SS |
7862 | { |
7863 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7864 | } | |
f718cd4a | 7865 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7866 | struct sysdev_class_attribute *attr, |
48f24c4d | 7867 | const char *buf, size_t count) |
5c45bf27 SS |
7868 | { |
7869 | return sched_power_savings_store(buf, count, 1); | |
7870 | } | |
f718cd4a AK |
7871 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7872 | sched_smt_power_savings_show, | |
6707de00 AB |
7873 | sched_smt_power_savings_store); |
7874 | #endif | |
7875 | ||
39aac648 | 7876 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7877 | { |
7878 | int err = 0; | |
7879 | ||
7880 | #ifdef CONFIG_SCHED_SMT | |
7881 | if (smt_capable()) | |
7882 | err = sysfs_create_file(&cls->kset.kobj, | |
7883 | &attr_sched_smt_power_savings.attr); | |
7884 | #endif | |
7885 | #ifdef CONFIG_SCHED_MC | |
7886 | if (!err && mc_capable()) | |
7887 | err = sysfs_create_file(&cls->kset.kobj, | |
7888 | &attr_sched_mc_power_savings.attr); | |
7889 | #endif | |
7890 | return err; | |
7891 | } | |
6d6bc0ad | 7892 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7893 | |
1da177e4 | 7894 | /* |
3a101d05 TH |
7895 | * Update cpusets according to cpu_active mask. If cpusets are |
7896 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7897 | * around partition_sched_domains(). | |
1da177e4 | 7898 | */ |
0b2e918a TH |
7899 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7900 | void *hcpu) | |
e761b772 | 7901 | { |
3a101d05 | 7902 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7903 | case CPU_ONLINE: |
6ad4c188 | 7904 | case CPU_DOWN_FAILED: |
3a101d05 | 7905 | cpuset_update_active_cpus(); |
e761b772 | 7906 | return NOTIFY_OK; |
3a101d05 TH |
7907 | default: |
7908 | return NOTIFY_DONE; | |
7909 | } | |
7910 | } | |
e761b772 | 7911 | |
0b2e918a TH |
7912 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7913 | void *hcpu) | |
3a101d05 TH |
7914 | { |
7915 | switch (action & ~CPU_TASKS_FROZEN) { | |
7916 | case CPU_DOWN_PREPARE: | |
7917 | cpuset_update_active_cpus(); | |
7918 | return NOTIFY_OK; | |
e761b772 MK |
7919 | default: |
7920 | return NOTIFY_DONE; | |
7921 | } | |
7922 | } | |
e761b772 MK |
7923 | |
7924 | static int update_runtime(struct notifier_block *nfb, | |
7925 | unsigned long action, void *hcpu) | |
1da177e4 | 7926 | { |
7def2be1 PZ |
7927 | int cpu = (int)(long)hcpu; |
7928 | ||
1da177e4 | 7929 | switch (action) { |
1da177e4 | 7930 | case CPU_DOWN_PREPARE: |
8bb78442 | 7931 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7932 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7933 | return NOTIFY_OK; |
7934 | ||
1da177e4 | 7935 | case CPU_DOWN_FAILED: |
8bb78442 | 7936 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7937 | case CPU_ONLINE: |
8bb78442 | 7938 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7939 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7940 | return NOTIFY_OK; |
7941 | ||
1da177e4 LT |
7942 | default: |
7943 | return NOTIFY_DONE; | |
7944 | } | |
1da177e4 | 7945 | } |
1da177e4 LT |
7946 | |
7947 | void __init sched_init_smp(void) | |
7948 | { | |
dcc30a35 RR |
7949 | cpumask_var_t non_isolated_cpus; |
7950 | ||
7951 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7952 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7953 | |
434d53b0 MT |
7954 | #if defined(CONFIG_NUMA) |
7955 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7956 | GFP_KERNEL); | |
7957 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7958 | #endif | |
95402b38 | 7959 | get_online_cpus(); |
712555ee | 7960 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7961 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7962 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7963 | if (cpumask_empty(non_isolated_cpus)) | |
7964 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7965 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7966 | put_online_cpus(); |
e761b772 | 7967 | |
3a101d05 TH |
7968 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7969 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7970 | |
7971 | /* RT runtime code needs to handle some hotplug events */ | |
7972 | hotcpu_notifier(update_runtime, 0); | |
7973 | ||
b328ca18 | 7974 | init_hrtick(); |
5c1e1767 NP |
7975 | |
7976 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7977 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7978 | BUG(); |
19978ca6 | 7979 | sched_init_granularity(); |
dcc30a35 | 7980 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7981 | |
0e3900e6 | 7982 | init_sched_rt_class(); |
1da177e4 LT |
7983 | } |
7984 | #else | |
7985 | void __init sched_init_smp(void) | |
7986 | { | |
19978ca6 | 7987 | sched_init_granularity(); |
1da177e4 LT |
7988 | } |
7989 | #endif /* CONFIG_SMP */ | |
7990 | ||
cd1bb94b AB |
7991 | const_debug unsigned int sysctl_timer_migration = 1; |
7992 | ||
1da177e4 LT |
7993 | int in_sched_functions(unsigned long addr) |
7994 | { | |
1da177e4 LT |
7995 | return in_lock_functions(addr) || |
7996 | (addr >= (unsigned long)__sched_text_start | |
7997 | && addr < (unsigned long)__sched_text_end); | |
7998 | } | |
7999 | ||
a9957449 | 8000 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8001 | { |
8002 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8003 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8004 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8005 | cfs_rq->rq = rq; | |
f07333bf | 8006 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 8007 | #ifdef CONFIG_SMP |
f07333bf | 8008 | cfs_rq->load_stamp = 1; |
6ea72f12 | 8009 | #endif |
dd41f596 | 8010 | #endif |
67e9fb2a | 8011 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8012 | } |
8013 | ||
fa85ae24 PZ |
8014 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8015 | { | |
8016 | struct rt_prio_array *array; | |
8017 | int i; | |
8018 | ||
8019 | array = &rt_rq->active; | |
8020 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8021 | INIT_LIST_HEAD(array->queue + i); | |
8022 | __clear_bit(i, array->bitmap); | |
8023 | } | |
8024 | /* delimiter for bitsearch: */ | |
8025 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8026 | ||
052f1dc7 | 8027 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8028 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8029 | #ifdef CONFIG_SMP |
e864c499 | 8030 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8031 | #endif |
48d5e258 | 8032 | #endif |
fa85ae24 PZ |
8033 | #ifdef CONFIG_SMP |
8034 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8035 | rt_rq->overloaded = 0; |
05fa785c | 8036 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8037 | #endif |
8038 | ||
8039 | rt_rq->rt_time = 0; | |
8040 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8041 | rt_rq->rt_runtime = 0; |
0986b11b | 8042 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 8043 | |
052f1dc7 | 8044 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8045 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8046 | rt_rq->rq = rq; |
8047 | #endif | |
fa85ae24 PZ |
8048 | } |
8049 | ||
6f505b16 | 8050 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 8051 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 8052 | struct sched_entity *se, int cpu, |
ec7dc8ac | 8053 | struct sched_entity *parent) |
6f505b16 | 8054 | { |
ec7dc8ac | 8055 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8056 | tg->cfs_rq[cpu] = cfs_rq; |
8057 | init_cfs_rq(cfs_rq, rq); | |
8058 | cfs_rq->tg = tg; | |
6f505b16 PZ |
8059 | |
8060 | tg->se[cpu] = se; | |
07e06b01 | 8061 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
8062 | if (!se) |
8063 | return; | |
8064 | ||
ec7dc8ac DG |
8065 | if (!parent) |
8066 | se->cfs_rq = &rq->cfs; | |
8067 | else | |
8068 | se->cfs_rq = parent->my_q; | |
8069 | ||
6f505b16 | 8070 | se->my_q = cfs_rq; |
9437178f | 8071 | update_load_set(&se->load, 0); |
ec7dc8ac | 8072 | se->parent = parent; |
6f505b16 | 8073 | } |
052f1dc7 | 8074 | #endif |
6f505b16 | 8075 | |
052f1dc7 | 8076 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 8077 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 8078 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 8079 | struct sched_rt_entity *parent) |
6f505b16 | 8080 | { |
ec7dc8ac DG |
8081 | struct rq *rq = cpu_rq(cpu); |
8082 | ||
6f505b16 PZ |
8083 | tg->rt_rq[cpu] = rt_rq; |
8084 | init_rt_rq(rt_rq, rq); | |
8085 | rt_rq->tg = tg; | |
ac086bc2 | 8086 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8087 | |
8088 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8089 | if (!rt_se) |
8090 | return; | |
8091 | ||
ec7dc8ac DG |
8092 | if (!parent) |
8093 | rt_se->rt_rq = &rq->rt; | |
8094 | else | |
8095 | rt_se->rt_rq = parent->my_q; | |
8096 | ||
6f505b16 | 8097 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8098 | rt_se->parent = parent; |
6f505b16 PZ |
8099 | INIT_LIST_HEAD(&rt_se->run_list); |
8100 | } | |
8101 | #endif | |
8102 | ||
1da177e4 LT |
8103 | void __init sched_init(void) |
8104 | { | |
dd41f596 | 8105 | int i, j; |
434d53b0 MT |
8106 | unsigned long alloc_size = 0, ptr; |
8107 | ||
8108 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8109 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8110 | #endif | |
8111 | #ifdef CONFIG_RT_GROUP_SCHED | |
8112 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8113 | #endif |
df7c8e84 | 8114 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8115 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8116 | #endif |
434d53b0 | 8117 | if (alloc_size) { |
36b7b6d4 | 8118 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8119 | |
8120 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8121 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8122 | ptr += nr_cpu_ids * sizeof(void **); |
8123 | ||
07e06b01 | 8124 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8125 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8126 | |
6d6bc0ad | 8127 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8128 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8129 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8130 | ptr += nr_cpu_ids * sizeof(void **); |
8131 | ||
07e06b01 | 8132 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8133 | ptr += nr_cpu_ids * sizeof(void **); |
8134 | ||
6d6bc0ad | 8135 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8136 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8137 | for_each_possible_cpu(i) { | |
8138 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8139 | ptr += cpumask_size(); | |
8140 | } | |
8141 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8142 | } |
dd41f596 | 8143 | |
57d885fe GH |
8144 | #ifdef CONFIG_SMP |
8145 | init_defrootdomain(); | |
8146 | #endif | |
8147 | ||
d0b27fa7 PZ |
8148 | init_rt_bandwidth(&def_rt_bandwidth, |
8149 | global_rt_period(), global_rt_runtime()); | |
8150 | ||
8151 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 8152 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8153 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8154 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8155 | |
7c941438 | 8156 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
8157 | list_add(&root_task_group.list, &task_groups); |
8158 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 8159 | autogroup_init(&init_task); |
7c941438 | 8160 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8161 | |
0a945022 | 8162 | for_each_possible_cpu(i) { |
70b97a7f | 8163 | struct rq *rq; |
1da177e4 LT |
8164 | |
8165 | rq = cpu_rq(i); | |
05fa785c | 8166 | raw_spin_lock_init(&rq->lock); |
7897986b | 8167 | rq->nr_running = 0; |
dce48a84 TG |
8168 | rq->calc_load_active = 0; |
8169 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8170 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8171 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8172 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 8173 | root_task_group.shares = root_task_group_load; |
6f505b16 | 8174 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 8175 | /* |
07e06b01 | 8176 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
8177 | * |
8178 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8179 | * gets 100% of the cpu resources in the system. This overall | |
8180 | * system cpu resource is divided among the tasks of | |
07e06b01 | 8181 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8182 | * based on each entity's (task or task-group's) weight |
8183 | * (se->load.weight). | |
8184 | * | |
07e06b01 | 8185 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
8186 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
8187 | * then A0's share of the cpu resource is: | |
8188 | * | |
0d905bca | 8189 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8190 | * |
07e06b01 YZ |
8191 | * We achieve this by letting root_task_group's tasks sit |
8192 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8193 | */ |
07e06b01 | 8194 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8195 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8196 | ||
8197 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8198 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8199 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 8200 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8201 | #endif |
1da177e4 | 8202 | |
dd41f596 IM |
8203 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8204 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8205 | |
8206 | rq->last_load_update_tick = jiffies; | |
8207 | ||
1da177e4 | 8208 | #ifdef CONFIG_SMP |
41c7ce9a | 8209 | rq->sd = NULL; |
57d885fe | 8210 | rq->rd = NULL; |
e51fd5e2 | 8211 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8212 | rq->post_schedule = 0; |
1da177e4 | 8213 | rq->active_balance = 0; |
dd41f596 | 8214 | rq->next_balance = jiffies; |
1da177e4 | 8215 | rq->push_cpu = 0; |
0a2966b4 | 8216 | rq->cpu = i; |
1f11eb6a | 8217 | rq->online = 0; |
eae0c9df MG |
8218 | rq->idle_stamp = 0; |
8219 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8220 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8221 | #ifdef CONFIG_NO_HZ |
8222 | rq->nohz_balance_kick = 0; | |
8223 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8224 | #endif | |
1da177e4 | 8225 | #endif |
8f4d37ec | 8226 | init_rq_hrtick(rq); |
1da177e4 | 8227 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8228 | } |
8229 | ||
2dd73a4f | 8230 | set_load_weight(&init_task); |
b50f60ce | 8231 | |
e107be36 AK |
8232 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8233 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8234 | #endif | |
8235 | ||
c9819f45 | 8236 | #ifdef CONFIG_SMP |
962cf36c | 8237 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8238 | #endif |
8239 | ||
b50f60ce | 8240 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8241 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8242 | #endif |
8243 | ||
1da177e4 LT |
8244 | /* |
8245 | * The boot idle thread does lazy MMU switching as well: | |
8246 | */ | |
8247 | atomic_inc(&init_mm.mm_count); | |
8248 | enter_lazy_tlb(&init_mm, current); | |
8249 | ||
8250 | /* | |
8251 | * Make us the idle thread. Technically, schedule() should not be | |
8252 | * called from this thread, however somewhere below it might be, | |
8253 | * but because we are the idle thread, we just pick up running again | |
8254 | * when this runqueue becomes "idle". | |
8255 | */ | |
8256 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8257 | |
8258 | calc_load_update = jiffies + LOAD_FREQ; | |
8259 | ||
dd41f596 IM |
8260 | /* |
8261 | * During early bootup we pretend to be a normal task: | |
8262 | */ | |
8263 | current->sched_class = &fair_sched_class; | |
6892b75e | 8264 | |
6a7b3dc3 | 8265 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8266 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8267 | #ifdef CONFIG_SMP |
7d1e6a9b | 8268 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8269 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8270 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8271 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8272 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8273 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8274 | #endif |
bdddd296 RR |
8275 | /* May be allocated at isolcpus cmdline parse time */ |
8276 | if (cpu_isolated_map == NULL) | |
8277 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8278 | #endif /* SMP */ |
6a7b3dc3 | 8279 | |
6892b75e | 8280 | scheduler_running = 1; |
1da177e4 LT |
8281 | } |
8282 | ||
8283 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8284 | static inline int preempt_count_equals(int preempt_offset) |
8285 | { | |
234da7bc | 8286 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8287 | |
4ba8216c | 8288 | return (nested == preempt_offset); |
e4aafea2 FW |
8289 | } |
8290 | ||
d894837f | 8291 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8292 | { |
48f24c4d | 8293 | #ifdef in_atomic |
1da177e4 LT |
8294 | static unsigned long prev_jiffy; /* ratelimiting */ |
8295 | ||
e4aafea2 FW |
8296 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8297 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8298 | return; |
8299 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8300 | return; | |
8301 | prev_jiffy = jiffies; | |
8302 | ||
3df0fc5b PZ |
8303 | printk(KERN_ERR |
8304 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8305 | file, line); | |
8306 | printk(KERN_ERR | |
8307 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8308 | in_atomic(), irqs_disabled(), | |
8309 | current->pid, current->comm); | |
aef745fc IM |
8310 | |
8311 | debug_show_held_locks(current); | |
8312 | if (irqs_disabled()) | |
8313 | print_irqtrace_events(current); | |
8314 | dump_stack(); | |
1da177e4 LT |
8315 | #endif |
8316 | } | |
8317 | EXPORT_SYMBOL(__might_sleep); | |
8318 | #endif | |
8319 | ||
8320 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8321 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8322 | { | |
da7a735e PZ |
8323 | const struct sched_class *prev_class = p->sched_class; |
8324 | int old_prio = p->prio; | |
3a5e4dc1 | 8325 | int on_rq; |
3e51f33f | 8326 | |
3a5e4dc1 AK |
8327 | on_rq = p->se.on_rq; |
8328 | if (on_rq) | |
8329 | deactivate_task(rq, p, 0); | |
8330 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8331 | if (on_rq) { | |
8332 | activate_task(rq, p, 0); | |
8333 | resched_task(rq->curr); | |
8334 | } | |
da7a735e PZ |
8335 | |
8336 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8337 | } |
8338 | ||
1da177e4 LT |
8339 | void normalize_rt_tasks(void) |
8340 | { | |
a0f98a1c | 8341 | struct task_struct *g, *p; |
1da177e4 | 8342 | unsigned long flags; |
70b97a7f | 8343 | struct rq *rq; |
1da177e4 | 8344 | |
4cf5d77a | 8345 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8346 | do_each_thread(g, p) { |
178be793 IM |
8347 | /* |
8348 | * Only normalize user tasks: | |
8349 | */ | |
8350 | if (!p->mm) | |
8351 | continue; | |
8352 | ||
6cfb0d5d | 8353 | p->se.exec_start = 0; |
6cfb0d5d | 8354 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8355 | p->se.statistics.wait_start = 0; |
8356 | p->se.statistics.sleep_start = 0; | |
8357 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8358 | #endif |
dd41f596 IM |
8359 | |
8360 | if (!rt_task(p)) { | |
8361 | /* | |
8362 | * Renice negative nice level userspace | |
8363 | * tasks back to 0: | |
8364 | */ | |
8365 | if (TASK_NICE(p) < 0 && p->mm) | |
8366 | set_user_nice(p, 0); | |
1da177e4 | 8367 | continue; |
dd41f596 | 8368 | } |
1da177e4 | 8369 | |
1d615482 | 8370 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8371 | rq = __task_rq_lock(p); |
1da177e4 | 8372 | |
178be793 | 8373 | normalize_task(rq, p); |
3a5e4dc1 | 8374 | |
b29739f9 | 8375 | __task_rq_unlock(rq); |
1d615482 | 8376 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8377 | } while_each_thread(g, p); |
8378 | ||
4cf5d77a | 8379 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8380 | } |
8381 | ||
8382 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8383 | |
67fc4e0c | 8384 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8385 | /* |
67fc4e0c | 8386 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8387 | * |
8388 | * They can only be called when the whole system has been | |
8389 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8390 | * activity can take place. Using them for anything else would | |
8391 | * be a serious bug, and as a result, they aren't even visible | |
8392 | * under any other configuration. | |
8393 | */ | |
8394 | ||
8395 | /** | |
8396 | * curr_task - return the current task for a given cpu. | |
8397 | * @cpu: the processor in question. | |
8398 | * | |
8399 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8400 | */ | |
36c8b586 | 8401 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8402 | { |
8403 | return cpu_curr(cpu); | |
8404 | } | |
8405 | ||
67fc4e0c JW |
8406 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8407 | ||
8408 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8409 | /** |
8410 | * set_curr_task - set the current task for a given cpu. | |
8411 | * @cpu: the processor in question. | |
8412 | * @p: the task pointer to set. | |
8413 | * | |
8414 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8415 | * are serviced on a separate stack. It allows the architecture to switch the |
8416 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8417 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8418 | * and caller must save the original value of the current task (see | |
8419 | * curr_task() above) and restore that value before reenabling interrupts and | |
8420 | * re-starting the system. | |
8421 | * | |
8422 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8423 | */ | |
36c8b586 | 8424 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8425 | { |
8426 | cpu_curr(cpu) = p; | |
8427 | } | |
8428 | ||
8429 | #endif | |
29f59db3 | 8430 | |
bccbe08a PZ |
8431 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8432 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8433 | { |
8434 | int i; | |
8435 | ||
8436 | for_each_possible_cpu(i) { | |
8437 | if (tg->cfs_rq) | |
8438 | kfree(tg->cfs_rq[i]); | |
8439 | if (tg->se) | |
8440 | kfree(tg->se[i]); | |
6f505b16 PZ |
8441 | } |
8442 | ||
8443 | kfree(tg->cfs_rq); | |
8444 | kfree(tg->se); | |
6f505b16 PZ |
8445 | } |
8446 | ||
ec7dc8ac DG |
8447 | static |
8448 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8449 | { |
29f59db3 | 8450 | struct cfs_rq *cfs_rq; |
eab17229 | 8451 | struct sched_entity *se; |
9b5b7751 | 8452 | struct rq *rq; |
29f59db3 SV |
8453 | int i; |
8454 | ||
434d53b0 | 8455 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8456 | if (!tg->cfs_rq) |
8457 | goto err; | |
434d53b0 | 8458 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8459 | if (!tg->se) |
8460 | goto err; | |
052f1dc7 PZ |
8461 | |
8462 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8463 | |
8464 | for_each_possible_cpu(i) { | |
9b5b7751 | 8465 | rq = cpu_rq(i); |
29f59db3 | 8466 | |
eab17229 LZ |
8467 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8468 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8469 | if (!cfs_rq) |
8470 | goto err; | |
8471 | ||
eab17229 LZ |
8472 | se = kzalloc_node(sizeof(struct sched_entity), |
8473 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8474 | if (!se) |
dfc12eb2 | 8475 | goto err_free_rq; |
29f59db3 | 8476 | |
3d4b47b4 | 8477 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8478 | } |
8479 | ||
8480 | return 1; | |
8481 | ||
49246274 | 8482 | err_free_rq: |
dfc12eb2 | 8483 | kfree(cfs_rq); |
49246274 | 8484 | err: |
bccbe08a PZ |
8485 | return 0; |
8486 | } | |
8487 | ||
bccbe08a PZ |
8488 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8489 | { | |
3d4b47b4 PZ |
8490 | struct rq *rq = cpu_rq(cpu); |
8491 | unsigned long flags; | |
3d4b47b4 PZ |
8492 | |
8493 | /* | |
8494 | * Only empty task groups can be destroyed; so we can speculatively | |
8495 | * check on_list without danger of it being re-added. | |
8496 | */ | |
8497 | if (!tg->cfs_rq[cpu]->on_list) | |
8498 | return; | |
8499 | ||
8500 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8501 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8502 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8503 | } |
6d6bc0ad | 8504 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8505 | static inline void free_fair_sched_group(struct task_group *tg) |
8506 | { | |
8507 | } | |
8508 | ||
ec7dc8ac DG |
8509 | static inline |
8510 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8511 | { |
8512 | return 1; | |
8513 | } | |
8514 | ||
bccbe08a PZ |
8515 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8516 | { | |
8517 | } | |
6d6bc0ad | 8518 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8519 | |
8520 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8521 | static void free_rt_sched_group(struct task_group *tg) |
8522 | { | |
8523 | int i; | |
8524 | ||
d0b27fa7 PZ |
8525 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8526 | ||
bccbe08a PZ |
8527 | for_each_possible_cpu(i) { |
8528 | if (tg->rt_rq) | |
8529 | kfree(tg->rt_rq[i]); | |
8530 | if (tg->rt_se) | |
8531 | kfree(tg->rt_se[i]); | |
8532 | } | |
8533 | ||
8534 | kfree(tg->rt_rq); | |
8535 | kfree(tg->rt_se); | |
8536 | } | |
8537 | ||
ec7dc8ac DG |
8538 | static |
8539 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8540 | { |
8541 | struct rt_rq *rt_rq; | |
eab17229 | 8542 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8543 | struct rq *rq; |
8544 | int i; | |
8545 | ||
434d53b0 | 8546 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8547 | if (!tg->rt_rq) |
8548 | goto err; | |
434d53b0 | 8549 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8550 | if (!tg->rt_se) |
8551 | goto err; | |
8552 | ||
d0b27fa7 PZ |
8553 | init_rt_bandwidth(&tg->rt_bandwidth, |
8554 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8555 | |
8556 | for_each_possible_cpu(i) { | |
8557 | rq = cpu_rq(i); | |
8558 | ||
eab17229 LZ |
8559 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8560 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8561 | if (!rt_rq) |
8562 | goto err; | |
29f59db3 | 8563 | |
eab17229 LZ |
8564 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8565 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8566 | if (!rt_se) |
dfc12eb2 | 8567 | goto err_free_rq; |
29f59db3 | 8568 | |
3d4b47b4 | 8569 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8570 | } |
8571 | ||
bccbe08a PZ |
8572 | return 1; |
8573 | ||
49246274 | 8574 | err_free_rq: |
dfc12eb2 | 8575 | kfree(rt_rq); |
49246274 | 8576 | err: |
bccbe08a PZ |
8577 | return 0; |
8578 | } | |
6d6bc0ad | 8579 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8580 | static inline void free_rt_sched_group(struct task_group *tg) |
8581 | { | |
8582 | } | |
8583 | ||
ec7dc8ac DG |
8584 | static inline |
8585 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8586 | { |
8587 | return 1; | |
8588 | } | |
6d6bc0ad | 8589 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8590 | |
7c941438 | 8591 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8592 | static void free_sched_group(struct task_group *tg) |
8593 | { | |
8594 | free_fair_sched_group(tg); | |
8595 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8596 | autogroup_free(tg); |
bccbe08a PZ |
8597 | kfree(tg); |
8598 | } | |
8599 | ||
8600 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8601 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8602 | { |
8603 | struct task_group *tg; | |
8604 | unsigned long flags; | |
bccbe08a PZ |
8605 | |
8606 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8607 | if (!tg) | |
8608 | return ERR_PTR(-ENOMEM); | |
8609 | ||
ec7dc8ac | 8610 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8611 | goto err; |
8612 | ||
ec7dc8ac | 8613 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8614 | goto err; |
8615 | ||
8ed36996 | 8616 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8617 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8618 | |
8619 | WARN_ON(!parent); /* root should already exist */ | |
8620 | ||
8621 | tg->parent = parent; | |
f473aa5e | 8622 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8623 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8624 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8625 | |
9b5b7751 | 8626 | return tg; |
29f59db3 SV |
8627 | |
8628 | err: | |
6f505b16 | 8629 | free_sched_group(tg); |
29f59db3 SV |
8630 | return ERR_PTR(-ENOMEM); |
8631 | } | |
8632 | ||
9b5b7751 | 8633 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8634 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8635 | { |
29f59db3 | 8636 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8637 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8638 | } |
8639 | ||
9b5b7751 | 8640 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8641 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8642 | { |
8ed36996 | 8643 | unsigned long flags; |
9b5b7751 | 8644 | int i; |
29f59db3 | 8645 | |
3d4b47b4 PZ |
8646 | /* end participation in shares distribution */ |
8647 | for_each_possible_cpu(i) | |
bccbe08a | 8648 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8649 | |
8650 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8651 | list_del_rcu(&tg->list); |
f473aa5e | 8652 | list_del_rcu(&tg->siblings); |
8ed36996 | 8653 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8654 | |
9b5b7751 | 8655 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8656 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8657 | } |
8658 | ||
9b5b7751 | 8659 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8660 | * The caller of this function should have put the task in its new group |
8661 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8662 | * reflect its new group. | |
9b5b7751 SV |
8663 | */ |
8664 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8665 | { |
8666 | int on_rq, running; | |
8667 | unsigned long flags; | |
8668 | struct rq *rq; | |
8669 | ||
8670 | rq = task_rq_lock(tsk, &flags); | |
8671 | ||
051a1d1a | 8672 | running = task_current(rq, tsk); |
29f59db3 SV |
8673 | on_rq = tsk->se.on_rq; |
8674 | ||
0e1f3483 | 8675 | if (on_rq) |
29f59db3 | 8676 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8677 | if (unlikely(running)) |
8678 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8679 | |
810b3817 | 8680 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8681 | if (tsk->sched_class->task_move_group) |
8682 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8683 | else | |
810b3817 | 8684 | #endif |
b2b5ce02 | 8685 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8686 | |
0e1f3483 HS |
8687 | if (unlikely(running)) |
8688 | tsk->sched_class->set_curr_task(rq); | |
8689 | if (on_rq) | |
371fd7e7 | 8690 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8691 | |
29f59db3 SV |
8692 | task_rq_unlock(rq, &flags); |
8693 | } | |
7c941438 | 8694 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8695 | |
052f1dc7 | 8696 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8697 | static DEFINE_MUTEX(shares_mutex); |
8698 | ||
4cf86d77 | 8699 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8700 | { |
8701 | int i; | |
8ed36996 | 8702 | unsigned long flags; |
c61935fd | 8703 | |
ec7dc8ac DG |
8704 | /* |
8705 | * We can't change the weight of the root cgroup. | |
8706 | */ | |
8707 | if (!tg->se[0]) | |
8708 | return -EINVAL; | |
8709 | ||
18d95a28 PZ |
8710 | if (shares < MIN_SHARES) |
8711 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8712 | else if (shares > MAX_SHARES) |
8713 | shares = MAX_SHARES; | |
62fb1851 | 8714 | |
8ed36996 | 8715 | mutex_lock(&shares_mutex); |
9b5b7751 | 8716 | if (tg->shares == shares) |
5cb350ba | 8717 | goto done; |
29f59db3 | 8718 | |
9b5b7751 | 8719 | tg->shares = shares; |
c09595f6 | 8720 | for_each_possible_cpu(i) { |
9437178f PT |
8721 | struct rq *rq = cpu_rq(i); |
8722 | struct sched_entity *se; | |
8723 | ||
8724 | se = tg->se[i]; | |
8725 | /* Propagate contribution to hierarchy */ | |
8726 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8727 | for_each_sched_entity(se) | |
6d5ab293 | 8728 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8729 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8730 | } |
29f59db3 | 8731 | |
5cb350ba | 8732 | done: |
8ed36996 | 8733 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8734 | return 0; |
29f59db3 SV |
8735 | } |
8736 | ||
5cb350ba DG |
8737 | unsigned long sched_group_shares(struct task_group *tg) |
8738 | { | |
8739 | return tg->shares; | |
8740 | } | |
052f1dc7 | 8741 | #endif |
5cb350ba | 8742 | |
052f1dc7 | 8743 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8744 | /* |
9f0c1e56 | 8745 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8746 | */ |
9f0c1e56 PZ |
8747 | static DEFINE_MUTEX(rt_constraints_mutex); |
8748 | ||
8749 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8750 | { | |
8751 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8752 | return 1ULL << 20; |
9f0c1e56 | 8753 | |
9a7e0b18 | 8754 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8755 | } |
8756 | ||
9a7e0b18 PZ |
8757 | /* Must be called with tasklist_lock held */ |
8758 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8759 | { |
9a7e0b18 | 8760 | struct task_struct *g, *p; |
b40b2e8e | 8761 | |
9a7e0b18 PZ |
8762 | do_each_thread(g, p) { |
8763 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8764 | return 1; | |
8765 | } while_each_thread(g, p); | |
b40b2e8e | 8766 | |
9a7e0b18 PZ |
8767 | return 0; |
8768 | } | |
b40b2e8e | 8769 | |
9a7e0b18 PZ |
8770 | struct rt_schedulable_data { |
8771 | struct task_group *tg; | |
8772 | u64 rt_period; | |
8773 | u64 rt_runtime; | |
8774 | }; | |
b40b2e8e | 8775 | |
9a7e0b18 PZ |
8776 | static int tg_schedulable(struct task_group *tg, void *data) |
8777 | { | |
8778 | struct rt_schedulable_data *d = data; | |
8779 | struct task_group *child; | |
8780 | unsigned long total, sum = 0; | |
8781 | u64 period, runtime; | |
b40b2e8e | 8782 | |
9a7e0b18 PZ |
8783 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8784 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8785 | |
9a7e0b18 PZ |
8786 | if (tg == d->tg) { |
8787 | period = d->rt_period; | |
8788 | runtime = d->rt_runtime; | |
b40b2e8e | 8789 | } |
b40b2e8e | 8790 | |
4653f803 PZ |
8791 | /* |
8792 | * Cannot have more runtime than the period. | |
8793 | */ | |
8794 | if (runtime > period && runtime != RUNTIME_INF) | |
8795 | return -EINVAL; | |
6f505b16 | 8796 | |
4653f803 PZ |
8797 | /* |
8798 | * Ensure we don't starve existing RT tasks. | |
8799 | */ | |
9a7e0b18 PZ |
8800 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8801 | return -EBUSY; | |
6f505b16 | 8802 | |
9a7e0b18 | 8803 | total = to_ratio(period, runtime); |
6f505b16 | 8804 | |
4653f803 PZ |
8805 | /* |
8806 | * Nobody can have more than the global setting allows. | |
8807 | */ | |
8808 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8809 | return -EINVAL; | |
6f505b16 | 8810 | |
4653f803 PZ |
8811 | /* |
8812 | * The sum of our children's runtime should not exceed our own. | |
8813 | */ | |
9a7e0b18 PZ |
8814 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8815 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8816 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8817 | |
9a7e0b18 PZ |
8818 | if (child == d->tg) { |
8819 | period = d->rt_period; | |
8820 | runtime = d->rt_runtime; | |
8821 | } | |
6f505b16 | 8822 | |
9a7e0b18 | 8823 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8824 | } |
6f505b16 | 8825 | |
9a7e0b18 PZ |
8826 | if (sum > total) |
8827 | return -EINVAL; | |
8828 | ||
8829 | return 0; | |
6f505b16 PZ |
8830 | } |
8831 | ||
9a7e0b18 | 8832 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8833 | { |
9a7e0b18 PZ |
8834 | struct rt_schedulable_data data = { |
8835 | .tg = tg, | |
8836 | .rt_period = period, | |
8837 | .rt_runtime = runtime, | |
8838 | }; | |
8839 | ||
8840 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8841 | } |
8842 | ||
d0b27fa7 PZ |
8843 | static int tg_set_bandwidth(struct task_group *tg, |
8844 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8845 | { |
ac086bc2 | 8846 | int i, err = 0; |
9f0c1e56 | 8847 | |
9f0c1e56 | 8848 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8849 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8850 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8851 | if (err) | |
9f0c1e56 | 8852 | goto unlock; |
ac086bc2 | 8853 | |
0986b11b | 8854 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8855 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8856 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8857 | |
8858 | for_each_possible_cpu(i) { | |
8859 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8860 | ||
0986b11b | 8861 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8862 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8863 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8864 | } |
0986b11b | 8865 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8866 | unlock: |
521f1a24 | 8867 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8868 | mutex_unlock(&rt_constraints_mutex); |
8869 | ||
8870 | return err; | |
6f505b16 PZ |
8871 | } |
8872 | ||
d0b27fa7 PZ |
8873 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8874 | { | |
8875 | u64 rt_runtime, rt_period; | |
8876 | ||
8877 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8878 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8879 | if (rt_runtime_us < 0) | |
8880 | rt_runtime = RUNTIME_INF; | |
8881 | ||
8882 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8883 | } | |
8884 | ||
9f0c1e56 PZ |
8885 | long sched_group_rt_runtime(struct task_group *tg) |
8886 | { | |
8887 | u64 rt_runtime_us; | |
8888 | ||
d0b27fa7 | 8889 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8890 | return -1; |
8891 | ||
d0b27fa7 | 8892 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8893 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8894 | return rt_runtime_us; | |
8895 | } | |
d0b27fa7 PZ |
8896 | |
8897 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8898 | { | |
8899 | u64 rt_runtime, rt_period; | |
8900 | ||
8901 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8902 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8903 | ||
619b0488 R |
8904 | if (rt_period == 0) |
8905 | return -EINVAL; | |
8906 | ||
d0b27fa7 PZ |
8907 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8908 | } | |
8909 | ||
8910 | long sched_group_rt_period(struct task_group *tg) | |
8911 | { | |
8912 | u64 rt_period_us; | |
8913 | ||
8914 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8915 | do_div(rt_period_us, NSEC_PER_USEC); | |
8916 | return rt_period_us; | |
8917 | } | |
8918 | ||
8919 | static int sched_rt_global_constraints(void) | |
8920 | { | |
4653f803 | 8921 | u64 runtime, period; |
d0b27fa7 PZ |
8922 | int ret = 0; |
8923 | ||
ec5d4989 HS |
8924 | if (sysctl_sched_rt_period <= 0) |
8925 | return -EINVAL; | |
8926 | ||
4653f803 PZ |
8927 | runtime = global_rt_runtime(); |
8928 | period = global_rt_period(); | |
8929 | ||
8930 | /* | |
8931 | * Sanity check on the sysctl variables. | |
8932 | */ | |
8933 | if (runtime > period && runtime != RUNTIME_INF) | |
8934 | return -EINVAL; | |
10b612f4 | 8935 | |
d0b27fa7 | 8936 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8937 | read_lock(&tasklist_lock); |
4653f803 | 8938 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8939 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8940 | mutex_unlock(&rt_constraints_mutex); |
8941 | ||
8942 | return ret; | |
8943 | } | |
54e99124 DG |
8944 | |
8945 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8946 | { | |
8947 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8948 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8949 | return 0; | |
8950 | ||
8951 | return 1; | |
8952 | } | |
8953 | ||
6d6bc0ad | 8954 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8955 | static int sched_rt_global_constraints(void) |
8956 | { | |
ac086bc2 PZ |
8957 | unsigned long flags; |
8958 | int i; | |
8959 | ||
ec5d4989 HS |
8960 | if (sysctl_sched_rt_period <= 0) |
8961 | return -EINVAL; | |
8962 | ||
60aa605d PZ |
8963 | /* |
8964 | * There's always some RT tasks in the root group | |
8965 | * -- migration, kstopmachine etc.. | |
8966 | */ | |
8967 | if (sysctl_sched_rt_runtime == 0) | |
8968 | return -EBUSY; | |
8969 | ||
0986b11b | 8970 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8971 | for_each_possible_cpu(i) { |
8972 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8973 | ||
0986b11b | 8974 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8975 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8976 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8977 | } |
0986b11b | 8978 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8979 | |
d0b27fa7 PZ |
8980 | return 0; |
8981 | } | |
6d6bc0ad | 8982 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8983 | |
8984 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8985 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8986 | loff_t *ppos) |
8987 | { | |
8988 | int ret; | |
8989 | int old_period, old_runtime; | |
8990 | static DEFINE_MUTEX(mutex); | |
8991 | ||
8992 | mutex_lock(&mutex); | |
8993 | old_period = sysctl_sched_rt_period; | |
8994 | old_runtime = sysctl_sched_rt_runtime; | |
8995 | ||
8d65af78 | 8996 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8997 | |
8998 | if (!ret && write) { | |
8999 | ret = sched_rt_global_constraints(); | |
9000 | if (ret) { | |
9001 | sysctl_sched_rt_period = old_period; | |
9002 | sysctl_sched_rt_runtime = old_runtime; | |
9003 | } else { | |
9004 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9005 | def_rt_bandwidth.rt_period = | |
9006 | ns_to_ktime(global_rt_period()); | |
9007 | } | |
9008 | } | |
9009 | mutex_unlock(&mutex); | |
9010 | ||
9011 | return ret; | |
9012 | } | |
68318b8e | 9013 | |
052f1dc7 | 9014 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9015 | |
9016 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9017 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9018 | { |
2b01dfe3 PM |
9019 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9020 | struct task_group, css); | |
68318b8e SV |
9021 | } |
9022 | ||
9023 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9024 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9025 | { |
ec7dc8ac | 9026 | struct task_group *tg, *parent; |
68318b8e | 9027 | |
2b01dfe3 | 9028 | if (!cgrp->parent) { |
68318b8e | 9029 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9030 | return &root_task_group.css; |
68318b8e SV |
9031 | } |
9032 | ||
ec7dc8ac DG |
9033 | parent = cgroup_tg(cgrp->parent); |
9034 | tg = sched_create_group(parent); | |
68318b8e SV |
9035 | if (IS_ERR(tg)) |
9036 | return ERR_PTR(-ENOMEM); | |
9037 | ||
68318b8e SV |
9038 | return &tg->css; |
9039 | } | |
9040 | ||
41a2d6cf IM |
9041 | static void |
9042 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9043 | { |
2b01dfe3 | 9044 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9045 | |
9046 | sched_destroy_group(tg); | |
9047 | } | |
9048 | ||
41a2d6cf | 9049 | static int |
be367d09 | 9050 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9051 | { |
b68aa230 | 9052 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9053 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9054 | return -EINVAL; |
9055 | #else | |
68318b8e SV |
9056 | /* We don't support RT-tasks being in separate groups */ |
9057 | if (tsk->sched_class != &fair_sched_class) | |
9058 | return -EINVAL; | |
b68aa230 | 9059 | #endif |
be367d09 BB |
9060 | return 0; |
9061 | } | |
68318b8e | 9062 | |
be367d09 BB |
9063 | static int |
9064 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9065 | struct task_struct *tsk, bool threadgroup) | |
9066 | { | |
9067 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9068 | if (retval) | |
9069 | return retval; | |
9070 | if (threadgroup) { | |
9071 | struct task_struct *c; | |
9072 | rcu_read_lock(); | |
9073 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9074 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9075 | if (retval) { | |
9076 | rcu_read_unlock(); | |
9077 | return retval; | |
9078 | } | |
9079 | } | |
9080 | rcu_read_unlock(); | |
9081 | } | |
68318b8e SV |
9082 | return 0; |
9083 | } | |
9084 | ||
9085 | static void | |
2b01dfe3 | 9086 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9087 | struct cgroup *old_cont, struct task_struct *tsk, |
9088 | bool threadgroup) | |
68318b8e SV |
9089 | { |
9090 | sched_move_task(tsk); | |
be367d09 BB |
9091 | if (threadgroup) { |
9092 | struct task_struct *c; | |
9093 | rcu_read_lock(); | |
9094 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9095 | sched_move_task(c); | |
9096 | } | |
9097 | rcu_read_unlock(); | |
9098 | } | |
68318b8e SV |
9099 | } |
9100 | ||
068c5cc5 | 9101 | static void |
d41d5a01 PZ |
9102 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
9103 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
9104 | { |
9105 | /* | |
9106 | * cgroup_exit() is called in the copy_process() failure path. | |
9107 | * Ignore this case since the task hasn't ran yet, this avoids | |
9108 | * trying to poke a half freed task state from generic code. | |
9109 | */ | |
9110 | if (!(task->flags & PF_EXITING)) | |
9111 | return; | |
9112 | ||
9113 | sched_move_task(task); | |
9114 | } | |
9115 | ||
052f1dc7 | 9116 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9117 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9118 | u64 shareval) |
68318b8e | 9119 | { |
2b01dfe3 | 9120 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9121 | } |
9122 | ||
f4c753b7 | 9123 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9124 | { |
2b01dfe3 | 9125 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9126 | |
9127 | return (u64) tg->shares; | |
9128 | } | |
6d6bc0ad | 9129 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9130 | |
052f1dc7 | 9131 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9132 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9133 | s64 val) |
6f505b16 | 9134 | { |
06ecb27c | 9135 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9136 | } |
9137 | ||
06ecb27c | 9138 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9139 | { |
06ecb27c | 9140 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9141 | } |
d0b27fa7 PZ |
9142 | |
9143 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9144 | u64 rt_period_us) | |
9145 | { | |
9146 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9147 | } | |
9148 | ||
9149 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9150 | { | |
9151 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9152 | } | |
6d6bc0ad | 9153 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9154 | |
fe5c7cc2 | 9155 | static struct cftype cpu_files[] = { |
052f1dc7 | 9156 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9157 | { |
9158 | .name = "shares", | |
f4c753b7 PM |
9159 | .read_u64 = cpu_shares_read_u64, |
9160 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9161 | }, |
052f1dc7 PZ |
9162 | #endif |
9163 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9164 | { |
9f0c1e56 | 9165 | .name = "rt_runtime_us", |
06ecb27c PM |
9166 | .read_s64 = cpu_rt_runtime_read, |
9167 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9168 | }, |
d0b27fa7 PZ |
9169 | { |
9170 | .name = "rt_period_us", | |
f4c753b7 PM |
9171 | .read_u64 = cpu_rt_period_read_uint, |
9172 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9173 | }, |
052f1dc7 | 9174 | #endif |
68318b8e SV |
9175 | }; |
9176 | ||
9177 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9178 | { | |
fe5c7cc2 | 9179 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9180 | } |
9181 | ||
9182 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9183 | .name = "cpu", |
9184 | .create = cpu_cgroup_create, | |
9185 | .destroy = cpu_cgroup_destroy, | |
9186 | .can_attach = cpu_cgroup_can_attach, | |
9187 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 9188 | .exit = cpu_cgroup_exit, |
38605cae IM |
9189 | .populate = cpu_cgroup_populate, |
9190 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9191 | .early_init = 1, |
9192 | }; | |
9193 | ||
052f1dc7 | 9194 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9195 | |
9196 | #ifdef CONFIG_CGROUP_CPUACCT | |
9197 | ||
9198 | /* | |
9199 | * CPU accounting code for task groups. | |
9200 | * | |
9201 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9202 | * (balbir@in.ibm.com). | |
9203 | */ | |
9204 | ||
934352f2 | 9205 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9206 | struct cpuacct { |
9207 | struct cgroup_subsys_state css; | |
9208 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9209 | u64 __percpu *cpuusage; |
ef12fefa | 9210 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9211 | struct cpuacct *parent; |
d842de87 SV |
9212 | }; |
9213 | ||
9214 | struct cgroup_subsys cpuacct_subsys; | |
9215 | ||
9216 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9217 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9218 | { |
32cd756a | 9219 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9220 | struct cpuacct, css); |
9221 | } | |
9222 | ||
9223 | /* return cpu accounting group to which this task belongs */ | |
9224 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9225 | { | |
9226 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9227 | struct cpuacct, css); | |
9228 | } | |
9229 | ||
9230 | /* create a new cpu accounting group */ | |
9231 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9232 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9233 | { |
9234 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9235 | int i; |
d842de87 SV |
9236 | |
9237 | if (!ca) | |
ef12fefa | 9238 | goto out; |
d842de87 SV |
9239 | |
9240 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9241 | if (!ca->cpuusage) |
9242 | goto out_free_ca; | |
9243 | ||
9244 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9245 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9246 | goto out_free_counters; | |
d842de87 | 9247 | |
934352f2 BR |
9248 | if (cgrp->parent) |
9249 | ca->parent = cgroup_ca(cgrp->parent); | |
9250 | ||
d842de87 | 9251 | return &ca->css; |
ef12fefa BR |
9252 | |
9253 | out_free_counters: | |
9254 | while (--i >= 0) | |
9255 | percpu_counter_destroy(&ca->cpustat[i]); | |
9256 | free_percpu(ca->cpuusage); | |
9257 | out_free_ca: | |
9258 | kfree(ca); | |
9259 | out: | |
9260 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9261 | } |
9262 | ||
9263 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9264 | static void |
32cd756a | 9265 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9266 | { |
32cd756a | 9267 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9268 | int i; |
d842de87 | 9269 | |
ef12fefa BR |
9270 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9271 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9272 | free_percpu(ca->cpuusage); |
9273 | kfree(ca); | |
9274 | } | |
9275 | ||
720f5498 KC |
9276 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9277 | { | |
b36128c8 | 9278 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9279 | u64 data; |
9280 | ||
9281 | #ifndef CONFIG_64BIT | |
9282 | /* | |
9283 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9284 | */ | |
05fa785c | 9285 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9286 | data = *cpuusage; |
05fa785c | 9287 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9288 | #else |
9289 | data = *cpuusage; | |
9290 | #endif | |
9291 | ||
9292 | return data; | |
9293 | } | |
9294 | ||
9295 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9296 | { | |
b36128c8 | 9297 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9298 | |
9299 | #ifndef CONFIG_64BIT | |
9300 | /* | |
9301 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9302 | */ | |
05fa785c | 9303 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9304 | *cpuusage = val; |
05fa785c | 9305 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9306 | #else |
9307 | *cpuusage = val; | |
9308 | #endif | |
9309 | } | |
9310 | ||
d842de87 | 9311 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9312 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9313 | { |
32cd756a | 9314 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9315 | u64 totalcpuusage = 0; |
9316 | int i; | |
9317 | ||
720f5498 KC |
9318 | for_each_present_cpu(i) |
9319 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9320 | |
9321 | return totalcpuusage; | |
9322 | } | |
9323 | ||
0297b803 DG |
9324 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9325 | u64 reset) | |
9326 | { | |
9327 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9328 | int err = 0; | |
9329 | int i; | |
9330 | ||
9331 | if (reset) { | |
9332 | err = -EINVAL; | |
9333 | goto out; | |
9334 | } | |
9335 | ||
720f5498 KC |
9336 | for_each_present_cpu(i) |
9337 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9338 | |
0297b803 DG |
9339 | out: |
9340 | return err; | |
9341 | } | |
9342 | ||
e9515c3c KC |
9343 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9344 | struct seq_file *m) | |
9345 | { | |
9346 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9347 | u64 percpu; | |
9348 | int i; | |
9349 | ||
9350 | for_each_present_cpu(i) { | |
9351 | percpu = cpuacct_cpuusage_read(ca, i); | |
9352 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9353 | } | |
9354 | seq_printf(m, "\n"); | |
9355 | return 0; | |
9356 | } | |
9357 | ||
ef12fefa BR |
9358 | static const char *cpuacct_stat_desc[] = { |
9359 | [CPUACCT_STAT_USER] = "user", | |
9360 | [CPUACCT_STAT_SYSTEM] = "system", | |
9361 | }; | |
9362 | ||
9363 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9364 | struct cgroup_map_cb *cb) | |
9365 | { | |
9366 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9367 | int i; | |
9368 | ||
9369 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9370 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9371 | val = cputime64_to_clock_t(val); | |
9372 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9373 | } | |
9374 | return 0; | |
9375 | } | |
9376 | ||
d842de87 SV |
9377 | static struct cftype files[] = { |
9378 | { | |
9379 | .name = "usage", | |
f4c753b7 PM |
9380 | .read_u64 = cpuusage_read, |
9381 | .write_u64 = cpuusage_write, | |
d842de87 | 9382 | }, |
e9515c3c KC |
9383 | { |
9384 | .name = "usage_percpu", | |
9385 | .read_seq_string = cpuacct_percpu_seq_read, | |
9386 | }, | |
ef12fefa BR |
9387 | { |
9388 | .name = "stat", | |
9389 | .read_map = cpuacct_stats_show, | |
9390 | }, | |
d842de87 SV |
9391 | }; |
9392 | ||
32cd756a | 9393 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9394 | { |
32cd756a | 9395 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9396 | } |
9397 | ||
9398 | /* | |
9399 | * charge this task's execution time to its accounting group. | |
9400 | * | |
9401 | * called with rq->lock held. | |
9402 | */ | |
9403 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9404 | { | |
9405 | struct cpuacct *ca; | |
934352f2 | 9406 | int cpu; |
d842de87 | 9407 | |
c40c6f85 | 9408 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9409 | return; |
9410 | ||
934352f2 | 9411 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9412 | |
9413 | rcu_read_lock(); | |
9414 | ||
d842de87 | 9415 | ca = task_ca(tsk); |
d842de87 | 9416 | |
934352f2 | 9417 | for (; ca; ca = ca->parent) { |
b36128c8 | 9418 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9419 | *cpuusage += cputime; |
9420 | } | |
a18b83b7 BR |
9421 | |
9422 | rcu_read_unlock(); | |
d842de87 SV |
9423 | } |
9424 | ||
fa535a77 AB |
9425 | /* |
9426 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9427 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9428 | * percpu_counter_add with values large enough to always overflow the | |
9429 | * per cpu batch limit causing bad SMP scalability. | |
9430 | * | |
9431 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9432 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9433 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9434 | */ | |
9435 | #ifdef CONFIG_SMP | |
9436 | #define CPUACCT_BATCH \ | |
9437 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9438 | #else | |
9439 | #define CPUACCT_BATCH 0 | |
9440 | #endif | |
9441 | ||
ef12fefa BR |
9442 | /* |
9443 | * Charge the system/user time to the task's accounting group. | |
9444 | */ | |
9445 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9446 | enum cpuacct_stat_index idx, cputime_t val) | |
9447 | { | |
9448 | struct cpuacct *ca; | |
fa535a77 | 9449 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9450 | |
9451 | if (unlikely(!cpuacct_subsys.active)) | |
9452 | return; | |
9453 | ||
9454 | rcu_read_lock(); | |
9455 | ca = task_ca(tsk); | |
9456 | ||
9457 | do { | |
fa535a77 | 9458 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9459 | ca = ca->parent; |
9460 | } while (ca); | |
9461 | rcu_read_unlock(); | |
9462 | } | |
9463 | ||
d842de87 SV |
9464 | struct cgroup_subsys cpuacct_subsys = { |
9465 | .name = "cpuacct", | |
9466 | .create = cpuacct_create, | |
9467 | .destroy = cpuacct_destroy, | |
9468 | .populate = cpuacct_populate, | |
9469 | .subsys_id = cpuacct_subsys_id, | |
9470 | }; | |
9471 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9472 |