]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
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> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
52f5684c | 76 | #include <linux/compiler.h> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
029632fb | 93 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 94 | { |
58088ad0 PT |
95 | unsigned long delta; |
96 | ktime_t soft, hard, now; | |
d0b27fa7 | 97 | |
58088ad0 PT |
98 | for (;;) { |
99 | if (hrtimer_active(period_timer)) | |
100 | break; | |
101 | ||
102 | now = hrtimer_cb_get_time(period_timer); | |
103 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 104 | |
58088ad0 PT |
105 | soft = hrtimer_get_softexpires(period_timer); |
106 | hard = hrtimer_get_expires(period_timer); | |
107 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
108 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
109 | HRTIMER_MODE_ABS_PINNED, 0); | |
110 | } | |
111 | } | |
112 | ||
029632fb PZ |
113 | DEFINE_MUTEX(sched_domains_mutex); |
114 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 115 | |
fe44d621 | 116 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 117 | |
029632fb | 118 | void update_rq_clock(struct rq *rq) |
3e51f33f | 119 | { |
fe44d621 | 120 | s64 delta; |
305e6835 | 121 | |
61eadef6 | 122 | if (rq->skip_clock_update > 0) |
f26f9aff | 123 | return; |
aa483808 | 124 | |
fe44d621 | 125 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
126 | if (delta < 0) |
127 | return; | |
fe44d621 PZ |
128 | rq->clock += delta; |
129 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
130 | } |
131 | ||
bf5c91ba IM |
132 | /* |
133 | * Debugging: various feature bits | |
134 | */ | |
f00b45c1 | 135 | |
f00b45c1 PZ |
136 | #define SCHED_FEAT(name, enabled) \ |
137 | (1UL << __SCHED_FEAT_##name) * enabled | | |
138 | ||
bf5c91ba | 139 | const_debug unsigned int sysctl_sched_features = |
391e43da | 140 | #include "features.h" |
f00b45c1 PZ |
141 | 0; |
142 | ||
143 | #undef SCHED_FEAT | |
144 | ||
145 | #ifdef CONFIG_SCHED_DEBUG | |
146 | #define SCHED_FEAT(name, enabled) \ | |
147 | #name , | |
148 | ||
1292531f | 149 | static const char * const sched_feat_names[] = { |
391e43da | 150 | #include "features.h" |
f00b45c1 PZ |
151 | }; |
152 | ||
153 | #undef SCHED_FEAT | |
154 | ||
34f3a814 | 155 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 156 | { |
f00b45c1 PZ |
157 | int i; |
158 | ||
f8b6d1cc | 159 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
160 | if (!(sysctl_sched_features & (1UL << i))) |
161 | seq_puts(m, "NO_"); | |
162 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 163 | } |
34f3a814 | 164 | seq_puts(m, "\n"); |
f00b45c1 | 165 | |
34f3a814 | 166 | return 0; |
f00b45c1 PZ |
167 | } |
168 | ||
f8b6d1cc PZ |
169 | #ifdef HAVE_JUMP_LABEL |
170 | ||
c5905afb IM |
171 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
172 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
173 | |
174 | #define SCHED_FEAT(name, enabled) \ | |
175 | jump_label_key__##enabled , | |
176 | ||
c5905afb | 177 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
178 | #include "features.h" |
179 | }; | |
180 | ||
181 | #undef SCHED_FEAT | |
182 | ||
183 | static void sched_feat_disable(int i) | |
184 | { | |
c5905afb IM |
185 | if (static_key_enabled(&sched_feat_keys[i])) |
186 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
187 | } |
188 | ||
189 | static void sched_feat_enable(int i) | |
190 | { | |
c5905afb IM |
191 | if (!static_key_enabled(&sched_feat_keys[i])) |
192 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
193 | } |
194 | #else | |
195 | static void sched_feat_disable(int i) { }; | |
196 | static void sched_feat_enable(int i) { }; | |
197 | #endif /* HAVE_JUMP_LABEL */ | |
198 | ||
1a687c2e | 199 | static int sched_feat_set(char *cmp) |
f00b45c1 | 200 | { |
f00b45c1 | 201 | int i; |
1a687c2e | 202 | int neg = 0; |
f00b45c1 | 203 | |
524429c3 | 204 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
205 | neg = 1; |
206 | cmp += 3; | |
207 | } | |
208 | ||
f8b6d1cc | 209 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 210 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 211 | if (neg) { |
f00b45c1 | 212 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_disable(i); |
214 | } else { | |
f00b45c1 | 215 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
216 | sched_feat_enable(i); |
217 | } | |
f00b45c1 PZ |
218 | break; |
219 | } | |
220 | } | |
221 | ||
1a687c2e MG |
222 | return i; |
223 | } | |
224 | ||
225 | static ssize_t | |
226 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
227 | size_t cnt, loff_t *ppos) | |
228 | { | |
229 | char buf[64]; | |
230 | char *cmp; | |
231 | int i; | |
5cd08fbf | 232 | struct inode *inode; |
1a687c2e MG |
233 | |
234 | if (cnt > 63) | |
235 | cnt = 63; | |
236 | ||
237 | if (copy_from_user(&buf, ubuf, cnt)) | |
238 | return -EFAULT; | |
239 | ||
240 | buf[cnt] = 0; | |
241 | cmp = strstrip(buf); | |
242 | ||
5cd08fbf JB |
243 | /* Ensure the static_key remains in a consistent state */ |
244 | inode = file_inode(filp); | |
245 | mutex_lock(&inode->i_mutex); | |
1a687c2e | 246 | i = sched_feat_set(cmp); |
5cd08fbf | 247 | mutex_unlock(&inode->i_mutex); |
f8b6d1cc | 248 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
249 | return -EINVAL; |
250 | ||
42994724 | 251 | *ppos += cnt; |
f00b45c1 PZ |
252 | |
253 | return cnt; | |
254 | } | |
255 | ||
34f3a814 LZ |
256 | static int sched_feat_open(struct inode *inode, struct file *filp) |
257 | { | |
258 | return single_open(filp, sched_feat_show, NULL); | |
259 | } | |
260 | ||
828c0950 | 261 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
262 | .open = sched_feat_open, |
263 | .write = sched_feat_write, | |
264 | .read = seq_read, | |
265 | .llseek = seq_lseek, | |
266 | .release = single_release, | |
f00b45c1 PZ |
267 | }; |
268 | ||
269 | static __init int sched_init_debug(void) | |
270 | { | |
f00b45c1 PZ |
271 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
272 | &sched_feat_fops); | |
273 | ||
274 | return 0; | |
275 | } | |
276 | late_initcall(sched_init_debug); | |
f8b6d1cc | 277 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 278 | |
b82d9fdd PZ |
279 | /* |
280 | * Number of tasks to iterate in a single balance run. | |
281 | * Limited because this is done with IRQs disabled. | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
284 | ||
e9e9250b PZ |
285 | /* |
286 | * period over which we average the RT time consumption, measured | |
287 | * in ms. | |
288 | * | |
289 | * default: 1s | |
290 | */ | |
291 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
292 | ||
fa85ae24 | 293 | /* |
9f0c1e56 | 294 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
295 | * default: 1s |
296 | */ | |
9f0c1e56 | 297 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 298 | |
029632fb | 299 | __read_mostly int scheduler_running; |
6892b75e | 300 | |
9f0c1e56 PZ |
301 | /* |
302 | * part of the period that we allow rt tasks to run in us. | |
303 | * default: 0.95s | |
304 | */ | |
305 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 306 | |
0970d299 | 307 | /* |
0122ec5b | 308 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 309 | */ |
70b97a7f | 310 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
311 | __acquires(rq->lock) |
312 | { | |
0970d299 PZ |
313 | struct rq *rq; |
314 | ||
0122ec5b PZ |
315 | lockdep_assert_held(&p->pi_lock); |
316 | ||
3a5c359a | 317 | for (;;) { |
0970d299 | 318 | rq = task_rq(p); |
05fa785c | 319 | raw_spin_lock(&rq->lock); |
cca26e80 | 320 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) |
3a5c359a | 321 | return rq; |
05fa785c | 322 | raw_spin_unlock(&rq->lock); |
cca26e80 KT |
323 | |
324 | while (unlikely(task_on_rq_migrating(p))) | |
325 | cpu_relax(); | |
b29739f9 | 326 | } |
b29739f9 IM |
327 | } |
328 | ||
1da177e4 | 329 | /* |
0122ec5b | 330 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 331 | */ |
70b97a7f | 332 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 333 | __acquires(p->pi_lock) |
1da177e4 LT |
334 | __acquires(rq->lock) |
335 | { | |
70b97a7f | 336 | struct rq *rq; |
1da177e4 | 337 | |
3a5c359a | 338 | for (;;) { |
0122ec5b | 339 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 340 | rq = task_rq(p); |
05fa785c | 341 | raw_spin_lock(&rq->lock); |
cca26e80 | 342 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) |
3a5c359a | 343 | return rq; |
0122ec5b PZ |
344 | raw_spin_unlock(&rq->lock); |
345 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
cca26e80 KT |
346 | |
347 | while (unlikely(task_on_rq_migrating(p))) | |
348 | cpu_relax(); | |
1da177e4 | 349 | } |
1da177e4 LT |
350 | } |
351 | ||
a9957449 | 352 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
353 | __releases(rq->lock) |
354 | { | |
05fa785c | 355 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
356 | } |
357 | ||
0122ec5b PZ |
358 | static inline void |
359 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 360 | __releases(rq->lock) |
0122ec5b | 361 | __releases(p->pi_lock) |
1da177e4 | 362 | { |
0122ec5b PZ |
363 | raw_spin_unlock(&rq->lock); |
364 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
365 | } |
366 | ||
1da177e4 | 367 | /* |
cc2a73b5 | 368 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 369 | */ |
a9957449 | 370 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
371 | __acquires(rq->lock) |
372 | { | |
70b97a7f | 373 | struct rq *rq; |
1da177e4 LT |
374 | |
375 | local_irq_disable(); | |
376 | rq = this_rq(); | |
05fa785c | 377 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
378 | |
379 | return rq; | |
380 | } | |
381 | ||
8f4d37ec PZ |
382 | #ifdef CONFIG_SCHED_HRTICK |
383 | /* | |
384 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 385 | */ |
8f4d37ec | 386 | |
8f4d37ec PZ |
387 | static void hrtick_clear(struct rq *rq) |
388 | { | |
389 | if (hrtimer_active(&rq->hrtick_timer)) | |
390 | hrtimer_cancel(&rq->hrtick_timer); | |
391 | } | |
392 | ||
8f4d37ec PZ |
393 | /* |
394 | * High-resolution timer tick. | |
395 | * Runs from hardirq context with interrupts disabled. | |
396 | */ | |
397 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
398 | { | |
399 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
400 | ||
401 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
402 | ||
05fa785c | 403 | raw_spin_lock(&rq->lock); |
3e51f33f | 404 | update_rq_clock(rq); |
8f4d37ec | 405 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 406 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
407 | |
408 | return HRTIMER_NORESTART; | |
409 | } | |
410 | ||
95e904c7 | 411 | #ifdef CONFIG_SMP |
971ee28c PZ |
412 | |
413 | static int __hrtick_restart(struct rq *rq) | |
414 | { | |
415 | struct hrtimer *timer = &rq->hrtick_timer; | |
416 | ktime_t time = hrtimer_get_softexpires(timer); | |
417 | ||
418 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
419 | } | |
420 | ||
31656519 PZ |
421 | /* |
422 | * called from hardirq (IPI) context | |
423 | */ | |
424 | static void __hrtick_start(void *arg) | |
b328ca18 | 425 | { |
31656519 | 426 | struct rq *rq = arg; |
b328ca18 | 427 | |
05fa785c | 428 | raw_spin_lock(&rq->lock); |
971ee28c | 429 | __hrtick_restart(rq); |
31656519 | 430 | rq->hrtick_csd_pending = 0; |
05fa785c | 431 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
432 | } |
433 | ||
31656519 PZ |
434 | /* |
435 | * Called to set the hrtick timer state. | |
436 | * | |
437 | * called with rq->lock held and irqs disabled | |
438 | */ | |
029632fb | 439 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 440 | { |
31656519 | 441 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 442 | ktime_t time; |
443 | s64 delta; | |
444 | ||
445 | /* | |
446 | * Don't schedule slices shorter than 10000ns, that just | |
447 | * doesn't make sense and can cause timer DoS. | |
448 | */ | |
449 | delta = max_t(s64, delay, 10000LL); | |
450 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 451 | |
cc584b21 | 452 | hrtimer_set_expires(timer, time); |
31656519 PZ |
453 | |
454 | if (rq == this_rq()) { | |
971ee28c | 455 | __hrtick_restart(rq); |
31656519 | 456 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 457 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
458 | rq->hrtick_csd_pending = 1; |
459 | } | |
b328ca18 PZ |
460 | } |
461 | ||
462 | static int | |
463 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
464 | { | |
465 | int cpu = (int)(long)hcpu; | |
466 | ||
467 | switch (action) { | |
468 | case CPU_UP_CANCELED: | |
469 | case CPU_UP_CANCELED_FROZEN: | |
470 | case CPU_DOWN_PREPARE: | |
471 | case CPU_DOWN_PREPARE_FROZEN: | |
472 | case CPU_DEAD: | |
473 | case CPU_DEAD_FROZEN: | |
31656519 | 474 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
475 | return NOTIFY_OK; |
476 | } | |
477 | ||
478 | return NOTIFY_DONE; | |
479 | } | |
480 | ||
fa748203 | 481 | static __init void init_hrtick(void) |
b328ca18 PZ |
482 | { |
483 | hotcpu_notifier(hotplug_hrtick, 0); | |
484 | } | |
31656519 PZ |
485 | #else |
486 | /* | |
487 | * Called to set the hrtick timer state. | |
488 | * | |
489 | * called with rq->lock held and irqs disabled | |
490 | */ | |
029632fb | 491 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 492 | { |
7f1e2ca9 | 493 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 494 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 495 | } |
b328ca18 | 496 | |
006c75f1 | 497 | static inline void init_hrtick(void) |
8f4d37ec | 498 | { |
8f4d37ec | 499 | } |
31656519 | 500 | #endif /* CONFIG_SMP */ |
8f4d37ec | 501 | |
31656519 | 502 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 503 | { |
31656519 PZ |
504 | #ifdef CONFIG_SMP |
505 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 506 | |
31656519 PZ |
507 | rq->hrtick_csd.flags = 0; |
508 | rq->hrtick_csd.func = __hrtick_start; | |
509 | rq->hrtick_csd.info = rq; | |
510 | #endif | |
8f4d37ec | 511 | |
31656519 PZ |
512 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
513 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 514 | } |
006c75f1 | 515 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
516 | static inline void hrtick_clear(struct rq *rq) |
517 | { | |
518 | } | |
519 | ||
8f4d37ec PZ |
520 | static inline void init_rq_hrtick(struct rq *rq) |
521 | { | |
522 | } | |
523 | ||
b328ca18 PZ |
524 | static inline void init_hrtick(void) |
525 | { | |
526 | } | |
006c75f1 | 527 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 528 | |
fd99f91a PZ |
529 | /* |
530 | * cmpxchg based fetch_or, macro so it works for different integer types | |
531 | */ | |
532 | #define fetch_or(ptr, val) \ | |
533 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
534 | for (;;) { \ | |
535 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
536 | if (__old == __val) \ | |
537 | break; \ | |
538 | __val = __old; \ | |
539 | } \ | |
540 | __old; \ | |
541 | }) | |
542 | ||
e3baac47 | 543 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
544 | /* |
545 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
546 | * this avoids any races wrt polling state changes and thereby avoids | |
547 | * spurious IPIs. | |
548 | */ | |
549 | static bool set_nr_and_not_polling(struct task_struct *p) | |
550 | { | |
551 | struct thread_info *ti = task_thread_info(p); | |
552 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
553 | } | |
e3baac47 PZ |
554 | |
555 | /* | |
556 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
557 | * | |
558 | * If this returns true, then the idle task promises to call | |
559 | * sched_ttwu_pending() and reschedule soon. | |
560 | */ | |
561 | static bool set_nr_if_polling(struct task_struct *p) | |
562 | { | |
563 | struct thread_info *ti = task_thread_info(p); | |
564 | typeof(ti->flags) old, val = ACCESS_ONCE(ti->flags); | |
565 | ||
566 | for (;;) { | |
567 | if (!(val & _TIF_POLLING_NRFLAG)) | |
568 | return false; | |
569 | if (val & _TIF_NEED_RESCHED) | |
570 | return true; | |
571 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
572 | if (old == val) | |
573 | break; | |
574 | val = old; | |
575 | } | |
576 | return true; | |
577 | } | |
578 | ||
fd99f91a PZ |
579 | #else |
580 | static bool set_nr_and_not_polling(struct task_struct *p) | |
581 | { | |
582 | set_tsk_need_resched(p); | |
583 | return true; | |
584 | } | |
e3baac47 PZ |
585 | |
586 | #ifdef CONFIG_SMP | |
587 | static bool set_nr_if_polling(struct task_struct *p) | |
588 | { | |
589 | return false; | |
590 | } | |
591 | #endif | |
fd99f91a PZ |
592 | #endif |
593 | ||
c24d20db | 594 | /* |
8875125e | 595 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
596 | * |
597 | * On UP this means the setting of the need_resched flag, on SMP it | |
598 | * might also involve a cross-CPU call to trigger the scheduler on | |
599 | * the target CPU. | |
600 | */ | |
8875125e | 601 | void resched_curr(struct rq *rq) |
c24d20db | 602 | { |
8875125e | 603 | struct task_struct *curr = rq->curr; |
c24d20db IM |
604 | int cpu; |
605 | ||
8875125e | 606 | lockdep_assert_held(&rq->lock); |
c24d20db | 607 | |
8875125e | 608 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
609 | return; |
610 | ||
8875125e | 611 | cpu = cpu_of(rq); |
fd99f91a | 612 | |
f27dde8d | 613 | if (cpu == smp_processor_id()) { |
8875125e | 614 | set_tsk_need_resched(curr); |
f27dde8d | 615 | set_preempt_need_resched(); |
c24d20db | 616 | return; |
f27dde8d | 617 | } |
c24d20db | 618 | |
8875125e | 619 | if (set_nr_and_not_polling(curr)) |
c24d20db | 620 | smp_send_reschedule(cpu); |
dfc68f29 AL |
621 | else |
622 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
623 | } |
624 | ||
029632fb | 625 | void resched_cpu(int cpu) |
c24d20db IM |
626 | { |
627 | struct rq *rq = cpu_rq(cpu); | |
628 | unsigned long flags; | |
629 | ||
05fa785c | 630 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 631 | return; |
8875125e | 632 | resched_curr(rq); |
05fa785c | 633 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 634 | } |
06d8308c | 635 | |
b021fe3e | 636 | #ifdef CONFIG_SMP |
3451d024 | 637 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
638 | /* |
639 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
640 | * from an idle cpu. This is good for power-savings. | |
641 | * | |
642 | * We don't do similar optimization for completely idle system, as | |
643 | * selecting an idle cpu will add more delays to the timers than intended | |
644 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
645 | */ | |
6201b4d6 | 646 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
647 | { |
648 | int cpu = smp_processor_id(); | |
649 | int i; | |
650 | struct sched_domain *sd; | |
651 | ||
6201b4d6 VK |
652 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
653 | return cpu; | |
654 | ||
057f3fad | 655 | rcu_read_lock(); |
83cd4fe2 | 656 | for_each_domain(cpu, sd) { |
057f3fad PZ |
657 | for_each_cpu(i, sched_domain_span(sd)) { |
658 | if (!idle_cpu(i)) { | |
659 | cpu = i; | |
660 | goto unlock; | |
661 | } | |
662 | } | |
83cd4fe2 | 663 | } |
057f3fad PZ |
664 | unlock: |
665 | rcu_read_unlock(); | |
83cd4fe2 VP |
666 | return cpu; |
667 | } | |
06d8308c TG |
668 | /* |
669 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
670 | * idle CPU then this timer might expire before the next timer event | |
671 | * which is scheduled to wake up that CPU. In case of a completely | |
672 | * idle system the next event might even be infinite time into the | |
673 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
674 | * leaves the inner idle loop so the newly added timer is taken into | |
675 | * account when the CPU goes back to idle and evaluates the timer | |
676 | * wheel for the next timer event. | |
677 | */ | |
1c20091e | 678 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
679 | { |
680 | struct rq *rq = cpu_rq(cpu); | |
681 | ||
682 | if (cpu == smp_processor_id()) | |
683 | return; | |
684 | ||
67b9ca70 | 685 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 686 | smp_send_reschedule(cpu); |
dfc68f29 AL |
687 | else |
688 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
689 | } |
690 | ||
c5bfece2 | 691 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 692 | { |
53c5fa16 FW |
693 | /* |
694 | * We just need the target to call irq_exit() and re-evaluate | |
695 | * the next tick. The nohz full kick at least implies that. | |
696 | * If needed we can still optimize that later with an | |
697 | * empty IRQ. | |
698 | */ | |
c5bfece2 | 699 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
700 | if (cpu != smp_processor_id() || |
701 | tick_nohz_tick_stopped()) | |
53c5fa16 | 702 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
703 | return true; |
704 | } | |
705 | ||
706 | return false; | |
707 | } | |
708 | ||
709 | void wake_up_nohz_cpu(int cpu) | |
710 | { | |
c5bfece2 | 711 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
712 | wake_up_idle_cpu(cpu); |
713 | } | |
714 | ||
ca38062e | 715 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 716 | { |
1c792db7 | 717 | int cpu = smp_processor_id(); |
873b4c65 VG |
718 | |
719 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
720 | return false; | |
721 | ||
722 | if (idle_cpu(cpu) && !need_resched()) | |
723 | return true; | |
724 | ||
725 | /* | |
726 | * We can't run Idle Load Balance on this CPU for this time so we | |
727 | * cancel it and clear NOHZ_BALANCE_KICK | |
728 | */ | |
729 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
730 | return false; | |
45bf76df IM |
731 | } |
732 | ||
3451d024 | 733 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 734 | |
ca38062e | 735 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 736 | { |
ca38062e | 737 | return false; |
2069dd75 PZ |
738 | } |
739 | ||
3451d024 | 740 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 741 | |
ce831b38 FW |
742 | #ifdef CONFIG_NO_HZ_FULL |
743 | bool sched_can_stop_tick(void) | |
744 | { | |
3882ec64 FW |
745 | /* |
746 | * More than one running task need preemption. | |
747 | * nr_running update is assumed to be visible | |
748 | * after IPI is sent from wakers. | |
749 | */ | |
541b8264 VK |
750 | if (this_rq()->nr_running > 1) |
751 | return false; | |
ce831b38 | 752 | |
541b8264 | 753 | return true; |
ce831b38 FW |
754 | } |
755 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 756 | |
029632fb | 757 | void sched_avg_update(struct rq *rq) |
18d95a28 | 758 | { |
e9e9250b PZ |
759 | s64 period = sched_avg_period(); |
760 | ||
78becc27 | 761 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
762 | /* |
763 | * Inline assembly required to prevent the compiler | |
764 | * optimising this loop into a divmod call. | |
765 | * See __iter_div_u64_rem() for another example of this. | |
766 | */ | |
767 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
768 | rq->age_stamp += period; |
769 | rq->rt_avg /= 2; | |
770 | } | |
18d95a28 PZ |
771 | } |
772 | ||
6d6bc0ad | 773 | #endif /* CONFIG_SMP */ |
18d95a28 | 774 | |
a790de99 PT |
775 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
776 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 777 | /* |
8277434e PT |
778 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
779 | * node and @up when leaving it for the final time. | |
780 | * | |
781 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 782 | */ |
029632fb | 783 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 784 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
785 | { |
786 | struct task_group *parent, *child; | |
eb755805 | 787 | int ret; |
c09595f6 | 788 | |
8277434e PT |
789 | parent = from; |
790 | ||
c09595f6 | 791 | down: |
eb755805 PZ |
792 | ret = (*down)(parent, data); |
793 | if (ret) | |
8277434e | 794 | goto out; |
c09595f6 PZ |
795 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
796 | parent = child; | |
797 | goto down; | |
798 | ||
799 | up: | |
800 | continue; | |
801 | } | |
eb755805 | 802 | ret = (*up)(parent, data); |
8277434e PT |
803 | if (ret || parent == from) |
804 | goto out; | |
c09595f6 PZ |
805 | |
806 | child = parent; | |
807 | parent = parent->parent; | |
808 | if (parent) | |
809 | goto up; | |
8277434e | 810 | out: |
eb755805 | 811 | return ret; |
c09595f6 PZ |
812 | } |
813 | ||
029632fb | 814 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 815 | { |
e2b245f8 | 816 | return 0; |
eb755805 | 817 | } |
18d95a28 PZ |
818 | #endif |
819 | ||
45bf76df IM |
820 | static void set_load_weight(struct task_struct *p) |
821 | { | |
f05998d4 NR |
822 | int prio = p->static_prio - MAX_RT_PRIO; |
823 | struct load_weight *load = &p->se.load; | |
824 | ||
dd41f596 IM |
825 | /* |
826 | * SCHED_IDLE tasks get minimal weight: | |
827 | */ | |
828 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 829 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 830 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
831 | return; |
832 | } | |
71f8bd46 | 833 | |
c8b28116 | 834 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 835 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
836 | } |
837 | ||
371fd7e7 | 838 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 839 | { |
a64692a3 | 840 | update_rq_clock(rq); |
43148951 | 841 | sched_info_queued(rq, p); |
371fd7e7 | 842 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
843 | } |
844 | ||
371fd7e7 | 845 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 846 | { |
a64692a3 | 847 | update_rq_clock(rq); |
43148951 | 848 | sched_info_dequeued(rq, p); |
371fd7e7 | 849 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
850 | } |
851 | ||
029632fb | 852 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
853 | { |
854 | if (task_contributes_to_load(p)) | |
855 | rq->nr_uninterruptible--; | |
856 | ||
371fd7e7 | 857 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
858 | } |
859 | ||
029632fb | 860 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
861 | { |
862 | if (task_contributes_to_load(p)) | |
863 | rq->nr_uninterruptible++; | |
864 | ||
371fd7e7 | 865 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
866 | } |
867 | ||
fe44d621 | 868 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 869 | { |
095c0aa8 GC |
870 | /* |
871 | * In theory, the compile should just see 0 here, and optimize out the call | |
872 | * to sched_rt_avg_update. But I don't trust it... | |
873 | */ | |
874 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
875 | s64 steal = 0, irq_delta = 0; | |
876 | #endif | |
877 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 878 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
879 | |
880 | /* | |
881 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
882 | * this case when a previous update_rq_clock() happened inside a | |
883 | * {soft,}irq region. | |
884 | * | |
885 | * When this happens, we stop ->clock_task and only update the | |
886 | * prev_irq_time stamp to account for the part that fit, so that a next | |
887 | * update will consume the rest. This ensures ->clock_task is | |
888 | * monotonic. | |
889 | * | |
890 | * It does however cause some slight miss-attribution of {soft,}irq | |
891 | * time, a more accurate solution would be to update the irq_time using | |
892 | * the current rq->clock timestamp, except that would require using | |
893 | * atomic ops. | |
894 | */ | |
895 | if (irq_delta > delta) | |
896 | irq_delta = delta; | |
897 | ||
898 | rq->prev_irq_time += irq_delta; | |
899 | delta -= irq_delta; | |
095c0aa8 GC |
900 | #endif |
901 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 902 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
903 | steal = paravirt_steal_clock(cpu_of(rq)); |
904 | steal -= rq->prev_steal_time_rq; | |
905 | ||
906 | if (unlikely(steal > delta)) | |
907 | steal = delta; | |
908 | ||
095c0aa8 | 909 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
910 | delta -= steal; |
911 | } | |
912 | #endif | |
913 | ||
fe44d621 PZ |
914 | rq->clock_task += delta; |
915 | ||
095c0aa8 | 916 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 917 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
918 | sched_rt_avg_update(rq, irq_delta + steal); |
919 | #endif | |
aa483808 VP |
920 | } |
921 | ||
34f971f6 PZ |
922 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
923 | { | |
924 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
925 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
926 | ||
927 | if (stop) { | |
928 | /* | |
929 | * Make it appear like a SCHED_FIFO task, its something | |
930 | * userspace knows about and won't get confused about. | |
931 | * | |
932 | * Also, it will make PI more or less work without too | |
933 | * much confusion -- but then, stop work should not | |
934 | * rely on PI working anyway. | |
935 | */ | |
936 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
937 | ||
938 | stop->sched_class = &stop_sched_class; | |
939 | } | |
940 | ||
941 | cpu_rq(cpu)->stop = stop; | |
942 | ||
943 | if (old_stop) { | |
944 | /* | |
945 | * Reset it back to a normal scheduling class so that | |
946 | * it can die in pieces. | |
947 | */ | |
948 | old_stop->sched_class = &rt_sched_class; | |
949 | } | |
950 | } | |
951 | ||
14531189 | 952 | /* |
dd41f596 | 953 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 954 | */ |
14531189 IM |
955 | static inline int __normal_prio(struct task_struct *p) |
956 | { | |
dd41f596 | 957 | return p->static_prio; |
14531189 IM |
958 | } |
959 | ||
b29739f9 IM |
960 | /* |
961 | * Calculate the expected normal priority: i.e. priority | |
962 | * without taking RT-inheritance into account. Might be | |
963 | * boosted by interactivity modifiers. Changes upon fork, | |
964 | * setprio syscalls, and whenever the interactivity | |
965 | * estimator recalculates. | |
966 | */ | |
36c8b586 | 967 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
968 | { |
969 | int prio; | |
970 | ||
aab03e05 DF |
971 | if (task_has_dl_policy(p)) |
972 | prio = MAX_DL_PRIO-1; | |
973 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
974 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
975 | else | |
976 | prio = __normal_prio(p); | |
977 | return prio; | |
978 | } | |
979 | ||
980 | /* | |
981 | * Calculate the current priority, i.e. the priority | |
982 | * taken into account by the scheduler. This value might | |
983 | * be boosted by RT tasks, or might be boosted by | |
984 | * interactivity modifiers. Will be RT if the task got | |
985 | * RT-boosted. If not then it returns p->normal_prio. | |
986 | */ | |
36c8b586 | 987 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
988 | { |
989 | p->normal_prio = normal_prio(p); | |
990 | /* | |
991 | * If we are RT tasks or we were boosted to RT priority, | |
992 | * keep the priority unchanged. Otherwise, update priority | |
993 | * to the normal priority: | |
994 | */ | |
995 | if (!rt_prio(p->prio)) | |
996 | return p->normal_prio; | |
997 | return p->prio; | |
998 | } | |
999 | ||
1da177e4 LT |
1000 | /** |
1001 | * task_curr - is this task currently executing on a CPU? | |
1002 | * @p: the task in question. | |
e69f6186 YB |
1003 | * |
1004 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1005 | */ |
36c8b586 | 1006 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1007 | { |
1008 | return cpu_curr(task_cpu(p)) == p; | |
1009 | } | |
1010 | ||
cb469845 SR |
1011 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1012 | const struct sched_class *prev_class, | |
da7a735e | 1013 | int oldprio) |
cb469845 SR |
1014 | { |
1015 | if (prev_class != p->sched_class) { | |
1016 | if (prev_class->switched_from) | |
da7a735e PZ |
1017 | prev_class->switched_from(rq, p); |
1018 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 1019 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1020 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1021 | } |
1022 | ||
029632fb | 1023 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1024 | { |
1025 | const struct sched_class *class; | |
1026 | ||
1027 | if (p->sched_class == rq->curr->sched_class) { | |
1028 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1029 | } else { | |
1030 | for_each_class(class) { | |
1031 | if (class == rq->curr->sched_class) | |
1032 | break; | |
1033 | if (class == p->sched_class) { | |
8875125e | 1034 | resched_curr(rq); |
1e5a7405 PZ |
1035 | break; |
1036 | } | |
1037 | } | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * A queue event has occurred, and we're going to schedule. In | |
1042 | * this case, we can save a useless back to back clock update. | |
1043 | */ | |
da0c1e65 | 1044 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1045 | rq->skip_clock_update = 1; |
1046 | } | |
1047 | ||
1da177e4 | 1048 | #ifdef CONFIG_SMP |
dd41f596 | 1049 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1050 | { |
e2912009 PZ |
1051 | #ifdef CONFIG_SCHED_DEBUG |
1052 | /* | |
1053 | * We should never call set_task_cpu() on a blocked task, | |
1054 | * ttwu() will sort out the placement. | |
1055 | */ | |
077614ee | 1056 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1057 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1058 | |
1059 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1060 | /* |
1061 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1062 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1063 | * | |
1064 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1065 | * see task_group(). |
6c6c54e1 PZ |
1066 | * |
1067 | * Furthermore, all task_rq users should acquire both locks, see | |
1068 | * task_rq_lock(). | |
1069 | */ | |
0122ec5b PZ |
1070 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1071 | lockdep_is_held(&task_rq(p)->lock))); | |
1072 | #endif | |
e2912009 PZ |
1073 | #endif |
1074 | ||
de1d7286 | 1075 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1076 | |
0c69774e | 1077 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1078 | if (p->sched_class->migrate_task_rq) |
1079 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1080 | p->se.nr_migrations++; |
a8b0ca17 | 1081 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1082 | } |
dd41f596 IM |
1083 | |
1084 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1085 | } |
1086 | ||
ac66f547 PZ |
1087 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1088 | { | |
da0c1e65 | 1089 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1090 | struct rq *src_rq, *dst_rq; |
1091 | ||
1092 | src_rq = task_rq(p); | |
1093 | dst_rq = cpu_rq(cpu); | |
1094 | ||
1095 | deactivate_task(src_rq, p, 0); | |
1096 | set_task_cpu(p, cpu); | |
1097 | activate_task(dst_rq, p, 0); | |
1098 | check_preempt_curr(dst_rq, p, 0); | |
1099 | } else { | |
1100 | /* | |
1101 | * Task isn't running anymore; make it appear like we migrated | |
1102 | * it before it went to sleep. This means on wakeup we make the | |
1103 | * previous cpu our targer instead of where it really is. | |
1104 | */ | |
1105 | p->wake_cpu = cpu; | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | struct migration_swap_arg { | |
1110 | struct task_struct *src_task, *dst_task; | |
1111 | int src_cpu, dst_cpu; | |
1112 | }; | |
1113 | ||
1114 | static int migrate_swap_stop(void *data) | |
1115 | { | |
1116 | struct migration_swap_arg *arg = data; | |
1117 | struct rq *src_rq, *dst_rq; | |
1118 | int ret = -EAGAIN; | |
1119 | ||
1120 | src_rq = cpu_rq(arg->src_cpu); | |
1121 | dst_rq = cpu_rq(arg->dst_cpu); | |
1122 | ||
74602315 PZ |
1123 | double_raw_lock(&arg->src_task->pi_lock, |
1124 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1125 | double_rq_lock(src_rq, dst_rq); |
1126 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1127 | goto unlock; | |
1128 | ||
1129 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1130 | goto unlock; | |
1131 | ||
1132 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1133 | goto unlock; | |
1134 | ||
1135 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1136 | goto unlock; | |
1137 | ||
1138 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1139 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1140 | ||
1141 | ret = 0; | |
1142 | ||
1143 | unlock: | |
1144 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1145 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1146 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1147 | |
1148 | return ret; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * Cross migrate two tasks | |
1153 | */ | |
1154 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1155 | { | |
1156 | struct migration_swap_arg arg; | |
1157 | int ret = -EINVAL; | |
1158 | ||
ac66f547 PZ |
1159 | arg = (struct migration_swap_arg){ |
1160 | .src_task = cur, | |
1161 | .src_cpu = task_cpu(cur), | |
1162 | .dst_task = p, | |
1163 | .dst_cpu = task_cpu(p), | |
1164 | }; | |
1165 | ||
1166 | if (arg.src_cpu == arg.dst_cpu) | |
1167 | goto out; | |
1168 | ||
6acce3ef PZ |
1169 | /* |
1170 | * These three tests are all lockless; this is OK since all of them | |
1171 | * will be re-checked with proper locks held further down the line. | |
1172 | */ | |
ac66f547 PZ |
1173 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1174 | goto out; | |
1175 | ||
1176 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1177 | goto out; | |
1178 | ||
1179 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1180 | goto out; | |
1181 | ||
286549dc | 1182 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1183 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1184 | ||
1185 | out: | |
ac66f547 PZ |
1186 | return ret; |
1187 | } | |
1188 | ||
969c7921 | 1189 | struct migration_arg { |
36c8b586 | 1190 | struct task_struct *task; |
1da177e4 | 1191 | int dest_cpu; |
70b97a7f | 1192 | }; |
1da177e4 | 1193 | |
969c7921 TH |
1194 | static int migration_cpu_stop(void *data); |
1195 | ||
1da177e4 LT |
1196 | /* |
1197 | * wait_task_inactive - wait for a thread to unschedule. | |
1198 | * | |
85ba2d86 RM |
1199 | * If @match_state is nonzero, it's the @p->state value just checked and |
1200 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1201 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1202 | * we return a positive number (its total switch count). If a second call | |
1203 | * a short while later returns the same number, the caller can be sure that | |
1204 | * @p has remained unscheduled the whole time. | |
1205 | * | |
1da177e4 LT |
1206 | * The caller must ensure that the task *will* unschedule sometime soon, |
1207 | * else this function might spin for a *long* time. This function can't | |
1208 | * be called with interrupts off, or it may introduce deadlock with | |
1209 | * smp_call_function() if an IPI is sent by the same process we are | |
1210 | * waiting to become inactive. | |
1211 | */ | |
85ba2d86 | 1212 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1213 | { |
1214 | unsigned long flags; | |
da0c1e65 | 1215 | int running, queued; |
85ba2d86 | 1216 | unsigned long ncsw; |
70b97a7f | 1217 | struct rq *rq; |
1da177e4 | 1218 | |
3a5c359a AK |
1219 | for (;;) { |
1220 | /* | |
1221 | * We do the initial early heuristics without holding | |
1222 | * any task-queue locks at all. We'll only try to get | |
1223 | * the runqueue lock when things look like they will | |
1224 | * work out! | |
1225 | */ | |
1226 | rq = task_rq(p); | |
fa490cfd | 1227 | |
3a5c359a AK |
1228 | /* |
1229 | * If the task is actively running on another CPU | |
1230 | * still, just relax and busy-wait without holding | |
1231 | * any locks. | |
1232 | * | |
1233 | * NOTE! Since we don't hold any locks, it's not | |
1234 | * even sure that "rq" stays as the right runqueue! | |
1235 | * But we don't care, since "task_running()" will | |
1236 | * return false if the runqueue has changed and p | |
1237 | * is actually now running somewhere else! | |
1238 | */ | |
85ba2d86 RM |
1239 | while (task_running(rq, p)) { |
1240 | if (match_state && unlikely(p->state != match_state)) | |
1241 | return 0; | |
3a5c359a | 1242 | cpu_relax(); |
85ba2d86 | 1243 | } |
fa490cfd | 1244 | |
3a5c359a AK |
1245 | /* |
1246 | * Ok, time to look more closely! We need the rq | |
1247 | * lock now, to be *sure*. If we're wrong, we'll | |
1248 | * just go back and repeat. | |
1249 | */ | |
1250 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1251 | trace_sched_wait_task(p); |
3a5c359a | 1252 | running = task_running(rq, p); |
da0c1e65 | 1253 | queued = task_on_rq_queued(p); |
85ba2d86 | 1254 | ncsw = 0; |
f31e11d8 | 1255 | if (!match_state || p->state == match_state) |
93dcf55f | 1256 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1257 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1258 | |
85ba2d86 RM |
1259 | /* |
1260 | * If it changed from the expected state, bail out now. | |
1261 | */ | |
1262 | if (unlikely(!ncsw)) | |
1263 | break; | |
1264 | ||
3a5c359a AK |
1265 | /* |
1266 | * Was it really running after all now that we | |
1267 | * checked with the proper locks actually held? | |
1268 | * | |
1269 | * Oops. Go back and try again.. | |
1270 | */ | |
1271 | if (unlikely(running)) { | |
1272 | cpu_relax(); | |
1273 | continue; | |
1274 | } | |
fa490cfd | 1275 | |
3a5c359a AK |
1276 | /* |
1277 | * It's not enough that it's not actively running, | |
1278 | * it must be off the runqueue _entirely_, and not | |
1279 | * preempted! | |
1280 | * | |
80dd99b3 | 1281 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1282 | * running right now), it's preempted, and we should |
1283 | * yield - it could be a while. | |
1284 | */ | |
da0c1e65 | 1285 | if (unlikely(queued)) { |
8eb90c30 TG |
1286 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1287 | ||
1288 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1289 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1290 | continue; |
1291 | } | |
fa490cfd | 1292 | |
3a5c359a AK |
1293 | /* |
1294 | * Ahh, all good. It wasn't running, and it wasn't | |
1295 | * runnable, which means that it will never become | |
1296 | * running in the future either. We're all done! | |
1297 | */ | |
1298 | break; | |
1299 | } | |
85ba2d86 RM |
1300 | |
1301 | return ncsw; | |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | /*** | |
1305 | * kick_process - kick a running thread to enter/exit the kernel | |
1306 | * @p: the to-be-kicked thread | |
1307 | * | |
1308 | * Cause a process which is running on another CPU to enter | |
1309 | * kernel-mode, without any delay. (to get signals handled.) | |
1310 | * | |
25985edc | 1311 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1312 | * because all it wants to ensure is that the remote task enters |
1313 | * the kernel. If the IPI races and the task has been migrated | |
1314 | * to another CPU then no harm is done and the purpose has been | |
1315 | * achieved as well. | |
1316 | */ | |
36c8b586 | 1317 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1318 | { |
1319 | int cpu; | |
1320 | ||
1321 | preempt_disable(); | |
1322 | cpu = task_cpu(p); | |
1323 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1324 | smp_send_reschedule(cpu); | |
1325 | preempt_enable(); | |
1326 | } | |
b43e3521 | 1327 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1328 | #endif /* CONFIG_SMP */ |
1da177e4 | 1329 | |
970b13ba | 1330 | #ifdef CONFIG_SMP |
30da688e | 1331 | /* |
013fdb80 | 1332 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1333 | */ |
5da9a0fb PZ |
1334 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1335 | { | |
aa00d89c TC |
1336 | int nid = cpu_to_node(cpu); |
1337 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1338 | enum { cpuset, possible, fail } state = cpuset; |
1339 | int dest_cpu; | |
5da9a0fb | 1340 | |
aa00d89c TC |
1341 | /* |
1342 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1343 | * will return -1. There is no cpu on the node, and we should | |
1344 | * select the cpu on the other node. | |
1345 | */ | |
1346 | if (nid != -1) { | |
1347 | nodemask = cpumask_of_node(nid); | |
1348 | ||
1349 | /* Look for allowed, online CPU in same node. */ | |
1350 | for_each_cpu(dest_cpu, nodemask) { | |
1351 | if (!cpu_online(dest_cpu)) | |
1352 | continue; | |
1353 | if (!cpu_active(dest_cpu)) | |
1354 | continue; | |
1355 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1356 | return dest_cpu; | |
1357 | } | |
2baab4e9 | 1358 | } |
5da9a0fb | 1359 | |
2baab4e9 PZ |
1360 | for (;;) { |
1361 | /* Any allowed, online CPU? */ | |
e3831edd | 1362 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1363 | if (!cpu_online(dest_cpu)) |
1364 | continue; | |
1365 | if (!cpu_active(dest_cpu)) | |
1366 | continue; | |
1367 | goto out; | |
1368 | } | |
5da9a0fb | 1369 | |
2baab4e9 PZ |
1370 | switch (state) { |
1371 | case cpuset: | |
1372 | /* No more Mr. Nice Guy. */ | |
1373 | cpuset_cpus_allowed_fallback(p); | |
1374 | state = possible; | |
1375 | break; | |
1376 | ||
1377 | case possible: | |
1378 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1379 | state = fail; | |
1380 | break; | |
1381 | ||
1382 | case fail: | |
1383 | BUG(); | |
1384 | break; | |
1385 | } | |
1386 | } | |
1387 | ||
1388 | out: | |
1389 | if (state != cpuset) { | |
1390 | /* | |
1391 | * Don't tell them about moving exiting tasks or | |
1392 | * kernel threads (both mm NULL), since they never | |
1393 | * leave kernel. | |
1394 | */ | |
1395 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1396 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1397 | task_pid_nr(p), p->comm, cpu); |
1398 | } | |
5da9a0fb PZ |
1399 | } |
1400 | ||
1401 | return dest_cpu; | |
1402 | } | |
1403 | ||
e2912009 | 1404 | /* |
013fdb80 | 1405 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1406 | */ |
970b13ba | 1407 | static inline |
ac66f547 | 1408 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1409 | { |
ac66f547 | 1410 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1411 | |
1412 | /* | |
1413 | * In order not to call set_task_cpu() on a blocking task we need | |
1414 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1415 | * cpu. | |
1416 | * | |
1417 | * Since this is common to all placement strategies, this lives here. | |
1418 | * | |
1419 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1420 | * not worry about this generic constraint ] | |
1421 | */ | |
fa17b507 | 1422 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1423 | !cpu_online(cpu))) |
5da9a0fb | 1424 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1425 | |
1426 | return cpu; | |
970b13ba | 1427 | } |
09a40af5 MG |
1428 | |
1429 | static void update_avg(u64 *avg, u64 sample) | |
1430 | { | |
1431 | s64 diff = sample - *avg; | |
1432 | *avg += diff >> 3; | |
1433 | } | |
970b13ba PZ |
1434 | #endif |
1435 | ||
d7c01d27 | 1436 | static void |
b84cb5df | 1437 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1438 | { |
d7c01d27 | 1439 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1440 | struct rq *rq = this_rq(); |
1441 | ||
d7c01d27 PZ |
1442 | #ifdef CONFIG_SMP |
1443 | int this_cpu = smp_processor_id(); | |
1444 | ||
1445 | if (cpu == this_cpu) { | |
1446 | schedstat_inc(rq, ttwu_local); | |
1447 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1448 | } else { | |
1449 | struct sched_domain *sd; | |
1450 | ||
1451 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1452 | rcu_read_lock(); |
d7c01d27 PZ |
1453 | for_each_domain(this_cpu, sd) { |
1454 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1455 | schedstat_inc(sd, ttwu_wake_remote); | |
1456 | break; | |
1457 | } | |
1458 | } | |
057f3fad | 1459 | rcu_read_unlock(); |
d7c01d27 | 1460 | } |
f339b9dc PZ |
1461 | |
1462 | if (wake_flags & WF_MIGRATED) | |
1463 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1464 | ||
d7c01d27 PZ |
1465 | #endif /* CONFIG_SMP */ |
1466 | ||
1467 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1468 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1469 | |
1470 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1471 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1472 | |
d7c01d27 PZ |
1473 | #endif /* CONFIG_SCHEDSTATS */ |
1474 | } | |
1475 | ||
1476 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1477 | { | |
9ed3811a | 1478 | activate_task(rq, p, en_flags); |
da0c1e65 | 1479 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1480 | |
1481 | /* if a worker is waking up, notify workqueue */ | |
1482 | if (p->flags & PF_WQ_WORKER) | |
1483 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1484 | } |
1485 | ||
23f41eeb PZ |
1486 | /* |
1487 | * Mark the task runnable and perform wakeup-preemption. | |
1488 | */ | |
89363381 | 1489 | static void |
23f41eeb | 1490 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1491 | { |
9ed3811a | 1492 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1493 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1494 | |
1495 | p->state = TASK_RUNNING; | |
1496 | #ifdef CONFIG_SMP | |
1497 | if (p->sched_class->task_woken) | |
1498 | p->sched_class->task_woken(rq, p); | |
1499 | ||
e69c6341 | 1500 | if (rq->idle_stamp) { |
78becc27 | 1501 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1502 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1503 | |
abfafa54 JL |
1504 | update_avg(&rq->avg_idle, delta); |
1505 | ||
1506 | if (rq->avg_idle > max) | |
9ed3811a | 1507 | rq->avg_idle = max; |
abfafa54 | 1508 | |
9ed3811a TH |
1509 | rq->idle_stamp = 0; |
1510 | } | |
1511 | #endif | |
1512 | } | |
1513 | ||
c05fbafb PZ |
1514 | static void |
1515 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1516 | { | |
1517 | #ifdef CONFIG_SMP | |
1518 | if (p->sched_contributes_to_load) | |
1519 | rq->nr_uninterruptible--; | |
1520 | #endif | |
1521 | ||
1522 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1523 | ttwu_do_wakeup(rq, p, wake_flags); | |
1524 | } | |
1525 | ||
1526 | /* | |
1527 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1528 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1529 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1530 | * the task is still ->on_rq. | |
1531 | */ | |
1532 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1533 | { | |
1534 | struct rq *rq; | |
1535 | int ret = 0; | |
1536 | ||
1537 | rq = __task_rq_lock(p); | |
da0c1e65 | 1538 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1539 | /* check_preempt_curr() may use rq clock */ |
1540 | update_rq_clock(rq); | |
c05fbafb PZ |
1541 | ttwu_do_wakeup(rq, p, wake_flags); |
1542 | ret = 1; | |
1543 | } | |
1544 | __task_rq_unlock(rq); | |
1545 | ||
1546 | return ret; | |
1547 | } | |
1548 | ||
317f3941 | 1549 | #ifdef CONFIG_SMP |
e3baac47 | 1550 | void sched_ttwu_pending(void) |
317f3941 PZ |
1551 | { |
1552 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1553 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1554 | struct task_struct *p; | |
e3baac47 | 1555 | unsigned long flags; |
317f3941 | 1556 | |
e3baac47 PZ |
1557 | if (!llist) |
1558 | return; | |
1559 | ||
1560 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1561 | |
fa14ff4a PZ |
1562 | while (llist) { |
1563 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1564 | llist = llist_next(llist); | |
317f3941 PZ |
1565 | ttwu_do_activate(rq, p, 0); |
1566 | } | |
1567 | ||
e3baac47 | 1568 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1569 | } |
1570 | ||
1571 | void scheduler_ipi(void) | |
1572 | { | |
f27dde8d PZ |
1573 | /* |
1574 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1575 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1576 | * this IPI. | |
1577 | */ | |
8cb75e0c | 1578 | preempt_fold_need_resched(); |
f27dde8d | 1579 | |
fd2ac4f4 | 1580 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1581 | return; |
1582 | ||
1583 | /* | |
1584 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1585 | * traditionally all their work was done from the interrupt return | |
1586 | * path. Now that we actually do some work, we need to make sure | |
1587 | * we do call them. | |
1588 | * | |
1589 | * Some archs already do call them, luckily irq_enter/exit nest | |
1590 | * properly. | |
1591 | * | |
1592 | * Arguably we should visit all archs and update all handlers, | |
1593 | * however a fair share of IPIs are still resched only so this would | |
1594 | * somewhat pessimize the simple resched case. | |
1595 | */ | |
1596 | irq_enter(); | |
fa14ff4a | 1597 | sched_ttwu_pending(); |
ca38062e SS |
1598 | |
1599 | /* | |
1600 | * Check if someone kicked us for doing the nohz idle load balance. | |
1601 | */ | |
873b4c65 | 1602 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1603 | this_rq()->idle_balance = 1; |
ca38062e | 1604 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1605 | } |
c5d753a5 | 1606 | irq_exit(); |
317f3941 PZ |
1607 | } |
1608 | ||
1609 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1610 | { | |
e3baac47 PZ |
1611 | struct rq *rq = cpu_rq(cpu); |
1612 | ||
1613 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1614 | if (!set_nr_if_polling(rq->idle)) | |
1615 | smp_send_reschedule(cpu); | |
1616 | else | |
1617 | trace_sched_wake_idle_without_ipi(cpu); | |
1618 | } | |
317f3941 | 1619 | } |
d6aa8f85 | 1620 | |
f6be8af1 CL |
1621 | void wake_up_if_idle(int cpu) |
1622 | { | |
1623 | struct rq *rq = cpu_rq(cpu); | |
1624 | unsigned long flags; | |
1625 | ||
1626 | if (!is_idle_task(rq->curr)) | |
1627 | return; | |
1628 | ||
1629 | if (set_nr_if_polling(rq->idle)) { | |
1630 | trace_sched_wake_idle_without_ipi(cpu); | |
1631 | } else { | |
1632 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1633 | if (is_idle_task(rq->curr)) | |
1634 | smp_send_reschedule(cpu); | |
1635 | /* Else cpu is not in idle, do nothing here */ | |
1636 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1637 | } | |
1638 | } | |
1639 | ||
39be3501 | 1640 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1641 | { |
1642 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1643 | } | |
d6aa8f85 | 1644 | #endif /* CONFIG_SMP */ |
317f3941 | 1645 | |
c05fbafb PZ |
1646 | static void ttwu_queue(struct task_struct *p, int cpu) |
1647 | { | |
1648 | struct rq *rq = cpu_rq(cpu); | |
1649 | ||
17d9f311 | 1650 | #if defined(CONFIG_SMP) |
39be3501 | 1651 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1652 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1653 | ttwu_queue_remote(p, cpu); |
1654 | return; | |
1655 | } | |
1656 | #endif | |
1657 | ||
c05fbafb PZ |
1658 | raw_spin_lock(&rq->lock); |
1659 | ttwu_do_activate(rq, p, 0); | |
1660 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1661 | } |
1662 | ||
1663 | /** | |
1da177e4 | 1664 | * try_to_wake_up - wake up a thread |
9ed3811a | 1665 | * @p: the thread to be awakened |
1da177e4 | 1666 | * @state: the mask of task states that can be woken |
9ed3811a | 1667 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1668 | * |
1669 | * Put it on the run-queue if it's not already there. The "current" | |
1670 | * thread is always on the run-queue (except when the actual | |
1671 | * re-schedule is in progress), and as such you're allowed to do | |
1672 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1673 | * runnable without the overhead of this. | |
1674 | * | |
e69f6186 | 1675 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1676 | * or @state didn't match @p's state. |
1da177e4 | 1677 | */ |
e4a52bcb PZ |
1678 | static int |
1679 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1680 | { |
1da177e4 | 1681 | unsigned long flags; |
c05fbafb | 1682 | int cpu, success = 0; |
2398f2c6 | 1683 | |
e0acd0a6 ON |
1684 | /* |
1685 | * If we are going to wake up a thread waiting for CONDITION we | |
1686 | * need to ensure that CONDITION=1 done by the caller can not be | |
1687 | * reordered with p->state check below. This pairs with mb() in | |
1688 | * set_current_state() the waiting thread does. | |
1689 | */ | |
1690 | smp_mb__before_spinlock(); | |
013fdb80 | 1691 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1692 | if (!(p->state & state)) |
1da177e4 LT |
1693 | goto out; |
1694 | ||
c05fbafb | 1695 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1696 | cpu = task_cpu(p); |
1da177e4 | 1697 | |
c05fbafb PZ |
1698 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1699 | goto stat; | |
1da177e4 | 1700 | |
1da177e4 | 1701 | #ifdef CONFIG_SMP |
e9c84311 | 1702 | /* |
c05fbafb PZ |
1703 | * If the owning (remote) cpu is still in the middle of schedule() with |
1704 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1705 | */ |
f3e94786 | 1706 | while (p->on_cpu) |
e4a52bcb | 1707 | cpu_relax(); |
0970d299 | 1708 | /* |
e4a52bcb | 1709 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1710 | */ |
e4a52bcb | 1711 | smp_rmb(); |
1da177e4 | 1712 | |
a8e4f2ea | 1713 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1714 | p->state = TASK_WAKING; |
e7693a36 | 1715 | |
e4a52bcb | 1716 | if (p->sched_class->task_waking) |
74f8e4b2 | 1717 | p->sched_class->task_waking(p); |
efbbd05a | 1718 | |
ac66f547 | 1719 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1720 | if (task_cpu(p) != cpu) { |
1721 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1722 | set_task_cpu(p, cpu); |
f339b9dc | 1723 | } |
1da177e4 | 1724 | #endif /* CONFIG_SMP */ |
1da177e4 | 1725 | |
c05fbafb PZ |
1726 | ttwu_queue(p, cpu); |
1727 | stat: | |
b84cb5df | 1728 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1729 | out: |
013fdb80 | 1730 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1731 | |
1732 | return success; | |
1733 | } | |
1734 | ||
21aa9af0 TH |
1735 | /** |
1736 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1737 | * @p: the thread to be awakened | |
1738 | * | |
2acca55e | 1739 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1740 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1741 | * the current task. |
21aa9af0 TH |
1742 | */ |
1743 | static void try_to_wake_up_local(struct task_struct *p) | |
1744 | { | |
1745 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1746 | |
383efcd0 TH |
1747 | if (WARN_ON_ONCE(rq != this_rq()) || |
1748 | WARN_ON_ONCE(p == current)) | |
1749 | return; | |
1750 | ||
21aa9af0 TH |
1751 | lockdep_assert_held(&rq->lock); |
1752 | ||
2acca55e PZ |
1753 | if (!raw_spin_trylock(&p->pi_lock)) { |
1754 | raw_spin_unlock(&rq->lock); | |
1755 | raw_spin_lock(&p->pi_lock); | |
1756 | raw_spin_lock(&rq->lock); | |
1757 | } | |
1758 | ||
21aa9af0 | 1759 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1760 | goto out; |
21aa9af0 | 1761 | |
da0c1e65 | 1762 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
1763 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1764 | ||
23f41eeb | 1765 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1766 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1767 | out: |
1768 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1769 | } |
1770 | ||
50fa610a DH |
1771 | /** |
1772 | * wake_up_process - Wake up a specific process | |
1773 | * @p: The process to be woken up. | |
1774 | * | |
1775 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1776 | * processes. |
1777 | * | |
1778 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1779 | * |
1780 | * It may be assumed that this function implies a write memory barrier before | |
1781 | * changing the task state if and only if any tasks are woken up. | |
1782 | */ | |
7ad5b3a5 | 1783 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1784 | { |
9067ac85 ON |
1785 | WARN_ON(task_is_stopped_or_traced(p)); |
1786 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1787 | } |
1da177e4 LT |
1788 | EXPORT_SYMBOL(wake_up_process); |
1789 | ||
7ad5b3a5 | 1790 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1791 | { |
1792 | return try_to_wake_up(p, state, 0); | |
1793 | } | |
1794 | ||
a5e7be3b JL |
1795 | /* |
1796 | * This function clears the sched_dl_entity static params. | |
1797 | */ | |
1798 | void __dl_clear_params(struct task_struct *p) | |
1799 | { | |
1800 | struct sched_dl_entity *dl_se = &p->dl; | |
1801 | ||
1802 | dl_se->dl_runtime = 0; | |
1803 | dl_se->dl_deadline = 0; | |
1804 | dl_se->dl_period = 0; | |
1805 | dl_se->flags = 0; | |
1806 | dl_se->dl_bw = 0; | |
1807 | } | |
1808 | ||
1da177e4 LT |
1809 | /* |
1810 | * Perform scheduler related setup for a newly forked process p. | |
1811 | * p is forked by current. | |
dd41f596 IM |
1812 | * |
1813 | * __sched_fork() is basic setup used by init_idle() too: | |
1814 | */ | |
5e1576ed | 1815 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1816 | { |
fd2f4419 PZ |
1817 | p->on_rq = 0; |
1818 | ||
1819 | p->se.on_rq = 0; | |
dd41f596 IM |
1820 | p->se.exec_start = 0; |
1821 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1822 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1823 | p->se.nr_migrations = 0; |
da7a735e | 1824 | p->se.vruntime = 0; |
fd2f4419 | 1825 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1826 | |
1827 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1828 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1829 | #endif |
476d139c | 1830 | |
aab03e05 DF |
1831 | RB_CLEAR_NODE(&p->dl.rb_node); |
1832 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
a5e7be3b | 1833 | __dl_clear_params(p); |
aab03e05 | 1834 | |
fa717060 | 1835 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1836 | |
e107be36 AK |
1837 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1838 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1839 | #endif | |
cbee9f88 PZ |
1840 | |
1841 | #ifdef CONFIG_NUMA_BALANCING | |
1842 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1843 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1844 | p->mm->numa_scan_seq = 0; |
1845 | } | |
1846 | ||
5e1576ed RR |
1847 | if (clone_flags & CLONE_VM) |
1848 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1849 | else | |
1850 | p->numa_preferred_nid = -1; | |
1851 | ||
cbee9f88 PZ |
1852 | p->node_stamp = 0ULL; |
1853 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1854 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1855 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1856 | p->numa_faults_memory = NULL; |
1857 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1858 | p->last_task_numa_placement = 0; |
1859 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1860 | |
1861 | INIT_LIST_HEAD(&p->numa_entry); | |
1862 | p->numa_group = NULL; | |
cbee9f88 | 1863 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1864 | } |
1865 | ||
1a687c2e | 1866 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1867 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1868 | void set_numabalancing_state(bool enabled) |
1869 | { | |
1870 | if (enabled) | |
1871 | sched_feat_set("NUMA"); | |
1872 | else | |
1873 | sched_feat_set("NO_NUMA"); | |
1874 | } | |
3105b86a MG |
1875 | #else |
1876 | __read_mostly bool numabalancing_enabled; | |
1877 | ||
1878 | void set_numabalancing_state(bool enabled) | |
1879 | { | |
1880 | numabalancing_enabled = enabled; | |
dd41f596 | 1881 | } |
3105b86a | 1882 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1883 | |
1884 | #ifdef CONFIG_PROC_SYSCTL | |
1885 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1886 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1887 | { | |
1888 | struct ctl_table t; | |
1889 | int err; | |
1890 | int state = numabalancing_enabled; | |
1891 | ||
1892 | if (write && !capable(CAP_SYS_ADMIN)) | |
1893 | return -EPERM; | |
1894 | ||
1895 | t = *table; | |
1896 | t.data = &state; | |
1897 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1898 | if (err < 0) | |
1899 | return err; | |
1900 | if (write) | |
1901 | set_numabalancing_state(state); | |
1902 | return err; | |
1903 | } | |
1904 | #endif | |
1905 | #endif | |
dd41f596 IM |
1906 | |
1907 | /* | |
1908 | * fork()/clone()-time setup: | |
1909 | */ | |
aab03e05 | 1910 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1911 | { |
0122ec5b | 1912 | unsigned long flags; |
dd41f596 IM |
1913 | int cpu = get_cpu(); |
1914 | ||
5e1576ed | 1915 | __sched_fork(clone_flags, p); |
06b83b5f | 1916 | /* |
0017d735 | 1917 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1918 | * nobody will actually run it, and a signal or other external |
1919 | * event cannot wake it up and insert it on the runqueue either. | |
1920 | */ | |
0017d735 | 1921 | p->state = TASK_RUNNING; |
dd41f596 | 1922 | |
c350a04e MG |
1923 | /* |
1924 | * Make sure we do not leak PI boosting priority to the child. | |
1925 | */ | |
1926 | p->prio = current->normal_prio; | |
1927 | ||
b9dc29e7 MG |
1928 | /* |
1929 | * Revert to default priority/policy on fork if requested. | |
1930 | */ | |
1931 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1932 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1933 | p->policy = SCHED_NORMAL; |
6c697bdf | 1934 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1935 | p->rt_priority = 0; |
1936 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1937 | p->static_prio = NICE_TO_PRIO(0); | |
1938 | ||
1939 | p->prio = p->normal_prio = __normal_prio(p); | |
1940 | set_load_weight(p); | |
6c697bdf | 1941 | |
b9dc29e7 MG |
1942 | /* |
1943 | * We don't need the reset flag anymore after the fork. It has | |
1944 | * fulfilled its duty: | |
1945 | */ | |
1946 | p->sched_reset_on_fork = 0; | |
1947 | } | |
ca94c442 | 1948 | |
aab03e05 DF |
1949 | if (dl_prio(p->prio)) { |
1950 | put_cpu(); | |
1951 | return -EAGAIN; | |
1952 | } else if (rt_prio(p->prio)) { | |
1953 | p->sched_class = &rt_sched_class; | |
1954 | } else { | |
2ddbf952 | 1955 | p->sched_class = &fair_sched_class; |
aab03e05 | 1956 | } |
b29739f9 | 1957 | |
cd29fe6f PZ |
1958 | if (p->sched_class->task_fork) |
1959 | p->sched_class->task_fork(p); | |
1960 | ||
86951599 PZ |
1961 | /* |
1962 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1963 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1964 | * is ran before sched_fork(). | |
1965 | * | |
1966 | * Silence PROVE_RCU. | |
1967 | */ | |
0122ec5b | 1968 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1969 | set_task_cpu(p, cpu); |
0122ec5b | 1970 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1971 | |
52f17b6c | 1972 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1973 | if (likely(sched_info_on())) |
52f17b6c | 1974 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1975 | #endif |
3ca7a440 PZ |
1976 | #if defined(CONFIG_SMP) |
1977 | p->on_cpu = 0; | |
4866cde0 | 1978 | #endif |
01028747 | 1979 | init_task_preempt_count(p); |
806c09a7 | 1980 | #ifdef CONFIG_SMP |
917b627d | 1981 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1982 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1983 | #endif |
917b627d | 1984 | |
476d139c | 1985 | put_cpu(); |
aab03e05 | 1986 | return 0; |
1da177e4 LT |
1987 | } |
1988 | ||
332ac17e DF |
1989 | unsigned long to_ratio(u64 period, u64 runtime) |
1990 | { | |
1991 | if (runtime == RUNTIME_INF) | |
1992 | return 1ULL << 20; | |
1993 | ||
1994 | /* | |
1995 | * Doing this here saves a lot of checks in all | |
1996 | * the calling paths, and returning zero seems | |
1997 | * safe for them anyway. | |
1998 | */ | |
1999 | if (period == 0) | |
2000 | return 0; | |
2001 | ||
2002 | return div64_u64(runtime << 20, period); | |
2003 | } | |
2004 | ||
2005 | #ifdef CONFIG_SMP | |
2006 | inline struct dl_bw *dl_bw_of(int i) | |
2007 | { | |
66339c31 KT |
2008 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2009 | "sched RCU must be held"); | |
332ac17e DF |
2010 | return &cpu_rq(i)->rd->dl_bw; |
2011 | } | |
2012 | ||
de212f18 | 2013 | static inline int dl_bw_cpus(int i) |
332ac17e | 2014 | { |
de212f18 PZ |
2015 | struct root_domain *rd = cpu_rq(i)->rd; |
2016 | int cpus = 0; | |
2017 | ||
66339c31 KT |
2018 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2019 | "sched RCU must be held"); | |
de212f18 PZ |
2020 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2021 | cpus++; | |
2022 | ||
2023 | return cpus; | |
332ac17e DF |
2024 | } |
2025 | #else | |
2026 | inline struct dl_bw *dl_bw_of(int i) | |
2027 | { | |
2028 | return &cpu_rq(i)->dl.dl_bw; | |
2029 | } | |
2030 | ||
de212f18 | 2031 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2032 | { |
2033 | return 1; | |
2034 | } | |
2035 | #endif | |
2036 | ||
2037 | static inline | |
2038 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
2039 | { | |
2040 | dl_b->total_bw -= tsk_bw; | |
2041 | } | |
2042 | ||
2043 | static inline | |
2044 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
2045 | { | |
2046 | dl_b->total_bw += tsk_bw; | |
2047 | } | |
2048 | ||
2049 | static inline | |
2050 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
2051 | { | |
2052 | return dl_b->bw != -1 && | |
2053 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * We must be sure that accepting a new task (or allowing changing the | |
2058 | * parameters of an existing one) is consistent with the bandwidth | |
2059 | * constraints. If yes, this function also accordingly updates the currently | |
2060 | * allocated bandwidth to reflect the new situation. | |
2061 | * | |
2062 | * This function is called while holding p's rq->lock. | |
2063 | */ | |
2064 | static int dl_overflow(struct task_struct *p, int policy, | |
2065 | const struct sched_attr *attr) | |
2066 | { | |
2067 | ||
2068 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2069 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2070 | u64 runtime = attr->sched_runtime; |
2071 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2072 | int cpus, err = -1; |
332ac17e DF |
2073 | |
2074 | if (new_bw == p->dl.dl_bw) | |
2075 | return 0; | |
2076 | ||
2077 | /* | |
2078 | * Either if a task, enters, leave, or stays -deadline but changes | |
2079 | * its parameters, we may need to update accordingly the total | |
2080 | * allocated bandwidth of the container. | |
2081 | */ | |
2082 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2083 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2084 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2085 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2086 | __dl_add(dl_b, new_bw); | |
2087 | err = 0; | |
2088 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2089 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2090 | __dl_clear(dl_b, p->dl.dl_bw); | |
2091 | __dl_add(dl_b, new_bw); | |
2092 | err = 0; | |
2093 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2094 | __dl_clear(dl_b, p->dl.dl_bw); | |
2095 | err = 0; | |
2096 | } | |
2097 | raw_spin_unlock(&dl_b->lock); | |
2098 | ||
2099 | return err; | |
2100 | } | |
2101 | ||
2102 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2103 | ||
1da177e4 LT |
2104 | /* |
2105 | * wake_up_new_task - wake up a newly created task for the first time. | |
2106 | * | |
2107 | * This function will do some initial scheduler statistics housekeeping | |
2108 | * that must be done for every newly created context, then puts the task | |
2109 | * on the runqueue and wakes it. | |
2110 | */ | |
3e51e3ed | 2111 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2112 | { |
2113 | unsigned long flags; | |
dd41f596 | 2114 | struct rq *rq; |
fabf318e | 2115 | |
ab2515c4 | 2116 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2117 | #ifdef CONFIG_SMP |
2118 | /* | |
2119 | * Fork balancing, do it here and not earlier because: | |
2120 | * - cpus_allowed can change in the fork path | |
2121 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2122 | */ |
ac66f547 | 2123 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2124 | #endif |
2125 | ||
a75cdaa9 AS |
2126 | /* Initialize new task's runnable average */ |
2127 | init_task_runnable_average(p); | |
ab2515c4 | 2128 | rq = __task_rq_lock(p); |
cd29fe6f | 2129 | activate_task(rq, p, 0); |
da0c1e65 | 2130 | p->on_rq = TASK_ON_RQ_QUEUED; |
89363381 | 2131 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2132 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2133 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2134 | if (p->sched_class->task_woken) |
2135 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2136 | #endif |
0122ec5b | 2137 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2138 | } |
2139 | ||
e107be36 AK |
2140 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2141 | ||
2142 | /** | |
80dd99b3 | 2143 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2144 | * @notifier: notifier struct to register |
e107be36 AK |
2145 | */ |
2146 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2147 | { | |
2148 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2149 | } | |
2150 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2151 | ||
2152 | /** | |
2153 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2154 | * @notifier: notifier struct to unregister |
e107be36 AK |
2155 | * |
2156 | * This is safe to call from within a preemption notifier. | |
2157 | */ | |
2158 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2159 | { | |
2160 | hlist_del(¬ifier->link); | |
2161 | } | |
2162 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2163 | ||
2164 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2165 | { | |
2166 | struct preempt_notifier *notifier; | |
e107be36 | 2167 | |
b67bfe0d | 2168 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2169 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2170 | } | |
2171 | ||
2172 | static void | |
2173 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2174 | struct task_struct *next) | |
2175 | { | |
2176 | struct preempt_notifier *notifier; | |
e107be36 | 2177 | |
b67bfe0d | 2178 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2179 | notifier->ops->sched_out(notifier, next); |
2180 | } | |
2181 | ||
6d6bc0ad | 2182 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2183 | |
2184 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2185 | { | |
2186 | } | |
2187 | ||
2188 | static void | |
2189 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2190 | struct task_struct *next) | |
2191 | { | |
2192 | } | |
2193 | ||
6d6bc0ad | 2194 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2195 | |
4866cde0 NP |
2196 | /** |
2197 | * prepare_task_switch - prepare to switch tasks | |
2198 | * @rq: the runqueue preparing to switch | |
421cee29 | 2199 | * @prev: the current task that is being switched out |
4866cde0 NP |
2200 | * @next: the task we are going to switch to. |
2201 | * | |
2202 | * This is called with the rq lock held and interrupts off. It must | |
2203 | * be paired with a subsequent finish_task_switch after the context | |
2204 | * switch. | |
2205 | * | |
2206 | * prepare_task_switch sets up locking and calls architecture specific | |
2207 | * hooks. | |
2208 | */ | |
e107be36 AK |
2209 | static inline void |
2210 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2211 | struct task_struct *next) | |
4866cde0 | 2212 | { |
895dd92c | 2213 | trace_sched_switch(prev, next); |
43148951 | 2214 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2215 | perf_event_task_sched_out(prev, next); |
e107be36 | 2216 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2217 | prepare_lock_switch(rq, next); |
2218 | prepare_arch_switch(next); | |
2219 | } | |
2220 | ||
1da177e4 LT |
2221 | /** |
2222 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2223 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2224 | * @prev: the thread we just switched away from. |
2225 | * | |
4866cde0 NP |
2226 | * finish_task_switch must be called after the context switch, paired |
2227 | * with a prepare_task_switch call before the context switch. | |
2228 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2229 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2230 | * |
2231 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2232 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2233 | * with the lock held can cause deadlocks; see schedule() for |
2234 | * details.) | |
2235 | */ | |
a9957449 | 2236 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2237 | __releases(rq->lock) |
2238 | { | |
1da177e4 | 2239 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2240 | long prev_state; |
1da177e4 LT |
2241 | |
2242 | rq->prev_mm = NULL; | |
2243 | ||
2244 | /* | |
2245 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2246 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2247 | * schedule one last time. The schedule call will never return, and |
2248 | * the scheduled task must drop that reference. | |
c394cc9f | 2249 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2250 | * still held, otherwise prev could be scheduled on another cpu, die |
2251 | * there before we look at prev->state, and then the reference would | |
2252 | * be dropped twice. | |
2253 | * Manfred Spraul <manfred@colorfullife.com> | |
2254 | */ | |
55a101f8 | 2255 | prev_state = prev->state; |
bf9fae9f | 2256 | vtime_task_switch(prev); |
4866cde0 | 2257 | finish_arch_switch(prev); |
a8d757ef | 2258 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2259 | finish_lock_switch(rq, prev); |
01f23e16 | 2260 | finish_arch_post_lock_switch(); |
e8fa1362 | 2261 | |
e107be36 | 2262 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2263 | if (mm) |
2264 | mmdrop(mm); | |
c394cc9f | 2265 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2266 | if (prev->sched_class->task_dead) |
2267 | prev->sched_class->task_dead(prev); | |
2268 | ||
c6fd91f0 | 2269 | /* |
2270 | * Remove function-return probe instances associated with this | |
2271 | * task and put them back on the free list. | |
9761eea8 | 2272 | */ |
c6fd91f0 | 2273 | kprobe_flush_task(prev); |
1da177e4 | 2274 | put_task_struct(prev); |
c6fd91f0 | 2275 | } |
99e5ada9 FW |
2276 | |
2277 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2278 | } |
2279 | ||
3f029d3c GH |
2280 | #ifdef CONFIG_SMP |
2281 | ||
3f029d3c GH |
2282 | /* rq->lock is NOT held, but preemption is disabled */ |
2283 | static inline void post_schedule(struct rq *rq) | |
2284 | { | |
2285 | if (rq->post_schedule) { | |
2286 | unsigned long flags; | |
2287 | ||
05fa785c | 2288 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2289 | if (rq->curr->sched_class->post_schedule) |
2290 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2291 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2292 | |
2293 | rq->post_schedule = 0; | |
2294 | } | |
2295 | } | |
2296 | ||
2297 | #else | |
da19ab51 | 2298 | |
3f029d3c GH |
2299 | static inline void post_schedule(struct rq *rq) |
2300 | { | |
1da177e4 LT |
2301 | } |
2302 | ||
3f029d3c GH |
2303 | #endif |
2304 | ||
1da177e4 LT |
2305 | /** |
2306 | * schedule_tail - first thing a freshly forked thread must call. | |
2307 | * @prev: the thread we just switched away from. | |
2308 | */ | |
722a9f92 | 2309 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2310 | __releases(rq->lock) |
2311 | { | |
70b97a7f IM |
2312 | struct rq *rq = this_rq(); |
2313 | ||
4866cde0 | 2314 | finish_task_switch(rq, prev); |
da19ab51 | 2315 | |
3f029d3c GH |
2316 | /* |
2317 | * FIXME: do we need to worry about rq being invalidated by the | |
2318 | * task_switch? | |
2319 | */ | |
2320 | post_schedule(rq); | |
70b97a7f | 2321 | |
1da177e4 | 2322 | if (current->set_child_tid) |
b488893a | 2323 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2324 | } |
2325 | ||
2326 | /* | |
2327 | * context_switch - switch to the new MM and the new | |
2328 | * thread's register state. | |
2329 | */ | |
dd41f596 | 2330 | static inline void |
70b97a7f | 2331 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2332 | struct task_struct *next) |
1da177e4 | 2333 | { |
dd41f596 | 2334 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2335 | |
e107be36 | 2336 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2337 | |
dd41f596 IM |
2338 | mm = next->mm; |
2339 | oldmm = prev->active_mm; | |
9226d125 ZA |
2340 | /* |
2341 | * For paravirt, this is coupled with an exit in switch_to to | |
2342 | * combine the page table reload and the switch backend into | |
2343 | * one hypercall. | |
2344 | */ | |
224101ed | 2345 | arch_start_context_switch(prev); |
9226d125 | 2346 | |
31915ab4 | 2347 | if (!mm) { |
1da177e4 LT |
2348 | next->active_mm = oldmm; |
2349 | atomic_inc(&oldmm->mm_count); | |
2350 | enter_lazy_tlb(oldmm, next); | |
2351 | } else | |
2352 | switch_mm(oldmm, mm, next); | |
2353 | ||
31915ab4 | 2354 | if (!prev->mm) { |
1da177e4 | 2355 | prev->active_mm = NULL; |
1da177e4 LT |
2356 | rq->prev_mm = oldmm; |
2357 | } | |
3a5f5e48 IM |
2358 | /* |
2359 | * Since the runqueue lock will be released by the next | |
2360 | * task (which is an invalid locking op but in the case | |
2361 | * of the scheduler it's an obvious special-case), so we | |
2362 | * do an early lockdep release here: | |
2363 | */ | |
8a25d5de | 2364 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 | 2365 | |
91d1aa43 | 2366 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2367 | /* Here we just switch the register state and the stack. */ |
2368 | switch_to(prev, next, prev); | |
2369 | ||
dd41f596 IM |
2370 | barrier(); |
2371 | /* | |
2372 | * this_rq must be evaluated again because prev may have moved | |
2373 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2374 | * frame will be invalid. | |
2375 | */ | |
2376 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2377 | } |
2378 | ||
2379 | /* | |
1c3e8264 | 2380 | * nr_running and nr_context_switches: |
1da177e4 LT |
2381 | * |
2382 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2383 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2384 | */ |
2385 | unsigned long nr_running(void) | |
2386 | { | |
2387 | unsigned long i, sum = 0; | |
2388 | ||
2389 | for_each_online_cpu(i) | |
2390 | sum += cpu_rq(i)->nr_running; | |
2391 | ||
2392 | return sum; | |
f711f609 | 2393 | } |
1da177e4 | 2394 | |
2ee507c4 TC |
2395 | /* |
2396 | * Check if only the current task is running on the cpu. | |
2397 | */ | |
2398 | bool single_task_running(void) | |
2399 | { | |
2400 | if (cpu_rq(smp_processor_id())->nr_running == 1) | |
2401 | return true; | |
2402 | else | |
2403 | return false; | |
2404 | } | |
2405 | EXPORT_SYMBOL(single_task_running); | |
2406 | ||
1da177e4 | 2407 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2408 | { |
cc94abfc SR |
2409 | int i; |
2410 | unsigned long long sum = 0; | |
46cb4b7c | 2411 | |
0a945022 | 2412 | for_each_possible_cpu(i) |
1da177e4 | 2413 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2414 | |
1da177e4 LT |
2415 | return sum; |
2416 | } | |
483b4ee6 | 2417 | |
1da177e4 LT |
2418 | unsigned long nr_iowait(void) |
2419 | { | |
2420 | unsigned long i, sum = 0; | |
483b4ee6 | 2421 | |
0a945022 | 2422 | for_each_possible_cpu(i) |
1da177e4 | 2423 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2424 | |
1da177e4 LT |
2425 | return sum; |
2426 | } | |
483b4ee6 | 2427 | |
8c215bd3 | 2428 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2429 | { |
8c215bd3 | 2430 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2431 | return atomic_read(&this->nr_iowait); |
2432 | } | |
46cb4b7c | 2433 | |
372ba8cb MG |
2434 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2435 | { | |
2436 | struct rq *this = this_rq(); | |
2437 | *nr_waiters = atomic_read(&this->nr_iowait); | |
2438 | *load = this->cpu_load[0]; | |
2439 | } | |
2440 | ||
dd41f596 | 2441 | #ifdef CONFIG_SMP |
8a0be9ef | 2442 | |
46cb4b7c | 2443 | /* |
38022906 PZ |
2444 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2445 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2446 | */ |
38022906 | 2447 | void sched_exec(void) |
46cb4b7c | 2448 | { |
38022906 | 2449 | struct task_struct *p = current; |
1da177e4 | 2450 | unsigned long flags; |
0017d735 | 2451 | int dest_cpu; |
46cb4b7c | 2452 | |
8f42ced9 | 2453 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2454 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2455 | if (dest_cpu == smp_processor_id()) |
2456 | goto unlock; | |
38022906 | 2457 | |
8f42ced9 | 2458 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2459 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2460 | |
8f42ced9 PZ |
2461 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2462 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2463 | return; |
2464 | } | |
0017d735 | 2465 | unlock: |
8f42ced9 | 2466 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2467 | } |
dd41f596 | 2468 | |
1da177e4 LT |
2469 | #endif |
2470 | ||
1da177e4 | 2471 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2472 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2473 | |
2474 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2475 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2476 | |
2477 | /* | |
c5f8d995 | 2478 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2479 | * @p in case that task is currently running. |
c5f8d995 HS |
2480 | * |
2481 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2482 | */ |
c5f8d995 HS |
2483 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2484 | { | |
2485 | u64 ns = 0; | |
2486 | ||
4036ac15 MG |
2487 | /* |
2488 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2489 | * project cycles that may never be accounted to this | |
2490 | * thread, breaking clock_gettime(). | |
2491 | */ | |
da0c1e65 | 2492 | if (task_current(rq, p) && task_on_rq_queued(p)) { |
c5f8d995 | 2493 | update_rq_clock(rq); |
78becc27 | 2494 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2495 | if ((s64)ns < 0) |
2496 | ns = 0; | |
2497 | } | |
2498 | ||
2499 | return ns; | |
2500 | } | |
2501 | ||
bb34d92f | 2502 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2503 | { |
1da177e4 | 2504 | unsigned long flags; |
41b86e9c | 2505 | struct rq *rq; |
bb34d92f | 2506 | u64 ns = 0; |
48f24c4d | 2507 | |
41b86e9c | 2508 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2509 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2510 | task_rq_unlock(rq, p, &flags); |
1508487e | 2511 | |
c5f8d995 HS |
2512 | return ns; |
2513 | } | |
f06febc9 | 2514 | |
c5f8d995 HS |
2515 | /* |
2516 | * Return accounted runtime for the task. | |
2517 | * In case the task is currently running, return the runtime plus current's | |
2518 | * pending runtime that have not been accounted yet. | |
2519 | */ | |
2520 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2521 | { | |
2522 | unsigned long flags; | |
2523 | struct rq *rq; | |
2524 | u64 ns = 0; | |
2525 | ||
911b2898 PZ |
2526 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2527 | /* | |
2528 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2529 | * So we have a optimization chance when the task's delta_exec is 0. | |
2530 | * Reading ->on_cpu is racy, but this is ok. | |
2531 | * | |
2532 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2533 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2534 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2535 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2536 | * been accounted, so we're correct here as well. | |
911b2898 | 2537 | */ |
da0c1e65 | 2538 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2539 | return p->se.sum_exec_runtime; |
2540 | #endif | |
2541 | ||
c5f8d995 HS |
2542 | rq = task_rq_lock(p, &flags); |
2543 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2544 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2545 | |
2546 | return ns; | |
2547 | } | |
48f24c4d | 2548 | |
7835b98b CL |
2549 | /* |
2550 | * This function gets called by the timer code, with HZ frequency. | |
2551 | * We call it with interrupts disabled. | |
7835b98b CL |
2552 | */ |
2553 | void scheduler_tick(void) | |
2554 | { | |
7835b98b CL |
2555 | int cpu = smp_processor_id(); |
2556 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2557 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2558 | |
2559 | sched_clock_tick(); | |
dd41f596 | 2560 | |
05fa785c | 2561 | raw_spin_lock(&rq->lock); |
3e51f33f | 2562 | update_rq_clock(rq); |
fa85ae24 | 2563 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2564 | update_cpu_load_active(rq); |
05fa785c | 2565 | raw_spin_unlock(&rq->lock); |
7835b98b | 2566 | |
e9d2b064 | 2567 | perf_event_task_tick(); |
e220d2dc | 2568 | |
e418e1c2 | 2569 | #ifdef CONFIG_SMP |
6eb57e0d | 2570 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2571 | trigger_load_balance(rq); |
e418e1c2 | 2572 | #endif |
265f22a9 | 2573 | rq_last_tick_reset(rq); |
1da177e4 LT |
2574 | } |
2575 | ||
265f22a9 FW |
2576 | #ifdef CONFIG_NO_HZ_FULL |
2577 | /** | |
2578 | * scheduler_tick_max_deferment | |
2579 | * | |
2580 | * Keep at least one tick per second when a single | |
2581 | * active task is running because the scheduler doesn't | |
2582 | * yet completely support full dynticks environment. | |
2583 | * | |
2584 | * This makes sure that uptime, CFS vruntime, load | |
2585 | * balancing, etc... continue to move forward, even | |
2586 | * with a very low granularity. | |
e69f6186 YB |
2587 | * |
2588 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2589 | */ |
2590 | u64 scheduler_tick_max_deferment(void) | |
2591 | { | |
2592 | struct rq *rq = this_rq(); | |
2593 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2594 | ||
2595 | next = rq->last_sched_tick + HZ; | |
2596 | ||
2597 | if (time_before_eq(next, now)) | |
2598 | return 0; | |
2599 | ||
8fe8ff09 | 2600 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2601 | } |
265f22a9 | 2602 | #endif |
1da177e4 | 2603 | |
132380a0 | 2604 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2605 | { |
2606 | if (in_lock_functions(addr)) { | |
2607 | addr = CALLER_ADDR2; | |
2608 | if (in_lock_functions(addr)) | |
2609 | addr = CALLER_ADDR3; | |
2610 | } | |
2611 | return addr; | |
2612 | } | |
1da177e4 | 2613 | |
7e49fcce SR |
2614 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2615 | defined(CONFIG_PREEMPT_TRACER)) | |
2616 | ||
edafe3a5 | 2617 | void preempt_count_add(int val) |
1da177e4 | 2618 | { |
6cd8a4bb | 2619 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2620 | /* |
2621 | * Underflow? | |
2622 | */ | |
9a11b49a IM |
2623 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2624 | return; | |
6cd8a4bb | 2625 | #endif |
bdb43806 | 2626 | __preempt_count_add(val); |
6cd8a4bb | 2627 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2628 | /* |
2629 | * Spinlock count overflowing soon? | |
2630 | */ | |
33859f7f MOS |
2631 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2632 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2633 | #endif |
8f47b187 TG |
2634 | if (preempt_count() == val) { |
2635 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2636 | #ifdef CONFIG_DEBUG_PREEMPT | |
2637 | current->preempt_disable_ip = ip; | |
2638 | #endif | |
2639 | trace_preempt_off(CALLER_ADDR0, ip); | |
2640 | } | |
1da177e4 | 2641 | } |
bdb43806 | 2642 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2643 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2644 | |
edafe3a5 | 2645 | void preempt_count_sub(int val) |
1da177e4 | 2646 | { |
6cd8a4bb | 2647 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2648 | /* |
2649 | * Underflow? | |
2650 | */ | |
01e3eb82 | 2651 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2652 | return; |
1da177e4 LT |
2653 | /* |
2654 | * Is the spinlock portion underflowing? | |
2655 | */ | |
9a11b49a IM |
2656 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2657 | !(preempt_count() & PREEMPT_MASK))) | |
2658 | return; | |
6cd8a4bb | 2659 | #endif |
9a11b49a | 2660 | |
6cd8a4bb SR |
2661 | if (preempt_count() == val) |
2662 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2663 | __preempt_count_sub(val); |
1da177e4 | 2664 | } |
bdb43806 | 2665 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2666 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2667 | |
2668 | #endif | |
2669 | ||
2670 | /* | |
dd41f596 | 2671 | * Print scheduling while atomic bug: |
1da177e4 | 2672 | */ |
dd41f596 | 2673 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2674 | { |
664dfa65 DJ |
2675 | if (oops_in_progress) |
2676 | return; | |
2677 | ||
3df0fc5b PZ |
2678 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2679 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2680 | |
dd41f596 | 2681 | debug_show_held_locks(prev); |
e21f5b15 | 2682 | print_modules(); |
dd41f596 IM |
2683 | if (irqs_disabled()) |
2684 | print_irqtrace_events(prev); | |
8f47b187 TG |
2685 | #ifdef CONFIG_DEBUG_PREEMPT |
2686 | if (in_atomic_preempt_off()) { | |
2687 | pr_err("Preemption disabled at:"); | |
2688 | print_ip_sym(current->preempt_disable_ip); | |
2689 | pr_cont("\n"); | |
2690 | } | |
2691 | #endif | |
6135fc1e | 2692 | dump_stack(); |
373d4d09 | 2693 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2694 | } |
1da177e4 | 2695 | |
dd41f596 IM |
2696 | /* |
2697 | * Various schedule()-time debugging checks and statistics: | |
2698 | */ | |
2699 | static inline void schedule_debug(struct task_struct *prev) | |
2700 | { | |
0d9e2632 AT |
2701 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
2702 | BUG_ON(unlikely(task_stack_end_corrupted(prev))); | |
2703 | #endif | |
1da177e4 | 2704 | /* |
41a2d6cf | 2705 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2706 | * schedule() atomically, we ignore that path. Otherwise whine |
2707 | * if we are scheduling when we should not. | |
1da177e4 | 2708 | */ |
192301e7 | 2709 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2710 | __schedule_bug(prev); |
b3fbab05 | 2711 | rcu_sleep_check(); |
dd41f596 | 2712 | |
1da177e4 LT |
2713 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2714 | ||
2d72376b | 2715 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2716 | } |
2717 | ||
2718 | /* | |
2719 | * Pick up the highest-prio task: | |
2720 | */ | |
2721 | static inline struct task_struct * | |
606dba2e | 2722 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2723 | { |
37e117c0 | 2724 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2725 | struct task_struct *p; |
1da177e4 LT |
2726 | |
2727 | /* | |
dd41f596 IM |
2728 | * Optimization: we know that if all tasks are in |
2729 | * the fair class we can call that function directly: | |
1da177e4 | 2730 | */ |
37e117c0 | 2731 | if (likely(prev->sched_class == class && |
38033c37 | 2732 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2733 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2734 | if (unlikely(p == RETRY_TASK)) |
2735 | goto again; | |
2736 | ||
2737 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2738 | if (unlikely(!p)) | |
2739 | p = idle_sched_class.pick_next_task(rq, prev); | |
2740 | ||
2741 | return p; | |
1da177e4 LT |
2742 | } |
2743 | ||
37e117c0 | 2744 | again: |
34f971f6 | 2745 | for_each_class(class) { |
606dba2e | 2746 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2747 | if (p) { |
2748 | if (unlikely(p == RETRY_TASK)) | |
2749 | goto again; | |
dd41f596 | 2750 | return p; |
37e117c0 | 2751 | } |
dd41f596 | 2752 | } |
34f971f6 PZ |
2753 | |
2754 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2755 | } |
1da177e4 | 2756 | |
dd41f596 | 2757 | /* |
c259e01a | 2758 | * __schedule() is the main scheduler function. |
edde96ea PE |
2759 | * |
2760 | * The main means of driving the scheduler and thus entering this function are: | |
2761 | * | |
2762 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2763 | * | |
2764 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2765 | * paths. For example, see arch/x86/entry_64.S. | |
2766 | * | |
2767 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2768 | * interrupt handler scheduler_tick(). | |
2769 | * | |
2770 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2771 | * task to the run-queue and that's it. | |
2772 | * | |
2773 | * Now, if the new task added to the run-queue preempts the current | |
2774 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2775 | * called on the nearest possible occasion: | |
2776 | * | |
2777 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2778 | * | |
2779 | * - in syscall or exception context, at the next outmost | |
2780 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2781 | * spin_unlock()!) | |
2782 | * | |
2783 | * - in IRQ context, return from interrupt-handler to | |
2784 | * preemptible context | |
2785 | * | |
2786 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2787 | * then at the next: | |
2788 | * | |
2789 | * - cond_resched() call | |
2790 | * - explicit schedule() call | |
2791 | * - return from syscall or exception to user-space | |
2792 | * - return from interrupt-handler to user-space | |
dd41f596 | 2793 | */ |
c259e01a | 2794 | static void __sched __schedule(void) |
dd41f596 IM |
2795 | { |
2796 | struct task_struct *prev, *next; | |
67ca7bde | 2797 | unsigned long *switch_count; |
dd41f596 | 2798 | struct rq *rq; |
31656519 | 2799 | int cpu; |
dd41f596 | 2800 | |
ff743345 PZ |
2801 | need_resched: |
2802 | preempt_disable(); | |
dd41f596 IM |
2803 | cpu = smp_processor_id(); |
2804 | rq = cpu_rq(cpu); | |
25502a6c | 2805 | rcu_note_context_switch(cpu); |
dd41f596 | 2806 | prev = rq->curr; |
dd41f596 | 2807 | |
dd41f596 | 2808 | schedule_debug(prev); |
1da177e4 | 2809 | |
31656519 | 2810 | if (sched_feat(HRTICK)) |
f333fdc9 | 2811 | hrtick_clear(rq); |
8f4d37ec | 2812 | |
e0acd0a6 ON |
2813 | /* |
2814 | * Make sure that signal_pending_state()->signal_pending() below | |
2815 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2816 | * done by the caller to avoid the race with signal_wake_up(). | |
2817 | */ | |
2818 | smp_mb__before_spinlock(); | |
05fa785c | 2819 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2820 | |
246d86b5 | 2821 | switch_count = &prev->nivcsw; |
1da177e4 | 2822 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2823 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2824 | prev->state = TASK_RUNNING; |
21aa9af0 | 2825 | } else { |
2acca55e PZ |
2826 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2827 | prev->on_rq = 0; | |
2828 | ||
21aa9af0 | 2829 | /* |
2acca55e PZ |
2830 | * If a worker went to sleep, notify and ask workqueue |
2831 | * whether it wants to wake up a task to maintain | |
2832 | * concurrency. | |
21aa9af0 TH |
2833 | */ |
2834 | if (prev->flags & PF_WQ_WORKER) { | |
2835 | struct task_struct *to_wakeup; | |
2836 | ||
2837 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2838 | if (to_wakeup) | |
2839 | try_to_wake_up_local(to_wakeup); | |
2840 | } | |
21aa9af0 | 2841 | } |
dd41f596 | 2842 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2843 | } |
2844 | ||
da0c1e65 | 2845 | if (task_on_rq_queued(prev) || rq->skip_clock_update < 0) |
606dba2e PZ |
2846 | update_rq_clock(rq); |
2847 | ||
2848 | next = pick_next_task(rq, prev); | |
f26f9aff | 2849 | clear_tsk_need_resched(prev); |
f27dde8d | 2850 | clear_preempt_need_resched(); |
f26f9aff | 2851 | rq->skip_clock_update = 0; |
1da177e4 | 2852 | |
1da177e4 | 2853 | if (likely(prev != next)) { |
1da177e4 LT |
2854 | rq->nr_switches++; |
2855 | rq->curr = next; | |
2856 | ++*switch_count; | |
2857 | ||
dd41f596 | 2858 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2859 | /* |
246d86b5 ON |
2860 | * The context switch have flipped the stack from under us |
2861 | * and restored the local variables which were saved when | |
2862 | * this task called schedule() in the past. prev == current | |
2863 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2864 | */ |
2865 | cpu = smp_processor_id(); | |
2866 | rq = cpu_rq(cpu); | |
1da177e4 | 2867 | } else |
05fa785c | 2868 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2869 | |
3f029d3c | 2870 | post_schedule(rq); |
1da177e4 | 2871 | |
ba74c144 | 2872 | sched_preempt_enable_no_resched(); |
ff743345 | 2873 | if (need_resched()) |
1da177e4 LT |
2874 | goto need_resched; |
2875 | } | |
c259e01a | 2876 | |
9c40cef2 TG |
2877 | static inline void sched_submit_work(struct task_struct *tsk) |
2878 | { | |
3c7d5184 | 2879 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2880 | return; |
2881 | /* | |
2882 | * If we are going to sleep and we have plugged IO queued, | |
2883 | * make sure to submit it to avoid deadlocks. | |
2884 | */ | |
2885 | if (blk_needs_flush_plug(tsk)) | |
2886 | blk_schedule_flush_plug(tsk); | |
2887 | } | |
2888 | ||
722a9f92 | 2889 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2890 | { |
9c40cef2 TG |
2891 | struct task_struct *tsk = current; |
2892 | ||
2893 | sched_submit_work(tsk); | |
c259e01a TG |
2894 | __schedule(); |
2895 | } | |
1da177e4 LT |
2896 | EXPORT_SYMBOL(schedule); |
2897 | ||
91d1aa43 | 2898 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2899 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2900 | { |
2901 | /* | |
2902 | * If we come here after a random call to set_need_resched(), | |
2903 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2904 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2905 | * we find a better solution. | |
2906 | */ | |
91d1aa43 | 2907 | user_exit(); |
20ab65e3 | 2908 | schedule(); |
91d1aa43 | 2909 | user_enter(); |
20ab65e3 FW |
2910 | } |
2911 | #endif | |
2912 | ||
c5491ea7 TG |
2913 | /** |
2914 | * schedule_preempt_disabled - called with preemption disabled | |
2915 | * | |
2916 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2917 | */ | |
2918 | void __sched schedule_preempt_disabled(void) | |
2919 | { | |
ba74c144 | 2920 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2921 | schedule(); |
2922 | preempt_disable(); | |
2923 | } | |
2924 | ||
1da177e4 LT |
2925 | #ifdef CONFIG_PREEMPT |
2926 | /* | |
2ed6e34f | 2927 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2928 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2929 | * occur there and call schedule directly. |
2930 | */ | |
722a9f92 | 2931 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2932 | { |
1da177e4 LT |
2933 | /* |
2934 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2935 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2936 | */ |
fbb00b56 | 2937 | if (likely(!preemptible())) |
1da177e4 LT |
2938 | return; |
2939 | ||
3a5c359a | 2940 | do { |
bdb43806 | 2941 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2942 | __schedule(); |
bdb43806 | 2943 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2944 | |
3a5c359a AK |
2945 | /* |
2946 | * Check again in case we missed a preemption opportunity | |
2947 | * between schedule and now. | |
2948 | */ | |
2949 | barrier(); | |
5ed0cec0 | 2950 | } while (need_resched()); |
1da177e4 | 2951 | } |
376e2424 | 2952 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2953 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 ON |
2954 | |
2955 | #ifdef CONFIG_CONTEXT_TRACKING | |
2956 | /** | |
2957 | * preempt_schedule_context - preempt_schedule called by tracing | |
2958 | * | |
2959 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
2960 | * recursion and tracing preempt enabling caused by the tracing | |
2961 | * infrastructure itself. But as tracing can happen in areas coming | |
2962 | * from userspace or just about to enter userspace, a preempt enable | |
2963 | * can occur before user_exit() is called. This will cause the scheduler | |
2964 | * to be called when the system is still in usermode. | |
2965 | * | |
2966 | * To prevent this, the preempt_enable_notrace will use this function | |
2967 | * instead of preempt_schedule() to exit user context if needed before | |
2968 | * calling the scheduler. | |
2969 | */ | |
2970 | asmlinkage __visible void __sched notrace preempt_schedule_context(void) | |
2971 | { | |
2972 | enum ctx_state prev_ctx; | |
2973 | ||
2974 | if (likely(!preemptible())) | |
2975 | return; | |
2976 | ||
2977 | do { | |
2978 | __preempt_count_add(PREEMPT_ACTIVE); | |
2979 | /* | |
2980 | * Needs preempt disabled in case user_exit() is traced | |
2981 | * and the tracer calls preempt_enable_notrace() causing | |
2982 | * an infinite recursion. | |
2983 | */ | |
2984 | prev_ctx = exception_enter(); | |
2985 | __schedule(); | |
2986 | exception_exit(prev_ctx); | |
2987 | ||
2988 | __preempt_count_sub(PREEMPT_ACTIVE); | |
2989 | barrier(); | |
2990 | } while (need_resched()); | |
2991 | } | |
2992 | EXPORT_SYMBOL_GPL(preempt_schedule_context); | |
2993 | #endif /* CONFIG_CONTEXT_TRACKING */ | |
2994 | ||
32e475d7 | 2995 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2996 | |
2997 | /* | |
2ed6e34f | 2998 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2999 | * off of irq context. |
3000 | * Note, that this is called and return with irqs disabled. This will | |
3001 | * protect us against recursive calling from irq. | |
3002 | */ | |
722a9f92 | 3003 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3004 | { |
b22366cd | 3005 | enum ctx_state prev_state; |
6478d880 | 3006 | |
2ed6e34f | 3007 | /* Catch callers which need to be fixed */ |
f27dde8d | 3008 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3009 | |
b22366cd FW |
3010 | prev_state = exception_enter(); |
3011 | ||
3a5c359a | 3012 | do { |
bdb43806 | 3013 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 3014 | local_irq_enable(); |
c259e01a | 3015 | __schedule(); |
3a5c359a | 3016 | local_irq_disable(); |
bdb43806 | 3017 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 3018 | |
3a5c359a AK |
3019 | /* |
3020 | * Check again in case we missed a preemption opportunity | |
3021 | * between schedule and now. | |
3022 | */ | |
3023 | barrier(); | |
5ed0cec0 | 3024 | } while (need_resched()); |
b22366cd FW |
3025 | |
3026 | exception_exit(prev_state); | |
1da177e4 LT |
3027 | } |
3028 | ||
63859d4f | 3029 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3030 | void *key) |
1da177e4 | 3031 | { |
63859d4f | 3032 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3033 | } |
1da177e4 LT |
3034 | EXPORT_SYMBOL(default_wake_function); |
3035 | ||
b29739f9 IM |
3036 | #ifdef CONFIG_RT_MUTEXES |
3037 | ||
3038 | /* | |
3039 | * rt_mutex_setprio - set the current priority of a task | |
3040 | * @p: task | |
3041 | * @prio: prio value (kernel-internal form) | |
3042 | * | |
3043 | * This function changes the 'effective' priority of a task. It does | |
3044 | * not touch ->normal_prio like __setscheduler(). | |
3045 | * | |
c365c292 TG |
3046 | * Used by the rt_mutex code to implement priority inheritance |
3047 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3048 | */ |
36c8b586 | 3049 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3050 | { |
da0c1e65 | 3051 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 3052 | struct rq *rq; |
83ab0aa0 | 3053 | const struct sched_class *prev_class; |
b29739f9 | 3054 | |
aab03e05 | 3055 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3056 | |
0122ec5b | 3057 | rq = __task_rq_lock(p); |
b29739f9 | 3058 | |
1c4dd99b TG |
3059 | /* |
3060 | * Idle task boosting is a nono in general. There is one | |
3061 | * exception, when PREEMPT_RT and NOHZ is active: | |
3062 | * | |
3063 | * The idle task calls get_next_timer_interrupt() and holds | |
3064 | * the timer wheel base->lock on the CPU and another CPU wants | |
3065 | * to access the timer (probably to cancel it). We can safely | |
3066 | * ignore the boosting request, as the idle CPU runs this code | |
3067 | * with interrupts disabled and will complete the lock | |
3068 | * protected section without being interrupted. So there is no | |
3069 | * real need to boost. | |
3070 | */ | |
3071 | if (unlikely(p == rq->idle)) { | |
3072 | WARN_ON(p != rq->curr); | |
3073 | WARN_ON(p->pi_blocked_on); | |
3074 | goto out_unlock; | |
3075 | } | |
3076 | ||
a8027073 | 3077 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3078 | oldprio = p->prio; |
83ab0aa0 | 3079 | prev_class = p->sched_class; |
da0c1e65 | 3080 | queued = task_on_rq_queued(p); |
051a1d1a | 3081 | running = task_current(rq, p); |
da0c1e65 | 3082 | if (queued) |
69be72c1 | 3083 | dequeue_task(rq, p, 0); |
0e1f3483 | 3084 | if (running) |
f3cd1c4e | 3085 | put_prev_task(rq, p); |
dd41f596 | 3086 | |
2d3d891d DF |
3087 | /* |
3088 | * Boosting condition are: | |
3089 | * 1. -rt task is running and holds mutex A | |
3090 | * --> -dl task blocks on mutex A | |
3091 | * | |
3092 | * 2. -dl task is running and holds mutex A | |
3093 | * --> -dl task blocks on mutex A and could preempt the | |
3094 | * running task | |
3095 | */ | |
3096 | if (dl_prio(prio)) { | |
466af29b ON |
3097 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3098 | if (!dl_prio(p->normal_prio) || | |
3099 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3100 | p->dl.dl_boosted = 1; |
3101 | p->dl.dl_throttled = 0; | |
3102 | enqueue_flag = ENQUEUE_REPLENISH; | |
3103 | } else | |
3104 | p->dl.dl_boosted = 0; | |
aab03e05 | 3105 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3106 | } else if (rt_prio(prio)) { |
3107 | if (dl_prio(oldprio)) | |
3108 | p->dl.dl_boosted = 0; | |
3109 | if (oldprio < prio) | |
3110 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3111 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3112 | } else { |
3113 | if (dl_prio(oldprio)) | |
3114 | p->dl.dl_boosted = 0; | |
dd41f596 | 3115 | p->sched_class = &fair_sched_class; |
2d3d891d | 3116 | } |
dd41f596 | 3117 | |
b29739f9 IM |
3118 | p->prio = prio; |
3119 | ||
0e1f3483 HS |
3120 | if (running) |
3121 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3122 | if (queued) |
2d3d891d | 3123 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3124 | |
da7a735e | 3125 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3126 | out_unlock: |
0122ec5b | 3127 | __task_rq_unlock(rq); |
b29739f9 | 3128 | } |
b29739f9 | 3129 | #endif |
d50dde5a | 3130 | |
36c8b586 | 3131 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3132 | { |
da0c1e65 | 3133 | int old_prio, delta, queued; |
1da177e4 | 3134 | unsigned long flags; |
70b97a7f | 3135 | struct rq *rq; |
1da177e4 | 3136 | |
75e45d51 | 3137 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3138 | return; |
3139 | /* | |
3140 | * We have to be careful, if called from sys_setpriority(), | |
3141 | * the task might be in the middle of scheduling on another CPU. | |
3142 | */ | |
3143 | rq = task_rq_lock(p, &flags); | |
3144 | /* | |
3145 | * The RT priorities are set via sched_setscheduler(), but we still | |
3146 | * allow the 'normal' nice value to be set - but as expected | |
3147 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3148 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3149 | */ |
aab03e05 | 3150 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3151 | p->static_prio = NICE_TO_PRIO(nice); |
3152 | goto out_unlock; | |
3153 | } | |
da0c1e65 KT |
3154 | queued = task_on_rq_queued(p); |
3155 | if (queued) | |
69be72c1 | 3156 | dequeue_task(rq, p, 0); |
1da177e4 | 3157 | |
1da177e4 | 3158 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3159 | set_load_weight(p); |
b29739f9 IM |
3160 | old_prio = p->prio; |
3161 | p->prio = effective_prio(p); | |
3162 | delta = p->prio - old_prio; | |
1da177e4 | 3163 | |
da0c1e65 | 3164 | if (queued) { |
371fd7e7 | 3165 | enqueue_task(rq, p, 0); |
1da177e4 | 3166 | /* |
d5f9f942 AM |
3167 | * If the task increased its priority or is running and |
3168 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3169 | */ |
d5f9f942 | 3170 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3171 | resched_curr(rq); |
1da177e4 LT |
3172 | } |
3173 | out_unlock: | |
0122ec5b | 3174 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3175 | } |
1da177e4 LT |
3176 | EXPORT_SYMBOL(set_user_nice); |
3177 | ||
e43379f1 MM |
3178 | /* |
3179 | * can_nice - check if a task can reduce its nice value | |
3180 | * @p: task | |
3181 | * @nice: nice value | |
3182 | */ | |
36c8b586 | 3183 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3184 | { |
024f4747 | 3185 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3186 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3187 | |
78d7d407 | 3188 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3189 | capable(CAP_SYS_NICE)); |
3190 | } | |
3191 | ||
1da177e4 LT |
3192 | #ifdef __ARCH_WANT_SYS_NICE |
3193 | ||
3194 | /* | |
3195 | * sys_nice - change the priority of the current process. | |
3196 | * @increment: priority increment | |
3197 | * | |
3198 | * sys_setpriority is a more generic, but much slower function that | |
3199 | * does similar things. | |
3200 | */ | |
5add95d4 | 3201 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3202 | { |
48f24c4d | 3203 | long nice, retval; |
1da177e4 LT |
3204 | |
3205 | /* | |
3206 | * Setpriority might change our priority at the same moment. | |
3207 | * We don't have to worry. Conceptually one call occurs first | |
3208 | * and we have a single winner. | |
3209 | */ | |
a9467fa3 | 3210 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3211 | nice = task_nice(current) + increment; |
1da177e4 | 3212 | |
a9467fa3 | 3213 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3214 | if (increment < 0 && !can_nice(current, nice)) |
3215 | return -EPERM; | |
3216 | ||
1da177e4 LT |
3217 | retval = security_task_setnice(current, nice); |
3218 | if (retval) | |
3219 | return retval; | |
3220 | ||
3221 | set_user_nice(current, nice); | |
3222 | return 0; | |
3223 | } | |
3224 | ||
3225 | #endif | |
3226 | ||
3227 | /** | |
3228 | * task_prio - return the priority value of a given task. | |
3229 | * @p: the task in question. | |
3230 | * | |
e69f6186 | 3231 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3232 | * RT tasks are offset by -200. Normal tasks are centered |
3233 | * around 0, value goes from -16 to +15. | |
3234 | */ | |
36c8b586 | 3235 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3236 | { |
3237 | return p->prio - MAX_RT_PRIO; | |
3238 | } | |
3239 | ||
1da177e4 LT |
3240 | /** |
3241 | * idle_cpu - is a given cpu idle currently? | |
3242 | * @cpu: the processor in question. | |
e69f6186 YB |
3243 | * |
3244 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3245 | */ |
3246 | int idle_cpu(int cpu) | |
3247 | { | |
908a3283 TG |
3248 | struct rq *rq = cpu_rq(cpu); |
3249 | ||
3250 | if (rq->curr != rq->idle) | |
3251 | return 0; | |
3252 | ||
3253 | if (rq->nr_running) | |
3254 | return 0; | |
3255 | ||
3256 | #ifdef CONFIG_SMP | |
3257 | if (!llist_empty(&rq->wake_list)) | |
3258 | return 0; | |
3259 | #endif | |
3260 | ||
3261 | return 1; | |
1da177e4 LT |
3262 | } |
3263 | ||
1da177e4 LT |
3264 | /** |
3265 | * idle_task - return the idle task for a given cpu. | |
3266 | * @cpu: the processor in question. | |
e69f6186 YB |
3267 | * |
3268 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3269 | */ |
36c8b586 | 3270 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3271 | { |
3272 | return cpu_rq(cpu)->idle; | |
3273 | } | |
3274 | ||
3275 | /** | |
3276 | * find_process_by_pid - find a process with a matching PID value. | |
3277 | * @pid: the pid in question. | |
e69f6186 YB |
3278 | * |
3279 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3280 | */ |
a9957449 | 3281 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3282 | { |
228ebcbe | 3283 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3284 | } |
3285 | ||
aab03e05 DF |
3286 | /* |
3287 | * This function initializes the sched_dl_entity of a newly becoming | |
3288 | * SCHED_DEADLINE task. | |
3289 | * | |
3290 | * Only the static values are considered here, the actual runtime and the | |
3291 | * absolute deadline will be properly calculated when the task is enqueued | |
3292 | * for the first time with its new policy. | |
3293 | */ | |
3294 | static void | |
3295 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3296 | { | |
3297 | struct sched_dl_entity *dl_se = &p->dl; | |
3298 | ||
3299 | init_dl_task_timer(dl_se); | |
3300 | dl_se->dl_runtime = attr->sched_runtime; | |
3301 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3302 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3303 | dl_se->flags = attr->sched_flags; |
332ac17e | 3304 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3305 | dl_se->dl_throttled = 0; |
3306 | dl_se->dl_new = 1; | |
5bfd126e | 3307 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3308 | } |
3309 | ||
c13db6b1 SR |
3310 | /* |
3311 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3312 | * it calls know not to change it. | |
3313 | */ | |
3314 | #define SETPARAM_POLICY -1 | |
3315 | ||
c365c292 TG |
3316 | static void __setscheduler_params(struct task_struct *p, |
3317 | const struct sched_attr *attr) | |
1da177e4 | 3318 | { |
d50dde5a DF |
3319 | int policy = attr->sched_policy; |
3320 | ||
c13db6b1 | 3321 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3322 | policy = p->policy; |
3323 | ||
1da177e4 | 3324 | p->policy = policy; |
d50dde5a | 3325 | |
aab03e05 DF |
3326 | if (dl_policy(policy)) |
3327 | __setparam_dl(p, attr); | |
39fd8fd2 | 3328 | else if (fair_policy(policy)) |
d50dde5a DF |
3329 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3330 | ||
39fd8fd2 PZ |
3331 | /* |
3332 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3333 | * !rt_policy. Always setting this ensures that things like | |
3334 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3335 | */ | |
3336 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3337 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3338 | set_load_weight(p); |
3339 | } | |
39fd8fd2 | 3340 | |
c365c292 TG |
3341 | /* Actually do priority change: must hold pi & rq lock. */ |
3342 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3343 | const struct sched_attr *attr) | |
3344 | { | |
3345 | __setscheduler_params(p, attr); | |
d50dde5a | 3346 | |
383afd09 SR |
3347 | /* |
3348 | * If we get here, there was no pi waiters boosting the | |
3349 | * task. It is safe to use the normal prio. | |
3350 | */ | |
3351 | p->prio = normal_prio(p); | |
3352 | ||
aab03e05 DF |
3353 | if (dl_prio(p->prio)) |
3354 | p->sched_class = &dl_sched_class; | |
3355 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3356 | p->sched_class = &rt_sched_class; |
3357 | else | |
3358 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3359 | } |
aab03e05 DF |
3360 | |
3361 | static void | |
3362 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3363 | { | |
3364 | struct sched_dl_entity *dl_se = &p->dl; | |
3365 | ||
3366 | attr->sched_priority = p->rt_priority; | |
3367 | attr->sched_runtime = dl_se->dl_runtime; | |
3368 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3369 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3370 | attr->sched_flags = dl_se->flags; |
3371 | } | |
3372 | ||
3373 | /* | |
3374 | * This function validates the new parameters of a -deadline task. | |
3375 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3376 | * than the runtime, as well as the period of being zero or |
332ac17e | 3377 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3378 | * user parameters are above the internal resolution of 1us (we |
3379 | * check sched_runtime only since it is always the smaller one) and | |
3380 | * below 2^63 ns (we have to check both sched_deadline and | |
3381 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3382 | */ |
3383 | static bool | |
3384 | __checkparam_dl(const struct sched_attr *attr) | |
3385 | { | |
b0827819 JL |
3386 | /* deadline != 0 */ |
3387 | if (attr->sched_deadline == 0) | |
3388 | return false; | |
3389 | ||
3390 | /* | |
3391 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3392 | * that big. | |
3393 | */ | |
3394 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3395 | return false; | |
3396 | ||
3397 | /* | |
3398 | * Since we use the MSB for wrap-around and sign issues, make | |
3399 | * sure it's not set (mind that period can be equal to zero). | |
3400 | */ | |
3401 | if (attr->sched_deadline & (1ULL << 63) || | |
3402 | attr->sched_period & (1ULL << 63)) | |
3403 | return false; | |
3404 | ||
3405 | /* runtime <= deadline <= period (if period != 0) */ | |
3406 | if ((attr->sched_period != 0 && | |
3407 | attr->sched_period < attr->sched_deadline) || | |
3408 | attr->sched_deadline < attr->sched_runtime) | |
3409 | return false; | |
3410 | ||
3411 | return true; | |
aab03e05 DF |
3412 | } |
3413 | ||
c69e8d9c DH |
3414 | /* |
3415 | * check the target process has a UID that matches the current process's | |
3416 | */ | |
3417 | static bool check_same_owner(struct task_struct *p) | |
3418 | { | |
3419 | const struct cred *cred = current_cred(), *pcred; | |
3420 | bool match; | |
3421 | ||
3422 | rcu_read_lock(); | |
3423 | pcred = __task_cred(p); | |
9c806aa0 EB |
3424 | match = (uid_eq(cred->euid, pcred->euid) || |
3425 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3426 | rcu_read_unlock(); |
3427 | return match; | |
3428 | } | |
3429 | ||
d50dde5a DF |
3430 | static int __sched_setscheduler(struct task_struct *p, |
3431 | const struct sched_attr *attr, | |
3432 | bool user) | |
1da177e4 | 3433 | { |
383afd09 SR |
3434 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3435 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3436 | int retval, oldprio, oldpolicy = -1, queued, running; |
d50dde5a | 3437 | int policy = attr->sched_policy; |
1da177e4 | 3438 | unsigned long flags; |
83ab0aa0 | 3439 | const struct sched_class *prev_class; |
70b97a7f | 3440 | struct rq *rq; |
ca94c442 | 3441 | int reset_on_fork; |
1da177e4 | 3442 | |
66e5393a SR |
3443 | /* may grab non-irq protected spin_locks */ |
3444 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3445 | recheck: |
3446 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3447 | if (policy < 0) { |
3448 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3449 | policy = oldpolicy = p->policy; |
ca94c442 | 3450 | } else { |
7479f3c9 | 3451 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3452 | |
aab03e05 DF |
3453 | if (policy != SCHED_DEADLINE && |
3454 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3455 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3456 | policy != SCHED_IDLE) | |
3457 | return -EINVAL; | |
3458 | } | |
3459 | ||
7479f3c9 PZ |
3460 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3461 | return -EINVAL; | |
3462 | ||
1da177e4 LT |
3463 | /* |
3464 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3465 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3466 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3467 | */ |
0bb040a4 | 3468 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3469 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3470 | return -EINVAL; |
aab03e05 DF |
3471 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3472 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3473 | return -EINVAL; |
3474 | ||
37e4ab3f OC |
3475 | /* |
3476 | * Allow unprivileged RT tasks to decrease priority: | |
3477 | */ | |
961ccddd | 3478 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3479 | if (fair_policy(policy)) { |
d0ea0268 | 3480 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3481 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3482 | return -EPERM; |
3483 | } | |
3484 | ||
e05606d3 | 3485 | if (rt_policy(policy)) { |
a44702e8 ON |
3486 | unsigned long rlim_rtprio = |
3487 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3488 | |
3489 | /* can't set/change the rt policy */ | |
3490 | if (policy != p->policy && !rlim_rtprio) | |
3491 | return -EPERM; | |
3492 | ||
3493 | /* can't increase priority */ | |
d50dde5a DF |
3494 | if (attr->sched_priority > p->rt_priority && |
3495 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3496 | return -EPERM; |
3497 | } | |
c02aa73b | 3498 | |
d44753b8 JL |
3499 | /* |
3500 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3501 | * (safest behavior); in the future we would like to allow | |
3502 | * unprivileged DL tasks to increase their relative deadline | |
3503 | * or reduce their runtime (both ways reducing utilization) | |
3504 | */ | |
3505 | if (dl_policy(policy)) | |
3506 | return -EPERM; | |
3507 | ||
dd41f596 | 3508 | /* |
c02aa73b DH |
3509 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3510 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3511 | */ |
c02aa73b | 3512 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3513 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3514 | return -EPERM; |
3515 | } | |
5fe1d75f | 3516 | |
37e4ab3f | 3517 | /* can't change other user's priorities */ |
c69e8d9c | 3518 | if (!check_same_owner(p)) |
37e4ab3f | 3519 | return -EPERM; |
ca94c442 LP |
3520 | |
3521 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3522 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3523 | return -EPERM; | |
37e4ab3f | 3524 | } |
1da177e4 | 3525 | |
725aad24 | 3526 | if (user) { |
b0ae1981 | 3527 | retval = security_task_setscheduler(p); |
725aad24 JF |
3528 | if (retval) |
3529 | return retval; | |
3530 | } | |
3531 | ||
b29739f9 IM |
3532 | /* |
3533 | * make sure no PI-waiters arrive (or leave) while we are | |
3534 | * changing the priority of the task: | |
0122ec5b | 3535 | * |
25985edc | 3536 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3537 | * runqueue lock must be held. |
3538 | */ | |
0122ec5b | 3539 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3540 | |
34f971f6 PZ |
3541 | /* |
3542 | * Changing the policy of the stop threads its a very bad idea | |
3543 | */ | |
3544 | if (p == rq->stop) { | |
0122ec5b | 3545 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3546 | return -EINVAL; |
3547 | } | |
3548 | ||
a51e9198 | 3549 | /* |
d6b1e911 TG |
3550 | * If not changing anything there's no need to proceed further, |
3551 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3552 | */ |
d50dde5a | 3553 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3554 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3555 | goto change; |
3556 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3557 | goto change; | |
aab03e05 DF |
3558 | if (dl_policy(policy)) |
3559 | goto change; | |
d50dde5a | 3560 | |
d6b1e911 | 3561 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3562 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3563 | return 0; |
3564 | } | |
d50dde5a | 3565 | change: |
a51e9198 | 3566 | |
dc61b1d6 | 3567 | if (user) { |
332ac17e | 3568 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3569 | /* |
3570 | * Do not allow realtime tasks into groups that have no runtime | |
3571 | * assigned. | |
3572 | */ | |
3573 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3574 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3575 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3576 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3577 | return -EPERM; |
3578 | } | |
dc61b1d6 | 3579 | #endif |
332ac17e DF |
3580 | #ifdef CONFIG_SMP |
3581 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3582 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3583 | |
3584 | /* | |
3585 | * Don't allow tasks with an affinity mask smaller than | |
3586 | * the entire root_domain to become SCHED_DEADLINE. We | |
3587 | * will also fail if there's no bandwidth available. | |
3588 | */ | |
e4099a5e PZ |
3589 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3590 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3591 | task_rq_unlock(rq, p, &flags); |
3592 | return -EPERM; | |
3593 | } | |
3594 | } | |
3595 | #endif | |
3596 | } | |
dc61b1d6 | 3597 | |
1da177e4 LT |
3598 | /* recheck policy now with rq lock held */ |
3599 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3600 | policy = oldpolicy = -1; | |
0122ec5b | 3601 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3602 | goto recheck; |
3603 | } | |
332ac17e DF |
3604 | |
3605 | /* | |
3606 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3607 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3608 | * is available. | |
3609 | */ | |
e4099a5e | 3610 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3611 | task_rq_unlock(rq, p, &flags); |
3612 | return -EBUSY; | |
3613 | } | |
3614 | ||
c365c292 TG |
3615 | p->sched_reset_on_fork = reset_on_fork; |
3616 | oldprio = p->prio; | |
3617 | ||
3618 | /* | |
3619 | * Special case for priority boosted tasks. | |
3620 | * | |
3621 | * If the new priority is lower or equal (user space view) | |
3622 | * than the current (boosted) priority, we just store the new | |
3623 | * normal parameters and do not touch the scheduler class and | |
3624 | * the runqueue. This will be done when the task deboost | |
3625 | * itself. | |
3626 | */ | |
3627 | if (rt_mutex_check_prio(p, newprio)) { | |
3628 | __setscheduler_params(p, attr); | |
3629 | task_rq_unlock(rq, p, &flags); | |
3630 | return 0; | |
3631 | } | |
3632 | ||
da0c1e65 | 3633 | queued = task_on_rq_queued(p); |
051a1d1a | 3634 | running = task_current(rq, p); |
da0c1e65 | 3635 | if (queued) |
4ca9b72b | 3636 | dequeue_task(rq, p, 0); |
0e1f3483 | 3637 | if (running) |
f3cd1c4e | 3638 | put_prev_task(rq, p); |
f6b53205 | 3639 | |
83ab0aa0 | 3640 | prev_class = p->sched_class; |
d50dde5a | 3641 | __setscheduler(rq, p, attr); |
f6b53205 | 3642 | |
0e1f3483 HS |
3643 | if (running) |
3644 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3645 | if (queued) { |
81a44c54 TG |
3646 | /* |
3647 | * We enqueue to tail when the priority of a task is | |
3648 | * increased (user space view). | |
3649 | */ | |
3650 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3651 | } | |
cb469845 | 3652 | |
da7a735e | 3653 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3654 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3655 | |
95e02ca9 TG |
3656 | rt_mutex_adjust_pi(p); |
3657 | ||
1da177e4 LT |
3658 | return 0; |
3659 | } | |
961ccddd | 3660 | |
7479f3c9 PZ |
3661 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3662 | const struct sched_param *param, bool check) | |
3663 | { | |
3664 | struct sched_attr attr = { | |
3665 | .sched_policy = policy, | |
3666 | .sched_priority = param->sched_priority, | |
3667 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3668 | }; | |
3669 | ||
c13db6b1 SR |
3670 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3671 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3672 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3673 | policy &= ~SCHED_RESET_ON_FORK; | |
3674 | attr.sched_policy = policy; | |
3675 | } | |
3676 | ||
3677 | return __sched_setscheduler(p, &attr, check); | |
3678 | } | |
961ccddd RR |
3679 | /** |
3680 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3681 | * @p: the task in question. | |
3682 | * @policy: new policy. | |
3683 | * @param: structure containing the new RT priority. | |
3684 | * | |
e69f6186 YB |
3685 | * Return: 0 on success. An error code otherwise. |
3686 | * | |
961ccddd RR |
3687 | * NOTE that the task may be already dead. |
3688 | */ | |
3689 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3690 | const struct sched_param *param) |
961ccddd | 3691 | { |
7479f3c9 | 3692 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3693 | } |
1da177e4 LT |
3694 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3695 | ||
d50dde5a DF |
3696 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3697 | { | |
3698 | return __sched_setscheduler(p, attr, true); | |
3699 | } | |
3700 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3701 | ||
961ccddd RR |
3702 | /** |
3703 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3704 | * @p: the task in question. | |
3705 | * @policy: new policy. | |
3706 | * @param: structure containing the new RT priority. | |
3707 | * | |
3708 | * Just like sched_setscheduler, only don't bother checking if the | |
3709 | * current context has permission. For example, this is needed in | |
3710 | * stop_machine(): we create temporary high priority worker threads, | |
3711 | * but our caller might not have that capability. | |
e69f6186 YB |
3712 | * |
3713 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3714 | */ |
3715 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3716 | const struct sched_param *param) |
961ccddd | 3717 | { |
7479f3c9 | 3718 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3719 | } |
3720 | ||
95cdf3b7 IM |
3721 | static int |
3722 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3723 | { |
1da177e4 LT |
3724 | struct sched_param lparam; |
3725 | struct task_struct *p; | |
36c8b586 | 3726 | int retval; |
1da177e4 LT |
3727 | |
3728 | if (!param || pid < 0) | |
3729 | return -EINVAL; | |
3730 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3731 | return -EFAULT; | |
5fe1d75f ON |
3732 | |
3733 | rcu_read_lock(); | |
3734 | retval = -ESRCH; | |
1da177e4 | 3735 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3736 | if (p != NULL) |
3737 | retval = sched_setscheduler(p, policy, &lparam); | |
3738 | rcu_read_unlock(); | |
36c8b586 | 3739 | |
1da177e4 LT |
3740 | return retval; |
3741 | } | |
3742 | ||
d50dde5a DF |
3743 | /* |
3744 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3745 | */ | |
3746 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3747 | struct sched_attr *attr) | |
3748 | { | |
3749 | u32 size; | |
3750 | int ret; | |
3751 | ||
3752 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3753 | return -EFAULT; | |
3754 | ||
3755 | /* | |
3756 | * zero the full structure, so that a short copy will be nice. | |
3757 | */ | |
3758 | memset(attr, 0, sizeof(*attr)); | |
3759 | ||
3760 | ret = get_user(size, &uattr->size); | |
3761 | if (ret) | |
3762 | return ret; | |
3763 | ||
3764 | if (size > PAGE_SIZE) /* silly large */ | |
3765 | goto err_size; | |
3766 | ||
3767 | if (!size) /* abi compat */ | |
3768 | size = SCHED_ATTR_SIZE_VER0; | |
3769 | ||
3770 | if (size < SCHED_ATTR_SIZE_VER0) | |
3771 | goto err_size; | |
3772 | ||
3773 | /* | |
3774 | * If we're handed a bigger struct than we know of, | |
3775 | * ensure all the unknown bits are 0 - i.e. new | |
3776 | * user-space does not rely on any kernel feature | |
3777 | * extensions we dont know about yet. | |
3778 | */ | |
3779 | if (size > sizeof(*attr)) { | |
3780 | unsigned char __user *addr; | |
3781 | unsigned char __user *end; | |
3782 | unsigned char val; | |
3783 | ||
3784 | addr = (void __user *)uattr + sizeof(*attr); | |
3785 | end = (void __user *)uattr + size; | |
3786 | ||
3787 | for (; addr < end; addr++) { | |
3788 | ret = get_user(val, addr); | |
3789 | if (ret) | |
3790 | return ret; | |
3791 | if (val) | |
3792 | goto err_size; | |
3793 | } | |
3794 | size = sizeof(*attr); | |
3795 | } | |
3796 | ||
3797 | ret = copy_from_user(attr, uattr, size); | |
3798 | if (ret) | |
3799 | return -EFAULT; | |
3800 | ||
3801 | /* | |
3802 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3803 | * to be strict and return an error on out-of-bounds values? | |
3804 | */ | |
75e45d51 | 3805 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3806 | |
e78c7bca | 3807 | return 0; |
d50dde5a DF |
3808 | |
3809 | err_size: | |
3810 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3811 | return -E2BIG; |
d50dde5a DF |
3812 | } |
3813 | ||
1da177e4 LT |
3814 | /** |
3815 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3816 | * @pid: the pid in question. | |
3817 | * @policy: new policy. | |
3818 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3819 | * |
3820 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3821 | */ |
5add95d4 HC |
3822 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3823 | struct sched_param __user *, param) | |
1da177e4 | 3824 | { |
c21761f1 JB |
3825 | /* negative values for policy are not valid */ |
3826 | if (policy < 0) | |
3827 | return -EINVAL; | |
3828 | ||
1da177e4 LT |
3829 | return do_sched_setscheduler(pid, policy, param); |
3830 | } | |
3831 | ||
3832 | /** | |
3833 | * sys_sched_setparam - set/change the RT priority of a thread | |
3834 | * @pid: the pid in question. | |
3835 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3836 | * |
3837 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3838 | */ |
5add95d4 | 3839 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3840 | { |
c13db6b1 | 3841 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
3842 | } |
3843 | ||
d50dde5a DF |
3844 | /** |
3845 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3846 | * @pid: the pid in question. | |
5778fccf | 3847 | * @uattr: structure containing the extended parameters. |
db66d756 | 3848 | * @flags: for future extension. |
d50dde5a | 3849 | */ |
6d35ab48 PZ |
3850 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3851 | unsigned int, flags) | |
d50dde5a DF |
3852 | { |
3853 | struct sched_attr attr; | |
3854 | struct task_struct *p; | |
3855 | int retval; | |
3856 | ||
6d35ab48 | 3857 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3858 | return -EINVAL; |
3859 | ||
143cf23d MK |
3860 | retval = sched_copy_attr(uattr, &attr); |
3861 | if (retval) | |
3862 | return retval; | |
d50dde5a | 3863 | |
b14ed2c2 | 3864 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3865 | return -EINVAL; |
d50dde5a DF |
3866 | |
3867 | rcu_read_lock(); | |
3868 | retval = -ESRCH; | |
3869 | p = find_process_by_pid(pid); | |
3870 | if (p != NULL) | |
3871 | retval = sched_setattr(p, &attr); | |
3872 | rcu_read_unlock(); | |
3873 | ||
3874 | return retval; | |
3875 | } | |
3876 | ||
1da177e4 LT |
3877 | /** |
3878 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3879 | * @pid: the pid in question. | |
e69f6186 YB |
3880 | * |
3881 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3882 | * code. | |
1da177e4 | 3883 | */ |
5add95d4 | 3884 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3885 | { |
36c8b586 | 3886 | struct task_struct *p; |
3a5c359a | 3887 | int retval; |
1da177e4 LT |
3888 | |
3889 | if (pid < 0) | |
3a5c359a | 3890 | return -EINVAL; |
1da177e4 LT |
3891 | |
3892 | retval = -ESRCH; | |
5fe85be0 | 3893 | rcu_read_lock(); |
1da177e4 LT |
3894 | p = find_process_by_pid(pid); |
3895 | if (p) { | |
3896 | retval = security_task_getscheduler(p); | |
3897 | if (!retval) | |
ca94c442 LP |
3898 | retval = p->policy |
3899 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3900 | } |
5fe85be0 | 3901 | rcu_read_unlock(); |
1da177e4 LT |
3902 | return retval; |
3903 | } | |
3904 | ||
3905 | /** | |
ca94c442 | 3906 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3907 | * @pid: the pid in question. |
3908 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3909 | * |
3910 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3911 | * code. | |
1da177e4 | 3912 | */ |
5add95d4 | 3913 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3914 | { |
ce5f7f82 | 3915 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3916 | struct task_struct *p; |
3a5c359a | 3917 | int retval; |
1da177e4 LT |
3918 | |
3919 | if (!param || pid < 0) | |
3a5c359a | 3920 | return -EINVAL; |
1da177e4 | 3921 | |
5fe85be0 | 3922 | rcu_read_lock(); |
1da177e4 LT |
3923 | p = find_process_by_pid(pid); |
3924 | retval = -ESRCH; | |
3925 | if (!p) | |
3926 | goto out_unlock; | |
3927 | ||
3928 | retval = security_task_getscheduler(p); | |
3929 | if (retval) | |
3930 | goto out_unlock; | |
3931 | ||
ce5f7f82 PZ |
3932 | if (task_has_rt_policy(p)) |
3933 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3934 | rcu_read_unlock(); |
1da177e4 LT |
3935 | |
3936 | /* | |
3937 | * This one might sleep, we cannot do it with a spinlock held ... | |
3938 | */ | |
3939 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3940 | ||
1da177e4 LT |
3941 | return retval; |
3942 | ||
3943 | out_unlock: | |
5fe85be0 | 3944 | rcu_read_unlock(); |
1da177e4 LT |
3945 | return retval; |
3946 | } | |
3947 | ||
d50dde5a DF |
3948 | static int sched_read_attr(struct sched_attr __user *uattr, |
3949 | struct sched_attr *attr, | |
3950 | unsigned int usize) | |
3951 | { | |
3952 | int ret; | |
3953 | ||
3954 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3955 | return -EFAULT; | |
3956 | ||
3957 | /* | |
3958 | * If we're handed a smaller struct than we know of, | |
3959 | * ensure all the unknown bits are 0 - i.e. old | |
3960 | * user-space does not get uncomplete information. | |
3961 | */ | |
3962 | if (usize < sizeof(*attr)) { | |
3963 | unsigned char *addr; | |
3964 | unsigned char *end; | |
3965 | ||
3966 | addr = (void *)attr + usize; | |
3967 | end = (void *)attr + sizeof(*attr); | |
3968 | ||
3969 | for (; addr < end; addr++) { | |
3970 | if (*addr) | |
22400674 | 3971 | return -EFBIG; |
d50dde5a DF |
3972 | } |
3973 | ||
3974 | attr->size = usize; | |
3975 | } | |
3976 | ||
4efbc454 | 3977 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3978 | if (ret) |
3979 | return -EFAULT; | |
3980 | ||
22400674 | 3981 | return 0; |
d50dde5a DF |
3982 | } |
3983 | ||
3984 | /** | |
aab03e05 | 3985 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3986 | * @pid: the pid in question. |
5778fccf | 3987 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3988 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3989 | * @flags: for future extension. |
d50dde5a | 3990 | */ |
6d35ab48 PZ |
3991 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3992 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3993 | { |
3994 | struct sched_attr attr = { | |
3995 | .size = sizeof(struct sched_attr), | |
3996 | }; | |
3997 | struct task_struct *p; | |
3998 | int retval; | |
3999 | ||
4000 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4001 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4002 | return -EINVAL; |
4003 | ||
4004 | rcu_read_lock(); | |
4005 | p = find_process_by_pid(pid); | |
4006 | retval = -ESRCH; | |
4007 | if (!p) | |
4008 | goto out_unlock; | |
4009 | ||
4010 | retval = security_task_getscheduler(p); | |
4011 | if (retval) | |
4012 | goto out_unlock; | |
4013 | ||
4014 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4015 | if (p->sched_reset_on_fork) |
4016 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4017 | if (task_has_dl_policy(p)) |
4018 | __getparam_dl(p, &attr); | |
4019 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4020 | attr.sched_priority = p->rt_priority; |
4021 | else | |
d0ea0268 | 4022 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4023 | |
4024 | rcu_read_unlock(); | |
4025 | ||
4026 | retval = sched_read_attr(uattr, &attr, size); | |
4027 | return retval; | |
4028 | ||
4029 | out_unlock: | |
4030 | rcu_read_unlock(); | |
4031 | return retval; | |
4032 | } | |
4033 | ||
96f874e2 | 4034 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4035 | { |
5a16f3d3 | 4036 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4037 | struct task_struct *p; |
4038 | int retval; | |
1da177e4 | 4039 | |
23f5d142 | 4040 | rcu_read_lock(); |
1da177e4 LT |
4041 | |
4042 | p = find_process_by_pid(pid); | |
4043 | if (!p) { | |
23f5d142 | 4044 | rcu_read_unlock(); |
1da177e4 LT |
4045 | return -ESRCH; |
4046 | } | |
4047 | ||
23f5d142 | 4048 | /* Prevent p going away */ |
1da177e4 | 4049 | get_task_struct(p); |
23f5d142 | 4050 | rcu_read_unlock(); |
1da177e4 | 4051 | |
14a40ffc TH |
4052 | if (p->flags & PF_NO_SETAFFINITY) { |
4053 | retval = -EINVAL; | |
4054 | goto out_put_task; | |
4055 | } | |
5a16f3d3 RR |
4056 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4057 | retval = -ENOMEM; | |
4058 | goto out_put_task; | |
4059 | } | |
4060 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4061 | retval = -ENOMEM; | |
4062 | goto out_free_cpus_allowed; | |
4063 | } | |
1da177e4 | 4064 | retval = -EPERM; |
4c44aaaf EB |
4065 | if (!check_same_owner(p)) { |
4066 | rcu_read_lock(); | |
4067 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4068 | rcu_read_unlock(); | |
16303ab2 | 4069 | goto out_free_new_mask; |
4c44aaaf EB |
4070 | } |
4071 | rcu_read_unlock(); | |
4072 | } | |
1da177e4 | 4073 | |
b0ae1981 | 4074 | retval = security_task_setscheduler(p); |
e7834f8f | 4075 | if (retval) |
16303ab2 | 4076 | goto out_free_new_mask; |
e7834f8f | 4077 | |
e4099a5e PZ |
4078 | |
4079 | cpuset_cpus_allowed(p, cpus_allowed); | |
4080 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4081 | ||
332ac17e DF |
4082 | /* |
4083 | * Since bandwidth control happens on root_domain basis, | |
4084 | * if admission test is enabled, we only admit -deadline | |
4085 | * tasks allowed to run on all the CPUs in the task's | |
4086 | * root_domain. | |
4087 | */ | |
4088 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4089 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4090 | rcu_read_lock(); | |
4091 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4092 | retval = -EBUSY; |
f1e3a093 | 4093 | rcu_read_unlock(); |
16303ab2 | 4094 | goto out_free_new_mask; |
332ac17e | 4095 | } |
f1e3a093 | 4096 | rcu_read_unlock(); |
332ac17e DF |
4097 | } |
4098 | #endif | |
49246274 | 4099 | again: |
5a16f3d3 | 4100 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4101 | |
8707d8b8 | 4102 | if (!retval) { |
5a16f3d3 RR |
4103 | cpuset_cpus_allowed(p, cpus_allowed); |
4104 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4105 | /* |
4106 | * We must have raced with a concurrent cpuset | |
4107 | * update. Just reset the cpus_allowed to the | |
4108 | * cpuset's cpus_allowed | |
4109 | */ | |
5a16f3d3 | 4110 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4111 | goto again; |
4112 | } | |
4113 | } | |
16303ab2 | 4114 | out_free_new_mask: |
5a16f3d3 RR |
4115 | free_cpumask_var(new_mask); |
4116 | out_free_cpus_allowed: | |
4117 | free_cpumask_var(cpus_allowed); | |
4118 | out_put_task: | |
1da177e4 | 4119 | put_task_struct(p); |
1da177e4 LT |
4120 | return retval; |
4121 | } | |
4122 | ||
4123 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4124 | struct cpumask *new_mask) |
1da177e4 | 4125 | { |
96f874e2 RR |
4126 | if (len < cpumask_size()) |
4127 | cpumask_clear(new_mask); | |
4128 | else if (len > cpumask_size()) | |
4129 | len = cpumask_size(); | |
4130 | ||
1da177e4 LT |
4131 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4132 | } | |
4133 | ||
4134 | /** | |
4135 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4136 | * @pid: pid of the process | |
4137 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4138 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4139 | * |
4140 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4141 | */ |
5add95d4 HC |
4142 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4143 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4144 | { |
5a16f3d3 | 4145 | cpumask_var_t new_mask; |
1da177e4 LT |
4146 | int retval; |
4147 | ||
5a16f3d3 RR |
4148 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4149 | return -ENOMEM; | |
1da177e4 | 4150 | |
5a16f3d3 RR |
4151 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4152 | if (retval == 0) | |
4153 | retval = sched_setaffinity(pid, new_mask); | |
4154 | free_cpumask_var(new_mask); | |
4155 | return retval; | |
1da177e4 LT |
4156 | } |
4157 | ||
96f874e2 | 4158 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4159 | { |
36c8b586 | 4160 | struct task_struct *p; |
31605683 | 4161 | unsigned long flags; |
1da177e4 | 4162 | int retval; |
1da177e4 | 4163 | |
23f5d142 | 4164 | rcu_read_lock(); |
1da177e4 LT |
4165 | |
4166 | retval = -ESRCH; | |
4167 | p = find_process_by_pid(pid); | |
4168 | if (!p) | |
4169 | goto out_unlock; | |
4170 | ||
e7834f8f DQ |
4171 | retval = security_task_getscheduler(p); |
4172 | if (retval) | |
4173 | goto out_unlock; | |
4174 | ||
013fdb80 | 4175 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4176 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4177 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4178 | |
4179 | out_unlock: | |
23f5d142 | 4180 | rcu_read_unlock(); |
1da177e4 | 4181 | |
9531b62f | 4182 | return retval; |
1da177e4 LT |
4183 | } |
4184 | ||
4185 | /** | |
4186 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4187 | * @pid: pid of the process | |
4188 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4189 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4190 | * |
4191 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4192 | */ |
5add95d4 HC |
4193 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4194 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4195 | { |
4196 | int ret; | |
f17c8607 | 4197 | cpumask_var_t mask; |
1da177e4 | 4198 | |
84fba5ec | 4199 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4200 | return -EINVAL; |
4201 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4202 | return -EINVAL; |
4203 | ||
f17c8607 RR |
4204 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4205 | return -ENOMEM; | |
1da177e4 | 4206 | |
f17c8607 RR |
4207 | ret = sched_getaffinity(pid, mask); |
4208 | if (ret == 0) { | |
8bc037fb | 4209 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4210 | |
4211 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4212 | ret = -EFAULT; |
4213 | else | |
cd3d8031 | 4214 | ret = retlen; |
f17c8607 RR |
4215 | } |
4216 | free_cpumask_var(mask); | |
1da177e4 | 4217 | |
f17c8607 | 4218 | return ret; |
1da177e4 LT |
4219 | } |
4220 | ||
4221 | /** | |
4222 | * sys_sched_yield - yield the current processor to other threads. | |
4223 | * | |
dd41f596 IM |
4224 | * This function yields the current CPU to other tasks. If there are no |
4225 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4226 | * |
4227 | * Return: 0. | |
1da177e4 | 4228 | */ |
5add95d4 | 4229 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4230 | { |
70b97a7f | 4231 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4232 | |
2d72376b | 4233 | schedstat_inc(rq, yld_count); |
4530d7ab | 4234 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4235 | |
4236 | /* | |
4237 | * Since we are going to call schedule() anyway, there's | |
4238 | * no need to preempt or enable interrupts: | |
4239 | */ | |
4240 | __release(rq->lock); | |
8a25d5de | 4241 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4242 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4243 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4244 | |
4245 | schedule(); | |
4246 | ||
4247 | return 0; | |
4248 | } | |
4249 | ||
e7b38404 | 4250 | static void __cond_resched(void) |
1da177e4 | 4251 | { |
bdb43806 | 4252 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4253 | __schedule(); |
bdb43806 | 4254 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4255 | } |
4256 | ||
02b67cc3 | 4257 | int __sched _cond_resched(void) |
1da177e4 | 4258 | { |
d86ee480 | 4259 | if (should_resched()) { |
1da177e4 LT |
4260 | __cond_resched(); |
4261 | return 1; | |
4262 | } | |
4263 | return 0; | |
4264 | } | |
02b67cc3 | 4265 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4266 | |
4267 | /* | |
613afbf8 | 4268 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4269 | * call schedule, and on return reacquire the lock. |
4270 | * | |
41a2d6cf | 4271 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4272 | * operations here to prevent schedule() from being called twice (once via |
4273 | * spin_unlock(), once by hand). | |
4274 | */ | |
613afbf8 | 4275 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4276 | { |
d86ee480 | 4277 | int resched = should_resched(); |
6df3cecb JK |
4278 | int ret = 0; |
4279 | ||
f607c668 PZ |
4280 | lockdep_assert_held(lock); |
4281 | ||
4a81e832 | 4282 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4283 | spin_unlock(lock); |
d86ee480 | 4284 | if (resched) |
95c354fe NP |
4285 | __cond_resched(); |
4286 | else | |
4287 | cpu_relax(); | |
6df3cecb | 4288 | ret = 1; |
1da177e4 | 4289 | spin_lock(lock); |
1da177e4 | 4290 | } |
6df3cecb | 4291 | return ret; |
1da177e4 | 4292 | } |
613afbf8 | 4293 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4294 | |
613afbf8 | 4295 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4296 | { |
4297 | BUG_ON(!in_softirq()); | |
4298 | ||
d86ee480 | 4299 | if (should_resched()) { |
98d82567 | 4300 | local_bh_enable(); |
1da177e4 LT |
4301 | __cond_resched(); |
4302 | local_bh_disable(); | |
4303 | return 1; | |
4304 | } | |
4305 | return 0; | |
4306 | } | |
613afbf8 | 4307 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4308 | |
1da177e4 LT |
4309 | /** |
4310 | * yield - yield the current processor to other threads. | |
4311 | * | |
8e3fabfd PZ |
4312 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4313 | * | |
4314 | * The scheduler is at all times free to pick the calling task as the most | |
4315 | * eligible task to run, if removing the yield() call from your code breaks | |
4316 | * it, its already broken. | |
4317 | * | |
4318 | * Typical broken usage is: | |
4319 | * | |
4320 | * while (!event) | |
4321 | * yield(); | |
4322 | * | |
4323 | * where one assumes that yield() will let 'the other' process run that will | |
4324 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4325 | * happen. Never use yield() as a progress guarantee!! | |
4326 | * | |
4327 | * If you want to use yield() to wait for something, use wait_event(). | |
4328 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4329 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4330 | */ |
4331 | void __sched yield(void) | |
4332 | { | |
4333 | set_current_state(TASK_RUNNING); | |
4334 | sys_sched_yield(); | |
4335 | } | |
1da177e4 LT |
4336 | EXPORT_SYMBOL(yield); |
4337 | ||
d95f4122 MG |
4338 | /** |
4339 | * yield_to - yield the current processor to another thread in | |
4340 | * your thread group, or accelerate that thread toward the | |
4341 | * processor it's on. | |
16addf95 RD |
4342 | * @p: target task |
4343 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4344 | * |
4345 | * It's the caller's job to ensure that the target task struct | |
4346 | * can't go away on us before we can do any checks. | |
4347 | * | |
e69f6186 | 4348 | * Return: |
7b270f60 PZ |
4349 | * true (>0) if we indeed boosted the target task. |
4350 | * false (0) if we failed to boost the target. | |
4351 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4352 | */ |
fa93384f | 4353 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4354 | { |
4355 | struct task_struct *curr = current; | |
4356 | struct rq *rq, *p_rq; | |
4357 | unsigned long flags; | |
c3c18640 | 4358 | int yielded = 0; |
d95f4122 MG |
4359 | |
4360 | local_irq_save(flags); | |
4361 | rq = this_rq(); | |
4362 | ||
4363 | again: | |
4364 | p_rq = task_rq(p); | |
7b270f60 PZ |
4365 | /* |
4366 | * If we're the only runnable task on the rq and target rq also | |
4367 | * has only one task, there's absolutely no point in yielding. | |
4368 | */ | |
4369 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4370 | yielded = -ESRCH; | |
4371 | goto out_irq; | |
4372 | } | |
4373 | ||
d95f4122 | 4374 | double_rq_lock(rq, p_rq); |
39e24d8f | 4375 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4376 | double_rq_unlock(rq, p_rq); |
4377 | goto again; | |
4378 | } | |
4379 | ||
4380 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4381 | goto out_unlock; |
d95f4122 MG |
4382 | |
4383 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4384 | goto out_unlock; |
d95f4122 MG |
4385 | |
4386 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4387 | goto out_unlock; |
d95f4122 MG |
4388 | |
4389 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4390 | if (yielded) { |
d95f4122 | 4391 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4392 | /* |
4393 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4394 | * fairness. | |
4395 | */ | |
4396 | if (preempt && rq != p_rq) | |
8875125e | 4397 | resched_curr(p_rq); |
6d1cafd8 | 4398 | } |
d95f4122 | 4399 | |
7b270f60 | 4400 | out_unlock: |
d95f4122 | 4401 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4402 | out_irq: |
d95f4122 MG |
4403 | local_irq_restore(flags); |
4404 | ||
7b270f60 | 4405 | if (yielded > 0) |
d95f4122 MG |
4406 | schedule(); |
4407 | ||
4408 | return yielded; | |
4409 | } | |
4410 | EXPORT_SYMBOL_GPL(yield_to); | |
4411 | ||
1da177e4 | 4412 | /* |
41a2d6cf | 4413 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4414 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4415 | */ |
4416 | void __sched io_schedule(void) | |
4417 | { | |
54d35f29 | 4418 | struct rq *rq = raw_rq(); |
1da177e4 | 4419 | |
0ff92245 | 4420 | delayacct_blkio_start(); |
1da177e4 | 4421 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4422 | blk_flush_plug(current); |
8f0dfc34 | 4423 | current->in_iowait = 1; |
1da177e4 | 4424 | schedule(); |
8f0dfc34 | 4425 | current->in_iowait = 0; |
1da177e4 | 4426 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4427 | delayacct_blkio_end(); |
1da177e4 | 4428 | } |
1da177e4 LT |
4429 | EXPORT_SYMBOL(io_schedule); |
4430 | ||
4431 | long __sched io_schedule_timeout(long timeout) | |
4432 | { | |
54d35f29 | 4433 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4434 | long ret; |
4435 | ||
0ff92245 | 4436 | delayacct_blkio_start(); |
1da177e4 | 4437 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4438 | blk_flush_plug(current); |
8f0dfc34 | 4439 | current->in_iowait = 1; |
1da177e4 | 4440 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4441 | current->in_iowait = 0; |
1da177e4 | 4442 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4443 | delayacct_blkio_end(); |
1da177e4 LT |
4444 | return ret; |
4445 | } | |
4446 | ||
4447 | /** | |
4448 | * sys_sched_get_priority_max - return maximum RT priority. | |
4449 | * @policy: scheduling class. | |
4450 | * | |
e69f6186 YB |
4451 | * Return: On success, this syscall returns the maximum |
4452 | * rt_priority that can be used by a given scheduling class. | |
4453 | * On failure, a negative error code is returned. | |
1da177e4 | 4454 | */ |
5add95d4 | 4455 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4456 | { |
4457 | int ret = -EINVAL; | |
4458 | ||
4459 | switch (policy) { | |
4460 | case SCHED_FIFO: | |
4461 | case SCHED_RR: | |
4462 | ret = MAX_USER_RT_PRIO-1; | |
4463 | break; | |
aab03e05 | 4464 | case SCHED_DEADLINE: |
1da177e4 | 4465 | case SCHED_NORMAL: |
b0a9499c | 4466 | case SCHED_BATCH: |
dd41f596 | 4467 | case SCHED_IDLE: |
1da177e4 LT |
4468 | ret = 0; |
4469 | break; | |
4470 | } | |
4471 | return ret; | |
4472 | } | |
4473 | ||
4474 | /** | |
4475 | * sys_sched_get_priority_min - return minimum RT priority. | |
4476 | * @policy: scheduling class. | |
4477 | * | |
e69f6186 YB |
4478 | * Return: On success, this syscall returns the minimum |
4479 | * rt_priority that can be used by a given scheduling class. | |
4480 | * On failure, a negative error code is returned. | |
1da177e4 | 4481 | */ |
5add95d4 | 4482 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4483 | { |
4484 | int ret = -EINVAL; | |
4485 | ||
4486 | switch (policy) { | |
4487 | case SCHED_FIFO: | |
4488 | case SCHED_RR: | |
4489 | ret = 1; | |
4490 | break; | |
aab03e05 | 4491 | case SCHED_DEADLINE: |
1da177e4 | 4492 | case SCHED_NORMAL: |
b0a9499c | 4493 | case SCHED_BATCH: |
dd41f596 | 4494 | case SCHED_IDLE: |
1da177e4 LT |
4495 | ret = 0; |
4496 | } | |
4497 | return ret; | |
4498 | } | |
4499 | ||
4500 | /** | |
4501 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4502 | * @pid: pid of the process. | |
4503 | * @interval: userspace pointer to the timeslice value. | |
4504 | * | |
4505 | * this syscall writes the default timeslice value of a given process | |
4506 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4507 | * |
4508 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4509 | * an error code. | |
1da177e4 | 4510 | */ |
17da2bd9 | 4511 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4512 | struct timespec __user *, interval) |
1da177e4 | 4513 | { |
36c8b586 | 4514 | struct task_struct *p; |
a4ec24b4 | 4515 | unsigned int time_slice; |
dba091b9 TG |
4516 | unsigned long flags; |
4517 | struct rq *rq; | |
3a5c359a | 4518 | int retval; |
1da177e4 | 4519 | struct timespec t; |
1da177e4 LT |
4520 | |
4521 | if (pid < 0) | |
3a5c359a | 4522 | return -EINVAL; |
1da177e4 LT |
4523 | |
4524 | retval = -ESRCH; | |
1a551ae7 | 4525 | rcu_read_lock(); |
1da177e4 LT |
4526 | p = find_process_by_pid(pid); |
4527 | if (!p) | |
4528 | goto out_unlock; | |
4529 | ||
4530 | retval = security_task_getscheduler(p); | |
4531 | if (retval) | |
4532 | goto out_unlock; | |
4533 | ||
dba091b9 | 4534 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4535 | time_slice = 0; |
4536 | if (p->sched_class->get_rr_interval) | |
4537 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4538 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4539 | |
1a551ae7 | 4540 | rcu_read_unlock(); |
a4ec24b4 | 4541 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4542 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4543 | return retval; |
3a5c359a | 4544 | |
1da177e4 | 4545 | out_unlock: |
1a551ae7 | 4546 | rcu_read_unlock(); |
1da177e4 LT |
4547 | return retval; |
4548 | } | |
4549 | ||
7c731e0a | 4550 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4551 | |
82a1fcb9 | 4552 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4553 | { |
1da177e4 | 4554 | unsigned long free = 0; |
4e79752c | 4555 | int ppid; |
36c8b586 | 4556 | unsigned state; |
1da177e4 | 4557 | |
1da177e4 | 4558 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4559 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4560 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4561 | #if BITS_PER_LONG == 32 |
1da177e4 | 4562 | if (state == TASK_RUNNING) |
3df0fc5b | 4563 | printk(KERN_CONT " running "); |
1da177e4 | 4564 | else |
3df0fc5b | 4565 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4566 | #else |
4567 | if (state == TASK_RUNNING) | |
3df0fc5b | 4568 | printk(KERN_CONT " running task "); |
1da177e4 | 4569 | else |
3df0fc5b | 4570 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4571 | #endif |
4572 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4573 | free = stack_not_used(p); |
1da177e4 | 4574 | #endif |
4e79752c PM |
4575 | rcu_read_lock(); |
4576 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4577 | rcu_read_unlock(); | |
3df0fc5b | 4578 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4579 | task_pid_nr(p), ppid, |
aa47b7e0 | 4580 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4581 | |
3d1cb205 | 4582 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4583 | show_stack(p, NULL); |
1da177e4 LT |
4584 | } |
4585 | ||
e59e2ae2 | 4586 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4587 | { |
36c8b586 | 4588 | struct task_struct *g, *p; |
1da177e4 | 4589 | |
4bd77321 | 4590 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4591 | printk(KERN_INFO |
4592 | " task PC stack pid father\n"); | |
1da177e4 | 4593 | #else |
3df0fc5b PZ |
4594 | printk(KERN_INFO |
4595 | " task PC stack pid father\n"); | |
1da177e4 | 4596 | #endif |
510f5acc | 4597 | rcu_read_lock(); |
5d07f420 | 4598 | for_each_process_thread(g, p) { |
1da177e4 LT |
4599 | /* |
4600 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4601 | * console might take a lot of time: |
1da177e4 LT |
4602 | */ |
4603 | touch_nmi_watchdog(); | |
39bc89fd | 4604 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4605 | sched_show_task(p); |
5d07f420 | 4606 | } |
1da177e4 | 4607 | |
04c9167f JF |
4608 | touch_all_softlockup_watchdogs(); |
4609 | ||
dd41f596 IM |
4610 | #ifdef CONFIG_SCHED_DEBUG |
4611 | sysrq_sched_debug_show(); | |
4612 | #endif | |
510f5acc | 4613 | rcu_read_unlock(); |
e59e2ae2 IM |
4614 | /* |
4615 | * Only show locks if all tasks are dumped: | |
4616 | */ | |
93335a21 | 4617 | if (!state_filter) |
e59e2ae2 | 4618 | debug_show_all_locks(); |
1da177e4 LT |
4619 | } |
4620 | ||
0db0628d | 4621 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4622 | { |
dd41f596 | 4623 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4624 | } |
4625 | ||
f340c0d1 IM |
4626 | /** |
4627 | * init_idle - set up an idle thread for a given CPU | |
4628 | * @idle: task in question | |
4629 | * @cpu: cpu the idle task belongs to | |
4630 | * | |
4631 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4632 | * flag, to make booting more robust. | |
4633 | */ | |
0db0628d | 4634 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4635 | { |
70b97a7f | 4636 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4637 | unsigned long flags; |
4638 | ||
05fa785c | 4639 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4640 | |
5e1576ed | 4641 | __sched_fork(0, idle); |
06b83b5f | 4642 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4643 | idle->se.exec_start = sched_clock(); |
4644 | ||
1e1b6c51 | 4645 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4646 | /* |
4647 | * We're having a chicken and egg problem, even though we are | |
4648 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4649 | * lockdep check in task_group() will fail. | |
4650 | * | |
4651 | * Similar case to sched_fork(). / Alternatively we could | |
4652 | * use task_rq_lock() here and obtain the other rq->lock. | |
4653 | * | |
4654 | * Silence PROVE_RCU | |
4655 | */ | |
4656 | rcu_read_lock(); | |
dd41f596 | 4657 | __set_task_cpu(idle, cpu); |
6506cf6c | 4658 | rcu_read_unlock(); |
1da177e4 | 4659 | |
1da177e4 | 4660 | rq->curr = rq->idle = idle; |
da0c1e65 | 4661 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4662 | #if defined(CONFIG_SMP) |
4663 | idle->on_cpu = 1; | |
4866cde0 | 4664 | #endif |
05fa785c | 4665 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4666 | |
4667 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4668 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4669 | |
dd41f596 IM |
4670 | /* |
4671 | * The idle tasks have their own, simple scheduling class: | |
4672 | */ | |
4673 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4674 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4675 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4676 | #if defined(CONFIG_SMP) |
4677 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4678 | #endif | |
19978ca6 IM |
4679 | } |
4680 | ||
1da177e4 | 4681 | #ifdef CONFIG_SMP |
a15b12ac KT |
4682 | /* |
4683 | * move_queued_task - move a queued task to new rq. | |
4684 | * | |
4685 | * Returns (locked) new rq. Old rq's lock is released. | |
4686 | */ | |
4687 | static struct rq *move_queued_task(struct task_struct *p, int new_cpu) | |
4688 | { | |
4689 | struct rq *rq = task_rq(p); | |
4690 | ||
4691 | lockdep_assert_held(&rq->lock); | |
4692 | ||
4693 | dequeue_task(rq, p, 0); | |
4694 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
4695 | set_task_cpu(p, new_cpu); | |
4696 | raw_spin_unlock(&rq->lock); | |
4697 | ||
4698 | rq = cpu_rq(new_cpu); | |
4699 | ||
4700 | raw_spin_lock(&rq->lock); | |
4701 | BUG_ON(task_cpu(p) != new_cpu); | |
4702 | p->on_rq = TASK_ON_RQ_QUEUED; | |
4703 | enqueue_task(rq, p, 0); | |
4704 | check_preempt_curr(rq, p, 0); | |
4705 | ||
4706 | return rq; | |
4707 | } | |
4708 | ||
1e1b6c51 KM |
4709 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4710 | { | |
4711 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4712 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4713 | |
4714 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4715 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4716 | } |
4717 | ||
1da177e4 LT |
4718 | /* |
4719 | * This is how migration works: | |
4720 | * | |
969c7921 TH |
4721 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4722 | * stop_one_cpu(). | |
4723 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4724 | * off the CPU) | |
4725 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4726 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4727 | * it and puts it into the right queue. |
969c7921 TH |
4728 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4729 | * is done. | |
1da177e4 LT |
4730 | */ |
4731 | ||
4732 | /* | |
4733 | * Change a given task's CPU affinity. Migrate the thread to a | |
4734 | * proper CPU and schedule it away if the CPU it's executing on | |
4735 | * is removed from the allowed bitmask. | |
4736 | * | |
4737 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4738 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4739 | * call is not atomic; no spinlocks may be held. |
4740 | */ | |
96f874e2 | 4741 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4742 | { |
4743 | unsigned long flags; | |
70b97a7f | 4744 | struct rq *rq; |
969c7921 | 4745 | unsigned int dest_cpu; |
48f24c4d | 4746 | int ret = 0; |
1da177e4 LT |
4747 | |
4748 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4749 | |
db44fc01 YZ |
4750 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4751 | goto out; | |
4752 | ||
6ad4c188 | 4753 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4754 | ret = -EINVAL; |
4755 | goto out; | |
4756 | } | |
4757 | ||
1e1b6c51 | 4758 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4759 | |
1da177e4 | 4760 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4761 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4762 | goto out; |
4763 | ||
969c7921 | 4764 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
a15b12ac | 4765 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
969c7921 | 4766 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4767 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4768 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4769 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4770 | tlb_migrate_finish(p->mm); |
4771 | return 0; | |
a15b12ac KT |
4772 | } else if (task_on_rq_queued(p)) |
4773 | rq = move_queued_task(p, dest_cpu); | |
1da177e4 | 4774 | out: |
0122ec5b | 4775 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4776 | |
1da177e4 LT |
4777 | return ret; |
4778 | } | |
cd8ba7cd | 4779 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4780 | |
4781 | /* | |
41a2d6cf | 4782 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4783 | * this because either it can't run here any more (set_cpus_allowed() |
4784 | * away from this CPU, or CPU going down), or because we're | |
4785 | * attempting to rebalance this task on exec (sched_exec). | |
4786 | * | |
4787 | * So we race with normal scheduler movements, but that's OK, as long | |
4788 | * as the task is no longer on this CPU. | |
efc30814 KK |
4789 | * |
4790 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4791 | */ |
efc30814 | 4792 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4793 | { |
a1e01829 | 4794 | struct rq *rq; |
e2912009 | 4795 | int ret = 0; |
1da177e4 | 4796 | |
e761b772 | 4797 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4798 | return ret; |
1da177e4 | 4799 | |
a1e01829 | 4800 | rq = cpu_rq(src_cpu); |
1da177e4 | 4801 | |
0122ec5b | 4802 | raw_spin_lock(&p->pi_lock); |
a1e01829 | 4803 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
4804 | /* Already moved. */ |
4805 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4806 | goto done; |
a1e01829 | 4807 | |
1da177e4 | 4808 | /* Affinity changed (again). */ |
fa17b507 | 4809 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4810 | goto fail; |
1da177e4 | 4811 | |
e2912009 PZ |
4812 | /* |
4813 | * If we're not on a rq, the next wake-up will ensure we're | |
4814 | * placed properly. | |
4815 | */ | |
a15b12ac KT |
4816 | if (task_on_rq_queued(p)) |
4817 | rq = move_queued_task(p, dest_cpu); | |
b1e38734 | 4818 | done: |
efc30814 | 4819 | ret = 1; |
b1e38734 | 4820 | fail: |
a1e01829 | 4821 | raw_spin_unlock(&rq->lock); |
0122ec5b | 4822 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4823 | return ret; |
1da177e4 LT |
4824 | } |
4825 | ||
e6628d5b MG |
4826 | #ifdef CONFIG_NUMA_BALANCING |
4827 | /* Migrate current task p to target_cpu */ | |
4828 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4829 | { | |
4830 | struct migration_arg arg = { p, target_cpu }; | |
4831 | int curr_cpu = task_cpu(p); | |
4832 | ||
4833 | if (curr_cpu == target_cpu) | |
4834 | return 0; | |
4835 | ||
4836 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4837 | return -EINVAL; | |
4838 | ||
4839 | /* TODO: This is not properly updating schedstats */ | |
4840 | ||
286549dc | 4841 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4842 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4843 | } | |
0ec8aa00 PZ |
4844 | |
4845 | /* | |
4846 | * Requeue a task on a given node and accurately track the number of NUMA | |
4847 | * tasks on the runqueues | |
4848 | */ | |
4849 | void sched_setnuma(struct task_struct *p, int nid) | |
4850 | { | |
4851 | struct rq *rq; | |
4852 | unsigned long flags; | |
da0c1e65 | 4853 | bool queued, running; |
0ec8aa00 PZ |
4854 | |
4855 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 4856 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
4857 | running = task_current(rq, p); |
4858 | ||
da0c1e65 | 4859 | if (queued) |
0ec8aa00 PZ |
4860 | dequeue_task(rq, p, 0); |
4861 | if (running) | |
f3cd1c4e | 4862 | put_prev_task(rq, p); |
0ec8aa00 PZ |
4863 | |
4864 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4865 | |
4866 | if (running) | |
4867 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4868 | if (queued) |
0ec8aa00 PZ |
4869 | enqueue_task(rq, p, 0); |
4870 | task_rq_unlock(rq, p, &flags); | |
4871 | } | |
e6628d5b MG |
4872 | #endif |
4873 | ||
1da177e4 | 4874 | /* |
969c7921 TH |
4875 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4876 | * and performs thread migration by bumping thread off CPU then | |
4877 | * 'pushing' onto another runqueue. | |
1da177e4 | 4878 | */ |
969c7921 | 4879 | static int migration_cpu_stop(void *data) |
1da177e4 | 4880 | { |
969c7921 | 4881 | struct migration_arg *arg = data; |
f7b4cddc | 4882 | |
969c7921 TH |
4883 | /* |
4884 | * The original target cpu might have gone down and we might | |
4885 | * be on another cpu but it doesn't matter. | |
4886 | */ | |
f7b4cddc | 4887 | local_irq_disable(); |
5cd038f5 LJ |
4888 | /* |
4889 | * We need to explicitly wake pending tasks before running | |
4890 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
4891 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
4892 | */ | |
4893 | sched_ttwu_pending(); | |
969c7921 | 4894 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4895 | local_irq_enable(); |
1da177e4 | 4896 | return 0; |
f7b4cddc ON |
4897 | } |
4898 | ||
1da177e4 | 4899 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4900 | |
054b9108 | 4901 | /* |
48c5ccae PZ |
4902 | * Ensures that the idle task is using init_mm right before its cpu goes |
4903 | * offline. | |
054b9108 | 4904 | */ |
48c5ccae | 4905 | void idle_task_exit(void) |
1da177e4 | 4906 | { |
48c5ccae | 4907 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4908 | |
48c5ccae | 4909 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4910 | |
a53efe5f | 4911 | if (mm != &init_mm) { |
48c5ccae | 4912 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4913 | finish_arch_post_lock_switch(); |
4914 | } | |
48c5ccae | 4915 | mmdrop(mm); |
1da177e4 LT |
4916 | } |
4917 | ||
4918 | /* | |
5d180232 PZ |
4919 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4920 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4921 | * nr_active count is stable. | |
4922 | * | |
4923 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4924 | */ |
5d180232 | 4925 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4926 | { |
5d180232 PZ |
4927 | long delta = calc_load_fold_active(rq); |
4928 | if (delta) | |
4929 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4930 | } |
4931 | ||
3f1d2a31 PZ |
4932 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4933 | { | |
4934 | } | |
4935 | ||
4936 | static const struct sched_class fake_sched_class = { | |
4937 | .put_prev_task = put_prev_task_fake, | |
4938 | }; | |
4939 | ||
4940 | static struct task_struct fake_task = { | |
4941 | /* | |
4942 | * Avoid pull_{rt,dl}_task() | |
4943 | */ | |
4944 | .prio = MAX_PRIO + 1, | |
4945 | .sched_class = &fake_sched_class, | |
4946 | }; | |
4947 | ||
48f24c4d | 4948 | /* |
48c5ccae PZ |
4949 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4950 | * try_to_wake_up()->select_task_rq(). | |
4951 | * | |
4952 | * Called with rq->lock held even though we'er in stop_machine() and | |
4953 | * there's no concurrency possible, we hold the required locks anyway | |
4954 | * because of lock validation efforts. | |
1da177e4 | 4955 | */ |
48c5ccae | 4956 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4957 | { |
70b97a7f | 4958 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4959 | struct task_struct *next, *stop = rq->stop; |
4960 | int dest_cpu; | |
1da177e4 LT |
4961 | |
4962 | /* | |
48c5ccae PZ |
4963 | * Fudge the rq selection such that the below task selection loop |
4964 | * doesn't get stuck on the currently eligible stop task. | |
4965 | * | |
4966 | * We're currently inside stop_machine() and the rq is either stuck | |
4967 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4968 | * either way we should never end up calling schedule() until we're | |
4969 | * done here. | |
1da177e4 | 4970 | */ |
48c5ccae | 4971 | rq->stop = NULL; |
48f24c4d | 4972 | |
77bd3970 FW |
4973 | /* |
4974 | * put_prev_task() and pick_next_task() sched | |
4975 | * class method both need to have an up-to-date | |
4976 | * value of rq->clock[_task] | |
4977 | */ | |
4978 | update_rq_clock(rq); | |
4979 | ||
dd41f596 | 4980 | for ( ; ; ) { |
48c5ccae PZ |
4981 | /* |
4982 | * There's this thread running, bail when that's the only | |
4983 | * remaining thread. | |
4984 | */ | |
4985 | if (rq->nr_running == 1) | |
dd41f596 | 4986 | break; |
48c5ccae | 4987 | |
3f1d2a31 | 4988 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4989 | BUG_ON(!next); |
79c53799 | 4990 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4991 | |
48c5ccae PZ |
4992 | /* Find suitable destination for @next, with force if needed. */ |
4993 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4994 | raw_spin_unlock(&rq->lock); | |
4995 | ||
4996 | __migrate_task(next, dead_cpu, dest_cpu); | |
4997 | ||
4998 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4999 | } |
dce48a84 | 5000 | |
48c5ccae | 5001 | rq->stop = stop; |
dce48a84 | 5002 | } |
48c5ccae | 5003 | |
1da177e4 LT |
5004 | #endif /* CONFIG_HOTPLUG_CPU */ |
5005 | ||
e692ab53 NP |
5006 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5007 | ||
5008 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5009 | { |
5010 | .procname = "sched_domain", | |
c57baf1e | 5011 | .mode = 0555, |
e0361851 | 5012 | }, |
56992309 | 5013 | {} |
e692ab53 NP |
5014 | }; |
5015 | ||
5016 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5017 | { |
5018 | .procname = "kernel", | |
c57baf1e | 5019 | .mode = 0555, |
e0361851 AD |
5020 | .child = sd_ctl_dir, |
5021 | }, | |
56992309 | 5022 | {} |
e692ab53 NP |
5023 | }; |
5024 | ||
5025 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5026 | { | |
5027 | struct ctl_table *entry = | |
5cf9f062 | 5028 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5029 | |
e692ab53 NP |
5030 | return entry; |
5031 | } | |
5032 | ||
6382bc90 MM |
5033 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5034 | { | |
cd790076 | 5035 | struct ctl_table *entry; |
6382bc90 | 5036 | |
cd790076 MM |
5037 | /* |
5038 | * In the intermediate directories, both the child directory and | |
5039 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5040 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5041 | * static strings and all have proc handlers. |
5042 | */ | |
5043 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5044 | if (entry->child) |
5045 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5046 | if (entry->proc_handler == NULL) |
5047 | kfree(entry->procname); | |
5048 | } | |
6382bc90 MM |
5049 | |
5050 | kfree(*tablep); | |
5051 | *tablep = NULL; | |
5052 | } | |
5053 | ||
201c373e | 5054 | static int min_load_idx = 0; |
fd9b86d3 | 5055 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5056 | |
e692ab53 | 5057 | static void |
e0361851 | 5058 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5059 | const char *procname, void *data, int maxlen, |
201c373e NK |
5060 | umode_t mode, proc_handler *proc_handler, |
5061 | bool load_idx) | |
e692ab53 | 5062 | { |
e692ab53 NP |
5063 | entry->procname = procname; |
5064 | entry->data = data; | |
5065 | entry->maxlen = maxlen; | |
5066 | entry->mode = mode; | |
5067 | entry->proc_handler = proc_handler; | |
201c373e NK |
5068 | |
5069 | if (load_idx) { | |
5070 | entry->extra1 = &min_load_idx; | |
5071 | entry->extra2 = &max_load_idx; | |
5072 | } | |
e692ab53 NP |
5073 | } |
5074 | ||
5075 | static struct ctl_table * | |
5076 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5077 | { | |
37e6bae8 | 5078 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 5079 | |
ad1cdc1d MM |
5080 | if (table == NULL) |
5081 | return NULL; | |
5082 | ||
e0361851 | 5083 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5084 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5085 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5086 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5087 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5088 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5089 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5090 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5091 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5092 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5093 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5094 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5095 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5096 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5097 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5098 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5099 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5100 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5101 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5102 | &sd->cache_nice_tries, |
201c373e | 5103 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5104 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5105 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
5106 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5107 | &sd->max_newidle_lb_cost, | |
5108 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5109 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5110 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5111 | /* &table[13] is terminator */ |
e692ab53 NP |
5112 | |
5113 | return table; | |
5114 | } | |
5115 | ||
be7002e6 | 5116 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5117 | { |
5118 | struct ctl_table *entry, *table; | |
5119 | struct sched_domain *sd; | |
5120 | int domain_num = 0, i; | |
5121 | char buf[32]; | |
5122 | ||
5123 | for_each_domain(cpu, sd) | |
5124 | domain_num++; | |
5125 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5126 | if (table == NULL) |
5127 | return NULL; | |
e692ab53 NP |
5128 | |
5129 | i = 0; | |
5130 | for_each_domain(cpu, sd) { | |
5131 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5132 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5133 | entry->mode = 0555; |
e692ab53 NP |
5134 | entry->child = sd_alloc_ctl_domain_table(sd); |
5135 | entry++; | |
5136 | i++; | |
5137 | } | |
5138 | return table; | |
5139 | } | |
5140 | ||
5141 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5142 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5143 | { |
6ad4c188 | 5144 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5145 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5146 | char buf[32]; | |
5147 | ||
7378547f MM |
5148 | WARN_ON(sd_ctl_dir[0].child); |
5149 | sd_ctl_dir[0].child = entry; | |
5150 | ||
ad1cdc1d MM |
5151 | if (entry == NULL) |
5152 | return; | |
5153 | ||
6ad4c188 | 5154 | for_each_possible_cpu(i) { |
e692ab53 | 5155 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5156 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5157 | entry->mode = 0555; |
e692ab53 | 5158 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5159 | entry++; |
e692ab53 | 5160 | } |
7378547f MM |
5161 | |
5162 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5163 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5164 | } | |
6382bc90 | 5165 | |
7378547f | 5166 | /* may be called multiple times per register */ |
6382bc90 MM |
5167 | static void unregister_sched_domain_sysctl(void) |
5168 | { | |
7378547f MM |
5169 | if (sd_sysctl_header) |
5170 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5171 | sd_sysctl_header = NULL; |
7378547f MM |
5172 | if (sd_ctl_dir[0].child) |
5173 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5174 | } |
e692ab53 | 5175 | #else |
6382bc90 MM |
5176 | static void register_sched_domain_sysctl(void) |
5177 | { | |
5178 | } | |
5179 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5180 | { |
5181 | } | |
5182 | #endif | |
5183 | ||
1f11eb6a GH |
5184 | static void set_rq_online(struct rq *rq) |
5185 | { | |
5186 | if (!rq->online) { | |
5187 | const struct sched_class *class; | |
5188 | ||
c6c4927b | 5189 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5190 | rq->online = 1; |
5191 | ||
5192 | for_each_class(class) { | |
5193 | if (class->rq_online) | |
5194 | class->rq_online(rq); | |
5195 | } | |
5196 | } | |
5197 | } | |
5198 | ||
5199 | static void set_rq_offline(struct rq *rq) | |
5200 | { | |
5201 | if (rq->online) { | |
5202 | const struct sched_class *class; | |
5203 | ||
5204 | for_each_class(class) { | |
5205 | if (class->rq_offline) | |
5206 | class->rq_offline(rq); | |
5207 | } | |
5208 | ||
c6c4927b | 5209 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5210 | rq->online = 0; |
5211 | } | |
5212 | } | |
5213 | ||
1da177e4 LT |
5214 | /* |
5215 | * migration_call - callback that gets triggered when a CPU is added. | |
5216 | * Here we can start up the necessary migration thread for the new CPU. | |
5217 | */ | |
0db0628d | 5218 | static int |
48f24c4d | 5219 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5220 | { |
48f24c4d | 5221 | int cpu = (long)hcpu; |
1da177e4 | 5222 | unsigned long flags; |
969c7921 | 5223 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5224 | |
48c5ccae | 5225 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5226 | |
1da177e4 | 5227 | case CPU_UP_PREPARE: |
a468d389 | 5228 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5229 | break; |
48f24c4d | 5230 | |
1da177e4 | 5231 | case CPU_ONLINE: |
1f94ef59 | 5232 | /* Update our root-domain */ |
05fa785c | 5233 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5234 | if (rq->rd) { |
c6c4927b | 5235 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5236 | |
5237 | set_rq_online(rq); | |
1f94ef59 | 5238 | } |
05fa785c | 5239 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5240 | break; |
48f24c4d | 5241 | |
1da177e4 | 5242 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5243 | case CPU_DYING: |
317f3941 | 5244 | sched_ttwu_pending(); |
57d885fe | 5245 | /* Update our root-domain */ |
05fa785c | 5246 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5247 | if (rq->rd) { |
c6c4927b | 5248 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5249 | set_rq_offline(rq); |
57d885fe | 5250 | } |
48c5ccae PZ |
5251 | migrate_tasks(cpu); |
5252 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5253 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5254 | break; |
48c5ccae | 5255 | |
5d180232 | 5256 | case CPU_DEAD: |
f319da0c | 5257 | calc_load_migrate(rq); |
57d885fe | 5258 | break; |
1da177e4 LT |
5259 | #endif |
5260 | } | |
49c022e6 PZ |
5261 | |
5262 | update_max_interval(); | |
5263 | ||
1da177e4 LT |
5264 | return NOTIFY_OK; |
5265 | } | |
5266 | ||
f38b0820 PM |
5267 | /* |
5268 | * Register at high priority so that task migration (migrate_all_tasks) | |
5269 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5270 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5271 | */ |
0db0628d | 5272 | static struct notifier_block migration_notifier = { |
1da177e4 | 5273 | .notifier_call = migration_call, |
50a323b7 | 5274 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5275 | }; |
5276 | ||
a803f026 CM |
5277 | static void __cpuinit set_cpu_rq_start_time(void) |
5278 | { | |
5279 | int cpu = smp_processor_id(); | |
5280 | struct rq *rq = cpu_rq(cpu); | |
5281 | rq->age_stamp = sched_clock_cpu(cpu); | |
5282 | } | |
5283 | ||
0db0628d | 5284 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5285 | unsigned long action, void *hcpu) |
5286 | { | |
5287 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5288 | case CPU_STARTING: |
5289 | set_cpu_rq_start_time(); | |
5290 | return NOTIFY_OK; | |
3a101d05 TH |
5291 | case CPU_DOWN_FAILED: |
5292 | set_cpu_active((long)hcpu, true); | |
5293 | return NOTIFY_OK; | |
5294 | default: | |
5295 | return NOTIFY_DONE; | |
5296 | } | |
5297 | } | |
5298 | ||
0db0628d | 5299 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5300 | unsigned long action, void *hcpu) |
5301 | { | |
de212f18 PZ |
5302 | unsigned long flags; |
5303 | long cpu = (long)hcpu; | |
f10e00f4 | 5304 | struct dl_bw *dl_b; |
de212f18 | 5305 | |
3a101d05 TH |
5306 | switch (action & ~CPU_TASKS_FROZEN) { |
5307 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5308 | set_cpu_active(cpu, false); |
5309 | ||
5310 | /* explicitly allow suspend */ | |
5311 | if (!(action & CPU_TASKS_FROZEN)) { | |
de212f18 PZ |
5312 | bool overflow; |
5313 | int cpus; | |
5314 | ||
f10e00f4 KT |
5315 | rcu_read_lock_sched(); |
5316 | dl_b = dl_bw_of(cpu); | |
5317 | ||
de212f18 PZ |
5318 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5319 | cpus = dl_bw_cpus(cpu); | |
5320 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5321 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5322 | ||
f10e00f4 KT |
5323 | rcu_read_unlock_sched(); |
5324 | ||
de212f18 PZ |
5325 | if (overflow) |
5326 | return notifier_from_errno(-EBUSY); | |
5327 | } | |
3a101d05 | 5328 | return NOTIFY_OK; |
3a101d05 | 5329 | } |
de212f18 PZ |
5330 | |
5331 | return NOTIFY_DONE; | |
3a101d05 TH |
5332 | } |
5333 | ||
7babe8db | 5334 | static int __init migration_init(void) |
1da177e4 LT |
5335 | { |
5336 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5337 | int err; |
48f24c4d | 5338 | |
3a101d05 | 5339 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5340 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5341 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5342 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5343 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5344 | |
3a101d05 TH |
5345 | /* Register cpu active notifiers */ |
5346 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5347 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5348 | ||
a004cd42 | 5349 | return 0; |
1da177e4 | 5350 | } |
7babe8db | 5351 | early_initcall(migration_init); |
1da177e4 LT |
5352 | #endif |
5353 | ||
5354 | #ifdef CONFIG_SMP | |
476f3534 | 5355 | |
4cb98839 PZ |
5356 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5357 | ||
3e9830dc | 5358 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5359 | |
d039ac60 | 5360 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5361 | |
d039ac60 | 5362 | static int __init sched_debug_setup(char *str) |
f6630114 | 5363 | { |
d039ac60 | 5364 | sched_debug_enabled = 1; |
f6630114 MT |
5365 | |
5366 | return 0; | |
5367 | } | |
d039ac60 PZ |
5368 | early_param("sched_debug", sched_debug_setup); |
5369 | ||
5370 | static inline bool sched_debug(void) | |
5371 | { | |
5372 | return sched_debug_enabled; | |
5373 | } | |
f6630114 | 5374 | |
7c16ec58 | 5375 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5376 | struct cpumask *groupmask) |
1da177e4 | 5377 | { |
4dcf6aff | 5378 | struct sched_group *group = sd->groups; |
434d53b0 | 5379 | char str[256]; |
1da177e4 | 5380 | |
968ea6d8 | 5381 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5382 | cpumask_clear(groupmask); |
4dcf6aff IM |
5383 | |
5384 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5385 | ||
5386 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5387 | printk("does not load-balance\n"); |
4dcf6aff | 5388 | if (sd->parent) |
3df0fc5b PZ |
5389 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5390 | " has parent"); | |
4dcf6aff | 5391 | return -1; |
41c7ce9a NP |
5392 | } |
5393 | ||
3df0fc5b | 5394 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5395 | |
758b2cdc | 5396 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5397 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5398 | "CPU%d\n", cpu); | |
4dcf6aff | 5399 | } |
758b2cdc | 5400 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5401 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5402 | " CPU%d\n", cpu); | |
4dcf6aff | 5403 | } |
1da177e4 | 5404 | |
4dcf6aff | 5405 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5406 | do { |
4dcf6aff | 5407 | if (!group) { |
3df0fc5b PZ |
5408 | printk("\n"); |
5409 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5410 | break; |
5411 | } | |
5412 | ||
c3decf0d | 5413 | /* |
63b2ca30 NP |
5414 | * Even though we initialize ->capacity to something semi-sane, |
5415 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5416 | * domain iteration is still funny without causing /0 traps. |
5417 | */ | |
63b2ca30 | 5418 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5419 | printk(KERN_CONT "\n"); |
63b2ca30 | 5420 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5421 | break; |
5422 | } | |
1da177e4 | 5423 | |
758b2cdc | 5424 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5425 | printk(KERN_CONT "\n"); |
5426 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5427 | break; |
5428 | } | |
1da177e4 | 5429 | |
cb83b629 PZ |
5430 | if (!(sd->flags & SD_OVERLAP) && |
5431 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5432 | printk(KERN_CONT "\n"); |
5433 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5434 | break; |
5435 | } | |
1da177e4 | 5436 | |
758b2cdc | 5437 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5438 | |
968ea6d8 | 5439 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5440 | |
3df0fc5b | 5441 | printk(KERN_CONT " %s", str); |
ca8ce3d0 | 5442 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5443 | printk(KERN_CONT " (cpu_capacity = %d)", |
5444 | group->sgc->capacity); | |
381512cf | 5445 | } |
1da177e4 | 5446 | |
4dcf6aff IM |
5447 | group = group->next; |
5448 | } while (group != sd->groups); | |
3df0fc5b | 5449 | printk(KERN_CONT "\n"); |
1da177e4 | 5450 | |
758b2cdc | 5451 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5452 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5453 | |
758b2cdc RR |
5454 | if (sd->parent && |
5455 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5456 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5457 | "of domain->span\n"); | |
4dcf6aff IM |
5458 | return 0; |
5459 | } | |
1da177e4 | 5460 | |
4dcf6aff IM |
5461 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5462 | { | |
5463 | int level = 0; | |
1da177e4 | 5464 | |
d039ac60 | 5465 | if (!sched_debug_enabled) |
f6630114 MT |
5466 | return; |
5467 | ||
4dcf6aff IM |
5468 | if (!sd) { |
5469 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5470 | return; | |
5471 | } | |
1da177e4 | 5472 | |
4dcf6aff IM |
5473 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5474 | ||
5475 | for (;;) { | |
4cb98839 | 5476 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5477 | break; |
1da177e4 LT |
5478 | level++; |
5479 | sd = sd->parent; | |
33859f7f | 5480 | if (!sd) |
4dcf6aff IM |
5481 | break; |
5482 | } | |
1da177e4 | 5483 | } |
6d6bc0ad | 5484 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5485 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5486 | static inline bool sched_debug(void) |
5487 | { | |
5488 | return false; | |
5489 | } | |
6d6bc0ad | 5490 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5491 | |
1a20ff27 | 5492 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5493 | { |
758b2cdc | 5494 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5495 | return 1; |
5496 | ||
5497 | /* Following flags need at least 2 groups */ | |
5498 | if (sd->flags & (SD_LOAD_BALANCE | | |
5499 | SD_BALANCE_NEWIDLE | | |
5500 | SD_BALANCE_FORK | | |
89c4710e | 5501 | SD_BALANCE_EXEC | |
5d4dfddd | 5502 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5503 | SD_SHARE_PKG_RESOURCES | |
5504 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5505 | if (sd->groups != sd->groups->next) |
5506 | return 0; | |
5507 | } | |
5508 | ||
5509 | /* Following flags don't use groups */ | |
c88d5910 | 5510 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5511 | return 0; |
5512 | ||
5513 | return 1; | |
5514 | } | |
5515 | ||
48f24c4d IM |
5516 | static int |
5517 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5518 | { |
5519 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5520 | ||
5521 | if (sd_degenerate(parent)) | |
5522 | return 1; | |
5523 | ||
758b2cdc | 5524 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5525 | return 0; |
5526 | ||
245af2c7 SS |
5527 | /* Flags needing groups don't count if only 1 group in parent */ |
5528 | if (parent->groups == parent->groups->next) { | |
5529 | pflags &= ~(SD_LOAD_BALANCE | | |
5530 | SD_BALANCE_NEWIDLE | | |
5531 | SD_BALANCE_FORK | | |
89c4710e | 5532 | SD_BALANCE_EXEC | |
5d4dfddd | 5533 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5534 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5535 | SD_PREFER_SIBLING | |
5536 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5537 | if (nr_node_ids == 1) |
5538 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5539 | } |
5540 | if (~cflags & pflags) | |
5541 | return 0; | |
5542 | ||
5543 | return 1; | |
5544 | } | |
5545 | ||
dce840a0 | 5546 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5547 | { |
dce840a0 | 5548 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5549 | |
68e74568 | 5550 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5551 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5552 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5553 | free_cpumask_var(rd->rto_mask); |
5554 | free_cpumask_var(rd->online); | |
5555 | free_cpumask_var(rd->span); | |
5556 | kfree(rd); | |
5557 | } | |
5558 | ||
57d885fe GH |
5559 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5560 | { | |
a0490fa3 | 5561 | struct root_domain *old_rd = NULL; |
57d885fe | 5562 | unsigned long flags; |
57d885fe | 5563 | |
05fa785c | 5564 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5565 | |
5566 | if (rq->rd) { | |
a0490fa3 | 5567 | old_rd = rq->rd; |
57d885fe | 5568 | |
c6c4927b | 5569 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5570 | set_rq_offline(rq); |
57d885fe | 5571 | |
c6c4927b | 5572 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5573 | |
a0490fa3 | 5574 | /* |
0515973f | 5575 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5576 | * set old_rd to NULL to skip the freeing later |
5577 | * in this function: | |
5578 | */ | |
5579 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5580 | old_rd = NULL; | |
57d885fe GH |
5581 | } |
5582 | ||
5583 | atomic_inc(&rd->refcount); | |
5584 | rq->rd = rd; | |
5585 | ||
c6c4927b | 5586 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5587 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5588 | set_rq_online(rq); |
57d885fe | 5589 | |
05fa785c | 5590 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5591 | |
5592 | if (old_rd) | |
dce840a0 | 5593 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5594 | } |
5595 | ||
68c38fc3 | 5596 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5597 | { |
5598 | memset(rd, 0, sizeof(*rd)); | |
5599 | ||
68c38fc3 | 5600 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5601 | goto out; |
68c38fc3 | 5602 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5603 | goto free_span; |
1baca4ce | 5604 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5605 | goto free_online; |
1baca4ce JL |
5606 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5607 | goto free_dlo_mask; | |
6e0534f2 | 5608 | |
332ac17e | 5609 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5610 | if (cpudl_init(&rd->cpudl) != 0) |
5611 | goto free_dlo_mask; | |
332ac17e | 5612 | |
68c38fc3 | 5613 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5614 | goto free_rto_mask; |
c6c4927b | 5615 | return 0; |
6e0534f2 | 5616 | |
68e74568 RR |
5617 | free_rto_mask: |
5618 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5619 | free_dlo_mask: |
5620 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5621 | free_online: |
5622 | free_cpumask_var(rd->online); | |
5623 | free_span: | |
5624 | free_cpumask_var(rd->span); | |
0c910d28 | 5625 | out: |
c6c4927b | 5626 | return -ENOMEM; |
57d885fe GH |
5627 | } |
5628 | ||
029632fb PZ |
5629 | /* |
5630 | * By default the system creates a single root-domain with all cpus as | |
5631 | * members (mimicking the global state we have today). | |
5632 | */ | |
5633 | struct root_domain def_root_domain; | |
5634 | ||
57d885fe GH |
5635 | static void init_defrootdomain(void) |
5636 | { | |
68c38fc3 | 5637 | init_rootdomain(&def_root_domain); |
c6c4927b | 5638 | |
57d885fe GH |
5639 | atomic_set(&def_root_domain.refcount, 1); |
5640 | } | |
5641 | ||
dc938520 | 5642 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5643 | { |
5644 | struct root_domain *rd; | |
5645 | ||
5646 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5647 | if (!rd) | |
5648 | return NULL; | |
5649 | ||
68c38fc3 | 5650 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5651 | kfree(rd); |
5652 | return NULL; | |
5653 | } | |
57d885fe GH |
5654 | |
5655 | return rd; | |
5656 | } | |
5657 | ||
63b2ca30 | 5658 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5659 | { |
5660 | struct sched_group *tmp, *first; | |
5661 | ||
5662 | if (!sg) | |
5663 | return; | |
5664 | ||
5665 | first = sg; | |
5666 | do { | |
5667 | tmp = sg->next; | |
5668 | ||
63b2ca30 NP |
5669 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5670 | kfree(sg->sgc); | |
e3589f6c PZ |
5671 | |
5672 | kfree(sg); | |
5673 | sg = tmp; | |
5674 | } while (sg != first); | |
5675 | } | |
5676 | ||
dce840a0 PZ |
5677 | static void free_sched_domain(struct rcu_head *rcu) |
5678 | { | |
5679 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5680 | |
5681 | /* | |
5682 | * If its an overlapping domain it has private groups, iterate and | |
5683 | * nuke them all. | |
5684 | */ | |
5685 | if (sd->flags & SD_OVERLAP) { | |
5686 | free_sched_groups(sd->groups, 1); | |
5687 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5688 | kfree(sd->groups->sgc); |
dce840a0 | 5689 | kfree(sd->groups); |
9c3f75cb | 5690 | } |
dce840a0 PZ |
5691 | kfree(sd); |
5692 | } | |
5693 | ||
5694 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5695 | { | |
5696 | call_rcu(&sd->rcu, free_sched_domain); | |
5697 | } | |
5698 | ||
5699 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5700 | { | |
5701 | for (; sd; sd = sd->parent) | |
5702 | destroy_sched_domain(sd, cpu); | |
5703 | } | |
5704 | ||
518cd623 PZ |
5705 | /* |
5706 | * Keep a special pointer to the highest sched_domain that has | |
5707 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5708 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5709 | * | |
5710 | * Also keep a unique ID per domain (we use the first cpu number in | |
5711 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5712 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5713 | */ |
5714 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5715 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5716 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5717 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5718 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5719 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5720 | |
5721 | static void update_top_cache_domain(int cpu) | |
5722 | { | |
5723 | struct sched_domain *sd; | |
5d4cf996 | 5724 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5725 | int id = cpu; |
7d9ffa89 | 5726 | int size = 1; |
518cd623 PZ |
5727 | |
5728 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5729 | if (sd) { |
518cd623 | 5730 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5731 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5732 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5733 | } |
5d4cf996 | 5734 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5735 | |
5736 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5737 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5738 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5739 | |
5740 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5741 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5742 | |
5743 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5744 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5745 | } |
5746 | ||
1da177e4 | 5747 | /* |
0eab9146 | 5748 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5749 | * hold the hotplug lock. |
5750 | */ | |
0eab9146 IM |
5751 | static void |
5752 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5753 | { |
70b97a7f | 5754 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5755 | struct sched_domain *tmp; |
5756 | ||
5757 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5758 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5759 | struct sched_domain *parent = tmp->parent; |
5760 | if (!parent) | |
5761 | break; | |
f29c9b1c | 5762 | |
1a848870 | 5763 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5764 | tmp->parent = parent->parent; |
1a848870 SS |
5765 | if (parent->parent) |
5766 | parent->parent->child = tmp; | |
10866e62 PZ |
5767 | /* |
5768 | * Transfer SD_PREFER_SIBLING down in case of a | |
5769 | * degenerate parent; the spans match for this | |
5770 | * so the property transfers. | |
5771 | */ | |
5772 | if (parent->flags & SD_PREFER_SIBLING) | |
5773 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5774 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5775 | } else |
5776 | tmp = tmp->parent; | |
245af2c7 SS |
5777 | } |
5778 | ||
1a848870 | 5779 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5780 | tmp = sd; |
245af2c7 | 5781 | sd = sd->parent; |
dce840a0 | 5782 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5783 | if (sd) |
5784 | sd->child = NULL; | |
5785 | } | |
1da177e4 | 5786 | |
4cb98839 | 5787 | sched_domain_debug(sd, cpu); |
1da177e4 | 5788 | |
57d885fe | 5789 | rq_attach_root(rq, rd); |
dce840a0 | 5790 | tmp = rq->sd; |
674311d5 | 5791 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5792 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5793 | |
5794 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5795 | } |
5796 | ||
5797 | /* cpus with isolated domains */ | |
dcc30a35 | 5798 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5799 | |
5800 | /* Setup the mask of cpus configured for isolated domains */ | |
5801 | static int __init isolated_cpu_setup(char *str) | |
5802 | { | |
bdddd296 | 5803 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5804 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5805 | return 1; |
5806 | } | |
5807 | ||
8927f494 | 5808 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5809 | |
49a02c51 | 5810 | struct s_data { |
21d42ccf | 5811 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5812 | struct root_domain *rd; |
5813 | }; | |
5814 | ||
2109b99e | 5815 | enum s_alloc { |
2109b99e | 5816 | sa_rootdomain, |
21d42ccf | 5817 | sa_sd, |
dce840a0 | 5818 | sa_sd_storage, |
2109b99e AH |
5819 | sa_none, |
5820 | }; | |
5821 | ||
c1174876 PZ |
5822 | /* |
5823 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5824 | * domain traversal. | |
5825 | * | |
5826 | * Asymmetric node setups can result in situations where the domain tree is of | |
5827 | * unequal depth, make sure to skip domains that already cover the entire | |
5828 | * range. | |
5829 | * | |
5830 | * In that case build_sched_domains() will have terminated the iteration early | |
5831 | * and our sibling sd spans will be empty. Domains should always include the | |
5832 | * cpu they're built on, so check that. | |
5833 | * | |
5834 | */ | |
5835 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5836 | { | |
5837 | const struct cpumask *span = sched_domain_span(sd); | |
5838 | struct sd_data *sdd = sd->private; | |
5839 | struct sched_domain *sibling; | |
5840 | int i; | |
5841 | ||
5842 | for_each_cpu(i, span) { | |
5843 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5844 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5845 | continue; | |
5846 | ||
5847 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5848 | } | |
5849 | } | |
5850 | ||
5851 | /* | |
5852 | * Return the canonical balance cpu for this group, this is the first cpu | |
5853 | * of this group that's also in the iteration mask. | |
5854 | */ | |
5855 | int group_balance_cpu(struct sched_group *sg) | |
5856 | { | |
5857 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5858 | } | |
5859 | ||
e3589f6c PZ |
5860 | static int |
5861 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5862 | { | |
5863 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5864 | const struct cpumask *span = sched_domain_span(sd); | |
5865 | struct cpumask *covered = sched_domains_tmpmask; | |
5866 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5867 | struct sched_domain *sibling; |
e3589f6c PZ |
5868 | int i; |
5869 | ||
5870 | cpumask_clear(covered); | |
5871 | ||
5872 | for_each_cpu(i, span) { | |
5873 | struct cpumask *sg_span; | |
5874 | ||
5875 | if (cpumask_test_cpu(i, covered)) | |
5876 | continue; | |
5877 | ||
aaecac4a | 5878 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5879 | |
5880 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5881 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5882 | continue; |
5883 | ||
e3589f6c | 5884 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5885 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5886 | |
5887 | if (!sg) | |
5888 | goto fail; | |
5889 | ||
5890 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5891 | if (sibling->child) |
5892 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5893 | else | |
e3589f6c PZ |
5894 | cpumask_set_cpu(i, sg_span); |
5895 | ||
5896 | cpumask_or(covered, covered, sg_span); | |
5897 | ||
63b2ca30 NP |
5898 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5899 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5900 | build_group_mask(sd, sg); |
5901 | ||
c3decf0d | 5902 | /* |
63b2ca30 | 5903 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5904 | * domains and no possible iteration will get us here, we won't |
5905 | * die on a /0 trap. | |
5906 | */ | |
ca8ce3d0 | 5907 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5908 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5909 | |
c1174876 PZ |
5910 | /* |
5911 | * Make sure the first group of this domain contains the | |
5912 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5913 | * breaks. See update_sg_lb_stats(). | |
5914 | */ | |
74a5ce20 | 5915 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5916 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5917 | groups = sg; |
5918 | ||
5919 | if (!first) | |
5920 | first = sg; | |
5921 | if (last) | |
5922 | last->next = sg; | |
5923 | last = sg; | |
5924 | last->next = first; | |
5925 | } | |
5926 | sd->groups = groups; | |
5927 | ||
5928 | return 0; | |
5929 | ||
5930 | fail: | |
5931 | free_sched_groups(first, 0); | |
5932 | ||
5933 | return -ENOMEM; | |
5934 | } | |
5935 | ||
dce840a0 | 5936 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5937 | { |
dce840a0 PZ |
5938 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5939 | struct sched_domain *child = sd->child; | |
1da177e4 | 5940 | |
dce840a0 PZ |
5941 | if (child) |
5942 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5943 | |
9c3f75cb | 5944 | if (sg) { |
dce840a0 | 5945 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5946 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5947 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5948 | } |
dce840a0 PZ |
5949 | |
5950 | return cpu; | |
1e9f28fa | 5951 | } |
1e9f28fa | 5952 | |
01a08546 | 5953 | /* |
dce840a0 PZ |
5954 | * build_sched_groups will build a circular linked list of the groups |
5955 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5956 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5957 | * |
5958 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5959 | */ |
e3589f6c PZ |
5960 | static int |
5961 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5962 | { |
dce840a0 PZ |
5963 | struct sched_group *first = NULL, *last = NULL; |
5964 | struct sd_data *sdd = sd->private; | |
5965 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5966 | struct cpumask *covered; |
dce840a0 | 5967 | int i; |
9c1cfda2 | 5968 | |
e3589f6c PZ |
5969 | get_group(cpu, sdd, &sd->groups); |
5970 | atomic_inc(&sd->groups->ref); | |
5971 | ||
0936629f | 5972 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5973 | return 0; |
5974 | ||
f96225fd PZ |
5975 | lockdep_assert_held(&sched_domains_mutex); |
5976 | covered = sched_domains_tmpmask; | |
5977 | ||
dce840a0 | 5978 | cpumask_clear(covered); |
6711cab4 | 5979 | |
dce840a0 PZ |
5980 | for_each_cpu(i, span) { |
5981 | struct sched_group *sg; | |
cd08e923 | 5982 | int group, j; |
6711cab4 | 5983 | |
dce840a0 PZ |
5984 | if (cpumask_test_cpu(i, covered)) |
5985 | continue; | |
6711cab4 | 5986 | |
cd08e923 | 5987 | group = get_group(i, sdd, &sg); |
c1174876 | 5988 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5989 | |
dce840a0 PZ |
5990 | for_each_cpu(j, span) { |
5991 | if (get_group(j, sdd, NULL) != group) | |
5992 | continue; | |
0601a88d | 5993 | |
dce840a0 PZ |
5994 | cpumask_set_cpu(j, covered); |
5995 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5996 | } | |
0601a88d | 5997 | |
dce840a0 PZ |
5998 | if (!first) |
5999 | first = sg; | |
6000 | if (last) | |
6001 | last->next = sg; | |
6002 | last = sg; | |
6003 | } | |
6004 | last->next = first; | |
e3589f6c PZ |
6005 | |
6006 | return 0; | |
0601a88d | 6007 | } |
51888ca2 | 6008 | |
89c4710e | 6009 | /* |
63b2ca30 | 6010 | * Initialize sched groups cpu_capacity. |
89c4710e | 6011 | * |
63b2ca30 | 6012 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6013 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6014 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6015 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6016 | * group having more cpu_capacity will pickup more load compared to the | |
6017 | * group having less cpu_capacity. | |
89c4710e | 6018 | */ |
63b2ca30 | 6019 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6020 | { |
e3589f6c | 6021 | struct sched_group *sg = sd->groups; |
89c4710e | 6022 | |
94c95ba6 | 6023 | WARN_ON(!sg); |
e3589f6c PZ |
6024 | |
6025 | do { | |
6026 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6027 | sg = sg->next; | |
6028 | } while (sg != sd->groups); | |
89c4710e | 6029 | |
c1174876 | 6030 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6031 | return; |
aae6d3dd | 6032 | |
63b2ca30 NP |
6033 | update_group_capacity(sd, cpu); |
6034 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6035 | } |
6036 | ||
7c16ec58 MT |
6037 | /* |
6038 | * Initializers for schedule domains | |
6039 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6040 | */ | |
6041 | ||
1d3504fc | 6042 | static int default_relax_domain_level = -1; |
60495e77 | 6043 | int sched_domain_level_max; |
1d3504fc HS |
6044 | |
6045 | static int __init setup_relax_domain_level(char *str) | |
6046 | { | |
a841f8ce DS |
6047 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6048 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6049 | |
1d3504fc HS |
6050 | return 1; |
6051 | } | |
6052 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6053 | ||
6054 | static void set_domain_attribute(struct sched_domain *sd, | |
6055 | struct sched_domain_attr *attr) | |
6056 | { | |
6057 | int request; | |
6058 | ||
6059 | if (!attr || attr->relax_domain_level < 0) { | |
6060 | if (default_relax_domain_level < 0) | |
6061 | return; | |
6062 | else | |
6063 | request = default_relax_domain_level; | |
6064 | } else | |
6065 | request = attr->relax_domain_level; | |
6066 | if (request < sd->level) { | |
6067 | /* turn off idle balance on this domain */ | |
c88d5910 | 6068 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6069 | } else { |
6070 | /* turn on idle balance on this domain */ | |
c88d5910 | 6071 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6072 | } |
6073 | } | |
6074 | ||
54ab4ff4 PZ |
6075 | static void __sdt_free(const struct cpumask *cpu_map); |
6076 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6077 | ||
2109b99e AH |
6078 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6079 | const struct cpumask *cpu_map) | |
6080 | { | |
6081 | switch (what) { | |
2109b99e | 6082 | case sa_rootdomain: |
822ff793 PZ |
6083 | if (!atomic_read(&d->rd->refcount)) |
6084 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6085 | case sa_sd: |
6086 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6087 | case sa_sd_storage: |
54ab4ff4 | 6088 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6089 | case sa_none: |
6090 | break; | |
6091 | } | |
6092 | } | |
3404c8d9 | 6093 | |
2109b99e AH |
6094 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6095 | const struct cpumask *cpu_map) | |
6096 | { | |
dce840a0 PZ |
6097 | memset(d, 0, sizeof(*d)); |
6098 | ||
54ab4ff4 PZ |
6099 | if (__sdt_alloc(cpu_map)) |
6100 | return sa_sd_storage; | |
dce840a0 PZ |
6101 | d->sd = alloc_percpu(struct sched_domain *); |
6102 | if (!d->sd) | |
6103 | return sa_sd_storage; | |
2109b99e | 6104 | d->rd = alloc_rootdomain(); |
dce840a0 | 6105 | if (!d->rd) |
21d42ccf | 6106 | return sa_sd; |
2109b99e AH |
6107 | return sa_rootdomain; |
6108 | } | |
57d885fe | 6109 | |
dce840a0 PZ |
6110 | /* |
6111 | * NULL the sd_data elements we've used to build the sched_domain and | |
6112 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6113 | * will not free the data we're using. | |
6114 | */ | |
6115 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6116 | { | |
6117 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6118 | |
6119 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6120 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6121 | ||
e3589f6c | 6122 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6123 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6124 | |
63b2ca30 NP |
6125 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6126 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6127 | } |
6128 | ||
cb83b629 | 6129 | #ifdef CONFIG_NUMA |
cb83b629 | 6130 | static int sched_domains_numa_levels; |
e3fe70b1 | 6131 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6132 | static int *sched_domains_numa_distance; |
9942f79b | 6133 | int sched_max_numa_distance; |
cb83b629 PZ |
6134 | static struct cpumask ***sched_domains_numa_masks; |
6135 | static int sched_domains_curr_level; | |
143e1e28 | 6136 | #endif |
cb83b629 | 6137 | |
143e1e28 VG |
6138 | /* |
6139 | * SD_flags allowed in topology descriptions. | |
6140 | * | |
5d4dfddd | 6141 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6142 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6143 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6144 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6145 | * |
6146 | * Odd one out: | |
6147 | * SD_ASYM_PACKING - describes SMT quirks | |
6148 | */ | |
6149 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6150 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6151 | SD_SHARE_PKG_RESOURCES | \ |
6152 | SD_NUMA | \ | |
d77b3ed5 VG |
6153 | SD_ASYM_PACKING | \ |
6154 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6155 | |
6156 | static struct sched_domain * | |
143e1e28 | 6157 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6158 | { |
6159 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6160 | int sd_weight, sd_flags = 0; |
6161 | ||
6162 | #ifdef CONFIG_NUMA | |
6163 | /* | |
6164 | * Ugly hack to pass state to sd_numa_mask()... | |
6165 | */ | |
6166 | sched_domains_curr_level = tl->numa_level; | |
6167 | #endif | |
6168 | ||
6169 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6170 | ||
6171 | if (tl->sd_flags) | |
6172 | sd_flags = (*tl->sd_flags)(); | |
6173 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6174 | "wrong sd_flags in topology description\n")) | |
6175 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6176 | |
6177 | *sd = (struct sched_domain){ | |
6178 | .min_interval = sd_weight, | |
6179 | .max_interval = 2*sd_weight, | |
6180 | .busy_factor = 32, | |
870a0bb5 | 6181 | .imbalance_pct = 125, |
143e1e28 VG |
6182 | |
6183 | .cache_nice_tries = 0, | |
6184 | .busy_idx = 0, | |
6185 | .idle_idx = 0, | |
cb83b629 PZ |
6186 | .newidle_idx = 0, |
6187 | .wake_idx = 0, | |
6188 | .forkexec_idx = 0, | |
6189 | ||
6190 | .flags = 1*SD_LOAD_BALANCE | |
6191 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6192 | | 1*SD_BALANCE_EXEC |
6193 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6194 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6195 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6196 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6197 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6198 | | 0*SD_SERIALIZE |
cb83b629 | 6199 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6200 | | 0*SD_NUMA |
6201 | | sd_flags | |
cb83b629 | 6202 | , |
143e1e28 | 6203 | |
cb83b629 PZ |
6204 | .last_balance = jiffies, |
6205 | .balance_interval = sd_weight, | |
143e1e28 | 6206 | .smt_gain = 0, |
2b4cfe64 JL |
6207 | .max_newidle_lb_cost = 0, |
6208 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6209 | #ifdef CONFIG_SCHED_DEBUG |
6210 | .name = tl->name, | |
6211 | #endif | |
cb83b629 | 6212 | }; |
cb83b629 PZ |
6213 | |
6214 | /* | |
143e1e28 | 6215 | * Convert topological properties into behaviour. |
cb83b629 | 6216 | */ |
143e1e28 | 6217 | |
5d4dfddd | 6218 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6219 | sd->imbalance_pct = 110; |
6220 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6221 | |
6222 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6223 | sd->imbalance_pct = 117; | |
6224 | sd->cache_nice_tries = 1; | |
6225 | sd->busy_idx = 2; | |
6226 | ||
6227 | #ifdef CONFIG_NUMA | |
6228 | } else if (sd->flags & SD_NUMA) { | |
6229 | sd->cache_nice_tries = 2; | |
6230 | sd->busy_idx = 3; | |
6231 | sd->idle_idx = 2; | |
6232 | ||
6233 | sd->flags |= SD_SERIALIZE; | |
6234 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6235 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6236 | SD_BALANCE_FORK | | |
6237 | SD_WAKE_AFFINE); | |
6238 | } | |
6239 | ||
6240 | #endif | |
6241 | } else { | |
6242 | sd->flags |= SD_PREFER_SIBLING; | |
6243 | sd->cache_nice_tries = 1; | |
6244 | sd->busy_idx = 2; | |
6245 | sd->idle_idx = 1; | |
6246 | } | |
6247 | ||
6248 | sd->private = &tl->data; | |
cb83b629 PZ |
6249 | |
6250 | return sd; | |
6251 | } | |
6252 | ||
143e1e28 VG |
6253 | /* |
6254 | * Topology list, bottom-up. | |
6255 | */ | |
6256 | static struct sched_domain_topology_level default_topology[] = { | |
6257 | #ifdef CONFIG_SCHED_SMT | |
6258 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6259 | #endif | |
6260 | #ifdef CONFIG_SCHED_MC | |
6261 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6262 | #endif |
6263 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6264 | { NULL, }, | |
6265 | }; | |
6266 | ||
6267 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6268 | ||
6269 | #define for_each_sd_topology(tl) \ | |
6270 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6271 | ||
6272 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6273 | { | |
6274 | sched_domain_topology = tl; | |
6275 | } | |
6276 | ||
6277 | #ifdef CONFIG_NUMA | |
6278 | ||
cb83b629 PZ |
6279 | static const struct cpumask *sd_numa_mask(int cpu) |
6280 | { | |
6281 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6282 | } | |
6283 | ||
d039ac60 PZ |
6284 | static void sched_numa_warn(const char *str) |
6285 | { | |
6286 | static int done = false; | |
6287 | int i,j; | |
6288 | ||
6289 | if (done) | |
6290 | return; | |
6291 | ||
6292 | done = true; | |
6293 | ||
6294 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6295 | ||
6296 | for (i = 0; i < nr_node_ids; i++) { | |
6297 | printk(KERN_WARNING " "); | |
6298 | for (j = 0; j < nr_node_ids; j++) | |
6299 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6300 | printk(KERN_CONT "\n"); | |
6301 | } | |
6302 | printk(KERN_WARNING "\n"); | |
6303 | } | |
6304 | ||
9942f79b | 6305 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6306 | { |
6307 | int i; | |
6308 | ||
6309 | if (distance == node_distance(0, 0)) | |
6310 | return true; | |
6311 | ||
6312 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6313 | if (sched_domains_numa_distance[i] == distance) | |
6314 | return true; | |
6315 | } | |
6316 | ||
6317 | return false; | |
6318 | } | |
6319 | ||
e3fe70b1 RR |
6320 | /* |
6321 | * A system can have three types of NUMA topology: | |
6322 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6323 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6324 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6325 | * | |
6326 | * The difference between a glueless mesh topology and a backplane | |
6327 | * topology lies in whether communication between not directly | |
6328 | * connected nodes goes through intermediary nodes (where programs | |
6329 | * could run), or through backplane controllers. This affects | |
6330 | * placement of programs. | |
6331 | * | |
6332 | * The type of topology can be discerned with the following tests: | |
6333 | * - If the maximum distance between any nodes is 1 hop, the system | |
6334 | * is directly connected. | |
6335 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6336 | * there is an intermediary node C, which is < N hops away from both | |
6337 | * nodes A and B, the system is a glueless mesh. | |
6338 | */ | |
6339 | static void init_numa_topology_type(void) | |
6340 | { | |
6341 | int a, b, c, n; | |
6342 | ||
6343 | n = sched_max_numa_distance; | |
6344 | ||
6345 | if (n <= 1) | |
6346 | sched_numa_topology_type = NUMA_DIRECT; | |
6347 | ||
6348 | for_each_online_node(a) { | |
6349 | for_each_online_node(b) { | |
6350 | /* Find two nodes furthest removed from each other. */ | |
6351 | if (node_distance(a, b) < n) | |
6352 | continue; | |
6353 | ||
6354 | /* Is there an intermediary node between a and b? */ | |
6355 | for_each_online_node(c) { | |
6356 | if (node_distance(a, c) < n && | |
6357 | node_distance(b, c) < n) { | |
6358 | sched_numa_topology_type = | |
6359 | NUMA_GLUELESS_MESH; | |
6360 | return; | |
6361 | } | |
6362 | } | |
6363 | ||
6364 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6365 | return; | |
6366 | } | |
6367 | } | |
6368 | } | |
6369 | ||
cb83b629 PZ |
6370 | static void sched_init_numa(void) |
6371 | { | |
6372 | int next_distance, curr_distance = node_distance(0, 0); | |
6373 | struct sched_domain_topology_level *tl; | |
6374 | int level = 0; | |
6375 | int i, j, k; | |
6376 | ||
cb83b629 PZ |
6377 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6378 | if (!sched_domains_numa_distance) | |
6379 | return; | |
6380 | ||
6381 | /* | |
6382 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6383 | * unique distances in the node_distance() table. | |
6384 | * | |
6385 | * Assumes node_distance(0,j) includes all distances in | |
6386 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6387 | */ |
6388 | next_distance = curr_distance; | |
6389 | for (i = 0; i < nr_node_ids; i++) { | |
6390 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6391 | for (k = 0; k < nr_node_ids; k++) { |
6392 | int distance = node_distance(i, k); | |
6393 | ||
6394 | if (distance > curr_distance && | |
6395 | (distance < next_distance || | |
6396 | next_distance == curr_distance)) | |
6397 | next_distance = distance; | |
6398 | ||
6399 | /* | |
6400 | * While not a strong assumption it would be nice to know | |
6401 | * about cases where if node A is connected to B, B is not | |
6402 | * equally connected to A. | |
6403 | */ | |
6404 | if (sched_debug() && node_distance(k, i) != distance) | |
6405 | sched_numa_warn("Node-distance not symmetric"); | |
6406 | ||
6407 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6408 | sched_numa_warn("Node-0 not representative"); | |
6409 | } | |
6410 | if (next_distance != curr_distance) { | |
6411 | sched_domains_numa_distance[level++] = next_distance; | |
6412 | sched_domains_numa_levels = level; | |
6413 | curr_distance = next_distance; | |
6414 | } else break; | |
cb83b629 | 6415 | } |
d039ac60 PZ |
6416 | |
6417 | /* | |
6418 | * In case of sched_debug() we verify the above assumption. | |
6419 | */ | |
6420 | if (!sched_debug()) | |
6421 | break; | |
cb83b629 PZ |
6422 | } |
6423 | /* | |
6424 | * 'level' contains the number of unique distances, excluding the | |
6425 | * identity distance node_distance(i,i). | |
6426 | * | |
28b4a521 | 6427 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6428 | * numbers. |
6429 | */ | |
6430 | ||
5f7865f3 TC |
6431 | /* |
6432 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6433 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6434 | * the array will contain less then 'level' members. This could be | |
6435 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6436 | * in other functions. | |
6437 | * | |
6438 | * We reset it to 'level' at the end of this function. | |
6439 | */ | |
6440 | sched_domains_numa_levels = 0; | |
6441 | ||
cb83b629 PZ |
6442 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6443 | if (!sched_domains_numa_masks) | |
6444 | return; | |
6445 | ||
6446 | /* | |
6447 | * Now for each level, construct a mask per node which contains all | |
6448 | * cpus of nodes that are that many hops away from us. | |
6449 | */ | |
6450 | for (i = 0; i < level; i++) { | |
6451 | sched_domains_numa_masks[i] = | |
6452 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6453 | if (!sched_domains_numa_masks[i]) | |
6454 | return; | |
6455 | ||
6456 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6457 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6458 | if (!mask) |
6459 | return; | |
6460 | ||
6461 | sched_domains_numa_masks[i][j] = mask; | |
6462 | ||
6463 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6464 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6465 | continue; |
6466 | ||
6467 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6468 | } | |
6469 | } | |
6470 | } | |
6471 | ||
143e1e28 VG |
6472 | /* Compute default topology size */ |
6473 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6474 | ||
c515db8c | 6475 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6476 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6477 | if (!tl) | |
6478 | return; | |
6479 | ||
6480 | /* | |
6481 | * Copy the default topology bits.. | |
6482 | */ | |
143e1e28 VG |
6483 | for (i = 0; sched_domain_topology[i].mask; i++) |
6484 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6485 | |
6486 | /* | |
6487 | * .. and append 'j' levels of NUMA goodness. | |
6488 | */ | |
6489 | for (j = 0; j < level; i++, j++) { | |
6490 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6491 | .mask = sd_numa_mask, |
143e1e28 | 6492 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6493 | .flags = SDTL_OVERLAP, |
6494 | .numa_level = j, | |
143e1e28 | 6495 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6496 | }; |
6497 | } | |
6498 | ||
6499 | sched_domain_topology = tl; | |
5f7865f3 TC |
6500 | |
6501 | sched_domains_numa_levels = level; | |
9942f79b | 6502 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6503 | |
6504 | init_numa_topology_type(); | |
cb83b629 | 6505 | } |
301a5cba TC |
6506 | |
6507 | static void sched_domains_numa_masks_set(int cpu) | |
6508 | { | |
6509 | int i, j; | |
6510 | int node = cpu_to_node(cpu); | |
6511 | ||
6512 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6513 | for (j = 0; j < nr_node_ids; j++) { | |
6514 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6515 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6516 | } | |
6517 | } | |
6518 | } | |
6519 | ||
6520 | static void sched_domains_numa_masks_clear(int cpu) | |
6521 | { | |
6522 | int i, j; | |
6523 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6524 | for (j = 0; j < nr_node_ids; j++) | |
6525 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6526 | } | |
6527 | } | |
6528 | ||
6529 | /* | |
6530 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6531 | * are onlined. | |
6532 | */ | |
6533 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6534 | unsigned long action, | |
6535 | void *hcpu) | |
6536 | { | |
6537 | int cpu = (long)hcpu; | |
6538 | ||
6539 | switch (action & ~CPU_TASKS_FROZEN) { | |
6540 | case CPU_ONLINE: | |
6541 | sched_domains_numa_masks_set(cpu); | |
6542 | break; | |
6543 | ||
6544 | case CPU_DEAD: | |
6545 | sched_domains_numa_masks_clear(cpu); | |
6546 | break; | |
6547 | ||
6548 | default: | |
6549 | return NOTIFY_DONE; | |
6550 | } | |
6551 | ||
6552 | return NOTIFY_OK; | |
cb83b629 PZ |
6553 | } |
6554 | #else | |
6555 | static inline void sched_init_numa(void) | |
6556 | { | |
6557 | } | |
301a5cba TC |
6558 | |
6559 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6560 | unsigned long action, | |
6561 | void *hcpu) | |
6562 | { | |
6563 | return 0; | |
6564 | } | |
cb83b629 PZ |
6565 | #endif /* CONFIG_NUMA */ |
6566 | ||
54ab4ff4 PZ |
6567 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6568 | { | |
6569 | struct sched_domain_topology_level *tl; | |
6570 | int j; | |
6571 | ||
27723a68 | 6572 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6573 | struct sd_data *sdd = &tl->data; |
6574 | ||
6575 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6576 | if (!sdd->sd) | |
6577 | return -ENOMEM; | |
6578 | ||
6579 | sdd->sg = alloc_percpu(struct sched_group *); | |
6580 | if (!sdd->sg) | |
6581 | return -ENOMEM; | |
6582 | ||
63b2ca30 NP |
6583 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6584 | if (!sdd->sgc) | |
9c3f75cb PZ |
6585 | return -ENOMEM; |
6586 | ||
54ab4ff4 PZ |
6587 | for_each_cpu(j, cpu_map) { |
6588 | struct sched_domain *sd; | |
6589 | struct sched_group *sg; | |
63b2ca30 | 6590 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6591 | |
6592 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6593 | GFP_KERNEL, cpu_to_node(j)); | |
6594 | if (!sd) | |
6595 | return -ENOMEM; | |
6596 | ||
6597 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6598 | ||
6599 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6600 | GFP_KERNEL, cpu_to_node(j)); | |
6601 | if (!sg) | |
6602 | return -ENOMEM; | |
6603 | ||
30b4e9eb IM |
6604 | sg->next = sg; |
6605 | ||
54ab4ff4 | 6606 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6607 | |
63b2ca30 | 6608 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6609 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6610 | if (!sgc) |
9c3f75cb PZ |
6611 | return -ENOMEM; |
6612 | ||
63b2ca30 | 6613 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6614 | } |
6615 | } | |
6616 | ||
6617 | return 0; | |
6618 | } | |
6619 | ||
6620 | static void __sdt_free(const struct cpumask *cpu_map) | |
6621 | { | |
6622 | struct sched_domain_topology_level *tl; | |
6623 | int j; | |
6624 | ||
27723a68 | 6625 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6626 | struct sd_data *sdd = &tl->data; |
6627 | ||
6628 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6629 | struct sched_domain *sd; |
6630 | ||
6631 | if (sdd->sd) { | |
6632 | sd = *per_cpu_ptr(sdd->sd, j); | |
6633 | if (sd && (sd->flags & SD_OVERLAP)) | |
6634 | free_sched_groups(sd->groups, 0); | |
6635 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6636 | } | |
6637 | ||
6638 | if (sdd->sg) | |
6639 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6640 | if (sdd->sgc) |
6641 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6642 | } |
6643 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6644 | sdd->sd = NULL; |
54ab4ff4 | 6645 | free_percpu(sdd->sg); |
fb2cf2c6 | 6646 | sdd->sg = NULL; |
63b2ca30 NP |
6647 | free_percpu(sdd->sgc); |
6648 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6649 | } |
6650 | } | |
6651 | ||
2c402dc3 | 6652 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6653 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6654 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6655 | { |
143e1e28 | 6656 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6657 | if (!sd) |
d069b916 | 6658 | return child; |
2c402dc3 | 6659 | |
2c402dc3 | 6660 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6661 | if (child) { |
6662 | sd->level = child->level + 1; | |
6663 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6664 | child->parent = sd; |
c75e0128 | 6665 | sd->child = child; |
6ae72dff PZ |
6666 | |
6667 | if (!cpumask_subset(sched_domain_span(child), | |
6668 | sched_domain_span(sd))) { | |
6669 | pr_err("BUG: arch topology borken\n"); | |
6670 | #ifdef CONFIG_SCHED_DEBUG | |
6671 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6672 | child->name, sd->name); | |
6673 | #endif | |
6674 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6675 | cpumask_or(sched_domain_span(sd), | |
6676 | sched_domain_span(sd), | |
6677 | sched_domain_span(child)); | |
6678 | } | |
6679 | ||
60495e77 | 6680 | } |
a841f8ce | 6681 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6682 | |
6683 | return sd; | |
6684 | } | |
6685 | ||
2109b99e AH |
6686 | /* |
6687 | * Build sched domains for a given set of cpus and attach the sched domains | |
6688 | * to the individual cpus | |
6689 | */ | |
dce840a0 PZ |
6690 | static int build_sched_domains(const struct cpumask *cpu_map, |
6691 | struct sched_domain_attr *attr) | |
2109b99e | 6692 | { |
1c632169 | 6693 | enum s_alloc alloc_state; |
dce840a0 | 6694 | struct sched_domain *sd; |
2109b99e | 6695 | struct s_data d; |
822ff793 | 6696 | int i, ret = -ENOMEM; |
9c1cfda2 | 6697 | |
2109b99e AH |
6698 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6699 | if (alloc_state != sa_rootdomain) | |
6700 | goto error; | |
9c1cfda2 | 6701 | |
dce840a0 | 6702 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6703 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6704 | struct sched_domain_topology_level *tl; |
6705 | ||
3bd65a80 | 6706 | sd = NULL; |
27723a68 | 6707 | for_each_sd_topology(tl) { |
4a850cbe | 6708 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6709 | if (tl == sched_domain_topology) |
6710 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6711 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6712 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6713 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6714 | break; | |
e3589f6c | 6715 | } |
dce840a0 PZ |
6716 | } |
6717 | ||
6718 | /* Build the groups for the domains */ | |
6719 | for_each_cpu(i, cpu_map) { | |
6720 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6721 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6722 | if (sd->flags & SD_OVERLAP) { |
6723 | if (build_overlap_sched_groups(sd, i)) | |
6724 | goto error; | |
6725 | } else { | |
6726 | if (build_sched_groups(sd, i)) | |
6727 | goto error; | |
6728 | } | |
1cf51902 | 6729 | } |
a06dadbe | 6730 | } |
9c1cfda2 | 6731 | |
ced549fa | 6732 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6733 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6734 | if (!cpumask_test_cpu(i, cpu_map)) | |
6735 | continue; | |
9c1cfda2 | 6736 | |
dce840a0 PZ |
6737 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6738 | claim_allocations(i, sd); | |
63b2ca30 | 6739 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6740 | } |
f712c0c7 | 6741 | } |
9c1cfda2 | 6742 | |
1da177e4 | 6743 | /* Attach the domains */ |
dce840a0 | 6744 | rcu_read_lock(); |
abcd083a | 6745 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6746 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6747 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6748 | } |
dce840a0 | 6749 | rcu_read_unlock(); |
51888ca2 | 6750 | |
822ff793 | 6751 | ret = 0; |
51888ca2 | 6752 | error: |
2109b99e | 6753 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6754 | return ret; |
1da177e4 | 6755 | } |
029190c5 | 6756 | |
acc3f5d7 | 6757 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6758 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6759 | static struct sched_domain_attr *dattr_cur; |
6760 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6761 | |
6762 | /* | |
6763 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6764 | * cpumask) fails, then fallback to a single sched domain, |
6765 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6766 | */ |
4212823f | 6767 | static cpumask_var_t fallback_doms; |
029190c5 | 6768 | |
ee79d1bd HC |
6769 | /* |
6770 | * arch_update_cpu_topology lets virtualized architectures update the | |
6771 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6772 | * or 0 if it stayed the same. | |
6773 | */ | |
52f5684c | 6774 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6775 | { |
ee79d1bd | 6776 | return 0; |
22e52b07 HC |
6777 | } |
6778 | ||
acc3f5d7 RR |
6779 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6780 | { | |
6781 | int i; | |
6782 | cpumask_var_t *doms; | |
6783 | ||
6784 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6785 | if (!doms) | |
6786 | return NULL; | |
6787 | for (i = 0; i < ndoms; i++) { | |
6788 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6789 | free_sched_domains(doms, i); | |
6790 | return NULL; | |
6791 | } | |
6792 | } | |
6793 | return doms; | |
6794 | } | |
6795 | ||
6796 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6797 | { | |
6798 | unsigned int i; | |
6799 | for (i = 0; i < ndoms; i++) | |
6800 | free_cpumask_var(doms[i]); | |
6801 | kfree(doms); | |
6802 | } | |
6803 | ||
1a20ff27 | 6804 | /* |
41a2d6cf | 6805 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6806 | * For now this just excludes isolated cpus, but could be used to |
6807 | * exclude other special cases in the future. | |
1a20ff27 | 6808 | */ |
c4a8849a | 6809 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6810 | { |
7378547f MM |
6811 | int err; |
6812 | ||
22e52b07 | 6813 | arch_update_cpu_topology(); |
029190c5 | 6814 | ndoms_cur = 1; |
acc3f5d7 | 6815 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6816 | if (!doms_cur) |
acc3f5d7 RR |
6817 | doms_cur = &fallback_doms; |
6818 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6819 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6820 | register_sched_domain_sysctl(); |
7378547f MM |
6821 | |
6822 | return err; | |
1a20ff27 DG |
6823 | } |
6824 | ||
1a20ff27 DG |
6825 | /* |
6826 | * Detach sched domains from a group of cpus specified in cpu_map | |
6827 | * These cpus will now be attached to the NULL domain | |
6828 | */ | |
96f874e2 | 6829 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6830 | { |
6831 | int i; | |
6832 | ||
dce840a0 | 6833 | rcu_read_lock(); |
abcd083a | 6834 | for_each_cpu(i, cpu_map) |
57d885fe | 6835 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6836 | rcu_read_unlock(); |
1a20ff27 DG |
6837 | } |
6838 | ||
1d3504fc HS |
6839 | /* handle null as "default" */ |
6840 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6841 | struct sched_domain_attr *new, int idx_new) | |
6842 | { | |
6843 | struct sched_domain_attr tmp; | |
6844 | ||
6845 | /* fast path */ | |
6846 | if (!new && !cur) | |
6847 | return 1; | |
6848 | ||
6849 | tmp = SD_ATTR_INIT; | |
6850 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6851 | new ? (new + idx_new) : &tmp, | |
6852 | sizeof(struct sched_domain_attr)); | |
6853 | } | |
6854 | ||
029190c5 PJ |
6855 | /* |
6856 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6857 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6858 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6859 | * It destroys each deleted domain and builds each new domain. | |
6860 | * | |
acc3f5d7 | 6861 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6862 | * The masks don't intersect (don't overlap.) We should setup one |
6863 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6864 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6865 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6866 | * it as it is. | |
6867 | * | |
acc3f5d7 RR |
6868 | * The passed in 'doms_new' should be allocated using |
6869 | * alloc_sched_domains. This routine takes ownership of it and will | |
6870 | * free_sched_domains it when done with it. If the caller failed the | |
6871 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6872 | * and partition_sched_domains() will fallback to the single partition | |
6873 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6874 | * |
96f874e2 | 6875 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6876 | * ndoms_new == 0 is a special case for destroying existing domains, |
6877 | * and it will not create the default domain. | |
dfb512ec | 6878 | * |
029190c5 PJ |
6879 | * Call with hotplug lock held |
6880 | */ | |
acc3f5d7 | 6881 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6882 | struct sched_domain_attr *dattr_new) |
029190c5 | 6883 | { |
dfb512ec | 6884 | int i, j, n; |
d65bd5ec | 6885 | int new_topology; |
029190c5 | 6886 | |
712555ee | 6887 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6888 | |
7378547f MM |
6889 | /* always unregister in case we don't destroy any domains */ |
6890 | unregister_sched_domain_sysctl(); | |
6891 | ||
d65bd5ec HC |
6892 | /* Let architecture update cpu core mappings. */ |
6893 | new_topology = arch_update_cpu_topology(); | |
6894 | ||
dfb512ec | 6895 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6896 | |
6897 | /* Destroy deleted domains */ | |
6898 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6899 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6900 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6901 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6902 | goto match1; |
6903 | } | |
6904 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6905 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6906 | match1: |
6907 | ; | |
6908 | } | |
6909 | ||
c8d2d47a | 6910 | n = ndoms_cur; |
e761b772 | 6911 | if (doms_new == NULL) { |
c8d2d47a | 6912 | n = 0; |
acc3f5d7 | 6913 | doms_new = &fallback_doms; |
6ad4c188 | 6914 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6915 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6916 | } |
6917 | ||
029190c5 PJ |
6918 | /* Build new domains */ |
6919 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6920 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6921 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6922 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6923 | goto match2; |
6924 | } | |
6925 | /* no match - add a new doms_new */ | |
dce840a0 | 6926 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6927 | match2: |
6928 | ; | |
6929 | } | |
6930 | ||
6931 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6932 | if (doms_cur != &fallback_doms) |
6933 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6934 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6935 | doms_cur = doms_new; |
1d3504fc | 6936 | dattr_cur = dattr_new; |
029190c5 | 6937 | ndoms_cur = ndoms_new; |
7378547f MM |
6938 | |
6939 | register_sched_domain_sysctl(); | |
a1835615 | 6940 | |
712555ee | 6941 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6942 | } |
6943 | ||
d35be8ba SB |
6944 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6945 | ||
1da177e4 | 6946 | /* |
3a101d05 TH |
6947 | * Update cpusets according to cpu_active mask. If cpusets are |
6948 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6949 | * around partition_sched_domains(). | |
d35be8ba SB |
6950 | * |
6951 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6952 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6953 | */ |
0b2e918a TH |
6954 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6955 | void *hcpu) | |
e761b772 | 6956 | { |
d35be8ba SB |
6957 | switch (action) { |
6958 | case CPU_ONLINE_FROZEN: | |
6959 | case CPU_DOWN_FAILED_FROZEN: | |
6960 | ||
6961 | /* | |
6962 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6963 | * resume sequence. As long as this is not the last online | |
6964 | * operation in the resume sequence, just build a single sched | |
6965 | * domain, ignoring cpusets. | |
6966 | */ | |
6967 | num_cpus_frozen--; | |
6968 | if (likely(num_cpus_frozen)) { | |
6969 | partition_sched_domains(1, NULL, NULL); | |
6970 | break; | |
6971 | } | |
6972 | ||
6973 | /* | |
6974 | * This is the last CPU online operation. So fall through and | |
6975 | * restore the original sched domains by considering the | |
6976 | * cpuset configurations. | |
6977 | */ | |
6978 | ||
e761b772 | 6979 | case CPU_ONLINE: |
6ad4c188 | 6980 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6981 | cpuset_update_active_cpus(true); |
d35be8ba | 6982 | break; |
3a101d05 TH |
6983 | default: |
6984 | return NOTIFY_DONE; | |
6985 | } | |
d35be8ba | 6986 | return NOTIFY_OK; |
3a101d05 | 6987 | } |
e761b772 | 6988 | |
0b2e918a TH |
6989 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6990 | void *hcpu) | |
3a101d05 | 6991 | { |
d35be8ba | 6992 | switch (action) { |
3a101d05 | 6993 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6994 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6995 | break; |
6996 | case CPU_DOWN_PREPARE_FROZEN: | |
6997 | num_cpus_frozen++; | |
6998 | partition_sched_domains(1, NULL, NULL); | |
6999 | break; | |
e761b772 MK |
7000 | default: |
7001 | return NOTIFY_DONE; | |
7002 | } | |
d35be8ba | 7003 | return NOTIFY_OK; |
e761b772 | 7004 | } |
e761b772 | 7005 | |
1da177e4 LT |
7006 | void __init sched_init_smp(void) |
7007 | { | |
dcc30a35 RR |
7008 | cpumask_var_t non_isolated_cpus; |
7009 | ||
7010 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7011 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7012 | |
cb83b629 PZ |
7013 | sched_init_numa(); |
7014 | ||
6acce3ef PZ |
7015 | /* |
7016 | * There's no userspace yet to cause hotplug operations; hence all the | |
7017 | * cpu masks are stable and all blatant races in the below code cannot | |
7018 | * happen. | |
7019 | */ | |
712555ee | 7020 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7021 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7022 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7023 | if (cpumask_empty(non_isolated_cpus)) | |
7024 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7025 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7026 | |
301a5cba | 7027 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7028 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7029 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7030 | |
b328ca18 | 7031 | init_hrtick(); |
5c1e1767 NP |
7032 | |
7033 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7034 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7035 | BUG(); |
19978ca6 | 7036 | sched_init_granularity(); |
dcc30a35 | 7037 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7038 | |
0e3900e6 | 7039 | init_sched_rt_class(); |
1baca4ce | 7040 | init_sched_dl_class(); |
1da177e4 LT |
7041 | } |
7042 | #else | |
7043 | void __init sched_init_smp(void) | |
7044 | { | |
19978ca6 | 7045 | sched_init_granularity(); |
1da177e4 LT |
7046 | } |
7047 | #endif /* CONFIG_SMP */ | |
7048 | ||
cd1bb94b AB |
7049 | const_debug unsigned int sysctl_timer_migration = 1; |
7050 | ||
1da177e4 LT |
7051 | int in_sched_functions(unsigned long addr) |
7052 | { | |
1da177e4 LT |
7053 | return in_lock_functions(addr) || |
7054 | (addr >= (unsigned long)__sched_text_start | |
7055 | && addr < (unsigned long)__sched_text_end); | |
7056 | } | |
7057 | ||
029632fb | 7058 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7059 | /* |
7060 | * Default task group. | |
7061 | * Every task in system belongs to this group at bootup. | |
7062 | */ | |
029632fb | 7063 | struct task_group root_task_group; |
35cf4e50 | 7064 | LIST_HEAD(task_groups); |
052f1dc7 | 7065 | #endif |
6f505b16 | 7066 | |
e6252c3e | 7067 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7068 | |
1da177e4 LT |
7069 | void __init sched_init(void) |
7070 | { | |
dd41f596 | 7071 | int i, j; |
434d53b0 MT |
7072 | unsigned long alloc_size = 0, ptr; |
7073 | ||
7074 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7075 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7076 | #endif | |
7077 | #ifdef CONFIG_RT_GROUP_SCHED | |
7078 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7079 | #endif |
df7c8e84 | 7080 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7081 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7082 | #endif |
434d53b0 | 7083 | if (alloc_size) { |
36b7b6d4 | 7084 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7085 | |
7086 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7087 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7088 | ptr += nr_cpu_ids * sizeof(void **); |
7089 | ||
07e06b01 | 7090 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7091 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7092 | |
6d6bc0ad | 7093 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7094 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7095 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7096 | ptr += nr_cpu_ids * sizeof(void **); |
7097 | ||
07e06b01 | 7098 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7099 | ptr += nr_cpu_ids * sizeof(void **); |
7100 | ||
6d6bc0ad | 7101 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7102 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7103 | for_each_possible_cpu(i) { | |
e6252c3e | 7104 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
7105 | ptr += cpumask_size(); |
7106 | } | |
7107 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7108 | } |
dd41f596 | 7109 | |
332ac17e DF |
7110 | init_rt_bandwidth(&def_rt_bandwidth, |
7111 | global_rt_period(), global_rt_runtime()); | |
7112 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7113 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7114 | |
57d885fe GH |
7115 | #ifdef CONFIG_SMP |
7116 | init_defrootdomain(); | |
7117 | #endif | |
7118 | ||
d0b27fa7 | 7119 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7120 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7121 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7122 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7123 | |
7c941438 | 7124 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7125 | list_add(&root_task_group.list, &task_groups); |
7126 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7127 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7128 | autogroup_init(&init_task); |
54c707e9 | 7129 | |
7c941438 | 7130 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7131 | |
0a945022 | 7132 | for_each_possible_cpu(i) { |
70b97a7f | 7133 | struct rq *rq; |
1da177e4 LT |
7134 | |
7135 | rq = cpu_rq(i); | |
05fa785c | 7136 | raw_spin_lock_init(&rq->lock); |
7897986b | 7137 | rq->nr_running = 0; |
dce48a84 TG |
7138 | rq->calc_load_active = 0; |
7139 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7140 | init_cfs_rq(&rq->cfs); |
6f505b16 | 7141 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 7142 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 7143 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7144 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7145 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7146 | /* |
07e06b01 | 7147 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7148 | * |
7149 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7150 | * gets 100% of the cpu resources in the system. This overall | |
7151 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7152 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7153 | * based on each entity's (task or task-group's) weight |
7154 | * (se->load.weight). | |
7155 | * | |
07e06b01 | 7156 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7157 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7158 | * then A0's share of the cpu resource is: | |
7159 | * | |
0d905bca | 7160 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7161 | * |
07e06b01 YZ |
7162 | * We achieve this by letting root_task_group's tasks sit |
7163 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7164 | */ |
ab84d31e | 7165 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7166 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7167 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7168 | ||
7169 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7170 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7171 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7172 | #endif |
1da177e4 | 7173 | |
dd41f596 IM |
7174 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7175 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7176 | |
7177 | rq->last_load_update_tick = jiffies; | |
7178 | ||
1da177e4 | 7179 | #ifdef CONFIG_SMP |
41c7ce9a | 7180 | rq->sd = NULL; |
57d885fe | 7181 | rq->rd = NULL; |
ca8ce3d0 | 7182 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 7183 | rq->post_schedule = 0; |
1da177e4 | 7184 | rq->active_balance = 0; |
dd41f596 | 7185 | rq->next_balance = jiffies; |
1da177e4 | 7186 | rq->push_cpu = 0; |
0a2966b4 | 7187 | rq->cpu = i; |
1f11eb6a | 7188 | rq->online = 0; |
eae0c9df MG |
7189 | rq->idle_stamp = 0; |
7190 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7191 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7192 | |
7193 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7194 | ||
dc938520 | 7195 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7196 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7197 | rq->nohz_flags = 0; |
83cd4fe2 | 7198 | #endif |
265f22a9 FW |
7199 | #ifdef CONFIG_NO_HZ_FULL |
7200 | rq->last_sched_tick = 0; | |
7201 | #endif | |
1da177e4 | 7202 | #endif |
8f4d37ec | 7203 | init_rq_hrtick(rq); |
1da177e4 | 7204 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7205 | } |
7206 | ||
2dd73a4f | 7207 | set_load_weight(&init_task); |
b50f60ce | 7208 | |
e107be36 AK |
7209 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7210 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7211 | #endif | |
7212 | ||
1da177e4 LT |
7213 | /* |
7214 | * The boot idle thread does lazy MMU switching as well: | |
7215 | */ | |
7216 | atomic_inc(&init_mm.mm_count); | |
7217 | enter_lazy_tlb(&init_mm, current); | |
7218 | ||
7219 | /* | |
7220 | * Make us the idle thread. Technically, schedule() should not be | |
7221 | * called from this thread, however somewhere below it might be, | |
7222 | * but because we are the idle thread, we just pick up running again | |
7223 | * when this runqueue becomes "idle". | |
7224 | */ | |
7225 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7226 | |
7227 | calc_load_update = jiffies + LOAD_FREQ; | |
7228 | ||
dd41f596 IM |
7229 | /* |
7230 | * During early bootup we pretend to be a normal task: | |
7231 | */ | |
7232 | current->sched_class = &fair_sched_class; | |
6892b75e | 7233 | |
bf4d83f6 | 7234 | #ifdef CONFIG_SMP |
4cb98839 | 7235 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7236 | /* May be allocated at isolcpus cmdline parse time */ |
7237 | if (cpu_isolated_map == NULL) | |
7238 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7239 | idle_thread_set_boot_cpu(); |
a803f026 | 7240 | set_cpu_rq_start_time(); |
029632fb PZ |
7241 | #endif |
7242 | init_sched_fair_class(); | |
6a7b3dc3 | 7243 | |
6892b75e | 7244 | scheduler_running = 1; |
1da177e4 LT |
7245 | } |
7246 | ||
d902db1e | 7247 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7248 | static inline int preempt_count_equals(int preempt_offset) |
7249 | { | |
234da7bc | 7250 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7251 | |
4ba8216c | 7252 | return (nested == preempt_offset); |
e4aafea2 FW |
7253 | } |
7254 | ||
d894837f | 7255 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7256 | { |
1da177e4 LT |
7257 | static unsigned long prev_jiffy; /* ratelimiting */ |
7258 | ||
b3fbab05 | 7259 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7260 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7261 | !is_idle_task(current)) || | |
e4aafea2 | 7262 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7263 | return; |
7264 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7265 | return; | |
7266 | prev_jiffy = jiffies; | |
7267 | ||
3df0fc5b PZ |
7268 | printk(KERN_ERR |
7269 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7270 | file, line); | |
7271 | printk(KERN_ERR | |
7272 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7273 | in_atomic(), irqs_disabled(), | |
7274 | current->pid, current->comm); | |
aef745fc IM |
7275 | |
7276 | debug_show_held_locks(current); | |
7277 | if (irqs_disabled()) | |
7278 | print_irqtrace_events(current); | |
8f47b187 TG |
7279 | #ifdef CONFIG_DEBUG_PREEMPT |
7280 | if (!preempt_count_equals(preempt_offset)) { | |
7281 | pr_err("Preemption disabled at:"); | |
7282 | print_ip_sym(current->preempt_disable_ip); | |
7283 | pr_cont("\n"); | |
7284 | } | |
7285 | #endif | |
aef745fc | 7286 | dump_stack(); |
1da177e4 LT |
7287 | } |
7288 | EXPORT_SYMBOL(__might_sleep); | |
7289 | #endif | |
7290 | ||
7291 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7292 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7293 | { | |
da7a735e | 7294 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7295 | struct sched_attr attr = { |
7296 | .sched_policy = SCHED_NORMAL, | |
7297 | }; | |
da7a735e | 7298 | int old_prio = p->prio; |
da0c1e65 | 7299 | int queued; |
3e51f33f | 7300 | |
da0c1e65 KT |
7301 | queued = task_on_rq_queued(p); |
7302 | if (queued) | |
4ca9b72b | 7303 | dequeue_task(rq, p, 0); |
d50dde5a | 7304 | __setscheduler(rq, p, &attr); |
da0c1e65 | 7305 | if (queued) { |
4ca9b72b | 7306 | enqueue_task(rq, p, 0); |
8875125e | 7307 | resched_curr(rq); |
3a5e4dc1 | 7308 | } |
da7a735e PZ |
7309 | |
7310 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7311 | } |
7312 | ||
1da177e4 LT |
7313 | void normalize_rt_tasks(void) |
7314 | { | |
a0f98a1c | 7315 | struct task_struct *g, *p; |
1da177e4 | 7316 | unsigned long flags; |
70b97a7f | 7317 | struct rq *rq; |
1da177e4 | 7318 | |
3472eaa1 | 7319 | read_lock(&tasklist_lock); |
5d07f420 | 7320 | for_each_process_thread(g, p) { |
178be793 IM |
7321 | /* |
7322 | * Only normalize user tasks: | |
7323 | */ | |
3472eaa1 | 7324 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7325 | continue; |
7326 | ||
6cfb0d5d | 7327 | p->se.exec_start = 0; |
6cfb0d5d | 7328 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7329 | p->se.statistics.wait_start = 0; |
7330 | p->se.statistics.sleep_start = 0; | |
7331 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7332 | #endif |
dd41f596 | 7333 | |
aab03e05 | 7334 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7335 | /* |
7336 | * Renice negative nice level userspace | |
7337 | * tasks back to 0: | |
7338 | */ | |
3472eaa1 | 7339 | if (task_nice(p) < 0) |
dd41f596 | 7340 | set_user_nice(p, 0); |
1da177e4 | 7341 | continue; |
dd41f596 | 7342 | } |
1da177e4 | 7343 | |
3472eaa1 | 7344 | rq = task_rq_lock(p, &flags); |
178be793 | 7345 | normalize_task(rq, p); |
3472eaa1 | 7346 | task_rq_unlock(rq, p, &flags); |
5d07f420 | 7347 | } |
3472eaa1 | 7348 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7349 | } |
7350 | ||
7351 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7352 | |
67fc4e0c | 7353 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7354 | /* |
67fc4e0c | 7355 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7356 | * |
7357 | * They can only be called when the whole system has been | |
7358 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7359 | * activity can take place. Using them for anything else would | |
7360 | * be a serious bug, and as a result, they aren't even visible | |
7361 | * under any other configuration. | |
7362 | */ | |
7363 | ||
7364 | /** | |
7365 | * curr_task - return the current task for a given cpu. | |
7366 | * @cpu: the processor in question. | |
7367 | * | |
7368 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7369 | * |
7370 | * Return: The current task for @cpu. | |
1df5c10a | 7371 | */ |
36c8b586 | 7372 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7373 | { |
7374 | return cpu_curr(cpu); | |
7375 | } | |
7376 | ||
67fc4e0c JW |
7377 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7378 | ||
7379 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7380 | /** |
7381 | * set_curr_task - set the current task for a given cpu. | |
7382 | * @cpu: the processor in question. | |
7383 | * @p: the task pointer to set. | |
7384 | * | |
7385 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7386 | * are serviced on a separate stack. It allows the architecture to switch the |
7387 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7388 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7389 | * and caller must save the original value of the current task (see | |
7390 | * curr_task() above) and restore that value before reenabling interrupts and | |
7391 | * re-starting the system. | |
7392 | * | |
7393 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7394 | */ | |
36c8b586 | 7395 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7396 | { |
7397 | cpu_curr(cpu) = p; | |
7398 | } | |
7399 | ||
7400 | #endif | |
29f59db3 | 7401 | |
7c941438 | 7402 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7403 | /* task_group_lock serializes the addition/removal of task groups */ |
7404 | static DEFINE_SPINLOCK(task_group_lock); | |
7405 | ||
bccbe08a PZ |
7406 | static void free_sched_group(struct task_group *tg) |
7407 | { | |
7408 | free_fair_sched_group(tg); | |
7409 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7410 | autogroup_free(tg); |
bccbe08a PZ |
7411 | kfree(tg); |
7412 | } | |
7413 | ||
7414 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7415 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7416 | { |
7417 | struct task_group *tg; | |
bccbe08a PZ |
7418 | |
7419 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7420 | if (!tg) | |
7421 | return ERR_PTR(-ENOMEM); | |
7422 | ||
ec7dc8ac | 7423 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7424 | goto err; |
7425 | ||
ec7dc8ac | 7426 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7427 | goto err; |
7428 | ||
ace783b9 LZ |
7429 | return tg; |
7430 | ||
7431 | err: | |
7432 | free_sched_group(tg); | |
7433 | return ERR_PTR(-ENOMEM); | |
7434 | } | |
7435 | ||
7436 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7437 | { | |
7438 | unsigned long flags; | |
7439 | ||
8ed36996 | 7440 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7441 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7442 | |
7443 | WARN_ON(!parent); /* root should already exist */ | |
7444 | ||
7445 | tg->parent = parent; | |
f473aa5e | 7446 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7447 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7448 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7449 | } |
7450 | ||
9b5b7751 | 7451 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7452 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7453 | { |
29f59db3 | 7454 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7455 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7456 | } |
7457 | ||
9b5b7751 | 7458 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7459 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7460 | { |
7461 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7462 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7463 | } | |
7464 | ||
7465 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7466 | { |
8ed36996 | 7467 | unsigned long flags; |
9b5b7751 | 7468 | int i; |
29f59db3 | 7469 | |
3d4b47b4 PZ |
7470 | /* end participation in shares distribution */ |
7471 | for_each_possible_cpu(i) | |
bccbe08a | 7472 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7473 | |
7474 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7475 | list_del_rcu(&tg->list); |
f473aa5e | 7476 | list_del_rcu(&tg->siblings); |
8ed36996 | 7477 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7478 | } |
7479 | ||
9b5b7751 | 7480 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7481 | * The caller of this function should have put the task in its new group |
7482 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7483 | * reflect its new group. | |
9b5b7751 SV |
7484 | */ |
7485 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7486 | { |
8323f26c | 7487 | struct task_group *tg; |
da0c1e65 | 7488 | int queued, running; |
29f59db3 SV |
7489 | unsigned long flags; |
7490 | struct rq *rq; | |
7491 | ||
7492 | rq = task_rq_lock(tsk, &flags); | |
7493 | ||
051a1d1a | 7494 | running = task_current(rq, tsk); |
da0c1e65 | 7495 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7496 | |
da0c1e65 | 7497 | if (queued) |
29f59db3 | 7498 | dequeue_task(rq, tsk, 0); |
0e1f3483 | 7499 | if (unlikely(running)) |
f3cd1c4e | 7500 | put_prev_task(rq, tsk); |
29f59db3 | 7501 | |
073219e9 | 7502 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7503 | lockdep_is_held(&tsk->sighand->siglock)), |
7504 | struct task_group, css); | |
7505 | tg = autogroup_task_group(tsk, tg); | |
7506 | tsk->sched_task_group = tg; | |
7507 | ||
810b3817 | 7508 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7509 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7510 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7511 | else |
810b3817 | 7512 | #endif |
b2b5ce02 | 7513 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7514 | |
0e1f3483 HS |
7515 | if (unlikely(running)) |
7516 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7517 | if (queued) |
371fd7e7 | 7518 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7519 | |
0122ec5b | 7520 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7521 | } |
7c941438 | 7522 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7523 | |
a790de99 PT |
7524 | #ifdef CONFIG_RT_GROUP_SCHED |
7525 | /* | |
7526 | * Ensure that the real time constraints are schedulable. | |
7527 | */ | |
7528 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7529 | |
9a7e0b18 PZ |
7530 | /* Must be called with tasklist_lock held */ |
7531 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7532 | { |
9a7e0b18 | 7533 | struct task_struct *g, *p; |
b40b2e8e | 7534 | |
5d07f420 | 7535 | for_each_process_thread(g, p) { |
8651c658 | 7536 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7537 | return 1; |
5d07f420 | 7538 | } |
b40b2e8e | 7539 | |
9a7e0b18 PZ |
7540 | return 0; |
7541 | } | |
b40b2e8e | 7542 | |
9a7e0b18 PZ |
7543 | struct rt_schedulable_data { |
7544 | struct task_group *tg; | |
7545 | u64 rt_period; | |
7546 | u64 rt_runtime; | |
7547 | }; | |
b40b2e8e | 7548 | |
a790de99 | 7549 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7550 | { |
7551 | struct rt_schedulable_data *d = data; | |
7552 | struct task_group *child; | |
7553 | unsigned long total, sum = 0; | |
7554 | u64 period, runtime; | |
b40b2e8e | 7555 | |
9a7e0b18 PZ |
7556 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7557 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7558 | |
9a7e0b18 PZ |
7559 | if (tg == d->tg) { |
7560 | period = d->rt_period; | |
7561 | runtime = d->rt_runtime; | |
b40b2e8e | 7562 | } |
b40b2e8e | 7563 | |
4653f803 PZ |
7564 | /* |
7565 | * Cannot have more runtime than the period. | |
7566 | */ | |
7567 | if (runtime > period && runtime != RUNTIME_INF) | |
7568 | return -EINVAL; | |
6f505b16 | 7569 | |
4653f803 PZ |
7570 | /* |
7571 | * Ensure we don't starve existing RT tasks. | |
7572 | */ | |
9a7e0b18 PZ |
7573 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7574 | return -EBUSY; | |
6f505b16 | 7575 | |
9a7e0b18 | 7576 | total = to_ratio(period, runtime); |
6f505b16 | 7577 | |
4653f803 PZ |
7578 | /* |
7579 | * Nobody can have more than the global setting allows. | |
7580 | */ | |
7581 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7582 | return -EINVAL; | |
6f505b16 | 7583 | |
4653f803 PZ |
7584 | /* |
7585 | * The sum of our children's runtime should not exceed our own. | |
7586 | */ | |
9a7e0b18 PZ |
7587 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7588 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7589 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7590 | |
9a7e0b18 PZ |
7591 | if (child == d->tg) { |
7592 | period = d->rt_period; | |
7593 | runtime = d->rt_runtime; | |
7594 | } | |
6f505b16 | 7595 | |
9a7e0b18 | 7596 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7597 | } |
6f505b16 | 7598 | |
9a7e0b18 PZ |
7599 | if (sum > total) |
7600 | return -EINVAL; | |
7601 | ||
7602 | return 0; | |
6f505b16 PZ |
7603 | } |
7604 | ||
9a7e0b18 | 7605 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7606 | { |
8277434e PT |
7607 | int ret; |
7608 | ||
9a7e0b18 PZ |
7609 | struct rt_schedulable_data data = { |
7610 | .tg = tg, | |
7611 | .rt_period = period, | |
7612 | .rt_runtime = runtime, | |
7613 | }; | |
7614 | ||
8277434e PT |
7615 | rcu_read_lock(); |
7616 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7617 | rcu_read_unlock(); | |
7618 | ||
7619 | return ret; | |
521f1a24 DG |
7620 | } |
7621 | ||
ab84d31e | 7622 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7623 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7624 | { |
ac086bc2 | 7625 | int i, err = 0; |
9f0c1e56 | 7626 | |
9f0c1e56 | 7627 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7628 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7629 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7630 | if (err) | |
9f0c1e56 | 7631 | goto unlock; |
ac086bc2 | 7632 | |
0986b11b | 7633 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7634 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7635 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7636 | |
7637 | for_each_possible_cpu(i) { | |
7638 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7639 | ||
0986b11b | 7640 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7641 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7642 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7643 | } |
0986b11b | 7644 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7645 | unlock: |
521f1a24 | 7646 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7647 | mutex_unlock(&rt_constraints_mutex); |
7648 | ||
7649 | return err; | |
6f505b16 PZ |
7650 | } |
7651 | ||
25cc7da7 | 7652 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7653 | { |
7654 | u64 rt_runtime, rt_period; | |
7655 | ||
7656 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7657 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7658 | if (rt_runtime_us < 0) | |
7659 | rt_runtime = RUNTIME_INF; | |
7660 | ||
ab84d31e | 7661 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7662 | } |
7663 | ||
25cc7da7 | 7664 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7665 | { |
7666 | u64 rt_runtime_us; | |
7667 | ||
d0b27fa7 | 7668 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7669 | return -1; |
7670 | ||
d0b27fa7 | 7671 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7672 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7673 | return rt_runtime_us; | |
7674 | } | |
d0b27fa7 | 7675 | |
25cc7da7 | 7676 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7677 | { |
7678 | u64 rt_runtime, rt_period; | |
7679 | ||
7680 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7681 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7682 | ||
619b0488 R |
7683 | if (rt_period == 0) |
7684 | return -EINVAL; | |
7685 | ||
ab84d31e | 7686 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7687 | } |
7688 | ||
25cc7da7 | 7689 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7690 | { |
7691 | u64 rt_period_us; | |
7692 | ||
7693 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7694 | do_div(rt_period_us, NSEC_PER_USEC); | |
7695 | return rt_period_us; | |
7696 | } | |
332ac17e | 7697 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7698 | |
332ac17e | 7699 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7700 | static int sched_rt_global_constraints(void) |
7701 | { | |
7702 | int ret = 0; | |
7703 | ||
7704 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7705 | read_lock(&tasklist_lock); |
4653f803 | 7706 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7707 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7708 | mutex_unlock(&rt_constraints_mutex); |
7709 | ||
7710 | return ret; | |
7711 | } | |
54e99124 | 7712 | |
25cc7da7 | 7713 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7714 | { |
7715 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7716 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7717 | return 0; | |
7718 | ||
7719 | return 1; | |
7720 | } | |
7721 | ||
6d6bc0ad | 7722 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7723 | static int sched_rt_global_constraints(void) |
7724 | { | |
ac086bc2 | 7725 | unsigned long flags; |
332ac17e | 7726 | int i, ret = 0; |
ec5d4989 | 7727 | |
0986b11b | 7728 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7729 | for_each_possible_cpu(i) { |
7730 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7731 | ||
0986b11b | 7732 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7733 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7734 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7735 | } |
0986b11b | 7736 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7737 | |
332ac17e | 7738 | return ret; |
d0b27fa7 | 7739 | } |
6d6bc0ad | 7740 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7741 | |
332ac17e DF |
7742 | static int sched_dl_global_constraints(void) |
7743 | { | |
1724813d PZ |
7744 | u64 runtime = global_rt_runtime(); |
7745 | u64 period = global_rt_period(); | |
332ac17e | 7746 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7747 | struct dl_bw *dl_b; |
1724813d | 7748 | int cpu, ret = 0; |
49516342 | 7749 | unsigned long flags; |
332ac17e DF |
7750 | |
7751 | /* | |
7752 | * Here we want to check the bandwidth not being set to some | |
7753 | * value smaller than the currently allocated bandwidth in | |
7754 | * any of the root_domains. | |
7755 | * | |
7756 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7757 | * cycling on root_domains... Discussion on different/better | |
7758 | * solutions is welcome! | |
7759 | */ | |
1724813d | 7760 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7761 | rcu_read_lock_sched(); |
7762 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7763 | |
49516342 | 7764 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7765 | if (new_bw < dl_b->total_bw) |
7766 | ret = -EBUSY; | |
49516342 | 7767 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7768 | |
f10e00f4 KT |
7769 | rcu_read_unlock_sched(); |
7770 | ||
1724813d PZ |
7771 | if (ret) |
7772 | break; | |
332ac17e DF |
7773 | } |
7774 | ||
1724813d | 7775 | return ret; |
332ac17e DF |
7776 | } |
7777 | ||
1724813d | 7778 | static void sched_dl_do_global(void) |
ce0dbbbb | 7779 | { |
1724813d | 7780 | u64 new_bw = -1; |
f10e00f4 | 7781 | struct dl_bw *dl_b; |
1724813d | 7782 | int cpu; |
49516342 | 7783 | unsigned long flags; |
ce0dbbbb | 7784 | |
1724813d PZ |
7785 | def_dl_bandwidth.dl_period = global_rt_period(); |
7786 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7787 | ||
7788 | if (global_rt_runtime() != RUNTIME_INF) | |
7789 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7790 | ||
7791 | /* | |
7792 | * FIXME: As above... | |
7793 | */ | |
7794 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7795 | rcu_read_lock_sched(); |
7796 | dl_b = dl_bw_of(cpu); | |
1724813d | 7797 | |
49516342 | 7798 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7799 | dl_b->bw = new_bw; |
49516342 | 7800 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
7801 | |
7802 | rcu_read_unlock_sched(); | |
ce0dbbbb | 7803 | } |
1724813d PZ |
7804 | } |
7805 | ||
7806 | static int sched_rt_global_validate(void) | |
7807 | { | |
7808 | if (sysctl_sched_rt_period <= 0) | |
7809 | return -EINVAL; | |
7810 | ||
e9e7cb38 JL |
7811 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7812 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7813 | return -EINVAL; |
7814 | ||
7815 | return 0; | |
7816 | } | |
7817 | ||
7818 | static void sched_rt_do_global(void) | |
7819 | { | |
7820 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7821 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7822 | } |
7823 | ||
d0b27fa7 | 7824 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7825 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7826 | loff_t *ppos) |
7827 | { | |
d0b27fa7 PZ |
7828 | int old_period, old_runtime; |
7829 | static DEFINE_MUTEX(mutex); | |
1724813d | 7830 | int ret; |
d0b27fa7 PZ |
7831 | |
7832 | mutex_lock(&mutex); | |
7833 | old_period = sysctl_sched_rt_period; | |
7834 | old_runtime = sysctl_sched_rt_runtime; | |
7835 | ||
8d65af78 | 7836 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7837 | |
7838 | if (!ret && write) { | |
1724813d PZ |
7839 | ret = sched_rt_global_validate(); |
7840 | if (ret) | |
7841 | goto undo; | |
7842 | ||
d0b27fa7 | 7843 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7844 | if (ret) |
7845 | goto undo; | |
7846 | ||
7847 | ret = sched_dl_global_constraints(); | |
7848 | if (ret) | |
7849 | goto undo; | |
7850 | ||
7851 | sched_rt_do_global(); | |
7852 | sched_dl_do_global(); | |
7853 | } | |
7854 | if (0) { | |
7855 | undo: | |
7856 | sysctl_sched_rt_period = old_period; | |
7857 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7858 | } |
7859 | mutex_unlock(&mutex); | |
7860 | ||
7861 | return ret; | |
7862 | } | |
68318b8e | 7863 | |
1724813d | 7864 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7865 | void __user *buffer, size_t *lenp, |
7866 | loff_t *ppos) | |
7867 | { | |
7868 | int ret; | |
332ac17e | 7869 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7870 | |
7871 | mutex_lock(&mutex); | |
332ac17e | 7872 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7873 | /* make sure that internally we keep jiffies */ |
7874 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7875 | if (!ret && write) { |
1724813d PZ |
7876 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7877 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7878 | } |
7879 | mutex_unlock(&mutex); | |
332ac17e DF |
7880 | return ret; |
7881 | } | |
7882 | ||
052f1dc7 | 7883 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7884 | |
a7c6d554 | 7885 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7886 | { |
a7c6d554 | 7887 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7888 | } |
7889 | ||
eb95419b TH |
7890 | static struct cgroup_subsys_state * |
7891 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7892 | { |
eb95419b TH |
7893 | struct task_group *parent = css_tg(parent_css); |
7894 | struct task_group *tg; | |
68318b8e | 7895 | |
eb95419b | 7896 | if (!parent) { |
68318b8e | 7897 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7898 | return &root_task_group.css; |
68318b8e SV |
7899 | } |
7900 | ||
ec7dc8ac | 7901 | tg = sched_create_group(parent); |
68318b8e SV |
7902 | if (IS_ERR(tg)) |
7903 | return ERR_PTR(-ENOMEM); | |
7904 | ||
68318b8e SV |
7905 | return &tg->css; |
7906 | } | |
7907 | ||
eb95419b | 7908 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7909 | { |
eb95419b | 7910 | struct task_group *tg = css_tg(css); |
5c9d535b | 7911 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 7912 | |
63876986 TH |
7913 | if (parent) |
7914 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7915 | return 0; |
7916 | } | |
7917 | ||
eb95419b | 7918 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7919 | { |
eb95419b | 7920 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7921 | |
7922 | sched_destroy_group(tg); | |
7923 | } | |
7924 | ||
eb95419b | 7925 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7926 | { |
eb95419b | 7927 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7928 | |
7929 | sched_offline_group(tg); | |
7930 | } | |
7931 | ||
eeb61e53 KT |
7932 | static void cpu_cgroup_fork(struct task_struct *task) |
7933 | { | |
7934 | sched_move_task(task); | |
7935 | } | |
7936 | ||
eb95419b | 7937 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7938 | struct cgroup_taskset *tset) |
68318b8e | 7939 | { |
bb9d97b6 TH |
7940 | struct task_struct *task; |
7941 | ||
924f0d9a | 7942 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7943 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7944 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7945 | return -EINVAL; |
b68aa230 | 7946 | #else |
bb9d97b6 TH |
7947 | /* We don't support RT-tasks being in separate groups */ |
7948 | if (task->sched_class != &fair_sched_class) | |
7949 | return -EINVAL; | |
b68aa230 | 7950 | #endif |
bb9d97b6 | 7951 | } |
be367d09 BB |
7952 | return 0; |
7953 | } | |
68318b8e | 7954 | |
eb95419b | 7955 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7956 | struct cgroup_taskset *tset) |
68318b8e | 7957 | { |
bb9d97b6 TH |
7958 | struct task_struct *task; |
7959 | ||
924f0d9a | 7960 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7961 | sched_move_task(task); |
68318b8e SV |
7962 | } |
7963 | ||
eb95419b TH |
7964 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7965 | struct cgroup_subsys_state *old_css, | |
7966 | struct task_struct *task) | |
068c5cc5 PZ |
7967 | { |
7968 | /* | |
7969 | * cgroup_exit() is called in the copy_process() failure path. | |
7970 | * Ignore this case since the task hasn't ran yet, this avoids | |
7971 | * trying to poke a half freed task state from generic code. | |
7972 | */ | |
7973 | if (!(task->flags & PF_EXITING)) | |
7974 | return; | |
7975 | ||
7976 | sched_move_task(task); | |
7977 | } | |
7978 | ||
052f1dc7 | 7979 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7980 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7981 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7982 | { |
182446d0 | 7983 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7984 | } |
7985 | ||
182446d0 TH |
7986 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7987 | struct cftype *cft) | |
68318b8e | 7988 | { |
182446d0 | 7989 | struct task_group *tg = css_tg(css); |
68318b8e | 7990 | |
c8b28116 | 7991 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7992 | } |
ab84d31e PT |
7993 | |
7994 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7995 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7996 | ||
ab84d31e PT |
7997 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7998 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7999 | ||
a790de99 PT |
8000 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8001 | ||
ab84d31e PT |
8002 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8003 | { | |
56f570e5 | 8004 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8005 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8006 | |
8007 | if (tg == &root_task_group) | |
8008 | return -EINVAL; | |
8009 | ||
8010 | /* | |
8011 | * Ensure we have at some amount of bandwidth every period. This is | |
8012 | * to prevent reaching a state of large arrears when throttled via | |
8013 | * entity_tick() resulting in prolonged exit starvation. | |
8014 | */ | |
8015 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8016 | return -EINVAL; | |
8017 | ||
8018 | /* | |
8019 | * Likewise, bound things on the otherside by preventing insane quota | |
8020 | * periods. This also allows us to normalize in computing quota | |
8021 | * feasibility. | |
8022 | */ | |
8023 | if (period > max_cfs_quota_period) | |
8024 | return -EINVAL; | |
8025 | ||
0e59bdae KT |
8026 | /* |
8027 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8028 | * unthrottle_offline_cfs_rqs(). | |
8029 | */ | |
8030 | get_online_cpus(); | |
a790de99 PT |
8031 | mutex_lock(&cfs_constraints_mutex); |
8032 | ret = __cfs_schedulable(tg, period, quota); | |
8033 | if (ret) | |
8034 | goto out_unlock; | |
8035 | ||
58088ad0 | 8036 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8037 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8038 | /* |
8039 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8040 | * before making related changes, and on->off must occur afterwards | |
8041 | */ | |
8042 | if (runtime_enabled && !runtime_was_enabled) | |
8043 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8044 | raw_spin_lock_irq(&cfs_b->lock); |
8045 | cfs_b->period = ns_to_ktime(period); | |
8046 | cfs_b->quota = quota; | |
58088ad0 | 8047 | |
a9cf55b2 | 8048 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
8049 | /* restart the period timer (if active) to handle new period expiry */ |
8050 | if (runtime_enabled && cfs_b->timer_active) { | |
8051 | /* force a reprogram */ | |
09dc4ab0 | 8052 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 8053 | } |
ab84d31e PT |
8054 | raw_spin_unlock_irq(&cfs_b->lock); |
8055 | ||
0e59bdae | 8056 | for_each_online_cpu(i) { |
ab84d31e | 8057 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8058 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8059 | |
8060 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8061 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8062 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8063 | |
029632fb | 8064 | if (cfs_rq->throttled) |
671fd9da | 8065 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8066 | raw_spin_unlock_irq(&rq->lock); |
8067 | } | |
1ee14e6c BS |
8068 | if (runtime_was_enabled && !runtime_enabled) |
8069 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8070 | out_unlock: |
8071 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8072 | put_online_cpus(); |
ab84d31e | 8073 | |
a790de99 | 8074 | return ret; |
ab84d31e PT |
8075 | } |
8076 | ||
8077 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8078 | { | |
8079 | u64 quota, period; | |
8080 | ||
029632fb | 8081 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8082 | if (cfs_quota_us < 0) |
8083 | quota = RUNTIME_INF; | |
8084 | else | |
8085 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8086 | ||
8087 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8088 | } | |
8089 | ||
8090 | long tg_get_cfs_quota(struct task_group *tg) | |
8091 | { | |
8092 | u64 quota_us; | |
8093 | ||
029632fb | 8094 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8095 | return -1; |
8096 | ||
029632fb | 8097 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8098 | do_div(quota_us, NSEC_PER_USEC); |
8099 | ||
8100 | return quota_us; | |
8101 | } | |
8102 | ||
8103 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8104 | { | |
8105 | u64 quota, period; | |
8106 | ||
8107 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8108 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8109 | |
ab84d31e PT |
8110 | return tg_set_cfs_bandwidth(tg, period, quota); |
8111 | } | |
8112 | ||
8113 | long tg_get_cfs_period(struct task_group *tg) | |
8114 | { | |
8115 | u64 cfs_period_us; | |
8116 | ||
029632fb | 8117 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8118 | do_div(cfs_period_us, NSEC_PER_USEC); |
8119 | ||
8120 | return cfs_period_us; | |
8121 | } | |
8122 | ||
182446d0 TH |
8123 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8124 | struct cftype *cft) | |
ab84d31e | 8125 | { |
182446d0 | 8126 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8127 | } |
8128 | ||
182446d0 TH |
8129 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8130 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8131 | { |
182446d0 | 8132 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8133 | } |
8134 | ||
182446d0 TH |
8135 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8136 | struct cftype *cft) | |
ab84d31e | 8137 | { |
182446d0 | 8138 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8139 | } |
8140 | ||
182446d0 TH |
8141 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8142 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8143 | { |
182446d0 | 8144 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8145 | } |
8146 | ||
a790de99 PT |
8147 | struct cfs_schedulable_data { |
8148 | struct task_group *tg; | |
8149 | u64 period, quota; | |
8150 | }; | |
8151 | ||
8152 | /* | |
8153 | * normalize group quota/period to be quota/max_period | |
8154 | * note: units are usecs | |
8155 | */ | |
8156 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8157 | struct cfs_schedulable_data *d) | |
8158 | { | |
8159 | u64 quota, period; | |
8160 | ||
8161 | if (tg == d->tg) { | |
8162 | period = d->period; | |
8163 | quota = d->quota; | |
8164 | } else { | |
8165 | period = tg_get_cfs_period(tg); | |
8166 | quota = tg_get_cfs_quota(tg); | |
8167 | } | |
8168 | ||
8169 | /* note: these should typically be equivalent */ | |
8170 | if (quota == RUNTIME_INF || quota == -1) | |
8171 | return RUNTIME_INF; | |
8172 | ||
8173 | return to_ratio(period, quota); | |
8174 | } | |
8175 | ||
8176 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8177 | { | |
8178 | struct cfs_schedulable_data *d = data; | |
029632fb | 8179 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8180 | s64 quota = 0, parent_quota = -1; |
8181 | ||
8182 | if (!tg->parent) { | |
8183 | quota = RUNTIME_INF; | |
8184 | } else { | |
029632fb | 8185 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8186 | |
8187 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8188 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8189 | |
8190 | /* | |
8191 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8192 | * limit is set | |
8193 | */ | |
8194 | if (quota == RUNTIME_INF) | |
8195 | quota = parent_quota; | |
8196 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8197 | return -EINVAL; | |
8198 | } | |
9c58c79a | 8199 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8200 | |
8201 | return 0; | |
8202 | } | |
8203 | ||
8204 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8205 | { | |
8277434e | 8206 | int ret; |
a790de99 PT |
8207 | struct cfs_schedulable_data data = { |
8208 | .tg = tg, | |
8209 | .period = period, | |
8210 | .quota = quota, | |
8211 | }; | |
8212 | ||
8213 | if (quota != RUNTIME_INF) { | |
8214 | do_div(data.period, NSEC_PER_USEC); | |
8215 | do_div(data.quota, NSEC_PER_USEC); | |
8216 | } | |
8217 | ||
8277434e PT |
8218 | rcu_read_lock(); |
8219 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8220 | rcu_read_unlock(); | |
8221 | ||
8222 | return ret; | |
a790de99 | 8223 | } |
e8da1b18 | 8224 | |
2da8ca82 | 8225 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8226 | { |
2da8ca82 | 8227 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8228 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8229 | |
44ffc75b TH |
8230 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8231 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8232 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8233 | |
8234 | return 0; | |
8235 | } | |
ab84d31e | 8236 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8237 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8238 | |
052f1dc7 | 8239 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8240 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8241 | struct cftype *cft, s64 val) | |
6f505b16 | 8242 | { |
182446d0 | 8243 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8244 | } |
8245 | ||
182446d0 TH |
8246 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8247 | struct cftype *cft) | |
6f505b16 | 8248 | { |
182446d0 | 8249 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8250 | } |
d0b27fa7 | 8251 | |
182446d0 TH |
8252 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8253 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8254 | { |
182446d0 | 8255 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8256 | } |
8257 | ||
182446d0 TH |
8258 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8259 | struct cftype *cft) | |
d0b27fa7 | 8260 | { |
182446d0 | 8261 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8262 | } |
6d6bc0ad | 8263 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8264 | |
fe5c7cc2 | 8265 | static struct cftype cpu_files[] = { |
052f1dc7 | 8266 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8267 | { |
8268 | .name = "shares", | |
f4c753b7 PM |
8269 | .read_u64 = cpu_shares_read_u64, |
8270 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8271 | }, |
052f1dc7 | 8272 | #endif |
ab84d31e PT |
8273 | #ifdef CONFIG_CFS_BANDWIDTH |
8274 | { | |
8275 | .name = "cfs_quota_us", | |
8276 | .read_s64 = cpu_cfs_quota_read_s64, | |
8277 | .write_s64 = cpu_cfs_quota_write_s64, | |
8278 | }, | |
8279 | { | |
8280 | .name = "cfs_period_us", | |
8281 | .read_u64 = cpu_cfs_period_read_u64, | |
8282 | .write_u64 = cpu_cfs_period_write_u64, | |
8283 | }, | |
e8da1b18 NR |
8284 | { |
8285 | .name = "stat", | |
2da8ca82 | 8286 | .seq_show = cpu_stats_show, |
e8da1b18 | 8287 | }, |
ab84d31e | 8288 | #endif |
052f1dc7 | 8289 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8290 | { |
9f0c1e56 | 8291 | .name = "rt_runtime_us", |
06ecb27c PM |
8292 | .read_s64 = cpu_rt_runtime_read, |
8293 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8294 | }, |
d0b27fa7 PZ |
8295 | { |
8296 | .name = "rt_period_us", | |
f4c753b7 PM |
8297 | .read_u64 = cpu_rt_period_read_uint, |
8298 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8299 | }, |
052f1dc7 | 8300 | #endif |
4baf6e33 | 8301 | { } /* terminate */ |
68318b8e SV |
8302 | }; |
8303 | ||
073219e9 | 8304 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8305 | .css_alloc = cpu_cgroup_css_alloc, |
8306 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8307 | .css_online = cpu_cgroup_css_online, |
8308 | .css_offline = cpu_cgroup_css_offline, | |
eeb61e53 | 8309 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8310 | .can_attach = cpu_cgroup_can_attach, |
8311 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8312 | .exit = cpu_cgroup_exit, |
5577964e | 8313 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8314 | .early_init = 1, |
8315 | }; | |
8316 | ||
052f1dc7 | 8317 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8318 | |
b637a328 PM |
8319 | void dump_cpu_task(int cpu) |
8320 | { | |
8321 | pr_info("Task dump for CPU %d:\n", cpu); | |
8322 | sched_show_task(cpu_curr(cpu)); | |
8323 | } |