]>
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 | ||
e1b77c92 | 29 | #include <linux/kasan.h> |
1da177e4 LT |
30 | #include <linux/mm.h> |
31 | #include <linux/module.h> | |
32 | #include <linux/nmi.h> | |
33 | #include <linux/init.h> | |
dff06c15 | 34 | #include <linux/uaccess.h> |
1da177e4 | 35 | #include <linux/highmem.h> |
f98db601 | 36 | #include <linux/mmu_context.h> |
1da177e4 | 37 | #include <linux/interrupt.h> |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 | 70 | #include <linux/ctype.h> |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
91d1aa43 | 74 | #include <linux/context_tracking.h> |
52f5684c | 75 | #include <linux/compiler.h> |
8e05e96a | 76 | #include <linux/frame.h> |
6075620b | 77 | #include <linux/prefetch.h> |
1da177e4 | 78 | |
96f951ed | 79 | #include <asm/switch_to.h> |
5517d86b | 80 | #include <asm/tlb.h> |
838225b4 | 81 | #include <asm/irq_regs.h> |
db7e527d | 82 | #include <asm/mutex.h> |
e6e6685a GC |
83 | #ifdef CONFIG_PARAVIRT |
84 | #include <asm/paravirt.h> | |
85 | #endif | |
1da177e4 | 86 | |
029632fb | 87 | #include "sched.h" |
ea138446 | 88 | #include "../workqueue_internal.h" |
29d5e047 | 89 | #include "../smpboot.h" |
6e0534f2 | 90 | |
a8d154b0 | 91 | #define CREATE_TRACE_POINTS |
ad8d75ff | 92 | #include <trace/events/sched.h> |
a8d154b0 | 93 | |
029632fb PZ |
94 | DEFINE_MUTEX(sched_domains_mutex); |
95 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 96 | |
fe44d621 | 97 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 98 | |
029632fb | 99 | void update_rq_clock(struct rq *rq) |
3e51f33f | 100 | { |
fe44d621 | 101 | s64 delta; |
305e6835 | 102 | |
9edfbfed PZ |
103 | lockdep_assert_held(&rq->lock); |
104 | ||
105 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 106 | return; |
aa483808 | 107 | |
fe44d621 | 108 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
109 | if (delta < 0) |
110 | return; | |
fe44d621 PZ |
111 | rq->clock += delta; |
112 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
113 | } |
114 | ||
bf5c91ba IM |
115 | /* |
116 | * Debugging: various feature bits | |
117 | */ | |
f00b45c1 | 118 | |
f00b45c1 PZ |
119 | #define SCHED_FEAT(name, enabled) \ |
120 | (1UL << __SCHED_FEAT_##name) * enabled | | |
121 | ||
bf5c91ba | 122 | const_debug unsigned int sysctl_sched_features = |
391e43da | 123 | #include "features.h" |
f00b45c1 PZ |
124 | 0; |
125 | ||
126 | #undef SCHED_FEAT | |
127 | ||
b82d9fdd PZ |
128 | /* |
129 | * Number of tasks to iterate in a single balance run. | |
130 | * Limited because this is done with IRQs disabled. | |
131 | */ | |
132 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
133 | ||
e9e9250b PZ |
134 | /* |
135 | * period over which we average the RT time consumption, measured | |
136 | * in ms. | |
137 | * | |
138 | * default: 1s | |
139 | */ | |
140 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
141 | ||
fa85ae24 | 142 | /* |
9f0c1e56 | 143 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
144 | * default: 1s |
145 | */ | |
9f0c1e56 | 146 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 147 | |
029632fb | 148 | __read_mostly int scheduler_running; |
6892b75e | 149 | |
9f0c1e56 PZ |
150 | /* |
151 | * part of the period that we allow rt tasks to run in us. | |
152 | * default: 0.95s | |
153 | */ | |
154 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 155 | |
3fa0818b RR |
156 | /* cpus with isolated domains */ |
157 | cpumask_var_t cpu_isolated_map; | |
158 | ||
1da177e4 | 159 | /* |
cc2a73b5 | 160 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 161 | */ |
a9957449 | 162 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
163 | __acquires(rq->lock) |
164 | { | |
70b97a7f | 165 | struct rq *rq; |
1da177e4 LT |
166 | |
167 | local_irq_disable(); | |
168 | rq = this_rq(); | |
05fa785c | 169 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
170 | |
171 | return rq; | |
172 | } | |
173 | ||
3e71a462 PZ |
174 | /* |
175 | * __task_rq_lock - lock the rq @p resides on. | |
176 | */ | |
eb580751 | 177 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
178 | __acquires(rq->lock) |
179 | { | |
180 | struct rq *rq; | |
181 | ||
182 | lockdep_assert_held(&p->pi_lock); | |
183 | ||
184 | for (;;) { | |
185 | rq = task_rq(p); | |
186 | raw_spin_lock(&rq->lock); | |
187 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
e7904a28 | 188 | rf->cookie = lockdep_pin_lock(&rq->lock); |
3e71a462 PZ |
189 | return rq; |
190 | } | |
191 | raw_spin_unlock(&rq->lock); | |
192 | ||
193 | while (unlikely(task_on_rq_migrating(p))) | |
194 | cpu_relax(); | |
195 | } | |
196 | } | |
197 | ||
198 | /* | |
199 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
200 | */ | |
eb580751 | 201 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
202 | __acquires(p->pi_lock) |
203 | __acquires(rq->lock) | |
204 | { | |
205 | struct rq *rq; | |
206 | ||
207 | for (;;) { | |
eb580751 | 208 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
209 | rq = task_rq(p); |
210 | raw_spin_lock(&rq->lock); | |
211 | /* | |
212 | * move_queued_task() task_rq_lock() | |
213 | * | |
214 | * ACQUIRE (rq->lock) | |
215 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
216 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
217 | * [S] ->cpu = new_cpu [L] task_rq() | |
218 | * [L] ->on_rq | |
219 | * RELEASE (rq->lock) | |
220 | * | |
221 | * If we observe the old cpu in task_rq_lock, the acquire of | |
222 | * the old rq->lock will fully serialize against the stores. | |
223 | * | |
224 | * If we observe the new cpu in task_rq_lock, the acquire will | |
225 | * pair with the WMB to ensure we must then also see migrating. | |
226 | */ | |
227 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
e7904a28 | 228 | rf->cookie = lockdep_pin_lock(&rq->lock); |
3e71a462 PZ |
229 | return rq; |
230 | } | |
231 | raw_spin_unlock(&rq->lock); | |
eb580751 | 232 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
233 | |
234 | while (unlikely(task_on_rq_migrating(p))) | |
235 | cpu_relax(); | |
236 | } | |
237 | } | |
238 | ||
8f4d37ec PZ |
239 | #ifdef CONFIG_SCHED_HRTICK |
240 | /* | |
241 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 242 | */ |
8f4d37ec | 243 | |
8f4d37ec PZ |
244 | static void hrtick_clear(struct rq *rq) |
245 | { | |
246 | if (hrtimer_active(&rq->hrtick_timer)) | |
247 | hrtimer_cancel(&rq->hrtick_timer); | |
248 | } | |
249 | ||
8f4d37ec PZ |
250 | /* |
251 | * High-resolution timer tick. | |
252 | * Runs from hardirq context with interrupts disabled. | |
253 | */ | |
254 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
255 | { | |
256 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
257 | ||
258 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
259 | ||
05fa785c | 260 | raw_spin_lock(&rq->lock); |
3e51f33f | 261 | update_rq_clock(rq); |
8f4d37ec | 262 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 263 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
264 | |
265 | return HRTIMER_NORESTART; | |
266 | } | |
267 | ||
95e904c7 | 268 | #ifdef CONFIG_SMP |
971ee28c | 269 | |
4961b6e1 | 270 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
271 | { |
272 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 273 | |
4961b6e1 | 274 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
275 | } |
276 | ||
31656519 PZ |
277 | /* |
278 | * called from hardirq (IPI) context | |
279 | */ | |
280 | static void __hrtick_start(void *arg) | |
b328ca18 | 281 | { |
31656519 | 282 | struct rq *rq = arg; |
b328ca18 | 283 | |
05fa785c | 284 | raw_spin_lock(&rq->lock); |
971ee28c | 285 | __hrtick_restart(rq); |
31656519 | 286 | rq->hrtick_csd_pending = 0; |
05fa785c | 287 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
288 | } |
289 | ||
31656519 PZ |
290 | /* |
291 | * Called to set the hrtick timer state. | |
292 | * | |
293 | * called with rq->lock held and irqs disabled | |
294 | */ | |
029632fb | 295 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 296 | { |
31656519 | 297 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 298 | ktime_t time; |
299 | s64 delta; | |
300 | ||
301 | /* | |
302 | * Don't schedule slices shorter than 10000ns, that just | |
303 | * doesn't make sense and can cause timer DoS. | |
304 | */ | |
305 | delta = max_t(s64, delay, 10000LL); | |
306 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 307 | |
cc584b21 | 308 | hrtimer_set_expires(timer, time); |
31656519 PZ |
309 | |
310 | if (rq == this_rq()) { | |
971ee28c | 311 | __hrtick_restart(rq); |
31656519 | 312 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 313 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
314 | rq->hrtick_csd_pending = 1; |
315 | } | |
b328ca18 PZ |
316 | } |
317 | ||
31656519 PZ |
318 | #else |
319 | /* | |
320 | * Called to set the hrtick timer state. | |
321 | * | |
322 | * called with rq->lock held and irqs disabled | |
323 | */ | |
029632fb | 324 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 325 | { |
86893335 WL |
326 | /* |
327 | * Don't schedule slices shorter than 10000ns, that just | |
328 | * doesn't make sense. Rely on vruntime for fairness. | |
329 | */ | |
330 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
331 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
332 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 333 | } |
31656519 | 334 | #endif /* CONFIG_SMP */ |
8f4d37ec | 335 | |
31656519 | 336 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 337 | { |
31656519 PZ |
338 | #ifdef CONFIG_SMP |
339 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 340 | |
31656519 PZ |
341 | rq->hrtick_csd.flags = 0; |
342 | rq->hrtick_csd.func = __hrtick_start; | |
343 | rq->hrtick_csd.info = rq; | |
344 | #endif | |
8f4d37ec | 345 | |
31656519 PZ |
346 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
347 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 348 | } |
006c75f1 | 349 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
350 | static inline void hrtick_clear(struct rq *rq) |
351 | { | |
352 | } | |
353 | ||
8f4d37ec PZ |
354 | static inline void init_rq_hrtick(struct rq *rq) |
355 | { | |
356 | } | |
006c75f1 | 357 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 358 | |
5529578a FW |
359 | /* |
360 | * cmpxchg based fetch_or, macro so it works for different integer types | |
361 | */ | |
362 | #define fetch_or(ptr, mask) \ | |
363 | ({ \ | |
364 | typeof(ptr) _ptr = (ptr); \ | |
365 | typeof(mask) _mask = (mask); \ | |
366 | typeof(*_ptr) _old, _val = *_ptr; \ | |
367 | \ | |
368 | for (;;) { \ | |
369 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
370 | if (_old == _val) \ | |
371 | break; \ | |
372 | _val = _old; \ | |
373 | } \ | |
374 | _old; \ | |
375 | }) | |
376 | ||
e3baac47 | 377 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
378 | /* |
379 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
380 | * this avoids any races wrt polling state changes and thereby avoids | |
381 | * spurious IPIs. | |
382 | */ | |
383 | static bool set_nr_and_not_polling(struct task_struct *p) | |
384 | { | |
385 | struct thread_info *ti = task_thread_info(p); | |
386 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
387 | } | |
e3baac47 PZ |
388 | |
389 | /* | |
390 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
391 | * | |
392 | * If this returns true, then the idle task promises to call | |
393 | * sched_ttwu_pending() and reschedule soon. | |
394 | */ | |
395 | static bool set_nr_if_polling(struct task_struct *p) | |
396 | { | |
397 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 398 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
399 | |
400 | for (;;) { | |
401 | if (!(val & _TIF_POLLING_NRFLAG)) | |
402 | return false; | |
403 | if (val & _TIF_NEED_RESCHED) | |
404 | return true; | |
405 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
406 | if (old == val) | |
407 | break; | |
408 | val = old; | |
409 | } | |
410 | return true; | |
411 | } | |
412 | ||
fd99f91a PZ |
413 | #else |
414 | static bool set_nr_and_not_polling(struct task_struct *p) | |
415 | { | |
416 | set_tsk_need_resched(p); | |
417 | return true; | |
418 | } | |
e3baac47 PZ |
419 | |
420 | #ifdef CONFIG_SMP | |
421 | static bool set_nr_if_polling(struct task_struct *p) | |
422 | { | |
423 | return false; | |
424 | } | |
425 | #endif | |
fd99f91a PZ |
426 | #endif |
427 | ||
76751049 PZ |
428 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
429 | { | |
430 | struct wake_q_node *node = &task->wake_q; | |
431 | ||
432 | /* | |
433 | * Atomically grab the task, if ->wake_q is !nil already it means | |
434 | * its already queued (either by us or someone else) and will get the | |
435 | * wakeup due to that. | |
436 | * | |
437 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 438 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
439 | */ |
440 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
441 | return; | |
442 | ||
443 | get_task_struct(task); | |
444 | ||
445 | /* | |
446 | * The head is context local, there can be no concurrency. | |
447 | */ | |
448 | *head->lastp = node; | |
449 | head->lastp = &node->next; | |
450 | } | |
451 | ||
452 | void wake_up_q(struct wake_q_head *head) | |
453 | { | |
454 | struct wake_q_node *node = head->first; | |
455 | ||
456 | while (node != WAKE_Q_TAIL) { | |
457 | struct task_struct *task; | |
458 | ||
459 | task = container_of(node, struct task_struct, wake_q); | |
460 | BUG_ON(!task); | |
461 | /* task can safely be re-inserted now */ | |
462 | node = node->next; | |
463 | task->wake_q.next = NULL; | |
464 | ||
465 | /* | |
466 | * wake_up_process() implies a wmb() to pair with the queueing | |
467 | * in wake_q_add() so as not to miss wakeups. | |
468 | */ | |
469 | wake_up_process(task); | |
470 | put_task_struct(task); | |
471 | } | |
472 | } | |
473 | ||
c24d20db | 474 | /* |
8875125e | 475 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
476 | * |
477 | * On UP this means the setting of the need_resched flag, on SMP it | |
478 | * might also involve a cross-CPU call to trigger the scheduler on | |
479 | * the target CPU. | |
480 | */ | |
8875125e | 481 | void resched_curr(struct rq *rq) |
c24d20db | 482 | { |
8875125e | 483 | struct task_struct *curr = rq->curr; |
c24d20db IM |
484 | int cpu; |
485 | ||
8875125e | 486 | lockdep_assert_held(&rq->lock); |
c24d20db | 487 | |
8875125e | 488 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
489 | return; |
490 | ||
8875125e | 491 | cpu = cpu_of(rq); |
fd99f91a | 492 | |
f27dde8d | 493 | if (cpu == smp_processor_id()) { |
8875125e | 494 | set_tsk_need_resched(curr); |
f27dde8d | 495 | set_preempt_need_resched(); |
c24d20db | 496 | return; |
f27dde8d | 497 | } |
c24d20db | 498 | |
8875125e | 499 | if (set_nr_and_not_polling(curr)) |
c24d20db | 500 | smp_send_reschedule(cpu); |
dfc68f29 AL |
501 | else |
502 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
503 | } |
504 | ||
029632fb | 505 | void resched_cpu(int cpu) |
c24d20db IM |
506 | { |
507 | struct rq *rq = cpu_rq(cpu); | |
508 | unsigned long flags; | |
509 | ||
05fa785c | 510 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 511 | return; |
8875125e | 512 | resched_curr(rq); |
05fa785c | 513 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 514 | } |
06d8308c | 515 | |
b021fe3e | 516 | #ifdef CONFIG_SMP |
3451d024 | 517 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
518 | /* |
519 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
520 | * from an idle cpu. This is good for power-savings. | |
521 | * | |
522 | * We don't do similar optimization for completely idle system, as | |
523 | * selecting an idle cpu will add more delays to the timers than intended | |
524 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
525 | */ | |
bc7a34b8 | 526 | int get_nohz_timer_target(void) |
83cd4fe2 | 527 | { |
bc7a34b8 | 528 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
529 | struct sched_domain *sd; |
530 | ||
9642d18e | 531 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
532 | return cpu; |
533 | ||
057f3fad | 534 | rcu_read_lock(); |
83cd4fe2 | 535 | for_each_domain(cpu, sd) { |
057f3fad | 536 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
537 | if (cpu == i) |
538 | continue; | |
539 | ||
540 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
541 | cpu = i; |
542 | goto unlock; | |
543 | } | |
544 | } | |
83cd4fe2 | 545 | } |
9642d18e VH |
546 | |
547 | if (!is_housekeeping_cpu(cpu)) | |
548 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
549 | unlock: |
550 | rcu_read_unlock(); | |
83cd4fe2 VP |
551 | return cpu; |
552 | } | |
06d8308c TG |
553 | /* |
554 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
555 | * idle CPU then this timer might expire before the next timer event | |
556 | * which is scheduled to wake up that CPU. In case of a completely | |
557 | * idle system the next event might even be infinite time into the | |
558 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
559 | * leaves the inner idle loop so the newly added timer is taken into | |
560 | * account when the CPU goes back to idle and evaluates the timer | |
561 | * wheel for the next timer event. | |
562 | */ | |
1c20091e | 563 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
564 | { |
565 | struct rq *rq = cpu_rq(cpu); | |
566 | ||
567 | if (cpu == smp_processor_id()) | |
568 | return; | |
569 | ||
67b9ca70 | 570 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 571 | smp_send_reschedule(cpu); |
dfc68f29 AL |
572 | else |
573 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
574 | } |
575 | ||
c5bfece2 | 576 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 577 | { |
53c5fa16 FW |
578 | /* |
579 | * We just need the target to call irq_exit() and re-evaluate | |
580 | * the next tick. The nohz full kick at least implies that. | |
581 | * If needed we can still optimize that later with an | |
582 | * empty IRQ. | |
583 | */ | |
c5bfece2 | 584 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
585 | if (cpu != smp_processor_id() || |
586 | tick_nohz_tick_stopped()) | |
53c5fa16 | 587 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
588 | return true; |
589 | } | |
590 | ||
591 | return false; | |
592 | } | |
593 | ||
594 | void wake_up_nohz_cpu(int cpu) | |
595 | { | |
c5bfece2 | 596 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
597 | wake_up_idle_cpu(cpu); |
598 | } | |
599 | ||
ca38062e | 600 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 601 | { |
1c792db7 | 602 | int cpu = smp_processor_id(); |
873b4c65 VG |
603 | |
604 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
605 | return false; | |
606 | ||
607 | if (idle_cpu(cpu) && !need_resched()) | |
608 | return true; | |
609 | ||
610 | /* | |
611 | * We can't run Idle Load Balance on this CPU for this time so we | |
612 | * cancel it and clear NOHZ_BALANCE_KICK | |
613 | */ | |
614 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
615 | return false; | |
45bf76df IM |
616 | } |
617 | ||
3451d024 | 618 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 619 | |
ca38062e | 620 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 621 | { |
ca38062e | 622 | return false; |
2069dd75 PZ |
623 | } |
624 | ||
3451d024 | 625 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 626 | |
ce831b38 | 627 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 628 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 629 | { |
76d92ac3 FW |
630 | int fifo_nr_running; |
631 | ||
632 | /* Deadline tasks, even if single, need the tick */ | |
633 | if (rq->dl.dl_nr_running) | |
634 | return false; | |
635 | ||
1e78cdbd | 636 | /* |
2548d546 PZ |
637 | * If there are more than one RR tasks, we need the tick to effect the |
638 | * actual RR behaviour. | |
1e78cdbd | 639 | */ |
76d92ac3 FW |
640 | if (rq->rt.rr_nr_running) { |
641 | if (rq->rt.rr_nr_running == 1) | |
642 | return true; | |
643 | else | |
644 | return false; | |
1e78cdbd RR |
645 | } |
646 | ||
2548d546 PZ |
647 | /* |
648 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
649 | * forced preemption between FIFO tasks. | |
650 | */ | |
651 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
652 | if (fifo_nr_running) | |
653 | return true; | |
654 | ||
655 | /* | |
656 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
657 | * if there's more than one we need the tick for involuntary | |
658 | * preemption. | |
659 | */ | |
660 | if (rq->nr_running > 1) | |
541b8264 | 661 | return false; |
ce831b38 | 662 | |
541b8264 | 663 | return true; |
ce831b38 FW |
664 | } |
665 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 666 | |
029632fb | 667 | void sched_avg_update(struct rq *rq) |
18d95a28 | 668 | { |
e9e9250b PZ |
669 | s64 period = sched_avg_period(); |
670 | ||
78becc27 | 671 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
672 | /* |
673 | * Inline assembly required to prevent the compiler | |
674 | * optimising this loop into a divmod call. | |
675 | * See __iter_div_u64_rem() for another example of this. | |
676 | */ | |
677 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
678 | rq->age_stamp += period; |
679 | rq->rt_avg /= 2; | |
680 | } | |
18d95a28 PZ |
681 | } |
682 | ||
6d6bc0ad | 683 | #endif /* CONFIG_SMP */ |
18d95a28 | 684 | |
a790de99 PT |
685 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
686 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 687 | /* |
8277434e PT |
688 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
689 | * node and @up when leaving it for the final time. | |
690 | * | |
691 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 692 | */ |
029632fb | 693 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 694 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
695 | { |
696 | struct task_group *parent, *child; | |
eb755805 | 697 | int ret; |
c09595f6 | 698 | |
8277434e PT |
699 | parent = from; |
700 | ||
c09595f6 | 701 | down: |
eb755805 PZ |
702 | ret = (*down)(parent, data); |
703 | if (ret) | |
8277434e | 704 | goto out; |
c09595f6 PZ |
705 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
706 | parent = child; | |
707 | goto down; | |
708 | ||
709 | up: | |
710 | continue; | |
711 | } | |
eb755805 | 712 | ret = (*up)(parent, data); |
8277434e PT |
713 | if (ret || parent == from) |
714 | goto out; | |
c09595f6 PZ |
715 | |
716 | child = parent; | |
717 | parent = parent->parent; | |
718 | if (parent) | |
719 | goto up; | |
8277434e | 720 | out: |
eb755805 | 721 | return ret; |
c09595f6 PZ |
722 | } |
723 | ||
029632fb | 724 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 725 | { |
e2b245f8 | 726 | return 0; |
eb755805 | 727 | } |
18d95a28 PZ |
728 | #endif |
729 | ||
45bf76df IM |
730 | static void set_load_weight(struct task_struct *p) |
731 | { | |
f05998d4 NR |
732 | int prio = p->static_prio - MAX_RT_PRIO; |
733 | struct load_weight *load = &p->se.load; | |
734 | ||
dd41f596 IM |
735 | /* |
736 | * SCHED_IDLE tasks get minimal weight: | |
737 | */ | |
20f9cd2a | 738 | if (idle_policy(p->policy)) { |
c8b28116 | 739 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 740 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
741 | return; |
742 | } | |
71f8bd46 | 743 | |
ed82b8a1 AK |
744 | load->weight = scale_load(sched_prio_to_weight[prio]); |
745 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
746 | } |
747 | ||
1de64443 | 748 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 749 | { |
a64692a3 | 750 | update_rq_clock(rq); |
1de64443 PZ |
751 | if (!(flags & ENQUEUE_RESTORE)) |
752 | sched_info_queued(rq, p); | |
371fd7e7 | 753 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
754 | } |
755 | ||
1de64443 | 756 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 757 | { |
a64692a3 | 758 | update_rq_clock(rq); |
1de64443 PZ |
759 | if (!(flags & DEQUEUE_SAVE)) |
760 | sched_info_dequeued(rq, p); | |
371fd7e7 | 761 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
762 | } |
763 | ||
029632fb | 764 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
765 | { |
766 | if (task_contributes_to_load(p)) | |
767 | rq->nr_uninterruptible--; | |
768 | ||
371fd7e7 | 769 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
770 | } |
771 | ||
029632fb | 772 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
773 | { |
774 | if (task_contributes_to_load(p)) | |
775 | rq->nr_uninterruptible++; | |
776 | ||
371fd7e7 | 777 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
778 | } |
779 | ||
fe44d621 | 780 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 781 | { |
095c0aa8 GC |
782 | /* |
783 | * In theory, the compile should just see 0 here, and optimize out the call | |
784 | * to sched_rt_avg_update. But I don't trust it... | |
785 | */ | |
786 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
787 | s64 steal = 0, irq_delta = 0; | |
788 | #endif | |
789 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 790 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
791 | |
792 | /* | |
793 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
794 | * this case when a previous update_rq_clock() happened inside a | |
795 | * {soft,}irq region. | |
796 | * | |
797 | * When this happens, we stop ->clock_task and only update the | |
798 | * prev_irq_time stamp to account for the part that fit, so that a next | |
799 | * update will consume the rest. This ensures ->clock_task is | |
800 | * monotonic. | |
801 | * | |
802 | * It does however cause some slight miss-attribution of {soft,}irq | |
803 | * time, a more accurate solution would be to update the irq_time using | |
804 | * the current rq->clock timestamp, except that would require using | |
805 | * atomic ops. | |
806 | */ | |
807 | if (irq_delta > delta) | |
808 | irq_delta = delta; | |
809 | ||
810 | rq->prev_irq_time += irq_delta; | |
811 | delta -= irq_delta; | |
095c0aa8 GC |
812 | #endif |
813 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 814 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
815 | steal = paravirt_steal_clock(cpu_of(rq)); |
816 | steal -= rq->prev_steal_time_rq; | |
817 | ||
818 | if (unlikely(steal > delta)) | |
819 | steal = delta; | |
820 | ||
095c0aa8 | 821 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
822 | delta -= steal; |
823 | } | |
824 | #endif | |
825 | ||
fe44d621 PZ |
826 | rq->clock_task += delta; |
827 | ||
095c0aa8 | 828 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 829 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
830 | sched_rt_avg_update(rq, irq_delta + steal); |
831 | #endif | |
aa483808 VP |
832 | } |
833 | ||
34f971f6 PZ |
834 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
835 | { | |
836 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
837 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
838 | ||
839 | if (stop) { | |
840 | /* | |
841 | * Make it appear like a SCHED_FIFO task, its something | |
842 | * userspace knows about and won't get confused about. | |
843 | * | |
844 | * Also, it will make PI more or less work without too | |
845 | * much confusion -- but then, stop work should not | |
846 | * rely on PI working anyway. | |
847 | */ | |
848 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
849 | ||
850 | stop->sched_class = &stop_sched_class; | |
851 | } | |
852 | ||
853 | cpu_rq(cpu)->stop = stop; | |
854 | ||
855 | if (old_stop) { | |
856 | /* | |
857 | * Reset it back to a normal scheduling class so that | |
858 | * it can die in pieces. | |
859 | */ | |
860 | old_stop->sched_class = &rt_sched_class; | |
861 | } | |
862 | } | |
863 | ||
14531189 | 864 | /* |
dd41f596 | 865 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 866 | */ |
14531189 IM |
867 | static inline int __normal_prio(struct task_struct *p) |
868 | { | |
dd41f596 | 869 | return p->static_prio; |
14531189 IM |
870 | } |
871 | ||
b29739f9 IM |
872 | /* |
873 | * Calculate the expected normal priority: i.e. priority | |
874 | * without taking RT-inheritance into account. Might be | |
875 | * boosted by interactivity modifiers. Changes upon fork, | |
876 | * setprio syscalls, and whenever the interactivity | |
877 | * estimator recalculates. | |
878 | */ | |
36c8b586 | 879 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
880 | { |
881 | int prio; | |
882 | ||
aab03e05 DF |
883 | if (task_has_dl_policy(p)) |
884 | prio = MAX_DL_PRIO-1; | |
885 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
886 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
887 | else | |
888 | prio = __normal_prio(p); | |
889 | return prio; | |
890 | } | |
891 | ||
892 | /* | |
893 | * Calculate the current priority, i.e. the priority | |
894 | * taken into account by the scheduler. This value might | |
895 | * be boosted by RT tasks, or might be boosted by | |
896 | * interactivity modifiers. Will be RT if the task got | |
897 | * RT-boosted. If not then it returns p->normal_prio. | |
898 | */ | |
36c8b586 | 899 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
900 | { |
901 | p->normal_prio = normal_prio(p); | |
902 | /* | |
903 | * If we are RT tasks or we were boosted to RT priority, | |
904 | * keep the priority unchanged. Otherwise, update priority | |
905 | * to the normal priority: | |
906 | */ | |
907 | if (!rt_prio(p->prio)) | |
908 | return p->normal_prio; | |
909 | return p->prio; | |
910 | } | |
911 | ||
1da177e4 LT |
912 | /** |
913 | * task_curr - is this task currently executing on a CPU? | |
914 | * @p: the task in question. | |
e69f6186 YB |
915 | * |
916 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 917 | */ |
36c8b586 | 918 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
919 | { |
920 | return cpu_curr(task_cpu(p)) == p; | |
921 | } | |
922 | ||
67dfa1b7 | 923 | /* |
4c9a4bc8 PZ |
924 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
925 | * use the balance_callback list if you want balancing. | |
926 | * | |
927 | * this means any call to check_class_changed() must be followed by a call to | |
928 | * balance_callback(). | |
67dfa1b7 | 929 | */ |
cb469845 SR |
930 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
931 | const struct sched_class *prev_class, | |
da7a735e | 932 | int oldprio) |
cb469845 SR |
933 | { |
934 | if (prev_class != p->sched_class) { | |
935 | if (prev_class->switched_from) | |
da7a735e | 936 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 937 | |
da7a735e | 938 | p->sched_class->switched_to(rq, p); |
2d3d891d | 939 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 940 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
941 | } |
942 | ||
029632fb | 943 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
944 | { |
945 | const struct sched_class *class; | |
946 | ||
947 | if (p->sched_class == rq->curr->sched_class) { | |
948 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
949 | } else { | |
950 | for_each_class(class) { | |
951 | if (class == rq->curr->sched_class) | |
952 | break; | |
953 | if (class == p->sched_class) { | |
8875125e | 954 | resched_curr(rq); |
1e5a7405 PZ |
955 | break; |
956 | } | |
957 | } | |
958 | } | |
959 | ||
960 | /* | |
961 | * A queue event has occurred, and we're going to schedule. In | |
962 | * this case, we can save a useless back to back clock update. | |
963 | */ | |
da0c1e65 | 964 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 965 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
966 | } |
967 | ||
1da177e4 | 968 | #ifdef CONFIG_SMP |
5cc389bc PZ |
969 | /* |
970 | * This is how migration works: | |
971 | * | |
972 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
973 | * stop_one_cpu(). | |
974 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
975 | * off the CPU) | |
976 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
977 | * 4) if it's in the wrong runqueue then the migration thread removes | |
978 | * it and puts it into the right queue. | |
979 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
980 | * is done. | |
981 | */ | |
982 | ||
983 | /* | |
984 | * move_queued_task - move a queued task to new rq. | |
985 | * | |
986 | * Returns (locked) new rq. Old rq's lock is released. | |
987 | */ | |
5e16bbc2 | 988 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 989 | { |
5cc389bc PZ |
990 | lockdep_assert_held(&rq->lock); |
991 | ||
5cc389bc | 992 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 993 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
994 | set_task_cpu(p, new_cpu); |
995 | raw_spin_unlock(&rq->lock); | |
996 | ||
997 | rq = cpu_rq(new_cpu); | |
998 | ||
999 | raw_spin_lock(&rq->lock); | |
1000 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 1001 | enqueue_task(rq, p, 0); |
3ea94de1 | 1002 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
1003 | check_preempt_curr(rq, p, 0); |
1004 | ||
1005 | return rq; | |
1006 | } | |
1007 | ||
1008 | struct migration_arg { | |
1009 | struct task_struct *task; | |
1010 | int dest_cpu; | |
1011 | }; | |
1012 | ||
1013 | /* | |
1014 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1015 | * this because either it can't run here any more (set_cpus_allowed() | |
1016 | * away from this CPU, or CPU going down), or because we're | |
1017 | * attempting to rebalance this task on exec (sched_exec). | |
1018 | * | |
1019 | * So we race with normal scheduler movements, but that's OK, as long | |
1020 | * as the task is no longer on this CPU. | |
5cc389bc | 1021 | */ |
5e16bbc2 | 1022 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1023 | { |
5cc389bc | 1024 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1025 | return rq; |
5cc389bc PZ |
1026 | |
1027 | /* Affinity changed (again). */ | |
1028 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1029 | return rq; |
5cc389bc | 1030 | |
5e16bbc2 PZ |
1031 | rq = move_queued_task(rq, p, dest_cpu); |
1032 | ||
1033 | return rq; | |
5cc389bc PZ |
1034 | } |
1035 | ||
1036 | /* | |
1037 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1038 | * and performs thread migration by bumping thread off CPU then | |
1039 | * 'pushing' onto another runqueue. | |
1040 | */ | |
1041 | static int migration_cpu_stop(void *data) | |
1042 | { | |
1043 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1044 | struct task_struct *p = arg->task; |
1045 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1046 | |
1047 | /* | |
1048 | * The original target cpu might have gone down and we might | |
1049 | * be on another cpu but it doesn't matter. | |
1050 | */ | |
1051 | local_irq_disable(); | |
1052 | /* | |
1053 | * We need to explicitly wake pending tasks before running | |
1054 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1055 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1056 | */ | |
1057 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1058 | |
1059 | raw_spin_lock(&p->pi_lock); | |
1060 | raw_spin_lock(&rq->lock); | |
1061 | /* | |
1062 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1063 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1064 | * we're holding p->pi_lock. | |
1065 | */ | |
1066 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1067 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1068 | raw_spin_unlock(&rq->lock); | |
1069 | raw_spin_unlock(&p->pi_lock); | |
1070 | ||
5cc389bc PZ |
1071 | local_irq_enable(); |
1072 | return 0; | |
1073 | } | |
1074 | ||
c5b28038 PZ |
1075 | /* |
1076 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1077 | * actually call this function. | |
1078 | */ | |
1079 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1080 | { |
5cc389bc PZ |
1081 | cpumask_copy(&p->cpus_allowed, new_mask); |
1082 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1083 | } | |
1084 | ||
c5b28038 PZ |
1085 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1086 | { | |
6c37067e PZ |
1087 | struct rq *rq = task_rq(p); |
1088 | bool queued, running; | |
1089 | ||
c5b28038 | 1090 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1091 | |
1092 | queued = task_on_rq_queued(p); | |
1093 | running = task_current(rq, p); | |
1094 | ||
1095 | if (queued) { | |
1096 | /* | |
1097 | * Because __kthread_bind() calls this on blocked tasks without | |
1098 | * holding rq->lock. | |
1099 | */ | |
1100 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1101 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1102 | } |
1103 | if (running) | |
1104 | put_prev_task(rq, p); | |
1105 | ||
c5b28038 | 1106 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e PZ |
1107 | |
1108 | if (running) | |
1109 | p->sched_class->set_curr_task(rq); | |
1110 | if (queued) | |
1de64443 | 1111 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
c5b28038 PZ |
1112 | } |
1113 | ||
5cc389bc PZ |
1114 | /* |
1115 | * Change a given task's CPU affinity. Migrate the thread to a | |
1116 | * proper CPU and schedule it away if the CPU it's executing on | |
1117 | * is removed from the allowed bitmask. | |
1118 | * | |
1119 | * NOTE: the caller must have a valid reference to the task, the | |
1120 | * task must not exit() & deallocate itself prematurely. The | |
1121 | * call is not atomic; no spinlocks may be held. | |
1122 | */ | |
25834c73 PZ |
1123 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1124 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1125 | { |
e9d867a6 | 1126 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1127 | unsigned int dest_cpu; |
eb580751 PZ |
1128 | struct rq_flags rf; |
1129 | struct rq *rq; | |
5cc389bc PZ |
1130 | int ret = 0; |
1131 | ||
eb580751 | 1132 | rq = task_rq_lock(p, &rf); |
5cc389bc | 1133 | |
e9d867a6 PZI |
1134 | if (p->flags & PF_KTHREAD) { |
1135 | /* | |
1136 | * Kernel threads are allowed on online && !active CPUs | |
1137 | */ | |
1138 | cpu_valid_mask = cpu_online_mask; | |
1139 | } | |
1140 | ||
25834c73 PZ |
1141 | /* |
1142 | * Must re-check here, to close a race against __kthread_bind(), | |
1143 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1144 | */ | |
1145 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1146 | ret = -EINVAL; | |
1147 | goto out; | |
1148 | } | |
1149 | ||
5cc389bc PZ |
1150 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1151 | goto out; | |
1152 | ||
e9d867a6 | 1153 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1154 | ret = -EINVAL; |
1155 | goto out; | |
1156 | } | |
1157 | ||
1158 | do_set_cpus_allowed(p, new_mask); | |
1159 | ||
e9d867a6 PZI |
1160 | if (p->flags & PF_KTHREAD) { |
1161 | /* | |
1162 | * For kernel threads that do indeed end up on online && | |
1163 | * !active we want to ensure they are strict per-cpu threads. | |
1164 | */ | |
1165 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1166 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1167 | p->nr_cpus_allowed != 1); | |
1168 | } | |
1169 | ||
5cc389bc PZ |
1170 | /* Can the task run on the task's current CPU? If so, we're done */ |
1171 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1172 | goto out; | |
1173 | ||
e9d867a6 | 1174 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1175 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1176 | struct migration_arg arg = { p, dest_cpu }; | |
1177 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1178 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1179 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1180 | tlb_migrate_finish(p->mm); | |
1181 | return 0; | |
cbce1a68 PZ |
1182 | } else if (task_on_rq_queued(p)) { |
1183 | /* | |
1184 | * OK, since we're going to drop the lock immediately | |
1185 | * afterwards anyway. | |
1186 | */ | |
e7904a28 | 1187 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
5e16bbc2 | 1188 | rq = move_queued_task(rq, p, dest_cpu); |
e7904a28 | 1189 | lockdep_repin_lock(&rq->lock, rf.cookie); |
cbce1a68 | 1190 | } |
5cc389bc | 1191 | out: |
eb580751 | 1192 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1193 | |
1194 | return ret; | |
1195 | } | |
25834c73 PZ |
1196 | |
1197 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1198 | { | |
1199 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1200 | } | |
5cc389bc PZ |
1201 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1202 | ||
dd41f596 | 1203 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1204 | { |
e2912009 PZ |
1205 | #ifdef CONFIG_SCHED_DEBUG |
1206 | /* | |
1207 | * We should never call set_task_cpu() on a blocked task, | |
1208 | * ttwu() will sort out the placement. | |
1209 | */ | |
077614ee | 1210 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1211 | !p->on_rq); |
0122ec5b | 1212 | |
3ea94de1 JP |
1213 | /* |
1214 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1215 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1216 | * time relying on p->on_rq. | |
1217 | */ | |
1218 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1219 | p->sched_class == &fair_sched_class && | |
1220 | (p->on_rq && !task_on_rq_migrating(p))); | |
1221 | ||
0122ec5b | 1222 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1223 | /* |
1224 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1225 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1226 | * | |
1227 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1228 | * see task_group(). |
6c6c54e1 PZ |
1229 | * |
1230 | * Furthermore, all task_rq users should acquire both locks, see | |
1231 | * task_rq_lock(). | |
1232 | */ | |
0122ec5b PZ |
1233 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1234 | lockdep_is_held(&task_rq(p)->lock))); | |
1235 | #endif | |
e2912009 PZ |
1236 | #endif |
1237 | ||
de1d7286 | 1238 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1239 | |
0c69774e | 1240 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1241 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1242 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1243 | p->se.nr_migrations++; |
ff303e66 | 1244 | perf_event_task_migrate(p); |
0c69774e | 1245 | } |
dd41f596 IM |
1246 | |
1247 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1248 | } |
1249 | ||
ac66f547 PZ |
1250 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1251 | { | |
da0c1e65 | 1252 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1253 | struct rq *src_rq, *dst_rq; |
1254 | ||
1255 | src_rq = task_rq(p); | |
1256 | dst_rq = cpu_rq(cpu); | |
1257 | ||
3ea94de1 | 1258 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1259 | deactivate_task(src_rq, p, 0); |
1260 | set_task_cpu(p, cpu); | |
1261 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1262 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1263 | check_preempt_curr(dst_rq, p, 0); |
1264 | } else { | |
1265 | /* | |
1266 | * Task isn't running anymore; make it appear like we migrated | |
1267 | * it before it went to sleep. This means on wakeup we make the | |
a1fd4656 | 1268 | * previous cpu our target instead of where it really is. |
ac66f547 PZ |
1269 | */ |
1270 | p->wake_cpu = cpu; | |
1271 | } | |
1272 | } | |
1273 | ||
1274 | struct migration_swap_arg { | |
1275 | struct task_struct *src_task, *dst_task; | |
1276 | int src_cpu, dst_cpu; | |
1277 | }; | |
1278 | ||
1279 | static int migrate_swap_stop(void *data) | |
1280 | { | |
1281 | struct migration_swap_arg *arg = data; | |
1282 | struct rq *src_rq, *dst_rq; | |
1283 | int ret = -EAGAIN; | |
1284 | ||
62694cd5 PZ |
1285 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1286 | return -EAGAIN; | |
1287 | ||
ac66f547 PZ |
1288 | src_rq = cpu_rq(arg->src_cpu); |
1289 | dst_rq = cpu_rq(arg->dst_cpu); | |
1290 | ||
74602315 PZ |
1291 | double_raw_lock(&arg->src_task->pi_lock, |
1292 | &arg->dst_task->pi_lock); | |
ac66f547 | 1293 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1294 | |
ac66f547 PZ |
1295 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1296 | goto unlock; | |
1297 | ||
1298 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1299 | goto unlock; | |
1300 | ||
1301 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1302 | goto unlock; | |
1303 | ||
1304 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1305 | goto unlock; | |
1306 | ||
1307 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1308 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1309 | ||
1310 | ret = 0; | |
1311 | ||
1312 | unlock: | |
1313 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1314 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1315 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1316 | |
1317 | return ret; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Cross migrate two tasks | |
1322 | */ | |
1323 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1324 | { | |
1325 | struct migration_swap_arg arg; | |
1326 | int ret = -EINVAL; | |
1327 | ||
ac66f547 PZ |
1328 | arg = (struct migration_swap_arg){ |
1329 | .src_task = cur, | |
1330 | .src_cpu = task_cpu(cur), | |
1331 | .dst_task = p, | |
1332 | .dst_cpu = task_cpu(p), | |
1333 | }; | |
1334 | ||
1335 | if (arg.src_cpu == arg.dst_cpu) | |
1336 | goto out; | |
1337 | ||
6acce3ef PZ |
1338 | /* |
1339 | * These three tests are all lockless; this is OK since all of them | |
1340 | * will be re-checked with proper locks held further down the line. | |
1341 | */ | |
ac66f547 PZ |
1342 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1343 | goto out; | |
1344 | ||
1345 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1346 | goto out; | |
1347 | ||
1348 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1349 | goto out; | |
1350 | ||
286549dc | 1351 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1352 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1353 | ||
1354 | out: | |
ac66f547 PZ |
1355 | return ret; |
1356 | } | |
1357 | ||
1da177e4 LT |
1358 | /* |
1359 | * wait_task_inactive - wait for a thread to unschedule. | |
1360 | * | |
85ba2d86 RM |
1361 | * If @match_state is nonzero, it's the @p->state value just checked and |
1362 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1363 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1364 | * we return a positive number (its total switch count). If a second call | |
1365 | * a short while later returns the same number, the caller can be sure that | |
1366 | * @p has remained unscheduled the whole time. | |
1367 | * | |
1da177e4 LT |
1368 | * The caller must ensure that the task *will* unschedule sometime soon, |
1369 | * else this function might spin for a *long* time. This function can't | |
1370 | * be called with interrupts off, or it may introduce deadlock with | |
1371 | * smp_call_function() if an IPI is sent by the same process we are | |
1372 | * waiting to become inactive. | |
1373 | */ | |
85ba2d86 | 1374 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1375 | { |
da0c1e65 | 1376 | int running, queued; |
eb580751 | 1377 | struct rq_flags rf; |
85ba2d86 | 1378 | unsigned long ncsw; |
70b97a7f | 1379 | struct rq *rq; |
1da177e4 | 1380 | |
3a5c359a AK |
1381 | for (;;) { |
1382 | /* | |
1383 | * We do the initial early heuristics without holding | |
1384 | * any task-queue locks at all. We'll only try to get | |
1385 | * the runqueue lock when things look like they will | |
1386 | * work out! | |
1387 | */ | |
1388 | rq = task_rq(p); | |
fa490cfd | 1389 | |
3a5c359a AK |
1390 | /* |
1391 | * If the task is actively running on another CPU | |
1392 | * still, just relax and busy-wait without holding | |
1393 | * any locks. | |
1394 | * | |
1395 | * NOTE! Since we don't hold any locks, it's not | |
1396 | * even sure that "rq" stays as the right runqueue! | |
1397 | * But we don't care, since "task_running()" will | |
1398 | * return false if the runqueue has changed and p | |
1399 | * is actually now running somewhere else! | |
1400 | */ | |
85ba2d86 RM |
1401 | while (task_running(rq, p)) { |
1402 | if (match_state && unlikely(p->state != match_state)) | |
1403 | return 0; | |
3a5c359a | 1404 | cpu_relax(); |
85ba2d86 | 1405 | } |
fa490cfd | 1406 | |
3a5c359a AK |
1407 | /* |
1408 | * Ok, time to look more closely! We need the rq | |
1409 | * lock now, to be *sure*. If we're wrong, we'll | |
1410 | * just go back and repeat. | |
1411 | */ | |
eb580751 | 1412 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1413 | trace_sched_wait_task(p); |
3a5c359a | 1414 | running = task_running(rq, p); |
da0c1e65 | 1415 | queued = task_on_rq_queued(p); |
85ba2d86 | 1416 | ncsw = 0; |
f31e11d8 | 1417 | if (!match_state || p->state == match_state) |
93dcf55f | 1418 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1419 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1420 | |
85ba2d86 RM |
1421 | /* |
1422 | * If it changed from the expected state, bail out now. | |
1423 | */ | |
1424 | if (unlikely(!ncsw)) | |
1425 | break; | |
1426 | ||
3a5c359a AK |
1427 | /* |
1428 | * Was it really running after all now that we | |
1429 | * checked with the proper locks actually held? | |
1430 | * | |
1431 | * Oops. Go back and try again.. | |
1432 | */ | |
1433 | if (unlikely(running)) { | |
1434 | cpu_relax(); | |
1435 | continue; | |
1436 | } | |
fa490cfd | 1437 | |
3a5c359a AK |
1438 | /* |
1439 | * It's not enough that it's not actively running, | |
1440 | * it must be off the runqueue _entirely_, and not | |
1441 | * preempted! | |
1442 | * | |
80dd99b3 | 1443 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1444 | * running right now), it's preempted, and we should |
1445 | * yield - it could be a while. | |
1446 | */ | |
da0c1e65 | 1447 | if (unlikely(queued)) { |
8eb90c30 TG |
1448 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1449 | ||
1450 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1451 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1452 | continue; |
1453 | } | |
fa490cfd | 1454 | |
3a5c359a AK |
1455 | /* |
1456 | * Ahh, all good. It wasn't running, and it wasn't | |
1457 | * runnable, which means that it will never become | |
1458 | * running in the future either. We're all done! | |
1459 | */ | |
1460 | break; | |
1461 | } | |
85ba2d86 RM |
1462 | |
1463 | return ncsw; | |
1da177e4 LT |
1464 | } |
1465 | ||
1466 | /*** | |
1467 | * kick_process - kick a running thread to enter/exit the kernel | |
1468 | * @p: the to-be-kicked thread | |
1469 | * | |
1470 | * Cause a process which is running on another CPU to enter | |
1471 | * kernel-mode, without any delay. (to get signals handled.) | |
1472 | * | |
25985edc | 1473 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1474 | * because all it wants to ensure is that the remote task enters |
1475 | * the kernel. If the IPI races and the task has been migrated | |
1476 | * to another CPU then no harm is done and the purpose has been | |
1477 | * achieved as well. | |
1478 | */ | |
36c8b586 | 1479 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1480 | { |
1481 | int cpu; | |
1482 | ||
1483 | preempt_disable(); | |
1484 | cpu = task_cpu(p); | |
1485 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1486 | smp_send_reschedule(cpu); | |
1487 | preempt_enable(); | |
1488 | } | |
b43e3521 | 1489 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1490 | |
30da688e | 1491 | /* |
013fdb80 | 1492 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1493 | * |
1494 | * A few notes on cpu_active vs cpu_online: | |
1495 | * | |
1496 | * - cpu_active must be a subset of cpu_online | |
1497 | * | |
1498 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1499 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
1500 | * cpu isn't yet part of the sched domains, and balancing will not | |
1501 | * see it. | |
1502 | * | |
1503 | * - on cpu-down we clear cpu_active() to mask the sched domains and | |
1504 | * avoid the load balancer to place new tasks on the to be removed | |
1505 | * cpu. Existing tasks will remain running there and will be taken | |
1506 | * off. | |
1507 | * | |
1508 | * This means that fallback selection must not select !active CPUs. | |
1509 | * And can assume that any active CPU must be online. Conversely | |
1510 | * select_task_rq() below may allow selection of !active CPUs in order | |
1511 | * to satisfy the above rules. | |
30da688e | 1512 | */ |
5da9a0fb PZ |
1513 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1514 | { | |
aa00d89c TC |
1515 | int nid = cpu_to_node(cpu); |
1516 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1517 | enum { cpuset, possible, fail } state = cpuset; |
1518 | int dest_cpu; | |
5da9a0fb | 1519 | |
aa00d89c TC |
1520 | /* |
1521 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1522 | * will return -1. There is no cpu on the node, and we should | |
1523 | * select the cpu on the other node. | |
1524 | */ | |
1525 | if (nid != -1) { | |
1526 | nodemask = cpumask_of_node(nid); | |
1527 | ||
1528 | /* Look for allowed, online CPU in same node. */ | |
1529 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1530 | if (!cpu_active(dest_cpu)) |
1531 | continue; | |
1532 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1533 | return dest_cpu; | |
1534 | } | |
2baab4e9 | 1535 | } |
5da9a0fb | 1536 | |
2baab4e9 PZ |
1537 | for (;;) { |
1538 | /* Any allowed, online CPU? */ | |
e3831edd | 1539 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
feb245e3 TH |
1540 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
1541 | continue; | |
1542 | if (!cpu_online(dest_cpu)) | |
2baab4e9 PZ |
1543 | continue; |
1544 | goto out; | |
1545 | } | |
5da9a0fb | 1546 | |
e73e85f0 | 1547 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1548 | switch (state) { |
1549 | case cpuset: | |
e73e85f0 ON |
1550 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1551 | cpuset_cpus_allowed_fallback(p); | |
1552 | state = possible; | |
1553 | break; | |
1554 | } | |
1555 | /* fall-through */ | |
2baab4e9 PZ |
1556 | case possible: |
1557 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1558 | state = fail; | |
1559 | break; | |
1560 | ||
1561 | case fail: | |
1562 | BUG(); | |
1563 | break; | |
1564 | } | |
1565 | } | |
1566 | ||
1567 | out: | |
1568 | if (state != cpuset) { | |
1569 | /* | |
1570 | * Don't tell them about moving exiting tasks or | |
1571 | * kernel threads (both mm NULL), since they never | |
1572 | * leave kernel. | |
1573 | */ | |
1574 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1575 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1576 | task_pid_nr(p), p->comm, cpu); |
1577 | } | |
5da9a0fb PZ |
1578 | } |
1579 | ||
1580 | return dest_cpu; | |
1581 | } | |
1582 | ||
e2912009 | 1583 | /* |
013fdb80 | 1584 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1585 | */ |
970b13ba | 1586 | static inline |
ac66f547 | 1587 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1588 | { |
cbce1a68 PZ |
1589 | lockdep_assert_held(&p->pi_lock); |
1590 | ||
50605ffb | 1591 | if (tsk_nr_cpus_allowed(p) > 1) |
6c1d9410 | 1592 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 PZI |
1593 | else |
1594 | cpu = cpumask_any(tsk_cpus_allowed(p)); | |
e2912009 PZ |
1595 | |
1596 | /* | |
1597 | * In order not to call set_task_cpu() on a blocking task we need | |
1598 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1599 | * cpu. | |
1600 | * | |
1601 | * Since this is common to all placement strategies, this lives here. | |
1602 | * | |
1603 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1604 | * not worry about this generic constraint ] | |
1605 | */ | |
fa17b507 | 1606 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1607 | !cpu_online(cpu))) |
5da9a0fb | 1608 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1609 | |
1610 | return cpu; | |
970b13ba | 1611 | } |
09a40af5 MG |
1612 | |
1613 | static void update_avg(u64 *avg, u64 sample) | |
1614 | { | |
1615 | s64 diff = sample - *avg; | |
1616 | *avg += diff >> 3; | |
1617 | } | |
25834c73 PZ |
1618 | |
1619 | #else | |
1620 | ||
1621 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1622 | const struct cpumask *new_mask, bool check) | |
1623 | { | |
1624 | return set_cpus_allowed_ptr(p, new_mask); | |
1625 | } | |
1626 | ||
5cc389bc | 1627 | #endif /* CONFIG_SMP */ |
970b13ba | 1628 | |
d7c01d27 | 1629 | static void |
b84cb5df | 1630 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1631 | { |
4fa8d299 | 1632 | struct rq *rq; |
b84cb5df | 1633 | |
4fa8d299 JP |
1634 | if (!schedstat_enabled()) |
1635 | return; | |
1636 | ||
1637 | rq = this_rq(); | |
d7c01d27 | 1638 | |
4fa8d299 JP |
1639 | #ifdef CONFIG_SMP |
1640 | if (cpu == rq->cpu) { | |
ae92882e JP |
1641 | schedstat_inc(rq->ttwu_local); |
1642 | schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1643 | } else { |
1644 | struct sched_domain *sd; | |
1645 | ||
ae92882e | 1646 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1647 | rcu_read_lock(); |
4fa8d299 | 1648 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1649 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
ae92882e | 1650 | schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1651 | break; |
1652 | } | |
1653 | } | |
057f3fad | 1654 | rcu_read_unlock(); |
d7c01d27 | 1655 | } |
f339b9dc PZ |
1656 | |
1657 | if (wake_flags & WF_MIGRATED) | |
ae92882e | 1658 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1659 | #endif /* CONFIG_SMP */ |
1660 | ||
ae92882e JP |
1661 | schedstat_inc(rq->ttwu_count); |
1662 | schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1663 | |
1664 | if (wake_flags & WF_SYNC) | |
ae92882e | 1665 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1666 | } |
1667 | ||
1de64443 | 1668 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1669 | { |
9ed3811a | 1670 | activate_task(rq, p, en_flags); |
da0c1e65 | 1671 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1672 | |
1673 | /* if a worker is waking up, notify workqueue */ | |
1674 | if (p->flags & PF_WQ_WORKER) | |
1675 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1676 | } |
1677 | ||
23f41eeb PZ |
1678 | /* |
1679 | * Mark the task runnable and perform wakeup-preemption. | |
1680 | */ | |
e7904a28 PZ |
1681 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
1682 | struct pin_cookie cookie) | |
9ed3811a | 1683 | { |
9ed3811a | 1684 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1685 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1686 | trace_sched_wakeup(p); |
1687 | ||
9ed3811a | 1688 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1689 | if (p->sched_class->task_woken) { |
1690 | /* | |
cbce1a68 PZ |
1691 | * Our task @p is fully woken up and running; so its safe to |
1692 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1693 | */ |
e7904a28 | 1694 | lockdep_unpin_lock(&rq->lock, cookie); |
9ed3811a | 1695 | p->sched_class->task_woken(rq, p); |
e7904a28 | 1696 | lockdep_repin_lock(&rq->lock, cookie); |
4c9a4bc8 | 1697 | } |
9ed3811a | 1698 | |
e69c6341 | 1699 | if (rq->idle_stamp) { |
78becc27 | 1700 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1701 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1702 | |
abfafa54 JL |
1703 | update_avg(&rq->avg_idle, delta); |
1704 | ||
1705 | if (rq->avg_idle > max) | |
9ed3811a | 1706 | rq->avg_idle = max; |
abfafa54 | 1707 | |
9ed3811a TH |
1708 | rq->idle_stamp = 0; |
1709 | } | |
1710 | #endif | |
1711 | } | |
1712 | ||
c05fbafb | 1713 | static void |
e7904a28 PZ |
1714 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
1715 | struct pin_cookie cookie) | |
c05fbafb | 1716 | { |
b5179ac7 PZ |
1717 | int en_flags = ENQUEUE_WAKEUP; |
1718 | ||
cbce1a68 PZ |
1719 | lockdep_assert_held(&rq->lock); |
1720 | ||
c05fbafb PZ |
1721 | #ifdef CONFIG_SMP |
1722 | if (p->sched_contributes_to_load) | |
1723 | rq->nr_uninterruptible--; | |
b5179ac7 | 1724 | |
b5179ac7 | 1725 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1726 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1727 | #endif |
1728 | ||
b5179ac7 | 1729 | ttwu_activate(rq, p, en_flags); |
e7904a28 | 1730 | ttwu_do_wakeup(rq, p, wake_flags, cookie); |
c05fbafb PZ |
1731 | } |
1732 | ||
1733 | /* | |
1734 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1735 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1736 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1737 | * the task is still ->on_rq. | |
1738 | */ | |
1739 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1740 | { | |
eb580751 | 1741 | struct rq_flags rf; |
c05fbafb PZ |
1742 | struct rq *rq; |
1743 | int ret = 0; | |
1744 | ||
eb580751 | 1745 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1746 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1747 | /* check_preempt_curr() may use rq clock */ |
1748 | update_rq_clock(rq); | |
e7904a28 | 1749 | ttwu_do_wakeup(rq, p, wake_flags, rf.cookie); |
c05fbafb PZ |
1750 | ret = 1; |
1751 | } | |
eb580751 | 1752 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1753 | |
1754 | return ret; | |
1755 | } | |
1756 | ||
317f3941 | 1757 | #ifdef CONFIG_SMP |
e3baac47 | 1758 | void sched_ttwu_pending(void) |
317f3941 PZ |
1759 | { |
1760 | struct rq *rq = this_rq(); | |
fa14ff4a | 1761 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
e7904a28 | 1762 | struct pin_cookie cookie; |
fa14ff4a | 1763 | struct task_struct *p; |
e3baac47 | 1764 | unsigned long flags; |
317f3941 | 1765 | |
e3baac47 PZ |
1766 | if (!llist) |
1767 | return; | |
1768 | ||
1769 | raw_spin_lock_irqsave(&rq->lock, flags); | |
e7904a28 | 1770 | cookie = lockdep_pin_lock(&rq->lock); |
317f3941 | 1771 | |
fa14ff4a | 1772 | while (llist) { |
b7e7ade3 PZ |
1773 | int wake_flags = 0; |
1774 | ||
fa14ff4a PZ |
1775 | p = llist_entry(llist, struct task_struct, wake_entry); |
1776 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1777 | |
1778 | if (p->sched_remote_wakeup) | |
1779 | wake_flags = WF_MIGRATED; | |
1780 | ||
1781 | ttwu_do_activate(rq, p, wake_flags, cookie); | |
317f3941 PZ |
1782 | } |
1783 | ||
e7904a28 | 1784 | lockdep_unpin_lock(&rq->lock, cookie); |
e3baac47 | 1785 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1786 | } |
1787 | ||
1788 | void scheduler_ipi(void) | |
1789 | { | |
f27dde8d PZ |
1790 | /* |
1791 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1792 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1793 | * this IPI. | |
1794 | */ | |
8cb75e0c | 1795 | preempt_fold_need_resched(); |
f27dde8d | 1796 | |
fd2ac4f4 | 1797 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1798 | return; |
1799 | ||
1800 | /* | |
1801 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1802 | * traditionally all their work was done from the interrupt return | |
1803 | * path. Now that we actually do some work, we need to make sure | |
1804 | * we do call them. | |
1805 | * | |
1806 | * Some archs already do call them, luckily irq_enter/exit nest | |
1807 | * properly. | |
1808 | * | |
1809 | * Arguably we should visit all archs and update all handlers, | |
1810 | * however a fair share of IPIs are still resched only so this would | |
1811 | * somewhat pessimize the simple resched case. | |
1812 | */ | |
1813 | irq_enter(); | |
fa14ff4a | 1814 | sched_ttwu_pending(); |
ca38062e SS |
1815 | |
1816 | /* | |
1817 | * Check if someone kicked us for doing the nohz idle load balance. | |
1818 | */ | |
873b4c65 | 1819 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1820 | this_rq()->idle_balance = 1; |
ca38062e | 1821 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1822 | } |
c5d753a5 | 1823 | irq_exit(); |
317f3941 PZ |
1824 | } |
1825 | ||
b7e7ade3 | 1826 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1827 | { |
e3baac47 PZ |
1828 | struct rq *rq = cpu_rq(cpu); |
1829 | ||
b7e7ade3 PZ |
1830 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1831 | ||
e3baac47 PZ |
1832 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1833 | if (!set_nr_if_polling(rq->idle)) | |
1834 | smp_send_reschedule(cpu); | |
1835 | else | |
1836 | trace_sched_wake_idle_without_ipi(cpu); | |
1837 | } | |
317f3941 | 1838 | } |
d6aa8f85 | 1839 | |
f6be8af1 CL |
1840 | void wake_up_if_idle(int cpu) |
1841 | { | |
1842 | struct rq *rq = cpu_rq(cpu); | |
1843 | unsigned long flags; | |
1844 | ||
fd7de1e8 AL |
1845 | rcu_read_lock(); |
1846 | ||
1847 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1848 | goto out; | |
f6be8af1 CL |
1849 | |
1850 | if (set_nr_if_polling(rq->idle)) { | |
1851 | trace_sched_wake_idle_without_ipi(cpu); | |
1852 | } else { | |
1853 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1854 | if (is_idle_task(rq->curr)) | |
1855 | smp_send_reschedule(cpu); | |
1856 | /* Else cpu is not in idle, do nothing here */ | |
1857 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1858 | } | |
fd7de1e8 AL |
1859 | |
1860 | out: | |
1861 | rcu_read_unlock(); | |
f6be8af1 CL |
1862 | } |
1863 | ||
39be3501 | 1864 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1865 | { |
1866 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1867 | } | |
d6aa8f85 | 1868 | #endif /* CONFIG_SMP */ |
317f3941 | 1869 | |
b5179ac7 | 1870 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1871 | { |
1872 | struct rq *rq = cpu_rq(cpu); | |
e7904a28 | 1873 | struct pin_cookie cookie; |
c05fbafb | 1874 | |
17d9f311 | 1875 | #if defined(CONFIG_SMP) |
39be3501 | 1876 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1877 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
b7e7ade3 | 1878 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1879 | return; |
1880 | } | |
1881 | #endif | |
1882 | ||
c05fbafb | 1883 | raw_spin_lock(&rq->lock); |
e7904a28 | 1884 | cookie = lockdep_pin_lock(&rq->lock); |
b5179ac7 | 1885 | ttwu_do_activate(rq, p, wake_flags, cookie); |
e7904a28 | 1886 | lockdep_unpin_lock(&rq->lock, cookie); |
c05fbafb | 1887 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1888 | } |
1889 | ||
8643cda5 PZ |
1890 | /* |
1891 | * Notes on Program-Order guarantees on SMP systems. | |
1892 | * | |
1893 | * MIGRATION | |
1894 | * | |
1895 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1896 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1897 | * execution on its new cpu [c1]. | |
1898 | * | |
1899 | * For migration (of runnable tasks) this is provided by the following means: | |
1900 | * | |
1901 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1902 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1903 | * rq(c1)->lock (if not at the same time, then in that order). | |
1904 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1905 | * | |
1906 | * Transitivity guarantees that B happens after A and C after B. | |
1907 | * Note: we only require RCpc transitivity. | |
1908 | * Note: the cpu doing B need not be c0 or c1 | |
1909 | * | |
1910 | * Example: | |
1911 | * | |
1912 | * CPU0 CPU1 CPU2 | |
1913 | * | |
1914 | * LOCK rq(0)->lock | |
1915 | * sched-out X | |
1916 | * sched-in Y | |
1917 | * UNLOCK rq(0)->lock | |
1918 | * | |
1919 | * LOCK rq(0)->lock // orders against CPU0 | |
1920 | * dequeue X | |
1921 | * UNLOCK rq(0)->lock | |
1922 | * | |
1923 | * LOCK rq(1)->lock | |
1924 | * enqueue X | |
1925 | * UNLOCK rq(1)->lock | |
1926 | * | |
1927 | * LOCK rq(1)->lock // orders against CPU2 | |
1928 | * sched-out Z | |
1929 | * sched-in X | |
1930 | * UNLOCK rq(1)->lock | |
1931 | * | |
1932 | * | |
1933 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1934 | * | |
1935 | * For blocking we (obviously) need to provide the same guarantee as for | |
1936 | * migration. However the means are completely different as there is no lock | |
1937 | * chain to provide order. Instead we do: | |
1938 | * | |
1939 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1940 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1941 | * |
1942 | * Example: | |
1943 | * | |
1944 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1945 | * | |
1946 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1947 | * dequeue X | |
1948 | * sched-out X | |
1949 | * smp_store_release(X->on_cpu, 0); | |
1950 | * | |
1f03e8d2 | 1951 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1952 | * X->state = WAKING |
1953 | * set_task_cpu(X,2) | |
1954 | * | |
1955 | * LOCK rq(2)->lock | |
1956 | * enqueue X | |
1957 | * X->state = RUNNING | |
1958 | * UNLOCK rq(2)->lock | |
1959 | * | |
1960 | * LOCK rq(2)->lock // orders against CPU1 | |
1961 | * sched-out Z | |
1962 | * sched-in X | |
1963 | * UNLOCK rq(2)->lock | |
1964 | * | |
1965 | * UNLOCK X->pi_lock | |
1966 | * UNLOCK rq(0)->lock | |
1967 | * | |
1968 | * | |
1969 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1970 | * must observe the state that lead to our wakeup. That is, not only must our | |
1971 | * task observe its own prior state, it must also observe the stores prior to | |
1972 | * its wakeup. | |
1973 | * | |
1974 | * This means that any means of doing remote wakeups must order the CPU doing | |
1975 | * the wakeup against the CPU the task is going to end up running on. This, | |
1976 | * however, is already required for the regular Program-Order guarantee above, | |
1f03e8d2 | 1977 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
8643cda5 PZ |
1978 | * |
1979 | */ | |
1980 | ||
9ed3811a | 1981 | /** |
1da177e4 | 1982 | * try_to_wake_up - wake up a thread |
9ed3811a | 1983 | * @p: the thread to be awakened |
1da177e4 | 1984 | * @state: the mask of task states that can be woken |
9ed3811a | 1985 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1986 | * |
1987 | * Put it on the run-queue if it's not already there. The "current" | |
1988 | * thread is always on the run-queue (except when the actual | |
1989 | * re-schedule is in progress), and as such you're allowed to do | |
1990 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1991 | * runnable without the overhead of this. | |
1992 | * | |
e69f6186 | 1993 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1994 | * or @state didn't match @p's state. |
1da177e4 | 1995 | */ |
e4a52bcb PZ |
1996 | static int |
1997 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1998 | { |
1da177e4 | 1999 | unsigned long flags; |
c05fbafb | 2000 | int cpu, success = 0; |
2398f2c6 | 2001 | |
e0acd0a6 ON |
2002 | /* |
2003 | * If we are going to wake up a thread waiting for CONDITION we | |
2004 | * need to ensure that CONDITION=1 done by the caller can not be | |
2005 | * reordered with p->state check below. This pairs with mb() in | |
2006 | * set_current_state() the waiting thread does. | |
2007 | */ | |
2008 | smp_mb__before_spinlock(); | |
013fdb80 | 2009 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2010 | if (!(p->state & state)) |
1da177e4 LT |
2011 | goto out; |
2012 | ||
fbd705a0 PZ |
2013 | trace_sched_waking(p); |
2014 | ||
c05fbafb | 2015 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2016 | cpu = task_cpu(p); |
1da177e4 | 2017 | |
135e8c92 BS |
2018 | /* |
2019 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2020 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2021 | * in smp_cond_load_acquire() below. | |
2022 | * | |
2023 | * sched_ttwu_pending() try_to_wake_up() | |
2024 | * [S] p->on_rq = 1; [L] P->state | |
2025 | * UNLOCK rq->lock -----. | |
2026 | * \ | |
2027 | * +--- RMB | |
2028 | * schedule() / | |
2029 | * LOCK rq->lock -----' | |
2030 | * UNLOCK rq->lock | |
2031 | * | |
2032 | * [task p] | |
2033 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq | |
2034 | * | |
2035 | * Pairs with the UNLOCK+LOCK on rq->lock from the | |
2036 | * last wakeup of our task and the schedule that got our task | |
2037 | * current. | |
2038 | */ | |
2039 | smp_rmb(); | |
c05fbafb PZ |
2040 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2041 | goto stat; | |
1da177e4 | 2042 | |
1da177e4 | 2043 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2044 | /* |
2045 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2046 | * possible to, falsely, observe p->on_cpu == 0. | |
2047 | * | |
2048 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2049 | * from the runqueue. | |
2050 | * | |
2051 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2052 | * UNLOCK rq->lock | |
2053 | * RMB | |
2054 | * LOCK rq->lock | |
2055 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2056 | * | |
2057 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2058 | * from the consecutive calls to schedule(); the first switching to our | |
2059 | * task, the second putting it to sleep. | |
2060 | */ | |
2061 | smp_rmb(); | |
2062 | ||
e9c84311 | 2063 | /* |
c05fbafb PZ |
2064 | * If the owning (remote) cpu is still in the middle of schedule() with |
2065 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
2066 | * |
2067 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2068 | * | |
2069 | * This ensures that tasks getting woken will be fully ordered against | |
2070 | * their previous state and preserve Program Order. | |
0970d299 | 2071 | */ |
1f03e8d2 | 2072 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2073 | |
a8e4f2ea | 2074 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2075 | p->state = TASK_WAKING; |
e7693a36 | 2076 | |
ac66f547 | 2077 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2078 | if (task_cpu(p) != cpu) { |
2079 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2080 | set_task_cpu(p, cpu); |
f339b9dc | 2081 | } |
1da177e4 | 2082 | #endif /* CONFIG_SMP */ |
1da177e4 | 2083 | |
b5179ac7 | 2084 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2085 | stat: |
4fa8d299 | 2086 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2087 | out: |
013fdb80 | 2088 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2089 | |
2090 | return success; | |
2091 | } | |
2092 | ||
21aa9af0 TH |
2093 | /** |
2094 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2095 | * @p: the thread to be awakened | |
9279e0d2 | 2096 | * @cookie: context's cookie for pinning |
21aa9af0 | 2097 | * |
2acca55e | 2098 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2099 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2100 | * the current task. |
21aa9af0 | 2101 | */ |
e7904a28 | 2102 | static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie) |
21aa9af0 TH |
2103 | { |
2104 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2105 | |
383efcd0 TH |
2106 | if (WARN_ON_ONCE(rq != this_rq()) || |
2107 | WARN_ON_ONCE(p == current)) | |
2108 | return; | |
2109 | ||
21aa9af0 TH |
2110 | lockdep_assert_held(&rq->lock); |
2111 | ||
2acca55e | 2112 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2113 | /* |
2114 | * This is OK, because current is on_cpu, which avoids it being | |
2115 | * picked for load-balance and preemption/IRQs are still | |
2116 | * disabled avoiding further scheduler activity on it and we've | |
2117 | * not yet picked a replacement task. | |
2118 | */ | |
e7904a28 | 2119 | lockdep_unpin_lock(&rq->lock, cookie); |
2acca55e PZ |
2120 | raw_spin_unlock(&rq->lock); |
2121 | raw_spin_lock(&p->pi_lock); | |
2122 | raw_spin_lock(&rq->lock); | |
e7904a28 | 2123 | lockdep_repin_lock(&rq->lock, cookie); |
2acca55e PZ |
2124 | } |
2125 | ||
21aa9af0 | 2126 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2127 | goto out; |
21aa9af0 | 2128 | |
fbd705a0 PZ |
2129 | trace_sched_waking(p); |
2130 | ||
da0c1e65 | 2131 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2132 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2133 | ||
e7904a28 | 2134 | ttwu_do_wakeup(rq, p, 0, cookie); |
4fa8d299 | 2135 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2136 | out: |
2137 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2138 | } |
2139 | ||
50fa610a DH |
2140 | /** |
2141 | * wake_up_process - Wake up a specific process | |
2142 | * @p: The process to be woken up. | |
2143 | * | |
2144 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2145 | * processes. |
2146 | * | |
2147 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2148 | * |
2149 | * It may be assumed that this function implies a write memory barrier before | |
2150 | * changing the task state if and only if any tasks are woken up. | |
2151 | */ | |
7ad5b3a5 | 2152 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2153 | { |
9067ac85 | 2154 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2155 | } |
1da177e4 LT |
2156 | EXPORT_SYMBOL(wake_up_process); |
2157 | ||
7ad5b3a5 | 2158 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2159 | { |
2160 | return try_to_wake_up(p, state, 0); | |
2161 | } | |
2162 | ||
a5e7be3b JL |
2163 | /* |
2164 | * This function clears the sched_dl_entity static params. | |
2165 | */ | |
2166 | void __dl_clear_params(struct task_struct *p) | |
2167 | { | |
2168 | struct sched_dl_entity *dl_se = &p->dl; | |
2169 | ||
2170 | dl_se->dl_runtime = 0; | |
2171 | dl_se->dl_deadline = 0; | |
2172 | dl_se->dl_period = 0; | |
2173 | dl_se->flags = 0; | |
2174 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2175 | |
2176 | dl_se->dl_throttled = 0; | |
40767b0d | 2177 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2178 | } |
2179 | ||
1da177e4 LT |
2180 | /* |
2181 | * Perform scheduler related setup for a newly forked process p. | |
2182 | * p is forked by current. | |
dd41f596 IM |
2183 | * |
2184 | * __sched_fork() is basic setup used by init_idle() too: | |
2185 | */ | |
5e1576ed | 2186 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2187 | { |
fd2f4419 PZ |
2188 | p->on_rq = 0; |
2189 | ||
2190 | p->se.on_rq = 0; | |
dd41f596 IM |
2191 | p->se.exec_start = 0; |
2192 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2193 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2194 | p->se.nr_migrations = 0; |
da7a735e | 2195 | p->se.vruntime = 0; |
fd2f4419 | 2196 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2197 | |
ad936d86 BP |
2198 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2199 | p->se.cfs_rq = NULL; | |
2200 | #endif | |
2201 | ||
6cfb0d5d | 2202 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2203 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2204 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2205 | #endif |
476d139c | 2206 | |
aab03e05 | 2207 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2208 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2209 | __dl_clear_params(p); |
aab03e05 | 2210 | |
fa717060 | 2211 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2212 | p->rt.timeout = 0; |
2213 | p->rt.time_slice = sched_rr_timeslice; | |
2214 | p->rt.on_rq = 0; | |
2215 | p->rt.on_list = 0; | |
476d139c | 2216 | |
e107be36 AK |
2217 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2218 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2219 | #endif | |
cbee9f88 PZ |
2220 | |
2221 | #ifdef CONFIG_NUMA_BALANCING | |
2222 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2223 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2224 | p->mm->numa_scan_seq = 0; |
2225 | } | |
2226 | ||
5e1576ed RR |
2227 | if (clone_flags & CLONE_VM) |
2228 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2229 | else | |
2230 | p->numa_preferred_nid = -1; | |
2231 | ||
cbee9f88 PZ |
2232 | p->node_stamp = 0ULL; |
2233 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2234 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2235 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2236 | p->numa_faults = NULL; |
7e2703e6 RR |
2237 | p->last_task_numa_placement = 0; |
2238 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2239 | |
8c8a743c | 2240 | p->numa_group = NULL; |
cbee9f88 | 2241 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2242 | } |
2243 | ||
2a595721 SD |
2244 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2245 | ||
1a687c2e | 2246 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2247 | |
1a687c2e MG |
2248 | void set_numabalancing_state(bool enabled) |
2249 | { | |
2250 | if (enabled) | |
2a595721 | 2251 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2252 | else |
2a595721 | 2253 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2254 | } |
54a43d54 AK |
2255 | |
2256 | #ifdef CONFIG_PROC_SYSCTL | |
2257 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2258 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2259 | { | |
2260 | struct ctl_table t; | |
2261 | int err; | |
2a595721 | 2262 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2263 | |
2264 | if (write && !capable(CAP_SYS_ADMIN)) | |
2265 | return -EPERM; | |
2266 | ||
2267 | t = *table; | |
2268 | t.data = &state; | |
2269 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2270 | if (err < 0) | |
2271 | return err; | |
2272 | if (write) | |
2273 | set_numabalancing_state(state); | |
2274 | return err; | |
2275 | } | |
2276 | #endif | |
2277 | #endif | |
dd41f596 | 2278 | |
4698f88c JP |
2279 | #ifdef CONFIG_SCHEDSTATS |
2280 | ||
cb251765 | 2281 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2282 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2283 | |
cb251765 MG |
2284 | static void set_schedstats(bool enabled) |
2285 | { | |
2286 | if (enabled) | |
2287 | static_branch_enable(&sched_schedstats); | |
2288 | else | |
2289 | static_branch_disable(&sched_schedstats); | |
2290 | } | |
2291 | ||
2292 | void force_schedstat_enabled(void) | |
2293 | { | |
2294 | if (!schedstat_enabled()) { | |
2295 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2296 | static_branch_enable(&sched_schedstats); | |
2297 | } | |
2298 | } | |
2299 | ||
2300 | static int __init setup_schedstats(char *str) | |
2301 | { | |
2302 | int ret = 0; | |
2303 | if (!str) | |
2304 | goto out; | |
2305 | ||
4698f88c JP |
2306 | /* |
2307 | * This code is called before jump labels have been set up, so we can't | |
2308 | * change the static branch directly just yet. Instead set a temporary | |
2309 | * variable so init_schedstats() can do it later. | |
2310 | */ | |
cb251765 | 2311 | if (!strcmp(str, "enable")) { |
4698f88c | 2312 | __sched_schedstats = true; |
cb251765 MG |
2313 | ret = 1; |
2314 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2315 | __sched_schedstats = false; |
cb251765 MG |
2316 | ret = 1; |
2317 | } | |
2318 | out: | |
2319 | if (!ret) | |
2320 | pr_warn("Unable to parse schedstats=\n"); | |
2321 | ||
2322 | return ret; | |
2323 | } | |
2324 | __setup("schedstats=", setup_schedstats); | |
2325 | ||
4698f88c JP |
2326 | static void __init init_schedstats(void) |
2327 | { | |
2328 | set_schedstats(__sched_schedstats); | |
2329 | } | |
2330 | ||
cb251765 MG |
2331 | #ifdef CONFIG_PROC_SYSCTL |
2332 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2333 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2334 | { | |
2335 | struct ctl_table t; | |
2336 | int err; | |
2337 | int state = static_branch_likely(&sched_schedstats); | |
2338 | ||
2339 | if (write && !capable(CAP_SYS_ADMIN)) | |
2340 | return -EPERM; | |
2341 | ||
2342 | t = *table; | |
2343 | t.data = &state; | |
2344 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2345 | if (err < 0) | |
2346 | return err; | |
2347 | if (write) | |
2348 | set_schedstats(state); | |
2349 | return err; | |
2350 | } | |
4698f88c JP |
2351 | #endif /* CONFIG_PROC_SYSCTL */ |
2352 | #else /* !CONFIG_SCHEDSTATS */ | |
2353 | static inline void init_schedstats(void) {} | |
2354 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2355 | |
2356 | /* | |
2357 | * fork()/clone()-time setup: | |
2358 | */ | |
aab03e05 | 2359 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2360 | { |
0122ec5b | 2361 | unsigned long flags; |
dd41f596 IM |
2362 | int cpu = get_cpu(); |
2363 | ||
5e1576ed | 2364 | __sched_fork(clone_flags, p); |
06b83b5f | 2365 | /* |
7dc603c9 | 2366 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2367 | * nobody will actually run it, and a signal or other external |
2368 | * event cannot wake it up and insert it on the runqueue either. | |
2369 | */ | |
7dc603c9 | 2370 | p->state = TASK_NEW; |
dd41f596 | 2371 | |
c350a04e MG |
2372 | /* |
2373 | * Make sure we do not leak PI boosting priority to the child. | |
2374 | */ | |
2375 | p->prio = current->normal_prio; | |
2376 | ||
b9dc29e7 MG |
2377 | /* |
2378 | * Revert to default priority/policy on fork if requested. | |
2379 | */ | |
2380 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2381 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2382 | p->policy = SCHED_NORMAL; |
6c697bdf | 2383 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2384 | p->rt_priority = 0; |
2385 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2386 | p->static_prio = NICE_TO_PRIO(0); | |
2387 | ||
2388 | p->prio = p->normal_prio = __normal_prio(p); | |
2389 | set_load_weight(p); | |
6c697bdf | 2390 | |
b9dc29e7 MG |
2391 | /* |
2392 | * We don't need the reset flag anymore after the fork. It has | |
2393 | * fulfilled its duty: | |
2394 | */ | |
2395 | p->sched_reset_on_fork = 0; | |
2396 | } | |
ca94c442 | 2397 | |
aab03e05 DF |
2398 | if (dl_prio(p->prio)) { |
2399 | put_cpu(); | |
2400 | return -EAGAIN; | |
2401 | } else if (rt_prio(p->prio)) { | |
2402 | p->sched_class = &rt_sched_class; | |
2403 | } else { | |
2ddbf952 | 2404 | p->sched_class = &fair_sched_class; |
aab03e05 | 2405 | } |
b29739f9 | 2406 | |
7dc603c9 | 2407 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2408 | |
86951599 PZ |
2409 | /* |
2410 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2411 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2412 | * is ran before sched_fork(). | |
2413 | * | |
2414 | * Silence PROVE_RCU. | |
2415 | */ | |
0122ec5b | 2416 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd PZ |
2417 | /* |
2418 | * We're setting the cpu for the first time, we don't migrate, | |
2419 | * so use __set_task_cpu(). | |
2420 | */ | |
2421 | __set_task_cpu(p, cpu); | |
2422 | if (p->sched_class->task_fork) | |
2423 | p->sched_class->task_fork(p); | |
0122ec5b | 2424 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2425 | |
f6db8347 | 2426 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2427 | if (likely(sched_info_on())) |
52f17b6c | 2428 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2429 | #endif |
3ca7a440 PZ |
2430 | #if defined(CONFIG_SMP) |
2431 | p->on_cpu = 0; | |
4866cde0 | 2432 | #endif |
01028747 | 2433 | init_task_preempt_count(p); |
806c09a7 | 2434 | #ifdef CONFIG_SMP |
917b627d | 2435 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2436 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2437 | #endif |
917b627d | 2438 | |
476d139c | 2439 | put_cpu(); |
aab03e05 | 2440 | return 0; |
1da177e4 LT |
2441 | } |
2442 | ||
332ac17e DF |
2443 | unsigned long to_ratio(u64 period, u64 runtime) |
2444 | { | |
2445 | if (runtime == RUNTIME_INF) | |
2446 | return 1ULL << 20; | |
2447 | ||
2448 | /* | |
2449 | * Doing this here saves a lot of checks in all | |
2450 | * the calling paths, and returning zero seems | |
2451 | * safe for them anyway. | |
2452 | */ | |
2453 | if (period == 0) | |
2454 | return 0; | |
2455 | ||
2456 | return div64_u64(runtime << 20, period); | |
2457 | } | |
2458 | ||
2459 | #ifdef CONFIG_SMP | |
2460 | inline struct dl_bw *dl_bw_of(int i) | |
2461 | { | |
f78f5b90 PM |
2462 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2463 | "sched RCU must be held"); | |
332ac17e DF |
2464 | return &cpu_rq(i)->rd->dl_bw; |
2465 | } | |
2466 | ||
de212f18 | 2467 | static inline int dl_bw_cpus(int i) |
332ac17e | 2468 | { |
de212f18 PZ |
2469 | struct root_domain *rd = cpu_rq(i)->rd; |
2470 | int cpus = 0; | |
2471 | ||
f78f5b90 PM |
2472 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2473 | "sched RCU must be held"); | |
de212f18 PZ |
2474 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2475 | cpus++; | |
2476 | ||
2477 | return cpus; | |
332ac17e DF |
2478 | } |
2479 | #else | |
2480 | inline struct dl_bw *dl_bw_of(int i) | |
2481 | { | |
2482 | return &cpu_rq(i)->dl.dl_bw; | |
2483 | } | |
2484 | ||
de212f18 | 2485 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2486 | { |
2487 | return 1; | |
2488 | } | |
2489 | #endif | |
2490 | ||
332ac17e DF |
2491 | /* |
2492 | * We must be sure that accepting a new task (or allowing changing the | |
2493 | * parameters of an existing one) is consistent with the bandwidth | |
2494 | * constraints. If yes, this function also accordingly updates the currently | |
2495 | * allocated bandwidth to reflect the new situation. | |
2496 | * | |
2497 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2498 | * |
2499 | * XXX we should delay bw change until the task's 0-lag point, see | |
2500 | * __setparam_dl(). | |
332ac17e DF |
2501 | */ |
2502 | static int dl_overflow(struct task_struct *p, int policy, | |
2503 | const struct sched_attr *attr) | |
2504 | { | |
2505 | ||
2506 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2507 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2508 | u64 runtime = attr->sched_runtime; |
2509 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2510 | int cpus, err = -1; |
332ac17e | 2511 | |
fec148c0 XP |
2512 | /* !deadline task may carry old deadline bandwidth */ |
2513 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2514 | return 0; |
2515 | ||
2516 | /* | |
2517 | * Either if a task, enters, leave, or stays -deadline but changes | |
2518 | * its parameters, we may need to update accordingly the total | |
2519 | * allocated bandwidth of the container. | |
2520 | */ | |
2521 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2522 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2523 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2524 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2525 | __dl_add(dl_b, new_bw); | |
2526 | err = 0; | |
2527 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2528 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2529 | __dl_clear(dl_b, p->dl.dl_bw); | |
2530 | __dl_add(dl_b, new_bw); | |
2531 | err = 0; | |
2532 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2533 | __dl_clear(dl_b, p->dl.dl_bw); | |
2534 | err = 0; | |
2535 | } | |
2536 | raw_spin_unlock(&dl_b->lock); | |
2537 | ||
2538 | return err; | |
2539 | } | |
2540 | ||
2541 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2542 | ||
1da177e4 LT |
2543 | /* |
2544 | * wake_up_new_task - wake up a newly created task for the first time. | |
2545 | * | |
2546 | * This function will do some initial scheduler statistics housekeeping | |
2547 | * that must be done for every newly created context, then puts the task | |
2548 | * on the runqueue and wakes it. | |
2549 | */ | |
3e51e3ed | 2550 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2551 | { |
eb580751 | 2552 | struct rq_flags rf; |
dd41f596 | 2553 | struct rq *rq; |
fabf318e | 2554 | |
eb580751 | 2555 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2556 | p->state = TASK_RUNNING; |
fabf318e PZ |
2557 | #ifdef CONFIG_SMP |
2558 | /* | |
2559 | * Fork balancing, do it here and not earlier because: | |
2560 | * - cpus_allowed can change in the fork path | |
2561 | * - any previously selected cpu might disappear through hotplug | |
e210bffd PZ |
2562 | * |
2563 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2564 | * as we're not fully set-up yet. | |
fabf318e | 2565 | */ |
e210bffd | 2566 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2567 | #endif |
b7fa30c9 | 2568 | rq = __task_rq_lock(p, &rf); |
2b8c41da | 2569 | post_init_entity_util_avg(&p->se); |
0017d735 | 2570 | |
cd29fe6f | 2571 | activate_task(rq, p, 0); |
da0c1e65 | 2572 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2573 | trace_sched_wakeup_new(p); |
a7558e01 | 2574 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2575 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2576 | if (p->sched_class->task_woken) { |
2577 | /* | |
2578 | * Nothing relies on rq->lock after this, so its fine to | |
2579 | * drop it. | |
2580 | */ | |
e7904a28 | 2581 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
efbbd05a | 2582 | p->sched_class->task_woken(rq, p); |
e7904a28 | 2583 | lockdep_repin_lock(&rq->lock, rf.cookie); |
0aaafaab | 2584 | } |
9a897c5a | 2585 | #endif |
eb580751 | 2586 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2587 | } |
2588 | ||
e107be36 AK |
2589 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2590 | ||
1cde2930 PZ |
2591 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2592 | ||
2ecd9d29 PZ |
2593 | void preempt_notifier_inc(void) |
2594 | { | |
2595 | static_key_slow_inc(&preempt_notifier_key); | |
2596 | } | |
2597 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2598 | ||
2599 | void preempt_notifier_dec(void) | |
2600 | { | |
2601 | static_key_slow_dec(&preempt_notifier_key); | |
2602 | } | |
2603 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2604 | ||
e107be36 | 2605 | /** |
80dd99b3 | 2606 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2607 | * @notifier: notifier struct to register |
e107be36 AK |
2608 | */ |
2609 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2610 | { | |
2ecd9d29 PZ |
2611 | if (!static_key_false(&preempt_notifier_key)) |
2612 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2613 | ||
e107be36 AK |
2614 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2615 | } | |
2616 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2617 | ||
2618 | /** | |
2619 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2620 | * @notifier: notifier struct to unregister |
e107be36 | 2621 | * |
d84525a8 | 2622 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2623 | */ |
2624 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2625 | { | |
2626 | hlist_del(¬ifier->link); | |
2627 | } | |
2628 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2629 | ||
1cde2930 | 2630 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2631 | { |
2632 | struct preempt_notifier *notifier; | |
e107be36 | 2633 | |
b67bfe0d | 2634 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2635 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2636 | } | |
2637 | ||
1cde2930 PZ |
2638 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2639 | { | |
2640 | if (static_key_false(&preempt_notifier_key)) | |
2641 | __fire_sched_in_preempt_notifiers(curr); | |
2642 | } | |
2643 | ||
e107be36 | 2644 | static void |
1cde2930 PZ |
2645 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2646 | struct task_struct *next) | |
e107be36 AK |
2647 | { |
2648 | struct preempt_notifier *notifier; | |
e107be36 | 2649 | |
b67bfe0d | 2650 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2651 | notifier->ops->sched_out(notifier, next); |
2652 | } | |
2653 | ||
1cde2930 PZ |
2654 | static __always_inline void |
2655 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2656 | struct task_struct *next) | |
2657 | { | |
2658 | if (static_key_false(&preempt_notifier_key)) | |
2659 | __fire_sched_out_preempt_notifiers(curr, next); | |
2660 | } | |
2661 | ||
6d6bc0ad | 2662 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2663 | |
1cde2930 | 2664 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2665 | { |
2666 | } | |
2667 | ||
1cde2930 | 2668 | static inline void |
e107be36 AK |
2669 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2670 | struct task_struct *next) | |
2671 | { | |
2672 | } | |
2673 | ||
6d6bc0ad | 2674 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2675 | |
4866cde0 NP |
2676 | /** |
2677 | * prepare_task_switch - prepare to switch tasks | |
2678 | * @rq: the runqueue preparing to switch | |
421cee29 | 2679 | * @prev: the current task that is being switched out |
4866cde0 NP |
2680 | * @next: the task we are going to switch to. |
2681 | * | |
2682 | * This is called with the rq lock held and interrupts off. It must | |
2683 | * be paired with a subsequent finish_task_switch after the context | |
2684 | * switch. | |
2685 | * | |
2686 | * prepare_task_switch sets up locking and calls architecture specific | |
2687 | * hooks. | |
2688 | */ | |
e107be36 AK |
2689 | static inline void |
2690 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2691 | struct task_struct *next) | |
4866cde0 | 2692 | { |
43148951 | 2693 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2694 | perf_event_task_sched_out(prev, next); |
e107be36 | 2695 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2696 | prepare_lock_switch(rq, next); |
2697 | prepare_arch_switch(next); | |
2698 | } | |
2699 | ||
1da177e4 LT |
2700 | /** |
2701 | * finish_task_switch - clean up after a task-switch | |
2702 | * @prev: the thread we just switched away from. | |
2703 | * | |
4866cde0 NP |
2704 | * finish_task_switch must be called after the context switch, paired |
2705 | * with a prepare_task_switch call before the context switch. | |
2706 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2707 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2708 | * |
2709 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2710 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2711 | * with the lock held can cause deadlocks; see schedule() for |
2712 | * details.) | |
dfa50b60 ON |
2713 | * |
2714 | * The context switch have flipped the stack from under us and restored the | |
2715 | * local variables which were saved when this task called schedule() in the | |
2716 | * past. prev == current is still correct but we need to recalculate this_rq | |
2717 | * because prev may have moved to another CPU. | |
1da177e4 | 2718 | */ |
dfa50b60 | 2719 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2720 | __releases(rq->lock) |
2721 | { | |
dfa50b60 | 2722 | struct rq *rq = this_rq(); |
1da177e4 | 2723 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2724 | long prev_state; |
1da177e4 | 2725 | |
609ca066 PZ |
2726 | /* |
2727 | * The previous task will have left us with a preempt_count of 2 | |
2728 | * because it left us after: | |
2729 | * | |
2730 | * schedule() | |
2731 | * preempt_disable(); // 1 | |
2732 | * __schedule() | |
2733 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2734 | * | |
2735 | * Also, see FORK_PREEMPT_COUNT. | |
2736 | */ | |
e2bf1c4b PZ |
2737 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2738 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2739 | current->comm, current->pid, preempt_count())) | |
2740 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2741 | |
1da177e4 LT |
2742 | rq->prev_mm = NULL; |
2743 | ||
2744 | /* | |
2745 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2746 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2747 | * schedule one last time. The schedule call will never return, and |
2748 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2749 | * |
2750 | * We must observe prev->state before clearing prev->on_cpu (in | |
2751 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2752 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2753 | * transition, resulting in a double drop. | |
1da177e4 | 2754 | */ |
55a101f8 | 2755 | prev_state = prev->state; |
bf9fae9f | 2756 | vtime_task_switch(prev); |
a8d757ef | 2757 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2758 | finish_lock_switch(rq, prev); |
01f23e16 | 2759 | finish_arch_post_lock_switch(); |
e8fa1362 | 2760 | |
e107be36 | 2761 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2762 | if (mm) |
2763 | mmdrop(mm); | |
c394cc9f | 2764 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2765 | if (prev->sched_class->task_dead) |
2766 | prev->sched_class->task_dead(prev); | |
2767 | ||
c6fd91f0 | 2768 | /* |
2769 | * Remove function-return probe instances associated with this | |
2770 | * task and put them back on the free list. | |
9761eea8 | 2771 | */ |
c6fd91f0 | 2772 | kprobe_flush_task(prev); |
1da177e4 | 2773 | put_task_struct(prev); |
c6fd91f0 | 2774 | } |
99e5ada9 | 2775 | |
de734f89 | 2776 | tick_nohz_task_switch(); |
dfa50b60 | 2777 | return rq; |
1da177e4 LT |
2778 | } |
2779 | ||
3f029d3c GH |
2780 | #ifdef CONFIG_SMP |
2781 | ||
3f029d3c | 2782 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2783 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2784 | { |
e3fca9e7 PZ |
2785 | struct callback_head *head, *next; |
2786 | void (*func)(struct rq *rq); | |
2787 | unsigned long flags; | |
3f029d3c | 2788 | |
e3fca9e7 PZ |
2789 | raw_spin_lock_irqsave(&rq->lock, flags); |
2790 | head = rq->balance_callback; | |
2791 | rq->balance_callback = NULL; | |
2792 | while (head) { | |
2793 | func = (void (*)(struct rq *))head->func; | |
2794 | next = head->next; | |
2795 | head->next = NULL; | |
2796 | head = next; | |
3f029d3c | 2797 | |
e3fca9e7 | 2798 | func(rq); |
3f029d3c | 2799 | } |
e3fca9e7 PZ |
2800 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2801 | } | |
2802 | ||
2803 | static inline void balance_callback(struct rq *rq) | |
2804 | { | |
2805 | if (unlikely(rq->balance_callback)) | |
2806 | __balance_callback(rq); | |
3f029d3c GH |
2807 | } |
2808 | ||
2809 | #else | |
da19ab51 | 2810 | |
e3fca9e7 | 2811 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2812 | { |
1da177e4 LT |
2813 | } |
2814 | ||
3f029d3c GH |
2815 | #endif |
2816 | ||
1da177e4 LT |
2817 | /** |
2818 | * schedule_tail - first thing a freshly forked thread must call. | |
2819 | * @prev: the thread we just switched away from. | |
2820 | */ | |
722a9f92 | 2821 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2822 | __releases(rq->lock) |
2823 | { | |
1a43a14a | 2824 | struct rq *rq; |
da19ab51 | 2825 | |
609ca066 PZ |
2826 | /* |
2827 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2828 | * finish_task_switch() for details. | |
2829 | * | |
2830 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2831 | * and the preempt_enable() will end up enabling preemption (on | |
2832 | * PREEMPT_COUNT kernels). | |
2833 | */ | |
2834 | ||
dfa50b60 | 2835 | rq = finish_task_switch(prev); |
e3fca9e7 | 2836 | balance_callback(rq); |
1a43a14a | 2837 | preempt_enable(); |
70b97a7f | 2838 | |
1da177e4 | 2839 | if (current->set_child_tid) |
b488893a | 2840 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2841 | } |
2842 | ||
2843 | /* | |
dfa50b60 | 2844 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2845 | */ |
04936948 | 2846 | static __always_inline struct rq * |
70b97a7f | 2847 | context_switch(struct rq *rq, struct task_struct *prev, |
e7904a28 | 2848 | struct task_struct *next, struct pin_cookie cookie) |
1da177e4 | 2849 | { |
dd41f596 | 2850 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2851 | |
e107be36 | 2852 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2853 | |
dd41f596 IM |
2854 | mm = next->mm; |
2855 | oldmm = prev->active_mm; | |
9226d125 ZA |
2856 | /* |
2857 | * For paravirt, this is coupled with an exit in switch_to to | |
2858 | * combine the page table reload and the switch backend into | |
2859 | * one hypercall. | |
2860 | */ | |
224101ed | 2861 | arch_start_context_switch(prev); |
9226d125 | 2862 | |
31915ab4 | 2863 | if (!mm) { |
1da177e4 LT |
2864 | next->active_mm = oldmm; |
2865 | atomic_inc(&oldmm->mm_count); | |
2866 | enter_lazy_tlb(oldmm, next); | |
2867 | } else | |
f98db601 | 2868 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2869 | |
31915ab4 | 2870 | if (!prev->mm) { |
1da177e4 | 2871 | prev->active_mm = NULL; |
1da177e4 LT |
2872 | rq->prev_mm = oldmm; |
2873 | } | |
3a5f5e48 IM |
2874 | /* |
2875 | * Since the runqueue lock will be released by the next | |
2876 | * task (which is an invalid locking op but in the case | |
2877 | * of the scheduler it's an obvious special-case), so we | |
2878 | * do an early lockdep release here: | |
2879 | */ | |
e7904a28 | 2880 | lockdep_unpin_lock(&rq->lock, cookie); |
8a25d5de | 2881 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2882 | |
2883 | /* Here we just switch the register state and the stack. */ | |
2884 | switch_to(prev, next, prev); | |
dd41f596 | 2885 | barrier(); |
dfa50b60 ON |
2886 | |
2887 | return finish_task_switch(prev); | |
1da177e4 LT |
2888 | } |
2889 | ||
2890 | /* | |
1c3e8264 | 2891 | * nr_running and nr_context_switches: |
1da177e4 LT |
2892 | * |
2893 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2894 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2895 | */ |
2896 | unsigned long nr_running(void) | |
2897 | { | |
2898 | unsigned long i, sum = 0; | |
2899 | ||
2900 | for_each_online_cpu(i) | |
2901 | sum += cpu_rq(i)->nr_running; | |
2902 | ||
2903 | return sum; | |
f711f609 | 2904 | } |
1da177e4 | 2905 | |
2ee507c4 TC |
2906 | /* |
2907 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2908 | * |
2909 | * Caution: this function does not check that the caller has disabled | |
2910 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2911 | * race. The caller is responsible to use it correctly, for example: | |
2912 | * | |
2913 | * - from a non-preemptable section (of course) | |
2914 | * | |
2915 | * - from a thread that is bound to a single CPU | |
2916 | * | |
2917 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2918 | */ |
2919 | bool single_task_running(void) | |
2920 | { | |
00cc1633 | 2921 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2922 | } |
2923 | EXPORT_SYMBOL(single_task_running); | |
2924 | ||
1da177e4 | 2925 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2926 | { |
cc94abfc SR |
2927 | int i; |
2928 | unsigned long long sum = 0; | |
46cb4b7c | 2929 | |
0a945022 | 2930 | for_each_possible_cpu(i) |
1da177e4 | 2931 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2932 | |
1da177e4 LT |
2933 | return sum; |
2934 | } | |
483b4ee6 | 2935 | |
1da177e4 LT |
2936 | unsigned long nr_iowait(void) |
2937 | { | |
2938 | unsigned long i, sum = 0; | |
483b4ee6 | 2939 | |
0a945022 | 2940 | for_each_possible_cpu(i) |
1da177e4 | 2941 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2942 | |
1da177e4 LT |
2943 | return sum; |
2944 | } | |
483b4ee6 | 2945 | |
8c215bd3 | 2946 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2947 | { |
8c215bd3 | 2948 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2949 | return atomic_read(&this->nr_iowait); |
2950 | } | |
46cb4b7c | 2951 | |
372ba8cb MG |
2952 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2953 | { | |
3289bdb4 PZ |
2954 | struct rq *rq = this_rq(); |
2955 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2956 | *load = rq->load.weight; | |
372ba8cb MG |
2957 | } |
2958 | ||
dd41f596 | 2959 | #ifdef CONFIG_SMP |
8a0be9ef | 2960 | |
46cb4b7c | 2961 | /* |
38022906 PZ |
2962 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2963 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2964 | */ |
38022906 | 2965 | void sched_exec(void) |
46cb4b7c | 2966 | { |
38022906 | 2967 | struct task_struct *p = current; |
1da177e4 | 2968 | unsigned long flags; |
0017d735 | 2969 | int dest_cpu; |
46cb4b7c | 2970 | |
8f42ced9 | 2971 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2972 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2973 | if (dest_cpu == smp_processor_id()) |
2974 | goto unlock; | |
38022906 | 2975 | |
8f42ced9 | 2976 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2977 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2978 | |
8f42ced9 PZ |
2979 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2980 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2981 | return; |
2982 | } | |
0017d735 | 2983 | unlock: |
8f42ced9 | 2984 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2985 | } |
dd41f596 | 2986 | |
1da177e4 LT |
2987 | #endif |
2988 | ||
1da177e4 | 2989 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2990 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2991 | |
2992 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2993 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2994 | |
6075620b GG |
2995 | /* |
2996 | * The function fair_sched_class.update_curr accesses the struct curr | |
2997 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
2998 | * we observe a high rate of cache misses in practice. | |
2999 | * Prefetching this data results in improved performance. | |
3000 | */ | |
3001 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3002 | { | |
3003 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3004 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3005 | #else | |
3006 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3007 | #endif | |
3008 | prefetch(curr); | |
3009 | prefetch(&curr->exec_start); | |
3010 | } | |
3011 | ||
c5f8d995 HS |
3012 | /* |
3013 | * Return accounted runtime for the task. | |
3014 | * In case the task is currently running, return the runtime plus current's | |
3015 | * pending runtime that have not been accounted yet. | |
3016 | */ | |
3017 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3018 | { | |
eb580751 | 3019 | struct rq_flags rf; |
c5f8d995 | 3020 | struct rq *rq; |
6e998916 | 3021 | u64 ns; |
c5f8d995 | 3022 | |
911b2898 PZ |
3023 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3024 | /* | |
3025 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
3026 | * So we have a optimization chance when the task's delta_exec is 0. | |
3027 | * Reading ->on_cpu is racy, but this is ok. | |
3028 | * | |
3029 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
3030 | * If we race with it entering cpu, unaccounted time is 0. This is | |
3031 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
3032 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3033 | * been accounted, so we're correct here as well. | |
911b2898 | 3034 | */ |
da0c1e65 | 3035 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3036 | return p->se.sum_exec_runtime; |
3037 | #endif | |
3038 | ||
eb580751 | 3039 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3040 | /* |
3041 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3042 | * project cycles that may never be accounted to this | |
3043 | * thread, breaking clock_gettime(). | |
3044 | */ | |
3045 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3046 | prefetch_curr_exec_start(p); |
6e998916 SG |
3047 | update_rq_clock(rq); |
3048 | p->sched_class->update_curr(rq); | |
3049 | } | |
3050 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3051 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3052 | |
3053 | return ns; | |
3054 | } | |
48f24c4d | 3055 | |
7835b98b CL |
3056 | /* |
3057 | * This function gets called by the timer code, with HZ frequency. | |
3058 | * We call it with interrupts disabled. | |
7835b98b CL |
3059 | */ |
3060 | void scheduler_tick(void) | |
3061 | { | |
7835b98b CL |
3062 | int cpu = smp_processor_id(); |
3063 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3064 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3065 | |
3066 | sched_clock_tick(); | |
dd41f596 | 3067 | |
05fa785c | 3068 | raw_spin_lock(&rq->lock); |
3e51f33f | 3069 | update_rq_clock(rq); |
fa85ae24 | 3070 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3071 | cpu_load_update_active(rq); |
3289bdb4 | 3072 | calc_global_load_tick(rq); |
05fa785c | 3073 | raw_spin_unlock(&rq->lock); |
7835b98b | 3074 | |
e9d2b064 | 3075 | perf_event_task_tick(); |
e220d2dc | 3076 | |
e418e1c2 | 3077 | #ifdef CONFIG_SMP |
6eb57e0d | 3078 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3079 | trigger_load_balance(rq); |
e418e1c2 | 3080 | #endif |
265f22a9 | 3081 | rq_last_tick_reset(rq); |
1da177e4 LT |
3082 | } |
3083 | ||
265f22a9 FW |
3084 | #ifdef CONFIG_NO_HZ_FULL |
3085 | /** | |
3086 | * scheduler_tick_max_deferment | |
3087 | * | |
3088 | * Keep at least one tick per second when a single | |
3089 | * active task is running because the scheduler doesn't | |
3090 | * yet completely support full dynticks environment. | |
3091 | * | |
3092 | * This makes sure that uptime, CFS vruntime, load | |
3093 | * balancing, etc... continue to move forward, even | |
3094 | * with a very low granularity. | |
e69f6186 YB |
3095 | * |
3096 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3097 | */ |
3098 | u64 scheduler_tick_max_deferment(void) | |
3099 | { | |
3100 | struct rq *rq = this_rq(); | |
316c1608 | 3101 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3102 | |
3103 | next = rq->last_sched_tick + HZ; | |
3104 | ||
3105 | if (time_before_eq(next, now)) | |
3106 | return 0; | |
3107 | ||
8fe8ff09 | 3108 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3109 | } |
265f22a9 | 3110 | #endif |
1da177e4 | 3111 | |
7e49fcce SR |
3112 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3113 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3114 | /* |
3115 | * If the value passed in is equal to the current preempt count | |
3116 | * then we just disabled preemption. Start timing the latency. | |
3117 | */ | |
3118 | static inline void preempt_latency_start(int val) | |
3119 | { | |
3120 | if (preempt_count() == val) { | |
3121 | unsigned long ip = get_lock_parent_ip(); | |
3122 | #ifdef CONFIG_DEBUG_PREEMPT | |
3123 | current->preempt_disable_ip = ip; | |
3124 | #endif | |
3125 | trace_preempt_off(CALLER_ADDR0, ip); | |
3126 | } | |
3127 | } | |
7e49fcce | 3128 | |
edafe3a5 | 3129 | void preempt_count_add(int val) |
1da177e4 | 3130 | { |
6cd8a4bb | 3131 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3132 | /* |
3133 | * Underflow? | |
3134 | */ | |
9a11b49a IM |
3135 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3136 | return; | |
6cd8a4bb | 3137 | #endif |
bdb43806 | 3138 | __preempt_count_add(val); |
6cd8a4bb | 3139 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3140 | /* |
3141 | * Spinlock count overflowing soon? | |
3142 | */ | |
33859f7f MOS |
3143 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3144 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3145 | #endif |
47252cfb | 3146 | preempt_latency_start(val); |
1da177e4 | 3147 | } |
bdb43806 | 3148 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3149 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3150 | |
47252cfb SR |
3151 | /* |
3152 | * If the value passed in equals to the current preempt count | |
3153 | * then we just enabled preemption. Stop timing the latency. | |
3154 | */ | |
3155 | static inline void preempt_latency_stop(int val) | |
3156 | { | |
3157 | if (preempt_count() == val) | |
3158 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3159 | } | |
3160 | ||
edafe3a5 | 3161 | void preempt_count_sub(int val) |
1da177e4 | 3162 | { |
6cd8a4bb | 3163 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3164 | /* |
3165 | * Underflow? | |
3166 | */ | |
01e3eb82 | 3167 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3168 | return; |
1da177e4 LT |
3169 | /* |
3170 | * Is the spinlock portion underflowing? | |
3171 | */ | |
9a11b49a IM |
3172 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3173 | !(preempt_count() & PREEMPT_MASK))) | |
3174 | return; | |
6cd8a4bb | 3175 | #endif |
9a11b49a | 3176 | |
47252cfb | 3177 | preempt_latency_stop(val); |
bdb43806 | 3178 | __preempt_count_sub(val); |
1da177e4 | 3179 | } |
bdb43806 | 3180 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3181 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3182 | |
47252cfb SR |
3183 | #else |
3184 | static inline void preempt_latency_start(int val) { } | |
3185 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3186 | #endif |
3187 | ||
3188 | /* | |
dd41f596 | 3189 | * Print scheduling while atomic bug: |
1da177e4 | 3190 | */ |
dd41f596 | 3191 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3192 | { |
d1c6d149 VN |
3193 | /* Save this before calling printk(), since that will clobber it */ |
3194 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3195 | ||
664dfa65 DJ |
3196 | if (oops_in_progress) |
3197 | return; | |
3198 | ||
3df0fc5b PZ |
3199 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3200 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3201 | |
dd41f596 | 3202 | debug_show_held_locks(prev); |
e21f5b15 | 3203 | print_modules(); |
dd41f596 IM |
3204 | if (irqs_disabled()) |
3205 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3206 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3207 | && in_atomic_preempt_off()) { | |
8f47b187 | 3208 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3209 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3210 | pr_cont("\n"); |
3211 | } | |
748c7201 DBO |
3212 | if (panic_on_warn) |
3213 | panic("scheduling while atomic\n"); | |
3214 | ||
6135fc1e | 3215 | dump_stack(); |
373d4d09 | 3216 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3217 | } |
1da177e4 | 3218 | |
dd41f596 IM |
3219 | /* |
3220 | * Various schedule()-time debugging checks and statistics: | |
3221 | */ | |
3222 | static inline void schedule_debug(struct task_struct *prev) | |
3223 | { | |
0d9e2632 | 3224 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3225 | if (task_stack_end_corrupted(prev)) |
3226 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3227 | #endif |
b99def8b | 3228 | |
1dc0fffc | 3229 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3230 | __schedule_bug(prev); |
1dc0fffc PZ |
3231 | preempt_count_set(PREEMPT_DISABLED); |
3232 | } | |
b3fbab05 | 3233 | rcu_sleep_check(); |
dd41f596 | 3234 | |
1da177e4 LT |
3235 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3236 | ||
ae92882e | 3237 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3238 | } |
3239 | ||
3240 | /* | |
3241 | * Pick up the highest-prio task: | |
3242 | */ | |
3243 | static inline struct task_struct * | |
e7904a28 | 3244 | pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie) |
dd41f596 | 3245 | { |
37e117c0 | 3246 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 3247 | struct task_struct *p; |
1da177e4 LT |
3248 | |
3249 | /* | |
dd41f596 IM |
3250 | * Optimization: we know that if all tasks are in |
3251 | * the fair class we can call that function directly: | |
1da177e4 | 3252 | */ |
37e117c0 | 3253 | if (likely(prev->sched_class == class && |
38033c37 | 3254 | rq->nr_running == rq->cfs.h_nr_running)) { |
e7904a28 | 3255 | p = fair_sched_class.pick_next_task(rq, prev, cookie); |
6ccdc84b PZ |
3256 | if (unlikely(p == RETRY_TASK)) |
3257 | goto again; | |
3258 | ||
3259 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3260 | if (unlikely(!p)) | |
e7904a28 | 3261 | p = idle_sched_class.pick_next_task(rq, prev, cookie); |
6ccdc84b PZ |
3262 | |
3263 | return p; | |
1da177e4 LT |
3264 | } |
3265 | ||
37e117c0 | 3266 | again: |
34f971f6 | 3267 | for_each_class(class) { |
e7904a28 | 3268 | p = class->pick_next_task(rq, prev, cookie); |
37e117c0 PZ |
3269 | if (p) { |
3270 | if (unlikely(p == RETRY_TASK)) | |
3271 | goto again; | |
dd41f596 | 3272 | return p; |
37e117c0 | 3273 | } |
dd41f596 | 3274 | } |
34f971f6 PZ |
3275 | |
3276 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3277 | } |
1da177e4 | 3278 | |
dd41f596 | 3279 | /* |
c259e01a | 3280 | * __schedule() is the main scheduler function. |
edde96ea PE |
3281 | * |
3282 | * The main means of driving the scheduler and thus entering this function are: | |
3283 | * | |
3284 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3285 | * | |
3286 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3287 | * paths. For example, see arch/x86/entry_64.S. | |
3288 | * | |
3289 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3290 | * interrupt handler scheduler_tick(). | |
3291 | * | |
3292 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3293 | * task to the run-queue and that's it. | |
3294 | * | |
3295 | * Now, if the new task added to the run-queue preempts the current | |
3296 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3297 | * called on the nearest possible occasion: | |
3298 | * | |
3299 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3300 | * | |
3301 | * - in syscall or exception context, at the next outmost | |
3302 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3303 | * spin_unlock()!) | |
3304 | * | |
3305 | * - in IRQ context, return from interrupt-handler to | |
3306 | * preemptible context | |
3307 | * | |
3308 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3309 | * then at the next: | |
3310 | * | |
3311 | * - cond_resched() call | |
3312 | * - explicit schedule() call | |
3313 | * - return from syscall or exception to user-space | |
3314 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3315 | * |
b30f0e3f | 3316 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3317 | */ |
499d7955 | 3318 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3319 | { |
3320 | struct task_struct *prev, *next; | |
67ca7bde | 3321 | unsigned long *switch_count; |
e7904a28 | 3322 | struct pin_cookie cookie; |
dd41f596 | 3323 | struct rq *rq; |
31656519 | 3324 | int cpu; |
dd41f596 | 3325 | |
dd41f596 IM |
3326 | cpu = smp_processor_id(); |
3327 | rq = cpu_rq(cpu); | |
dd41f596 | 3328 | prev = rq->curr; |
dd41f596 | 3329 | |
b99def8b PZ |
3330 | /* |
3331 | * do_exit() calls schedule() with preemption disabled as an exception; | |
3332 | * however we must fix that up, otherwise the next task will see an | |
3333 | * inconsistent (higher) preempt count. | |
3334 | * | |
3335 | * It also avoids the below schedule_debug() test from complaining | |
3336 | * about this. | |
3337 | */ | |
3338 | if (unlikely(prev->state == TASK_DEAD)) | |
3339 | preempt_enable_no_resched_notrace(); | |
3340 | ||
dd41f596 | 3341 | schedule_debug(prev); |
1da177e4 | 3342 | |
31656519 | 3343 | if (sched_feat(HRTICK)) |
f333fdc9 | 3344 | hrtick_clear(rq); |
8f4d37ec | 3345 | |
46a5d164 PM |
3346 | local_irq_disable(); |
3347 | rcu_note_context_switch(); | |
3348 | ||
e0acd0a6 ON |
3349 | /* |
3350 | * Make sure that signal_pending_state()->signal_pending() below | |
3351 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3352 | * done by the caller to avoid the race with signal_wake_up(). | |
3353 | */ | |
3354 | smp_mb__before_spinlock(); | |
46a5d164 | 3355 | raw_spin_lock(&rq->lock); |
e7904a28 | 3356 | cookie = lockdep_pin_lock(&rq->lock); |
1da177e4 | 3357 | |
9edfbfed PZ |
3358 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3359 | ||
246d86b5 | 3360 | switch_count = &prev->nivcsw; |
fc13aeba | 3361 | if (!preempt && prev->state) { |
21aa9af0 | 3362 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3363 | prev->state = TASK_RUNNING; |
21aa9af0 | 3364 | } else { |
2acca55e PZ |
3365 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3366 | prev->on_rq = 0; | |
3367 | ||
21aa9af0 | 3368 | /* |
2acca55e PZ |
3369 | * If a worker went to sleep, notify and ask workqueue |
3370 | * whether it wants to wake up a task to maintain | |
3371 | * concurrency. | |
21aa9af0 TH |
3372 | */ |
3373 | if (prev->flags & PF_WQ_WORKER) { | |
3374 | struct task_struct *to_wakeup; | |
3375 | ||
9b7f6597 | 3376 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3377 | if (to_wakeup) |
e7904a28 | 3378 | try_to_wake_up_local(to_wakeup, cookie); |
21aa9af0 | 3379 | } |
21aa9af0 | 3380 | } |
dd41f596 | 3381 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3382 | } |
3383 | ||
9edfbfed | 3384 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3385 | update_rq_clock(rq); |
3386 | ||
e7904a28 | 3387 | next = pick_next_task(rq, prev, cookie); |
f26f9aff | 3388 | clear_tsk_need_resched(prev); |
f27dde8d | 3389 | clear_preempt_need_resched(); |
9edfbfed | 3390 | rq->clock_skip_update = 0; |
1da177e4 | 3391 | |
1da177e4 | 3392 | if (likely(prev != next)) { |
1da177e4 LT |
3393 | rq->nr_switches++; |
3394 | rq->curr = next; | |
3395 | ++*switch_count; | |
3396 | ||
c73464b1 | 3397 | trace_sched_switch(preempt, prev, next); |
e7904a28 | 3398 | rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */ |
cbce1a68 | 3399 | } else { |
e7904a28 | 3400 | lockdep_unpin_lock(&rq->lock, cookie); |
05fa785c | 3401 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3402 | } |
1da177e4 | 3403 | |
e3fca9e7 | 3404 | balance_callback(rq); |
1da177e4 | 3405 | } |
8e05e96a | 3406 | STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */ |
c259e01a | 3407 | |
9c40cef2 TG |
3408 | static inline void sched_submit_work(struct task_struct *tsk) |
3409 | { | |
3c7d5184 | 3410 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3411 | return; |
3412 | /* | |
3413 | * If we are going to sleep and we have plugged IO queued, | |
3414 | * make sure to submit it to avoid deadlocks. | |
3415 | */ | |
3416 | if (blk_needs_flush_plug(tsk)) | |
3417 | blk_schedule_flush_plug(tsk); | |
3418 | } | |
3419 | ||
722a9f92 | 3420 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3421 | { |
9c40cef2 TG |
3422 | struct task_struct *tsk = current; |
3423 | ||
3424 | sched_submit_work(tsk); | |
bfd9b2b5 | 3425 | do { |
b30f0e3f | 3426 | preempt_disable(); |
fc13aeba | 3427 | __schedule(false); |
b30f0e3f | 3428 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3429 | } while (need_resched()); |
c259e01a | 3430 | } |
1da177e4 LT |
3431 | EXPORT_SYMBOL(schedule); |
3432 | ||
91d1aa43 | 3433 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3434 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3435 | { |
3436 | /* | |
3437 | * If we come here after a random call to set_need_resched(), | |
3438 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3439 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3440 | * we find a better solution. | |
7cc78f8f AL |
3441 | * |
3442 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3443 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3444 | * too frequently to make sense yet. |
20ab65e3 | 3445 | */ |
7cc78f8f | 3446 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3447 | schedule(); |
7cc78f8f | 3448 | exception_exit(prev_state); |
20ab65e3 FW |
3449 | } |
3450 | #endif | |
3451 | ||
c5491ea7 TG |
3452 | /** |
3453 | * schedule_preempt_disabled - called with preemption disabled | |
3454 | * | |
3455 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3456 | */ | |
3457 | void __sched schedule_preempt_disabled(void) | |
3458 | { | |
ba74c144 | 3459 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3460 | schedule(); |
3461 | preempt_disable(); | |
3462 | } | |
3463 | ||
06b1f808 | 3464 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3465 | { |
3466 | do { | |
47252cfb SR |
3467 | /* |
3468 | * Because the function tracer can trace preempt_count_sub() | |
3469 | * and it also uses preempt_enable/disable_notrace(), if | |
3470 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3471 | * by the function tracer will call this function again and | |
3472 | * cause infinite recursion. | |
3473 | * | |
3474 | * Preemption must be disabled here before the function | |
3475 | * tracer can trace. Break up preempt_disable() into two | |
3476 | * calls. One to disable preemption without fear of being | |
3477 | * traced. The other to still record the preemption latency, | |
3478 | * which can also be traced by the function tracer. | |
3479 | */ | |
499d7955 | 3480 | preempt_disable_notrace(); |
47252cfb | 3481 | preempt_latency_start(1); |
fc13aeba | 3482 | __schedule(true); |
47252cfb | 3483 | preempt_latency_stop(1); |
499d7955 | 3484 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3485 | |
3486 | /* | |
3487 | * Check again in case we missed a preemption opportunity | |
3488 | * between schedule and now. | |
3489 | */ | |
a18b5d01 FW |
3490 | } while (need_resched()); |
3491 | } | |
3492 | ||
1da177e4 LT |
3493 | #ifdef CONFIG_PREEMPT |
3494 | /* | |
2ed6e34f | 3495 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3496 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3497 | * occur there and call schedule directly. |
3498 | */ | |
722a9f92 | 3499 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3500 | { |
1da177e4 LT |
3501 | /* |
3502 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3503 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3504 | */ |
fbb00b56 | 3505 | if (likely(!preemptible())) |
1da177e4 LT |
3506 | return; |
3507 | ||
a18b5d01 | 3508 | preempt_schedule_common(); |
1da177e4 | 3509 | } |
376e2424 | 3510 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3511 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3512 | |
009f60e2 | 3513 | /** |
4eaca0a8 | 3514 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3515 | * |
3516 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3517 | * recursion and tracing preempt enabling caused by the tracing | |
3518 | * infrastructure itself. But as tracing can happen in areas coming | |
3519 | * from userspace or just about to enter userspace, a preempt enable | |
3520 | * can occur before user_exit() is called. This will cause the scheduler | |
3521 | * to be called when the system is still in usermode. | |
3522 | * | |
3523 | * To prevent this, the preempt_enable_notrace will use this function | |
3524 | * instead of preempt_schedule() to exit user context if needed before | |
3525 | * calling the scheduler. | |
3526 | */ | |
4eaca0a8 | 3527 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3528 | { |
3529 | enum ctx_state prev_ctx; | |
3530 | ||
3531 | if (likely(!preemptible())) | |
3532 | return; | |
3533 | ||
3534 | do { | |
47252cfb SR |
3535 | /* |
3536 | * Because the function tracer can trace preempt_count_sub() | |
3537 | * and it also uses preempt_enable/disable_notrace(), if | |
3538 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3539 | * by the function tracer will call this function again and | |
3540 | * cause infinite recursion. | |
3541 | * | |
3542 | * Preemption must be disabled here before the function | |
3543 | * tracer can trace. Break up preempt_disable() into two | |
3544 | * calls. One to disable preemption without fear of being | |
3545 | * traced. The other to still record the preemption latency, | |
3546 | * which can also be traced by the function tracer. | |
3547 | */ | |
3d8f74dd | 3548 | preempt_disable_notrace(); |
47252cfb | 3549 | preempt_latency_start(1); |
009f60e2 ON |
3550 | /* |
3551 | * Needs preempt disabled in case user_exit() is traced | |
3552 | * and the tracer calls preempt_enable_notrace() causing | |
3553 | * an infinite recursion. | |
3554 | */ | |
3555 | prev_ctx = exception_enter(); | |
fc13aeba | 3556 | __schedule(true); |
009f60e2 ON |
3557 | exception_exit(prev_ctx); |
3558 | ||
47252cfb | 3559 | preempt_latency_stop(1); |
3d8f74dd | 3560 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3561 | } while (need_resched()); |
3562 | } | |
4eaca0a8 | 3563 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3564 | |
32e475d7 | 3565 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3566 | |
3567 | /* | |
2ed6e34f | 3568 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3569 | * off of irq context. |
3570 | * Note, that this is called and return with irqs disabled. This will | |
3571 | * protect us against recursive calling from irq. | |
3572 | */ | |
722a9f92 | 3573 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3574 | { |
b22366cd | 3575 | enum ctx_state prev_state; |
6478d880 | 3576 | |
2ed6e34f | 3577 | /* Catch callers which need to be fixed */ |
f27dde8d | 3578 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3579 | |
b22366cd FW |
3580 | prev_state = exception_enter(); |
3581 | ||
3a5c359a | 3582 | do { |
3d8f74dd | 3583 | preempt_disable(); |
3a5c359a | 3584 | local_irq_enable(); |
fc13aeba | 3585 | __schedule(true); |
3a5c359a | 3586 | local_irq_disable(); |
3d8f74dd | 3587 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3588 | } while (need_resched()); |
b22366cd FW |
3589 | |
3590 | exception_exit(prev_state); | |
1da177e4 LT |
3591 | } |
3592 | ||
63859d4f | 3593 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3594 | void *key) |
1da177e4 | 3595 | { |
63859d4f | 3596 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3597 | } |
1da177e4 LT |
3598 | EXPORT_SYMBOL(default_wake_function); |
3599 | ||
b29739f9 IM |
3600 | #ifdef CONFIG_RT_MUTEXES |
3601 | ||
3602 | /* | |
3603 | * rt_mutex_setprio - set the current priority of a task | |
3604 | * @p: task | |
3605 | * @prio: prio value (kernel-internal form) | |
3606 | * | |
3607 | * This function changes the 'effective' priority of a task. It does | |
3608 | * not touch ->normal_prio like __setscheduler(). | |
3609 | * | |
c365c292 TG |
3610 | * Used by the rt_mutex code to implement priority inheritance |
3611 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3612 | */ |
36c8b586 | 3613 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3614 | { |
ff77e468 | 3615 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3616 | const struct sched_class *prev_class; |
eb580751 PZ |
3617 | struct rq_flags rf; |
3618 | struct rq *rq; | |
b29739f9 | 3619 | |
aab03e05 | 3620 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3621 | |
eb580751 | 3622 | rq = __task_rq_lock(p, &rf); |
b29739f9 | 3623 | |
1c4dd99b TG |
3624 | /* |
3625 | * Idle task boosting is a nono in general. There is one | |
3626 | * exception, when PREEMPT_RT and NOHZ is active: | |
3627 | * | |
3628 | * The idle task calls get_next_timer_interrupt() and holds | |
3629 | * the timer wheel base->lock on the CPU and another CPU wants | |
3630 | * to access the timer (probably to cancel it). We can safely | |
3631 | * ignore the boosting request, as the idle CPU runs this code | |
3632 | * with interrupts disabled and will complete the lock | |
3633 | * protected section without being interrupted. So there is no | |
3634 | * real need to boost. | |
3635 | */ | |
3636 | if (unlikely(p == rq->idle)) { | |
3637 | WARN_ON(p != rq->curr); | |
3638 | WARN_ON(p->pi_blocked_on); | |
3639 | goto out_unlock; | |
3640 | } | |
3641 | ||
a8027073 | 3642 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3643 | oldprio = p->prio; |
ff77e468 PZ |
3644 | |
3645 | if (oldprio == prio) | |
3646 | queue_flag &= ~DEQUEUE_MOVE; | |
3647 | ||
83ab0aa0 | 3648 | prev_class = p->sched_class; |
da0c1e65 | 3649 | queued = task_on_rq_queued(p); |
051a1d1a | 3650 | running = task_current(rq, p); |
da0c1e65 | 3651 | if (queued) |
ff77e468 | 3652 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3653 | if (running) |
f3cd1c4e | 3654 | put_prev_task(rq, p); |
dd41f596 | 3655 | |
2d3d891d DF |
3656 | /* |
3657 | * Boosting condition are: | |
3658 | * 1. -rt task is running and holds mutex A | |
3659 | * --> -dl task blocks on mutex A | |
3660 | * | |
3661 | * 2. -dl task is running and holds mutex A | |
3662 | * --> -dl task blocks on mutex A and could preempt the | |
3663 | * running task | |
3664 | */ | |
3665 | if (dl_prio(prio)) { | |
466af29b ON |
3666 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3667 | if (!dl_prio(p->normal_prio) || | |
3668 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3669 | p->dl.dl_boosted = 1; |
ff77e468 | 3670 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3671 | } else |
3672 | p->dl.dl_boosted = 0; | |
aab03e05 | 3673 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3674 | } else if (rt_prio(prio)) { |
3675 | if (dl_prio(oldprio)) | |
3676 | p->dl.dl_boosted = 0; | |
3677 | if (oldprio < prio) | |
ff77e468 | 3678 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3679 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3680 | } else { |
3681 | if (dl_prio(oldprio)) | |
3682 | p->dl.dl_boosted = 0; | |
746db944 BS |
3683 | if (rt_prio(oldprio)) |
3684 | p->rt.timeout = 0; | |
dd41f596 | 3685 | p->sched_class = &fair_sched_class; |
2d3d891d | 3686 | } |
dd41f596 | 3687 | |
b29739f9 IM |
3688 | p->prio = prio; |
3689 | ||
0e1f3483 HS |
3690 | if (running) |
3691 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3692 | if (queued) |
ff77e468 | 3693 | enqueue_task(rq, p, queue_flag); |
cb469845 | 3694 | |
da7a735e | 3695 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3696 | out_unlock: |
4c9a4bc8 | 3697 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 3698 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3699 | |
3700 | balance_callback(rq); | |
3701 | preempt_enable(); | |
b29739f9 | 3702 | } |
b29739f9 | 3703 | #endif |
d50dde5a | 3704 | |
36c8b586 | 3705 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3706 | { |
da0c1e65 | 3707 | int old_prio, delta, queued; |
eb580751 | 3708 | struct rq_flags rf; |
70b97a7f | 3709 | struct rq *rq; |
1da177e4 | 3710 | |
75e45d51 | 3711 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3712 | return; |
3713 | /* | |
3714 | * We have to be careful, if called from sys_setpriority(), | |
3715 | * the task might be in the middle of scheduling on another CPU. | |
3716 | */ | |
eb580751 | 3717 | rq = task_rq_lock(p, &rf); |
1da177e4 LT |
3718 | /* |
3719 | * The RT priorities are set via sched_setscheduler(), but we still | |
3720 | * allow the 'normal' nice value to be set - but as expected | |
3721 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3722 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3723 | */ |
aab03e05 | 3724 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3725 | p->static_prio = NICE_TO_PRIO(nice); |
3726 | goto out_unlock; | |
3727 | } | |
da0c1e65 KT |
3728 | queued = task_on_rq_queued(p); |
3729 | if (queued) | |
1de64443 | 3730 | dequeue_task(rq, p, DEQUEUE_SAVE); |
1da177e4 | 3731 | |
1da177e4 | 3732 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3733 | set_load_weight(p); |
b29739f9 IM |
3734 | old_prio = p->prio; |
3735 | p->prio = effective_prio(p); | |
3736 | delta = p->prio - old_prio; | |
1da177e4 | 3737 | |
da0c1e65 | 3738 | if (queued) { |
1de64443 | 3739 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3740 | /* |
d5f9f942 AM |
3741 | * If the task increased its priority or is running and |
3742 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3743 | */ |
d5f9f942 | 3744 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3745 | resched_curr(rq); |
1da177e4 LT |
3746 | } |
3747 | out_unlock: | |
eb580751 | 3748 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3749 | } |
1da177e4 LT |
3750 | EXPORT_SYMBOL(set_user_nice); |
3751 | ||
e43379f1 MM |
3752 | /* |
3753 | * can_nice - check if a task can reduce its nice value | |
3754 | * @p: task | |
3755 | * @nice: nice value | |
3756 | */ | |
36c8b586 | 3757 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3758 | { |
024f4747 | 3759 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3760 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3761 | |
78d7d407 | 3762 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3763 | capable(CAP_SYS_NICE)); |
3764 | } | |
3765 | ||
1da177e4 LT |
3766 | #ifdef __ARCH_WANT_SYS_NICE |
3767 | ||
3768 | /* | |
3769 | * sys_nice - change the priority of the current process. | |
3770 | * @increment: priority increment | |
3771 | * | |
3772 | * sys_setpriority is a more generic, but much slower function that | |
3773 | * does similar things. | |
3774 | */ | |
5add95d4 | 3775 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3776 | { |
48f24c4d | 3777 | long nice, retval; |
1da177e4 LT |
3778 | |
3779 | /* | |
3780 | * Setpriority might change our priority at the same moment. | |
3781 | * We don't have to worry. Conceptually one call occurs first | |
3782 | * and we have a single winner. | |
3783 | */ | |
a9467fa3 | 3784 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3785 | nice = task_nice(current) + increment; |
1da177e4 | 3786 | |
a9467fa3 | 3787 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3788 | if (increment < 0 && !can_nice(current, nice)) |
3789 | return -EPERM; | |
3790 | ||
1da177e4 LT |
3791 | retval = security_task_setnice(current, nice); |
3792 | if (retval) | |
3793 | return retval; | |
3794 | ||
3795 | set_user_nice(current, nice); | |
3796 | return 0; | |
3797 | } | |
3798 | ||
3799 | #endif | |
3800 | ||
3801 | /** | |
3802 | * task_prio - return the priority value of a given task. | |
3803 | * @p: the task in question. | |
3804 | * | |
e69f6186 | 3805 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3806 | * RT tasks are offset by -200. Normal tasks are centered |
3807 | * around 0, value goes from -16 to +15. | |
3808 | */ | |
36c8b586 | 3809 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3810 | { |
3811 | return p->prio - MAX_RT_PRIO; | |
3812 | } | |
3813 | ||
1da177e4 LT |
3814 | /** |
3815 | * idle_cpu - is a given cpu idle currently? | |
3816 | * @cpu: the processor in question. | |
e69f6186 YB |
3817 | * |
3818 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3819 | */ |
3820 | int idle_cpu(int cpu) | |
3821 | { | |
908a3283 TG |
3822 | struct rq *rq = cpu_rq(cpu); |
3823 | ||
3824 | if (rq->curr != rq->idle) | |
3825 | return 0; | |
3826 | ||
3827 | if (rq->nr_running) | |
3828 | return 0; | |
3829 | ||
3830 | #ifdef CONFIG_SMP | |
3831 | if (!llist_empty(&rq->wake_list)) | |
3832 | return 0; | |
3833 | #endif | |
3834 | ||
3835 | return 1; | |
1da177e4 LT |
3836 | } |
3837 | ||
1da177e4 LT |
3838 | /** |
3839 | * idle_task - return the idle task for a given cpu. | |
3840 | * @cpu: the processor in question. | |
e69f6186 YB |
3841 | * |
3842 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3843 | */ |
36c8b586 | 3844 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3845 | { |
3846 | return cpu_rq(cpu)->idle; | |
3847 | } | |
3848 | ||
3849 | /** | |
3850 | * find_process_by_pid - find a process with a matching PID value. | |
3851 | * @pid: the pid in question. | |
e69f6186 YB |
3852 | * |
3853 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3854 | */ |
a9957449 | 3855 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3856 | { |
228ebcbe | 3857 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3858 | } |
3859 | ||
aab03e05 DF |
3860 | /* |
3861 | * This function initializes the sched_dl_entity of a newly becoming | |
3862 | * SCHED_DEADLINE task. | |
3863 | * | |
3864 | * Only the static values are considered here, the actual runtime and the | |
3865 | * absolute deadline will be properly calculated when the task is enqueued | |
3866 | * for the first time with its new policy. | |
3867 | */ | |
3868 | static void | |
3869 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3870 | { | |
3871 | struct sched_dl_entity *dl_se = &p->dl; | |
3872 | ||
aab03e05 DF |
3873 | dl_se->dl_runtime = attr->sched_runtime; |
3874 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3875 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3876 | dl_se->flags = attr->sched_flags; |
332ac17e | 3877 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3878 | |
3879 | /* | |
3880 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3881 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3882 | * | |
3883 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3884 | * point. This would include retaining the task_struct until that time | |
3885 | * and change dl_overflow() to not immediately decrement the current | |
3886 | * amount. | |
3887 | * | |
3888 | * Instead we retain the current runtime/deadline and let the new | |
3889 | * parameters take effect after the current reservation period lapses. | |
3890 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3891 | * before the current scheduling deadline. | |
3892 | * | |
3893 | * We can still have temporary overloads because we do not delay the | |
3894 | * change in bandwidth until that time; so admission control is | |
3895 | * not on the safe side. It does however guarantee tasks will never | |
3896 | * consume more than promised. | |
3897 | */ | |
aab03e05 DF |
3898 | } |
3899 | ||
c13db6b1 SR |
3900 | /* |
3901 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3902 | * it calls know not to change it. | |
3903 | */ | |
3904 | #define SETPARAM_POLICY -1 | |
3905 | ||
c365c292 TG |
3906 | static void __setscheduler_params(struct task_struct *p, |
3907 | const struct sched_attr *attr) | |
1da177e4 | 3908 | { |
d50dde5a DF |
3909 | int policy = attr->sched_policy; |
3910 | ||
c13db6b1 | 3911 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3912 | policy = p->policy; |
3913 | ||
1da177e4 | 3914 | p->policy = policy; |
d50dde5a | 3915 | |
aab03e05 DF |
3916 | if (dl_policy(policy)) |
3917 | __setparam_dl(p, attr); | |
39fd8fd2 | 3918 | else if (fair_policy(policy)) |
d50dde5a DF |
3919 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3920 | ||
39fd8fd2 PZ |
3921 | /* |
3922 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3923 | * !rt_policy. Always setting this ensures that things like | |
3924 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3925 | */ | |
3926 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3927 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3928 | set_load_weight(p); |
3929 | } | |
39fd8fd2 | 3930 | |
c365c292 TG |
3931 | /* Actually do priority change: must hold pi & rq lock. */ |
3932 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3933 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3934 | { |
3935 | __setscheduler_params(p, attr); | |
d50dde5a | 3936 | |
383afd09 | 3937 | /* |
0782e63b TG |
3938 | * Keep a potential priority boosting if called from |
3939 | * sched_setscheduler(). | |
383afd09 | 3940 | */ |
0782e63b TG |
3941 | if (keep_boost) |
3942 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3943 | else | |
3944 | p->prio = normal_prio(p); | |
383afd09 | 3945 | |
aab03e05 DF |
3946 | if (dl_prio(p->prio)) |
3947 | p->sched_class = &dl_sched_class; | |
3948 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3949 | p->sched_class = &rt_sched_class; |
3950 | else | |
3951 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3952 | } |
aab03e05 DF |
3953 | |
3954 | static void | |
3955 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3956 | { | |
3957 | struct sched_dl_entity *dl_se = &p->dl; | |
3958 | ||
3959 | attr->sched_priority = p->rt_priority; | |
3960 | attr->sched_runtime = dl_se->dl_runtime; | |
3961 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3962 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3963 | attr->sched_flags = dl_se->flags; |
3964 | } | |
3965 | ||
3966 | /* | |
3967 | * This function validates the new parameters of a -deadline task. | |
3968 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3969 | * than the runtime, as well as the period of being zero or |
332ac17e | 3970 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3971 | * user parameters are above the internal resolution of 1us (we |
3972 | * check sched_runtime only since it is always the smaller one) and | |
3973 | * below 2^63 ns (we have to check both sched_deadline and | |
3974 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3975 | */ |
3976 | static bool | |
3977 | __checkparam_dl(const struct sched_attr *attr) | |
3978 | { | |
b0827819 JL |
3979 | /* deadline != 0 */ |
3980 | if (attr->sched_deadline == 0) | |
3981 | return false; | |
3982 | ||
3983 | /* | |
3984 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3985 | * that big. | |
3986 | */ | |
3987 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3988 | return false; | |
3989 | ||
3990 | /* | |
3991 | * Since we use the MSB for wrap-around and sign issues, make | |
3992 | * sure it's not set (mind that period can be equal to zero). | |
3993 | */ | |
3994 | if (attr->sched_deadline & (1ULL << 63) || | |
3995 | attr->sched_period & (1ULL << 63)) | |
3996 | return false; | |
3997 | ||
3998 | /* runtime <= deadline <= period (if period != 0) */ | |
3999 | if ((attr->sched_period != 0 && | |
4000 | attr->sched_period < attr->sched_deadline) || | |
4001 | attr->sched_deadline < attr->sched_runtime) | |
4002 | return false; | |
4003 | ||
4004 | return true; | |
aab03e05 DF |
4005 | } |
4006 | ||
c69e8d9c DH |
4007 | /* |
4008 | * check the target process has a UID that matches the current process's | |
4009 | */ | |
4010 | static bool check_same_owner(struct task_struct *p) | |
4011 | { | |
4012 | const struct cred *cred = current_cred(), *pcred; | |
4013 | bool match; | |
4014 | ||
4015 | rcu_read_lock(); | |
4016 | pcred = __task_cred(p); | |
9c806aa0 EB |
4017 | match = (uid_eq(cred->euid, pcred->euid) || |
4018 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4019 | rcu_read_unlock(); |
4020 | return match; | |
4021 | } | |
4022 | ||
75381608 WL |
4023 | static bool dl_param_changed(struct task_struct *p, |
4024 | const struct sched_attr *attr) | |
4025 | { | |
4026 | struct sched_dl_entity *dl_se = &p->dl; | |
4027 | ||
4028 | if (dl_se->dl_runtime != attr->sched_runtime || | |
4029 | dl_se->dl_deadline != attr->sched_deadline || | |
4030 | dl_se->dl_period != attr->sched_period || | |
4031 | dl_se->flags != attr->sched_flags) | |
4032 | return true; | |
4033 | ||
4034 | return false; | |
4035 | } | |
4036 | ||
d50dde5a DF |
4037 | static int __sched_setscheduler(struct task_struct *p, |
4038 | const struct sched_attr *attr, | |
dbc7f069 | 4039 | bool user, bool pi) |
1da177e4 | 4040 | { |
383afd09 SR |
4041 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4042 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4043 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4044 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4045 | const struct sched_class *prev_class; |
eb580751 | 4046 | struct rq_flags rf; |
ca94c442 | 4047 | int reset_on_fork; |
ff77e468 | 4048 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 4049 | struct rq *rq; |
1da177e4 | 4050 | |
66e5393a SR |
4051 | /* may grab non-irq protected spin_locks */ |
4052 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4053 | recheck: |
4054 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4055 | if (policy < 0) { |
4056 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4057 | policy = oldpolicy = p->policy; |
ca94c442 | 4058 | } else { |
7479f3c9 | 4059 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4060 | |
20f9cd2a | 4061 | if (!valid_policy(policy)) |
ca94c442 LP |
4062 | return -EINVAL; |
4063 | } | |
4064 | ||
7479f3c9 PZ |
4065 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
4066 | return -EINVAL; | |
4067 | ||
1da177e4 LT |
4068 | /* |
4069 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4070 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4071 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4072 | */ |
0bb040a4 | 4073 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4074 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4075 | return -EINVAL; |
aab03e05 DF |
4076 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4077 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4078 | return -EINVAL; |
4079 | ||
37e4ab3f OC |
4080 | /* |
4081 | * Allow unprivileged RT tasks to decrease priority: | |
4082 | */ | |
961ccddd | 4083 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4084 | if (fair_policy(policy)) { |
d0ea0268 | 4085 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4086 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4087 | return -EPERM; |
4088 | } | |
4089 | ||
e05606d3 | 4090 | if (rt_policy(policy)) { |
a44702e8 ON |
4091 | unsigned long rlim_rtprio = |
4092 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4093 | |
4094 | /* can't set/change the rt policy */ | |
4095 | if (policy != p->policy && !rlim_rtprio) | |
4096 | return -EPERM; | |
4097 | ||
4098 | /* can't increase priority */ | |
d50dde5a DF |
4099 | if (attr->sched_priority > p->rt_priority && |
4100 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4101 | return -EPERM; |
4102 | } | |
c02aa73b | 4103 | |
d44753b8 JL |
4104 | /* |
4105 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4106 | * (safest behavior); in the future we would like to allow | |
4107 | * unprivileged DL tasks to increase their relative deadline | |
4108 | * or reduce their runtime (both ways reducing utilization) | |
4109 | */ | |
4110 | if (dl_policy(policy)) | |
4111 | return -EPERM; | |
4112 | ||
dd41f596 | 4113 | /* |
c02aa73b DH |
4114 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4115 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4116 | */ |
20f9cd2a | 4117 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4118 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4119 | return -EPERM; |
4120 | } | |
5fe1d75f | 4121 | |
37e4ab3f | 4122 | /* can't change other user's priorities */ |
c69e8d9c | 4123 | if (!check_same_owner(p)) |
37e4ab3f | 4124 | return -EPERM; |
ca94c442 LP |
4125 | |
4126 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4127 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4128 | return -EPERM; | |
37e4ab3f | 4129 | } |
1da177e4 | 4130 | |
725aad24 | 4131 | if (user) { |
b0ae1981 | 4132 | retval = security_task_setscheduler(p); |
725aad24 JF |
4133 | if (retval) |
4134 | return retval; | |
4135 | } | |
4136 | ||
b29739f9 IM |
4137 | /* |
4138 | * make sure no PI-waiters arrive (or leave) while we are | |
4139 | * changing the priority of the task: | |
0122ec5b | 4140 | * |
25985edc | 4141 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4142 | * runqueue lock must be held. |
4143 | */ | |
eb580751 | 4144 | rq = task_rq_lock(p, &rf); |
dc61b1d6 | 4145 | |
34f971f6 PZ |
4146 | /* |
4147 | * Changing the policy of the stop threads its a very bad idea | |
4148 | */ | |
4149 | if (p == rq->stop) { | |
eb580751 | 4150 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4151 | return -EINVAL; |
4152 | } | |
4153 | ||
a51e9198 | 4154 | /* |
d6b1e911 TG |
4155 | * If not changing anything there's no need to proceed further, |
4156 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4157 | */ |
d50dde5a | 4158 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4159 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4160 | goto change; |
4161 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4162 | goto change; | |
75381608 | 4163 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4164 | goto change; |
d50dde5a | 4165 | |
d6b1e911 | 4166 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4167 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4168 | return 0; |
4169 | } | |
d50dde5a | 4170 | change: |
a51e9198 | 4171 | |
dc61b1d6 | 4172 | if (user) { |
332ac17e | 4173 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4174 | /* |
4175 | * Do not allow realtime tasks into groups that have no runtime | |
4176 | * assigned. | |
4177 | */ | |
4178 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4179 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4180 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4181 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4182 | return -EPERM; |
4183 | } | |
dc61b1d6 | 4184 | #endif |
332ac17e DF |
4185 | #ifdef CONFIG_SMP |
4186 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4187 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4188 | |
4189 | /* | |
4190 | * Don't allow tasks with an affinity mask smaller than | |
4191 | * the entire root_domain to become SCHED_DEADLINE. We | |
4192 | * will also fail if there's no bandwidth available. | |
4193 | */ | |
e4099a5e PZ |
4194 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4195 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4196 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4197 | return -EPERM; |
4198 | } | |
4199 | } | |
4200 | #endif | |
4201 | } | |
dc61b1d6 | 4202 | |
1da177e4 LT |
4203 | /* recheck policy now with rq lock held */ |
4204 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4205 | policy = oldpolicy = -1; | |
eb580751 | 4206 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4207 | goto recheck; |
4208 | } | |
332ac17e DF |
4209 | |
4210 | /* | |
4211 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4212 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4213 | * is available. | |
4214 | */ | |
e4099a5e | 4215 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4216 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4217 | return -EBUSY; |
4218 | } | |
4219 | ||
c365c292 TG |
4220 | p->sched_reset_on_fork = reset_on_fork; |
4221 | oldprio = p->prio; | |
4222 | ||
dbc7f069 PZ |
4223 | if (pi) { |
4224 | /* | |
4225 | * Take priority boosted tasks into account. If the new | |
4226 | * effective priority is unchanged, we just store the new | |
4227 | * normal parameters and do not touch the scheduler class and | |
4228 | * the runqueue. This will be done when the task deboost | |
4229 | * itself. | |
4230 | */ | |
4231 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4232 | if (new_effective_prio == oldprio) |
4233 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4234 | } |
4235 | ||
da0c1e65 | 4236 | queued = task_on_rq_queued(p); |
051a1d1a | 4237 | running = task_current(rq, p); |
da0c1e65 | 4238 | if (queued) |
ff77e468 | 4239 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4240 | if (running) |
f3cd1c4e | 4241 | put_prev_task(rq, p); |
f6b53205 | 4242 | |
83ab0aa0 | 4243 | prev_class = p->sched_class; |
dbc7f069 | 4244 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4245 | |
0e1f3483 HS |
4246 | if (running) |
4247 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4248 | if (queued) { |
81a44c54 TG |
4249 | /* |
4250 | * We enqueue to tail when the priority of a task is | |
4251 | * increased (user space view). | |
4252 | */ | |
ff77e468 PZ |
4253 | if (oldprio < p->prio) |
4254 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4255 | |
ff77e468 | 4256 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4257 | } |
cb469845 | 4258 | |
da7a735e | 4259 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4260 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 4261 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4262 | |
dbc7f069 PZ |
4263 | if (pi) |
4264 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4265 | |
4c9a4bc8 PZ |
4266 | /* |
4267 | * Run balance callbacks after we've adjusted the PI chain. | |
4268 | */ | |
4269 | balance_callback(rq); | |
4270 | preempt_enable(); | |
95e02ca9 | 4271 | |
1da177e4 LT |
4272 | return 0; |
4273 | } | |
961ccddd | 4274 | |
7479f3c9 PZ |
4275 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4276 | const struct sched_param *param, bool check) | |
4277 | { | |
4278 | struct sched_attr attr = { | |
4279 | .sched_policy = policy, | |
4280 | .sched_priority = param->sched_priority, | |
4281 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4282 | }; | |
4283 | ||
c13db6b1 SR |
4284 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4285 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4286 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4287 | policy &= ~SCHED_RESET_ON_FORK; | |
4288 | attr.sched_policy = policy; | |
4289 | } | |
4290 | ||
dbc7f069 | 4291 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4292 | } |
961ccddd RR |
4293 | /** |
4294 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4295 | * @p: the task in question. | |
4296 | * @policy: new policy. | |
4297 | * @param: structure containing the new RT priority. | |
4298 | * | |
e69f6186 YB |
4299 | * Return: 0 on success. An error code otherwise. |
4300 | * | |
961ccddd RR |
4301 | * NOTE that the task may be already dead. |
4302 | */ | |
4303 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4304 | const struct sched_param *param) |
961ccddd | 4305 | { |
7479f3c9 | 4306 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4307 | } |
1da177e4 LT |
4308 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4309 | ||
d50dde5a DF |
4310 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4311 | { | |
dbc7f069 | 4312 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4313 | } |
4314 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4315 | ||
961ccddd RR |
4316 | /** |
4317 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4318 | * @p: the task in question. | |
4319 | * @policy: new policy. | |
4320 | * @param: structure containing the new RT priority. | |
4321 | * | |
4322 | * Just like sched_setscheduler, only don't bother checking if the | |
4323 | * current context has permission. For example, this is needed in | |
4324 | * stop_machine(): we create temporary high priority worker threads, | |
4325 | * but our caller might not have that capability. | |
e69f6186 YB |
4326 | * |
4327 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4328 | */ |
4329 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4330 | const struct sched_param *param) |
961ccddd | 4331 | { |
7479f3c9 | 4332 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4333 | } |
84778472 | 4334 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4335 | |
95cdf3b7 IM |
4336 | static int |
4337 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4338 | { |
1da177e4 LT |
4339 | struct sched_param lparam; |
4340 | struct task_struct *p; | |
36c8b586 | 4341 | int retval; |
1da177e4 LT |
4342 | |
4343 | if (!param || pid < 0) | |
4344 | return -EINVAL; | |
4345 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4346 | return -EFAULT; | |
5fe1d75f ON |
4347 | |
4348 | rcu_read_lock(); | |
4349 | retval = -ESRCH; | |
1da177e4 | 4350 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4351 | if (p != NULL) |
4352 | retval = sched_setscheduler(p, policy, &lparam); | |
4353 | rcu_read_unlock(); | |
36c8b586 | 4354 | |
1da177e4 LT |
4355 | return retval; |
4356 | } | |
4357 | ||
d50dde5a DF |
4358 | /* |
4359 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4360 | */ | |
4361 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4362 | struct sched_attr *attr) | |
4363 | { | |
4364 | u32 size; | |
4365 | int ret; | |
4366 | ||
4367 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4368 | return -EFAULT; | |
4369 | ||
4370 | /* | |
4371 | * zero the full structure, so that a short copy will be nice. | |
4372 | */ | |
4373 | memset(attr, 0, sizeof(*attr)); | |
4374 | ||
4375 | ret = get_user(size, &uattr->size); | |
4376 | if (ret) | |
4377 | return ret; | |
4378 | ||
4379 | if (size > PAGE_SIZE) /* silly large */ | |
4380 | goto err_size; | |
4381 | ||
4382 | if (!size) /* abi compat */ | |
4383 | size = SCHED_ATTR_SIZE_VER0; | |
4384 | ||
4385 | if (size < SCHED_ATTR_SIZE_VER0) | |
4386 | goto err_size; | |
4387 | ||
4388 | /* | |
4389 | * If we're handed a bigger struct than we know of, | |
4390 | * ensure all the unknown bits are 0 - i.e. new | |
4391 | * user-space does not rely on any kernel feature | |
4392 | * extensions we dont know about yet. | |
4393 | */ | |
4394 | if (size > sizeof(*attr)) { | |
4395 | unsigned char __user *addr; | |
4396 | unsigned char __user *end; | |
4397 | unsigned char val; | |
4398 | ||
4399 | addr = (void __user *)uattr + sizeof(*attr); | |
4400 | end = (void __user *)uattr + size; | |
4401 | ||
4402 | for (; addr < end; addr++) { | |
4403 | ret = get_user(val, addr); | |
4404 | if (ret) | |
4405 | return ret; | |
4406 | if (val) | |
4407 | goto err_size; | |
4408 | } | |
4409 | size = sizeof(*attr); | |
4410 | } | |
4411 | ||
4412 | ret = copy_from_user(attr, uattr, size); | |
4413 | if (ret) | |
4414 | return -EFAULT; | |
4415 | ||
4416 | /* | |
4417 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4418 | * to be strict and return an error on out-of-bounds values? | |
4419 | */ | |
75e45d51 | 4420 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4421 | |
e78c7bca | 4422 | return 0; |
d50dde5a DF |
4423 | |
4424 | err_size: | |
4425 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4426 | return -E2BIG; |
d50dde5a DF |
4427 | } |
4428 | ||
1da177e4 LT |
4429 | /** |
4430 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4431 | * @pid: the pid in question. | |
4432 | * @policy: new policy. | |
4433 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4434 | * |
4435 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4436 | */ |
5add95d4 HC |
4437 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4438 | struct sched_param __user *, param) | |
1da177e4 | 4439 | { |
c21761f1 JB |
4440 | /* negative values for policy are not valid */ |
4441 | if (policy < 0) | |
4442 | return -EINVAL; | |
4443 | ||
1da177e4 LT |
4444 | return do_sched_setscheduler(pid, policy, param); |
4445 | } | |
4446 | ||
4447 | /** | |
4448 | * sys_sched_setparam - set/change the RT priority of a thread | |
4449 | * @pid: the pid in question. | |
4450 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4451 | * |
4452 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4453 | */ |
5add95d4 | 4454 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4455 | { |
c13db6b1 | 4456 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4457 | } |
4458 | ||
d50dde5a DF |
4459 | /** |
4460 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4461 | * @pid: the pid in question. | |
5778fccf | 4462 | * @uattr: structure containing the extended parameters. |
db66d756 | 4463 | * @flags: for future extension. |
d50dde5a | 4464 | */ |
6d35ab48 PZ |
4465 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4466 | unsigned int, flags) | |
d50dde5a DF |
4467 | { |
4468 | struct sched_attr attr; | |
4469 | struct task_struct *p; | |
4470 | int retval; | |
4471 | ||
6d35ab48 | 4472 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4473 | return -EINVAL; |
4474 | ||
143cf23d MK |
4475 | retval = sched_copy_attr(uattr, &attr); |
4476 | if (retval) | |
4477 | return retval; | |
d50dde5a | 4478 | |
b14ed2c2 | 4479 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4480 | return -EINVAL; |
d50dde5a DF |
4481 | |
4482 | rcu_read_lock(); | |
4483 | retval = -ESRCH; | |
4484 | p = find_process_by_pid(pid); | |
4485 | if (p != NULL) | |
4486 | retval = sched_setattr(p, &attr); | |
4487 | rcu_read_unlock(); | |
4488 | ||
4489 | return retval; | |
4490 | } | |
4491 | ||
1da177e4 LT |
4492 | /** |
4493 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4494 | * @pid: the pid in question. | |
e69f6186 YB |
4495 | * |
4496 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4497 | * code. | |
1da177e4 | 4498 | */ |
5add95d4 | 4499 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4500 | { |
36c8b586 | 4501 | struct task_struct *p; |
3a5c359a | 4502 | int retval; |
1da177e4 LT |
4503 | |
4504 | if (pid < 0) | |
3a5c359a | 4505 | return -EINVAL; |
1da177e4 LT |
4506 | |
4507 | retval = -ESRCH; | |
5fe85be0 | 4508 | rcu_read_lock(); |
1da177e4 LT |
4509 | p = find_process_by_pid(pid); |
4510 | if (p) { | |
4511 | retval = security_task_getscheduler(p); | |
4512 | if (!retval) | |
ca94c442 LP |
4513 | retval = p->policy |
4514 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4515 | } |
5fe85be0 | 4516 | rcu_read_unlock(); |
1da177e4 LT |
4517 | return retval; |
4518 | } | |
4519 | ||
4520 | /** | |
ca94c442 | 4521 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4522 | * @pid: the pid in question. |
4523 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4524 | * |
4525 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4526 | * code. | |
1da177e4 | 4527 | */ |
5add95d4 | 4528 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4529 | { |
ce5f7f82 | 4530 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4531 | struct task_struct *p; |
3a5c359a | 4532 | int retval; |
1da177e4 LT |
4533 | |
4534 | if (!param || pid < 0) | |
3a5c359a | 4535 | return -EINVAL; |
1da177e4 | 4536 | |
5fe85be0 | 4537 | rcu_read_lock(); |
1da177e4 LT |
4538 | p = find_process_by_pid(pid); |
4539 | retval = -ESRCH; | |
4540 | if (!p) | |
4541 | goto out_unlock; | |
4542 | ||
4543 | retval = security_task_getscheduler(p); | |
4544 | if (retval) | |
4545 | goto out_unlock; | |
4546 | ||
ce5f7f82 PZ |
4547 | if (task_has_rt_policy(p)) |
4548 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4549 | rcu_read_unlock(); |
1da177e4 LT |
4550 | |
4551 | /* | |
4552 | * This one might sleep, we cannot do it with a spinlock held ... | |
4553 | */ | |
4554 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4555 | ||
1da177e4 LT |
4556 | return retval; |
4557 | ||
4558 | out_unlock: | |
5fe85be0 | 4559 | rcu_read_unlock(); |
1da177e4 LT |
4560 | return retval; |
4561 | } | |
4562 | ||
d50dde5a DF |
4563 | static int sched_read_attr(struct sched_attr __user *uattr, |
4564 | struct sched_attr *attr, | |
4565 | unsigned int usize) | |
4566 | { | |
4567 | int ret; | |
4568 | ||
4569 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4570 | return -EFAULT; | |
4571 | ||
4572 | /* | |
4573 | * If we're handed a smaller struct than we know of, | |
4574 | * ensure all the unknown bits are 0 - i.e. old | |
4575 | * user-space does not get uncomplete information. | |
4576 | */ | |
4577 | if (usize < sizeof(*attr)) { | |
4578 | unsigned char *addr; | |
4579 | unsigned char *end; | |
4580 | ||
4581 | addr = (void *)attr + usize; | |
4582 | end = (void *)attr + sizeof(*attr); | |
4583 | ||
4584 | for (; addr < end; addr++) { | |
4585 | if (*addr) | |
22400674 | 4586 | return -EFBIG; |
d50dde5a DF |
4587 | } |
4588 | ||
4589 | attr->size = usize; | |
4590 | } | |
4591 | ||
4efbc454 | 4592 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4593 | if (ret) |
4594 | return -EFAULT; | |
4595 | ||
22400674 | 4596 | return 0; |
d50dde5a DF |
4597 | } |
4598 | ||
4599 | /** | |
aab03e05 | 4600 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4601 | * @pid: the pid in question. |
5778fccf | 4602 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4603 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4604 | * @flags: for future extension. |
d50dde5a | 4605 | */ |
6d35ab48 PZ |
4606 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4607 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4608 | { |
4609 | struct sched_attr attr = { | |
4610 | .size = sizeof(struct sched_attr), | |
4611 | }; | |
4612 | struct task_struct *p; | |
4613 | int retval; | |
4614 | ||
4615 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4616 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4617 | return -EINVAL; |
4618 | ||
4619 | rcu_read_lock(); | |
4620 | p = find_process_by_pid(pid); | |
4621 | retval = -ESRCH; | |
4622 | if (!p) | |
4623 | goto out_unlock; | |
4624 | ||
4625 | retval = security_task_getscheduler(p); | |
4626 | if (retval) | |
4627 | goto out_unlock; | |
4628 | ||
4629 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4630 | if (p->sched_reset_on_fork) |
4631 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4632 | if (task_has_dl_policy(p)) |
4633 | __getparam_dl(p, &attr); | |
4634 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4635 | attr.sched_priority = p->rt_priority; |
4636 | else | |
d0ea0268 | 4637 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4638 | |
4639 | rcu_read_unlock(); | |
4640 | ||
4641 | retval = sched_read_attr(uattr, &attr, size); | |
4642 | return retval; | |
4643 | ||
4644 | out_unlock: | |
4645 | rcu_read_unlock(); | |
4646 | return retval; | |
4647 | } | |
4648 | ||
96f874e2 | 4649 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4650 | { |
5a16f3d3 | 4651 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4652 | struct task_struct *p; |
4653 | int retval; | |
1da177e4 | 4654 | |
23f5d142 | 4655 | rcu_read_lock(); |
1da177e4 LT |
4656 | |
4657 | p = find_process_by_pid(pid); | |
4658 | if (!p) { | |
23f5d142 | 4659 | rcu_read_unlock(); |
1da177e4 LT |
4660 | return -ESRCH; |
4661 | } | |
4662 | ||
23f5d142 | 4663 | /* Prevent p going away */ |
1da177e4 | 4664 | get_task_struct(p); |
23f5d142 | 4665 | rcu_read_unlock(); |
1da177e4 | 4666 | |
14a40ffc TH |
4667 | if (p->flags & PF_NO_SETAFFINITY) { |
4668 | retval = -EINVAL; | |
4669 | goto out_put_task; | |
4670 | } | |
5a16f3d3 RR |
4671 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4672 | retval = -ENOMEM; | |
4673 | goto out_put_task; | |
4674 | } | |
4675 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4676 | retval = -ENOMEM; | |
4677 | goto out_free_cpus_allowed; | |
4678 | } | |
1da177e4 | 4679 | retval = -EPERM; |
4c44aaaf EB |
4680 | if (!check_same_owner(p)) { |
4681 | rcu_read_lock(); | |
4682 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4683 | rcu_read_unlock(); | |
16303ab2 | 4684 | goto out_free_new_mask; |
4c44aaaf EB |
4685 | } |
4686 | rcu_read_unlock(); | |
4687 | } | |
1da177e4 | 4688 | |
b0ae1981 | 4689 | retval = security_task_setscheduler(p); |
e7834f8f | 4690 | if (retval) |
16303ab2 | 4691 | goto out_free_new_mask; |
e7834f8f | 4692 | |
e4099a5e PZ |
4693 | |
4694 | cpuset_cpus_allowed(p, cpus_allowed); | |
4695 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4696 | ||
332ac17e DF |
4697 | /* |
4698 | * Since bandwidth control happens on root_domain basis, | |
4699 | * if admission test is enabled, we only admit -deadline | |
4700 | * tasks allowed to run on all the CPUs in the task's | |
4701 | * root_domain. | |
4702 | */ | |
4703 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4704 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4705 | rcu_read_lock(); | |
4706 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4707 | retval = -EBUSY; |
f1e3a093 | 4708 | rcu_read_unlock(); |
16303ab2 | 4709 | goto out_free_new_mask; |
332ac17e | 4710 | } |
f1e3a093 | 4711 | rcu_read_unlock(); |
332ac17e DF |
4712 | } |
4713 | #endif | |
49246274 | 4714 | again: |
25834c73 | 4715 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4716 | |
8707d8b8 | 4717 | if (!retval) { |
5a16f3d3 RR |
4718 | cpuset_cpus_allowed(p, cpus_allowed); |
4719 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4720 | /* |
4721 | * We must have raced with a concurrent cpuset | |
4722 | * update. Just reset the cpus_allowed to the | |
4723 | * cpuset's cpus_allowed | |
4724 | */ | |
5a16f3d3 | 4725 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4726 | goto again; |
4727 | } | |
4728 | } | |
16303ab2 | 4729 | out_free_new_mask: |
5a16f3d3 RR |
4730 | free_cpumask_var(new_mask); |
4731 | out_free_cpus_allowed: | |
4732 | free_cpumask_var(cpus_allowed); | |
4733 | out_put_task: | |
1da177e4 | 4734 | put_task_struct(p); |
1da177e4 LT |
4735 | return retval; |
4736 | } | |
4737 | ||
4738 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4739 | struct cpumask *new_mask) |
1da177e4 | 4740 | { |
96f874e2 RR |
4741 | if (len < cpumask_size()) |
4742 | cpumask_clear(new_mask); | |
4743 | else if (len > cpumask_size()) | |
4744 | len = cpumask_size(); | |
4745 | ||
1da177e4 LT |
4746 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4747 | } | |
4748 | ||
4749 | /** | |
4750 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4751 | * @pid: pid of the process | |
4752 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4753 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4754 | * |
4755 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4756 | */ |
5add95d4 HC |
4757 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4758 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4759 | { |
5a16f3d3 | 4760 | cpumask_var_t new_mask; |
1da177e4 LT |
4761 | int retval; |
4762 | ||
5a16f3d3 RR |
4763 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4764 | return -ENOMEM; | |
1da177e4 | 4765 | |
5a16f3d3 RR |
4766 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4767 | if (retval == 0) | |
4768 | retval = sched_setaffinity(pid, new_mask); | |
4769 | free_cpumask_var(new_mask); | |
4770 | return retval; | |
1da177e4 LT |
4771 | } |
4772 | ||
96f874e2 | 4773 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4774 | { |
36c8b586 | 4775 | struct task_struct *p; |
31605683 | 4776 | unsigned long flags; |
1da177e4 | 4777 | int retval; |
1da177e4 | 4778 | |
23f5d142 | 4779 | rcu_read_lock(); |
1da177e4 LT |
4780 | |
4781 | retval = -ESRCH; | |
4782 | p = find_process_by_pid(pid); | |
4783 | if (!p) | |
4784 | goto out_unlock; | |
4785 | ||
e7834f8f DQ |
4786 | retval = security_task_getscheduler(p); |
4787 | if (retval) | |
4788 | goto out_unlock; | |
4789 | ||
013fdb80 | 4790 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4791 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4792 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4793 | |
4794 | out_unlock: | |
23f5d142 | 4795 | rcu_read_unlock(); |
1da177e4 | 4796 | |
9531b62f | 4797 | return retval; |
1da177e4 LT |
4798 | } |
4799 | ||
4800 | /** | |
4801 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4802 | * @pid: pid of the process | |
4803 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4804 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 | 4805 | * |
599b4840 ZW |
4806 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4807 | * error code otherwise. | |
1da177e4 | 4808 | */ |
5add95d4 HC |
4809 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4810 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4811 | { |
4812 | int ret; | |
f17c8607 | 4813 | cpumask_var_t mask; |
1da177e4 | 4814 | |
84fba5ec | 4815 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4816 | return -EINVAL; |
4817 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4818 | return -EINVAL; |
4819 | ||
f17c8607 RR |
4820 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4821 | return -ENOMEM; | |
1da177e4 | 4822 | |
f17c8607 RR |
4823 | ret = sched_getaffinity(pid, mask); |
4824 | if (ret == 0) { | |
8bc037fb | 4825 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4826 | |
4827 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4828 | ret = -EFAULT; |
4829 | else | |
cd3d8031 | 4830 | ret = retlen; |
f17c8607 RR |
4831 | } |
4832 | free_cpumask_var(mask); | |
1da177e4 | 4833 | |
f17c8607 | 4834 | return ret; |
1da177e4 LT |
4835 | } |
4836 | ||
4837 | /** | |
4838 | * sys_sched_yield - yield the current processor to other threads. | |
4839 | * | |
dd41f596 IM |
4840 | * This function yields the current CPU to other tasks. If there are no |
4841 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4842 | * |
4843 | * Return: 0. | |
1da177e4 | 4844 | */ |
5add95d4 | 4845 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4846 | { |
70b97a7f | 4847 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4848 | |
ae92882e | 4849 | schedstat_inc(rq->yld_count); |
4530d7ab | 4850 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4851 | |
4852 | /* | |
4853 | * Since we are going to call schedule() anyway, there's | |
4854 | * no need to preempt or enable interrupts: | |
4855 | */ | |
4856 | __release(rq->lock); | |
8a25d5de | 4857 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4858 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4859 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4860 | |
4861 | schedule(); | |
4862 | ||
4863 | return 0; | |
4864 | } | |
4865 | ||
02b67cc3 | 4866 | int __sched _cond_resched(void) |
1da177e4 | 4867 | { |
fe32d3cd | 4868 | if (should_resched(0)) { |
a18b5d01 | 4869 | preempt_schedule_common(); |
1da177e4 LT |
4870 | return 1; |
4871 | } | |
4872 | return 0; | |
4873 | } | |
02b67cc3 | 4874 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4875 | |
4876 | /* | |
613afbf8 | 4877 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4878 | * call schedule, and on return reacquire the lock. |
4879 | * | |
41a2d6cf | 4880 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4881 | * operations here to prevent schedule() from being called twice (once via |
4882 | * spin_unlock(), once by hand). | |
4883 | */ | |
613afbf8 | 4884 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4885 | { |
fe32d3cd | 4886 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4887 | int ret = 0; |
4888 | ||
f607c668 PZ |
4889 | lockdep_assert_held(lock); |
4890 | ||
4a81e832 | 4891 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4892 | spin_unlock(lock); |
d86ee480 | 4893 | if (resched) |
a18b5d01 | 4894 | preempt_schedule_common(); |
95c354fe NP |
4895 | else |
4896 | cpu_relax(); | |
6df3cecb | 4897 | ret = 1; |
1da177e4 | 4898 | spin_lock(lock); |
1da177e4 | 4899 | } |
6df3cecb | 4900 | return ret; |
1da177e4 | 4901 | } |
613afbf8 | 4902 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4903 | |
613afbf8 | 4904 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4905 | { |
4906 | BUG_ON(!in_softirq()); | |
4907 | ||
fe32d3cd | 4908 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4909 | local_bh_enable(); |
a18b5d01 | 4910 | preempt_schedule_common(); |
1da177e4 LT |
4911 | local_bh_disable(); |
4912 | return 1; | |
4913 | } | |
4914 | return 0; | |
4915 | } | |
613afbf8 | 4916 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4917 | |
1da177e4 LT |
4918 | /** |
4919 | * yield - yield the current processor to other threads. | |
4920 | * | |
8e3fabfd PZ |
4921 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4922 | * | |
4923 | * The scheduler is at all times free to pick the calling task as the most | |
4924 | * eligible task to run, if removing the yield() call from your code breaks | |
4925 | * it, its already broken. | |
4926 | * | |
4927 | * Typical broken usage is: | |
4928 | * | |
4929 | * while (!event) | |
4930 | * yield(); | |
4931 | * | |
4932 | * where one assumes that yield() will let 'the other' process run that will | |
4933 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4934 | * happen. Never use yield() as a progress guarantee!! | |
4935 | * | |
4936 | * If you want to use yield() to wait for something, use wait_event(). | |
4937 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4938 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4939 | */ |
4940 | void __sched yield(void) | |
4941 | { | |
4942 | set_current_state(TASK_RUNNING); | |
4943 | sys_sched_yield(); | |
4944 | } | |
1da177e4 LT |
4945 | EXPORT_SYMBOL(yield); |
4946 | ||
d95f4122 MG |
4947 | /** |
4948 | * yield_to - yield the current processor to another thread in | |
4949 | * your thread group, or accelerate that thread toward the | |
4950 | * processor it's on. | |
16addf95 RD |
4951 | * @p: target task |
4952 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4953 | * |
4954 | * It's the caller's job to ensure that the target task struct | |
4955 | * can't go away on us before we can do any checks. | |
4956 | * | |
e69f6186 | 4957 | * Return: |
7b270f60 PZ |
4958 | * true (>0) if we indeed boosted the target task. |
4959 | * false (0) if we failed to boost the target. | |
4960 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4961 | */ |
fa93384f | 4962 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4963 | { |
4964 | struct task_struct *curr = current; | |
4965 | struct rq *rq, *p_rq; | |
4966 | unsigned long flags; | |
c3c18640 | 4967 | int yielded = 0; |
d95f4122 MG |
4968 | |
4969 | local_irq_save(flags); | |
4970 | rq = this_rq(); | |
4971 | ||
4972 | again: | |
4973 | p_rq = task_rq(p); | |
7b270f60 PZ |
4974 | /* |
4975 | * If we're the only runnable task on the rq and target rq also | |
4976 | * has only one task, there's absolutely no point in yielding. | |
4977 | */ | |
4978 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4979 | yielded = -ESRCH; | |
4980 | goto out_irq; | |
4981 | } | |
4982 | ||
d95f4122 | 4983 | double_rq_lock(rq, p_rq); |
39e24d8f | 4984 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4985 | double_rq_unlock(rq, p_rq); |
4986 | goto again; | |
4987 | } | |
4988 | ||
4989 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4990 | goto out_unlock; |
d95f4122 MG |
4991 | |
4992 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4993 | goto out_unlock; |
d95f4122 MG |
4994 | |
4995 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4996 | goto out_unlock; |
d95f4122 MG |
4997 | |
4998 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4999 | if (yielded) { |
ae92882e | 5000 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5001 | /* |
5002 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5003 | * fairness. | |
5004 | */ | |
5005 | if (preempt && rq != p_rq) | |
8875125e | 5006 | resched_curr(p_rq); |
6d1cafd8 | 5007 | } |
d95f4122 | 5008 | |
7b270f60 | 5009 | out_unlock: |
d95f4122 | 5010 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5011 | out_irq: |
d95f4122 MG |
5012 | local_irq_restore(flags); |
5013 | ||
7b270f60 | 5014 | if (yielded > 0) |
d95f4122 MG |
5015 | schedule(); |
5016 | ||
5017 | return yielded; | |
5018 | } | |
5019 | EXPORT_SYMBOL_GPL(yield_to); | |
5020 | ||
1da177e4 | 5021 | /* |
41a2d6cf | 5022 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5023 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5024 | */ |
1da177e4 LT |
5025 | long __sched io_schedule_timeout(long timeout) |
5026 | { | |
9cff8ade N |
5027 | int old_iowait = current->in_iowait; |
5028 | struct rq *rq; | |
1da177e4 LT |
5029 | long ret; |
5030 | ||
9cff8ade | 5031 | current->in_iowait = 1; |
10d784ea | 5032 | blk_schedule_flush_plug(current); |
9cff8ade | 5033 | |
0ff92245 | 5034 | delayacct_blkio_start(); |
9cff8ade | 5035 | rq = raw_rq(); |
1da177e4 LT |
5036 | atomic_inc(&rq->nr_iowait); |
5037 | ret = schedule_timeout(timeout); | |
9cff8ade | 5038 | current->in_iowait = old_iowait; |
1da177e4 | 5039 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5040 | delayacct_blkio_end(); |
9cff8ade | 5041 | |
1da177e4 LT |
5042 | return ret; |
5043 | } | |
9cff8ade | 5044 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
5045 | |
5046 | /** | |
5047 | * sys_sched_get_priority_max - return maximum RT priority. | |
5048 | * @policy: scheduling class. | |
5049 | * | |
e69f6186 YB |
5050 | * Return: On success, this syscall returns the maximum |
5051 | * rt_priority that can be used by a given scheduling class. | |
5052 | * On failure, a negative error code is returned. | |
1da177e4 | 5053 | */ |
5add95d4 | 5054 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5055 | { |
5056 | int ret = -EINVAL; | |
5057 | ||
5058 | switch (policy) { | |
5059 | case SCHED_FIFO: | |
5060 | case SCHED_RR: | |
5061 | ret = MAX_USER_RT_PRIO-1; | |
5062 | break; | |
aab03e05 | 5063 | case SCHED_DEADLINE: |
1da177e4 | 5064 | case SCHED_NORMAL: |
b0a9499c | 5065 | case SCHED_BATCH: |
dd41f596 | 5066 | case SCHED_IDLE: |
1da177e4 LT |
5067 | ret = 0; |
5068 | break; | |
5069 | } | |
5070 | return ret; | |
5071 | } | |
5072 | ||
5073 | /** | |
5074 | * sys_sched_get_priority_min - return minimum RT priority. | |
5075 | * @policy: scheduling class. | |
5076 | * | |
e69f6186 YB |
5077 | * Return: On success, this syscall returns the minimum |
5078 | * rt_priority that can be used by a given scheduling class. | |
5079 | * On failure, a negative error code is returned. | |
1da177e4 | 5080 | */ |
5add95d4 | 5081 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5082 | { |
5083 | int ret = -EINVAL; | |
5084 | ||
5085 | switch (policy) { | |
5086 | case SCHED_FIFO: | |
5087 | case SCHED_RR: | |
5088 | ret = 1; | |
5089 | break; | |
aab03e05 | 5090 | case SCHED_DEADLINE: |
1da177e4 | 5091 | case SCHED_NORMAL: |
b0a9499c | 5092 | case SCHED_BATCH: |
dd41f596 | 5093 | case SCHED_IDLE: |
1da177e4 LT |
5094 | ret = 0; |
5095 | } | |
5096 | return ret; | |
5097 | } | |
5098 | ||
5099 | /** | |
5100 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5101 | * @pid: pid of the process. | |
5102 | * @interval: userspace pointer to the timeslice value. | |
5103 | * | |
5104 | * this syscall writes the default timeslice value of a given process | |
5105 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5106 | * |
5107 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5108 | * an error code. | |
1da177e4 | 5109 | */ |
17da2bd9 | 5110 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5111 | struct timespec __user *, interval) |
1da177e4 | 5112 | { |
36c8b586 | 5113 | struct task_struct *p; |
a4ec24b4 | 5114 | unsigned int time_slice; |
eb580751 PZ |
5115 | struct rq_flags rf; |
5116 | struct timespec t; | |
dba091b9 | 5117 | struct rq *rq; |
3a5c359a | 5118 | int retval; |
1da177e4 LT |
5119 | |
5120 | if (pid < 0) | |
3a5c359a | 5121 | return -EINVAL; |
1da177e4 LT |
5122 | |
5123 | retval = -ESRCH; | |
1a551ae7 | 5124 | rcu_read_lock(); |
1da177e4 LT |
5125 | p = find_process_by_pid(pid); |
5126 | if (!p) | |
5127 | goto out_unlock; | |
5128 | ||
5129 | retval = security_task_getscheduler(p); | |
5130 | if (retval) | |
5131 | goto out_unlock; | |
5132 | ||
eb580751 | 5133 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5134 | time_slice = 0; |
5135 | if (p->sched_class->get_rr_interval) | |
5136 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5137 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5138 | |
1a551ae7 | 5139 | rcu_read_unlock(); |
a4ec24b4 | 5140 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5141 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5142 | return retval; |
3a5c359a | 5143 | |
1da177e4 | 5144 | out_unlock: |
1a551ae7 | 5145 | rcu_read_unlock(); |
1da177e4 LT |
5146 | return retval; |
5147 | } | |
5148 | ||
7c731e0a | 5149 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5150 | |
82a1fcb9 | 5151 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5152 | { |
1da177e4 | 5153 | unsigned long free = 0; |
4e79752c | 5154 | int ppid; |
1f8a7633 | 5155 | unsigned long state = p->state; |
1da177e4 | 5156 | |
1f8a7633 TH |
5157 | if (state) |
5158 | state = __ffs(state) + 1; | |
28d0686c | 5159 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5160 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5161 | #if BITS_PER_LONG == 32 |
1da177e4 | 5162 | if (state == TASK_RUNNING) |
3df0fc5b | 5163 | printk(KERN_CONT " running "); |
1da177e4 | 5164 | else |
3df0fc5b | 5165 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5166 | #else |
5167 | if (state == TASK_RUNNING) | |
3df0fc5b | 5168 | printk(KERN_CONT " running task "); |
1da177e4 | 5169 | else |
3df0fc5b | 5170 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5171 | #endif |
5172 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5173 | free = stack_not_used(p); |
1da177e4 | 5174 | #endif |
a90e984c | 5175 | ppid = 0; |
4e79752c | 5176 | rcu_read_lock(); |
a90e984c ON |
5177 | if (pid_alive(p)) |
5178 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5179 | rcu_read_unlock(); |
3df0fc5b | 5180 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5181 | task_pid_nr(p), ppid, |
aa47b7e0 | 5182 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5183 | |
3d1cb205 | 5184 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5185 | show_stack(p, NULL); |
1da177e4 LT |
5186 | } |
5187 | ||
e59e2ae2 | 5188 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5189 | { |
36c8b586 | 5190 | struct task_struct *g, *p; |
1da177e4 | 5191 | |
4bd77321 | 5192 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5193 | printk(KERN_INFO |
5194 | " task PC stack pid father\n"); | |
1da177e4 | 5195 | #else |
3df0fc5b PZ |
5196 | printk(KERN_INFO |
5197 | " task PC stack pid father\n"); | |
1da177e4 | 5198 | #endif |
510f5acc | 5199 | rcu_read_lock(); |
5d07f420 | 5200 | for_each_process_thread(g, p) { |
1da177e4 LT |
5201 | /* |
5202 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5203 | * console might take a lot of time: |
57675cb9 AR |
5204 | * Also, reset softlockup watchdogs on all CPUs, because |
5205 | * another CPU might be blocked waiting for us to process | |
5206 | * an IPI. | |
1da177e4 LT |
5207 | */ |
5208 | touch_nmi_watchdog(); | |
57675cb9 | 5209 | touch_all_softlockup_watchdogs(); |
39bc89fd | 5210 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5211 | sched_show_task(p); |
5d07f420 | 5212 | } |
1da177e4 | 5213 | |
dd41f596 | 5214 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5215 | if (!state_filter) |
5216 | sysrq_sched_debug_show(); | |
dd41f596 | 5217 | #endif |
510f5acc | 5218 | rcu_read_unlock(); |
e59e2ae2 IM |
5219 | /* |
5220 | * Only show locks if all tasks are dumped: | |
5221 | */ | |
93335a21 | 5222 | if (!state_filter) |
e59e2ae2 | 5223 | debug_show_all_locks(); |
1da177e4 LT |
5224 | } |
5225 | ||
0db0628d | 5226 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5227 | { |
dd41f596 | 5228 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5229 | } |
5230 | ||
f340c0d1 IM |
5231 | /** |
5232 | * init_idle - set up an idle thread for a given CPU | |
5233 | * @idle: task in question | |
5234 | * @cpu: cpu the idle task belongs to | |
5235 | * | |
5236 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5237 | * flag, to make booting more robust. | |
5238 | */ | |
0db0628d | 5239 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5240 | { |
70b97a7f | 5241 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5242 | unsigned long flags; |
5243 | ||
25834c73 PZ |
5244 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5245 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5246 | |
5e1576ed | 5247 | __sched_fork(0, idle); |
06b83b5f | 5248 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5249 | idle->se.exec_start = sched_clock(); |
5250 | ||
e1b77c92 MR |
5251 | kasan_unpoison_task_stack(idle); |
5252 | ||
de9b8f5d PZ |
5253 | #ifdef CONFIG_SMP |
5254 | /* | |
5255 | * Its possible that init_idle() gets called multiple times on a task, | |
5256 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5257 | * | |
5258 | * And since this is boot we can forgo the serialization. | |
5259 | */ | |
5260 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5261 | #endif | |
6506cf6c PZ |
5262 | /* |
5263 | * We're having a chicken and egg problem, even though we are | |
5264 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5265 | * lockdep check in task_group() will fail. | |
5266 | * | |
5267 | * Similar case to sched_fork(). / Alternatively we could | |
5268 | * use task_rq_lock() here and obtain the other rq->lock. | |
5269 | * | |
5270 | * Silence PROVE_RCU | |
5271 | */ | |
5272 | rcu_read_lock(); | |
dd41f596 | 5273 | __set_task_cpu(idle, cpu); |
6506cf6c | 5274 | rcu_read_unlock(); |
1da177e4 | 5275 | |
1da177e4 | 5276 | rq->curr = rq->idle = idle; |
da0c1e65 | 5277 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5278 | #ifdef CONFIG_SMP |
3ca7a440 | 5279 | idle->on_cpu = 1; |
4866cde0 | 5280 | #endif |
25834c73 PZ |
5281 | raw_spin_unlock(&rq->lock); |
5282 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5283 | |
5284 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5285 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5286 | |
dd41f596 IM |
5287 | /* |
5288 | * The idle tasks have their own, simple scheduling class: | |
5289 | */ | |
5290 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5291 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5292 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5293 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5294 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5295 | #endif | |
19978ca6 IM |
5296 | } |
5297 | ||
f82f8042 JL |
5298 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5299 | const struct cpumask *trial) | |
5300 | { | |
5301 | int ret = 1, trial_cpus; | |
5302 | struct dl_bw *cur_dl_b; | |
5303 | unsigned long flags; | |
5304 | ||
bb2bc55a MG |
5305 | if (!cpumask_weight(cur)) |
5306 | return ret; | |
5307 | ||
75e23e49 | 5308 | rcu_read_lock_sched(); |
f82f8042 JL |
5309 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5310 | trial_cpus = cpumask_weight(trial); | |
5311 | ||
5312 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5313 | if (cur_dl_b->bw != -1 && | |
5314 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5315 | ret = 0; | |
5316 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5317 | rcu_read_unlock_sched(); |
f82f8042 JL |
5318 | |
5319 | return ret; | |
5320 | } | |
5321 | ||
7f51412a JL |
5322 | int task_can_attach(struct task_struct *p, |
5323 | const struct cpumask *cs_cpus_allowed) | |
5324 | { | |
5325 | int ret = 0; | |
5326 | ||
5327 | /* | |
5328 | * Kthreads which disallow setaffinity shouldn't be moved | |
5329 | * to a new cpuset; we don't want to change their cpu | |
5330 | * affinity and isolating such threads by their set of | |
5331 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5332 | * applicable for such threads. This prevents checking for | |
5333 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5334 | * before cpus_allowed may be changed. | |
5335 | */ | |
5336 | if (p->flags & PF_NO_SETAFFINITY) { | |
5337 | ret = -EINVAL; | |
5338 | goto out; | |
5339 | } | |
5340 | ||
5341 | #ifdef CONFIG_SMP | |
5342 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5343 | cs_cpus_allowed)) { | |
5344 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5345 | cs_cpus_allowed); | |
75e23e49 | 5346 | struct dl_bw *dl_b; |
7f51412a JL |
5347 | bool overflow; |
5348 | int cpus; | |
5349 | unsigned long flags; | |
5350 | ||
75e23e49 JL |
5351 | rcu_read_lock_sched(); |
5352 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5353 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5354 | cpus = dl_bw_cpus(dest_cpu); | |
5355 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5356 | if (overflow) | |
5357 | ret = -EBUSY; | |
5358 | else { | |
5359 | /* | |
5360 | * We reserve space for this task in the destination | |
5361 | * root_domain, as we can't fail after this point. | |
5362 | * We will free resources in the source root_domain | |
5363 | * later on (see set_cpus_allowed_dl()). | |
5364 | */ | |
5365 | __dl_add(dl_b, p->dl.dl_bw); | |
5366 | } | |
5367 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5368 | rcu_read_unlock_sched(); |
7f51412a JL |
5369 | |
5370 | } | |
5371 | #endif | |
5372 | out: | |
5373 | return ret; | |
5374 | } | |
5375 | ||
1da177e4 | 5376 | #ifdef CONFIG_SMP |
1da177e4 | 5377 | |
e26fbffd TG |
5378 | static bool sched_smp_initialized __read_mostly; |
5379 | ||
e6628d5b MG |
5380 | #ifdef CONFIG_NUMA_BALANCING |
5381 | /* Migrate current task p to target_cpu */ | |
5382 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5383 | { | |
5384 | struct migration_arg arg = { p, target_cpu }; | |
5385 | int curr_cpu = task_cpu(p); | |
5386 | ||
5387 | if (curr_cpu == target_cpu) | |
5388 | return 0; | |
5389 | ||
5390 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5391 | return -EINVAL; | |
5392 | ||
5393 | /* TODO: This is not properly updating schedstats */ | |
5394 | ||
286549dc | 5395 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5396 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5397 | } | |
0ec8aa00 PZ |
5398 | |
5399 | /* | |
5400 | * Requeue a task on a given node and accurately track the number of NUMA | |
5401 | * tasks on the runqueues | |
5402 | */ | |
5403 | void sched_setnuma(struct task_struct *p, int nid) | |
5404 | { | |
da0c1e65 | 5405 | bool queued, running; |
eb580751 PZ |
5406 | struct rq_flags rf; |
5407 | struct rq *rq; | |
0ec8aa00 | 5408 | |
eb580751 | 5409 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5410 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5411 | running = task_current(rq, p); |
5412 | ||
da0c1e65 | 5413 | if (queued) |
1de64443 | 5414 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5415 | if (running) |
f3cd1c4e | 5416 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5417 | |
5418 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5419 | |
5420 | if (running) | |
5421 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5422 | if (queued) |
1de64443 | 5423 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
eb580751 | 5424 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5425 | } |
5cc389bc | 5426 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5427 | |
1da177e4 | 5428 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5429 | /* |
48c5ccae PZ |
5430 | * Ensures that the idle task is using init_mm right before its cpu goes |
5431 | * offline. | |
054b9108 | 5432 | */ |
48c5ccae | 5433 | void idle_task_exit(void) |
1da177e4 | 5434 | { |
48c5ccae | 5435 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5436 | |
48c5ccae | 5437 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5438 | |
a53efe5f | 5439 | if (mm != &init_mm) { |
f98db601 | 5440 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5441 | finish_arch_post_lock_switch(); |
5442 | } | |
48c5ccae | 5443 | mmdrop(mm); |
1da177e4 LT |
5444 | } |
5445 | ||
5446 | /* | |
5d180232 PZ |
5447 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5448 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5449 | * nr_active count is stable. We need to take the teardown thread which |
5450 | * is calling this into account, so we hand in adjust = 1 to the load | |
5451 | * calculation. | |
5d180232 PZ |
5452 | * |
5453 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5454 | */ |
5d180232 | 5455 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5456 | { |
d60585c5 | 5457 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5458 | if (delta) |
5459 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5460 | } |
5461 | ||
3f1d2a31 PZ |
5462 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5463 | { | |
5464 | } | |
5465 | ||
5466 | static const struct sched_class fake_sched_class = { | |
5467 | .put_prev_task = put_prev_task_fake, | |
5468 | }; | |
5469 | ||
5470 | static struct task_struct fake_task = { | |
5471 | /* | |
5472 | * Avoid pull_{rt,dl}_task() | |
5473 | */ | |
5474 | .prio = MAX_PRIO + 1, | |
5475 | .sched_class = &fake_sched_class, | |
5476 | }; | |
5477 | ||
48f24c4d | 5478 | /* |
48c5ccae PZ |
5479 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5480 | * try_to_wake_up()->select_task_rq(). | |
5481 | * | |
5482 | * Called with rq->lock held even though we'er in stop_machine() and | |
5483 | * there's no concurrency possible, we hold the required locks anyway | |
5484 | * because of lock validation efforts. | |
1da177e4 | 5485 | */ |
5e16bbc2 | 5486 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5487 | { |
5e16bbc2 | 5488 | struct rq *rq = dead_rq; |
48c5ccae | 5489 | struct task_struct *next, *stop = rq->stop; |
e7904a28 | 5490 | struct pin_cookie cookie; |
48c5ccae | 5491 | int dest_cpu; |
1da177e4 LT |
5492 | |
5493 | /* | |
48c5ccae PZ |
5494 | * Fudge the rq selection such that the below task selection loop |
5495 | * doesn't get stuck on the currently eligible stop task. | |
5496 | * | |
5497 | * We're currently inside stop_machine() and the rq is either stuck | |
5498 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5499 | * either way we should never end up calling schedule() until we're | |
5500 | * done here. | |
1da177e4 | 5501 | */ |
48c5ccae | 5502 | rq->stop = NULL; |
48f24c4d | 5503 | |
77bd3970 FW |
5504 | /* |
5505 | * put_prev_task() and pick_next_task() sched | |
5506 | * class method both need to have an up-to-date | |
5507 | * value of rq->clock[_task] | |
5508 | */ | |
5509 | update_rq_clock(rq); | |
5510 | ||
5e16bbc2 | 5511 | for (;;) { |
48c5ccae PZ |
5512 | /* |
5513 | * There's this thread running, bail when that's the only | |
5514 | * remaining thread. | |
5515 | */ | |
5516 | if (rq->nr_running == 1) | |
dd41f596 | 5517 | break; |
48c5ccae | 5518 | |
cbce1a68 | 5519 | /* |
5473e0cc | 5520 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 | 5521 | */ |
e7904a28 PZ |
5522 | cookie = lockdep_pin_lock(&rq->lock); |
5523 | next = pick_next_task(rq, &fake_task, cookie); | |
48c5ccae | 5524 | BUG_ON(!next); |
79c53799 | 5525 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5526 | |
5473e0cc WL |
5527 | /* |
5528 | * Rules for changing task_struct::cpus_allowed are holding | |
5529 | * both pi_lock and rq->lock, such that holding either | |
5530 | * stabilizes the mask. | |
5531 | * | |
5532 | * Drop rq->lock is not quite as disastrous as it usually is | |
5533 | * because !cpu_active at this point, which means load-balance | |
5534 | * will not interfere. Also, stop-machine. | |
5535 | */ | |
e7904a28 | 5536 | lockdep_unpin_lock(&rq->lock, cookie); |
5473e0cc WL |
5537 | raw_spin_unlock(&rq->lock); |
5538 | raw_spin_lock(&next->pi_lock); | |
5539 | raw_spin_lock(&rq->lock); | |
5540 | ||
5541 | /* | |
5542 | * Since we're inside stop-machine, _nothing_ should have | |
5543 | * changed the task, WARN if weird stuff happened, because in | |
5544 | * that case the above rq->lock drop is a fail too. | |
5545 | */ | |
5546 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5547 | raw_spin_unlock(&next->pi_lock); | |
5548 | continue; | |
5549 | } | |
5550 | ||
48c5ccae | 5551 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5552 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5553 | |
5e16bbc2 PZ |
5554 | rq = __migrate_task(rq, next, dest_cpu); |
5555 | if (rq != dead_rq) { | |
5556 | raw_spin_unlock(&rq->lock); | |
5557 | rq = dead_rq; | |
5558 | raw_spin_lock(&rq->lock); | |
5559 | } | |
5473e0cc | 5560 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5561 | } |
dce48a84 | 5562 | |
48c5ccae | 5563 | rq->stop = stop; |
dce48a84 | 5564 | } |
1da177e4 LT |
5565 | #endif /* CONFIG_HOTPLUG_CPU */ |
5566 | ||
1f11eb6a GH |
5567 | static void set_rq_online(struct rq *rq) |
5568 | { | |
5569 | if (!rq->online) { | |
5570 | const struct sched_class *class; | |
5571 | ||
c6c4927b | 5572 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5573 | rq->online = 1; |
5574 | ||
5575 | for_each_class(class) { | |
5576 | if (class->rq_online) | |
5577 | class->rq_online(rq); | |
5578 | } | |
5579 | } | |
5580 | } | |
5581 | ||
5582 | static void set_rq_offline(struct rq *rq) | |
5583 | { | |
5584 | if (rq->online) { | |
5585 | const struct sched_class *class; | |
5586 | ||
5587 | for_each_class(class) { | |
5588 | if (class->rq_offline) | |
5589 | class->rq_offline(rq); | |
5590 | } | |
5591 | ||
c6c4927b | 5592 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5593 | rq->online = 0; |
5594 | } | |
5595 | } | |
5596 | ||
9cf7243d | 5597 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5598 | { |
969c7921 | 5599 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5600 | |
a803f026 CM |
5601 | rq->age_stamp = sched_clock_cpu(cpu); |
5602 | } | |
5603 | ||
4cb98839 PZ |
5604 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5605 | ||
3e9830dc | 5606 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5607 | |
d039ac60 | 5608 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5609 | |
d039ac60 | 5610 | static int __init sched_debug_setup(char *str) |
f6630114 | 5611 | { |
d039ac60 | 5612 | sched_debug_enabled = 1; |
f6630114 MT |
5613 | |
5614 | return 0; | |
5615 | } | |
d039ac60 PZ |
5616 | early_param("sched_debug", sched_debug_setup); |
5617 | ||
5618 | static inline bool sched_debug(void) | |
5619 | { | |
5620 | return sched_debug_enabled; | |
5621 | } | |
f6630114 | 5622 | |
7c16ec58 | 5623 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5624 | struct cpumask *groupmask) |
1da177e4 | 5625 | { |
4dcf6aff | 5626 | struct sched_group *group = sd->groups; |
1da177e4 | 5627 | |
96f874e2 | 5628 | cpumask_clear(groupmask); |
4dcf6aff IM |
5629 | |
5630 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5631 | ||
5632 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5633 | printk("does not load-balance\n"); |
4dcf6aff | 5634 | if (sd->parent) |
3df0fc5b PZ |
5635 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5636 | " has parent"); | |
4dcf6aff | 5637 | return -1; |
41c7ce9a NP |
5638 | } |
5639 | ||
333470ee TH |
5640 | printk(KERN_CONT "span %*pbl level %s\n", |
5641 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5642 | |
758b2cdc | 5643 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5644 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5645 | "CPU%d\n", cpu); | |
4dcf6aff | 5646 | } |
758b2cdc | 5647 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5648 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5649 | " CPU%d\n", cpu); | |
4dcf6aff | 5650 | } |
1da177e4 | 5651 | |
4dcf6aff | 5652 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5653 | do { |
4dcf6aff | 5654 | if (!group) { |
3df0fc5b PZ |
5655 | printk("\n"); |
5656 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5657 | break; |
5658 | } | |
5659 | ||
758b2cdc | 5660 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5661 | printk(KERN_CONT "\n"); |
5662 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5663 | break; |
5664 | } | |
1da177e4 | 5665 | |
cb83b629 PZ |
5666 | if (!(sd->flags & SD_OVERLAP) && |
5667 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5668 | printk(KERN_CONT "\n"); |
5669 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5670 | break; |
5671 | } | |
1da177e4 | 5672 | |
758b2cdc | 5673 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5674 | |
333470ee TH |
5675 | printk(KERN_CONT " %*pbl", |
5676 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5677 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5678 | printk(KERN_CONT " (cpu_capacity = %d)", |
5679 | group->sgc->capacity); | |
381512cf | 5680 | } |
1da177e4 | 5681 | |
4dcf6aff IM |
5682 | group = group->next; |
5683 | } while (group != sd->groups); | |
3df0fc5b | 5684 | printk(KERN_CONT "\n"); |
1da177e4 | 5685 | |
758b2cdc | 5686 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5687 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5688 | |
758b2cdc RR |
5689 | if (sd->parent && |
5690 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5691 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5692 | "of domain->span\n"); | |
4dcf6aff IM |
5693 | return 0; |
5694 | } | |
1da177e4 | 5695 | |
4dcf6aff IM |
5696 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5697 | { | |
5698 | int level = 0; | |
1da177e4 | 5699 | |
d039ac60 | 5700 | if (!sched_debug_enabled) |
f6630114 MT |
5701 | return; |
5702 | ||
4dcf6aff IM |
5703 | if (!sd) { |
5704 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5705 | return; | |
5706 | } | |
1da177e4 | 5707 | |
4dcf6aff IM |
5708 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5709 | ||
5710 | for (;;) { | |
4cb98839 | 5711 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5712 | break; |
1da177e4 LT |
5713 | level++; |
5714 | sd = sd->parent; | |
33859f7f | 5715 | if (!sd) |
4dcf6aff IM |
5716 | break; |
5717 | } | |
1da177e4 | 5718 | } |
6d6bc0ad | 5719 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5720 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5721 | static inline bool sched_debug(void) |
5722 | { | |
5723 | return false; | |
5724 | } | |
6d6bc0ad | 5725 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5726 | |
1a20ff27 | 5727 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5728 | { |
758b2cdc | 5729 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5730 | return 1; |
5731 | ||
5732 | /* Following flags need at least 2 groups */ | |
5733 | if (sd->flags & (SD_LOAD_BALANCE | | |
5734 | SD_BALANCE_NEWIDLE | | |
5735 | SD_BALANCE_FORK | | |
89c4710e | 5736 | SD_BALANCE_EXEC | |
5d4dfddd | 5737 | SD_SHARE_CPUCAPACITY | |
1f6e6c7c | 5738 | SD_ASYM_CPUCAPACITY | |
d77b3ed5 VG |
5739 | SD_SHARE_PKG_RESOURCES | |
5740 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5741 | if (sd->groups != sd->groups->next) |
5742 | return 0; | |
5743 | } | |
5744 | ||
5745 | /* Following flags don't use groups */ | |
c88d5910 | 5746 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5747 | return 0; |
5748 | ||
5749 | return 1; | |
5750 | } | |
5751 | ||
48f24c4d IM |
5752 | static int |
5753 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5754 | { |
5755 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5756 | ||
5757 | if (sd_degenerate(parent)) | |
5758 | return 1; | |
5759 | ||
758b2cdc | 5760 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5761 | return 0; |
5762 | ||
245af2c7 SS |
5763 | /* Flags needing groups don't count if only 1 group in parent */ |
5764 | if (parent->groups == parent->groups->next) { | |
5765 | pflags &= ~(SD_LOAD_BALANCE | | |
5766 | SD_BALANCE_NEWIDLE | | |
5767 | SD_BALANCE_FORK | | |
89c4710e | 5768 | SD_BALANCE_EXEC | |
1f6e6c7c | 5769 | SD_ASYM_CPUCAPACITY | |
5d4dfddd | 5770 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5771 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5772 | SD_PREFER_SIBLING | |
5773 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5774 | if (nr_node_ids == 1) |
5775 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5776 | } |
5777 | if (~cflags & pflags) | |
5778 | return 0; | |
5779 | ||
5780 | return 1; | |
5781 | } | |
5782 | ||
dce840a0 | 5783 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5784 | { |
dce840a0 | 5785 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5786 | |
68e74568 | 5787 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5788 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5789 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5790 | free_cpumask_var(rd->rto_mask); |
5791 | free_cpumask_var(rd->online); | |
5792 | free_cpumask_var(rd->span); | |
5793 | kfree(rd); | |
5794 | } | |
5795 | ||
57d885fe GH |
5796 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5797 | { | |
a0490fa3 | 5798 | struct root_domain *old_rd = NULL; |
57d885fe | 5799 | unsigned long flags; |
57d885fe | 5800 | |
05fa785c | 5801 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5802 | |
5803 | if (rq->rd) { | |
a0490fa3 | 5804 | old_rd = rq->rd; |
57d885fe | 5805 | |
c6c4927b | 5806 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5807 | set_rq_offline(rq); |
57d885fe | 5808 | |
c6c4927b | 5809 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5810 | |
a0490fa3 | 5811 | /* |
0515973f | 5812 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5813 | * set old_rd to NULL to skip the freeing later |
5814 | * in this function: | |
5815 | */ | |
5816 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5817 | old_rd = NULL; | |
57d885fe GH |
5818 | } |
5819 | ||
5820 | atomic_inc(&rd->refcount); | |
5821 | rq->rd = rd; | |
5822 | ||
c6c4927b | 5823 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5824 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5825 | set_rq_online(rq); |
57d885fe | 5826 | |
05fa785c | 5827 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5828 | |
5829 | if (old_rd) | |
dce840a0 | 5830 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5831 | } |
5832 | ||
68c38fc3 | 5833 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5834 | { |
5835 | memset(rd, 0, sizeof(*rd)); | |
5836 | ||
8295c699 | 5837 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5838 | goto out; |
8295c699 | 5839 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5840 | goto free_span; |
8295c699 | 5841 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5842 | goto free_online; |
8295c699 | 5843 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5844 | goto free_dlo_mask; |
6e0534f2 | 5845 | |
332ac17e | 5846 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5847 | if (cpudl_init(&rd->cpudl) != 0) |
5848 | goto free_dlo_mask; | |
332ac17e | 5849 | |
68c38fc3 | 5850 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5851 | goto free_rto_mask; |
c6c4927b | 5852 | return 0; |
6e0534f2 | 5853 | |
68e74568 RR |
5854 | free_rto_mask: |
5855 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5856 | free_dlo_mask: |
5857 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5858 | free_online: |
5859 | free_cpumask_var(rd->online); | |
5860 | free_span: | |
5861 | free_cpumask_var(rd->span); | |
0c910d28 | 5862 | out: |
c6c4927b | 5863 | return -ENOMEM; |
57d885fe GH |
5864 | } |
5865 | ||
029632fb PZ |
5866 | /* |
5867 | * By default the system creates a single root-domain with all cpus as | |
5868 | * members (mimicking the global state we have today). | |
5869 | */ | |
5870 | struct root_domain def_root_domain; | |
5871 | ||
57d885fe GH |
5872 | static void init_defrootdomain(void) |
5873 | { | |
68c38fc3 | 5874 | init_rootdomain(&def_root_domain); |
c6c4927b | 5875 | |
57d885fe GH |
5876 | atomic_set(&def_root_domain.refcount, 1); |
5877 | } | |
5878 | ||
dc938520 | 5879 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5880 | { |
5881 | struct root_domain *rd; | |
5882 | ||
5883 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5884 | if (!rd) | |
5885 | return NULL; | |
5886 | ||
68c38fc3 | 5887 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5888 | kfree(rd); |
5889 | return NULL; | |
5890 | } | |
57d885fe GH |
5891 | |
5892 | return rd; | |
5893 | } | |
5894 | ||
63b2ca30 | 5895 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5896 | { |
5897 | struct sched_group *tmp, *first; | |
5898 | ||
5899 | if (!sg) | |
5900 | return; | |
5901 | ||
5902 | first = sg; | |
5903 | do { | |
5904 | tmp = sg->next; | |
5905 | ||
63b2ca30 NP |
5906 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5907 | kfree(sg->sgc); | |
e3589f6c PZ |
5908 | |
5909 | kfree(sg); | |
5910 | sg = tmp; | |
5911 | } while (sg != first); | |
5912 | } | |
5913 | ||
dce840a0 PZ |
5914 | static void free_sched_domain(struct rcu_head *rcu) |
5915 | { | |
5916 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5917 | |
5918 | /* | |
5919 | * If its an overlapping domain it has private groups, iterate and | |
5920 | * nuke them all. | |
5921 | */ | |
5922 | if (sd->flags & SD_OVERLAP) { | |
5923 | free_sched_groups(sd->groups, 1); | |
5924 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5925 | kfree(sd->groups->sgc); |
dce840a0 | 5926 | kfree(sd->groups); |
9c3f75cb | 5927 | } |
dce840a0 PZ |
5928 | kfree(sd); |
5929 | } | |
5930 | ||
5931 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5932 | { | |
5933 | call_rcu(&sd->rcu, free_sched_domain); | |
5934 | } | |
5935 | ||
5936 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5937 | { | |
5938 | for (; sd; sd = sd->parent) | |
5939 | destroy_sched_domain(sd, cpu); | |
5940 | } | |
5941 | ||
518cd623 PZ |
5942 | /* |
5943 | * Keep a special pointer to the highest sched_domain that has | |
5944 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5945 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5946 | * | |
5947 | * Also keep a unique ID per domain (we use the first cpu number in | |
5948 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5949 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5950 | */ |
5951 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5952 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5953 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5954 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5955 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5956 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5957 | |
5958 | static void update_top_cache_domain(int cpu) | |
5959 | { | |
5960 | struct sched_domain *sd; | |
5d4cf996 | 5961 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5962 | int id = cpu; |
7d9ffa89 | 5963 | int size = 1; |
518cd623 PZ |
5964 | |
5965 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5966 | if (sd) { |
518cd623 | 5967 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5968 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5969 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5970 | } |
5d4cf996 | 5971 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5972 | |
5973 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5974 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5975 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5976 | |
5977 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5978 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5979 | |
5980 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5981 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5982 | } |
5983 | ||
1da177e4 | 5984 | /* |
0eab9146 | 5985 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5986 | * hold the hotplug lock. |
5987 | */ | |
0eab9146 IM |
5988 | static void |
5989 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5990 | { |
70b97a7f | 5991 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5992 | struct sched_domain *tmp; |
5993 | ||
5994 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5995 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5996 | struct sched_domain *parent = tmp->parent; |
5997 | if (!parent) | |
5998 | break; | |
f29c9b1c | 5999 | |
1a848870 | 6000 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6001 | tmp->parent = parent->parent; |
1a848870 SS |
6002 | if (parent->parent) |
6003 | parent->parent->child = tmp; | |
10866e62 PZ |
6004 | /* |
6005 | * Transfer SD_PREFER_SIBLING down in case of a | |
6006 | * degenerate parent; the spans match for this | |
6007 | * so the property transfers. | |
6008 | */ | |
6009 | if (parent->flags & SD_PREFER_SIBLING) | |
6010 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 6011 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6012 | } else |
6013 | tmp = tmp->parent; | |
245af2c7 SS |
6014 | } |
6015 | ||
1a848870 | 6016 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6017 | tmp = sd; |
245af2c7 | 6018 | sd = sd->parent; |
dce840a0 | 6019 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6020 | if (sd) |
6021 | sd->child = NULL; | |
6022 | } | |
1da177e4 | 6023 | |
4cb98839 | 6024 | sched_domain_debug(sd, cpu); |
1da177e4 | 6025 | |
57d885fe | 6026 | rq_attach_root(rq, rd); |
dce840a0 | 6027 | tmp = rq->sd; |
674311d5 | 6028 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6029 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
6030 | |
6031 | update_top_cache_domain(cpu); | |
1da177e4 LT |
6032 | } |
6033 | ||
1da177e4 LT |
6034 | /* Setup the mask of cpus configured for isolated domains */ |
6035 | static int __init isolated_cpu_setup(char *str) | |
6036 | { | |
a6e4491c PB |
6037 | int ret; |
6038 | ||
bdddd296 | 6039 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
6040 | ret = cpulist_parse(str, cpu_isolated_map); |
6041 | if (ret) { | |
6042 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
6043 | return 0; | |
6044 | } | |
1da177e4 LT |
6045 | return 1; |
6046 | } | |
8927f494 | 6047 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6048 | |
49a02c51 | 6049 | struct s_data { |
21d42ccf | 6050 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6051 | struct root_domain *rd; |
6052 | }; | |
6053 | ||
2109b99e | 6054 | enum s_alloc { |
2109b99e | 6055 | sa_rootdomain, |
21d42ccf | 6056 | sa_sd, |
dce840a0 | 6057 | sa_sd_storage, |
2109b99e AH |
6058 | sa_none, |
6059 | }; | |
6060 | ||
c1174876 PZ |
6061 | /* |
6062 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
6063 | * domain traversal. | |
6064 | * | |
6065 | * Asymmetric node setups can result in situations where the domain tree is of | |
6066 | * unequal depth, make sure to skip domains that already cover the entire | |
6067 | * range. | |
6068 | * | |
6069 | * In that case build_sched_domains() will have terminated the iteration early | |
6070 | * and our sibling sd spans will be empty. Domains should always include the | |
6071 | * cpu they're built on, so check that. | |
6072 | * | |
6073 | */ | |
6074 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6075 | { | |
6076 | const struct cpumask *span = sched_domain_span(sd); | |
6077 | struct sd_data *sdd = sd->private; | |
6078 | struct sched_domain *sibling; | |
6079 | int i; | |
6080 | ||
6081 | for_each_cpu(i, span) { | |
6082 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6083 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6084 | continue; | |
6085 | ||
6086 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6087 | } | |
6088 | } | |
6089 | ||
6090 | /* | |
6091 | * Return the canonical balance cpu for this group, this is the first cpu | |
6092 | * of this group that's also in the iteration mask. | |
6093 | */ | |
6094 | int group_balance_cpu(struct sched_group *sg) | |
6095 | { | |
6096 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6097 | } | |
6098 | ||
e3589f6c PZ |
6099 | static int |
6100 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6101 | { | |
6102 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6103 | const struct cpumask *span = sched_domain_span(sd); | |
6104 | struct cpumask *covered = sched_domains_tmpmask; | |
6105 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6106 | struct sched_domain *sibling; |
e3589f6c PZ |
6107 | int i; |
6108 | ||
6109 | cpumask_clear(covered); | |
6110 | ||
6111 | for_each_cpu(i, span) { | |
6112 | struct cpumask *sg_span; | |
6113 | ||
6114 | if (cpumask_test_cpu(i, covered)) | |
6115 | continue; | |
6116 | ||
aaecac4a | 6117 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6118 | |
6119 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6120 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6121 | continue; |
6122 | ||
e3589f6c | 6123 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6124 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6125 | |
6126 | if (!sg) | |
6127 | goto fail; | |
6128 | ||
6129 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6130 | if (sibling->child) |
6131 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6132 | else | |
e3589f6c PZ |
6133 | cpumask_set_cpu(i, sg_span); |
6134 | ||
6135 | cpumask_or(covered, covered, sg_span); | |
6136 | ||
63b2ca30 NP |
6137 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6138 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6139 | build_group_mask(sd, sg); |
6140 | ||
c3decf0d | 6141 | /* |
63b2ca30 | 6142 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6143 | * domains and no possible iteration will get us here, we won't |
6144 | * die on a /0 trap. | |
6145 | */ | |
ca8ce3d0 | 6146 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 6147 | |
c1174876 PZ |
6148 | /* |
6149 | * Make sure the first group of this domain contains the | |
6150 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6151 | * breaks. See update_sg_lb_stats(). | |
6152 | */ | |
74a5ce20 | 6153 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6154 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6155 | groups = sg; |
6156 | ||
6157 | if (!first) | |
6158 | first = sg; | |
6159 | if (last) | |
6160 | last->next = sg; | |
6161 | last = sg; | |
6162 | last->next = first; | |
6163 | } | |
6164 | sd->groups = groups; | |
6165 | ||
6166 | return 0; | |
6167 | ||
6168 | fail: | |
6169 | free_sched_groups(first, 0); | |
6170 | ||
6171 | return -ENOMEM; | |
6172 | } | |
6173 | ||
dce840a0 | 6174 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6175 | { |
dce840a0 PZ |
6176 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6177 | struct sched_domain *child = sd->child; | |
1da177e4 | 6178 | |
dce840a0 PZ |
6179 | if (child) |
6180 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6181 | |
9c3f75cb | 6182 | if (sg) { |
dce840a0 | 6183 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6184 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6185 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6186 | } |
dce840a0 PZ |
6187 | |
6188 | return cpu; | |
1e9f28fa | 6189 | } |
1e9f28fa | 6190 | |
01a08546 | 6191 | /* |
dce840a0 PZ |
6192 | * build_sched_groups will build a circular linked list of the groups |
6193 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6194 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6195 | * |
6196 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6197 | */ |
e3589f6c PZ |
6198 | static int |
6199 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6200 | { |
dce840a0 PZ |
6201 | struct sched_group *first = NULL, *last = NULL; |
6202 | struct sd_data *sdd = sd->private; | |
6203 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6204 | struct cpumask *covered; |
dce840a0 | 6205 | int i; |
9c1cfda2 | 6206 | |
e3589f6c PZ |
6207 | get_group(cpu, sdd, &sd->groups); |
6208 | atomic_inc(&sd->groups->ref); | |
6209 | ||
0936629f | 6210 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6211 | return 0; |
6212 | ||
f96225fd PZ |
6213 | lockdep_assert_held(&sched_domains_mutex); |
6214 | covered = sched_domains_tmpmask; | |
6215 | ||
dce840a0 | 6216 | cpumask_clear(covered); |
6711cab4 | 6217 | |
dce840a0 PZ |
6218 | for_each_cpu(i, span) { |
6219 | struct sched_group *sg; | |
cd08e923 | 6220 | int group, j; |
6711cab4 | 6221 | |
dce840a0 PZ |
6222 | if (cpumask_test_cpu(i, covered)) |
6223 | continue; | |
6711cab4 | 6224 | |
cd08e923 | 6225 | group = get_group(i, sdd, &sg); |
c1174876 | 6226 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6227 | |
dce840a0 PZ |
6228 | for_each_cpu(j, span) { |
6229 | if (get_group(j, sdd, NULL) != group) | |
6230 | continue; | |
0601a88d | 6231 | |
dce840a0 PZ |
6232 | cpumask_set_cpu(j, covered); |
6233 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6234 | } | |
0601a88d | 6235 | |
dce840a0 PZ |
6236 | if (!first) |
6237 | first = sg; | |
6238 | if (last) | |
6239 | last->next = sg; | |
6240 | last = sg; | |
6241 | } | |
6242 | last->next = first; | |
e3589f6c PZ |
6243 | |
6244 | return 0; | |
0601a88d | 6245 | } |
51888ca2 | 6246 | |
89c4710e | 6247 | /* |
63b2ca30 | 6248 | * Initialize sched groups cpu_capacity. |
89c4710e | 6249 | * |
63b2ca30 | 6250 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6251 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6252 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6253 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6254 | * group having more cpu_capacity will pickup more load compared to the | |
6255 | * group having less cpu_capacity. | |
89c4710e | 6256 | */ |
63b2ca30 | 6257 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6258 | { |
e3589f6c | 6259 | struct sched_group *sg = sd->groups; |
89c4710e | 6260 | |
94c95ba6 | 6261 | WARN_ON(!sg); |
e3589f6c PZ |
6262 | |
6263 | do { | |
6264 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6265 | sg = sg->next; | |
6266 | } while (sg != sd->groups); | |
89c4710e | 6267 | |
c1174876 | 6268 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6269 | return; |
aae6d3dd | 6270 | |
63b2ca30 NP |
6271 | update_group_capacity(sd, cpu); |
6272 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6273 | } |
6274 | ||
7c16ec58 MT |
6275 | /* |
6276 | * Initializers for schedule domains | |
6277 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6278 | */ | |
6279 | ||
1d3504fc | 6280 | static int default_relax_domain_level = -1; |
60495e77 | 6281 | int sched_domain_level_max; |
1d3504fc HS |
6282 | |
6283 | static int __init setup_relax_domain_level(char *str) | |
6284 | { | |
a841f8ce DS |
6285 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6286 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6287 | |
1d3504fc HS |
6288 | return 1; |
6289 | } | |
6290 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6291 | ||
6292 | static void set_domain_attribute(struct sched_domain *sd, | |
6293 | struct sched_domain_attr *attr) | |
6294 | { | |
6295 | int request; | |
6296 | ||
6297 | if (!attr || attr->relax_domain_level < 0) { | |
6298 | if (default_relax_domain_level < 0) | |
6299 | return; | |
6300 | else | |
6301 | request = default_relax_domain_level; | |
6302 | } else | |
6303 | request = attr->relax_domain_level; | |
6304 | if (request < sd->level) { | |
6305 | /* turn off idle balance on this domain */ | |
c88d5910 | 6306 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6307 | } else { |
6308 | /* turn on idle balance on this domain */ | |
c88d5910 | 6309 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6310 | } |
6311 | } | |
6312 | ||
54ab4ff4 PZ |
6313 | static void __sdt_free(const struct cpumask *cpu_map); |
6314 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6315 | ||
2109b99e AH |
6316 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6317 | const struct cpumask *cpu_map) | |
6318 | { | |
6319 | switch (what) { | |
2109b99e | 6320 | case sa_rootdomain: |
822ff793 PZ |
6321 | if (!atomic_read(&d->rd->refcount)) |
6322 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6323 | case sa_sd: |
6324 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6325 | case sa_sd_storage: |
54ab4ff4 | 6326 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6327 | case sa_none: |
6328 | break; | |
6329 | } | |
6330 | } | |
3404c8d9 | 6331 | |
2109b99e AH |
6332 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6333 | const struct cpumask *cpu_map) | |
6334 | { | |
dce840a0 PZ |
6335 | memset(d, 0, sizeof(*d)); |
6336 | ||
54ab4ff4 PZ |
6337 | if (__sdt_alloc(cpu_map)) |
6338 | return sa_sd_storage; | |
dce840a0 PZ |
6339 | d->sd = alloc_percpu(struct sched_domain *); |
6340 | if (!d->sd) | |
6341 | return sa_sd_storage; | |
2109b99e | 6342 | d->rd = alloc_rootdomain(); |
dce840a0 | 6343 | if (!d->rd) |
21d42ccf | 6344 | return sa_sd; |
2109b99e AH |
6345 | return sa_rootdomain; |
6346 | } | |
57d885fe | 6347 | |
dce840a0 PZ |
6348 | /* |
6349 | * NULL the sd_data elements we've used to build the sched_domain and | |
6350 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6351 | * will not free the data we're using. | |
6352 | */ | |
6353 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6354 | { | |
6355 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6356 | |
6357 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6358 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6359 | ||
e3589f6c | 6360 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6361 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6362 | |
63b2ca30 NP |
6363 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6364 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6365 | } |
6366 | ||
cb83b629 | 6367 | #ifdef CONFIG_NUMA |
cb83b629 | 6368 | static int sched_domains_numa_levels; |
e3fe70b1 | 6369 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6370 | static int *sched_domains_numa_distance; |
9942f79b | 6371 | int sched_max_numa_distance; |
cb83b629 PZ |
6372 | static struct cpumask ***sched_domains_numa_masks; |
6373 | static int sched_domains_curr_level; | |
143e1e28 | 6374 | #endif |
cb83b629 | 6375 | |
143e1e28 VG |
6376 | /* |
6377 | * SD_flags allowed in topology descriptions. | |
6378 | * | |
94f438c8 PZ |
6379 | * These flags are purely descriptive of the topology and do not prescribe |
6380 | * behaviour. Behaviour is artificial and mapped in the below sd_init() | |
6381 | * function: | |
143e1e28 | 6382 | * |
94f438c8 PZ |
6383 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
6384 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
6385 | * SD_NUMA - describes NUMA topologies | |
6386 | * SD_SHARE_POWERDOMAIN - describes shared power domain | |
1f6e6c7c | 6387 | * SD_ASYM_CPUCAPACITY - describes mixed capacity topologies |
94f438c8 PZ |
6388 | * |
6389 | * Odd one out, which beside describing the topology has a quirk also | |
6390 | * prescribes the desired behaviour that goes along with it: | |
6391 | * | |
6392 | * SD_ASYM_PACKING - describes SMT quirks | |
143e1e28 VG |
6393 | */ |
6394 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6395 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6396 | SD_SHARE_PKG_RESOURCES | \ |
6397 | SD_NUMA | \ | |
d77b3ed5 | 6398 | SD_ASYM_PACKING | \ |
1f6e6c7c | 6399 | SD_ASYM_CPUCAPACITY | \ |
d77b3ed5 | 6400 | SD_SHARE_POWERDOMAIN) |
cb83b629 PZ |
6401 | |
6402 | static struct sched_domain * | |
3676b13e MR |
6403 | sd_init(struct sched_domain_topology_level *tl, |
6404 | struct sched_domain *child, int cpu) | |
cb83b629 PZ |
6405 | { |
6406 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6407 | int sd_weight, sd_flags = 0; |
6408 | ||
6409 | #ifdef CONFIG_NUMA | |
6410 | /* | |
6411 | * Ugly hack to pass state to sd_numa_mask()... | |
6412 | */ | |
6413 | sched_domains_curr_level = tl->numa_level; | |
6414 | #endif | |
6415 | ||
6416 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6417 | ||
6418 | if (tl->sd_flags) | |
6419 | sd_flags = (*tl->sd_flags)(); | |
6420 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6421 | "wrong sd_flags in topology description\n")) | |
6422 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6423 | |
6424 | *sd = (struct sched_domain){ | |
6425 | .min_interval = sd_weight, | |
6426 | .max_interval = 2*sd_weight, | |
6427 | .busy_factor = 32, | |
870a0bb5 | 6428 | .imbalance_pct = 125, |
143e1e28 VG |
6429 | |
6430 | .cache_nice_tries = 0, | |
6431 | .busy_idx = 0, | |
6432 | .idle_idx = 0, | |
cb83b629 PZ |
6433 | .newidle_idx = 0, |
6434 | .wake_idx = 0, | |
6435 | .forkexec_idx = 0, | |
6436 | ||
6437 | .flags = 1*SD_LOAD_BALANCE | |
6438 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6439 | | 1*SD_BALANCE_EXEC |
6440 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6441 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6442 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6443 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6444 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6445 | | 0*SD_SERIALIZE |
cb83b629 | 6446 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6447 | | 0*SD_NUMA |
6448 | | sd_flags | |
cb83b629 | 6449 | , |
143e1e28 | 6450 | |
cb83b629 PZ |
6451 | .last_balance = jiffies, |
6452 | .balance_interval = sd_weight, | |
143e1e28 | 6453 | .smt_gain = 0, |
2b4cfe64 JL |
6454 | .max_newidle_lb_cost = 0, |
6455 | .next_decay_max_lb_cost = jiffies, | |
3676b13e | 6456 | .child = child, |
143e1e28 VG |
6457 | #ifdef CONFIG_SCHED_DEBUG |
6458 | .name = tl->name, | |
6459 | #endif | |
cb83b629 | 6460 | }; |
cb83b629 PZ |
6461 | |
6462 | /* | |
143e1e28 | 6463 | * Convert topological properties into behaviour. |
cb83b629 | 6464 | */ |
143e1e28 | 6465 | |
9ee1cda5 MR |
6466 | if (sd->flags & SD_ASYM_CPUCAPACITY) { |
6467 | struct sched_domain *t = sd; | |
6468 | ||
6469 | for_each_lower_domain(t) | |
6470 | t->flags |= SD_BALANCE_WAKE; | |
6471 | } | |
6472 | ||
5d4dfddd | 6473 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6474 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6475 | sd->imbalance_pct = 110; |
6476 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6477 | |
6478 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6479 | sd->imbalance_pct = 117; | |
6480 | sd->cache_nice_tries = 1; | |
6481 | sd->busy_idx = 2; | |
6482 | ||
6483 | #ifdef CONFIG_NUMA | |
6484 | } else if (sd->flags & SD_NUMA) { | |
6485 | sd->cache_nice_tries = 2; | |
6486 | sd->busy_idx = 3; | |
6487 | sd->idle_idx = 2; | |
6488 | ||
6489 | sd->flags |= SD_SERIALIZE; | |
6490 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6491 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6492 | SD_BALANCE_FORK | | |
6493 | SD_WAKE_AFFINE); | |
6494 | } | |
6495 | ||
6496 | #endif | |
6497 | } else { | |
6498 | sd->flags |= SD_PREFER_SIBLING; | |
6499 | sd->cache_nice_tries = 1; | |
6500 | sd->busy_idx = 2; | |
6501 | sd->idle_idx = 1; | |
6502 | } | |
6503 | ||
6504 | sd->private = &tl->data; | |
cb83b629 PZ |
6505 | |
6506 | return sd; | |
6507 | } | |
6508 | ||
143e1e28 VG |
6509 | /* |
6510 | * Topology list, bottom-up. | |
6511 | */ | |
6512 | static struct sched_domain_topology_level default_topology[] = { | |
6513 | #ifdef CONFIG_SCHED_SMT | |
6514 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6515 | #endif | |
6516 | #ifdef CONFIG_SCHED_MC | |
6517 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6518 | #endif |
6519 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6520 | { NULL, }, | |
6521 | }; | |
6522 | ||
c6e1e7b5 JG |
6523 | static struct sched_domain_topology_level *sched_domain_topology = |
6524 | default_topology; | |
143e1e28 VG |
6525 | |
6526 | #define for_each_sd_topology(tl) \ | |
6527 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6528 | ||
6529 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6530 | { | |
6531 | sched_domain_topology = tl; | |
6532 | } | |
6533 | ||
6534 | #ifdef CONFIG_NUMA | |
6535 | ||
cb83b629 PZ |
6536 | static const struct cpumask *sd_numa_mask(int cpu) |
6537 | { | |
6538 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6539 | } | |
6540 | ||
d039ac60 PZ |
6541 | static void sched_numa_warn(const char *str) |
6542 | { | |
6543 | static int done = false; | |
6544 | int i,j; | |
6545 | ||
6546 | if (done) | |
6547 | return; | |
6548 | ||
6549 | done = true; | |
6550 | ||
6551 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6552 | ||
6553 | for (i = 0; i < nr_node_ids; i++) { | |
6554 | printk(KERN_WARNING " "); | |
6555 | for (j = 0; j < nr_node_ids; j++) | |
6556 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6557 | printk(KERN_CONT "\n"); | |
6558 | } | |
6559 | printk(KERN_WARNING "\n"); | |
6560 | } | |
6561 | ||
9942f79b | 6562 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6563 | { |
6564 | int i; | |
6565 | ||
6566 | if (distance == node_distance(0, 0)) | |
6567 | return true; | |
6568 | ||
6569 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6570 | if (sched_domains_numa_distance[i] == distance) | |
6571 | return true; | |
6572 | } | |
6573 | ||
6574 | return false; | |
6575 | } | |
6576 | ||
e3fe70b1 RR |
6577 | /* |
6578 | * A system can have three types of NUMA topology: | |
6579 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6580 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6581 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6582 | * | |
6583 | * The difference between a glueless mesh topology and a backplane | |
6584 | * topology lies in whether communication between not directly | |
6585 | * connected nodes goes through intermediary nodes (where programs | |
6586 | * could run), or through backplane controllers. This affects | |
6587 | * placement of programs. | |
6588 | * | |
6589 | * The type of topology can be discerned with the following tests: | |
6590 | * - If the maximum distance between any nodes is 1 hop, the system | |
6591 | * is directly connected. | |
6592 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6593 | * there is an intermediary node C, which is < N hops away from both | |
6594 | * nodes A and B, the system is a glueless mesh. | |
6595 | */ | |
6596 | static void init_numa_topology_type(void) | |
6597 | { | |
6598 | int a, b, c, n; | |
6599 | ||
6600 | n = sched_max_numa_distance; | |
6601 | ||
e237882b | 6602 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6603 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6604 | return; |
6605 | } | |
e3fe70b1 RR |
6606 | |
6607 | for_each_online_node(a) { | |
6608 | for_each_online_node(b) { | |
6609 | /* Find two nodes furthest removed from each other. */ | |
6610 | if (node_distance(a, b) < n) | |
6611 | continue; | |
6612 | ||
6613 | /* Is there an intermediary node between a and b? */ | |
6614 | for_each_online_node(c) { | |
6615 | if (node_distance(a, c) < n && | |
6616 | node_distance(b, c) < n) { | |
6617 | sched_numa_topology_type = | |
6618 | NUMA_GLUELESS_MESH; | |
6619 | return; | |
6620 | } | |
6621 | } | |
6622 | ||
6623 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6624 | return; | |
6625 | } | |
6626 | } | |
6627 | } | |
6628 | ||
cb83b629 PZ |
6629 | static void sched_init_numa(void) |
6630 | { | |
6631 | int next_distance, curr_distance = node_distance(0, 0); | |
6632 | struct sched_domain_topology_level *tl; | |
6633 | int level = 0; | |
6634 | int i, j, k; | |
6635 | ||
cb83b629 PZ |
6636 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6637 | if (!sched_domains_numa_distance) | |
6638 | return; | |
6639 | ||
6640 | /* | |
6641 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6642 | * unique distances in the node_distance() table. | |
6643 | * | |
6644 | * Assumes node_distance(0,j) includes all distances in | |
6645 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6646 | */ |
6647 | next_distance = curr_distance; | |
6648 | for (i = 0; i < nr_node_ids; i++) { | |
6649 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6650 | for (k = 0; k < nr_node_ids; k++) { |
6651 | int distance = node_distance(i, k); | |
6652 | ||
6653 | if (distance > curr_distance && | |
6654 | (distance < next_distance || | |
6655 | next_distance == curr_distance)) | |
6656 | next_distance = distance; | |
6657 | ||
6658 | /* | |
6659 | * While not a strong assumption it would be nice to know | |
6660 | * about cases where if node A is connected to B, B is not | |
6661 | * equally connected to A. | |
6662 | */ | |
6663 | if (sched_debug() && node_distance(k, i) != distance) | |
6664 | sched_numa_warn("Node-distance not symmetric"); | |
6665 | ||
6666 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6667 | sched_numa_warn("Node-0 not representative"); | |
6668 | } | |
6669 | if (next_distance != curr_distance) { | |
6670 | sched_domains_numa_distance[level++] = next_distance; | |
6671 | sched_domains_numa_levels = level; | |
6672 | curr_distance = next_distance; | |
6673 | } else break; | |
cb83b629 | 6674 | } |
d039ac60 PZ |
6675 | |
6676 | /* | |
6677 | * In case of sched_debug() we verify the above assumption. | |
6678 | */ | |
6679 | if (!sched_debug()) | |
6680 | break; | |
cb83b629 | 6681 | } |
c123588b AR |
6682 | |
6683 | if (!level) | |
6684 | return; | |
6685 | ||
cb83b629 PZ |
6686 | /* |
6687 | * 'level' contains the number of unique distances, excluding the | |
6688 | * identity distance node_distance(i,i). | |
6689 | * | |
28b4a521 | 6690 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6691 | * numbers. |
6692 | */ | |
6693 | ||
5f7865f3 TC |
6694 | /* |
6695 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6696 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6697 | * the array will contain less then 'level' members. This could be | |
6698 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6699 | * in other functions. | |
6700 | * | |
6701 | * We reset it to 'level' at the end of this function. | |
6702 | */ | |
6703 | sched_domains_numa_levels = 0; | |
6704 | ||
cb83b629 PZ |
6705 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6706 | if (!sched_domains_numa_masks) | |
6707 | return; | |
6708 | ||
6709 | /* | |
6710 | * Now for each level, construct a mask per node which contains all | |
6711 | * cpus of nodes that are that many hops away from us. | |
6712 | */ | |
6713 | for (i = 0; i < level; i++) { | |
6714 | sched_domains_numa_masks[i] = | |
6715 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6716 | if (!sched_domains_numa_masks[i]) | |
6717 | return; | |
6718 | ||
6719 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6720 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6721 | if (!mask) |
6722 | return; | |
6723 | ||
6724 | sched_domains_numa_masks[i][j] = mask; | |
6725 | ||
9c03ee14 | 6726 | for_each_node(k) { |
dd7d8634 | 6727 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6728 | continue; |
6729 | ||
6730 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6731 | } | |
6732 | } | |
6733 | } | |
6734 | ||
143e1e28 VG |
6735 | /* Compute default topology size */ |
6736 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6737 | ||
c515db8c | 6738 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6739 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6740 | if (!tl) | |
6741 | return; | |
6742 | ||
6743 | /* | |
6744 | * Copy the default topology bits.. | |
6745 | */ | |
143e1e28 VG |
6746 | for (i = 0; sched_domain_topology[i].mask; i++) |
6747 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6748 | |
6749 | /* | |
6750 | * .. and append 'j' levels of NUMA goodness. | |
6751 | */ | |
6752 | for (j = 0; j < level; i++, j++) { | |
6753 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6754 | .mask = sd_numa_mask, |
143e1e28 | 6755 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6756 | .flags = SDTL_OVERLAP, |
6757 | .numa_level = j, | |
143e1e28 | 6758 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6759 | }; |
6760 | } | |
6761 | ||
6762 | sched_domain_topology = tl; | |
5f7865f3 TC |
6763 | |
6764 | sched_domains_numa_levels = level; | |
9942f79b | 6765 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6766 | |
6767 | init_numa_topology_type(); | |
cb83b629 | 6768 | } |
301a5cba | 6769 | |
135fb3e1 | 6770 | static void sched_domains_numa_masks_set(unsigned int cpu) |
301a5cba | 6771 | { |
301a5cba | 6772 | int node = cpu_to_node(cpu); |
135fb3e1 | 6773 | int i, j; |
301a5cba TC |
6774 | |
6775 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6776 | for (j = 0; j < nr_node_ids; j++) { | |
6777 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6778 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6779 | } | |
6780 | } | |
6781 | } | |
6782 | ||
135fb3e1 | 6783 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
301a5cba TC |
6784 | { |
6785 | int i, j; | |
135fb3e1 | 6786 | |
301a5cba TC |
6787 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6788 | for (j = 0; j < nr_node_ids; j++) | |
6789 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6790 | } | |
6791 | } | |
6792 | ||
cb83b629 | 6793 | #else |
135fb3e1 TG |
6794 | static inline void sched_init_numa(void) { } |
6795 | static void sched_domains_numa_masks_set(unsigned int cpu) { } | |
6796 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
cb83b629 PZ |
6797 | #endif /* CONFIG_NUMA */ |
6798 | ||
54ab4ff4 PZ |
6799 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6800 | { | |
6801 | struct sched_domain_topology_level *tl; | |
6802 | int j; | |
6803 | ||
27723a68 | 6804 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6805 | struct sd_data *sdd = &tl->data; |
6806 | ||
6807 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6808 | if (!sdd->sd) | |
6809 | return -ENOMEM; | |
6810 | ||
6811 | sdd->sg = alloc_percpu(struct sched_group *); | |
6812 | if (!sdd->sg) | |
6813 | return -ENOMEM; | |
6814 | ||
63b2ca30 NP |
6815 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6816 | if (!sdd->sgc) | |
9c3f75cb PZ |
6817 | return -ENOMEM; |
6818 | ||
54ab4ff4 PZ |
6819 | for_each_cpu(j, cpu_map) { |
6820 | struct sched_domain *sd; | |
6821 | struct sched_group *sg; | |
63b2ca30 | 6822 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6823 | |
5cc389bc | 6824 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6825 | GFP_KERNEL, cpu_to_node(j)); |
6826 | if (!sd) | |
6827 | return -ENOMEM; | |
6828 | ||
6829 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6830 | ||
6831 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6832 | GFP_KERNEL, cpu_to_node(j)); | |
6833 | if (!sg) | |
6834 | return -ENOMEM; | |
6835 | ||
30b4e9eb IM |
6836 | sg->next = sg; |
6837 | ||
54ab4ff4 | 6838 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6839 | |
63b2ca30 | 6840 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6841 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6842 | if (!sgc) |
9c3f75cb PZ |
6843 | return -ENOMEM; |
6844 | ||
63b2ca30 | 6845 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6846 | } |
6847 | } | |
6848 | ||
6849 | return 0; | |
6850 | } | |
6851 | ||
6852 | static void __sdt_free(const struct cpumask *cpu_map) | |
6853 | { | |
6854 | struct sched_domain_topology_level *tl; | |
6855 | int j; | |
6856 | ||
27723a68 | 6857 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6858 | struct sd_data *sdd = &tl->data; |
6859 | ||
6860 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6861 | struct sched_domain *sd; |
6862 | ||
6863 | if (sdd->sd) { | |
6864 | sd = *per_cpu_ptr(sdd->sd, j); | |
6865 | if (sd && (sd->flags & SD_OVERLAP)) | |
6866 | free_sched_groups(sd->groups, 0); | |
6867 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6868 | } | |
6869 | ||
6870 | if (sdd->sg) | |
6871 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6872 | if (sdd->sgc) |
6873 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6874 | } |
6875 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6876 | sdd->sd = NULL; |
54ab4ff4 | 6877 | free_percpu(sdd->sg); |
fb2cf2c6 | 6878 | sdd->sg = NULL; |
63b2ca30 NP |
6879 | free_percpu(sdd->sgc); |
6880 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6881 | } |
6882 | } | |
6883 | ||
2c402dc3 | 6884 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6885 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6886 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6887 | { |
3676b13e | 6888 | struct sched_domain *sd = sd_init(tl, child, cpu); |
2c402dc3 | 6889 | |
2c402dc3 | 6890 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6891 | if (child) { |
6892 | sd->level = child->level + 1; | |
6893 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6894 | child->parent = sd; |
6ae72dff PZ |
6895 | |
6896 | if (!cpumask_subset(sched_domain_span(child), | |
6897 | sched_domain_span(sd))) { | |
6898 | pr_err("BUG: arch topology borken\n"); | |
6899 | #ifdef CONFIG_SCHED_DEBUG | |
6900 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6901 | child->name, sd->name); | |
6902 | #endif | |
6903 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6904 | cpumask_or(sched_domain_span(sd), | |
6905 | sched_domain_span(sd), | |
6906 | sched_domain_span(child)); | |
6907 | } | |
6908 | ||
60495e77 | 6909 | } |
a841f8ce | 6910 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6911 | |
6912 | return sd; | |
6913 | } | |
6914 | ||
2109b99e AH |
6915 | /* |
6916 | * Build sched domains for a given set of cpus and attach the sched domains | |
6917 | * to the individual cpus | |
6918 | */ | |
dce840a0 PZ |
6919 | static int build_sched_domains(const struct cpumask *cpu_map, |
6920 | struct sched_domain_attr *attr) | |
2109b99e | 6921 | { |
1c632169 | 6922 | enum s_alloc alloc_state; |
dce840a0 | 6923 | struct sched_domain *sd; |
2109b99e | 6924 | struct s_data d; |
cd92bfd3 | 6925 | struct rq *rq = NULL; |
822ff793 | 6926 | int i, ret = -ENOMEM; |
9c1cfda2 | 6927 | |
2109b99e AH |
6928 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6929 | if (alloc_state != sa_rootdomain) | |
6930 | goto error; | |
9c1cfda2 | 6931 | |
dce840a0 | 6932 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6933 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6934 | struct sched_domain_topology_level *tl; |
6935 | ||
3bd65a80 | 6936 | sd = NULL; |
27723a68 | 6937 | for_each_sd_topology(tl) { |
4a850cbe | 6938 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6939 | if (tl == sched_domain_topology) |
6940 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6941 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6942 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6943 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6944 | break; | |
e3589f6c | 6945 | } |
dce840a0 PZ |
6946 | } |
6947 | ||
6948 | /* Build the groups for the domains */ | |
6949 | for_each_cpu(i, cpu_map) { | |
6950 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6951 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6952 | if (sd->flags & SD_OVERLAP) { |
6953 | if (build_overlap_sched_groups(sd, i)) | |
6954 | goto error; | |
6955 | } else { | |
6956 | if (build_sched_groups(sd, i)) | |
6957 | goto error; | |
6958 | } | |
1cf51902 | 6959 | } |
a06dadbe | 6960 | } |
9c1cfda2 | 6961 | |
ced549fa | 6962 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6963 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6964 | if (!cpumask_test_cpu(i, cpu_map)) | |
6965 | continue; | |
9c1cfda2 | 6966 | |
dce840a0 PZ |
6967 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6968 | claim_allocations(i, sd); | |
63b2ca30 | 6969 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6970 | } |
f712c0c7 | 6971 | } |
9c1cfda2 | 6972 | |
1da177e4 | 6973 | /* Attach the domains */ |
dce840a0 | 6974 | rcu_read_lock(); |
abcd083a | 6975 | for_each_cpu(i, cpu_map) { |
cd92bfd3 | 6976 | rq = cpu_rq(i); |
21d42ccf | 6977 | sd = *per_cpu_ptr(d.sd, i); |
cd92bfd3 DE |
6978 | |
6979 | /* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */ | |
6980 | if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity)) | |
6981 | WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig); | |
6982 | ||
49a02c51 | 6983 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6984 | } |
dce840a0 | 6985 | rcu_read_unlock(); |
51888ca2 | 6986 | |
cd92bfd3 DE |
6987 | if (rq) { |
6988 | pr_info("span: %*pbl (max cpu_capacity = %lu)\n", | |
6989 | cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity); | |
6990 | } | |
6991 | ||
822ff793 | 6992 | ret = 0; |
51888ca2 | 6993 | error: |
2109b99e | 6994 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6995 | return ret; |
1da177e4 | 6996 | } |
029190c5 | 6997 | |
acc3f5d7 | 6998 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6999 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7000 | static struct sched_domain_attr *dattr_cur; |
7001 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7002 | |
7003 | /* | |
7004 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7005 | * cpumask) fails, then fallback to a single sched domain, |
7006 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7007 | */ |
4212823f | 7008 | static cpumask_var_t fallback_doms; |
029190c5 | 7009 | |
ee79d1bd HC |
7010 | /* |
7011 | * arch_update_cpu_topology lets virtualized architectures update the | |
7012 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7013 | * or 0 if it stayed the same. | |
7014 | */ | |
52f5684c | 7015 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 7016 | { |
ee79d1bd | 7017 | return 0; |
22e52b07 HC |
7018 | } |
7019 | ||
acc3f5d7 RR |
7020 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7021 | { | |
7022 | int i; | |
7023 | cpumask_var_t *doms; | |
7024 | ||
7025 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7026 | if (!doms) | |
7027 | return NULL; | |
7028 | for (i = 0; i < ndoms; i++) { | |
7029 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7030 | free_sched_domains(doms, i); | |
7031 | return NULL; | |
7032 | } | |
7033 | } | |
7034 | return doms; | |
7035 | } | |
7036 | ||
7037 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7038 | { | |
7039 | unsigned int i; | |
7040 | for (i = 0; i < ndoms; i++) | |
7041 | free_cpumask_var(doms[i]); | |
7042 | kfree(doms); | |
7043 | } | |
7044 | ||
1a20ff27 | 7045 | /* |
41a2d6cf | 7046 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7047 | * For now this just excludes isolated cpus, but could be used to |
7048 | * exclude other special cases in the future. | |
1a20ff27 | 7049 | */ |
c4a8849a | 7050 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7051 | { |
7378547f MM |
7052 | int err; |
7053 | ||
22e52b07 | 7054 | arch_update_cpu_topology(); |
029190c5 | 7055 | ndoms_cur = 1; |
acc3f5d7 | 7056 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7057 | if (!doms_cur) |
acc3f5d7 RR |
7058 | doms_cur = &fallback_doms; |
7059 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 7060 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7061 | register_sched_domain_sysctl(); |
7378547f MM |
7062 | |
7063 | return err; | |
1a20ff27 DG |
7064 | } |
7065 | ||
1a20ff27 DG |
7066 | /* |
7067 | * Detach sched domains from a group of cpus specified in cpu_map | |
7068 | * These cpus will now be attached to the NULL domain | |
7069 | */ | |
96f874e2 | 7070 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7071 | { |
7072 | int i; | |
7073 | ||
dce840a0 | 7074 | rcu_read_lock(); |
abcd083a | 7075 | for_each_cpu(i, cpu_map) |
57d885fe | 7076 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7077 | rcu_read_unlock(); |
1a20ff27 DG |
7078 | } |
7079 | ||
1d3504fc HS |
7080 | /* handle null as "default" */ |
7081 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7082 | struct sched_domain_attr *new, int idx_new) | |
7083 | { | |
7084 | struct sched_domain_attr tmp; | |
7085 | ||
7086 | /* fast path */ | |
7087 | if (!new && !cur) | |
7088 | return 1; | |
7089 | ||
7090 | tmp = SD_ATTR_INIT; | |
7091 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7092 | new ? (new + idx_new) : &tmp, | |
7093 | sizeof(struct sched_domain_attr)); | |
7094 | } | |
7095 | ||
029190c5 PJ |
7096 | /* |
7097 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7098 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7099 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7100 | * It destroys each deleted domain and builds each new domain. | |
7101 | * | |
acc3f5d7 | 7102 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7103 | * The masks don't intersect (don't overlap.) We should setup one |
7104 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7105 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7106 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7107 | * it as it is. | |
7108 | * | |
acc3f5d7 RR |
7109 | * The passed in 'doms_new' should be allocated using |
7110 | * alloc_sched_domains. This routine takes ownership of it and will | |
7111 | * free_sched_domains it when done with it. If the caller failed the | |
7112 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7113 | * and partition_sched_domains() will fallback to the single partition | |
7114 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7115 | * |
96f874e2 | 7116 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7117 | * ndoms_new == 0 is a special case for destroying existing domains, |
7118 | * and it will not create the default domain. | |
dfb512ec | 7119 | * |
029190c5 PJ |
7120 | * Call with hotplug lock held |
7121 | */ | |
acc3f5d7 | 7122 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7123 | struct sched_domain_attr *dattr_new) |
029190c5 | 7124 | { |
dfb512ec | 7125 | int i, j, n; |
d65bd5ec | 7126 | int new_topology; |
029190c5 | 7127 | |
712555ee | 7128 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7129 | |
7378547f MM |
7130 | /* always unregister in case we don't destroy any domains */ |
7131 | unregister_sched_domain_sysctl(); | |
7132 | ||
d65bd5ec HC |
7133 | /* Let architecture update cpu core mappings. */ |
7134 | new_topology = arch_update_cpu_topology(); | |
7135 | ||
dfb512ec | 7136 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7137 | |
7138 | /* Destroy deleted domains */ | |
7139 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7140 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7141 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7142 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7143 | goto match1; |
7144 | } | |
7145 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7146 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7147 | match1: |
7148 | ; | |
7149 | } | |
7150 | ||
c8d2d47a | 7151 | n = ndoms_cur; |
e761b772 | 7152 | if (doms_new == NULL) { |
c8d2d47a | 7153 | n = 0; |
acc3f5d7 | 7154 | doms_new = &fallback_doms; |
6ad4c188 | 7155 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7156 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7157 | } |
7158 | ||
029190c5 PJ |
7159 | /* Build new domains */ |
7160 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7161 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7162 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7163 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7164 | goto match2; |
7165 | } | |
7166 | /* no match - add a new doms_new */ | |
dce840a0 | 7167 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7168 | match2: |
7169 | ; | |
7170 | } | |
7171 | ||
7172 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7173 | if (doms_cur != &fallback_doms) |
7174 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7175 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7176 | doms_cur = doms_new; |
1d3504fc | 7177 | dattr_cur = dattr_new; |
029190c5 | 7178 | ndoms_cur = ndoms_new; |
7378547f MM |
7179 | |
7180 | register_sched_domain_sysctl(); | |
a1835615 | 7181 | |
712555ee | 7182 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7183 | } |
7184 | ||
d35be8ba SB |
7185 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7186 | ||
1da177e4 | 7187 | /* |
3a101d05 TH |
7188 | * Update cpusets according to cpu_active mask. If cpusets are |
7189 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7190 | * around partition_sched_domains(). | |
d35be8ba SB |
7191 | * |
7192 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7193 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7194 | */ |
40190a78 | 7195 | static void cpuset_cpu_active(void) |
e761b772 | 7196 | { |
40190a78 | 7197 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7198 | /* |
7199 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7200 | * resume sequence. As long as this is not the last online | |
7201 | * operation in the resume sequence, just build a single sched | |
7202 | * domain, ignoring cpusets. | |
7203 | */ | |
7204 | num_cpus_frozen--; | |
7205 | if (likely(num_cpus_frozen)) { | |
7206 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 7207 | return; |
d35be8ba | 7208 | } |
d35be8ba SB |
7209 | /* |
7210 | * This is the last CPU online operation. So fall through and | |
7211 | * restore the original sched domains by considering the | |
7212 | * cpuset configurations. | |
7213 | */ | |
3a101d05 | 7214 | } |
135fb3e1 | 7215 | cpuset_update_active_cpus(true); |
3a101d05 | 7216 | } |
e761b772 | 7217 | |
40190a78 | 7218 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7219 | { |
3c18d447 | 7220 | unsigned long flags; |
3c18d447 | 7221 | struct dl_bw *dl_b; |
533445c6 OS |
7222 | bool overflow; |
7223 | int cpus; | |
3c18d447 | 7224 | |
40190a78 | 7225 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
7226 | rcu_read_lock_sched(); |
7227 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7228 | |
533445c6 OS |
7229 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7230 | cpus = dl_bw_cpus(cpu); | |
7231 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7232 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7233 | |
533445c6 | 7234 | rcu_read_unlock_sched(); |
3c18d447 | 7235 | |
533445c6 | 7236 | if (overflow) |
135fb3e1 | 7237 | return -EBUSY; |
7ddf96b0 | 7238 | cpuset_update_active_cpus(false); |
135fb3e1 | 7239 | } else { |
d35be8ba SB |
7240 | num_cpus_frozen++; |
7241 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7242 | } |
135fb3e1 | 7243 | return 0; |
e761b772 | 7244 | } |
e761b772 | 7245 | |
40190a78 | 7246 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7247 | { |
7d976699 TG |
7248 | struct rq *rq = cpu_rq(cpu); |
7249 | unsigned long flags; | |
7250 | ||
40190a78 | 7251 | set_cpu_active(cpu, true); |
135fb3e1 | 7252 | |
40190a78 | 7253 | if (sched_smp_initialized) { |
135fb3e1 | 7254 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7255 | cpuset_cpu_active(); |
e761b772 | 7256 | } |
7d976699 TG |
7257 | |
7258 | /* | |
7259 | * Put the rq online, if not already. This happens: | |
7260 | * | |
7261 | * 1) In the early boot process, because we build the real domains | |
7262 | * after all cpus have been brought up. | |
7263 | * | |
7264 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7265 | * domains. | |
7266 | */ | |
7267 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7268 | if (rq->rd) { | |
7269 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7270 | set_rq_online(rq); | |
7271 | } | |
7272 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7273 | ||
7274 | update_max_interval(); | |
7275 | ||
40190a78 | 7276 | return 0; |
135fb3e1 TG |
7277 | } |
7278 | ||
40190a78 | 7279 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7280 | { |
135fb3e1 TG |
7281 | int ret; |
7282 | ||
40190a78 | 7283 | set_cpu_active(cpu, false); |
b2454caa PZ |
7284 | /* |
7285 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7286 | * users of this state to go away such that all new such users will | |
7287 | * observe it. | |
7288 | * | |
7289 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
7290 | * not imply sync_sched(), so wait for both. | |
7291 | * | |
7292 | * Do sync before park smpboot threads to take care the rcu boost case. | |
7293 | */ | |
7294 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
7295 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
7296 | else | |
7297 | synchronize_rcu(); | |
40190a78 TG |
7298 | |
7299 | if (!sched_smp_initialized) | |
7300 | return 0; | |
7301 | ||
7302 | ret = cpuset_cpu_inactive(cpu); | |
7303 | if (ret) { | |
7304 | set_cpu_active(cpu, true); | |
7305 | return ret; | |
135fb3e1 | 7306 | } |
40190a78 TG |
7307 | sched_domains_numa_masks_clear(cpu); |
7308 | return 0; | |
135fb3e1 TG |
7309 | } |
7310 | ||
94baf7a5 TG |
7311 | static void sched_rq_cpu_starting(unsigned int cpu) |
7312 | { | |
7313 | struct rq *rq = cpu_rq(cpu); | |
7314 | ||
7315 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7316 | update_max_interval(); |
7317 | } | |
7318 | ||
135fb3e1 TG |
7319 | int sched_cpu_starting(unsigned int cpu) |
7320 | { | |
7321 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 7322 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 7323 | return 0; |
e761b772 | 7324 | } |
e761b772 | 7325 | |
f2785ddb TG |
7326 | #ifdef CONFIG_HOTPLUG_CPU |
7327 | int sched_cpu_dying(unsigned int cpu) | |
7328 | { | |
7329 | struct rq *rq = cpu_rq(cpu); | |
7330 | unsigned long flags; | |
7331 | ||
7332 | /* Handle pending wakeups and then migrate everything off */ | |
7333 | sched_ttwu_pending(); | |
7334 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7335 | if (rq->rd) { | |
7336 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7337 | set_rq_offline(rq); | |
7338 | } | |
7339 | migrate_tasks(rq); | |
7340 | BUG_ON(rq->nr_running != 1); | |
7341 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7342 | calc_load_migrate(rq); | |
7343 | update_max_interval(); | |
20a5c8cc | 7344 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 7345 | hrtick_clear(rq); |
f2785ddb TG |
7346 | return 0; |
7347 | } | |
7348 | #endif | |
7349 | ||
1da177e4 LT |
7350 | void __init sched_init_smp(void) |
7351 | { | |
dcc30a35 RR |
7352 | cpumask_var_t non_isolated_cpus; |
7353 | ||
7354 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7355 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7356 | |
cb83b629 PZ |
7357 | sched_init_numa(); |
7358 | ||
6acce3ef PZ |
7359 | /* |
7360 | * There's no userspace yet to cause hotplug operations; hence all the | |
7361 | * cpu masks are stable and all blatant races in the below code cannot | |
7362 | * happen. | |
7363 | */ | |
712555ee | 7364 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7365 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7366 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7367 | if (cpumask_empty(non_isolated_cpus)) | |
7368 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7369 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7370 | |
5c1e1767 | 7371 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 7372 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7373 | BUG(); |
19978ca6 | 7374 | sched_init_granularity(); |
dcc30a35 | 7375 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7376 | |
0e3900e6 | 7377 | init_sched_rt_class(); |
1baca4ce | 7378 | init_sched_dl_class(); |
e26fbffd | 7379 | sched_smp_initialized = true; |
1da177e4 | 7380 | } |
e26fbffd TG |
7381 | |
7382 | static int __init migration_init(void) | |
7383 | { | |
94baf7a5 | 7384 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 7385 | return 0; |
1da177e4 | 7386 | } |
e26fbffd TG |
7387 | early_initcall(migration_init); |
7388 | ||
1da177e4 LT |
7389 | #else |
7390 | void __init sched_init_smp(void) | |
7391 | { | |
19978ca6 | 7392 | sched_init_granularity(); |
1da177e4 LT |
7393 | } |
7394 | #endif /* CONFIG_SMP */ | |
7395 | ||
7396 | int in_sched_functions(unsigned long addr) | |
7397 | { | |
1da177e4 LT |
7398 | return in_lock_functions(addr) || |
7399 | (addr >= (unsigned long)__sched_text_start | |
7400 | && addr < (unsigned long)__sched_text_end); | |
7401 | } | |
7402 | ||
029632fb | 7403 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7404 | /* |
7405 | * Default task group. | |
7406 | * Every task in system belongs to this group at bootup. | |
7407 | */ | |
029632fb | 7408 | struct task_group root_task_group; |
35cf4e50 | 7409 | LIST_HEAD(task_groups); |
b0367629 WL |
7410 | |
7411 | /* Cacheline aligned slab cache for task_group */ | |
7412 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7413 | #endif |
6f505b16 | 7414 | |
e6252c3e | 7415 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7416 | |
1da177e4 LT |
7417 | void __init sched_init(void) |
7418 | { | |
dd41f596 | 7419 | int i, j; |
434d53b0 MT |
7420 | unsigned long alloc_size = 0, ptr; |
7421 | ||
7422 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7423 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7424 | #endif | |
7425 | #ifdef CONFIG_RT_GROUP_SCHED | |
7426 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7427 | #endif | |
434d53b0 | 7428 | if (alloc_size) { |
36b7b6d4 | 7429 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7430 | |
7431 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7432 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7433 | ptr += nr_cpu_ids * sizeof(void **); |
7434 | ||
07e06b01 | 7435 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7436 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7437 | |
6d6bc0ad | 7438 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7439 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7440 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7441 | ptr += nr_cpu_ids * sizeof(void **); |
7442 | ||
07e06b01 | 7443 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7444 | ptr += nr_cpu_ids * sizeof(void **); |
7445 | ||
6d6bc0ad | 7446 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7447 | } |
df7c8e84 | 7448 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7449 | for_each_possible_cpu(i) { |
7450 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7451 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7452 | } |
b74e6278 | 7453 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7454 | |
332ac17e DF |
7455 | init_rt_bandwidth(&def_rt_bandwidth, |
7456 | global_rt_period(), global_rt_runtime()); | |
7457 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7458 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7459 | |
57d885fe GH |
7460 | #ifdef CONFIG_SMP |
7461 | init_defrootdomain(); | |
7462 | #endif | |
7463 | ||
d0b27fa7 | 7464 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7465 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7466 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7467 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7468 | |
7c941438 | 7469 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7470 | task_group_cache = KMEM_CACHE(task_group, 0); |
7471 | ||
07e06b01 YZ |
7472 | list_add(&root_task_group.list, &task_groups); |
7473 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7474 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7475 | autogroup_init(&init_task); |
7c941438 | 7476 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7477 | |
0a945022 | 7478 | for_each_possible_cpu(i) { |
70b97a7f | 7479 | struct rq *rq; |
1da177e4 LT |
7480 | |
7481 | rq = cpu_rq(i); | |
05fa785c | 7482 | raw_spin_lock_init(&rq->lock); |
7897986b | 7483 | rq->nr_running = 0; |
dce48a84 TG |
7484 | rq->calc_load_active = 0; |
7485 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7486 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7487 | init_rt_rq(&rq->rt); |
7488 | init_dl_rq(&rq->dl); | |
dd41f596 | 7489 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7490 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7491 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7492 | /* |
07e06b01 | 7493 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7494 | * |
7495 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7496 | * gets 100% of the cpu resources in the system. This overall | |
7497 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7498 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7499 | * based on each entity's (task or task-group's) weight |
7500 | * (se->load.weight). | |
7501 | * | |
07e06b01 | 7502 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7503 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7504 | * then A0's share of the cpu resource is: | |
7505 | * | |
0d905bca | 7506 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7507 | * |
07e06b01 YZ |
7508 | * We achieve this by letting root_task_group's tasks sit |
7509 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7510 | */ |
ab84d31e | 7511 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7512 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7513 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7514 | ||
7515 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7516 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7517 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7518 | #endif |
1da177e4 | 7519 | |
dd41f596 IM |
7520 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7521 | rq->cpu_load[j] = 0; | |
fdf3e95d | 7522 | |
1da177e4 | 7523 | #ifdef CONFIG_SMP |
41c7ce9a | 7524 | rq->sd = NULL; |
57d885fe | 7525 | rq->rd = NULL; |
ca6d75e6 | 7526 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7527 | rq->balance_callback = NULL; |
1da177e4 | 7528 | rq->active_balance = 0; |
dd41f596 | 7529 | rq->next_balance = jiffies; |
1da177e4 | 7530 | rq->push_cpu = 0; |
0a2966b4 | 7531 | rq->cpu = i; |
1f11eb6a | 7532 | rq->online = 0; |
eae0c9df MG |
7533 | rq->idle_stamp = 0; |
7534 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7535 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7536 | |
7537 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7538 | ||
dc938520 | 7539 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7540 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 7541 | rq->last_load_update_tick = jiffies; |
1c792db7 | 7542 | rq->nohz_flags = 0; |
83cd4fe2 | 7543 | #endif |
265f22a9 FW |
7544 | #ifdef CONFIG_NO_HZ_FULL |
7545 | rq->last_sched_tick = 0; | |
7546 | #endif | |
9fd81dd5 | 7547 | #endif /* CONFIG_SMP */ |
8f4d37ec | 7548 | init_rq_hrtick(rq); |
1da177e4 | 7549 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7550 | } |
7551 | ||
2dd73a4f | 7552 | set_load_weight(&init_task); |
b50f60ce | 7553 | |
1da177e4 LT |
7554 | /* |
7555 | * The boot idle thread does lazy MMU switching as well: | |
7556 | */ | |
7557 | atomic_inc(&init_mm.mm_count); | |
7558 | enter_lazy_tlb(&init_mm, current); | |
7559 | ||
7560 | /* | |
7561 | * Make us the idle thread. Technically, schedule() should not be | |
7562 | * called from this thread, however somewhere below it might be, | |
7563 | * but because we are the idle thread, we just pick up running again | |
7564 | * when this runqueue becomes "idle". | |
7565 | */ | |
7566 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7567 | |
7568 | calc_load_update = jiffies + LOAD_FREQ; | |
7569 | ||
bf4d83f6 | 7570 | #ifdef CONFIG_SMP |
4cb98839 | 7571 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7572 | /* May be allocated at isolcpus cmdline parse time */ |
7573 | if (cpu_isolated_map == NULL) | |
7574 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7575 | idle_thread_set_boot_cpu(); |
9cf7243d | 7576 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
7577 | #endif |
7578 | init_sched_fair_class(); | |
6a7b3dc3 | 7579 | |
4698f88c JP |
7580 | init_schedstats(); |
7581 | ||
6892b75e | 7582 | scheduler_running = 1; |
1da177e4 LT |
7583 | } |
7584 | ||
d902db1e | 7585 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7586 | static inline int preempt_count_equals(int preempt_offset) |
7587 | { | |
da7142e2 | 7588 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7589 | |
4ba8216c | 7590 | return (nested == preempt_offset); |
e4aafea2 FW |
7591 | } |
7592 | ||
d894837f | 7593 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7594 | { |
8eb23b9f PZ |
7595 | /* |
7596 | * Blocking primitives will set (and therefore destroy) current->state, | |
7597 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7598 | * otherwise we will destroy state. | |
7599 | */ | |
00845eb9 | 7600 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7601 | "do not call blocking ops when !TASK_RUNNING; " |
7602 | "state=%lx set at [<%p>] %pS\n", | |
7603 | current->state, | |
7604 | (void *)current->task_state_change, | |
00845eb9 | 7605 | (void *)current->task_state_change); |
8eb23b9f | 7606 | |
3427445a PZ |
7607 | ___might_sleep(file, line, preempt_offset); |
7608 | } | |
7609 | EXPORT_SYMBOL(__might_sleep); | |
7610 | ||
7611 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7612 | { |
1da177e4 | 7613 | static unsigned long prev_jiffy; /* ratelimiting */ |
d1c6d149 | 7614 | unsigned long preempt_disable_ip; |
1da177e4 | 7615 | |
b3fbab05 | 7616 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7617 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7618 | !is_idle_task(current)) || | |
e4aafea2 | 7619 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7620 | return; |
7621 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7622 | return; | |
7623 | prev_jiffy = jiffies; | |
7624 | ||
d1c6d149 VN |
7625 | /* Save this before calling printk(), since that will clobber it */ |
7626 | preempt_disable_ip = get_preempt_disable_ip(current); | |
7627 | ||
3df0fc5b PZ |
7628 | printk(KERN_ERR |
7629 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7630 | file, line); | |
7631 | printk(KERN_ERR | |
7632 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7633 | in_atomic(), irqs_disabled(), | |
7634 | current->pid, current->comm); | |
aef745fc | 7635 | |
a8b686b3 ES |
7636 | if (task_stack_end_corrupted(current)) |
7637 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7638 | ||
aef745fc IM |
7639 | debug_show_held_locks(current); |
7640 | if (irqs_disabled()) | |
7641 | print_irqtrace_events(current); | |
d1c6d149 VN |
7642 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7643 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7644 | pr_err("Preemption disabled at:"); |
d1c6d149 | 7645 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
7646 | pr_cont("\n"); |
7647 | } | |
aef745fc | 7648 | dump_stack(); |
f0b22e39 | 7649 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7650 | } |
3427445a | 7651 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7652 | #endif |
7653 | ||
7654 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7655 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7656 | { |
dbc7f069 | 7657 | struct task_struct *g, *p; |
d50dde5a DF |
7658 | struct sched_attr attr = { |
7659 | .sched_policy = SCHED_NORMAL, | |
7660 | }; | |
1da177e4 | 7661 | |
3472eaa1 | 7662 | read_lock(&tasklist_lock); |
5d07f420 | 7663 | for_each_process_thread(g, p) { |
178be793 IM |
7664 | /* |
7665 | * Only normalize user tasks: | |
7666 | */ | |
3472eaa1 | 7667 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7668 | continue; |
7669 | ||
4fa8d299 JP |
7670 | p->se.exec_start = 0; |
7671 | schedstat_set(p->se.statistics.wait_start, 0); | |
7672 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7673 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7674 | |
aab03e05 | 7675 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7676 | /* |
7677 | * Renice negative nice level userspace | |
7678 | * tasks back to 0: | |
7679 | */ | |
3472eaa1 | 7680 | if (task_nice(p) < 0) |
dd41f596 | 7681 | set_user_nice(p, 0); |
1da177e4 | 7682 | continue; |
dd41f596 | 7683 | } |
1da177e4 | 7684 | |
dbc7f069 | 7685 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7686 | } |
3472eaa1 | 7687 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7688 | } |
7689 | ||
7690 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7691 | |
67fc4e0c | 7692 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7693 | /* |
67fc4e0c | 7694 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7695 | * |
7696 | * They can only be called when the whole system has been | |
7697 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7698 | * activity can take place. Using them for anything else would | |
7699 | * be a serious bug, and as a result, they aren't even visible | |
7700 | * under any other configuration. | |
7701 | */ | |
7702 | ||
7703 | /** | |
7704 | * curr_task - return the current task for a given cpu. | |
7705 | * @cpu: the processor in question. | |
7706 | * | |
7707 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7708 | * |
7709 | * Return: The current task for @cpu. | |
1df5c10a | 7710 | */ |
36c8b586 | 7711 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7712 | { |
7713 | return cpu_curr(cpu); | |
7714 | } | |
7715 | ||
67fc4e0c JW |
7716 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7717 | ||
7718 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7719 | /** |
7720 | * set_curr_task - set the current task for a given cpu. | |
7721 | * @cpu: the processor in question. | |
7722 | * @p: the task pointer to set. | |
7723 | * | |
7724 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7725 | * are serviced on a separate stack. It allows the architecture to switch the |
7726 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7727 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7728 | * and caller must save the original value of the current task (see | |
7729 | * curr_task() above) and restore that value before reenabling interrupts and | |
7730 | * re-starting the system. | |
7731 | * | |
7732 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7733 | */ | |
36c8b586 | 7734 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7735 | { |
7736 | cpu_curr(cpu) = p; | |
7737 | } | |
7738 | ||
7739 | #endif | |
29f59db3 | 7740 | |
7c941438 | 7741 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7742 | /* task_group_lock serializes the addition/removal of task groups */ |
7743 | static DEFINE_SPINLOCK(task_group_lock); | |
7744 | ||
2f5177f0 | 7745 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7746 | { |
7747 | free_fair_sched_group(tg); | |
7748 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7749 | autogroup_free(tg); |
b0367629 | 7750 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7751 | } |
7752 | ||
7753 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7754 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7755 | { |
7756 | struct task_group *tg; | |
bccbe08a | 7757 | |
b0367629 | 7758 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7759 | if (!tg) |
7760 | return ERR_PTR(-ENOMEM); | |
7761 | ||
ec7dc8ac | 7762 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7763 | goto err; |
7764 | ||
ec7dc8ac | 7765 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7766 | goto err; |
7767 | ||
ace783b9 LZ |
7768 | return tg; |
7769 | ||
7770 | err: | |
2f5177f0 | 7771 | sched_free_group(tg); |
ace783b9 LZ |
7772 | return ERR_PTR(-ENOMEM); |
7773 | } | |
7774 | ||
7775 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7776 | { | |
7777 | unsigned long flags; | |
7778 | ||
8ed36996 | 7779 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7780 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7781 | |
7782 | WARN_ON(!parent); /* root should already exist */ | |
7783 | ||
7784 | tg->parent = parent; | |
f473aa5e | 7785 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7786 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7787 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
7788 | |
7789 | online_fair_sched_group(tg); | |
29f59db3 SV |
7790 | } |
7791 | ||
9b5b7751 | 7792 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7793 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7794 | { |
29f59db3 | 7795 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 7796 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7797 | } |
7798 | ||
4cf86d77 | 7799 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7800 | { |
7801 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 7802 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7803 | } |
7804 | ||
7805 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7806 | { |
8ed36996 | 7807 | unsigned long flags; |
29f59db3 | 7808 | |
3d4b47b4 | 7809 | /* end participation in shares distribution */ |
6fe1f348 | 7810 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7811 | |
7812 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7813 | list_del_rcu(&tg->list); |
f473aa5e | 7814 | list_del_rcu(&tg->siblings); |
8ed36996 | 7815 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7816 | } |
7817 | ||
ea86cb4b | 7818 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 7819 | { |
8323f26c | 7820 | struct task_group *tg; |
29f59db3 | 7821 | |
f7b8a47d KT |
7822 | /* |
7823 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7824 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7825 | * to prevent lockdep warnings. | |
7826 | */ | |
7827 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7828 | struct task_group, css); |
7829 | tg = autogroup_task_group(tsk, tg); | |
7830 | tsk->sched_task_group = tg; | |
7831 | ||
810b3817 | 7832 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
7833 | if (tsk->sched_class->task_change_group) |
7834 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 7835 | else |
810b3817 | 7836 | #endif |
b2b5ce02 | 7837 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
7838 | } |
7839 | ||
7840 | /* | |
7841 | * Change task's runqueue when it moves between groups. | |
7842 | * | |
7843 | * The caller of this function should have put the task in its new group by | |
7844 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
7845 | * its new group. | |
7846 | */ | |
7847 | void sched_move_task(struct task_struct *tsk) | |
7848 | { | |
7849 | int queued, running; | |
7850 | struct rq_flags rf; | |
7851 | struct rq *rq; | |
7852 | ||
7853 | rq = task_rq_lock(tsk, &rf); | |
7854 | ||
7855 | running = task_current(rq, tsk); | |
7856 | queued = task_on_rq_queued(tsk); | |
7857 | ||
7858 | if (queued) | |
7859 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); | |
7860 | if (unlikely(running)) | |
7861 | put_prev_task(rq, tsk); | |
7862 | ||
7863 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 7864 | |
0e1f3483 HS |
7865 | if (unlikely(running)) |
7866 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7867 | if (queued) |
ff77e468 | 7868 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
29f59db3 | 7869 | |
eb580751 | 7870 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 7871 | } |
7c941438 | 7872 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7873 | |
a790de99 PT |
7874 | #ifdef CONFIG_RT_GROUP_SCHED |
7875 | /* | |
7876 | * Ensure that the real time constraints are schedulable. | |
7877 | */ | |
7878 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7879 | |
9a7e0b18 PZ |
7880 | /* Must be called with tasklist_lock held */ |
7881 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7882 | { |
9a7e0b18 | 7883 | struct task_struct *g, *p; |
b40b2e8e | 7884 | |
1fe89e1b PZ |
7885 | /* |
7886 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7887 | */ | |
7888 | if (task_group_is_autogroup(tg)) | |
7889 | return 0; | |
7890 | ||
5d07f420 | 7891 | for_each_process_thread(g, p) { |
8651c658 | 7892 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7893 | return 1; |
5d07f420 | 7894 | } |
b40b2e8e | 7895 | |
9a7e0b18 PZ |
7896 | return 0; |
7897 | } | |
b40b2e8e | 7898 | |
9a7e0b18 PZ |
7899 | struct rt_schedulable_data { |
7900 | struct task_group *tg; | |
7901 | u64 rt_period; | |
7902 | u64 rt_runtime; | |
7903 | }; | |
b40b2e8e | 7904 | |
a790de99 | 7905 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7906 | { |
7907 | struct rt_schedulable_data *d = data; | |
7908 | struct task_group *child; | |
7909 | unsigned long total, sum = 0; | |
7910 | u64 period, runtime; | |
b40b2e8e | 7911 | |
9a7e0b18 PZ |
7912 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7913 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7914 | |
9a7e0b18 PZ |
7915 | if (tg == d->tg) { |
7916 | period = d->rt_period; | |
7917 | runtime = d->rt_runtime; | |
b40b2e8e | 7918 | } |
b40b2e8e | 7919 | |
4653f803 PZ |
7920 | /* |
7921 | * Cannot have more runtime than the period. | |
7922 | */ | |
7923 | if (runtime > period && runtime != RUNTIME_INF) | |
7924 | return -EINVAL; | |
6f505b16 | 7925 | |
4653f803 PZ |
7926 | /* |
7927 | * Ensure we don't starve existing RT tasks. | |
7928 | */ | |
9a7e0b18 PZ |
7929 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7930 | return -EBUSY; | |
6f505b16 | 7931 | |
9a7e0b18 | 7932 | total = to_ratio(period, runtime); |
6f505b16 | 7933 | |
4653f803 PZ |
7934 | /* |
7935 | * Nobody can have more than the global setting allows. | |
7936 | */ | |
7937 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7938 | return -EINVAL; | |
6f505b16 | 7939 | |
4653f803 PZ |
7940 | /* |
7941 | * The sum of our children's runtime should not exceed our own. | |
7942 | */ | |
9a7e0b18 PZ |
7943 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7944 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7945 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7946 | |
9a7e0b18 PZ |
7947 | if (child == d->tg) { |
7948 | period = d->rt_period; | |
7949 | runtime = d->rt_runtime; | |
7950 | } | |
6f505b16 | 7951 | |
9a7e0b18 | 7952 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7953 | } |
6f505b16 | 7954 | |
9a7e0b18 PZ |
7955 | if (sum > total) |
7956 | return -EINVAL; | |
7957 | ||
7958 | return 0; | |
6f505b16 PZ |
7959 | } |
7960 | ||
9a7e0b18 | 7961 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7962 | { |
8277434e PT |
7963 | int ret; |
7964 | ||
9a7e0b18 PZ |
7965 | struct rt_schedulable_data data = { |
7966 | .tg = tg, | |
7967 | .rt_period = period, | |
7968 | .rt_runtime = runtime, | |
7969 | }; | |
7970 | ||
8277434e PT |
7971 | rcu_read_lock(); |
7972 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7973 | rcu_read_unlock(); | |
7974 | ||
7975 | return ret; | |
521f1a24 DG |
7976 | } |
7977 | ||
ab84d31e | 7978 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7979 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7980 | { |
ac086bc2 | 7981 | int i, err = 0; |
9f0c1e56 | 7982 | |
2636ed5f PZ |
7983 | /* |
7984 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7985 | * kernel creating (and or operating) RT threads. | |
7986 | */ | |
7987 | if (tg == &root_task_group && rt_runtime == 0) | |
7988 | return -EINVAL; | |
7989 | ||
7990 | /* No period doesn't make any sense. */ | |
7991 | if (rt_period == 0) | |
7992 | return -EINVAL; | |
7993 | ||
9f0c1e56 | 7994 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7995 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7996 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7997 | if (err) | |
9f0c1e56 | 7998 | goto unlock; |
ac086bc2 | 7999 | |
0986b11b | 8000 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8001 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8002 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8003 | |
8004 | for_each_possible_cpu(i) { | |
8005 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8006 | ||
0986b11b | 8007 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8008 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8009 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8010 | } |
0986b11b | 8011 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8012 | unlock: |
521f1a24 | 8013 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8014 | mutex_unlock(&rt_constraints_mutex); |
8015 | ||
8016 | return err; | |
6f505b16 PZ |
8017 | } |
8018 | ||
25cc7da7 | 8019 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
8020 | { |
8021 | u64 rt_runtime, rt_period; | |
8022 | ||
8023 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8024 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8025 | if (rt_runtime_us < 0) | |
8026 | rt_runtime = RUNTIME_INF; | |
8027 | ||
ab84d31e | 8028 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8029 | } |
8030 | ||
25cc7da7 | 8031 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
8032 | { |
8033 | u64 rt_runtime_us; | |
8034 | ||
d0b27fa7 | 8035 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8036 | return -1; |
8037 | ||
d0b27fa7 | 8038 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8039 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8040 | return rt_runtime_us; | |
8041 | } | |
d0b27fa7 | 8042 | |
ce2f5fe4 | 8043 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
8044 | { |
8045 | u64 rt_runtime, rt_period; | |
8046 | ||
ce2f5fe4 | 8047 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
8048 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
8049 | ||
ab84d31e | 8050 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8051 | } |
8052 | ||
25cc7da7 | 8053 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
8054 | { |
8055 | u64 rt_period_us; | |
8056 | ||
8057 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8058 | do_div(rt_period_us, NSEC_PER_USEC); | |
8059 | return rt_period_us; | |
8060 | } | |
332ac17e | 8061 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8062 | |
332ac17e | 8063 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
8064 | static int sched_rt_global_constraints(void) |
8065 | { | |
8066 | int ret = 0; | |
8067 | ||
8068 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 8069 | read_lock(&tasklist_lock); |
4653f803 | 8070 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8071 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8072 | mutex_unlock(&rt_constraints_mutex); |
8073 | ||
8074 | return ret; | |
8075 | } | |
54e99124 | 8076 | |
25cc7da7 | 8077 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
8078 | { |
8079 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8080 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8081 | return 0; | |
8082 | ||
8083 | return 1; | |
8084 | } | |
8085 | ||
6d6bc0ad | 8086 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8087 | static int sched_rt_global_constraints(void) |
8088 | { | |
ac086bc2 | 8089 | unsigned long flags; |
8c5e9554 | 8090 | int i; |
ec5d4989 | 8091 | |
0986b11b | 8092 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8093 | for_each_possible_cpu(i) { |
8094 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8095 | ||
0986b11b | 8096 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8097 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8098 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8099 | } |
0986b11b | 8100 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8101 | |
8c5e9554 | 8102 | return 0; |
d0b27fa7 | 8103 | } |
6d6bc0ad | 8104 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8105 | |
a1963b81 | 8106 | static int sched_dl_global_validate(void) |
332ac17e | 8107 | { |
1724813d PZ |
8108 | u64 runtime = global_rt_runtime(); |
8109 | u64 period = global_rt_period(); | |
332ac17e | 8110 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 8111 | struct dl_bw *dl_b; |
1724813d | 8112 | int cpu, ret = 0; |
49516342 | 8113 | unsigned long flags; |
332ac17e DF |
8114 | |
8115 | /* | |
8116 | * Here we want to check the bandwidth not being set to some | |
8117 | * value smaller than the currently allocated bandwidth in | |
8118 | * any of the root_domains. | |
8119 | * | |
8120 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
8121 | * cycling on root_domains... Discussion on different/better | |
8122 | * solutions is welcome! | |
8123 | */ | |
1724813d | 8124 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
8125 | rcu_read_lock_sched(); |
8126 | dl_b = dl_bw_of(cpu); | |
332ac17e | 8127 | |
49516342 | 8128 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
8129 | if (new_bw < dl_b->total_bw) |
8130 | ret = -EBUSY; | |
49516342 | 8131 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 8132 | |
f10e00f4 KT |
8133 | rcu_read_unlock_sched(); |
8134 | ||
1724813d PZ |
8135 | if (ret) |
8136 | break; | |
332ac17e DF |
8137 | } |
8138 | ||
1724813d | 8139 | return ret; |
332ac17e DF |
8140 | } |
8141 | ||
1724813d | 8142 | static void sched_dl_do_global(void) |
ce0dbbbb | 8143 | { |
1724813d | 8144 | u64 new_bw = -1; |
f10e00f4 | 8145 | struct dl_bw *dl_b; |
1724813d | 8146 | int cpu; |
49516342 | 8147 | unsigned long flags; |
ce0dbbbb | 8148 | |
1724813d PZ |
8149 | def_dl_bandwidth.dl_period = global_rt_period(); |
8150 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
8151 | ||
8152 | if (global_rt_runtime() != RUNTIME_INF) | |
8153 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
8154 | ||
8155 | /* | |
8156 | * FIXME: As above... | |
8157 | */ | |
8158 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
8159 | rcu_read_lock_sched(); |
8160 | dl_b = dl_bw_of(cpu); | |
1724813d | 8161 | |
49516342 | 8162 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8163 | dl_b->bw = new_bw; |
49516342 | 8164 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8165 | |
8166 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8167 | } |
1724813d PZ |
8168 | } |
8169 | ||
8170 | static int sched_rt_global_validate(void) | |
8171 | { | |
8172 | if (sysctl_sched_rt_period <= 0) | |
8173 | return -EINVAL; | |
8174 | ||
e9e7cb38 JL |
8175 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8176 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8177 | return -EINVAL; |
8178 | ||
8179 | return 0; | |
8180 | } | |
8181 | ||
8182 | static void sched_rt_do_global(void) | |
8183 | { | |
8184 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8185 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8186 | } |
8187 | ||
d0b27fa7 | 8188 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8189 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8190 | loff_t *ppos) |
8191 | { | |
d0b27fa7 PZ |
8192 | int old_period, old_runtime; |
8193 | static DEFINE_MUTEX(mutex); | |
1724813d | 8194 | int ret; |
d0b27fa7 PZ |
8195 | |
8196 | mutex_lock(&mutex); | |
8197 | old_period = sysctl_sched_rt_period; | |
8198 | old_runtime = sysctl_sched_rt_runtime; | |
8199 | ||
8d65af78 | 8200 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8201 | |
8202 | if (!ret && write) { | |
1724813d PZ |
8203 | ret = sched_rt_global_validate(); |
8204 | if (ret) | |
8205 | goto undo; | |
8206 | ||
a1963b81 | 8207 | ret = sched_dl_global_validate(); |
1724813d PZ |
8208 | if (ret) |
8209 | goto undo; | |
8210 | ||
a1963b81 | 8211 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8212 | if (ret) |
8213 | goto undo; | |
8214 | ||
8215 | sched_rt_do_global(); | |
8216 | sched_dl_do_global(); | |
8217 | } | |
8218 | if (0) { | |
8219 | undo: | |
8220 | sysctl_sched_rt_period = old_period; | |
8221 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8222 | } |
8223 | mutex_unlock(&mutex); | |
8224 | ||
8225 | return ret; | |
8226 | } | |
68318b8e | 8227 | |
1724813d | 8228 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8229 | void __user *buffer, size_t *lenp, |
8230 | loff_t *ppos) | |
8231 | { | |
8232 | int ret; | |
332ac17e | 8233 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8234 | |
8235 | mutex_lock(&mutex); | |
332ac17e | 8236 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8237 | /* make sure that internally we keep jiffies */ |
8238 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8239 | if (!ret && write) { |
1724813d PZ |
8240 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8241 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8242 | } |
8243 | mutex_unlock(&mutex); | |
332ac17e DF |
8244 | return ret; |
8245 | } | |
8246 | ||
052f1dc7 | 8247 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8248 | |
a7c6d554 | 8249 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8250 | { |
a7c6d554 | 8251 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8252 | } |
8253 | ||
eb95419b TH |
8254 | static struct cgroup_subsys_state * |
8255 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8256 | { |
eb95419b TH |
8257 | struct task_group *parent = css_tg(parent_css); |
8258 | struct task_group *tg; | |
68318b8e | 8259 | |
eb95419b | 8260 | if (!parent) { |
68318b8e | 8261 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8262 | return &root_task_group.css; |
68318b8e SV |
8263 | } |
8264 | ||
ec7dc8ac | 8265 | tg = sched_create_group(parent); |
68318b8e SV |
8266 | if (IS_ERR(tg)) |
8267 | return ERR_PTR(-ENOMEM); | |
8268 | ||
2f5177f0 PZ |
8269 | sched_online_group(tg, parent); |
8270 | ||
68318b8e SV |
8271 | return &tg->css; |
8272 | } | |
8273 | ||
2f5177f0 | 8274 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8275 | { |
eb95419b | 8276 | struct task_group *tg = css_tg(css); |
ace783b9 | 8277 | |
2f5177f0 | 8278 | sched_offline_group(tg); |
ace783b9 LZ |
8279 | } |
8280 | ||
eb95419b | 8281 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8282 | { |
eb95419b | 8283 | struct task_group *tg = css_tg(css); |
68318b8e | 8284 | |
2f5177f0 PZ |
8285 | /* |
8286 | * Relies on the RCU grace period between css_released() and this. | |
8287 | */ | |
8288 | sched_free_group(tg); | |
ace783b9 LZ |
8289 | } |
8290 | ||
ea86cb4b VG |
8291 | /* |
8292 | * This is called before wake_up_new_task(), therefore we really only | |
8293 | * have to set its group bits, all the other stuff does not apply. | |
8294 | */ | |
b53202e6 | 8295 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8296 | { |
ea86cb4b VG |
8297 | struct rq_flags rf; |
8298 | struct rq *rq; | |
8299 | ||
8300 | rq = task_rq_lock(task, &rf); | |
8301 | ||
8302 | sched_change_group(task, TASK_SET_GROUP); | |
8303 | ||
8304 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8305 | } |
8306 | ||
1f7dd3e5 | 8307 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8308 | { |
bb9d97b6 | 8309 | struct task_struct *task; |
1f7dd3e5 | 8310 | struct cgroup_subsys_state *css; |
7dc603c9 | 8311 | int ret = 0; |
bb9d97b6 | 8312 | |
1f7dd3e5 | 8313 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8314 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8315 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8316 | return -EINVAL; |
b68aa230 | 8317 | #else |
bb9d97b6 TH |
8318 | /* We don't support RT-tasks being in separate groups */ |
8319 | if (task->sched_class != &fair_sched_class) | |
8320 | return -EINVAL; | |
b68aa230 | 8321 | #endif |
7dc603c9 PZ |
8322 | /* |
8323 | * Serialize against wake_up_new_task() such that if its | |
8324 | * running, we're sure to observe its full state. | |
8325 | */ | |
8326 | raw_spin_lock_irq(&task->pi_lock); | |
8327 | /* | |
8328 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8329 | * has happened. This would lead to problems with PELT, due to | |
8330 | * move wanting to detach+attach while we're not attached yet. | |
8331 | */ | |
8332 | if (task->state == TASK_NEW) | |
8333 | ret = -EINVAL; | |
8334 | raw_spin_unlock_irq(&task->pi_lock); | |
8335 | ||
8336 | if (ret) | |
8337 | break; | |
bb9d97b6 | 8338 | } |
7dc603c9 | 8339 | return ret; |
be367d09 | 8340 | } |
68318b8e | 8341 | |
1f7dd3e5 | 8342 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8343 | { |
bb9d97b6 | 8344 | struct task_struct *task; |
1f7dd3e5 | 8345 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8346 | |
1f7dd3e5 | 8347 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8348 | sched_move_task(task); |
68318b8e SV |
8349 | } |
8350 | ||
052f1dc7 | 8351 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8352 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8353 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8354 | { |
182446d0 | 8355 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8356 | } |
8357 | ||
182446d0 TH |
8358 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8359 | struct cftype *cft) | |
68318b8e | 8360 | { |
182446d0 | 8361 | struct task_group *tg = css_tg(css); |
68318b8e | 8362 | |
c8b28116 | 8363 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8364 | } |
ab84d31e PT |
8365 | |
8366 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8367 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8368 | ||
ab84d31e PT |
8369 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8370 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8371 | ||
a790de99 PT |
8372 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8373 | ||
ab84d31e PT |
8374 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8375 | { | |
56f570e5 | 8376 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8377 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8378 | |
8379 | if (tg == &root_task_group) | |
8380 | return -EINVAL; | |
8381 | ||
8382 | /* | |
8383 | * Ensure we have at some amount of bandwidth every period. This is | |
8384 | * to prevent reaching a state of large arrears when throttled via | |
8385 | * entity_tick() resulting in prolonged exit starvation. | |
8386 | */ | |
8387 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8388 | return -EINVAL; | |
8389 | ||
8390 | /* | |
8391 | * Likewise, bound things on the otherside by preventing insane quota | |
8392 | * periods. This also allows us to normalize in computing quota | |
8393 | * feasibility. | |
8394 | */ | |
8395 | if (period > max_cfs_quota_period) | |
8396 | return -EINVAL; | |
8397 | ||
0e59bdae KT |
8398 | /* |
8399 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8400 | * unthrottle_offline_cfs_rqs(). | |
8401 | */ | |
8402 | get_online_cpus(); | |
a790de99 PT |
8403 | mutex_lock(&cfs_constraints_mutex); |
8404 | ret = __cfs_schedulable(tg, period, quota); | |
8405 | if (ret) | |
8406 | goto out_unlock; | |
8407 | ||
58088ad0 | 8408 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8409 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8410 | /* |
8411 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8412 | * before making related changes, and on->off must occur afterwards | |
8413 | */ | |
8414 | if (runtime_enabled && !runtime_was_enabled) | |
8415 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8416 | raw_spin_lock_irq(&cfs_b->lock); |
8417 | cfs_b->period = ns_to_ktime(period); | |
8418 | cfs_b->quota = quota; | |
58088ad0 | 8419 | |
a9cf55b2 | 8420 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8421 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8422 | if (runtime_enabled) |
8423 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8424 | raw_spin_unlock_irq(&cfs_b->lock); |
8425 | ||
0e59bdae | 8426 | for_each_online_cpu(i) { |
ab84d31e | 8427 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8428 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8429 | |
8430 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8431 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8432 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8433 | |
029632fb | 8434 | if (cfs_rq->throttled) |
671fd9da | 8435 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8436 | raw_spin_unlock_irq(&rq->lock); |
8437 | } | |
1ee14e6c BS |
8438 | if (runtime_was_enabled && !runtime_enabled) |
8439 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8440 | out_unlock: |
8441 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8442 | put_online_cpus(); |
ab84d31e | 8443 | |
a790de99 | 8444 | return ret; |
ab84d31e PT |
8445 | } |
8446 | ||
8447 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8448 | { | |
8449 | u64 quota, period; | |
8450 | ||
029632fb | 8451 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8452 | if (cfs_quota_us < 0) |
8453 | quota = RUNTIME_INF; | |
8454 | else | |
8455 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8456 | ||
8457 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8458 | } | |
8459 | ||
8460 | long tg_get_cfs_quota(struct task_group *tg) | |
8461 | { | |
8462 | u64 quota_us; | |
8463 | ||
029632fb | 8464 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8465 | return -1; |
8466 | ||
029632fb | 8467 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8468 | do_div(quota_us, NSEC_PER_USEC); |
8469 | ||
8470 | return quota_us; | |
8471 | } | |
8472 | ||
8473 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8474 | { | |
8475 | u64 quota, period; | |
8476 | ||
8477 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8478 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8479 | |
ab84d31e PT |
8480 | return tg_set_cfs_bandwidth(tg, period, quota); |
8481 | } | |
8482 | ||
8483 | long tg_get_cfs_period(struct task_group *tg) | |
8484 | { | |
8485 | u64 cfs_period_us; | |
8486 | ||
029632fb | 8487 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8488 | do_div(cfs_period_us, NSEC_PER_USEC); |
8489 | ||
8490 | return cfs_period_us; | |
8491 | } | |
8492 | ||
182446d0 TH |
8493 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8494 | struct cftype *cft) | |
ab84d31e | 8495 | { |
182446d0 | 8496 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8497 | } |
8498 | ||
182446d0 TH |
8499 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8500 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8501 | { |
182446d0 | 8502 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8503 | } |
8504 | ||
182446d0 TH |
8505 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8506 | struct cftype *cft) | |
ab84d31e | 8507 | { |
182446d0 | 8508 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8509 | } |
8510 | ||
182446d0 TH |
8511 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8512 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8513 | { |
182446d0 | 8514 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8515 | } |
8516 | ||
a790de99 PT |
8517 | struct cfs_schedulable_data { |
8518 | struct task_group *tg; | |
8519 | u64 period, quota; | |
8520 | }; | |
8521 | ||
8522 | /* | |
8523 | * normalize group quota/period to be quota/max_period | |
8524 | * note: units are usecs | |
8525 | */ | |
8526 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8527 | struct cfs_schedulable_data *d) | |
8528 | { | |
8529 | u64 quota, period; | |
8530 | ||
8531 | if (tg == d->tg) { | |
8532 | period = d->period; | |
8533 | quota = d->quota; | |
8534 | } else { | |
8535 | period = tg_get_cfs_period(tg); | |
8536 | quota = tg_get_cfs_quota(tg); | |
8537 | } | |
8538 | ||
8539 | /* note: these should typically be equivalent */ | |
8540 | if (quota == RUNTIME_INF || quota == -1) | |
8541 | return RUNTIME_INF; | |
8542 | ||
8543 | return to_ratio(period, quota); | |
8544 | } | |
8545 | ||
8546 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8547 | { | |
8548 | struct cfs_schedulable_data *d = data; | |
029632fb | 8549 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8550 | s64 quota = 0, parent_quota = -1; |
8551 | ||
8552 | if (!tg->parent) { | |
8553 | quota = RUNTIME_INF; | |
8554 | } else { | |
029632fb | 8555 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8556 | |
8557 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8558 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8559 | |
8560 | /* | |
8561 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8562 | * limit is set | |
8563 | */ | |
8564 | if (quota == RUNTIME_INF) | |
8565 | quota = parent_quota; | |
8566 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8567 | return -EINVAL; | |
8568 | } | |
9c58c79a | 8569 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8570 | |
8571 | return 0; | |
8572 | } | |
8573 | ||
8574 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8575 | { | |
8277434e | 8576 | int ret; |
a790de99 PT |
8577 | struct cfs_schedulable_data data = { |
8578 | .tg = tg, | |
8579 | .period = period, | |
8580 | .quota = quota, | |
8581 | }; | |
8582 | ||
8583 | if (quota != RUNTIME_INF) { | |
8584 | do_div(data.period, NSEC_PER_USEC); | |
8585 | do_div(data.quota, NSEC_PER_USEC); | |
8586 | } | |
8587 | ||
8277434e PT |
8588 | rcu_read_lock(); |
8589 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8590 | rcu_read_unlock(); | |
8591 | ||
8592 | return ret; | |
a790de99 | 8593 | } |
e8da1b18 | 8594 | |
2da8ca82 | 8595 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8596 | { |
2da8ca82 | 8597 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8598 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8599 | |
44ffc75b TH |
8600 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8601 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8602 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8603 | |
8604 | return 0; | |
8605 | } | |
ab84d31e | 8606 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8607 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8608 | |
052f1dc7 | 8609 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8610 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8611 | struct cftype *cft, s64 val) | |
6f505b16 | 8612 | { |
182446d0 | 8613 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8614 | } |
8615 | ||
182446d0 TH |
8616 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8617 | struct cftype *cft) | |
6f505b16 | 8618 | { |
182446d0 | 8619 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8620 | } |
d0b27fa7 | 8621 | |
182446d0 TH |
8622 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8623 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8624 | { |
182446d0 | 8625 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8626 | } |
8627 | ||
182446d0 TH |
8628 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8629 | struct cftype *cft) | |
d0b27fa7 | 8630 | { |
182446d0 | 8631 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8632 | } |
6d6bc0ad | 8633 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8634 | |
fe5c7cc2 | 8635 | static struct cftype cpu_files[] = { |
052f1dc7 | 8636 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8637 | { |
8638 | .name = "shares", | |
f4c753b7 PM |
8639 | .read_u64 = cpu_shares_read_u64, |
8640 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8641 | }, |
052f1dc7 | 8642 | #endif |
ab84d31e PT |
8643 | #ifdef CONFIG_CFS_BANDWIDTH |
8644 | { | |
8645 | .name = "cfs_quota_us", | |
8646 | .read_s64 = cpu_cfs_quota_read_s64, | |
8647 | .write_s64 = cpu_cfs_quota_write_s64, | |
8648 | }, | |
8649 | { | |
8650 | .name = "cfs_period_us", | |
8651 | .read_u64 = cpu_cfs_period_read_u64, | |
8652 | .write_u64 = cpu_cfs_period_write_u64, | |
8653 | }, | |
e8da1b18 NR |
8654 | { |
8655 | .name = "stat", | |
2da8ca82 | 8656 | .seq_show = cpu_stats_show, |
e8da1b18 | 8657 | }, |
ab84d31e | 8658 | #endif |
052f1dc7 | 8659 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8660 | { |
9f0c1e56 | 8661 | .name = "rt_runtime_us", |
06ecb27c PM |
8662 | .read_s64 = cpu_rt_runtime_read, |
8663 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8664 | }, |
d0b27fa7 PZ |
8665 | { |
8666 | .name = "rt_period_us", | |
f4c753b7 PM |
8667 | .read_u64 = cpu_rt_period_read_uint, |
8668 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8669 | }, |
052f1dc7 | 8670 | #endif |
4baf6e33 | 8671 | { } /* terminate */ |
68318b8e SV |
8672 | }; |
8673 | ||
073219e9 | 8674 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8675 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8676 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8677 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8678 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8679 | .can_attach = cpu_cgroup_can_attach, |
8680 | .attach = cpu_cgroup_attach, | |
5577964e | 8681 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8682 | .early_init = true, |
68318b8e SV |
8683 | }; |
8684 | ||
052f1dc7 | 8685 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8686 | |
b637a328 PM |
8687 | void dump_cpu_task(int cpu) |
8688 | { | |
8689 | pr_info("Task dump for CPU %d:\n", cpu); | |
8690 | sched_show_task(cpu_curr(cpu)); | |
8691 | } | |
ed82b8a1 AK |
8692 | |
8693 | /* | |
8694 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8695 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8696 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8697 | * that remained on nice 0. | |
8698 | * | |
8699 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8700 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8701 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8702 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8703 | * the relative distance between them is ~25%.) | |
8704 | */ | |
8705 | const int sched_prio_to_weight[40] = { | |
8706 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8707 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8708 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8709 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8710 | /* 0 */ 1024, 820, 655, 526, 423, | |
8711 | /* 5 */ 335, 272, 215, 172, 137, | |
8712 | /* 10 */ 110, 87, 70, 56, 45, | |
8713 | /* 15 */ 36, 29, 23, 18, 15, | |
8714 | }; | |
8715 | ||
8716 | /* | |
8717 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8718 | * | |
8719 | * In cases where the weight does not change often, we can use the | |
8720 | * precalculated inverse to speed up arithmetics by turning divisions | |
8721 | * into multiplications: | |
8722 | */ | |
8723 | const u32 sched_prio_to_wmult[40] = { | |
8724 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8725 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8726 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8727 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8728 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8729 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8730 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8731 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8732 | }; |