]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
e1b77c92 | 29 | #include <linux/kasan.h> |
1da177e4 LT |
30 | #include <linux/mm.h> |
31 | #include <linux/module.h> | |
32 | #include <linux/nmi.h> | |
33 | #include <linux/init.h> | |
dff06c15 | 34 | #include <linux/uaccess.h> |
1da177e4 | 35 | #include <linux/highmem.h> |
f98db601 | 36 | #include <linux/mmu_context.h> |
1da177e4 | 37 | #include <linux/interrupt.h> |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 | 70 | #include <linux/ctype.h> |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
91d1aa43 | 74 | #include <linux/context_tracking.h> |
52f5684c | 75 | #include <linux/compiler.h> |
8e05e96a | 76 | #include <linux/frame.h> |
6075620b | 77 | #include <linux/prefetch.h> |
3ca0ff57 | 78 | #include <linux/mutex.h> |
1da177e4 | 79 | |
96f951ed | 80 | #include <asm/switch_to.h> |
5517d86b | 81 | #include <asm/tlb.h> |
838225b4 | 82 | #include <asm/irq_regs.h> |
e6e6685a GC |
83 | #ifdef CONFIG_PARAVIRT |
84 | #include <asm/paravirt.h> | |
85 | #endif | |
1da177e4 | 86 | |
029632fb | 87 | #include "sched.h" |
ea138446 | 88 | #include "../workqueue_internal.h" |
29d5e047 | 89 | #include "../smpboot.h" |
6e0534f2 | 90 | |
a8d154b0 | 91 | #define CREATE_TRACE_POINTS |
ad8d75ff | 92 | #include <trace/events/sched.h> |
a8d154b0 | 93 | |
029632fb PZ |
94 | DEFINE_MUTEX(sched_domains_mutex); |
95 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 96 | |
fe44d621 | 97 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 98 | |
029632fb | 99 | void update_rq_clock(struct rq *rq) |
3e51f33f | 100 | { |
fe44d621 | 101 | s64 delta; |
305e6835 | 102 | |
9edfbfed PZ |
103 | lockdep_assert_held(&rq->lock); |
104 | ||
105 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 106 | return; |
aa483808 | 107 | |
fe44d621 | 108 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
109 | if (delta < 0) |
110 | return; | |
fe44d621 PZ |
111 | rq->clock += delta; |
112 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
113 | } |
114 | ||
bf5c91ba IM |
115 | /* |
116 | * Debugging: various feature bits | |
117 | */ | |
f00b45c1 | 118 | |
f00b45c1 PZ |
119 | #define SCHED_FEAT(name, enabled) \ |
120 | (1UL << __SCHED_FEAT_##name) * enabled | | |
121 | ||
bf5c91ba | 122 | const_debug unsigned int sysctl_sched_features = |
391e43da | 123 | #include "features.h" |
f00b45c1 PZ |
124 | 0; |
125 | ||
126 | #undef SCHED_FEAT | |
127 | ||
b82d9fdd PZ |
128 | /* |
129 | * Number of tasks to iterate in a single balance run. | |
130 | * Limited because this is done with IRQs disabled. | |
131 | */ | |
132 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
133 | ||
e9e9250b PZ |
134 | /* |
135 | * period over which we average the RT time consumption, measured | |
136 | * in ms. | |
137 | * | |
138 | * default: 1s | |
139 | */ | |
140 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
141 | ||
fa85ae24 | 142 | /* |
9f0c1e56 | 143 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
144 | * default: 1s |
145 | */ | |
9f0c1e56 | 146 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 147 | |
029632fb | 148 | __read_mostly int scheduler_running; |
6892b75e | 149 | |
9f0c1e56 PZ |
150 | /* |
151 | * part of the period that we allow rt tasks to run in us. | |
152 | * default: 0.95s | |
153 | */ | |
154 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 155 | |
3fa0818b RR |
156 | /* cpus with isolated domains */ |
157 | cpumask_var_t cpu_isolated_map; | |
158 | ||
1da177e4 | 159 | /* |
cc2a73b5 | 160 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 161 | */ |
a9957449 | 162 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
163 | __acquires(rq->lock) |
164 | { | |
70b97a7f | 165 | struct rq *rq; |
1da177e4 LT |
166 | |
167 | local_irq_disable(); | |
168 | rq = this_rq(); | |
05fa785c | 169 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
170 | |
171 | return rq; | |
172 | } | |
173 | ||
3e71a462 PZ |
174 | /* |
175 | * __task_rq_lock - lock the rq @p resides on. | |
176 | */ | |
eb580751 | 177 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
178 | __acquires(rq->lock) |
179 | { | |
180 | struct rq *rq; | |
181 | ||
182 | lockdep_assert_held(&p->pi_lock); | |
183 | ||
184 | for (;;) { | |
185 | rq = task_rq(p); | |
186 | raw_spin_lock(&rq->lock); | |
187 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 188 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
189 | return rq; |
190 | } | |
191 | raw_spin_unlock(&rq->lock); | |
192 | ||
193 | while (unlikely(task_on_rq_migrating(p))) | |
194 | cpu_relax(); | |
195 | } | |
196 | } | |
197 | ||
198 | /* | |
199 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
200 | */ | |
eb580751 | 201 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
202 | __acquires(p->pi_lock) |
203 | __acquires(rq->lock) | |
204 | { | |
205 | struct rq *rq; | |
206 | ||
207 | for (;;) { | |
eb580751 | 208 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
209 | rq = task_rq(p); |
210 | raw_spin_lock(&rq->lock); | |
211 | /* | |
212 | * move_queued_task() task_rq_lock() | |
213 | * | |
214 | * ACQUIRE (rq->lock) | |
215 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
216 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
217 | * [S] ->cpu = new_cpu [L] task_rq() | |
218 | * [L] ->on_rq | |
219 | * RELEASE (rq->lock) | |
220 | * | |
221 | * If we observe the old cpu in task_rq_lock, the acquire of | |
222 | * the old rq->lock will fully serialize against the stores. | |
223 | * | |
224 | * If we observe the new cpu in task_rq_lock, the acquire will | |
225 | * pair with the WMB to ensure we must then also see migrating. | |
226 | */ | |
227 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 228 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
229 | return rq; |
230 | } | |
231 | raw_spin_unlock(&rq->lock); | |
eb580751 | 232 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
233 | |
234 | while (unlikely(task_on_rq_migrating(p))) | |
235 | cpu_relax(); | |
236 | } | |
237 | } | |
238 | ||
8f4d37ec PZ |
239 | #ifdef CONFIG_SCHED_HRTICK |
240 | /* | |
241 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 242 | */ |
8f4d37ec | 243 | |
8f4d37ec PZ |
244 | static void hrtick_clear(struct rq *rq) |
245 | { | |
246 | if (hrtimer_active(&rq->hrtick_timer)) | |
247 | hrtimer_cancel(&rq->hrtick_timer); | |
248 | } | |
249 | ||
8f4d37ec PZ |
250 | /* |
251 | * High-resolution timer tick. | |
252 | * Runs from hardirq context with interrupts disabled. | |
253 | */ | |
254 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
255 | { | |
256 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
257 | ||
258 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
259 | ||
05fa785c | 260 | raw_spin_lock(&rq->lock); |
3e51f33f | 261 | update_rq_clock(rq); |
8f4d37ec | 262 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 263 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
264 | |
265 | return HRTIMER_NORESTART; | |
266 | } | |
267 | ||
95e904c7 | 268 | #ifdef CONFIG_SMP |
971ee28c | 269 | |
4961b6e1 | 270 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
271 | { |
272 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 273 | |
4961b6e1 | 274 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
275 | } |
276 | ||
31656519 PZ |
277 | /* |
278 | * called from hardirq (IPI) context | |
279 | */ | |
280 | static void __hrtick_start(void *arg) | |
b328ca18 | 281 | { |
31656519 | 282 | struct rq *rq = arg; |
b328ca18 | 283 | |
05fa785c | 284 | raw_spin_lock(&rq->lock); |
971ee28c | 285 | __hrtick_restart(rq); |
31656519 | 286 | rq->hrtick_csd_pending = 0; |
05fa785c | 287 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
288 | } |
289 | ||
31656519 PZ |
290 | /* |
291 | * Called to set the hrtick timer state. | |
292 | * | |
293 | * called with rq->lock held and irqs disabled | |
294 | */ | |
029632fb | 295 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 296 | { |
31656519 | 297 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 298 | ktime_t time; |
299 | s64 delta; | |
300 | ||
301 | /* | |
302 | * Don't schedule slices shorter than 10000ns, that just | |
303 | * doesn't make sense and can cause timer DoS. | |
304 | */ | |
305 | delta = max_t(s64, delay, 10000LL); | |
306 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 307 | |
cc584b21 | 308 | hrtimer_set_expires(timer, time); |
31656519 PZ |
309 | |
310 | if (rq == this_rq()) { | |
971ee28c | 311 | __hrtick_restart(rq); |
31656519 | 312 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 313 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
314 | rq->hrtick_csd_pending = 1; |
315 | } | |
b328ca18 PZ |
316 | } |
317 | ||
31656519 PZ |
318 | #else |
319 | /* | |
320 | * Called to set the hrtick timer state. | |
321 | * | |
322 | * called with rq->lock held and irqs disabled | |
323 | */ | |
029632fb | 324 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 325 | { |
86893335 WL |
326 | /* |
327 | * Don't schedule slices shorter than 10000ns, that just | |
328 | * doesn't make sense. Rely on vruntime for fairness. | |
329 | */ | |
330 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
331 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
332 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 333 | } |
31656519 | 334 | #endif /* CONFIG_SMP */ |
8f4d37ec | 335 | |
31656519 | 336 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 337 | { |
31656519 PZ |
338 | #ifdef CONFIG_SMP |
339 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 340 | |
31656519 PZ |
341 | rq->hrtick_csd.flags = 0; |
342 | rq->hrtick_csd.func = __hrtick_start; | |
343 | rq->hrtick_csd.info = rq; | |
344 | #endif | |
8f4d37ec | 345 | |
31656519 PZ |
346 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
347 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 348 | } |
006c75f1 | 349 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
350 | static inline void hrtick_clear(struct rq *rq) |
351 | { | |
352 | } | |
353 | ||
8f4d37ec PZ |
354 | static inline void init_rq_hrtick(struct rq *rq) |
355 | { | |
356 | } | |
006c75f1 | 357 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 358 | |
5529578a FW |
359 | /* |
360 | * cmpxchg based fetch_or, macro so it works for different integer types | |
361 | */ | |
362 | #define fetch_or(ptr, mask) \ | |
363 | ({ \ | |
364 | typeof(ptr) _ptr = (ptr); \ | |
365 | typeof(mask) _mask = (mask); \ | |
366 | typeof(*_ptr) _old, _val = *_ptr; \ | |
367 | \ | |
368 | for (;;) { \ | |
369 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
370 | if (_old == _val) \ | |
371 | break; \ | |
372 | _val = _old; \ | |
373 | } \ | |
374 | _old; \ | |
375 | }) | |
376 | ||
e3baac47 | 377 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
378 | /* |
379 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
380 | * this avoids any races wrt polling state changes and thereby avoids | |
381 | * spurious IPIs. | |
382 | */ | |
383 | static bool set_nr_and_not_polling(struct task_struct *p) | |
384 | { | |
385 | struct thread_info *ti = task_thread_info(p); | |
386 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
387 | } | |
e3baac47 PZ |
388 | |
389 | /* | |
390 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
391 | * | |
392 | * If this returns true, then the idle task promises to call | |
393 | * sched_ttwu_pending() and reschedule soon. | |
394 | */ | |
395 | static bool set_nr_if_polling(struct task_struct *p) | |
396 | { | |
397 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 398 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
399 | |
400 | for (;;) { | |
401 | if (!(val & _TIF_POLLING_NRFLAG)) | |
402 | return false; | |
403 | if (val & _TIF_NEED_RESCHED) | |
404 | return true; | |
405 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
406 | if (old == val) | |
407 | break; | |
408 | val = old; | |
409 | } | |
410 | return true; | |
411 | } | |
412 | ||
fd99f91a PZ |
413 | #else |
414 | static bool set_nr_and_not_polling(struct task_struct *p) | |
415 | { | |
416 | set_tsk_need_resched(p); | |
417 | return true; | |
418 | } | |
e3baac47 PZ |
419 | |
420 | #ifdef CONFIG_SMP | |
421 | static bool set_nr_if_polling(struct task_struct *p) | |
422 | { | |
423 | return false; | |
424 | } | |
425 | #endif | |
fd99f91a PZ |
426 | #endif |
427 | ||
76751049 PZ |
428 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
429 | { | |
430 | struct wake_q_node *node = &task->wake_q; | |
431 | ||
432 | /* | |
433 | * Atomically grab the task, if ->wake_q is !nil already it means | |
434 | * its already queued (either by us or someone else) and will get the | |
435 | * wakeup due to that. | |
436 | * | |
437 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 438 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
439 | */ |
440 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
441 | return; | |
442 | ||
443 | get_task_struct(task); | |
444 | ||
445 | /* | |
446 | * The head is context local, there can be no concurrency. | |
447 | */ | |
448 | *head->lastp = node; | |
449 | head->lastp = &node->next; | |
450 | } | |
451 | ||
452 | void wake_up_q(struct wake_q_head *head) | |
453 | { | |
454 | struct wake_q_node *node = head->first; | |
455 | ||
456 | while (node != WAKE_Q_TAIL) { | |
457 | struct task_struct *task; | |
458 | ||
459 | task = container_of(node, struct task_struct, wake_q); | |
460 | BUG_ON(!task); | |
461 | /* task can safely be re-inserted now */ | |
462 | node = node->next; | |
463 | task->wake_q.next = NULL; | |
464 | ||
465 | /* | |
466 | * wake_up_process() implies a wmb() to pair with the queueing | |
467 | * in wake_q_add() so as not to miss wakeups. | |
468 | */ | |
469 | wake_up_process(task); | |
470 | put_task_struct(task); | |
471 | } | |
472 | } | |
473 | ||
c24d20db | 474 | /* |
8875125e | 475 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
476 | * |
477 | * On UP this means the setting of the need_resched flag, on SMP it | |
478 | * might also involve a cross-CPU call to trigger the scheduler on | |
479 | * the target CPU. | |
480 | */ | |
8875125e | 481 | void resched_curr(struct rq *rq) |
c24d20db | 482 | { |
8875125e | 483 | struct task_struct *curr = rq->curr; |
c24d20db IM |
484 | int cpu; |
485 | ||
8875125e | 486 | lockdep_assert_held(&rq->lock); |
c24d20db | 487 | |
8875125e | 488 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
489 | return; |
490 | ||
8875125e | 491 | cpu = cpu_of(rq); |
fd99f91a | 492 | |
f27dde8d | 493 | if (cpu == smp_processor_id()) { |
8875125e | 494 | set_tsk_need_resched(curr); |
f27dde8d | 495 | set_preempt_need_resched(); |
c24d20db | 496 | return; |
f27dde8d | 497 | } |
c24d20db | 498 | |
8875125e | 499 | if (set_nr_and_not_polling(curr)) |
c24d20db | 500 | smp_send_reschedule(cpu); |
dfc68f29 AL |
501 | else |
502 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
503 | } |
504 | ||
029632fb | 505 | void resched_cpu(int cpu) |
c24d20db IM |
506 | { |
507 | struct rq *rq = cpu_rq(cpu); | |
508 | unsigned long flags; | |
509 | ||
05fa785c | 510 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 511 | return; |
8875125e | 512 | resched_curr(rq); |
05fa785c | 513 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 514 | } |
06d8308c | 515 | |
b021fe3e | 516 | #ifdef CONFIG_SMP |
3451d024 | 517 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
518 | /* |
519 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
520 | * from an idle cpu. This is good for power-savings. | |
521 | * | |
522 | * We don't do similar optimization for completely idle system, as | |
523 | * selecting an idle cpu will add more delays to the timers than intended | |
524 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
525 | */ | |
bc7a34b8 | 526 | int get_nohz_timer_target(void) |
83cd4fe2 | 527 | { |
bc7a34b8 | 528 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
529 | struct sched_domain *sd; |
530 | ||
9642d18e | 531 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
532 | return cpu; |
533 | ||
057f3fad | 534 | rcu_read_lock(); |
83cd4fe2 | 535 | for_each_domain(cpu, sd) { |
057f3fad | 536 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
537 | if (cpu == i) |
538 | continue; | |
539 | ||
540 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
541 | cpu = i; |
542 | goto unlock; | |
543 | } | |
544 | } | |
83cd4fe2 | 545 | } |
9642d18e VH |
546 | |
547 | if (!is_housekeeping_cpu(cpu)) | |
548 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
549 | unlock: |
550 | rcu_read_unlock(); | |
83cd4fe2 VP |
551 | return cpu; |
552 | } | |
06d8308c TG |
553 | /* |
554 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
555 | * idle CPU then this timer might expire before the next timer event | |
556 | * which is scheduled to wake up that CPU. In case of a completely | |
557 | * idle system the next event might even be infinite time into the | |
558 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
559 | * leaves the inner idle loop so the newly added timer is taken into | |
560 | * account when the CPU goes back to idle and evaluates the timer | |
561 | * wheel for the next timer event. | |
562 | */ | |
1c20091e | 563 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
564 | { |
565 | struct rq *rq = cpu_rq(cpu); | |
566 | ||
567 | if (cpu == smp_processor_id()) | |
568 | return; | |
569 | ||
67b9ca70 | 570 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 571 | smp_send_reschedule(cpu); |
dfc68f29 AL |
572 | else |
573 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
574 | } |
575 | ||
c5bfece2 | 576 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 577 | { |
53c5fa16 FW |
578 | /* |
579 | * We just need the target to call irq_exit() and re-evaluate | |
580 | * the next tick. The nohz full kick at least implies that. | |
581 | * If needed we can still optimize that later with an | |
582 | * empty IRQ. | |
583 | */ | |
379d9ecb PM |
584 | if (cpu_is_offline(cpu)) |
585 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 586 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
587 | if (cpu != smp_processor_id() || |
588 | tick_nohz_tick_stopped()) | |
53c5fa16 | 589 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
590 | return true; |
591 | } | |
592 | ||
593 | return false; | |
594 | } | |
595 | ||
379d9ecb PM |
596 | /* |
597 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
598 | * caller's responsibility to deal with the lost wakeup, for example, | |
599 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
600 | */ | |
1c20091e FW |
601 | void wake_up_nohz_cpu(int cpu) |
602 | { | |
c5bfece2 | 603 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
604 | wake_up_idle_cpu(cpu); |
605 | } | |
606 | ||
ca38062e | 607 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 608 | { |
1c792db7 | 609 | int cpu = smp_processor_id(); |
873b4c65 VG |
610 | |
611 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
612 | return false; | |
613 | ||
614 | if (idle_cpu(cpu) && !need_resched()) | |
615 | return true; | |
616 | ||
617 | /* | |
618 | * We can't run Idle Load Balance on this CPU for this time so we | |
619 | * cancel it and clear NOHZ_BALANCE_KICK | |
620 | */ | |
621 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
622 | return false; | |
45bf76df IM |
623 | } |
624 | ||
3451d024 | 625 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 626 | |
ca38062e | 627 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 628 | { |
ca38062e | 629 | return false; |
2069dd75 PZ |
630 | } |
631 | ||
3451d024 | 632 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 633 | |
ce831b38 | 634 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 635 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 636 | { |
76d92ac3 FW |
637 | int fifo_nr_running; |
638 | ||
639 | /* Deadline tasks, even if single, need the tick */ | |
640 | if (rq->dl.dl_nr_running) | |
641 | return false; | |
642 | ||
1e78cdbd | 643 | /* |
2548d546 PZ |
644 | * If there are more than one RR tasks, we need the tick to effect the |
645 | * actual RR behaviour. | |
1e78cdbd | 646 | */ |
76d92ac3 FW |
647 | if (rq->rt.rr_nr_running) { |
648 | if (rq->rt.rr_nr_running == 1) | |
649 | return true; | |
650 | else | |
651 | return false; | |
1e78cdbd RR |
652 | } |
653 | ||
2548d546 PZ |
654 | /* |
655 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
656 | * forced preemption between FIFO tasks. | |
657 | */ | |
658 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
659 | if (fifo_nr_running) | |
660 | return true; | |
661 | ||
662 | /* | |
663 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
664 | * if there's more than one we need the tick for involuntary | |
665 | * preemption. | |
666 | */ | |
667 | if (rq->nr_running > 1) | |
541b8264 | 668 | return false; |
ce831b38 | 669 | |
541b8264 | 670 | return true; |
ce831b38 FW |
671 | } |
672 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 673 | |
029632fb | 674 | void sched_avg_update(struct rq *rq) |
18d95a28 | 675 | { |
e9e9250b PZ |
676 | s64 period = sched_avg_period(); |
677 | ||
78becc27 | 678 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
679 | /* |
680 | * Inline assembly required to prevent the compiler | |
681 | * optimising this loop into a divmod call. | |
682 | * See __iter_div_u64_rem() for another example of this. | |
683 | */ | |
684 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
685 | rq->age_stamp += period; |
686 | rq->rt_avg /= 2; | |
687 | } | |
18d95a28 PZ |
688 | } |
689 | ||
6d6bc0ad | 690 | #endif /* CONFIG_SMP */ |
18d95a28 | 691 | |
a790de99 PT |
692 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
693 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 694 | /* |
8277434e PT |
695 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
696 | * node and @up when leaving it for the final time. | |
697 | * | |
698 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 699 | */ |
029632fb | 700 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 701 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
702 | { |
703 | struct task_group *parent, *child; | |
eb755805 | 704 | int ret; |
c09595f6 | 705 | |
8277434e PT |
706 | parent = from; |
707 | ||
c09595f6 | 708 | down: |
eb755805 PZ |
709 | ret = (*down)(parent, data); |
710 | if (ret) | |
8277434e | 711 | goto out; |
c09595f6 PZ |
712 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
713 | parent = child; | |
714 | goto down; | |
715 | ||
716 | up: | |
717 | continue; | |
718 | } | |
eb755805 | 719 | ret = (*up)(parent, data); |
8277434e PT |
720 | if (ret || parent == from) |
721 | goto out; | |
c09595f6 PZ |
722 | |
723 | child = parent; | |
724 | parent = parent->parent; | |
725 | if (parent) | |
726 | goto up; | |
8277434e | 727 | out: |
eb755805 | 728 | return ret; |
c09595f6 PZ |
729 | } |
730 | ||
029632fb | 731 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 732 | { |
e2b245f8 | 733 | return 0; |
eb755805 | 734 | } |
18d95a28 PZ |
735 | #endif |
736 | ||
45bf76df IM |
737 | static void set_load_weight(struct task_struct *p) |
738 | { | |
f05998d4 NR |
739 | int prio = p->static_prio - MAX_RT_PRIO; |
740 | struct load_weight *load = &p->se.load; | |
741 | ||
dd41f596 IM |
742 | /* |
743 | * SCHED_IDLE tasks get minimal weight: | |
744 | */ | |
20f9cd2a | 745 | if (idle_policy(p->policy)) { |
c8b28116 | 746 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 747 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
748 | return; |
749 | } | |
71f8bd46 | 750 | |
ed82b8a1 AK |
751 | load->weight = scale_load(sched_prio_to_weight[prio]); |
752 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
753 | } |
754 | ||
1de64443 | 755 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 756 | { |
a64692a3 | 757 | update_rq_clock(rq); |
1de64443 PZ |
758 | if (!(flags & ENQUEUE_RESTORE)) |
759 | sched_info_queued(rq, p); | |
371fd7e7 | 760 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
761 | } |
762 | ||
1de64443 | 763 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 764 | { |
a64692a3 | 765 | update_rq_clock(rq); |
1de64443 PZ |
766 | if (!(flags & DEQUEUE_SAVE)) |
767 | sched_info_dequeued(rq, p); | |
371fd7e7 | 768 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
769 | } |
770 | ||
029632fb | 771 | void activate_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 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
777 | } |
778 | ||
029632fb | 779 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
780 | { |
781 | if (task_contributes_to_load(p)) | |
782 | rq->nr_uninterruptible++; | |
783 | ||
371fd7e7 | 784 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
785 | } |
786 | ||
fe44d621 | 787 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 788 | { |
095c0aa8 GC |
789 | /* |
790 | * In theory, the compile should just see 0 here, and optimize out the call | |
791 | * to sched_rt_avg_update. But I don't trust it... | |
792 | */ | |
793 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
794 | s64 steal = 0, irq_delta = 0; | |
795 | #endif | |
796 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 797 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
798 | |
799 | /* | |
800 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
801 | * this case when a previous update_rq_clock() happened inside a | |
802 | * {soft,}irq region. | |
803 | * | |
804 | * When this happens, we stop ->clock_task and only update the | |
805 | * prev_irq_time stamp to account for the part that fit, so that a next | |
806 | * update will consume the rest. This ensures ->clock_task is | |
807 | * monotonic. | |
808 | * | |
809 | * It does however cause some slight miss-attribution of {soft,}irq | |
810 | * time, a more accurate solution would be to update the irq_time using | |
811 | * the current rq->clock timestamp, except that would require using | |
812 | * atomic ops. | |
813 | */ | |
814 | if (irq_delta > delta) | |
815 | irq_delta = delta; | |
816 | ||
817 | rq->prev_irq_time += irq_delta; | |
818 | delta -= irq_delta; | |
095c0aa8 GC |
819 | #endif |
820 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 821 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
822 | steal = paravirt_steal_clock(cpu_of(rq)); |
823 | steal -= rq->prev_steal_time_rq; | |
824 | ||
825 | if (unlikely(steal > delta)) | |
826 | steal = delta; | |
827 | ||
095c0aa8 | 828 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
829 | delta -= steal; |
830 | } | |
831 | #endif | |
832 | ||
fe44d621 PZ |
833 | rq->clock_task += delta; |
834 | ||
095c0aa8 | 835 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 836 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
837 | sched_rt_avg_update(rq, irq_delta + steal); |
838 | #endif | |
aa483808 VP |
839 | } |
840 | ||
34f971f6 PZ |
841 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
842 | { | |
843 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
844 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
845 | ||
846 | if (stop) { | |
847 | /* | |
848 | * Make it appear like a SCHED_FIFO task, its something | |
849 | * userspace knows about and won't get confused about. | |
850 | * | |
851 | * Also, it will make PI more or less work without too | |
852 | * much confusion -- but then, stop work should not | |
853 | * rely on PI working anyway. | |
854 | */ | |
855 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
856 | ||
857 | stop->sched_class = &stop_sched_class; | |
858 | } | |
859 | ||
860 | cpu_rq(cpu)->stop = stop; | |
861 | ||
862 | if (old_stop) { | |
863 | /* | |
864 | * Reset it back to a normal scheduling class so that | |
865 | * it can die in pieces. | |
866 | */ | |
867 | old_stop->sched_class = &rt_sched_class; | |
868 | } | |
869 | } | |
870 | ||
14531189 | 871 | /* |
dd41f596 | 872 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 873 | */ |
14531189 IM |
874 | static inline int __normal_prio(struct task_struct *p) |
875 | { | |
dd41f596 | 876 | return p->static_prio; |
14531189 IM |
877 | } |
878 | ||
b29739f9 IM |
879 | /* |
880 | * Calculate the expected normal priority: i.e. priority | |
881 | * without taking RT-inheritance into account. Might be | |
882 | * boosted by interactivity modifiers. Changes upon fork, | |
883 | * setprio syscalls, and whenever the interactivity | |
884 | * estimator recalculates. | |
885 | */ | |
36c8b586 | 886 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
887 | { |
888 | int prio; | |
889 | ||
aab03e05 DF |
890 | if (task_has_dl_policy(p)) |
891 | prio = MAX_DL_PRIO-1; | |
892 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
893 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
894 | else | |
895 | prio = __normal_prio(p); | |
896 | return prio; | |
897 | } | |
898 | ||
899 | /* | |
900 | * Calculate the current priority, i.e. the priority | |
901 | * taken into account by the scheduler. This value might | |
902 | * be boosted by RT tasks, or might be boosted by | |
903 | * interactivity modifiers. Will be RT if the task got | |
904 | * RT-boosted. If not then it returns p->normal_prio. | |
905 | */ | |
36c8b586 | 906 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
907 | { |
908 | p->normal_prio = normal_prio(p); | |
909 | /* | |
910 | * If we are RT tasks or we were boosted to RT priority, | |
911 | * keep the priority unchanged. Otherwise, update priority | |
912 | * to the normal priority: | |
913 | */ | |
914 | if (!rt_prio(p->prio)) | |
915 | return p->normal_prio; | |
916 | return p->prio; | |
917 | } | |
918 | ||
1da177e4 LT |
919 | /** |
920 | * task_curr - is this task currently executing on a CPU? | |
921 | * @p: the task in question. | |
e69f6186 YB |
922 | * |
923 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 924 | */ |
36c8b586 | 925 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
926 | { |
927 | return cpu_curr(task_cpu(p)) == p; | |
928 | } | |
929 | ||
67dfa1b7 | 930 | /* |
4c9a4bc8 PZ |
931 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
932 | * use the balance_callback list if you want balancing. | |
933 | * | |
934 | * this means any call to check_class_changed() must be followed by a call to | |
935 | * balance_callback(). | |
67dfa1b7 | 936 | */ |
cb469845 SR |
937 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
938 | const struct sched_class *prev_class, | |
da7a735e | 939 | int oldprio) |
cb469845 SR |
940 | { |
941 | if (prev_class != p->sched_class) { | |
942 | if (prev_class->switched_from) | |
da7a735e | 943 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 944 | |
da7a735e | 945 | p->sched_class->switched_to(rq, p); |
2d3d891d | 946 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 947 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
948 | } |
949 | ||
029632fb | 950 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
951 | { |
952 | const struct sched_class *class; | |
953 | ||
954 | if (p->sched_class == rq->curr->sched_class) { | |
955 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
956 | } else { | |
957 | for_each_class(class) { | |
958 | if (class == rq->curr->sched_class) | |
959 | break; | |
960 | if (class == p->sched_class) { | |
8875125e | 961 | resched_curr(rq); |
1e5a7405 PZ |
962 | break; |
963 | } | |
964 | } | |
965 | } | |
966 | ||
967 | /* | |
968 | * A queue event has occurred, and we're going to schedule. In | |
969 | * this case, we can save a useless back to back clock update. | |
970 | */ | |
da0c1e65 | 971 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 972 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
973 | } |
974 | ||
1da177e4 | 975 | #ifdef CONFIG_SMP |
5cc389bc PZ |
976 | /* |
977 | * This is how migration works: | |
978 | * | |
979 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
980 | * stop_one_cpu(). | |
981 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
982 | * off the CPU) | |
983 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
984 | * 4) if it's in the wrong runqueue then the migration thread removes | |
985 | * it and puts it into the right queue. | |
986 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
987 | * is done. | |
988 | */ | |
989 | ||
990 | /* | |
991 | * move_queued_task - move a queued task to new rq. | |
992 | * | |
993 | * Returns (locked) new rq. Old rq's lock is released. | |
994 | */ | |
5e16bbc2 | 995 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 996 | { |
5cc389bc PZ |
997 | lockdep_assert_held(&rq->lock); |
998 | ||
5cc389bc | 999 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 1000 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
1001 | set_task_cpu(p, new_cpu); |
1002 | raw_spin_unlock(&rq->lock); | |
1003 | ||
1004 | rq = cpu_rq(new_cpu); | |
1005 | ||
1006 | raw_spin_lock(&rq->lock); | |
1007 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 1008 | enqueue_task(rq, p, 0); |
3ea94de1 | 1009 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
1010 | check_preempt_curr(rq, p, 0); |
1011 | ||
1012 | return rq; | |
1013 | } | |
1014 | ||
1015 | struct migration_arg { | |
1016 | struct task_struct *task; | |
1017 | int dest_cpu; | |
1018 | }; | |
1019 | ||
1020 | /* | |
1021 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1022 | * this because either it can't run here any more (set_cpus_allowed() | |
1023 | * away from this CPU, or CPU going down), or because we're | |
1024 | * attempting to rebalance this task on exec (sched_exec). | |
1025 | * | |
1026 | * So we race with normal scheduler movements, but that's OK, as long | |
1027 | * as the task is no longer on this CPU. | |
5cc389bc | 1028 | */ |
5e16bbc2 | 1029 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1030 | { |
5cc389bc | 1031 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1032 | return rq; |
5cc389bc PZ |
1033 | |
1034 | /* Affinity changed (again). */ | |
1035 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1036 | return rq; |
5cc389bc | 1037 | |
5e16bbc2 PZ |
1038 | rq = move_queued_task(rq, p, dest_cpu); |
1039 | ||
1040 | return rq; | |
5cc389bc PZ |
1041 | } |
1042 | ||
1043 | /* | |
1044 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1045 | * and performs thread migration by bumping thread off CPU then | |
1046 | * 'pushing' onto another runqueue. | |
1047 | */ | |
1048 | static int migration_cpu_stop(void *data) | |
1049 | { | |
1050 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1051 | struct task_struct *p = arg->task; |
1052 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1053 | |
1054 | /* | |
1055 | * The original target cpu might have gone down and we might | |
1056 | * be on another cpu but it doesn't matter. | |
1057 | */ | |
1058 | local_irq_disable(); | |
1059 | /* | |
1060 | * We need to explicitly wake pending tasks before running | |
1061 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1062 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1063 | */ | |
1064 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1065 | |
1066 | raw_spin_lock(&p->pi_lock); | |
1067 | raw_spin_lock(&rq->lock); | |
1068 | /* | |
1069 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1070 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1071 | * we're holding p->pi_lock. | |
1072 | */ | |
bf89a304 CC |
1073 | if (task_rq(p) == rq) { |
1074 | if (task_on_rq_queued(p)) | |
1075 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1076 | else | |
1077 | p->wake_cpu = arg->dest_cpu; | |
1078 | } | |
5e16bbc2 PZ |
1079 | raw_spin_unlock(&rq->lock); |
1080 | raw_spin_unlock(&p->pi_lock); | |
1081 | ||
5cc389bc PZ |
1082 | local_irq_enable(); |
1083 | return 0; | |
1084 | } | |
1085 | ||
c5b28038 PZ |
1086 | /* |
1087 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1088 | * actually call this function. | |
1089 | */ | |
1090 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1091 | { |
5cc389bc PZ |
1092 | cpumask_copy(&p->cpus_allowed, new_mask); |
1093 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1094 | } | |
1095 | ||
c5b28038 PZ |
1096 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1097 | { | |
6c37067e PZ |
1098 | struct rq *rq = task_rq(p); |
1099 | bool queued, running; | |
1100 | ||
c5b28038 | 1101 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1102 | |
1103 | queued = task_on_rq_queued(p); | |
1104 | running = task_current(rq, p); | |
1105 | ||
1106 | if (queued) { | |
1107 | /* | |
1108 | * Because __kthread_bind() calls this on blocked tasks without | |
1109 | * holding rq->lock. | |
1110 | */ | |
1111 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1112 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1113 | } |
1114 | if (running) | |
1115 | put_prev_task(rq, p); | |
1116 | ||
c5b28038 | 1117 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1118 | |
6c37067e | 1119 | if (queued) |
1de64443 | 1120 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 1121 | if (running) |
b2bf6c31 | 1122 | set_curr_task(rq, p); |
c5b28038 PZ |
1123 | } |
1124 | ||
5cc389bc PZ |
1125 | /* |
1126 | * Change a given task's CPU affinity. Migrate the thread to a | |
1127 | * proper CPU and schedule it away if the CPU it's executing on | |
1128 | * is removed from the allowed bitmask. | |
1129 | * | |
1130 | * NOTE: the caller must have a valid reference to the task, the | |
1131 | * task must not exit() & deallocate itself prematurely. The | |
1132 | * call is not atomic; no spinlocks may be held. | |
1133 | */ | |
25834c73 PZ |
1134 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1135 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1136 | { |
e9d867a6 | 1137 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1138 | unsigned int dest_cpu; |
eb580751 PZ |
1139 | struct rq_flags rf; |
1140 | struct rq *rq; | |
5cc389bc PZ |
1141 | int ret = 0; |
1142 | ||
eb580751 | 1143 | rq = task_rq_lock(p, &rf); |
5cc389bc | 1144 | |
e9d867a6 PZI |
1145 | if (p->flags & PF_KTHREAD) { |
1146 | /* | |
1147 | * Kernel threads are allowed on online && !active CPUs | |
1148 | */ | |
1149 | cpu_valid_mask = cpu_online_mask; | |
1150 | } | |
1151 | ||
25834c73 PZ |
1152 | /* |
1153 | * Must re-check here, to close a race against __kthread_bind(), | |
1154 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1155 | */ | |
1156 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1157 | ret = -EINVAL; | |
1158 | goto out; | |
1159 | } | |
1160 | ||
5cc389bc PZ |
1161 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1162 | goto out; | |
1163 | ||
e9d867a6 | 1164 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1165 | ret = -EINVAL; |
1166 | goto out; | |
1167 | } | |
1168 | ||
1169 | do_set_cpus_allowed(p, new_mask); | |
1170 | ||
e9d867a6 PZI |
1171 | if (p->flags & PF_KTHREAD) { |
1172 | /* | |
1173 | * For kernel threads that do indeed end up on online && | |
1174 | * !active we want to ensure they are strict per-cpu threads. | |
1175 | */ | |
1176 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1177 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1178 | p->nr_cpus_allowed != 1); | |
1179 | } | |
1180 | ||
5cc389bc PZ |
1181 | /* Can the task run on the task's current CPU? If so, we're done */ |
1182 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1183 | goto out; | |
1184 | ||
e9d867a6 | 1185 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1186 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1187 | struct migration_arg arg = { p, dest_cpu }; | |
1188 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1189 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1190 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1191 | tlb_migrate_finish(p->mm); | |
1192 | return 0; | |
cbce1a68 PZ |
1193 | } else if (task_on_rq_queued(p)) { |
1194 | /* | |
1195 | * OK, since we're going to drop the lock immediately | |
1196 | * afterwards anyway. | |
1197 | */ | |
d8ac8971 | 1198 | rq_unpin_lock(rq, &rf); |
5e16bbc2 | 1199 | rq = move_queued_task(rq, p, dest_cpu); |
d8ac8971 | 1200 | rq_repin_lock(rq, &rf); |
cbce1a68 | 1201 | } |
5cc389bc | 1202 | out: |
eb580751 | 1203 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1204 | |
1205 | return ret; | |
1206 | } | |
25834c73 PZ |
1207 | |
1208 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1209 | { | |
1210 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1211 | } | |
5cc389bc PZ |
1212 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1213 | ||
dd41f596 | 1214 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1215 | { |
e2912009 PZ |
1216 | #ifdef CONFIG_SCHED_DEBUG |
1217 | /* | |
1218 | * We should never call set_task_cpu() on a blocked task, | |
1219 | * ttwu() will sort out the placement. | |
1220 | */ | |
077614ee | 1221 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1222 | !p->on_rq); |
0122ec5b | 1223 | |
3ea94de1 JP |
1224 | /* |
1225 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1226 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1227 | * time relying on p->on_rq. | |
1228 | */ | |
1229 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1230 | p->sched_class == &fair_sched_class && | |
1231 | (p->on_rq && !task_on_rq_migrating(p))); | |
1232 | ||
0122ec5b | 1233 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1234 | /* |
1235 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1236 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1237 | * | |
1238 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1239 | * see task_group(). |
6c6c54e1 PZ |
1240 | * |
1241 | * Furthermore, all task_rq users should acquire both locks, see | |
1242 | * task_rq_lock(). | |
1243 | */ | |
0122ec5b PZ |
1244 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1245 | lockdep_is_held(&task_rq(p)->lock))); | |
1246 | #endif | |
e2912009 PZ |
1247 | #endif |
1248 | ||
de1d7286 | 1249 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1250 | |
0c69774e | 1251 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1252 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1253 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1254 | p->se.nr_migrations++; |
ff303e66 | 1255 | perf_event_task_migrate(p); |
0c69774e | 1256 | } |
dd41f596 IM |
1257 | |
1258 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1259 | } |
1260 | ||
ac66f547 PZ |
1261 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1262 | { | |
da0c1e65 | 1263 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1264 | struct rq *src_rq, *dst_rq; |
1265 | ||
1266 | src_rq = task_rq(p); | |
1267 | dst_rq = cpu_rq(cpu); | |
1268 | ||
3ea94de1 | 1269 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1270 | deactivate_task(src_rq, p, 0); |
1271 | set_task_cpu(p, cpu); | |
1272 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1273 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1274 | check_preempt_curr(dst_rq, p, 0); |
1275 | } else { | |
1276 | /* | |
1277 | * Task isn't running anymore; make it appear like we migrated | |
1278 | * it before it went to sleep. This means on wakeup we make the | |
a1fd4656 | 1279 | * previous cpu our target instead of where it really is. |
ac66f547 PZ |
1280 | */ |
1281 | p->wake_cpu = cpu; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | struct migration_swap_arg { | |
1286 | struct task_struct *src_task, *dst_task; | |
1287 | int src_cpu, dst_cpu; | |
1288 | }; | |
1289 | ||
1290 | static int migrate_swap_stop(void *data) | |
1291 | { | |
1292 | struct migration_swap_arg *arg = data; | |
1293 | struct rq *src_rq, *dst_rq; | |
1294 | int ret = -EAGAIN; | |
1295 | ||
62694cd5 PZ |
1296 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1297 | return -EAGAIN; | |
1298 | ||
ac66f547 PZ |
1299 | src_rq = cpu_rq(arg->src_cpu); |
1300 | dst_rq = cpu_rq(arg->dst_cpu); | |
1301 | ||
74602315 PZ |
1302 | double_raw_lock(&arg->src_task->pi_lock, |
1303 | &arg->dst_task->pi_lock); | |
ac66f547 | 1304 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1305 | |
ac66f547 PZ |
1306 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1307 | goto unlock; | |
1308 | ||
1309 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1310 | goto unlock; | |
1311 | ||
1312 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1313 | goto unlock; | |
1314 | ||
1315 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1316 | goto unlock; | |
1317 | ||
1318 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1319 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1320 | ||
1321 | ret = 0; | |
1322 | ||
1323 | unlock: | |
1324 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1325 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1326 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1327 | |
1328 | return ret; | |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * Cross migrate two tasks | |
1333 | */ | |
1334 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1335 | { | |
1336 | struct migration_swap_arg arg; | |
1337 | int ret = -EINVAL; | |
1338 | ||
ac66f547 PZ |
1339 | arg = (struct migration_swap_arg){ |
1340 | .src_task = cur, | |
1341 | .src_cpu = task_cpu(cur), | |
1342 | .dst_task = p, | |
1343 | .dst_cpu = task_cpu(p), | |
1344 | }; | |
1345 | ||
1346 | if (arg.src_cpu == arg.dst_cpu) | |
1347 | goto out; | |
1348 | ||
6acce3ef PZ |
1349 | /* |
1350 | * These three tests are all lockless; this is OK since all of them | |
1351 | * will be re-checked with proper locks held further down the line. | |
1352 | */ | |
ac66f547 PZ |
1353 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1354 | goto out; | |
1355 | ||
1356 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1357 | goto out; | |
1358 | ||
1359 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1360 | goto out; | |
1361 | ||
286549dc | 1362 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1363 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1364 | ||
1365 | out: | |
ac66f547 PZ |
1366 | return ret; |
1367 | } | |
1368 | ||
1da177e4 LT |
1369 | /* |
1370 | * wait_task_inactive - wait for a thread to unschedule. | |
1371 | * | |
85ba2d86 RM |
1372 | * If @match_state is nonzero, it's the @p->state value just checked and |
1373 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1374 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1375 | * we return a positive number (its total switch count). If a second call | |
1376 | * a short while later returns the same number, the caller can be sure that | |
1377 | * @p has remained unscheduled the whole time. | |
1378 | * | |
1da177e4 LT |
1379 | * The caller must ensure that the task *will* unschedule sometime soon, |
1380 | * else this function might spin for a *long* time. This function can't | |
1381 | * be called with interrupts off, or it may introduce deadlock with | |
1382 | * smp_call_function() if an IPI is sent by the same process we are | |
1383 | * waiting to become inactive. | |
1384 | */ | |
85ba2d86 | 1385 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1386 | { |
da0c1e65 | 1387 | int running, queued; |
eb580751 | 1388 | struct rq_flags rf; |
85ba2d86 | 1389 | unsigned long ncsw; |
70b97a7f | 1390 | struct rq *rq; |
1da177e4 | 1391 | |
3a5c359a AK |
1392 | for (;;) { |
1393 | /* | |
1394 | * We do the initial early heuristics without holding | |
1395 | * any task-queue locks at all. We'll only try to get | |
1396 | * the runqueue lock when things look like they will | |
1397 | * work out! | |
1398 | */ | |
1399 | rq = task_rq(p); | |
fa490cfd | 1400 | |
3a5c359a AK |
1401 | /* |
1402 | * If the task is actively running on another CPU | |
1403 | * still, just relax and busy-wait without holding | |
1404 | * any locks. | |
1405 | * | |
1406 | * NOTE! Since we don't hold any locks, it's not | |
1407 | * even sure that "rq" stays as the right runqueue! | |
1408 | * But we don't care, since "task_running()" will | |
1409 | * return false if the runqueue has changed and p | |
1410 | * is actually now running somewhere else! | |
1411 | */ | |
85ba2d86 RM |
1412 | while (task_running(rq, p)) { |
1413 | if (match_state && unlikely(p->state != match_state)) | |
1414 | return 0; | |
3a5c359a | 1415 | cpu_relax(); |
85ba2d86 | 1416 | } |
fa490cfd | 1417 | |
3a5c359a AK |
1418 | /* |
1419 | * Ok, time to look more closely! We need the rq | |
1420 | * lock now, to be *sure*. If we're wrong, we'll | |
1421 | * just go back and repeat. | |
1422 | */ | |
eb580751 | 1423 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1424 | trace_sched_wait_task(p); |
3a5c359a | 1425 | running = task_running(rq, p); |
da0c1e65 | 1426 | queued = task_on_rq_queued(p); |
85ba2d86 | 1427 | ncsw = 0; |
f31e11d8 | 1428 | if (!match_state || p->state == match_state) |
93dcf55f | 1429 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1430 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1431 | |
85ba2d86 RM |
1432 | /* |
1433 | * If it changed from the expected state, bail out now. | |
1434 | */ | |
1435 | if (unlikely(!ncsw)) | |
1436 | break; | |
1437 | ||
3a5c359a AK |
1438 | /* |
1439 | * Was it really running after all now that we | |
1440 | * checked with the proper locks actually held? | |
1441 | * | |
1442 | * Oops. Go back and try again.. | |
1443 | */ | |
1444 | if (unlikely(running)) { | |
1445 | cpu_relax(); | |
1446 | continue; | |
1447 | } | |
fa490cfd | 1448 | |
3a5c359a AK |
1449 | /* |
1450 | * It's not enough that it's not actively running, | |
1451 | * it must be off the runqueue _entirely_, and not | |
1452 | * preempted! | |
1453 | * | |
80dd99b3 | 1454 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1455 | * running right now), it's preempted, and we should |
1456 | * yield - it could be a while. | |
1457 | */ | |
da0c1e65 | 1458 | if (unlikely(queued)) { |
8b0e1953 | 1459 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1460 | |
1461 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1462 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1463 | continue; |
1464 | } | |
fa490cfd | 1465 | |
3a5c359a AK |
1466 | /* |
1467 | * Ahh, all good. It wasn't running, and it wasn't | |
1468 | * runnable, which means that it will never become | |
1469 | * running in the future either. We're all done! | |
1470 | */ | |
1471 | break; | |
1472 | } | |
85ba2d86 RM |
1473 | |
1474 | return ncsw; | |
1da177e4 LT |
1475 | } |
1476 | ||
1477 | /*** | |
1478 | * kick_process - kick a running thread to enter/exit the kernel | |
1479 | * @p: the to-be-kicked thread | |
1480 | * | |
1481 | * Cause a process which is running on another CPU to enter | |
1482 | * kernel-mode, without any delay. (to get signals handled.) | |
1483 | * | |
25985edc | 1484 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1485 | * because all it wants to ensure is that the remote task enters |
1486 | * the kernel. If the IPI races and the task has been migrated | |
1487 | * to another CPU then no harm is done and the purpose has been | |
1488 | * achieved as well. | |
1489 | */ | |
36c8b586 | 1490 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1491 | { |
1492 | int cpu; | |
1493 | ||
1494 | preempt_disable(); | |
1495 | cpu = task_cpu(p); | |
1496 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1497 | smp_send_reschedule(cpu); | |
1498 | preempt_enable(); | |
1499 | } | |
b43e3521 | 1500 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1501 | |
30da688e | 1502 | /* |
013fdb80 | 1503 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1504 | * |
1505 | * A few notes on cpu_active vs cpu_online: | |
1506 | * | |
1507 | * - cpu_active must be a subset of cpu_online | |
1508 | * | |
1509 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1510 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
1511 | * cpu isn't yet part of the sched domains, and balancing will not | |
1512 | * see it. | |
1513 | * | |
1514 | * - on cpu-down we clear cpu_active() to mask the sched domains and | |
1515 | * avoid the load balancer to place new tasks on the to be removed | |
1516 | * cpu. Existing tasks will remain running there and will be taken | |
1517 | * off. | |
1518 | * | |
1519 | * This means that fallback selection must not select !active CPUs. | |
1520 | * And can assume that any active CPU must be online. Conversely | |
1521 | * select_task_rq() below may allow selection of !active CPUs in order | |
1522 | * to satisfy the above rules. | |
30da688e | 1523 | */ |
5da9a0fb PZ |
1524 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1525 | { | |
aa00d89c TC |
1526 | int nid = cpu_to_node(cpu); |
1527 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1528 | enum { cpuset, possible, fail } state = cpuset; |
1529 | int dest_cpu; | |
5da9a0fb | 1530 | |
aa00d89c TC |
1531 | /* |
1532 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1533 | * will return -1. There is no cpu on the node, and we should | |
1534 | * select the cpu on the other node. | |
1535 | */ | |
1536 | if (nid != -1) { | |
1537 | nodemask = cpumask_of_node(nid); | |
1538 | ||
1539 | /* Look for allowed, online CPU in same node. */ | |
1540 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1541 | if (!cpu_active(dest_cpu)) |
1542 | continue; | |
1543 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1544 | return dest_cpu; | |
1545 | } | |
2baab4e9 | 1546 | } |
5da9a0fb | 1547 | |
2baab4e9 PZ |
1548 | for (;;) { |
1549 | /* Any allowed, online CPU? */ | |
e3831edd | 1550 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
feb245e3 TH |
1551 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
1552 | continue; | |
1553 | if (!cpu_online(dest_cpu)) | |
2baab4e9 PZ |
1554 | continue; |
1555 | goto out; | |
1556 | } | |
5da9a0fb | 1557 | |
e73e85f0 | 1558 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1559 | switch (state) { |
1560 | case cpuset: | |
e73e85f0 ON |
1561 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1562 | cpuset_cpus_allowed_fallback(p); | |
1563 | state = possible; | |
1564 | break; | |
1565 | } | |
1566 | /* fall-through */ | |
2baab4e9 PZ |
1567 | case possible: |
1568 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1569 | state = fail; | |
1570 | break; | |
1571 | ||
1572 | case fail: | |
1573 | BUG(); | |
1574 | break; | |
1575 | } | |
1576 | } | |
1577 | ||
1578 | out: | |
1579 | if (state != cpuset) { | |
1580 | /* | |
1581 | * Don't tell them about moving exiting tasks or | |
1582 | * kernel threads (both mm NULL), since they never | |
1583 | * leave kernel. | |
1584 | */ | |
1585 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1586 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1587 | task_pid_nr(p), p->comm, cpu); |
1588 | } | |
5da9a0fb PZ |
1589 | } |
1590 | ||
1591 | return dest_cpu; | |
1592 | } | |
1593 | ||
e2912009 | 1594 | /* |
013fdb80 | 1595 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1596 | */ |
970b13ba | 1597 | static inline |
ac66f547 | 1598 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1599 | { |
cbce1a68 PZ |
1600 | lockdep_assert_held(&p->pi_lock); |
1601 | ||
50605ffb | 1602 | if (tsk_nr_cpus_allowed(p) > 1) |
6c1d9410 | 1603 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 PZI |
1604 | else |
1605 | cpu = cpumask_any(tsk_cpus_allowed(p)); | |
e2912009 PZ |
1606 | |
1607 | /* | |
1608 | * In order not to call set_task_cpu() on a blocking task we need | |
1609 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1610 | * cpu. | |
1611 | * | |
1612 | * Since this is common to all placement strategies, this lives here. | |
1613 | * | |
1614 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1615 | * not worry about this generic constraint ] | |
1616 | */ | |
fa17b507 | 1617 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1618 | !cpu_online(cpu))) |
5da9a0fb | 1619 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1620 | |
1621 | return cpu; | |
970b13ba | 1622 | } |
09a40af5 MG |
1623 | |
1624 | static void update_avg(u64 *avg, u64 sample) | |
1625 | { | |
1626 | s64 diff = sample - *avg; | |
1627 | *avg += diff >> 3; | |
1628 | } | |
25834c73 PZ |
1629 | |
1630 | #else | |
1631 | ||
1632 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1633 | const struct cpumask *new_mask, bool check) | |
1634 | { | |
1635 | return set_cpus_allowed_ptr(p, new_mask); | |
1636 | } | |
1637 | ||
5cc389bc | 1638 | #endif /* CONFIG_SMP */ |
970b13ba | 1639 | |
d7c01d27 | 1640 | static void |
b84cb5df | 1641 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1642 | { |
4fa8d299 | 1643 | struct rq *rq; |
b84cb5df | 1644 | |
4fa8d299 JP |
1645 | if (!schedstat_enabled()) |
1646 | return; | |
1647 | ||
1648 | rq = this_rq(); | |
d7c01d27 | 1649 | |
4fa8d299 JP |
1650 | #ifdef CONFIG_SMP |
1651 | if (cpu == rq->cpu) { | |
ae92882e JP |
1652 | schedstat_inc(rq->ttwu_local); |
1653 | schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1654 | } else { |
1655 | struct sched_domain *sd; | |
1656 | ||
ae92882e | 1657 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1658 | rcu_read_lock(); |
4fa8d299 | 1659 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1660 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
ae92882e | 1661 | schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1662 | break; |
1663 | } | |
1664 | } | |
057f3fad | 1665 | rcu_read_unlock(); |
d7c01d27 | 1666 | } |
f339b9dc PZ |
1667 | |
1668 | if (wake_flags & WF_MIGRATED) | |
ae92882e | 1669 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1670 | #endif /* CONFIG_SMP */ |
1671 | ||
ae92882e JP |
1672 | schedstat_inc(rq->ttwu_count); |
1673 | schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1674 | |
1675 | if (wake_flags & WF_SYNC) | |
ae92882e | 1676 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1677 | } |
1678 | ||
1de64443 | 1679 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1680 | { |
9ed3811a | 1681 | activate_task(rq, p, en_flags); |
da0c1e65 | 1682 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1683 | |
1684 | /* if a worker is waking up, notify workqueue */ | |
1685 | if (p->flags & PF_WQ_WORKER) | |
1686 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1687 | } |
1688 | ||
23f41eeb PZ |
1689 | /* |
1690 | * Mark the task runnable and perform wakeup-preemption. | |
1691 | */ | |
e7904a28 | 1692 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1693 | struct rq_flags *rf) |
9ed3811a | 1694 | { |
9ed3811a | 1695 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1696 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1697 | trace_sched_wakeup(p); |
1698 | ||
9ed3811a | 1699 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1700 | if (p->sched_class->task_woken) { |
1701 | /* | |
cbce1a68 PZ |
1702 | * Our task @p is fully woken up and running; so its safe to |
1703 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1704 | */ |
d8ac8971 | 1705 | rq_unpin_lock(rq, rf); |
9ed3811a | 1706 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 1707 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 1708 | } |
9ed3811a | 1709 | |
e69c6341 | 1710 | if (rq->idle_stamp) { |
78becc27 | 1711 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1712 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1713 | |
abfafa54 JL |
1714 | update_avg(&rq->avg_idle, delta); |
1715 | ||
1716 | if (rq->avg_idle > max) | |
9ed3811a | 1717 | rq->avg_idle = max; |
abfafa54 | 1718 | |
9ed3811a TH |
1719 | rq->idle_stamp = 0; |
1720 | } | |
1721 | #endif | |
1722 | } | |
1723 | ||
c05fbafb | 1724 | static void |
e7904a28 | 1725 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1726 | struct rq_flags *rf) |
c05fbafb | 1727 | { |
b5179ac7 PZ |
1728 | int en_flags = ENQUEUE_WAKEUP; |
1729 | ||
cbce1a68 PZ |
1730 | lockdep_assert_held(&rq->lock); |
1731 | ||
c05fbafb PZ |
1732 | #ifdef CONFIG_SMP |
1733 | if (p->sched_contributes_to_load) | |
1734 | rq->nr_uninterruptible--; | |
b5179ac7 | 1735 | |
b5179ac7 | 1736 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1737 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1738 | #endif |
1739 | ||
b5179ac7 | 1740 | ttwu_activate(rq, p, en_flags); |
d8ac8971 | 1741 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1746 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1747 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1748 | * the task is still ->on_rq. | |
1749 | */ | |
1750 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1751 | { | |
eb580751 | 1752 | struct rq_flags rf; |
c05fbafb PZ |
1753 | struct rq *rq; |
1754 | int ret = 0; | |
1755 | ||
eb580751 | 1756 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1757 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1758 | /* check_preempt_curr() may use rq clock */ |
1759 | update_rq_clock(rq); | |
d8ac8971 | 1760 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
1761 | ret = 1; |
1762 | } | |
eb580751 | 1763 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1764 | |
1765 | return ret; | |
1766 | } | |
1767 | ||
317f3941 | 1768 | #ifdef CONFIG_SMP |
e3baac47 | 1769 | void sched_ttwu_pending(void) |
317f3941 PZ |
1770 | { |
1771 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1772 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1773 | struct task_struct *p; | |
e3baac47 | 1774 | unsigned long flags; |
d8ac8971 | 1775 | struct rq_flags rf; |
317f3941 | 1776 | |
e3baac47 PZ |
1777 | if (!llist) |
1778 | return; | |
1779 | ||
1780 | raw_spin_lock_irqsave(&rq->lock, flags); | |
d8ac8971 | 1781 | rq_pin_lock(rq, &rf); |
317f3941 | 1782 | |
fa14ff4a | 1783 | while (llist) { |
b7e7ade3 PZ |
1784 | int wake_flags = 0; |
1785 | ||
fa14ff4a PZ |
1786 | p = llist_entry(llist, struct task_struct, wake_entry); |
1787 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1788 | |
1789 | if (p->sched_remote_wakeup) | |
1790 | wake_flags = WF_MIGRATED; | |
1791 | ||
d8ac8971 | 1792 | ttwu_do_activate(rq, p, wake_flags, &rf); |
317f3941 PZ |
1793 | } |
1794 | ||
d8ac8971 | 1795 | rq_unpin_lock(rq, &rf); |
e3baac47 | 1796 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1797 | } |
1798 | ||
1799 | void scheduler_ipi(void) | |
1800 | { | |
f27dde8d PZ |
1801 | /* |
1802 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1803 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1804 | * this IPI. | |
1805 | */ | |
8cb75e0c | 1806 | preempt_fold_need_resched(); |
f27dde8d | 1807 | |
fd2ac4f4 | 1808 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1809 | return; |
1810 | ||
1811 | /* | |
1812 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1813 | * traditionally all their work was done from the interrupt return | |
1814 | * path. Now that we actually do some work, we need to make sure | |
1815 | * we do call them. | |
1816 | * | |
1817 | * Some archs already do call them, luckily irq_enter/exit nest | |
1818 | * properly. | |
1819 | * | |
1820 | * Arguably we should visit all archs and update all handlers, | |
1821 | * however a fair share of IPIs are still resched only so this would | |
1822 | * somewhat pessimize the simple resched case. | |
1823 | */ | |
1824 | irq_enter(); | |
fa14ff4a | 1825 | sched_ttwu_pending(); |
ca38062e SS |
1826 | |
1827 | /* | |
1828 | * Check if someone kicked us for doing the nohz idle load balance. | |
1829 | */ | |
873b4c65 | 1830 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1831 | this_rq()->idle_balance = 1; |
ca38062e | 1832 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1833 | } |
c5d753a5 | 1834 | irq_exit(); |
317f3941 PZ |
1835 | } |
1836 | ||
b7e7ade3 | 1837 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1838 | { |
e3baac47 PZ |
1839 | struct rq *rq = cpu_rq(cpu); |
1840 | ||
b7e7ade3 PZ |
1841 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1842 | ||
e3baac47 PZ |
1843 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1844 | if (!set_nr_if_polling(rq->idle)) | |
1845 | smp_send_reschedule(cpu); | |
1846 | else | |
1847 | trace_sched_wake_idle_without_ipi(cpu); | |
1848 | } | |
317f3941 | 1849 | } |
d6aa8f85 | 1850 | |
f6be8af1 CL |
1851 | void wake_up_if_idle(int cpu) |
1852 | { | |
1853 | struct rq *rq = cpu_rq(cpu); | |
1854 | unsigned long flags; | |
1855 | ||
fd7de1e8 AL |
1856 | rcu_read_lock(); |
1857 | ||
1858 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1859 | goto out; | |
f6be8af1 CL |
1860 | |
1861 | if (set_nr_if_polling(rq->idle)) { | |
1862 | trace_sched_wake_idle_without_ipi(cpu); | |
1863 | } else { | |
1864 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1865 | if (is_idle_task(rq->curr)) | |
1866 | smp_send_reschedule(cpu); | |
1867 | /* Else cpu is not in idle, do nothing here */ | |
1868 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1869 | } | |
fd7de1e8 AL |
1870 | |
1871 | out: | |
1872 | rcu_read_unlock(); | |
f6be8af1 CL |
1873 | } |
1874 | ||
39be3501 | 1875 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1876 | { |
1877 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1878 | } | |
d6aa8f85 | 1879 | #endif /* CONFIG_SMP */ |
317f3941 | 1880 | |
b5179ac7 | 1881 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1882 | { |
1883 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 1884 | struct rq_flags rf; |
c05fbafb | 1885 | |
17d9f311 | 1886 | #if defined(CONFIG_SMP) |
39be3501 | 1887 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1888 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
b7e7ade3 | 1889 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1890 | return; |
1891 | } | |
1892 | #endif | |
1893 | ||
c05fbafb | 1894 | raw_spin_lock(&rq->lock); |
d8ac8971 MF |
1895 | rq_pin_lock(rq, &rf); |
1896 | ttwu_do_activate(rq, p, wake_flags, &rf); | |
1897 | rq_unpin_lock(rq, &rf); | |
c05fbafb | 1898 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1899 | } |
1900 | ||
8643cda5 PZ |
1901 | /* |
1902 | * Notes on Program-Order guarantees on SMP systems. | |
1903 | * | |
1904 | * MIGRATION | |
1905 | * | |
1906 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1907 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1908 | * execution on its new cpu [c1]. | |
1909 | * | |
1910 | * For migration (of runnable tasks) this is provided by the following means: | |
1911 | * | |
1912 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1913 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1914 | * rq(c1)->lock (if not at the same time, then in that order). | |
1915 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1916 | * | |
1917 | * Transitivity guarantees that B happens after A and C after B. | |
1918 | * Note: we only require RCpc transitivity. | |
1919 | * Note: the cpu doing B need not be c0 or c1 | |
1920 | * | |
1921 | * Example: | |
1922 | * | |
1923 | * CPU0 CPU1 CPU2 | |
1924 | * | |
1925 | * LOCK rq(0)->lock | |
1926 | * sched-out X | |
1927 | * sched-in Y | |
1928 | * UNLOCK rq(0)->lock | |
1929 | * | |
1930 | * LOCK rq(0)->lock // orders against CPU0 | |
1931 | * dequeue X | |
1932 | * UNLOCK rq(0)->lock | |
1933 | * | |
1934 | * LOCK rq(1)->lock | |
1935 | * enqueue X | |
1936 | * UNLOCK rq(1)->lock | |
1937 | * | |
1938 | * LOCK rq(1)->lock // orders against CPU2 | |
1939 | * sched-out Z | |
1940 | * sched-in X | |
1941 | * UNLOCK rq(1)->lock | |
1942 | * | |
1943 | * | |
1944 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1945 | * | |
1946 | * For blocking we (obviously) need to provide the same guarantee as for | |
1947 | * migration. However the means are completely different as there is no lock | |
1948 | * chain to provide order. Instead we do: | |
1949 | * | |
1950 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1951 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1952 | * |
1953 | * Example: | |
1954 | * | |
1955 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1956 | * | |
1957 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1958 | * dequeue X | |
1959 | * sched-out X | |
1960 | * smp_store_release(X->on_cpu, 0); | |
1961 | * | |
1f03e8d2 | 1962 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1963 | * X->state = WAKING |
1964 | * set_task_cpu(X,2) | |
1965 | * | |
1966 | * LOCK rq(2)->lock | |
1967 | * enqueue X | |
1968 | * X->state = RUNNING | |
1969 | * UNLOCK rq(2)->lock | |
1970 | * | |
1971 | * LOCK rq(2)->lock // orders against CPU1 | |
1972 | * sched-out Z | |
1973 | * sched-in X | |
1974 | * UNLOCK rq(2)->lock | |
1975 | * | |
1976 | * UNLOCK X->pi_lock | |
1977 | * UNLOCK rq(0)->lock | |
1978 | * | |
1979 | * | |
1980 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1981 | * must observe the state that lead to our wakeup. That is, not only must our | |
1982 | * task observe its own prior state, it must also observe the stores prior to | |
1983 | * its wakeup. | |
1984 | * | |
1985 | * This means that any means of doing remote wakeups must order the CPU doing | |
1986 | * the wakeup against the CPU the task is going to end up running on. This, | |
1987 | * however, is already required for the regular Program-Order guarantee above, | |
1f03e8d2 | 1988 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
8643cda5 PZ |
1989 | * |
1990 | */ | |
1991 | ||
9ed3811a | 1992 | /** |
1da177e4 | 1993 | * try_to_wake_up - wake up a thread |
9ed3811a | 1994 | * @p: the thread to be awakened |
1da177e4 | 1995 | * @state: the mask of task states that can be woken |
9ed3811a | 1996 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 1997 | * |
a2250238 | 1998 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 1999 | * |
a2250238 PZ |
2000 | * If the task was not queued/runnable, also place it back on a runqueue. |
2001 | * | |
2002 | * Atomic against schedule() which would dequeue a task, also see | |
2003 | * set_current_state(). | |
2004 | * | |
2005 | * Return: %true if @p->state changes (an actual wakeup was done), | |
2006 | * %false otherwise. | |
1da177e4 | 2007 | */ |
e4a52bcb PZ |
2008 | static int |
2009 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2010 | { |
1da177e4 | 2011 | unsigned long flags; |
c05fbafb | 2012 | int cpu, success = 0; |
2398f2c6 | 2013 | |
e0acd0a6 ON |
2014 | /* |
2015 | * If we are going to wake up a thread waiting for CONDITION we | |
2016 | * need to ensure that CONDITION=1 done by the caller can not be | |
2017 | * reordered with p->state check below. This pairs with mb() in | |
2018 | * set_current_state() the waiting thread does. | |
2019 | */ | |
2020 | smp_mb__before_spinlock(); | |
013fdb80 | 2021 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2022 | if (!(p->state & state)) |
1da177e4 LT |
2023 | goto out; |
2024 | ||
fbd705a0 PZ |
2025 | trace_sched_waking(p); |
2026 | ||
c05fbafb | 2027 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2028 | cpu = task_cpu(p); |
1da177e4 | 2029 | |
135e8c92 BS |
2030 | /* |
2031 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2032 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2033 | * in smp_cond_load_acquire() below. | |
2034 | * | |
2035 | * sched_ttwu_pending() try_to_wake_up() | |
2036 | * [S] p->on_rq = 1; [L] P->state | |
2037 | * UNLOCK rq->lock -----. | |
2038 | * \ | |
2039 | * +--- RMB | |
2040 | * schedule() / | |
2041 | * LOCK rq->lock -----' | |
2042 | * UNLOCK rq->lock | |
2043 | * | |
2044 | * [task p] | |
2045 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq | |
2046 | * | |
2047 | * Pairs with the UNLOCK+LOCK on rq->lock from the | |
2048 | * last wakeup of our task and the schedule that got our task | |
2049 | * current. | |
2050 | */ | |
2051 | smp_rmb(); | |
c05fbafb PZ |
2052 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2053 | goto stat; | |
1da177e4 | 2054 | |
1da177e4 | 2055 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2056 | /* |
2057 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2058 | * possible to, falsely, observe p->on_cpu == 0. | |
2059 | * | |
2060 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2061 | * from the runqueue. | |
2062 | * | |
2063 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2064 | * UNLOCK rq->lock | |
2065 | * RMB | |
2066 | * LOCK rq->lock | |
2067 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2068 | * | |
2069 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2070 | * from the consecutive calls to schedule(); the first switching to our | |
2071 | * task, the second putting it to sleep. | |
2072 | */ | |
2073 | smp_rmb(); | |
2074 | ||
e9c84311 | 2075 | /* |
c05fbafb PZ |
2076 | * If the owning (remote) cpu is still in the middle of schedule() with |
2077 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
2078 | * |
2079 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2080 | * | |
2081 | * This ensures that tasks getting woken will be fully ordered against | |
2082 | * their previous state and preserve Program Order. | |
0970d299 | 2083 | */ |
1f03e8d2 | 2084 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2085 | |
a8e4f2ea | 2086 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2087 | p->state = TASK_WAKING; |
e7693a36 | 2088 | |
ac66f547 | 2089 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2090 | if (task_cpu(p) != cpu) { |
2091 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2092 | set_task_cpu(p, cpu); |
f339b9dc | 2093 | } |
1da177e4 | 2094 | #endif /* CONFIG_SMP */ |
1da177e4 | 2095 | |
b5179ac7 | 2096 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2097 | stat: |
4fa8d299 | 2098 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2099 | out: |
013fdb80 | 2100 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2101 | |
2102 | return success; | |
2103 | } | |
2104 | ||
21aa9af0 TH |
2105 | /** |
2106 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2107 | * @p: the thread to be awakened | |
9279e0d2 | 2108 | * @cookie: context's cookie for pinning |
21aa9af0 | 2109 | * |
2acca55e | 2110 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2111 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2112 | * the current task. |
21aa9af0 | 2113 | */ |
d8ac8971 | 2114 | static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf) |
21aa9af0 TH |
2115 | { |
2116 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2117 | |
383efcd0 TH |
2118 | if (WARN_ON_ONCE(rq != this_rq()) || |
2119 | WARN_ON_ONCE(p == current)) | |
2120 | return; | |
2121 | ||
21aa9af0 TH |
2122 | lockdep_assert_held(&rq->lock); |
2123 | ||
2acca55e | 2124 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2125 | /* |
2126 | * This is OK, because current is on_cpu, which avoids it being | |
2127 | * picked for load-balance and preemption/IRQs are still | |
2128 | * disabled avoiding further scheduler activity on it and we've | |
2129 | * not yet picked a replacement task. | |
2130 | */ | |
d8ac8971 | 2131 | rq_unpin_lock(rq, rf); |
2acca55e PZ |
2132 | raw_spin_unlock(&rq->lock); |
2133 | raw_spin_lock(&p->pi_lock); | |
2134 | raw_spin_lock(&rq->lock); | |
d8ac8971 | 2135 | rq_repin_lock(rq, rf); |
2acca55e PZ |
2136 | } |
2137 | ||
21aa9af0 | 2138 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2139 | goto out; |
21aa9af0 | 2140 | |
fbd705a0 PZ |
2141 | trace_sched_waking(p); |
2142 | ||
da0c1e65 | 2143 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2144 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2145 | ||
d8ac8971 | 2146 | ttwu_do_wakeup(rq, p, 0, rf); |
4fa8d299 | 2147 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2148 | out: |
2149 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2150 | } |
2151 | ||
50fa610a DH |
2152 | /** |
2153 | * wake_up_process - Wake up a specific process | |
2154 | * @p: The process to be woken up. | |
2155 | * | |
2156 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2157 | * processes. |
2158 | * | |
2159 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2160 | * |
2161 | * It may be assumed that this function implies a write memory barrier before | |
2162 | * changing the task state if and only if any tasks are woken up. | |
2163 | */ | |
7ad5b3a5 | 2164 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2165 | { |
9067ac85 | 2166 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2167 | } |
1da177e4 LT |
2168 | EXPORT_SYMBOL(wake_up_process); |
2169 | ||
7ad5b3a5 | 2170 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2171 | { |
2172 | return try_to_wake_up(p, state, 0); | |
2173 | } | |
2174 | ||
a5e7be3b JL |
2175 | /* |
2176 | * This function clears the sched_dl_entity static params. | |
2177 | */ | |
2178 | void __dl_clear_params(struct task_struct *p) | |
2179 | { | |
2180 | struct sched_dl_entity *dl_se = &p->dl; | |
2181 | ||
2182 | dl_se->dl_runtime = 0; | |
2183 | dl_se->dl_deadline = 0; | |
2184 | dl_se->dl_period = 0; | |
2185 | dl_se->flags = 0; | |
2186 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2187 | |
2188 | dl_se->dl_throttled = 0; | |
40767b0d | 2189 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2190 | } |
2191 | ||
1da177e4 LT |
2192 | /* |
2193 | * Perform scheduler related setup for a newly forked process p. | |
2194 | * p is forked by current. | |
dd41f596 IM |
2195 | * |
2196 | * __sched_fork() is basic setup used by init_idle() too: | |
2197 | */ | |
5e1576ed | 2198 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2199 | { |
fd2f4419 PZ |
2200 | p->on_rq = 0; |
2201 | ||
2202 | p->se.on_rq = 0; | |
dd41f596 IM |
2203 | p->se.exec_start = 0; |
2204 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2205 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2206 | p->se.nr_migrations = 0; |
da7a735e | 2207 | p->se.vruntime = 0; |
fd2f4419 | 2208 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2209 | |
ad936d86 BP |
2210 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2211 | p->se.cfs_rq = NULL; | |
2212 | #endif | |
2213 | ||
6cfb0d5d | 2214 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2215 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2216 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2217 | #endif |
476d139c | 2218 | |
aab03e05 | 2219 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2220 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2221 | __dl_clear_params(p); |
aab03e05 | 2222 | |
fa717060 | 2223 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2224 | p->rt.timeout = 0; |
2225 | p->rt.time_slice = sched_rr_timeslice; | |
2226 | p->rt.on_rq = 0; | |
2227 | p->rt.on_list = 0; | |
476d139c | 2228 | |
e107be36 AK |
2229 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2230 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2231 | #endif | |
cbee9f88 PZ |
2232 | |
2233 | #ifdef CONFIG_NUMA_BALANCING | |
2234 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2235 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2236 | p->mm->numa_scan_seq = 0; |
2237 | } | |
2238 | ||
5e1576ed RR |
2239 | if (clone_flags & CLONE_VM) |
2240 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2241 | else | |
2242 | p->numa_preferred_nid = -1; | |
2243 | ||
cbee9f88 PZ |
2244 | p->node_stamp = 0ULL; |
2245 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2246 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2247 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2248 | p->numa_faults = NULL; |
7e2703e6 RR |
2249 | p->last_task_numa_placement = 0; |
2250 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2251 | |
8c8a743c | 2252 | p->numa_group = NULL; |
cbee9f88 | 2253 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2254 | } |
2255 | ||
2a595721 SD |
2256 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2257 | ||
1a687c2e | 2258 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2259 | |
1a687c2e MG |
2260 | void set_numabalancing_state(bool enabled) |
2261 | { | |
2262 | if (enabled) | |
2a595721 | 2263 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2264 | else |
2a595721 | 2265 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2266 | } |
54a43d54 AK |
2267 | |
2268 | #ifdef CONFIG_PROC_SYSCTL | |
2269 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2270 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2271 | { | |
2272 | struct ctl_table t; | |
2273 | int err; | |
2a595721 | 2274 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2275 | |
2276 | if (write && !capable(CAP_SYS_ADMIN)) | |
2277 | return -EPERM; | |
2278 | ||
2279 | t = *table; | |
2280 | t.data = &state; | |
2281 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2282 | if (err < 0) | |
2283 | return err; | |
2284 | if (write) | |
2285 | set_numabalancing_state(state); | |
2286 | return err; | |
2287 | } | |
2288 | #endif | |
2289 | #endif | |
dd41f596 | 2290 | |
4698f88c JP |
2291 | #ifdef CONFIG_SCHEDSTATS |
2292 | ||
cb251765 | 2293 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2294 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2295 | |
cb251765 MG |
2296 | static void set_schedstats(bool enabled) |
2297 | { | |
2298 | if (enabled) | |
2299 | static_branch_enable(&sched_schedstats); | |
2300 | else | |
2301 | static_branch_disable(&sched_schedstats); | |
2302 | } | |
2303 | ||
2304 | void force_schedstat_enabled(void) | |
2305 | { | |
2306 | if (!schedstat_enabled()) { | |
2307 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2308 | static_branch_enable(&sched_schedstats); | |
2309 | } | |
2310 | } | |
2311 | ||
2312 | static int __init setup_schedstats(char *str) | |
2313 | { | |
2314 | int ret = 0; | |
2315 | if (!str) | |
2316 | goto out; | |
2317 | ||
4698f88c JP |
2318 | /* |
2319 | * This code is called before jump labels have been set up, so we can't | |
2320 | * change the static branch directly just yet. Instead set a temporary | |
2321 | * variable so init_schedstats() can do it later. | |
2322 | */ | |
cb251765 | 2323 | if (!strcmp(str, "enable")) { |
4698f88c | 2324 | __sched_schedstats = true; |
cb251765 MG |
2325 | ret = 1; |
2326 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2327 | __sched_schedstats = false; |
cb251765 MG |
2328 | ret = 1; |
2329 | } | |
2330 | out: | |
2331 | if (!ret) | |
2332 | pr_warn("Unable to parse schedstats=\n"); | |
2333 | ||
2334 | return ret; | |
2335 | } | |
2336 | __setup("schedstats=", setup_schedstats); | |
2337 | ||
4698f88c JP |
2338 | static void __init init_schedstats(void) |
2339 | { | |
2340 | set_schedstats(__sched_schedstats); | |
2341 | } | |
2342 | ||
cb251765 MG |
2343 | #ifdef CONFIG_PROC_SYSCTL |
2344 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2345 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2346 | { | |
2347 | struct ctl_table t; | |
2348 | int err; | |
2349 | int state = static_branch_likely(&sched_schedstats); | |
2350 | ||
2351 | if (write && !capable(CAP_SYS_ADMIN)) | |
2352 | return -EPERM; | |
2353 | ||
2354 | t = *table; | |
2355 | t.data = &state; | |
2356 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2357 | if (err < 0) | |
2358 | return err; | |
2359 | if (write) | |
2360 | set_schedstats(state); | |
2361 | return err; | |
2362 | } | |
4698f88c JP |
2363 | #endif /* CONFIG_PROC_SYSCTL */ |
2364 | #else /* !CONFIG_SCHEDSTATS */ | |
2365 | static inline void init_schedstats(void) {} | |
2366 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2367 | |
2368 | /* | |
2369 | * fork()/clone()-time setup: | |
2370 | */ | |
aab03e05 | 2371 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2372 | { |
0122ec5b | 2373 | unsigned long flags; |
dd41f596 IM |
2374 | int cpu = get_cpu(); |
2375 | ||
5e1576ed | 2376 | __sched_fork(clone_flags, p); |
06b83b5f | 2377 | /* |
7dc603c9 | 2378 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2379 | * nobody will actually run it, and a signal or other external |
2380 | * event cannot wake it up and insert it on the runqueue either. | |
2381 | */ | |
7dc603c9 | 2382 | p->state = TASK_NEW; |
dd41f596 | 2383 | |
c350a04e MG |
2384 | /* |
2385 | * Make sure we do not leak PI boosting priority to the child. | |
2386 | */ | |
2387 | p->prio = current->normal_prio; | |
2388 | ||
b9dc29e7 MG |
2389 | /* |
2390 | * Revert to default priority/policy on fork if requested. | |
2391 | */ | |
2392 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2393 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2394 | p->policy = SCHED_NORMAL; |
6c697bdf | 2395 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2396 | p->rt_priority = 0; |
2397 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2398 | p->static_prio = NICE_TO_PRIO(0); | |
2399 | ||
2400 | p->prio = p->normal_prio = __normal_prio(p); | |
2401 | set_load_weight(p); | |
6c697bdf | 2402 | |
b9dc29e7 MG |
2403 | /* |
2404 | * We don't need the reset flag anymore after the fork. It has | |
2405 | * fulfilled its duty: | |
2406 | */ | |
2407 | p->sched_reset_on_fork = 0; | |
2408 | } | |
ca94c442 | 2409 | |
aab03e05 DF |
2410 | if (dl_prio(p->prio)) { |
2411 | put_cpu(); | |
2412 | return -EAGAIN; | |
2413 | } else if (rt_prio(p->prio)) { | |
2414 | p->sched_class = &rt_sched_class; | |
2415 | } else { | |
2ddbf952 | 2416 | p->sched_class = &fair_sched_class; |
aab03e05 | 2417 | } |
b29739f9 | 2418 | |
7dc603c9 | 2419 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2420 | |
86951599 PZ |
2421 | /* |
2422 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2423 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2424 | * is ran before sched_fork(). | |
2425 | * | |
2426 | * Silence PROVE_RCU. | |
2427 | */ | |
0122ec5b | 2428 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd PZ |
2429 | /* |
2430 | * We're setting the cpu for the first time, we don't migrate, | |
2431 | * so use __set_task_cpu(). | |
2432 | */ | |
2433 | __set_task_cpu(p, cpu); | |
2434 | if (p->sched_class->task_fork) | |
2435 | p->sched_class->task_fork(p); | |
0122ec5b | 2436 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2437 | |
f6db8347 | 2438 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2439 | if (likely(sched_info_on())) |
52f17b6c | 2440 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2441 | #endif |
3ca7a440 PZ |
2442 | #if defined(CONFIG_SMP) |
2443 | p->on_cpu = 0; | |
4866cde0 | 2444 | #endif |
01028747 | 2445 | init_task_preempt_count(p); |
806c09a7 | 2446 | #ifdef CONFIG_SMP |
917b627d | 2447 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2448 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2449 | #endif |
917b627d | 2450 | |
476d139c | 2451 | put_cpu(); |
aab03e05 | 2452 | return 0; |
1da177e4 LT |
2453 | } |
2454 | ||
332ac17e DF |
2455 | unsigned long to_ratio(u64 period, u64 runtime) |
2456 | { | |
2457 | if (runtime == RUNTIME_INF) | |
2458 | return 1ULL << 20; | |
2459 | ||
2460 | /* | |
2461 | * Doing this here saves a lot of checks in all | |
2462 | * the calling paths, and returning zero seems | |
2463 | * safe for them anyway. | |
2464 | */ | |
2465 | if (period == 0) | |
2466 | return 0; | |
2467 | ||
2468 | return div64_u64(runtime << 20, period); | |
2469 | } | |
2470 | ||
2471 | #ifdef CONFIG_SMP | |
2472 | inline struct dl_bw *dl_bw_of(int i) | |
2473 | { | |
f78f5b90 PM |
2474 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2475 | "sched RCU must be held"); | |
332ac17e DF |
2476 | return &cpu_rq(i)->rd->dl_bw; |
2477 | } | |
2478 | ||
de212f18 | 2479 | static inline int dl_bw_cpus(int i) |
332ac17e | 2480 | { |
de212f18 PZ |
2481 | struct root_domain *rd = cpu_rq(i)->rd; |
2482 | int cpus = 0; | |
2483 | ||
f78f5b90 PM |
2484 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2485 | "sched RCU must be held"); | |
de212f18 PZ |
2486 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2487 | cpus++; | |
2488 | ||
2489 | return cpus; | |
332ac17e DF |
2490 | } |
2491 | #else | |
2492 | inline struct dl_bw *dl_bw_of(int i) | |
2493 | { | |
2494 | return &cpu_rq(i)->dl.dl_bw; | |
2495 | } | |
2496 | ||
de212f18 | 2497 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2498 | { |
2499 | return 1; | |
2500 | } | |
2501 | #endif | |
2502 | ||
332ac17e DF |
2503 | /* |
2504 | * We must be sure that accepting a new task (or allowing changing the | |
2505 | * parameters of an existing one) is consistent with the bandwidth | |
2506 | * constraints. If yes, this function also accordingly updates the currently | |
2507 | * allocated bandwidth to reflect the new situation. | |
2508 | * | |
2509 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2510 | * |
2511 | * XXX we should delay bw change until the task's 0-lag point, see | |
2512 | * __setparam_dl(). | |
332ac17e DF |
2513 | */ |
2514 | static int dl_overflow(struct task_struct *p, int policy, | |
2515 | const struct sched_attr *attr) | |
2516 | { | |
2517 | ||
2518 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2519 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2520 | u64 runtime = attr->sched_runtime; |
2521 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2522 | int cpus, err = -1; |
332ac17e | 2523 | |
fec148c0 XP |
2524 | /* !deadline task may carry old deadline bandwidth */ |
2525 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2526 | return 0; |
2527 | ||
2528 | /* | |
2529 | * Either if a task, enters, leave, or stays -deadline but changes | |
2530 | * its parameters, we may need to update accordingly the total | |
2531 | * allocated bandwidth of the container. | |
2532 | */ | |
2533 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2534 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2535 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2536 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2537 | __dl_add(dl_b, new_bw); | |
2538 | err = 0; | |
2539 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2540 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2541 | __dl_clear(dl_b, p->dl.dl_bw); | |
2542 | __dl_add(dl_b, new_bw); | |
2543 | err = 0; | |
2544 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2545 | __dl_clear(dl_b, p->dl.dl_bw); | |
2546 | err = 0; | |
2547 | } | |
2548 | raw_spin_unlock(&dl_b->lock); | |
2549 | ||
2550 | return err; | |
2551 | } | |
2552 | ||
2553 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2554 | ||
1da177e4 LT |
2555 | /* |
2556 | * wake_up_new_task - wake up a newly created task for the first time. | |
2557 | * | |
2558 | * This function will do some initial scheduler statistics housekeeping | |
2559 | * that must be done for every newly created context, then puts the task | |
2560 | * on the runqueue and wakes it. | |
2561 | */ | |
3e51e3ed | 2562 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2563 | { |
eb580751 | 2564 | struct rq_flags rf; |
dd41f596 | 2565 | struct rq *rq; |
fabf318e | 2566 | |
eb580751 | 2567 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2568 | p->state = TASK_RUNNING; |
fabf318e PZ |
2569 | #ifdef CONFIG_SMP |
2570 | /* | |
2571 | * Fork balancing, do it here and not earlier because: | |
2572 | * - cpus_allowed can change in the fork path | |
2573 | * - any previously selected cpu might disappear through hotplug | |
e210bffd PZ |
2574 | * |
2575 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2576 | * as we're not fully set-up yet. | |
fabf318e | 2577 | */ |
e210bffd | 2578 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2579 | #endif |
b7fa30c9 | 2580 | rq = __task_rq_lock(p, &rf); |
2b8c41da | 2581 | post_init_entity_util_avg(&p->se); |
0017d735 | 2582 | |
cd29fe6f | 2583 | activate_task(rq, p, 0); |
da0c1e65 | 2584 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2585 | trace_sched_wakeup_new(p); |
a7558e01 | 2586 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2587 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2588 | if (p->sched_class->task_woken) { |
2589 | /* | |
2590 | * Nothing relies on rq->lock after this, so its fine to | |
2591 | * drop it. | |
2592 | */ | |
d8ac8971 | 2593 | rq_unpin_lock(rq, &rf); |
efbbd05a | 2594 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2595 | rq_repin_lock(rq, &rf); |
0aaafaab | 2596 | } |
9a897c5a | 2597 | #endif |
eb580751 | 2598 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2599 | } |
2600 | ||
e107be36 AK |
2601 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2602 | ||
1cde2930 PZ |
2603 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2604 | ||
2ecd9d29 PZ |
2605 | void preempt_notifier_inc(void) |
2606 | { | |
2607 | static_key_slow_inc(&preempt_notifier_key); | |
2608 | } | |
2609 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2610 | ||
2611 | void preempt_notifier_dec(void) | |
2612 | { | |
2613 | static_key_slow_dec(&preempt_notifier_key); | |
2614 | } | |
2615 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2616 | ||
e107be36 | 2617 | /** |
80dd99b3 | 2618 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2619 | * @notifier: notifier struct to register |
e107be36 AK |
2620 | */ |
2621 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2622 | { | |
2ecd9d29 PZ |
2623 | if (!static_key_false(&preempt_notifier_key)) |
2624 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2625 | ||
e107be36 AK |
2626 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2627 | } | |
2628 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2629 | ||
2630 | /** | |
2631 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2632 | * @notifier: notifier struct to unregister |
e107be36 | 2633 | * |
d84525a8 | 2634 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2635 | */ |
2636 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2637 | { | |
2638 | hlist_del(¬ifier->link); | |
2639 | } | |
2640 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2641 | ||
1cde2930 | 2642 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2643 | { |
2644 | struct preempt_notifier *notifier; | |
e107be36 | 2645 | |
b67bfe0d | 2646 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2647 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2648 | } | |
2649 | ||
1cde2930 PZ |
2650 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2651 | { | |
2652 | if (static_key_false(&preempt_notifier_key)) | |
2653 | __fire_sched_in_preempt_notifiers(curr); | |
2654 | } | |
2655 | ||
e107be36 | 2656 | static void |
1cde2930 PZ |
2657 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2658 | struct task_struct *next) | |
e107be36 AK |
2659 | { |
2660 | struct preempt_notifier *notifier; | |
e107be36 | 2661 | |
b67bfe0d | 2662 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2663 | notifier->ops->sched_out(notifier, next); |
2664 | } | |
2665 | ||
1cde2930 PZ |
2666 | static __always_inline void |
2667 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2668 | struct task_struct *next) | |
2669 | { | |
2670 | if (static_key_false(&preempt_notifier_key)) | |
2671 | __fire_sched_out_preempt_notifiers(curr, next); | |
2672 | } | |
2673 | ||
6d6bc0ad | 2674 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2675 | |
1cde2930 | 2676 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2677 | { |
2678 | } | |
2679 | ||
1cde2930 | 2680 | static inline void |
e107be36 AK |
2681 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2682 | struct task_struct *next) | |
2683 | { | |
2684 | } | |
2685 | ||
6d6bc0ad | 2686 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2687 | |
4866cde0 NP |
2688 | /** |
2689 | * prepare_task_switch - prepare to switch tasks | |
2690 | * @rq: the runqueue preparing to switch | |
421cee29 | 2691 | * @prev: the current task that is being switched out |
4866cde0 NP |
2692 | * @next: the task we are going to switch to. |
2693 | * | |
2694 | * This is called with the rq lock held and interrupts off. It must | |
2695 | * be paired with a subsequent finish_task_switch after the context | |
2696 | * switch. | |
2697 | * | |
2698 | * prepare_task_switch sets up locking and calls architecture specific | |
2699 | * hooks. | |
2700 | */ | |
e107be36 AK |
2701 | static inline void |
2702 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2703 | struct task_struct *next) | |
4866cde0 | 2704 | { |
43148951 | 2705 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2706 | perf_event_task_sched_out(prev, next); |
e107be36 | 2707 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2708 | prepare_lock_switch(rq, next); |
2709 | prepare_arch_switch(next); | |
2710 | } | |
2711 | ||
1da177e4 LT |
2712 | /** |
2713 | * finish_task_switch - clean up after a task-switch | |
2714 | * @prev: the thread we just switched away from. | |
2715 | * | |
4866cde0 NP |
2716 | * finish_task_switch must be called after the context switch, paired |
2717 | * with a prepare_task_switch call before the context switch. | |
2718 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2719 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2720 | * |
2721 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2722 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2723 | * with the lock held can cause deadlocks; see schedule() for |
2724 | * details.) | |
dfa50b60 ON |
2725 | * |
2726 | * The context switch have flipped the stack from under us and restored the | |
2727 | * local variables which were saved when this task called schedule() in the | |
2728 | * past. prev == current is still correct but we need to recalculate this_rq | |
2729 | * because prev may have moved to another CPU. | |
1da177e4 | 2730 | */ |
dfa50b60 | 2731 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2732 | __releases(rq->lock) |
2733 | { | |
dfa50b60 | 2734 | struct rq *rq = this_rq(); |
1da177e4 | 2735 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2736 | long prev_state; |
1da177e4 | 2737 | |
609ca066 PZ |
2738 | /* |
2739 | * The previous task will have left us with a preempt_count of 2 | |
2740 | * because it left us after: | |
2741 | * | |
2742 | * schedule() | |
2743 | * preempt_disable(); // 1 | |
2744 | * __schedule() | |
2745 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2746 | * | |
2747 | * Also, see FORK_PREEMPT_COUNT. | |
2748 | */ | |
e2bf1c4b PZ |
2749 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2750 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2751 | current->comm, current->pid, preempt_count())) | |
2752 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2753 | |
1da177e4 LT |
2754 | rq->prev_mm = NULL; |
2755 | ||
2756 | /* | |
2757 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2758 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2759 | * schedule one last time. The schedule call will never return, and |
2760 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2761 | * |
2762 | * We must observe prev->state before clearing prev->on_cpu (in | |
2763 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2764 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2765 | * transition, resulting in a double drop. | |
1da177e4 | 2766 | */ |
55a101f8 | 2767 | prev_state = prev->state; |
bf9fae9f | 2768 | vtime_task_switch(prev); |
a8d757ef | 2769 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2770 | finish_lock_switch(rq, prev); |
01f23e16 | 2771 | finish_arch_post_lock_switch(); |
e8fa1362 | 2772 | |
e107be36 | 2773 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2774 | if (mm) |
2775 | mmdrop(mm); | |
c394cc9f | 2776 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2777 | if (prev->sched_class->task_dead) |
2778 | prev->sched_class->task_dead(prev); | |
2779 | ||
c6fd91f0 | 2780 | /* |
2781 | * Remove function-return probe instances associated with this | |
2782 | * task and put them back on the free list. | |
9761eea8 | 2783 | */ |
c6fd91f0 | 2784 | kprobe_flush_task(prev); |
68f24b08 AL |
2785 | |
2786 | /* Task is done with its stack. */ | |
2787 | put_task_stack(prev); | |
2788 | ||
1da177e4 | 2789 | put_task_struct(prev); |
c6fd91f0 | 2790 | } |
99e5ada9 | 2791 | |
de734f89 | 2792 | tick_nohz_task_switch(); |
dfa50b60 | 2793 | return rq; |
1da177e4 LT |
2794 | } |
2795 | ||
3f029d3c GH |
2796 | #ifdef CONFIG_SMP |
2797 | ||
3f029d3c | 2798 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2799 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2800 | { |
e3fca9e7 PZ |
2801 | struct callback_head *head, *next; |
2802 | void (*func)(struct rq *rq); | |
2803 | unsigned long flags; | |
3f029d3c | 2804 | |
e3fca9e7 PZ |
2805 | raw_spin_lock_irqsave(&rq->lock, flags); |
2806 | head = rq->balance_callback; | |
2807 | rq->balance_callback = NULL; | |
2808 | while (head) { | |
2809 | func = (void (*)(struct rq *))head->func; | |
2810 | next = head->next; | |
2811 | head->next = NULL; | |
2812 | head = next; | |
3f029d3c | 2813 | |
e3fca9e7 | 2814 | func(rq); |
3f029d3c | 2815 | } |
e3fca9e7 PZ |
2816 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2817 | } | |
2818 | ||
2819 | static inline void balance_callback(struct rq *rq) | |
2820 | { | |
2821 | if (unlikely(rq->balance_callback)) | |
2822 | __balance_callback(rq); | |
3f029d3c GH |
2823 | } |
2824 | ||
2825 | #else | |
da19ab51 | 2826 | |
e3fca9e7 | 2827 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2828 | { |
1da177e4 LT |
2829 | } |
2830 | ||
3f029d3c GH |
2831 | #endif |
2832 | ||
1da177e4 LT |
2833 | /** |
2834 | * schedule_tail - first thing a freshly forked thread must call. | |
2835 | * @prev: the thread we just switched away from. | |
2836 | */ | |
722a9f92 | 2837 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2838 | __releases(rq->lock) |
2839 | { | |
1a43a14a | 2840 | struct rq *rq; |
da19ab51 | 2841 | |
609ca066 PZ |
2842 | /* |
2843 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2844 | * finish_task_switch() for details. | |
2845 | * | |
2846 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2847 | * and the preempt_enable() will end up enabling preemption (on | |
2848 | * PREEMPT_COUNT kernels). | |
2849 | */ | |
2850 | ||
dfa50b60 | 2851 | rq = finish_task_switch(prev); |
e3fca9e7 | 2852 | balance_callback(rq); |
1a43a14a | 2853 | preempt_enable(); |
70b97a7f | 2854 | |
1da177e4 | 2855 | if (current->set_child_tid) |
b488893a | 2856 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2857 | } |
2858 | ||
2859 | /* | |
dfa50b60 | 2860 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2861 | */ |
04936948 | 2862 | static __always_inline struct rq * |
70b97a7f | 2863 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 2864 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 2865 | { |
dd41f596 | 2866 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2867 | |
e107be36 | 2868 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2869 | |
dd41f596 IM |
2870 | mm = next->mm; |
2871 | oldmm = prev->active_mm; | |
9226d125 ZA |
2872 | /* |
2873 | * For paravirt, this is coupled with an exit in switch_to to | |
2874 | * combine the page table reload and the switch backend into | |
2875 | * one hypercall. | |
2876 | */ | |
224101ed | 2877 | arch_start_context_switch(prev); |
9226d125 | 2878 | |
31915ab4 | 2879 | if (!mm) { |
1da177e4 LT |
2880 | next->active_mm = oldmm; |
2881 | atomic_inc(&oldmm->mm_count); | |
2882 | enter_lazy_tlb(oldmm, next); | |
2883 | } else | |
f98db601 | 2884 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2885 | |
31915ab4 | 2886 | if (!prev->mm) { |
1da177e4 | 2887 | prev->active_mm = NULL; |
1da177e4 LT |
2888 | rq->prev_mm = oldmm; |
2889 | } | |
92509b73 MF |
2890 | |
2891 | rq->clock_skip_update = 0; | |
2892 | ||
3a5f5e48 IM |
2893 | /* |
2894 | * Since the runqueue lock will be released by the next | |
2895 | * task (which is an invalid locking op but in the case | |
2896 | * of the scheduler it's an obvious special-case), so we | |
2897 | * do an early lockdep release here: | |
2898 | */ | |
d8ac8971 | 2899 | rq_unpin_lock(rq, rf); |
8a25d5de | 2900 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2901 | |
2902 | /* Here we just switch the register state and the stack. */ | |
2903 | switch_to(prev, next, prev); | |
dd41f596 | 2904 | barrier(); |
dfa50b60 ON |
2905 | |
2906 | return finish_task_switch(prev); | |
1da177e4 LT |
2907 | } |
2908 | ||
2909 | /* | |
1c3e8264 | 2910 | * nr_running and nr_context_switches: |
1da177e4 LT |
2911 | * |
2912 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2913 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2914 | */ |
2915 | unsigned long nr_running(void) | |
2916 | { | |
2917 | unsigned long i, sum = 0; | |
2918 | ||
2919 | for_each_online_cpu(i) | |
2920 | sum += cpu_rq(i)->nr_running; | |
2921 | ||
2922 | return sum; | |
f711f609 | 2923 | } |
1da177e4 | 2924 | |
2ee507c4 TC |
2925 | /* |
2926 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2927 | * |
2928 | * Caution: this function does not check that the caller has disabled | |
2929 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2930 | * race. The caller is responsible to use it correctly, for example: | |
2931 | * | |
2932 | * - from a non-preemptable section (of course) | |
2933 | * | |
2934 | * - from a thread that is bound to a single CPU | |
2935 | * | |
2936 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2937 | */ |
2938 | bool single_task_running(void) | |
2939 | { | |
00cc1633 | 2940 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2941 | } |
2942 | EXPORT_SYMBOL(single_task_running); | |
2943 | ||
1da177e4 | 2944 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2945 | { |
cc94abfc SR |
2946 | int i; |
2947 | unsigned long long sum = 0; | |
46cb4b7c | 2948 | |
0a945022 | 2949 | for_each_possible_cpu(i) |
1da177e4 | 2950 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2951 | |
1da177e4 LT |
2952 | return sum; |
2953 | } | |
483b4ee6 | 2954 | |
1da177e4 LT |
2955 | unsigned long nr_iowait(void) |
2956 | { | |
2957 | unsigned long i, sum = 0; | |
483b4ee6 | 2958 | |
0a945022 | 2959 | for_each_possible_cpu(i) |
1da177e4 | 2960 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2961 | |
1da177e4 LT |
2962 | return sum; |
2963 | } | |
483b4ee6 | 2964 | |
8c215bd3 | 2965 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2966 | { |
8c215bd3 | 2967 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2968 | return atomic_read(&this->nr_iowait); |
2969 | } | |
46cb4b7c | 2970 | |
372ba8cb MG |
2971 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2972 | { | |
3289bdb4 PZ |
2973 | struct rq *rq = this_rq(); |
2974 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2975 | *load = rq->load.weight; | |
372ba8cb MG |
2976 | } |
2977 | ||
dd41f596 | 2978 | #ifdef CONFIG_SMP |
8a0be9ef | 2979 | |
46cb4b7c | 2980 | /* |
38022906 PZ |
2981 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2982 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2983 | */ |
38022906 | 2984 | void sched_exec(void) |
46cb4b7c | 2985 | { |
38022906 | 2986 | struct task_struct *p = current; |
1da177e4 | 2987 | unsigned long flags; |
0017d735 | 2988 | int dest_cpu; |
46cb4b7c | 2989 | |
8f42ced9 | 2990 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2991 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2992 | if (dest_cpu == smp_processor_id()) |
2993 | goto unlock; | |
38022906 | 2994 | |
8f42ced9 | 2995 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2996 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2997 | |
8f42ced9 PZ |
2998 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2999 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3000 | return; |
3001 | } | |
0017d735 | 3002 | unlock: |
8f42ced9 | 3003 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3004 | } |
dd41f596 | 3005 | |
1da177e4 LT |
3006 | #endif |
3007 | ||
1da177e4 | 3008 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3009 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3010 | |
3011 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3012 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3013 | |
6075620b GG |
3014 | /* |
3015 | * The function fair_sched_class.update_curr accesses the struct curr | |
3016 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3017 | * we observe a high rate of cache misses in practice. | |
3018 | * Prefetching this data results in improved performance. | |
3019 | */ | |
3020 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3021 | { | |
3022 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3023 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3024 | #else | |
3025 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3026 | #endif | |
3027 | prefetch(curr); | |
3028 | prefetch(&curr->exec_start); | |
3029 | } | |
3030 | ||
c5f8d995 HS |
3031 | /* |
3032 | * Return accounted runtime for the task. | |
3033 | * In case the task is currently running, return the runtime plus current's | |
3034 | * pending runtime that have not been accounted yet. | |
3035 | */ | |
3036 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3037 | { | |
eb580751 | 3038 | struct rq_flags rf; |
c5f8d995 | 3039 | struct rq *rq; |
6e998916 | 3040 | u64 ns; |
c5f8d995 | 3041 | |
911b2898 PZ |
3042 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3043 | /* | |
3044 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
3045 | * So we have a optimization chance when the task's delta_exec is 0. | |
3046 | * Reading ->on_cpu is racy, but this is ok. | |
3047 | * | |
3048 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
3049 | * If we race with it entering cpu, unaccounted time is 0. This is | |
3050 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
3051 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3052 | * been accounted, so we're correct here as well. | |
911b2898 | 3053 | */ |
da0c1e65 | 3054 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3055 | return p->se.sum_exec_runtime; |
3056 | #endif | |
3057 | ||
eb580751 | 3058 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3059 | /* |
3060 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3061 | * project cycles that may never be accounted to this | |
3062 | * thread, breaking clock_gettime(). | |
3063 | */ | |
3064 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3065 | prefetch_curr_exec_start(p); |
6e998916 SG |
3066 | update_rq_clock(rq); |
3067 | p->sched_class->update_curr(rq); | |
3068 | } | |
3069 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3070 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3071 | |
3072 | return ns; | |
3073 | } | |
48f24c4d | 3074 | |
7835b98b CL |
3075 | /* |
3076 | * This function gets called by the timer code, with HZ frequency. | |
3077 | * We call it with interrupts disabled. | |
7835b98b CL |
3078 | */ |
3079 | void scheduler_tick(void) | |
3080 | { | |
7835b98b CL |
3081 | int cpu = smp_processor_id(); |
3082 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3083 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3084 | |
3085 | sched_clock_tick(); | |
dd41f596 | 3086 | |
05fa785c | 3087 | raw_spin_lock(&rq->lock); |
3e51f33f | 3088 | update_rq_clock(rq); |
fa85ae24 | 3089 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3090 | cpu_load_update_active(rq); |
3289bdb4 | 3091 | calc_global_load_tick(rq); |
05fa785c | 3092 | raw_spin_unlock(&rq->lock); |
7835b98b | 3093 | |
e9d2b064 | 3094 | perf_event_task_tick(); |
e220d2dc | 3095 | |
e418e1c2 | 3096 | #ifdef CONFIG_SMP |
6eb57e0d | 3097 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3098 | trigger_load_balance(rq); |
e418e1c2 | 3099 | #endif |
265f22a9 | 3100 | rq_last_tick_reset(rq); |
1da177e4 LT |
3101 | } |
3102 | ||
265f22a9 FW |
3103 | #ifdef CONFIG_NO_HZ_FULL |
3104 | /** | |
3105 | * scheduler_tick_max_deferment | |
3106 | * | |
3107 | * Keep at least one tick per second when a single | |
3108 | * active task is running because the scheduler doesn't | |
3109 | * yet completely support full dynticks environment. | |
3110 | * | |
3111 | * This makes sure that uptime, CFS vruntime, load | |
3112 | * balancing, etc... continue to move forward, even | |
3113 | * with a very low granularity. | |
e69f6186 YB |
3114 | * |
3115 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3116 | */ |
3117 | u64 scheduler_tick_max_deferment(void) | |
3118 | { | |
3119 | struct rq *rq = this_rq(); | |
316c1608 | 3120 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3121 | |
3122 | next = rq->last_sched_tick + HZ; | |
3123 | ||
3124 | if (time_before_eq(next, now)) | |
3125 | return 0; | |
3126 | ||
8fe8ff09 | 3127 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3128 | } |
265f22a9 | 3129 | #endif |
1da177e4 | 3130 | |
7e49fcce SR |
3131 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3132 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3133 | /* |
3134 | * If the value passed in is equal to the current preempt count | |
3135 | * then we just disabled preemption. Start timing the latency. | |
3136 | */ | |
3137 | static inline void preempt_latency_start(int val) | |
3138 | { | |
3139 | if (preempt_count() == val) { | |
3140 | unsigned long ip = get_lock_parent_ip(); | |
3141 | #ifdef CONFIG_DEBUG_PREEMPT | |
3142 | current->preempt_disable_ip = ip; | |
3143 | #endif | |
3144 | trace_preempt_off(CALLER_ADDR0, ip); | |
3145 | } | |
3146 | } | |
7e49fcce | 3147 | |
edafe3a5 | 3148 | void preempt_count_add(int val) |
1da177e4 | 3149 | { |
6cd8a4bb | 3150 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3151 | /* |
3152 | * Underflow? | |
3153 | */ | |
9a11b49a IM |
3154 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3155 | return; | |
6cd8a4bb | 3156 | #endif |
bdb43806 | 3157 | __preempt_count_add(val); |
6cd8a4bb | 3158 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3159 | /* |
3160 | * Spinlock count overflowing soon? | |
3161 | */ | |
33859f7f MOS |
3162 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3163 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3164 | #endif |
47252cfb | 3165 | preempt_latency_start(val); |
1da177e4 | 3166 | } |
bdb43806 | 3167 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3168 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3169 | |
47252cfb SR |
3170 | /* |
3171 | * If the value passed in equals to the current preempt count | |
3172 | * then we just enabled preemption. Stop timing the latency. | |
3173 | */ | |
3174 | static inline void preempt_latency_stop(int val) | |
3175 | { | |
3176 | if (preempt_count() == val) | |
3177 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3178 | } | |
3179 | ||
edafe3a5 | 3180 | void preempt_count_sub(int val) |
1da177e4 | 3181 | { |
6cd8a4bb | 3182 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3183 | /* |
3184 | * Underflow? | |
3185 | */ | |
01e3eb82 | 3186 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3187 | return; |
1da177e4 LT |
3188 | /* |
3189 | * Is the spinlock portion underflowing? | |
3190 | */ | |
9a11b49a IM |
3191 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3192 | !(preempt_count() & PREEMPT_MASK))) | |
3193 | return; | |
6cd8a4bb | 3194 | #endif |
9a11b49a | 3195 | |
47252cfb | 3196 | preempt_latency_stop(val); |
bdb43806 | 3197 | __preempt_count_sub(val); |
1da177e4 | 3198 | } |
bdb43806 | 3199 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3200 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3201 | |
47252cfb SR |
3202 | #else |
3203 | static inline void preempt_latency_start(int val) { } | |
3204 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3205 | #endif |
3206 | ||
3207 | /* | |
dd41f596 | 3208 | * Print scheduling while atomic bug: |
1da177e4 | 3209 | */ |
dd41f596 | 3210 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3211 | { |
d1c6d149 VN |
3212 | /* Save this before calling printk(), since that will clobber it */ |
3213 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3214 | ||
664dfa65 DJ |
3215 | if (oops_in_progress) |
3216 | return; | |
3217 | ||
3df0fc5b PZ |
3218 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3219 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3220 | |
dd41f596 | 3221 | debug_show_held_locks(prev); |
e21f5b15 | 3222 | print_modules(); |
dd41f596 IM |
3223 | if (irqs_disabled()) |
3224 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3225 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3226 | && in_atomic_preempt_off()) { | |
8f47b187 | 3227 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3228 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3229 | pr_cont("\n"); |
3230 | } | |
748c7201 DBO |
3231 | if (panic_on_warn) |
3232 | panic("scheduling while atomic\n"); | |
3233 | ||
6135fc1e | 3234 | dump_stack(); |
373d4d09 | 3235 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3236 | } |
1da177e4 | 3237 | |
dd41f596 IM |
3238 | /* |
3239 | * Various schedule()-time debugging checks and statistics: | |
3240 | */ | |
3241 | static inline void schedule_debug(struct task_struct *prev) | |
3242 | { | |
0d9e2632 | 3243 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3244 | if (task_stack_end_corrupted(prev)) |
3245 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3246 | #endif |
b99def8b | 3247 | |
1dc0fffc | 3248 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3249 | __schedule_bug(prev); |
1dc0fffc PZ |
3250 | preempt_count_set(PREEMPT_DISABLED); |
3251 | } | |
b3fbab05 | 3252 | rcu_sleep_check(); |
dd41f596 | 3253 | |
1da177e4 LT |
3254 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3255 | ||
ae92882e | 3256 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3257 | } |
3258 | ||
3259 | /* | |
3260 | * Pick up the highest-prio task: | |
3261 | */ | |
3262 | static inline struct task_struct * | |
d8ac8971 | 3263 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3264 | { |
37e117c0 | 3265 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 3266 | struct task_struct *p; |
1da177e4 LT |
3267 | |
3268 | /* | |
dd41f596 IM |
3269 | * Optimization: we know that if all tasks are in |
3270 | * the fair class we can call that function directly: | |
1da177e4 | 3271 | */ |
37e117c0 | 3272 | if (likely(prev->sched_class == class && |
38033c37 | 3273 | rq->nr_running == rq->cfs.h_nr_running)) { |
d8ac8971 | 3274 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3275 | if (unlikely(p == RETRY_TASK)) |
3276 | goto again; | |
3277 | ||
3278 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3279 | if (unlikely(!p)) | |
d8ac8971 | 3280 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3281 | |
3282 | return p; | |
1da177e4 LT |
3283 | } |
3284 | ||
37e117c0 | 3285 | again: |
34f971f6 | 3286 | for_each_class(class) { |
d8ac8971 | 3287 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3288 | if (p) { |
3289 | if (unlikely(p == RETRY_TASK)) | |
3290 | goto again; | |
dd41f596 | 3291 | return p; |
37e117c0 | 3292 | } |
dd41f596 | 3293 | } |
34f971f6 PZ |
3294 | |
3295 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3296 | } |
1da177e4 | 3297 | |
dd41f596 | 3298 | /* |
c259e01a | 3299 | * __schedule() is the main scheduler function. |
edde96ea PE |
3300 | * |
3301 | * The main means of driving the scheduler and thus entering this function are: | |
3302 | * | |
3303 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3304 | * | |
3305 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3306 | * paths. For example, see arch/x86/entry_64.S. | |
3307 | * | |
3308 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3309 | * interrupt handler scheduler_tick(). | |
3310 | * | |
3311 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3312 | * task to the run-queue and that's it. | |
3313 | * | |
3314 | * Now, if the new task added to the run-queue preempts the current | |
3315 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3316 | * called on the nearest possible occasion: | |
3317 | * | |
3318 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3319 | * | |
3320 | * - in syscall or exception context, at the next outmost | |
3321 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3322 | * spin_unlock()!) | |
3323 | * | |
3324 | * - in IRQ context, return from interrupt-handler to | |
3325 | * preemptible context | |
3326 | * | |
3327 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3328 | * then at the next: | |
3329 | * | |
3330 | * - cond_resched() call | |
3331 | * - explicit schedule() call | |
3332 | * - return from syscall or exception to user-space | |
3333 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3334 | * |
b30f0e3f | 3335 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3336 | */ |
499d7955 | 3337 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3338 | { |
3339 | struct task_struct *prev, *next; | |
67ca7bde | 3340 | unsigned long *switch_count; |
d8ac8971 | 3341 | struct rq_flags rf; |
dd41f596 | 3342 | struct rq *rq; |
31656519 | 3343 | int cpu; |
dd41f596 | 3344 | |
dd41f596 IM |
3345 | cpu = smp_processor_id(); |
3346 | rq = cpu_rq(cpu); | |
dd41f596 | 3347 | prev = rq->curr; |
dd41f596 | 3348 | |
dd41f596 | 3349 | schedule_debug(prev); |
1da177e4 | 3350 | |
31656519 | 3351 | if (sched_feat(HRTICK)) |
f333fdc9 | 3352 | hrtick_clear(rq); |
8f4d37ec | 3353 | |
46a5d164 PM |
3354 | local_irq_disable(); |
3355 | rcu_note_context_switch(); | |
3356 | ||
e0acd0a6 ON |
3357 | /* |
3358 | * Make sure that signal_pending_state()->signal_pending() below | |
3359 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3360 | * done by the caller to avoid the race with signal_wake_up(). | |
3361 | */ | |
3362 | smp_mb__before_spinlock(); | |
46a5d164 | 3363 | raw_spin_lock(&rq->lock); |
d8ac8971 | 3364 | rq_pin_lock(rq, &rf); |
1da177e4 | 3365 | |
9edfbfed PZ |
3366 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3367 | ||
246d86b5 | 3368 | switch_count = &prev->nivcsw; |
fc13aeba | 3369 | if (!preempt && prev->state) { |
21aa9af0 | 3370 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3371 | prev->state = TASK_RUNNING; |
21aa9af0 | 3372 | } else { |
2acca55e PZ |
3373 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3374 | prev->on_rq = 0; | |
3375 | ||
21aa9af0 | 3376 | /* |
2acca55e PZ |
3377 | * If a worker went to sleep, notify and ask workqueue |
3378 | * whether it wants to wake up a task to maintain | |
3379 | * concurrency. | |
21aa9af0 TH |
3380 | */ |
3381 | if (prev->flags & PF_WQ_WORKER) { | |
3382 | struct task_struct *to_wakeup; | |
3383 | ||
9b7f6597 | 3384 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3385 | if (to_wakeup) |
d8ac8971 | 3386 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3387 | } |
21aa9af0 | 3388 | } |
dd41f596 | 3389 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3390 | } |
3391 | ||
9edfbfed | 3392 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3393 | update_rq_clock(rq); |
3394 | ||
d8ac8971 | 3395 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3396 | clear_tsk_need_resched(prev); |
f27dde8d | 3397 | clear_preempt_need_resched(); |
1da177e4 | 3398 | |
1da177e4 | 3399 | if (likely(prev != next)) { |
1da177e4 LT |
3400 | rq->nr_switches++; |
3401 | rq->curr = next; | |
3402 | ++*switch_count; | |
3403 | ||
c73464b1 | 3404 | trace_sched_switch(preempt, prev, next); |
d8ac8971 | 3405 | rq = context_switch(rq, prev, next, &rf); /* unlocks the rq */ |
cbce1a68 | 3406 | } else { |
92509b73 | 3407 | rq->clock_skip_update = 0; |
d8ac8971 | 3408 | rq_unpin_lock(rq, &rf); |
05fa785c | 3409 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3410 | } |
1da177e4 | 3411 | |
e3fca9e7 | 3412 | balance_callback(rq); |
1da177e4 | 3413 | } |
c259e01a | 3414 | |
9af6528e PZ |
3415 | void __noreturn do_task_dead(void) |
3416 | { | |
3417 | /* | |
3418 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed | |
3419 | * when the following two conditions become true. | |
3420 | * - There is race condition of mmap_sem (It is acquired by | |
3421 | * exit_mm()), and | |
3422 | * - SMI occurs before setting TASK_RUNINNG. | |
3423 | * (or hypervisor of virtual machine switches to other guest) | |
3424 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD | |
3425 | * | |
3426 | * To avoid it, we have to wait for releasing tsk->pi_lock which | |
3427 | * is held by try_to_wake_up() | |
3428 | */ | |
3429 | smp_mb(); | |
3430 | raw_spin_unlock_wait(¤t->pi_lock); | |
3431 | ||
3432 | /* causes final put_task_struct in finish_task_switch(). */ | |
3433 | __set_current_state(TASK_DEAD); | |
3434 | current->flags |= PF_NOFREEZE; /* tell freezer to ignore us */ | |
3435 | __schedule(false); | |
3436 | BUG(); | |
3437 | /* Avoid "noreturn function does return". */ | |
3438 | for (;;) | |
3439 | cpu_relax(); /* For when BUG is null */ | |
3440 | } | |
3441 | ||
9c40cef2 TG |
3442 | static inline void sched_submit_work(struct task_struct *tsk) |
3443 | { | |
3c7d5184 | 3444 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3445 | return; |
3446 | /* | |
3447 | * If we are going to sleep and we have plugged IO queued, | |
3448 | * make sure to submit it to avoid deadlocks. | |
3449 | */ | |
3450 | if (blk_needs_flush_plug(tsk)) | |
3451 | blk_schedule_flush_plug(tsk); | |
3452 | } | |
3453 | ||
722a9f92 | 3454 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3455 | { |
9c40cef2 TG |
3456 | struct task_struct *tsk = current; |
3457 | ||
3458 | sched_submit_work(tsk); | |
bfd9b2b5 | 3459 | do { |
b30f0e3f | 3460 | preempt_disable(); |
fc13aeba | 3461 | __schedule(false); |
b30f0e3f | 3462 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3463 | } while (need_resched()); |
c259e01a | 3464 | } |
1da177e4 LT |
3465 | EXPORT_SYMBOL(schedule); |
3466 | ||
91d1aa43 | 3467 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3468 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3469 | { |
3470 | /* | |
3471 | * If we come here after a random call to set_need_resched(), | |
3472 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3473 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3474 | * we find a better solution. | |
7cc78f8f AL |
3475 | * |
3476 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3477 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3478 | * too frequently to make sense yet. |
20ab65e3 | 3479 | */ |
7cc78f8f | 3480 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3481 | schedule(); |
7cc78f8f | 3482 | exception_exit(prev_state); |
20ab65e3 FW |
3483 | } |
3484 | #endif | |
3485 | ||
c5491ea7 TG |
3486 | /** |
3487 | * schedule_preempt_disabled - called with preemption disabled | |
3488 | * | |
3489 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3490 | */ | |
3491 | void __sched schedule_preempt_disabled(void) | |
3492 | { | |
ba74c144 | 3493 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3494 | schedule(); |
3495 | preempt_disable(); | |
3496 | } | |
3497 | ||
06b1f808 | 3498 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3499 | { |
3500 | do { | |
47252cfb SR |
3501 | /* |
3502 | * Because the function tracer can trace preempt_count_sub() | |
3503 | * and it also uses preempt_enable/disable_notrace(), if | |
3504 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3505 | * by the function tracer will call this function again and | |
3506 | * cause infinite recursion. | |
3507 | * | |
3508 | * Preemption must be disabled here before the function | |
3509 | * tracer can trace. Break up preempt_disable() into two | |
3510 | * calls. One to disable preemption without fear of being | |
3511 | * traced. The other to still record the preemption latency, | |
3512 | * which can also be traced by the function tracer. | |
3513 | */ | |
499d7955 | 3514 | preempt_disable_notrace(); |
47252cfb | 3515 | preempt_latency_start(1); |
fc13aeba | 3516 | __schedule(true); |
47252cfb | 3517 | preempt_latency_stop(1); |
499d7955 | 3518 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3519 | |
3520 | /* | |
3521 | * Check again in case we missed a preemption opportunity | |
3522 | * between schedule and now. | |
3523 | */ | |
a18b5d01 FW |
3524 | } while (need_resched()); |
3525 | } | |
3526 | ||
1da177e4 LT |
3527 | #ifdef CONFIG_PREEMPT |
3528 | /* | |
2ed6e34f | 3529 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3530 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3531 | * occur there and call schedule directly. |
3532 | */ | |
722a9f92 | 3533 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3534 | { |
1da177e4 LT |
3535 | /* |
3536 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3537 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3538 | */ |
fbb00b56 | 3539 | if (likely(!preemptible())) |
1da177e4 LT |
3540 | return; |
3541 | ||
a18b5d01 | 3542 | preempt_schedule_common(); |
1da177e4 | 3543 | } |
376e2424 | 3544 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3545 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3546 | |
009f60e2 | 3547 | /** |
4eaca0a8 | 3548 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3549 | * |
3550 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3551 | * recursion and tracing preempt enabling caused by the tracing | |
3552 | * infrastructure itself. But as tracing can happen in areas coming | |
3553 | * from userspace or just about to enter userspace, a preempt enable | |
3554 | * can occur before user_exit() is called. This will cause the scheduler | |
3555 | * to be called when the system is still in usermode. | |
3556 | * | |
3557 | * To prevent this, the preempt_enable_notrace will use this function | |
3558 | * instead of preempt_schedule() to exit user context if needed before | |
3559 | * calling the scheduler. | |
3560 | */ | |
4eaca0a8 | 3561 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3562 | { |
3563 | enum ctx_state prev_ctx; | |
3564 | ||
3565 | if (likely(!preemptible())) | |
3566 | return; | |
3567 | ||
3568 | do { | |
47252cfb SR |
3569 | /* |
3570 | * Because the function tracer can trace preempt_count_sub() | |
3571 | * and it also uses preempt_enable/disable_notrace(), if | |
3572 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3573 | * by the function tracer will call this function again and | |
3574 | * cause infinite recursion. | |
3575 | * | |
3576 | * Preemption must be disabled here before the function | |
3577 | * tracer can trace. Break up preempt_disable() into two | |
3578 | * calls. One to disable preemption without fear of being | |
3579 | * traced. The other to still record the preemption latency, | |
3580 | * which can also be traced by the function tracer. | |
3581 | */ | |
3d8f74dd | 3582 | preempt_disable_notrace(); |
47252cfb | 3583 | preempt_latency_start(1); |
009f60e2 ON |
3584 | /* |
3585 | * Needs preempt disabled in case user_exit() is traced | |
3586 | * and the tracer calls preempt_enable_notrace() causing | |
3587 | * an infinite recursion. | |
3588 | */ | |
3589 | prev_ctx = exception_enter(); | |
fc13aeba | 3590 | __schedule(true); |
009f60e2 ON |
3591 | exception_exit(prev_ctx); |
3592 | ||
47252cfb | 3593 | preempt_latency_stop(1); |
3d8f74dd | 3594 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3595 | } while (need_resched()); |
3596 | } | |
4eaca0a8 | 3597 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3598 | |
32e475d7 | 3599 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3600 | |
3601 | /* | |
2ed6e34f | 3602 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3603 | * off of irq context. |
3604 | * Note, that this is called and return with irqs disabled. This will | |
3605 | * protect us against recursive calling from irq. | |
3606 | */ | |
722a9f92 | 3607 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3608 | { |
b22366cd | 3609 | enum ctx_state prev_state; |
6478d880 | 3610 | |
2ed6e34f | 3611 | /* Catch callers which need to be fixed */ |
f27dde8d | 3612 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3613 | |
b22366cd FW |
3614 | prev_state = exception_enter(); |
3615 | ||
3a5c359a | 3616 | do { |
3d8f74dd | 3617 | preempt_disable(); |
3a5c359a | 3618 | local_irq_enable(); |
fc13aeba | 3619 | __schedule(true); |
3a5c359a | 3620 | local_irq_disable(); |
3d8f74dd | 3621 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3622 | } while (need_resched()); |
b22366cd FW |
3623 | |
3624 | exception_exit(prev_state); | |
1da177e4 LT |
3625 | } |
3626 | ||
63859d4f | 3627 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3628 | void *key) |
1da177e4 | 3629 | { |
63859d4f | 3630 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3631 | } |
1da177e4 LT |
3632 | EXPORT_SYMBOL(default_wake_function); |
3633 | ||
b29739f9 IM |
3634 | #ifdef CONFIG_RT_MUTEXES |
3635 | ||
3636 | /* | |
3637 | * rt_mutex_setprio - set the current priority of a task | |
3638 | * @p: task | |
3639 | * @prio: prio value (kernel-internal form) | |
3640 | * | |
3641 | * This function changes the 'effective' priority of a task. It does | |
3642 | * not touch ->normal_prio like __setscheduler(). | |
3643 | * | |
c365c292 TG |
3644 | * Used by the rt_mutex code to implement priority inheritance |
3645 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3646 | */ |
36c8b586 | 3647 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3648 | { |
ff77e468 | 3649 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3650 | const struct sched_class *prev_class; |
eb580751 PZ |
3651 | struct rq_flags rf; |
3652 | struct rq *rq; | |
b29739f9 | 3653 | |
aab03e05 | 3654 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3655 | |
eb580751 | 3656 | rq = __task_rq_lock(p, &rf); |
b29739f9 | 3657 | |
1c4dd99b TG |
3658 | /* |
3659 | * Idle task boosting is a nono in general. There is one | |
3660 | * exception, when PREEMPT_RT and NOHZ is active: | |
3661 | * | |
3662 | * The idle task calls get_next_timer_interrupt() and holds | |
3663 | * the timer wheel base->lock on the CPU and another CPU wants | |
3664 | * to access the timer (probably to cancel it). We can safely | |
3665 | * ignore the boosting request, as the idle CPU runs this code | |
3666 | * with interrupts disabled and will complete the lock | |
3667 | * protected section without being interrupted. So there is no | |
3668 | * real need to boost. | |
3669 | */ | |
3670 | if (unlikely(p == rq->idle)) { | |
3671 | WARN_ON(p != rq->curr); | |
3672 | WARN_ON(p->pi_blocked_on); | |
3673 | goto out_unlock; | |
3674 | } | |
3675 | ||
a8027073 | 3676 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3677 | oldprio = p->prio; |
ff77e468 PZ |
3678 | |
3679 | if (oldprio == prio) | |
3680 | queue_flag &= ~DEQUEUE_MOVE; | |
3681 | ||
83ab0aa0 | 3682 | prev_class = p->sched_class; |
da0c1e65 | 3683 | queued = task_on_rq_queued(p); |
051a1d1a | 3684 | running = task_current(rq, p); |
da0c1e65 | 3685 | if (queued) |
ff77e468 | 3686 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3687 | if (running) |
f3cd1c4e | 3688 | put_prev_task(rq, p); |
dd41f596 | 3689 | |
2d3d891d DF |
3690 | /* |
3691 | * Boosting condition are: | |
3692 | * 1. -rt task is running and holds mutex A | |
3693 | * --> -dl task blocks on mutex A | |
3694 | * | |
3695 | * 2. -dl task is running and holds mutex A | |
3696 | * --> -dl task blocks on mutex A and could preempt the | |
3697 | * running task | |
3698 | */ | |
3699 | if (dl_prio(prio)) { | |
466af29b ON |
3700 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3701 | if (!dl_prio(p->normal_prio) || | |
3702 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3703 | p->dl.dl_boosted = 1; |
ff77e468 | 3704 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3705 | } else |
3706 | p->dl.dl_boosted = 0; | |
aab03e05 | 3707 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3708 | } else if (rt_prio(prio)) { |
3709 | if (dl_prio(oldprio)) | |
3710 | p->dl.dl_boosted = 0; | |
3711 | if (oldprio < prio) | |
ff77e468 | 3712 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3713 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3714 | } else { |
3715 | if (dl_prio(oldprio)) | |
3716 | p->dl.dl_boosted = 0; | |
746db944 BS |
3717 | if (rt_prio(oldprio)) |
3718 | p->rt.timeout = 0; | |
dd41f596 | 3719 | p->sched_class = &fair_sched_class; |
2d3d891d | 3720 | } |
dd41f596 | 3721 | |
b29739f9 IM |
3722 | p->prio = prio; |
3723 | ||
da0c1e65 | 3724 | if (queued) |
ff77e468 | 3725 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3726 | if (running) |
b2bf6c31 | 3727 | set_curr_task(rq, p); |
cb469845 | 3728 | |
da7a735e | 3729 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3730 | out_unlock: |
4c9a4bc8 | 3731 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 3732 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3733 | |
3734 | balance_callback(rq); | |
3735 | preempt_enable(); | |
b29739f9 | 3736 | } |
b29739f9 | 3737 | #endif |
d50dde5a | 3738 | |
36c8b586 | 3739 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3740 | { |
49bd21ef PZ |
3741 | bool queued, running; |
3742 | int old_prio, delta; | |
eb580751 | 3743 | struct rq_flags rf; |
70b97a7f | 3744 | struct rq *rq; |
1da177e4 | 3745 | |
75e45d51 | 3746 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3747 | return; |
3748 | /* | |
3749 | * We have to be careful, if called from sys_setpriority(), | |
3750 | * the task might be in the middle of scheduling on another CPU. | |
3751 | */ | |
eb580751 | 3752 | rq = task_rq_lock(p, &rf); |
1da177e4 LT |
3753 | /* |
3754 | * The RT priorities are set via sched_setscheduler(), but we still | |
3755 | * allow the 'normal' nice value to be set - but as expected | |
3756 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3757 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3758 | */ |
aab03e05 | 3759 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3760 | p->static_prio = NICE_TO_PRIO(nice); |
3761 | goto out_unlock; | |
3762 | } | |
da0c1e65 | 3763 | queued = task_on_rq_queued(p); |
49bd21ef | 3764 | running = task_current(rq, p); |
da0c1e65 | 3765 | if (queued) |
1de64443 | 3766 | dequeue_task(rq, p, DEQUEUE_SAVE); |
49bd21ef PZ |
3767 | if (running) |
3768 | put_prev_task(rq, p); | |
1da177e4 | 3769 | |
1da177e4 | 3770 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3771 | set_load_weight(p); |
b29739f9 IM |
3772 | old_prio = p->prio; |
3773 | p->prio = effective_prio(p); | |
3774 | delta = p->prio - old_prio; | |
1da177e4 | 3775 | |
da0c1e65 | 3776 | if (queued) { |
1de64443 | 3777 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3778 | /* |
d5f9f942 AM |
3779 | * If the task increased its priority or is running and |
3780 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3781 | */ |
d5f9f942 | 3782 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3783 | resched_curr(rq); |
1da177e4 | 3784 | } |
49bd21ef PZ |
3785 | if (running) |
3786 | set_curr_task(rq, p); | |
1da177e4 | 3787 | out_unlock: |
eb580751 | 3788 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3789 | } |
1da177e4 LT |
3790 | EXPORT_SYMBOL(set_user_nice); |
3791 | ||
e43379f1 MM |
3792 | /* |
3793 | * can_nice - check if a task can reduce its nice value | |
3794 | * @p: task | |
3795 | * @nice: nice value | |
3796 | */ | |
36c8b586 | 3797 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3798 | { |
024f4747 | 3799 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3800 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3801 | |
78d7d407 | 3802 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3803 | capable(CAP_SYS_NICE)); |
3804 | } | |
3805 | ||
1da177e4 LT |
3806 | #ifdef __ARCH_WANT_SYS_NICE |
3807 | ||
3808 | /* | |
3809 | * sys_nice - change the priority of the current process. | |
3810 | * @increment: priority increment | |
3811 | * | |
3812 | * sys_setpriority is a more generic, but much slower function that | |
3813 | * does similar things. | |
3814 | */ | |
5add95d4 | 3815 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3816 | { |
48f24c4d | 3817 | long nice, retval; |
1da177e4 LT |
3818 | |
3819 | /* | |
3820 | * Setpriority might change our priority at the same moment. | |
3821 | * We don't have to worry. Conceptually one call occurs first | |
3822 | * and we have a single winner. | |
3823 | */ | |
a9467fa3 | 3824 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3825 | nice = task_nice(current) + increment; |
1da177e4 | 3826 | |
a9467fa3 | 3827 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3828 | if (increment < 0 && !can_nice(current, nice)) |
3829 | return -EPERM; | |
3830 | ||
1da177e4 LT |
3831 | retval = security_task_setnice(current, nice); |
3832 | if (retval) | |
3833 | return retval; | |
3834 | ||
3835 | set_user_nice(current, nice); | |
3836 | return 0; | |
3837 | } | |
3838 | ||
3839 | #endif | |
3840 | ||
3841 | /** | |
3842 | * task_prio - return the priority value of a given task. | |
3843 | * @p: the task in question. | |
3844 | * | |
e69f6186 | 3845 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3846 | * RT tasks are offset by -200. Normal tasks are centered |
3847 | * around 0, value goes from -16 to +15. | |
3848 | */ | |
36c8b586 | 3849 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3850 | { |
3851 | return p->prio - MAX_RT_PRIO; | |
3852 | } | |
3853 | ||
1da177e4 LT |
3854 | /** |
3855 | * idle_cpu - is a given cpu idle currently? | |
3856 | * @cpu: the processor in question. | |
e69f6186 YB |
3857 | * |
3858 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3859 | */ |
3860 | int idle_cpu(int cpu) | |
3861 | { | |
908a3283 TG |
3862 | struct rq *rq = cpu_rq(cpu); |
3863 | ||
3864 | if (rq->curr != rq->idle) | |
3865 | return 0; | |
3866 | ||
3867 | if (rq->nr_running) | |
3868 | return 0; | |
3869 | ||
3870 | #ifdef CONFIG_SMP | |
3871 | if (!llist_empty(&rq->wake_list)) | |
3872 | return 0; | |
3873 | #endif | |
3874 | ||
3875 | return 1; | |
1da177e4 LT |
3876 | } |
3877 | ||
1da177e4 LT |
3878 | /** |
3879 | * idle_task - return the idle task for a given cpu. | |
3880 | * @cpu: the processor in question. | |
e69f6186 YB |
3881 | * |
3882 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3883 | */ |
36c8b586 | 3884 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3885 | { |
3886 | return cpu_rq(cpu)->idle; | |
3887 | } | |
3888 | ||
3889 | /** | |
3890 | * find_process_by_pid - find a process with a matching PID value. | |
3891 | * @pid: the pid in question. | |
e69f6186 YB |
3892 | * |
3893 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3894 | */ |
a9957449 | 3895 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3896 | { |
228ebcbe | 3897 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3898 | } |
3899 | ||
aab03e05 DF |
3900 | /* |
3901 | * This function initializes the sched_dl_entity of a newly becoming | |
3902 | * SCHED_DEADLINE task. | |
3903 | * | |
3904 | * Only the static values are considered here, the actual runtime and the | |
3905 | * absolute deadline will be properly calculated when the task is enqueued | |
3906 | * for the first time with its new policy. | |
3907 | */ | |
3908 | static void | |
3909 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3910 | { | |
3911 | struct sched_dl_entity *dl_se = &p->dl; | |
3912 | ||
aab03e05 DF |
3913 | dl_se->dl_runtime = attr->sched_runtime; |
3914 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3915 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3916 | dl_se->flags = attr->sched_flags; |
332ac17e | 3917 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3918 | |
3919 | /* | |
3920 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3921 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3922 | * | |
3923 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3924 | * point. This would include retaining the task_struct until that time | |
3925 | * and change dl_overflow() to not immediately decrement the current | |
3926 | * amount. | |
3927 | * | |
3928 | * Instead we retain the current runtime/deadline and let the new | |
3929 | * parameters take effect after the current reservation period lapses. | |
3930 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3931 | * before the current scheduling deadline. | |
3932 | * | |
3933 | * We can still have temporary overloads because we do not delay the | |
3934 | * change in bandwidth until that time; so admission control is | |
3935 | * not on the safe side. It does however guarantee tasks will never | |
3936 | * consume more than promised. | |
3937 | */ | |
aab03e05 DF |
3938 | } |
3939 | ||
c13db6b1 SR |
3940 | /* |
3941 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3942 | * it calls know not to change it. | |
3943 | */ | |
3944 | #define SETPARAM_POLICY -1 | |
3945 | ||
c365c292 TG |
3946 | static void __setscheduler_params(struct task_struct *p, |
3947 | const struct sched_attr *attr) | |
1da177e4 | 3948 | { |
d50dde5a DF |
3949 | int policy = attr->sched_policy; |
3950 | ||
c13db6b1 | 3951 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3952 | policy = p->policy; |
3953 | ||
1da177e4 | 3954 | p->policy = policy; |
d50dde5a | 3955 | |
aab03e05 DF |
3956 | if (dl_policy(policy)) |
3957 | __setparam_dl(p, attr); | |
39fd8fd2 | 3958 | else if (fair_policy(policy)) |
d50dde5a DF |
3959 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3960 | ||
39fd8fd2 PZ |
3961 | /* |
3962 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3963 | * !rt_policy. Always setting this ensures that things like | |
3964 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3965 | */ | |
3966 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3967 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3968 | set_load_weight(p); |
3969 | } | |
39fd8fd2 | 3970 | |
c365c292 TG |
3971 | /* Actually do priority change: must hold pi & rq lock. */ |
3972 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3973 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3974 | { |
3975 | __setscheduler_params(p, attr); | |
d50dde5a | 3976 | |
383afd09 | 3977 | /* |
0782e63b TG |
3978 | * Keep a potential priority boosting if called from |
3979 | * sched_setscheduler(). | |
383afd09 | 3980 | */ |
0782e63b TG |
3981 | if (keep_boost) |
3982 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3983 | else | |
3984 | p->prio = normal_prio(p); | |
383afd09 | 3985 | |
aab03e05 DF |
3986 | if (dl_prio(p->prio)) |
3987 | p->sched_class = &dl_sched_class; | |
3988 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3989 | p->sched_class = &rt_sched_class; |
3990 | else | |
3991 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3992 | } |
aab03e05 DF |
3993 | |
3994 | static void | |
3995 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3996 | { | |
3997 | struct sched_dl_entity *dl_se = &p->dl; | |
3998 | ||
3999 | attr->sched_priority = p->rt_priority; | |
4000 | attr->sched_runtime = dl_se->dl_runtime; | |
4001 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 4002 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
4003 | attr->sched_flags = dl_se->flags; |
4004 | } | |
4005 | ||
4006 | /* | |
4007 | * This function validates the new parameters of a -deadline task. | |
4008 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 4009 | * than the runtime, as well as the period of being zero or |
332ac17e | 4010 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
4011 | * user parameters are above the internal resolution of 1us (we |
4012 | * check sched_runtime only since it is always the smaller one) and | |
4013 | * below 2^63 ns (we have to check both sched_deadline and | |
4014 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
4015 | */ |
4016 | static bool | |
4017 | __checkparam_dl(const struct sched_attr *attr) | |
4018 | { | |
b0827819 JL |
4019 | /* deadline != 0 */ |
4020 | if (attr->sched_deadline == 0) | |
4021 | return false; | |
4022 | ||
4023 | /* | |
4024 | * Since we truncate DL_SCALE bits, make sure we're at least | |
4025 | * that big. | |
4026 | */ | |
4027 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
4028 | return false; | |
4029 | ||
4030 | /* | |
4031 | * Since we use the MSB for wrap-around and sign issues, make | |
4032 | * sure it's not set (mind that period can be equal to zero). | |
4033 | */ | |
4034 | if (attr->sched_deadline & (1ULL << 63) || | |
4035 | attr->sched_period & (1ULL << 63)) | |
4036 | return false; | |
4037 | ||
4038 | /* runtime <= deadline <= period (if period != 0) */ | |
4039 | if ((attr->sched_period != 0 && | |
4040 | attr->sched_period < attr->sched_deadline) || | |
4041 | attr->sched_deadline < attr->sched_runtime) | |
4042 | return false; | |
4043 | ||
4044 | return true; | |
aab03e05 DF |
4045 | } |
4046 | ||
c69e8d9c DH |
4047 | /* |
4048 | * check the target process has a UID that matches the current process's | |
4049 | */ | |
4050 | static bool check_same_owner(struct task_struct *p) | |
4051 | { | |
4052 | const struct cred *cred = current_cred(), *pcred; | |
4053 | bool match; | |
4054 | ||
4055 | rcu_read_lock(); | |
4056 | pcred = __task_cred(p); | |
9c806aa0 EB |
4057 | match = (uid_eq(cred->euid, pcred->euid) || |
4058 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4059 | rcu_read_unlock(); |
4060 | return match; | |
4061 | } | |
4062 | ||
75381608 WL |
4063 | static bool dl_param_changed(struct task_struct *p, |
4064 | const struct sched_attr *attr) | |
4065 | { | |
4066 | struct sched_dl_entity *dl_se = &p->dl; | |
4067 | ||
4068 | if (dl_se->dl_runtime != attr->sched_runtime || | |
4069 | dl_se->dl_deadline != attr->sched_deadline || | |
4070 | dl_se->dl_period != attr->sched_period || | |
4071 | dl_se->flags != attr->sched_flags) | |
4072 | return true; | |
4073 | ||
4074 | return false; | |
4075 | } | |
4076 | ||
d50dde5a DF |
4077 | static int __sched_setscheduler(struct task_struct *p, |
4078 | const struct sched_attr *attr, | |
dbc7f069 | 4079 | bool user, bool pi) |
1da177e4 | 4080 | { |
383afd09 SR |
4081 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4082 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4083 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4084 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4085 | const struct sched_class *prev_class; |
eb580751 | 4086 | struct rq_flags rf; |
ca94c442 | 4087 | int reset_on_fork; |
ff77e468 | 4088 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 4089 | struct rq *rq; |
1da177e4 | 4090 | |
66e5393a SR |
4091 | /* may grab non-irq protected spin_locks */ |
4092 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4093 | recheck: |
4094 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4095 | if (policy < 0) { |
4096 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4097 | policy = oldpolicy = p->policy; |
ca94c442 | 4098 | } else { |
7479f3c9 | 4099 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4100 | |
20f9cd2a | 4101 | if (!valid_policy(policy)) |
ca94c442 LP |
4102 | return -EINVAL; |
4103 | } | |
4104 | ||
7479f3c9 PZ |
4105 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
4106 | return -EINVAL; | |
4107 | ||
1da177e4 LT |
4108 | /* |
4109 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4110 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4111 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4112 | */ |
0bb040a4 | 4113 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4114 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4115 | return -EINVAL; |
aab03e05 DF |
4116 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4117 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4118 | return -EINVAL; |
4119 | ||
37e4ab3f OC |
4120 | /* |
4121 | * Allow unprivileged RT tasks to decrease priority: | |
4122 | */ | |
961ccddd | 4123 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4124 | if (fair_policy(policy)) { |
d0ea0268 | 4125 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4126 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4127 | return -EPERM; |
4128 | } | |
4129 | ||
e05606d3 | 4130 | if (rt_policy(policy)) { |
a44702e8 ON |
4131 | unsigned long rlim_rtprio = |
4132 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4133 | |
4134 | /* can't set/change the rt policy */ | |
4135 | if (policy != p->policy && !rlim_rtprio) | |
4136 | return -EPERM; | |
4137 | ||
4138 | /* can't increase priority */ | |
d50dde5a DF |
4139 | if (attr->sched_priority > p->rt_priority && |
4140 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4141 | return -EPERM; |
4142 | } | |
c02aa73b | 4143 | |
d44753b8 JL |
4144 | /* |
4145 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4146 | * (safest behavior); in the future we would like to allow | |
4147 | * unprivileged DL tasks to increase their relative deadline | |
4148 | * or reduce their runtime (both ways reducing utilization) | |
4149 | */ | |
4150 | if (dl_policy(policy)) | |
4151 | return -EPERM; | |
4152 | ||
dd41f596 | 4153 | /* |
c02aa73b DH |
4154 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4155 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4156 | */ |
20f9cd2a | 4157 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4158 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4159 | return -EPERM; |
4160 | } | |
5fe1d75f | 4161 | |
37e4ab3f | 4162 | /* can't change other user's priorities */ |
c69e8d9c | 4163 | if (!check_same_owner(p)) |
37e4ab3f | 4164 | return -EPERM; |
ca94c442 LP |
4165 | |
4166 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4167 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4168 | return -EPERM; | |
37e4ab3f | 4169 | } |
1da177e4 | 4170 | |
725aad24 | 4171 | if (user) { |
b0ae1981 | 4172 | retval = security_task_setscheduler(p); |
725aad24 JF |
4173 | if (retval) |
4174 | return retval; | |
4175 | } | |
4176 | ||
b29739f9 IM |
4177 | /* |
4178 | * make sure no PI-waiters arrive (or leave) while we are | |
4179 | * changing the priority of the task: | |
0122ec5b | 4180 | * |
25985edc | 4181 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4182 | * runqueue lock must be held. |
4183 | */ | |
eb580751 | 4184 | rq = task_rq_lock(p, &rf); |
dc61b1d6 | 4185 | |
34f971f6 PZ |
4186 | /* |
4187 | * Changing the policy of the stop threads its a very bad idea | |
4188 | */ | |
4189 | if (p == rq->stop) { | |
eb580751 | 4190 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4191 | return -EINVAL; |
4192 | } | |
4193 | ||
a51e9198 | 4194 | /* |
d6b1e911 TG |
4195 | * If not changing anything there's no need to proceed further, |
4196 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4197 | */ |
d50dde5a | 4198 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4199 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4200 | goto change; |
4201 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4202 | goto change; | |
75381608 | 4203 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4204 | goto change; |
d50dde5a | 4205 | |
d6b1e911 | 4206 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4207 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4208 | return 0; |
4209 | } | |
d50dde5a | 4210 | change: |
a51e9198 | 4211 | |
dc61b1d6 | 4212 | if (user) { |
332ac17e | 4213 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4214 | /* |
4215 | * Do not allow realtime tasks into groups that have no runtime | |
4216 | * assigned. | |
4217 | */ | |
4218 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4219 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4220 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4221 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4222 | return -EPERM; |
4223 | } | |
dc61b1d6 | 4224 | #endif |
332ac17e DF |
4225 | #ifdef CONFIG_SMP |
4226 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4227 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4228 | |
4229 | /* | |
4230 | * Don't allow tasks with an affinity mask smaller than | |
4231 | * the entire root_domain to become SCHED_DEADLINE. We | |
4232 | * will also fail if there's no bandwidth available. | |
4233 | */ | |
e4099a5e PZ |
4234 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4235 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4236 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4237 | return -EPERM; |
4238 | } | |
4239 | } | |
4240 | #endif | |
4241 | } | |
dc61b1d6 | 4242 | |
1da177e4 LT |
4243 | /* recheck policy now with rq lock held */ |
4244 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4245 | policy = oldpolicy = -1; | |
eb580751 | 4246 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4247 | goto recheck; |
4248 | } | |
332ac17e DF |
4249 | |
4250 | /* | |
4251 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4252 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4253 | * is available. | |
4254 | */ | |
e4099a5e | 4255 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4256 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4257 | return -EBUSY; |
4258 | } | |
4259 | ||
c365c292 TG |
4260 | p->sched_reset_on_fork = reset_on_fork; |
4261 | oldprio = p->prio; | |
4262 | ||
dbc7f069 PZ |
4263 | if (pi) { |
4264 | /* | |
4265 | * Take priority boosted tasks into account. If the new | |
4266 | * effective priority is unchanged, we just store the new | |
4267 | * normal parameters and do not touch the scheduler class and | |
4268 | * the runqueue. This will be done when the task deboost | |
4269 | * itself. | |
4270 | */ | |
4271 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4272 | if (new_effective_prio == oldprio) |
4273 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4274 | } |
4275 | ||
da0c1e65 | 4276 | queued = task_on_rq_queued(p); |
051a1d1a | 4277 | running = task_current(rq, p); |
da0c1e65 | 4278 | if (queued) |
ff77e468 | 4279 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4280 | if (running) |
f3cd1c4e | 4281 | put_prev_task(rq, p); |
f6b53205 | 4282 | |
83ab0aa0 | 4283 | prev_class = p->sched_class; |
dbc7f069 | 4284 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4285 | |
da0c1e65 | 4286 | if (queued) { |
81a44c54 TG |
4287 | /* |
4288 | * We enqueue to tail when the priority of a task is | |
4289 | * increased (user space view). | |
4290 | */ | |
ff77e468 PZ |
4291 | if (oldprio < p->prio) |
4292 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4293 | |
ff77e468 | 4294 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4295 | } |
a399d233 | 4296 | if (running) |
b2bf6c31 | 4297 | set_curr_task(rq, p); |
cb469845 | 4298 | |
da7a735e | 4299 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4300 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 4301 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4302 | |
dbc7f069 PZ |
4303 | if (pi) |
4304 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4305 | |
4c9a4bc8 PZ |
4306 | /* |
4307 | * Run balance callbacks after we've adjusted the PI chain. | |
4308 | */ | |
4309 | balance_callback(rq); | |
4310 | preempt_enable(); | |
95e02ca9 | 4311 | |
1da177e4 LT |
4312 | return 0; |
4313 | } | |
961ccddd | 4314 | |
7479f3c9 PZ |
4315 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4316 | const struct sched_param *param, bool check) | |
4317 | { | |
4318 | struct sched_attr attr = { | |
4319 | .sched_policy = policy, | |
4320 | .sched_priority = param->sched_priority, | |
4321 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4322 | }; | |
4323 | ||
c13db6b1 SR |
4324 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4325 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4326 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4327 | policy &= ~SCHED_RESET_ON_FORK; | |
4328 | attr.sched_policy = policy; | |
4329 | } | |
4330 | ||
dbc7f069 | 4331 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4332 | } |
961ccddd RR |
4333 | /** |
4334 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4335 | * @p: the task in question. | |
4336 | * @policy: new policy. | |
4337 | * @param: structure containing the new RT priority. | |
4338 | * | |
e69f6186 YB |
4339 | * Return: 0 on success. An error code otherwise. |
4340 | * | |
961ccddd RR |
4341 | * NOTE that the task may be already dead. |
4342 | */ | |
4343 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4344 | const struct sched_param *param) |
961ccddd | 4345 | { |
7479f3c9 | 4346 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4347 | } |
1da177e4 LT |
4348 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4349 | ||
d50dde5a DF |
4350 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4351 | { | |
dbc7f069 | 4352 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4353 | } |
4354 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4355 | ||
961ccddd RR |
4356 | /** |
4357 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4358 | * @p: the task in question. | |
4359 | * @policy: new policy. | |
4360 | * @param: structure containing the new RT priority. | |
4361 | * | |
4362 | * Just like sched_setscheduler, only don't bother checking if the | |
4363 | * current context has permission. For example, this is needed in | |
4364 | * stop_machine(): we create temporary high priority worker threads, | |
4365 | * but our caller might not have that capability. | |
e69f6186 YB |
4366 | * |
4367 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4368 | */ |
4369 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4370 | const struct sched_param *param) |
961ccddd | 4371 | { |
7479f3c9 | 4372 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4373 | } |
84778472 | 4374 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4375 | |
95cdf3b7 IM |
4376 | static int |
4377 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4378 | { |
1da177e4 LT |
4379 | struct sched_param lparam; |
4380 | struct task_struct *p; | |
36c8b586 | 4381 | int retval; |
1da177e4 LT |
4382 | |
4383 | if (!param || pid < 0) | |
4384 | return -EINVAL; | |
4385 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4386 | return -EFAULT; | |
5fe1d75f ON |
4387 | |
4388 | rcu_read_lock(); | |
4389 | retval = -ESRCH; | |
1da177e4 | 4390 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4391 | if (p != NULL) |
4392 | retval = sched_setscheduler(p, policy, &lparam); | |
4393 | rcu_read_unlock(); | |
36c8b586 | 4394 | |
1da177e4 LT |
4395 | return retval; |
4396 | } | |
4397 | ||
d50dde5a DF |
4398 | /* |
4399 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4400 | */ | |
4401 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4402 | struct sched_attr *attr) | |
4403 | { | |
4404 | u32 size; | |
4405 | int ret; | |
4406 | ||
4407 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4408 | return -EFAULT; | |
4409 | ||
4410 | /* | |
4411 | * zero the full structure, so that a short copy will be nice. | |
4412 | */ | |
4413 | memset(attr, 0, sizeof(*attr)); | |
4414 | ||
4415 | ret = get_user(size, &uattr->size); | |
4416 | if (ret) | |
4417 | return ret; | |
4418 | ||
4419 | if (size > PAGE_SIZE) /* silly large */ | |
4420 | goto err_size; | |
4421 | ||
4422 | if (!size) /* abi compat */ | |
4423 | size = SCHED_ATTR_SIZE_VER0; | |
4424 | ||
4425 | if (size < SCHED_ATTR_SIZE_VER0) | |
4426 | goto err_size; | |
4427 | ||
4428 | /* | |
4429 | * If we're handed a bigger struct than we know of, | |
4430 | * ensure all the unknown bits are 0 - i.e. new | |
4431 | * user-space does not rely on any kernel feature | |
4432 | * extensions we dont know about yet. | |
4433 | */ | |
4434 | if (size > sizeof(*attr)) { | |
4435 | unsigned char __user *addr; | |
4436 | unsigned char __user *end; | |
4437 | unsigned char val; | |
4438 | ||
4439 | addr = (void __user *)uattr + sizeof(*attr); | |
4440 | end = (void __user *)uattr + size; | |
4441 | ||
4442 | for (; addr < end; addr++) { | |
4443 | ret = get_user(val, addr); | |
4444 | if (ret) | |
4445 | return ret; | |
4446 | if (val) | |
4447 | goto err_size; | |
4448 | } | |
4449 | size = sizeof(*attr); | |
4450 | } | |
4451 | ||
4452 | ret = copy_from_user(attr, uattr, size); | |
4453 | if (ret) | |
4454 | return -EFAULT; | |
4455 | ||
4456 | /* | |
4457 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4458 | * to be strict and return an error on out-of-bounds values? | |
4459 | */ | |
75e45d51 | 4460 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4461 | |
e78c7bca | 4462 | return 0; |
d50dde5a DF |
4463 | |
4464 | err_size: | |
4465 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4466 | return -E2BIG; |
d50dde5a DF |
4467 | } |
4468 | ||
1da177e4 LT |
4469 | /** |
4470 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4471 | * @pid: the pid in question. | |
4472 | * @policy: new policy. | |
4473 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4474 | * |
4475 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4476 | */ |
5add95d4 HC |
4477 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4478 | struct sched_param __user *, param) | |
1da177e4 | 4479 | { |
c21761f1 JB |
4480 | /* negative values for policy are not valid */ |
4481 | if (policy < 0) | |
4482 | return -EINVAL; | |
4483 | ||
1da177e4 LT |
4484 | return do_sched_setscheduler(pid, policy, param); |
4485 | } | |
4486 | ||
4487 | /** | |
4488 | * sys_sched_setparam - set/change the RT priority of a thread | |
4489 | * @pid: the pid in question. | |
4490 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4491 | * |
4492 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4493 | */ |
5add95d4 | 4494 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4495 | { |
c13db6b1 | 4496 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4497 | } |
4498 | ||
d50dde5a DF |
4499 | /** |
4500 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4501 | * @pid: the pid in question. | |
5778fccf | 4502 | * @uattr: structure containing the extended parameters. |
db66d756 | 4503 | * @flags: for future extension. |
d50dde5a | 4504 | */ |
6d35ab48 PZ |
4505 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4506 | unsigned int, flags) | |
d50dde5a DF |
4507 | { |
4508 | struct sched_attr attr; | |
4509 | struct task_struct *p; | |
4510 | int retval; | |
4511 | ||
6d35ab48 | 4512 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4513 | return -EINVAL; |
4514 | ||
143cf23d MK |
4515 | retval = sched_copy_attr(uattr, &attr); |
4516 | if (retval) | |
4517 | return retval; | |
d50dde5a | 4518 | |
b14ed2c2 | 4519 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4520 | return -EINVAL; |
d50dde5a DF |
4521 | |
4522 | rcu_read_lock(); | |
4523 | retval = -ESRCH; | |
4524 | p = find_process_by_pid(pid); | |
4525 | if (p != NULL) | |
4526 | retval = sched_setattr(p, &attr); | |
4527 | rcu_read_unlock(); | |
4528 | ||
4529 | return retval; | |
4530 | } | |
4531 | ||
1da177e4 LT |
4532 | /** |
4533 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4534 | * @pid: the pid in question. | |
e69f6186 YB |
4535 | * |
4536 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4537 | * code. | |
1da177e4 | 4538 | */ |
5add95d4 | 4539 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4540 | { |
36c8b586 | 4541 | struct task_struct *p; |
3a5c359a | 4542 | int retval; |
1da177e4 LT |
4543 | |
4544 | if (pid < 0) | |
3a5c359a | 4545 | return -EINVAL; |
1da177e4 LT |
4546 | |
4547 | retval = -ESRCH; | |
5fe85be0 | 4548 | rcu_read_lock(); |
1da177e4 LT |
4549 | p = find_process_by_pid(pid); |
4550 | if (p) { | |
4551 | retval = security_task_getscheduler(p); | |
4552 | if (!retval) | |
ca94c442 LP |
4553 | retval = p->policy |
4554 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4555 | } |
5fe85be0 | 4556 | rcu_read_unlock(); |
1da177e4 LT |
4557 | return retval; |
4558 | } | |
4559 | ||
4560 | /** | |
ca94c442 | 4561 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4562 | * @pid: the pid in question. |
4563 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4564 | * |
4565 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4566 | * code. | |
1da177e4 | 4567 | */ |
5add95d4 | 4568 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4569 | { |
ce5f7f82 | 4570 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4571 | struct task_struct *p; |
3a5c359a | 4572 | int retval; |
1da177e4 LT |
4573 | |
4574 | if (!param || pid < 0) | |
3a5c359a | 4575 | return -EINVAL; |
1da177e4 | 4576 | |
5fe85be0 | 4577 | rcu_read_lock(); |
1da177e4 LT |
4578 | p = find_process_by_pid(pid); |
4579 | retval = -ESRCH; | |
4580 | if (!p) | |
4581 | goto out_unlock; | |
4582 | ||
4583 | retval = security_task_getscheduler(p); | |
4584 | if (retval) | |
4585 | goto out_unlock; | |
4586 | ||
ce5f7f82 PZ |
4587 | if (task_has_rt_policy(p)) |
4588 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4589 | rcu_read_unlock(); |
1da177e4 LT |
4590 | |
4591 | /* | |
4592 | * This one might sleep, we cannot do it with a spinlock held ... | |
4593 | */ | |
4594 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4595 | ||
1da177e4 LT |
4596 | return retval; |
4597 | ||
4598 | out_unlock: | |
5fe85be0 | 4599 | rcu_read_unlock(); |
1da177e4 LT |
4600 | return retval; |
4601 | } | |
4602 | ||
d50dde5a DF |
4603 | static int sched_read_attr(struct sched_attr __user *uattr, |
4604 | struct sched_attr *attr, | |
4605 | unsigned int usize) | |
4606 | { | |
4607 | int ret; | |
4608 | ||
4609 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4610 | return -EFAULT; | |
4611 | ||
4612 | /* | |
4613 | * If we're handed a smaller struct than we know of, | |
4614 | * ensure all the unknown bits are 0 - i.e. old | |
4615 | * user-space does not get uncomplete information. | |
4616 | */ | |
4617 | if (usize < sizeof(*attr)) { | |
4618 | unsigned char *addr; | |
4619 | unsigned char *end; | |
4620 | ||
4621 | addr = (void *)attr + usize; | |
4622 | end = (void *)attr + sizeof(*attr); | |
4623 | ||
4624 | for (; addr < end; addr++) { | |
4625 | if (*addr) | |
22400674 | 4626 | return -EFBIG; |
d50dde5a DF |
4627 | } |
4628 | ||
4629 | attr->size = usize; | |
4630 | } | |
4631 | ||
4efbc454 | 4632 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4633 | if (ret) |
4634 | return -EFAULT; | |
4635 | ||
22400674 | 4636 | return 0; |
d50dde5a DF |
4637 | } |
4638 | ||
4639 | /** | |
aab03e05 | 4640 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4641 | * @pid: the pid in question. |
5778fccf | 4642 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4643 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4644 | * @flags: for future extension. |
d50dde5a | 4645 | */ |
6d35ab48 PZ |
4646 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4647 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4648 | { |
4649 | struct sched_attr attr = { | |
4650 | .size = sizeof(struct sched_attr), | |
4651 | }; | |
4652 | struct task_struct *p; | |
4653 | int retval; | |
4654 | ||
4655 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4656 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4657 | return -EINVAL; |
4658 | ||
4659 | rcu_read_lock(); | |
4660 | p = find_process_by_pid(pid); | |
4661 | retval = -ESRCH; | |
4662 | if (!p) | |
4663 | goto out_unlock; | |
4664 | ||
4665 | retval = security_task_getscheduler(p); | |
4666 | if (retval) | |
4667 | goto out_unlock; | |
4668 | ||
4669 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4670 | if (p->sched_reset_on_fork) |
4671 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4672 | if (task_has_dl_policy(p)) |
4673 | __getparam_dl(p, &attr); | |
4674 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4675 | attr.sched_priority = p->rt_priority; |
4676 | else | |
d0ea0268 | 4677 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4678 | |
4679 | rcu_read_unlock(); | |
4680 | ||
4681 | retval = sched_read_attr(uattr, &attr, size); | |
4682 | return retval; | |
4683 | ||
4684 | out_unlock: | |
4685 | rcu_read_unlock(); | |
4686 | return retval; | |
4687 | } | |
4688 | ||
96f874e2 | 4689 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4690 | { |
5a16f3d3 | 4691 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4692 | struct task_struct *p; |
4693 | int retval; | |
1da177e4 | 4694 | |
23f5d142 | 4695 | rcu_read_lock(); |
1da177e4 LT |
4696 | |
4697 | p = find_process_by_pid(pid); | |
4698 | if (!p) { | |
23f5d142 | 4699 | rcu_read_unlock(); |
1da177e4 LT |
4700 | return -ESRCH; |
4701 | } | |
4702 | ||
23f5d142 | 4703 | /* Prevent p going away */ |
1da177e4 | 4704 | get_task_struct(p); |
23f5d142 | 4705 | rcu_read_unlock(); |
1da177e4 | 4706 | |
14a40ffc TH |
4707 | if (p->flags & PF_NO_SETAFFINITY) { |
4708 | retval = -EINVAL; | |
4709 | goto out_put_task; | |
4710 | } | |
5a16f3d3 RR |
4711 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4712 | retval = -ENOMEM; | |
4713 | goto out_put_task; | |
4714 | } | |
4715 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4716 | retval = -ENOMEM; | |
4717 | goto out_free_cpus_allowed; | |
4718 | } | |
1da177e4 | 4719 | retval = -EPERM; |
4c44aaaf EB |
4720 | if (!check_same_owner(p)) { |
4721 | rcu_read_lock(); | |
4722 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4723 | rcu_read_unlock(); | |
16303ab2 | 4724 | goto out_free_new_mask; |
4c44aaaf EB |
4725 | } |
4726 | rcu_read_unlock(); | |
4727 | } | |
1da177e4 | 4728 | |
b0ae1981 | 4729 | retval = security_task_setscheduler(p); |
e7834f8f | 4730 | if (retval) |
16303ab2 | 4731 | goto out_free_new_mask; |
e7834f8f | 4732 | |
e4099a5e PZ |
4733 | |
4734 | cpuset_cpus_allowed(p, cpus_allowed); | |
4735 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4736 | ||
332ac17e DF |
4737 | /* |
4738 | * Since bandwidth control happens on root_domain basis, | |
4739 | * if admission test is enabled, we only admit -deadline | |
4740 | * tasks allowed to run on all the CPUs in the task's | |
4741 | * root_domain. | |
4742 | */ | |
4743 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4744 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4745 | rcu_read_lock(); | |
4746 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4747 | retval = -EBUSY; |
f1e3a093 | 4748 | rcu_read_unlock(); |
16303ab2 | 4749 | goto out_free_new_mask; |
332ac17e | 4750 | } |
f1e3a093 | 4751 | rcu_read_unlock(); |
332ac17e DF |
4752 | } |
4753 | #endif | |
49246274 | 4754 | again: |
25834c73 | 4755 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4756 | |
8707d8b8 | 4757 | if (!retval) { |
5a16f3d3 RR |
4758 | cpuset_cpus_allowed(p, cpus_allowed); |
4759 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4760 | /* |
4761 | * We must have raced with a concurrent cpuset | |
4762 | * update. Just reset the cpus_allowed to the | |
4763 | * cpuset's cpus_allowed | |
4764 | */ | |
5a16f3d3 | 4765 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4766 | goto again; |
4767 | } | |
4768 | } | |
16303ab2 | 4769 | out_free_new_mask: |
5a16f3d3 RR |
4770 | free_cpumask_var(new_mask); |
4771 | out_free_cpus_allowed: | |
4772 | free_cpumask_var(cpus_allowed); | |
4773 | out_put_task: | |
1da177e4 | 4774 | put_task_struct(p); |
1da177e4 LT |
4775 | return retval; |
4776 | } | |
4777 | ||
4778 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4779 | struct cpumask *new_mask) |
1da177e4 | 4780 | { |
96f874e2 RR |
4781 | if (len < cpumask_size()) |
4782 | cpumask_clear(new_mask); | |
4783 | else if (len > cpumask_size()) | |
4784 | len = cpumask_size(); | |
4785 | ||
1da177e4 LT |
4786 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4787 | } | |
4788 | ||
4789 | /** | |
4790 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4791 | * @pid: pid of the process | |
4792 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4793 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4794 | * |
4795 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4796 | */ |
5add95d4 HC |
4797 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4798 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4799 | { |
5a16f3d3 | 4800 | cpumask_var_t new_mask; |
1da177e4 LT |
4801 | int retval; |
4802 | ||
5a16f3d3 RR |
4803 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4804 | return -ENOMEM; | |
1da177e4 | 4805 | |
5a16f3d3 RR |
4806 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4807 | if (retval == 0) | |
4808 | retval = sched_setaffinity(pid, new_mask); | |
4809 | free_cpumask_var(new_mask); | |
4810 | return retval; | |
1da177e4 LT |
4811 | } |
4812 | ||
96f874e2 | 4813 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4814 | { |
36c8b586 | 4815 | struct task_struct *p; |
31605683 | 4816 | unsigned long flags; |
1da177e4 | 4817 | int retval; |
1da177e4 | 4818 | |
23f5d142 | 4819 | rcu_read_lock(); |
1da177e4 LT |
4820 | |
4821 | retval = -ESRCH; | |
4822 | p = find_process_by_pid(pid); | |
4823 | if (!p) | |
4824 | goto out_unlock; | |
4825 | ||
e7834f8f DQ |
4826 | retval = security_task_getscheduler(p); |
4827 | if (retval) | |
4828 | goto out_unlock; | |
4829 | ||
013fdb80 | 4830 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4831 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4832 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4833 | |
4834 | out_unlock: | |
23f5d142 | 4835 | rcu_read_unlock(); |
1da177e4 | 4836 | |
9531b62f | 4837 | return retval; |
1da177e4 LT |
4838 | } |
4839 | ||
4840 | /** | |
4841 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4842 | * @pid: pid of the process | |
4843 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4844 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 | 4845 | * |
599b4840 ZW |
4846 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4847 | * error code otherwise. | |
1da177e4 | 4848 | */ |
5add95d4 HC |
4849 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4850 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4851 | { |
4852 | int ret; | |
f17c8607 | 4853 | cpumask_var_t mask; |
1da177e4 | 4854 | |
84fba5ec | 4855 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4856 | return -EINVAL; |
4857 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4858 | return -EINVAL; |
4859 | ||
f17c8607 RR |
4860 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4861 | return -ENOMEM; | |
1da177e4 | 4862 | |
f17c8607 RR |
4863 | ret = sched_getaffinity(pid, mask); |
4864 | if (ret == 0) { | |
8bc037fb | 4865 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4866 | |
4867 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4868 | ret = -EFAULT; |
4869 | else | |
cd3d8031 | 4870 | ret = retlen; |
f17c8607 RR |
4871 | } |
4872 | free_cpumask_var(mask); | |
1da177e4 | 4873 | |
f17c8607 | 4874 | return ret; |
1da177e4 LT |
4875 | } |
4876 | ||
4877 | /** | |
4878 | * sys_sched_yield - yield the current processor to other threads. | |
4879 | * | |
dd41f596 IM |
4880 | * This function yields the current CPU to other tasks. If there are no |
4881 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4882 | * |
4883 | * Return: 0. | |
1da177e4 | 4884 | */ |
5add95d4 | 4885 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4886 | { |
70b97a7f | 4887 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4888 | |
ae92882e | 4889 | schedstat_inc(rq->yld_count); |
4530d7ab | 4890 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4891 | |
4892 | /* | |
4893 | * Since we are going to call schedule() anyway, there's | |
4894 | * no need to preempt or enable interrupts: | |
4895 | */ | |
4896 | __release(rq->lock); | |
8a25d5de | 4897 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4898 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4899 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4900 | |
4901 | schedule(); | |
4902 | ||
4903 | return 0; | |
4904 | } | |
4905 | ||
35a773a0 | 4906 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4907 | int __sched _cond_resched(void) |
1da177e4 | 4908 | { |
fe32d3cd | 4909 | if (should_resched(0)) { |
a18b5d01 | 4910 | preempt_schedule_common(); |
1da177e4 LT |
4911 | return 1; |
4912 | } | |
4913 | return 0; | |
4914 | } | |
02b67cc3 | 4915 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4916 | #endif |
1da177e4 LT |
4917 | |
4918 | /* | |
613afbf8 | 4919 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4920 | * call schedule, and on return reacquire the lock. |
4921 | * | |
41a2d6cf | 4922 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4923 | * operations here to prevent schedule() from being called twice (once via |
4924 | * spin_unlock(), once by hand). | |
4925 | */ | |
613afbf8 | 4926 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4927 | { |
fe32d3cd | 4928 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4929 | int ret = 0; |
4930 | ||
f607c668 PZ |
4931 | lockdep_assert_held(lock); |
4932 | ||
4a81e832 | 4933 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4934 | spin_unlock(lock); |
d86ee480 | 4935 | if (resched) |
a18b5d01 | 4936 | preempt_schedule_common(); |
95c354fe NP |
4937 | else |
4938 | cpu_relax(); | |
6df3cecb | 4939 | ret = 1; |
1da177e4 | 4940 | spin_lock(lock); |
1da177e4 | 4941 | } |
6df3cecb | 4942 | return ret; |
1da177e4 | 4943 | } |
613afbf8 | 4944 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4945 | |
613afbf8 | 4946 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4947 | { |
4948 | BUG_ON(!in_softirq()); | |
4949 | ||
fe32d3cd | 4950 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4951 | local_bh_enable(); |
a18b5d01 | 4952 | preempt_schedule_common(); |
1da177e4 LT |
4953 | local_bh_disable(); |
4954 | return 1; | |
4955 | } | |
4956 | return 0; | |
4957 | } | |
613afbf8 | 4958 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4959 | |
1da177e4 LT |
4960 | /** |
4961 | * yield - yield the current processor to other threads. | |
4962 | * | |
8e3fabfd PZ |
4963 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4964 | * | |
4965 | * The scheduler is at all times free to pick the calling task as the most | |
4966 | * eligible task to run, if removing the yield() call from your code breaks | |
4967 | * it, its already broken. | |
4968 | * | |
4969 | * Typical broken usage is: | |
4970 | * | |
4971 | * while (!event) | |
4972 | * yield(); | |
4973 | * | |
4974 | * where one assumes that yield() will let 'the other' process run that will | |
4975 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4976 | * happen. Never use yield() as a progress guarantee!! | |
4977 | * | |
4978 | * If you want to use yield() to wait for something, use wait_event(). | |
4979 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4980 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4981 | */ |
4982 | void __sched yield(void) | |
4983 | { | |
4984 | set_current_state(TASK_RUNNING); | |
4985 | sys_sched_yield(); | |
4986 | } | |
1da177e4 LT |
4987 | EXPORT_SYMBOL(yield); |
4988 | ||
d95f4122 MG |
4989 | /** |
4990 | * yield_to - yield the current processor to another thread in | |
4991 | * your thread group, or accelerate that thread toward the | |
4992 | * processor it's on. | |
16addf95 RD |
4993 | * @p: target task |
4994 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4995 | * |
4996 | * It's the caller's job to ensure that the target task struct | |
4997 | * can't go away on us before we can do any checks. | |
4998 | * | |
e69f6186 | 4999 | * Return: |
7b270f60 PZ |
5000 | * true (>0) if we indeed boosted the target task. |
5001 | * false (0) if we failed to boost the target. | |
5002 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5003 | */ |
fa93384f | 5004 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5005 | { |
5006 | struct task_struct *curr = current; | |
5007 | struct rq *rq, *p_rq; | |
5008 | unsigned long flags; | |
c3c18640 | 5009 | int yielded = 0; |
d95f4122 MG |
5010 | |
5011 | local_irq_save(flags); | |
5012 | rq = this_rq(); | |
5013 | ||
5014 | again: | |
5015 | p_rq = task_rq(p); | |
7b270f60 PZ |
5016 | /* |
5017 | * If we're the only runnable task on the rq and target rq also | |
5018 | * has only one task, there's absolutely no point in yielding. | |
5019 | */ | |
5020 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5021 | yielded = -ESRCH; | |
5022 | goto out_irq; | |
5023 | } | |
5024 | ||
d95f4122 | 5025 | double_rq_lock(rq, p_rq); |
39e24d8f | 5026 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5027 | double_rq_unlock(rq, p_rq); |
5028 | goto again; | |
5029 | } | |
5030 | ||
5031 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5032 | goto out_unlock; |
d95f4122 MG |
5033 | |
5034 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5035 | goto out_unlock; |
d95f4122 MG |
5036 | |
5037 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5038 | goto out_unlock; |
d95f4122 MG |
5039 | |
5040 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5041 | if (yielded) { |
ae92882e | 5042 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5043 | /* |
5044 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5045 | * fairness. | |
5046 | */ | |
5047 | if (preempt && rq != p_rq) | |
8875125e | 5048 | resched_curr(p_rq); |
6d1cafd8 | 5049 | } |
d95f4122 | 5050 | |
7b270f60 | 5051 | out_unlock: |
d95f4122 | 5052 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5053 | out_irq: |
d95f4122 MG |
5054 | local_irq_restore(flags); |
5055 | ||
7b270f60 | 5056 | if (yielded > 0) |
d95f4122 MG |
5057 | schedule(); |
5058 | ||
5059 | return yielded; | |
5060 | } | |
5061 | EXPORT_SYMBOL_GPL(yield_to); | |
5062 | ||
1da177e4 | 5063 | /* |
41a2d6cf | 5064 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5065 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5066 | */ |
1da177e4 LT |
5067 | long __sched io_schedule_timeout(long timeout) |
5068 | { | |
9cff8ade N |
5069 | int old_iowait = current->in_iowait; |
5070 | struct rq *rq; | |
1da177e4 LT |
5071 | long ret; |
5072 | ||
9cff8ade | 5073 | current->in_iowait = 1; |
10d784ea | 5074 | blk_schedule_flush_plug(current); |
9cff8ade | 5075 | |
0ff92245 | 5076 | delayacct_blkio_start(); |
9cff8ade | 5077 | rq = raw_rq(); |
1da177e4 LT |
5078 | atomic_inc(&rq->nr_iowait); |
5079 | ret = schedule_timeout(timeout); | |
9cff8ade | 5080 | current->in_iowait = old_iowait; |
1da177e4 | 5081 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5082 | delayacct_blkio_end(); |
9cff8ade | 5083 | |
1da177e4 LT |
5084 | return ret; |
5085 | } | |
9cff8ade | 5086 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
5087 | |
5088 | /** | |
5089 | * sys_sched_get_priority_max - return maximum RT priority. | |
5090 | * @policy: scheduling class. | |
5091 | * | |
e69f6186 YB |
5092 | * Return: On success, this syscall returns the maximum |
5093 | * rt_priority that can be used by a given scheduling class. | |
5094 | * On failure, a negative error code is returned. | |
1da177e4 | 5095 | */ |
5add95d4 | 5096 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5097 | { |
5098 | int ret = -EINVAL; | |
5099 | ||
5100 | switch (policy) { | |
5101 | case SCHED_FIFO: | |
5102 | case SCHED_RR: | |
5103 | ret = MAX_USER_RT_PRIO-1; | |
5104 | break; | |
aab03e05 | 5105 | case SCHED_DEADLINE: |
1da177e4 | 5106 | case SCHED_NORMAL: |
b0a9499c | 5107 | case SCHED_BATCH: |
dd41f596 | 5108 | case SCHED_IDLE: |
1da177e4 LT |
5109 | ret = 0; |
5110 | break; | |
5111 | } | |
5112 | return ret; | |
5113 | } | |
5114 | ||
5115 | /** | |
5116 | * sys_sched_get_priority_min - return minimum RT priority. | |
5117 | * @policy: scheduling class. | |
5118 | * | |
e69f6186 YB |
5119 | * Return: On success, this syscall returns the minimum |
5120 | * rt_priority that can be used by a given scheduling class. | |
5121 | * On failure, a negative error code is returned. | |
1da177e4 | 5122 | */ |
5add95d4 | 5123 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5124 | { |
5125 | int ret = -EINVAL; | |
5126 | ||
5127 | switch (policy) { | |
5128 | case SCHED_FIFO: | |
5129 | case SCHED_RR: | |
5130 | ret = 1; | |
5131 | break; | |
aab03e05 | 5132 | case SCHED_DEADLINE: |
1da177e4 | 5133 | case SCHED_NORMAL: |
b0a9499c | 5134 | case SCHED_BATCH: |
dd41f596 | 5135 | case SCHED_IDLE: |
1da177e4 LT |
5136 | ret = 0; |
5137 | } | |
5138 | return ret; | |
5139 | } | |
5140 | ||
5141 | /** | |
5142 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5143 | * @pid: pid of the process. | |
5144 | * @interval: userspace pointer to the timeslice value. | |
5145 | * | |
5146 | * this syscall writes the default timeslice value of a given process | |
5147 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5148 | * |
5149 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5150 | * an error code. | |
1da177e4 | 5151 | */ |
17da2bd9 | 5152 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5153 | struct timespec __user *, interval) |
1da177e4 | 5154 | { |
36c8b586 | 5155 | struct task_struct *p; |
a4ec24b4 | 5156 | unsigned int time_slice; |
eb580751 PZ |
5157 | struct rq_flags rf; |
5158 | struct timespec t; | |
dba091b9 | 5159 | struct rq *rq; |
3a5c359a | 5160 | int retval; |
1da177e4 LT |
5161 | |
5162 | if (pid < 0) | |
3a5c359a | 5163 | return -EINVAL; |
1da177e4 LT |
5164 | |
5165 | retval = -ESRCH; | |
1a551ae7 | 5166 | rcu_read_lock(); |
1da177e4 LT |
5167 | p = find_process_by_pid(pid); |
5168 | if (!p) | |
5169 | goto out_unlock; | |
5170 | ||
5171 | retval = security_task_getscheduler(p); | |
5172 | if (retval) | |
5173 | goto out_unlock; | |
5174 | ||
eb580751 | 5175 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5176 | time_slice = 0; |
5177 | if (p->sched_class->get_rr_interval) | |
5178 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5179 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5180 | |
1a551ae7 | 5181 | rcu_read_unlock(); |
a4ec24b4 | 5182 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5183 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5184 | return retval; |
3a5c359a | 5185 | |
1da177e4 | 5186 | out_unlock: |
1a551ae7 | 5187 | rcu_read_unlock(); |
1da177e4 LT |
5188 | return retval; |
5189 | } | |
5190 | ||
7c731e0a | 5191 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5192 | |
82a1fcb9 | 5193 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5194 | { |
1da177e4 | 5195 | unsigned long free = 0; |
4e79752c | 5196 | int ppid; |
1f8a7633 | 5197 | unsigned long state = p->state; |
1da177e4 | 5198 | |
38200502 TH |
5199 | if (!try_get_task_stack(p)) |
5200 | return; | |
1f8a7633 TH |
5201 | if (state) |
5202 | state = __ffs(state) + 1; | |
28d0686c | 5203 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5204 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
1da177e4 | 5205 | if (state == TASK_RUNNING) |
3df0fc5b | 5206 | printk(KERN_CONT " running task "); |
1da177e4 | 5207 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5208 | free = stack_not_used(p); |
1da177e4 | 5209 | #endif |
a90e984c | 5210 | ppid = 0; |
4e79752c | 5211 | rcu_read_lock(); |
a90e984c ON |
5212 | if (pid_alive(p)) |
5213 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5214 | rcu_read_unlock(); |
3df0fc5b | 5215 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5216 | task_pid_nr(p), ppid, |
aa47b7e0 | 5217 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5218 | |
3d1cb205 | 5219 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5220 | show_stack(p, NULL); |
38200502 | 5221 | put_task_stack(p); |
1da177e4 LT |
5222 | } |
5223 | ||
e59e2ae2 | 5224 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5225 | { |
36c8b586 | 5226 | struct task_struct *g, *p; |
1da177e4 | 5227 | |
4bd77321 | 5228 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5229 | printk(KERN_INFO |
5230 | " task PC stack pid father\n"); | |
1da177e4 | 5231 | #else |
3df0fc5b PZ |
5232 | printk(KERN_INFO |
5233 | " task PC stack pid father\n"); | |
1da177e4 | 5234 | #endif |
510f5acc | 5235 | rcu_read_lock(); |
5d07f420 | 5236 | for_each_process_thread(g, p) { |
1da177e4 LT |
5237 | /* |
5238 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5239 | * console might take a lot of time: |
57675cb9 AR |
5240 | * Also, reset softlockup watchdogs on all CPUs, because |
5241 | * another CPU might be blocked waiting for us to process | |
5242 | * an IPI. | |
1da177e4 LT |
5243 | */ |
5244 | touch_nmi_watchdog(); | |
57675cb9 | 5245 | touch_all_softlockup_watchdogs(); |
39bc89fd | 5246 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5247 | sched_show_task(p); |
5d07f420 | 5248 | } |
1da177e4 | 5249 | |
dd41f596 | 5250 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5251 | if (!state_filter) |
5252 | sysrq_sched_debug_show(); | |
dd41f596 | 5253 | #endif |
510f5acc | 5254 | rcu_read_unlock(); |
e59e2ae2 IM |
5255 | /* |
5256 | * Only show locks if all tasks are dumped: | |
5257 | */ | |
93335a21 | 5258 | if (!state_filter) |
e59e2ae2 | 5259 | debug_show_all_locks(); |
1da177e4 LT |
5260 | } |
5261 | ||
0db0628d | 5262 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5263 | { |
dd41f596 | 5264 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5265 | } |
5266 | ||
f340c0d1 IM |
5267 | /** |
5268 | * init_idle - set up an idle thread for a given CPU | |
5269 | * @idle: task in question | |
5270 | * @cpu: cpu the idle task belongs to | |
5271 | * | |
5272 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5273 | * flag, to make booting more robust. | |
5274 | */ | |
0db0628d | 5275 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5276 | { |
70b97a7f | 5277 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5278 | unsigned long flags; |
5279 | ||
25834c73 PZ |
5280 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5281 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5282 | |
5e1576ed | 5283 | __sched_fork(0, idle); |
06b83b5f | 5284 | idle->state = TASK_RUNNING; |
dd41f596 | 5285 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5286 | idle->flags |= PF_IDLE; |
dd41f596 | 5287 | |
e1b77c92 MR |
5288 | kasan_unpoison_task_stack(idle); |
5289 | ||
de9b8f5d PZ |
5290 | #ifdef CONFIG_SMP |
5291 | /* | |
5292 | * Its possible that init_idle() gets called multiple times on a task, | |
5293 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5294 | * | |
5295 | * And since this is boot we can forgo the serialization. | |
5296 | */ | |
5297 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5298 | #endif | |
6506cf6c PZ |
5299 | /* |
5300 | * We're having a chicken and egg problem, even though we are | |
5301 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5302 | * lockdep check in task_group() will fail. | |
5303 | * | |
5304 | * Similar case to sched_fork(). / Alternatively we could | |
5305 | * use task_rq_lock() here and obtain the other rq->lock. | |
5306 | * | |
5307 | * Silence PROVE_RCU | |
5308 | */ | |
5309 | rcu_read_lock(); | |
dd41f596 | 5310 | __set_task_cpu(idle, cpu); |
6506cf6c | 5311 | rcu_read_unlock(); |
1da177e4 | 5312 | |
1da177e4 | 5313 | rq->curr = rq->idle = idle; |
da0c1e65 | 5314 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5315 | #ifdef CONFIG_SMP |
3ca7a440 | 5316 | idle->on_cpu = 1; |
4866cde0 | 5317 | #endif |
25834c73 PZ |
5318 | raw_spin_unlock(&rq->lock); |
5319 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5320 | |
5321 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5322 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5323 | |
dd41f596 IM |
5324 | /* |
5325 | * The idle tasks have their own, simple scheduling class: | |
5326 | */ | |
5327 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5328 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5329 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5330 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5331 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5332 | #endif | |
19978ca6 IM |
5333 | } |
5334 | ||
f82f8042 JL |
5335 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5336 | const struct cpumask *trial) | |
5337 | { | |
5338 | int ret = 1, trial_cpus; | |
5339 | struct dl_bw *cur_dl_b; | |
5340 | unsigned long flags; | |
5341 | ||
bb2bc55a MG |
5342 | if (!cpumask_weight(cur)) |
5343 | return ret; | |
5344 | ||
75e23e49 | 5345 | rcu_read_lock_sched(); |
f82f8042 JL |
5346 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5347 | trial_cpus = cpumask_weight(trial); | |
5348 | ||
5349 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5350 | if (cur_dl_b->bw != -1 && | |
5351 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5352 | ret = 0; | |
5353 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5354 | rcu_read_unlock_sched(); |
f82f8042 JL |
5355 | |
5356 | return ret; | |
5357 | } | |
5358 | ||
7f51412a JL |
5359 | int task_can_attach(struct task_struct *p, |
5360 | const struct cpumask *cs_cpus_allowed) | |
5361 | { | |
5362 | int ret = 0; | |
5363 | ||
5364 | /* | |
5365 | * Kthreads which disallow setaffinity shouldn't be moved | |
5366 | * to a new cpuset; we don't want to change their cpu | |
5367 | * affinity and isolating such threads by their set of | |
5368 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5369 | * applicable for such threads. This prevents checking for | |
5370 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5371 | * before cpus_allowed may be changed. | |
5372 | */ | |
5373 | if (p->flags & PF_NO_SETAFFINITY) { | |
5374 | ret = -EINVAL; | |
5375 | goto out; | |
5376 | } | |
5377 | ||
5378 | #ifdef CONFIG_SMP | |
5379 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5380 | cs_cpus_allowed)) { | |
5381 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5382 | cs_cpus_allowed); | |
75e23e49 | 5383 | struct dl_bw *dl_b; |
7f51412a JL |
5384 | bool overflow; |
5385 | int cpus; | |
5386 | unsigned long flags; | |
5387 | ||
75e23e49 JL |
5388 | rcu_read_lock_sched(); |
5389 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5390 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5391 | cpus = dl_bw_cpus(dest_cpu); | |
5392 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5393 | if (overflow) | |
5394 | ret = -EBUSY; | |
5395 | else { | |
5396 | /* | |
5397 | * We reserve space for this task in the destination | |
5398 | * root_domain, as we can't fail after this point. | |
5399 | * We will free resources in the source root_domain | |
5400 | * later on (see set_cpus_allowed_dl()). | |
5401 | */ | |
5402 | __dl_add(dl_b, p->dl.dl_bw); | |
5403 | } | |
5404 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5405 | rcu_read_unlock_sched(); |
7f51412a JL |
5406 | |
5407 | } | |
5408 | #endif | |
5409 | out: | |
5410 | return ret; | |
5411 | } | |
5412 | ||
1da177e4 | 5413 | #ifdef CONFIG_SMP |
1da177e4 | 5414 | |
e26fbffd TG |
5415 | static bool sched_smp_initialized __read_mostly; |
5416 | ||
e6628d5b MG |
5417 | #ifdef CONFIG_NUMA_BALANCING |
5418 | /* Migrate current task p to target_cpu */ | |
5419 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5420 | { | |
5421 | struct migration_arg arg = { p, target_cpu }; | |
5422 | int curr_cpu = task_cpu(p); | |
5423 | ||
5424 | if (curr_cpu == target_cpu) | |
5425 | return 0; | |
5426 | ||
5427 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5428 | return -EINVAL; | |
5429 | ||
5430 | /* TODO: This is not properly updating schedstats */ | |
5431 | ||
286549dc | 5432 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5433 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5434 | } | |
0ec8aa00 PZ |
5435 | |
5436 | /* | |
5437 | * Requeue a task on a given node and accurately track the number of NUMA | |
5438 | * tasks on the runqueues | |
5439 | */ | |
5440 | void sched_setnuma(struct task_struct *p, int nid) | |
5441 | { | |
da0c1e65 | 5442 | bool queued, running; |
eb580751 PZ |
5443 | struct rq_flags rf; |
5444 | struct rq *rq; | |
0ec8aa00 | 5445 | |
eb580751 | 5446 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5447 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5448 | running = task_current(rq, p); |
5449 | ||
da0c1e65 | 5450 | if (queued) |
1de64443 | 5451 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5452 | if (running) |
f3cd1c4e | 5453 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5454 | |
5455 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5456 | |
da0c1e65 | 5457 | if (queued) |
1de64443 | 5458 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 5459 | if (running) |
b2bf6c31 | 5460 | set_curr_task(rq, p); |
eb580751 | 5461 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5462 | } |
5cc389bc | 5463 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5464 | |
1da177e4 | 5465 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5466 | /* |
48c5ccae PZ |
5467 | * Ensures that the idle task is using init_mm right before its cpu goes |
5468 | * offline. | |
054b9108 | 5469 | */ |
48c5ccae | 5470 | void idle_task_exit(void) |
1da177e4 | 5471 | { |
48c5ccae | 5472 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5473 | |
48c5ccae | 5474 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5475 | |
a53efe5f | 5476 | if (mm != &init_mm) { |
f98db601 | 5477 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5478 | finish_arch_post_lock_switch(); |
5479 | } | |
48c5ccae | 5480 | mmdrop(mm); |
1da177e4 LT |
5481 | } |
5482 | ||
5483 | /* | |
5d180232 PZ |
5484 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5485 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5486 | * nr_active count is stable. We need to take the teardown thread which |
5487 | * is calling this into account, so we hand in adjust = 1 to the load | |
5488 | * calculation. | |
5d180232 PZ |
5489 | * |
5490 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5491 | */ |
5d180232 | 5492 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5493 | { |
d60585c5 | 5494 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5495 | if (delta) |
5496 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5497 | } |
5498 | ||
3f1d2a31 PZ |
5499 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5500 | { | |
5501 | } | |
5502 | ||
5503 | static const struct sched_class fake_sched_class = { | |
5504 | .put_prev_task = put_prev_task_fake, | |
5505 | }; | |
5506 | ||
5507 | static struct task_struct fake_task = { | |
5508 | /* | |
5509 | * Avoid pull_{rt,dl}_task() | |
5510 | */ | |
5511 | .prio = MAX_PRIO + 1, | |
5512 | .sched_class = &fake_sched_class, | |
5513 | }; | |
5514 | ||
48f24c4d | 5515 | /* |
48c5ccae PZ |
5516 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5517 | * try_to_wake_up()->select_task_rq(). | |
5518 | * | |
5519 | * Called with rq->lock held even though we'er in stop_machine() and | |
5520 | * there's no concurrency possible, we hold the required locks anyway | |
5521 | * because of lock validation efforts. | |
1da177e4 | 5522 | */ |
5e16bbc2 | 5523 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5524 | { |
5e16bbc2 | 5525 | struct rq *rq = dead_rq; |
48c5ccae | 5526 | struct task_struct *next, *stop = rq->stop; |
d8ac8971 | 5527 | struct rq_flags rf; |
48c5ccae | 5528 | int dest_cpu; |
1da177e4 LT |
5529 | |
5530 | /* | |
48c5ccae PZ |
5531 | * Fudge the rq selection such that the below task selection loop |
5532 | * doesn't get stuck on the currently eligible stop task. | |
5533 | * | |
5534 | * We're currently inside stop_machine() and the rq is either stuck | |
5535 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5536 | * either way we should never end up calling schedule() until we're | |
5537 | * done here. | |
1da177e4 | 5538 | */ |
48c5ccae | 5539 | rq->stop = NULL; |
48f24c4d | 5540 | |
77bd3970 FW |
5541 | /* |
5542 | * put_prev_task() and pick_next_task() sched | |
5543 | * class method both need to have an up-to-date | |
5544 | * value of rq->clock[_task] | |
5545 | */ | |
5546 | update_rq_clock(rq); | |
5547 | ||
5e16bbc2 | 5548 | for (;;) { |
48c5ccae PZ |
5549 | /* |
5550 | * There's this thread running, bail when that's the only | |
5551 | * remaining thread. | |
5552 | */ | |
5553 | if (rq->nr_running == 1) | |
dd41f596 | 5554 | break; |
48c5ccae | 5555 | |
cbce1a68 | 5556 | /* |
5473e0cc | 5557 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 | 5558 | */ |
d8ac8971 MF |
5559 | rq_pin_lock(rq, &rf); |
5560 | next = pick_next_task(rq, &fake_task, &rf); | |
48c5ccae | 5561 | BUG_ON(!next); |
79c53799 | 5562 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5563 | |
5473e0cc WL |
5564 | /* |
5565 | * Rules for changing task_struct::cpus_allowed are holding | |
5566 | * both pi_lock and rq->lock, such that holding either | |
5567 | * stabilizes the mask. | |
5568 | * | |
5569 | * Drop rq->lock is not quite as disastrous as it usually is | |
5570 | * because !cpu_active at this point, which means load-balance | |
5571 | * will not interfere. Also, stop-machine. | |
5572 | */ | |
d8ac8971 | 5573 | rq_unpin_lock(rq, &rf); |
5473e0cc WL |
5574 | raw_spin_unlock(&rq->lock); |
5575 | raw_spin_lock(&next->pi_lock); | |
5576 | raw_spin_lock(&rq->lock); | |
5577 | ||
5578 | /* | |
5579 | * Since we're inside stop-machine, _nothing_ should have | |
5580 | * changed the task, WARN if weird stuff happened, because in | |
5581 | * that case the above rq->lock drop is a fail too. | |
5582 | */ | |
5583 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5584 | raw_spin_unlock(&next->pi_lock); | |
5585 | continue; | |
5586 | } | |
5587 | ||
48c5ccae | 5588 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5589 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5590 | |
5e16bbc2 PZ |
5591 | rq = __migrate_task(rq, next, dest_cpu); |
5592 | if (rq != dead_rq) { | |
5593 | raw_spin_unlock(&rq->lock); | |
5594 | rq = dead_rq; | |
5595 | raw_spin_lock(&rq->lock); | |
5596 | } | |
5473e0cc | 5597 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5598 | } |
dce48a84 | 5599 | |
48c5ccae | 5600 | rq->stop = stop; |
dce48a84 | 5601 | } |
1da177e4 LT |
5602 | #endif /* CONFIG_HOTPLUG_CPU */ |
5603 | ||
1f11eb6a GH |
5604 | static void set_rq_online(struct rq *rq) |
5605 | { | |
5606 | if (!rq->online) { | |
5607 | const struct sched_class *class; | |
5608 | ||
c6c4927b | 5609 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5610 | rq->online = 1; |
5611 | ||
5612 | for_each_class(class) { | |
5613 | if (class->rq_online) | |
5614 | class->rq_online(rq); | |
5615 | } | |
5616 | } | |
5617 | } | |
5618 | ||
5619 | static void set_rq_offline(struct rq *rq) | |
5620 | { | |
5621 | if (rq->online) { | |
5622 | const struct sched_class *class; | |
5623 | ||
5624 | for_each_class(class) { | |
5625 | if (class->rq_offline) | |
5626 | class->rq_offline(rq); | |
5627 | } | |
5628 | ||
c6c4927b | 5629 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5630 | rq->online = 0; |
5631 | } | |
5632 | } | |
5633 | ||
9cf7243d | 5634 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5635 | { |
969c7921 | 5636 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5637 | |
a803f026 CM |
5638 | rq->age_stamp = sched_clock_cpu(cpu); |
5639 | } | |
5640 | ||
4cb98839 PZ |
5641 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5642 | ||
3e9830dc | 5643 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5644 | |
d039ac60 | 5645 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5646 | |
d039ac60 | 5647 | static int __init sched_debug_setup(char *str) |
f6630114 | 5648 | { |
d039ac60 | 5649 | sched_debug_enabled = 1; |
f6630114 MT |
5650 | |
5651 | return 0; | |
5652 | } | |
d039ac60 PZ |
5653 | early_param("sched_debug", sched_debug_setup); |
5654 | ||
5655 | static inline bool sched_debug(void) | |
5656 | { | |
5657 | return sched_debug_enabled; | |
5658 | } | |
f6630114 | 5659 | |
7c16ec58 | 5660 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5661 | struct cpumask *groupmask) |
1da177e4 | 5662 | { |
4dcf6aff | 5663 | struct sched_group *group = sd->groups; |
1da177e4 | 5664 | |
96f874e2 | 5665 | cpumask_clear(groupmask); |
4dcf6aff IM |
5666 | |
5667 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5668 | ||
5669 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5670 | printk("does not load-balance\n"); |
4dcf6aff | 5671 | if (sd->parent) |
3df0fc5b PZ |
5672 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5673 | " has parent"); | |
4dcf6aff | 5674 | return -1; |
41c7ce9a NP |
5675 | } |
5676 | ||
333470ee TH |
5677 | printk(KERN_CONT "span %*pbl level %s\n", |
5678 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5679 | |
758b2cdc | 5680 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5681 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5682 | "CPU%d\n", cpu); | |
4dcf6aff | 5683 | } |
758b2cdc | 5684 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5685 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5686 | " CPU%d\n", cpu); | |
4dcf6aff | 5687 | } |
1da177e4 | 5688 | |
4dcf6aff | 5689 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5690 | do { |
4dcf6aff | 5691 | if (!group) { |
3df0fc5b PZ |
5692 | printk("\n"); |
5693 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5694 | break; |
5695 | } | |
5696 | ||
758b2cdc | 5697 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5698 | printk(KERN_CONT "\n"); |
5699 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5700 | break; |
5701 | } | |
1da177e4 | 5702 | |
cb83b629 PZ |
5703 | if (!(sd->flags & SD_OVERLAP) && |
5704 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5705 | printk(KERN_CONT "\n"); |
5706 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5707 | break; |
5708 | } | |
1da177e4 | 5709 | |
758b2cdc | 5710 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5711 | |
333470ee TH |
5712 | printk(KERN_CONT " %*pbl", |
5713 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5714 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
bf475ce0 | 5715 | printk(KERN_CONT " (cpu_capacity = %lu)", |
63b2ca30 | 5716 | group->sgc->capacity); |
381512cf | 5717 | } |
1da177e4 | 5718 | |
4dcf6aff IM |
5719 | group = group->next; |
5720 | } while (group != sd->groups); | |
3df0fc5b | 5721 | printk(KERN_CONT "\n"); |
1da177e4 | 5722 | |
758b2cdc | 5723 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5724 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5725 | |
758b2cdc RR |
5726 | if (sd->parent && |
5727 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5728 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5729 | "of domain->span\n"); | |
4dcf6aff IM |
5730 | return 0; |
5731 | } | |
1da177e4 | 5732 | |
4dcf6aff IM |
5733 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5734 | { | |
5735 | int level = 0; | |
1da177e4 | 5736 | |
d039ac60 | 5737 | if (!sched_debug_enabled) |
f6630114 MT |
5738 | return; |
5739 | ||
4dcf6aff IM |
5740 | if (!sd) { |
5741 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5742 | return; | |
5743 | } | |
1da177e4 | 5744 | |
4dcf6aff IM |
5745 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5746 | ||
5747 | for (;;) { | |
4cb98839 | 5748 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5749 | break; |
1da177e4 LT |
5750 | level++; |
5751 | sd = sd->parent; | |
33859f7f | 5752 | if (!sd) |
4dcf6aff IM |
5753 | break; |
5754 | } | |
1da177e4 | 5755 | } |
6d6bc0ad | 5756 | #else /* !CONFIG_SCHED_DEBUG */ |
a18a579e PZ |
5757 | |
5758 | # define sched_debug_enabled 0 | |
48f24c4d | 5759 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5760 | static inline bool sched_debug(void) |
5761 | { | |
5762 | return false; | |
5763 | } | |
6d6bc0ad | 5764 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5765 | |
1a20ff27 | 5766 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5767 | { |
758b2cdc | 5768 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5769 | return 1; |
5770 | ||
5771 | /* Following flags need at least 2 groups */ | |
5772 | if (sd->flags & (SD_LOAD_BALANCE | | |
5773 | SD_BALANCE_NEWIDLE | | |
5774 | SD_BALANCE_FORK | | |
89c4710e | 5775 | SD_BALANCE_EXEC | |
5d4dfddd | 5776 | SD_SHARE_CPUCAPACITY | |
1f6e6c7c | 5777 | SD_ASYM_CPUCAPACITY | |
d77b3ed5 VG |
5778 | SD_SHARE_PKG_RESOURCES | |
5779 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5780 | if (sd->groups != sd->groups->next) |
5781 | return 0; | |
5782 | } | |
5783 | ||
5784 | /* Following flags don't use groups */ | |
c88d5910 | 5785 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5786 | return 0; |
5787 | ||
5788 | return 1; | |
5789 | } | |
5790 | ||
48f24c4d IM |
5791 | static int |
5792 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5793 | { |
5794 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5795 | ||
5796 | if (sd_degenerate(parent)) | |
5797 | return 1; | |
5798 | ||
758b2cdc | 5799 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5800 | return 0; |
5801 | ||
245af2c7 SS |
5802 | /* Flags needing groups don't count if only 1 group in parent */ |
5803 | if (parent->groups == parent->groups->next) { | |
5804 | pflags &= ~(SD_LOAD_BALANCE | | |
5805 | SD_BALANCE_NEWIDLE | | |
5806 | SD_BALANCE_FORK | | |
89c4710e | 5807 | SD_BALANCE_EXEC | |
1f6e6c7c | 5808 | SD_ASYM_CPUCAPACITY | |
5d4dfddd | 5809 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5810 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5811 | SD_PREFER_SIBLING | |
5812 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5813 | if (nr_node_ids == 1) |
5814 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5815 | } |
5816 | if (~cflags & pflags) | |
5817 | return 0; | |
5818 | ||
5819 | return 1; | |
5820 | } | |
5821 | ||
dce840a0 | 5822 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5823 | { |
dce840a0 | 5824 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5825 | |
68e74568 | 5826 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5827 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5828 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5829 | free_cpumask_var(rd->rto_mask); |
5830 | free_cpumask_var(rd->online); | |
5831 | free_cpumask_var(rd->span); | |
5832 | kfree(rd); | |
5833 | } | |
5834 | ||
57d885fe GH |
5835 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5836 | { | |
a0490fa3 | 5837 | struct root_domain *old_rd = NULL; |
57d885fe | 5838 | unsigned long flags; |
57d885fe | 5839 | |
05fa785c | 5840 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5841 | |
5842 | if (rq->rd) { | |
a0490fa3 | 5843 | old_rd = rq->rd; |
57d885fe | 5844 | |
c6c4927b | 5845 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5846 | set_rq_offline(rq); |
57d885fe | 5847 | |
c6c4927b | 5848 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5849 | |
a0490fa3 | 5850 | /* |
0515973f | 5851 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5852 | * set old_rd to NULL to skip the freeing later |
5853 | * in this function: | |
5854 | */ | |
5855 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5856 | old_rd = NULL; | |
57d885fe GH |
5857 | } |
5858 | ||
5859 | atomic_inc(&rd->refcount); | |
5860 | rq->rd = rd; | |
5861 | ||
c6c4927b | 5862 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5863 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5864 | set_rq_online(rq); |
57d885fe | 5865 | |
05fa785c | 5866 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5867 | |
5868 | if (old_rd) | |
dce840a0 | 5869 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5870 | } |
5871 | ||
68c38fc3 | 5872 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5873 | { |
5874 | memset(rd, 0, sizeof(*rd)); | |
5875 | ||
8295c699 | 5876 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5877 | goto out; |
8295c699 | 5878 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5879 | goto free_span; |
8295c699 | 5880 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5881 | goto free_online; |
8295c699 | 5882 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5883 | goto free_dlo_mask; |
6e0534f2 | 5884 | |
332ac17e | 5885 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5886 | if (cpudl_init(&rd->cpudl) != 0) |
5887 | goto free_dlo_mask; | |
332ac17e | 5888 | |
68c38fc3 | 5889 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5890 | goto free_rto_mask; |
c6c4927b | 5891 | return 0; |
6e0534f2 | 5892 | |
68e74568 RR |
5893 | free_rto_mask: |
5894 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5895 | free_dlo_mask: |
5896 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5897 | free_online: |
5898 | free_cpumask_var(rd->online); | |
5899 | free_span: | |
5900 | free_cpumask_var(rd->span); | |
0c910d28 | 5901 | out: |
c6c4927b | 5902 | return -ENOMEM; |
57d885fe GH |
5903 | } |
5904 | ||
029632fb PZ |
5905 | /* |
5906 | * By default the system creates a single root-domain with all cpus as | |
5907 | * members (mimicking the global state we have today). | |
5908 | */ | |
5909 | struct root_domain def_root_domain; | |
5910 | ||
57d885fe GH |
5911 | static void init_defrootdomain(void) |
5912 | { | |
68c38fc3 | 5913 | init_rootdomain(&def_root_domain); |
c6c4927b | 5914 | |
57d885fe GH |
5915 | atomic_set(&def_root_domain.refcount, 1); |
5916 | } | |
5917 | ||
dc938520 | 5918 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5919 | { |
5920 | struct root_domain *rd; | |
5921 | ||
5922 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5923 | if (!rd) | |
5924 | return NULL; | |
5925 | ||
68c38fc3 | 5926 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5927 | kfree(rd); |
5928 | return NULL; | |
5929 | } | |
57d885fe GH |
5930 | |
5931 | return rd; | |
5932 | } | |
5933 | ||
63b2ca30 | 5934 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5935 | { |
5936 | struct sched_group *tmp, *first; | |
5937 | ||
5938 | if (!sg) | |
5939 | return; | |
5940 | ||
5941 | first = sg; | |
5942 | do { | |
5943 | tmp = sg->next; | |
5944 | ||
63b2ca30 NP |
5945 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5946 | kfree(sg->sgc); | |
e3589f6c PZ |
5947 | |
5948 | kfree(sg); | |
5949 | sg = tmp; | |
5950 | } while (sg != first); | |
5951 | } | |
5952 | ||
16f3ef46 | 5953 | static void destroy_sched_domain(struct sched_domain *sd) |
dce840a0 | 5954 | { |
e3589f6c PZ |
5955 | /* |
5956 | * If its an overlapping domain it has private groups, iterate and | |
5957 | * nuke them all. | |
5958 | */ | |
5959 | if (sd->flags & SD_OVERLAP) { | |
5960 | free_sched_groups(sd->groups, 1); | |
5961 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5962 | kfree(sd->groups->sgc); |
dce840a0 | 5963 | kfree(sd->groups); |
9c3f75cb | 5964 | } |
24fc7edb PZ |
5965 | if (sd->shared && atomic_dec_and_test(&sd->shared->ref)) |
5966 | kfree(sd->shared); | |
dce840a0 PZ |
5967 | kfree(sd); |
5968 | } | |
5969 | ||
16f3ef46 | 5970 | static void destroy_sched_domains_rcu(struct rcu_head *rcu) |
dce840a0 | 5971 | { |
16f3ef46 PZ |
5972 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); |
5973 | ||
5974 | while (sd) { | |
5975 | struct sched_domain *parent = sd->parent; | |
5976 | destroy_sched_domain(sd); | |
5977 | sd = parent; | |
5978 | } | |
dce840a0 PZ |
5979 | } |
5980 | ||
f39180ef | 5981 | static void destroy_sched_domains(struct sched_domain *sd) |
dce840a0 | 5982 | { |
16f3ef46 PZ |
5983 | if (sd) |
5984 | call_rcu(&sd->rcu, destroy_sched_domains_rcu); | |
dce840a0 PZ |
5985 | } |
5986 | ||
518cd623 PZ |
5987 | /* |
5988 | * Keep a special pointer to the highest sched_domain that has | |
5989 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5990 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5991 | * | |
5992 | * Also keep a unique ID per domain (we use the first cpu number in | |
5993 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5994 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5995 | */ |
5996 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5997 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5998 | DEFINE_PER_CPU(int, sd_llc_id); |
0e369d75 | 5999 | DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 6000 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 6001 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 PZ |
6002 | |
6003 | static void update_top_cache_domain(int cpu) | |
6004 | { | |
0e369d75 | 6005 | struct sched_domain_shared *sds = NULL; |
518cd623 PZ |
6006 | struct sched_domain *sd; |
6007 | int id = cpu; | |
7d9ffa89 | 6008 | int size = 1; |
518cd623 PZ |
6009 | |
6010 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 6011 | if (sd) { |
518cd623 | 6012 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 6013 | size = cpumask_weight(sched_domain_span(sd)); |
0e369d75 | 6014 | sds = sd->shared; |
7d9ffa89 | 6015 | } |
518cd623 PZ |
6016 | |
6017 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 6018 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 6019 | per_cpu(sd_llc_id, cpu) = id; |
0e369d75 | 6020 | rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds); |
fb13c7ee MG |
6021 | |
6022 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
6023 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
6024 | |
6025 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
6026 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
6027 | } |
6028 | ||
1da177e4 | 6029 | /* |
0eab9146 | 6030 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6031 | * hold the hotplug lock. |
6032 | */ | |
0eab9146 IM |
6033 | static void |
6034 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6035 | { |
70b97a7f | 6036 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6037 | struct sched_domain *tmp; |
6038 | ||
6039 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6040 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6041 | struct sched_domain *parent = tmp->parent; |
6042 | if (!parent) | |
6043 | break; | |
f29c9b1c | 6044 | |
1a848870 | 6045 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6046 | tmp->parent = parent->parent; |
1a848870 SS |
6047 | if (parent->parent) |
6048 | parent->parent->child = tmp; | |
10866e62 PZ |
6049 | /* |
6050 | * Transfer SD_PREFER_SIBLING down in case of a | |
6051 | * degenerate parent; the spans match for this | |
6052 | * so the property transfers. | |
6053 | */ | |
6054 | if (parent->flags & SD_PREFER_SIBLING) | |
6055 | tmp->flags |= SD_PREFER_SIBLING; | |
f39180ef | 6056 | destroy_sched_domain(parent); |
f29c9b1c LZ |
6057 | } else |
6058 | tmp = tmp->parent; | |
245af2c7 SS |
6059 | } |
6060 | ||
1a848870 | 6061 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6062 | tmp = sd; |
245af2c7 | 6063 | sd = sd->parent; |
f39180ef | 6064 | destroy_sched_domain(tmp); |
1a848870 SS |
6065 | if (sd) |
6066 | sd->child = NULL; | |
6067 | } | |
1da177e4 | 6068 | |
4cb98839 | 6069 | sched_domain_debug(sd, cpu); |
1da177e4 | 6070 | |
57d885fe | 6071 | rq_attach_root(rq, rd); |
dce840a0 | 6072 | tmp = rq->sd; |
674311d5 | 6073 | rcu_assign_pointer(rq->sd, sd); |
f39180ef | 6074 | destroy_sched_domains(tmp); |
518cd623 PZ |
6075 | |
6076 | update_top_cache_domain(cpu); | |
1da177e4 LT |
6077 | } |
6078 | ||
1da177e4 LT |
6079 | /* Setup the mask of cpus configured for isolated domains */ |
6080 | static int __init isolated_cpu_setup(char *str) | |
6081 | { | |
a6e4491c PB |
6082 | int ret; |
6083 | ||
bdddd296 | 6084 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
6085 | ret = cpulist_parse(str, cpu_isolated_map); |
6086 | if (ret) { | |
6087 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
6088 | return 0; | |
6089 | } | |
1da177e4 LT |
6090 | return 1; |
6091 | } | |
8927f494 | 6092 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6093 | |
49a02c51 | 6094 | struct s_data { |
21d42ccf | 6095 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6096 | struct root_domain *rd; |
6097 | }; | |
6098 | ||
2109b99e | 6099 | enum s_alloc { |
2109b99e | 6100 | sa_rootdomain, |
21d42ccf | 6101 | sa_sd, |
dce840a0 | 6102 | sa_sd_storage, |
2109b99e AH |
6103 | sa_none, |
6104 | }; | |
6105 | ||
c1174876 PZ |
6106 | /* |
6107 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
6108 | * domain traversal. | |
6109 | * | |
6110 | * Asymmetric node setups can result in situations where the domain tree is of | |
6111 | * unequal depth, make sure to skip domains that already cover the entire | |
6112 | * range. | |
6113 | * | |
6114 | * In that case build_sched_domains() will have terminated the iteration early | |
6115 | * and our sibling sd spans will be empty. Domains should always include the | |
6116 | * cpu they're built on, so check that. | |
6117 | * | |
6118 | */ | |
6119 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6120 | { | |
6121 | const struct cpumask *span = sched_domain_span(sd); | |
6122 | struct sd_data *sdd = sd->private; | |
6123 | struct sched_domain *sibling; | |
6124 | int i; | |
6125 | ||
6126 | for_each_cpu(i, span) { | |
6127 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6128 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6129 | continue; | |
6130 | ||
6131 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6132 | } | |
6133 | } | |
6134 | ||
6135 | /* | |
6136 | * Return the canonical balance cpu for this group, this is the first cpu | |
6137 | * of this group that's also in the iteration mask. | |
6138 | */ | |
6139 | int group_balance_cpu(struct sched_group *sg) | |
6140 | { | |
6141 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6142 | } | |
6143 | ||
e3589f6c PZ |
6144 | static int |
6145 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6146 | { | |
6147 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6148 | const struct cpumask *span = sched_domain_span(sd); | |
6149 | struct cpumask *covered = sched_domains_tmpmask; | |
6150 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6151 | struct sched_domain *sibling; |
e3589f6c PZ |
6152 | int i; |
6153 | ||
6154 | cpumask_clear(covered); | |
6155 | ||
6156 | for_each_cpu(i, span) { | |
6157 | struct cpumask *sg_span; | |
6158 | ||
6159 | if (cpumask_test_cpu(i, covered)) | |
6160 | continue; | |
6161 | ||
aaecac4a | 6162 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6163 | |
6164 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6165 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6166 | continue; |
6167 | ||
e3589f6c | 6168 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6169 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6170 | |
6171 | if (!sg) | |
6172 | goto fail; | |
6173 | ||
6174 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6175 | if (sibling->child) |
6176 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6177 | else | |
e3589f6c PZ |
6178 | cpumask_set_cpu(i, sg_span); |
6179 | ||
6180 | cpumask_or(covered, covered, sg_span); | |
6181 | ||
63b2ca30 NP |
6182 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6183 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6184 | build_group_mask(sd, sg); |
6185 | ||
c3decf0d | 6186 | /* |
63b2ca30 | 6187 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6188 | * domains and no possible iteration will get us here, we won't |
6189 | * die on a /0 trap. | |
6190 | */ | |
ca8ce3d0 | 6191 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
bf475ce0 | 6192 | sg->sgc->min_capacity = SCHED_CAPACITY_SCALE; |
e3589f6c | 6193 | |
c1174876 PZ |
6194 | /* |
6195 | * Make sure the first group of this domain contains the | |
6196 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6197 | * breaks. See update_sg_lb_stats(). | |
6198 | */ | |
74a5ce20 | 6199 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6200 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6201 | groups = sg; |
6202 | ||
6203 | if (!first) | |
6204 | first = sg; | |
6205 | if (last) | |
6206 | last->next = sg; | |
6207 | last = sg; | |
6208 | last->next = first; | |
6209 | } | |
6210 | sd->groups = groups; | |
6211 | ||
6212 | return 0; | |
6213 | ||
6214 | fail: | |
6215 | free_sched_groups(first, 0); | |
6216 | ||
6217 | return -ENOMEM; | |
6218 | } | |
6219 | ||
dce840a0 | 6220 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6221 | { |
dce840a0 PZ |
6222 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6223 | struct sched_domain *child = sd->child; | |
1da177e4 | 6224 | |
dce840a0 PZ |
6225 | if (child) |
6226 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6227 | |
9c3f75cb | 6228 | if (sg) { |
dce840a0 | 6229 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6230 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6231 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6232 | } |
dce840a0 PZ |
6233 | |
6234 | return cpu; | |
1e9f28fa | 6235 | } |
1e9f28fa | 6236 | |
01a08546 | 6237 | /* |
dce840a0 PZ |
6238 | * build_sched_groups will build a circular linked list of the groups |
6239 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6240 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6241 | * |
6242 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6243 | */ |
e3589f6c PZ |
6244 | static int |
6245 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6246 | { |
dce840a0 PZ |
6247 | struct sched_group *first = NULL, *last = NULL; |
6248 | struct sd_data *sdd = sd->private; | |
6249 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6250 | struct cpumask *covered; |
dce840a0 | 6251 | int i; |
9c1cfda2 | 6252 | |
e3589f6c PZ |
6253 | get_group(cpu, sdd, &sd->groups); |
6254 | atomic_inc(&sd->groups->ref); | |
6255 | ||
0936629f | 6256 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6257 | return 0; |
6258 | ||
f96225fd PZ |
6259 | lockdep_assert_held(&sched_domains_mutex); |
6260 | covered = sched_domains_tmpmask; | |
6261 | ||
dce840a0 | 6262 | cpumask_clear(covered); |
6711cab4 | 6263 | |
dce840a0 PZ |
6264 | for_each_cpu(i, span) { |
6265 | struct sched_group *sg; | |
cd08e923 | 6266 | int group, j; |
6711cab4 | 6267 | |
dce840a0 PZ |
6268 | if (cpumask_test_cpu(i, covered)) |
6269 | continue; | |
6711cab4 | 6270 | |
cd08e923 | 6271 | group = get_group(i, sdd, &sg); |
c1174876 | 6272 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6273 | |
dce840a0 PZ |
6274 | for_each_cpu(j, span) { |
6275 | if (get_group(j, sdd, NULL) != group) | |
6276 | continue; | |
0601a88d | 6277 | |
dce840a0 PZ |
6278 | cpumask_set_cpu(j, covered); |
6279 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6280 | } | |
0601a88d | 6281 | |
dce840a0 PZ |
6282 | if (!first) |
6283 | first = sg; | |
6284 | if (last) | |
6285 | last->next = sg; | |
6286 | last = sg; | |
6287 | } | |
6288 | last->next = first; | |
e3589f6c PZ |
6289 | |
6290 | return 0; | |
0601a88d | 6291 | } |
51888ca2 | 6292 | |
89c4710e | 6293 | /* |
63b2ca30 | 6294 | * Initialize sched groups cpu_capacity. |
89c4710e | 6295 | * |
63b2ca30 | 6296 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6297 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6298 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6299 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6300 | * group having more cpu_capacity will pickup more load compared to the | |
6301 | * group having less cpu_capacity. | |
89c4710e | 6302 | */ |
63b2ca30 | 6303 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6304 | { |
e3589f6c | 6305 | struct sched_group *sg = sd->groups; |
89c4710e | 6306 | |
94c95ba6 | 6307 | WARN_ON(!sg); |
e3589f6c PZ |
6308 | |
6309 | do { | |
afe06efd TC |
6310 | int cpu, max_cpu = -1; |
6311 | ||
e3589f6c | 6312 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); |
afe06efd TC |
6313 | |
6314 | if (!(sd->flags & SD_ASYM_PACKING)) | |
6315 | goto next; | |
6316 | ||
6317 | for_each_cpu(cpu, sched_group_cpus(sg)) { | |
6318 | if (max_cpu < 0) | |
6319 | max_cpu = cpu; | |
6320 | else if (sched_asym_prefer(cpu, max_cpu)) | |
6321 | max_cpu = cpu; | |
6322 | } | |
6323 | sg->asym_prefer_cpu = max_cpu; | |
6324 | ||
6325 | next: | |
e3589f6c PZ |
6326 | sg = sg->next; |
6327 | } while (sg != sd->groups); | |
89c4710e | 6328 | |
c1174876 | 6329 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6330 | return; |
aae6d3dd | 6331 | |
63b2ca30 | 6332 | update_group_capacity(sd, cpu); |
89c4710e SS |
6333 | } |
6334 | ||
7c16ec58 MT |
6335 | /* |
6336 | * Initializers for schedule domains | |
6337 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6338 | */ | |
6339 | ||
1d3504fc | 6340 | static int default_relax_domain_level = -1; |
60495e77 | 6341 | int sched_domain_level_max; |
1d3504fc HS |
6342 | |
6343 | static int __init setup_relax_domain_level(char *str) | |
6344 | { | |
a841f8ce DS |
6345 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6346 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6347 | |
1d3504fc HS |
6348 | return 1; |
6349 | } | |
6350 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6351 | ||
6352 | static void set_domain_attribute(struct sched_domain *sd, | |
6353 | struct sched_domain_attr *attr) | |
6354 | { | |
6355 | int request; | |
6356 | ||
6357 | if (!attr || attr->relax_domain_level < 0) { | |
6358 | if (default_relax_domain_level < 0) | |
6359 | return; | |
6360 | else | |
6361 | request = default_relax_domain_level; | |
6362 | } else | |
6363 | request = attr->relax_domain_level; | |
6364 | if (request < sd->level) { | |
6365 | /* turn off idle balance on this domain */ | |
c88d5910 | 6366 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6367 | } else { |
6368 | /* turn on idle balance on this domain */ | |
c88d5910 | 6369 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6370 | } |
6371 | } | |
6372 | ||
54ab4ff4 PZ |
6373 | static void __sdt_free(const struct cpumask *cpu_map); |
6374 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6375 | ||
2109b99e AH |
6376 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6377 | const struct cpumask *cpu_map) | |
6378 | { | |
6379 | switch (what) { | |
2109b99e | 6380 | case sa_rootdomain: |
822ff793 PZ |
6381 | if (!atomic_read(&d->rd->refcount)) |
6382 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6383 | case sa_sd: |
6384 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6385 | case sa_sd_storage: |
54ab4ff4 | 6386 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6387 | case sa_none: |
6388 | break; | |
6389 | } | |
6390 | } | |
3404c8d9 | 6391 | |
2109b99e AH |
6392 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6393 | const struct cpumask *cpu_map) | |
6394 | { | |
dce840a0 PZ |
6395 | memset(d, 0, sizeof(*d)); |
6396 | ||
54ab4ff4 PZ |
6397 | if (__sdt_alloc(cpu_map)) |
6398 | return sa_sd_storage; | |
dce840a0 PZ |
6399 | d->sd = alloc_percpu(struct sched_domain *); |
6400 | if (!d->sd) | |
6401 | return sa_sd_storage; | |
2109b99e | 6402 | d->rd = alloc_rootdomain(); |
dce840a0 | 6403 | if (!d->rd) |
21d42ccf | 6404 | return sa_sd; |
2109b99e AH |
6405 | return sa_rootdomain; |
6406 | } | |
57d885fe | 6407 | |
dce840a0 PZ |
6408 | /* |
6409 | * NULL the sd_data elements we've used to build the sched_domain and | |
6410 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6411 | * will not free the data we're using. | |
6412 | */ | |
6413 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6414 | { | |
6415 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6416 | |
6417 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6418 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6419 | ||
24fc7edb PZ |
6420 | if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref)) |
6421 | *per_cpu_ptr(sdd->sds, cpu) = NULL; | |
6422 | ||
e3589f6c | 6423 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6424 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6425 | |
63b2ca30 NP |
6426 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6427 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6428 | } |
6429 | ||
cb83b629 | 6430 | #ifdef CONFIG_NUMA |
cb83b629 | 6431 | static int sched_domains_numa_levels; |
e3fe70b1 | 6432 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6433 | static int *sched_domains_numa_distance; |
9942f79b | 6434 | int sched_max_numa_distance; |
cb83b629 PZ |
6435 | static struct cpumask ***sched_domains_numa_masks; |
6436 | static int sched_domains_curr_level; | |
143e1e28 | 6437 | #endif |
cb83b629 | 6438 | |
143e1e28 VG |
6439 | /* |
6440 | * SD_flags allowed in topology descriptions. | |
6441 | * | |
94f438c8 PZ |
6442 | * These flags are purely descriptive of the topology and do not prescribe |
6443 | * behaviour. Behaviour is artificial and mapped in the below sd_init() | |
6444 | * function: | |
143e1e28 | 6445 | * |
94f438c8 PZ |
6446 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
6447 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
6448 | * SD_NUMA - describes NUMA topologies | |
6449 | * SD_SHARE_POWERDOMAIN - describes shared power domain | |
1f6e6c7c | 6450 | * SD_ASYM_CPUCAPACITY - describes mixed capacity topologies |
94f438c8 PZ |
6451 | * |
6452 | * Odd one out, which beside describing the topology has a quirk also | |
6453 | * prescribes the desired behaviour that goes along with it: | |
143e1e28 | 6454 | * |
94f438c8 | 6455 | * SD_ASYM_PACKING - describes SMT quirks |
143e1e28 VG |
6456 | */ |
6457 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6458 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6459 | SD_SHARE_PKG_RESOURCES | \ |
6460 | SD_NUMA | \ | |
d77b3ed5 | 6461 | SD_ASYM_PACKING | \ |
1f6e6c7c | 6462 | SD_ASYM_CPUCAPACITY | \ |
d77b3ed5 | 6463 | SD_SHARE_POWERDOMAIN) |
cb83b629 PZ |
6464 | |
6465 | static struct sched_domain * | |
3676b13e | 6466 | sd_init(struct sched_domain_topology_level *tl, |
24fc7edb | 6467 | const struct cpumask *cpu_map, |
3676b13e | 6468 | struct sched_domain *child, int cpu) |
cb83b629 | 6469 | { |
24fc7edb PZ |
6470 | struct sd_data *sdd = &tl->data; |
6471 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); | |
6472 | int sd_id, sd_weight, sd_flags = 0; | |
143e1e28 VG |
6473 | |
6474 | #ifdef CONFIG_NUMA | |
6475 | /* | |
6476 | * Ugly hack to pass state to sd_numa_mask()... | |
6477 | */ | |
6478 | sched_domains_curr_level = tl->numa_level; | |
6479 | #endif | |
6480 | ||
6481 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6482 | ||
6483 | if (tl->sd_flags) | |
6484 | sd_flags = (*tl->sd_flags)(); | |
6485 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6486 | "wrong sd_flags in topology description\n")) | |
6487 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6488 | |
6489 | *sd = (struct sched_domain){ | |
6490 | .min_interval = sd_weight, | |
6491 | .max_interval = 2*sd_weight, | |
6492 | .busy_factor = 32, | |
870a0bb5 | 6493 | .imbalance_pct = 125, |
143e1e28 VG |
6494 | |
6495 | .cache_nice_tries = 0, | |
6496 | .busy_idx = 0, | |
6497 | .idle_idx = 0, | |
cb83b629 PZ |
6498 | .newidle_idx = 0, |
6499 | .wake_idx = 0, | |
6500 | .forkexec_idx = 0, | |
6501 | ||
6502 | .flags = 1*SD_LOAD_BALANCE | |
6503 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6504 | | 1*SD_BALANCE_EXEC |
6505 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6506 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6507 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6508 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6509 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6510 | | 0*SD_SERIALIZE |
cb83b629 | 6511 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6512 | | 0*SD_NUMA |
6513 | | sd_flags | |
cb83b629 | 6514 | , |
143e1e28 | 6515 | |
cb83b629 PZ |
6516 | .last_balance = jiffies, |
6517 | .balance_interval = sd_weight, | |
143e1e28 | 6518 | .smt_gain = 0, |
2b4cfe64 JL |
6519 | .max_newidle_lb_cost = 0, |
6520 | .next_decay_max_lb_cost = jiffies, | |
3676b13e | 6521 | .child = child, |
143e1e28 VG |
6522 | #ifdef CONFIG_SCHED_DEBUG |
6523 | .name = tl->name, | |
6524 | #endif | |
cb83b629 | 6525 | }; |
cb83b629 | 6526 | |
24fc7edb PZ |
6527 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
6528 | sd_id = cpumask_first(sched_domain_span(sd)); | |
6529 | ||
cb83b629 | 6530 | /* |
143e1e28 | 6531 | * Convert topological properties into behaviour. |
cb83b629 | 6532 | */ |
143e1e28 | 6533 | |
9ee1cda5 MR |
6534 | if (sd->flags & SD_ASYM_CPUCAPACITY) { |
6535 | struct sched_domain *t = sd; | |
6536 | ||
6537 | for_each_lower_domain(t) | |
6538 | t->flags |= SD_BALANCE_WAKE; | |
6539 | } | |
6540 | ||
5d4dfddd | 6541 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6542 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6543 | sd->imbalance_pct = 110; |
6544 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6545 | |
6546 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6547 | sd->imbalance_pct = 117; | |
6548 | sd->cache_nice_tries = 1; | |
6549 | sd->busy_idx = 2; | |
6550 | ||
6551 | #ifdef CONFIG_NUMA | |
6552 | } else if (sd->flags & SD_NUMA) { | |
6553 | sd->cache_nice_tries = 2; | |
6554 | sd->busy_idx = 3; | |
6555 | sd->idle_idx = 2; | |
6556 | ||
6557 | sd->flags |= SD_SERIALIZE; | |
6558 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6559 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6560 | SD_BALANCE_FORK | | |
6561 | SD_WAKE_AFFINE); | |
6562 | } | |
6563 | ||
6564 | #endif | |
6565 | } else { | |
6566 | sd->flags |= SD_PREFER_SIBLING; | |
6567 | sd->cache_nice_tries = 1; | |
6568 | sd->busy_idx = 2; | |
6569 | sd->idle_idx = 1; | |
6570 | } | |
6571 | ||
24fc7edb PZ |
6572 | /* |
6573 | * For all levels sharing cache; connect a sched_domain_shared | |
6574 | * instance. | |
6575 | */ | |
6576 | if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6577 | sd->shared = *per_cpu_ptr(sdd->sds, sd_id); | |
6578 | atomic_inc(&sd->shared->ref); | |
0e369d75 | 6579 | atomic_set(&sd->shared->nr_busy_cpus, sd_weight); |
24fc7edb PZ |
6580 | } |
6581 | ||
6582 | sd->private = sdd; | |
cb83b629 PZ |
6583 | |
6584 | return sd; | |
6585 | } | |
6586 | ||
143e1e28 VG |
6587 | /* |
6588 | * Topology list, bottom-up. | |
6589 | */ | |
6590 | static struct sched_domain_topology_level default_topology[] = { | |
6591 | #ifdef CONFIG_SCHED_SMT | |
6592 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6593 | #endif | |
6594 | #ifdef CONFIG_SCHED_MC | |
6595 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6596 | #endif |
6597 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6598 | { NULL, }, | |
6599 | }; | |
6600 | ||
c6e1e7b5 JG |
6601 | static struct sched_domain_topology_level *sched_domain_topology = |
6602 | default_topology; | |
143e1e28 VG |
6603 | |
6604 | #define for_each_sd_topology(tl) \ | |
6605 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6606 | ||
6607 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6608 | { | |
8f37961c TC |
6609 | if (WARN_ON_ONCE(sched_smp_initialized)) |
6610 | return; | |
6611 | ||
143e1e28 VG |
6612 | sched_domain_topology = tl; |
6613 | } | |
6614 | ||
6615 | #ifdef CONFIG_NUMA | |
6616 | ||
cb83b629 PZ |
6617 | static const struct cpumask *sd_numa_mask(int cpu) |
6618 | { | |
6619 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6620 | } | |
6621 | ||
d039ac60 PZ |
6622 | static void sched_numa_warn(const char *str) |
6623 | { | |
6624 | static int done = false; | |
6625 | int i,j; | |
6626 | ||
6627 | if (done) | |
6628 | return; | |
6629 | ||
6630 | done = true; | |
6631 | ||
6632 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6633 | ||
6634 | for (i = 0; i < nr_node_ids; i++) { | |
6635 | printk(KERN_WARNING " "); | |
6636 | for (j = 0; j < nr_node_ids; j++) | |
6637 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6638 | printk(KERN_CONT "\n"); | |
6639 | } | |
6640 | printk(KERN_WARNING "\n"); | |
6641 | } | |
6642 | ||
9942f79b | 6643 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6644 | { |
6645 | int i; | |
6646 | ||
6647 | if (distance == node_distance(0, 0)) | |
6648 | return true; | |
6649 | ||
6650 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6651 | if (sched_domains_numa_distance[i] == distance) | |
6652 | return true; | |
6653 | } | |
6654 | ||
6655 | return false; | |
6656 | } | |
6657 | ||
e3fe70b1 RR |
6658 | /* |
6659 | * A system can have three types of NUMA topology: | |
6660 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6661 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6662 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6663 | * | |
6664 | * The difference between a glueless mesh topology and a backplane | |
6665 | * topology lies in whether communication between not directly | |
6666 | * connected nodes goes through intermediary nodes (where programs | |
6667 | * could run), or through backplane controllers. This affects | |
6668 | * placement of programs. | |
6669 | * | |
6670 | * The type of topology can be discerned with the following tests: | |
6671 | * - If the maximum distance between any nodes is 1 hop, the system | |
6672 | * is directly connected. | |
6673 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6674 | * there is an intermediary node C, which is < N hops away from both | |
6675 | * nodes A and B, the system is a glueless mesh. | |
6676 | */ | |
6677 | static void init_numa_topology_type(void) | |
6678 | { | |
6679 | int a, b, c, n; | |
6680 | ||
6681 | n = sched_max_numa_distance; | |
6682 | ||
e237882b | 6683 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6684 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6685 | return; |
6686 | } | |
e3fe70b1 RR |
6687 | |
6688 | for_each_online_node(a) { | |
6689 | for_each_online_node(b) { | |
6690 | /* Find two nodes furthest removed from each other. */ | |
6691 | if (node_distance(a, b) < n) | |
6692 | continue; | |
6693 | ||
6694 | /* Is there an intermediary node between a and b? */ | |
6695 | for_each_online_node(c) { | |
6696 | if (node_distance(a, c) < n && | |
6697 | node_distance(b, c) < n) { | |
6698 | sched_numa_topology_type = | |
6699 | NUMA_GLUELESS_MESH; | |
6700 | return; | |
6701 | } | |
6702 | } | |
6703 | ||
6704 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6705 | return; | |
6706 | } | |
6707 | } | |
6708 | } | |
6709 | ||
cb83b629 PZ |
6710 | static void sched_init_numa(void) |
6711 | { | |
6712 | int next_distance, curr_distance = node_distance(0, 0); | |
6713 | struct sched_domain_topology_level *tl; | |
6714 | int level = 0; | |
6715 | int i, j, k; | |
6716 | ||
cb83b629 PZ |
6717 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6718 | if (!sched_domains_numa_distance) | |
6719 | return; | |
6720 | ||
6721 | /* | |
6722 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6723 | * unique distances in the node_distance() table. | |
6724 | * | |
6725 | * Assumes node_distance(0,j) includes all distances in | |
6726 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6727 | */ |
6728 | next_distance = curr_distance; | |
6729 | for (i = 0; i < nr_node_ids; i++) { | |
6730 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6731 | for (k = 0; k < nr_node_ids; k++) { |
6732 | int distance = node_distance(i, k); | |
6733 | ||
6734 | if (distance > curr_distance && | |
6735 | (distance < next_distance || | |
6736 | next_distance == curr_distance)) | |
6737 | next_distance = distance; | |
6738 | ||
6739 | /* | |
6740 | * While not a strong assumption it would be nice to know | |
6741 | * about cases where if node A is connected to B, B is not | |
6742 | * equally connected to A. | |
6743 | */ | |
6744 | if (sched_debug() && node_distance(k, i) != distance) | |
6745 | sched_numa_warn("Node-distance not symmetric"); | |
6746 | ||
6747 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6748 | sched_numa_warn("Node-0 not representative"); | |
6749 | } | |
6750 | if (next_distance != curr_distance) { | |
6751 | sched_domains_numa_distance[level++] = next_distance; | |
6752 | sched_domains_numa_levels = level; | |
6753 | curr_distance = next_distance; | |
6754 | } else break; | |
cb83b629 | 6755 | } |
d039ac60 PZ |
6756 | |
6757 | /* | |
6758 | * In case of sched_debug() we verify the above assumption. | |
6759 | */ | |
6760 | if (!sched_debug()) | |
6761 | break; | |
cb83b629 | 6762 | } |
c123588b AR |
6763 | |
6764 | if (!level) | |
6765 | return; | |
6766 | ||
cb83b629 PZ |
6767 | /* |
6768 | * 'level' contains the number of unique distances, excluding the | |
6769 | * identity distance node_distance(i,i). | |
6770 | * | |
28b4a521 | 6771 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6772 | * numbers. |
6773 | */ | |
6774 | ||
5f7865f3 TC |
6775 | /* |
6776 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6777 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6778 | * the array will contain less then 'level' members. This could be | |
6779 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6780 | * in other functions. | |
6781 | * | |
6782 | * We reset it to 'level' at the end of this function. | |
6783 | */ | |
6784 | sched_domains_numa_levels = 0; | |
6785 | ||
cb83b629 PZ |
6786 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6787 | if (!sched_domains_numa_masks) | |
6788 | return; | |
6789 | ||
6790 | /* | |
6791 | * Now for each level, construct a mask per node which contains all | |
6792 | * cpus of nodes that are that many hops away from us. | |
6793 | */ | |
6794 | for (i = 0; i < level; i++) { | |
6795 | sched_domains_numa_masks[i] = | |
6796 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6797 | if (!sched_domains_numa_masks[i]) | |
6798 | return; | |
6799 | ||
6800 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6801 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6802 | if (!mask) |
6803 | return; | |
6804 | ||
6805 | sched_domains_numa_masks[i][j] = mask; | |
6806 | ||
9c03ee14 | 6807 | for_each_node(k) { |
dd7d8634 | 6808 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6809 | continue; |
6810 | ||
6811 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6812 | } | |
6813 | } | |
6814 | } | |
6815 | ||
143e1e28 VG |
6816 | /* Compute default topology size */ |
6817 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6818 | ||
c515db8c | 6819 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6820 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6821 | if (!tl) | |
6822 | return; | |
6823 | ||
6824 | /* | |
6825 | * Copy the default topology bits.. | |
6826 | */ | |
143e1e28 VG |
6827 | for (i = 0; sched_domain_topology[i].mask; i++) |
6828 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6829 | |
6830 | /* | |
6831 | * .. and append 'j' levels of NUMA goodness. | |
6832 | */ | |
6833 | for (j = 0; j < level; i++, j++) { | |
6834 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6835 | .mask = sd_numa_mask, |
143e1e28 | 6836 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6837 | .flags = SDTL_OVERLAP, |
6838 | .numa_level = j, | |
143e1e28 | 6839 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6840 | }; |
6841 | } | |
6842 | ||
6843 | sched_domain_topology = tl; | |
5f7865f3 TC |
6844 | |
6845 | sched_domains_numa_levels = level; | |
9942f79b | 6846 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6847 | |
6848 | init_numa_topology_type(); | |
cb83b629 | 6849 | } |
301a5cba | 6850 | |
135fb3e1 | 6851 | static void sched_domains_numa_masks_set(unsigned int cpu) |
301a5cba | 6852 | { |
301a5cba | 6853 | int node = cpu_to_node(cpu); |
135fb3e1 | 6854 | int i, j; |
301a5cba TC |
6855 | |
6856 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6857 | for (j = 0; j < nr_node_ids; j++) { | |
6858 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6859 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6860 | } | |
6861 | } | |
6862 | } | |
6863 | ||
135fb3e1 | 6864 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
301a5cba TC |
6865 | { |
6866 | int i, j; | |
135fb3e1 | 6867 | |
301a5cba TC |
6868 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6869 | for (j = 0; j < nr_node_ids; j++) | |
6870 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6871 | } | |
6872 | } | |
6873 | ||
cb83b629 | 6874 | #else |
135fb3e1 TG |
6875 | static inline void sched_init_numa(void) { } |
6876 | static void sched_domains_numa_masks_set(unsigned int cpu) { } | |
6877 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
cb83b629 PZ |
6878 | #endif /* CONFIG_NUMA */ |
6879 | ||
54ab4ff4 PZ |
6880 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6881 | { | |
6882 | struct sched_domain_topology_level *tl; | |
6883 | int j; | |
6884 | ||
27723a68 | 6885 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6886 | struct sd_data *sdd = &tl->data; |
6887 | ||
6888 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6889 | if (!sdd->sd) | |
6890 | return -ENOMEM; | |
6891 | ||
24fc7edb PZ |
6892 | sdd->sds = alloc_percpu(struct sched_domain_shared *); |
6893 | if (!sdd->sds) | |
6894 | return -ENOMEM; | |
6895 | ||
54ab4ff4 PZ |
6896 | sdd->sg = alloc_percpu(struct sched_group *); |
6897 | if (!sdd->sg) | |
6898 | return -ENOMEM; | |
6899 | ||
63b2ca30 NP |
6900 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6901 | if (!sdd->sgc) | |
9c3f75cb PZ |
6902 | return -ENOMEM; |
6903 | ||
54ab4ff4 PZ |
6904 | for_each_cpu(j, cpu_map) { |
6905 | struct sched_domain *sd; | |
24fc7edb | 6906 | struct sched_domain_shared *sds; |
54ab4ff4 | 6907 | struct sched_group *sg; |
63b2ca30 | 6908 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6909 | |
5cc389bc | 6910 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6911 | GFP_KERNEL, cpu_to_node(j)); |
6912 | if (!sd) | |
6913 | return -ENOMEM; | |
6914 | ||
6915 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6916 | ||
24fc7edb PZ |
6917 | sds = kzalloc_node(sizeof(struct sched_domain_shared), |
6918 | GFP_KERNEL, cpu_to_node(j)); | |
6919 | if (!sds) | |
6920 | return -ENOMEM; | |
6921 | ||
6922 | *per_cpu_ptr(sdd->sds, j) = sds; | |
6923 | ||
54ab4ff4 PZ |
6924 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
6925 | GFP_KERNEL, cpu_to_node(j)); | |
6926 | if (!sg) | |
6927 | return -ENOMEM; | |
6928 | ||
30b4e9eb IM |
6929 | sg->next = sg; |
6930 | ||
54ab4ff4 | 6931 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6932 | |
63b2ca30 | 6933 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6934 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6935 | if (!sgc) |
9c3f75cb PZ |
6936 | return -ENOMEM; |
6937 | ||
63b2ca30 | 6938 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6939 | } |
6940 | } | |
6941 | ||
6942 | return 0; | |
6943 | } | |
6944 | ||
6945 | static void __sdt_free(const struct cpumask *cpu_map) | |
6946 | { | |
6947 | struct sched_domain_topology_level *tl; | |
6948 | int j; | |
6949 | ||
27723a68 | 6950 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6951 | struct sd_data *sdd = &tl->data; |
6952 | ||
6953 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6954 | struct sched_domain *sd; |
6955 | ||
6956 | if (sdd->sd) { | |
6957 | sd = *per_cpu_ptr(sdd->sd, j); | |
6958 | if (sd && (sd->flags & SD_OVERLAP)) | |
6959 | free_sched_groups(sd->groups, 0); | |
6960 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6961 | } | |
6962 | ||
24fc7edb PZ |
6963 | if (sdd->sds) |
6964 | kfree(*per_cpu_ptr(sdd->sds, j)); | |
fb2cf2c6 | 6965 | if (sdd->sg) |
6966 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6967 | if (sdd->sgc) |
6968 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6969 | } |
6970 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6971 | sdd->sd = NULL; |
24fc7edb PZ |
6972 | free_percpu(sdd->sds); |
6973 | sdd->sds = NULL; | |
54ab4ff4 | 6974 | free_percpu(sdd->sg); |
fb2cf2c6 | 6975 | sdd->sg = NULL; |
63b2ca30 NP |
6976 | free_percpu(sdd->sgc); |
6977 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6978 | } |
6979 | } | |
6980 | ||
2c402dc3 | 6981 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6982 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6983 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6984 | { |
24fc7edb | 6985 | struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu); |
2c402dc3 | 6986 | |
60495e77 PZ |
6987 | if (child) { |
6988 | sd->level = child->level + 1; | |
6989 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6990 | child->parent = sd; |
6ae72dff PZ |
6991 | |
6992 | if (!cpumask_subset(sched_domain_span(child), | |
6993 | sched_domain_span(sd))) { | |
6994 | pr_err("BUG: arch topology borken\n"); | |
6995 | #ifdef CONFIG_SCHED_DEBUG | |
6996 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6997 | child->name, sd->name); | |
6998 | #endif | |
6999 | /* Fixup, ensure @sd has at least @child cpus. */ | |
7000 | cpumask_or(sched_domain_span(sd), | |
7001 | sched_domain_span(sd), | |
7002 | sched_domain_span(child)); | |
7003 | } | |
7004 | ||
60495e77 | 7005 | } |
a841f8ce | 7006 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
7007 | |
7008 | return sd; | |
7009 | } | |
7010 | ||
2109b99e AH |
7011 | /* |
7012 | * Build sched domains for a given set of cpus and attach the sched domains | |
7013 | * to the individual cpus | |
7014 | */ | |
dce840a0 PZ |
7015 | static int build_sched_domains(const struct cpumask *cpu_map, |
7016 | struct sched_domain_attr *attr) | |
2109b99e | 7017 | { |
1c632169 | 7018 | enum s_alloc alloc_state; |
dce840a0 | 7019 | struct sched_domain *sd; |
2109b99e | 7020 | struct s_data d; |
cd92bfd3 | 7021 | struct rq *rq = NULL; |
822ff793 | 7022 | int i, ret = -ENOMEM; |
9c1cfda2 | 7023 | |
2109b99e AH |
7024 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7025 | if (alloc_state != sa_rootdomain) | |
7026 | goto error; | |
9c1cfda2 | 7027 | |
dce840a0 | 7028 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7029 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7030 | struct sched_domain_topology_level *tl; |
7031 | ||
3bd65a80 | 7032 | sd = NULL; |
27723a68 | 7033 | for_each_sd_topology(tl) { |
4a850cbe | 7034 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
7035 | if (tl == sched_domain_topology) |
7036 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
7037 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
7038 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
7039 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
7040 | break; | |
e3589f6c | 7041 | } |
dce840a0 PZ |
7042 | } |
7043 | ||
7044 | /* Build the groups for the domains */ | |
7045 | for_each_cpu(i, cpu_map) { | |
7046 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7047 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
7048 | if (sd->flags & SD_OVERLAP) { |
7049 | if (build_overlap_sched_groups(sd, i)) | |
7050 | goto error; | |
7051 | } else { | |
7052 | if (build_sched_groups(sd, i)) | |
7053 | goto error; | |
7054 | } | |
1cf51902 | 7055 | } |
a06dadbe | 7056 | } |
9c1cfda2 | 7057 | |
ced549fa | 7058 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
7059 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7060 | if (!cpumask_test_cpu(i, cpu_map)) | |
7061 | continue; | |
9c1cfda2 | 7062 | |
dce840a0 PZ |
7063 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7064 | claim_allocations(i, sd); | |
63b2ca30 | 7065 | init_sched_groups_capacity(i, sd); |
dce840a0 | 7066 | } |
f712c0c7 | 7067 | } |
9c1cfda2 | 7068 | |
1da177e4 | 7069 | /* Attach the domains */ |
dce840a0 | 7070 | rcu_read_lock(); |
abcd083a | 7071 | for_each_cpu(i, cpu_map) { |
cd92bfd3 | 7072 | rq = cpu_rq(i); |
21d42ccf | 7073 | sd = *per_cpu_ptr(d.sd, i); |
cd92bfd3 DE |
7074 | |
7075 | /* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */ | |
7076 | if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity)) | |
7077 | WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig); | |
7078 | ||
49a02c51 | 7079 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7080 | } |
dce840a0 | 7081 | rcu_read_unlock(); |
51888ca2 | 7082 | |
a18a579e | 7083 | if (rq && sched_debug_enabled) { |
cd92bfd3 DE |
7084 | pr_info("span: %*pbl (max cpu_capacity = %lu)\n", |
7085 | cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity); | |
7086 | } | |
7087 | ||
822ff793 | 7088 | ret = 0; |
51888ca2 | 7089 | error: |
2109b99e | 7090 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7091 | return ret; |
1da177e4 | 7092 | } |
029190c5 | 7093 | |
acc3f5d7 | 7094 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7095 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7096 | static struct sched_domain_attr *dattr_cur; |
7097 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7098 | |
7099 | /* | |
7100 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7101 | * cpumask) fails, then fallback to a single sched domain, |
7102 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7103 | */ |
4212823f | 7104 | static cpumask_var_t fallback_doms; |
029190c5 | 7105 | |
ee79d1bd HC |
7106 | /* |
7107 | * arch_update_cpu_topology lets virtualized architectures update the | |
7108 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7109 | * or 0 if it stayed the same. | |
7110 | */ | |
52f5684c | 7111 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 7112 | { |
ee79d1bd | 7113 | return 0; |
22e52b07 HC |
7114 | } |
7115 | ||
acc3f5d7 RR |
7116 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7117 | { | |
7118 | int i; | |
7119 | cpumask_var_t *doms; | |
7120 | ||
7121 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7122 | if (!doms) | |
7123 | return NULL; | |
7124 | for (i = 0; i < ndoms; i++) { | |
7125 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7126 | free_sched_domains(doms, i); | |
7127 | return NULL; | |
7128 | } | |
7129 | } | |
7130 | return doms; | |
7131 | } | |
7132 | ||
7133 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7134 | { | |
7135 | unsigned int i; | |
7136 | for (i = 0; i < ndoms; i++) | |
7137 | free_cpumask_var(doms[i]); | |
7138 | kfree(doms); | |
7139 | } | |
7140 | ||
1a20ff27 | 7141 | /* |
41a2d6cf | 7142 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7143 | * For now this just excludes isolated cpus, but could be used to |
7144 | * exclude other special cases in the future. | |
1a20ff27 | 7145 | */ |
c4a8849a | 7146 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7147 | { |
7378547f MM |
7148 | int err; |
7149 | ||
22e52b07 | 7150 | arch_update_cpu_topology(); |
029190c5 | 7151 | ndoms_cur = 1; |
acc3f5d7 | 7152 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7153 | if (!doms_cur) |
acc3f5d7 RR |
7154 | doms_cur = &fallback_doms; |
7155 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 7156 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7157 | register_sched_domain_sysctl(); |
7378547f MM |
7158 | |
7159 | return err; | |
1a20ff27 DG |
7160 | } |
7161 | ||
1a20ff27 DG |
7162 | /* |
7163 | * Detach sched domains from a group of cpus specified in cpu_map | |
7164 | * These cpus will now be attached to the NULL domain | |
7165 | */ | |
96f874e2 | 7166 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7167 | { |
7168 | int i; | |
7169 | ||
dce840a0 | 7170 | rcu_read_lock(); |
abcd083a | 7171 | for_each_cpu(i, cpu_map) |
57d885fe | 7172 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7173 | rcu_read_unlock(); |
1a20ff27 DG |
7174 | } |
7175 | ||
1d3504fc HS |
7176 | /* handle null as "default" */ |
7177 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7178 | struct sched_domain_attr *new, int idx_new) | |
7179 | { | |
7180 | struct sched_domain_attr tmp; | |
7181 | ||
7182 | /* fast path */ | |
7183 | if (!new && !cur) | |
7184 | return 1; | |
7185 | ||
7186 | tmp = SD_ATTR_INIT; | |
7187 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7188 | new ? (new + idx_new) : &tmp, | |
7189 | sizeof(struct sched_domain_attr)); | |
7190 | } | |
7191 | ||
029190c5 PJ |
7192 | /* |
7193 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7194 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7195 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7196 | * It destroys each deleted domain and builds each new domain. | |
7197 | * | |
acc3f5d7 | 7198 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7199 | * The masks don't intersect (don't overlap.) We should setup one |
7200 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7201 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7202 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7203 | * it as it is. | |
7204 | * | |
acc3f5d7 RR |
7205 | * The passed in 'doms_new' should be allocated using |
7206 | * alloc_sched_domains. This routine takes ownership of it and will | |
7207 | * free_sched_domains it when done with it. If the caller failed the | |
7208 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7209 | * and partition_sched_domains() will fallback to the single partition | |
7210 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7211 | * |
96f874e2 | 7212 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7213 | * ndoms_new == 0 is a special case for destroying existing domains, |
7214 | * and it will not create the default domain. | |
dfb512ec | 7215 | * |
029190c5 PJ |
7216 | * Call with hotplug lock held |
7217 | */ | |
acc3f5d7 | 7218 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7219 | struct sched_domain_attr *dattr_new) |
029190c5 | 7220 | { |
dfb512ec | 7221 | int i, j, n; |
d65bd5ec | 7222 | int new_topology; |
029190c5 | 7223 | |
712555ee | 7224 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7225 | |
7378547f MM |
7226 | /* always unregister in case we don't destroy any domains */ |
7227 | unregister_sched_domain_sysctl(); | |
7228 | ||
d65bd5ec HC |
7229 | /* Let architecture update cpu core mappings. */ |
7230 | new_topology = arch_update_cpu_topology(); | |
7231 | ||
dfb512ec | 7232 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7233 | |
7234 | /* Destroy deleted domains */ | |
7235 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7236 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7237 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7238 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7239 | goto match1; |
7240 | } | |
7241 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7242 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7243 | match1: |
7244 | ; | |
7245 | } | |
7246 | ||
c8d2d47a | 7247 | n = ndoms_cur; |
e761b772 | 7248 | if (doms_new == NULL) { |
c8d2d47a | 7249 | n = 0; |
acc3f5d7 | 7250 | doms_new = &fallback_doms; |
6ad4c188 | 7251 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7252 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7253 | } |
7254 | ||
029190c5 PJ |
7255 | /* Build new domains */ |
7256 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7257 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7258 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7259 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7260 | goto match2; |
7261 | } | |
7262 | /* no match - add a new doms_new */ | |
dce840a0 | 7263 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7264 | match2: |
7265 | ; | |
7266 | } | |
7267 | ||
7268 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7269 | if (doms_cur != &fallback_doms) |
7270 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7271 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7272 | doms_cur = doms_new; |
1d3504fc | 7273 | dattr_cur = dattr_new; |
029190c5 | 7274 | ndoms_cur = ndoms_new; |
7378547f MM |
7275 | |
7276 | register_sched_domain_sysctl(); | |
a1835615 | 7277 | |
712555ee | 7278 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7279 | } |
7280 | ||
d35be8ba SB |
7281 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7282 | ||
1da177e4 | 7283 | /* |
3a101d05 TH |
7284 | * Update cpusets according to cpu_active mask. If cpusets are |
7285 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7286 | * around partition_sched_domains(). | |
d35be8ba SB |
7287 | * |
7288 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7289 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7290 | */ |
40190a78 | 7291 | static void cpuset_cpu_active(void) |
e761b772 | 7292 | { |
40190a78 | 7293 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7294 | /* |
7295 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7296 | * resume sequence. As long as this is not the last online | |
7297 | * operation in the resume sequence, just build a single sched | |
7298 | * domain, ignoring cpusets. | |
7299 | */ | |
7300 | num_cpus_frozen--; | |
7301 | if (likely(num_cpus_frozen)) { | |
7302 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 7303 | return; |
d35be8ba | 7304 | } |
d35be8ba SB |
7305 | /* |
7306 | * This is the last CPU online operation. So fall through and | |
7307 | * restore the original sched domains by considering the | |
7308 | * cpuset configurations. | |
7309 | */ | |
3a101d05 | 7310 | } |
135fb3e1 | 7311 | cpuset_update_active_cpus(true); |
3a101d05 | 7312 | } |
e761b772 | 7313 | |
40190a78 | 7314 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7315 | { |
3c18d447 | 7316 | unsigned long flags; |
3c18d447 | 7317 | struct dl_bw *dl_b; |
533445c6 OS |
7318 | bool overflow; |
7319 | int cpus; | |
3c18d447 | 7320 | |
40190a78 | 7321 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
7322 | rcu_read_lock_sched(); |
7323 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7324 | |
533445c6 OS |
7325 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7326 | cpus = dl_bw_cpus(cpu); | |
7327 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7328 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7329 | |
533445c6 | 7330 | rcu_read_unlock_sched(); |
3c18d447 | 7331 | |
533445c6 | 7332 | if (overflow) |
135fb3e1 | 7333 | return -EBUSY; |
7ddf96b0 | 7334 | cpuset_update_active_cpus(false); |
135fb3e1 | 7335 | } else { |
d35be8ba SB |
7336 | num_cpus_frozen++; |
7337 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7338 | } |
135fb3e1 | 7339 | return 0; |
e761b772 | 7340 | } |
e761b772 | 7341 | |
40190a78 | 7342 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7343 | { |
7d976699 TG |
7344 | struct rq *rq = cpu_rq(cpu); |
7345 | unsigned long flags; | |
7346 | ||
40190a78 | 7347 | set_cpu_active(cpu, true); |
135fb3e1 | 7348 | |
40190a78 | 7349 | if (sched_smp_initialized) { |
135fb3e1 | 7350 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7351 | cpuset_cpu_active(); |
e761b772 | 7352 | } |
7d976699 TG |
7353 | |
7354 | /* | |
7355 | * Put the rq online, if not already. This happens: | |
7356 | * | |
7357 | * 1) In the early boot process, because we build the real domains | |
7358 | * after all cpus have been brought up. | |
7359 | * | |
7360 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7361 | * domains. | |
7362 | */ | |
7363 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7364 | if (rq->rd) { | |
7365 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7366 | set_rq_online(rq); | |
7367 | } | |
7368 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7369 | ||
7370 | update_max_interval(); | |
7371 | ||
40190a78 | 7372 | return 0; |
135fb3e1 TG |
7373 | } |
7374 | ||
40190a78 | 7375 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7376 | { |
135fb3e1 TG |
7377 | int ret; |
7378 | ||
40190a78 | 7379 | set_cpu_active(cpu, false); |
b2454caa PZ |
7380 | /* |
7381 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7382 | * users of this state to go away such that all new such users will | |
7383 | * observe it. | |
7384 | * | |
7385 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
7386 | * not imply sync_sched(), so wait for both. | |
7387 | * | |
7388 | * Do sync before park smpboot threads to take care the rcu boost case. | |
7389 | */ | |
7390 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
7391 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
7392 | else | |
7393 | synchronize_rcu(); | |
40190a78 TG |
7394 | |
7395 | if (!sched_smp_initialized) | |
7396 | return 0; | |
7397 | ||
7398 | ret = cpuset_cpu_inactive(cpu); | |
7399 | if (ret) { | |
7400 | set_cpu_active(cpu, true); | |
7401 | return ret; | |
135fb3e1 | 7402 | } |
40190a78 TG |
7403 | sched_domains_numa_masks_clear(cpu); |
7404 | return 0; | |
135fb3e1 TG |
7405 | } |
7406 | ||
94baf7a5 TG |
7407 | static void sched_rq_cpu_starting(unsigned int cpu) |
7408 | { | |
7409 | struct rq *rq = cpu_rq(cpu); | |
7410 | ||
7411 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7412 | update_max_interval(); |
7413 | } | |
7414 | ||
135fb3e1 TG |
7415 | int sched_cpu_starting(unsigned int cpu) |
7416 | { | |
7417 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 7418 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 7419 | return 0; |
e761b772 | 7420 | } |
e761b772 | 7421 | |
f2785ddb TG |
7422 | #ifdef CONFIG_HOTPLUG_CPU |
7423 | int sched_cpu_dying(unsigned int cpu) | |
7424 | { | |
7425 | struct rq *rq = cpu_rq(cpu); | |
7426 | unsigned long flags; | |
7427 | ||
7428 | /* Handle pending wakeups and then migrate everything off */ | |
7429 | sched_ttwu_pending(); | |
7430 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7431 | if (rq->rd) { | |
7432 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7433 | set_rq_offline(rq); | |
7434 | } | |
7435 | migrate_tasks(rq); | |
7436 | BUG_ON(rq->nr_running != 1); | |
7437 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7438 | calc_load_migrate(rq); | |
7439 | update_max_interval(); | |
20a5c8cc | 7440 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 7441 | hrtick_clear(rq); |
f2785ddb TG |
7442 | return 0; |
7443 | } | |
7444 | #endif | |
7445 | ||
1b568f0a PZ |
7446 | #ifdef CONFIG_SCHED_SMT |
7447 | DEFINE_STATIC_KEY_FALSE(sched_smt_present); | |
7448 | ||
7449 | static void sched_init_smt(void) | |
7450 | { | |
7451 | /* | |
7452 | * We've enumerated all CPUs and will assume that if any CPU | |
7453 | * has SMT siblings, CPU0 will too. | |
7454 | */ | |
7455 | if (cpumask_weight(cpu_smt_mask(0)) > 1) | |
7456 | static_branch_enable(&sched_smt_present); | |
7457 | } | |
7458 | #else | |
7459 | static inline void sched_init_smt(void) { } | |
7460 | #endif | |
7461 | ||
1da177e4 LT |
7462 | void __init sched_init_smp(void) |
7463 | { | |
dcc30a35 RR |
7464 | cpumask_var_t non_isolated_cpus; |
7465 | ||
7466 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7467 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7468 | |
cb83b629 PZ |
7469 | sched_init_numa(); |
7470 | ||
6acce3ef PZ |
7471 | /* |
7472 | * There's no userspace yet to cause hotplug operations; hence all the | |
7473 | * cpu masks are stable and all blatant races in the below code cannot | |
7474 | * happen. | |
7475 | */ | |
712555ee | 7476 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7477 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7478 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7479 | if (cpumask_empty(non_isolated_cpus)) | |
7480 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7481 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7482 | |
5c1e1767 | 7483 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 7484 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7485 | BUG(); |
19978ca6 | 7486 | sched_init_granularity(); |
dcc30a35 | 7487 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7488 | |
0e3900e6 | 7489 | init_sched_rt_class(); |
1baca4ce | 7490 | init_sched_dl_class(); |
1b568f0a PZ |
7491 | |
7492 | sched_init_smt(); | |
7493 | ||
e26fbffd | 7494 | sched_smp_initialized = true; |
1da177e4 | 7495 | } |
e26fbffd TG |
7496 | |
7497 | static int __init migration_init(void) | |
7498 | { | |
94baf7a5 | 7499 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 7500 | return 0; |
1da177e4 | 7501 | } |
e26fbffd TG |
7502 | early_initcall(migration_init); |
7503 | ||
1da177e4 LT |
7504 | #else |
7505 | void __init sched_init_smp(void) | |
7506 | { | |
19978ca6 | 7507 | sched_init_granularity(); |
1da177e4 LT |
7508 | } |
7509 | #endif /* CONFIG_SMP */ | |
7510 | ||
7511 | int in_sched_functions(unsigned long addr) | |
7512 | { | |
1da177e4 LT |
7513 | return in_lock_functions(addr) || |
7514 | (addr >= (unsigned long)__sched_text_start | |
7515 | && addr < (unsigned long)__sched_text_end); | |
7516 | } | |
7517 | ||
029632fb | 7518 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7519 | /* |
7520 | * Default task group. | |
7521 | * Every task in system belongs to this group at bootup. | |
7522 | */ | |
029632fb | 7523 | struct task_group root_task_group; |
35cf4e50 | 7524 | LIST_HEAD(task_groups); |
b0367629 WL |
7525 | |
7526 | /* Cacheline aligned slab cache for task_group */ | |
7527 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7528 | #endif |
6f505b16 | 7529 | |
e6252c3e | 7530 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7531 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7532 | |
9dcb8b68 LT |
7533 | #define WAIT_TABLE_BITS 8 |
7534 | #define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS) | |
7535 | static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned; | |
7536 | ||
7537 | wait_queue_head_t *bit_waitqueue(void *word, int bit) | |
7538 | { | |
7539 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; | |
7540 | unsigned long val = (unsigned long)word << shift | bit; | |
7541 | ||
7542 | return bit_wait_table + hash_long(val, WAIT_TABLE_BITS); | |
7543 | } | |
7544 | EXPORT_SYMBOL(bit_waitqueue); | |
7545 | ||
1da177e4 LT |
7546 | void __init sched_init(void) |
7547 | { | |
dd41f596 | 7548 | int i, j; |
434d53b0 MT |
7549 | unsigned long alloc_size = 0, ptr; |
7550 | ||
9dcb8b68 LT |
7551 | for (i = 0; i < WAIT_TABLE_SIZE; i++) |
7552 | init_waitqueue_head(bit_wait_table + i); | |
7553 | ||
434d53b0 MT |
7554 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7555 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7556 | #endif | |
7557 | #ifdef CONFIG_RT_GROUP_SCHED | |
7558 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7559 | #endif | |
434d53b0 | 7560 | if (alloc_size) { |
36b7b6d4 | 7561 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7562 | |
7563 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7564 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7565 | ptr += nr_cpu_ids * sizeof(void **); |
7566 | ||
07e06b01 | 7567 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7568 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7569 | |
6d6bc0ad | 7570 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7571 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7572 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7573 | ptr += nr_cpu_ids * sizeof(void **); |
7574 | ||
07e06b01 | 7575 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7576 | ptr += nr_cpu_ids * sizeof(void **); |
7577 | ||
6d6bc0ad | 7578 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7579 | } |
df7c8e84 | 7580 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7581 | for_each_possible_cpu(i) { |
7582 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7583 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7584 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7585 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7586 | } |
b74e6278 | 7587 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7588 | |
332ac17e DF |
7589 | init_rt_bandwidth(&def_rt_bandwidth, |
7590 | global_rt_period(), global_rt_runtime()); | |
7591 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7592 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7593 | |
57d885fe GH |
7594 | #ifdef CONFIG_SMP |
7595 | init_defrootdomain(); | |
7596 | #endif | |
7597 | ||
d0b27fa7 | 7598 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7599 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7600 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7601 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7602 | |
7c941438 | 7603 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7604 | task_group_cache = KMEM_CACHE(task_group, 0); |
7605 | ||
07e06b01 YZ |
7606 | list_add(&root_task_group.list, &task_groups); |
7607 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7608 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7609 | autogroup_init(&init_task); |
7c941438 | 7610 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7611 | |
0a945022 | 7612 | for_each_possible_cpu(i) { |
70b97a7f | 7613 | struct rq *rq; |
1da177e4 LT |
7614 | |
7615 | rq = cpu_rq(i); | |
05fa785c | 7616 | raw_spin_lock_init(&rq->lock); |
7897986b | 7617 | rq->nr_running = 0; |
dce48a84 TG |
7618 | rq->calc_load_active = 0; |
7619 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7620 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7621 | init_rt_rq(&rq->rt); |
7622 | init_dl_rq(&rq->dl); | |
dd41f596 | 7623 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7624 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7625 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7626 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7627 | /* |
07e06b01 | 7628 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7629 | * |
7630 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7631 | * gets 100% of the cpu resources in the system. This overall | |
7632 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7633 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7634 | * based on each entity's (task or task-group's) weight |
7635 | * (se->load.weight). | |
7636 | * | |
07e06b01 | 7637 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7638 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7639 | * then A0's share of the cpu resource is: | |
7640 | * | |
0d905bca | 7641 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7642 | * |
07e06b01 YZ |
7643 | * We achieve this by letting root_task_group's tasks sit |
7644 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7645 | */ |
ab84d31e | 7646 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7647 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7648 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7649 | ||
7650 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7651 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7652 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7653 | #endif |
1da177e4 | 7654 | |
dd41f596 IM |
7655 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7656 | rq->cpu_load[j] = 0; | |
fdf3e95d | 7657 | |
1da177e4 | 7658 | #ifdef CONFIG_SMP |
41c7ce9a | 7659 | rq->sd = NULL; |
57d885fe | 7660 | rq->rd = NULL; |
ca6d75e6 | 7661 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7662 | rq->balance_callback = NULL; |
1da177e4 | 7663 | rq->active_balance = 0; |
dd41f596 | 7664 | rq->next_balance = jiffies; |
1da177e4 | 7665 | rq->push_cpu = 0; |
0a2966b4 | 7666 | rq->cpu = i; |
1f11eb6a | 7667 | rq->online = 0; |
eae0c9df MG |
7668 | rq->idle_stamp = 0; |
7669 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7670 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7671 | |
7672 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7673 | ||
dc938520 | 7674 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7675 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 7676 | rq->last_load_update_tick = jiffies; |
1c792db7 | 7677 | rq->nohz_flags = 0; |
83cd4fe2 | 7678 | #endif |
265f22a9 FW |
7679 | #ifdef CONFIG_NO_HZ_FULL |
7680 | rq->last_sched_tick = 0; | |
7681 | #endif | |
9fd81dd5 | 7682 | #endif /* CONFIG_SMP */ |
8f4d37ec | 7683 | init_rq_hrtick(rq); |
1da177e4 | 7684 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7685 | } |
7686 | ||
2dd73a4f | 7687 | set_load_weight(&init_task); |
b50f60ce | 7688 | |
1da177e4 LT |
7689 | /* |
7690 | * The boot idle thread does lazy MMU switching as well: | |
7691 | */ | |
7692 | atomic_inc(&init_mm.mm_count); | |
7693 | enter_lazy_tlb(&init_mm, current); | |
7694 | ||
7695 | /* | |
7696 | * Make us the idle thread. Technically, schedule() should not be | |
7697 | * called from this thread, however somewhere below it might be, | |
7698 | * but because we are the idle thread, we just pick up running again | |
7699 | * when this runqueue becomes "idle". | |
7700 | */ | |
7701 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7702 | |
7703 | calc_load_update = jiffies + LOAD_FREQ; | |
7704 | ||
bf4d83f6 | 7705 | #ifdef CONFIG_SMP |
4cb98839 | 7706 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7707 | /* May be allocated at isolcpus cmdline parse time */ |
7708 | if (cpu_isolated_map == NULL) | |
7709 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7710 | idle_thread_set_boot_cpu(); |
9cf7243d | 7711 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
7712 | #endif |
7713 | init_sched_fair_class(); | |
6a7b3dc3 | 7714 | |
4698f88c JP |
7715 | init_schedstats(); |
7716 | ||
6892b75e | 7717 | scheduler_running = 1; |
1da177e4 LT |
7718 | } |
7719 | ||
d902db1e | 7720 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7721 | static inline int preempt_count_equals(int preempt_offset) |
7722 | { | |
da7142e2 | 7723 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7724 | |
4ba8216c | 7725 | return (nested == preempt_offset); |
e4aafea2 FW |
7726 | } |
7727 | ||
d894837f | 7728 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7729 | { |
8eb23b9f PZ |
7730 | /* |
7731 | * Blocking primitives will set (and therefore destroy) current->state, | |
7732 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7733 | * otherwise we will destroy state. | |
7734 | */ | |
00845eb9 | 7735 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7736 | "do not call blocking ops when !TASK_RUNNING; " |
7737 | "state=%lx set at [<%p>] %pS\n", | |
7738 | current->state, | |
7739 | (void *)current->task_state_change, | |
00845eb9 | 7740 | (void *)current->task_state_change); |
8eb23b9f | 7741 | |
3427445a PZ |
7742 | ___might_sleep(file, line, preempt_offset); |
7743 | } | |
7744 | EXPORT_SYMBOL(__might_sleep); | |
7745 | ||
7746 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7747 | { |
1da177e4 | 7748 | static unsigned long prev_jiffy; /* ratelimiting */ |
d1c6d149 | 7749 | unsigned long preempt_disable_ip; |
1da177e4 | 7750 | |
b3fbab05 | 7751 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7752 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7753 | !is_idle_task(current)) || | |
e4aafea2 | 7754 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7755 | return; |
7756 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7757 | return; | |
7758 | prev_jiffy = jiffies; | |
7759 | ||
d1c6d149 VN |
7760 | /* Save this before calling printk(), since that will clobber it */ |
7761 | preempt_disable_ip = get_preempt_disable_ip(current); | |
7762 | ||
3df0fc5b PZ |
7763 | printk(KERN_ERR |
7764 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7765 | file, line); | |
7766 | printk(KERN_ERR | |
7767 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7768 | in_atomic(), irqs_disabled(), | |
7769 | current->pid, current->comm); | |
aef745fc | 7770 | |
a8b686b3 ES |
7771 | if (task_stack_end_corrupted(current)) |
7772 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7773 | ||
aef745fc IM |
7774 | debug_show_held_locks(current); |
7775 | if (irqs_disabled()) | |
7776 | print_irqtrace_events(current); | |
d1c6d149 VN |
7777 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7778 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7779 | pr_err("Preemption disabled at:"); |
d1c6d149 | 7780 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
7781 | pr_cont("\n"); |
7782 | } | |
aef745fc | 7783 | dump_stack(); |
f0b22e39 | 7784 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7785 | } |
3427445a | 7786 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7787 | #endif |
7788 | ||
7789 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7790 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7791 | { |
dbc7f069 | 7792 | struct task_struct *g, *p; |
d50dde5a DF |
7793 | struct sched_attr attr = { |
7794 | .sched_policy = SCHED_NORMAL, | |
7795 | }; | |
1da177e4 | 7796 | |
3472eaa1 | 7797 | read_lock(&tasklist_lock); |
5d07f420 | 7798 | for_each_process_thread(g, p) { |
178be793 IM |
7799 | /* |
7800 | * Only normalize user tasks: | |
7801 | */ | |
3472eaa1 | 7802 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7803 | continue; |
7804 | ||
4fa8d299 JP |
7805 | p->se.exec_start = 0; |
7806 | schedstat_set(p->se.statistics.wait_start, 0); | |
7807 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7808 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7809 | |
aab03e05 | 7810 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7811 | /* |
7812 | * Renice negative nice level userspace | |
7813 | * tasks back to 0: | |
7814 | */ | |
3472eaa1 | 7815 | if (task_nice(p) < 0) |
dd41f596 | 7816 | set_user_nice(p, 0); |
1da177e4 | 7817 | continue; |
dd41f596 | 7818 | } |
1da177e4 | 7819 | |
dbc7f069 | 7820 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7821 | } |
3472eaa1 | 7822 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7823 | } |
7824 | ||
7825 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7826 | |
67fc4e0c | 7827 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7828 | /* |
67fc4e0c | 7829 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7830 | * |
7831 | * They can only be called when the whole system has been | |
7832 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7833 | * activity can take place. Using them for anything else would | |
7834 | * be a serious bug, and as a result, they aren't even visible | |
7835 | * under any other configuration. | |
7836 | */ | |
7837 | ||
7838 | /** | |
7839 | * curr_task - return the current task for a given cpu. | |
7840 | * @cpu: the processor in question. | |
7841 | * | |
7842 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7843 | * |
7844 | * Return: The current task for @cpu. | |
1df5c10a | 7845 | */ |
36c8b586 | 7846 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7847 | { |
7848 | return cpu_curr(cpu); | |
7849 | } | |
7850 | ||
67fc4e0c JW |
7851 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7852 | ||
7853 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7854 | /** |
7855 | * set_curr_task - set the current task for a given cpu. | |
7856 | * @cpu: the processor in question. | |
7857 | * @p: the task pointer to set. | |
7858 | * | |
7859 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7860 | * are serviced on a separate stack. It allows the architecture to switch the |
7861 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7862 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7863 | * and caller must save the original value of the current task (see | |
7864 | * curr_task() above) and restore that value before reenabling interrupts and | |
7865 | * re-starting the system. | |
7866 | * | |
7867 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7868 | */ | |
a458ae2e | 7869 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7870 | { |
7871 | cpu_curr(cpu) = p; | |
7872 | } | |
7873 | ||
7874 | #endif | |
29f59db3 | 7875 | |
7c941438 | 7876 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7877 | /* task_group_lock serializes the addition/removal of task groups */ |
7878 | static DEFINE_SPINLOCK(task_group_lock); | |
7879 | ||
2f5177f0 | 7880 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7881 | { |
7882 | free_fair_sched_group(tg); | |
7883 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7884 | autogroup_free(tg); |
b0367629 | 7885 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7886 | } |
7887 | ||
7888 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7889 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7890 | { |
7891 | struct task_group *tg; | |
bccbe08a | 7892 | |
b0367629 | 7893 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7894 | if (!tg) |
7895 | return ERR_PTR(-ENOMEM); | |
7896 | ||
ec7dc8ac | 7897 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7898 | goto err; |
7899 | ||
ec7dc8ac | 7900 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7901 | goto err; |
7902 | ||
ace783b9 LZ |
7903 | return tg; |
7904 | ||
7905 | err: | |
2f5177f0 | 7906 | sched_free_group(tg); |
ace783b9 LZ |
7907 | return ERR_PTR(-ENOMEM); |
7908 | } | |
7909 | ||
7910 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7911 | { | |
7912 | unsigned long flags; | |
7913 | ||
8ed36996 | 7914 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7915 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7916 | |
7917 | WARN_ON(!parent); /* root should already exist */ | |
7918 | ||
7919 | tg->parent = parent; | |
f473aa5e | 7920 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7921 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7922 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
7923 | |
7924 | online_fair_sched_group(tg); | |
29f59db3 SV |
7925 | } |
7926 | ||
9b5b7751 | 7927 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7928 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7929 | { |
29f59db3 | 7930 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 7931 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7932 | } |
7933 | ||
4cf86d77 | 7934 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7935 | { |
7936 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 7937 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7938 | } |
7939 | ||
7940 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7941 | { |
8ed36996 | 7942 | unsigned long flags; |
29f59db3 | 7943 | |
3d4b47b4 | 7944 | /* end participation in shares distribution */ |
6fe1f348 | 7945 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7946 | |
7947 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7948 | list_del_rcu(&tg->list); |
f473aa5e | 7949 | list_del_rcu(&tg->siblings); |
8ed36996 | 7950 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7951 | } |
7952 | ||
ea86cb4b | 7953 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 7954 | { |
8323f26c | 7955 | struct task_group *tg; |
29f59db3 | 7956 | |
f7b8a47d KT |
7957 | /* |
7958 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7959 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7960 | * to prevent lockdep warnings. | |
7961 | */ | |
7962 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7963 | struct task_group, css); |
7964 | tg = autogroup_task_group(tsk, tg); | |
7965 | tsk->sched_task_group = tg; | |
7966 | ||
810b3817 | 7967 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
7968 | if (tsk->sched_class->task_change_group) |
7969 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 7970 | else |
810b3817 | 7971 | #endif |
b2b5ce02 | 7972 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
7973 | } |
7974 | ||
7975 | /* | |
7976 | * Change task's runqueue when it moves between groups. | |
7977 | * | |
7978 | * The caller of this function should have put the task in its new group by | |
7979 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
7980 | * its new group. | |
7981 | */ | |
7982 | void sched_move_task(struct task_struct *tsk) | |
7983 | { | |
7984 | int queued, running; | |
7985 | struct rq_flags rf; | |
7986 | struct rq *rq; | |
7987 | ||
7988 | rq = task_rq_lock(tsk, &rf); | |
7989 | ||
7990 | running = task_current(rq, tsk); | |
7991 | queued = task_on_rq_queued(tsk); | |
7992 | ||
7993 | if (queued) | |
7994 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); | |
7995 | if (unlikely(running)) | |
7996 | put_prev_task(rq, tsk); | |
7997 | ||
7998 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 7999 | |
da0c1e65 | 8000 | if (queued) |
ff77e468 | 8001 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
a399d233 | 8002 | if (unlikely(running)) |
b2bf6c31 | 8003 | set_curr_task(rq, tsk); |
29f59db3 | 8004 | |
eb580751 | 8005 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8006 | } |
7c941438 | 8007 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8008 | |
a790de99 PT |
8009 | #ifdef CONFIG_RT_GROUP_SCHED |
8010 | /* | |
8011 | * Ensure that the real time constraints are schedulable. | |
8012 | */ | |
8013 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 8014 | |
9a7e0b18 PZ |
8015 | /* Must be called with tasklist_lock held */ |
8016 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8017 | { |
9a7e0b18 | 8018 | struct task_struct *g, *p; |
b40b2e8e | 8019 | |
1fe89e1b PZ |
8020 | /* |
8021 | * Autogroups do not have RT tasks; see autogroup_create(). | |
8022 | */ | |
8023 | if (task_group_is_autogroup(tg)) | |
8024 | return 0; | |
8025 | ||
5d07f420 | 8026 | for_each_process_thread(g, p) { |
8651c658 | 8027 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 8028 | return 1; |
5d07f420 | 8029 | } |
b40b2e8e | 8030 | |
9a7e0b18 PZ |
8031 | return 0; |
8032 | } | |
b40b2e8e | 8033 | |
9a7e0b18 PZ |
8034 | struct rt_schedulable_data { |
8035 | struct task_group *tg; | |
8036 | u64 rt_period; | |
8037 | u64 rt_runtime; | |
8038 | }; | |
b40b2e8e | 8039 | |
a790de99 | 8040 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
8041 | { |
8042 | struct rt_schedulable_data *d = data; | |
8043 | struct task_group *child; | |
8044 | unsigned long total, sum = 0; | |
8045 | u64 period, runtime; | |
b40b2e8e | 8046 | |
9a7e0b18 PZ |
8047 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8048 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8049 | |
9a7e0b18 PZ |
8050 | if (tg == d->tg) { |
8051 | period = d->rt_period; | |
8052 | runtime = d->rt_runtime; | |
b40b2e8e | 8053 | } |
b40b2e8e | 8054 | |
4653f803 PZ |
8055 | /* |
8056 | * Cannot have more runtime than the period. | |
8057 | */ | |
8058 | if (runtime > period && runtime != RUNTIME_INF) | |
8059 | return -EINVAL; | |
6f505b16 | 8060 | |
4653f803 PZ |
8061 | /* |
8062 | * Ensure we don't starve existing RT tasks. | |
8063 | */ | |
9a7e0b18 PZ |
8064 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8065 | return -EBUSY; | |
6f505b16 | 8066 | |
9a7e0b18 | 8067 | total = to_ratio(period, runtime); |
6f505b16 | 8068 | |
4653f803 PZ |
8069 | /* |
8070 | * Nobody can have more than the global setting allows. | |
8071 | */ | |
8072 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8073 | return -EINVAL; | |
6f505b16 | 8074 | |
4653f803 PZ |
8075 | /* |
8076 | * The sum of our children's runtime should not exceed our own. | |
8077 | */ | |
9a7e0b18 PZ |
8078 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8079 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8080 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8081 | |
9a7e0b18 PZ |
8082 | if (child == d->tg) { |
8083 | period = d->rt_period; | |
8084 | runtime = d->rt_runtime; | |
8085 | } | |
6f505b16 | 8086 | |
9a7e0b18 | 8087 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8088 | } |
6f505b16 | 8089 | |
9a7e0b18 PZ |
8090 | if (sum > total) |
8091 | return -EINVAL; | |
8092 | ||
8093 | return 0; | |
6f505b16 PZ |
8094 | } |
8095 | ||
9a7e0b18 | 8096 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8097 | { |
8277434e PT |
8098 | int ret; |
8099 | ||
9a7e0b18 PZ |
8100 | struct rt_schedulable_data data = { |
8101 | .tg = tg, | |
8102 | .rt_period = period, | |
8103 | .rt_runtime = runtime, | |
8104 | }; | |
8105 | ||
8277434e PT |
8106 | rcu_read_lock(); |
8107 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
8108 | rcu_read_unlock(); | |
8109 | ||
8110 | return ret; | |
521f1a24 DG |
8111 | } |
8112 | ||
ab84d31e | 8113 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 8114 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 8115 | { |
ac086bc2 | 8116 | int i, err = 0; |
9f0c1e56 | 8117 | |
2636ed5f PZ |
8118 | /* |
8119 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
8120 | * kernel creating (and or operating) RT threads. | |
8121 | */ | |
8122 | if (tg == &root_task_group && rt_runtime == 0) | |
8123 | return -EINVAL; | |
8124 | ||
8125 | /* No period doesn't make any sense. */ | |
8126 | if (rt_period == 0) | |
8127 | return -EINVAL; | |
8128 | ||
9f0c1e56 | 8129 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8130 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8131 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8132 | if (err) | |
9f0c1e56 | 8133 | goto unlock; |
ac086bc2 | 8134 | |
0986b11b | 8135 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8136 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8137 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8138 | |
8139 | for_each_possible_cpu(i) { | |
8140 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8141 | ||
0986b11b | 8142 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8143 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8144 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8145 | } |
0986b11b | 8146 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8147 | unlock: |
521f1a24 | 8148 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8149 | mutex_unlock(&rt_constraints_mutex); |
8150 | ||
8151 | return err; | |
6f505b16 PZ |
8152 | } |
8153 | ||
25cc7da7 | 8154 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
8155 | { |
8156 | u64 rt_runtime, rt_period; | |
8157 | ||
8158 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8159 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8160 | if (rt_runtime_us < 0) | |
8161 | rt_runtime = RUNTIME_INF; | |
8162 | ||
ab84d31e | 8163 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8164 | } |
8165 | ||
25cc7da7 | 8166 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
8167 | { |
8168 | u64 rt_runtime_us; | |
8169 | ||
d0b27fa7 | 8170 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8171 | return -1; |
8172 | ||
d0b27fa7 | 8173 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8174 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8175 | return rt_runtime_us; | |
8176 | } | |
d0b27fa7 | 8177 | |
ce2f5fe4 | 8178 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
8179 | { |
8180 | u64 rt_runtime, rt_period; | |
8181 | ||
ce2f5fe4 | 8182 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
8183 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
8184 | ||
ab84d31e | 8185 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8186 | } |
8187 | ||
25cc7da7 | 8188 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
8189 | { |
8190 | u64 rt_period_us; | |
8191 | ||
8192 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8193 | do_div(rt_period_us, NSEC_PER_USEC); | |
8194 | return rt_period_us; | |
8195 | } | |
332ac17e | 8196 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8197 | |
332ac17e | 8198 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
8199 | static int sched_rt_global_constraints(void) |
8200 | { | |
8201 | int ret = 0; | |
8202 | ||
8203 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 8204 | read_lock(&tasklist_lock); |
4653f803 | 8205 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8206 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8207 | mutex_unlock(&rt_constraints_mutex); |
8208 | ||
8209 | return ret; | |
8210 | } | |
54e99124 | 8211 | |
25cc7da7 | 8212 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
8213 | { |
8214 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8215 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8216 | return 0; | |
8217 | ||
8218 | return 1; | |
8219 | } | |
8220 | ||
6d6bc0ad | 8221 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8222 | static int sched_rt_global_constraints(void) |
8223 | { | |
ac086bc2 | 8224 | unsigned long flags; |
8c5e9554 | 8225 | int i; |
ec5d4989 | 8226 | |
0986b11b | 8227 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8228 | for_each_possible_cpu(i) { |
8229 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8230 | ||
0986b11b | 8231 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8232 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8233 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8234 | } |
0986b11b | 8235 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8236 | |
8c5e9554 | 8237 | return 0; |
d0b27fa7 | 8238 | } |
6d6bc0ad | 8239 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8240 | |
a1963b81 | 8241 | static int sched_dl_global_validate(void) |
332ac17e | 8242 | { |
1724813d PZ |
8243 | u64 runtime = global_rt_runtime(); |
8244 | u64 period = global_rt_period(); | |
332ac17e | 8245 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 8246 | struct dl_bw *dl_b; |
1724813d | 8247 | int cpu, ret = 0; |
49516342 | 8248 | unsigned long flags; |
332ac17e DF |
8249 | |
8250 | /* | |
8251 | * Here we want to check the bandwidth not being set to some | |
8252 | * value smaller than the currently allocated bandwidth in | |
8253 | * any of the root_domains. | |
8254 | * | |
8255 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
8256 | * cycling on root_domains... Discussion on different/better | |
8257 | * solutions is welcome! | |
8258 | */ | |
1724813d | 8259 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
8260 | rcu_read_lock_sched(); |
8261 | dl_b = dl_bw_of(cpu); | |
332ac17e | 8262 | |
49516342 | 8263 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
8264 | if (new_bw < dl_b->total_bw) |
8265 | ret = -EBUSY; | |
49516342 | 8266 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 8267 | |
f10e00f4 KT |
8268 | rcu_read_unlock_sched(); |
8269 | ||
1724813d PZ |
8270 | if (ret) |
8271 | break; | |
332ac17e DF |
8272 | } |
8273 | ||
1724813d | 8274 | return ret; |
332ac17e DF |
8275 | } |
8276 | ||
1724813d | 8277 | static void sched_dl_do_global(void) |
ce0dbbbb | 8278 | { |
1724813d | 8279 | u64 new_bw = -1; |
f10e00f4 | 8280 | struct dl_bw *dl_b; |
1724813d | 8281 | int cpu; |
49516342 | 8282 | unsigned long flags; |
ce0dbbbb | 8283 | |
1724813d PZ |
8284 | def_dl_bandwidth.dl_period = global_rt_period(); |
8285 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
8286 | ||
8287 | if (global_rt_runtime() != RUNTIME_INF) | |
8288 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
8289 | ||
8290 | /* | |
8291 | * FIXME: As above... | |
8292 | */ | |
8293 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
8294 | rcu_read_lock_sched(); |
8295 | dl_b = dl_bw_of(cpu); | |
1724813d | 8296 | |
49516342 | 8297 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8298 | dl_b->bw = new_bw; |
49516342 | 8299 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8300 | |
8301 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8302 | } |
1724813d PZ |
8303 | } |
8304 | ||
8305 | static int sched_rt_global_validate(void) | |
8306 | { | |
8307 | if (sysctl_sched_rt_period <= 0) | |
8308 | return -EINVAL; | |
8309 | ||
e9e7cb38 JL |
8310 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8311 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8312 | return -EINVAL; |
8313 | ||
8314 | return 0; | |
8315 | } | |
8316 | ||
8317 | static void sched_rt_do_global(void) | |
8318 | { | |
8319 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8320 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8321 | } |
8322 | ||
d0b27fa7 | 8323 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8324 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8325 | loff_t *ppos) |
8326 | { | |
d0b27fa7 PZ |
8327 | int old_period, old_runtime; |
8328 | static DEFINE_MUTEX(mutex); | |
1724813d | 8329 | int ret; |
d0b27fa7 PZ |
8330 | |
8331 | mutex_lock(&mutex); | |
8332 | old_period = sysctl_sched_rt_period; | |
8333 | old_runtime = sysctl_sched_rt_runtime; | |
8334 | ||
8d65af78 | 8335 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8336 | |
8337 | if (!ret && write) { | |
1724813d PZ |
8338 | ret = sched_rt_global_validate(); |
8339 | if (ret) | |
8340 | goto undo; | |
8341 | ||
a1963b81 | 8342 | ret = sched_dl_global_validate(); |
1724813d PZ |
8343 | if (ret) |
8344 | goto undo; | |
8345 | ||
a1963b81 | 8346 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8347 | if (ret) |
8348 | goto undo; | |
8349 | ||
8350 | sched_rt_do_global(); | |
8351 | sched_dl_do_global(); | |
8352 | } | |
8353 | if (0) { | |
8354 | undo: | |
8355 | sysctl_sched_rt_period = old_period; | |
8356 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8357 | } |
8358 | mutex_unlock(&mutex); | |
8359 | ||
8360 | return ret; | |
8361 | } | |
68318b8e | 8362 | |
1724813d | 8363 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8364 | void __user *buffer, size_t *lenp, |
8365 | loff_t *ppos) | |
8366 | { | |
8367 | int ret; | |
332ac17e | 8368 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8369 | |
8370 | mutex_lock(&mutex); | |
332ac17e | 8371 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8372 | /* make sure that internally we keep jiffies */ |
8373 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8374 | if (!ret && write) { |
1724813d PZ |
8375 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8376 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8377 | } |
8378 | mutex_unlock(&mutex); | |
332ac17e DF |
8379 | return ret; |
8380 | } | |
8381 | ||
052f1dc7 | 8382 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8383 | |
a7c6d554 | 8384 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8385 | { |
a7c6d554 | 8386 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8387 | } |
8388 | ||
eb95419b TH |
8389 | static struct cgroup_subsys_state * |
8390 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8391 | { |
eb95419b TH |
8392 | struct task_group *parent = css_tg(parent_css); |
8393 | struct task_group *tg; | |
68318b8e | 8394 | |
eb95419b | 8395 | if (!parent) { |
68318b8e | 8396 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8397 | return &root_task_group.css; |
68318b8e SV |
8398 | } |
8399 | ||
ec7dc8ac | 8400 | tg = sched_create_group(parent); |
68318b8e SV |
8401 | if (IS_ERR(tg)) |
8402 | return ERR_PTR(-ENOMEM); | |
8403 | ||
2f5177f0 PZ |
8404 | sched_online_group(tg, parent); |
8405 | ||
68318b8e SV |
8406 | return &tg->css; |
8407 | } | |
8408 | ||
2f5177f0 | 8409 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8410 | { |
eb95419b | 8411 | struct task_group *tg = css_tg(css); |
ace783b9 | 8412 | |
2f5177f0 | 8413 | sched_offline_group(tg); |
ace783b9 LZ |
8414 | } |
8415 | ||
eb95419b | 8416 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8417 | { |
eb95419b | 8418 | struct task_group *tg = css_tg(css); |
68318b8e | 8419 | |
2f5177f0 PZ |
8420 | /* |
8421 | * Relies on the RCU grace period between css_released() and this. | |
8422 | */ | |
8423 | sched_free_group(tg); | |
ace783b9 LZ |
8424 | } |
8425 | ||
ea86cb4b VG |
8426 | /* |
8427 | * This is called before wake_up_new_task(), therefore we really only | |
8428 | * have to set its group bits, all the other stuff does not apply. | |
8429 | */ | |
b53202e6 | 8430 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8431 | { |
ea86cb4b VG |
8432 | struct rq_flags rf; |
8433 | struct rq *rq; | |
8434 | ||
8435 | rq = task_rq_lock(task, &rf); | |
8436 | ||
8437 | sched_change_group(task, TASK_SET_GROUP); | |
8438 | ||
8439 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8440 | } |
8441 | ||
1f7dd3e5 | 8442 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8443 | { |
bb9d97b6 | 8444 | struct task_struct *task; |
1f7dd3e5 | 8445 | struct cgroup_subsys_state *css; |
7dc603c9 | 8446 | int ret = 0; |
bb9d97b6 | 8447 | |
1f7dd3e5 | 8448 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8449 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8450 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8451 | return -EINVAL; |
b68aa230 | 8452 | #else |
bb9d97b6 TH |
8453 | /* We don't support RT-tasks being in separate groups */ |
8454 | if (task->sched_class != &fair_sched_class) | |
8455 | return -EINVAL; | |
b68aa230 | 8456 | #endif |
7dc603c9 PZ |
8457 | /* |
8458 | * Serialize against wake_up_new_task() such that if its | |
8459 | * running, we're sure to observe its full state. | |
8460 | */ | |
8461 | raw_spin_lock_irq(&task->pi_lock); | |
8462 | /* | |
8463 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8464 | * has happened. This would lead to problems with PELT, due to | |
8465 | * move wanting to detach+attach while we're not attached yet. | |
8466 | */ | |
8467 | if (task->state == TASK_NEW) | |
8468 | ret = -EINVAL; | |
8469 | raw_spin_unlock_irq(&task->pi_lock); | |
8470 | ||
8471 | if (ret) | |
8472 | break; | |
bb9d97b6 | 8473 | } |
7dc603c9 | 8474 | return ret; |
be367d09 | 8475 | } |
68318b8e | 8476 | |
1f7dd3e5 | 8477 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8478 | { |
bb9d97b6 | 8479 | struct task_struct *task; |
1f7dd3e5 | 8480 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8481 | |
1f7dd3e5 | 8482 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8483 | sched_move_task(task); |
68318b8e SV |
8484 | } |
8485 | ||
052f1dc7 | 8486 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8487 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8488 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8489 | { |
182446d0 | 8490 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8491 | } |
8492 | ||
182446d0 TH |
8493 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8494 | struct cftype *cft) | |
68318b8e | 8495 | { |
182446d0 | 8496 | struct task_group *tg = css_tg(css); |
68318b8e | 8497 | |
c8b28116 | 8498 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8499 | } |
ab84d31e PT |
8500 | |
8501 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8502 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8503 | ||
ab84d31e PT |
8504 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8505 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8506 | ||
a790de99 PT |
8507 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8508 | ||
ab84d31e PT |
8509 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8510 | { | |
56f570e5 | 8511 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8512 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8513 | |
8514 | if (tg == &root_task_group) | |
8515 | return -EINVAL; | |
8516 | ||
8517 | /* | |
8518 | * Ensure we have at some amount of bandwidth every period. This is | |
8519 | * to prevent reaching a state of large arrears when throttled via | |
8520 | * entity_tick() resulting in prolonged exit starvation. | |
8521 | */ | |
8522 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8523 | return -EINVAL; | |
8524 | ||
8525 | /* | |
8526 | * Likewise, bound things on the otherside by preventing insane quota | |
8527 | * periods. This also allows us to normalize in computing quota | |
8528 | * feasibility. | |
8529 | */ | |
8530 | if (period > max_cfs_quota_period) | |
8531 | return -EINVAL; | |
8532 | ||
0e59bdae KT |
8533 | /* |
8534 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8535 | * unthrottle_offline_cfs_rqs(). | |
8536 | */ | |
8537 | get_online_cpus(); | |
a790de99 PT |
8538 | mutex_lock(&cfs_constraints_mutex); |
8539 | ret = __cfs_schedulable(tg, period, quota); | |
8540 | if (ret) | |
8541 | goto out_unlock; | |
8542 | ||
58088ad0 | 8543 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8544 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8545 | /* |
8546 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8547 | * before making related changes, and on->off must occur afterwards | |
8548 | */ | |
8549 | if (runtime_enabled && !runtime_was_enabled) | |
8550 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8551 | raw_spin_lock_irq(&cfs_b->lock); |
8552 | cfs_b->period = ns_to_ktime(period); | |
8553 | cfs_b->quota = quota; | |
58088ad0 | 8554 | |
a9cf55b2 | 8555 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8556 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8557 | if (runtime_enabled) |
8558 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8559 | raw_spin_unlock_irq(&cfs_b->lock); |
8560 | ||
0e59bdae | 8561 | for_each_online_cpu(i) { |
ab84d31e | 8562 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8563 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8564 | |
8565 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8566 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8567 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8568 | |
029632fb | 8569 | if (cfs_rq->throttled) |
671fd9da | 8570 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8571 | raw_spin_unlock_irq(&rq->lock); |
8572 | } | |
1ee14e6c BS |
8573 | if (runtime_was_enabled && !runtime_enabled) |
8574 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8575 | out_unlock: |
8576 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8577 | put_online_cpus(); |
ab84d31e | 8578 | |
a790de99 | 8579 | return ret; |
ab84d31e PT |
8580 | } |
8581 | ||
8582 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8583 | { | |
8584 | u64 quota, period; | |
8585 | ||
029632fb | 8586 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8587 | if (cfs_quota_us < 0) |
8588 | quota = RUNTIME_INF; | |
8589 | else | |
8590 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8591 | ||
8592 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8593 | } | |
8594 | ||
8595 | long tg_get_cfs_quota(struct task_group *tg) | |
8596 | { | |
8597 | u64 quota_us; | |
8598 | ||
029632fb | 8599 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8600 | return -1; |
8601 | ||
029632fb | 8602 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8603 | do_div(quota_us, NSEC_PER_USEC); |
8604 | ||
8605 | return quota_us; | |
8606 | } | |
8607 | ||
8608 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8609 | { | |
8610 | u64 quota, period; | |
8611 | ||
8612 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8613 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8614 | |
ab84d31e PT |
8615 | return tg_set_cfs_bandwidth(tg, period, quota); |
8616 | } | |
8617 | ||
8618 | long tg_get_cfs_period(struct task_group *tg) | |
8619 | { | |
8620 | u64 cfs_period_us; | |
8621 | ||
029632fb | 8622 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8623 | do_div(cfs_period_us, NSEC_PER_USEC); |
8624 | ||
8625 | return cfs_period_us; | |
8626 | } | |
8627 | ||
182446d0 TH |
8628 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8629 | struct cftype *cft) | |
ab84d31e | 8630 | { |
182446d0 | 8631 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8632 | } |
8633 | ||
182446d0 TH |
8634 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8635 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8636 | { |
182446d0 | 8637 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8638 | } |
8639 | ||
182446d0 TH |
8640 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8641 | struct cftype *cft) | |
ab84d31e | 8642 | { |
182446d0 | 8643 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8644 | } |
8645 | ||
182446d0 TH |
8646 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8647 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8648 | { |
182446d0 | 8649 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8650 | } |
8651 | ||
a790de99 PT |
8652 | struct cfs_schedulable_data { |
8653 | struct task_group *tg; | |
8654 | u64 period, quota; | |
8655 | }; | |
8656 | ||
8657 | /* | |
8658 | * normalize group quota/period to be quota/max_period | |
8659 | * note: units are usecs | |
8660 | */ | |
8661 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8662 | struct cfs_schedulable_data *d) | |
8663 | { | |
8664 | u64 quota, period; | |
8665 | ||
8666 | if (tg == d->tg) { | |
8667 | period = d->period; | |
8668 | quota = d->quota; | |
8669 | } else { | |
8670 | period = tg_get_cfs_period(tg); | |
8671 | quota = tg_get_cfs_quota(tg); | |
8672 | } | |
8673 | ||
8674 | /* note: these should typically be equivalent */ | |
8675 | if (quota == RUNTIME_INF || quota == -1) | |
8676 | return RUNTIME_INF; | |
8677 | ||
8678 | return to_ratio(period, quota); | |
8679 | } | |
8680 | ||
8681 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8682 | { | |
8683 | struct cfs_schedulable_data *d = data; | |
029632fb | 8684 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8685 | s64 quota = 0, parent_quota = -1; |
8686 | ||
8687 | if (!tg->parent) { | |
8688 | quota = RUNTIME_INF; | |
8689 | } else { | |
029632fb | 8690 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8691 | |
8692 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8693 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8694 | |
8695 | /* | |
8696 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8697 | * limit is set | |
8698 | */ | |
8699 | if (quota == RUNTIME_INF) | |
8700 | quota = parent_quota; | |
8701 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8702 | return -EINVAL; | |
8703 | } | |
9c58c79a | 8704 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8705 | |
8706 | return 0; | |
8707 | } | |
8708 | ||
8709 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8710 | { | |
8277434e | 8711 | int ret; |
a790de99 PT |
8712 | struct cfs_schedulable_data data = { |
8713 | .tg = tg, | |
8714 | .period = period, | |
8715 | .quota = quota, | |
8716 | }; | |
8717 | ||
8718 | if (quota != RUNTIME_INF) { | |
8719 | do_div(data.period, NSEC_PER_USEC); | |
8720 | do_div(data.quota, NSEC_PER_USEC); | |
8721 | } | |
8722 | ||
8277434e PT |
8723 | rcu_read_lock(); |
8724 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8725 | rcu_read_unlock(); | |
8726 | ||
8727 | return ret; | |
a790de99 | 8728 | } |
e8da1b18 | 8729 | |
2da8ca82 | 8730 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8731 | { |
2da8ca82 | 8732 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8733 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8734 | |
44ffc75b TH |
8735 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8736 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8737 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8738 | |
8739 | return 0; | |
8740 | } | |
ab84d31e | 8741 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8742 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8743 | |
052f1dc7 | 8744 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8745 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8746 | struct cftype *cft, s64 val) | |
6f505b16 | 8747 | { |
182446d0 | 8748 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8749 | } |
8750 | ||
182446d0 TH |
8751 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8752 | struct cftype *cft) | |
6f505b16 | 8753 | { |
182446d0 | 8754 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8755 | } |
d0b27fa7 | 8756 | |
182446d0 TH |
8757 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8758 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8759 | { |
182446d0 | 8760 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8761 | } |
8762 | ||
182446d0 TH |
8763 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8764 | struct cftype *cft) | |
d0b27fa7 | 8765 | { |
182446d0 | 8766 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8767 | } |
6d6bc0ad | 8768 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8769 | |
fe5c7cc2 | 8770 | static struct cftype cpu_files[] = { |
052f1dc7 | 8771 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8772 | { |
8773 | .name = "shares", | |
f4c753b7 PM |
8774 | .read_u64 = cpu_shares_read_u64, |
8775 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8776 | }, |
052f1dc7 | 8777 | #endif |
ab84d31e PT |
8778 | #ifdef CONFIG_CFS_BANDWIDTH |
8779 | { | |
8780 | .name = "cfs_quota_us", | |
8781 | .read_s64 = cpu_cfs_quota_read_s64, | |
8782 | .write_s64 = cpu_cfs_quota_write_s64, | |
8783 | }, | |
8784 | { | |
8785 | .name = "cfs_period_us", | |
8786 | .read_u64 = cpu_cfs_period_read_u64, | |
8787 | .write_u64 = cpu_cfs_period_write_u64, | |
8788 | }, | |
e8da1b18 NR |
8789 | { |
8790 | .name = "stat", | |
2da8ca82 | 8791 | .seq_show = cpu_stats_show, |
e8da1b18 | 8792 | }, |
ab84d31e | 8793 | #endif |
052f1dc7 | 8794 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8795 | { |
9f0c1e56 | 8796 | .name = "rt_runtime_us", |
06ecb27c PM |
8797 | .read_s64 = cpu_rt_runtime_read, |
8798 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8799 | }, |
d0b27fa7 PZ |
8800 | { |
8801 | .name = "rt_period_us", | |
f4c753b7 PM |
8802 | .read_u64 = cpu_rt_period_read_uint, |
8803 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8804 | }, |
052f1dc7 | 8805 | #endif |
4baf6e33 | 8806 | { } /* terminate */ |
68318b8e SV |
8807 | }; |
8808 | ||
073219e9 | 8809 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8810 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8811 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8812 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8813 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8814 | .can_attach = cpu_cgroup_can_attach, |
8815 | .attach = cpu_cgroup_attach, | |
5577964e | 8816 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8817 | .early_init = true, |
68318b8e SV |
8818 | }; |
8819 | ||
052f1dc7 | 8820 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8821 | |
b637a328 PM |
8822 | void dump_cpu_task(int cpu) |
8823 | { | |
8824 | pr_info("Task dump for CPU %d:\n", cpu); | |
8825 | sched_show_task(cpu_curr(cpu)); | |
8826 | } | |
ed82b8a1 AK |
8827 | |
8828 | /* | |
8829 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8830 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8831 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8832 | * that remained on nice 0. | |
8833 | * | |
8834 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8835 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8836 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8837 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8838 | * the relative distance between them is ~25%.) | |
8839 | */ | |
8840 | const int sched_prio_to_weight[40] = { | |
8841 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8842 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8843 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8844 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8845 | /* 0 */ 1024, 820, 655, 526, 423, | |
8846 | /* 5 */ 335, 272, 215, 172, 137, | |
8847 | /* 10 */ 110, 87, 70, 56, 45, | |
8848 | /* 15 */ 36, 29, 23, 18, 15, | |
8849 | }; | |
8850 | ||
8851 | /* | |
8852 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8853 | * | |
8854 | * In cases where the weight does not change often, we can use the | |
8855 | * precalculated inverse to speed up arithmetics by turning divisions | |
8856 | * into multiplications: | |
8857 | */ | |
8858 | const u32 sched_prio_to_wmult[40] = { | |
8859 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8860 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8861 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8862 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8863 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8864 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8865 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8866 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8867 | }; |