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> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
029632fb PZ |
93 | DEFINE_MUTEX(sched_domains_mutex); |
94 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 95 | |
fe44d621 | 96 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 97 | |
029632fb | 98 | void update_rq_clock(struct rq *rq) |
3e51f33f | 99 | { |
fe44d621 | 100 | s64 delta; |
305e6835 | 101 | |
9edfbfed PZ |
102 | lockdep_assert_held(&rq->lock); |
103 | ||
104 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 105 | return; |
aa483808 | 106 | |
fe44d621 | 107 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
108 | if (delta < 0) |
109 | return; | |
fe44d621 PZ |
110 | rq->clock += delta; |
111 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
112 | } |
113 | ||
bf5c91ba IM |
114 | /* |
115 | * Debugging: various feature bits | |
116 | */ | |
f00b45c1 | 117 | |
f00b45c1 PZ |
118 | #define SCHED_FEAT(name, enabled) \ |
119 | (1UL << __SCHED_FEAT_##name) * enabled | | |
120 | ||
bf5c91ba | 121 | const_debug unsigned int sysctl_sched_features = |
391e43da | 122 | #include "features.h" |
f00b45c1 PZ |
123 | 0; |
124 | ||
125 | #undef SCHED_FEAT | |
126 | ||
b82d9fdd PZ |
127 | /* |
128 | * Number of tasks to iterate in a single balance run. | |
129 | * Limited because this is done with IRQs disabled. | |
130 | */ | |
131 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
132 | ||
e9e9250b PZ |
133 | /* |
134 | * period over which we average the RT time consumption, measured | |
135 | * in ms. | |
136 | * | |
137 | * default: 1s | |
138 | */ | |
139 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
140 | ||
fa85ae24 | 141 | /* |
9f0c1e56 | 142 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
143 | * default: 1s |
144 | */ | |
9f0c1e56 | 145 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 146 | |
029632fb | 147 | __read_mostly int scheduler_running; |
6892b75e | 148 | |
9f0c1e56 PZ |
149 | /* |
150 | * part of the period that we allow rt tasks to run in us. | |
151 | * default: 0.95s | |
152 | */ | |
153 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 154 | |
3fa0818b RR |
155 | /* cpus with isolated domains */ |
156 | cpumask_var_t cpu_isolated_map; | |
157 | ||
1da177e4 | 158 | /* |
cc2a73b5 | 159 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 160 | */ |
a9957449 | 161 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
162 | __acquires(rq->lock) |
163 | { | |
70b97a7f | 164 | struct rq *rq; |
1da177e4 LT |
165 | |
166 | local_irq_disable(); | |
167 | rq = this_rq(); | |
05fa785c | 168 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
169 | |
170 | return rq; | |
171 | } | |
172 | ||
3e71a462 PZ |
173 | /* |
174 | * __task_rq_lock - lock the rq @p resides on. | |
175 | */ | |
eb580751 | 176 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
177 | __acquires(rq->lock) |
178 | { | |
179 | struct rq *rq; | |
180 | ||
181 | lockdep_assert_held(&p->pi_lock); | |
182 | ||
183 | for (;;) { | |
184 | rq = task_rq(p); | |
185 | raw_spin_lock(&rq->lock); | |
186 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
e7904a28 | 187 | rf->cookie = lockdep_pin_lock(&rq->lock); |
3e71a462 PZ |
188 | return rq; |
189 | } | |
190 | raw_spin_unlock(&rq->lock); | |
191 | ||
192 | while (unlikely(task_on_rq_migrating(p))) | |
193 | cpu_relax(); | |
194 | } | |
195 | } | |
196 | ||
197 | /* | |
198 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
199 | */ | |
eb580751 | 200 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
201 | __acquires(p->pi_lock) |
202 | __acquires(rq->lock) | |
203 | { | |
204 | struct rq *rq; | |
205 | ||
206 | for (;;) { | |
eb580751 | 207 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
208 | rq = task_rq(p); |
209 | raw_spin_lock(&rq->lock); | |
210 | /* | |
211 | * move_queued_task() task_rq_lock() | |
212 | * | |
213 | * ACQUIRE (rq->lock) | |
214 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
215 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
216 | * [S] ->cpu = new_cpu [L] task_rq() | |
217 | * [L] ->on_rq | |
218 | * RELEASE (rq->lock) | |
219 | * | |
220 | * If we observe the old cpu in task_rq_lock, the acquire of | |
221 | * the old rq->lock will fully serialize against the stores. | |
222 | * | |
223 | * If we observe the new cpu in task_rq_lock, the acquire will | |
224 | * pair with the WMB to ensure we must then also see migrating. | |
225 | */ | |
226 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
e7904a28 | 227 | rf->cookie = lockdep_pin_lock(&rq->lock); |
3e71a462 PZ |
228 | return rq; |
229 | } | |
230 | raw_spin_unlock(&rq->lock); | |
eb580751 | 231 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
232 | |
233 | while (unlikely(task_on_rq_migrating(p))) | |
234 | cpu_relax(); | |
235 | } | |
236 | } | |
237 | ||
8f4d37ec PZ |
238 | #ifdef CONFIG_SCHED_HRTICK |
239 | /* | |
240 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 241 | */ |
8f4d37ec | 242 | |
8f4d37ec PZ |
243 | static void hrtick_clear(struct rq *rq) |
244 | { | |
245 | if (hrtimer_active(&rq->hrtick_timer)) | |
246 | hrtimer_cancel(&rq->hrtick_timer); | |
247 | } | |
248 | ||
8f4d37ec PZ |
249 | /* |
250 | * High-resolution timer tick. | |
251 | * Runs from hardirq context with interrupts disabled. | |
252 | */ | |
253 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
254 | { | |
255 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
256 | ||
257 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
258 | ||
05fa785c | 259 | raw_spin_lock(&rq->lock); |
3e51f33f | 260 | update_rq_clock(rq); |
8f4d37ec | 261 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 262 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
263 | |
264 | return HRTIMER_NORESTART; | |
265 | } | |
266 | ||
95e904c7 | 267 | #ifdef CONFIG_SMP |
971ee28c | 268 | |
4961b6e1 | 269 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
270 | { |
271 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 272 | |
4961b6e1 | 273 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
274 | } |
275 | ||
31656519 PZ |
276 | /* |
277 | * called from hardirq (IPI) context | |
278 | */ | |
279 | static void __hrtick_start(void *arg) | |
b328ca18 | 280 | { |
31656519 | 281 | struct rq *rq = arg; |
b328ca18 | 282 | |
05fa785c | 283 | raw_spin_lock(&rq->lock); |
971ee28c | 284 | __hrtick_restart(rq); |
31656519 | 285 | rq->hrtick_csd_pending = 0; |
05fa785c | 286 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
287 | } |
288 | ||
31656519 PZ |
289 | /* |
290 | * Called to set the hrtick timer state. | |
291 | * | |
292 | * called with rq->lock held and irqs disabled | |
293 | */ | |
029632fb | 294 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 295 | { |
31656519 | 296 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 297 | ktime_t time; |
298 | s64 delta; | |
299 | ||
300 | /* | |
301 | * Don't schedule slices shorter than 10000ns, that just | |
302 | * doesn't make sense and can cause timer DoS. | |
303 | */ | |
304 | delta = max_t(s64, delay, 10000LL); | |
305 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 306 | |
cc584b21 | 307 | hrtimer_set_expires(timer, time); |
31656519 PZ |
308 | |
309 | if (rq == this_rq()) { | |
971ee28c | 310 | __hrtick_restart(rq); |
31656519 | 311 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 312 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
313 | rq->hrtick_csd_pending = 1; |
314 | } | |
b328ca18 PZ |
315 | } |
316 | ||
31656519 PZ |
317 | #else |
318 | /* | |
319 | * Called to set the hrtick timer state. | |
320 | * | |
321 | * called with rq->lock held and irqs disabled | |
322 | */ | |
029632fb | 323 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 324 | { |
86893335 WL |
325 | /* |
326 | * Don't schedule slices shorter than 10000ns, that just | |
327 | * doesn't make sense. Rely on vruntime for fairness. | |
328 | */ | |
329 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
330 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
331 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 332 | } |
31656519 | 333 | #endif /* CONFIG_SMP */ |
8f4d37ec | 334 | |
31656519 | 335 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 336 | { |
31656519 PZ |
337 | #ifdef CONFIG_SMP |
338 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 339 | |
31656519 PZ |
340 | rq->hrtick_csd.flags = 0; |
341 | rq->hrtick_csd.func = __hrtick_start; | |
342 | rq->hrtick_csd.info = rq; | |
343 | #endif | |
8f4d37ec | 344 | |
31656519 PZ |
345 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
346 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 347 | } |
006c75f1 | 348 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
349 | static inline void hrtick_clear(struct rq *rq) |
350 | { | |
351 | } | |
352 | ||
8f4d37ec PZ |
353 | static inline void init_rq_hrtick(struct rq *rq) |
354 | { | |
355 | } | |
006c75f1 | 356 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 357 | |
5529578a FW |
358 | /* |
359 | * cmpxchg based fetch_or, macro so it works for different integer types | |
360 | */ | |
361 | #define fetch_or(ptr, mask) \ | |
362 | ({ \ | |
363 | typeof(ptr) _ptr = (ptr); \ | |
364 | typeof(mask) _mask = (mask); \ | |
365 | typeof(*_ptr) _old, _val = *_ptr; \ | |
366 | \ | |
367 | for (;;) { \ | |
368 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
369 | if (_old == _val) \ | |
370 | break; \ | |
371 | _val = _old; \ | |
372 | } \ | |
373 | _old; \ | |
374 | }) | |
375 | ||
e3baac47 | 376 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
377 | /* |
378 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
379 | * this avoids any races wrt polling state changes and thereby avoids | |
380 | * spurious IPIs. | |
381 | */ | |
382 | static bool set_nr_and_not_polling(struct task_struct *p) | |
383 | { | |
384 | struct thread_info *ti = task_thread_info(p); | |
385 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
386 | } | |
e3baac47 PZ |
387 | |
388 | /* | |
389 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
390 | * | |
391 | * If this returns true, then the idle task promises to call | |
392 | * sched_ttwu_pending() and reschedule soon. | |
393 | */ | |
394 | static bool set_nr_if_polling(struct task_struct *p) | |
395 | { | |
396 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 397 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
398 | |
399 | for (;;) { | |
400 | if (!(val & _TIF_POLLING_NRFLAG)) | |
401 | return false; | |
402 | if (val & _TIF_NEED_RESCHED) | |
403 | return true; | |
404 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
405 | if (old == val) | |
406 | break; | |
407 | val = old; | |
408 | } | |
409 | return true; | |
410 | } | |
411 | ||
fd99f91a PZ |
412 | #else |
413 | static bool set_nr_and_not_polling(struct task_struct *p) | |
414 | { | |
415 | set_tsk_need_resched(p); | |
416 | return true; | |
417 | } | |
e3baac47 PZ |
418 | |
419 | #ifdef CONFIG_SMP | |
420 | static bool set_nr_if_polling(struct task_struct *p) | |
421 | { | |
422 | return false; | |
423 | } | |
424 | #endif | |
fd99f91a PZ |
425 | #endif |
426 | ||
76751049 PZ |
427 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
428 | { | |
429 | struct wake_q_node *node = &task->wake_q; | |
430 | ||
431 | /* | |
432 | * Atomically grab the task, if ->wake_q is !nil already it means | |
433 | * its already queued (either by us or someone else) and will get the | |
434 | * wakeup due to that. | |
435 | * | |
436 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 437 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
438 | */ |
439 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
440 | return; | |
441 | ||
442 | get_task_struct(task); | |
443 | ||
444 | /* | |
445 | * The head is context local, there can be no concurrency. | |
446 | */ | |
447 | *head->lastp = node; | |
448 | head->lastp = &node->next; | |
449 | } | |
450 | ||
451 | void wake_up_q(struct wake_q_head *head) | |
452 | { | |
453 | struct wake_q_node *node = head->first; | |
454 | ||
455 | while (node != WAKE_Q_TAIL) { | |
456 | struct task_struct *task; | |
457 | ||
458 | task = container_of(node, struct task_struct, wake_q); | |
459 | BUG_ON(!task); | |
460 | /* task can safely be re-inserted now */ | |
461 | node = node->next; | |
462 | task->wake_q.next = NULL; | |
463 | ||
464 | /* | |
465 | * wake_up_process() implies a wmb() to pair with the queueing | |
466 | * in wake_q_add() so as not to miss wakeups. | |
467 | */ | |
468 | wake_up_process(task); | |
469 | put_task_struct(task); | |
470 | } | |
471 | } | |
472 | ||
c24d20db | 473 | /* |
8875125e | 474 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
475 | * |
476 | * On UP this means the setting of the need_resched flag, on SMP it | |
477 | * might also involve a cross-CPU call to trigger the scheduler on | |
478 | * the target CPU. | |
479 | */ | |
8875125e | 480 | void resched_curr(struct rq *rq) |
c24d20db | 481 | { |
8875125e | 482 | struct task_struct *curr = rq->curr; |
c24d20db IM |
483 | int cpu; |
484 | ||
8875125e | 485 | lockdep_assert_held(&rq->lock); |
c24d20db | 486 | |
8875125e | 487 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
488 | return; |
489 | ||
8875125e | 490 | cpu = cpu_of(rq); |
fd99f91a | 491 | |
f27dde8d | 492 | if (cpu == smp_processor_id()) { |
8875125e | 493 | set_tsk_need_resched(curr); |
f27dde8d | 494 | set_preempt_need_resched(); |
c24d20db | 495 | return; |
f27dde8d | 496 | } |
c24d20db | 497 | |
8875125e | 498 | if (set_nr_and_not_polling(curr)) |
c24d20db | 499 | smp_send_reschedule(cpu); |
dfc68f29 AL |
500 | else |
501 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
502 | } |
503 | ||
029632fb | 504 | void resched_cpu(int cpu) |
c24d20db IM |
505 | { |
506 | struct rq *rq = cpu_rq(cpu); | |
507 | unsigned long flags; | |
508 | ||
05fa785c | 509 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 510 | return; |
8875125e | 511 | resched_curr(rq); |
05fa785c | 512 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 513 | } |
06d8308c | 514 | |
b021fe3e | 515 | #ifdef CONFIG_SMP |
3451d024 | 516 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
517 | /* |
518 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
519 | * from an idle cpu. This is good for power-savings. | |
520 | * | |
521 | * We don't do similar optimization for completely idle system, as | |
522 | * selecting an idle cpu will add more delays to the timers than intended | |
523 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
524 | */ | |
bc7a34b8 | 525 | int get_nohz_timer_target(void) |
83cd4fe2 | 526 | { |
bc7a34b8 | 527 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
528 | struct sched_domain *sd; |
529 | ||
9642d18e | 530 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
531 | return cpu; |
532 | ||
057f3fad | 533 | rcu_read_lock(); |
83cd4fe2 | 534 | for_each_domain(cpu, sd) { |
057f3fad | 535 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
536 | if (cpu == i) |
537 | continue; | |
538 | ||
539 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
540 | cpu = i; |
541 | goto unlock; | |
542 | } | |
543 | } | |
83cd4fe2 | 544 | } |
9642d18e VH |
545 | |
546 | if (!is_housekeeping_cpu(cpu)) | |
547 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
548 | unlock: |
549 | rcu_read_unlock(); | |
83cd4fe2 VP |
550 | return cpu; |
551 | } | |
06d8308c TG |
552 | /* |
553 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
554 | * idle CPU then this timer might expire before the next timer event | |
555 | * which is scheduled to wake up that CPU. In case of a completely | |
556 | * idle system the next event might even be infinite time into the | |
557 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
558 | * leaves the inner idle loop so the newly added timer is taken into | |
559 | * account when the CPU goes back to idle and evaluates the timer | |
560 | * wheel for the next timer event. | |
561 | */ | |
1c20091e | 562 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
563 | { |
564 | struct rq *rq = cpu_rq(cpu); | |
565 | ||
566 | if (cpu == smp_processor_id()) | |
567 | return; | |
568 | ||
67b9ca70 | 569 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 570 | smp_send_reschedule(cpu); |
dfc68f29 AL |
571 | else |
572 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
573 | } |
574 | ||
c5bfece2 | 575 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 576 | { |
53c5fa16 FW |
577 | /* |
578 | * We just need the target to call irq_exit() and re-evaluate | |
579 | * the next tick. The nohz full kick at least implies that. | |
580 | * If needed we can still optimize that later with an | |
581 | * empty IRQ. | |
582 | */ | |
c5bfece2 | 583 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
584 | if (cpu != smp_processor_id() || |
585 | tick_nohz_tick_stopped()) | |
53c5fa16 | 586 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
587 | return true; |
588 | } | |
589 | ||
590 | return false; | |
591 | } | |
592 | ||
593 | void wake_up_nohz_cpu(int cpu) | |
594 | { | |
c5bfece2 | 595 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
596 | wake_up_idle_cpu(cpu); |
597 | } | |
598 | ||
ca38062e | 599 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 600 | { |
1c792db7 | 601 | int cpu = smp_processor_id(); |
873b4c65 VG |
602 | |
603 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
604 | return false; | |
605 | ||
606 | if (idle_cpu(cpu) && !need_resched()) | |
607 | return true; | |
608 | ||
609 | /* | |
610 | * We can't run Idle Load Balance on this CPU for this time so we | |
611 | * cancel it and clear NOHZ_BALANCE_KICK | |
612 | */ | |
613 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
614 | return false; | |
45bf76df IM |
615 | } |
616 | ||
3451d024 | 617 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 618 | |
ca38062e | 619 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 620 | { |
ca38062e | 621 | return false; |
2069dd75 PZ |
622 | } |
623 | ||
3451d024 | 624 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 625 | |
ce831b38 | 626 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 627 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 628 | { |
76d92ac3 FW |
629 | int fifo_nr_running; |
630 | ||
631 | /* Deadline tasks, even if single, need the tick */ | |
632 | if (rq->dl.dl_nr_running) | |
633 | return false; | |
634 | ||
1e78cdbd | 635 | /* |
2548d546 PZ |
636 | * If there are more than one RR tasks, we need the tick to effect the |
637 | * actual RR behaviour. | |
1e78cdbd | 638 | */ |
76d92ac3 FW |
639 | if (rq->rt.rr_nr_running) { |
640 | if (rq->rt.rr_nr_running == 1) | |
641 | return true; | |
642 | else | |
643 | return false; | |
1e78cdbd RR |
644 | } |
645 | ||
2548d546 PZ |
646 | /* |
647 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
648 | * forced preemption between FIFO tasks. | |
649 | */ | |
650 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
651 | if (fifo_nr_running) | |
652 | return true; | |
653 | ||
654 | /* | |
655 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
656 | * if there's more than one we need the tick for involuntary | |
657 | * preemption. | |
658 | */ | |
659 | if (rq->nr_running > 1) | |
541b8264 | 660 | return false; |
ce831b38 | 661 | |
541b8264 | 662 | return true; |
ce831b38 FW |
663 | } |
664 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 665 | |
029632fb | 666 | void sched_avg_update(struct rq *rq) |
18d95a28 | 667 | { |
e9e9250b PZ |
668 | s64 period = sched_avg_period(); |
669 | ||
78becc27 | 670 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
671 | /* |
672 | * Inline assembly required to prevent the compiler | |
673 | * optimising this loop into a divmod call. | |
674 | * See __iter_div_u64_rem() for another example of this. | |
675 | */ | |
676 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
677 | rq->age_stamp += period; |
678 | rq->rt_avg /= 2; | |
679 | } | |
18d95a28 PZ |
680 | } |
681 | ||
6d6bc0ad | 682 | #endif /* CONFIG_SMP */ |
18d95a28 | 683 | |
a790de99 PT |
684 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
685 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 686 | /* |
8277434e PT |
687 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
688 | * node and @up when leaving it for the final time. | |
689 | * | |
690 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 691 | */ |
029632fb | 692 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 693 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
694 | { |
695 | struct task_group *parent, *child; | |
eb755805 | 696 | int ret; |
c09595f6 | 697 | |
8277434e PT |
698 | parent = from; |
699 | ||
c09595f6 | 700 | down: |
eb755805 PZ |
701 | ret = (*down)(parent, data); |
702 | if (ret) | |
8277434e | 703 | goto out; |
c09595f6 PZ |
704 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
705 | parent = child; | |
706 | goto down; | |
707 | ||
708 | up: | |
709 | continue; | |
710 | } | |
eb755805 | 711 | ret = (*up)(parent, data); |
8277434e PT |
712 | if (ret || parent == from) |
713 | goto out; | |
c09595f6 PZ |
714 | |
715 | child = parent; | |
716 | parent = parent->parent; | |
717 | if (parent) | |
718 | goto up; | |
8277434e | 719 | out: |
eb755805 | 720 | return ret; |
c09595f6 PZ |
721 | } |
722 | ||
029632fb | 723 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 724 | { |
e2b245f8 | 725 | return 0; |
eb755805 | 726 | } |
18d95a28 PZ |
727 | #endif |
728 | ||
45bf76df IM |
729 | static void set_load_weight(struct task_struct *p) |
730 | { | |
f05998d4 NR |
731 | int prio = p->static_prio - MAX_RT_PRIO; |
732 | struct load_weight *load = &p->se.load; | |
733 | ||
dd41f596 IM |
734 | /* |
735 | * SCHED_IDLE tasks get minimal weight: | |
736 | */ | |
20f9cd2a | 737 | if (idle_policy(p->policy)) { |
c8b28116 | 738 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 739 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
740 | return; |
741 | } | |
71f8bd46 | 742 | |
ed82b8a1 AK |
743 | load->weight = scale_load(sched_prio_to_weight[prio]); |
744 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
745 | } |
746 | ||
1de64443 | 747 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 748 | { |
a64692a3 | 749 | update_rq_clock(rq); |
1de64443 PZ |
750 | if (!(flags & ENQUEUE_RESTORE)) |
751 | sched_info_queued(rq, p); | |
371fd7e7 | 752 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
753 | } |
754 | ||
1de64443 | 755 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 756 | { |
a64692a3 | 757 | update_rq_clock(rq); |
1de64443 PZ |
758 | if (!(flags & DEQUEUE_SAVE)) |
759 | sched_info_dequeued(rq, p); | |
371fd7e7 | 760 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
761 | } |
762 | ||
029632fb | 763 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
764 | { |
765 | if (task_contributes_to_load(p)) | |
766 | rq->nr_uninterruptible--; | |
767 | ||
371fd7e7 | 768 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
769 | } |
770 | ||
029632fb | 771 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
772 | { |
773 | if (task_contributes_to_load(p)) | |
774 | rq->nr_uninterruptible++; | |
775 | ||
371fd7e7 | 776 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
777 | } |
778 | ||
fe44d621 | 779 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 780 | { |
095c0aa8 GC |
781 | /* |
782 | * In theory, the compile should just see 0 here, and optimize out the call | |
783 | * to sched_rt_avg_update. But I don't trust it... | |
784 | */ | |
785 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
786 | s64 steal = 0, irq_delta = 0; | |
787 | #endif | |
788 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 789 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
790 | |
791 | /* | |
792 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
793 | * this case when a previous update_rq_clock() happened inside a | |
794 | * {soft,}irq region. | |
795 | * | |
796 | * When this happens, we stop ->clock_task and only update the | |
797 | * prev_irq_time stamp to account for the part that fit, so that a next | |
798 | * update will consume the rest. This ensures ->clock_task is | |
799 | * monotonic. | |
800 | * | |
801 | * It does however cause some slight miss-attribution of {soft,}irq | |
802 | * time, a more accurate solution would be to update the irq_time using | |
803 | * the current rq->clock timestamp, except that would require using | |
804 | * atomic ops. | |
805 | */ | |
806 | if (irq_delta > delta) | |
807 | irq_delta = delta; | |
808 | ||
809 | rq->prev_irq_time += irq_delta; | |
810 | delta -= irq_delta; | |
095c0aa8 GC |
811 | #endif |
812 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 813 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
814 | steal = paravirt_steal_clock(cpu_of(rq)); |
815 | steal -= rq->prev_steal_time_rq; | |
816 | ||
817 | if (unlikely(steal > delta)) | |
818 | steal = delta; | |
819 | ||
095c0aa8 | 820 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
821 | delta -= steal; |
822 | } | |
823 | #endif | |
824 | ||
fe44d621 PZ |
825 | rq->clock_task += delta; |
826 | ||
095c0aa8 | 827 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 828 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
829 | sched_rt_avg_update(rq, irq_delta + steal); |
830 | #endif | |
aa483808 VP |
831 | } |
832 | ||
34f971f6 PZ |
833 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
834 | { | |
835 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
836 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
837 | ||
838 | if (stop) { | |
839 | /* | |
840 | * Make it appear like a SCHED_FIFO task, its something | |
841 | * userspace knows about and won't get confused about. | |
842 | * | |
843 | * Also, it will make PI more or less work without too | |
844 | * much confusion -- but then, stop work should not | |
845 | * rely on PI working anyway. | |
846 | */ | |
847 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
848 | ||
849 | stop->sched_class = &stop_sched_class; | |
850 | } | |
851 | ||
852 | cpu_rq(cpu)->stop = stop; | |
853 | ||
854 | if (old_stop) { | |
855 | /* | |
856 | * Reset it back to a normal scheduling class so that | |
857 | * it can die in pieces. | |
858 | */ | |
859 | old_stop->sched_class = &rt_sched_class; | |
860 | } | |
861 | } | |
862 | ||
14531189 | 863 | /* |
dd41f596 | 864 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 865 | */ |
14531189 IM |
866 | static inline int __normal_prio(struct task_struct *p) |
867 | { | |
dd41f596 | 868 | return p->static_prio; |
14531189 IM |
869 | } |
870 | ||
b29739f9 IM |
871 | /* |
872 | * Calculate the expected normal priority: i.e. priority | |
873 | * without taking RT-inheritance into account. Might be | |
874 | * boosted by interactivity modifiers. Changes upon fork, | |
875 | * setprio syscalls, and whenever the interactivity | |
876 | * estimator recalculates. | |
877 | */ | |
36c8b586 | 878 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
879 | { |
880 | int prio; | |
881 | ||
aab03e05 DF |
882 | if (task_has_dl_policy(p)) |
883 | prio = MAX_DL_PRIO-1; | |
884 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
885 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
886 | else | |
887 | prio = __normal_prio(p); | |
888 | return prio; | |
889 | } | |
890 | ||
891 | /* | |
892 | * Calculate the current priority, i.e. the priority | |
893 | * taken into account by the scheduler. This value might | |
894 | * be boosted by RT tasks, or might be boosted by | |
895 | * interactivity modifiers. Will be RT if the task got | |
896 | * RT-boosted. If not then it returns p->normal_prio. | |
897 | */ | |
36c8b586 | 898 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
899 | { |
900 | p->normal_prio = normal_prio(p); | |
901 | /* | |
902 | * If we are RT tasks or we were boosted to RT priority, | |
903 | * keep the priority unchanged. Otherwise, update priority | |
904 | * to the normal priority: | |
905 | */ | |
906 | if (!rt_prio(p->prio)) | |
907 | return p->normal_prio; | |
908 | return p->prio; | |
909 | } | |
910 | ||
1da177e4 LT |
911 | /** |
912 | * task_curr - is this task currently executing on a CPU? | |
913 | * @p: the task in question. | |
e69f6186 YB |
914 | * |
915 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 916 | */ |
36c8b586 | 917 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
918 | { |
919 | return cpu_curr(task_cpu(p)) == p; | |
920 | } | |
921 | ||
67dfa1b7 | 922 | /* |
4c9a4bc8 PZ |
923 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
924 | * use the balance_callback list if you want balancing. | |
925 | * | |
926 | * this means any call to check_class_changed() must be followed by a call to | |
927 | * balance_callback(). | |
67dfa1b7 | 928 | */ |
cb469845 SR |
929 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
930 | const struct sched_class *prev_class, | |
da7a735e | 931 | int oldprio) |
cb469845 SR |
932 | { |
933 | if (prev_class != p->sched_class) { | |
934 | if (prev_class->switched_from) | |
da7a735e | 935 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 936 | |
da7a735e | 937 | p->sched_class->switched_to(rq, p); |
2d3d891d | 938 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 939 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
940 | } |
941 | ||
029632fb | 942 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
943 | { |
944 | const struct sched_class *class; | |
945 | ||
946 | if (p->sched_class == rq->curr->sched_class) { | |
947 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
948 | } else { | |
949 | for_each_class(class) { | |
950 | if (class == rq->curr->sched_class) | |
951 | break; | |
952 | if (class == p->sched_class) { | |
8875125e | 953 | resched_curr(rq); |
1e5a7405 PZ |
954 | break; |
955 | } | |
956 | } | |
957 | } | |
958 | ||
959 | /* | |
960 | * A queue event has occurred, and we're going to schedule. In | |
961 | * this case, we can save a useless back to back clock update. | |
962 | */ | |
da0c1e65 | 963 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 964 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
965 | } |
966 | ||
1da177e4 | 967 | #ifdef CONFIG_SMP |
5cc389bc PZ |
968 | /* |
969 | * This is how migration works: | |
970 | * | |
971 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
972 | * stop_one_cpu(). | |
973 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
974 | * off the CPU) | |
975 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
976 | * 4) if it's in the wrong runqueue then the migration thread removes | |
977 | * it and puts it into the right queue. | |
978 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
979 | * is done. | |
980 | */ | |
981 | ||
982 | /* | |
983 | * move_queued_task - move a queued task to new rq. | |
984 | * | |
985 | * Returns (locked) new rq. Old rq's lock is released. | |
986 | */ | |
5e16bbc2 | 987 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 988 | { |
5cc389bc PZ |
989 | lockdep_assert_held(&rq->lock); |
990 | ||
5cc389bc | 991 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 992 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
993 | set_task_cpu(p, new_cpu); |
994 | raw_spin_unlock(&rq->lock); | |
995 | ||
996 | rq = cpu_rq(new_cpu); | |
997 | ||
998 | raw_spin_lock(&rq->lock); | |
999 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 1000 | enqueue_task(rq, p, 0); |
3ea94de1 | 1001 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
1002 | check_preempt_curr(rq, p, 0); |
1003 | ||
1004 | return rq; | |
1005 | } | |
1006 | ||
1007 | struct migration_arg { | |
1008 | struct task_struct *task; | |
1009 | int dest_cpu; | |
1010 | }; | |
1011 | ||
1012 | /* | |
1013 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1014 | * this because either it can't run here any more (set_cpus_allowed() | |
1015 | * away from this CPU, or CPU going down), or because we're | |
1016 | * attempting to rebalance this task on exec (sched_exec). | |
1017 | * | |
1018 | * So we race with normal scheduler movements, but that's OK, as long | |
1019 | * as the task is no longer on this CPU. | |
5cc389bc | 1020 | */ |
5e16bbc2 | 1021 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1022 | { |
5cc389bc | 1023 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1024 | return rq; |
5cc389bc PZ |
1025 | |
1026 | /* Affinity changed (again). */ | |
1027 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1028 | return rq; |
5cc389bc | 1029 | |
5e16bbc2 PZ |
1030 | rq = move_queued_task(rq, p, dest_cpu); |
1031 | ||
1032 | return rq; | |
5cc389bc PZ |
1033 | } |
1034 | ||
1035 | /* | |
1036 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1037 | * and performs thread migration by bumping thread off CPU then | |
1038 | * 'pushing' onto another runqueue. | |
1039 | */ | |
1040 | static int migration_cpu_stop(void *data) | |
1041 | { | |
1042 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1043 | struct task_struct *p = arg->task; |
1044 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1045 | |
1046 | /* | |
1047 | * The original target cpu might have gone down and we might | |
1048 | * be on another cpu but it doesn't matter. | |
1049 | */ | |
1050 | local_irq_disable(); | |
1051 | /* | |
1052 | * We need to explicitly wake pending tasks before running | |
1053 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1054 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1055 | */ | |
1056 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1057 | |
1058 | raw_spin_lock(&p->pi_lock); | |
1059 | raw_spin_lock(&rq->lock); | |
1060 | /* | |
1061 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1062 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1063 | * we're holding p->pi_lock. | |
1064 | */ | |
1065 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1066 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1067 | raw_spin_unlock(&rq->lock); | |
1068 | raw_spin_unlock(&p->pi_lock); | |
1069 | ||
5cc389bc PZ |
1070 | local_irq_enable(); |
1071 | return 0; | |
1072 | } | |
1073 | ||
c5b28038 PZ |
1074 | /* |
1075 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1076 | * actually call this function. | |
1077 | */ | |
1078 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1079 | { |
5cc389bc PZ |
1080 | cpumask_copy(&p->cpus_allowed, new_mask); |
1081 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1082 | } | |
1083 | ||
c5b28038 PZ |
1084 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1085 | { | |
6c37067e PZ |
1086 | struct rq *rq = task_rq(p); |
1087 | bool queued, running; | |
1088 | ||
c5b28038 | 1089 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1090 | |
1091 | queued = task_on_rq_queued(p); | |
1092 | running = task_current(rq, p); | |
1093 | ||
1094 | if (queued) { | |
1095 | /* | |
1096 | * Because __kthread_bind() calls this on blocked tasks without | |
1097 | * holding rq->lock. | |
1098 | */ | |
1099 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1100 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1101 | } |
1102 | if (running) | |
1103 | put_prev_task(rq, p); | |
1104 | ||
c5b28038 | 1105 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e PZ |
1106 | |
1107 | if (running) | |
1108 | p->sched_class->set_curr_task(rq); | |
1109 | if (queued) | |
1de64443 | 1110 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
c5b28038 PZ |
1111 | } |
1112 | ||
5cc389bc PZ |
1113 | /* |
1114 | * Change a given task's CPU affinity. Migrate the thread to a | |
1115 | * proper CPU and schedule it away if the CPU it's executing on | |
1116 | * is removed from the allowed bitmask. | |
1117 | * | |
1118 | * NOTE: the caller must have a valid reference to the task, the | |
1119 | * task must not exit() & deallocate itself prematurely. The | |
1120 | * call is not atomic; no spinlocks may be held. | |
1121 | */ | |
25834c73 PZ |
1122 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1123 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1124 | { |
e9d867a6 | 1125 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1126 | unsigned int dest_cpu; |
eb580751 PZ |
1127 | struct rq_flags rf; |
1128 | struct rq *rq; | |
5cc389bc PZ |
1129 | int ret = 0; |
1130 | ||
eb580751 | 1131 | rq = task_rq_lock(p, &rf); |
5cc389bc | 1132 | |
e9d867a6 PZI |
1133 | if (p->flags & PF_KTHREAD) { |
1134 | /* | |
1135 | * Kernel threads are allowed on online && !active CPUs | |
1136 | */ | |
1137 | cpu_valid_mask = cpu_online_mask; | |
1138 | } | |
1139 | ||
25834c73 PZ |
1140 | /* |
1141 | * Must re-check here, to close a race against __kthread_bind(), | |
1142 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1143 | */ | |
1144 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1145 | ret = -EINVAL; | |
1146 | goto out; | |
1147 | } | |
1148 | ||
5cc389bc PZ |
1149 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1150 | goto out; | |
1151 | ||
e9d867a6 | 1152 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1153 | ret = -EINVAL; |
1154 | goto out; | |
1155 | } | |
1156 | ||
1157 | do_set_cpus_allowed(p, new_mask); | |
1158 | ||
e9d867a6 PZI |
1159 | if (p->flags & PF_KTHREAD) { |
1160 | /* | |
1161 | * For kernel threads that do indeed end up on online && | |
1162 | * !active we want to ensure they are strict per-cpu threads. | |
1163 | */ | |
1164 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1165 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1166 | p->nr_cpus_allowed != 1); | |
1167 | } | |
1168 | ||
5cc389bc PZ |
1169 | /* Can the task run on the task's current CPU? If so, we're done */ |
1170 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1171 | goto out; | |
1172 | ||
e9d867a6 | 1173 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1174 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1175 | struct migration_arg arg = { p, dest_cpu }; | |
1176 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1177 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1178 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1179 | tlb_migrate_finish(p->mm); | |
1180 | return 0; | |
cbce1a68 PZ |
1181 | } else if (task_on_rq_queued(p)) { |
1182 | /* | |
1183 | * OK, since we're going to drop the lock immediately | |
1184 | * afterwards anyway. | |
1185 | */ | |
e7904a28 | 1186 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
5e16bbc2 | 1187 | rq = move_queued_task(rq, p, dest_cpu); |
e7904a28 | 1188 | lockdep_repin_lock(&rq->lock, rf.cookie); |
cbce1a68 | 1189 | } |
5cc389bc | 1190 | out: |
eb580751 | 1191 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1192 | |
1193 | return ret; | |
1194 | } | |
25834c73 PZ |
1195 | |
1196 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1197 | { | |
1198 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1199 | } | |
5cc389bc PZ |
1200 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1201 | ||
dd41f596 | 1202 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1203 | { |
e2912009 PZ |
1204 | #ifdef CONFIG_SCHED_DEBUG |
1205 | /* | |
1206 | * We should never call set_task_cpu() on a blocked task, | |
1207 | * ttwu() will sort out the placement. | |
1208 | */ | |
077614ee | 1209 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1210 | !p->on_rq); |
0122ec5b | 1211 | |
3ea94de1 JP |
1212 | /* |
1213 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1214 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1215 | * time relying on p->on_rq. | |
1216 | */ | |
1217 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1218 | p->sched_class == &fair_sched_class && | |
1219 | (p->on_rq && !task_on_rq_migrating(p))); | |
1220 | ||
0122ec5b | 1221 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1222 | /* |
1223 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1224 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1225 | * | |
1226 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1227 | * see task_group(). |
6c6c54e1 PZ |
1228 | * |
1229 | * Furthermore, all task_rq users should acquire both locks, see | |
1230 | * task_rq_lock(). | |
1231 | */ | |
0122ec5b PZ |
1232 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1233 | lockdep_is_held(&task_rq(p)->lock))); | |
1234 | #endif | |
e2912009 PZ |
1235 | #endif |
1236 | ||
de1d7286 | 1237 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1238 | |
0c69774e | 1239 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1240 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1241 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1242 | p->se.nr_migrations++; |
ff303e66 | 1243 | perf_event_task_migrate(p); |
0c69774e | 1244 | } |
dd41f596 IM |
1245 | |
1246 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1247 | } |
1248 | ||
ac66f547 PZ |
1249 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1250 | { | |
da0c1e65 | 1251 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1252 | struct rq *src_rq, *dst_rq; |
1253 | ||
1254 | src_rq = task_rq(p); | |
1255 | dst_rq = cpu_rq(cpu); | |
1256 | ||
3ea94de1 | 1257 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1258 | deactivate_task(src_rq, p, 0); |
1259 | set_task_cpu(p, cpu); | |
1260 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1261 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1262 | check_preempt_curr(dst_rq, p, 0); |
1263 | } else { | |
1264 | /* | |
1265 | * Task isn't running anymore; make it appear like we migrated | |
1266 | * it before it went to sleep. This means on wakeup we make the | |
1267 | * previous cpu our targer instead of where it really is. | |
1268 | */ | |
1269 | p->wake_cpu = cpu; | |
1270 | } | |
1271 | } | |
1272 | ||
1273 | struct migration_swap_arg { | |
1274 | struct task_struct *src_task, *dst_task; | |
1275 | int src_cpu, dst_cpu; | |
1276 | }; | |
1277 | ||
1278 | static int migrate_swap_stop(void *data) | |
1279 | { | |
1280 | struct migration_swap_arg *arg = data; | |
1281 | struct rq *src_rq, *dst_rq; | |
1282 | int ret = -EAGAIN; | |
1283 | ||
62694cd5 PZ |
1284 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1285 | return -EAGAIN; | |
1286 | ||
ac66f547 PZ |
1287 | src_rq = cpu_rq(arg->src_cpu); |
1288 | dst_rq = cpu_rq(arg->dst_cpu); | |
1289 | ||
74602315 PZ |
1290 | double_raw_lock(&arg->src_task->pi_lock, |
1291 | &arg->dst_task->pi_lock); | |
ac66f547 | 1292 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1293 | |
ac66f547 PZ |
1294 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1295 | goto unlock; | |
1296 | ||
1297 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1298 | goto unlock; | |
1299 | ||
1300 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1301 | goto unlock; | |
1302 | ||
1303 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1304 | goto unlock; | |
1305 | ||
1306 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1307 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1308 | ||
1309 | ret = 0; | |
1310 | ||
1311 | unlock: | |
1312 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1313 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1314 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1315 | |
1316 | return ret; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Cross migrate two tasks | |
1321 | */ | |
1322 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1323 | { | |
1324 | struct migration_swap_arg arg; | |
1325 | int ret = -EINVAL; | |
1326 | ||
ac66f547 PZ |
1327 | arg = (struct migration_swap_arg){ |
1328 | .src_task = cur, | |
1329 | .src_cpu = task_cpu(cur), | |
1330 | .dst_task = p, | |
1331 | .dst_cpu = task_cpu(p), | |
1332 | }; | |
1333 | ||
1334 | if (arg.src_cpu == arg.dst_cpu) | |
1335 | goto out; | |
1336 | ||
6acce3ef PZ |
1337 | /* |
1338 | * These three tests are all lockless; this is OK since all of them | |
1339 | * will be re-checked with proper locks held further down the line. | |
1340 | */ | |
ac66f547 PZ |
1341 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1342 | goto out; | |
1343 | ||
1344 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1345 | goto out; | |
1346 | ||
1347 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1348 | goto out; | |
1349 | ||
286549dc | 1350 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1351 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1352 | ||
1353 | out: | |
ac66f547 PZ |
1354 | return ret; |
1355 | } | |
1356 | ||
1da177e4 LT |
1357 | /* |
1358 | * wait_task_inactive - wait for a thread to unschedule. | |
1359 | * | |
85ba2d86 RM |
1360 | * If @match_state is nonzero, it's the @p->state value just checked and |
1361 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1362 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1363 | * we return a positive number (its total switch count). If a second call | |
1364 | * a short while later returns the same number, the caller can be sure that | |
1365 | * @p has remained unscheduled the whole time. | |
1366 | * | |
1da177e4 LT |
1367 | * The caller must ensure that the task *will* unschedule sometime soon, |
1368 | * else this function might spin for a *long* time. This function can't | |
1369 | * be called with interrupts off, or it may introduce deadlock with | |
1370 | * smp_call_function() if an IPI is sent by the same process we are | |
1371 | * waiting to become inactive. | |
1372 | */ | |
85ba2d86 | 1373 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1374 | { |
da0c1e65 | 1375 | int running, queued; |
eb580751 | 1376 | struct rq_flags rf; |
85ba2d86 | 1377 | unsigned long ncsw; |
70b97a7f | 1378 | struct rq *rq; |
1da177e4 | 1379 | |
3a5c359a AK |
1380 | for (;;) { |
1381 | /* | |
1382 | * We do the initial early heuristics without holding | |
1383 | * any task-queue locks at all. We'll only try to get | |
1384 | * the runqueue lock when things look like they will | |
1385 | * work out! | |
1386 | */ | |
1387 | rq = task_rq(p); | |
fa490cfd | 1388 | |
3a5c359a AK |
1389 | /* |
1390 | * If the task is actively running on another CPU | |
1391 | * still, just relax and busy-wait without holding | |
1392 | * any locks. | |
1393 | * | |
1394 | * NOTE! Since we don't hold any locks, it's not | |
1395 | * even sure that "rq" stays as the right runqueue! | |
1396 | * But we don't care, since "task_running()" will | |
1397 | * return false if the runqueue has changed and p | |
1398 | * is actually now running somewhere else! | |
1399 | */ | |
85ba2d86 RM |
1400 | while (task_running(rq, p)) { |
1401 | if (match_state && unlikely(p->state != match_state)) | |
1402 | return 0; | |
3a5c359a | 1403 | cpu_relax(); |
85ba2d86 | 1404 | } |
fa490cfd | 1405 | |
3a5c359a AK |
1406 | /* |
1407 | * Ok, time to look more closely! We need the rq | |
1408 | * lock now, to be *sure*. If we're wrong, we'll | |
1409 | * just go back and repeat. | |
1410 | */ | |
eb580751 | 1411 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1412 | trace_sched_wait_task(p); |
3a5c359a | 1413 | running = task_running(rq, p); |
da0c1e65 | 1414 | queued = task_on_rq_queued(p); |
85ba2d86 | 1415 | ncsw = 0; |
f31e11d8 | 1416 | if (!match_state || p->state == match_state) |
93dcf55f | 1417 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1418 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1419 | |
85ba2d86 RM |
1420 | /* |
1421 | * If it changed from the expected state, bail out now. | |
1422 | */ | |
1423 | if (unlikely(!ncsw)) | |
1424 | break; | |
1425 | ||
3a5c359a AK |
1426 | /* |
1427 | * Was it really running after all now that we | |
1428 | * checked with the proper locks actually held? | |
1429 | * | |
1430 | * Oops. Go back and try again.. | |
1431 | */ | |
1432 | if (unlikely(running)) { | |
1433 | cpu_relax(); | |
1434 | continue; | |
1435 | } | |
fa490cfd | 1436 | |
3a5c359a AK |
1437 | /* |
1438 | * It's not enough that it's not actively running, | |
1439 | * it must be off the runqueue _entirely_, and not | |
1440 | * preempted! | |
1441 | * | |
80dd99b3 | 1442 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1443 | * running right now), it's preempted, and we should |
1444 | * yield - it could be a while. | |
1445 | */ | |
da0c1e65 | 1446 | if (unlikely(queued)) { |
8eb90c30 TG |
1447 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1448 | ||
1449 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1450 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1451 | continue; |
1452 | } | |
fa490cfd | 1453 | |
3a5c359a AK |
1454 | /* |
1455 | * Ahh, all good. It wasn't running, and it wasn't | |
1456 | * runnable, which means that it will never become | |
1457 | * running in the future either. We're all done! | |
1458 | */ | |
1459 | break; | |
1460 | } | |
85ba2d86 RM |
1461 | |
1462 | return ncsw; | |
1da177e4 LT |
1463 | } |
1464 | ||
1465 | /*** | |
1466 | * kick_process - kick a running thread to enter/exit the kernel | |
1467 | * @p: the to-be-kicked thread | |
1468 | * | |
1469 | * Cause a process which is running on another CPU to enter | |
1470 | * kernel-mode, without any delay. (to get signals handled.) | |
1471 | * | |
25985edc | 1472 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1473 | * because all it wants to ensure is that the remote task enters |
1474 | * the kernel. If the IPI races and the task has been migrated | |
1475 | * to another CPU then no harm is done and the purpose has been | |
1476 | * achieved as well. | |
1477 | */ | |
36c8b586 | 1478 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1479 | { |
1480 | int cpu; | |
1481 | ||
1482 | preempt_disable(); | |
1483 | cpu = task_cpu(p); | |
1484 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1485 | smp_send_reschedule(cpu); | |
1486 | preempt_enable(); | |
1487 | } | |
b43e3521 | 1488 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1489 | |
30da688e | 1490 | /* |
013fdb80 | 1491 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1492 | * |
1493 | * A few notes on cpu_active vs cpu_online: | |
1494 | * | |
1495 | * - cpu_active must be a subset of cpu_online | |
1496 | * | |
1497 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1498 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
1499 | * cpu isn't yet part of the sched domains, and balancing will not | |
1500 | * see it. | |
1501 | * | |
1502 | * - on cpu-down we clear cpu_active() to mask the sched domains and | |
1503 | * avoid the load balancer to place new tasks on the to be removed | |
1504 | * cpu. Existing tasks will remain running there and will be taken | |
1505 | * off. | |
1506 | * | |
1507 | * This means that fallback selection must not select !active CPUs. | |
1508 | * And can assume that any active CPU must be online. Conversely | |
1509 | * select_task_rq() below may allow selection of !active CPUs in order | |
1510 | * to satisfy the above rules. | |
30da688e | 1511 | */ |
5da9a0fb PZ |
1512 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1513 | { | |
aa00d89c TC |
1514 | int nid = cpu_to_node(cpu); |
1515 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1516 | enum { cpuset, possible, fail } state = cpuset; |
1517 | int dest_cpu; | |
5da9a0fb | 1518 | |
aa00d89c TC |
1519 | /* |
1520 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1521 | * will return -1. There is no cpu on the node, and we should | |
1522 | * select the cpu on the other node. | |
1523 | */ | |
1524 | if (nid != -1) { | |
1525 | nodemask = cpumask_of_node(nid); | |
1526 | ||
1527 | /* Look for allowed, online CPU in same node. */ | |
1528 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1529 | if (!cpu_active(dest_cpu)) |
1530 | continue; | |
1531 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1532 | return dest_cpu; | |
1533 | } | |
2baab4e9 | 1534 | } |
5da9a0fb | 1535 | |
2baab4e9 PZ |
1536 | for (;;) { |
1537 | /* Any allowed, online CPU? */ | |
e3831edd | 1538 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1539 | if (!cpu_active(dest_cpu)) |
1540 | continue; | |
1541 | goto out; | |
1542 | } | |
5da9a0fb | 1543 | |
e73e85f0 | 1544 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1545 | switch (state) { |
1546 | case cpuset: | |
e73e85f0 ON |
1547 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1548 | cpuset_cpus_allowed_fallback(p); | |
1549 | state = possible; | |
1550 | break; | |
1551 | } | |
1552 | /* fall-through */ | |
2baab4e9 PZ |
1553 | case possible: |
1554 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1555 | state = fail; | |
1556 | break; | |
1557 | ||
1558 | case fail: | |
1559 | BUG(); | |
1560 | break; | |
1561 | } | |
1562 | } | |
1563 | ||
1564 | out: | |
1565 | if (state != cpuset) { | |
1566 | /* | |
1567 | * Don't tell them about moving exiting tasks or | |
1568 | * kernel threads (both mm NULL), since they never | |
1569 | * leave kernel. | |
1570 | */ | |
1571 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1572 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1573 | task_pid_nr(p), p->comm, cpu); |
1574 | } | |
5da9a0fb PZ |
1575 | } |
1576 | ||
1577 | return dest_cpu; | |
1578 | } | |
1579 | ||
e2912009 | 1580 | /* |
013fdb80 | 1581 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1582 | */ |
970b13ba | 1583 | static inline |
ac66f547 | 1584 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1585 | { |
cbce1a68 PZ |
1586 | lockdep_assert_held(&p->pi_lock); |
1587 | ||
50605ffb | 1588 | if (tsk_nr_cpus_allowed(p) > 1) |
6c1d9410 | 1589 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 PZI |
1590 | else |
1591 | cpu = cpumask_any(tsk_cpus_allowed(p)); | |
e2912009 PZ |
1592 | |
1593 | /* | |
1594 | * In order not to call set_task_cpu() on a blocking task we need | |
1595 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1596 | * cpu. | |
1597 | * | |
1598 | * Since this is common to all placement strategies, this lives here. | |
1599 | * | |
1600 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1601 | * not worry about this generic constraint ] | |
1602 | */ | |
fa17b507 | 1603 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1604 | !cpu_online(cpu))) |
5da9a0fb | 1605 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1606 | |
1607 | return cpu; | |
970b13ba | 1608 | } |
09a40af5 MG |
1609 | |
1610 | static void update_avg(u64 *avg, u64 sample) | |
1611 | { | |
1612 | s64 diff = sample - *avg; | |
1613 | *avg += diff >> 3; | |
1614 | } | |
25834c73 PZ |
1615 | |
1616 | #else | |
1617 | ||
1618 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1619 | const struct cpumask *new_mask, bool check) | |
1620 | { | |
1621 | return set_cpus_allowed_ptr(p, new_mask); | |
1622 | } | |
1623 | ||
5cc389bc | 1624 | #endif /* CONFIG_SMP */ |
970b13ba | 1625 | |
d7c01d27 | 1626 | static void |
b84cb5df | 1627 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1628 | { |
d7c01d27 | 1629 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1630 | struct rq *rq = this_rq(); |
1631 | ||
d7c01d27 PZ |
1632 | #ifdef CONFIG_SMP |
1633 | int this_cpu = smp_processor_id(); | |
1634 | ||
1635 | if (cpu == this_cpu) { | |
1636 | schedstat_inc(rq, ttwu_local); | |
1637 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1638 | } else { | |
1639 | struct sched_domain *sd; | |
1640 | ||
1641 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1642 | rcu_read_lock(); |
d7c01d27 PZ |
1643 | for_each_domain(this_cpu, sd) { |
1644 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1645 | schedstat_inc(sd, ttwu_wake_remote); | |
1646 | break; | |
1647 | } | |
1648 | } | |
057f3fad | 1649 | rcu_read_unlock(); |
d7c01d27 | 1650 | } |
f339b9dc PZ |
1651 | |
1652 | if (wake_flags & WF_MIGRATED) | |
1653 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1654 | ||
d7c01d27 PZ |
1655 | #endif /* CONFIG_SMP */ |
1656 | ||
1657 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1658 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1659 | |
1660 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1661 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1662 | |
d7c01d27 PZ |
1663 | #endif /* CONFIG_SCHEDSTATS */ |
1664 | } | |
1665 | ||
1de64443 | 1666 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1667 | { |
9ed3811a | 1668 | activate_task(rq, p, en_flags); |
da0c1e65 | 1669 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1670 | |
1671 | /* if a worker is waking up, notify workqueue */ | |
1672 | if (p->flags & PF_WQ_WORKER) | |
1673 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1674 | } |
1675 | ||
23f41eeb PZ |
1676 | /* |
1677 | * Mark the task runnable and perform wakeup-preemption. | |
1678 | */ | |
e7904a28 PZ |
1679 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
1680 | struct pin_cookie cookie) | |
9ed3811a | 1681 | { |
9ed3811a | 1682 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1683 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1684 | trace_sched_wakeup(p); |
1685 | ||
9ed3811a | 1686 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1687 | if (p->sched_class->task_woken) { |
1688 | /* | |
cbce1a68 PZ |
1689 | * Our task @p is fully woken up and running; so its safe to |
1690 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1691 | */ |
e7904a28 | 1692 | lockdep_unpin_lock(&rq->lock, cookie); |
9ed3811a | 1693 | p->sched_class->task_woken(rq, p); |
e7904a28 | 1694 | lockdep_repin_lock(&rq->lock, cookie); |
4c9a4bc8 | 1695 | } |
9ed3811a | 1696 | |
e69c6341 | 1697 | if (rq->idle_stamp) { |
78becc27 | 1698 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1699 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1700 | |
abfafa54 JL |
1701 | update_avg(&rq->avg_idle, delta); |
1702 | ||
1703 | if (rq->avg_idle > max) | |
9ed3811a | 1704 | rq->avg_idle = max; |
abfafa54 | 1705 | |
9ed3811a TH |
1706 | rq->idle_stamp = 0; |
1707 | } | |
1708 | #endif | |
1709 | } | |
1710 | ||
c05fbafb | 1711 | static void |
e7904a28 PZ |
1712 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
1713 | struct pin_cookie cookie) | |
c05fbafb | 1714 | { |
b5179ac7 PZ |
1715 | int en_flags = ENQUEUE_WAKEUP; |
1716 | ||
cbce1a68 PZ |
1717 | lockdep_assert_held(&rq->lock); |
1718 | ||
c05fbafb PZ |
1719 | #ifdef CONFIG_SMP |
1720 | if (p->sched_contributes_to_load) | |
1721 | rq->nr_uninterruptible--; | |
b5179ac7 | 1722 | |
b5179ac7 | 1723 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1724 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1725 | #endif |
1726 | ||
b5179ac7 | 1727 | ttwu_activate(rq, p, en_flags); |
e7904a28 | 1728 | ttwu_do_wakeup(rq, p, wake_flags, cookie); |
c05fbafb PZ |
1729 | } |
1730 | ||
1731 | /* | |
1732 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1733 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1734 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1735 | * the task is still ->on_rq. | |
1736 | */ | |
1737 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1738 | { | |
eb580751 | 1739 | struct rq_flags rf; |
c05fbafb PZ |
1740 | struct rq *rq; |
1741 | int ret = 0; | |
1742 | ||
eb580751 | 1743 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1744 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1745 | /* check_preempt_curr() may use rq clock */ |
1746 | update_rq_clock(rq); | |
e7904a28 | 1747 | ttwu_do_wakeup(rq, p, wake_flags, rf.cookie); |
c05fbafb PZ |
1748 | ret = 1; |
1749 | } | |
eb580751 | 1750 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1751 | |
1752 | return ret; | |
1753 | } | |
1754 | ||
317f3941 | 1755 | #ifdef CONFIG_SMP |
e3baac47 | 1756 | void sched_ttwu_pending(void) |
317f3941 PZ |
1757 | { |
1758 | struct rq *rq = this_rq(); | |
fa14ff4a | 1759 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
e7904a28 | 1760 | struct pin_cookie cookie; |
fa14ff4a | 1761 | struct task_struct *p; |
e3baac47 | 1762 | unsigned long flags; |
317f3941 | 1763 | |
e3baac47 PZ |
1764 | if (!llist) |
1765 | return; | |
1766 | ||
1767 | raw_spin_lock_irqsave(&rq->lock, flags); | |
e7904a28 | 1768 | cookie = lockdep_pin_lock(&rq->lock); |
317f3941 | 1769 | |
fa14ff4a | 1770 | while (llist) { |
b7e7ade3 PZ |
1771 | int wake_flags = 0; |
1772 | ||
fa14ff4a PZ |
1773 | p = llist_entry(llist, struct task_struct, wake_entry); |
1774 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1775 | |
1776 | if (p->sched_remote_wakeup) | |
1777 | wake_flags = WF_MIGRATED; | |
1778 | ||
1779 | ttwu_do_activate(rq, p, wake_flags, cookie); | |
317f3941 PZ |
1780 | } |
1781 | ||
e7904a28 | 1782 | lockdep_unpin_lock(&rq->lock, cookie); |
e3baac47 | 1783 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1784 | } |
1785 | ||
1786 | void scheduler_ipi(void) | |
1787 | { | |
f27dde8d PZ |
1788 | /* |
1789 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1790 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1791 | * this IPI. | |
1792 | */ | |
8cb75e0c | 1793 | preempt_fold_need_resched(); |
f27dde8d | 1794 | |
fd2ac4f4 | 1795 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1796 | return; |
1797 | ||
1798 | /* | |
1799 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1800 | * traditionally all their work was done from the interrupt return | |
1801 | * path. Now that we actually do some work, we need to make sure | |
1802 | * we do call them. | |
1803 | * | |
1804 | * Some archs already do call them, luckily irq_enter/exit nest | |
1805 | * properly. | |
1806 | * | |
1807 | * Arguably we should visit all archs and update all handlers, | |
1808 | * however a fair share of IPIs are still resched only so this would | |
1809 | * somewhat pessimize the simple resched case. | |
1810 | */ | |
1811 | irq_enter(); | |
fa14ff4a | 1812 | sched_ttwu_pending(); |
ca38062e SS |
1813 | |
1814 | /* | |
1815 | * Check if someone kicked us for doing the nohz idle load balance. | |
1816 | */ | |
873b4c65 | 1817 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1818 | this_rq()->idle_balance = 1; |
ca38062e | 1819 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1820 | } |
c5d753a5 | 1821 | irq_exit(); |
317f3941 PZ |
1822 | } |
1823 | ||
b7e7ade3 | 1824 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1825 | { |
e3baac47 PZ |
1826 | struct rq *rq = cpu_rq(cpu); |
1827 | ||
b7e7ade3 PZ |
1828 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1829 | ||
e3baac47 PZ |
1830 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1831 | if (!set_nr_if_polling(rq->idle)) | |
1832 | smp_send_reschedule(cpu); | |
1833 | else | |
1834 | trace_sched_wake_idle_without_ipi(cpu); | |
1835 | } | |
317f3941 | 1836 | } |
d6aa8f85 | 1837 | |
f6be8af1 CL |
1838 | void wake_up_if_idle(int cpu) |
1839 | { | |
1840 | struct rq *rq = cpu_rq(cpu); | |
1841 | unsigned long flags; | |
1842 | ||
fd7de1e8 AL |
1843 | rcu_read_lock(); |
1844 | ||
1845 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1846 | goto out; | |
f6be8af1 CL |
1847 | |
1848 | if (set_nr_if_polling(rq->idle)) { | |
1849 | trace_sched_wake_idle_without_ipi(cpu); | |
1850 | } else { | |
1851 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1852 | if (is_idle_task(rq->curr)) | |
1853 | smp_send_reschedule(cpu); | |
1854 | /* Else cpu is not in idle, do nothing here */ | |
1855 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1856 | } | |
fd7de1e8 AL |
1857 | |
1858 | out: | |
1859 | rcu_read_unlock(); | |
f6be8af1 CL |
1860 | } |
1861 | ||
39be3501 | 1862 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1863 | { |
1864 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1865 | } | |
d6aa8f85 | 1866 | #endif /* CONFIG_SMP */ |
317f3941 | 1867 | |
b5179ac7 | 1868 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1869 | { |
1870 | struct rq *rq = cpu_rq(cpu); | |
e7904a28 | 1871 | struct pin_cookie cookie; |
c05fbafb | 1872 | |
17d9f311 | 1873 | #if defined(CONFIG_SMP) |
39be3501 | 1874 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1875 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
b7e7ade3 | 1876 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1877 | return; |
1878 | } | |
1879 | #endif | |
1880 | ||
c05fbafb | 1881 | raw_spin_lock(&rq->lock); |
e7904a28 | 1882 | cookie = lockdep_pin_lock(&rq->lock); |
b5179ac7 | 1883 | ttwu_do_activate(rq, p, wake_flags, cookie); |
e7904a28 | 1884 | lockdep_unpin_lock(&rq->lock, cookie); |
c05fbafb | 1885 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1886 | } |
1887 | ||
8643cda5 PZ |
1888 | /* |
1889 | * Notes on Program-Order guarantees on SMP systems. | |
1890 | * | |
1891 | * MIGRATION | |
1892 | * | |
1893 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1894 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1895 | * execution on its new cpu [c1]. | |
1896 | * | |
1897 | * For migration (of runnable tasks) this is provided by the following means: | |
1898 | * | |
1899 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1900 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1901 | * rq(c1)->lock (if not at the same time, then in that order). | |
1902 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1903 | * | |
1904 | * Transitivity guarantees that B happens after A and C after B. | |
1905 | * Note: we only require RCpc transitivity. | |
1906 | * Note: the cpu doing B need not be c0 or c1 | |
1907 | * | |
1908 | * Example: | |
1909 | * | |
1910 | * CPU0 CPU1 CPU2 | |
1911 | * | |
1912 | * LOCK rq(0)->lock | |
1913 | * sched-out X | |
1914 | * sched-in Y | |
1915 | * UNLOCK rq(0)->lock | |
1916 | * | |
1917 | * LOCK rq(0)->lock // orders against CPU0 | |
1918 | * dequeue X | |
1919 | * UNLOCK rq(0)->lock | |
1920 | * | |
1921 | * LOCK rq(1)->lock | |
1922 | * enqueue X | |
1923 | * UNLOCK rq(1)->lock | |
1924 | * | |
1925 | * LOCK rq(1)->lock // orders against CPU2 | |
1926 | * sched-out Z | |
1927 | * sched-in X | |
1928 | * UNLOCK rq(1)->lock | |
1929 | * | |
1930 | * | |
1931 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1932 | * | |
1933 | * For blocking we (obviously) need to provide the same guarantee as for | |
1934 | * migration. However the means are completely different as there is no lock | |
1935 | * chain to provide order. Instead we do: | |
1936 | * | |
1937 | * 1) smp_store_release(X->on_cpu, 0) | |
1938 | * 2) smp_cond_acquire(!X->on_cpu) | |
1939 | * | |
1940 | * Example: | |
1941 | * | |
1942 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1943 | * | |
1944 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1945 | * dequeue X | |
1946 | * sched-out X | |
1947 | * smp_store_release(X->on_cpu, 0); | |
1948 | * | |
1949 | * smp_cond_acquire(!X->on_cpu); | |
1950 | * X->state = WAKING | |
1951 | * set_task_cpu(X,2) | |
1952 | * | |
1953 | * LOCK rq(2)->lock | |
1954 | * enqueue X | |
1955 | * X->state = RUNNING | |
1956 | * UNLOCK rq(2)->lock | |
1957 | * | |
1958 | * LOCK rq(2)->lock // orders against CPU1 | |
1959 | * sched-out Z | |
1960 | * sched-in X | |
1961 | * UNLOCK rq(2)->lock | |
1962 | * | |
1963 | * UNLOCK X->pi_lock | |
1964 | * UNLOCK rq(0)->lock | |
1965 | * | |
1966 | * | |
1967 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1968 | * must observe the state that lead to our wakeup. That is, not only must our | |
1969 | * task observe its own prior state, it must also observe the stores prior to | |
1970 | * its wakeup. | |
1971 | * | |
1972 | * This means that any means of doing remote wakeups must order the CPU doing | |
1973 | * the wakeup against the CPU the task is going to end up running on. This, | |
1974 | * however, is already required for the regular Program-Order guarantee above, | |
1975 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_acquire). | |
1976 | * | |
1977 | */ | |
1978 | ||
9ed3811a | 1979 | /** |
1da177e4 | 1980 | * try_to_wake_up - wake up a thread |
9ed3811a | 1981 | * @p: the thread to be awakened |
1da177e4 | 1982 | * @state: the mask of task states that can be woken |
9ed3811a | 1983 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1984 | * |
1985 | * Put it on the run-queue if it's not already there. The "current" | |
1986 | * thread is always on the run-queue (except when the actual | |
1987 | * re-schedule is in progress), and as such you're allowed to do | |
1988 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1989 | * runnable without the overhead of this. | |
1990 | * | |
e69f6186 | 1991 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1992 | * or @state didn't match @p's state. |
1da177e4 | 1993 | */ |
e4a52bcb PZ |
1994 | static int |
1995 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1996 | { |
1da177e4 | 1997 | unsigned long flags; |
c05fbafb | 1998 | int cpu, success = 0; |
2398f2c6 | 1999 | |
e0acd0a6 ON |
2000 | /* |
2001 | * If we are going to wake up a thread waiting for CONDITION we | |
2002 | * need to ensure that CONDITION=1 done by the caller can not be | |
2003 | * reordered with p->state check below. This pairs with mb() in | |
2004 | * set_current_state() the waiting thread does. | |
2005 | */ | |
2006 | smp_mb__before_spinlock(); | |
013fdb80 | 2007 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2008 | if (!(p->state & state)) |
1da177e4 LT |
2009 | goto out; |
2010 | ||
fbd705a0 PZ |
2011 | trace_sched_waking(p); |
2012 | ||
c05fbafb | 2013 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2014 | cpu = task_cpu(p); |
1da177e4 | 2015 | |
c05fbafb PZ |
2016 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2017 | goto stat; | |
1da177e4 | 2018 | |
1da177e4 | 2019 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2020 | /* |
2021 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2022 | * possible to, falsely, observe p->on_cpu == 0. | |
2023 | * | |
2024 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2025 | * from the runqueue. | |
2026 | * | |
2027 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2028 | * UNLOCK rq->lock | |
2029 | * RMB | |
2030 | * LOCK rq->lock | |
2031 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2032 | * | |
2033 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2034 | * from the consecutive calls to schedule(); the first switching to our | |
2035 | * task, the second putting it to sleep. | |
2036 | */ | |
2037 | smp_rmb(); | |
2038 | ||
e9c84311 | 2039 | /* |
c05fbafb PZ |
2040 | * If the owning (remote) cpu is still in the middle of schedule() with |
2041 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
2042 | * |
2043 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2044 | * | |
2045 | * This ensures that tasks getting woken will be fully ordered against | |
2046 | * their previous state and preserve Program Order. | |
0970d299 | 2047 | */ |
b3e0b1b6 | 2048 | smp_cond_acquire(!p->on_cpu); |
1da177e4 | 2049 | |
a8e4f2ea | 2050 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2051 | p->state = TASK_WAKING; |
e7693a36 | 2052 | |
ac66f547 | 2053 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2054 | if (task_cpu(p) != cpu) { |
2055 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2056 | set_task_cpu(p, cpu); |
f339b9dc | 2057 | } |
1da177e4 | 2058 | #endif /* CONFIG_SMP */ |
1da177e4 | 2059 | |
b5179ac7 | 2060 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2061 | stat: |
cb251765 MG |
2062 | if (schedstat_enabled()) |
2063 | ttwu_stat(p, cpu, wake_flags); | |
1da177e4 | 2064 | out: |
013fdb80 | 2065 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2066 | |
2067 | return success; | |
2068 | } | |
2069 | ||
21aa9af0 TH |
2070 | /** |
2071 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2072 | * @p: the thread to be awakened | |
2073 | * | |
2acca55e | 2074 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2075 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2076 | * the current task. |
21aa9af0 | 2077 | */ |
e7904a28 | 2078 | static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie) |
21aa9af0 TH |
2079 | { |
2080 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2081 | |
383efcd0 TH |
2082 | if (WARN_ON_ONCE(rq != this_rq()) || |
2083 | WARN_ON_ONCE(p == current)) | |
2084 | return; | |
2085 | ||
21aa9af0 TH |
2086 | lockdep_assert_held(&rq->lock); |
2087 | ||
2acca55e | 2088 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2089 | /* |
2090 | * This is OK, because current is on_cpu, which avoids it being | |
2091 | * picked for load-balance and preemption/IRQs are still | |
2092 | * disabled avoiding further scheduler activity on it and we've | |
2093 | * not yet picked a replacement task. | |
2094 | */ | |
e7904a28 | 2095 | lockdep_unpin_lock(&rq->lock, cookie); |
2acca55e PZ |
2096 | raw_spin_unlock(&rq->lock); |
2097 | raw_spin_lock(&p->pi_lock); | |
2098 | raw_spin_lock(&rq->lock); | |
e7904a28 | 2099 | lockdep_repin_lock(&rq->lock, cookie); |
2acca55e PZ |
2100 | } |
2101 | ||
21aa9af0 | 2102 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2103 | goto out; |
21aa9af0 | 2104 | |
fbd705a0 PZ |
2105 | trace_sched_waking(p); |
2106 | ||
da0c1e65 | 2107 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2108 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2109 | ||
e7904a28 | 2110 | ttwu_do_wakeup(rq, p, 0, cookie); |
cb251765 MG |
2111 | if (schedstat_enabled()) |
2112 | ttwu_stat(p, smp_processor_id(), 0); | |
2acca55e PZ |
2113 | out: |
2114 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2115 | } |
2116 | ||
50fa610a DH |
2117 | /** |
2118 | * wake_up_process - Wake up a specific process | |
2119 | * @p: The process to be woken up. | |
2120 | * | |
2121 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2122 | * processes. |
2123 | * | |
2124 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2125 | * |
2126 | * It may be assumed that this function implies a write memory barrier before | |
2127 | * changing the task state if and only if any tasks are woken up. | |
2128 | */ | |
7ad5b3a5 | 2129 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2130 | { |
9067ac85 | 2131 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2132 | } |
1da177e4 LT |
2133 | EXPORT_SYMBOL(wake_up_process); |
2134 | ||
7ad5b3a5 | 2135 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2136 | { |
2137 | return try_to_wake_up(p, state, 0); | |
2138 | } | |
2139 | ||
a5e7be3b JL |
2140 | /* |
2141 | * This function clears the sched_dl_entity static params. | |
2142 | */ | |
2143 | void __dl_clear_params(struct task_struct *p) | |
2144 | { | |
2145 | struct sched_dl_entity *dl_se = &p->dl; | |
2146 | ||
2147 | dl_se->dl_runtime = 0; | |
2148 | dl_se->dl_deadline = 0; | |
2149 | dl_se->dl_period = 0; | |
2150 | dl_se->flags = 0; | |
2151 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2152 | |
2153 | dl_se->dl_throttled = 0; | |
40767b0d | 2154 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2155 | } |
2156 | ||
1da177e4 LT |
2157 | /* |
2158 | * Perform scheduler related setup for a newly forked process p. | |
2159 | * p is forked by current. | |
dd41f596 IM |
2160 | * |
2161 | * __sched_fork() is basic setup used by init_idle() too: | |
2162 | */ | |
5e1576ed | 2163 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2164 | { |
fd2f4419 PZ |
2165 | p->on_rq = 0; |
2166 | ||
2167 | p->se.on_rq = 0; | |
dd41f596 IM |
2168 | p->se.exec_start = 0; |
2169 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2170 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2171 | p->se.nr_migrations = 0; |
da7a735e | 2172 | p->se.vruntime = 0; |
fd2f4419 | 2173 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2174 | |
ad936d86 BP |
2175 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2176 | p->se.cfs_rq = NULL; | |
2177 | #endif | |
2178 | ||
6cfb0d5d | 2179 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2180 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2181 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2182 | #endif |
476d139c | 2183 | |
aab03e05 | 2184 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2185 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2186 | __dl_clear_params(p); |
aab03e05 | 2187 | |
fa717060 | 2188 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2189 | p->rt.timeout = 0; |
2190 | p->rt.time_slice = sched_rr_timeslice; | |
2191 | p->rt.on_rq = 0; | |
2192 | p->rt.on_list = 0; | |
476d139c | 2193 | |
e107be36 AK |
2194 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2195 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2196 | #endif | |
cbee9f88 PZ |
2197 | |
2198 | #ifdef CONFIG_NUMA_BALANCING | |
2199 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2200 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2201 | p->mm->numa_scan_seq = 0; |
2202 | } | |
2203 | ||
5e1576ed RR |
2204 | if (clone_flags & CLONE_VM) |
2205 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2206 | else | |
2207 | p->numa_preferred_nid = -1; | |
2208 | ||
cbee9f88 PZ |
2209 | p->node_stamp = 0ULL; |
2210 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2211 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2212 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2213 | p->numa_faults = NULL; |
7e2703e6 RR |
2214 | p->last_task_numa_placement = 0; |
2215 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2216 | |
8c8a743c | 2217 | p->numa_group = NULL; |
cbee9f88 | 2218 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2219 | } |
2220 | ||
2a595721 SD |
2221 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2222 | ||
1a687c2e | 2223 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2224 | |
1a687c2e MG |
2225 | void set_numabalancing_state(bool enabled) |
2226 | { | |
2227 | if (enabled) | |
2a595721 | 2228 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2229 | else |
2a595721 | 2230 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2231 | } |
54a43d54 AK |
2232 | |
2233 | #ifdef CONFIG_PROC_SYSCTL | |
2234 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2235 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2236 | { | |
2237 | struct ctl_table t; | |
2238 | int err; | |
2a595721 | 2239 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2240 | |
2241 | if (write && !capable(CAP_SYS_ADMIN)) | |
2242 | return -EPERM; | |
2243 | ||
2244 | t = *table; | |
2245 | t.data = &state; | |
2246 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2247 | if (err < 0) | |
2248 | return err; | |
2249 | if (write) | |
2250 | set_numabalancing_state(state); | |
2251 | return err; | |
2252 | } | |
2253 | #endif | |
2254 | #endif | |
dd41f596 | 2255 | |
cb251765 MG |
2256 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
2257 | ||
2258 | #ifdef CONFIG_SCHEDSTATS | |
2259 | static void set_schedstats(bool enabled) | |
2260 | { | |
2261 | if (enabled) | |
2262 | static_branch_enable(&sched_schedstats); | |
2263 | else | |
2264 | static_branch_disable(&sched_schedstats); | |
2265 | } | |
2266 | ||
2267 | void force_schedstat_enabled(void) | |
2268 | { | |
2269 | if (!schedstat_enabled()) { | |
2270 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2271 | static_branch_enable(&sched_schedstats); | |
2272 | } | |
2273 | } | |
2274 | ||
2275 | static int __init setup_schedstats(char *str) | |
2276 | { | |
2277 | int ret = 0; | |
2278 | if (!str) | |
2279 | goto out; | |
2280 | ||
2281 | if (!strcmp(str, "enable")) { | |
2282 | set_schedstats(true); | |
2283 | ret = 1; | |
2284 | } else if (!strcmp(str, "disable")) { | |
2285 | set_schedstats(false); | |
2286 | ret = 1; | |
2287 | } | |
2288 | out: | |
2289 | if (!ret) | |
2290 | pr_warn("Unable to parse schedstats=\n"); | |
2291 | ||
2292 | return ret; | |
2293 | } | |
2294 | __setup("schedstats=", setup_schedstats); | |
2295 | ||
2296 | #ifdef CONFIG_PROC_SYSCTL | |
2297 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2298 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2299 | { | |
2300 | struct ctl_table t; | |
2301 | int err; | |
2302 | int state = static_branch_likely(&sched_schedstats); | |
2303 | ||
2304 | if (write && !capable(CAP_SYS_ADMIN)) | |
2305 | return -EPERM; | |
2306 | ||
2307 | t = *table; | |
2308 | t.data = &state; | |
2309 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2310 | if (err < 0) | |
2311 | return err; | |
2312 | if (write) | |
2313 | set_schedstats(state); | |
2314 | return err; | |
2315 | } | |
2316 | #endif | |
2317 | #endif | |
dd41f596 IM |
2318 | |
2319 | /* | |
2320 | * fork()/clone()-time setup: | |
2321 | */ | |
aab03e05 | 2322 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2323 | { |
0122ec5b | 2324 | unsigned long flags; |
dd41f596 IM |
2325 | int cpu = get_cpu(); |
2326 | ||
5e1576ed | 2327 | __sched_fork(clone_flags, p); |
06b83b5f | 2328 | /* |
0017d735 | 2329 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2330 | * nobody will actually run it, and a signal or other external |
2331 | * event cannot wake it up and insert it on the runqueue either. | |
2332 | */ | |
0017d735 | 2333 | p->state = TASK_RUNNING; |
dd41f596 | 2334 | |
c350a04e MG |
2335 | /* |
2336 | * Make sure we do not leak PI boosting priority to the child. | |
2337 | */ | |
2338 | p->prio = current->normal_prio; | |
2339 | ||
b9dc29e7 MG |
2340 | /* |
2341 | * Revert to default priority/policy on fork if requested. | |
2342 | */ | |
2343 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2344 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2345 | p->policy = SCHED_NORMAL; |
6c697bdf | 2346 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2347 | p->rt_priority = 0; |
2348 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2349 | p->static_prio = NICE_TO_PRIO(0); | |
2350 | ||
2351 | p->prio = p->normal_prio = __normal_prio(p); | |
2352 | set_load_weight(p); | |
6c697bdf | 2353 | |
b9dc29e7 MG |
2354 | /* |
2355 | * We don't need the reset flag anymore after the fork. It has | |
2356 | * fulfilled its duty: | |
2357 | */ | |
2358 | p->sched_reset_on_fork = 0; | |
2359 | } | |
ca94c442 | 2360 | |
aab03e05 DF |
2361 | if (dl_prio(p->prio)) { |
2362 | put_cpu(); | |
2363 | return -EAGAIN; | |
2364 | } else if (rt_prio(p->prio)) { | |
2365 | p->sched_class = &rt_sched_class; | |
2366 | } else { | |
2ddbf952 | 2367 | p->sched_class = &fair_sched_class; |
aab03e05 | 2368 | } |
b29739f9 | 2369 | |
cd29fe6f PZ |
2370 | if (p->sched_class->task_fork) |
2371 | p->sched_class->task_fork(p); | |
2372 | ||
86951599 PZ |
2373 | /* |
2374 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2375 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2376 | * is ran before sched_fork(). | |
2377 | * | |
2378 | * Silence PROVE_RCU. | |
2379 | */ | |
0122ec5b | 2380 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2381 | set_task_cpu(p, cpu); |
0122ec5b | 2382 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2383 | |
f6db8347 | 2384 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2385 | if (likely(sched_info_on())) |
52f17b6c | 2386 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2387 | #endif |
3ca7a440 PZ |
2388 | #if defined(CONFIG_SMP) |
2389 | p->on_cpu = 0; | |
4866cde0 | 2390 | #endif |
01028747 | 2391 | init_task_preempt_count(p); |
806c09a7 | 2392 | #ifdef CONFIG_SMP |
917b627d | 2393 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2394 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2395 | #endif |
917b627d | 2396 | |
476d139c | 2397 | put_cpu(); |
aab03e05 | 2398 | return 0; |
1da177e4 LT |
2399 | } |
2400 | ||
332ac17e DF |
2401 | unsigned long to_ratio(u64 period, u64 runtime) |
2402 | { | |
2403 | if (runtime == RUNTIME_INF) | |
2404 | return 1ULL << 20; | |
2405 | ||
2406 | /* | |
2407 | * Doing this here saves a lot of checks in all | |
2408 | * the calling paths, and returning zero seems | |
2409 | * safe for them anyway. | |
2410 | */ | |
2411 | if (period == 0) | |
2412 | return 0; | |
2413 | ||
2414 | return div64_u64(runtime << 20, period); | |
2415 | } | |
2416 | ||
2417 | #ifdef CONFIG_SMP | |
2418 | inline struct dl_bw *dl_bw_of(int i) | |
2419 | { | |
f78f5b90 PM |
2420 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2421 | "sched RCU must be held"); | |
332ac17e DF |
2422 | return &cpu_rq(i)->rd->dl_bw; |
2423 | } | |
2424 | ||
de212f18 | 2425 | static inline int dl_bw_cpus(int i) |
332ac17e | 2426 | { |
de212f18 PZ |
2427 | struct root_domain *rd = cpu_rq(i)->rd; |
2428 | int cpus = 0; | |
2429 | ||
f78f5b90 PM |
2430 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2431 | "sched RCU must be held"); | |
de212f18 PZ |
2432 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2433 | cpus++; | |
2434 | ||
2435 | return cpus; | |
332ac17e DF |
2436 | } |
2437 | #else | |
2438 | inline struct dl_bw *dl_bw_of(int i) | |
2439 | { | |
2440 | return &cpu_rq(i)->dl.dl_bw; | |
2441 | } | |
2442 | ||
de212f18 | 2443 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2444 | { |
2445 | return 1; | |
2446 | } | |
2447 | #endif | |
2448 | ||
332ac17e DF |
2449 | /* |
2450 | * We must be sure that accepting a new task (or allowing changing the | |
2451 | * parameters of an existing one) is consistent with the bandwidth | |
2452 | * constraints. If yes, this function also accordingly updates the currently | |
2453 | * allocated bandwidth to reflect the new situation. | |
2454 | * | |
2455 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2456 | * |
2457 | * XXX we should delay bw change until the task's 0-lag point, see | |
2458 | * __setparam_dl(). | |
332ac17e DF |
2459 | */ |
2460 | static int dl_overflow(struct task_struct *p, int policy, | |
2461 | const struct sched_attr *attr) | |
2462 | { | |
2463 | ||
2464 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2465 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2466 | u64 runtime = attr->sched_runtime; |
2467 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2468 | int cpus, err = -1; |
332ac17e | 2469 | |
fec148c0 XP |
2470 | /* !deadline task may carry old deadline bandwidth */ |
2471 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2472 | return 0; |
2473 | ||
2474 | /* | |
2475 | * Either if a task, enters, leave, or stays -deadline but changes | |
2476 | * its parameters, we may need to update accordingly the total | |
2477 | * allocated bandwidth of the container. | |
2478 | */ | |
2479 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2480 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2481 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2482 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2483 | __dl_add(dl_b, new_bw); | |
2484 | err = 0; | |
2485 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2486 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2487 | __dl_clear(dl_b, p->dl.dl_bw); | |
2488 | __dl_add(dl_b, new_bw); | |
2489 | err = 0; | |
2490 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2491 | __dl_clear(dl_b, p->dl.dl_bw); | |
2492 | err = 0; | |
2493 | } | |
2494 | raw_spin_unlock(&dl_b->lock); | |
2495 | ||
2496 | return err; | |
2497 | } | |
2498 | ||
2499 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2500 | ||
1da177e4 LT |
2501 | /* |
2502 | * wake_up_new_task - wake up a newly created task for the first time. | |
2503 | * | |
2504 | * This function will do some initial scheduler statistics housekeeping | |
2505 | * that must be done for every newly created context, then puts the task | |
2506 | * on the runqueue and wakes it. | |
2507 | */ | |
3e51e3ed | 2508 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2509 | { |
eb580751 | 2510 | struct rq_flags rf; |
dd41f596 | 2511 | struct rq *rq; |
fabf318e | 2512 | |
98d8fd81 MR |
2513 | /* Initialize new task's runnable average */ |
2514 | init_entity_runnable_average(&p->se); | |
eb580751 | 2515 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
fabf318e PZ |
2516 | #ifdef CONFIG_SMP |
2517 | /* | |
2518 | * Fork balancing, do it here and not earlier because: | |
2519 | * - cpus_allowed can change in the fork path | |
2520 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2521 | */ |
ac66f547 | 2522 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2523 | #endif |
2b8c41da YD |
2524 | /* Post initialize new task's util average when its cfs_rq is set */ |
2525 | post_init_entity_util_avg(&p->se); | |
0017d735 | 2526 | |
eb580751 | 2527 | rq = __task_rq_lock(p, &rf); |
cd29fe6f | 2528 | activate_task(rq, p, 0); |
da0c1e65 | 2529 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2530 | trace_sched_wakeup_new(p); |
a7558e01 | 2531 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2532 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2533 | if (p->sched_class->task_woken) { |
2534 | /* | |
2535 | * Nothing relies on rq->lock after this, so its fine to | |
2536 | * drop it. | |
2537 | */ | |
e7904a28 | 2538 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
efbbd05a | 2539 | p->sched_class->task_woken(rq, p); |
e7904a28 | 2540 | lockdep_repin_lock(&rq->lock, rf.cookie); |
0aaafaab | 2541 | } |
9a897c5a | 2542 | #endif |
eb580751 | 2543 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2544 | } |
2545 | ||
e107be36 AK |
2546 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2547 | ||
1cde2930 PZ |
2548 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2549 | ||
2ecd9d29 PZ |
2550 | void preempt_notifier_inc(void) |
2551 | { | |
2552 | static_key_slow_inc(&preempt_notifier_key); | |
2553 | } | |
2554 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2555 | ||
2556 | void preempt_notifier_dec(void) | |
2557 | { | |
2558 | static_key_slow_dec(&preempt_notifier_key); | |
2559 | } | |
2560 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2561 | ||
e107be36 | 2562 | /** |
80dd99b3 | 2563 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2564 | * @notifier: notifier struct to register |
e107be36 AK |
2565 | */ |
2566 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2567 | { | |
2ecd9d29 PZ |
2568 | if (!static_key_false(&preempt_notifier_key)) |
2569 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2570 | ||
e107be36 AK |
2571 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2572 | } | |
2573 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2574 | ||
2575 | /** | |
2576 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2577 | * @notifier: notifier struct to unregister |
e107be36 | 2578 | * |
d84525a8 | 2579 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2580 | */ |
2581 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2582 | { | |
2583 | hlist_del(¬ifier->link); | |
2584 | } | |
2585 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2586 | ||
1cde2930 | 2587 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2588 | { |
2589 | struct preempt_notifier *notifier; | |
e107be36 | 2590 | |
b67bfe0d | 2591 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2592 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2593 | } | |
2594 | ||
1cde2930 PZ |
2595 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2596 | { | |
2597 | if (static_key_false(&preempt_notifier_key)) | |
2598 | __fire_sched_in_preempt_notifiers(curr); | |
2599 | } | |
2600 | ||
e107be36 | 2601 | static void |
1cde2930 PZ |
2602 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2603 | struct task_struct *next) | |
e107be36 AK |
2604 | { |
2605 | struct preempt_notifier *notifier; | |
e107be36 | 2606 | |
b67bfe0d | 2607 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2608 | notifier->ops->sched_out(notifier, next); |
2609 | } | |
2610 | ||
1cde2930 PZ |
2611 | static __always_inline void |
2612 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2613 | struct task_struct *next) | |
2614 | { | |
2615 | if (static_key_false(&preempt_notifier_key)) | |
2616 | __fire_sched_out_preempt_notifiers(curr, next); | |
2617 | } | |
2618 | ||
6d6bc0ad | 2619 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2620 | |
1cde2930 | 2621 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2622 | { |
2623 | } | |
2624 | ||
1cde2930 | 2625 | static inline void |
e107be36 AK |
2626 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2627 | struct task_struct *next) | |
2628 | { | |
2629 | } | |
2630 | ||
6d6bc0ad | 2631 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2632 | |
4866cde0 NP |
2633 | /** |
2634 | * prepare_task_switch - prepare to switch tasks | |
2635 | * @rq: the runqueue preparing to switch | |
421cee29 | 2636 | * @prev: the current task that is being switched out |
4866cde0 NP |
2637 | * @next: the task we are going to switch to. |
2638 | * | |
2639 | * This is called with the rq lock held and interrupts off. It must | |
2640 | * be paired with a subsequent finish_task_switch after the context | |
2641 | * switch. | |
2642 | * | |
2643 | * prepare_task_switch sets up locking and calls architecture specific | |
2644 | * hooks. | |
2645 | */ | |
e107be36 AK |
2646 | static inline void |
2647 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2648 | struct task_struct *next) | |
4866cde0 | 2649 | { |
43148951 | 2650 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2651 | perf_event_task_sched_out(prev, next); |
e107be36 | 2652 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2653 | prepare_lock_switch(rq, next); |
2654 | prepare_arch_switch(next); | |
2655 | } | |
2656 | ||
1da177e4 LT |
2657 | /** |
2658 | * finish_task_switch - clean up after a task-switch | |
2659 | * @prev: the thread we just switched away from. | |
2660 | * | |
4866cde0 NP |
2661 | * finish_task_switch must be called after the context switch, paired |
2662 | * with a prepare_task_switch call before the context switch. | |
2663 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2664 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2665 | * |
2666 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2667 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2668 | * with the lock held can cause deadlocks; see schedule() for |
2669 | * details.) | |
dfa50b60 ON |
2670 | * |
2671 | * The context switch have flipped the stack from under us and restored the | |
2672 | * local variables which were saved when this task called schedule() in the | |
2673 | * past. prev == current is still correct but we need to recalculate this_rq | |
2674 | * because prev may have moved to another CPU. | |
1da177e4 | 2675 | */ |
dfa50b60 | 2676 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2677 | __releases(rq->lock) |
2678 | { | |
dfa50b60 | 2679 | struct rq *rq = this_rq(); |
1da177e4 | 2680 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2681 | long prev_state; |
1da177e4 | 2682 | |
609ca066 PZ |
2683 | /* |
2684 | * The previous task will have left us with a preempt_count of 2 | |
2685 | * because it left us after: | |
2686 | * | |
2687 | * schedule() | |
2688 | * preempt_disable(); // 1 | |
2689 | * __schedule() | |
2690 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2691 | * | |
2692 | * Also, see FORK_PREEMPT_COUNT. | |
2693 | */ | |
e2bf1c4b PZ |
2694 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2695 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2696 | current->comm, current->pid, preempt_count())) | |
2697 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2698 | |
1da177e4 LT |
2699 | rq->prev_mm = NULL; |
2700 | ||
2701 | /* | |
2702 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2703 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2704 | * schedule one last time. The schedule call will never return, and |
2705 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2706 | * |
2707 | * We must observe prev->state before clearing prev->on_cpu (in | |
2708 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2709 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2710 | * transition, resulting in a double drop. | |
1da177e4 | 2711 | */ |
55a101f8 | 2712 | prev_state = prev->state; |
bf9fae9f | 2713 | vtime_task_switch(prev); |
a8d757ef | 2714 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2715 | finish_lock_switch(rq, prev); |
01f23e16 | 2716 | finish_arch_post_lock_switch(); |
e8fa1362 | 2717 | |
e107be36 | 2718 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2719 | if (mm) |
2720 | mmdrop(mm); | |
c394cc9f | 2721 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2722 | if (prev->sched_class->task_dead) |
2723 | prev->sched_class->task_dead(prev); | |
2724 | ||
c6fd91f0 | 2725 | /* |
2726 | * Remove function-return probe instances associated with this | |
2727 | * task and put them back on the free list. | |
9761eea8 | 2728 | */ |
c6fd91f0 | 2729 | kprobe_flush_task(prev); |
1da177e4 | 2730 | put_task_struct(prev); |
c6fd91f0 | 2731 | } |
99e5ada9 | 2732 | |
de734f89 | 2733 | tick_nohz_task_switch(); |
dfa50b60 | 2734 | return rq; |
1da177e4 LT |
2735 | } |
2736 | ||
3f029d3c GH |
2737 | #ifdef CONFIG_SMP |
2738 | ||
3f029d3c | 2739 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2740 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2741 | { |
e3fca9e7 PZ |
2742 | struct callback_head *head, *next; |
2743 | void (*func)(struct rq *rq); | |
2744 | unsigned long flags; | |
3f029d3c | 2745 | |
e3fca9e7 PZ |
2746 | raw_spin_lock_irqsave(&rq->lock, flags); |
2747 | head = rq->balance_callback; | |
2748 | rq->balance_callback = NULL; | |
2749 | while (head) { | |
2750 | func = (void (*)(struct rq *))head->func; | |
2751 | next = head->next; | |
2752 | head->next = NULL; | |
2753 | head = next; | |
3f029d3c | 2754 | |
e3fca9e7 | 2755 | func(rq); |
3f029d3c | 2756 | } |
e3fca9e7 PZ |
2757 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2758 | } | |
2759 | ||
2760 | static inline void balance_callback(struct rq *rq) | |
2761 | { | |
2762 | if (unlikely(rq->balance_callback)) | |
2763 | __balance_callback(rq); | |
3f029d3c GH |
2764 | } |
2765 | ||
2766 | #else | |
da19ab51 | 2767 | |
e3fca9e7 | 2768 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2769 | { |
1da177e4 LT |
2770 | } |
2771 | ||
3f029d3c GH |
2772 | #endif |
2773 | ||
1da177e4 LT |
2774 | /** |
2775 | * schedule_tail - first thing a freshly forked thread must call. | |
2776 | * @prev: the thread we just switched away from. | |
2777 | */ | |
722a9f92 | 2778 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2779 | __releases(rq->lock) |
2780 | { | |
1a43a14a | 2781 | struct rq *rq; |
da19ab51 | 2782 | |
609ca066 PZ |
2783 | /* |
2784 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2785 | * finish_task_switch() for details. | |
2786 | * | |
2787 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2788 | * and the preempt_enable() will end up enabling preemption (on | |
2789 | * PREEMPT_COUNT kernels). | |
2790 | */ | |
2791 | ||
dfa50b60 | 2792 | rq = finish_task_switch(prev); |
e3fca9e7 | 2793 | balance_callback(rq); |
1a43a14a | 2794 | preempt_enable(); |
70b97a7f | 2795 | |
1da177e4 | 2796 | if (current->set_child_tid) |
b488893a | 2797 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2798 | } |
2799 | ||
2800 | /* | |
dfa50b60 | 2801 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2802 | */ |
04936948 | 2803 | static __always_inline struct rq * |
70b97a7f | 2804 | context_switch(struct rq *rq, struct task_struct *prev, |
e7904a28 | 2805 | struct task_struct *next, struct pin_cookie cookie) |
1da177e4 | 2806 | { |
dd41f596 | 2807 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2808 | |
e107be36 | 2809 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2810 | |
dd41f596 IM |
2811 | mm = next->mm; |
2812 | oldmm = prev->active_mm; | |
9226d125 ZA |
2813 | /* |
2814 | * For paravirt, this is coupled with an exit in switch_to to | |
2815 | * combine the page table reload and the switch backend into | |
2816 | * one hypercall. | |
2817 | */ | |
224101ed | 2818 | arch_start_context_switch(prev); |
9226d125 | 2819 | |
31915ab4 | 2820 | if (!mm) { |
1da177e4 LT |
2821 | next->active_mm = oldmm; |
2822 | atomic_inc(&oldmm->mm_count); | |
2823 | enter_lazy_tlb(oldmm, next); | |
2824 | } else | |
f98db601 | 2825 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2826 | |
31915ab4 | 2827 | if (!prev->mm) { |
1da177e4 | 2828 | prev->active_mm = NULL; |
1da177e4 LT |
2829 | rq->prev_mm = oldmm; |
2830 | } | |
3a5f5e48 IM |
2831 | /* |
2832 | * Since the runqueue lock will be released by the next | |
2833 | * task (which is an invalid locking op but in the case | |
2834 | * of the scheduler it's an obvious special-case), so we | |
2835 | * do an early lockdep release here: | |
2836 | */ | |
e7904a28 | 2837 | lockdep_unpin_lock(&rq->lock, cookie); |
8a25d5de | 2838 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2839 | |
2840 | /* Here we just switch the register state and the stack. */ | |
2841 | switch_to(prev, next, prev); | |
dd41f596 | 2842 | barrier(); |
dfa50b60 ON |
2843 | |
2844 | return finish_task_switch(prev); | |
1da177e4 LT |
2845 | } |
2846 | ||
2847 | /* | |
1c3e8264 | 2848 | * nr_running and nr_context_switches: |
1da177e4 LT |
2849 | * |
2850 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2851 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2852 | */ |
2853 | unsigned long nr_running(void) | |
2854 | { | |
2855 | unsigned long i, sum = 0; | |
2856 | ||
2857 | for_each_online_cpu(i) | |
2858 | sum += cpu_rq(i)->nr_running; | |
2859 | ||
2860 | return sum; | |
f711f609 | 2861 | } |
1da177e4 | 2862 | |
2ee507c4 TC |
2863 | /* |
2864 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2865 | * |
2866 | * Caution: this function does not check that the caller has disabled | |
2867 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2868 | * race. The caller is responsible to use it correctly, for example: | |
2869 | * | |
2870 | * - from a non-preemptable section (of course) | |
2871 | * | |
2872 | * - from a thread that is bound to a single CPU | |
2873 | * | |
2874 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2875 | */ |
2876 | bool single_task_running(void) | |
2877 | { | |
00cc1633 | 2878 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2879 | } |
2880 | EXPORT_SYMBOL(single_task_running); | |
2881 | ||
1da177e4 | 2882 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2883 | { |
cc94abfc SR |
2884 | int i; |
2885 | unsigned long long sum = 0; | |
46cb4b7c | 2886 | |
0a945022 | 2887 | for_each_possible_cpu(i) |
1da177e4 | 2888 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2889 | |
1da177e4 LT |
2890 | return sum; |
2891 | } | |
483b4ee6 | 2892 | |
1da177e4 LT |
2893 | unsigned long nr_iowait(void) |
2894 | { | |
2895 | unsigned long i, sum = 0; | |
483b4ee6 | 2896 | |
0a945022 | 2897 | for_each_possible_cpu(i) |
1da177e4 | 2898 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2899 | |
1da177e4 LT |
2900 | return sum; |
2901 | } | |
483b4ee6 | 2902 | |
8c215bd3 | 2903 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2904 | { |
8c215bd3 | 2905 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2906 | return atomic_read(&this->nr_iowait); |
2907 | } | |
46cb4b7c | 2908 | |
372ba8cb MG |
2909 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2910 | { | |
3289bdb4 PZ |
2911 | struct rq *rq = this_rq(); |
2912 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2913 | *load = rq->load.weight; | |
372ba8cb MG |
2914 | } |
2915 | ||
dd41f596 | 2916 | #ifdef CONFIG_SMP |
8a0be9ef | 2917 | |
46cb4b7c | 2918 | /* |
38022906 PZ |
2919 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2920 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2921 | */ |
38022906 | 2922 | void sched_exec(void) |
46cb4b7c | 2923 | { |
38022906 | 2924 | struct task_struct *p = current; |
1da177e4 | 2925 | unsigned long flags; |
0017d735 | 2926 | int dest_cpu; |
46cb4b7c | 2927 | |
8f42ced9 | 2928 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2929 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2930 | if (dest_cpu == smp_processor_id()) |
2931 | goto unlock; | |
38022906 | 2932 | |
8f42ced9 | 2933 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2934 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2935 | |
8f42ced9 PZ |
2936 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2937 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2938 | return; |
2939 | } | |
0017d735 | 2940 | unlock: |
8f42ced9 | 2941 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2942 | } |
dd41f596 | 2943 | |
1da177e4 LT |
2944 | #endif |
2945 | ||
1da177e4 | 2946 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2947 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2948 | |
2949 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2950 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2951 | |
c5f8d995 HS |
2952 | /* |
2953 | * Return accounted runtime for the task. | |
2954 | * In case the task is currently running, return the runtime plus current's | |
2955 | * pending runtime that have not been accounted yet. | |
2956 | */ | |
2957 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2958 | { | |
eb580751 | 2959 | struct rq_flags rf; |
c5f8d995 | 2960 | struct rq *rq; |
6e998916 | 2961 | u64 ns; |
c5f8d995 | 2962 | |
911b2898 PZ |
2963 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2964 | /* | |
2965 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2966 | * So we have a optimization chance when the task's delta_exec is 0. | |
2967 | * Reading ->on_cpu is racy, but this is ok. | |
2968 | * | |
2969 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2970 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2971 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2972 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2973 | * been accounted, so we're correct here as well. | |
911b2898 | 2974 | */ |
da0c1e65 | 2975 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2976 | return p->se.sum_exec_runtime; |
2977 | #endif | |
2978 | ||
eb580751 | 2979 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
2980 | /* |
2981 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2982 | * project cycles that may never be accounted to this | |
2983 | * thread, breaking clock_gettime(). | |
2984 | */ | |
2985 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2986 | update_rq_clock(rq); | |
2987 | p->sched_class->update_curr(rq); | |
2988 | } | |
2989 | ns = p->se.sum_exec_runtime; | |
eb580751 | 2990 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
2991 | |
2992 | return ns; | |
2993 | } | |
48f24c4d | 2994 | |
7835b98b CL |
2995 | /* |
2996 | * This function gets called by the timer code, with HZ frequency. | |
2997 | * We call it with interrupts disabled. | |
7835b98b CL |
2998 | */ |
2999 | void scheduler_tick(void) | |
3000 | { | |
7835b98b CL |
3001 | int cpu = smp_processor_id(); |
3002 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3003 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3004 | |
3005 | sched_clock_tick(); | |
dd41f596 | 3006 | |
05fa785c | 3007 | raw_spin_lock(&rq->lock); |
3e51f33f | 3008 | update_rq_clock(rq); |
fa85ae24 | 3009 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3010 | cpu_load_update_active(rq); |
3289bdb4 | 3011 | calc_global_load_tick(rq); |
05fa785c | 3012 | raw_spin_unlock(&rq->lock); |
7835b98b | 3013 | |
e9d2b064 | 3014 | perf_event_task_tick(); |
e220d2dc | 3015 | |
e418e1c2 | 3016 | #ifdef CONFIG_SMP |
6eb57e0d | 3017 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3018 | trigger_load_balance(rq); |
e418e1c2 | 3019 | #endif |
265f22a9 | 3020 | rq_last_tick_reset(rq); |
1da177e4 LT |
3021 | } |
3022 | ||
265f22a9 FW |
3023 | #ifdef CONFIG_NO_HZ_FULL |
3024 | /** | |
3025 | * scheduler_tick_max_deferment | |
3026 | * | |
3027 | * Keep at least one tick per second when a single | |
3028 | * active task is running because the scheduler doesn't | |
3029 | * yet completely support full dynticks environment. | |
3030 | * | |
3031 | * This makes sure that uptime, CFS vruntime, load | |
3032 | * balancing, etc... continue to move forward, even | |
3033 | * with a very low granularity. | |
e69f6186 YB |
3034 | * |
3035 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3036 | */ |
3037 | u64 scheduler_tick_max_deferment(void) | |
3038 | { | |
3039 | struct rq *rq = this_rq(); | |
316c1608 | 3040 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3041 | |
3042 | next = rq->last_sched_tick + HZ; | |
3043 | ||
3044 | if (time_before_eq(next, now)) | |
3045 | return 0; | |
3046 | ||
8fe8ff09 | 3047 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3048 | } |
265f22a9 | 3049 | #endif |
1da177e4 | 3050 | |
7e49fcce SR |
3051 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3052 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3053 | /* |
3054 | * If the value passed in is equal to the current preempt count | |
3055 | * then we just disabled preemption. Start timing the latency. | |
3056 | */ | |
3057 | static inline void preempt_latency_start(int val) | |
3058 | { | |
3059 | if (preempt_count() == val) { | |
3060 | unsigned long ip = get_lock_parent_ip(); | |
3061 | #ifdef CONFIG_DEBUG_PREEMPT | |
3062 | current->preempt_disable_ip = ip; | |
3063 | #endif | |
3064 | trace_preempt_off(CALLER_ADDR0, ip); | |
3065 | } | |
3066 | } | |
7e49fcce | 3067 | |
edafe3a5 | 3068 | void preempt_count_add(int val) |
1da177e4 | 3069 | { |
6cd8a4bb | 3070 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3071 | /* |
3072 | * Underflow? | |
3073 | */ | |
9a11b49a IM |
3074 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3075 | return; | |
6cd8a4bb | 3076 | #endif |
bdb43806 | 3077 | __preempt_count_add(val); |
6cd8a4bb | 3078 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3079 | /* |
3080 | * Spinlock count overflowing soon? | |
3081 | */ | |
33859f7f MOS |
3082 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3083 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3084 | #endif |
47252cfb | 3085 | preempt_latency_start(val); |
1da177e4 | 3086 | } |
bdb43806 | 3087 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3088 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3089 | |
47252cfb SR |
3090 | /* |
3091 | * If the value passed in equals to the current preempt count | |
3092 | * then we just enabled preemption. Stop timing the latency. | |
3093 | */ | |
3094 | static inline void preempt_latency_stop(int val) | |
3095 | { | |
3096 | if (preempt_count() == val) | |
3097 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3098 | } | |
3099 | ||
edafe3a5 | 3100 | void preempt_count_sub(int val) |
1da177e4 | 3101 | { |
6cd8a4bb | 3102 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3103 | /* |
3104 | * Underflow? | |
3105 | */ | |
01e3eb82 | 3106 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3107 | return; |
1da177e4 LT |
3108 | /* |
3109 | * Is the spinlock portion underflowing? | |
3110 | */ | |
9a11b49a IM |
3111 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3112 | !(preempt_count() & PREEMPT_MASK))) | |
3113 | return; | |
6cd8a4bb | 3114 | #endif |
9a11b49a | 3115 | |
47252cfb | 3116 | preempt_latency_stop(val); |
bdb43806 | 3117 | __preempt_count_sub(val); |
1da177e4 | 3118 | } |
bdb43806 | 3119 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3120 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3121 | |
47252cfb SR |
3122 | #else |
3123 | static inline void preempt_latency_start(int val) { } | |
3124 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3125 | #endif |
3126 | ||
3127 | /* | |
dd41f596 | 3128 | * Print scheduling while atomic bug: |
1da177e4 | 3129 | */ |
dd41f596 | 3130 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3131 | { |
664dfa65 DJ |
3132 | if (oops_in_progress) |
3133 | return; | |
3134 | ||
3df0fc5b PZ |
3135 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3136 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3137 | |
dd41f596 | 3138 | debug_show_held_locks(prev); |
e21f5b15 | 3139 | print_modules(); |
dd41f596 IM |
3140 | if (irqs_disabled()) |
3141 | print_irqtrace_events(prev); | |
8f47b187 TG |
3142 | #ifdef CONFIG_DEBUG_PREEMPT |
3143 | if (in_atomic_preempt_off()) { | |
3144 | pr_err("Preemption disabled at:"); | |
3145 | print_ip_sym(current->preempt_disable_ip); | |
3146 | pr_cont("\n"); | |
3147 | } | |
3148 | #endif | |
6135fc1e | 3149 | dump_stack(); |
373d4d09 | 3150 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3151 | } |
1da177e4 | 3152 | |
dd41f596 IM |
3153 | /* |
3154 | * Various schedule()-time debugging checks and statistics: | |
3155 | */ | |
3156 | static inline void schedule_debug(struct task_struct *prev) | |
3157 | { | |
0d9e2632 | 3158 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
ce03e413 | 3159 | BUG_ON(task_stack_end_corrupted(prev)); |
0d9e2632 | 3160 | #endif |
b99def8b | 3161 | |
1dc0fffc | 3162 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3163 | __schedule_bug(prev); |
1dc0fffc PZ |
3164 | preempt_count_set(PREEMPT_DISABLED); |
3165 | } | |
b3fbab05 | 3166 | rcu_sleep_check(); |
dd41f596 | 3167 | |
1da177e4 LT |
3168 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3169 | ||
2d72376b | 3170 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3171 | } |
3172 | ||
3173 | /* | |
3174 | * Pick up the highest-prio task: | |
3175 | */ | |
3176 | static inline struct task_struct * | |
e7904a28 | 3177 | pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie) |
dd41f596 | 3178 | { |
37e117c0 | 3179 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 3180 | struct task_struct *p; |
1da177e4 LT |
3181 | |
3182 | /* | |
dd41f596 IM |
3183 | * Optimization: we know that if all tasks are in |
3184 | * the fair class we can call that function directly: | |
1da177e4 | 3185 | */ |
37e117c0 | 3186 | if (likely(prev->sched_class == class && |
38033c37 | 3187 | rq->nr_running == rq->cfs.h_nr_running)) { |
e7904a28 | 3188 | p = fair_sched_class.pick_next_task(rq, prev, cookie); |
6ccdc84b PZ |
3189 | if (unlikely(p == RETRY_TASK)) |
3190 | goto again; | |
3191 | ||
3192 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3193 | if (unlikely(!p)) | |
e7904a28 | 3194 | p = idle_sched_class.pick_next_task(rq, prev, cookie); |
6ccdc84b PZ |
3195 | |
3196 | return p; | |
1da177e4 LT |
3197 | } |
3198 | ||
37e117c0 | 3199 | again: |
34f971f6 | 3200 | for_each_class(class) { |
e7904a28 | 3201 | p = class->pick_next_task(rq, prev, cookie); |
37e117c0 PZ |
3202 | if (p) { |
3203 | if (unlikely(p == RETRY_TASK)) | |
3204 | goto again; | |
dd41f596 | 3205 | return p; |
37e117c0 | 3206 | } |
dd41f596 | 3207 | } |
34f971f6 PZ |
3208 | |
3209 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3210 | } |
1da177e4 | 3211 | |
dd41f596 | 3212 | /* |
c259e01a | 3213 | * __schedule() is the main scheduler function. |
edde96ea PE |
3214 | * |
3215 | * The main means of driving the scheduler and thus entering this function are: | |
3216 | * | |
3217 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3218 | * | |
3219 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3220 | * paths. For example, see arch/x86/entry_64.S. | |
3221 | * | |
3222 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3223 | * interrupt handler scheduler_tick(). | |
3224 | * | |
3225 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3226 | * task to the run-queue and that's it. | |
3227 | * | |
3228 | * Now, if the new task added to the run-queue preempts the current | |
3229 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3230 | * called on the nearest possible occasion: | |
3231 | * | |
3232 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3233 | * | |
3234 | * - in syscall or exception context, at the next outmost | |
3235 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3236 | * spin_unlock()!) | |
3237 | * | |
3238 | * - in IRQ context, return from interrupt-handler to | |
3239 | * preemptible context | |
3240 | * | |
3241 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3242 | * then at the next: | |
3243 | * | |
3244 | * - cond_resched() call | |
3245 | * - explicit schedule() call | |
3246 | * - return from syscall or exception to user-space | |
3247 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3248 | * |
b30f0e3f | 3249 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3250 | */ |
499d7955 | 3251 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3252 | { |
3253 | struct task_struct *prev, *next; | |
67ca7bde | 3254 | unsigned long *switch_count; |
e7904a28 | 3255 | struct pin_cookie cookie; |
dd41f596 | 3256 | struct rq *rq; |
31656519 | 3257 | int cpu; |
dd41f596 | 3258 | |
dd41f596 IM |
3259 | cpu = smp_processor_id(); |
3260 | rq = cpu_rq(cpu); | |
dd41f596 | 3261 | prev = rq->curr; |
dd41f596 | 3262 | |
b99def8b PZ |
3263 | /* |
3264 | * do_exit() calls schedule() with preemption disabled as an exception; | |
3265 | * however we must fix that up, otherwise the next task will see an | |
3266 | * inconsistent (higher) preempt count. | |
3267 | * | |
3268 | * It also avoids the below schedule_debug() test from complaining | |
3269 | * about this. | |
3270 | */ | |
3271 | if (unlikely(prev->state == TASK_DEAD)) | |
3272 | preempt_enable_no_resched_notrace(); | |
3273 | ||
dd41f596 | 3274 | schedule_debug(prev); |
1da177e4 | 3275 | |
31656519 | 3276 | if (sched_feat(HRTICK)) |
f333fdc9 | 3277 | hrtick_clear(rq); |
8f4d37ec | 3278 | |
46a5d164 PM |
3279 | local_irq_disable(); |
3280 | rcu_note_context_switch(); | |
3281 | ||
e0acd0a6 ON |
3282 | /* |
3283 | * Make sure that signal_pending_state()->signal_pending() below | |
3284 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3285 | * done by the caller to avoid the race with signal_wake_up(). | |
3286 | */ | |
3287 | smp_mb__before_spinlock(); | |
46a5d164 | 3288 | raw_spin_lock(&rq->lock); |
e7904a28 | 3289 | cookie = lockdep_pin_lock(&rq->lock); |
1da177e4 | 3290 | |
9edfbfed PZ |
3291 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3292 | ||
246d86b5 | 3293 | switch_count = &prev->nivcsw; |
fc13aeba | 3294 | if (!preempt && prev->state) { |
21aa9af0 | 3295 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3296 | prev->state = TASK_RUNNING; |
21aa9af0 | 3297 | } else { |
2acca55e PZ |
3298 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3299 | prev->on_rq = 0; | |
3300 | ||
21aa9af0 | 3301 | /* |
2acca55e PZ |
3302 | * If a worker went to sleep, notify and ask workqueue |
3303 | * whether it wants to wake up a task to maintain | |
3304 | * concurrency. | |
21aa9af0 TH |
3305 | */ |
3306 | if (prev->flags & PF_WQ_WORKER) { | |
3307 | struct task_struct *to_wakeup; | |
3308 | ||
9b7f6597 | 3309 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3310 | if (to_wakeup) |
e7904a28 | 3311 | try_to_wake_up_local(to_wakeup, cookie); |
21aa9af0 | 3312 | } |
21aa9af0 | 3313 | } |
dd41f596 | 3314 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3315 | } |
3316 | ||
9edfbfed | 3317 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3318 | update_rq_clock(rq); |
3319 | ||
e7904a28 | 3320 | next = pick_next_task(rq, prev, cookie); |
f26f9aff | 3321 | clear_tsk_need_resched(prev); |
f27dde8d | 3322 | clear_preempt_need_resched(); |
9edfbfed | 3323 | rq->clock_skip_update = 0; |
1da177e4 | 3324 | |
1da177e4 | 3325 | if (likely(prev != next)) { |
1da177e4 LT |
3326 | rq->nr_switches++; |
3327 | rq->curr = next; | |
3328 | ++*switch_count; | |
3329 | ||
c73464b1 | 3330 | trace_sched_switch(preempt, prev, next); |
e7904a28 | 3331 | rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */ |
cbce1a68 | 3332 | } else { |
e7904a28 | 3333 | lockdep_unpin_lock(&rq->lock, cookie); |
05fa785c | 3334 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3335 | } |
1da177e4 | 3336 | |
e3fca9e7 | 3337 | balance_callback(rq); |
1da177e4 | 3338 | } |
8e05e96a | 3339 | STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */ |
c259e01a | 3340 | |
9c40cef2 TG |
3341 | static inline void sched_submit_work(struct task_struct *tsk) |
3342 | { | |
3c7d5184 | 3343 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3344 | return; |
3345 | /* | |
3346 | * If we are going to sleep and we have plugged IO queued, | |
3347 | * make sure to submit it to avoid deadlocks. | |
3348 | */ | |
3349 | if (blk_needs_flush_plug(tsk)) | |
3350 | blk_schedule_flush_plug(tsk); | |
3351 | } | |
3352 | ||
722a9f92 | 3353 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3354 | { |
9c40cef2 TG |
3355 | struct task_struct *tsk = current; |
3356 | ||
3357 | sched_submit_work(tsk); | |
bfd9b2b5 | 3358 | do { |
b30f0e3f | 3359 | preempt_disable(); |
fc13aeba | 3360 | __schedule(false); |
b30f0e3f | 3361 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3362 | } while (need_resched()); |
c259e01a | 3363 | } |
1da177e4 LT |
3364 | EXPORT_SYMBOL(schedule); |
3365 | ||
91d1aa43 | 3366 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3367 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3368 | { |
3369 | /* | |
3370 | * If we come here after a random call to set_need_resched(), | |
3371 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3372 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3373 | * we find a better solution. | |
7cc78f8f AL |
3374 | * |
3375 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3376 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3377 | * too frequently to make sense yet. |
20ab65e3 | 3378 | */ |
7cc78f8f | 3379 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3380 | schedule(); |
7cc78f8f | 3381 | exception_exit(prev_state); |
20ab65e3 FW |
3382 | } |
3383 | #endif | |
3384 | ||
c5491ea7 TG |
3385 | /** |
3386 | * schedule_preempt_disabled - called with preemption disabled | |
3387 | * | |
3388 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3389 | */ | |
3390 | void __sched schedule_preempt_disabled(void) | |
3391 | { | |
ba74c144 | 3392 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3393 | schedule(); |
3394 | preempt_disable(); | |
3395 | } | |
3396 | ||
06b1f808 | 3397 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3398 | { |
3399 | do { | |
47252cfb SR |
3400 | /* |
3401 | * Because the function tracer can trace preempt_count_sub() | |
3402 | * and it also uses preempt_enable/disable_notrace(), if | |
3403 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3404 | * by the function tracer will call this function again and | |
3405 | * cause infinite recursion. | |
3406 | * | |
3407 | * Preemption must be disabled here before the function | |
3408 | * tracer can trace. Break up preempt_disable() into two | |
3409 | * calls. One to disable preemption without fear of being | |
3410 | * traced. The other to still record the preemption latency, | |
3411 | * which can also be traced by the function tracer. | |
3412 | */ | |
499d7955 | 3413 | preempt_disable_notrace(); |
47252cfb | 3414 | preempt_latency_start(1); |
fc13aeba | 3415 | __schedule(true); |
47252cfb | 3416 | preempt_latency_stop(1); |
499d7955 | 3417 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3418 | |
3419 | /* | |
3420 | * Check again in case we missed a preemption opportunity | |
3421 | * between schedule and now. | |
3422 | */ | |
a18b5d01 FW |
3423 | } while (need_resched()); |
3424 | } | |
3425 | ||
1da177e4 LT |
3426 | #ifdef CONFIG_PREEMPT |
3427 | /* | |
2ed6e34f | 3428 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3429 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3430 | * occur there and call schedule directly. |
3431 | */ | |
722a9f92 | 3432 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3433 | { |
1da177e4 LT |
3434 | /* |
3435 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3436 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3437 | */ |
fbb00b56 | 3438 | if (likely(!preemptible())) |
1da177e4 LT |
3439 | return; |
3440 | ||
a18b5d01 | 3441 | preempt_schedule_common(); |
1da177e4 | 3442 | } |
376e2424 | 3443 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3444 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3445 | |
009f60e2 | 3446 | /** |
4eaca0a8 | 3447 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3448 | * |
3449 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3450 | * recursion and tracing preempt enabling caused by the tracing | |
3451 | * infrastructure itself. But as tracing can happen in areas coming | |
3452 | * from userspace or just about to enter userspace, a preempt enable | |
3453 | * can occur before user_exit() is called. This will cause the scheduler | |
3454 | * to be called when the system is still in usermode. | |
3455 | * | |
3456 | * To prevent this, the preempt_enable_notrace will use this function | |
3457 | * instead of preempt_schedule() to exit user context if needed before | |
3458 | * calling the scheduler. | |
3459 | */ | |
4eaca0a8 | 3460 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3461 | { |
3462 | enum ctx_state prev_ctx; | |
3463 | ||
3464 | if (likely(!preemptible())) | |
3465 | return; | |
3466 | ||
3467 | do { | |
47252cfb SR |
3468 | /* |
3469 | * Because the function tracer can trace preempt_count_sub() | |
3470 | * and it also uses preempt_enable/disable_notrace(), if | |
3471 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3472 | * by the function tracer will call this function again and | |
3473 | * cause infinite recursion. | |
3474 | * | |
3475 | * Preemption must be disabled here before the function | |
3476 | * tracer can trace. Break up preempt_disable() into two | |
3477 | * calls. One to disable preemption without fear of being | |
3478 | * traced. The other to still record the preemption latency, | |
3479 | * which can also be traced by the function tracer. | |
3480 | */ | |
3d8f74dd | 3481 | preempt_disable_notrace(); |
47252cfb | 3482 | preempt_latency_start(1); |
009f60e2 ON |
3483 | /* |
3484 | * Needs preempt disabled in case user_exit() is traced | |
3485 | * and the tracer calls preempt_enable_notrace() causing | |
3486 | * an infinite recursion. | |
3487 | */ | |
3488 | prev_ctx = exception_enter(); | |
fc13aeba | 3489 | __schedule(true); |
009f60e2 ON |
3490 | exception_exit(prev_ctx); |
3491 | ||
47252cfb | 3492 | preempt_latency_stop(1); |
3d8f74dd | 3493 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3494 | } while (need_resched()); |
3495 | } | |
4eaca0a8 | 3496 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3497 | |
32e475d7 | 3498 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3499 | |
3500 | /* | |
2ed6e34f | 3501 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3502 | * off of irq context. |
3503 | * Note, that this is called and return with irqs disabled. This will | |
3504 | * protect us against recursive calling from irq. | |
3505 | */ | |
722a9f92 | 3506 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3507 | { |
b22366cd | 3508 | enum ctx_state prev_state; |
6478d880 | 3509 | |
2ed6e34f | 3510 | /* Catch callers which need to be fixed */ |
f27dde8d | 3511 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3512 | |
b22366cd FW |
3513 | prev_state = exception_enter(); |
3514 | ||
3a5c359a | 3515 | do { |
3d8f74dd | 3516 | preempt_disable(); |
3a5c359a | 3517 | local_irq_enable(); |
fc13aeba | 3518 | __schedule(true); |
3a5c359a | 3519 | local_irq_disable(); |
3d8f74dd | 3520 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3521 | } while (need_resched()); |
b22366cd FW |
3522 | |
3523 | exception_exit(prev_state); | |
1da177e4 LT |
3524 | } |
3525 | ||
63859d4f | 3526 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3527 | void *key) |
1da177e4 | 3528 | { |
63859d4f | 3529 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3530 | } |
1da177e4 LT |
3531 | EXPORT_SYMBOL(default_wake_function); |
3532 | ||
b29739f9 IM |
3533 | #ifdef CONFIG_RT_MUTEXES |
3534 | ||
3535 | /* | |
3536 | * rt_mutex_setprio - set the current priority of a task | |
3537 | * @p: task | |
3538 | * @prio: prio value (kernel-internal form) | |
3539 | * | |
3540 | * This function changes the 'effective' priority of a task. It does | |
3541 | * not touch ->normal_prio like __setscheduler(). | |
3542 | * | |
c365c292 TG |
3543 | * Used by the rt_mutex code to implement priority inheritance |
3544 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3545 | */ |
36c8b586 | 3546 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3547 | { |
ff77e468 | 3548 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3549 | const struct sched_class *prev_class; |
eb580751 PZ |
3550 | struct rq_flags rf; |
3551 | struct rq *rq; | |
b29739f9 | 3552 | |
aab03e05 | 3553 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3554 | |
eb580751 | 3555 | rq = __task_rq_lock(p, &rf); |
b29739f9 | 3556 | |
1c4dd99b TG |
3557 | /* |
3558 | * Idle task boosting is a nono in general. There is one | |
3559 | * exception, when PREEMPT_RT and NOHZ is active: | |
3560 | * | |
3561 | * The idle task calls get_next_timer_interrupt() and holds | |
3562 | * the timer wheel base->lock on the CPU and another CPU wants | |
3563 | * to access the timer (probably to cancel it). We can safely | |
3564 | * ignore the boosting request, as the idle CPU runs this code | |
3565 | * with interrupts disabled and will complete the lock | |
3566 | * protected section without being interrupted. So there is no | |
3567 | * real need to boost. | |
3568 | */ | |
3569 | if (unlikely(p == rq->idle)) { | |
3570 | WARN_ON(p != rq->curr); | |
3571 | WARN_ON(p->pi_blocked_on); | |
3572 | goto out_unlock; | |
3573 | } | |
3574 | ||
a8027073 | 3575 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3576 | oldprio = p->prio; |
ff77e468 PZ |
3577 | |
3578 | if (oldprio == prio) | |
3579 | queue_flag &= ~DEQUEUE_MOVE; | |
3580 | ||
83ab0aa0 | 3581 | prev_class = p->sched_class; |
da0c1e65 | 3582 | queued = task_on_rq_queued(p); |
051a1d1a | 3583 | running = task_current(rq, p); |
da0c1e65 | 3584 | if (queued) |
ff77e468 | 3585 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3586 | if (running) |
f3cd1c4e | 3587 | put_prev_task(rq, p); |
dd41f596 | 3588 | |
2d3d891d DF |
3589 | /* |
3590 | * Boosting condition are: | |
3591 | * 1. -rt task is running and holds mutex A | |
3592 | * --> -dl task blocks on mutex A | |
3593 | * | |
3594 | * 2. -dl task is running and holds mutex A | |
3595 | * --> -dl task blocks on mutex A and could preempt the | |
3596 | * running task | |
3597 | */ | |
3598 | if (dl_prio(prio)) { | |
466af29b ON |
3599 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3600 | if (!dl_prio(p->normal_prio) || | |
3601 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3602 | p->dl.dl_boosted = 1; |
ff77e468 | 3603 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3604 | } else |
3605 | p->dl.dl_boosted = 0; | |
aab03e05 | 3606 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3607 | } else if (rt_prio(prio)) { |
3608 | if (dl_prio(oldprio)) | |
3609 | p->dl.dl_boosted = 0; | |
3610 | if (oldprio < prio) | |
ff77e468 | 3611 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3612 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3613 | } else { |
3614 | if (dl_prio(oldprio)) | |
3615 | p->dl.dl_boosted = 0; | |
746db944 BS |
3616 | if (rt_prio(oldprio)) |
3617 | p->rt.timeout = 0; | |
dd41f596 | 3618 | p->sched_class = &fair_sched_class; |
2d3d891d | 3619 | } |
dd41f596 | 3620 | |
b29739f9 IM |
3621 | p->prio = prio; |
3622 | ||
0e1f3483 HS |
3623 | if (running) |
3624 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3625 | if (queued) |
ff77e468 | 3626 | enqueue_task(rq, p, queue_flag); |
cb469845 | 3627 | |
da7a735e | 3628 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3629 | out_unlock: |
4c9a4bc8 | 3630 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 3631 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3632 | |
3633 | balance_callback(rq); | |
3634 | preempt_enable(); | |
b29739f9 | 3635 | } |
b29739f9 | 3636 | #endif |
d50dde5a | 3637 | |
36c8b586 | 3638 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3639 | { |
da0c1e65 | 3640 | int old_prio, delta, queued; |
eb580751 | 3641 | struct rq_flags rf; |
70b97a7f | 3642 | struct rq *rq; |
1da177e4 | 3643 | |
75e45d51 | 3644 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3645 | return; |
3646 | /* | |
3647 | * We have to be careful, if called from sys_setpriority(), | |
3648 | * the task might be in the middle of scheduling on another CPU. | |
3649 | */ | |
eb580751 | 3650 | rq = task_rq_lock(p, &rf); |
1da177e4 LT |
3651 | /* |
3652 | * The RT priorities are set via sched_setscheduler(), but we still | |
3653 | * allow the 'normal' nice value to be set - but as expected | |
3654 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3655 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3656 | */ |
aab03e05 | 3657 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3658 | p->static_prio = NICE_TO_PRIO(nice); |
3659 | goto out_unlock; | |
3660 | } | |
da0c1e65 KT |
3661 | queued = task_on_rq_queued(p); |
3662 | if (queued) | |
1de64443 | 3663 | dequeue_task(rq, p, DEQUEUE_SAVE); |
1da177e4 | 3664 | |
1da177e4 | 3665 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3666 | set_load_weight(p); |
b29739f9 IM |
3667 | old_prio = p->prio; |
3668 | p->prio = effective_prio(p); | |
3669 | delta = p->prio - old_prio; | |
1da177e4 | 3670 | |
da0c1e65 | 3671 | if (queued) { |
1de64443 | 3672 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3673 | /* |
d5f9f942 AM |
3674 | * If the task increased its priority or is running and |
3675 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3676 | */ |
d5f9f942 | 3677 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3678 | resched_curr(rq); |
1da177e4 LT |
3679 | } |
3680 | out_unlock: | |
eb580751 | 3681 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3682 | } |
1da177e4 LT |
3683 | EXPORT_SYMBOL(set_user_nice); |
3684 | ||
e43379f1 MM |
3685 | /* |
3686 | * can_nice - check if a task can reduce its nice value | |
3687 | * @p: task | |
3688 | * @nice: nice value | |
3689 | */ | |
36c8b586 | 3690 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3691 | { |
024f4747 | 3692 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3693 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3694 | |
78d7d407 | 3695 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3696 | capable(CAP_SYS_NICE)); |
3697 | } | |
3698 | ||
1da177e4 LT |
3699 | #ifdef __ARCH_WANT_SYS_NICE |
3700 | ||
3701 | /* | |
3702 | * sys_nice - change the priority of the current process. | |
3703 | * @increment: priority increment | |
3704 | * | |
3705 | * sys_setpriority is a more generic, but much slower function that | |
3706 | * does similar things. | |
3707 | */ | |
5add95d4 | 3708 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3709 | { |
48f24c4d | 3710 | long nice, retval; |
1da177e4 LT |
3711 | |
3712 | /* | |
3713 | * Setpriority might change our priority at the same moment. | |
3714 | * We don't have to worry. Conceptually one call occurs first | |
3715 | * and we have a single winner. | |
3716 | */ | |
a9467fa3 | 3717 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3718 | nice = task_nice(current) + increment; |
1da177e4 | 3719 | |
a9467fa3 | 3720 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3721 | if (increment < 0 && !can_nice(current, nice)) |
3722 | return -EPERM; | |
3723 | ||
1da177e4 LT |
3724 | retval = security_task_setnice(current, nice); |
3725 | if (retval) | |
3726 | return retval; | |
3727 | ||
3728 | set_user_nice(current, nice); | |
3729 | return 0; | |
3730 | } | |
3731 | ||
3732 | #endif | |
3733 | ||
3734 | /** | |
3735 | * task_prio - return the priority value of a given task. | |
3736 | * @p: the task in question. | |
3737 | * | |
e69f6186 | 3738 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3739 | * RT tasks are offset by -200. Normal tasks are centered |
3740 | * around 0, value goes from -16 to +15. | |
3741 | */ | |
36c8b586 | 3742 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3743 | { |
3744 | return p->prio - MAX_RT_PRIO; | |
3745 | } | |
3746 | ||
1da177e4 LT |
3747 | /** |
3748 | * idle_cpu - is a given cpu idle currently? | |
3749 | * @cpu: the processor in question. | |
e69f6186 YB |
3750 | * |
3751 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3752 | */ |
3753 | int idle_cpu(int cpu) | |
3754 | { | |
908a3283 TG |
3755 | struct rq *rq = cpu_rq(cpu); |
3756 | ||
3757 | if (rq->curr != rq->idle) | |
3758 | return 0; | |
3759 | ||
3760 | if (rq->nr_running) | |
3761 | return 0; | |
3762 | ||
3763 | #ifdef CONFIG_SMP | |
3764 | if (!llist_empty(&rq->wake_list)) | |
3765 | return 0; | |
3766 | #endif | |
3767 | ||
3768 | return 1; | |
1da177e4 LT |
3769 | } |
3770 | ||
1da177e4 LT |
3771 | /** |
3772 | * idle_task - return the idle task for a given cpu. | |
3773 | * @cpu: the processor in question. | |
e69f6186 YB |
3774 | * |
3775 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3776 | */ |
36c8b586 | 3777 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3778 | { |
3779 | return cpu_rq(cpu)->idle; | |
3780 | } | |
3781 | ||
3782 | /** | |
3783 | * find_process_by_pid - find a process with a matching PID value. | |
3784 | * @pid: the pid in question. | |
e69f6186 YB |
3785 | * |
3786 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3787 | */ |
a9957449 | 3788 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3789 | { |
228ebcbe | 3790 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3791 | } |
3792 | ||
aab03e05 DF |
3793 | /* |
3794 | * This function initializes the sched_dl_entity of a newly becoming | |
3795 | * SCHED_DEADLINE task. | |
3796 | * | |
3797 | * Only the static values are considered here, the actual runtime and the | |
3798 | * absolute deadline will be properly calculated when the task is enqueued | |
3799 | * for the first time with its new policy. | |
3800 | */ | |
3801 | static void | |
3802 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3803 | { | |
3804 | struct sched_dl_entity *dl_se = &p->dl; | |
3805 | ||
aab03e05 DF |
3806 | dl_se->dl_runtime = attr->sched_runtime; |
3807 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3808 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3809 | dl_se->flags = attr->sched_flags; |
332ac17e | 3810 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3811 | |
3812 | /* | |
3813 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3814 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3815 | * | |
3816 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3817 | * point. This would include retaining the task_struct until that time | |
3818 | * and change dl_overflow() to not immediately decrement the current | |
3819 | * amount. | |
3820 | * | |
3821 | * Instead we retain the current runtime/deadline and let the new | |
3822 | * parameters take effect after the current reservation period lapses. | |
3823 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3824 | * before the current scheduling deadline. | |
3825 | * | |
3826 | * We can still have temporary overloads because we do not delay the | |
3827 | * change in bandwidth until that time; so admission control is | |
3828 | * not on the safe side. It does however guarantee tasks will never | |
3829 | * consume more than promised. | |
3830 | */ | |
aab03e05 DF |
3831 | } |
3832 | ||
c13db6b1 SR |
3833 | /* |
3834 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3835 | * it calls know not to change it. | |
3836 | */ | |
3837 | #define SETPARAM_POLICY -1 | |
3838 | ||
c365c292 TG |
3839 | static void __setscheduler_params(struct task_struct *p, |
3840 | const struct sched_attr *attr) | |
1da177e4 | 3841 | { |
d50dde5a DF |
3842 | int policy = attr->sched_policy; |
3843 | ||
c13db6b1 | 3844 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3845 | policy = p->policy; |
3846 | ||
1da177e4 | 3847 | p->policy = policy; |
d50dde5a | 3848 | |
aab03e05 DF |
3849 | if (dl_policy(policy)) |
3850 | __setparam_dl(p, attr); | |
39fd8fd2 | 3851 | else if (fair_policy(policy)) |
d50dde5a DF |
3852 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3853 | ||
39fd8fd2 PZ |
3854 | /* |
3855 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3856 | * !rt_policy. Always setting this ensures that things like | |
3857 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3858 | */ | |
3859 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3860 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3861 | set_load_weight(p); |
3862 | } | |
39fd8fd2 | 3863 | |
c365c292 TG |
3864 | /* Actually do priority change: must hold pi & rq lock. */ |
3865 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3866 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3867 | { |
3868 | __setscheduler_params(p, attr); | |
d50dde5a | 3869 | |
383afd09 | 3870 | /* |
0782e63b TG |
3871 | * Keep a potential priority boosting if called from |
3872 | * sched_setscheduler(). | |
383afd09 | 3873 | */ |
0782e63b TG |
3874 | if (keep_boost) |
3875 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3876 | else | |
3877 | p->prio = normal_prio(p); | |
383afd09 | 3878 | |
aab03e05 DF |
3879 | if (dl_prio(p->prio)) |
3880 | p->sched_class = &dl_sched_class; | |
3881 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3882 | p->sched_class = &rt_sched_class; |
3883 | else | |
3884 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3885 | } |
aab03e05 DF |
3886 | |
3887 | static void | |
3888 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3889 | { | |
3890 | struct sched_dl_entity *dl_se = &p->dl; | |
3891 | ||
3892 | attr->sched_priority = p->rt_priority; | |
3893 | attr->sched_runtime = dl_se->dl_runtime; | |
3894 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3895 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3896 | attr->sched_flags = dl_se->flags; |
3897 | } | |
3898 | ||
3899 | /* | |
3900 | * This function validates the new parameters of a -deadline task. | |
3901 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3902 | * than the runtime, as well as the period of being zero or |
332ac17e | 3903 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3904 | * user parameters are above the internal resolution of 1us (we |
3905 | * check sched_runtime only since it is always the smaller one) and | |
3906 | * below 2^63 ns (we have to check both sched_deadline and | |
3907 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3908 | */ |
3909 | static bool | |
3910 | __checkparam_dl(const struct sched_attr *attr) | |
3911 | { | |
b0827819 JL |
3912 | /* deadline != 0 */ |
3913 | if (attr->sched_deadline == 0) | |
3914 | return false; | |
3915 | ||
3916 | /* | |
3917 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3918 | * that big. | |
3919 | */ | |
3920 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3921 | return false; | |
3922 | ||
3923 | /* | |
3924 | * Since we use the MSB for wrap-around and sign issues, make | |
3925 | * sure it's not set (mind that period can be equal to zero). | |
3926 | */ | |
3927 | if (attr->sched_deadline & (1ULL << 63) || | |
3928 | attr->sched_period & (1ULL << 63)) | |
3929 | return false; | |
3930 | ||
3931 | /* runtime <= deadline <= period (if period != 0) */ | |
3932 | if ((attr->sched_period != 0 && | |
3933 | attr->sched_period < attr->sched_deadline) || | |
3934 | attr->sched_deadline < attr->sched_runtime) | |
3935 | return false; | |
3936 | ||
3937 | return true; | |
aab03e05 DF |
3938 | } |
3939 | ||
c69e8d9c DH |
3940 | /* |
3941 | * check the target process has a UID that matches the current process's | |
3942 | */ | |
3943 | static bool check_same_owner(struct task_struct *p) | |
3944 | { | |
3945 | const struct cred *cred = current_cred(), *pcred; | |
3946 | bool match; | |
3947 | ||
3948 | rcu_read_lock(); | |
3949 | pcred = __task_cred(p); | |
9c806aa0 EB |
3950 | match = (uid_eq(cred->euid, pcred->euid) || |
3951 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3952 | rcu_read_unlock(); |
3953 | return match; | |
3954 | } | |
3955 | ||
75381608 WL |
3956 | static bool dl_param_changed(struct task_struct *p, |
3957 | const struct sched_attr *attr) | |
3958 | { | |
3959 | struct sched_dl_entity *dl_se = &p->dl; | |
3960 | ||
3961 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3962 | dl_se->dl_deadline != attr->sched_deadline || | |
3963 | dl_se->dl_period != attr->sched_period || | |
3964 | dl_se->flags != attr->sched_flags) | |
3965 | return true; | |
3966 | ||
3967 | return false; | |
3968 | } | |
3969 | ||
d50dde5a DF |
3970 | static int __sched_setscheduler(struct task_struct *p, |
3971 | const struct sched_attr *attr, | |
dbc7f069 | 3972 | bool user, bool pi) |
1da177e4 | 3973 | { |
383afd09 SR |
3974 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3975 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3976 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3977 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 3978 | const struct sched_class *prev_class; |
eb580751 | 3979 | struct rq_flags rf; |
ca94c442 | 3980 | int reset_on_fork; |
ff77e468 | 3981 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 3982 | struct rq *rq; |
1da177e4 | 3983 | |
66e5393a SR |
3984 | /* may grab non-irq protected spin_locks */ |
3985 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3986 | recheck: |
3987 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3988 | if (policy < 0) { |
3989 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3990 | policy = oldpolicy = p->policy; |
ca94c442 | 3991 | } else { |
7479f3c9 | 3992 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3993 | |
20f9cd2a | 3994 | if (!valid_policy(policy)) |
ca94c442 LP |
3995 | return -EINVAL; |
3996 | } | |
3997 | ||
7479f3c9 PZ |
3998 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3999 | return -EINVAL; | |
4000 | ||
1da177e4 LT |
4001 | /* |
4002 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4003 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4004 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4005 | */ |
0bb040a4 | 4006 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4007 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4008 | return -EINVAL; |
aab03e05 DF |
4009 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4010 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4011 | return -EINVAL; |
4012 | ||
37e4ab3f OC |
4013 | /* |
4014 | * Allow unprivileged RT tasks to decrease priority: | |
4015 | */ | |
961ccddd | 4016 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4017 | if (fair_policy(policy)) { |
d0ea0268 | 4018 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4019 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4020 | return -EPERM; |
4021 | } | |
4022 | ||
e05606d3 | 4023 | if (rt_policy(policy)) { |
a44702e8 ON |
4024 | unsigned long rlim_rtprio = |
4025 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4026 | |
4027 | /* can't set/change the rt policy */ | |
4028 | if (policy != p->policy && !rlim_rtprio) | |
4029 | return -EPERM; | |
4030 | ||
4031 | /* can't increase priority */ | |
d50dde5a DF |
4032 | if (attr->sched_priority > p->rt_priority && |
4033 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4034 | return -EPERM; |
4035 | } | |
c02aa73b | 4036 | |
d44753b8 JL |
4037 | /* |
4038 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4039 | * (safest behavior); in the future we would like to allow | |
4040 | * unprivileged DL tasks to increase their relative deadline | |
4041 | * or reduce their runtime (both ways reducing utilization) | |
4042 | */ | |
4043 | if (dl_policy(policy)) | |
4044 | return -EPERM; | |
4045 | ||
dd41f596 | 4046 | /* |
c02aa73b DH |
4047 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4048 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4049 | */ |
20f9cd2a | 4050 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4051 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4052 | return -EPERM; |
4053 | } | |
5fe1d75f | 4054 | |
37e4ab3f | 4055 | /* can't change other user's priorities */ |
c69e8d9c | 4056 | if (!check_same_owner(p)) |
37e4ab3f | 4057 | return -EPERM; |
ca94c442 LP |
4058 | |
4059 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4060 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4061 | return -EPERM; | |
37e4ab3f | 4062 | } |
1da177e4 | 4063 | |
725aad24 | 4064 | if (user) { |
b0ae1981 | 4065 | retval = security_task_setscheduler(p); |
725aad24 JF |
4066 | if (retval) |
4067 | return retval; | |
4068 | } | |
4069 | ||
b29739f9 IM |
4070 | /* |
4071 | * make sure no PI-waiters arrive (or leave) while we are | |
4072 | * changing the priority of the task: | |
0122ec5b | 4073 | * |
25985edc | 4074 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4075 | * runqueue lock must be held. |
4076 | */ | |
eb580751 | 4077 | rq = task_rq_lock(p, &rf); |
dc61b1d6 | 4078 | |
34f971f6 PZ |
4079 | /* |
4080 | * Changing the policy of the stop threads its a very bad idea | |
4081 | */ | |
4082 | if (p == rq->stop) { | |
eb580751 | 4083 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4084 | return -EINVAL; |
4085 | } | |
4086 | ||
a51e9198 | 4087 | /* |
d6b1e911 TG |
4088 | * If not changing anything there's no need to proceed further, |
4089 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4090 | */ |
d50dde5a | 4091 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4092 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4093 | goto change; |
4094 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4095 | goto change; | |
75381608 | 4096 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4097 | goto change; |
d50dde5a | 4098 | |
d6b1e911 | 4099 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4100 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4101 | return 0; |
4102 | } | |
d50dde5a | 4103 | change: |
a51e9198 | 4104 | |
dc61b1d6 | 4105 | if (user) { |
332ac17e | 4106 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4107 | /* |
4108 | * Do not allow realtime tasks into groups that have no runtime | |
4109 | * assigned. | |
4110 | */ | |
4111 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4112 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4113 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4114 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4115 | return -EPERM; |
4116 | } | |
dc61b1d6 | 4117 | #endif |
332ac17e DF |
4118 | #ifdef CONFIG_SMP |
4119 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4120 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4121 | |
4122 | /* | |
4123 | * Don't allow tasks with an affinity mask smaller than | |
4124 | * the entire root_domain to become SCHED_DEADLINE. We | |
4125 | * will also fail if there's no bandwidth available. | |
4126 | */ | |
e4099a5e PZ |
4127 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4128 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4129 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4130 | return -EPERM; |
4131 | } | |
4132 | } | |
4133 | #endif | |
4134 | } | |
dc61b1d6 | 4135 | |
1da177e4 LT |
4136 | /* recheck policy now with rq lock held */ |
4137 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4138 | policy = oldpolicy = -1; | |
eb580751 | 4139 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4140 | goto recheck; |
4141 | } | |
332ac17e DF |
4142 | |
4143 | /* | |
4144 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4145 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4146 | * is available. | |
4147 | */ | |
e4099a5e | 4148 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4149 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4150 | return -EBUSY; |
4151 | } | |
4152 | ||
c365c292 TG |
4153 | p->sched_reset_on_fork = reset_on_fork; |
4154 | oldprio = p->prio; | |
4155 | ||
dbc7f069 PZ |
4156 | if (pi) { |
4157 | /* | |
4158 | * Take priority boosted tasks into account. If the new | |
4159 | * effective priority is unchanged, we just store the new | |
4160 | * normal parameters and do not touch the scheduler class and | |
4161 | * the runqueue. This will be done when the task deboost | |
4162 | * itself. | |
4163 | */ | |
4164 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4165 | if (new_effective_prio == oldprio) |
4166 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4167 | } |
4168 | ||
da0c1e65 | 4169 | queued = task_on_rq_queued(p); |
051a1d1a | 4170 | running = task_current(rq, p); |
da0c1e65 | 4171 | if (queued) |
ff77e468 | 4172 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4173 | if (running) |
f3cd1c4e | 4174 | put_prev_task(rq, p); |
f6b53205 | 4175 | |
83ab0aa0 | 4176 | prev_class = p->sched_class; |
dbc7f069 | 4177 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4178 | |
0e1f3483 HS |
4179 | if (running) |
4180 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4181 | if (queued) { |
81a44c54 TG |
4182 | /* |
4183 | * We enqueue to tail when the priority of a task is | |
4184 | * increased (user space view). | |
4185 | */ | |
ff77e468 PZ |
4186 | if (oldprio < p->prio) |
4187 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4188 | |
ff77e468 | 4189 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4190 | } |
cb469845 | 4191 | |
da7a735e | 4192 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4193 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 4194 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4195 | |
dbc7f069 PZ |
4196 | if (pi) |
4197 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4198 | |
4c9a4bc8 PZ |
4199 | /* |
4200 | * Run balance callbacks after we've adjusted the PI chain. | |
4201 | */ | |
4202 | balance_callback(rq); | |
4203 | preempt_enable(); | |
95e02ca9 | 4204 | |
1da177e4 LT |
4205 | return 0; |
4206 | } | |
961ccddd | 4207 | |
7479f3c9 PZ |
4208 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4209 | const struct sched_param *param, bool check) | |
4210 | { | |
4211 | struct sched_attr attr = { | |
4212 | .sched_policy = policy, | |
4213 | .sched_priority = param->sched_priority, | |
4214 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4215 | }; | |
4216 | ||
c13db6b1 SR |
4217 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4218 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4219 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4220 | policy &= ~SCHED_RESET_ON_FORK; | |
4221 | attr.sched_policy = policy; | |
4222 | } | |
4223 | ||
dbc7f069 | 4224 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4225 | } |
961ccddd RR |
4226 | /** |
4227 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4228 | * @p: the task in question. | |
4229 | * @policy: new policy. | |
4230 | * @param: structure containing the new RT priority. | |
4231 | * | |
e69f6186 YB |
4232 | * Return: 0 on success. An error code otherwise. |
4233 | * | |
961ccddd RR |
4234 | * NOTE that the task may be already dead. |
4235 | */ | |
4236 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4237 | const struct sched_param *param) |
961ccddd | 4238 | { |
7479f3c9 | 4239 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4240 | } |
1da177e4 LT |
4241 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4242 | ||
d50dde5a DF |
4243 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4244 | { | |
dbc7f069 | 4245 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4246 | } |
4247 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4248 | ||
961ccddd RR |
4249 | /** |
4250 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4251 | * @p: the task in question. | |
4252 | * @policy: new policy. | |
4253 | * @param: structure containing the new RT priority. | |
4254 | * | |
4255 | * Just like sched_setscheduler, only don't bother checking if the | |
4256 | * current context has permission. For example, this is needed in | |
4257 | * stop_machine(): we create temporary high priority worker threads, | |
4258 | * but our caller might not have that capability. | |
e69f6186 YB |
4259 | * |
4260 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4261 | */ |
4262 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4263 | const struct sched_param *param) |
961ccddd | 4264 | { |
7479f3c9 | 4265 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4266 | } |
84778472 | 4267 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4268 | |
95cdf3b7 IM |
4269 | static int |
4270 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4271 | { |
1da177e4 LT |
4272 | struct sched_param lparam; |
4273 | struct task_struct *p; | |
36c8b586 | 4274 | int retval; |
1da177e4 LT |
4275 | |
4276 | if (!param || pid < 0) | |
4277 | return -EINVAL; | |
4278 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4279 | return -EFAULT; | |
5fe1d75f ON |
4280 | |
4281 | rcu_read_lock(); | |
4282 | retval = -ESRCH; | |
1da177e4 | 4283 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4284 | if (p != NULL) |
4285 | retval = sched_setscheduler(p, policy, &lparam); | |
4286 | rcu_read_unlock(); | |
36c8b586 | 4287 | |
1da177e4 LT |
4288 | return retval; |
4289 | } | |
4290 | ||
d50dde5a DF |
4291 | /* |
4292 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4293 | */ | |
4294 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4295 | struct sched_attr *attr) | |
4296 | { | |
4297 | u32 size; | |
4298 | int ret; | |
4299 | ||
4300 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4301 | return -EFAULT; | |
4302 | ||
4303 | /* | |
4304 | * zero the full structure, so that a short copy will be nice. | |
4305 | */ | |
4306 | memset(attr, 0, sizeof(*attr)); | |
4307 | ||
4308 | ret = get_user(size, &uattr->size); | |
4309 | if (ret) | |
4310 | return ret; | |
4311 | ||
4312 | if (size > PAGE_SIZE) /* silly large */ | |
4313 | goto err_size; | |
4314 | ||
4315 | if (!size) /* abi compat */ | |
4316 | size = SCHED_ATTR_SIZE_VER0; | |
4317 | ||
4318 | if (size < SCHED_ATTR_SIZE_VER0) | |
4319 | goto err_size; | |
4320 | ||
4321 | /* | |
4322 | * If we're handed a bigger struct than we know of, | |
4323 | * ensure all the unknown bits are 0 - i.e. new | |
4324 | * user-space does not rely on any kernel feature | |
4325 | * extensions we dont know about yet. | |
4326 | */ | |
4327 | if (size > sizeof(*attr)) { | |
4328 | unsigned char __user *addr; | |
4329 | unsigned char __user *end; | |
4330 | unsigned char val; | |
4331 | ||
4332 | addr = (void __user *)uattr + sizeof(*attr); | |
4333 | end = (void __user *)uattr + size; | |
4334 | ||
4335 | for (; addr < end; addr++) { | |
4336 | ret = get_user(val, addr); | |
4337 | if (ret) | |
4338 | return ret; | |
4339 | if (val) | |
4340 | goto err_size; | |
4341 | } | |
4342 | size = sizeof(*attr); | |
4343 | } | |
4344 | ||
4345 | ret = copy_from_user(attr, uattr, size); | |
4346 | if (ret) | |
4347 | return -EFAULT; | |
4348 | ||
4349 | /* | |
4350 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4351 | * to be strict and return an error on out-of-bounds values? | |
4352 | */ | |
75e45d51 | 4353 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4354 | |
e78c7bca | 4355 | return 0; |
d50dde5a DF |
4356 | |
4357 | err_size: | |
4358 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4359 | return -E2BIG; |
d50dde5a DF |
4360 | } |
4361 | ||
1da177e4 LT |
4362 | /** |
4363 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4364 | * @pid: the pid in question. | |
4365 | * @policy: new policy. | |
4366 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4367 | * |
4368 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4369 | */ |
5add95d4 HC |
4370 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4371 | struct sched_param __user *, param) | |
1da177e4 | 4372 | { |
c21761f1 JB |
4373 | /* negative values for policy are not valid */ |
4374 | if (policy < 0) | |
4375 | return -EINVAL; | |
4376 | ||
1da177e4 LT |
4377 | return do_sched_setscheduler(pid, policy, param); |
4378 | } | |
4379 | ||
4380 | /** | |
4381 | * sys_sched_setparam - set/change the RT priority of a thread | |
4382 | * @pid: the pid in question. | |
4383 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4384 | * |
4385 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4386 | */ |
5add95d4 | 4387 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4388 | { |
c13db6b1 | 4389 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4390 | } |
4391 | ||
d50dde5a DF |
4392 | /** |
4393 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4394 | * @pid: the pid in question. | |
5778fccf | 4395 | * @uattr: structure containing the extended parameters. |
db66d756 | 4396 | * @flags: for future extension. |
d50dde5a | 4397 | */ |
6d35ab48 PZ |
4398 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4399 | unsigned int, flags) | |
d50dde5a DF |
4400 | { |
4401 | struct sched_attr attr; | |
4402 | struct task_struct *p; | |
4403 | int retval; | |
4404 | ||
6d35ab48 | 4405 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4406 | return -EINVAL; |
4407 | ||
143cf23d MK |
4408 | retval = sched_copy_attr(uattr, &attr); |
4409 | if (retval) | |
4410 | return retval; | |
d50dde5a | 4411 | |
b14ed2c2 | 4412 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4413 | return -EINVAL; |
d50dde5a DF |
4414 | |
4415 | rcu_read_lock(); | |
4416 | retval = -ESRCH; | |
4417 | p = find_process_by_pid(pid); | |
4418 | if (p != NULL) | |
4419 | retval = sched_setattr(p, &attr); | |
4420 | rcu_read_unlock(); | |
4421 | ||
4422 | return retval; | |
4423 | } | |
4424 | ||
1da177e4 LT |
4425 | /** |
4426 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4427 | * @pid: the pid in question. | |
e69f6186 YB |
4428 | * |
4429 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4430 | * code. | |
1da177e4 | 4431 | */ |
5add95d4 | 4432 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4433 | { |
36c8b586 | 4434 | struct task_struct *p; |
3a5c359a | 4435 | int retval; |
1da177e4 LT |
4436 | |
4437 | if (pid < 0) | |
3a5c359a | 4438 | return -EINVAL; |
1da177e4 LT |
4439 | |
4440 | retval = -ESRCH; | |
5fe85be0 | 4441 | rcu_read_lock(); |
1da177e4 LT |
4442 | p = find_process_by_pid(pid); |
4443 | if (p) { | |
4444 | retval = security_task_getscheduler(p); | |
4445 | if (!retval) | |
ca94c442 LP |
4446 | retval = p->policy |
4447 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4448 | } |
5fe85be0 | 4449 | rcu_read_unlock(); |
1da177e4 LT |
4450 | return retval; |
4451 | } | |
4452 | ||
4453 | /** | |
ca94c442 | 4454 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4455 | * @pid: the pid in question. |
4456 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4457 | * |
4458 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4459 | * code. | |
1da177e4 | 4460 | */ |
5add95d4 | 4461 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4462 | { |
ce5f7f82 | 4463 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4464 | struct task_struct *p; |
3a5c359a | 4465 | int retval; |
1da177e4 LT |
4466 | |
4467 | if (!param || pid < 0) | |
3a5c359a | 4468 | return -EINVAL; |
1da177e4 | 4469 | |
5fe85be0 | 4470 | rcu_read_lock(); |
1da177e4 LT |
4471 | p = find_process_by_pid(pid); |
4472 | retval = -ESRCH; | |
4473 | if (!p) | |
4474 | goto out_unlock; | |
4475 | ||
4476 | retval = security_task_getscheduler(p); | |
4477 | if (retval) | |
4478 | goto out_unlock; | |
4479 | ||
ce5f7f82 PZ |
4480 | if (task_has_rt_policy(p)) |
4481 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4482 | rcu_read_unlock(); |
1da177e4 LT |
4483 | |
4484 | /* | |
4485 | * This one might sleep, we cannot do it with a spinlock held ... | |
4486 | */ | |
4487 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4488 | ||
1da177e4 LT |
4489 | return retval; |
4490 | ||
4491 | out_unlock: | |
5fe85be0 | 4492 | rcu_read_unlock(); |
1da177e4 LT |
4493 | return retval; |
4494 | } | |
4495 | ||
d50dde5a DF |
4496 | static int sched_read_attr(struct sched_attr __user *uattr, |
4497 | struct sched_attr *attr, | |
4498 | unsigned int usize) | |
4499 | { | |
4500 | int ret; | |
4501 | ||
4502 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4503 | return -EFAULT; | |
4504 | ||
4505 | /* | |
4506 | * If we're handed a smaller struct than we know of, | |
4507 | * ensure all the unknown bits are 0 - i.e. old | |
4508 | * user-space does not get uncomplete information. | |
4509 | */ | |
4510 | if (usize < sizeof(*attr)) { | |
4511 | unsigned char *addr; | |
4512 | unsigned char *end; | |
4513 | ||
4514 | addr = (void *)attr + usize; | |
4515 | end = (void *)attr + sizeof(*attr); | |
4516 | ||
4517 | for (; addr < end; addr++) { | |
4518 | if (*addr) | |
22400674 | 4519 | return -EFBIG; |
d50dde5a DF |
4520 | } |
4521 | ||
4522 | attr->size = usize; | |
4523 | } | |
4524 | ||
4efbc454 | 4525 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4526 | if (ret) |
4527 | return -EFAULT; | |
4528 | ||
22400674 | 4529 | return 0; |
d50dde5a DF |
4530 | } |
4531 | ||
4532 | /** | |
aab03e05 | 4533 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4534 | * @pid: the pid in question. |
5778fccf | 4535 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4536 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4537 | * @flags: for future extension. |
d50dde5a | 4538 | */ |
6d35ab48 PZ |
4539 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4540 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4541 | { |
4542 | struct sched_attr attr = { | |
4543 | .size = sizeof(struct sched_attr), | |
4544 | }; | |
4545 | struct task_struct *p; | |
4546 | int retval; | |
4547 | ||
4548 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4549 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4550 | return -EINVAL; |
4551 | ||
4552 | rcu_read_lock(); | |
4553 | p = find_process_by_pid(pid); | |
4554 | retval = -ESRCH; | |
4555 | if (!p) | |
4556 | goto out_unlock; | |
4557 | ||
4558 | retval = security_task_getscheduler(p); | |
4559 | if (retval) | |
4560 | goto out_unlock; | |
4561 | ||
4562 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4563 | if (p->sched_reset_on_fork) |
4564 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4565 | if (task_has_dl_policy(p)) |
4566 | __getparam_dl(p, &attr); | |
4567 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4568 | attr.sched_priority = p->rt_priority; |
4569 | else | |
d0ea0268 | 4570 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4571 | |
4572 | rcu_read_unlock(); | |
4573 | ||
4574 | retval = sched_read_attr(uattr, &attr, size); | |
4575 | return retval; | |
4576 | ||
4577 | out_unlock: | |
4578 | rcu_read_unlock(); | |
4579 | return retval; | |
4580 | } | |
4581 | ||
96f874e2 | 4582 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4583 | { |
5a16f3d3 | 4584 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4585 | struct task_struct *p; |
4586 | int retval; | |
1da177e4 | 4587 | |
23f5d142 | 4588 | rcu_read_lock(); |
1da177e4 LT |
4589 | |
4590 | p = find_process_by_pid(pid); | |
4591 | if (!p) { | |
23f5d142 | 4592 | rcu_read_unlock(); |
1da177e4 LT |
4593 | return -ESRCH; |
4594 | } | |
4595 | ||
23f5d142 | 4596 | /* Prevent p going away */ |
1da177e4 | 4597 | get_task_struct(p); |
23f5d142 | 4598 | rcu_read_unlock(); |
1da177e4 | 4599 | |
14a40ffc TH |
4600 | if (p->flags & PF_NO_SETAFFINITY) { |
4601 | retval = -EINVAL; | |
4602 | goto out_put_task; | |
4603 | } | |
5a16f3d3 RR |
4604 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4605 | retval = -ENOMEM; | |
4606 | goto out_put_task; | |
4607 | } | |
4608 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4609 | retval = -ENOMEM; | |
4610 | goto out_free_cpus_allowed; | |
4611 | } | |
1da177e4 | 4612 | retval = -EPERM; |
4c44aaaf EB |
4613 | if (!check_same_owner(p)) { |
4614 | rcu_read_lock(); | |
4615 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4616 | rcu_read_unlock(); | |
16303ab2 | 4617 | goto out_free_new_mask; |
4c44aaaf EB |
4618 | } |
4619 | rcu_read_unlock(); | |
4620 | } | |
1da177e4 | 4621 | |
b0ae1981 | 4622 | retval = security_task_setscheduler(p); |
e7834f8f | 4623 | if (retval) |
16303ab2 | 4624 | goto out_free_new_mask; |
e7834f8f | 4625 | |
e4099a5e PZ |
4626 | |
4627 | cpuset_cpus_allowed(p, cpus_allowed); | |
4628 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4629 | ||
332ac17e DF |
4630 | /* |
4631 | * Since bandwidth control happens on root_domain basis, | |
4632 | * if admission test is enabled, we only admit -deadline | |
4633 | * tasks allowed to run on all the CPUs in the task's | |
4634 | * root_domain. | |
4635 | */ | |
4636 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4637 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4638 | rcu_read_lock(); | |
4639 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4640 | retval = -EBUSY; |
f1e3a093 | 4641 | rcu_read_unlock(); |
16303ab2 | 4642 | goto out_free_new_mask; |
332ac17e | 4643 | } |
f1e3a093 | 4644 | rcu_read_unlock(); |
332ac17e DF |
4645 | } |
4646 | #endif | |
49246274 | 4647 | again: |
25834c73 | 4648 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4649 | |
8707d8b8 | 4650 | if (!retval) { |
5a16f3d3 RR |
4651 | cpuset_cpus_allowed(p, cpus_allowed); |
4652 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4653 | /* |
4654 | * We must have raced with a concurrent cpuset | |
4655 | * update. Just reset the cpus_allowed to the | |
4656 | * cpuset's cpus_allowed | |
4657 | */ | |
5a16f3d3 | 4658 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4659 | goto again; |
4660 | } | |
4661 | } | |
16303ab2 | 4662 | out_free_new_mask: |
5a16f3d3 RR |
4663 | free_cpumask_var(new_mask); |
4664 | out_free_cpus_allowed: | |
4665 | free_cpumask_var(cpus_allowed); | |
4666 | out_put_task: | |
1da177e4 | 4667 | put_task_struct(p); |
1da177e4 LT |
4668 | return retval; |
4669 | } | |
4670 | ||
4671 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4672 | struct cpumask *new_mask) |
1da177e4 | 4673 | { |
96f874e2 RR |
4674 | if (len < cpumask_size()) |
4675 | cpumask_clear(new_mask); | |
4676 | else if (len > cpumask_size()) | |
4677 | len = cpumask_size(); | |
4678 | ||
1da177e4 LT |
4679 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4680 | } | |
4681 | ||
4682 | /** | |
4683 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4684 | * @pid: pid of the process | |
4685 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4686 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4687 | * |
4688 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4689 | */ |
5add95d4 HC |
4690 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4691 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4692 | { |
5a16f3d3 | 4693 | cpumask_var_t new_mask; |
1da177e4 LT |
4694 | int retval; |
4695 | ||
5a16f3d3 RR |
4696 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4697 | return -ENOMEM; | |
1da177e4 | 4698 | |
5a16f3d3 RR |
4699 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4700 | if (retval == 0) | |
4701 | retval = sched_setaffinity(pid, new_mask); | |
4702 | free_cpumask_var(new_mask); | |
4703 | return retval; | |
1da177e4 LT |
4704 | } |
4705 | ||
96f874e2 | 4706 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4707 | { |
36c8b586 | 4708 | struct task_struct *p; |
31605683 | 4709 | unsigned long flags; |
1da177e4 | 4710 | int retval; |
1da177e4 | 4711 | |
23f5d142 | 4712 | rcu_read_lock(); |
1da177e4 LT |
4713 | |
4714 | retval = -ESRCH; | |
4715 | p = find_process_by_pid(pid); | |
4716 | if (!p) | |
4717 | goto out_unlock; | |
4718 | ||
e7834f8f DQ |
4719 | retval = security_task_getscheduler(p); |
4720 | if (retval) | |
4721 | goto out_unlock; | |
4722 | ||
013fdb80 | 4723 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4724 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4725 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4726 | |
4727 | out_unlock: | |
23f5d142 | 4728 | rcu_read_unlock(); |
1da177e4 | 4729 | |
9531b62f | 4730 | return retval; |
1da177e4 LT |
4731 | } |
4732 | ||
4733 | /** | |
4734 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4735 | * @pid: pid of the process | |
4736 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4737 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4738 | * |
4739 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4740 | */ |
5add95d4 HC |
4741 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4742 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4743 | { |
4744 | int ret; | |
f17c8607 | 4745 | cpumask_var_t mask; |
1da177e4 | 4746 | |
84fba5ec | 4747 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4748 | return -EINVAL; |
4749 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4750 | return -EINVAL; |
4751 | ||
f17c8607 RR |
4752 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4753 | return -ENOMEM; | |
1da177e4 | 4754 | |
f17c8607 RR |
4755 | ret = sched_getaffinity(pid, mask); |
4756 | if (ret == 0) { | |
8bc037fb | 4757 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4758 | |
4759 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4760 | ret = -EFAULT; |
4761 | else | |
cd3d8031 | 4762 | ret = retlen; |
f17c8607 RR |
4763 | } |
4764 | free_cpumask_var(mask); | |
1da177e4 | 4765 | |
f17c8607 | 4766 | return ret; |
1da177e4 LT |
4767 | } |
4768 | ||
4769 | /** | |
4770 | * sys_sched_yield - yield the current processor to other threads. | |
4771 | * | |
dd41f596 IM |
4772 | * This function yields the current CPU to other tasks. If there are no |
4773 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4774 | * |
4775 | * Return: 0. | |
1da177e4 | 4776 | */ |
5add95d4 | 4777 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4778 | { |
70b97a7f | 4779 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4780 | |
2d72376b | 4781 | schedstat_inc(rq, yld_count); |
4530d7ab | 4782 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4783 | |
4784 | /* | |
4785 | * Since we are going to call schedule() anyway, there's | |
4786 | * no need to preempt or enable interrupts: | |
4787 | */ | |
4788 | __release(rq->lock); | |
8a25d5de | 4789 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4790 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4791 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4792 | |
4793 | schedule(); | |
4794 | ||
4795 | return 0; | |
4796 | } | |
4797 | ||
02b67cc3 | 4798 | int __sched _cond_resched(void) |
1da177e4 | 4799 | { |
fe32d3cd | 4800 | if (should_resched(0)) { |
a18b5d01 | 4801 | preempt_schedule_common(); |
1da177e4 LT |
4802 | return 1; |
4803 | } | |
4804 | return 0; | |
4805 | } | |
02b67cc3 | 4806 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4807 | |
4808 | /* | |
613afbf8 | 4809 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4810 | * call schedule, and on return reacquire the lock. |
4811 | * | |
41a2d6cf | 4812 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4813 | * operations here to prevent schedule() from being called twice (once via |
4814 | * spin_unlock(), once by hand). | |
4815 | */ | |
613afbf8 | 4816 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4817 | { |
fe32d3cd | 4818 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4819 | int ret = 0; |
4820 | ||
f607c668 PZ |
4821 | lockdep_assert_held(lock); |
4822 | ||
4a81e832 | 4823 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4824 | spin_unlock(lock); |
d86ee480 | 4825 | if (resched) |
a18b5d01 | 4826 | preempt_schedule_common(); |
95c354fe NP |
4827 | else |
4828 | cpu_relax(); | |
6df3cecb | 4829 | ret = 1; |
1da177e4 | 4830 | spin_lock(lock); |
1da177e4 | 4831 | } |
6df3cecb | 4832 | return ret; |
1da177e4 | 4833 | } |
613afbf8 | 4834 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4835 | |
613afbf8 | 4836 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4837 | { |
4838 | BUG_ON(!in_softirq()); | |
4839 | ||
fe32d3cd | 4840 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4841 | local_bh_enable(); |
a18b5d01 | 4842 | preempt_schedule_common(); |
1da177e4 LT |
4843 | local_bh_disable(); |
4844 | return 1; | |
4845 | } | |
4846 | return 0; | |
4847 | } | |
613afbf8 | 4848 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4849 | |
1da177e4 LT |
4850 | /** |
4851 | * yield - yield the current processor to other threads. | |
4852 | * | |
8e3fabfd PZ |
4853 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4854 | * | |
4855 | * The scheduler is at all times free to pick the calling task as the most | |
4856 | * eligible task to run, if removing the yield() call from your code breaks | |
4857 | * it, its already broken. | |
4858 | * | |
4859 | * Typical broken usage is: | |
4860 | * | |
4861 | * while (!event) | |
4862 | * yield(); | |
4863 | * | |
4864 | * where one assumes that yield() will let 'the other' process run that will | |
4865 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4866 | * happen. Never use yield() as a progress guarantee!! | |
4867 | * | |
4868 | * If you want to use yield() to wait for something, use wait_event(). | |
4869 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4870 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4871 | */ |
4872 | void __sched yield(void) | |
4873 | { | |
4874 | set_current_state(TASK_RUNNING); | |
4875 | sys_sched_yield(); | |
4876 | } | |
1da177e4 LT |
4877 | EXPORT_SYMBOL(yield); |
4878 | ||
d95f4122 MG |
4879 | /** |
4880 | * yield_to - yield the current processor to another thread in | |
4881 | * your thread group, or accelerate that thread toward the | |
4882 | * processor it's on. | |
16addf95 RD |
4883 | * @p: target task |
4884 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4885 | * |
4886 | * It's the caller's job to ensure that the target task struct | |
4887 | * can't go away on us before we can do any checks. | |
4888 | * | |
e69f6186 | 4889 | * Return: |
7b270f60 PZ |
4890 | * true (>0) if we indeed boosted the target task. |
4891 | * false (0) if we failed to boost the target. | |
4892 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4893 | */ |
fa93384f | 4894 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4895 | { |
4896 | struct task_struct *curr = current; | |
4897 | struct rq *rq, *p_rq; | |
4898 | unsigned long flags; | |
c3c18640 | 4899 | int yielded = 0; |
d95f4122 MG |
4900 | |
4901 | local_irq_save(flags); | |
4902 | rq = this_rq(); | |
4903 | ||
4904 | again: | |
4905 | p_rq = task_rq(p); | |
7b270f60 PZ |
4906 | /* |
4907 | * If we're the only runnable task on the rq and target rq also | |
4908 | * has only one task, there's absolutely no point in yielding. | |
4909 | */ | |
4910 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4911 | yielded = -ESRCH; | |
4912 | goto out_irq; | |
4913 | } | |
4914 | ||
d95f4122 | 4915 | double_rq_lock(rq, p_rq); |
39e24d8f | 4916 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4917 | double_rq_unlock(rq, p_rq); |
4918 | goto again; | |
4919 | } | |
4920 | ||
4921 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4922 | goto out_unlock; |
d95f4122 MG |
4923 | |
4924 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4925 | goto out_unlock; |
d95f4122 MG |
4926 | |
4927 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4928 | goto out_unlock; |
d95f4122 MG |
4929 | |
4930 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4931 | if (yielded) { |
d95f4122 | 4932 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4933 | /* |
4934 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4935 | * fairness. | |
4936 | */ | |
4937 | if (preempt && rq != p_rq) | |
8875125e | 4938 | resched_curr(p_rq); |
6d1cafd8 | 4939 | } |
d95f4122 | 4940 | |
7b270f60 | 4941 | out_unlock: |
d95f4122 | 4942 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4943 | out_irq: |
d95f4122 MG |
4944 | local_irq_restore(flags); |
4945 | ||
7b270f60 | 4946 | if (yielded > 0) |
d95f4122 MG |
4947 | schedule(); |
4948 | ||
4949 | return yielded; | |
4950 | } | |
4951 | EXPORT_SYMBOL_GPL(yield_to); | |
4952 | ||
1da177e4 | 4953 | /* |
41a2d6cf | 4954 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4955 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4956 | */ |
1da177e4 LT |
4957 | long __sched io_schedule_timeout(long timeout) |
4958 | { | |
9cff8ade N |
4959 | int old_iowait = current->in_iowait; |
4960 | struct rq *rq; | |
1da177e4 LT |
4961 | long ret; |
4962 | ||
9cff8ade | 4963 | current->in_iowait = 1; |
10d784ea | 4964 | blk_schedule_flush_plug(current); |
9cff8ade | 4965 | |
0ff92245 | 4966 | delayacct_blkio_start(); |
9cff8ade | 4967 | rq = raw_rq(); |
1da177e4 LT |
4968 | atomic_inc(&rq->nr_iowait); |
4969 | ret = schedule_timeout(timeout); | |
9cff8ade | 4970 | current->in_iowait = old_iowait; |
1da177e4 | 4971 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4972 | delayacct_blkio_end(); |
9cff8ade | 4973 | |
1da177e4 LT |
4974 | return ret; |
4975 | } | |
9cff8ade | 4976 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4977 | |
4978 | /** | |
4979 | * sys_sched_get_priority_max - return maximum RT priority. | |
4980 | * @policy: scheduling class. | |
4981 | * | |
e69f6186 YB |
4982 | * Return: On success, this syscall returns the maximum |
4983 | * rt_priority that can be used by a given scheduling class. | |
4984 | * On failure, a negative error code is returned. | |
1da177e4 | 4985 | */ |
5add95d4 | 4986 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4987 | { |
4988 | int ret = -EINVAL; | |
4989 | ||
4990 | switch (policy) { | |
4991 | case SCHED_FIFO: | |
4992 | case SCHED_RR: | |
4993 | ret = MAX_USER_RT_PRIO-1; | |
4994 | break; | |
aab03e05 | 4995 | case SCHED_DEADLINE: |
1da177e4 | 4996 | case SCHED_NORMAL: |
b0a9499c | 4997 | case SCHED_BATCH: |
dd41f596 | 4998 | case SCHED_IDLE: |
1da177e4 LT |
4999 | ret = 0; |
5000 | break; | |
5001 | } | |
5002 | return ret; | |
5003 | } | |
5004 | ||
5005 | /** | |
5006 | * sys_sched_get_priority_min - return minimum RT priority. | |
5007 | * @policy: scheduling class. | |
5008 | * | |
e69f6186 YB |
5009 | * Return: On success, this syscall returns the minimum |
5010 | * rt_priority that can be used by a given scheduling class. | |
5011 | * On failure, a negative error code is returned. | |
1da177e4 | 5012 | */ |
5add95d4 | 5013 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5014 | { |
5015 | int ret = -EINVAL; | |
5016 | ||
5017 | switch (policy) { | |
5018 | case SCHED_FIFO: | |
5019 | case SCHED_RR: | |
5020 | ret = 1; | |
5021 | break; | |
aab03e05 | 5022 | case SCHED_DEADLINE: |
1da177e4 | 5023 | case SCHED_NORMAL: |
b0a9499c | 5024 | case SCHED_BATCH: |
dd41f596 | 5025 | case SCHED_IDLE: |
1da177e4 LT |
5026 | ret = 0; |
5027 | } | |
5028 | return ret; | |
5029 | } | |
5030 | ||
5031 | /** | |
5032 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5033 | * @pid: pid of the process. | |
5034 | * @interval: userspace pointer to the timeslice value. | |
5035 | * | |
5036 | * this syscall writes the default timeslice value of a given process | |
5037 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5038 | * |
5039 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5040 | * an error code. | |
1da177e4 | 5041 | */ |
17da2bd9 | 5042 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5043 | struct timespec __user *, interval) |
1da177e4 | 5044 | { |
36c8b586 | 5045 | struct task_struct *p; |
a4ec24b4 | 5046 | unsigned int time_slice; |
eb580751 PZ |
5047 | struct rq_flags rf; |
5048 | struct timespec t; | |
dba091b9 | 5049 | struct rq *rq; |
3a5c359a | 5050 | int retval; |
1da177e4 LT |
5051 | |
5052 | if (pid < 0) | |
3a5c359a | 5053 | return -EINVAL; |
1da177e4 LT |
5054 | |
5055 | retval = -ESRCH; | |
1a551ae7 | 5056 | rcu_read_lock(); |
1da177e4 LT |
5057 | p = find_process_by_pid(pid); |
5058 | if (!p) | |
5059 | goto out_unlock; | |
5060 | ||
5061 | retval = security_task_getscheduler(p); | |
5062 | if (retval) | |
5063 | goto out_unlock; | |
5064 | ||
eb580751 | 5065 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5066 | time_slice = 0; |
5067 | if (p->sched_class->get_rr_interval) | |
5068 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5069 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5070 | |
1a551ae7 | 5071 | rcu_read_unlock(); |
a4ec24b4 | 5072 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5073 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5074 | return retval; |
3a5c359a | 5075 | |
1da177e4 | 5076 | out_unlock: |
1a551ae7 | 5077 | rcu_read_unlock(); |
1da177e4 LT |
5078 | return retval; |
5079 | } | |
5080 | ||
7c731e0a | 5081 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5082 | |
82a1fcb9 | 5083 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5084 | { |
1da177e4 | 5085 | unsigned long free = 0; |
4e79752c | 5086 | int ppid; |
1f8a7633 | 5087 | unsigned long state = p->state; |
1da177e4 | 5088 | |
1f8a7633 TH |
5089 | if (state) |
5090 | state = __ffs(state) + 1; | |
28d0686c | 5091 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5092 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5093 | #if BITS_PER_LONG == 32 |
1da177e4 | 5094 | if (state == TASK_RUNNING) |
3df0fc5b | 5095 | printk(KERN_CONT " running "); |
1da177e4 | 5096 | else |
3df0fc5b | 5097 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5098 | #else |
5099 | if (state == TASK_RUNNING) | |
3df0fc5b | 5100 | printk(KERN_CONT " running task "); |
1da177e4 | 5101 | else |
3df0fc5b | 5102 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5103 | #endif |
5104 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5105 | free = stack_not_used(p); |
1da177e4 | 5106 | #endif |
a90e984c | 5107 | ppid = 0; |
4e79752c | 5108 | rcu_read_lock(); |
a90e984c ON |
5109 | if (pid_alive(p)) |
5110 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5111 | rcu_read_unlock(); |
3df0fc5b | 5112 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5113 | task_pid_nr(p), ppid, |
aa47b7e0 | 5114 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5115 | |
3d1cb205 | 5116 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5117 | show_stack(p, NULL); |
1da177e4 LT |
5118 | } |
5119 | ||
e59e2ae2 | 5120 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5121 | { |
36c8b586 | 5122 | struct task_struct *g, *p; |
1da177e4 | 5123 | |
4bd77321 | 5124 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5125 | printk(KERN_INFO |
5126 | " task PC stack pid father\n"); | |
1da177e4 | 5127 | #else |
3df0fc5b PZ |
5128 | printk(KERN_INFO |
5129 | " task PC stack pid father\n"); | |
1da177e4 | 5130 | #endif |
510f5acc | 5131 | rcu_read_lock(); |
5d07f420 | 5132 | for_each_process_thread(g, p) { |
1da177e4 LT |
5133 | /* |
5134 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5135 | * console might take a lot of time: |
1da177e4 LT |
5136 | */ |
5137 | touch_nmi_watchdog(); | |
39bc89fd | 5138 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5139 | sched_show_task(p); |
5d07f420 | 5140 | } |
1da177e4 | 5141 | |
04c9167f JF |
5142 | touch_all_softlockup_watchdogs(); |
5143 | ||
dd41f596 | 5144 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5145 | if (!state_filter) |
5146 | sysrq_sched_debug_show(); | |
dd41f596 | 5147 | #endif |
510f5acc | 5148 | rcu_read_unlock(); |
e59e2ae2 IM |
5149 | /* |
5150 | * Only show locks if all tasks are dumped: | |
5151 | */ | |
93335a21 | 5152 | if (!state_filter) |
e59e2ae2 | 5153 | debug_show_all_locks(); |
1da177e4 LT |
5154 | } |
5155 | ||
0db0628d | 5156 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5157 | { |
dd41f596 | 5158 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5159 | } |
5160 | ||
f340c0d1 IM |
5161 | /** |
5162 | * init_idle - set up an idle thread for a given CPU | |
5163 | * @idle: task in question | |
5164 | * @cpu: cpu the idle task belongs to | |
5165 | * | |
5166 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5167 | * flag, to make booting more robust. | |
5168 | */ | |
0db0628d | 5169 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5170 | { |
70b97a7f | 5171 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5172 | unsigned long flags; |
5173 | ||
25834c73 PZ |
5174 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5175 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5176 | |
5e1576ed | 5177 | __sched_fork(0, idle); |
06b83b5f | 5178 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5179 | idle->se.exec_start = sched_clock(); |
5180 | ||
e1b77c92 MR |
5181 | kasan_unpoison_task_stack(idle); |
5182 | ||
de9b8f5d PZ |
5183 | #ifdef CONFIG_SMP |
5184 | /* | |
5185 | * Its possible that init_idle() gets called multiple times on a task, | |
5186 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5187 | * | |
5188 | * And since this is boot we can forgo the serialization. | |
5189 | */ | |
5190 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5191 | #endif | |
6506cf6c PZ |
5192 | /* |
5193 | * We're having a chicken and egg problem, even though we are | |
5194 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5195 | * lockdep check in task_group() will fail. | |
5196 | * | |
5197 | * Similar case to sched_fork(). / Alternatively we could | |
5198 | * use task_rq_lock() here and obtain the other rq->lock. | |
5199 | * | |
5200 | * Silence PROVE_RCU | |
5201 | */ | |
5202 | rcu_read_lock(); | |
dd41f596 | 5203 | __set_task_cpu(idle, cpu); |
6506cf6c | 5204 | rcu_read_unlock(); |
1da177e4 | 5205 | |
1da177e4 | 5206 | rq->curr = rq->idle = idle; |
da0c1e65 | 5207 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5208 | #ifdef CONFIG_SMP |
3ca7a440 | 5209 | idle->on_cpu = 1; |
4866cde0 | 5210 | #endif |
25834c73 PZ |
5211 | raw_spin_unlock(&rq->lock); |
5212 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5213 | |
5214 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5215 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5216 | |
dd41f596 IM |
5217 | /* |
5218 | * The idle tasks have their own, simple scheduling class: | |
5219 | */ | |
5220 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5221 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5222 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5223 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5224 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5225 | #endif | |
19978ca6 IM |
5226 | } |
5227 | ||
f82f8042 JL |
5228 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5229 | const struct cpumask *trial) | |
5230 | { | |
5231 | int ret = 1, trial_cpus; | |
5232 | struct dl_bw *cur_dl_b; | |
5233 | unsigned long flags; | |
5234 | ||
bb2bc55a MG |
5235 | if (!cpumask_weight(cur)) |
5236 | return ret; | |
5237 | ||
75e23e49 | 5238 | rcu_read_lock_sched(); |
f82f8042 JL |
5239 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5240 | trial_cpus = cpumask_weight(trial); | |
5241 | ||
5242 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5243 | if (cur_dl_b->bw != -1 && | |
5244 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5245 | ret = 0; | |
5246 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5247 | rcu_read_unlock_sched(); |
f82f8042 JL |
5248 | |
5249 | return ret; | |
5250 | } | |
5251 | ||
7f51412a JL |
5252 | int task_can_attach(struct task_struct *p, |
5253 | const struct cpumask *cs_cpus_allowed) | |
5254 | { | |
5255 | int ret = 0; | |
5256 | ||
5257 | /* | |
5258 | * Kthreads which disallow setaffinity shouldn't be moved | |
5259 | * to a new cpuset; we don't want to change their cpu | |
5260 | * affinity and isolating such threads by their set of | |
5261 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5262 | * applicable for such threads. This prevents checking for | |
5263 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5264 | * before cpus_allowed may be changed. | |
5265 | */ | |
5266 | if (p->flags & PF_NO_SETAFFINITY) { | |
5267 | ret = -EINVAL; | |
5268 | goto out; | |
5269 | } | |
5270 | ||
5271 | #ifdef CONFIG_SMP | |
5272 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5273 | cs_cpus_allowed)) { | |
5274 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5275 | cs_cpus_allowed); | |
75e23e49 | 5276 | struct dl_bw *dl_b; |
7f51412a JL |
5277 | bool overflow; |
5278 | int cpus; | |
5279 | unsigned long flags; | |
5280 | ||
75e23e49 JL |
5281 | rcu_read_lock_sched(); |
5282 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5283 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5284 | cpus = dl_bw_cpus(dest_cpu); | |
5285 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5286 | if (overflow) | |
5287 | ret = -EBUSY; | |
5288 | else { | |
5289 | /* | |
5290 | * We reserve space for this task in the destination | |
5291 | * root_domain, as we can't fail after this point. | |
5292 | * We will free resources in the source root_domain | |
5293 | * later on (see set_cpus_allowed_dl()). | |
5294 | */ | |
5295 | __dl_add(dl_b, p->dl.dl_bw); | |
5296 | } | |
5297 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5298 | rcu_read_unlock_sched(); |
7f51412a JL |
5299 | |
5300 | } | |
5301 | #endif | |
5302 | out: | |
5303 | return ret; | |
5304 | } | |
5305 | ||
1da177e4 | 5306 | #ifdef CONFIG_SMP |
1da177e4 | 5307 | |
e26fbffd TG |
5308 | static bool sched_smp_initialized __read_mostly; |
5309 | ||
e6628d5b MG |
5310 | #ifdef CONFIG_NUMA_BALANCING |
5311 | /* Migrate current task p to target_cpu */ | |
5312 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5313 | { | |
5314 | struct migration_arg arg = { p, target_cpu }; | |
5315 | int curr_cpu = task_cpu(p); | |
5316 | ||
5317 | if (curr_cpu == target_cpu) | |
5318 | return 0; | |
5319 | ||
5320 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5321 | return -EINVAL; | |
5322 | ||
5323 | /* TODO: This is not properly updating schedstats */ | |
5324 | ||
286549dc | 5325 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5326 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5327 | } | |
0ec8aa00 PZ |
5328 | |
5329 | /* | |
5330 | * Requeue a task on a given node and accurately track the number of NUMA | |
5331 | * tasks on the runqueues | |
5332 | */ | |
5333 | void sched_setnuma(struct task_struct *p, int nid) | |
5334 | { | |
da0c1e65 | 5335 | bool queued, running; |
eb580751 PZ |
5336 | struct rq_flags rf; |
5337 | struct rq *rq; | |
0ec8aa00 | 5338 | |
eb580751 | 5339 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5340 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5341 | running = task_current(rq, p); |
5342 | ||
da0c1e65 | 5343 | if (queued) |
1de64443 | 5344 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5345 | if (running) |
f3cd1c4e | 5346 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5347 | |
5348 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5349 | |
5350 | if (running) | |
5351 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5352 | if (queued) |
1de64443 | 5353 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
eb580751 | 5354 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5355 | } |
5cc389bc | 5356 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5357 | |
1da177e4 | 5358 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5359 | /* |
48c5ccae PZ |
5360 | * Ensures that the idle task is using init_mm right before its cpu goes |
5361 | * offline. | |
054b9108 | 5362 | */ |
48c5ccae | 5363 | void idle_task_exit(void) |
1da177e4 | 5364 | { |
48c5ccae | 5365 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5366 | |
48c5ccae | 5367 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5368 | |
a53efe5f | 5369 | if (mm != &init_mm) { |
f98db601 | 5370 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5371 | finish_arch_post_lock_switch(); |
5372 | } | |
48c5ccae | 5373 | mmdrop(mm); |
1da177e4 LT |
5374 | } |
5375 | ||
5376 | /* | |
5d180232 PZ |
5377 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5378 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5379 | * nr_active count is stable. | |
5380 | * | |
5381 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5382 | */ |
5d180232 | 5383 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5384 | { |
5d180232 PZ |
5385 | long delta = calc_load_fold_active(rq); |
5386 | if (delta) | |
5387 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5388 | } |
5389 | ||
3f1d2a31 PZ |
5390 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5391 | { | |
5392 | } | |
5393 | ||
5394 | static const struct sched_class fake_sched_class = { | |
5395 | .put_prev_task = put_prev_task_fake, | |
5396 | }; | |
5397 | ||
5398 | static struct task_struct fake_task = { | |
5399 | /* | |
5400 | * Avoid pull_{rt,dl}_task() | |
5401 | */ | |
5402 | .prio = MAX_PRIO + 1, | |
5403 | .sched_class = &fake_sched_class, | |
5404 | }; | |
5405 | ||
48f24c4d | 5406 | /* |
48c5ccae PZ |
5407 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5408 | * try_to_wake_up()->select_task_rq(). | |
5409 | * | |
5410 | * Called with rq->lock held even though we'er in stop_machine() and | |
5411 | * there's no concurrency possible, we hold the required locks anyway | |
5412 | * because of lock validation efforts. | |
1da177e4 | 5413 | */ |
5e16bbc2 | 5414 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5415 | { |
5e16bbc2 | 5416 | struct rq *rq = dead_rq; |
48c5ccae | 5417 | struct task_struct *next, *stop = rq->stop; |
e7904a28 | 5418 | struct pin_cookie cookie; |
48c5ccae | 5419 | int dest_cpu; |
1da177e4 LT |
5420 | |
5421 | /* | |
48c5ccae PZ |
5422 | * Fudge the rq selection such that the below task selection loop |
5423 | * doesn't get stuck on the currently eligible stop task. | |
5424 | * | |
5425 | * We're currently inside stop_machine() and the rq is either stuck | |
5426 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5427 | * either way we should never end up calling schedule() until we're | |
5428 | * done here. | |
1da177e4 | 5429 | */ |
48c5ccae | 5430 | rq->stop = NULL; |
48f24c4d | 5431 | |
77bd3970 FW |
5432 | /* |
5433 | * put_prev_task() and pick_next_task() sched | |
5434 | * class method both need to have an up-to-date | |
5435 | * value of rq->clock[_task] | |
5436 | */ | |
5437 | update_rq_clock(rq); | |
5438 | ||
5e16bbc2 | 5439 | for (;;) { |
48c5ccae PZ |
5440 | /* |
5441 | * There's this thread running, bail when that's the only | |
5442 | * remaining thread. | |
5443 | */ | |
5444 | if (rq->nr_running == 1) | |
dd41f596 | 5445 | break; |
48c5ccae | 5446 | |
cbce1a68 | 5447 | /* |
5473e0cc | 5448 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 | 5449 | */ |
e7904a28 PZ |
5450 | cookie = lockdep_pin_lock(&rq->lock); |
5451 | next = pick_next_task(rq, &fake_task, cookie); | |
48c5ccae | 5452 | BUG_ON(!next); |
79c53799 | 5453 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5454 | |
5473e0cc WL |
5455 | /* |
5456 | * Rules for changing task_struct::cpus_allowed are holding | |
5457 | * both pi_lock and rq->lock, such that holding either | |
5458 | * stabilizes the mask. | |
5459 | * | |
5460 | * Drop rq->lock is not quite as disastrous as it usually is | |
5461 | * because !cpu_active at this point, which means load-balance | |
5462 | * will not interfere. Also, stop-machine. | |
5463 | */ | |
e7904a28 | 5464 | lockdep_unpin_lock(&rq->lock, cookie); |
5473e0cc WL |
5465 | raw_spin_unlock(&rq->lock); |
5466 | raw_spin_lock(&next->pi_lock); | |
5467 | raw_spin_lock(&rq->lock); | |
5468 | ||
5469 | /* | |
5470 | * Since we're inside stop-machine, _nothing_ should have | |
5471 | * changed the task, WARN if weird stuff happened, because in | |
5472 | * that case the above rq->lock drop is a fail too. | |
5473 | */ | |
5474 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5475 | raw_spin_unlock(&next->pi_lock); | |
5476 | continue; | |
5477 | } | |
5478 | ||
48c5ccae | 5479 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5480 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5481 | |
5e16bbc2 PZ |
5482 | rq = __migrate_task(rq, next, dest_cpu); |
5483 | if (rq != dead_rq) { | |
5484 | raw_spin_unlock(&rq->lock); | |
5485 | rq = dead_rq; | |
5486 | raw_spin_lock(&rq->lock); | |
5487 | } | |
5473e0cc | 5488 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5489 | } |
dce48a84 | 5490 | |
48c5ccae | 5491 | rq->stop = stop; |
dce48a84 | 5492 | } |
1da177e4 LT |
5493 | #endif /* CONFIG_HOTPLUG_CPU */ |
5494 | ||
1f11eb6a GH |
5495 | static void set_rq_online(struct rq *rq) |
5496 | { | |
5497 | if (!rq->online) { | |
5498 | const struct sched_class *class; | |
5499 | ||
c6c4927b | 5500 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5501 | rq->online = 1; |
5502 | ||
5503 | for_each_class(class) { | |
5504 | if (class->rq_online) | |
5505 | class->rq_online(rq); | |
5506 | } | |
5507 | } | |
5508 | } | |
5509 | ||
5510 | static void set_rq_offline(struct rq *rq) | |
5511 | { | |
5512 | if (rq->online) { | |
5513 | const struct sched_class *class; | |
5514 | ||
5515 | for_each_class(class) { | |
5516 | if (class->rq_offline) | |
5517 | class->rq_offline(rq); | |
5518 | } | |
5519 | ||
c6c4927b | 5520 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5521 | rq->online = 0; |
5522 | } | |
5523 | } | |
5524 | ||
9cf7243d | 5525 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5526 | { |
969c7921 | 5527 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5528 | |
a803f026 CM |
5529 | rq->age_stamp = sched_clock_cpu(cpu); |
5530 | } | |
5531 | ||
4cb98839 PZ |
5532 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5533 | ||
3e9830dc | 5534 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5535 | |
d039ac60 | 5536 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5537 | |
d039ac60 | 5538 | static int __init sched_debug_setup(char *str) |
f6630114 | 5539 | { |
d039ac60 | 5540 | sched_debug_enabled = 1; |
f6630114 MT |
5541 | |
5542 | return 0; | |
5543 | } | |
d039ac60 PZ |
5544 | early_param("sched_debug", sched_debug_setup); |
5545 | ||
5546 | static inline bool sched_debug(void) | |
5547 | { | |
5548 | return sched_debug_enabled; | |
5549 | } | |
f6630114 | 5550 | |
7c16ec58 | 5551 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5552 | struct cpumask *groupmask) |
1da177e4 | 5553 | { |
4dcf6aff | 5554 | struct sched_group *group = sd->groups; |
1da177e4 | 5555 | |
96f874e2 | 5556 | cpumask_clear(groupmask); |
4dcf6aff IM |
5557 | |
5558 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5559 | ||
5560 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5561 | printk("does not load-balance\n"); |
4dcf6aff | 5562 | if (sd->parent) |
3df0fc5b PZ |
5563 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5564 | " has parent"); | |
4dcf6aff | 5565 | return -1; |
41c7ce9a NP |
5566 | } |
5567 | ||
333470ee TH |
5568 | printk(KERN_CONT "span %*pbl level %s\n", |
5569 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5570 | |
758b2cdc | 5571 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5572 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5573 | "CPU%d\n", cpu); | |
4dcf6aff | 5574 | } |
758b2cdc | 5575 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5576 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5577 | " CPU%d\n", cpu); | |
4dcf6aff | 5578 | } |
1da177e4 | 5579 | |
4dcf6aff | 5580 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5581 | do { |
4dcf6aff | 5582 | if (!group) { |
3df0fc5b PZ |
5583 | printk("\n"); |
5584 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5585 | break; |
5586 | } | |
5587 | ||
758b2cdc | 5588 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5589 | printk(KERN_CONT "\n"); |
5590 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5591 | break; |
5592 | } | |
1da177e4 | 5593 | |
cb83b629 PZ |
5594 | if (!(sd->flags & SD_OVERLAP) && |
5595 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5596 | printk(KERN_CONT "\n"); |
5597 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5598 | break; |
5599 | } | |
1da177e4 | 5600 | |
758b2cdc | 5601 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5602 | |
333470ee TH |
5603 | printk(KERN_CONT " %*pbl", |
5604 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5605 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5606 | printk(KERN_CONT " (cpu_capacity = %d)", |
5607 | group->sgc->capacity); | |
381512cf | 5608 | } |
1da177e4 | 5609 | |
4dcf6aff IM |
5610 | group = group->next; |
5611 | } while (group != sd->groups); | |
3df0fc5b | 5612 | printk(KERN_CONT "\n"); |
1da177e4 | 5613 | |
758b2cdc | 5614 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5615 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5616 | |
758b2cdc RR |
5617 | if (sd->parent && |
5618 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5619 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5620 | "of domain->span\n"); | |
4dcf6aff IM |
5621 | return 0; |
5622 | } | |
1da177e4 | 5623 | |
4dcf6aff IM |
5624 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5625 | { | |
5626 | int level = 0; | |
1da177e4 | 5627 | |
d039ac60 | 5628 | if (!sched_debug_enabled) |
f6630114 MT |
5629 | return; |
5630 | ||
4dcf6aff IM |
5631 | if (!sd) { |
5632 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5633 | return; | |
5634 | } | |
1da177e4 | 5635 | |
4dcf6aff IM |
5636 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5637 | ||
5638 | for (;;) { | |
4cb98839 | 5639 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5640 | break; |
1da177e4 LT |
5641 | level++; |
5642 | sd = sd->parent; | |
33859f7f | 5643 | if (!sd) |
4dcf6aff IM |
5644 | break; |
5645 | } | |
1da177e4 | 5646 | } |
6d6bc0ad | 5647 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5648 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5649 | static inline bool sched_debug(void) |
5650 | { | |
5651 | return false; | |
5652 | } | |
6d6bc0ad | 5653 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5654 | |
1a20ff27 | 5655 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5656 | { |
758b2cdc | 5657 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5658 | return 1; |
5659 | ||
5660 | /* Following flags need at least 2 groups */ | |
5661 | if (sd->flags & (SD_LOAD_BALANCE | | |
5662 | SD_BALANCE_NEWIDLE | | |
5663 | SD_BALANCE_FORK | | |
89c4710e | 5664 | SD_BALANCE_EXEC | |
5d4dfddd | 5665 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5666 | SD_SHARE_PKG_RESOURCES | |
5667 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5668 | if (sd->groups != sd->groups->next) |
5669 | return 0; | |
5670 | } | |
5671 | ||
5672 | /* Following flags don't use groups */ | |
c88d5910 | 5673 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5674 | return 0; |
5675 | ||
5676 | return 1; | |
5677 | } | |
5678 | ||
48f24c4d IM |
5679 | static int |
5680 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5681 | { |
5682 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5683 | ||
5684 | if (sd_degenerate(parent)) | |
5685 | return 1; | |
5686 | ||
758b2cdc | 5687 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5688 | return 0; |
5689 | ||
245af2c7 SS |
5690 | /* Flags needing groups don't count if only 1 group in parent */ |
5691 | if (parent->groups == parent->groups->next) { | |
5692 | pflags &= ~(SD_LOAD_BALANCE | | |
5693 | SD_BALANCE_NEWIDLE | | |
5694 | SD_BALANCE_FORK | | |
89c4710e | 5695 | SD_BALANCE_EXEC | |
5d4dfddd | 5696 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5697 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5698 | SD_PREFER_SIBLING | |
5699 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5700 | if (nr_node_ids == 1) |
5701 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5702 | } |
5703 | if (~cflags & pflags) | |
5704 | return 0; | |
5705 | ||
5706 | return 1; | |
5707 | } | |
5708 | ||
dce840a0 | 5709 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5710 | { |
dce840a0 | 5711 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5712 | |
68e74568 | 5713 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5714 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5715 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5716 | free_cpumask_var(rd->rto_mask); |
5717 | free_cpumask_var(rd->online); | |
5718 | free_cpumask_var(rd->span); | |
5719 | kfree(rd); | |
5720 | } | |
5721 | ||
57d885fe GH |
5722 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5723 | { | |
a0490fa3 | 5724 | struct root_domain *old_rd = NULL; |
57d885fe | 5725 | unsigned long flags; |
57d885fe | 5726 | |
05fa785c | 5727 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5728 | |
5729 | if (rq->rd) { | |
a0490fa3 | 5730 | old_rd = rq->rd; |
57d885fe | 5731 | |
c6c4927b | 5732 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5733 | set_rq_offline(rq); |
57d885fe | 5734 | |
c6c4927b | 5735 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5736 | |
a0490fa3 | 5737 | /* |
0515973f | 5738 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5739 | * set old_rd to NULL to skip the freeing later |
5740 | * in this function: | |
5741 | */ | |
5742 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5743 | old_rd = NULL; | |
57d885fe GH |
5744 | } |
5745 | ||
5746 | atomic_inc(&rd->refcount); | |
5747 | rq->rd = rd; | |
5748 | ||
c6c4927b | 5749 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5750 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5751 | set_rq_online(rq); |
57d885fe | 5752 | |
05fa785c | 5753 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5754 | |
5755 | if (old_rd) | |
dce840a0 | 5756 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5757 | } |
5758 | ||
68c38fc3 | 5759 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5760 | { |
5761 | memset(rd, 0, sizeof(*rd)); | |
5762 | ||
8295c699 | 5763 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5764 | goto out; |
8295c699 | 5765 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5766 | goto free_span; |
8295c699 | 5767 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5768 | goto free_online; |
8295c699 | 5769 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5770 | goto free_dlo_mask; |
6e0534f2 | 5771 | |
332ac17e | 5772 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5773 | if (cpudl_init(&rd->cpudl) != 0) |
5774 | goto free_dlo_mask; | |
332ac17e | 5775 | |
68c38fc3 | 5776 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5777 | goto free_rto_mask; |
c6c4927b | 5778 | return 0; |
6e0534f2 | 5779 | |
68e74568 RR |
5780 | free_rto_mask: |
5781 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5782 | free_dlo_mask: |
5783 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5784 | free_online: |
5785 | free_cpumask_var(rd->online); | |
5786 | free_span: | |
5787 | free_cpumask_var(rd->span); | |
0c910d28 | 5788 | out: |
c6c4927b | 5789 | return -ENOMEM; |
57d885fe GH |
5790 | } |
5791 | ||
029632fb PZ |
5792 | /* |
5793 | * By default the system creates a single root-domain with all cpus as | |
5794 | * members (mimicking the global state we have today). | |
5795 | */ | |
5796 | struct root_domain def_root_domain; | |
5797 | ||
57d885fe GH |
5798 | static void init_defrootdomain(void) |
5799 | { | |
68c38fc3 | 5800 | init_rootdomain(&def_root_domain); |
c6c4927b | 5801 | |
57d885fe GH |
5802 | atomic_set(&def_root_domain.refcount, 1); |
5803 | } | |
5804 | ||
dc938520 | 5805 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5806 | { |
5807 | struct root_domain *rd; | |
5808 | ||
5809 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5810 | if (!rd) | |
5811 | return NULL; | |
5812 | ||
68c38fc3 | 5813 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5814 | kfree(rd); |
5815 | return NULL; | |
5816 | } | |
57d885fe GH |
5817 | |
5818 | return rd; | |
5819 | } | |
5820 | ||
63b2ca30 | 5821 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5822 | { |
5823 | struct sched_group *tmp, *first; | |
5824 | ||
5825 | if (!sg) | |
5826 | return; | |
5827 | ||
5828 | first = sg; | |
5829 | do { | |
5830 | tmp = sg->next; | |
5831 | ||
63b2ca30 NP |
5832 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5833 | kfree(sg->sgc); | |
e3589f6c PZ |
5834 | |
5835 | kfree(sg); | |
5836 | sg = tmp; | |
5837 | } while (sg != first); | |
5838 | } | |
5839 | ||
dce840a0 PZ |
5840 | static void free_sched_domain(struct rcu_head *rcu) |
5841 | { | |
5842 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5843 | |
5844 | /* | |
5845 | * If its an overlapping domain it has private groups, iterate and | |
5846 | * nuke them all. | |
5847 | */ | |
5848 | if (sd->flags & SD_OVERLAP) { | |
5849 | free_sched_groups(sd->groups, 1); | |
5850 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5851 | kfree(sd->groups->sgc); |
dce840a0 | 5852 | kfree(sd->groups); |
9c3f75cb | 5853 | } |
dce840a0 PZ |
5854 | kfree(sd); |
5855 | } | |
5856 | ||
5857 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5858 | { | |
5859 | call_rcu(&sd->rcu, free_sched_domain); | |
5860 | } | |
5861 | ||
5862 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5863 | { | |
5864 | for (; sd; sd = sd->parent) | |
5865 | destroy_sched_domain(sd, cpu); | |
5866 | } | |
5867 | ||
518cd623 PZ |
5868 | /* |
5869 | * Keep a special pointer to the highest sched_domain that has | |
5870 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5871 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5872 | * | |
5873 | * Also keep a unique ID per domain (we use the first cpu number in | |
5874 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5875 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5876 | */ |
5877 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5878 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5879 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5880 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5881 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5882 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5883 | |
5884 | static void update_top_cache_domain(int cpu) | |
5885 | { | |
5886 | struct sched_domain *sd; | |
5d4cf996 | 5887 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5888 | int id = cpu; |
7d9ffa89 | 5889 | int size = 1; |
518cd623 PZ |
5890 | |
5891 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5892 | if (sd) { |
518cd623 | 5893 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5894 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5895 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5896 | } |
5d4cf996 | 5897 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5898 | |
5899 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5900 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5901 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5902 | |
5903 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5904 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5905 | |
5906 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5907 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5908 | } |
5909 | ||
1da177e4 | 5910 | /* |
0eab9146 | 5911 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5912 | * hold the hotplug lock. |
5913 | */ | |
0eab9146 IM |
5914 | static void |
5915 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5916 | { |
70b97a7f | 5917 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5918 | struct sched_domain *tmp; |
5919 | ||
5920 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5921 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5922 | struct sched_domain *parent = tmp->parent; |
5923 | if (!parent) | |
5924 | break; | |
f29c9b1c | 5925 | |
1a848870 | 5926 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5927 | tmp->parent = parent->parent; |
1a848870 SS |
5928 | if (parent->parent) |
5929 | parent->parent->child = tmp; | |
10866e62 PZ |
5930 | /* |
5931 | * Transfer SD_PREFER_SIBLING down in case of a | |
5932 | * degenerate parent; the spans match for this | |
5933 | * so the property transfers. | |
5934 | */ | |
5935 | if (parent->flags & SD_PREFER_SIBLING) | |
5936 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5937 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5938 | } else |
5939 | tmp = tmp->parent; | |
245af2c7 SS |
5940 | } |
5941 | ||
1a848870 | 5942 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5943 | tmp = sd; |
245af2c7 | 5944 | sd = sd->parent; |
dce840a0 | 5945 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5946 | if (sd) |
5947 | sd->child = NULL; | |
5948 | } | |
1da177e4 | 5949 | |
4cb98839 | 5950 | sched_domain_debug(sd, cpu); |
1da177e4 | 5951 | |
57d885fe | 5952 | rq_attach_root(rq, rd); |
dce840a0 | 5953 | tmp = rq->sd; |
674311d5 | 5954 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5955 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5956 | |
5957 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5958 | } |
5959 | ||
1da177e4 LT |
5960 | /* Setup the mask of cpus configured for isolated domains */ |
5961 | static int __init isolated_cpu_setup(char *str) | |
5962 | { | |
a6e4491c PB |
5963 | int ret; |
5964 | ||
bdddd296 | 5965 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
5966 | ret = cpulist_parse(str, cpu_isolated_map); |
5967 | if (ret) { | |
5968 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
5969 | return 0; | |
5970 | } | |
1da177e4 LT |
5971 | return 1; |
5972 | } | |
8927f494 | 5973 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5974 | |
49a02c51 | 5975 | struct s_data { |
21d42ccf | 5976 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5977 | struct root_domain *rd; |
5978 | }; | |
5979 | ||
2109b99e | 5980 | enum s_alloc { |
2109b99e | 5981 | sa_rootdomain, |
21d42ccf | 5982 | sa_sd, |
dce840a0 | 5983 | sa_sd_storage, |
2109b99e AH |
5984 | sa_none, |
5985 | }; | |
5986 | ||
c1174876 PZ |
5987 | /* |
5988 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5989 | * domain traversal. | |
5990 | * | |
5991 | * Asymmetric node setups can result in situations where the domain tree is of | |
5992 | * unequal depth, make sure to skip domains that already cover the entire | |
5993 | * range. | |
5994 | * | |
5995 | * In that case build_sched_domains() will have terminated the iteration early | |
5996 | * and our sibling sd spans will be empty. Domains should always include the | |
5997 | * cpu they're built on, so check that. | |
5998 | * | |
5999 | */ | |
6000 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6001 | { | |
6002 | const struct cpumask *span = sched_domain_span(sd); | |
6003 | struct sd_data *sdd = sd->private; | |
6004 | struct sched_domain *sibling; | |
6005 | int i; | |
6006 | ||
6007 | for_each_cpu(i, span) { | |
6008 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6009 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6010 | continue; | |
6011 | ||
6012 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6013 | } | |
6014 | } | |
6015 | ||
6016 | /* | |
6017 | * Return the canonical balance cpu for this group, this is the first cpu | |
6018 | * of this group that's also in the iteration mask. | |
6019 | */ | |
6020 | int group_balance_cpu(struct sched_group *sg) | |
6021 | { | |
6022 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6023 | } | |
6024 | ||
e3589f6c PZ |
6025 | static int |
6026 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6027 | { | |
6028 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6029 | const struct cpumask *span = sched_domain_span(sd); | |
6030 | struct cpumask *covered = sched_domains_tmpmask; | |
6031 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6032 | struct sched_domain *sibling; |
e3589f6c PZ |
6033 | int i; |
6034 | ||
6035 | cpumask_clear(covered); | |
6036 | ||
6037 | for_each_cpu(i, span) { | |
6038 | struct cpumask *sg_span; | |
6039 | ||
6040 | if (cpumask_test_cpu(i, covered)) | |
6041 | continue; | |
6042 | ||
aaecac4a | 6043 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6044 | |
6045 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6046 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6047 | continue; |
6048 | ||
e3589f6c | 6049 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6050 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6051 | |
6052 | if (!sg) | |
6053 | goto fail; | |
6054 | ||
6055 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6056 | if (sibling->child) |
6057 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6058 | else | |
e3589f6c PZ |
6059 | cpumask_set_cpu(i, sg_span); |
6060 | ||
6061 | cpumask_or(covered, covered, sg_span); | |
6062 | ||
63b2ca30 NP |
6063 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6064 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6065 | build_group_mask(sd, sg); |
6066 | ||
c3decf0d | 6067 | /* |
63b2ca30 | 6068 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6069 | * domains and no possible iteration will get us here, we won't |
6070 | * die on a /0 trap. | |
6071 | */ | |
ca8ce3d0 | 6072 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 6073 | |
c1174876 PZ |
6074 | /* |
6075 | * Make sure the first group of this domain contains the | |
6076 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6077 | * breaks. See update_sg_lb_stats(). | |
6078 | */ | |
74a5ce20 | 6079 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6080 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6081 | groups = sg; |
6082 | ||
6083 | if (!first) | |
6084 | first = sg; | |
6085 | if (last) | |
6086 | last->next = sg; | |
6087 | last = sg; | |
6088 | last->next = first; | |
6089 | } | |
6090 | sd->groups = groups; | |
6091 | ||
6092 | return 0; | |
6093 | ||
6094 | fail: | |
6095 | free_sched_groups(first, 0); | |
6096 | ||
6097 | return -ENOMEM; | |
6098 | } | |
6099 | ||
dce840a0 | 6100 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6101 | { |
dce840a0 PZ |
6102 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6103 | struct sched_domain *child = sd->child; | |
1da177e4 | 6104 | |
dce840a0 PZ |
6105 | if (child) |
6106 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6107 | |
9c3f75cb | 6108 | if (sg) { |
dce840a0 | 6109 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6110 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6111 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6112 | } |
dce840a0 PZ |
6113 | |
6114 | return cpu; | |
1e9f28fa | 6115 | } |
1e9f28fa | 6116 | |
01a08546 | 6117 | /* |
dce840a0 PZ |
6118 | * build_sched_groups will build a circular linked list of the groups |
6119 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6120 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6121 | * |
6122 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6123 | */ |
e3589f6c PZ |
6124 | static int |
6125 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6126 | { |
dce840a0 PZ |
6127 | struct sched_group *first = NULL, *last = NULL; |
6128 | struct sd_data *sdd = sd->private; | |
6129 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6130 | struct cpumask *covered; |
dce840a0 | 6131 | int i; |
9c1cfda2 | 6132 | |
e3589f6c PZ |
6133 | get_group(cpu, sdd, &sd->groups); |
6134 | atomic_inc(&sd->groups->ref); | |
6135 | ||
0936629f | 6136 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6137 | return 0; |
6138 | ||
f96225fd PZ |
6139 | lockdep_assert_held(&sched_domains_mutex); |
6140 | covered = sched_domains_tmpmask; | |
6141 | ||
dce840a0 | 6142 | cpumask_clear(covered); |
6711cab4 | 6143 | |
dce840a0 PZ |
6144 | for_each_cpu(i, span) { |
6145 | struct sched_group *sg; | |
cd08e923 | 6146 | int group, j; |
6711cab4 | 6147 | |
dce840a0 PZ |
6148 | if (cpumask_test_cpu(i, covered)) |
6149 | continue; | |
6711cab4 | 6150 | |
cd08e923 | 6151 | group = get_group(i, sdd, &sg); |
c1174876 | 6152 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6153 | |
dce840a0 PZ |
6154 | for_each_cpu(j, span) { |
6155 | if (get_group(j, sdd, NULL) != group) | |
6156 | continue; | |
0601a88d | 6157 | |
dce840a0 PZ |
6158 | cpumask_set_cpu(j, covered); |
6159 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6160 | } | |
0601a88d | 6161 | |
dce840a0 PZ |
6162 | if (!first) |
6163 | first = sg; | |
6164 | if (last) | |
6165 | last->next = sg; | |
6166 | last = sg; | |
6167 | } | |
6168 | last->next = first; | |
e3589f6c PZ |
6169 | |
6170 | return 0; | |
0601a88d | 6171 | } |
51888ca2 | 6172 | |
89c4710e | 6173 | /* |
63b2ca30 | 6174 | * Initialize sched groups cpu_capacity. |
89c4710e | 6175 | * |
63b2ca30 | 6176 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6177 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6178 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6179 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6180 | * group having more cpu_capacity will pickup more load compared to the | |
6181 | * group having less cpu_capacity. | |
89c4710e | 6182 | */ |
63b2ca30 | 6183 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6184 | { |
e3589f6c | 6185 | struct sched_group *sg = sd->groups; |
89c4710e | 6186 | |
94c95ba6 | 6187 | WARN_ON(!sg); |
e3589f6c PZ |
6188 | |
6189 | do { | |
6190 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6191 | sg = sg->next; | |
6192 | } while (sg != sd->groups); | |
89c4710e | 6193 | |
c1174876 | 6194 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6195 | return; |
aae6d3dd | 6196 | |
63b2ca30 NP |
6197 | update_group_capacity(sd, cpu); |
6198 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6199 | } |
6200 | ||
7c16ec58 MT |
6201 | /* |
6202 | * Initializers for schedule domains | |
6203 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6204 | */ | |
6205 | ||
1d3504fc | 6206 | static int default_relax_domain_level = -1; |
60495e77 | 6207 | int sched_domain_level_max; |
1d3504fc HS |
6208 | |
6209 | static int __init setup_relax_domain_level(char *str) | |
6210 | { | |
a841f8ce DS |
6211 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6212 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6213 | |
1d3504fc HS |
6214 | return 1; |
6215 | } | |
6216 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6217 | ||
6218 | static void set_domain_attribute(struct sched_domain *sd, | |
6219 | struct sched_domain_attr *attr) | |
6220 | { | |
6221 | int request; | |
6222 | ||
6223 | if (!attr || attr->relax_domain_level < 0) { | |
6224 | if (default_relax_domain_level < 0) | |
6225 | return; | |
6226 | else | |
6227 | request = default_relax_domain_level; | |
6228 | } else | |
6229 | request = attr->relax_domain_level; | |
6230 | if (request < sd->level) { | |
6231 | /* turn off idle balance on this domain */ | |
c88d5910 | 6232 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6233 | } else { |
6234 | /* turn on idle balance on this domain */ | |
c88d5910 | 6235 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6236 | } |
6237 | } | |
6238 | ||
54ab4ff4 PZ |
6239 | static void __sdt_free(const struct cpumask *cpu_map); |
6240 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6241 | ||
2109b99e AH |
6242 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6243 | const struct cpumask *cpu_map) | |
6244 | { | |
6245 | switch (what) { | |
2109b99e | 6246 | case sa_rootdomain: |
822ff793 PZ |
6247 | if (!atomic_read(&d->rd->refcount)) |
6248 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6249 | case sa_sd: |
6250 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6251 | case sa_sd_storage: |
54ab4ff4 | 6252 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6253 | case sa_none: |
6254 | break; | |
6255 | } | |
6256 | } | |
3404c8d9 | 6257 | |
2109b99e AH |
6258 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6259 | const struct cpumask *cpu_map) | |
6260 | { | |
dce840a0 PZ |
6261 | memset(d, 0, sizeof(*d)); |
6262 | ||
54ab4ff4 PZ |
6263 | if (__sdt_alloc(cpu_map)) |
6264 | return sa_sd_storage; | |
dce840a0 PZ |
6265 | d->sd = alloc_percpu(struct sched_domain *); |
6266 | if (!d->sd) | |
6267 | return sa_sd_storage; | |
2109b99e | 6268 | d->rd = alloc_rootdomain(); |
dce840a0 | 6269 | if (!d->rd) |
21d42ccf | 6270 | return sa_sd; |
2109b99e AH |
6271 | return sa_rootdomain; |
6272 | } | |
57d885fe | 6273 | |
dce840a0 PZ |
6274 | /* |
6275 | * NULL the sd_data elements we've used to build the sched_domain and | |
6276 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6277 | * will not free the data we're using. | |
6278 | */ | |
6279 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6280 | { | |
6281 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6282 | |
6283 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6284 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6285 | ||
e3589f6c | 6286 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6287 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6288 | |
63b2ca30 NP |
6289 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6290 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6291 | } |
6292 | ||
cb83b629 | 6293 | #ifdef CONFIG_NUMA |
cb83b629 | 6294 | static int sched_domains_numa_levels; |
e3fe70b1 | 6295 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6296 | static int *sched_domains_numa_distance; |
9942f79b | 6297 | int sched_max_numa_distance; |
cb83b629 PZ |
6298 | static struct cpumask ***sched_domains_numa_masks; |
6299 | static int sched_domains_curr_level; | |
143e1e28 | 6300 | #endif |
cb83b629 | 6301 | |
143e1e28 VG |
6302 | /* |
6303 | * SD_flags allowed in topology descriptions. | |
6304 | * | |
5d4dfddd | 6305 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6306 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6307 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6308 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6309 | * |
6310 | * Odd one out: | |
6311 | * SD_ASYM_PACKING - describes SMT quirks | |
6312 | */ | |
6313 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6314 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6315 | SD_SHARE_PKG_RESOURCES | \ |
6316 | SD_NUMA | \ | |
d77b3ed5 VG |
6317 | SD_ASYM_PACKING | \ |
6318 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6319 | |
6320 | static struct sched_domain * | |
143e1e28 | 6321 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6322 | { |
6323 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6324 | int sd_weight, sd_flags = 0; |
6325 | ||
6326 | #ifdef CONFIG_NUMA | |
6327 | /* | |
6328 | * Ugly hack to pass state to sd_numa_mask()... | |
6329 | */ | |
6330 | sched_domains_curr_level = tl->numa_level; | |
6331 | #endif | |
6332 | ||
6333 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6334 | ||
6335 | if (tl->sd_flags) | |
6336 | sd_flags = (*tl->sd_flags)(); | |
6337 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6338 | "wrong sd_flags in topology description\n")) | |
6339 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6340 | |
6341 | *sd = (struct sched_domain){ | |
6342 | .min_interval = sd_weight, | |
6343 | .max_interval = 2*sd_weight, | |
6344 | .busy_factor = 32, | |
870a0bb5 | 6345 | .imbalance_pct = 125, |
143e1e28 VG |
6346 | |
6347 | .cache_nice_tries = 0, | |
6348 | .busy_idx = 0, | |
6349 | .idle_idx = 0, | |
cb83b629 PZ |
6350 | .newidle_idx = 0, |
6351 | .wake_idx = 0, | |
6352 | .forkexec_idx = 0, | |
6353 | ||
6354 | .flags = 1*SD_LOAD_BALANCE | |
6355 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6356 | | 1*SD_BALANCE_EXEC |
6357 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6358 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6359 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6360 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6361 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6362 | | 0*SD_SERIALIZE |
cb83b629 | 6363 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6364 | | 0*SD_NUMA |
6365 | | sd_flags | |
cb83b629 | 6366 | , |
143e1e28 | 6367 | |
cb83b629 PZ |
6368 | .last_balance = jiffies, |
6369 | .balance_interval = sd_weight, | |
143e1e28 | 6370 | .smt_gain = 0, |
2b4cfe64 JL |
6371 | .max_newidle_lb_cost = 0, |
6372 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6373 | #ifdef CONFIG_SCHED_DEBUG |
6374 | .name = tl->name, | |
6375 | #endif | |
cb83b629 | 6376 | }; |
cb83b629 PZ |
6377 | |
6378 | /* | |
143e1e28 | 6379 | * Convert topological properties into behaviour. |
cb83b629 | 6380 | */ |
143e1e28 | 6381 | |
5d4dfddd | 6382 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6383 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6384 | sd->imbalance_pct = 110; |
6385 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6386 | |
6387 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6388 | sd->imbalance_pct = 117; | |
6389 | sd->cache_nice_tries = 1; | |
6390 | sd->busy_idx = 2; | |
6391 | ||
6392 | #ifdef CONFIG_NUMA | |
6393 | } else if (sd->flags & SD_NUMA) { | |
6394 | sd->cache_nice_tries = 2; | |
6395 | sd->busy_idx = 3; | |
6396 | sd->idle_idx = 2; | |
6397 | ||
6398 | sd->flags |= SD_SERIALIZE; | |
6399 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6400 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6401 | SD_BALANCE_FORK | | |
6402 | SD_WAKE_AFFINE); | |
6403 | } | |
6404 | ||
6405 | #endif | |
6406 | } else { | |
6407 | sd->flags |= SD_PREFER_SIBLING; | |
6408 | sd->cache_nice_tries = 1; | |
6409 | sd->busy_idx = 2; | |
6410 | sd->idle_idx = 1; | |
6411 | } | |
6412 | ||
6413 | sd->private = &tl->data; | |
cb83b629 PZ |
6414 | |
6415 | return sd; | |
6416 | } | |
6417 | ||
143e1e28 VG |
6418 | /* |
6419 | * Topology list, bottom-up. | |
6420 | */ | |
6421 | static struct sched_domain_topology_level default_topology[] = { | |
6422 | #ifdef CONFIG_SCHED_SMT | |
6423 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6424 | #endif | |
6425 | #ifdef CONFIG_SCHED_MC | |
6426 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6427 | #endif |
6428 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6429 | { NULL, }, | |
6430 | }; | |
6431 | ||
c6e1e7b5 JG |
6432 | static struct sched_domain_topology_level *sched_domain_topology = |
6433 | default_topology; | |
143e1e28 VG |
6434 | |
6435 | #define for_each_sd_topology(tl) \ | |
6436 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6437 | ||
6438 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6439 | { | |
6440 | sched_domain_topology = tl; | |
6441 | } | |
6442 | ||
6443 | #ifdef CONFIG_NUMA | |
6444 | ||
cb83b629 PZ |
6445 | static const struct cpumask *sd_numa_mask(int cpu) |
6446 | { | |
6447 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6448 | } | |
6449 | ||
d039ac60 PZ |
6450 | static void sched_numa_warn(const char *str) |
6451 | { | |
6452 | static int done = false; | |
6453 | int i,j; | |
6454 | ||
6455 | if (done) | |
6456 | return; | |
6457 | ||
6458 | done = true; | |
6459 | ||
6460 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6461 | ||
6462 | for (i = 0; i < nr_node_ids; i++) { | |
6463 | printk(KERN_WARNING " "); | |
6464 | for (j = 0; j < nr_node_ids; j++) | |
6465 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6466 | printk(KERN_CONT "\n"); | |
6467 | } | |
6468 | printk(KERN_WARNING "\n"); | |
6469 | } | |
6470 | ||
9942f79b | 6471 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6472 | { |
6473 | int i; | |
6474 | ||
6475 | if (distance == node_distance(0, 0)) | |
6476 | return true; | |
6477 | ||
6478 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6479 | if (sched_domains_numa_distance[i] == distance) | |
6480 | return true; | |
6481 | } | |
6482 | ||
6483 | return false; | |
6484 | } | |
6485 | ||
e3fe70b1 RR |
6486 | /* |
6487 | * A system can have three types of NUMA topology: | |
6488 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6489 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6490 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6491 | * | |
6492 | * The difference between a glueless mesh topology and a backplane | |
6493 | * topology lies in whether communication between not directly | |
6494 | * connected nodes goes through intermediary nodes (where programs | |
6495 | * could run), or through backplane controllers. This affects | |
6496 | * placement of programs. | |
6497 | * | |
6498 | * The type of topology can be discerned with the following tests: | |
6499 | * - If the maximum distance between any nodes is 1 hop, the system | |
6500 | * is directly connected. | |
6501 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6502 | * there is an intermediary node C, which is < N hops away from both | |
6503 | * nodes A and B, the system is a glueless mesh. | |
6504 | */ | |
6505 | static void init_numa_topology_type(void) | |
6506 | { | |
6507 | int a, b, c, n; | |
6508 | ||
6509 | n = sched_max_numa_distance; | |
6510 | ||
e237882b | 6511 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6512 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6513 | return; |
6514 | } | |
e3fe70b1 RR |
6515 | |
6516 | for_each_online_node(a) { | |
6517 | for_each_online_node(b) { | |
6518 | /* Find two nodes furthest removed from each other. */ | |
6519 | if (node_distance(a, b) < n) | |
6520 | continue; | |
6521 | ||
6522 | /* Is there an intermediary node between a and b? */ | |
6523 | for_each_online_node(c) { | |
6524 | if (node_distance(a, c) < n && | |
6525 | node_distance(b, c) < n) { | |
6526 | sched_numa_topology_type = | |
6527 | NUMA_GLUELESS_MESH; | |
6528 | return; | |
6529 | } | |
6530 | } | |
6531 | ||
6532 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6533 | return; | |
6534 | } | |
6535 | } | |
6536 | } | |
6537 | ||
cb83b629 PZ |
6538 | static void sched_init_numa(void) |
6539 | { | |
6540 | int next_distance, curr_distance = node_distance(0, 0); | |
6541 | struct sched_domain_topology_level *tl; | |
6542 | int level = 0; | |
6543 | int i, j, k; | |
6544 | ||
cb83b629 PZ |
6545 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6546 | if (!sched_domains_numa_distance) | |
6547 | return; | |
6548 | ||
6549 | /* | |
6550 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6551 | * unique distances in the node_distance() table. | |
6552 | * | |
6553 | * Assumes node_distance(0,j) includes all distances in | |
6554 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6555 | */ |
6556 | next_distance = curr_distance; | |
6557 | for (i = 0; i < nr_node_ids; i++) { | |
6558 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6559 | for (k = 0; k < nr_node_ids; k++) { |
6560 | int distance = node_distance(i, k); | |
6561 | ||
6562 | if (distance > curr_distance && | |
6563 | (distance < next_distance || | |
6564 | next_distance == curr_distance)) | |
6565 | next_distance = distance; | |
6566 | ||
6567 | /* | |
6568 | * While not a strong assumption it would be nice to know | |
6569 | * about cases where if node A is connected to B, B is not | |
6570 | * equally connected to A. | |
6571 | */ | |
6572 | if (sched_debug() && node_distance(k, i) != distance) | |
6573 | sched_numa_warn("Node-distance not symmetric"); | |
6574 | ||
6575 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6576 | sched_numa_warn("Node-0 not representative"); | |
6577 | } | |
6578 | if (next_distance != curr_distance) { | |
6579 | sched_domains_numa_distance[level++] = next_distance; | |
6580 | sched_domains_numa_levels = level; | |
6581 | curr_distance = next_distance; | |
6582 | } else break; | |
cb83b629 | 6583 | } |
d039ac60 PZ |
6584 | |
6585 | /* | |
6586 | * In case of sched_debug() we verify the above assumption. | |
6587 | */ | |
6588 | if (!sched_debug()) | |
6589 | break; | |
cb83b629 | 6590 | } |
c123588b AR |
6591 | |
6592 | if (!level) | |
6593 | return; | |
6594 | ||
cb83b629 PZ |
6595 | /* |
6596 | * 'level' contains the number of unique distances, excluding the | |
6597 | * identity distance node_distance(i,i). | |
6598 | * | |
28b4a521 | 6599 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6600 | * numbers. |
6601 | */ | |
6602 | ||
5f7865f3 TC |
6603 | /* |
6604 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6605 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6606 | * the array will contain less then 'level' members. This could be | |
6607 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6608 | * in other functions. | |
6609 | * | |
6610 | * We reset it to 'level' at the end of this function. | |
6611 | */ | |
6612 | sched_domains_numa_levels = 0; | |
6613 | ||
cb83b629 PZ |
6614 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6615 | if (!sched_domains_numa_masks) | |
6616 | return; | |
6617 | ||
6618 | /* | |
6619 | * Now for each level, construct a mask per node which contains all | |
6620 | * cpus of nodes that are that many hops away from us. | |
6621 | */ | |
6622 | for (i = 0; i < level; i++) { | |
6623 | sched_domains_numa_masks[i] = | |
6624 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6625 | if (!sched_domains_numa_masks[i]) | |
6626 | return; | |
6627 | ||
6628 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6629 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6630 | if (!mask) |
6631 | return; | |
6632 | ||
6633 | sched_domains_numa_masks[i][j] = mask; | |
6634 | ||
9c03ee14 | 6635 | for_each_node(k) { |
dd7d8634 | 6636 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6637 | continue; |
6638 | ||
6639 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6640 | } | |
6641 | } | |
6642 | } | |
6643 | ||
143e1e28 VG |
6644 | /* Compute default topology size */ |
6645 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6646 | ||
c515db8c | 6647 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6648 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6649 | if (!tl) | |
6650 | return; | |
6651 | ||
6652 | /* | |
6653 | * Copy the default topology bits.. | |
6654 | */ | |
143e1e28 VG |
6655 | for (i = 0; sched_domain_topology[i].mask; i++) |
6656 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6657 | |
6658 | /* | |
6659 | * .. and append 'j' levels of NUMA goodness. | |
6660 | */ | |
6661 | for (j = 0; j < level; i++, j++) { | |
6662 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6663 | .mask = sd_numa_mask, |
143e1e28 | 6664 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6665 | .flags = SDTL_OVERLAP, |
6666 | .numa_level = j, | |
143e1e28 | 6667 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6668 | }; |
6669 | } | |
6670 | ||
6671 | sched_domain_topology = tl; | |
5f7865f3 TC |
6672 | |
6673 | sched_domains_numa_levels = level; | |
9942f79b | 6674 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6675 | |
6676 | init_numa_topology_type(); | |
cb83b629 | 6677 | } |
301a5cba | 6678 | |
135fb3e1 | 6679 | static void sched_domains_numa_masks_set(unsigned int cpu) |
301a5cba | 6680 | { |
301a5cba | 6681 | int node = cpu_to_node(cpu); |
135fb3e1 | 6682 | int i, j; |
301a5cba TC |
6683 | |
6684 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6685 | for (j = 0; j < nr_node_ids; j++) { | |
6686 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6687 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6688 | } | |
6689 | } | |
6690 | } | |
6691 | ||
135fb3e1 | 6692 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
301a5cba TC |
6693 | { |
6694 | int i, j; | |
135fb3e1 | 6695 | |
301a5cba TC |
6696 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6697 | for (j = 0; j < nr_node_ids; j++) | |
6698 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6699 | } | |
6700 | } | |
6701 | ||
cb83b629 | 6702 | #else |
135fb3e1 TG |
6703 | static inline void sched_init_numa(void) { } |
6704 | static void sched_domains_numa_masks_set(unsigned int cpu) { } | |
6705 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
cb83b629 PZ |
6706 | #endif /* CONFIG_NUMA */ |
6707 | ||
54ab4ff4 PZ |
6708 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6709 | { | |
6710 | struct sched_domain_topology_level *tl; | |
6711 | int j; | |
6712 | ||
27723a68 | 6713 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6714 | struct sd_data *sdd = &tl->data; |
6715 | ||
6716 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6717 | if (!sdd->sd) | |
6718 | return -ENOMEM; | |
6719 | ||
6720 | sdd->sg = alloc_percpu(struct sched_group *); | |
6721 | if (!sdd->sg) | |
6722 | return -ENOMEM; | |
6723 | ||
63b2ca30 NP |
6724 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6725 | if (!sdd->sgc) | |
9c3f75cb PZ |
6726 | return -ENOMEM; |
6727 | ||
54ab4ff4 PZ |
6728 | for_each_cpu(j, cpu_map) { |
6729 | struct sched_domain *sd; | |
6730 | struct sched_group *sg; | |
63b2ca30 | 6731 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6732 | |
5cc389bc | 6733 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6734 | GFP_KERNEL, cpu_to_node(j)); |
6735 | if (!sd) | |
6736 | return -ENOMEM; | |
6737 | ||
6738 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6739 | ||
6740 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6741 | GFP_KERNEL, cpu_to_node(j)); | |
6742 | if (!sg) | |
6743 | return -ENOMEM; | |
6744 | ||
30b4e9eb IM |
6745 | sg->next = sg; |
6746 | ||
54ab4ff4 | 6747 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6748 | |
63b2ca30 | 6749 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6750 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6751 | if (!sgc) |
9c3f75cb PZ |
6752 | return -ENOMEM; |
6753 | ||
63b2ca30 | 6754 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6755 | } |
6756 | } | |
6757 | ||
6758 | return 0; | |
6759 | } | |
6760 | ||
6761 | static void __sdt_free(const struct cpumask *cpu_map) | |
6762 | { | |
6763 | struct sched_domain_topology_level *tl; | |
6764 | int j; | |
6765 | ||
27723a68 | 6766 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6767 | struct sd_data *sdd = &tl->data; |
6768 | ||
6769 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6770 | struct sched_domain *sd; |
6771 | ||
6772 | if (sdd->sd) { | |
6773 | sd = *per_cpu_ptr(sdd->sd, j); | |
6774 | if (sd && (sd->flags & SD_OVERLAP)) | |
6775 | free_sched_groups(sd->groups, 0); | |
6776 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6777 | } | |
6778 | ||
6779 | if (sdd->sg) | |
6780 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6781 | if (sdd->sgc) |
6782 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6783 | } |
6784 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6785 | sdd->sd = NULL; |
54ab4ff4 | 6786 | free_percpu(sdd->sg); |
fb2cf2c6 | 6787 | sdd->sg = NULL; |
63b2ca30 NP |
6788 | free_percpu(sdd->sgc); |
6789 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6790 | } |
6791 | } | |
6792 | ||
2c402dc3 | 6793 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6794 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6795 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6796 | { |
143e1e28 | 6797 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6798 | if (!sd) |
d069b916 | 6799 | return child; |
2c402dc3 | 6800 | |
2c402dc3 | 6801 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6802 | if (child) { |
6803 | sd->level = child->level + 1; | |
6804 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6805 | child->parent = sd; |
c75e0128 | 6806 | sd->child = child; |
6ae72dff PZ |
6807 | |
6808 | if (!cpumask_subset(sched_domain_span(child), | |
6809 | sched_domain_span(sd))) { | |
6810 | pr_err("BUG: arch topology borken\n"); | |
6811 | #ifdef CONFIG_SCHED_DEBUG | |
6812 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6813 | child->name, sd->name); | |
6814 | #endif | |
6815 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6816 | cpumask_or(sched_domain_span(sd), | |
6817 | sched_domain_span(sd), | |
6818 | sched_domain_span(child)); | |
6819 | } | |
6820 | ||
60495e77 | 6821 | } |
a841f8ce | 6822 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6823 | |
6824 | return sd; | |
6825 | } | |
6826 | ||
2109b99e AH |
6827 | /* |
6828 | * Build sched domains for a given set of cpus and attach the sched domains | |
6829 | * to the individual cpus | |
6830 | */ | |
dce840a0 PZ |
6831 | static int build_sched_domains(const struct cpumask *cpu_map, |
6832 | struct sched_domain_attr *attr) | |
2109b99e | 6833 | { |
1c632169 | 6834 | enum s_alloc alloc_state; |
dce840a0 | 6835 | struct sched_domain *sd; |
2109b99e | 6836 | struct s_data d; |
822ff793 | 6837 | int i, ret = -ENOMEM; |
9c1cfda2 | 6838 | |
2109b99e AH |
6839 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6840 | if (alloc_state != sa_rootdomain) | |
6841 | goto error; | |
9c1cfda2 | 6842 | |
dce840a0 | 6843 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6844 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6845 | struct sched_domain_topology_level *tl; |
6846 | ||
3bd65a80 | 6847 | sd = NULL; |
27723a68 | 6848 | for_each_sd_topology(tl) { |
4a850cbe | 6849 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6850 | if (tl == sched_domain_topology) |
6851 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6852 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6853 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6854 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6855 | break; | |
e3589f6c | 6856 | } |
dce840a0 PZ |
6857 | } |
6858 | ||
6859 | /* Build the groups for the domains */ | |
6860 | for_each_cpu(i, cpu_map) { | |
6861 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6862 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6863 | if (sd->flags & SD_OVERLAP) { |
6864 | if (build_overlap_sched_groups(sd, i)) | |
6865 | goto error; | |
6866 | } else { | |
6867 | if (build_sched_groups(sd, i)) | |
6868 | goto error; | |
6869 | } | |
1cf51902 | 6870 | } |
a06dadbe | 6871 | } |
9c1cfda2 | 6872 | |
ced549fa | 6873 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6874 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6875 | if (!cpumask_test_cpu(i, cpu_map)) | |
6876 | continue; | |
9c1cfda2 | 6877 | |
dce840a0 PZ |
6878 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6879 | claim_allocations(i, sd); | |
63b2ca30 | 6880 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6881 | } |
f712c0c7 | 6882 | } |
9c1cfda2 | 6883 | |
1da177e4 | 6884 | /* Attach the domains */ |
dce840a0 | 6885 | rcu_read_lock(); |
abcd083a | 6886 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6887 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6888 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6889 | } |
dce840a0 | 6890 | rcu_read_unlock(); |
51888ca2 | 6891 | |
822ff793 | 6892 | ret = 0; |
51888ca2 | 6893 | error: |
2109b99e | 6894 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6895 | return ret; |
1da177e4 | 6896 | } |
029190c5 | 6897 | |
acc3f5d7 | 6898 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6899 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6900 | static struct sched_domain_attr *dattr_cur; |
6901 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6902 | |
6903 | /* | |
6904 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6905 | * cpumask) fails, then fallback to a single sched domain, |
6906 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6907 | */ |
4212823f | 6908 | static cpumask_var_t fallback_doms; |
029190c5 | 6909 | |
ee79d1bd HC |
6910 | /* |
6911 | * arch_update_cpu_topology lets virtualized architectures update the | |
6912 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6913 | * or 0 if it stayed the same. | |
6914 | */ | |
52f5684c | 6915 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6916 | { |
ee79d1bd | 6917 | return 0; |
22e52b07 HC |
6918 | } |
6919 | ||
acc3f5d7 RR |
6920 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6921 | { | |
6922 | int i; | |
6923 | cpumask_var_t *doms; | |
6924 | ||
6925 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6926 | if (!doms) | |
6927 | return NULL; | |
6928 | for (i = 0; i < ndoms; i++) { | |
6929 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6930 | free_sched_domains(doms, i); | |
6931 | return NULL; | |
6932 | } | |
6933 | } | |
6934 | return doms; | |
6935 | } | |
6936 | ||
6937 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6938 | { | |
6939 | unsigned int i; | |
6940 | for (i = 0; i < ndoms; i++) | |
6941 | free_cpumask_var(doms[i]); | |
6942 | kfree(doms); | |
6943 | } | |
6944 | ||
1a20ff27 | 6945 | /* |
41a2d6cf | 6946 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6947 | * For now this just excludes isolated cpus, but could be used to |
6948 | * exclude other special cases in the future. | |
1a20ff27 | 6949 | */ |
c4a8849a | 6950 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6951 | { |
7378547f MM |
6952 | int err; |
6953 | ||
22e52b07 | 6954 | arch_update_cpu_topology(); |
029190c5 | 6955 | ndoms_cur = 1; |
acc3f5d7 | 6956 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6957 | if (!doms_cur) |
acc3f5d7 RR |
6958 | doms_cur = &fallback_doms; |
6959 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6960 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6961 | register_sched_domain_sysctl(); |
7378547f MM |
6962 | |
6963 | return err; | |
1a20ff27 DG |
6964 | } |
6965 | ||
1a20ff27 DG |
6966 | /* |
6967 | * Detach sched domains from a group of cpus specified in cpu_map | |
6968 | * These cpus will now be attached to the NULL domain | |
6969 | */ | |
96f874e2 | 6970 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6971 | { |
6972 | int i; | |
6973 | ||
dce840a0 | 6974 | rcu_read_lock(); |
abcd083a | 6975 | for_each_cpu(i, cpu_map) |
57d885fe | 6976 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6977 | rcu_read_unlock(); |
1a20ff27 DG |
6978 | } |
6979 | ||
1d3504fc HS |
6980 | /* handle null as "default" */ |
6981 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6982 | struct sched_domain_attr *new, int idx_new) | |
6983 | { | |
6984 | struct sched_domain_attr tmp; | |
6985 | ||
6986 | /* fast path */ | |
6987 | if (!new && !cur) | |
6988 | return 1; | |
6989 | ||
6990 | tmp = SD_ATTR_INIT; | |
6991 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6992 | new ? (new + idx_new) : &tmp, | |
6993 | sizeof(struct sched_domain_attr)); | |
6994 | } | |
6995 | ||
029190c5 PJ |
6996 | /* |
6997 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6998 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6999 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7000 | * It destroys each deleted domain and builds each new domain. | |
7001 | * | |
acc3f5d7 | 7002 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7003 | * The masks don't intersect (don't overlap.) We should setup one |
7004 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7005 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7006 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7007 | * it as it is. | |
7008 | * | |
acc3f5d7 RR |
7009 | * The passed in 'doms_new' should be allocated using |
7010 | * alloc_sched_domains. This routine takes ownership of it and will | |
7011 | * free_sched_domains it when done with it. If the caller failed the | |
7012 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7013 | * and partition_sched_domains() will fallback to the single partition | |
7014 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7015 | * |
96f874e2 | 7016 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7017 | * ndoms_new == 0 is a special case for destroying existing domains, |
7018 | * and it will not create the default domain. | |
dfb512ec | 7019 | * |
029190c5 PJ |
7020 | * Call with hotplug lock held |
7021 | */ | |
acc3f5d7 | 7022 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7023 | struct sched_domain_attr *dattr_new) |
029190c5 | 7024 | { |
dfb512ec | 7025 | int i, j, n; |
d65bd5ec | 7026 | int new_topology; |
029190c5 | 7027 | |
712555ee | 7028 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7029 | |
7378547f MM |
7030 | /* always unregister in case we don't destroy any domains */ |
7031 | unregister_sched_domain_sysctl(); | |
7032 | ||
d65bd5ec HC |
7033 | /* Let architecture update cpu core mappings. */ |
7034 | new_topology = arch_update_cpu_topology(); | |
7035 | ||
dfb512ec | 7036 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7037 | |
7038 | /* Destroy deleted domains */ | |
7039 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7040 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7041 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7042 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7043 | goto match1; |
7044 | } | |
7045 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7046 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7047 | match1: |
7048 | ; | |
7049 | } | |
7050 | ||
c8d2d47a | 7051 | n = ndoms_cur; |
e761b772 | 7052 | if (doms_new == NULL) { |
c8d2d47a | 7053 | n = 0; |
acc3f5d7 | 7054 | doms_new = &fallback_doms; |
6ad4c188 | 7055 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7056 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7057 | } |
7058 | ||
029190c5 PJ |
7059 | /* Build new domains */ |
7060 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7061 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7062 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7063 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7064 | goto match2; |
7065 | } | |
7066 | /* no match - add a new doms_new */ | |
dce840a0 | 7067 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7068 | match2: |
7069 | ; | |
7070 | } | |
7071 | ||
7072 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7073 | if (doms_cur != &fallback_doms) |
7074 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7075 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7076 | doms_cur = doms_new; |
1d3504fc | 7077 | dattr_cur = dattr_new; |
029190c5 | 7078 | ndoms_cur = ndoms_new; |
7378547f MM |
7079 | |
7080 | register_sched_domain_sysctl(); | |
a1835615 | 7081 | |
712555ee | 7082 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7083 | } |
7084 | ||
d35be8ba SB |
7085 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7086 | ||
1da177e4 | 7087 | /* |
3a101d05 TH |
7088 | * Update cpusets according to cpu_active mask. If cpusets are |
7089 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7090 | * around partition_sched_domains(). | |
d35be8ba SB |
7091 | * |
7092 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7093 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7094 | */ |
40190a78 | 7095 | static void cpuset_cpu_active(void) |
e761b772 | 7096 | { |
40190a78 | 7097 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7098 | /* |
7099 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7100 | * resume sequence. As long as this is not the last online | |
7101 | * operation in the resume sequence, just build a single sched | |
7102 | * domain, ignoring cpusets. | |
7103 | */ | |
7104 | num_cpus_frozen--; | |
7105 | if (likely(num_cpus_frozen)) { | |
7106 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 7107 | return; |
d35be8ba | 7108 | } |
d35be8ba SB |
7109 | /* |
7110 | * This is the last CPU online operation. So fall through and | |
7111 | * restore the original sched domains by considering the | |
7112 | * cpuset configurations. | |
7113 | */ | |
3a101d05 | 7114 | } |
135fb3e1 | 7115 | cpuset_update_active_cpus(true); |
3a101d05 | 7116 | } |
e761b772 | 7117 | |
40190a78 | 7118 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7119 | { |
3c18d447 | 7120 | unsigned long flags; |
3c18d447 | 7121 | struct dl_bw *dl_b; |
533445c6 OS |
7122 | bool overflow; |
7123 | int cpus; | |
3c18d447 | 7124 | |
40190a78 | 7125 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
7126 | rcu_read_lock_sched(); |
7127 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7128 | |
533445c6 OS |
7129 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7130 | cpus = dl_bw_cpus(cpu); | |
7131 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7132 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7133 | |
533445c6 | 7134 | rcu_read_unlock_sched(); |
3c18d447 | 7135 | |
533445c6 | 7136 | if (overflow) |
135fb3e1 | 7137 | return -EBUSY; |
7ddf96b0 | 7138 | cpuset_update_active_cpus(false); |
135fb3e1 | 7139 | } else { |
d35be8ba SB |
7140 | num_cpus_frozen++; |
7141 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7142 | } |
135fb3e1 | 7143 | return 0; |
e761b772 | 7144 | } |
e761b772 | 7145 | |
40190a78 | 7146 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7147 | { |
7d976699 TG |
7148 | struct rq *rq = cpu_rq(cpu); |
7149 | unsigned long flags; | |
7150 | ||
40190a78 | 7151 | set_cpu_active(cpu, true); |
135fb3e1 | 7152 | |
40190a78 | 7153 | if (sched_smp_initialized) { |
135fb3e1 | 7154 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7155 | cpuset_cpu_active(); |
e761b772 | 7156 | } |
7d976699 TG |
7157 | |
7158 | /* | |
7159 | * Put the rq online, if not already. This happens: | |
7160 | * | |
7161 | * 1) In the early boot process, because we build the real domains | |
7162 | * after all cpus have been brought up. | |
7163 | * | |
7164 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7165 | * domains. | |
7166 | */ | |
7167 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7168 | if (rq->rd) { | |
7169 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7170 | set_rq_online(rq); | |
7171 | } | |
7172 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7173 | ||
7174 | update_max_interval(); | |
7175 | ||
40190a78 | 7176 | return 0; |
135fb3e1 TG |
7177 | } |
7178 | ||
40190a78 | 7179 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7180 | { |
135fb3e1 TG |
7181 | int ret; |
7182 | ||
40190a78 | 7183 | set_cpu_active(cpu, false); |
b2454caa PZ |
7184 | /* |
7185 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7186 | * users of this state to go away such that all new such users will | |
7187 | * observe it. | |
7188 | * | |
7189 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
7190 | * not imply sync_sched(), so wait for both. | |
7191 | * | |
7192 | * Do sync before park smpboot threads to take care the rcu boost case. | |
7193 | */ | |
7194 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
7195 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
7196 | else | |
7197 | synchronize_rcu(); | |
40190a78 TG |
7198 | |
7199 | if (!sched_smp_initialized) | |
7200 | return 0; | |
7201 | ||
7202 | ret = cpuset_cpu_inactive(cpu); | |
7203 | if (ret) { | |
7204 | set_cpu_active(cpu, true); | |
7205 | return ret; | |
135fb3e1 | 7206 | } |
40190a78 TG |
7207 | sched_domains_numa_masks_clear(cpu); |
7208 | return 0; | |
135fb3e1 TG |
7209 | } |
7210 | ||
94baf7a5 TG |
7211 | static void sched_rq_cpu_starting(unsigned int cpu) |
7212 | { | |
7213 | struct rq *rq = cpu_rq(cpu); | |
7214 | ||
7215 | rq->calc_load_update = calc_load_update; | |
7216 | account_reset_rq(rq); | |
7217 | update_max_interval(); | |
7218 | } | |
7219 | ||
135fb3e1 TG |
7220 | int sched_cpu_starting(unsigned int cpu) |
7221 | { | |
7222 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 7223 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 7224 | return 0; |
e761b772 | 7225 | } |
e761b772 | 7226 | |
f2785ddb TG |
7227 | #ifdef CONFIG_HOTPLUG_CPU |
7228 | int sched_cpu_dying(unsigned int cpu) | |
7229 | { | |
7230 | struct rq *rq = cpu_rq(cpu); | |
7231 | unsigned long flags; | |
7232 | ||
7233 | /* Handle pending wakeups and then migrate everything off */ | |
7234 | sched_ttwu_pending(); | |
7235 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7236 | if (rq->rd) { | |
7237 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7238 | set_rq_offline(rq); | |
7239 | } | |
7240 | migrate_tasks(rq); | |
7241 | BUG_ON(rq->nr_running != 1); | |
7242 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7243 | calc_load_migrate(rq); | |
7244 | update_max_interval(); | |
20a5c8cc | 7245 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 7246 | hrtick_clear(rq); |
f2785ddb TG |
7247 | return 0; |
7248 | } | |
7249 | #endif | |
7250 | ||
1da177e4 LT |
7251 | void __init sched_init_smp(void) |
7252 | { | |
dcc30a35 RR |
7253 | cpumask_var_t non_isolated_cpus; |
7254 | ||
7255 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7256 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7257 | |
cb83b629 PZ |
7258 | sched_init_numa(); |
7259 | ||
6acce3ef PZ |
7260 | /* |
7261 | * There's no userspace yet to cause hotplug operations; hence all the | |
7262 | * cpu masks are stable and all blatant races in the below code cannot | |
7263 | * happen. | |
7264 | */ | |
712555ee | 7265 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7266 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7267 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7268 | if (cpumask_empty(non_isolated_cpus)) | |
7269 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7270 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7271 | |
5c1e1767 | 7272 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 7273 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7274 | BUG(); |
19978ca6 | 7275 | sched_init_granularity(); |
dcc30a35 | 7276 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7277 | |
0e3900e6 | 7278 | init_sched_rt_class(); |
1baca4ce | 7279 | init_sched_dl_class(); |
e26fbffd | 7280 | sched_smp_initialized = true; |
1da177e4 | 7281 | } |
e26fbffd TG |
7282 | |
7283 | static int __init migration_init(void) | |
7284 | { | |
94baf7a5 | 7285 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 7286 | return 0; |
1da177e4 | 7287 | } |
e26fbffd TG |
7288 | early_initcall(migration_init); |
7289 | ||
1da177e4 LT |
7290 | #else |
7291 | void __init sched_init_smp(void) | |
7292 | { | |
19978ca6 | 7293 | sched_init_granularity(); |
1da177e4 LT |
7294 | } |
7295 | #endif /* CONFIG_SMP */ | |
7296 | ||
7297 | int in_sched_functions(unsigned long addr) | |
7298 | { | |
1da177e4 LT |
7299 | return in_lock_functions(addr) || |
7300 | (addr >= (unsigned long)__sched_text_start | |
7301 | && addr < (unsigned long)__sched_text_end); | |
7302 | } | |
7303 | ||
029632fb | 7304 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7305 | /* |
7306 | * Default task group. | |
7307 | * Every task in system belongs to this group at bootup. | |
7308 | */ | |
029632fb | 7309 | struct task_group root_task_group; |
35cf4e50 | 7310 | LIST_HEAD(task_groups); |
b0367629 WL |
7311 | |
7312 | /* Cacheline aligned slab cache for task_group */ | |
7313 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7314 | #endif |
6f505b16 | 7315 | |
e6252c3e | 7316 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7317 | |
1da177e4 LT |
7318 | void __init sched_init(void) |
7319 | { | |
dd41f596 | 7320 | int i, j; |
434d53b0 MT |
7321 | unsigned long alloc_size = 0, ptr; |
7322 | ||
7323 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7324 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7325 | #endif | |
7326 | #ifdef CONFIG_RT_GROUP_SCHED | |
7327 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7328 | #endif | |
434d53b0 | 7329 | if (alloc_size) { |
36b7b6d4 | 7330 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7331 | |
7332 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7333 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7334 | ptr += nr_cpu_ids * sizeof(void **); |
7335 | ||
07e06b01 | 7336 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7337 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7338 | |
6d6bc0ad | 7339 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7340 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7341 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7342 | ptr += nr_cpu_ids * sizeof(void **); |
7343 | ||
07e06b01 | 7344 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7345 | ptr += nr_cpu_ids * sizeof(void **); |
7346 | ||
6d6bc0ad | 7347 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7348 | } |
df7c8e84 | 7349 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7350 | for_each_possible_cpu(i) { |
7351 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7352 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7353 | } |
b74e6278 | 7354 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7355 | |
332ac17e DF |
7356 | init_rt_bandwidth(&def_rt_bandwidth, |
7357 | global_rt_period(), global_rt_runtime()); | |
7358 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7359 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7360 | |
57d885fe GH |
7361 | #ifdef CONFIG_SMP |
7362 | init_defrootdomain(); | |
7363 | #endif | |
7364 | ||
d0b27fa7 | 7365 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7366 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7367 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7368 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7369 | |
7c941438 | 7370 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7371 | task_group_cache = KMEM_CACHE(task_group, 0); |
7372 | ||
07e06b01 YZ |
7373 | list_add(&root_task_group.list, &task_groups); |
7374 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7375 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7376 | autogroup_init(&init_task); |
7c941438 | 7377 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7378 | |
0a945022 | 7379 | for_each_possible_cpu(i) { |
70b97a7f | 7380 | struct rq *rq; |
1da177e4 LT |
7381 | |
7382 | rq = cpu_rq(i); | |
05fa785c | 7383 | raw_spin_lock_init(&rq->lock); |
7897986b | 7384 | rq->nr_running = 0; |
dce48a84 TG |
7385 | rq->calc_load_active = 0; |
7386 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7387 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7388 | init_rt_rq(&rq->rt); |
7389 | init_dl_rq(&rq->dl); | |
dd41f596 | 7390 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7391 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7392 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7393 | /* |
07e06b01 | 7394 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7395 | * |
7396 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7397 | * gets 100% of the cpu resources in the system. This overall | |
7398 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7399 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7400 | * based on each entity's (task or task-group's) weight |
7401 | * (se->load.weight). | |
7402 | * | |
07e06b01 | 7403 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7404 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7405 | * then A0's share of the cpu resource is: | |
7406 | * | |
0d905bca | 7407 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7408 | * |
07e06b01 YZ |
7409 | * We achieve this by letting root_task_group's tasks sit |
7410 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7411 | */ |
ab84d31e | 7412 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7413 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7414 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7415 | ||
7416 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7417 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7418 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7419 | #endif |
1da177e4 | 7420 | |
dd41f596 IM |
7421 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7422 | rq->cpu_load[j] = 0; | |
fdf3e95d | 7423 | |
1da177e4 | 7424 | #ifdef CONFIG_SMP |
41c7ce9a | 7425 | rq->sd = NULL; |
57d885fe | 7426 | rq->rd = NULL; |
ca6d75e6 | 7427 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7428 | rq->balance_callback = NULL; |
1da177e4 | 7429 | rq->active_balance = 0; |
dd41f596 | 7430 | rq->next_balance = jiffies; |
1da177e4 | 7431 | rq->push_cpu = 0; |
0a2966b4 | 7432 | rq->cpu = i; |
1f11eb6a | 7433 | rq->online = 0; |
eae0c9df MG |
7434 | rq->idle_stamp = 0; |
7435 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7436 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7437 | |
7438 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7439 | ||
dc938520 | 7440 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7441 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 7442 | rq->last_load_update_tick = jiffies; |
1c792db7 | 7443 | rq->nohz_flags = 0; |
83cd4fe2 | 7444 | #endif |
265f22a9 FW |
7445 | #ifdef CONFIG_NO_HZ_FULL |
7446 | rq->last_sched_tick = 0; | |
7447 | #endif | |
9fd81dd5 | 7448 | #endif /* CONFIG_SMP */ |
8f4d37ec | 7449 | init_rq_hrtick(rq); |
1da177e4 | 7450 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7451 | } |
7452 | ||
2dd73a4f | 7453 | set_load_weight(&init_task); |
b50f60ce | 7454 | |
e107be36 AK |
7455 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7456 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7457 | #endif | |
7458 | ||
1da177e4 LT |
7459 | /* |
7460 | * The boot idle thread does lazy MMU switching as well: | |
7461 | */ | |
7462 | atomic_inc(&init_mm.mm_count); | |
7463 | enter_lazy_tlb(&init_mm, current); | |
7464 | ||
1b537c7d YD |
7465 | /* |
7466 | * During early bootup we pretend to be a normal task: | |
7467 | */ | |
7468 | current->sched_class = &fair_sched_class; | |
7469 | ||
1da177e4 LT |
7470 | /* |
7471 | * Make us the idle thread. Technically, schedule() should not be | |
7472 | * called from this thread, however somewhere below it might be, | |
7473 | * but because we are the idle thread, we just pick up running again | |
7474 | * when this runqueue becomes "idle". | |
7475 | */ | |
7476 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7477 | |
7478 | calc_load_update = jiffies + LOAD_FREQ; | |
7479 | ||
bf4d83f6 | 7480 | #ifdef CONFIG_SMP |
4cb98839 | 7481 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7482 | /* May be allocated at isolcpus cmdline parse time */ |
7483 | if (cpu_isolated_map == NULL) | |
7484 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7485 | idle_thread_set_boot_cpu(); |
9cf7243d | 7486 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
7487 | #endif |
7488 | init_sched_fair_class(); | |
6a7b3dc3 | 7489 | |
6892b75e | 7490 | scheduler_running = 1; |
1da177e4 LT |
7491 | } |
7492 | ||
d902db1e | 7493 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7494 | static inline int preempt_count_equals(int preempt_offset) |
7495 | { | |
da7142e2 | 7496 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7497 | |
4ba8216c | 7498 | return (nested == preempt_offset); |
e4aafea2 FW |
7499 | } |
7500 | ||
d894837f | 7501 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7502 | { |
8eb23b9f PZ |
7503 | /* |
7504 | * Blocking primitives will set (and therefore destroy) current->state, | |
7505 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7506 | * otherwise we will destroy state. | |
7507 | */ | |
00845eb9 | 7508 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7509 | "do not call blocking ops when !TASK_RUNNING; " |
7510 | "state=%lx set at [<%p>] %pS\n", | |
7511 | current->state, | |
7512 | (void *)current->task_state_change, | |
00845eb9 | 7513 | (void *)current->task_state_change); |
8eb23b9f | 7514 | |
3427445a PZ |
7515 | ___might_sleep(file, line, preempt_offset); |
7516 | } | |
7517 | EXPORT_SYMBOL(__might_sleep); | |
7518 | ||
7519 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7520 | { |
1da177e4 LT |
7521 | static unsigned long prev_jiffy; /* ratelimiting */ |
7522 | ||
b3fbab05 | 7523 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7524 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7525 | !is_idle_task(current)) || | |
e4aafea2 | 7526 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7527 | return; |
7528 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7529 | return; | |
7530 | prev_jiffy = jiffies; | |
7531 | ||
3df0fc5b PZ |
7532 | printk(KERN_ERR |
7533 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7534 | file, line); | |
7535 | printk(KERN_ERR | |
7536 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7537 | in_atomic(), irqs_disabled(), | |
7538 | current->pid, current->comm); | |
aef745fc | 7539 | |
a8b686b3 ES |
7540 | if (task_stack_end_corrupted(current)) |
7541 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7542 | ||
aef745fc IM |
7543 | debug_show_held_locks(current); |
7544 | if (irqs_disabled()) | |
7545 | print_irqtrace_events(current); | |
8f47b187 TG |
7546 | #ifdef CONFIG_DEBUG_PREEMPT |
7547 | if (!preempt_count_equals(preempt_offset)) { | |
7548 | pr_err("Preemption disabled at:"); | |
7549 | print_ip_sym(current->preempt_disable_ip); | |
7550 | pr_cont("\n"); | |
7551 | } | |
7552 | #endif | |
aef745fc | 7553 | dump_stack(); |
1da177e4 | 7554 | } |
3427445a | 7555 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7556 | #endif |
7557 | ||
7558 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7559 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7560 | { |
dbc7f069 | 7561 | struct task_struct *g, *p; |
d50dde5a DF |
7562 | struct sched_attr attr = { |
7563 | .sched_policy = SCHED_NORMAL, | |
7564 | }; | |
1da177e4 | 7565 | |
3472eaa1 | 7566 | read_lock(&tasklist_lock); |
5d07f420 | 7567 | for_each_process_thread(g, p) { |
178be793 IM |
7568 | /* |
7569 | * Only normalize user tasks: | |
7570 | */ | |
3472eaa1 | 7571 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7572 | continue; |
7573 | ||
6cfb0d5d | 7574 | p->se.exec_start = 0; |
6cfb0d5d | 7575 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7576 | p->se.statistics.wait_start = 0; |
7577 | p->se.statistics.sleep_start = 0; | |
7578 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7579 | #endif |
dd41f596 | 7580 | |
aab03e05 | 7581 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7582 | /* |
7583 | * Renice negative nice level userspace | |
7584 | * tasks back to 0: | |
7585 | */ | |
3472eaa1 | 7586 | if (task_nice(p) < 0) |
dd41f596 | 7587 | set_user_nice(p, 0); |
1da177e4 | 7588 | continue; |
dd41f596 | 7589 | } |
1da177e4 | 7590 | |
dbc7f069 | 7591 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7592 | } |
3472eaa1 | 7593 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7594 | } |
7595 | ||
7596 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7597 | |
67fc4e0c | 7598 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7599 | /* |
67fc4e0c | 7600 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7601 | * |
7602 | * They can only be called when the whole system has been | |
7603 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7604 | * activity can take place. Using them for anything else would | |
7605 | * be a serious bug, and as a result, they aren't even visible | |
7606 | * under any other configuration. | |
7607 | */ | |
7608 | ||
7609 | /** | |
7610 | * curr_task - return the current task for a given cpu. | |
7611 | * @cpu: the processor in question. | |
7612 | * | |
7613 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7614 | * |
7615 | * Return: The current task for @cpu. | |
1df5c10a | 7616 | */ |
36c8b586 | 7617 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7618 | { |
7619 | return cpu_curr(cpu); | |
7620 | } | |
7621 | ||
67fc4e0c JW |
7622 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7623 | ||
7624 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7625 | /** |
7626 | * set_curr_task - set the current task for a given cpu. | |
7627 | * @cpu: the processor in question. | |
7628 | * @p: the task pointer to set. | |
7629 | * | |
7630 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7631 | * are serviced on a separate stack. It allows the architecture to switch the |
7632 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7633 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7634 | * and caller must save the original value of the current task (see | |
7635 | * curr_task() above) and restore that value before reenabling interrupts and | |
7636 | * re-starting the system. | |
7637 | * | |
7638 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7639 | */ | |
36c8b586 | 7640 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7641 | { |
7642 | cpu_curr(cpu) = p; | |
7643 | } | |
7644 | ||
7645 | #endif | |
29f59db3 | 7646 | |
7c941438 | 7647 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7648 | /* task_group_lock serializes the addition/removal of task groups */ |
7649 | static DEFINE_SPINLOCK(task_group_lock); | |
7650 | ||
2f5177f0 | 7651 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7652 | { |
7653 | free_fair_sched_group(tg); | |
7654 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7655 | autogroup_free(tg); |
b0367629 | 7656 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7657 | } |
7658 | ||
7659 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7660 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7661 | { |
7662 | struct task_group *tg; | |
bccbe08a | 7663 | |
b0367629 | 7664 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7665 | if (!tg) |
7666 | return ERR_PTR(-ENOMEM); | |
7667 | ||
ec7dc8ac | 7668 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7669 | goto err; |
7670 | ||
ec7dc8ac | 7671 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7672 | goto err; |
7673 | ||
ace783b9 LZ |
7674 | return tg; |
7675 | ||
7676 | err: | |
2f5177f0 | 7677 | sched_free_group(tg); |
ace783b9 LZ |
7678 | return ERR_PTR(-ENOMEM); |
7679 | } | |
7680 | ||
7681 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7682 | { | |
7683 | unsigned long flags; | |
7684 | ||
8ed36996 | 7685 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7686 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7687 | |
7688 | WARN_ON(!parent); /* root should already exist */ | |
7689 | ||
7690 | tg->parent = parent; | |
f473aa5e | 7691 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7692 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7693 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7694 | } |
7695 | ||
9b5b7751 | 7696 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7697 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7698 | { |
29f59db3 | 7699 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 7700 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7701 | } |
7702 | ||
4cf86d77 | 7703 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7704 | { |
7705 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 7706 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7707 | } |
7708 | ||
7709 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7710 | { |
8ed36996 | 7711 | unsigned long flags; |
29f59db3 | 7712 | |
3d4b47b4 | 7713 | /* end participation in shares distribution */ |
6fe1f348 | 7714 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7715 | |
7716 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7717 | list_del_rcu(&tg->list); |
f473aa5e | 7718 | list_del_rcu(&tg->siblings); |
8ed36996 | 7719 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7720 | } |
7721 | ||
9b5b7751 | 7722 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7723 | * The caller of this function should have put the task in its new group |
7724 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7725 | * reflect its new group. | |
9b5b7751 SV |
7726 | */ |
7727 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7728 | { |
8323f26c | 7729 | struct task_group *tg; |
da0c1e65 | 7730 | int queued, running; |
eb580751 | 7731 | struct rq_flags rf; |
29f59db3 SV |
7732 | struct rq *rq; |
7733 | ||
eb580751 | 7734 | rq = task_rq_lock(tsk, &rf); |
29f59db3 | 7735 | |
051a1d1a | 7736 | running = task_current(rq, tsk); |
da0c1e65 | 7737 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7738 | |
da0c1e65 | 7739 | if (queued) |
ff77e468 | 7740 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); |
0e1f3483 | 7741 | if (unlikely(running)) |
f3cd1c4e | 7742 | put_prev_task(rq, tsk); |
29f59db3 | 7743 | |
f7b8a47d KT |
7744 | /* |
7745 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7746 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7747 | * to prevent lockdep warnings. | |
7748 | */ | |
7749 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7750 | struct task_group, css); |
7751 | tg = autogroup_task_group(tsk, tg); | |
7752 | tsk->sched_task_group = tg; | |
7753 | ||
810b3817 | 7754 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7755 | if (tsk->sched_class->task_move_group) |
bc54da21 | 7756 | tsk->sched_class->task_move_group(tsk); |
b2b5ce02 | 7757 | else |
810b3817 | 7758 | #endif |
b2b5ce02 | 7759 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7760 | |
0e1f3483 HS |
7761 | if (unlikely(running)) |
7762 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7763 | if (queued) |
ff77e468 | 7764 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
29f59db3 | 7765 | |
eb580751 | 7766 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 7767 | } |
7c941438 | 7768 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7769 | |
a790de99 PT |
7770 | #ifdef CONFIG_RT_GROUP_SCHED |
7771 | /* | |
7772 | * Ensure that the real time constraints are schedulable. | |
7773 | */ | |
7774 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7775 | |
9a7e0b18 PZ |
7776 | /* Must be called with tasklist_lock held */ |
7777 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7778 | { |
9a7e0b18 | 7779 | struct task_struct *g, *p; |
b40b2e8e | 7780 | |
1fe89e1b PZ |
7781 | /* |
7782 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7783 | */ | |
7784 | if (task_group_is_autogroup(tg)) | |
7785 | return 0; | |
7786 | ||
5d07f420 | 7787 | for_each_process_thread(g, p) { |
8651c658 | 7788 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7789 | return 1; |
5d07f420 | 7790 | } |
b40b2e8e | 7791 | |
9a7e0b18 PZ |
7792 | return 0; |
7793 | } | |
b40b2e8e | 7794 | |
9a7e0b18 PZ |
7795 | struct rt_schedulable_data { |
7796 | struct task_group *tg; | |
7797 | u64 rt_period; | |
7798 | u64 rt_runtime; | |
7799 | }; | |
b40b2e8e | 7800 | |
a790de99 | 7801 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7802 | { |
7803 | struct rt_schedulable_data *d = data; | |
7804 | struct task_group *child; | |
7805 | unsigned long total, sum = 0; | |
7806 | u64 period, runtime; | |
b40b2e8e | 7807 | |
9a7e0b18 PZ |
7808 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7809 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7810 | |
9a7e0b18 PZ |
7811 | if (tg == d->tg) { |
7812 | period = d->rt_period; | |
7813 | runtime = d->rt_runtime; | |
b40b2e8e | 7814 | } |
b40b2e8e | 7815 | |
4653f803 PZ |
7816 | /* |
7817 | * Cannot have more runtime than the period. | |
7818 | */ | |
7819 | if (runtime > period && runtime != RUNTIME_INF) | |
7820 | return -EINVAL; | |
6f505b16 | 7821 | |
4653f803 PZ |
7822 | /* |
7823 | * Ensure we don't starve existing RT tasks. | |
7824 | */ | |
9a7e0b18 PZ |
7825 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7826 | return -EBUSY; | |
6f505b16 | 7827 | |
9a7e0b18 | 7828 | total = to_ratio(period, runtime); |
6f505b16 | 7829 | |
4653f803 PZ |
7830 | /* |
7831 | * Nobody can have more than the global setting allows. | |
7832 | */ | |
7833 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7834 | return -EINVAL; | |
6f505b16 | 7835 | |
4653f803 PZ |
7836 | /* |
7837 | * The sum of our children's runtime should not exceed our own. | |
7838 | */ | |
9a7e0b18 PZ |
7839 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7840 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7841 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7842 | |
9a7e0b18 PZ |
7843 | if (child == d->tg) { |
7844 | period = d->rt_period; | |
7845 | runtime = d->rt_runtime; | |
7846 | } | |
6f505b16 | 7847 | |
9a7e0b18 | 7848 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7849 | } |
6f505b16 | 7850 | |
9a7e0b18 PZ |
7851 | if (sum > total) |
7852 | return -EINVAL; | |
7853 | ||
7854 | return 0; | |
6f505b16 PZ |
7855 | } |
7856 | ||
9a7e0b18 | 7857 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7858 | { |
8277434e PT |
7859 | int ret; |
7860 | ||
9a7e0b18 PZ |
7861 | struct rt_schedulable_data data = { |
7862 | .tg = tg, | |
7863 | .rt_period = period, | |
7864 | .rt_runtime = runtime, | |
7865 | }; | |
7866 | ||
8277434e PT |
7867 | rcu_read_lock(); |
7868 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7869 | rcu_read_unlock(); | |
7870 | ||
7871 | return ret; | |
521f1a24 DG |
7872 | } |
7873 | ||
ab84d31e | 7874 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7875 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7876 | { |
ac086bc2 | 7877 | int i, err = 0; |
9f0c1e56 | 7878 | |
2636ed5f PZ |
7879 | /* |
7880 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7881 | * kernel creating (and or operating) RT threads. | |
7882 | */ | |
7883 | if (tg == &root_task_group && rt_runtime == 0) | |
7884 | return -EINVAL; | |
7885 | ||
7886 | /* No period doesn't make any sense. */ | |
7887 | if (rt_period == 0) | |
7888 | return -EINVAL; | |
7889 | ||
9f0c1e56 | 7890 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7891 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7892 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7893 | if (err) | |
9f0c1e56 | 7894 | goto unlock; |
ac086bc2 | 7895 | |
0986b11b | 7896 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7897 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7898 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7899 | |
7900 | for_each_possible_cpu(i) { | |
7901 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7902 | ||
0986b11b | 7903 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7904 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7905 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7906 | } |
0986b11b | 7907 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7908 | unlock: |
521f1a24 | 7909 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7910 | mutex_unlock(&rt_constraints_mutex); |
7911 | ||
7912 | return err; | |
6f505b16 PZ |
7913 | } |
7914 | ||
25cc7da7 | 7915 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7916 | { |
7917 | u64 rt_runtime, rt_period; | |
7918 | ||
7919 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7920 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7921 | if (rt_runtime_us < 0) | |
7922 | rt_runtime = RUNTIME_INF; | |
7923 | ||
ab84d31e | 7924 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7925 | } |
7926 | ||
25cc7da7 | 7927 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7928 | { |
7929 | u64 rt_runtime_us; | |
7930 | ||
d0b27fa7 | 7931 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7932 | return -1; |
7933 | ||
d0b27fa7 | 7934 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7935 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7936 | return rt_runtime_us; | |
7937 | } | |
d0b27fa7 | 7938 | |
ce2f5fe4 | 7939 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7940 | { |
7941 | u64 rt_runtime, rt_period; | |
7942 | ||
ce2f5fe4 | 7943 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7944 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7945 | ||
ab84d31e | 7946 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7947 | } |
7948 | ||
25cc7da7 | 7949 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7950 | { |
7951 | u64 rt_period_us; | |
7952 | ||
7953 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7954 | do_div(rt_period_us, NSEC_PER_USEC); | |
7955 | return rt_period_us; | |
7956 | } | |
332ac17e | 7957 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7958 | |
332ac17e | 7959 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7960 | static int sched_rt_global_constraints(void) |
7961 | { | |
7962 | int ret = 0; | |
7963 | ||
7964 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7965 | read_lock(&tasklist_lock); |
4653f803 | 7966 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7967 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7968 | mutex_unlock(&rt_constraints_mutex); |
7969 | ||
7970 | return ret; | |
7971 | } | |
54e99124 | 7972 | |
25cc7da7 | 7973 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7974 | { |
7975 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7976 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7977 | return 0; | |
7978 | ||
7979 | return 1; | |
7980 | } | |
7981 | ||
6d6bc0ad | 7982 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7983 | static int sched_rt_global_constraints(void) |
7984 | { | |
ac086bc2 | 7985 | unsigned long flags; |
8c5e9554 | 7986 | int i; |
ec5d4989 | 7987 | |
0986b11b | 7988 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7989 | for_each_possible_cpu(i) { |
7990 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7991 | ||
0986b11b | 7992 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7993 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7994 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7995 | } |
0986b11b | 7996 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7997 | |
8c5e9554 | 7998 | return 0; |
d0b27fa7 | 7999 | } |
6d6bc0ad | 8000 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8001 | |
a1963b81 | 8002 | static int sched_dl_global_validate(void) |
332ac17e | 8003 | { |
1724813d PZ |
8004 | u64 runtime = global_rt_runtime(); |
8005 | u64 period = global_rt_period(); | |
332ac17e | 8006 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 8007 | struct dl_bw *dl_b; |
1724813d | 8008 | int cpu, ret = 0; |
49516342 | 8009 | unsigned long flags; |
332ac17e DF |
8010 | |
8011 | /* | |
8012 | * Here we want to check the bandwidth not being set to some | |
8013 | * value smaller than the currently allocated bandwidth in | |
8014 | * any of the root_domains. | |
8015 | * | |
8016 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
8017 | * cycling on root_domains... Discussion on different/better | |
8018 | * solutions is welcome! | |
8019 | */ | |
1724813d | 8020 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
8021 | rcu_read_lock_sched(); |
8022 | dl_b = dl_bw_of(cpu); | |
332ac17e | 8023 | |
49516342 | 8024 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
8025 | if (new_bw < dl_b->total_bw) |
8026 | ret = -EBUSY; | |
49516342 | 8027 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 8028 | |
f10e00f4 KT |
8029 | rcu_read_unlock_sched(); |
8030 | ||
1724813d PZ |
8031 | if (ret) |
8032 | break; | |
332ac17e DF |
8033 | } |
8034 | ||
1724813d | 8035 | return ret; |
332ac17e DF |
8036 | } |
8037 | ||
1724813d | 8038 | static void sched_dl_do_global(void) |
ce0dbbbb | 8039 | { |
1724813d | 8040 | u64 new_bw = -1; |
f10e00f4 | 8041 | struct dl_bw *dl_b; |
1724813d | 8042 | int cpu; |
49516342 | 8043 | unsigned long flags; |
ce0dbbbb | 8044 | |
1724813d PZ |
8045 | def_dl_bandwidth.dl_period = global_rt_period(); |
8046 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
8047 | ||
8048 | if (global_rt_runtime() != RUNTIME_INF) | |
8049 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
8050 | ||
8051 | /* | |
8052 | * FIXME: As above... | |
8053 | */ | |
8054 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
8055 | rcu_read_lock_sched(); |
8056 | dl_b = dl_bw_of(cpu); | |
1724813d | 8057 | |
49516342 | 8058 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8059 | dl_b->bw = new_bw; |
49516342 | 8060 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8061 | |
8062 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8063 | } |
1724813d PZ |
8064 | } |
8065 | ||
8066 | static int sched_rt_global_validate(void) | |
8067 | { | |
8068 | if (sysctl_sched_rt_period <= 0) | |
8069 | return -EINVAL; | |
8070 | ||
e9e7cb38 JL |
8071 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8072 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8073 | return -EINVAL; |
8074 | ||
8075 | return 0; | |
8076 | } | |
8077 | ||
8078 | static void sched_rt_do_global(void) | |
8079 | { | |
8080 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8081 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8082 | } |
8083 | ||
d0b27fa7 | 8084 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8085 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8086 | loff_t *ppos) |
8087 | { | |
d0b27fa7 PZ |
8088 | int old_period, old_runtime; |
8089 | static DEFINE_MUTEX(mutex); | |
1724813d | 8090 | int ret; |
d0b27fa7 PZ |
8091 | |
8092 | mutex_lock(&mutex); | |
8093 | old_period = sysctl_sched_rt_period; | |
8094 | old_runtime = sysctl_sched_rt_runtime; | |
8095 | ||
8d65af78 | 8096 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8097 | |
8098 | if (!ret && write) { | |
1724813d PZ |
8099 | ret = sched_rt_global_validate(); |
8100 | if (ret) | |
8101 | goto undo; | |
8102 | ||
a1963b81 | 8103 | ret = sched_dl_global_validate(); |
1724813d PZ |
8104 | if (ret) |
8105 | goto undo; | |
8106 | ||
a1963b81 | 8107 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8108 | if (ret) |
8109 | goto undo; | |
8110 | ||
8111 | sched_rt_do_global(); | |
8112 | sched_dl_do_global(); | |
8113 | } | |
8114 | if (0) { | |
8115 | undo: | |
8116 | sysctl_sched_rt_period = old_period; | |
8117 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8118 | } |
8119 | mutex_unlock(&mutex); | |
8120 | ||
8121 | return ret; | |
8122 | } | |
68318b8e | 8123 | |
1724813d | 8124 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8125 | void __user *buffer, size_t *lenp, |
8126 | loff_t *ppos) | |
8127 | { | |
8128 | int ret; | |
332ac17e | 8129 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8130 | |
8131 | mutex_lock(&mutex); | |
332ac17e | 8132 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8133 | /* make sure that internally we keep jiffies */ |
8134 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8135 | if (!ret && write) { |
1724813d PZ |
8136 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8137 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8138 | } |
8139 | mutex_unlock(&mutex); | |
332ac17e DF |
8140 | return ret; |
8141 | } | |
8142 | ||
052f1dc7 | 8143 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8144 | |
a7c6d554 | 8145 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8146 | { |
a7c6d554 | 8147 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8148 | } |
8149 | ||
eb95419b TH |
8150 | static struct cgroup_subsys_state * |
8151 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8152 | { |
eb95419b TH |
8153 | struct task_group *parent = css_tg(parent_css); |
8154 | struct task_group *tg; | |
68318b8e | 8155 | |
eb95419b | 8156 | if (!parent) { |
68318b8e | 8157 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8158 | return &root_task_group.css; |
68318b8e SV |
8159 | } |
8160 | ||
ec7dc8ac | 8161 | tg = sched_create_group(parent); |
68318b8e SV |
8162 | if (IS_ERR(tg)) |
8163 | return ERR_PTR(-ENOMEM); | |
8164 | ||
2f5177f0 PZ |
8165 | sched_online_group(tg, parent); |
8166 | ||
68318b8e SV |
8167 | return &tg->css; |
8168 | } | |
8169 | ||
2f5177f0 | 8170 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8171 | { |
eb95419b | 8172 | struct task_group *tg = css_tg(css); |
ace783b9 | 8173 | |
2f5177f0 | 8174 | sched_offline_group(tg); |
ace783b9 LZ |
8175 | } |
8176 | ||
eb95419b | 8177 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8178 | { |
eb95419b | 8179 | struct task_group *tg = css_tg(css); |
68318b8e | 8180 | |
2f5177f0 PZ |
8181 | /* |
8182 | * Relies on the RCU grace period between css_released() and this. | |
8183 | */ | |
8184 | sched_free_group(tg); | |
ace783b9 LZ |
8185 | } |
8186 | ||
b53202e6 | 8187 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 KT |
8188 | { |
8189 | sched_move_task(task); | |
8190 | } | |
8191 | ||
1f7dd3e5 | 8192 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8193 | { |
bb9d97b6 | 8194 | struct task_struct *task; |
1f7dd3e5 | 8195 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8196 | |
1f7dd3e5 | 8197 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8198 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8199 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8200 | return -EINVAL; |
b68aa230 | 8201 | #else |
bb9d97b6 TH |
8202 | /* We don't support RT-tasks being in separate groups */ |
8203 | if (task->sched_class != &fair_sched_class) | |
8204 | return -EINVAL; | |
b68aa230 | 8205 | #endif |
bb9d97b6 | 8206 | } |
be367d09 BB |
8207 | return 0; |
8208 | } | |
68318b8e | 8209 | |
1f7dd3e5 | 8210 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8211 | { |
bb9d97b6 | 8212 | struct task_struct *task; |
1f7dd3e5 | 8213 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8214 | |
1f7dd3e5 | 8215 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8216 | sched_move_task(task); |
68318b8e SV |
8217 | } |
8218 | ||
052f1dc7 | 8219 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8220 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8221 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8222 | { |
182446d0 | 8223 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8224 | } |
8225 | ||
182446d0 TH |
8226 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8227 | struct cftype *cft) | |
68318b8e | 8228 | { |
182446d0 | 8229 | struct task_group *tg = css_tg(css); |
68318b8e | 8230 | |
c8b28116 | 8231 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8232 | } |
ab84d31e PT |
8233 | |
8234 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8235 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8236 | ||
ab84d31e PT |
8237 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8238 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8239 | ||
a790de99 PT |
8240 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8241 | ||
ab84d31e PT |
8242 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8243 | { | |
56f570e5 | 8244 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8245 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8246 | |
8247 | if (tg == &root_task_group) | |
8248 | return -EINVAL; | |
8249 | ||
8250 | /* | |
8251 | * Ensure we have at some amount of bandwidth every period. This is | |
8252 | * to prevent reaching a state of large arrears when throttled via | |
8253 | * entity_tick() resulting in prolonged exit starvation. | |
8254 | */ | |
8255 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8256 | return -EINVAL; | |
8257 | ||
8258 | /* | |
8259 | * Likewise, bound things on the otherside by preventing insane quota | |
8260 | * periods. This also allows us to normalize in computing quota | |
8261 | * feasibility. | |
8262 | */ | |
8263 | if (period > max_cfs_quota_period) | |
8264 | return -EINVAL; | |
8265 | ||
0e59bdae KT |
8266 | /* |
8267 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8268 | * unthrottle_offline_cfs_rqs(). | |
8269 | */ | |
8270 | get_online_cpus(); | |
a790de99 PT |
8271 | mutex_lock(&cfs_constraints_mutex); |
8272 | ret = __cfs_schedulable(tg, period, quota); | |
8273 | if (ret) | |
8274 | goto out_unlock; | |
8275 | ||
58088ad0 | 8276 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8277 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8278 | /* |
8279 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8280 | * before making related changes, and on->off must occur afterwards | |
8281 | */ | |
8282 | if (runtime_enabled && !runtime_was_enabled) | |
8283 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8284 | raw_spin_lock_irq(&cfs_b->lock); |
8285 | cfs_b->period = ns_to_ktime(period); | |
8286 | cfs_b->quota = quota; | |
58088ad0 | 8287 | |
a9cf55b2 | 8288 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8289 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8290 | if (runtime_enabled) |
8291 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8292 | raw_spin_unlock_irq(&cfs_b->lock); |
8293 | ||
0e59bdae | 8294 | for_each_online_cpu(i) { |
ab84d31e | 8295 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8296 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8297 | |
8298 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8299 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8300 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8301 | |
029632fb | 8302 | if (cfs_rq->throttled) |
671fd9da | 8303 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8304 | raw_spin_unlock_irq(&rq->lock); |
8305 | } | |
1ee14e6c BS |
8306 | if (runtime_was_enabled && !runtime_enabled) |
8307 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8308 | out_unlock: |
8309 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8310 | put_online_cpus(); |
ab84d31e | 8311 | |
a790de99 | 8312 | return ret; |
ab84d31e PT |
8313 | } |
8314 | ||
8315 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8316 | { | |
8317 | u64 quota, period; | |
8318 | ||
029632fb | 8319 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8320 | if (cfs_quota_us < 0) |
8321 | quota = RUNTIME_INF; | |
8322 | else | |
8323 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8324 | ||
8325 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8326 | } | |
8327 | ||
8328 | long tg_get_cfs_quota(struct task_group *tg) | |
8329 | { | |
8330 | u64 quota_us; | |
8331 | ||
029632fb | 8332 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8333 | return -1; |
8334 | ||
029632fb | 8335 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8336 | do_div(quota_us, NSEC_PER_USEC); |
8337 | ||
8338 | return quota_us; | |
8339 | } | |
8340 | ||
8341 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8342 | { | |
8343 | u64 quota, period; | |
8344 | ||
8345 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8346 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8347 | |
ab84d31e PT |
8348 | return tg_set_cfs_bandwidth(tg, period, quota); |
8349 | } | |
8350 | ||
8351 | long tg_get_cfs_period(struct task_group *tg) | |
8352 | { | |
8353 | u64 cfs_period_us; | |
8354 | ||
029632fb | 8355 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8356 | do_div(cfs_period_us, NSEC_PER_USEC); |
8357 | ||
8358 | return cfs_period_us; | |
8359 | } | |
8360 | ||
182446d0 TH |
8361 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8362 | struct cftype *cft) | |
ab84d31e | 8363 | { |
182446d0 | 8364 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8365 | } |
8366 | ||
182446d0 TH |
8367 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8368 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8369 | { |
182446d0 | 8370 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8371 | } |
8372 | ||
182446d0 TH |
8373 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8374 | struct cftype *cft) | |
ab84d31e | 8375 | { |
182446d0 | 8376 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8377 | } |
8378 | ||
182446d0 TH |
8379 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8380 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8381 | { |
182446d0 | 8382 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8383 | } |
8384 | ||
a790de99 PT |
8385 | struct cfs_schedulable_data { |
8386 | struct task_group *tg; | |
8387 | u64 period, quota; | |
8388 | }; | |
8389 | ||
8390 | /* | |
8391 | * normalize group quota/period to be quota/max_period | |
8392 | * note: units are usecs | |
8393 | */ | |
8394 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8395 | struct cfs_schedulable_data *d) | |
8396 | { | |
8397 | u64 quota, period; | |
8398 | ||
8399 | if (tg == d->tg) { | |
8400 | period = d->period; | |
8401 | quota = d->quota; | |
8402 | } else { | |
8403 | period = tg_get_cfs_period(tg); | |
8404 | quota = tg_get_cfs_quota(tg); | |
8405 | } | |
8406 | ||
8407 | /* note: these should typically be equivalent */ | |
8408 | if (quota == RUNTIME_INF || quota == -1) | |
8409 | return RUNTIME_INF; | |
8410 | ||
8411 | return to_ratio(period, quota); | |
8412 | } | |
8413 | ||
8414 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8415 | { | |
8416 | struct cfs_schedulable_data *d = data; | |
029632fb | 8417 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8418 | s64 quota = 0, parent_quota = -1; |
8419 | ||
8420 | if (!tg->parent) { | |
8421 | quota = RUNTIME_INF; | |
8422 | } else { | |
029632fb | 8423 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8424 | |
8425 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8426 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8427 | |
8428 | /* | |
8429 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8430 | * limit is set | |
8431 | */ | |
8432 | if (quota == RUNTIME_INF) | |
8433 | quota = parent_quota; | |
8434 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8435 | return -EINVAL; | |
8436 | } | |
9c58c79a | 8437 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8438 | |
8439 | return 0; | |
8440 | } | |
8441 | ||
8442 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8443 | { | |
8277434e | 8444 | int ret; |
a790de99 PT |
8445 | struct cfs_schedulable_data data = { |
8446 | .tg = tg, | |
8447 | .period = period, | |
8448 | .quota = quota, | |
8449 | }; | |
8450 | ||
8451 | if (quota != RUNTIME_INF) { | |
8452 | do_div(data.period, NSEC_PER_USEC); | |
8453 | do_div(data.quota, NSEC_PER_USEC); | |
8454 | } | |
8455 | ||
8277434e PT |
8456 | rcu_read_lock(); |
8457 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8458 | rcu_read_unlock(); | |
8459 | ||
8460 | return ret; | |
a790de99 | 8461 | } |
e8da1b18 | 8462 | |
2da8ca82 | 8463 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8464 | { |
2da8ca82 | 8465 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8466 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8467 | |
44ffc75b TH |
8468 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8469 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8470 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8471 | |
8472 | return 0; | |
8473 | } | |
ab84d31e | 8474 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8475 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8476 | |
052f1dc7 | 8477 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8478 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8479 | struct cftype *cft, s64 val) | |
6f505b16 | 8480 | { |
182446d0 | 8481 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8482 | } |
8483 | ||
182446d0 TH |
8484 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8485 | struct cftype *cft) | |
6f505b16 | 8486 | { |
182446d0 | 8487 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8488 | } |
d0b27fa7 | 8489 | |
182446d0 TH |
8490 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8491 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8492 | { |
182446d0 | 8493 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8494 | } |
8495 | ||
182446d0 TH |
8496 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8497 | struct cftype *cft) | |
d0b27fa7 | 8498 | { |
182446d0 | 8499 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8500 | } |
6d6bc0ad | 8501 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8502 | |
fe5c7cc2 | 8503 | static struct cftype cpu_files[] = { |
052f1dc7 | 8504 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8505 | { |
8506 | .name = "shares", | |
f4c753b7 PM |
8507 | .read_u64 = cpu_shares_read_u64, |
8508 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8509 | }, |
052f1dc7 | 8510 | #endif |
ab84d31e PT |
8511 | #ifdef CONFIG_CFS_BANDWIDTH |
8512 | { | |
8513 | .name = "cfs_quota_us", | |
8514 | .read_s64 = cpu_cfs_quota_read_s64, | |
8515 | .write_s64 = cpu_cfs_quota_write_s64, | |
8516 | }, | |
8517 | { | |
8518 | .name = "cfs_period_us", | |
8519 | .read_u64 = cpu_cfs_period_read_u64, | |
8520 | .write_u64 = cpu_cfs_period_write_u64, | |
8521 | }, | |
e8da1b18 NR |
8522 | { |
8523 | .name = "stat", | |
2da8ca82 | 8524 | .seq_show = cpu_stats_show, |
e8da1b18 | 8525 | }, |
ab84d31e | 8526 | #endif |
052f1dc7 | 8527 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8528 | { |
9f0c1e56 | 8529 | .name = "rt_runtime_us", |
06ecb27c PM |
8530 | .read_s64 = cpu_rt_runtime_read, |
8531 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8532 | }, |
d0b27fa7 PZ |
8533 | { |
8534 | .name = "rt_period_us", | |
f4c753b7 PM |
8535 | .read_u64 = cpu_rt_period_read_uint, |
8536 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8537 | }, |
052f1dc7 | 8538 | #endif |
4baf6e33 | 8539 | { } /* terminate */ |
68318b8e SV |
8540 | }; |
8541 | ||
073219e9 | 8542 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8543 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8544 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8545 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8546 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8547 | .can_attach = cpu_cgroup_can_attach, |
8548 | .attach = cpu_cgroup_attach, | |
5577964e | 8549 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8550 | .early_init = true, |
68318b8e SV |
8551 | }; |
8552 | ||
052f1dc7 | 8553 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8554 | |
b637a328 PM |
8555 | void dump_cpu_task(int cpu) |
8556 | { | |
8557 | pr_info("Task dump for CPU %d:\n", cpu); | |
8558 | sched_show_task(cpu_curr(cpu)); | |
8559 | } | |
ed82b8a1 AK |
8560 | |
8561 | /* | |
8562 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8563 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8564 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8565 | * that remained on nice 0. | |
8566 | * | |
8567 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8568 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8569 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8570 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8571 | * the relative distance between them is ~25%.) | |
8572 | */ | |
8573 | const int sched_prio_to_weight[40] = { | |
8574 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8575 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8576 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8577 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8578 | /* 0 */ 1024, 820, 655, 526, 423, | |
8579 | /* 5 */ 335, 272, 215, 172, 137, | |
8580 | /* 10 */ 110, 87, 70, 56, 45, | |
8581 | /* 15 */ 36, 29, 23, 18, 15, | |
8582 | }; | |
8583 | ||
8584 | /* | |
8585 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8586 | * | |
8587 | * In cases where the weight does not change often, we can use the | |
8588 | * precalculated inverse to speed up arithmetics by turning divisions | |
8589 | * into multiplications: | |
8590 | */ | |
8591 | const u32 sched_prio_to_wmult[40] = { | |
8592 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8593 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8594 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8595 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8596 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8597 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8598 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8599 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8600 | }; |