]>
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 | ||
cb42c9a3 | 105 | if (rq->clock_update_flags & RQCF_ACT_SKIP) |
f26f9aff | 106 | return; |
aa483808 | 107 | |
cb42c9a3 MF |
108 | #ifdef CONFIG_SCHED_DEBUG |
109 | rq->clock_update_flags |= RQCF_UPDATED; | |
110 | #endif | |
fe44d621 | 111 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
112 | if (delta < 0) |
113 | return; | |
fe44d621 PZ |
114 | rq->clock += delta; |
115 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
116 | } |
117 | ||
bf5c91ba IM |
118 | /* |
119 | * Debugging: various feature bits | |
120 | */ | |
f00b45c1 | 121 | |
f00b45c1 PZ |
122 | #define SCHED_FEAT(name, enabled) \ |
123 | (1UL << __SCHED_FEAT_##name) * enabled | | |
124 | ||
bf5c91ba | 125 | const_debug unsigned int sysctl_sched_features = |
391e43da | 126 | #include "features.h" |
f00b45c1 PZ |
127 | 0; |
128 | ||
129 | #undef SCHED_FEAT | |
130 | ||
b82d9fdd PZ |
131 | /* |
132 | * Number of tasks to iterate in a single balance run. | |
133 | * Limited because this is done with IRQs disabled. | |
134 | */ | |
135 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
136 | ||
e9e9250b PZ |
137 | /* |
138 | * period over which we average the RT time consumption, measured | |
139 | * in ms. | |
140 | * | |
141 | * default: 1s | |
142 | */ | |
143 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
144 | ||
fa85ae24 | 145 | /* |
9f0c1e56 | 146 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
147 | * default: 1s |
148 | */ | |
9f0c1e56 | 149 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 150 | |
029632fb | 151 | __read_mostly int scheduler_running; |
6892b75e | 152 | |
9f0c1e56 PZ |
153 | /* |
154 | * part of the period that we allow rt tasks to run in us. | |
155 | * default: 0.95s | |
156 | */ | |
157 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 158 | |
3fa0818b RR |
159 | /* cpus with isolated domains */ |
160 | cpumask_var_t cpu_isolated_map; | |
161 | ||
1da177e4 | 162 | /* |
cc2a73b5 | 163 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 164 | */ |
a9957449 | 165 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
166 | __acquires(rq->lock) |
167 | { | |
70b97a7f | 168 | struct rq *rq; |
1da177e4 LT |
169 | |
170 | local_irq_disable(); | |
171 | rq = this_rq(); | |
05fa785c | 172 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
173 | |
174 | return rq; | |
175 | } | |
176 | ||
3e71a462 PZ |
177 | /* |
178 | * __task_rq_lock - lock the rq @p resides on. | |
179 | */ | |
eb580751 | 180 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
181 | __acquires(rq->lock) |
182 | { | |
183 | struct rq *rq; | |
184 | ||
185 | lockdep_assert_held(&p->pi_lock); | |
186 | ||
187 | for (;;) { | |
188 | rq = task_rq(p); | |
189 | raw_spin_lock(&rq->lock); | |
190 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 191 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
192 | return rq; |
193 | } | |
194 | raw_spin_unlock(&rq->lock); | |
195 | ||
196 | while (unlikely(task_on_rq_migrating(p))) | |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
203 | */ | |
eb580751 | 204 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
205 | __acquires(p->pi_lock) |
206 | __acquires(rq->lock) | |
207 | { | |
208 | struct rq *rq; | |
209 | ||
210 | for (;;) { | |
eb580751 | 211 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
212 | rq = task_rq(p); |
213 | raw_spin_lock(&rq->lock); | |
214 | /* | |
215 | * move_queued_task() task_rq_lock() | |
216 | * | |
217 | * ACQUIRE (rq->lock) | |
218 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
219 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
220 | * [S] ->cpu = new_cpu [L] task_rq() | |
221 | * [L] ->on_rq | |
222 | * RELEASE (rq->lock) | |
223 | * | |
224 | * If we observe the old cpu in task_rq_lock, the acquire of | |
225 | * the old rq->lock will fully serialize against the stores. | |
226 | * | |
227 | * If we observe the new cpu in task_rq_lock, the acquire will | |
228 | * pair with the WMB to ensure we must then also see migrating. | |
229 | */ | |
230 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 231 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
232 | return rq; |
233 | } | |
234 | raw_spin_unlock(&rq->lock); | |
eb580751 | 235 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
236 | |
237 | while (unlikely(task_on_rq_migrating(p))) | |
238 | cpu_relax(); | |
239 | } | |
240 | } | |
241 | ||
8f4d37ec PZ |
242 | #ifdef CONFIG_SCHED_HRTICK |
243 | /* | |
244 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 245 | */ |
8f4d37ec | 246 | |
8f4d37ec PZ |
247 | static void hrtick_clear(struct rq *rq) |
248 | { | |
249 | if (hrtimer_active(&rq->hrtick_timer)) | |
250 | hrtimer_cancel(&rq->hrtick_timer); | |
251 | } | |
252 | ||
8f4d37ec PZ |
253 | /* |
254 | * High-resolution timer tick. | |
255 | * Runs from hardirq context with interrupts disabled. | |
256 | */ | |
257 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
258 | { | |
259 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
260 | ||
261 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
262 | ||
05fa785c | 263 | raw_spin_lock(&rq->lock); |
3e51f33f | 264 | update_rq_clock(rq); |
8f4d37ec | 265 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 266 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
267 | |
268 | return HRTIMER_NORESTART; | |
269 | } | |
270 | ||
95e904c7 | 271 | #ifdef CONFIG_SMP |
971ee28c | 272 | |
4961b6e1 | 273 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
274 | { |
275 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 276 | |
4961b6e1 | 277 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
278 | } |
279 | ||
31656519 PZ |
280 | /* |
281 | * called from hardirq (IPI) context | |
282 | */ | |
283 | static void __hrtick_start(void *arg) | |
b328ca18 | 284 | { |
31656519 | 285 | struct rq *rq = arg; |
b328ca18 | 286 | |
05fa785c | 287 | raw_spin_lock(&rq->lock); |
971ee28c | 288 | __hrtick_restart(rq); |
31656519 | 289 | rq->hrtick_csd_pending = 0; |
05fa785c | 290 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
291 | } |
292 | ||
31656519 PZ |
293 | /* |
294 | * Called to set the hrtick timer state. | |
295 | * | |
296 | * called with rq->lock held and irqs disabled | |
297 | */ | |
029632fb | 298 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 299 | { |
31656519 | 300 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 301 | ktime_t time; |
302 | s64 delta; | |
303 | ||
304 | /* | |
305 | * Don't schedule slices shorter than 10000ns, that just | |
306 | * doesn't make sense and can cause timer DoS. | |
307 | */ | |
308 | delta = max_t(s64, delay, 10000LL); | |
309 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 310 | |
cc584b21 | 311 | hrtimer_set_expires(timer, time); |
31656519 PZ |
312 | |
313 | if (rq == this_rq()) { | |
971ee28c | 314 | __hrtick_restart(rq); |
31656519 | 315 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 316 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
317 | rq->hrtick_csd_pending = 1; |
318 | } | |
b328ca18 PZ |
319 | } |
320 | ||
31656519 PZ |
321 | #else |
322 | /* | |
323 | * Called to set the hrtick timer state. | |
324 | * | |
325 | * called with rq->lock held and irqs disabled | |
326 | */ | |
029632fb | 327 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 328 | { |
86893335 WL |
329 | /* |
330 | * Don't schedule slices shorter than 10000ns, that just | |
331 | * doesn't make sense. Rely on vruntime for fairness. | |
332 | */ | |
333 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
334 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
335 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 336 | } |
31656519 | 337 | #endif /* CONFIG_SMP */ |
8f4d37ec | 338 | |
31656519 | 339 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 340 | { |
31656519 PZ |
341 | #ifdef CONFIG_SMP |
342 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 343 | |
31656519 PZ |
344 | rq->hrtick_csd.flags = 0; |
345 | rq->hrtick_csd.func = __hrtick_start; | |
346 | rq->hrtick_csd.info = rq; | |
347 | #endif | |
8f4d37ec | 348 | |
31656519 PZ |
349 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
350 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 351 | } |
006c75f1 | 352 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
353 | static inline void hrtick_clear(struct rq *rq) |
354 | { | |
355 | } | |
356 | ||
8f4d37ec PZ |
357 | static inline void init_rq_hrtick(struct rq *rq) |
358 | { | |
359 | } | |
006c75f1 | 360 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 361 | |
5529578a FW |
362 | /* |
363 | * cmpxchg based fetch_or, macro so it works for different integer types | |
364 | */ | |
365 | #define fetch_or(ptr, mask) \ | |
366 | ({ \ | |
367 | typeof(ptr) _ptr = (ptr); \ | |
368 | typeof(mask) _mask = (mask); \ | |
369 | typeof(*_ptr) _old, _val = *_ptr; \ | |
370 | \ | |
371 | for (;;) { \ | |
372 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
373 | if (_old == _val) \ | |
374 | break; \ | |
375 | _val = _old; \ | |
376 | } \ | |
377 | _old; \ | |
378 | }) | |
379 | ||
e3baac47 | 380 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
381 | /* |
382 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
383 | * this avoids any races wrt polling state changes and thereby avoids | |
384 | * spurious IPIs. | |
385 | */ | |
386 | static bool set_nr_and_not_polling(struct task_struct *p) | |
387 | { | |
388 | struct thread_info *ti = task_thread_info(p); | |
389 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
390 | } | |
e3baac47 PZ |
391 | |
392 | /* | |
393 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
394 | * | |
395 | * If this returns true, then the idle task promises to call | |
396 | * sched_ttwu_pending() and reschedule soon. | |
397 | */ | |
398 | static bool set_nr_if_polling(struct task_struct *p) | |
399 | { | |
400 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 401 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
402 | |
403 | for (;;) { | |
404 | if (!(val & _TIF_POLLING_NRFLAG)) | |
405 | return false; | |
406 | if (val & _TIF_NEED_RESCHED) | |
407 | return true; | |
408 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
409 | if (old == val) | |
410 | break; | |
411 | val = old; | |
412 | } | |
413 | return true; | |
414 | } | |
415 | ||
fd99f91a PZ |
416 | #else |
417 | static bool set_nr_and_not_polling(struct task_struct *p) | |
418 | { | |
419 | set_tsk_need_resched(p); | |
420 | return true; | |
421 | } | |
e3baac47 PZ |
422 | |
423 | #ifdef CONFIG_SMP | |
424 | static bool set_nr_if_polling(struct task_struct *p) | |
425 | { | |
426 | return false; | |
427 | } | |
428 | #endif | |
fd99f91a PZ |
429 | #endif |
430 | ||
76751049 PZ |
431 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
432 | { | |
433 | struct wake_q_node *node = &task->wake_q; | |
434 | ||
435 | /* | |
436 | * Atomically grab the task, if ->wake_q is !nil already it means | |
437 | * its already queued (either by us or someone else) and will get the | |
438 | * wakeup due to that. | |
439 | * | |
440 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 441 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
442 | */ |
443 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
444 | return; | |
445 | ||
446 | get_task_struct(task); | |
447 | ||
448 | /* | |
449 | * The head is context local, there can be no concurrency. | |
450 | */ | |
451 | *head->lastp = node; | |
452 | head->lastp = &node->next; | |
453 | } | |
454 | ||
455 | void wake_up_q(struct wake_q_head *head) | |
456 | { | |
457 | struct wake_q_node *node = head->first; | |
458 | ||
459 | while (node != WAKE_Q_TAIL) { | |
460 | struct task_struct *task; | |
461 | ||
462 | task = container_of(node, struct task_struct, wake_q); | |
463 | BUG_ON(!task); | |
464 | /* task can safely be re-inserted now */ | |
465 | node = node->next; | |
466 | task->wake_q.next = NULL; | |
467 | ||
468 | /* | |
469 | * wake_up_process() implies a wmb() to pair with the queueing | |
470 | * in wake_q_add() so as not to miss wakeups. | |
471 | */ | |
472 | wake_up_process(task); | |
473 | put_task_struct(task); | |
474 | } | |
475 | } | |
476 | ||
c24d20db | 477 | /* |
8875125e | 478 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
479 | * |
480 | * On UP this means the setting of the need_resched flag, on SMP it | |
481 | * might also involve a cross-CPU call to trigger the scheduler on | |
482 | * the target CPU. | |
483 | */ | |
8875125e | 484 | void resched_curr(struct rq *rq) |
c24d20db | 485 | { |
8875125e | 486 | struct task_struct *curr = rq->curr; |
c24d20db IM |
487 | int cpu; |
488 | ||
8875125e | 489 | lockdep_assert_held(&rq->lock); |
c24d20db | 490 | |
8875125e | 491 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
492 | return; |
493 | ||
8875125e | 494 | cpu = cpu_of(rq); |
fd99f91a | 495 | |
f27dde8d | 496 | if (cpu == smp_processor_id()) { |
8875125e | 497 | set_tsk_need_resched(curr); |
f27dde8d | 498 | set_preempt_need_resched(); |
c24d20db | 499 | return; |
f27dde8d | 500 | } |
c24d20db | 501 | |
8875125e | 502 | if (set_nr_and_not_polling(curr)) |
c24d20db | 503 | smp_send_reschedule(cpu); |
dfc68f29 AL |
504 | else |
505 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
506 | } |
507 | ||
029632fb | 508 | void resched_cpu(int cpu) |
c24d20db IM |
509 | { |
510 | struct rq *rq = cpu_rq(cpu); | |
511 | unsigned long flags; | |
512 | ||
05fa785c | 513 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 514 | return; |
8875125e | 515 | resched_curr(rq); |
05fa785c | 516 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 517 | } |
06d8308c | 518 | |
b021fe3e | 519 | #ifdef CONFIG_SMP |
3451d024 | 520 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
521 | /* |
522 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
523 | * from an idle cpu. This is good for power-savings. | |
524 | * | |
525 | * We don't do similar optimization for completely idle system, as | |
526 | * selecting an idle cpu will add more delays to the timers than intended | |
527 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
528 | */ | |
bc7a34b8 | 529 | int get_nohz_timer_target(void) |
83cd4fe2 | 530 | { |
bc7a34b8 | 531 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
532 | struct sched_domain *sd; |
533 | ||
9642d18e | 534 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
535 | return cpu; |
536 | ||
057f3fad | 537 | rcu_read_lock(); |
83cd4fe2 | 538 | for_each_domain(cpu, sd) { |
057f3fad | 539 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
540 | if (cpu == i) |
541 | continue; | |
542 | ||
543 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
544 | cpu = i; |
545 | goto unlock; | |
546 | } | |
547 | } | |
83cd4fe2 | 548 | } |
9642d18e VH |
549 | |
550 | if (!is_housekeeping_cpu(cpu)) | |
551 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
552 | unlock: |
553 | rcu_read_unlock(); | |
83cd4fe2 VP |
554 | return cpu; |
555 | } | |
06d8308c TG |
556 | /* |
557 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
558 | * idle CPU then this timer might expire before the next timer event | |
559 | * which is scheduled to wake up that CPU. In case of a completely | |
560 | * idle system the next event might even be infinite time into the | |
561 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
562 | * leaves the inner idle loop so the newly added timer is taken into | |
563 | * account when the CPU goes back to idle and evaluates the timer | |
564 | * wheel for the next timer event. | |
565 | */ | |
1c20091e | 566 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
567 | { |
568 | struct rq *rq = cpu_rq(cpu); | |
569 | ||
570 | if (cpu == smp_processor_id()) | |
571 | return; | |
572 | ||
67b9ca70 | 573 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 574 | smp_send_reschedule(cpu); |
dfc68f29 AL |
575 | else |
576 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
577 | } |
578 | ||
c5bfece2 | 579 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 580 | { |
53c5fa16 FW |
581 | /* |
582 | * We just need the target to call irq_exit() and re-evaluate | |
583 | * the next tick. The nohz full kick at least implies that. | |
584 | * If needed we can still optimize that later with an | |
585 | * empty IRQ. | |
586 | */ | |
379d9ecb PM |
587 | if (cpu_is_offline(cpu)) |
588 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 589 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
590 | if (cpu != smp_processor_id() || |
591 | tick_nohz_tick_stopped()) | |
53c5fa16 | 592 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
593 | return true; |
594 | } | |
595 | ||
596 | return false; | |
597 | } | |
598 | ||
379d9ecb PM |
599 | /* |
600 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
601 | * caller's responsibility to deal with the lost wakeup, for example, | |
602 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
603 | */ | |
1c20091e FW |
604 | void wake_up_nohz_cpu(int cpu) |
605 | { | |
c5bfece2 | 606 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
607 | wake_up_idle_cpu(cpu); |
608 | } | |
609 | ||
ca38062e | 610 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 611 | { |
1c792db7 | 612 | int cpu = smp_processor_id(); |
873b4c65 VG |
613 | |
614 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
615 | return false; | |
616 | ||
617 | if (idle_cpu(cpu) && !need_resched()) | |
618 | return true; | |
619 | ||
620 | /* | |
621 | * We can't run Idle Load Balance on this CPU for this time so we | |
622 | * cancel it and clear NOHZ_BALANCE_KICK | |
623 | */ | |
624 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
625 | return false; | |
45bf76df IM |
626 | } |
627 | ||
3451d024 | 628 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 629 | |
ca38062e | 630 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 631 | { |
ca38062e | 632 | return false; |
2069dd75 PZ |
633 | } |
634 | ||
3451d024 | 635 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 636 | |
ce831b38 | 637 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 638 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 639 | { |
76d92ac3 FW |
640 | int fifo_nr_running; |
641 | ||
642 | /* Deadline tasks, even if single, need the tick */ | |
643 | if (rq->dl.dl_nr_running) | |
644 | return false; | |
645 | ||
1e78cdbd | 646 | /* |
2548d546 PZ |
647 | * If there are more than one RR tasks, we need the tick to effect the |
648 | * actual RR behaviour. | |
1e78cdbd | 649 | */ |
76d92ac3 FW |
650 | if (rq->rt.rr_nr_running) { |
651 | if (rq->rt.rr_nr_running == 1) | |
652 | return true; | |
653 | else | |
654 | return false; | |
1e78cdbd RR |
655 | } |
656 | ||
2548d546 PZ |
657 | /* |
658 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
659 | * forced preemption between FIFO tasks. | |
660 | */ | |
661 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
662 | if (fifo_nr_running) | |
663 | return true; | |
664 | ||
665 | /* | |
666 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
667 | * if there's more than one we need the tick for involuntary | |
668 | * preemption. | |
669 | */ | |
670 | if (rq->nr_running > 1) | |
541b8264 | 671 | return false; |
ce831b38 | 672 | |
541b8264 | 673 | return true; |
ce831b38 FW |
674 | } |
675 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 676 | |
029632fb | 677 | void sched_avg_update(struct rq *rq) |
18d95a28 | 678 | { |
e9e9250b PZ |
679 | s64 period = sched_avg_period(); |
680 | ||
78becc27 | 681 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
682 | /* |
683 | * Inline assembly required to prevent the compiler | |
684 | * optimising this loop into a divmod call. | |
685 | * See __iter_div_u64_rem() for another example of this. | |
686 | */ | |
687 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
688 | rq->age_stamp += period; |
689 | rq->rt_avg /= 2; | |
690 | } | |
18d95a28 PZ |
691 | } |
692 | ||
6d6bc0ad | 693 | #endif /* CONFIG_SMP */ |
18d95a28 | 694 | |
a790de99 PT |
695 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
696 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 697 | /* |
8277434e PT |
698 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
699 | * node and @up when leaving it for the final time. | |
700 | * | |
701 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 702 | */ |
029632fb | 703 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 704 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
705 | { |
706 | struct task_group *parent, *child; | |
eb755805 | 707 | int ret; |
c09595f6 | 708 | |
8277434e PT |
709 | parent = from; |
710 | ||
c09595f6 | 711 | down: |
eb755805 PZ |
712 | ret = (*down)(parent, data); |
713 | if (ret) | |
8277434e | 714 | goto out; |
c09595f6 PZ |
715 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
716 | parent = child; | |
717 | goto down; | |
718 | ||
719 | up: | |
720 | continue; | |
721 | } | |
eb755805 | 722 | ret = (*up)(parent, data); |
8277434e PT |
723 | if (ret || parent == from) |
724 | goto out; | |
c09595f6 PZ |
725 | |
726 | child = parent; | |
727 | parent = parent->parent; | |
728 | if (parent) | |
729 | goto up; | |
8277434e | 730 | out: |
eb755805 | 731 | return ret; |
c09595f6 PZ |
732 | } |
733 | ||
029632fb | 734 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 735 | { |
e2b245f8 | 736 | return 0; |
eb755805 | 737 | } |
18d95a28 PZ |
738 | #endif |
739 | ||
45bf76df IM |
740 | static void set_load_weight(struct task_struct *p) |
741 | { | |
f05998d4 NR |
742 | int prio = p->static_prio - MAX_RT_PRIO; |
743 | struct load_weight *load = &p->se.load; | |
744 | ||
dd41f596 IM |
745 | /* |
746 | * SCHED_IDLE tasks get minimal weight: | |
747 | */ | |
20f9cd2a | 748 | if (idle_policy(p->policy)) { |
c8b28116 | 749 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 750 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
751 | return; |
752 | } | |
71f8bd46 | 753 | |
ed82b8a1 AK |
754 | load->weight = scale_load(sched_prio_to_weight[prio]); |
755 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
756 | } |
757 | ||
1de64443 | 758 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 759 | { |
a64692a3 | 760 | update_rq_clock(rq); |
1de64443 PZ |
761 | if (!(flags & ENQUEUE_RESTORE)) |
762 | sched_info_queued(rq, p); | |
371fd7e7 | 763 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
764 | } |
765 | ||
1de64443 | 766 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 767 | { |
a64692a3 | 768 | update_rq_clock(rq); |
1de64443 PZ |
769 | if (!(flags & DEQUEUE_SAVE)) |
770 | sched_info_dequeued(rq, p); | |
371fd7e7 | 771 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
772 | } |
773 | ||
029632fb | 774 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
775 | { |
776 | if (task_contributes_to_load(p)) | |
777 | rq->nr_uninterruptible--; | |
778 | ||
371fd7e7 | 779 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
780 | } |
781 | ||
029632fb | 782 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
783 | { |
784 | if (task_contributes_to_load(p)) | |
785 | rq->nr_uninterruptible++; | |
786 | ||
371fd7e7 | 787 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
788 | } |
789 | ||
fe44d621 | 790 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 791 | { |
095c0aa8 GC |
792 | /* |
793 | * In theory, the compile should just see 0 here, and optimize out the call | |
794 | * to sched_rt_avg_update. But I don't trust it... | |
795 | */ | |
796 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
797 | s64 steal = 0, irq_delta = 0; | |
798 | #endif | |
799 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 800 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
801 | |
802 | /* | |
803 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
804 | * this case when a previous update_rq_clock() happened inside a | |
805 | * {soft,}irq region. | |
806 | * | |
807 | * When this happens, we stop ->clock_task and only update the | |
808 | * prev_irq_time stamp to account for the part that fit, so that a next | |
809 | * update will consume the rest. This ensures ->clock_task is | |
810 | * monotonic. | |
811 | * | |
812 | * It does however cause some slight miss-attribution of {soft,}irq | |
813 | * time, a more accurate solution would be to update the irq_time using | |
814 | * the current rq->clock timestamp, except that would require using | |
815 | * atomic ops. | |
816 | */ | |
817 | if (irq_delta > delta) | |
818 | irq_delta = delta; | |
819 | ||
820 | rq->prev_irq_time += irq_delta; | |
821 | delta -= irq_delta; | |
095c0aa8 GC |
822 | #endif |
823 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 824 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
825 | steal = paravirt_steal_clock(cpu_of(rq)); |
826 | steal -= rq->prev_steal_time_rq; | |
827 | ||
828 | if (unlikely(steal > delta)) | |
829 | steal = delta; | |
830 | ||
095c0aa8 | 831 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
832 | delta -= steal; |
833 | } | |
834 | #endif | |
835 | ||
fe44d621 PZ |
836 | rq->clock_task += delta; |
837 | ||
095c0aa8 | 838 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 839 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
840 | sched_rt_avg_update(rq, irq_delta + steal); |
841 | #endif | |
aa483808 VP |
842 | } |
843 | ||
34f971f6 PZ |
844 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
845 | { | |
846 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
847 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
848 | ||
849 | if (stop) { | |
850 | /* | |
851 | * Make it appear like a SCHED_FIFO task, its something | |
852 | * userspace knows about and won't get confused about. | |
853 | * | |
854 | * Also, it will make PI more or less work without too | |
855 | * much confusion -- but then, stop work should not | |
856 | * rely on PI working anyway. | |
857 | */ | |
858 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
859 | ||
860 | stop->sched_class = &stop_sched_class; | |
861 | } | |
862 | ||
863 | cpu_rq(cpu)->stop = stop; | |
864 | ||
865 | if (old_stop) { | |
866 | /* | |
867 | * Reset it back to a normal scheduling class so that | |
868 | * it can die in pieces. | |
869 | */ | |
870 | old_stop->sched_class = &rt_sched_class; | |
871 | } | |
872 | } | |
873 | ||
14531189 | 874 | /* |
dd41f596 | 875 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 876 | */ |
14531189 IM |
877 | static inline int __normal_prio(struct task_struct *p) |
878 | { | |
dd41f596 | 879 | return p->static_prio; |
14531189 IM |
880 | } |
881 | ||
b29739f9 IM |
882 | /* |
883 | * Calculate the expected normal priority: i.e. priority | |
884 | * without taking RT-inheritance into account. Might be | |
885 | * boosted by interactivity modifiers. Changes upon fork, | |
886 | * setprio syscalls, and whenever the interactivity | |
887 | * estimator recalculates. | |
888 | */ | |
36c8b586 | 889 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
890 | { |
891 | int prio; | |
892 | ||
aab03e05 DF |
893 | if (task_has_dl_policy(p)) |
894 | prio = MAX_DL_PRIO-1; | |
895 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
896 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
897 | else | |
898 | prio = __normal_prio(p); | |
899 | return prio; | |
900 | } | |
901 | ||
902 | /* | |
903 | * Calculate the current priority, i.e. the priority | |
904 | * taken into account by the scheduler. This value might | |
905 | * be boosted by RT tasks, or might be boosted by | |
906 | * interactivity modifiers. Will be RT if the task got | |
907 | * RT-boosted. If not then it returns p->normal_prio. | |
908 | */ | |
36c8b586 | 909 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
910 | { |
911 | p->normal_prio = normal_prio(p); | |
912 | /* | |
913 | * If we are RT tasks or we were boosted to RT priority, | |
914 | * keep the priority unchanged. Otherwise, update priority | |
915 | * to the normal priority: | |
916 | */ | |
917 | if (!rt_prio(p->prio)) | |
918 | return p->normal_prio; | |
919 | return p->prio; | |
920 | } | |
921 | ||
1da177e4 LT |
922 | /** |
923 | * task_curr - is this task currently executing on a CPU? | |
924 | * @p: the task in question. | |
e69f6186 YB |
925 | * |
926 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 927 | */ |
36c8b586 | 928 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
929 | { |
930 | return cpu_curr(task_cpu(p)) == p; | |
931 | } | |
932 | ||
67dfa1b7 | 933 | /* |
4c9a4bc8 PZ |
934 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
935 | * use the balance_callback list if you want balancing. | |
936 | * | |
937 | * this means any call to check_class_changed() must be followed by a call to | |
938 | * balance_callback(). | |
67dfa1b7 | 939 | */ |
cb469845 SR |
940 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
941 | const struct sched_class *prev_class, | |
da7a735e | 942 | int oldprio) |
cb469845 SR |
943 | { |
944 | if (prev_class != p->sched_class) { | |
945 | if (prev_class->switched_from) | |
da7a735e | 946 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 947 | |
da7a735e | 948 | p->sched_class->switched_to(rq, p); |
2d3d891d | 949 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 950 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
951 | } |
952 | ||
029632fb | 953 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
954 | { |
955 | const struct sched_class *class; | |
956 | ||
957 | if (p->sched_class == rq->curr->sched_class) { | |
958 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
959 | } else { | |
960 | for_each_class(class) { | |
961 | if (class == rq->curr->sched_class) | |
962 | break; | |
963 | if (class == p->sched_class) { | |
8875125e | 964 | resched_curr(rq); |
1e5a7405 PZ |
965 | break; |
966 | } | |
967 | } | |
968 | } | |
969 | ||
970 | /* | |
971 | * A queue event has occurred, and we're going to schedule. In | |
972 | * this case, we can save a useless back to back clock update. | |
973 | */ | |
da0c1e65 | 974 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 975 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
976 | } |
977 | ||
1da177e4 | 978 | #ifdef CONFIG_SMP |
5cc389bc PZ |
979 | /* |
980 | * This is how migration works: | |
981 | * | |
982 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
983 | * stop_one_cpu(). | |
984 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
985 | * off the CPU) | |
986 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
987 | * 4) if it's in the wrong runqueue then the migration thread removes | |
988 | * it and puts it into the right queue. | |
989 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
990 | * is done. | |
991 | */ | |
992 | ||
993 | /* | |
994 | * move_queued_task - move a queued task to new rq. | |
995 | * | |
996 | * Returns (locked) new rq. Old rq's lock is released. | |
997 | */ | |
5e16bbc2 | 998 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 999 | { |
5cc389bc PZ |
1000 | lockdep_assert_held(&rq->lock); |
1001 | ||
5cc389bc | 1002 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 1003 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
1004 | set_task_cpu(p, new_cpu); |
1005 | raw_spin_unlock(&rq->lock); | |
1006 | ||
1007 | rq = cpu_rq(new_cpu); | |
1008 | ||
1009 | raw_spin_lock(&rq->lock); | |
1010 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 1011 | enqueue_task(rq, p, 0); |
3ea94de1 | 1012 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
1013 | check_preempt_curr(rq, p, 0); |
1014 | ||
1015 | return rq; | |
1016 | } | |
1017 | ||
1018 | struct migration_arg { | |
1019 | struct task_struct *task; | |
1020 | int dest_cpu; | |
1021 | }; | |
1022 | ||
1023 | /* | |
1024 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1025 | * this because either it can't run here any more (set_cpus_allowed() | |
1026 | * away from this CPU, or CPU going down), or because we're | |
1027 | * attempting to rebalance this task on exec (sched_exec). | |
1028 | * | |
1029 | * So we race with normal scheduler movements, but that's OK, as long | |
1030 | * as the task is no longer on this CPU. | |
5cc389bc | 1031 | */ |
5e16bbc2 | 1032 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1033 | { |
5cc389bc | 1034 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1035 | return rq; |
5cc389bc PZ |
1036 | |
1037 | /* Affinity changed (again). */ | |
1038 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1039 | return rq; |
5cc389bc | 1040 | |
5e16bbc2 PZ |
1041 | rq = move_queued_task(rq, p, dest_cpu); |
1042 | ||
1043 | return rq; | |
5cc389bc PZ |
1044 | } |
1045 | ||
1046 | /* | |
1047 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1048 | * and performs thread migration by bumping thread off CPU then | |
1049 | * 'pushing' onto another runqueue. | |
1050 | */ | |
1051 | static int migration_cpu_stop(void *data) | |
1052 | { | |
1053 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1054 | struct task_struct *p = arg->task; |
1055 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1056 | |
1057 | /* | |
1058 | * The original target cpu might have gone down and we might | |
1059 | * be on another cpu but it doesn't matter. | |
1060 | */ | |
1061 | local_irq_disable(); | |
1062 | /* | |
1063 | * We need to explicitly wake pending tasks before running | |
1064 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1065 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1066 | */ | |
1067 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1068 | |
1069 | raw_spin_lock(&p->pi_lock); | |
1070 | raw_spin_lock(&rq->lock); | |
1071 | /* | |
1072 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1073 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1074 | * we're holding p->pi_lock. | |
1075 | */ | |
bf89a304 CC |
1076 | if (task_rq(p) == rq) { |
1077 | if (task_on_rq_queued(p)) | |
1078 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1079 | else | |
1080 | p->wake_cpu = arg->dest_cpu; | |
1081 | } | |
5e16bbc2 PZ |
1082 | raw_spin_unlock(&rq->lock); |
1083 | raw_spin_unlock(&p->pi_lock); | |
1084 | ||
5cc389bc PZ |
1085 | local_irq_enable(); |
1086 | return 0; | |
1087 | } | |
1088 | ||
c5b28038 PZ |
1089 | /* |
1090 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1091 | * actually call this function. | |
1092 | */ | |
1093 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1094 | { |
5cc389bc PZ |
1095 | cpumask_copy(&p->cpus_allowed, new_mask); |
1096 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1097 | } | |
1098 | ||
c5b28038 PZ |
1099 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1100 | { | |
6c37067e PZ |
1101 | struct rq *rq = task_rq(p); |
1102 | bool queued, running; | |
1103 | ||
c5b28038 | 1104 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1105 | |
1106 | queued = task_on_rq_queued(p); | |
1107 | running = task_current(rq, p); | |
1108 | ||
1109 | if (queued) { | |
1110 | /* | |
1111 | * Because __kthread_bind() calls this on blocked tasks without | |
1112 | * holding rq->lock. | |
1113 | */ | |
1114 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1115 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1116 | } |
1117 | if (running) | |
1118 | put_prev_task(rq, p); | |
1119 | ||
c5b28038 | 1120 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1121 | |
6c37067e | 1122 | if (queued) |
1de64443 | 1123 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 1124 | if (running) |
b2bf6c31 | 1125 | set_curr_task(rq, p); |
c5b28038 PZ |
1126 | } |
1127 | ||
5cc389bc PZ |
1128 | /* |
1129 | * Change a given task's CPU affinity. Migrate the thread to a | |
1130 | * proper CPU and schedule it away if the CPU it's executing on | |
1131 | * is removed from the allowed bitmask. | |
1132 | * | |
1133 | * NOTE: the caller must have a valid reference to the task, the | |
1134 | * task must not exit() & deallocate itself prematurely. The | |
1135 | * call is not atomic; no spinlocks may be held. | |
1136 | */ | |
25834c73 PZ |
1137 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1138 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1139 | { |
e9d867a6 | 1140 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1141 | unsigned int dest_cpu; |
eb580751 PZ |
1142 | struct rq_flags rf; |
1143 | struct rq *rq; | |
5cc389bc PZ |
1144 | int ret = 0; |
1145 | ||
eb580751 | 1146 | rq = task_rq_lock(p, &rf); |
5cc389bc | 1147 | |
e9d867a6 PZI |
1148 | if (p->flags & PF_KTHREAD) { |
1149 | /* | |
1150 | * Kernel threads are allowed on online && !active CPUs | |
1151 | */ | |
1152 | cpu_valid_mask = cpu_online_mask; | |
1153 | } | |
1154 | ||
25834c73 PZ |
1155 | /* |
1156 | * Must re-check here, to close a race against __kthread_bind(), | |
1157 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1158 | */ | |
1159 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1160 | ret = -EINVAL; | |
1161 | goto out; | |
1162 | } | |
1163 | ||
5cc389bc PZ |
1164 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1165 | goto out; | |
1166 | ||
e9d867a6 | 1167 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1168 | ret = -EINVAL; |
1169 | goto out; | |
1170 | } | |
1171 | ||
1172 | do_set_cpus_allowed(p, new_mask); | |
1173 | ||
e9d867a6 PZI |
1174 | if (p->flags & PF_KTHREAD) { |
1175 | /* | |
1176 | * For kernel threads that do indeed end up on online && | |
1177 | * !active we want to ensure they are strict per-cpu threads. | |
1178 | */ | |
1179 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1180 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1181 | p->nr_cpus_allowed != 1); | |
1182 | } | |
1183 | ||
5cc389bc PZ |
1184 | /* Can the task run on the task's current CPU? If so, we're done */ |
1185 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1186 | goto out; | |
1187 | ||
e9d867a6 | 1188 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1189 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1190 | struct migration_arg arg = { p, dest_cpu }; | |
1191 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1192 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1193 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1194 | tlb_migrate_finish(p->mm); | |
1195 | return 0; | |
cbce1a68 PZ |
1196 | } else if (task_on_rq_queued(p)) { |
1197 | /* | |
1198 | * OK, since we're going to drop the lock immediately | |
1199 | * afterwards anyway. | |
1200 | */ | |
d8ac8971 | 1201 | rq_unpin_lock(rq, &rf); |
5e16bbc2 | 1202 | rq = move_queued_task(rq, p, dest_cpu); |
d8ac8971 | 1203 | rq_repin_lock(rq, &rf); |
cbce1a68 | 1204 | } |
5cc389bc | 1205 | out: |
eb580751 | 1206 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1207 | |
1208 | return ret; | |
1209 | } | |
25834c73 PZ |
1210 | |
1211 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1212 | { | |
1213 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1214 | } | |
5cc389bc PZ |
1215 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1216 | ||
dd41f596 | 1217 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1218 | { |
e2912009 PZ |
1219 | #ifdef CONFIG_SCHED_DEBUG |
1220 | /* | |
1221 | * We should never call set_task_cpu() on a blocked task, | |
1222 | * ttwu() will sort out the placement. | |
1223 | */ | |
077614ee | 1224 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1225 | !p->on_rq); |
0122ec5b | 1226 | |
3ea94de1 JP |
1227 | /* |
1228 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1229 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1230 | * time relying on p->on_rq. | |
1231 | */ | |
1232 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1233 | p->sched_class == &fair_sched_class && | |
1234 | (p->on_rq && !task_on_rq_migrating(p))); | |
1235 | ||
0122ec5b | 1236 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1237 | /* |
1238 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1239 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1240 | * | |
1241 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1242 | * see task_group(). |
6c6c54e1 PZ |
1243 | * |
1244 | * Furthermore, all task_rq users should acquire both locks, see | |
1245 | * task_rq_lock(). | |
1246 | */ | |
0122ec5b PZ |
1247 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1248 | lockdep_is_held(&task_rq(p)->lock))); | |
1249 | #endif | |
e2912009 PZ |
1250 | #endif |
1251 | ||
de1d7286 | 1252 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1253 | |
0c69774e | 1254 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1255 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1256 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1257 | p->se.nr_migrations++; |
ff303e66 | 1258 | perf_event_task_migrate(p); |
0c69774e | 1259 | } |
dd41f596 IM |
1260 | |
1261 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1262 | } |
1263 | ||
ac66f547 PZ |
1264 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1265 | { | |
da0c1e65 | 1266 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1267 | struct rq *src_rq, *dst_rq; |
1268 | ||
1269 | src_rq = task_rq(p); | |
1270 | dst_rq = cpu_rq(cpu); | |
1271 | ||
3ea94de1 | 1272 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1273 | deactivate_task(src_rq, p, 0); |
1274 | set_task_cpu(p, cpu); | |
1275 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1276 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1277 | check_preempt_curr(dst_rq, p, 0); |
1278 | } else { | |
1279 | /* | |
1280 | * Task isn't running anymore; make it appear like we migrated | |
1281 | * it before it went to sleep. This means on wakeup we make the | |
a1fd4656 | 1282 | * previous cpu our target instead of where it really is. |
ac66f547 PZ |
1283 | */ |
1284 | p->wake_cpu = cpu; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | struct migration_swap_arg { | |
1289 | struct task_struct *src_task, *dst_task; | |
1290 | int src_cpu, dst_cpu; | |
1291 | }; | |
1292 | ||
1293 | static int migrate_swap_stop(void *data) | |
1294 | { | |
1295 | struct migration_swap_arg *arg = data; | |
1296 | struct rq *src_rq, *dst_rq; | |
1297 | int ret = -EAGAIN; | |
1298 | ||
62694cd5 PZ |
1299 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1300 | return -EAGAIN; | |
1301 | ||
ac66f547 PZ |
1302 | src_rq = cpu_rq(arg->src_cpu); |
1303 | dst_rq = cpu_rq(arg->dst_cpu); | |
1304 | ||
74602315 PZ |
1305 | double_raw_lock(&arg->src_task->pi_lock, |
1306 | &arg->dst_task->pi_lock); | |
ac66f547 | 1307 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1308 | |
ac66f547 PZ |
1309 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1310 | goto unlock; | |
1311 | ||
1312 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1313 | goto unlock; | |
1314 | ||
1315 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1316 | goto unlock; | |
1317 | ||
1318 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1319 | goto unlock; | |
1320 | ||
1321 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1322 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1323 | ||
1324 | ret = 0; | |
1325 | ||
1326 | unlock: | |
1327 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1328 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1329 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1330 | |
1331 | return ret; | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Cross migrate two tasks | |
1336 | */ | |
1337 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1338 | { | |
1339 | struct migration_swap_arg arg; | |
1340 | int ret = -EINVAL; | |
1341 | ||
ac66f547 PZ |
1342 | arg = (struct migration_swap_arg){ |
1343 | .src_task = cur, | |
1344 | .src_cpu = task_cpu(cur), | |
1345 | .dst_task = p, | |
1346 | .dst_cpu = task_cpu(p), | |
1347 | }; | |
1348 | ||
1349 | if (arg.src_cpu == arg.dst_cpu) | |
1350 | goto out; | |
1351 | ||
6acce3ef PZ |
1352 | /* |
1353 | * These three tests are all lockless; this is OK since all of them | |
1354 | * will be re-checked with proper locks held further down the line. | |
1355 | */ | |
ac66f547 PZ |
1356 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1357 | goto out; | |
1358 | ||
1359 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1360 | goto out; | |
1361 | ||
1362 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1363 | goto out; | |
1364 | ||
286549dc | 1365 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1366 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1367 | ||
1368 | out: | |
ac66f547 PZ |
1369 | return ret; |
1370 | } | |
1371 | ||
1da177e4 LT |
1372 | /* |
1373 | * wait_task_inactive - wait for a thread to unschedule. | |
1374 | * | |
85ba2d86 RM |
1375 | * If @match_state is nonzero, it's the @p->state value just checked and |
1376 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1377 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1378 | * we return a positive number (its total switch count). If a second call | |
1379 | * a short while later returns the same number, the caller can be sure that | |
1380 | * @p has remained unscheduled the whole time. | |
1381 | * | |
1da177e4 LT |
1382 | * The caller must ensure that the task *will* unschedule sometime soon, |
1383 | * else this function might spin for a *long* time. This function can't | |
1384 | * be called with interrupts off, or it may introduce deadlock with | |
1385 | * smp_call_function() if an IPI is sent by the same process we are | |
1386 | * waiting to become inactive. | |
1387 | */ | |
85ba2d86 | 1388 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1389 | { |
da0c1e65 | 1390 | int running, queued; |
eb580751 | 1391 | struct rq_flags rf; |
85ba2d86 | 1392 | unsigned long ncsw; |
70b97a7f | 1393 | struct rq *rq; |
1da177e4 | 1394 | |
3a5c359a AK |
1395 | for (;;) { |
1396 | /* | |
1397 | * We do the initial early heuristics without holding | |
1398 | * any task-queue locks at all. We'll only try to get | |
1399 | * the runqueue lock when things look like they will | |
1400 | * work out! | |
1401 | */ | |
1402 | rq = task_rq(p); | |
fa490cfd | 1403 | |
3a5c359a AK |
1404 | /* |
1405 | * If the task is actively running on another CPU | |
1406 | * still, just relax and busy-wait without holding | |
1407 | * any locks. | |
1408 | * | |
1409 | * NOTE! Since we don't hold any locks, it's not | |
1410 | * even sure that "rq" stays as the right runqueue! | |
1411 | * But we don't care, since "task_running()" will | |
1412 | * return false if the runqueue has changed and p | |
1413 | * is actually now running somewhere else! | |
1414 | */ | |
85ba2d86 RM |
1415 | while (task_running(rq, p)) { |
1416 | if (match_state && unlikely(p->state != match_state)) | |
1417 | return 0; | |
3a5c359a | 1418 | cpu_relax(); |
85ba2d86 | 1419 | } |
fa490cfd | 1420 | |
3a5c359a AK |
1421 | /* |
1422 | * Ok, time to look more closely! We need the rq | |
1423 | * lock now, to be *sure*. If we're wrong, we'll | |
1424 | * just go back and repeat. | |
1425 | */ | |
eb580751 | 1426 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1427 | trace_sched_wait_task(p); |
3a5c359a | 1428 | running = task_running(rq, p); |
da0c1e65 | 1429 | queued = task_on_rq_queued(p); |
85ba2d86 | 1430 | ncsw = 0; |
f31e11d8 | 1431 | if (!match_state || p->state == match_state) |
93dcf55f | 1432 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1433 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1434 | |
85ba2d86 RM |
1435 | /* |
1436 | * If it changed from the expected state, bail out now. | |
1437 | */ | |
1438 | if (unlikely(!ncsw)) | |
1439 | break; | |
1440 | ||
3a5c359a AK |
1441 | /* |
1442 | * Was it really running after all now that we | |
1443 | * checked with the proper locks actually held? | |
1444 | * | |
1445 | * Oops. Go back and try again.. | |
1446 | */ | |
1447 | if (unlikely(running)) { | |
1448 | cpu_relax(); | |
1449 | continue; | |
1450 | } | |
fa490cfd | 1451 | |
3a5c359a AK |
1452 | /* |
1453 | * It's not enough that it's not actively running, | |
1454 | * it must be off the runqueue _entirely_, and not | |
1455 | * preempted! | |
1456 | * | |
80dd99b3 | 1457 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1458 | * running right now), it's preempted, and we should |
1459 | * yield - it could be a while. | |
1460 | */ | |
da0c1e65 | 1461 | if (unlikely(queued)) { |
8b0e1953 | 1462 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1463 | |
1464 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1465 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1466 | continue; |
1467 | } | |
fa490cfd | 1468 | |
3a5c359a AK |
1469 | /* |
1470 | * Ahh, all good. It wasn't running, and it wasn't | |
1471 | * runnable, which means that it will never become | |
1472 | * running in the future either. We're all done! | |
1473 | */ | |
1474 | break; | |
1475 | } | |
85ba2d86 RM |
1476 | |
1477 | return ncsw; | |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | /*** | |
1481 | * kick_process - kick a running thread to enter/exit the kernel | |
1482 | * @p: the to-be-kicked thread | |
1483 | * | |
1484 | * Cause a process which is running on another CPU to enter | |
1485 | * kernel-mode, without any delay. (to get signals handled.) | |
1486 | * | |
25985edc | 1487 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1488 | * because all it wants to ensure is that the remote task enters |
1489 | * the kernel. If the IPI races and the task has been migrated | |
1490 | * to another CPU then no harm is done and the purpose has been | |
1491 | * achieved as well. | |
1492 | */ | |
36c8b586 | 1493 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1494 | { |
1495 | int cpu; | |
1496 | ||
1497 | preempt_disable(); | |
1498 | cpu = task_cpu(p); | |
1499 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1500 | smp_send_reschedule(cpu); | |
1501 | preempt_enable(); | |
1502 | } | |
b43e3521 | 1503 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1504 | |
30da688e | 1505 | /* |
013fdb80 | 1506 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1507 | * |
1508 | * A few notes on cpu_active vs cpu_online: | |
1509 | * | |
1510 | * - cpu_active must be a subset of cpu_online | |
1511 | * | |
1512 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1513 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
1514 | * cpu isn't yet part of the sched domains, and balancing will not | |
1515 | * see it. | |
1516 | * | |
1517 | * - on cpu-down we clear cpu_active() to mask the sched domains and | |
1518 | * avoid the load balancer to place new tasks on the to be removed | |
1519 | * cpu. Existing tasks will remain running there and will be taken | |
1520 | * off. | |
1521 | * | |
1522 | * This means that fallback selection must not select !active CPUs. | |
1523 | * And can assume that any active CPU must be online. Conversely | |
1524 | * select_task_rq() below may allow selection of !active CPUs in order | |
1525 | * to satisfy the above rules. | |
30da688e | 1526 | */ |
5da9a0fb PZ |
1527 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1528 | { | |
aa00d89c TC |
1529 | int nid = cpu_to_node(cpu); |
1530 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1531 | enum { cpuset, possible, fail } state = cpuset; |
1532 | int dest_cpu; | |
5da9a0fb | 1533 | |
aa00d89c TC |
1534 | /* |
1535 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1536 | * will return -1. There is no cpu on the node, and we should | |
1537 | * select the cpu on the other node. | |
1538 | */ | |
1539 | if (nid != -1) { | |
1540 | nodemask = cpumask_of_node(nid); | |
1541 | ||
1542 | /* Look for allowed, online CPU in same node. */ | |
1543 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1544 | if (!cpu_active(dest_cpu)) |
1545 | continue; | |
1546 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1547 | return dest_cpu; | |
1548 | } | |
2baab4e9 | 1549 | } |
5da9a0fb | 1550 | |
2baab4e9 PZ |
1551 | for (;;) { |
1552 | /* Any allowed, online CPU? */ | |
e3831edd | 1553 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
feb245e3 TH |
1554 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
1555 | continue; | |
1556 | if (!cpu_online(dest_cpu)) | |
2baab4e9 PZ |
1557 | continue; |
1558 | goto out; | |
1559 | } | |
5da9a0fb | 1560 | |
e73e85f0 | 1561 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1562 | switch (state) { |
1563 | case cpuset: | |
e73e85f0 ON |
1564 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1565 | cpuset_cpus_allowed_fallback(p); | |
1566 | state = possible; | |
1567 | break; | |
1568 | } | |
1569 | /* fall-through */ | |
2baab4e9 PZ |
1570 | case possible: |
1571 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1572 | state = fail; | |
1573 | break; | |
1574 | ||
1575 | case fail: | |
1576 | BUG(); | |
1577 | break; | |
1578 | } | |
1579 | } | |
1580 | ||
1581 | out: | |
1582 | if (state != cpuset) { | |
1583 | /* | |
1584 | * Don't tell them about moving exiting tasks or | |
1585 | * kernel threads (both mm NULL), since they never | |
1586 | * leave kernel. | |
1587 | */ | |
1588 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1589 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1590 | task_pid_nr(p), p->comm, cpu); |
1591 | } | |
5da9a0fb PZ |
1592 | } |
1593 | ||
1594 | return dest_cpu; | |
1595 | } | |
1596 | ||
e2912009 | 1597 | /* |
013fdb80 | 1598 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1599 | */ |
970b13ba | 1600 | static inline |
ac66f547 | 1601 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1602 | { |
cbce1a68 PZ |
1603 | lockdep_assert_held(&p->pi_lock); |
1604 | ||
50605ffb | 1605 | if (tsk_nr_cpus_allowed(p) > 1) |
6c1d9410 | 1606 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 PZI |
1607 | else |
1608 | cpu = cpumask_any(tsk_cpus_allowed(p)); | |
e2912009 PZ |
1609 | |
1610 | /* | |
1611 | * In order not to call set_task_cpu() on a blocking task we need | |
1612 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1613 | * cpu. | |
1614 | * | |
1615 | * Since this is common to all placement strategies, this lives here. | |
1616 | * | |
1617 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1618 | * not worry about this generic constraint ] | |
1619 | */ | |
fa17b507 | 1620 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1621 | !cpu_online(cpu))) |
5da9a0fb | 1622 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1623 | |
1624 | return cpu; | |
970b13ba | 1625 | } |
09a40af5 MG |
1626 | |
1627 | static void update_avg(u64 *avg, u64 sample) | |
1628 | { | |
1629 | s64 diff = sample - *avg; | |
1630 | *avg += diff >> 3; | |
1631 | } | |
25834c73 PZ |
1632 | |
1633 | #else | |
1634 | ||
1635 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1636 | const struct cpumask *new_mask, bool check) | |
1637 | { | |
1638 | return set_cpus_allowed_ptr(p, new_mask); | |
1639 | } | |
1640 | ||
5cc389bc | 1641 | #endif /* CONFIG_SMP */ |
970b13ba | 1642 | |
d7c01d27 | 1643 | static void |
b84cb5df | 1644 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1645 | { |
4fa8d299 | 1646 | struct rq *rq; |
b84cb5df | 1647 | |
4fa8d299 JP |
1648 | if (!schedstat_enabled()) |
1649 | return; | |
1650 | ||
1651 | rq = this_rq(); | |
d7c01d27 | 1652 | |
4fa8d299 JP |
1653 | #ifdef CONFIG_SMP |
1654 | if (cpu == rq->cpu) { | |
ae92882e JP |
1655 | schedstat_inc(rq->ttwu_local); |
1656 | schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1657 | } else { |
1658 | struct sched_domain *sd; | |
1659 | ||
ae92882e | 1660 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1661 | rcu_read_lock(); |
4fa8d299 | 1662 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1663 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
ae92882e | 1664 | schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1665 | break; |
1666 | } | |
1667 | } | |
057f3fad | 1668 | rcu_read_unlock(); |
d7c01d27 | 1669 | } |
f339b9dc PZ |
1670 | |
1671 | if (wake_flags & WF_MIGRATED) | |
ae92882e | 1672 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1673 | #endif /* CONFIG_SMP */ |
1674 | ||
ae92882e JP |
1675 | schedstat_inc(rq->ttwu_count); |
1676 | schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1677 | |
1678 | if (wake_flags & WF_SYNC) | |
ae92882e | 1679 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1680 | } |
1681 | ||
1de64443 | 1682 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1683 | { |
9ed3811a | 1684 | activate_task(rq, p, en_flags); |
da0c1e65 | 1685 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1686 | |
1687 | /* if a worker is waking up, notify workqueue */ | |
1688 | if (p->flags & PF_WQ_WORKER) | |
1689 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1690 | } |
1691 | ||
23f41eeb PZ |
1692 | /* |
1693 | * Mark the task runnable and perform wakeup-preemption. | |
1694 | */ | |
e7904a28 | 1695 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1696 | struct rq_flags *rf) |
9ed3811a | 1697 | { |
9ed3811a | 1698 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1699 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1700 | trace_sched_wakeup(p); |
1701 | ||
9ed3811a | 1702 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1703 | if (p->sched_class->task_woken) { |
1704 | /* | |
cbce1a68 PZ |
1705 | * Our task @p is fully woken up and running; so its safe to |
1706 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1707 | */ |
d8ac8971 | 1708 | rq_unpin_lock(rq, rf); |
9ed3811a | 1709 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 1710 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 1711 | } |
9ed3811a | 1712 | |
e69c6341 | 1713 | if (rq->idle_stamp) { |
78becc27 | 1714 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1715 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1716 | |
abfafa54 JL |
1717 | update_avg(&rq->avg_idle, delta); |
1718 | ||
1719 | if (rq->avg_idle > max) | |
9ed3811a | 1720 | rq->avg_idle = max; |
abfafa54 | 1721 | |
9ed3811a TH |
1722 | rq->idle_stamp = 0; |
1723 | } | |
1724 | #endif | |
1725 | } | |
1726 | ||
c05fbafb | 1727 | static void |
e7904a28 | 1728 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1729 | struct rq_flags *rf) |
c05fbafb | 1730 | { |
b5179ac7 PZ |
1731 | int en_flags = ENQUEUE_WAKEUP; |
1732 | ||
cbce1a68 PZ |
1733 | lockdep_assert_held(&rq->lock); |
1734 | ||
c05fbafb PZ |
1735 | #ifdef CONFIG_SMP |
1736 | if (p->sched_contributes_to_load) | |
1737 | rq->nr_uninterruptible--; | |
b5179ac7 | 1738 | |
b5179ac7 | 1739 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1740 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1741 | #endif |
1742 | ||
b5179ac7 | 1743 | ttwu_activate(rq, p, en_flags); |
d8ac8971 | 1744 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
1745 | } |
1746 | ||
1747 | /* | |
1748 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1749 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1750 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1751 | * the task is still ->on_rq. | |
1752 | */ | |
1753 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1754 | { | |
eb580751 | 1755 | struct rq_flags rf; |
c05fbafb PZ |
1756 | struct rq *rq; |
1757 | int ret = 0; | |
1758 | ||
eb580751 | 1759 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1760 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1761 | /* check_preempt_curr() may use rq clock */ |
1762 | update_rq_clock(rq); | |
d8ac8971 | 1763 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
1764 | ret = 1; |
1765 | } | |
eb580751 | 1766 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1767 | |
1768 | return ret; | |
1769 | } | |
1770 | ||
317f3941 | 1771 | #ifdef CONFIG_SMP |
e3baac47 | 1772 | void sched_ttwu_pending(void) |
317f3941 PZ |
1773 | { |
1774 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1775 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1776 | struct task_struct *p; | |
e3baac47 | 1777 | unsigned long flags; |
d8ac8971 | 1778 | struct rq_flags rf; |
317f3941 | 1779 | |
e3baac47 PZ |
1780 | if (!llist) |
1781 | return; | |
1782 | ||
1783 | raw_spin_lock_irqsave(&rq->lock, flags); | |
d8ac8971 | 1784 | rq_pin_lock(rq, &rf); |
317f3941 | 1785 | |
fa14ff4a | 1786 | while (llist) { |
b7e7ade3 PZ |
1787 | int wake_flags = 0; |
1788 | ||
fa14ff4a PZ |
1789 | p = llist_entry(llist, struct task_struct, wake_entry); |
1790 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1791 | |
1792 | if (p->sched_remote_wakeup) | |
1793 | wake_flags = WF_MIGRATED; | |
1794 | ||
d8ac8971 | 1795 | ttwu_do_activate(rq, p, wake_flags, &rf); |
317f3941 PZ |
1796 | } |
1797 | ||
d8ac8971 | 1798 | rq_unpin_lock(rq, &rf); |
e3baac47 | 1799 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1800 | } |
1801 | ||
1802 | void scheduler_ipi(void) | |
1803 | { | |
f27dde8d PZ |
1804 | /* |
1805 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1806 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1807 | * this IPI. | |
1808 | */ | |
8cb75e0c | 1809 | preempt_fold_need_resched(); |
f27dde8d | 1810 | |
fd2ac4f4 | 1811 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1812 | return; |
1813 | ||
1814 | /* | |
1815 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1816 | * traditionally all their work was done from the interrupt return | |
1817 | * path. Now that we actually do some work, we need to make sure | |
1818 | * we do call them. | |
1819 | * | |
1820 | * Some archs already do call them, luckily irq_enter/exit nest | |
1821 | * properly. | |
1822 | * | |
1823 | * Arguably we should visit all archs and update all handlers, | |
1824 | * however a fair share of IPIs are still resched only so this would | |
1825 | * somewhat pessimize the simple resched case. | |
1826 | */ | |
1827 | irq_enter(); | |
fa14ff4a | 1828 | sched_ttwu_pending(); |
ca38062e SS |
1829 | |
1830 | /* | |
1831 | * Check if someone kicked us for doing the nohz idle load balance. | |
1832 | */ | |
873b4c65 | 1833 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1834 | this_rq()->idle_balance = 1; |
ca38062e | 1835 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1836 | } |
c5d753a5 | 1837 | irq_exit(); |
317f3941 PZ |
1838 | } |
1839 | ||
b7e7ade3 | 1840 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1841 | { |
e3baac47 PZ |
1842 | struct rq *rq = cpu_rq(cpu); |
1843 | ||
b7e7ade3 PZ |
1844 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1845 | ||
e3baac47 PZ |
1846 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1847 | if (!set_nr_if_polling(rq->idle)) | |
1848 | smp_send_reschedule(cpu); | |
1849 | else | |
1850 | trace_sched_wake_idle_without_ipi(cpu); | |
1851 | } | |
317f3941 | 1852 | } |
d6aa8f85 | 1853 | |
f6be8af1 CL |
1854 | void wake_up_if_idle(int cpu) |
1855 | { | |
1856 | struct rq *rq = cpu_rq(cpu); | |
1857 | unsigned long flags; | |
1858 | ||
fd7de1e8 AL |
1859 | rcu_read_lock(); |
1860 | ||
1861 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1862 | goto out; | |
f6be8af1 CL |
1863 | |
1864 | if (set_nr_if_polling(rq->idle)) { | |
1865 | trace_sched_wake_idle_without_ipi(cpu); | |
1866 | } else { | |
1867 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1868 | if (is_idle_task(rq->curr)) | |
1869 | smp_send_reschedule(cpu); | |
1870 | /* Else cpu is not in idle, do nothing here */ | |
1871 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1872 | } | |
fd7de1e8 AL |
1873 | |
1874 | out: | |
1875 | rcu_read_unlock(); | |
f6be8af1 CL |
1876 | } |
1877 | ||
39be3501 | 1878 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1879 | { |
1880 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1881 | } | |
d6aa8f85 | 1882 | #endif /* CONFIG_SMP */ |
317f3941 | 1883 | |
b5179ac7 | 1884 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1885 | { |
1886 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 1887 | struct rq_flags rf; |
c05fbafb | 1888 | |
17d9f311 | 1889 | #if defined(CONFIG_SMP) |
39be3501 | 1890 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1891 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
b7e7ade3 | 1892 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1893 | return; |
1894 | } | |
1895 | #endif | |
1896 | ||
c05fbafb | 1897 | raw_spin_lock(&rq->lock); |
d8ac8971 MF |
1898 | rq_pin_lock(rq, &rf); |
1899 | ttwu_do_activate(rq, p, wake_flags, &rf); | |
1900 | rq_unpin_lock(rq, &rf); | |
c05fbafb | 1901 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1902 | } |
1903 | ||
8643cda5 PZ |
1904 | /* |
1905 | * Notes on Program-Order guarantees on SMP systems. | |
1906 | * | |
1907 | * MIGRATION | |
1908 | * | |
1909 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1910 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1911 | * execution on its new cpu [c1]. | |
1912 | * | |
1913 | * For migration (of runnable tasks) this is provided by the following means: | |
1914 | * | |
1915 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1916 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1917 | * rq(c1)->lock (if not at the same time, then in that order). | |
1918 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1919 | * | |
1920 | * Transitivity guarantees that B happens after A and C after B. | |
1921 | * Note: we only require RCpc transitivity. | |
1922 | * Note: the cpu doing B need not be c0 or c1 | |
1923 | * | |
1924 | * Example: | |
1925 | * | |
1926 | * CPU0 CPU1 CPU2 | |
1927 | * | |
1928 | * LOCK rq(0)->lock | |
1929 | * sched-out X | |
1930 | * sched-in Y | |
1931 | * UNLOCK rq(0)->lock | |
1932 | * | |
1933 | * LOCK rq(0)->lock // orders against CPU0 | |
1934 | * dequeue X | |
1935 | * UNLOCK rq(0)->lock | |
1936 | * | |
1937 | * LOCK rq(1)->lock | |
1938 | * enqueue X | |
1939 | * UNLOCK rq(1)->lock | |
1940 | * | |
1941 | * LOCK rq(1)->lock // orders against CPU2 | |
1942 | * sched-out Z | |
1943 | * sched-in X | |
1944 | * UNLOCK rq(1)->lock | |
1945 | * | |
1946 | * | |
1947 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1948 | * | |
1949 | * For blocking we (obviously) need to provide the same guarantee as for | |
1950 | * migration. However the means are completely different as there is no lock | |
1951 | * chain to provide order. Instead we do: | |
1952 | * | |
1953 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1954 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1955 | * |
1956 | * Example: | |
1957 | * | |
1958 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1959 | * | |
1960 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1961 | * dequeue X | |
1962 | * sched-out X | |
1963 | * smp_store_release(X->on_cpu, 0); | |
1964 | * | |
1f03e8d2 | 1965 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1966 | * X->state = WAKING |
1967 | * set_task_cpu(X,2) | |
1968 | * | |
1969 | * LOCK rq(2)->lock | |
1970 | * enqueue X | |
1971 | * X->state = RUNNING | |
1972 | * UNLOCK rq(2)->lock | |
1973 | * | |
1974 | * LOCK rq(2)->lock // orders against CPU1 | |
1975 | * sched-out Z | |
1976 | * sched-in X | |
1977 | * UNLOCK rq(2)->lock | |
1978 | * | |
1979 | * UNLOCK X->pi_lock | |
1980 | * UNLOCK rq(0)->lock | |
1981 | * | |
1982 | * | |
1983 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1984 | * must observe the state that lead to our wakeup. That is, not only must our | |
1985 | * task observe its own prior state, it must also observe the stores prior to | |
1986 | * its wakeup. | |
1987 | * | |
1988 | * This means that any means of doing remote wakeups must order the CPU doing | |
1989 | * the wakeup against the CPU the task is going to end up running on. This, | |
1990 | * however, is already required for the regular Program-Order guarantee above, | |
1f03e8d2 | 1991 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
8643cda5 PZ |
1992 | * |
1993 | */ | |
1994 | ||
9ed3811a | 1995 | /** |
1da177e4 | 1996 | * try_to_wake_up - wake up a thread |
9ed3811a | 1997 | * @p: the thread to be awakened |
1da177e4 | 1998 | * @state: the mask of task states that can be woken |
9ed3811a | 1999 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 2000 | * |
a2250238 | 2001 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 2002 | * |
a2250238 PZ |
2003 | * If the task was not queued/runnable, also place it back on a runqueue. |
2004 | * | |
2005 | * Atomic against schedule() which would dequeue a task, also see | |
2006 | * set_current_state(). | |
2007 | * | |
2008 | * Return: %true if @p->state changes (an actual wakeup was done), | |
2009 | * %false otherwise. | |
1da177e4 | 2010 | */ |
e4a52bcb PZ |
2011 | static int |
2012 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2013 | { |
1da177e4 | 2014 | unsigned long flags; |
c05fbafb | 2015 | int cpu, success = 0; |
2398f2c6 | 2016 | |
e0acd0a6 ON |
2017 | /* |
2018 | * If we are going to wake up a thread waiting for CONDITION we | |
2019 | * need to ensure that CONDITION=1 done by the caller can not be | |
2020 | * reordered with p->state check below. This pairs with mb() in | |
2021 | * set_current_state() the waiting thread does. | |
2022 | */ | |
2023 | smp_mb__before_spinlock(); | |
013fdb80 | 2024 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2025 | if (!(p->state & state)) |
1da177e4 LT |
2026 | goto out; |
2027 | ||
fbd705a0 PZ |
2028 | trace_sched_waking(p); |
2029 | ||
c05fbafb | 2030 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2031 | cpu = task_cpu(p); |
1da177e4 | 2032 | |
135e8c92 BS |
2033 | /* |
2034 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2035 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2036 | * in smp_cond_load_acquire() below. | |
2037 | * | |
2038 | * sched_ttwu_pending() try_to_wake_up() | |
2039 | * [S] p->on_rq = 1; [L] P->state | |
2040 | * UNLOCK rq->lock -----. | |
2041 | * \ | |
2042 | * +--- RMB | |
2043 | * schedule() / | |
2044 | * LOCK rq->lock -----' | |
2045 | * UNLOCK rq->lock | |
2046 | * | |
2047 | * [task p] | |
2048 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq | |
2049 | * | |
2050 | * Pairs with the UNLOCK+LOCK on rq->lock from the | |
2051 | * last wakeup of our task and the schedule that got our task | |
2052 | * current. | |
2053 | */ | |
2054 | smp_rmb(); | |
c05fbafb PZ |
2055 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2056 | goto stat; | |
1da177e4 | 2057 | |
1da177e4 | 2058 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2059 | /* |
2060 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2061 | * possible to, falsely, observe p->on_cpu == 0. | |
2062 | * | |
2063 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2064 | * from the runqueue. | |
2065 | * | |
2066 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2067 | * UNLOCK rq->lock | |
2068 | * RMB | |
2069 | * LOCK rq->lock | |
2070 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2071 | * | |
2072 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2073 | * from the consecutive calls to schedule(); the first switching to our | |
2074 | * task, the second putting it to sleep. | |
2075 | */ | |
2076 | smp_rmb(); | |
2077 | ||
e9c84311 | 2078 | /* |
c05fbafb PZ |
2079 | * If the owning (remote) cpu is still in the middle of schedule() with |
2080 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
2081 | * |
2082 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2083 | * | |
2084 | * This ensures that tasks getting woken will be fully ordered against | |
2085 | * their previous state and preserve Program Order. | |
0970d299 | 2086 | */ |
1f03e8d2 | 2087 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2088 | |
a8e4f2ea | 2089 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2090 | p->state = TASK_WAKING; |
e7693a36 | 2091 | |
e33a9bba TH |
2092 | if (p->in_iowait) { |
2093 | delayacct_blkio_end(); | |
2094 | atomic_dec(&task_rq(p)->nr_iowait); | |
2095 | } | |
2096 | ||
ac66f547 | 2097 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2098 | if (task_cpu(p) != cpu) { |
2099 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2100 | set_task_cpu(p, cpu); |
f339b9dc | 2101 | } |
e33a9bba TH |
2102 | |
2103 | #else /* CONFIG_SMP */ | |
2104 | ||
2105 | if (p->in_iowait) { | |
2106 | delayacct_blkio_end(); | |
2107 | atomic_dec(&task_rq(p)->nr_iowait); | |
2108 | } | |
2109 | ||
1da177e4 | 2110 | #endif /* CONFIG_SMP */ |
1da177e4 | 2111 | |
b5179ac7 | 2112 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2113 | stat: |
4fa8d299 | 2114 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2115 | out: |
013fdb80 | 2116 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2117 | |
2118 | return success; | |
2119 | } | |
2120 | ||
21aa9af0 TH |
2121 | /** |
2122 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2123 | * @p: the thread to be awakened | |
9279e0d2 | 2124 | * @cookie: context's cookie for pinning |
21aa9af0 | 2125 | * |
2acca55e | 2126 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2127 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2128 | * the current task. |
21aa9af0 | 2129 | */ |
d8ac8971 | 2130 | static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf) |
21aa9af0 TH |
2131 | { |
2132 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2133 | |
383efcd0 TH |
2134 | if (WARN_ON_ONCE(rq != this_rq()) || |
2135 | WARN_ON_ONCE(p == current)) | |
2136 | return; | |
2137 | ||
21aa9af0 TH |
2138 | lockdep_assert_held(&rq->lock); |
2139 | ||
2acca55e | 2140 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2141 | /* |
2142 | * This is OK, because current is on_cpu, which avoids it being | |
2143 | * picked for load-balance and preemption/IRQs are still | |
2144 | * disabled avoiding further scheduler activity on it and we've | |
2145 | * not yet picked a replacement task. | |
2146 | */ | |
d8ac8971 | 2147 | rq_unpin_lock(rq, rf); |
2acca55e PZ |
2148 | raw_spin_unlock(&rq->lock); |
2149 | raw_spin_lock(&p->pi_lock); | |
2150 | raw_spin_lock(&rq->lock); | |
d8ac8971 | 2151 | rq_repin_lock(rq, rf); |
2acca55e PZ |
2152 | } |
2153 | ||
21aa9af0 | 2154 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2155 | goto out; |
21aa9af0 | 2156 | |
fbd705a0 PZ |
2157 | trace_sched_waking(p); |
2158 | ||
e33a9bba TH |
2159 | if (!task_on_rq_queued(p)) { |
2160 | if (p->in_iowait) { | |
2161 | delayacct_blkio_end(); | |
2162 | atomic_dec(&rq->nr_iowait); | |
2163 | } | |
d7c01d27 | 2164 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
e33a9bba | 2165 | } |
d7c01d27 | 2166 | |
d8ac8971 | 2167 | ttwu_do_wakeup(rq, p, 0, rf); |
4fa8d299 | 2168 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2169 | out: |
2170 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2171 | } |
2172 | ||
50fa610a DH |
2173 | /** |
2174 | * wake_up_process - Wake up a specific process | |
2175 | * @p: The process to be woken up. | |
2176 | * | |
2177 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2178 | * processes. |
2179 | * | |
2180 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2181 | * |
2182 | * It may be assumed that this function implies a write memory barrier before | |
2183 | * changing the task state if and only if any tasks are woken up. | |
2184 | */ | |
7ad5b3a5 | 2185 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2186 | { |
9067ac85 | 2187 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2188 | } |
1da177e4 LT |
2189 | EXPORT_SYMBOL(wake_up_process); |
2190 | ||
7ad5b3a5 | 2191 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2192 | { |
2193 | return try_to_wake_up(p, state, 0); | |
2194 | } | |
2195 | ||
a5e7be3b JL |
2196 | /* |
2197 | * This function clears the sched_dl_entity static params. | |
2198 | */ | |
2199 | void __dl_clear_params(struct task_struct *p) | |
2200 | { | |
2201 | struct sched_dl_entity *dl_se = &p->dl; | |
2202 | ||
2203 | dl_se->dl_runtime = 0; | |
2204 | dl_se->dl_deadline = 0; | |
2205 | dl_se->dl_period = 0; | |
2206 | dl_se->flags = 0; | |
2207 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2208 | |
2209 | dl_se->dl_throttled = 0; | |
40767b0d | 2210 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2211 | } |
2212 | ||
1da177e4 LT |
2213 | /* |
2214 | * Perform scheduler related setup for a newly forked process p. | |
2215 | * p is forked by current. | |
dd41f596 IM |
2216 | * |
2217 | * __sched_fork() is basic setup used by init_idle() too: | |
2218 | */ | |
5e1576ed | 2219 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2220 | { |
fd2f4419 PZ |
2221 | p->on_rq = 0; |
2222 | ||
2223 | p->se.on_rq = 0; | |
dd41f596 IM |
2224 | p->se.exec_start = 0; |
2225 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2226 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2227 | p->se.nr_migrations = 0; |
da7a735e | 2228 | p->se.vruntime = 0; |
fd2f4419 | 2229 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2230 | |
ad936d86 BP |
2231 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2232 | p->se.cfs_rq = NULL; | |
2233 | #endif | |
2234 | ||
6cfb0d5d | 2235 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2236 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2237 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2238 | #endif |
476d139c | 2239 | |
aab03e05 | 2240 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2241 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2242 | __dl_clear_params(p); |
aab03e05 | 2243 | |
fa717060 | 2244 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2245 | p->rt.timeout = 0; |
2246 | p->rt.time_slice = sched_rr_timeslice; | |
2247 | p->rt.on_rq = 0; | |
2248 | p->rt.on_list = 0; | |
476d139c | 2249 | |
e107be36 AK |
2250 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2251 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2252 | #endif | |
cbee9f88 PZ |
2253 | |
2254 | #ifdef CONFIG_NUMA_BALANCING | |
2255 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2256 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2257 | p->mm->numa_scan_seq = 0; |
2258 | } | |
2259 | ||
5e1576ed RR |
2260 | if (clone_flags & CLONE_VM) |
2261 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2262 | else | |
2263 | p->numa_preferred_nid = -1; | |
2264 | ||
cbee9f88 PZ |
2265 | p->node_stamp = 0ULL; |
2266 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2267 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2268 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2269 | p->numa_faults = NULL; |
7e2703e6 RR |
2270 | p->last_task_numa_placement = 0; |
2271 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2272 | |
8c8a743c | 2273 | p->numa_group = NULL; |
cbee9f88 | 2274 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2275 | } |
2276 | ||
2a595721 SD |
2277 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2278 | ||
1a687c2e | 2279 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2280 | |
1a687c2e MG |
2281 | void set_numabalancing_state(bool enabled) |
2282 | { | |
2283 | if (enabled) | |
2a595721 | 2284 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2285 | else |
2a595721 | 2286 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2287 | } |
54a43d54 AK |
2288 | |
2289 | #ifdef CONFIG_PROC_SYSCTL | |
2290 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2291 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2292 | { | |
2293 | struct ctl_table t; | |
2294 | int err; | |
2a595721 | 2295 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2296 | |
2297 | if (write && !capable(CAP_SYS_ADMIN)) | |
2298 | return -EPERM; | |
2299 | ||
2300 | t = *table; | |
2301 | t.data = &state; | |
2302 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2303 | if (err < 0) | |
2304 | return err; | |
2305 | if (write) | |
2306 | set_numabalancing_state(state); | |
2307 | return err; | |
2308 | } | |
2309 | #endif | |
2310 | #endif | |
dd41f596 | 2311 | |
4698f88c JP |
2312 | #ifdef CONFIG_SCHEDSTATS |
2313 | ||
cb251765 | 2314 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2315 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2316 | |
cb251765 MG |
2317 | static void set_schedstats(bool enabled) |
2318 | { | |
2319 | if (enabled) | |
2320 | static_branch_enable(&sched_schedstats); | |
2321 | else | |
2322 | static_branch_disable(&sched_schedstats); | |
2323 | } | |
2324 | ||
2325 | void force_schedstat_enabled(void) | |
2326 | { | |
2327 | if (!schedstat_enabled()) { | |
2328 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2329 | static_branch_enable(&sched_schedstats); | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | static int __init setup_schedstats(char *str) | |
2334 | { | |
2335 | int ret = 0; | |
2336 | if (!str) | |
2337 | goto out; | |
2338 | ||
4698f88c JP |
2339 | /* |
2340 | * This code is called before jump labels have been set up, so we can't | |
2341 | * change the static branch directly just yet. Instead set a temporary | |
2342 | * variable so init_schedstats() can do it later. | |
2343 | */ | |
cb251765 | 2344 | if (!strcmp(str, "enable")) { |
4698f88c | 2345 | __sched_schedstats = true; |
cb251765 MG |
2346 | ret = 1; |
2347 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2348 | __sched_schedstats = false; |
cb251765 MG |
2349 | ret = 1; |
2350 | } | |
2351 | out: | |
2352 | if (!ret) | |
2353 | pr_warn("Unable to parse schedstats=\n"); | |
2354 | ||
2355 | return ret; | |
2356 | } | |
2357 | __setup("schedstats=", setup_schedstats); | |
2358 | ||
4698f88c JP |
2359 | static void __init init_schedstats(void) |
2360 | { | |
2361 | set_schedstats(__sched_schedstats); | |
2362 | } | |
2363 | ||
cb251765 MG |
2364 | #ifdef CONFIG_PROC_SYSCTL |
2365 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2366 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2367 | { | |
2368 | struct ctl_table t; | |
2369 | int err; | |
2370 | int state = static_branch_likely(&sched_schedstats); | |
2371 | ||
2372 | if (write && !capable(CAP_SYS_ADMIN)) | |
2373 | return -EPERM; | |
2374 | ||
2375 | t = *table; | |
2376 | t.data = &state; | |
2377 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2378 | if (err < 0) | |
2379 | return err; | |
2380 | if (write) | |
2381 | set_schedstats(state); | |
2382 | return err; | |
2383 | } | |
4698f88c JP |
2384 | #endif /* CONFIG_PROC_SYSCTL */ |
2385 | #else /* !CONFIG_SCHEDSTATS */ | |
2386 | static inline void init_schedstats(void) {} | |
2387 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2388 | |
2389 | /* | |
2390 | * fork()/clone()-time setup: | |
2391 | */ | |
aab03e05 | 2392 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2393 | { |
0122ec5b | 2394 | unsigned long flags; |
dd41f596 IM |
2395 | int cpu = get_cpu(); |
2396 | ||
5e1576ed | 2397 | __sched_fork(clone_flags, p); |
06b83b5f | 2398 | /* |
7dc603c9 | 2399 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2400 | * nobody will actually run it, and a signal or other external |
2401 | * event cannot wake it up and insert it on the runqueue either. | |
2402 | */ | |
7dc603c9 | 2403 | p->state = TASK_NEW; |
dd41f596 | 2404 | |
c350a04e MG |
2405 | /* |
2406 | * Make sure we do not leak PI boosting priority to the child. | |
2407 | */ | |
2408 | p->prio = current->normal_prio; | |
2409 | ||
b9dc29e7 MG |
2410 | /* |
2411 | * Revert to default priority/policy on fork if requested. | |
2412 | */ | |
2413 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2414 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2415 | p->policy = SCHED_NORMAL; |
6c697bdf | 2416 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2417 | p->rt_priority = 0; |
2418 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2419 | p->static_prio = NICE_TO_PRIO(0); | |
2420 | ||
2421 | p->prio = p->normal_prio = __normal_prio(p); | |
2422 | set_load_weight(p); | |
6c697bdf | 2423 | |
b9dc29e7 MG |
2424 | /* |
2425 | * We don't need the reset flag anymore after the fork. It has | |
2426 | * fulfilled its duty: | |
2427 | */ | |
2428 | p->sched_reset_on_fork = 0; | |
2429 | } | |
ca94c442 | 2430 | |
aab03e05 DF |
2431 | if (dl_prio(p->prio)) { |
2432 | put_cpu(); | |
2433 | return -EAGAIN; | |
2434 | } else if (rt_prio(p->prio)) { | |
2435 | p->sched_class = &rt_sched_class; | |
2436 | } else { | |
2ddbf952 | 2437 | p->sched_class = &fair_sched_class; |
aab03e05 | 2438 | } |
b29739f9 | 2439 | |
7dc603c9 | 2440 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2441 | |
86951599 PZ |
2442 | /* |
2443 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2444 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2445 | * is ran before sched_fork(). | |
2446 | * | |
2447 | * Silence PROVE_RCU. | |
2448 | */ | |
0122ec5b | 2449 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd PZ |
2450 | /* |
2451 | * We're setting the cpu for the first time, we don't migrate, | |
2452 | * so use __set_task_cpu(). | |
2453 | */ | |
2454 | __set_task_cpu(p, cpu); | |
2455 | if (p->sched_class->task_fork) | |
2456 | p->sched_class->task_fork(p); | |
0122ec5b | 2457 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2458 | |
f6db8347 | 2459 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2460 | if (likely(sched_info_on())) |
52f17b6c | 2461 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2462 | #endif |
3ca7a440 PZ |
2463 | #if defined(CONFIG_SMP) |
2464 | p->on_cpu = 0; | |
4866cde0 | 2465 | #endif |
01028747 | 2466 | init_task_preempt_count(p); |
806c09a7 | 2467 | #ifdef CONFIG_SMP |
917b627d | 2468 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2469 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2470 | #endif |
917b627d | 2471 | |
476d139c | 2472 | put_cpu(); |
aab03e05 | 2473 | return 0; |
1da177e4 LT |
2474 | } |
2475 | ||
332ac17e DF |
2476 | unsigned long to_ratio(u64 period, u64 runtime) |
2477 | { | |
2478 | if (runtime == RUNTIME_INF) | |
2479 | return 1ULL << 20; | |
2480 | ||
2481 | /* | |
2482 | * Doing this here saves a lot of checks in all | |
2483 | * the calling paths, and returning zero seems | |
2484 | * safe for them anyway. | |
2485 | */ | |
2486 | if (period == 0) | |
2487 | return 0; | |
2488 | ||
2489 | return div64_u64(runtime << 20, period); | |
2490 | } | |
2491 | ||
2492 | #ifdef CONFIG_SMP | |
2493 | inline struct dl_bw *dl_bw_of(int i) | |
2494 | { | |
f78f5b90 PM |
2495 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2496 | "sched RCU must be held"); | |
332ac17e DF |
2497 | return &cpu_rq(i)->rd->dl_bw; |
2498 | } | |
2499 | ||
de212f18 | 2500 | static inline int dl_bw_cpus(int i) |
332ac17e | 2501 | { |
de212f18 PZ |
2502 | struct root_domain *rd = cpu_rq(i)->rd; |
2503 | int cpus = 0; | |
2504 | ||
f78f5b90 PM |
2505 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2506 | "sched RCU must be held"); | |
de212f18 PZ |
2507 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2508 | cpus++; | |
2509 | ||
2510 | return cpus; | |
332ac17e DF |
2511 | } |
2512 | #else | |
2513 | inline struct dl_bw *dl_bw_of(int i) | |
2514 | { | |
2515 | return &cpu_rq(i)->dl.dl_bw; | |
2516 | } | |
2517 | ||
de212f18 | 2518 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2519 | { |
2520 | return 1; | |
2521 | } | |
2522 | #endif | |
2523 | ||
332ac17e DF |
2524 | /* |
2525 | * We must be sure that accepting a new task (or allowing changing the | |
2526 | * parameters of an existing one) is consistent with the bandwidth | |
2527 | * constraints. If yes, this function also accordingly updates the currently | |
2528 | * allocated bandwidth to reflect the new situation. | |
2529 | * | |
2530 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2531 | * |
2532 | * XXX we should delay bw change until the task's 0-lag point, see | |
2533 | * __setparam_dl(). | |
332ac17e DF |
2534 | */ |
2535 | static int dl_overflow(struct task_struct *p, int policy, | |
2536 | const struct sched_attr *attr) | |
2537 | { | |
2538 | ||
2539 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2540 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2541 | u64 runtime = attr->sched_runtime; |
2542 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2543 | int cpus, err = -1; |
332ac17e | 2544 | |
fec148c0 XP |
2545 | /* !deadline task may carry old deadline bandwidth */ |
2546 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2547 | return 0; |
2548 | ||
2549 | /* | |
2550 | * Either if a task, enters, leave, or stays -deadline but changes | |
2551 | * its parameters, we may need to update accordingly the total | |
2552 | * allocated bandwidth of the container. | |
2553 | */ | |
2554 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2555 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2556 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2557 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2558 | __dl_add(dl_b, new_bw); | |
2559 | err = 0; | |
2560 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2561 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2562 | __dl_clear(dl_b, p->dl.dl_bw); | |
2563 | __dl_add(dl_b, new_bw); | |
2564 | err = 0; | |
2565 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2566 | __dl_clear(dl_b, p->dl.dl_bw); | |
2567 | err = 0; | |
2568 | } | |
2569 | raw_spin_unlock(&dl_b->lock); | |
2570 | ||
2571 | return err; | |
2572 | } | |
2573 | ||
2574 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2575 | ||
1da177e4 LT |
2576 | /* |
2577 | * wake_up_new_task - wake up a newly created task for the first time. | |
2578 | * | |
2579 | * This function will do some initial scheduler statistics housekeeping | |
2580 | * that must be done for every newly created context, then puts the task | |
2581 | * on the runqueue and wakes it. | |
2582 | */ | |
3e51e3ed | 2583 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2584 | { |
eb580751 | 2585 | struct rq_flags rf; |
dd41f596 | 2586 | struct rq *rq; |
fabf318e | 2587 | |
eb580751 | 2588 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2589 | p->state = TASK_RUNNING; |
fabf318e PZ |
2590 | #ifdef CONFIG_SMP |
2591 | /* | |
2592 | * Fork balancing, do it here and not earlier because: | |
2593 | * - cpus_allowed can change in the fork path | |
2594 | * - any previously selected cpu might disappear through hotplug | |
e210bffd PZ |
2595 | * |
2596 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2597 | * as we're not fully set-up yet. | |
fabf318e | 2598 | */ |
e210bffd | 2599 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2600 | #endif |
b7fa30c9 | 2601 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 2602 | update_rq_clock(rq); |
2b8c41da | 2603 | post_init_entity_util_avg(&p->se); |
0017d735 | 2604 | |
cd29fe6f | 2605 | activate_task(rq, p, 0); |
da0c1e65 | 2606 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2607 | trace_sched_wakeup_new(p); |
a7558e01 | 2608 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2609 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2610 | if (p->sched_class->task_woken) { |
2611 | /* | |
2612 | * Nothing relies on rq->lock after this, so its fine to | |
2613 | * drop it. | |
2614 | */ | |
d8ac8971 | 2615 | rq_unpin_lock(rq, &rf); |
efbbd05a | 2616 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2617 | rq_repin_lock(rq, &rf); |
0aaafaab | 2618 | } |
9a897c5a | 2619 | #endif |
eb580751 | 2620 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2621 | } |
2622 | ||
e107be36 AK |
2623 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2624 | ||
1cde2930 PZ |
2625 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2626 | ||
2ecd9d29 PZ |
2627 | void preempt_notifier_inc(void) |
2628 | { | |
2629 | static_key_slow_inc(&preempt_notifier_key); | |
2630 | } | |
2631 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2632 | ||
2633 | void preempt_notifier_dec(void) | |
2634 | { | |
2635 | static_key_slow_dec(&preempt_notifier_key); | |
2636 | } | |
2637 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2638 | ||
e107be36 | 2639 | /** |
80dd99b3 | 2640 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2641 | * @notifier: notifier struct to register |
e107be36 AK |
2642 | */ |
2643 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2644 | { | |
2ecd9d29 PZ |
2645 | if (!static_key_false(&preempt_notifier_key)) |
2646 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2647 | ||
e107be36 AK |
2648 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2649 | } | |
2650 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2651 | ||
2652 | /** | |
2653 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2654 | * @notifier: notifier struct to unregister |
e107be36 | 2655 | * |
d84525a8 | 2656 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2657 | */ |
2658 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2659 | { | |
2660 | hlist_del(¬ifier->link); | |
2661 | } | |
2662 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2663 | ||
1cde2930 | 2664 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2665 | { |
2666 | struct preempt_notifier *notifier; | |
e107be36 | 2667 | |
b67bfe0d | 2668 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2669 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2670 | } | |
2671 | ||
1cde2930 PZ |
2672 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2673 | { | |
2674 | if (static_key_false(&preempt_notifier_key)) | |
2675 | __fire_sched_in_preempt_notifiers(curr); | |
2676 | } | |
2677 | ||
e107be36 | 2678 | static void |
1cde2930 PZ |
2679 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2680 | struct task_struct *next) | |
e107be36 AK |
2681 | { |
2682 | struct preempt_notifier *notifier; | |
e107be36 | 2683 | |
b67bfe0d | 2684 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2685 | notifier->ops->sched_out(notifier, next); |
2686 | } | |
2687 | ||
1cde2930 PZ |
2688 | static __always_inline void |
2689 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2690 | struct task_struct *next) | |
2691 | { | |
2692 | if (static_key_false(&preempt_notifier_key)) | |
2693 | __fire_sched_out_preempt_notifiers(curr, next); | |
2694 | } | |
2695 | ||
6d6bc0ad | 2696 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2697 | |
1cde2930 | 2698 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2699 | { |
2700 | } | |
2701 | ||
1cde2930 | 2702 | static inline void |
e107be36 AK |
2703 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2704 | struct task_struct *next) | |
2705 | { | |
2706 | } | |
2707 | ||
6d6bc0ad | 2708 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2709 | |
4866cde0 NP |
2710 | /** |
2711 | * prepare_task_switch - prepare to switch tasks | |
2712 | * @rq: the runqueue preparing to switch | |
421cee29 | 2713 | * @prev: the current task that is being switched out |
4866cde0 NP |
2714 | * @next: the task we are going to switch to. |
2715 | * | |
2716 | * This is called with the rq lock held and interrupts off. It must | |
2717 | * be paired with a subsequent finish_task_switch after the context | |
2718 | * switch. | |
2719 | * | |
2720 | * prepare_task_switch sets up locking and calls architecture specific | |
2721 | * hooks. | |
2722 | */ | |
e107be36 AK |
2723 | static inline void |
2724 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2725 | struct task_struct *next) | |
4866cde0 | 2726 | { |
43148951 | 2727 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2728 | perf_event_task_sched_out(prev, next); |
e107be36 | 2729 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2730 | prepare_lock_switch(rq, next); |
2731 | prepare_arch_switch(next); | |
2732 | } | |
2733 | ||
1da177e4 LT |
2734 | /** |
2735 | * finish_task_switch - clean up after a task-switch | |
2736 | * @prev: the thread we just switched away from. | |
2737 | * | |
4866cde0 NP |
2738 | * finish_task_switch must be called after the context switch, paired |
2739 | * with a prepare_task_switch call before the context switch. | |
2740 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2741 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2742 | * |
2743 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2744 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2745 | * with the lock held can cause deadlocks; see schedule() for |
2746 | * details.) | |
dfa50b60 ON |
2747 | * |
2748 | * The context switch have flipped the stack from under us and restored the | |
2749 | * local variables which were saved when this task called schedule() in the | |
2750 | * past. prev == current is still correct but we need to recalculate this_rq | |
2751 | * because prev may have moved to another CPU. | |
1da177e4 | 2752 | */ |
dfa50b60 | 2753 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2754 | __releases(rq->lock) |
2755 | { | |
dfa50b60 | 2756 | struct rq *rq = this_rq(); |
1da177e4 | 2757 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2758 | long prev_state; |
1da177e4 | 2759 | |
609ca066 PZ |
2760 | /* |
2761 | * The previous task will have left us with a preempt_count of 2 | |
2762 | * because it left us after: | |
2763 | * | |
2764 | * schedule() | |
2765 | * preempt_disable(); // 1 | |
2766 | * __schedule() | |
2767 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2768 | * | |
2769 | * Also, see FORK_PREEMPT_COUNT. | |
2770 | */ | |
e2bf1c4b PZ |
2771 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2772 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2773 | current->comm, current->pid, preempt_count())) | |
2774 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2775 | |
1da177e4 LT |
2776 | rq->prev_mm = NULL; |
2777 | ||
2778 | /* | |
2779 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2780 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2781 | * schedule one last time. The schedule call will never return, and |
2782 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2783 | * |
2784 | * We must observe prev->state before clearing prev->on_cpu (in | |
2785 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2786 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2787 | * transition, resulting in a double drop. | |
1da177e4 | 2788 | */ |
55a101f8 | 2789 | prev_state = prev->state; |
bf9fae9f | 2790 | vtime_task_switch(prev); |
a8d757ef | 2791 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2792 | finish_lock_switch(rq, prev); |
01f23e16 | 2793 | finish_arch_post_lock_switch(); |
e8fa1362 | 2794 | |
e107be36 | 2795 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2796 | if (mm) |
2797 | mmdrop(mm); | |
c394cc9f | 2798 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2799 | if (prev->sched_class->task_dead) |
2800 | prev->sched_class->task_dead(prev); | |
2801 | ||
c6fd91f0 | 2802 | /* |
2803 | * Remove function-return probe instances associated with this | |
2804 | * task and put them back on the free list. | |
9761eea8 | 2805 | */ |
c6fd91f0 | 2806 | kprobe_flush_task(prev); |
68f24b08 AL |
2807 | |
2808 | /* Task is done with its stack. */ | |
2809 | put_task_stack(prev); | |
2810 | ||
1da177e4 | 2811 | put_task_struct(prev); |
c6fd91f0 | 2812 | } |
99e5ada9 | 2813 | |
de734f89 | 2814 | tick_nohz_task_switch(); |
dfa50b60 | 2815 | return rq; |
1da177e4 LT |
2816 | } |
2817 | ||
3f029d3c GH |
2818 | #ifdef CONFIG_SMP |
2819 | ||
3f029d3c | 2820 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2821 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2822 | { |
e3fca9e7 PZ |
2823 | struct callback_head *head, *next; |
2824 | void (*func)(struct rq *rq); | |
2825 | unsigned long flags; | |
3f029d3c | 2826 | |
e3fca9e7 PZ |
2827 | raw_spin_lock_irqsave(&rq->lock, flags); |
2828 | head = rq->balance_callback; | |
2829 | rq->balance_callback = NULL; | |
2830 | while (head) { | |
2831 | func = (void (*)(struct rq *))head->func; | |
2832 | next = head->next; | |
2833 | head->next = NULL; | |
2834 | head = next; | |
3f029d3c | 2835 | |
e3fca9e7 | 2836 | func(rq); |
3f029d3c | 2837 | } |
e3fca9e7 PZ |
2838 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2839 | } | |
2840 | ||
2841 | static inline void balance_callback(struct rq *rq) | |
2842 | { | |
2843 | if (unlikely(rq->balance_callback)) | |
2844 | __balance_callback(rq); | |
3f029d3c GH |
2845 | } |
2846 | ||
2847 | #else | |
da19ab51 | 2848 | |
e3fca9e7 | 2849 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2850 | { |
1da177e4 LT |
2851 | } |
2852 | ||
3f029d3c GH |
2853 | #endif |
2854 | ||
1da177e4 LT |
2855 | /** |
2856 | * schedule_tail - first thing a freshly forked thread must call. | |
2857 | * @prev: the thread we just switched away from. | |
2858 | */ | |
722a9f92 | 2859 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2860 | __releases(rq->lock) |
2861 | { | |
1a43a14a | 2862 | struct rq *rq; |
da19ab51 | 2863 | |
609ca066 PZ |
2864 | /* |
2865 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2866 | * finish_task_switch() for details. | |
2867 | * | |
2868 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2869 | * and the preempt_enable() will end up enabling preemption (on | |
2870 | * PREEMPT_COUNT kernels). | |
2871 | */ | |
2872 | ||
dfa50b60 | 2873 | rq = finish_task_switch(prev); |
e3fca9e7 | 2874 | balance_callback(rq); |
1a43a14a | 2875 | preempt_enable(); |
70b97a7f | 2876 | |
1da177e4 | 2877 | if (current->set_child_tid) |
b488893a | 2878 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2879 | } |
2880 | ||
2881 | /* | |
dfa50b60 | 2882 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2883 | */ |
04936948 | 2884 | static __always_inline struct rq * |
70b97a7f | 2885 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 2886 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 2887 | { |
dd41f596 | 2888 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2889 | |
e107be36 | 2890 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2891 | |
dd41f596 IM |
2892 | mm = next->mm; |
2893 | oldmm = prev->active_mm; | |
9226d125 ZA |
2894 | /* |
2895 | * For paravirt, this is coupled with an exit in switch_to to | |
2896 | * combine the page table reload and the switch backend into | |
2897 | * one hypercall. | |
2898 | */ | |
224101ed | 2899 | arch_start_context_switch(prev); |
9226d125 | 2900 | |
31915ab4 | 2901 | if (!mm) { |
1da177e4 LT |
2902 | next->active_mm = oldmm; |
2903 | atomic_inc(&oldmm->mm_count); | |
2904 | enter_lazy_tlb(oldmm, next); | |
2905 | } else | |
f98db601 | 2906 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2907 | |
31915ab4 | 2908 | if (!prev->mm) { |
1da177e4 | 2909 | prev->active_mm = NULL; |
1da177e4 LT |
2910 | rq->prev_mm = oldmm; |
2911 | } | |
92509b73 | 2912 | |
cb42c9a3 | 2913 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 2914 | |
3a5f5e48 IM |
2915 | /* |
2916 | * Since the runqueue lock will be released by the next | |
2917 | * task (which is an invalid locking op but in the case | |
2918 | * of the scheduler it's an obvious special-case), so we | |
2919 | * do an early lockdep release here: | |
2920 | */ | |
d8ac8971 | 2921 | rq_unpin_lock(rq, rf); |
8a25d5de | 2922 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2923 | |
2924 | /* Here we just switch the register state and the stack. */ | |
2925 | switch_to(prev, next, prev); | |
dd41f596 | 2926 | barrier(); |
dfa50b60 ON |
2927 | |
2928 | return finish_task_switch(prev); | |
1da177e4 LT |
2929 | } |
2930 | ||
2931 | /* | |
1c3e8264 | 2932 | * nr_running and nr_context_switches: |
1da177e4 LT |
2933 | * |
2934 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2935 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2936 | */ |
2937 | unsigned long nr_running(void) | |
2938 | { | |
2939 | unsigned long i, sum = 0; | |
2940 | ||
2941 | for_each_online_cpu(i) | |
2942 | sum += cpu_rq(i)->nr_running; | |
2943 | ||
2944 | return sum; | |
f711f609 | 2945 | } |
1da177e4 | 2946 | |
2ee507c4 TC |
2947 | /* |
2948 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2949 | * |
2950 | * Caution: this function does not check that the caller has disabled | |
2951 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2952 | * race. The caller is responsible to use it correctly, for example: | |
2953 | * | |
2954 | * - from a non-preemptable section (of course) | |
2955 | * | |
2956 | * - from a thread that is bound to a single CPU | |
2957 | * | |
2958 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2959 | */ |
2960 | bool single_task_running(void) | |
2961 | { | |
00cc1633 | 2962 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2963 | } |
2964 | EXPORT_SYMBOL(single_task_running); | |
2965 | ||
1da177e4 | 2966 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2967 | { |
cc94abfc SR |
2968 | int i; |
2969 | unsigned long long sum = 0; | |
46cb4b7c | 2970 | |
0a945022 | 2971 | for_each_possible_cpu(i) |
1da177e4 | 2972 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2973 | |
1da177e4 LT |
2974 | return sum; |
2975 | } | |
483b4ee6 | 2976 | |
e33a9bba TH |
2977 | /* |
2978 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
2979 | * | |
2980 | * The idea behind IO-wait account is to account the idle time that we could | |
2981 | * have spend running if it were not for IO. That is, if we were to improve the | |
2982 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
2983 | * | |
2984 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
2985 | * idle time as IO-wait, because if the storage were faster, it could've been | |
2986 | * running and we'd not be idle. | |
2987 | * | |
2988 | * This has been extended to SMP, by doing the same for each CPU. This however | |
2989 | * is broken. | |
2990 | * | |
2991 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
2992 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
2993 | * though, if the storage were faster, both could've ran at the same time, | |
2994 | * utilising both CPUs. | |
2995 | * | |
2996 | * This means, that when looking globally, the current IO-wait accounting on | |
2997 | * SMP is a lower bound, by reason of under accounting. | |
2998 | * | |
2999 | * Worse, since the numbers are provided per CPU, they are sometimes | |
3000 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
3001 | * associated with any one particular CPU, it can wake to another CPU than it | |
3002 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
3003 | * | |
3004 | * Task CPU affinities can make all that even more 'interesting'. | |
3005 | */ | |
3006 | ||
1da177e4 LT |
3007 | unsigned long nr_iowait(void) |
3008 | { | |
3009 | unsigned long i, sum = 0; | |
483b4ee6 | 3010 | |
0a945022 | 3011 | for_each_possible_cpu(i) |
1da177e4 | 3012 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3013 | |
1da177e4 LT |
3014 | return sum; |
3015 | } | |
483b4ee6 | 3016 | |
e33a9bba TH |
3017 | /* |
3018 | * Consumers of these two interfaces, like for example the cpufreq menu | |
3019 | * governor are using nonsensical data. Boosting frequency for a CPU that has | |
3020 | * IO-wait which might not even end up running the task when it does become | |
3021 | * runnable. | |
3022 | */ | |
3023 | ||
8c215bd3 | 3024 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3025 | { |
8c215bd3 | 3026 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3027 | return atomic_read(&this->nr_iowait); |
3028 | } | |
46cb4b7c | 3029 | |
372ba8cb MG |
3030 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
3031 | { | |
3289bdb4 PZ |
3032 | struct rq *rq = this_rq(); |
3033 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
3034 | *load = rq->load.weight; | |
372ba8cb MG |
3035 | } |
3036 | ||
dd41f596 | 3037 | #ifdef CONFIG_SMP |
8a0be9ef | 3038 | |
46cb4b7c | 3039 | /* |
38022906 PZ |
3040 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3041 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3042 | */ |
38022906 | 3043 | void sched_exec(void) |
46cb4b7c | 3044 | { |
38022906 | 3045 | struct task_struct *p = current; |
1da177e4 | 3046 | unsigned long flags; |
0017d735 | 3047 | int dest_cpu; |
46cb4b7c | 3048 | |
8f42ced9 | 3049 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 3050 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3051 | if (dest_cpu == smp_processor_id()) |
3052 | goto unlock; | |
38022906 | 3053 | |
8f42ced9 | 3054 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3055 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3056 | |
8f42ced9 PZ |
3057 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3058 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3059 | return; |
3060 | } | |
0017d735 | 3061 | unlock: |
8f42ced9 | 3062 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3063 | } |
dd41f596 | 3064 | |
1da177e4 LT |
3065 | #endif |
3066 | ||
1da177e4 | 3067 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3068 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3069 | |
3070 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3071 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3072 | |
6075620b GG |
3073 | /* |
3074 | * The function fair_sched_class.update_curr accesses the struct curr | |
3075 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3076 | * we observe a high rate of cache misses in practice. | |
3077 | * Prefetching this data results in improved performance. | |
3078 | */ | |
3079 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3080 | { | |
3081 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3082 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3083 | #else | |
3084 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3085 | #endif | |
3086 | prefetch(curr); | |
3087 | prefetch(&curr->exec_start); | |
3088 | } | |
3089 | ||
c5f8d995 HS |
3090 | /* |
3091 | * Return accounted runtime for the task. | |
3092 | * In case the task is currently running, return the runtime plus current's | |
3093 | * pending runtime that have not been accounted yet. | |
3094 | */ | |
3095 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3096 | { | |
eb580751 | 3097 | struct rq_flags rf; |
c5f8d995 | 3098 | struct rq *rq; |
6e998916 | 3099 | u64 ns; |
c5f8d995 | 3100 | |
911b2898 PZ |
3101 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3102 | /* | |
3103 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
3104 | * So we have a optimization chance when the task's delta_exec is 0. | |
3105 | * Reading ->on_cpu is racy, but this is ok. | |
3106 | * | |
3107 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
3108 | * If we race with it entering cpu, unaccounted time is 0. This is | |
3109 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
3110 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3111 | * been accounted, so we're correct here as well. | |
911b2898 | 3112 | */ |
da0c1e65 | 3113 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3114 | return p->se.sum_exec_runtime; |
3115 | #endif | |
3116 | ||
eb580751 | 3117 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3118 | /* |
3119 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3120 | * project cycles that may never be accounted to this | |
3121 | * thread, breaking clock_gettime(). | |
3122 | */ | |
3123 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3124 | prefetch_curr_exec_start(p); |
6e998916 SG |
3125 | update_rq_clock(rq); |
3126 | p->sched_class->update_curr(rq); | |
3127 | } | |
3128 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3129 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3130 | |
3131 | return ns; | |
3132 | } | |
48f24c4d | 3133 | |
7835b98b CL |
3134 | /* |
3135 | * This function gets called by the timer code, with HZ frequency. | |
3136 | * We call it with interrupts disabled. | |
7835b98b CL |
3137 | */ |
3138 | void scheduler_tick(void) | |
3139 | { | |
7835b98b CL |
3140 | int cpu = smp_processor_id(); |
3141 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3142 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3143 | |
3144 | sched_clock_tick(); | |
dd41f596 | 3145 | |
05fa785c | 3146 | raw_spin_lock(&rq->lock); |
3e51f33f | 3147 | update_rq_clock(rq); |
fa85ae24 | 3148 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3149 | cpu_load_update_active(rq); |
3289bdb4 | 3150 | calc_global_load_tick(rq); |
05fa785c | 3151 | raw_spin_unlock(&rq->lock); |
7835b98b | 3152 | |
e9d2b064 | 3153 | perf_event_task_tick(); |
e220d2dc | 3154 | |
e418e1c2 | 3155 | #ifdef CONFIG_SMP |
6eb57e0d | 3156 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3157 | trigger_load_balance(rq); |
e418e1c2 | 3158 | #endif |
265f22a9 | 3159 | rq_last_tick_reset(rq); |
1da177e4 LT |
3160 | } |
3161 | ||
265f22a9 FW |
3162 | #ifdef CONFIG_NO_HZ_FULL |
3163 | /** | |
3164 | * scheduler_tick_max_deferment | |
3165 | * | |
3166 | * Keep at least one tick per second when a single | |
3167 | * active task is running because the scheduler doesn't | |
3168 | * yet completely support full dynticks environment. | |
3169 | * | |
3170 | * This makes sure that uptime, CFS vruntime, load | |
3171 | * balancing, etc... continue to move forward, even | |
3172 | * with a very low granularity. | |
e69f6186 YB |
3173 | * |
3174 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3175 | */ |
3176 | u64 scheduler_tick_max_deferment(void) | |
3177 | { | |
3178 | struct rq *rq = this_rq(); | |
316c1608 | 3179 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3180 | |
3181 | next = rq->last_sched_tick + HZ; | |
3182 | ||
3183 | if (time_before_eq(next, now)) | |
3184 | return 0; | |
3185 | ||
8fe8ff09 | 3186 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3187 | } |
265f22a9 | 3188 | #endif |
1da177e4 | 3189 | |
7e49fcce SR |
3190 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3191 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3192 | /* |
3193 | * If the value passed in is equal to the current preempt count | |
3194 | * then we just disabled preemption. Start timing the latency. | |
3195 | */ | |
3196 | static inline void preempt_latency_start(int val) | |
3197 | { | |
3198 | if (preempt_count() == val) { | |
3199 | unsigned long ip = get_lock_parent_ip(); | |
3200 | #ifdef CONFIG_DEBUG_PREEMPT | |
3201 | current->preempt_disable_ip = ip; | |
3202 | #endif | |
3203 | trace_preempt_off(CALLER_ADDR0, ip); | |
3204 | } | |
3205 | } | |
7e49fcce | 3206 | |
edafe3a5 | 3207 | void preempt_count_add(int val) |
1da177e4 | 3208 | { |
6cd8a4bb | 3209 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3210 | /* |
3211 | * Underflow? | |
3212 | */ | |
9a11b49a IM |
3213 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3214 | return; | |
6cd8a4bb | 3215 | #endif |
bdb43806 | 3216 | __preempt_count_add(val); |
6cd8a4bb | 3217 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3218 | /* |
3219 | * Spinlock count overflowing soon? | |
3220 | */ | |
33859f7f MOS |
3221 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3222 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3223 | #endif |
47252cfb | 3224 | preempt_latency_start(val); |
1da177e4 | 3225 | } |
bdb43806 | 3226 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3227 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3228 | |
47252cfb SR |
3229 | /* |
3230 | * If the value passed in equals to the current preempt count | |
3231 | * then we just enabled preemption. Stop timing the latency. | |
3232 | */ | |
3233 | static inline void preempt_latency_stop(int val) | |
3234 | { | |
3235 | if (preempt_count() == val) | |
3236 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3237 | } | |
3238 | ||
edafe3a5 | 3239 | void preempt_count_sub(int val) |
1da177e4 | 3240 | { |
6cd8a4bb | 3241 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3242 | /* |
3243 | * Underflow? | |
3244 | */ | |
01e3eb82 | 3245 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3246 | return; |
1da177e4 LT |
3247 | /* |
3248 | * Is the spinlock portion underflowing? | |
3249 | */ | |
9a11b49a IM |
3250 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3251 | !(preempt_count() & PREEMPT_MASK))) | |
3252 | return; | |
6cd8a4bb | 3253 | #endif |
9a11b49a | 3254 | |
47252cfb | 3255 | preempt_latency_stop(val); |
bdb43806 | 3256 | __preempt_count_sub(val); |
1da177e4 | 3257 | } |
bdb43806 | 3258 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3259 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3260 | |
47252cfb SR |
3261 | #else |
3262 | static inline void preempt_latency_start(int val) { } | |
3263 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3264 | #endif |
3265 | ||
3266 | /* | |
dd41f596 | 3267 | * Print scheduling while atomic bug: |
1da177e4 | 3268 | */ |
dd41f596 | 3269 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3270 | { |
d1c6d149 VN |
3271 | /* Save this before calling printk(), since that will clobber it */ |
3272 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3273 | ||
664dfa65 DJ |
3274 | if (oops_in_progress) |
3275 | return; | |
3276 | ||
3df0fc5b PZ |
3277 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3278 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3279 | |
dd41f596 | 3280 | debug_show_held_locks(prev); |
e21f5b15 | 3281 | print_modules(); |
dd41f596 IM |
3282 | if (irqs_disabled()) |
3283 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3284 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3285 | && in_atomic_preempt_off()) { | |
8f47b187 | 3286 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3287 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3288 | pr_cont("\n"); |
3289 | } | |
748c7201 DBO |
3290 | if (panic_on_warn) |
3291 | panic("scheduling while atomic\n"); | |
3292 | ||
6135fc1e | 3293 | dump_stack(); |
373d4d09 | 3294 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3295 | } |
1da177e4 | 3296 | |
dd41f596 IM |
3297 | /* |
3298 | * Various schedule()-time debugging checks and statistics: | |
3299 | */ | |
3300 | static inline void schedule_debug(struct task_struct *prev) | |
3301 | { | |
0d9e2632 | 3302 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3303 | if (task_stack_end_corrupted(prev)) |
3304 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3305 | #endif |
b99def8b | 3306 | |
1dc0fffc | 3307 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3308 | __schedule_bug(prev); |
1dc0fffc PZ |
3309 | preempt_count_set(PREEMPT_DISABLED); |
3310 | } | |
b3fbab05 | 3311 | rcu_sleep_check(); |
dd41f596 | 3312 | |
1da177e4 LT |
3313 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3314 | ||
ae92882e | 3315 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3316 | } |
3317 | ||
3318 | /* | |
3319 | * Pick up the highest-prio task: | |
3320 | */ | |
3321 | static inline struct task_struct * | |
d8ac8971 | 3322 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3323 | { |
49ee5768 | 3324 | const struct sched_class *class; |
dd41f596 | 3325 | struct task_struct *p; |
1da177e4 LT |
3326 | |
3327 | /* | |
dd41f596 IM |
3328 | * Optimization: we know that if all tasks are in |
3329 | * the fair class we can call that function directly: | |
1da177e4 | 3330 | */ |
49ee5768 | 3331 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
d8ac8971 | 3332 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3333 | if (unlikely(p == RETRY_TASK)) |
3334 | goto again; | |
3335 | ||
3336 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3337 | if (unlikely(!p)) | |
d8ac8971 | 3338 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3339 | |
3340 | return p; | |
1da177e4 LT |
3341 | } |
3342 | ||
37e117c0 | 3343 | again: |
34f971f6 | 3344 | for_each_class(class) { |
d8ac8971 | 3345 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3346 | if (p) { |
3347 | if (unlikely(p == RETRY_TASK)) | |
3348 | goto again; | |
dd41f596 | 3349 | return p; |
37e117c0 | 3350 | } |
dd41f596 | 3351 | } |
34f971f6 PZ |
3352 | |
3353 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3354 | } |
1da177e4 | 3355 | |
dd41f596 | 3356 | /* |
c259e01a | 3357 | * __schedule() is the main scheduler function. |
edde96ea PE |
3358 | * |
3359 | * The main means of driving the scheduler and thus entering this function are: | |
3360 | * | |
3361 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3362 | * | |
3363 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3364 | * paths. For example, see arch/x86/entry_64.S. | |
3365 | * | |
3366 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3367 | * interrupt handler scheduler_tick(). | |
3368 | * | |
3369 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3370 | * task to the run-queue and that's it. | |
3371 | * | |
3372 | * Now, if the new task added to the run-queue preempts the current | |
3373 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3374 | * called on the nearest possible occasion: | |
3375 | * | |
3376 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3377 | * | |
3378 | * - in syscall or exception context, at the next outmost | |
3379 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3380 | * spin_unlock()!) | |
3381 | * | |
3382 | * - in IRQ context, return from interrupt-handler to | |
3383 | * preemptible context | |
3384 | * | |
3385 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3386 | * then at the next: | |
3387 | * | |
3388 | * - cond_resched() call | |
3389 | * - explicit schedule() call | |
3390 | * - return from syscall or exception to user-space | |
3391 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3392 | * |
b30f0e3f | 3393 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3394 | */ |
499d7955 | 3395 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3396 | { |
3397 | struct task_struct *prev, *next; | |
67ca7bde | 3398 | unsigned long *switch_count; |
d8ac8971 | 3399 | struct rq_flags rf; |
dd41f596 | 3400 | struct rq *rq; |
31656519 | 3401 | int cpu; |
dd41f596 | 3402 | |
dd41f596 IM |
3403 | cpu = smp_processor_id(); |
3404 | rq = cpu_rq(cpu); | |
dd41f596 | 3405 | prev = rq->curr; |
dd41f596 | 3406 | |
dd41f596 | 3407 | schedule_debug(prev); |
1da177e4 | 3408 | |
31656519 | 3409 | if (sched_feat(HRTICK)) |
f333fdc9 | 3410 | hrtick_clear(rq); |
8f4d37ec | 3411 | |
46a5d164 PM |
3412 | local_irq_disable(); |
3413 | rcu_note_context_switch(); | |
3414 | ||
e0acd0a6 ON |
3415 | /* |
3416 | * Make sure that signal_pending_state()->signal_pending() below | |
3417 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3418 | * done by the caller to avoid the race with signal_wake_up(). | |
3419 | */ | |
3420 | smp_mb__before_spinlock(); | |
46a5d164 | 3421 | raw_spin_lock(&rq->lock); |
d8ac8971 | 3422 | rq_pin_lock(rq, &rf); |
1da177e4 | 3423 | |
cb42c9a3 | 3424 | rq->clock_update_flags <<= 1; /* promote REQ to ACT */ |
9edfbfed | 3425 | |
246d86b5 | 3426 | switch_count = &prev->nivcsw; |
fc13aeba | 3427 | if (!preempt && prev->state) { |
21aa9af0 | 3428 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3429 | prev->state = TASK_RUNNING; |
21aa9af0 | 3430 | } else { |
2acca55e PZ |
3431 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3432 | prev->on_rq = 0; | |
3433 | ||
e33a9bba TH |
3434 | if (prev->in_iowait) { |
3435 | atomic_inc(&rq->nr_iowait); | |
3436 | delayacct_blkio_start(); | |
3437 | } | |
3438 | ||
21aa9af0 | 3439 | /* |
2acca55e PZ |
3440 | * If a worker went to sleep, notify and ask workqueue |
3441 | * whether it wants to wake up a task to maintain | |
3442 | * concurrency. | |
21aa9af0 TH |
3443 | */ |
3444 | if (prev->flags & PF_WQ_WORKER) { | |
3445 | struct task_struct *to_wakeup; | |
3446 | ||
9b7f6597 | 3447 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3448 | if (to_wakeup) |
d8ac8971 | 3449 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3450 | } |
21aa9af0 | 3451 | } |
dd41f596 | 3452 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3453 | } |
3454 | ||
9edfbfed | 3455 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3456 | update_rq_clock(rq); |
3457 | ||
d8ac8971 | 3458 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3459 | clear_tsk_need_resched(prev); |
f27dde8d | 3460 | clear_preempt_need_resched(); |
1da177e4 | 3461 | |
1da177e4 | 3462 | if (likely(prev != next)) { |
1da177e4 LT |
3463 | rq->nr_switches++; |
3464 | rq->curr = next; | |
3465 | ++*switch_count; | |
3466 | ||
c73464b1 | 3467 | trace_sched_switch(preempt, prev, next); |
d8ac8971 | 3468 | rq = context_switch(rq, prev, next, &rf); /* unlocks the rq */ |
cbce1a68 | 3469 | } else { |
cb42c9a3 | 3470 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
d8ac8971 | 3471 | rq_unpin_lock(rq, &rf); |
05fa785c | 3472 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3473 | } |
1da177e4 | 3474 | |
e3fca9e7 | 3475 | balance_callback(rq); |
1da177e4 | 3476 | } |
c259e01a | 3477 | |
9af6528e PZ |
3478 | void __noreturn do_task_dead(void) |
3479 | { | |
3480 | /* | |
3481 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed | |
3482 | * when the following two conditions become true. | |
3483 | * - There is race condition of mmap_sem (It is acquired by | |
3484 | * exit_mm()), and | |
3485 | * - SMI occurs before setting TASK_RUNINNG. | |
3486 | * (or hypervisor of virtual machine switches to other guest) | |
3487 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD | |
3488 | * | |
3489 | * To avoid it, we have to wait for releasing tsk->pi_lock which | |
3490 | * is held by try_to_wake_up() | |
3491 | */ | |
3492 | smp_mb(); | |
3493 | raw_spin_unlock_wait(¤t->pi_lock); | |
3494 | ||
3495 | /* causes final put_task_struct in finish_task_switch(). */ | |
3496 | __set_current_state(TASK_DEAD); | |
3497 | current->flags |= PF_NOFREEZE; /* tell freezer to ignore us */ | |
3498 | __schedule(false); | |
3499 | BUG(); | |
3500 | /* Avoid "noreturn function does return". */ | |
3501 | for (;;) | |
3502 | cpu_relax(); /* For when BUG is null */ | |
3503 | } | |
3504 | ||
9c40cef2 TG |
3505 | static inline void sched_submit_work(struct task_struct *tsk) |
3506 | { | |
3c7d5184 | 3507 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3508 | return; |
3509 | /* | |
3510 | * If we are going to sleep and we have plugged IO queued, | |
3511 | * make sure to submit it to avoid deadlocks. | |
3512 | */ | |
3513 | if (blk_needs_flush_plug(tsk)) | |
3514 | blk_schedule_flush_plug(tsk); | |
3515 | } | |
3516 | ||
722a9f92 | 3517 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3518 | { |
9c40cef2 TG |
3519 | struct task_struct *tsk = current; |
3520 | ||
3521 | sched_submit_work(tsk); | |
bfd9b2b5 | 3522 | do { |
b30f0e3f | 3523 | preempt_disable(); |
fc13aeba | 3524 | __schedule(false); |
b30f0e3f | 3525 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3526 | } while (need_resched()); |
c259e01a | 3527 | } |
1da177e4 LT |
3528 | EXPORT_SYMBOL(schedule); |
3529 | ||
91d1aa43 | 3530 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3531 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3532 | { |
3533 | /* | |
3534 | * If we come here after a random call to set_need_resched(), | |
3535 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3536 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3537 | * we find a better solution. | |
7cc78f8f AL |
3538 | * |
3539 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3540 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3541 | * too frequently to make sense yet. |
20ab65e3 | 3542 | */ |
7cc78f8f | 3543 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3544 | schedule(); |
7cc78f8f | 3545 | exception_exit(prev_state); |
20ab65e3 FW |
3546 | } |
3547 | #endif | |
3548 | ||
c5491ea7 TG |
3549 | /** |
3550 | * schedule_preempt_disabled - called with preemption disabled | |
3551 | * | |
3552 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3553 | */ | |
3554 | void __sched schedule_preempt_disabled(void) | |
3555 | { | |
ba74c144 | 3556 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3557 | schedule(); |
3558 | preempt_disable(); | |
3559 | } | |
3560 | ||
06b1f808 | 3561 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3562 | { |
3563 | do { | |
47252cfb SR |
3564 | /* |
3565 | * Because the function tracer can trace preempt_count_sub() | |
3566 | * and it also uses preempt_enable/disable_notrace(), if | |
3567 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3568 | * by the function tracer will call this function again and | |
3569 | * cause infinite recursion. | |
3570 | * | |
3571 | * Preemption must be disabled here before the function | |
3572 | * tracer can trace. Break up preempt_disable() into two | |
3573 | * calls. One to disable preemption without fear of being | |
3574 | * traced. The other to still record the preemption latency, | |
3575 | * which can also be traced by the function tracer. | |
3576 | */ | |
499d7955 | 3577 | preempt_disable_notrace(); |
47252cfb | 3578 | preempt_latency_start(1); |
fc13aeba | 3579 | __schedule(true); |
47252cfb | 3580 | preempt_latency_stop(1); |
499d7955 | 3581 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3582 | |
3583 | /* | |
3584 | * Check again in case we missed a preemption opportunity | |
3585 | * between schedule and now. | |
3586 | */ | |
a18b5d01 FW |
3587 | } while (need_resched()); |
3588 | } | |
3589 | ||
1da177e4 LT |
3590 | #ifdef CONFIG_PREEMPT |
3591 | /* | |
2ed6e34f | 3592 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3593 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3594 | * occur there and call schedule directly. |
3595 | */ | |
722a9f92 | 3596 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3597 | { |
1da177e4 LT |
3598 | /* |
3599 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3600 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3601 | */ |
fbb00b56 | 3602 | if (likely(!preemptible())) |
1da177e4 LT |
3603 | return; |
3604 | ||
a18b5d01 | 3605 | preempt_schedule_common(); |
1da177e4 | 3606 | } |
376e2424 | 3607 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3608 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3609 | |
009f60e2 | 3610 | /** |
4eaca0a8 | 3611 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3612 | * |
3613 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3614 | * recursion and tracing preempt enabling caused by the tracing | |
3615 | * infrastructure itself. But as tracing can happen in areas coming | |
3616 | * from userspace or just about to enter userspace, a preempt enable | |
3617 | * can occur before user_exit() is called. This will cause the scheduler | |
3618 | * to be called when the system is still in usermode. | |
3619 | * | |
3620 | * To prevent this, the preempt_enable_notrace will use this function | |
3621 | * instead of preempt_schedule() to exit user context if needed before | |
3622 | * calling the scheduler. | |
3623 | */ | |
4eaca0a8 | 3624 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3625 | { |
3626 | enum ctx_state prev_ctx; | |
3627 | ||
3628 | if (likely(!preemptible())) | |
3629 | return; | |
3630 | ||
3631 | do { | |
47252cfb SR |
3632 | /* |
3633 | * Because the function tracer can trace preempt_count_sub() | |
3634 | * and it also uses preempt_enable/disable_notrace(), if | |
3635 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3636 | * by the function tracer will call this function again and | |
3637 | * cause infinite recursion. | |
3638 | * | |
3639 | * Preemption must be disabled here before the function | |
3640 | * tracer can trace. Break up preempt_disable() into two | |
3641 | * calls. One to disable preemption without fear of being | |
3642 | * traced. The other to still record the preemption latency, | |
3643 | * which can also be traced by the function tracer. | |
3644 | */ | |
3d8f74dd | 3645 | preempt_disable_notrace(); |
47252cfb | 3646 | preempt_latency_start(1); |
009f60e2 ON |
3647 | /* |
3648 | * Needs preempt disabled in case user_exit() is traced | |
3649 | * and the tracer calls preempt_enable_notrace() causing | |
3650 | * an infinite recursion. | |
3651 | */ | |
3652 | prev_ctx = exception_enter(); | |
fc13aeba | 3653 | __schedule(true); |
009f60e2 ON |
3654 | exception_exit(prev_ctx); |
3655 | ||
47252cfb | 3656 | preempt_latency_stop(1); |
3d8f74dd | 3657 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3658 | } while (need_resched()); |
3659 | } | |
4eaca0a8 | 3660 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3661 | |
32e475d7 | 3662 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3663 | |
3664 | /* | |
2ed6e34f | 3665 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3666 | * off of irq context. |
3667 | * Note, that this is called and return with irqs disabled. This will | |
3668 | * protect us against recursive calling from irq. | |
3669 | */ | |
722a9f92 | 3670 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3671 | { |
b22366cd | 3672 | enum ctx_state prev_state; |
6478d880 | 3673 | |
2ed6e34f | 3674 | /* Catch callers which need to be fixed */ |
f27dde8d | 3675 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3676 | |
b22366cd FW |
3677 | prev_state = exception_enter(); |
3678 | ||
3a5c359a | 3679 | do { |
3d8f74dd | 3680 | preempt_disable(); |
3a5c359a | 3681 | local_irq_enable(); |
fc13aeba | 3682 | __schedule(true); |
3a5c359a | 3683 | local_irq_disable(); |
3d8f74dd | 3684 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3685 | } while (need_resched()); |
b22366cd FW |
3686 | |
3687 | exception_exit(prev_state); | |
1da177e4 LT |
3688 | } |
3689 | ||
63859d4f | 3690 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3691 | void *key) |
1da177e4 | 3692 | { |
63859d4f | 3693 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3694 | } |
1da177e4 LT |
3695 | EXPORT_SYMBOL(default_wake_function); |
3696 | ||
b29739f9 IM |
3697 | #ifdef CONFIG_RT_MUTEXES |
3698 | ||
3699 | /* | |
3700 | * rt_mutex_setprio - set the current priority of a task | |
3701 | * @p: task | |
3702 | * @prio: prio value (kernel-internal form) | |
3703 | * | |
3704 | * This function changes the 'effective' priority of a task. It does | |
3705 | * not touch ->normal_prio like __setscheduler(). | |
3706 | * | |
c365c292 TG |
3707 | * Used by the rt_mutex code to implement priority inheritance |
3708 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3709 | */ |
36c8b586 | 3710 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3711 | { |
ff77e468 | 3712 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3713 | const struct sched_class *prev_class; |
eb580751 PZ |
3714 | struct rq_flags rf; |
3715 | struct rq *rq; | |
b29739f9 | 3716 | |
aab03e05 | 3717 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3718 | |
eb580751 | 3719 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 3720 | update_rq_clock(rq); |
b29739f9 | 3721 | |
1c4dd99b TG |
3722 | /* |
3723 | * Idle task boosting is a nono in general. There is one | |
3724 | * exception, when PREEMPT_RT and NOHZ is active: | |
3725 | * | |
3726 | * The idle task calls get_next_timer_interrupt() and holds | |
3727 | * the timer wheel base->lock on the CPU and another CPU wants | |
3728 | * to access the timer (probably to cancel it). We can safely | |
3729 | * ignore the boosting request, as the idle CPU runs this code | |
3730 | * with interrupts disabled and will complete the lock | |
3731 | * protected section without being interrupted. So there is no | |
3732 | * real need to boost. | |
3733 | */ | |
3734 | if (unlikely(p == rq->idle)) { | |
3735 | WARN_ON(p != rq->curr); | |
3736 | WARN_ON(p->pi_blocked_on); | |
3737 | goto out_unlock; | |
3738 | } | |
3739 | ||
a8027073 | 3740 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3741 | oldprio = p->prio; |
ff77e468 PZ |
3742 | |
3743 | if (oldprio == prio) | |
3744 | queue_flag &= ~DEQUEUE_MOVE; | |
3745 | ||
83ab0aa0 | 3746 | prev_class = p->sched_class; |
da0c1e65 | 3747 | queued = task_on_rq_queued(p); |
051a1d1a | 3748 | running = task_current(rq, p); |
da0c1e65 | 3749 | if (queued) |
ff77e468 | 3750 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3751 | if (running) |
f3cd1c4e | 3752 | put_prev_task(rq, p); |
dd41f596 | 3753 | |
2d3d891d DF |
3754 | /* |
3755 | * Boosting condition are: | |
3756 | * 1. -rt task is running and holds mutex A | |
3757 | * --> -dl task blocks on mutex A | |
3758 | * | |
3759 | * 2. -dl task is running and holds mutex A | |
3760 | * --> -dl task blocks on mutex A and could preempt the | |
3761 | * running task | |
3762 | */ | |
3763 | if (dl_prio(prio)) { | |
466af29b ON |
3764 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3765 | if (!dl_prio(p->normal_prio) || | |
3766 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3767 | p->dl.dl_boosted = 1; |
ff77e468 | 3768 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3769 | } else |
3770 | p->dl.dl_boosted = 0; | |
aab03e05 | 3771 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3772 | } else if (rt_prio(prio)) { |
3773 | if (dl_prio(oldprio)) | |
3774 | p->dl.dl_boosted = 0; | |
3775 | if (oldprio < prio) | |
ff77e468 | 3776 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3777 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3778 | } else { |
3779 | if (dl_prio(oldprio)) | |
3780 | p->dl.dl_boosted = 0; | |
746db944 BS |
3781 | if (rt_prio(oldprio)) |
3782 | p->rt.timeout = 0; | |
dd41f596 | 3783 | p->sched_class = &fair_sched_class; |
2d3d891d | 3784 | } |
dd41f596 | 3785 | |
b29739f9 IM |
3786 | p->prio = prio; |
3787 | ||
da0c1e65 | 3788 | if (queued) |
ff77e468 | 3789 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3790 | if (running) |
b2bf6c31 | 3791 | set_curr_task(rq, p); |
cb469845 | 3792 | |
da7a735e | 3793 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3794 | out_unlock: |
4c9a4bc8 | 3795 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 3796 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3797 | |
3798 | balance_callback(rq); | |
3799 | preempt_enable(); | |
b29739f9 | 3800 | } |
b29739f9 | 3801 | #endif |
d50dde5a | 3802 | |
36c8b586 | 3803 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3804 | { |
49bd21ef PZ |
3805 | bool queued, running; |
3806 | int old_prio, delta; | |
eb580751 | 3807 | struct rq_flags rf; |
70b97a7f | 3808 | struct rq *rq; |
1da177e4 | 3809 | |
75e45d51 | 3810 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3811 | return; |
3812 | /* | |
3813 | * We have to be careful, if called from sys_setpriority(), | |
3814 | * the task might be in the middle of scheduling on another CPU. | |
3815 | */ | |
eb580751 | 3816 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
3817 | update_rq_clock(rq); |
3818 | ||
1da177e4 LT |
3819 | /* |
3820 | * The RT priorities are set via sched_setscheduler(), but we still | |
3821 | * allow the 'normal' nice value to be set - but as expected | |
3822 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3823 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3824 | */ |
aab03e05 | 3825 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3826 | p->static_prio = NICE_TO_PRIO(nice); |
3827 | goto out_unlock; | |
3828 | } | |
da0c1e65 | 3829 | queued = task_on_rq_queued(p); |
49bd21ef | 3830 | running = task_current(rq, p); |
da0c1e65 | 3831 | if (queued) |
1de64443 | 3832 | dequeue_task(rq, p, DEQUEUE_SAVE); |
49bd21ef PZ |
3833 | if (running) |
3834 | put_prev_task(rq, p); | |
1da177e4 | 3835 | |
1da177e4 | 3836 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3837 | set_load_weight(p); |
b29739f9 IM |
3838 | old_prio = p->prio; |
3839 | p->prio = effective_prio(p); | |
3840 | delta = p->prio - old_prio; | |
1da177e4 | 3841 | |
da0c1e65 | 3842 | if (queued) { |
1de64443 | 3843 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3844 | /* |
d5f9f942 AM |
3845 | * If the task increased its priority or is running and |
3846 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3847 | */ |
d5f9f942 | 3848 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3849 | resched_curr(rq); |
1da177e4 | 3850 | } |
49bd21ef PZ |
3851 | if (running) |
3852 | set_curr_task(rq, p); | |
1da177e4 | 3853 | out_unlock: |
eb580751 | 3854 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3855 | } |
1da177e4 LT |
3856 | EXPORT_SYMBOL(set_user_nice); |
3857 | ||
e43379f1 MM |
3858 | /* |
3859 | * can_nice - check if a task can reduce its nice value | |
3860 | * @p: task | |
3861 | * @nice: nice value | |
3862 | */ | |
36c8b586 | 3863 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3864 | { |
024f4747 | 3865 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3866 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3867 | |
78d7d407 | 3868 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3869 | capable(CAP_SYS_NICE)); |
3870 | } | |
3871 | ||
1da177e4 LT |
3872 | #ifdef __ARCH_WANT_SYS_NICE |
3873 | ||
3874 | /* | |
3875 | * sys_nice - change the priority of the current process. | |
3876 | * @increment: priority increment | |
3877 | * | |
3878 | * sys_setpriority is a more generic, but much slower function that | |
3879 | * does similar things. | |
3880 | */ | |
5add95d4 | 3881 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3882 | { |
48f24c4d | 3883 | long nice, retval; |
1da177e4 LT |
3884 | |
3885 | /* | |
3886 | * Setpriority might change our priority at the same moment. | |
3887 | * We don't have to worry. Conceptually one call occurs first | |
3888 | * and we have a single winner. | |
3889 | */ | |
a9467fa3 | 3890 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3891 | nice = task_nice(current) + increment; |
1da177e4 | 3892 | |
a9467fa3 | 3893 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3894 | if (increment < 0 && !can_nice(current, nice)) |
3895 | return -EPERM; | |
3896 | ||
1da177e4 LT |
3897 | retval = security_task_setnice(current, nice); |
3898 | if (retval) | |
3899 | return retval; | |
3900 | ||
3901 | set_user_nice(current, nice); | |
3902 | return 0; | |
3903 | } | |
3904 | ||
3905 | #endif | |
3906 | ||
3907 | /** | |
3908 | * task_prio - return the priority value of a given task. | |
3909 | * @p: the task in question. | |
3910 | * | |
e69f6186 | 3911 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3912 | * RT tasks are offset by -200. Normal tasks are centered |
3913 | * around 0, value goes from -16 to +15. | |
3914 | */ | |
36c8b586 | 3915 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3916 | { |
3917 | return p->prio - MAX_RT_PRIO; | |
3918 | } | |
3919 | ||
1da177e4 LT |
3920 | /** |
3921 | * idle_cpu - is a given cpu idle currently? | |
3922 | * @cpu: the processor in question. | |
e69f6186 YB |
3923 | * |
3924 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3925 | */ |
3926 | int idle_cpu(int cpu) | |
3927 | { | |
908a3283 TG |
3928 | struct rq *rq = cpu_rq(cpu); |
3929 | ||
3930 | if (rq->curr != rq->idle) | |
3931 | return 0; | |
3932 | ||
3933 | if (rq->nr_running) | |
3934 | return 0; | |
3935 | ||
3936 | #ifdef CONFIG_SMP | |
3937 | if (!llist_empty(&rq->wake_list)) | |
3938 | return 0; | |
3939 | #endif | |
3940 | ||
3941 | return 1; | |
1da177e4 LT |
3942 | } |
3943 | ||
1da177e4 LT |
3944 | /** |
3945 | * idle_task - return the idle task for a given cpu. | |
3946 | * @cpu: the processor in question. | |
e69f6186 YB |
3947 | * |
3948 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3949 | */ |
36c8b586 | 3950 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3951 | { |
3952 | return cpu_rq(cpu)->idle; | |
3953 | } | |
3954 | ||
3955 | /** | |
3956 | * find_process_by_pid - find a process with a matching PID value. | |
3957 | * @pid: the pid in question. | |
e69f6186 YB |
3958 | * |
3959 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3960 | */ |
a9957449 | 3961 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3962 | { |
228ebcbe | 3963 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3964 | } |
3965 | ||
aab03e05 DF |
3966 | /* |
3967 | * This function initializes the sched_dl_entity of a newly becoming | |
3968 | * SCHED_DEADLINE task. | |
3969 | * | |
3970 | * Only the static values are considered here, the actual runtime and the | |
3971 | * absolute deadline will be properly calculated when the task is enqueued | |
3972 | * for the first time with its new policy. | |
3973 | */ | |
3974 | static void | |
3975 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3976 | { | |
3977 | struct sched_dl_entity *dl_se = &p->dl; | |
3978 | ||
aab03e05 DF |
3979 | dl_se->dl_runtime = attr->sched_runtime; |
3980 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3981 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3982 | dl_se->flags = attr->sched_flags; |
332ac17e | 3983 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3984 | |
3985 | /* | |
3986 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3987 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3988 | * | |
3989 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3990 | * point. This would include retaining the task_struct until that time | |
3991 | * and change dl_overflow() to not immediately decrement the current | |
3992 | * amount. | |
3993 | * | |
3994 | * Instead we retain the current runtime/deadline and let the new | |
3995 | * parameters take effect after the current reservation period lapses. | |
3996 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3997 | * before the current scheduling deadline. | |
3998 | * | |
3999 | * We can still have temporary overloads because we do not delay the | |
4000 | * change in bandwidth until that time; so admission control is | |
4001 | * not on the safe side. It does however guarantee tasks will never | |
4002 | * consume more than promised. | |
4003 | */ | |
aab03e05 DF |
4004 | } |
4005 | ||
c13db6b1 SR |
4006 | /* |
4007 | * sched_setparam() passes in -1 for its policy, to let the functions | |
4008 | * it calls know not to change it. | |
4009 | */ | |
4010 | #define SETPARAM_POLICY -1 | |
4011 | ||
c365c292 TG |
4012 | static void __setscheduler_params(struct task_struct *p, |
4013 | const struct sched_attr *attr) | |
1da177e4 | 4014 | { |
d50dde5a DF |
4015 | int policy = attr->sched_policy; |
4016 | ||
c13db6b1 | 4017 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
4018 | policy = p->policy; |
4019 | ||
1da177e4 | 4020 | p->policy = policy; |
d50dde5a | 4021 | |
aab03e05 DF |
4022 | if (dl_policy(policy)) |
4023 | __setparam_dl(p, attr); | |
39fd8fd2 | 4024 | else if (fair_policy(policy)) |
d50dde5a DF |
4025 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
4026 | ||
39fd8fd2 PZ |
4027 | /* |
4028 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
4029 | * !rt_policy. Always setting this ensures that things like | |
4030 | * getparam()/getattr() don't report silly values for !rt tasks. | |
4031 | */ | |
4032 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4033 | p->normal_prio = normal_prio(p); |
c365c292 TG |
4034 | set_load_weight(p); |
4035 | } | |
39fd8fd2 | 4036 | |
c365c292 TG |
4037 | /* Actually do priority change: must hold pi & rq lock. */ |
4038 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4039 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
4040 | { |
4041 | __setscheduler_params(p, attr); | |
d50dde5a | 4042 | |
383afd09 | 4043 | /* |
0782e63b TG |
4044 | * Keep a potential priority boosting if called from |
4045 | * sched_setscheduler(). | |
383afd09 | 4046 | */ |
0782e63b TG |
4047 | if (keep_boost) |
4048 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
4049 | else | |
4050 | p->prio = normal_prio(p); | |
383afd09 | 4051 | |
aab03e05 DF |
4052 | if (dl_prio(p->prio)) |
4053 | p->sched_class = &dl_sched_class; | |
4054 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4055 | p->sched_class = &rt_sched_class; |
4056 | else | |
4057 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4058 | } |
aab03e05 DF |
4059 | |
4060 | static void | |
4061 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
4062 | { | |
4063 | struct sched_dl_entity *dl_se = &p->dl; | |
4064 | ||
4065 | attr->sched_priority = p->rt_priority; | |
4066 | attr->sched_runtime = dl_se->dl_runtime; | |
4067 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 4068 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
4069 | attr->sched_flags = dl_se->flags; |
4070 | } | |
4071 | ||
4072 | /* | |
4073 | * This function validates the new parameters of a -deadline task. | |
4074 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 4075 | * than the runtime, as well as the period of being zero or |
332ac17e | 4076 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
4077 | * user parameters are above the internal resolution of 1us (we |
4078 | * check sched_runtime only since it is always the smaller one) and | |
4079 | * below 2^63 ns (we have to check both sched_deadline and | |
4080 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
4081 | */ |
4082 | static bool | |
4083 | __checkparam_dl(const struct sched_attr *attr) | |
4084 | { | |
b0827819 JL |
4085 | /* deadline != 0 */ |
4086 | if (attr->sched_deadline == 0) | |
4087 | return false; | |
4088 | ||
4089 | /* | |
4090 | * Since we truncate DL_SCALE bits, make sure we're at least | |
4091 | * that big. | |
4092 | */ | |
4093 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
4094 | return false; | |
4095 | ||
4096 | /* | |
4097 | * Since we use the MSB for wrap-around and sign issues, make | |
4098 | * sure it's not set (mind that period can be equal to zero). | |
4099 | */ | |
4100 | if (attr->sched_deadline & (1ULL << 63) || | |
4101 | attr->sched_period & (1ULL << 63)) | |
4102 | return false; | |
4103 | ||
4104 | /* runtime <= deadline <= period (if period != 0) */ | |
4105 | if ((attr->sched_period != 0 && | |
4106 | attr->sched_period < attr->sched_deadline) || | |
4107 | attr->sched_deadline < attr->sched_runtime) | |
4108 | return false; | |
4109 | ||
4110 | return true; | |
aab03e05 DF |
4111 | } |
4112 | ||
c69e8d9c DH |
4113 | /* |
4114 | * check the target process has a UID that matches the current process's | |
4115 | */ | |
4116 | static bool check_same_owner(struct task_struct *p) | |
4117 | { | |
4118 | const struct cred *cred = current_cred(), *pcred; | |
4119 | bool match; | |
4120 | ||
4121 | rcu_read_lock(); | |
4122 | pcred = __task_cred(p); | |
9c806aa0 EB |
4123 | match = (uid_eq(cred->euid, pcred->euid) || |
4124 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4125 | rcu_read_unlock(); |
4126 | return match; | |
4127 | } | |
4128 | ||
75381608 WL |
4129 | static bool dl_param_changed(struct task_struct *p, |
4130 | const struct sched_attr *attr) | |
4131 | { | |
4132 | struct sched_dl_entity *dl_se = &p->dl; | |
4133 | ||
4134 | if (dl_se->dl_runtime != attr->sched_runtime || | |
4135 | dl_se->dl_deadline != attr->sched_deadline || | |
4136 | dl_se->dl_period != attr->sched_period || | |
4137 | dl_se->flags != attr->sched_flags) | |
4138 | return true; | |
4139 | ||
4140 | return false; | |
4141 | } | |
4142 | ||
d50dde5a DF |
4143 | static int __sched_setscheduler(struct task_struct *p, |
4144 | const struct sched_attr *attr, | |
dbc7f069 | 4145 | bool user, bool pi) |
1da177e4 | 4146 | { |
383afd09 SR |
4147 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4148 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4149 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4150 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4151 | const struct sched_class *prev_class; |
eb580751 | 4152 | struct rq_flags rf; |
ca94c442 | 4153 | int reset_on_fork; |
ff77e468 | 4154 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 4155 | struct rq *rq; |
1da177e4 | 4156 | |
66e5393a SR |
4157 | /* may grab non-irq protected spin_locks */ |
4158 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4159 | recheck: |
4160 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4161 | if (policy < 0) { |
4162 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4163 | policy = oldpolicy = p->policy; |
ca94c442 | 4164 | } else { |
7479f3c9 | 4165 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4166 | |
20f9cd2a | 4167 | if (!valid_policy(policy)) |
ca94c442 LP |
4168 | return -EINVAL; |
4169 | } | |
4170 | ||
7479f3c9 PZ |
4171 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
4172 | return -EINVAL; | |
4173 | ||
1da177e4 LT |
4174 | /* |
4175 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4176 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4177 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4178 | */ |
0bb040a4 | 4179 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4180 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4181 | return -EINVAL; |
aab03e05 DF |
4182 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4183 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4184 | return -EINVAL; |
4185 | ||
37e4ab3f OC |
4186 | /* |
4187 | * Allow unprivileged RT tasks to decrease priority: | |
4188 | */ | |
961ccddd | 4189 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4190 | if (fair_policy(policy)) { |
d0ea0268 | 4191 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4192 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4193 | return -EPERM; |
4194 | } | |
4195 | ||
e05606d3 | 4196 | if (rt_policy(policy)) { |
a44702e8 ON |
4197 | unsigned long rlim_rtprio = |
4198 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4199 | |
4200 | /* can't set/change the rt policy */ | |
4201 | if (policy != p->policy && !rlim_rtprio) | |
4202 | return -EPERM; | |
4203 | ||
4204 | /* can't increase priority */ | |
d50dde5a DF |
4205 | if (attr->sched_priority > p->rt_priority && |
4206 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4207 | return -EPERM; |
4208 | } | |
c02aa73b | 4209 | |
d44753b8 JL |
4210 | /* |
4211 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4212 | * (safest behavior); in the future we would like to allow | |
4213 | * unprivileged DL tasks to increase their relative deadline | |
4214 | * or reduce their runtime (both ways reducing utilization) | |
4215 | */ | |
4216 | if (dl_policy(policy)) | |
4217 | return -EPERM; | |
4218 | ||
dd41f596 | 4219 | /* |
c02aa73b DH |
4220 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4221 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4222 | */ |
20f9cd2a | 4223 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4224 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4225 | return -EPERM; |
4226 | } | |
5fe1d75f | 4227 | |
37e4ab3f | 4228 | /* can't change other user's priorities */ |
c69e8d9c | 4229 | if (!check_same_owner(p)) |
37e4ab3f | 4230 | return -EPERM; |
ca94c442 LP |
4231 | |
4232 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4233 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4234 | return -EPERM; | |
37e4ab3f | 4235 | } |
1da177e4 | 4236 | |
725aad24 | 4237 | if (user) { |
b0ae1981 | 4238 | retval = security_task_setscheduler(p); |
725aad24 JF |
4239 | if (retval) |
4240 | return retval; | |
4241 | } | |
4242 | ||
b29739f9 IM |
4243 | /* |
4244 | * make sure no PI-waiters arrive (or leave) while we are | |
4245 | * changing the priority of the task: | |
0122ec5b | 4246 | * |
25985edc | 4247 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4248 | * runqueue lock must be held. |
4249 | */ | |
eb580751 | 4250 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4251 | update_rq_clock(rq); |
dc61b1d6 | 4252 | |
34f971f6 PZ |
4253 | /* |
4254 | * Changing the policy of the stop threads its a very bad idea | |
4255 | */ | |
4256 | if (p == rq->stop) { | |
eb580751 | 4257 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4258 | return -EINVAL; |
4259 | } | |
4260 | ||
a51e9198 | 4261 | /* |
d6b1e911 TG |
4262 | * If not changing anything there's no need to proceed further, |
4263 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4264 | */ |
d50dde5a | 4265 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4266 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4267 | goto change; |
4268 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4269 | goto change; | |
75381608 | 4270 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4271 | goto change; |
d50dde5a | 4272 | |
d6b1e911 | 4273 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4274 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4275 | return 0; |
4276 | } | |
d50dde5a | 4277 | change: |
a51e9198 | 4278 | |
dc61b1d6 | 4279 | if (user) { |
332ac17e | 4280 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4281 | /* |
4282 | * Do not allow realtime tasks into groups that have no runtime | |
4283 | * assigned. | |
4284 | */ | |
4285 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4286 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4287 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4288 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4289 | return -EPERM; |
4290 | } | |
dc61b1d6 | 4291 | #endif |
332ac17e DF |
4292 | #ifdef CONFIG_SMP |
4293 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4294 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4295 | |
4296 | /* | |
4297 | * Don't allow tasks with an affinity mask smaller than | |
4298 | * the entire root_domain to become SCHED_DEADLINE. We | |
4299 | * will also fail if there's no bandwidth available. | |
4300 | */ | |
e4099a5e PZ |
4301 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4302 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4303 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4304 | return -EPERM; |
4305 | } | |
4306 | } | |
4307 | #endif | |
4308 | } | |
dc61b1d6 | 4309 | |
1da177e4 LT |
4310 | /* recheck policy now with rq lock held */ |
4311 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4312 | policy = oldpolicy = -1; | |
eb580751 | 4313 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4314 | goto recheck; |
4315 | } | |
332ac17e DF |
4316 | |
4317 | /* | |
4318 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4319 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4320 | * is available. | |
4321 | */ | |
e4099a5e | 4322 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4323 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4324 | return -EBUSY; |
4325 | } | |
4326 | ||
c365c292 TG |
4327 | p->sched_reset_on_fork = reset_on_fork; |
4328 | oldprio = p->prio; | |
4329 | ||
dbc7f069 PZ |
4330 | if (pi) { |
4331 | /* | |
4332 | * Take priority boosted tasks into account. If the new | |
4333 | * effective priority is unchanged, we just store the new | |
4334 | * normal parameters and do not touch the scheduler class and | |
4335 | * the runqueue. This will be done when the task deboost | |
4336 | * itself. | |
4337 | */ | |
4338 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4339 | if (new_effective_prio == oldprio) |
4340 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4341 | } |
4342 | ||
da0c1e65 | 4343 | queued = task_on_rq_queued(p); |
051a1d1a | 4344 | running = task_current(rq, p); |
da0c1e65 | 4345 | if (queued) |
ff77e468 | 4346 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4347 | if (running) |
f3cd1c4e | 4348 | put_prev_task(rq, p); |
f6b53205 | 4349 | |
83ab0aa0 | 4350 | prev_class = p->sched_class; |
dbc7f069 | 4351 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4352 | |
da0c1e65 | 4353 | if (queued) { |
81a44c54 TG |
4354 | /* |
4355 | * We enqueue to tail when the priority of a task is | |
4356 | * increased (user space view). | |
4357 | */ | |
ff77e468 PZ |
4358 | if (oldprio < p->prio) |
4359 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4360 | |
ff77e468 | 4361 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4362 | } |
a399d233 | 4363 | if (running) |
b2bf6c31 | 4364 | set_curr_task(rq, p); |
cb469845 | 4365 | |
da7a735e | 4366 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4367 | preempt_disable(); /* avoid rq from going away on us */ |
eb580751 | 4368 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4369 | |
dbc7f069 PZ |
4370 | if (pi) |
4371 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4372 | |
4c9a4bc8 PZ |
4373 | /* |
4374 | * Run balance callbacks after we've adjusted the PI chain. | |
4375 | */ | |
4376 | balance_callback(rq); | |
4377 | preempt_enable(); | |
95e02ca9 | 4378 | |
1da177e4 LT |
4379 | return 0; |
4380 | } | |
961ccddd | 4381 | |
7479f3c9 PZ |
4382 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4383 | const struct sched_param *param, bool check) | |
4384 | { | |
4385 | struct sched_attr attr = { | |
4386 | .sched_policy = policy, | |
4387 | .sched_priority = param->sched_priority, | |
4388 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4389 | }; | |
4390 | ||
c13db6b1 SR |
4391 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4392 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4393 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4394 | policy &= ~SCHED_RESET_ON_FORK; | |
4395 | attr.sched_policy = policy; | |
4396 | } | |
4397 | ||
dbc7f069 | 4398 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4399 | } |
961ccddd RR |
4400 | /** |
4401 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4402 | * @p: the task in question. | |
4403 | * @policy: new policy. | |
4404 | * @param: structure containing the new RT priority. | |
4405 | * | |
e69f6186 YB |
4406 | * Return: 0 on success. An error code otherwise. |
4407 | * | |
961ccddd RR |
4408 | * NOTE that the task may be already dead. |
4409 | */ | |
4410 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4411 | const struct sched_param *param) |
961ccddd | 4412 | { |
7479f3c9 | 4413 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4414 | } |
1da177e4 LT |
4415 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4416 | ||
d50dde5a DF |
4417 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4418 | { | |
dbc7f069 | 4419 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4420 | } |
4421 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4422 | ||
961ccddd RR |
4423 | /** |
4424 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4425 | * @p: the task in question. | |
4426 | * @policy: new policy. | |
4427 | * @param: structure containing the new RT priority. | |
4428 | * | |
4429 | * Just like sched_setscheduler, only don't bother checking if the | |
4430 | * current context has permission. For example, this is needed in | |
4431 | * stop_machine(): we create temporary high priority worker threads, | |
4432 | * but our caller might not have that capability. | |
e69f6186 YB |
4433 | * |
4434 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4435 | */ |
4436 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4437 | const struct sched_param *param) |
961ccddd | 4438 | { |
7479f3c9 | 4439 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4440 | } |
84778472 | 4441 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4442 | |
95cdf3b7 IM |
4443 | static int |
4444 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4445 | { |
1da177e4 LT |
4446 | struct sched_param lparam; |
4447 | struct task_struct *p; | |
36c8b586 | 4448 | int retval; |
1da177e4 LT |
4449 | |
4450 | if (!param || pid < 0) | |
4451 | return -EINVAL; | |
4452 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4453 | return -EFAULT; | |
5fe1d75f ON |
4454 | |
4455 | rcu_read_lock(); | |
4456 | retval = -ESRCH; | |
1da177e4 | 4457 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4458 | if (p != NULL) |
4459 | retval = sched_setscheduler(p, policy, &lparam); | |
4460 | rcu_read_unlock(); | |
36c8b586 | 4461 | |
1da177e4 LT |
4462 | return retval; |
4463 | } | |
4464 | ||
d50dde5a DF |
4465 | /* |
4466 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4467 | */ | |
4468 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4469 | struct sched_attr *attr) | |
4470 | { | |
4471 | u32 size; | |
4472 | int ret; | |
4473 | ||
4474 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4475 | return -EFAULT; | |
4476 | ||
4477 | /* | |
4478 | * zero the full structure, so that a short copy will be nice. | |
4479 | */ | |
4480 | memset(attr, 0, sizeof(*attr)); | |
4481 | ||
4482 | ret = get_user(size, &uattr->size); | |
4483 | if (ret) | |
4484 | return ret; | |
4485 | ||
4486 | if (size > PAGE_SIZE) /* silly large */ | |
4487 | goto err_size; | |
4488 | ||
4489 | if (!size) /* abi compat */ | |
4490 | size = SCHED_ATTR_SIZE_VER0; | |
4491 | ||
4492 | if (size < SCHED_ATTR_SIZE_VER0) | |
4493 | goto err_size; | |
4494 | ||
4495 | /* | |
4496 | * If we're handed a bigger struct than we know of, | |
4497 | * ensure all the unknown bits are 0 - i.e. new | |
4498 | * user-space does not rely on any kernel feature | |
4499 | * extensions we dont know about yet. | |
4500 | */ | |
4501 | if (size > sizeof(*attr)) { | |
4502 | unsigned char __user *addr; | |
4503 | unsigned char __user *end; | |
4504 | unsigned char val; | |
4505 | ||
4506 | addr = (void __user *)uattr + sizeof(*attr); | |
4507 | end = (void __user *)uattr + size; | |
4508 | ||
4509 | for (; addr < end; addr++) { | |
4510 | ret = get_user(val, addr); | |
4511 | if (ret) | |
4512 | return ret; | |
4513 | if (val) | |
4514 | goto err_size; | |
4515 | } | |
4516 | size = sizeof(*attr); | |
4517 | } | |
4518 | ||
4519 | ret = copy_from_user(attr, uattr, size); | |
4520 | if (ret) | |
4521 | return -EFAULT; | |
4522 | ||
4523 | /* | |
4524 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4525 | * to be strict and return an error on out-of-bounds values? | |
4526 | */ | |
75e45d51 | 4527 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4528 | |
e78c7bca | 4529 | return 0; |
d50dde5a DF |
4530 | |
4531 | err_size: | |
4532 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4533 | return -E2BIG; |
d50dde5a DF |
4534 | } |
4535 | ||
1da177e4 LT |
4536 | /** |
4537 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4538 | * @pid: the pid in question. | |
4539 | * @policy: new policy. | |
4540 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4541 | * |
4542 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4543 | */ |
5add95d4 HC |
4544 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4545 | struct sched_param __user *, param) | |
1da177e4 | 4546 | { |
c21761f1 JB |
4547 | /* negative values for policy are not valid */ |
4548 | if (policy < 0) | |
4549 | return -EINVAL; | |
4550 | ||
1da177e4 LT |
4551 | return do_sched_setscheduler(pid, policy, param); |
4552 | } | |
4553 | ||
4554 | /** | |
4555 | * sys_sched_setparam - set/change the RT priority of a thread | |
4556 | * @pid: the pid in question. | |
4557 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4558 | * |
4559 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4560 | */ |
5add95d4 | 4561 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4562 | { |
c13db6b1 | 4563 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4564 | } |
4565 | ||
d50dde5a DF |
4566 | /** |
4567 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4568 | * @pid: the pid in question. | |
5778fccf | 4569 | * @uattr: structure containing the extended parameters. |
db66d756 | 4570 | * @flags: for future extension. |
d50dde5a | 4571 | */ |
6d35ab48 PZ |
4572 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4573 | unsigned int, flags) | |
d50dde5a DF |
4574 | { |
4575 | struct sched_attr attr; | |
4576 | struct task_struct *p; | |
4577 | int retval; | |
4578 | ||
6d35ab48 | 4579 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4580 | return -EINVAL; |
4581 | ||
143cf23d MK |
4582 | retval = sched_copy_attr(uattr, &attr); |
4583 | if (retval) | |
4584 | return retval; | |
d50dde5a | 4585 | |
b14ed2c2 | 4586 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4587 | return -EINVAL; |
d50dde5a DF |
4588 | |
4589 | rcu_read_lock(); | |
4590 | retval = -ESRCH; | |
4591 | p = find_process_by_pid(pid); | |
4592 | if (p != NULL) | |
4593 | retval = sched_setattr(p, &attr); | |
4594 | rcu_read_unlock(); | |
4595 | ||
4596 | return retval; | |
4597 | } | |
4598 | ||
1da177e4 LT |
4599 | /** |
4600 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4601 | * @pid: the pid in question. | |
e69f6186 YB |
4602 | * |
4603 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4604 | * code. | |
1da177e4 | 4605 | */ |
5add95d4 | 4606 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4607 | { |
36c8b586 | 4608 | struct task_struct *p; |
3a5c359a | 4609 | int retval; |
1da177e4 LT |
4610 | |
4611 | if (pid < 0) | |
3a5c359a | 4612 | return -EINVAL; |
1da177e4 LT |
4613 | |
4614 | retval = -ESRCH; | |
5fe85be0 | 4615 | rcu_read_lock(); |
1da177e4 LT |
4616 | p = find_process_by_pid(pid); |
4617 | if (p) { | |
4618 | retval = security_task_getscheduler(p); | |
4619 | if (!retval) | |
ca94c442 LP |
4620 | retval = p->policy |
4621 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4622 | } |
5fe85be0 | 4623 | rcu_read_unlock(); |
1da177e4 LT |
4624 | return retval; |
4625 | } | |
4626 | ||
4627 | /** | |
ca94c442 | 4628 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4629 | * @pid: the pid in question. |
4630 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4631 | * |
4632 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4633 | * code. | |
1da177e4 | 4634 | */ |
5add95d4 | 4635 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4636 | { |
ce5f7f82 | 4637 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4638 | struct task_struct *p; |
3a5c359a | 4639 | int retval; |
1da177e4 LT |
4640 | |
4641 | if (!param || pid < 0) | |
3a5c359a | 4642 | return -EINVAL; |
1da177e4 | 4643 | |
5fe85be0 | 4644 | rcu_read_lock(); |
1da177e4 LT |
4645 | p = find_process_by_pid(pid); |
4646 | retval = -ESRCH; | |
4647 | if (!p) | |
4648 | goto out_unlock; | |
4649 | ||
4650 | retval = security_task_getscheduler(p); | |
4651 | if (retval) | |
4652 | goto out_unlock; | |
4653 | ||
ce5f7f82 PZ |
4654 | if (task_has_rt_policy(p)) |
4655 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4656 | rcu_read_unlock(); |
1da177e4 LT |
4657 | |
4658 | /* | |
4659 | * This one might sleep, we cannot do it with a spinlock held ... | |
4660 | */ | |
4661 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4662 | ||
1da177e4 LT |
4663 | return retval; |
4664 | ||
4665 | out_unlock: | |
5fe85be0 | 4666 | rcu_read_unlock(); |
1da177e4 LT |
4667 | return retval; |
4668 | } | |
4669 | ||
d50dde5a DF |
4670 | static int sched_read_attr(struct sched_attr __user *uattr, |
4671 | struct sched_attr *attr, | |
4672 | unsigned int usize) | |
4673 | { | |
4674 | int ret; | |
4675 | ||
4676 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4677 | return -EFAULT; | |
4678 | ||
4679 | /* | |
4680 | * If we're handed a smaller struct than we know of, | |
4681 | * ensure all the unknown bits are 0 - i.e. old | |
4682 | * user-space does not get uncomplete information. | |
4683 | */ | |
4684 | if (usize < sizeof(*attr)) { | |
4685 | unsigned char *addr; | |
4686 | unsigned char *end; | |
4687 | ||
4688 | addr = (void *)attr + usize; | |
4689 | end = (void *)attr + sizeof(*attr); | |
4690 | ||
4691 | for (; addr < end; addr++) { | |
4692 | if (*addr) | |
22400674 | 4693 | return -EFBIG; |
d50dde5a DF |
4694 | } |
4695 | ||
4696 | attr->size = usize; | |
4697 | } | |
4698 | ||
4efbc454 | 4699 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4700 | if (ret) |
4701 | return -EFAULT; | |
4702 | ||
22400674 | 4703 | return 0; |
d50dde5a DF |
4704 | } |
4705 | ||
4706 | /** | |
aab03e05 | 4707 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4708 | * @pid: the pid in question. |
5778fccf | 4709 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4710 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4711 | * @flags: for future extension. |
d50dde5a | 4712 | */ |
6d35ab48 PZ |
4713 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4714 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4715 | { |
4716 | struct sched_attr attr = { | |
4717 | .size = sizeof(struct sched_attr), | |
4718 | }; | |
4719 | struct task_struct *p; | |
4720 | int retval; | |
4721 | ||
4722 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4723 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4724 | return -EINVAL; |
4725 | ||
4726 | rcu_read_lock(); | |
4727 | p = find_process_by_pid(pid); | |
4728 | retval = -ESRCH; | |
4729 | if (!p) | |
4730 | goto out_unlock; | |
4731 | ||
4732 | retval = security_task_getscheduler(p); | |
4733 | if (retval) | |
4734 | goto out_unlock; | |
4735 | ||
4736 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4737 | if (p->sched_reset_on_fork) |
4738 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4739 | if (task_has_dl_policy(p)) |
4740 | __getparam_dl(p, &attr); | |
4741 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4742 | attr.sched_priority = p->rt_priority; |
4743 | else | |
d0ea0268 | 4744 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4745 | |
4746 | rcu_read_unlock(); | |
4747 | ||
4748 | retval = sched_read_attr(uattr, &attr, size); | |
4749 | return retval; | |
4750 | ||
4751 | out_unlock: | |
4752 | rcu_read_unlock(); | |
4753 | return retval; | |
4754 | } | |
4755 | ||
96f874e2 | 4756 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4757 | { |
5a16f3d3 | 4758 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4759 | struct task_struct *p; |
4760 | int retval; | |
1da177e4 | 4761 | |
23f5d142 | 4762 | rcu_read_lock(); |
1da177e4 LT |
4763 | |
4764 | p = find_process_by_pid(pid); | |
4765 | if (!p) { | |
23f5d142 | 4766 | rcu_read_unlock(); |
1da177e4 LT |
4767 | return -ESRCH; |
4768 | } | |
4769 | ||
23f5d142 | 4770 | /* Prevent p going away */ |
1da177e4 | 4771 | get_task_struct(p); |
23f5d142 | 4772 | rcu_read_unlock(); |
1da177e4 | 4773 | |
14a40ffc TH |
4774 | if (p->flags & PF_NO_SETAFFINITY) { |
4775 | retval = -EINVAL; | |
4776 | goto out_put_task; | |
4777 | } | |
5a16f3d3 RR |
4778 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4779 | retval = -ENOMEM; | |
4780 | goto out_put_task; | |
4781 | } | |
4782 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4783 | retval = -ENOMEM; | |
4784 | goto out_free_cpus_allowed; | |
4785 | } | |
1da177e4 | 4786 | retval = -EPERM; |
4c44aaaf EB |
4787 | if (!check_same_owner(p)) { |
4788 | rcu_read_lock(); | |
4789 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4790 | rcu_read_unlock(); | |
16303ab2 | 4791 | goto out_free_new_mask; |
4c44aaaf EB |
4792 | } |
4793 | rcu_read_unlock(); | |
4794 | } | |
1da177e4 | 4795 | |
b0ae1981 | 4796 | retval = security_task_setscheduler(p); |
e7834f8f | 4797 | if (retval) |
16303ab2 | 4798 | goto out_free_new_mask; |
e7834f8f | 4799 | |
e4099a5e PZ |
4800 | |
4801 | cpuset_cpus_allowed(p, cpus_allowed); | |
4802 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4803 | ||
332ac17e DF |
4804 | /* |
4805 | * Since bandwidth control happens on root_domain basis, | |
4806 | * if admission test is enabled, we only admit -deadline | |
4807 | * tasks allowed to run on all the CPUs in the task's | |
4808 | * root_domain. | |
4809 | */ | |
4810 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4811 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4812 | rcu_read_lock(); | |
4813 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4814 | retval = -EBUSY; |
f1e3a093 | 4815 | rcu_read_unlock(); |
16303ab2 | 4816 | goto out_free_new_mask; |
332ac17e | 4817 | } |
f1e3a093 | 4818 | rcu_read_unlock(); |
332ac17e DF |
4819 | } |
4820 | #endif | |
49246274 | 4821 | again: |
25834c73 | 4822 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4823 | |
8707d8b8 | 4824 | if (!retval) { |
5a16f3d3 RR |
4825 | cpuset_cpus_allowed(p, cpus_allowed); |
4826 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4827 | /* |
4828 | * We must have raced with a concurrent cpuset | |
4829 | * update. Just reset the cpus_allowed to the | |
4830 | * cpuset's cpus_allowed | |
4831 | */ | |
5a16f3d3 | 4832 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4833 | goto again; |
4834 | } | |
4835 | } | |
16303ab2 | 4836 | out_free_new_mask: |
5a16f3d3 RR |
4837 | free_cpumask_var(new_mask); |
4838 | out_free_cpus_allowed: | |
4839 | free_cpumask_var(cpus_allowed); | |
4840 | out_put_task: | |
1da177e4 | 4841 | put_task_struct(p); |
1da177e4 LT |
4842 | return retval; |
4843 | } | |
4844 | ||
4845 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4846 | struct cpumask *new_mask) |
1da177e4 | 4847 | { |
96f874e2 RR |
4848 | if (len < cpumask_size()) |
4849 | cpumask_clear(new_mask); | |
4850 | else if (len > cpumask_size()) | |
4851 | len = cpumask_size(); | |
4852 | ||
1da177e4 LT |
4853 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4854 | } | |
4855 | ||
4856 | /** | |
4857 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4858 | * @pid: pid of the process | |
4859 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4860 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4861 | * |
4862 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4863 | */ |
5add95d4 HC |
4864 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4865 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4866 | { |
5a16f3d3 | 4867 | cpumask_var_t new_mask; |
1da177e4 LT |
4868 | int retval; |
4869 | ||
5a16f3d3 RR |
4870 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4871 | return -ENOMEM; | |
1da177e4 | 4872 | |
5a16f3d3 RR |
4873 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4874 | if (retval == 0) | |
4875 | retval = sched_setaffinity(pid, new_mask); | |
4876 | free_cpumask_var(new_mask); | |
4877 | return retval; | |
1da177e4 LT |
4878 | } |
4879 | ||
96f874e2 | 4880 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4881 | { |
36c8b586 | 4882 | struct task_struct *p; |
31605683 | 4883 | unsigned long flags; |
1da177e4 | 4884 | int retval; |
1da177e4 | 4885 | |
23f5d142 | 4886 | rcu_read_lock(); |
1da177e4 LT |
4887 | |
4888 | retval = -ESRCH; | |
4889 | p = find_process_by_pid(pid); | |
4890 | if (!p) | |
4891 | goto out_unlock; | |
4892 | ||
e7834f8f DQ |
4893 | retval = security_task_getscheduler(p); |
4894 | if (retval) | |
4895 | goto out_unlock; | |
4896 | ||
013fdb80 | 4897 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4898 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4899 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4900 | |
4901 | out_unlock: | |
23f5d142 | 4902 | rcu_read_unlock(); |
1da177e4 | 4903 | |
9531b62f | 4904 | return retval; |
1da177e4 LT |
4905 | } |
4906 | ||
4907 | /** | |
4908 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4909 | * @pid: pid of the process | |
4910 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4911 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 | 4912 | * |
599b4840 ZW |
4913 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4914 | * error code otherwise. | |
1da177e4 | 4915 | */ |
5add95d4 HC |
4916 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4917 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4918 | { |
4919 | int ret; | |
f17c8607 | 4920 | cpumask_var_t mask; |
1da177e4 | 4921 | |
84fba5ec | 4922 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4923 | return -EINVAL; |
4924 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4925 | return -EINVAL; |
4926 | ||
f17c8607 RR |
4927 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4928 | return -ENOMEM; | |
1da177e4 | 4929 | |
f17c8607 RR |
4930 | ret = sched_getaffinity(pid, mask); |
4931 | if (ret == 0) { | |
8bc037fb | 4932 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4933 | |
4934 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4935 | ret = -EFAULT; |
4936 | else | |
cd3d8031 | 4937 | ret = retlen; |
f17c8607 RR |
4938 | } |
4939 | free_cpumask_var(mask); | |
1da177e4 | 4940 | |
f17c8607 | 4941 | return ret; |
1da177e4 LT |
4942 | } |
4943 | ||
4944 | /** | |
4945 | * sys_sched_yield - yield the current processor to other threads. | |
4946 | * | |
dd41f596 IM |
4947 | * This function yields the current CPU to other tasks. If there are no |
4948 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4949 | * |
4950 | * Return: 0. | |
1da177e4 | 4951 | */ |
5add95d4 | 4952 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4953 | { |
70b97a7f | 4954 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4955 | |
ae92882e | 4956 | schedstat_inc(rq->yld_count); |
4530d7ab | 4957 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4958 | |
4959 | /* | |
4960 | * Since we are going to call schedule() anyway, there's | |
4961 | * no need to preempt or enable interrupts: | |
4962 | */ | |
4963 | __release(rq->lock); | |
8a25d5de | 4964 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4965 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4966 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4967 | |
4968 | schedule(); | |
4969 | ||
4970 | return 0; | |
4971 | } | |
4972 | ||
35a773a0 | 4973 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4974 | int __sched _cond_resched(void) |
1da177e4 | 4975 | { |
fe32d3cd | 4976 | if (should_resched(0)) { |
a18b5d01 | 4977 | preempt_schedule_common(); |
1da177e4 LT |
4978 | return 1; |
4979 | } | |
4980 | return 0; | |
4981 | } | |
02b67cc3 | 4982 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4983 | #endif |
1da177e4 LT |
4984 | |
4985 | /* | |
613afbf8 | 4986 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4987 | * call schedule, and on return reacquire the lock. |
4988 | * | |
41a2d6cf | 4989 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4990 | * operations here to prevent schedule() from being called twice (once via |
4991 | * spin_unlock(), once by hand). | |
4992 | */ | |
613afbf8 | 4993 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4994 | { |
fe32d3cd | 4995 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4996 | int ret = 0; |
4997 | ||
f607c668 PZ |
4998 | lockdep_assert_held(lock); |
4999 | ||
4a81e832 | 5000 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5001 | spin_unlock(lock); |
d86ee480 | 5002 | if (resched) |
a18b5d01 | 5003 | preempt_schedule_common(); |
95c354fe NP |
5004 | else |
5005 | cpu_relax(); | |
6df3cecb | 5006 | ret = 1; |
1da177e4 | 5007 | spin_lock(lock); |
1da177e4 | 5008 | } |
6df3cecb | 5009 | return ret; |
1da177e4 | 5010 | } |
613afbf8 | 5011 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5012 | |
613afbf8 | 5013 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5014 | { |
5015 | BUG_ON(!in_softirq()); | |
5016 | ||
fe32d3cd | 5017 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 5018 | local_bh_enable(); |
a18b5d01 | 5019 | preempt_schedule_common(); |
1da177e4 LT |
5020 | local_bh_disable(); |
5021 | return 1; | |
5022 | } | |
5023 | return 0; | |
5024 | } | |
613afbf8 | 5025 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5026 | |
1da177e4 LT |
5027 | /** |
5028 | * yield - yield the current processor to other threads. | |
5029 | * | |
8e3fabfd PZ |
5030 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
5031 | * | |
5032 | * The scheduler is at all times free to pick the calling task as the most | |
5033 | * eligible task to run, if removing the yield() call from your code breaks | |
5034 | * it, its already broken. | |
5035 | * | |
5036 | * Typical broken usage is: | |
5037 | * | |
5038 | * while (!event) | |
5039 | * yield(); | |
5040 | * | |
5041 | * where one assumes that yield() will let 'the other' process run that will | |
5042 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5043 | * happen. Never use yield() as a progress guarantee!! | |
5044 | * | |
5045 | * If you want to use yield() to wait for something, use wait_event(). | |
5046 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5047 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5048 | */ |
5049 | void __sched yield(void) | |
5050 | { | |
5051 | set_current_state(TASK_RUNNING); | |
5052 | sys_sched_yield(); | |
5053 | } | |
1da177e4 LT |
5054 | EXPORT_SYMBOL(yield); |
5055 | ||
d95f4122 MG |
5056 | /** |
5057 | * yield_to - yield the current processor to another thread in | |
5058 | * your thread group, or accelerate that thread toward the | |
5059 | * processor it's on. | |
16addf95 RD |
5060 | * @p: target task |
5061 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5062 | * |
5063 | * It's the caller's job to ensure that the target task struct | |
5064 | * can't go away on us before we can do any checks. | |
5065 | * | |
e69f6186 | 5066 | * Return: |
7b270f60 PZ |
5067 | * true (>0) if we indeed boosted the target task. |
5068 | * false (0) if we failed to boost the target. | |
5069 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5070 | */ |
fa93384f | 5071 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5072 | { |
5073 | struct task_struct *curr = current; | |
5074 | struct rq *rq, *p_rq; | |
5075 | unsigned long flags; | |
c3c18640 | 5076 | int yielded = 0; |
d95f4122 MG |
5077 | |
5078 | local_irq_save(flags); | |
5079 | rq = this_rq(); | |
5080 | ||
5081 | again: | |
5082 | p_rq = task_rq(p); | |
7b270f60 PZ |
5083 | /* |
5084 | * If we're the only runnable task on the rq and target rq also | |
5085 | * has only one task, there's absolutely no point in yielding. | |
5086 | */ | |
5087 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5088 | yielded = -ESRCH; | |
5089 | goto out_irq; | |
5090 | } | |
5091 | ||
d95f4122 | 5092 | double_rq_lock(rq, p_rq); |
39e24d8f | 5093 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5094 | double_rq_unlock(rq, p_rq); |
5095 | goto again; | |
5096 | } | |
5097 | ||
5098 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5099 | goto out_unlock; |
d95f4122 MG |
5100 | |
5101 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5102 | goto out_unlock; |
d95f4122 MG |
5103 | |
5104 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5105 | goto out_unlock; |
d95f4122 MG |
5106 | |
5107 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5108 | if (yielded) { |
ae92882e | 5109 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5110 | /* |
5111 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5112 | * fairness. | |
5113 | */ | |
5114 | if (preempt && rq != p_rq) | |
8875125e | 5115 | resched_curr(p_rq); |
6d1cafd8 | 5116 | } |
d95f4122 | 5117 | |
7b270f60 | 5118 | out_unlock: |
d95f4122 | 5119 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5120 | out_irq: |
d95f4122 MG |
5121 | local_irq_restore(flags); |
5122 | ||
7b270f60 | 5123 | if (yielded > 0) |
d95f4122 MG |
5124 | schedule(); |
5125 | ||
5126 | return yielded; | |
5127 | } | |
5128 | EXPORT_SYMBOL_GPL(yield_to); | |
5129 | ||
10ab5643 TH |
5130 | int io_schedule_prepare(void) |
5131 | { | |
5132 | int old_iowait = current->in_iowait; | |
5133 | ||
5134 | current->in_iowait = 1; | |
5135 | blk_schedule_flush_plug(current); | |
5136 | ||
5137 | return old_iowait; | |
5138 | } | |
5139 | ||
5140 | void io_schedule_finish(int token) | |
5141 | { | |
5142 | current->in_iowait = token; | |
5143 | } | |
5144 | ||
1da177e4 | 5145 | /* |
41a2d6cf | 5146 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5147 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5148 | */ |
1da177e4 LT |
5149 | long __sched io_schedule_timeout(long timeout) |
5150 | { | |
10ab5643 | 5151 | int token; |
1da177e4 LT |
5152 | long ret; |
5153 | ||
10ab5643 | 5154 | token = io_schedule_prepare(); |
1da177e4 | 5155 | ret = schedule_timeout(timeout); |
10ab5643 | 5156 | io_schedule_finish(token); |
9cff8ade | 5157 | |
1da177e4 LT |
5158 | return ret; |
5159 | } | |
9cff8ade | 5160 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5161 | |
10ab5643 TH |
5162 | void io_schedule(void) |
5163 | { | |
5164 | int token; | |
5165 | ||
5166 | token = io_schedule_prepare(); | |
5167 | schedule(); | |
5168 | io_schedule_finish(token); | |
5169 | } | |
5170 | EXPORT_SYMBOL(io_schedule); | |
5171 | ||
1da177e4 LT |
5172 | /** |
5173 | * sys_sched_get_priority_max - return maximum RT priority. | |
5174 | * @policy: scheduling class. | |
5175 | * | |
e69f6186 YB |
5176 | * Return: On success, this syscall returns the maximum |
5177 | * rt_priority that can be used by a given scheduling class. | |
5178 | * On failure, a negative error code is returned. | |
1da177e4 | 5179 | */ |
5add95d4 | 5180 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5181 | { |
5182 | int ret = -EINVAL; | |
5183 | ||
5184 | switch (policy) { | |
5185 | case SCHED_FIFO: | |
5186 | case SCHED_RR: | |
5187 | ret = MAX_USER_RT_PRIO-1; | |
5188 | break; | |
aab03e05 | 5189 | case SCHED_DEADLINE: |
1da177e4 | 5190 | case SCHED_NORMAL: |
b0a9499c | 5191 | case SCHED_BATCH: |
dd41f596 | 5192 | case SCHED_IDLE: |
1da177e4 LT |
5193 | ret = 0; |
5194 | break; | |
5195 | } | |
5196 | return ret; | |
5197 | } | |
5198 | ||
5199 | /** | |
5200 | * sys_sched_get_priority_min - return minimum RT priority. | |
5201 | * @policy: scheduling class. | |
5202 | * | |
e69f6186 YB |
5203 | * Return: On success, this syscall returns the minimum |
5204 | * rt_priority that can be used by a given scheduling class. | |
5205 | * On failure, a negative error code is returned. | |
1da177e4 | 5206 | */ |
5add95d4 | 5207 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5208 | { |
5209 | int ret = -EINVAL; | |
5210 | ||
5211 | switch (policy) { | |
5212 | case SCHED_FIFO: | |
5213 | case SCHED_RR: | |
5214 | ret = 1; | |
5215 | break; | |
aab03e05 | 5216 | case SCHED_DEADLINE: |
1da177e4 | 5217 | case SCHED_NORMAL: |
b0a9499c | 5218 | case SCHED_BATCH: |
dd41f596 | 5219 | case SCHED_IDLE: |
1da177e4 LT |
5220 | ret = 0; |
5221 | } | |
5222 | return ret; | |
5223 | } | |
5224 | ||
5225 | /** | |
5226 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5227 | * @pid: pid of the process. | |
5228 | * @interval: userspace pointer to the timeslice value. | |
5229 | * | |
5230 | * this syscall writes the default timeslice value of a given process | |
5231 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5232 | * |
5233 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5234 | * an error code. | |
1da177e4 | 5235 | */ |
17da2bd9 | 5236 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5237 | struct timespec __user *, interval) |
1da177e4 | 5238 | { |
36c8b586 | 5239 | struct task_struct *p; |
a4ec24b4 | 5240 | unsigned int time_slice; |
eb580751 PZ |
5241 | struct rq_flags rf; |
5242 | struct timespec t; | |
dba091b9 | 5243 | struct rq *rq; |
3a5c359a | 5244 | int retval; |
1da177e4 LT |
5245 | |
5246 | if (pid < 0) | |
3a5c359a | 5247 | return -EINVAL; |
1da177e4 LT |
5248 | |
5249 | retval = -ESRCH; | |
1a551ae7 | 5250 | rcu_read_lock(); |
1da177e4 LT |
5251 | p = find_process_by_pid(pid); |
5252 | if (!p) | |
5253 | goto out_unlock; | |
5254 | ||
5255 | retval = security_task_getscheduler(p); | |
5256 | if (retval) | |
5257 | goto out_unlock; | |
5258 | ||
eb580751 | 5259 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5260 | time_slice = 0; |
5261 | if (p->sched_class->get_rr_interval) | |
5262 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5263 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5264 | |
1a551ae7 | 5265 | rcu_read_unlock(); |
a4ec24b4 | 5266 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5267 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5268 | return retval; |
3a5c359a | 5269 | |
1da177e4 | 5270 | out_unlock: |
1a551ae7 | 5271 | rcu_read_unlock(); |
1da177e4 LT |
5272 | return retval; |
5273 | } | |
5274 | ||
7c731e0a | 5275 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5276 | |
82a1fcb9 | 5277 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5278 | { |
1da177e4 | 5279 | unsigned long free = 0; |
4e79752c | 5280 | int ppid; |
1f8a7633 | 5281 | unsigned long state = p->state; |
1da177e4 | 5282 | |
38200502 TH |
5283 | if (!try_get_task_stack(p)) |
5284 | return; | |
1f8a7633 TH |
5285 | if (state) |
5286 | state = __ffs(state) + 1; | |
28d0686c | 5287 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5288 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
1da177e4 | 5289 | if (state == TASK_RUNNING) |
3df0fc5b | 5290 | printk(KERN_CONT " running task "); |
1da177e4 | 5291 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5292 | free = stack_not_used(p); |
1da177e4 | 5293 | #endif |
a90e984c | 5294 | ppid = 0; |
4e79752c | 5295 | rcu_read_lock(); |
a90e984c ON |
5296 | if (pid_alive(p)) |
5297 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5298 | rcu_read_unlock(); |
3df0fc5b | 5299 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5300 | task_pid_nr(p), ppid, |
aa47b7e0 | 5301 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5302 | |
3d1cb205 | 5303 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5304 | show_stack(p, NULL); |
38200502 | 5305 | put_task_stack(p); |
1da177e4 LT |
5306 | } |
5307 | ||
e59e2ae2 | 5308 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5309 | { |
36c8b586 | 5310 | struct task_struct *g, *p; |
1da177e4 | 5311 | |
4bd77321 | 5312 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5313 | printk(KERN_INFO |
5314 | " task PC stack pid father\n"); | |
1da177e4 | 5315 | #else |
3df0fc5b PZ |
5316 | printk(KERN_INFO |
5317 | " task PC stack pid father\n"); | |
1da177e4 | 5318 | #endif |
510f5acc | 5319 | rcu_read_lock(); |
5d07f420 | 5320 | for_each_process_thread(g, p) { |
1da177e4 LT |
5321 | /* |
5322 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5323 | * console might take a lot of time: |
57675cb9 AR |
5324 | * Also, reset softlockup watchdogs on all CPUs, because |
5325 | * another CPU might be blocked waiting for us to process | |
5326 | * an IPI. | |
1da177e4 LT |
5327 | */ |
5328 | touch_nmi_watchdog(); | |
57675cb9 | 5329 | touch_all_softlockup_watchdogs(); |
39bc89fd | 5330 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5331 | sched_show_task(p); |
5d07f420 | 5332 | } |
1da177e4 | 5333 | |
dd41f596 | 5334 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5335 | if (!state_filter) |
5336 | sysrq_sched_debug_show(); | |
dd41f596 | 5337 | #endif |
510f5acc | 5338 | rcu_read_unlock(); |
e59e2ae2 IM |
5339 | /* |
5340 | * Only show locks if all tasks are dumped: | |
5341 | */ | |
93335a21 | 5342 | if (!state_filter) |
e59e2ae2 | 5343 | debug_show_all_locks(); |
1da177e4 LT |
5344 | } |
5345 | ||
0db0628d | 5346 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5347 | { |
dd41f596 | 5348 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5349 | } |
5350 | ||
f340c0d1 IM |
5351 | /** |
5352 | * init_idle - set up an idle thread for a given CPU | |
5353 | * @idle: task in question | |
5354 | * @cpu: cpu the idle task belongs to | |
5355 | * | |
5356 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5357 | * flag, to make booting more robust. | |
5358 | */ | |
0db0628d | 5359 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5360 | { |
70b97a7f | 5361 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5362 | unsigned long flags; |
5363 | ||
25834c73 PZ |
5364 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5365 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5366 | |
5e1576ed | 5367 | __sched_fork(0, idle); |
06b83b5f | 5368 | idle->state = TASK_RUNNING; |
dd41f596 | 5369 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5370 | idle->flags |= PF_IDLE; |
dd41f596 | 5371 | |
e1b77c92 MR |
5372 | kasan_unpoison_task_stack(idle); |
5373 | ||
de9b8f5d PZ |
5374 | #ifdef CONFIG_SMP |
5375 | /* | |
5376 | * Its possible that init_idle() gets called multiple times on a task, | |
5377 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5378 | * | |
5379 | * And since this is boot we can forgo the serialization. | |
5380 | */ | |
5381 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5382 | #endif | |
6506cf6c PZ |
5383 | /* |
5384 | * We're having a chicken and egg problem, even though we are | |
5385 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5386 | * lockdep check in task_group() will fail. | |
5387 | * | |
5388 | * Similar case to sched_fork(). / Alternatively we could | |
5389 | * use task_rq_lock() here and obtain the other rq->lock. | |
5390 | * | |
5391 | * Silence PROVE_RCU | |
5392 | */ | |
5393 | rcu_read_lock(); | |
dd41f596 | 5394 | __set_task_cpu(idle, cpu); |
6506cf6c | 5395 | rcu_read_unlock(); |
1da177e4 | 5396 | |
1da177e4 | 5397 | rq->curr = rq->idle = idle; |
da0c1e65 | 5398 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5399 | #ifdef CONFIG_SMP |
3ca7a440 | 5400 | idle->on_cpu = 1; |
4866cde0 | 5401 | #endif |
25834c73 PZ |
5402 | raw_spin_unlock(&rq->lock); |
5403 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5404 | |
5405 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5406 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5407 | |
dd41f596 IM |
5408 | /* |
5409 | * The idle tasks have their own, simple scheduling class: | |
5410 | */ | |
5411 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5412 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5413 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5414 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5415 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5416 | #endif | |
19978ca6 IM |
5417 | } |
5418 | ||
f82f8042 JL |
5419 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5420 | const struct cpumask *trial) | |
5421 | { | |
5422 | int ret = 1, trial_cpus; | |
5423 | struct dl_bw *cur_dl_b; | |
5424 | unsigned long flags; | |
5425 | ||
bb2bc55a MG |
5426 | if (!cpumask_weight(cur)) |
5427 | return ret; | |
5428 | ||
75e23e49 | 5429 | rcu_read_lock_sched(); |
f82f8042 JL |
5430 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5431 | trial_cpus = cpumask_weight(trial); | |
5432 | ||
5433 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5434 | if (cur_dl_b->bw != -1 && | |
5435 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5436 | ret = 0; | |
5437 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5438 | rcu_read_unlock_sched(); |
f82f8042 JL |
5439 | |
5440 | return ret; | |
5441 | } | |
5442 | ||
7f51412a JL |
5443 | int task_can_attach(struct task_struct *p, |
5444 | const struct cpumask *cs_cpus_allowed) | |
5445 | { | |
5446 | int ret = 0; | |
5447 | ||
5448 | /* | |
5449 | * Kthreads which disallow setaffinity shouldn't be moved | |
5450 | * to a new cpuset; we don't want to change their cpu | |
5451 | * affinity and isolating such threads by their set of | |
5452 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5453 | * applicable for such threads. This prevents checking for | |
5454 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5455 | * before cpus_allowed may be changed. | |
5456 | */ | |
5457 | if (p->flags & PF_NO_SETAFFINITY) { | |
5458 | ret = -EINVAL; | |
5459 | goto out; | |
5460 | } | |
5461 | ||
5462 | #ifdef CONFIG_SMP | |
5463 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5464 | cs_cpus_allowed)) { | |
5465 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5466 | cs_cpus_allowed); | |
75e23e49 | 5467 | struct dl_bw *dl_b; |
7f51412a JL |
5468 | bool overflow; |
5469 | int cpus; | |
5470 | unsigned long flags; | |
5471 | ||
75e23e49 JL |
5472 | rcu_read_lock_sched(); |
5473 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5474 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5475 | cpus = dl_bw_cpus(dest_cpu); | |
5476 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5477 | if (overflow) | |
5478 | ret = -EBUSY; | |
5479 | else { | |
5480 | /* | |
5481 | * We reserve space for this task in the destination | |
5482 | * root_domain, as we can't fail after this point. | |
5483 | * We will free resources in the source root_domain | |
5484 | * later on (see set_cpus_allowed_dl()). | |
5485 | */ | |
5486 | __dl_add(dl_b, p->dl.dl_bw); | |
5487 | } | |
5488 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5489 | rcu_read_unlock_sched(); |
7f51412a JL |
5490 | |
5491 | } | |
5492 | #endif | |
5493 | out: | |
5494 | return ret; | |
5495 | } | |
5496 | ||
1da177e4 | 5497 | #ifdef CONFIG_SMP |
1da177e4 | 5498 | |
e26fbffd TG |
5499 | static bool sched_smp_initialized __read_mostly; |
5500 | ||
e6628d5b MG |
5501 | #ifdef CONFIG_NUMA_BALANCING |
5502 | /* Migrate current task p to target_cpu */ | |
5503 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5504 | { | |
5505 | struct migration_arg arg = { p, target_cpu }; | |
5506 | int curr_cpu = task_cpu(p); | |
5507 | ||
5508 | if (curr_cpu == target_cpu) | |
5509 | return 0; | |
5510 | ||
5511 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5512 | return -EINVAL; | |
5513 | ||
5514 | /* TODO: This is not properly updating schedstats */ | |
5515 | ||
286549dc | 5516 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5517 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5518 | } | |
0ec8aa00 PZ |
5519 | |
5520 | /* | |
5521 | * Requeue a task on a given node and accurately track the number of NUMA | |
5522 | * tasks on the runqueues | |
5523 | */ | |
5524 | void sched_setnuma(struct task_struct *p, int nid) | |
5525 | { | |
da0c1e65 | 5526 | bool queued, running; |
eb580751 PZ |
5527 | struct rq_flags rf; |
5528 | struct rq *rq; | |
0ec8aa00 | 5529 | |
eb580751 | 5530 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5531 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5532 | running = task_current(rq, p); |
5533 | ||
da0c1e65 | 5534 | if (queued) |
1de64443 | 5535 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5536 | if (running) |
f3cd1c4e | 5537 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5538 | |
5539 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5540 | |
da0c1e65 | 5541 | if (queued) |
1de64443 | 5542 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 5543 | if (running) |
b2bf6c31 | 5544 | set_curr_task(rq, p); |
eb580751 | 5545 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5546 | } |
5cc389bc | 5547 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5548 | |
1da177e4 | 5549 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5550 | /* |
48c5ccae PZ |
5551 | * Ensures that the idle task is using init_mm right before its cpu goes |
5552 | * offline. | |
054b9108 | 5553 | */ |
48c5ccae | 5554 | void idle_task_exit(void) |
1da177e4 | 5555 | { |
48c5ccae | 5556 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5557 | |
48c5ccae | 5558 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5559 | |
a53efe5f | 5560 | if (mm != &init_mm) { |
f98db601 | 5561 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5562 | finish_arch_post_lock_switch(); |
5563 | } | |
48c5ccae | 5564 | mmdrop(mm); |
1da177e4 LT |
5565 | } |
5566 | ||
5567 | /* | |
5d180232 PZ |
5568 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5569 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5570 | * nr_active count is stable. We need to take the teardown thread which |
5571 | * is calling this into account, so we hand in adjust = 1 to the load | |
5572 | * calculation. | |
5d180232 PZ |
5573 | * |
5574 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5575 | */ |
5d180232 | 5576 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5577 | { |
d60585c5 | 5578 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5579 | if (delta) |
5580 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5581 | } |
5582 | ||
3f1d2a31 PZ |
5583 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5584 | { | |
5585 | } | |
5586 | ||
5587 | static const struct sched_class fake_sched_class = { | |
5588 | .put_prev_task = put_prev_task_fake, | |
5589 | }; | |
5590 | ||
5591 | static struct task_struct fake_task = { | |
5592 | /* | |
5593 | * Avoid pull_{rt,dl}_task() | |
5594 | */ | |
5595 | .prio = MAX_PRIO + 1, | |
5596 | .sched_class = &fake_sched_class, | |
5597 | }; | |
5598 | ||
48f24c4d | 5599 | /* |
48c5ccae PZ |
5600 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5601 | * try_to_wake_up()->select_task_rq(). | |
5602 | * | |
5603 | * Called with rq->lock held even though we'er in stop_machine() and | |
5604 | * there's no concurrency possible, we hold the required locks anyway | |
5605 | * because of lock validation efforts. | |
1da177e4 | 5606 | */ |
5e16bbc2 | 5607 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5608 | { |
5e16bbc2 | 5609 | struct rq *rq = dead_rq; |
48c5ccae | 5610 | struct task_struct *next, *stop = rq->stop; |
4d25b35e | 5611 | struct rq_flags rf, old_rf; |
48c5ccae | 5612 | int dest_cpu; |
1da177e4 LT |
5613 | |
5614 | /* | |
48c5ccae PZ |
5615 | * Fudge the rq selection such that the below task selection loop |
5616 | * doesn't get stuck on the currently eligible stop task. | |
5617 | * | |
5618 | * We're currently inside stop_machine() and the rq is either stuck | |
5619 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5620 | * either way we should never end up calling schedule() until we're | |
5621 | * done here. | |
1da177e4 | 5622 | */ |
48c5ccae | 5623 | rq->stop = NULL; |
48f24c4d | 5624 | |
77bd3970 FW |
5625 | /* |
5626 | * put_prev_task() and pick_next_task() sched | |
5627 | * class method both need to have an up-to-date | |
5628 | * value of rq->clock[_task] | |
5629 | */ | |
5630 | update_rq_clock(rq); | |
5631 | ||
5e16bbc2 | 5632 | for (;;) { |
48c5ccae PZ |
5633 | /* |
5634 | * There's this thread running, bail when that's the only | |
5635 | * remaining thread. | |
5636 | */ | |
5637 | if (rq->nr_running == 1) | |
dd41f596 | 5638 | break; |
48c5ccae | 5639 | |
cbce1a68 | 5640 | /* |
5473e0cc | 5641 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 | 5642 | */ |
d8ac8971 MF |
5643 | rq_pin_lock(rq, &rf); |
5644 | next = pick_next_task(rq, &fake_task, &rf); | |
48c5ccae | 5645 | BUG_ON(!next); |
79c53799 | 5646 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5647 | |
5473e0cc WL |
5648 | /* |
5649 | * Rules for changing task_struct::cpus_allowed are holding | |
5650 | * both pi_lock and rq->lock, such that holding either | |
5651 | * stabilizes the mask. | |
5652 | * | |
5653 | * Drop rq->lock is not quite as disastrous as it usually is | |
5654 | * because !cpu_active at this point, which means load-balance | |
5655 | * will not interfere. Also, stop-machine. | |
5656 | */ | |
d8ac8971 | 5657 | rq_unpin_lock(rq, &rf); |
5473e0cc WL |
5658 | raw_spin_unlock(&rq->lock); |
5659 | raw_spin_lock(&next->pi_lock); | |
5660 | raw_spin_lock(&rq->lock); | |
5661 | ||
5662 | /* | |
5663 | * Since we're inside stop-machine, _nothing_ should have | |
5664 | * changed the task, WARN if weird stuff happened, because in | |
5665 | * that case the above rq->lock drop is a fail too. | |
5666 | */ | |
5667 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5668 | raw_spin_unlock(&next->pi_lock); | |
5669 | continue; | |
5670 | } | |
5671 | ||
4d25b35e MF |
5672 | /* |
5673 | * __migrate_task() may return with a different | |
5674 | * rq->lock held and a new cookie in 'rf', but we need | |
5675 | * to preserve rf::clock_update_flags for 'dead_rq'. | |
5676 | */ | |
5677 | old_rf = rf; | |
5678 | ||
48c5ccae | 5679 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5680 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5681 | |
5e16bbc2 PZ |
5682 | rq = __migrate_task(rq, next, dest_cpu); |
5683 | if (rq != dead_rq) { | |
5684 | raw_spin_unlock(&rq->lock); | |
5685 | rq = dead_rq; | |
5686 | raw_spin_lock(&rq->lock); | |
4d25b35e | 5687 | rf = old_rf; |
5e16bbc2 | 5688 | } |
5473e0cc | 5689 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5690 | } |
dce48a84 | 5691 | |
48c5ccae | 5692 | rq->stop = stop; |
dce48a84 | 5693 | } |
1da177e4 LT |
5694 | #endif /* CONFIG_HOTPLUG_CPU */ |
5695 | ||
1f11eb6a GH |
5696 | static void set_rq_online(struct rq *rq) |
5697 | { | |
5698 | if (!rq->online) { | |
5699 | const struct sched_class *class; | |
5700 | ||
c6c4927b | 5701 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5702 | rq->online = 1; |
5703 | ||
5704 | for_each_class(class) { | |
5705 | if (class->rq_online) | |
5706 | class->rq_online(rq); | |
5707 | } | |
5708 | } | |
5709 | } | |
5710 | ||
5711 | static void set_rq_offline(struct rq *rq) | |
5712 | { | |
5713 | if (rq->online) { | |
5714 | const struct sched_class *class; | |
5715 | ||
5716 | for_each_class(class) { | |
5717 | if (class->rq_offline) | |
5718 | class->rq_offline(rq); | |
5719 | } | |
5720 | ||
c6c4927b | 5721 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5722 | rq->online = 0; |
5723 | } | |
5724 | } | |
5725 | ||
9cf7243d | 5726 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5727 | { |
969c7921 | 5728 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5729 | |
a803f026 CM |
5730 | rq->age_stamp = sched_clock_cpu(cpu); |
5731 | } | |
5732 | ||
4cb98839 PZ |
5733 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5734 | ||
3e9830dc | 5735 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5736 | |
d039ac60 | 5737 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5738 | |
d039ac60 | 5739 | static int __init sched_debug_setup(char *str) |
f6630114 | 5740 | { |
d039ac60 | 5741 | sched_debug_enabled = 1; |
f6630114 MT |
5742 | |
5743 | return 0; | |
5744 | } | |
d039ac60 PZ |
5745 | early_param("sched_debug", sched_debug_setup); |
5746 | ||
5747 | static inline bool sched_debug(void) | |
5748 | { | |
5749 | return sched_debug_enabled; | |
5750 | } | |
f6630114 | 5751 | |
7c16ec58 | 5752 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5753 | struct cpumask *groupmask) |
1da177e4 | 5754 | { |
4dcf6aff | 5755 | struct sched_group *group = sd->groups; |
1da177e4 | 5756 | |
96f874e2 | 5757 | cpumask_clear(groupmask); |
4dcf6aff IM |
5758 | |
5759 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5760 | ||
5761 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5762 | printk("does not load-balance\n"); |
4dcf6aff | 5763 | if (sd->parent) |
3df0fc5b PZ |
5764 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5765 | " has parent"); | |
4dcf6aff | 5766 | return -1; |
41c7ce9a NP |
5767 | } |
5768 | ||
333470ee TH |
5769 | printk(KERN_CONT "span %*pbl level %s\n", |
5770 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5771 | |
758b2cdc | 5772 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5773 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5774 | "CPU%d\n", cpu); | |
4dcf6aff | 5775 | } |
758b2cdc | 5776 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5777 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5778 | " CPU%d\n", cpu); | |
4dcf6aff | 5779 | } |
1da177e4 | 5780 | |
4dcf6aff | 5781 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5782 | do { |
4dcf6aff | 5783 | if (!group) { |
3df0fc5b PZ |
5784 | printk("\n"); |
5785 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5786 | break; |
5787 | } | |
5788 | ||
758b2cdc | 5789 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5790 | printk(KERN_CONT "\n"); |
5791 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5792 | break; |
5793 | } | |
1da177e4 | 5794 | |
cb83b629 PZ |
5795 | if (!(sd->flags & SD_OVERLAP) && |
5796 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5797 | printk(KERN_CONT "\n"); |
5798 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5799 | break; |
5800 | } | |
1da177e4 | 5801 | |
758b2cdc | 5802 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5803 | |
333470ee TH |
5804 | printk(KERN_CONT " %*pbl", |
5805 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5806 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
bf475ce0 | 5807 | printk(KERN_CONT " (cpu_capacity = %lu)", |
63b2ca30 | 5808 | group->sgc->capacity); |
381512cf | 5809 | } |
1da177e4 | 5810 | |
4dcf6aff IM |
5811 | group = group->next; |
5812 | } while (group != sd->groups); | |
3df0fc5b | 5813 | printk(KERN_CONT "\n"); |
1da177e4 | 5814 | |
758b2cdc | 5815 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5816 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5817 | |
758b2cdc RR |
5818 | if (sd->parent && |
5819 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5820 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5821 | "of domain->span\n"); | |
4dcf6aff IM |
5822 | return 0; |
5823 | } | |
1da177e4 | 5824 | |
4dcf6aff IM |
5825 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5826 | { | |
5827 | int level = 0; | |
1da177e4 | 5828 | |
d039ac60 | 5829 | if (!sched_debug_enabled) |
f6630114 MT |
5830 | return; |
5831 | ||
4dcf6aff IM |
5832 | if (!sd) { |
5833 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5834 | return; | |
5835 | } | |
1da177e4 | 5836 | |
4dcf6aff IM |
5837 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5838 | ||
5839 | for (;;) { | |
4cb98839 | 5840 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5841 | break; |
1da177e4 LT |
5842 | level++; |
5843 | sd = sd->parent; | |
33859f7f | 5844 | if (!sd) |
4dcf6aff IM |
5845 | break; |
5846 | } | |
1da177e4 | 5847 | } |
6d6bc0ad | 5848 | #else /* !CONFIG_SCHED_DEBUG */ |
a18a579e PZ |
5849 | |
5850 | # define sched_debug_enabled 0 | |
48f24c4d | 5851 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5852 | static inline bool sched_debug(void) |
5853 | { | |
5854 | return false; | |
5855 | } | |
6d6bc0ad | 5856 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5857 | |
1a20ff27 | 5858 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5859 | { |
758b2cdc | 5860 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5861 | return 1; |
5862 | ||
5863 | /* Following flags need at least 2 groups */ | |
5864 | if (sd->flags & (SD_LOAD_BALANCE | | |
5865 | SD_BALANCE_NEWIDLE | | |
5866 | SD_BALANCE_FORK | | |
89c4710e | 5867 | SD_BALANCE_EXEC | |
5d4dfddd | 5868 | SD_SHARE_CPUCAPACITY | |
1f6e6c7c | 5869 | SD_ASYM_CPUCAPACITY | |
d77b3ed5 VG |
5870 | SD_SHARE_PKG_RESOURCES | |
5871 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5872 | if (sd->groups != sd->groups->next) |
5873 | return 0; | |
5874 | } | |
5875 | ||
5876 | /* Following flags don't use groups */ | |
c88d5910 | 5877 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5878 | return 0; |
5879 | ||
5880 | return 1; | |
5881 | } | |
5882 | ||
48f24c4d IM |
5883 | static int |
5884 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5885 | { |
5886 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5887 | ||
5888 | if (sd_degenerate(parent)) | |
5889 | return 1; | |
5890 | ||
758b2cdc | 5891 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5892 | return 0; |
5893 | ||
245af2c7 SS |
5894 | /* Flags needing groups don't count if only 1 group in parent */ |
5895 | if (parent->groups == parent->groups->next) { | |
5896 | pflags &= ~(SD_LOAD_BALANCE | | |
5897 | SD_BALANCE_NEWIDLE | | |
5898 | SD_BALANCE_FORK | | |
89c4710e | 5899 | SD_BALANCE_EXEC | |
1f6e6c7c | 5900 | SD_ASYM_CPUCAPACITY | |
5d4dfddd | 5901 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5902 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5903 | SD_PREFER_SIBLING | |
5904 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5905 | if (nr_node_ids == 1) |
5906 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5907 | } |
5908 | if (~cflags & pflags) | |
5909 | return 0; | |
5910 | ||
5911 | return 1; | |
5912 | } | |
5913 | ||
dce840a0 | 5914 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5915 | { |
dce840a0 | 5916 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5917 | |
68e74568 | 5918 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5919 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5920 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5921 | free_cpumask_var(rd->rto_mask); |
5922 | free_cpumask_var(rd->online); | |
5923 | free_cpumask_var(rd->span); | |
5924 | kfree(rd); | |
5925 | } | |
5926 | ||
57d885fe GH |
5927 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5928 | { | |
a0490fa3 | 5929 | struct root_domain *old_rd = NULL; |
57d885fe | 5930 | unsigned long flags; |
57d885fe | 5931 | |
05fa785c | 5932 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5933 | |
5934 | if (rq->rd) { | |
a0490fa3 | 5935 | old_rd = rq->rd; |
57d885fe | 5936 | |
c6c4927b | 5937 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5938 | set_rq_offline(rq); |
57d885fe | 5939 | |
c6c4927b | 5940 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5941 | |
a0490fa3 | 5942 | /* |
0515973f | 5943 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5944 | * set old_rd to NULL to skip the freeing later |
5945 | * in this function: | |
5946 | */ | |
5947 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5948 | old_rd = NULL; | |
57d885fe GH |
5949 | } |
5950 | ||
5951 | atomic_inc(&rd->refcount); | |
5952 | rq->rd = rd; | |
5953 | ||
c6c4927b | 5954 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5955 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5956 | set_rq_online(rq); |
57d885fe | 5957 | |
05fa785c | 5958 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5959 | |
5960 | if (old_rd) | |
dce840a0 | 5961 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5962 | } |
5963 | ||
68c38fc3 | 5964 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5965 | { |
5966 | memset(rd, 0, sizeof(*rd)); | |
5967 | ||
8295c699 | 5968 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5969 | goto out; |
8295c699 | 5970 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5971 | goto free_span; |
8295c699 | 5972 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5973 | goto free_online; |
8295c699 | 5974 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5975 | goto free_dlo_mask; |
6e0534f2 | 5976 | |
332ac17e | 5977 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 | 5978 | if (cpudl_init(&rd->cpudl) != 0) |
92c99ac8 | 5979 | goto free_rto_mask; |
332ac17e | 5980 | |
68c38fc3 | 5981 | if (cpupri_init(&rd->cpupri) != 0) |
4b12db93 | 5982 | goto free_cpudl; |
c6c4927b | 5983 | return 0; |
6e0534f2 | 5984 | |
4b12db93 MP |
5985 | free_cpudl: |
5986 | cpudl_cleanup(&rd->cpudl); | |
68e74568 RR |
5987 | free_rto_mask: |
5988 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5989 | free_dlo_mask: |
5990 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5991 | free_online: |
5992 | free_cpumask_var(rd->online); | |
5993 | free_span: | |
5994 | free_cpumask_var(rd->span); | |
0c910d28 | 5995 | out: |
c6c4927b | 5996 | return -ENOMEM; |
57d885fe GH |
5997 | } |
5998 | ||
029632fb PZ |
5999 | /* |
6000 | * By default the system creates a single root-domain with all cpus as | |
6001 | * members (mimicking the global state we have today). | |
6002 | */ | |
6003 | struct root_domain def_root_domain; | |
6004 | ||
57d885fe GH |
6005 | static void init_defrootdomain(void) |
6006 | { | |
68c38fc3 | 6007 | init_rootdomain(&def_root_domain); |
c6c4927b | 6008 | |
57d885fe GH |
6009 | atomic_set(&def_root_domain.refcount, 1); |
6010 | } | |
6011 | ||
dc938520 | 6012 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6013 | { |
6014 | struct root_domain *rd; | |
6015 | ||
6016 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6017 | if (!rd) | |
6018 | return NULL; | |
6019 | ||
68c38fc3 | 6020 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6021 | kfree(rd); |
6022 | return NULL; | |
6023 | } | |
57d885fe GH |
6024 | |
6025 | return rd; | |
6026 | } | |
6027 | ||
63b2ca30 | 6028 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
6029 | { |
6030 | struct sched_group *tmp, *first; | |
6031 | ||
6032 | if (!sg) | |
6033 | return; | |
6034 | ||
6035 | first = sg; | |
6036 | do { | |
6037 | tmp = sg->next; | |
6038 | ||
63b2ca30 NP |
6039 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
6040 | kfree(sg->sgc); | |
e3589f6c PZ |
6041 | |
6042 | kfree(sg); | |
6043 | sg = tmp; | |
6044 | } while (sg != first); | |
6045 | } | |
6046 | ||
16f3ef46 | 6047 | static void destroy_sched_domain(struct sched_domain *sd) |
dce840a0 | 6048 | { |
e3589f6c PZ |
6049 | /* |
6050 | * If its an overlapping domain it has private groups, iterate and | |
6051 | * nuke them all. | |
6052 | */ | |
6053 | if (sd->flags & SD_OVERLAP) { | |
6054 | free_sched_groups(sd->groups, 1); | |
6055 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 6056 | kfree(sd->groups->sgc); |
dce840a0 | 6057 | kfree(sd->groups); |
9c3f75cb | 6058 | } |
24fc7edb PZ |
6059 | if (sd->shared && atomic_dec_and_test(&sd->shared->ref)) |
6060 | kfree(sd->shared); | |
dce840a0 PZ |
6061 | kfree(sd); |
6062 | } | |
6063 | ||
16f3ef46 | 6064 | static void destroy_sched_domains_rcu(struct rcu_head *rcu) |
dce840a0 | 6065 | { |
16f3ef46 PZ |
6066 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); |
6067 | ||
6068 | while (sd) { | |
6069 | struct sched_domain *parent = sd->parent; | |
6070 | destroy_sched_domain(sd); | |
6071 | sd = parent; | |
6072 | } | |
dce840a0 PZ |
6073 | } |
6074 | ||
f39180ef | 6075 | static void destroy_sched_domains(struct sched_domain *sd) |
dce840a0 | 6076 | { |
16f3ef46 PZ |
6077 | if (sd) |
6078 | call_rcu(&sd->rcu, destroy_sched_domains_rcu); | |
dce840a0 PZ |
6079 | } |
6080 | ||
518cd623 PZ |
6081 | /* |
6082 | * Keep a special pointer to the highest sched_domain that has | |
6083 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
6084 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
6085 | * | |
6086 | * Also keep a unique ID per domain (we use the first cpu number in | |
6087 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 6088 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
6089 | */ |
6090 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 6091 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 6092 | DEFINE_PER_CPU(int, sd_llc_id); |
0e369d75 | 6093 | DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 6094 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 6095 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 PZ |
6096 | |
6097 | static void update_top_cache_domain(int cpu) | |
6098 | { | |
0e369d75 | 6099 | struct sched_domain_shared *sds = NULL; |
518cd623 PZ |
6100 | struct sched_domain *sd; |
6101 | int id = cpu; | |
7d9ffa89 | 6102 | int size = 1; |
518cd623 PZ |
6103 | |
6104 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 6105 | if (sd) { |
518cd623 | 6106 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 6107 | size = cpumask_weight(sched_domain_span(sd)); |
0e369d75 | 6108 | sds = sd->shared; |
7d9ffa89 | 6109 | } |
518cd623 PZ |
6110 | |
6111 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 6112 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 6113 | per_cpu(sd_llc_id, cpu) = id; |
0e369d75 | 6114 | rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds); |
fb13c7ee MG |
6115 | |
6116 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
6117 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
6118 | |
6119 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
6120 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
6121 | } |
6122 | ||
1da177e4 | 6123 | /* |
0eab9146 | 6124 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6125 | * hold the hotplug lock. |
6126 | */ | |
0eab9146 IM |
6127 | static void |
6128 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6129 | { |
70b97a7f | 6130 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6131 | struct sched_domain *tmp; |
6132 | ||
6133 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6134 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6135 | struct sched_domain *parent = tmp->parent; |
6136 | if (!parent) | |
6137 | break; | |
f29c9b1c | 6138 | |
1a848870 | 6139 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6140 | tmp->parent = parent->parent; |
1a848870 SS |
6141 | if (parent->parent) |
6142 | parent->parent->child = tmp; | |
10866e62 PZ |
6143 | /* |
6144 | * Transfer SD_PREFER_SIBLING down in case of a | |
6145 | * degenerate parent; the spans match for this | |
6146 | * so the property transfers. | |
6147 | */ | |
6148 | if (parent->flags & SD_PREFER_SIBLING) | |
6149 | tmp->flags |= SD_PREFER_SIBLING; | |
f39180ef | 6150 | destroy_sched_domain(parent); |
f29c9b1c LZ |
6151 | } else |
6152 | tmp = tmp->parent; | |
245af2c7 SS |
6153 | } |
6154 | ||
1a848870 | 6155 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6156 | tmp = sd; |
245af2c7 | 6157 | sd = sd->parent; |
f39180ef | 6158 | destroy_sched_domain(tmp); |
1a848870 SS |
6159 | if (sd) |
6160 | sd->child = NULL; | |
6161 | } | |
1da177e4 | 6162 | |
4cb98839 | 6163 | sched_domain_debug(sd, cpu); |
1da177e4 | 6164 | |
57d885fe | 6165 | rq_attach_root(rq, rd); |
dce840a0 | 6166 | tmp = rq->sd; |
674311d5 | 6167 | rcu_assign_pointer(rq->sd, sd); |
f39180ef | 6168 | destroy_sched_domains(tmp); |
518cd623 PZ |
6169 | |
6170 | update_top_cache_domain(cpu); | |
1da177e4 LT |
6171 | } |
6172 | ||
1da177e4 LT |
6173 | /* Setup the mask of cpus configured for isolated domains */ |
6174 | static int __init isolated_cpu_setup(char *str) | |
6175 | { | |
a6e4491c PB |
6176 | int ret; |
6177 | ||
bdddd296 | 6178 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
6179 | ret = cpulist_parse(str, cpu_isolated_map); |
6180 | if (ret) { | |
6181 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
6182 | return 0; | |
6183 | } | |
1da177e4 LT |
6184 | return 1; |
6185 | } | |
8927f494 | 6186 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6187 | |
49a02c51 | 6188 | struct s_data { |
21d42ccf | 6189 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6190 | struct root_domain *rd; |
6191 | }; | |
6192 | ||
2109b99e | 6193 | enum s_alloc { |
2109b99e | 6194 | sa_rootdomain, |
21d42ccf | 6195 | sa_sd, |
dce840a0 | 6196 | sa_sd_storage, |
2109b99e AH |
6197 | sa_none, |
6198 | }; | |
6199 | ||
c1174876 PZ |
6200 | /* |
6201 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
6202 | * domain traversal. | |
6203 | * | |
6204 | * Asymmetric node setups can result in situations where the domain tree is of | |
6205 | * unequal depth, make sure to skip domains that already cover the entire | |
6206 | * range. | |
6207 | * | |
6208 | * In that case build_sched_domains() will have terminated the iteration early | |
6209 | * and our sibling sd spans will be empty. Domains should always include the | |
6210 | * cpu they're built on, so check that. | |
6211 | * | |
6212 | */ | |
6213 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6214 | { | |
6215 | const struct cpumask *span = sched_domain_span(sd); | |
6216 | struct sd_data *sdd = sd->private; | |
6217 | struct sched_domain *sibling; | |
6218 | int i; | |
6219 | ||
6220 | for_each_cpu(i, span) { | |
6221 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6222 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6223 | continue; | |
6224 | ||
6225 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6226 | } | |
6227 | } | |
6228 | ||
6229 | /* | |
6230 | * Return the canonical balance cpu for this group, this is the first cpu | |
6231 | * of this group that's also in the iteration mask. | |
6232 | */ | |
6233 | int group_balance_cpu(struct sched_group *sg) | |
6234 | { | |
6235 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6236 | } | |
6237 | ||
e3589f6c PZ |
6238 | static int |
6239 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6240 | { | |
6241 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6242 | const struct cpumask *span = sched_domain_span(sd); | |
6243 | struct cpumask *covered = sched_domains_tmpmask; | |
6244 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6245 | struct sched_domain *sibling; |
e3589f6c PZ |
6246 | int i; |
6247 | ||
6248 | cpumask_clear(covered); | |
6249 | ||
6250 | for_each_cpu(i, span) { | |
6251 | struct cpumask *sg_span; | |
6252 | ||
6253 | if (cpumask_test_cpu(i, covered)) | |
6254 | continue; | |
6255 | ||
aaecac4a | 6256 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6257 | |
6258 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6259 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6260 | continue; |
6261 | ||
e3589f6c | 6262 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6263 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6264 | |
6265 | if (!sg) | |
6266 | goto fail; | |
6267 | ||
6268 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6269 | if (sibling->child) |
6270 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6271 | else | |
e3589f6c PZ |
6272 | cpumask_set_cpu(i, sg_span); |
6273 | ||
6274 | cpumask_or(covered, covered, sg_span); | |
6275 | ||
63b2ca30 NP |
6276 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6277 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6278 | build_group_mask(sd, sg); |
6279 | ||
c3decf0d | 6280 | /* |
63b2ca30 | 6281 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6282 | * domains and no possible iteration will get us here, we won't |
6283 | * die on a /0 trap. | |
6284 | */ | |
ca8ce3d0 | 6285 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
bf475ce0 | 6286 | sg->sgc->min_capacity = SCHED_CAPACITY_SCALE; |
e3589f6c | 6287 | |
c1174876 PZ |
6288 | /* |
6289 | * Make sure the first group of this domain contains the | |
6290 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6291 | * breaks. See update_sg_lb_stats(). | |
6292 | */ | |
74a5ce20 | 6293 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6294 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6295 | groups = sg; |
6296 | ||
6297 | if (!first) | |
6298 | first = sg; | |
6299 | if (last) | |
6300 | last->next = sg; | |
6301 | last = sg; | |
6302 | last->next = first; | |
6303 | } | |
6304 | sd->groups = groups; | |
6305 | ||
6306 | return 0; | |
6307 | ||
6308 | fail: | |
6309 | free_sched_groups(first, 0); | |
6310 | ||
6311 | return -ENOMEM; | |
6312 | } | |
6313 | ||
dce840a0 | 6314 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6315 | { |
dce840a0 PZ |
6316 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6317 | struct sched_domain *child = sd->child; | |
1da177e4 | 6318 | |
dce840a0 PZ |
6319 | if (child) |
6320 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6321 | |
9c3f75cb | 6322 | if (sg) { |
dce840a0 | 6323 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6324 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6325 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6326 | } |
dce840a0 PZ |
6327 | |
6328 | return cpu; | |
1e9f28fa | 6329 | } |
1e9f28fa | 6330 | |
01a08546 | 6331 | /* |
dce840a0 PZ |
6332 | * build_sched_groups will build a circular linked list of the groups |
6333 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6334 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6335 | * |
6336 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6337 | */ |
e3589f6c PZ |
6338 | static int |
6339 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6340 | { |
dce840a0 PZ |
6341 | struct sched_group *first = NULL, *last = NULL; |
6342 | struct sd_data *sdd = sd->private; | |
6343 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6344 | struct cpumask *covered; |
dce840a0 | 6345 | int i; |
9c1cfda2 | 6346 | |
e3589f6c PZ |
6347 | get_group(cpu, sdd, &sd->groups); |
6348 | atomic_inc(&sd->groups->ref); | |
6349 | ||
0936629f | 6350 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6351 | return 0; |
6352 | ||
f96225fd PZ |
6353 | lockdep_assert_held(&sched_domains_mutex); |
6354 | covered = sched_domains_tmpmask; | |
6355 | ||
dce840a0 | 6356 | cpumask_clear(covered); |
6711cab4 | 6357 | |
dce840a0 PZ |
6358 | for_each_cpu(i, span) { |
6359 | struct sched_group *sg; | |
cd08e923 | 6360 | int group, j; |
6711cab4 | 6361 | |
dce840a0 PZ |
6362 | if (cpumask_test_cpu(i, covered)) |
6363 | continue; | |
6711cab4 | 6364 | |
cd08e923 | 6365 | group = get_group(i, sdd, &sg); |
c1174876 | 6366 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6367 | |
dce840a0 PZ |
6368 | for_each_cpu(j, span) { |
6369 | if (get_group(j, sdd, NULL) != group) | |
6370 | continue; | |
0601a88d | 6371 | |
dce840a0 PZ |
6372 | cpumask_set_cpu(j, covered); |
6373 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6374 | } | |
0601a88d | 6375 | |
dce840a0 PZ |
6376 | if (!first) |
6377 | first = sg; | |
6378 | if (last) | |
6379 | last->next = sg; | |
6380 | last = sg; | |
6381 | } | |
6382 | last->next = first; | |
e3589f6c PZ |
6383 | |
6384 | return 0; | |
0601a88d | 6385 | } |
51888ca2 | 6386 | |
89c4710e | 6387 | /* |
63b2ca30 | 6388 | * Initialize sched groups cpu_capacity. |
89c4710e | 6389 | * |
63b2ca30 | 6390 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6391 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6392 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6393 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6394 | * group having more cpu_capacity will pickup more load compared to the | |
6395 | * group having less cpu_capacity. | |
89c4710e | 6396 | */ |
63b2ca30 | 6397 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6398 | { |
e3589f6c | 6399 | struct sched_group *sg = sd->groups; |
89c4710e | 6400 | |
94c95ba6 | 6401 | WARN_ON(!sg); |
e3589f6c PZ |
6402 | |
6403 | do { | |
afe06efd TC |
6404 | int cpu, max_cpu = -1; |
6405 | ||
e3589f6c | 6406 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); |
afe06efd TC |
6407 | |
6408 | if (!(sd->flags & SD_ASYM_PACKING)) | |
6409 | goto next; | |
6410 | ||
6411 | for_each_cpu(cpu, sched_group_cpus(sg)) { | |
6412 | if (max_cpu < 0) | |
6413 | max_cpu = cpu; | |
6414 | else if (sched_asym_prefer(cpu, max_cpu)) | |
6415 | max_cpu = cpu; | |
6416 | } | |
6417 | sg->asym_prefer_cpu = max_cpu; | |
6418 | ||
6419 | next: | |
e3589f6c PZ |
6420 | sg = sg->next; |
6421 | } while (sg != sd->groups); | |
89c4710e | 6422 | |
c1174876 | 6423 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6424 | return; |
aae6d3dd | 6425 | |
63b2ca30 | 6426 | update_group_capacity(sd, cpu); |
89c4710e SS |
6427 | } |
6428 | ||
7c16ec58 MT |
6429 | /* |
6430 | * Initializers for schedule domains | |
6431 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6432 | */ | |
6433 | ||
1d3504fc | 6434 | static int default_relax_domain_level = -1; |
60495e77 | 6435 | int sched_domain_level_max; |
1d3504fc HS |
6436 | |
6437 | static int __init setup_relax_domain_level(char *str) | |
6438 | { | |
a841f8ce DS |
6439 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6440 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6441 | |
1d3504fc HS |
6442 | return 1; |
6443 | } | |
6444 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6445 | ||
6446 | static void set_domain_attribute(struct sched_domain *sd, | |
6447 | struct sched_domain_attr *attr) | |
6448 | { | |
6449 | int request; | |
6450 | ||
6451 | if (!attr || attr->relax_domain_level < 0) { | |
6452 | if (default_relax_domain_level < 0) | |
6453 | return; | |
6454 | else | |
6455 | request = default_relax_domain_level; | |
6456 | } else | |
6457 | request = attr->relax_domain_level; | |
6458 | if (request < sd->level) { | |
6459 | /* turn off idle balance on this domain */ | |
c88d5910 | 6460 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6461 | } else { |
6462 | /* turn on idle balance on this domain */ | |
c88d5910 | 6463 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6464 | } |
6465 | } | |
6466 | ||
54ab4ff4 PZ |
6467 | static void __sdt_free(const struct cpumask *cpu_map); |
6468 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6469 | ||
2109b99e AH |
6470 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6471 | const struct cpumask *cpu_map) | |
6472 | { | |
6473 | switch (what) { | |
2109b99e | 6474 | case sa_rootdomain: |
822ff793 PZ |
6475 | if (!atomic_read(&d->rd->refcount)) |
6476 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6477 | case sa_sd: |
6478 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6479 | case sa_sd_storage: |
54ab4ff4 | 6480 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6481 | case sa_none: |
6482 | break; | |
6483 | } | |
6484 | } | |
3404c8d9 | 6485 | |
2109b99e AH |
6486 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6487 | const struct cpumask *cpu_map) | |
6488 | { | |
dce840a0 PZ |
6489 | memset(d, 0, sizeof(*d)); |
6490 | ||
54ab4ff4 PZ |
6491 | if (__sdt_alloc(cpu_map)) |
6492 | return sa_sd_storage; | |
dce840a0 PZ |
6493 | d->sd = alloc_percpu(struct sched_domain *); |
6494 | if (!d->sd) | |
6495 | return sa_sd_storage; | |
2109b99e | 6496 | d->rd = alloc_rootdomain(); |
dce840a0 | 6497 | if (!d->rd) |
21d42ccf | 6498 | return sa_sd; |
2109b99e AH |
6499 | return sa_rootdomain; |
6500 | } | |
57d885fe | 6501 | |
dce840a0 PZ |
6502 | /* |
6503 | * NULL the sd_data elements we've used to build the sched_domain and | |
6504 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6505 | * will not free the data we're using. | |
6506 | */ | |
6507 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6508 | { | |
6509 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6510 | |
6511 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6512 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6513 | ||
24fc7edb PZ |
6514 | if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref)) |
6515 | *per_cpu_ptr(sdd->sds, cpu) = NULL; | |
6516 | ||
e3589f6c | 6517 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6518 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6519 | |
63b2ca30 NP |
6520 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6521 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6522 | } |
6523 | ||
cb83b629 | 6524 | #ifdef CONFIG_NUMA |
cb83b629 | 6525 | static int sched_domains_numa_levels; |
e3fe70b1 | 6526 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6527 | static int *sched_domains_numa_distance; |
9942f79b | 6528 | int sched_max_numa_distance; |
cb83b629 PZ |
6529 | static struct cpumask ***sched_domains_numa_masks; |
6530 | static int sched_domains_curr_level; | |
143e1e28 | 6531 | #endif |
cb83b629 | 6532 | |
143e1e28 VG |
6533 | /* |
6534 | * SD_flags allowed in topology descriptions. | |
6535 | * | |
94f438c8 PZ |
6536 | * These flags are purely descriptive of the topology and do not prescribe |
6537 | * behaviour. Behaviour is artificial and mapped in the below sd_init() | |
6538 | * function: | |
143e1e28 | 6539 | * |
94f438c8 PZ |
6540 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
6541 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
6542 | * SD_NUMA - describes NUMA topologies | |
6543 | * SD_SHARE_POWERDOMAIN - describes shared power domain | |
1f6e6c7c | 6544 | * SD_ASYM_CPUCAPACITY - describes mixed capacity topologies |
94f438c8 PZ |
6545 | * |
6546 | * Odd one out, which beside describing the topology has a quirk also | |
6547 | * prescribes the desired behaviour that goes along with it: | |
143e1e28 | 6548 | * |
94f438c8 | 6549 | * SD_ASYM_PACKING - describes SMT quirks |
143e1e28 VG |
6550 | */ |
6551 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6552 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6553 | SD_SHARE_PKG_RESOURCES | \ |
6554 | SD_NUMA | \ | |
d77b3ed5 | 6555 | SD_ASYM_PACKING | \ |
1f6e6c7c | 6556 | SD_ASYM_CPUCAPACITY | \ |
d77b3ed5 | 6557 | SD_SHARE_POWERDOMAIN) |
cb83b629 PZ |
6558 | |
6559 | static struct sched_domain * | |
3676b13e | 6560 | sd_init(struct sched_domain_topology_level *tl, |
24fc7edb | 6561 | const struct cpumask *cpu_map, |
3676b13e | 6562 | struct sched_domain *child, int cpu) |
cb83b629 | 6563 | { |
24fc7edb PZ |
6564 | struct sd_data *sdd = &tl->data; |
6565 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); | |
6566 | int sd_id, sd_weight, sd_flags = 0; | |
143e1e28 VG |
6567 | |
6568 | #ifdef CONFIG_NUMA | |
6569 | /* | |
6570 | * Ugly hack to pass state to sd_numa_mask()... | |
6571 | */ | |
6572 | sched_domains_curr_level = tl->numa_level; | |
6573 | #endif | |
6574 | ||
6575 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6576 | ||
6577 | if (tl->sd_flags) | |
6578 | sd_flags = (*tl->sd_flags)(); | |
6579 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6580 | "wrong sd_flags in topology description\n")) | |
6581 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6582 | |
6583 | *sd = (struct sched_domain){ | |
6584 | .min_interval = sd_weight, | |
6585 | .max_interval = 2*sd_weight, | |
6586 | .busy_factor = 32, | |
870a0bb5 | 6587 | .imbalance_pct = 125, |
143e1e28 VG |
6588 | |
6589 | .cache_nice_tries = 0, | |
6590 | .busy_idx = 0, | |
6591 | .idle_idx = 0, | |
cb83b629 PZ |
6592 | .newidle_idx = 0, |
6593 | .wake_idx = 0, | |
6594 | .forkexec_idx = 0, | |
6595 | ||
6596 | .flags = 1*SD_LOAD_BALANCE | |
6597 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6598 | | 1*SD_BALANCE_EXEC |
6599 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6600 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6601 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6602 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6603 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6604 | | 0*SD_SERIALIZE |
cb83b629 | 6605 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6606 | | 0*SD_NUMA |
6607 | | sd_flags | |
cb83b629 | 6608 | , |
143e1e28 | 6609 | |
cb83b629 PZ |
6610 | .last_balance = jiffies, |
6611 | .balance_interval = sd_weight, | |
143e1e28 | 6612 | .smt_gain = 0, |
2b4cfe64 JL |
6613 | .max_newidle_lb_cost = 0, |
6614 | .next_decay_max_lb_cost = jiffies, | |
3676b13e | 6615 | .child = child, |
143e1e28 VG |
6616 | #ifdef CONFIG_SCHED_DEBUG |
6617 | .name = tl->name, | |
6618 | #endif | |
cb83b629 | 6619 | }; |
cb83b629 | 6620 | |
24fc7edb PZ |
6621 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
6622 | sd_id = cpumask_first(sched_domain_span(sd)); | |
6623 | ||
cb83b629 | 6624 | /* |
143e1e28 | 6625 | * Convert topological properties into behaviour. |
cb83b629 | 6626 | */ |
143e1e28 | 6627 | |
9ee1cda5 MR |
6628 | if (sd->flags & SD_ASYM_CPUCAPACITY) { |
6629 | struct sched_domain *t = sd; | |
6630 | ||
6631 | for_each_lower_domain(t) | |
6632 | t->flags |= SD_BALANCE_WAKE; | |
6633 | } | |
6634 | ||
5d4dfddd | 6635 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6636 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6637 | sd->imbalance_pct = 110; |
6638 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6639 | |
6640 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6641 | sd->imbalance_pct = 117; | |
6642 | sd->cache_nice_tries = 1; | |
6643 | sd->busy_idx = 2; | |
6644 | ||
6645 | #ifdef CONFIG_NUMA | |
6646 | } else if (sd->flags & SD_NUMA) { | |
6647 | sd->cache_nice_tries = 2; | |
6648 | sd->busy_idx = 3; | |
6649 | sd->idle_idx = 2; | |
6650 | ||
6651 | sd->flags |= SD_SERIALIZE; | |
6652 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6653 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6654 | SD_BALANCE_FORK | | |
6655 | SD_WAKE_AFFINE); | |
6656 | } | |
6657 | ||
6658 | #endif | |
6659 | } else { | |
6660 | sd->flags |= SD_PREFER_SIBLING; | |
6661 | sd->cache_nice_tries = 1; | |
6662 | sd->busy_idx = 2; | |
6663 | sd->idle_idx = 1; | |
6664 | } | |
6665 | ||
24fc7edb PZ |
6666 | /* |
6667 | * For all levels sharing cache; connect a sched_domain_shared | |
6668 | * instance. | |
6669 | */ | |
6670 | if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6671 | sd->shared = *per_cpu_ptr(sdd->sds, sd_id); | |
6672 | atomic_inc(&sd->shared->ref); | |
0e369d75 | 6673 | atomic_set(&sd->shared->nr_busy_cpus, sd_weight); |
24fc7edb PZ |
6674 | } |
6675 | ||
6676 | sd->private = sdd; | |
cb83b629 PZ |
6677 | |
6678 | return sd; | |
6679 | } | |
6680 | ||
143e1e28 VG |
6681 | /* |
6682 | * Topology list, bottom-up. | |
6683 | */ | |
6684 | static struct sched_domain_topology_level default_topology[] = { | |
6685 | #ifdef CONFIG_SCHED_SMT | |
6686 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6687 | #endif | |
6688 | #ifdef CONFIG_SCHED_MC | |
6689 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6690 | #endif |
6691 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6692 | { NULL, }, | |
6693 | }; | |
6694 | ||
c6e1e7b5 JG |
6695 | static struct sched_domain_topology_level *sched_domain_topology = |
6696 | default_topology; | |
143e1e28 VG |
6697 | |
6698 | #define for_each_sd_topology(tl) \ | |
6699 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6700 | ||
6701 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6702 | { | |
8f37961c TC |
6703 | if (WARN_ON_ONCE(sched_smp_initialized)) |
6704 | return; | |
6705 | ||
143e1e28 VG |
6706 | sched_domain_topology = tl; |
6707 | } | |
6708 | ||
6709 | #ifdef CONFIG_NUMA | |
6710 | ||
cb83b629 PZ |
6711 | static const struct cpumask *sd_numa_mask(int cpu) |
6712 | { | |
6713 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6714 | } | |
6715 | ||
d039ac60 PZ |
6716 | static void sched_numa_warn(const char *str) |
6717 | { | |
6718 | static int done = false; | |
6719 | int i,j; | |
6720 | ||
6721 | if (done) | |
6722 | return; | |
6723 | ||
6724 | done = true; | |
6725 | ||
6726 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6727 | ||
6728 | for (i = 0; i < nr_node_ids; i++) { | |
6729 | printk(KERN_WARNING " "); | |
6730 | for (j = 0; j < nr_node_ids; j++) | |
6731 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6732 | printk(KERN_CONT "\n"); | |
6733 | } | |
6734 | printk(KERN_WARNING "\n"); | |
6735 | } | |
6736 | ||
9942f79b | 6737 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6738 | { |
6739 | int i; | |
6740 | ||
6741 | if (distance == node_distance(0, 0)) | |
6742 | return true; | |
6743 | ||
6744 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6745 | if (sched_domains_numa_distance[i] == distance) | |
6746 | return true; | |
6747 | } | |
6748 | ||
6749 | return false; | |
6750 | } | |
6751 | ||
e3fe70b1 RR |
6752 | /* |
6753 | * A system can have three types of NUMA topology: | |
6754 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6755 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6756 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6757 | * | |
6758 | * The difference between a glueless mesh topology and a backplane | |
6759 | * topology lies in whether communication between not directly | |
6760 | * connected nodes goes through intermediary nodes (where programs | |
6761 | * could run), or through backplane controllers. This affects | |
6762 | * placement of programs. | |
6763 | * | |
6764 | * The type of topology can be discerned with the following tests: | |
6765 | * - If the maximum distance between any nodes is 1 hop, the system | |
6766 | * is directly connected. | |
6767 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6768 | * there is an intermediary node C, which is < N hops away from both | |
6769 | * nodes A and B, the system is a glueless mesh. | |
6770 | */ | |
6771 | static void init_numa_topology_type(void) | |
6772 | { | |
6773 | int a, b, c, n; | |
6774 | ||
6775 | n = sched_max_numa_distance; | |
6776 | ||
e237882b | 6777 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6778 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6779 | return; |
6780 | } | |
e3fe70b1 RR |
6781 | |
6782 | for_each_online_node(a) { | |
6783 | for_each_online_node(b) { | |
6784 | /* Find two nodes furthest removed from each other. */ | |
6785 | if (node_distance(a, b) < n) | |
6786 | continue; | |
6787 | ||
6788 | /* Is there an intermediary node between a and b? */ | |
6789 | for_each_online_node(c) { | |
6790 | if (node_distance(a, c) < n && | |
6791 | node_distance(b, c) < n) { | |
6792 | sched_numa_topology_type = | |
6793 | NUMA_GLUELESS_MESH; | |
6794 | return; | |
6795 | } | |
6796 | } | |
6797 | ||
6798 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6799 | return; | |
6800 | } | |
6801 | } | |
6802 | } | |
6803 | ||
cb83b629 PZ |
6804 | static void sched_init_numa(void) |
6805 | { | |
6806 | int next_distance, curr_distance = node_distance(0, 0); | |
6807 | struct sched_domain_topology_level *tl; | |
6808 | int level = 0; | |
6809 | int i, j, k; | |
6810 | ||
cb83b629 PZ |
6811 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6812 | if (!sched_domains_numa_distance) | |
6813 | return; | |
6814 | ||
6815 | /* | |
6816 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6817 | * unique distances in the node_distance() table. | |
6818 | * | |
6819 | * Assumes node_distance(0,j) includes all distances in | |
6820 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6821 | */ |
6822 | next_distance = curr_distance; | |
6823 | for (i = 0; i < nr_node_ids; i++) { | |
6824 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6825 | for (k = 0; k < nr_node_ids; k++) { |
6826 | int distance = node_distance(i, k); | |
6827 | ||
6828 | if (distance > curr_distance && | |
6829 | (distance < next_distance || | |
6830 | next_distance == curr_distance)) | |
6831 | next_distance = distance; | |
6832 | ||
6833 | /* | |
6834 | * While not a strong assumption it would be nice to know | |
6835 | * about cases where if node A is connected to B, B is not | |
6836 | * equally connected to A. | |
6837 | */ | |
6838 | if (sched_debug() && node_distance(k, i) != distance) | |
6839 | sched_numa_warn("Node-distance not symmetric"); | |
6840 | ||
6841 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6842 | sched_numa_warn("Node-0 not representative"); | |
6843 | } | |
6844 | if (next_distance != curr_distance) { | |
6845 | sched_domains_numa_distance[level++] = next_distance; | |
6846 | sched_domains_numa_levels = level; | |
6847 | curr_distance = next_distance; | |
6848 | } else break; | |
cb83b629 | 6849 | } |
d039ac60 PZ |
6850 | |
6851 | /* | |
6852 | * In case of sched_debug() we verify the above assumption. | |
6853 | */ | |
6854 | if (!sched_debug()) | |
6855 | break; | |
cb83b629 | 6856 | } |
c123588b AR |
6857 | |
6858 | if (!level) | |
6859 | return; | |
6860 | ||
cb83b629 PZ |
6861 | /* |
6862 | * 'level' contains the number of unique distances, excluding the | |
6863 | * identity distance node_distance(i,i). | |
6864 | * | |
28b4a521 | 6865 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6866 | * numbers. |
6867 | */ | |
6868 | ||
5f7865f3 TC |
6869 | /* |
6870 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6871 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6872 | * the array will contain less then 'level' members. This could be | |
6873 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6874 | * in other functions. | |
6875 | * | |
6876 | * We reset it to 'level' at the end of this function. | |
6877 | */ | |
6878 | sched_domains_numa_levels = 0; | |
6879 | ||
cb83b629 PZ |
6880 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6881 | if (!sched_domains_numa_masks) | |
6882 | return; | |
6883 | ||
6884 | /* | |
6885 | * Now for each level, construct a mask per node which contains all | |
6886 | * cpus of nodes that are that many hops away from us. | |
6887 | */ | |
6888 | for (i = 0; i < level; i++) { | |
6889 | sched_domains_numa_masks[i] = | |
6890 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6891 | if (!sched_domains_numa_masks[i]) | |
6892 | return; | |
6893 | ||
6894 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6895 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6896 | if (!mask) |
6897 | return; | |
6898 | ||
6899 | sched_domains_numa_masks[i][j] = mask; | |
6900 | ||
9c03ee14 | 6901 | for_each_node(k) { |
dd7d8634 | 6902 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6903 | continue; |
6904 | ||
6905 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6906 | } | |
6907 | } | |
6908 | } | |
6909 | ||
143e1e28 VG |
6910 | /* Compute default topology size */ |
6911 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6912 | ||
c515db8c | 6913 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6914 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6915 | if (!tl) | |
6916 | return; | |
6917 | ||
6918 | /* | |
6919 | * Copy the default topology bits.. | |
6920 | */ | |
143e1e28 VG |
6921 | for (i = 0; sched_domain_topology[i].mask; i++) |
6922 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6923 | |
6924 | /* | |
6925 | * .. and append 'j' levels of NUMA goodness. | |
6926 | */ | |
6927 | for (j = 0; j < level; i++, j++) { | |
6928 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6929 | .mask = sd_numa_mask, |
143e1e28 | 6930 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6931 | .flags = SDTL_OVERLAP, |
6932 | .numa_level = j, | |
143e1e28 | 6933 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6934 | }; |
6935 | } | |
6936 | ||
6937 | sched_domain_topology = tl; | |
5f7865f3 TC |
6938 | |
6939 | sched_domains_numa_levels = level; | |
9942f79b | 6940 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6941 | |
6942 | init_numa_topology_type(); | |
cb83b629 | 6943 | } |
301a5cba | 6944 | |
135fb3e1 | 6945 | static void sched_domains_numa_masks_set(unsigned int cpu) |
301a5cba | 6946 | { |
301a5cba | 6947 | int node = cpu_to_node(cpu); |
135fb3e1 | 6948 | int i, j; |
301a5cba TC |
6949 | |
6950 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6951 | for (j = 0; j < nr_node_ids; j++) { | |
6952 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6953 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6954 | } | |
6955 | } | |
6956 | } | |
6957 | ||
135fb3e1 | 6958 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
301a5cba TC |
6959 | { |
6960 | int i, j; | |
135fb3e1 | 6961 | |
301a5cba TC |
6962 | for (i = 0; i < sched_domains_numa_levels; i++) { |
6963 | for (j = 0; j < nr_node_ids; j++) | |
6964 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6965 | } | |
6966 | } | |
6967 | ||
cb83b629 | 6968 | #else |
135fb3e1 TG |
6969 | static inline void sched_init_numa(void) { } |
6970 | static void sched_domains_numa_masks_set(unsigned int cpu) { } | |
6971 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
cb83b629 PZ |
6972 | #endif /* CONFIG_NUMA */ |
6973 | ||
54ab4ff4 PZ |
6974 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6975 | { | |
6976 | struct sched_domain_topology_level *tl; | |
6977 | int j; | |
6978 | ||
27723a68 | 6979 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6980 | struct sd_data *sdd = &tl->data; |
6981 | ||
6982 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6983 | if (!sdd->sd) | |
6984 | return -ENOMEM; | |
6985 | ||
24fc7edb PZ |
6986 | sdd->sds = alloc_percpu(struct sched_domain_shared *); |
6987 | if (!sdd->sds) | |
6988 | return -ENOMEM; | |
6989 | ||
54ab4ff4 PZ |
6990 | sdd->sg = alloc_percpu(struct sched_group *); |
6991 | if (!sdd->sg) | |
6992 | return -ENOMEM; | |
6993 | ||
63b2ca30 NP |
6994 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6995 | if (!sdd->sgc) | |
9c3f75cb PZ |
6996 | return -ENOMEM; |
6997 | ||
54ab4ff4 PZ |
6998 | for_each_cpu(j, cpu_map) { |
6999 | struct sched_domain *sd; | |
24fc7edb | 7000 | struct sched_domain_shared *sds; |
54ab4ff4 | 7001 | struct sched_group *sg; |
63b2ca30 | 7002 | struct sched_group_capacity *sgc; |
54ab4ff4 | 7003 | |
5cc389bc | 7004 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
7005 | GFP_KERNEL, cpu_to_node(j)); |
7006 | if (!sd) | |
7007 | return -ENOMEM; | |
7008 | ||
7009 | *per_cpu_ptr(sdd->sd, j) = sd; | |
7010 | ||
24fc7edb PZ |
7011 | sds = kzalloc_node(sizeof(struct sched_domain_shared), |
7012 | GFP_KERNEL, cpu_to_node(j)); | |
7013 | if (!sds) | |
7014 | return -ENOMEM; | |
7015 | ||
7016 | *per_cpu_ptr(sdd->sds, j) = sds; | |
7017 | ||
54ab4ff4 PZ |
7018 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
7019 | GFP_KERNEL, cpu_to_node(j)); | |
7020 | if (!sg) | |
7021 | return -ENOMEM; | |
7022 | ||
30b4e9eb IM |
7023 | sg->next = sg; |
7024 | ||
54ab4ff4 | 7025 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 7026 | |
63b2ca30 | 7027 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 7028 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 7029 | if (!sgc) |
9c3f75cb PZ |
7030 | return -ENOMEM; |
7031 | ||
63b2ca30 | 7032 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
7033 | } |
7034 | } | |
7035 | ||
7036 | return 0; | |
7037 | } | |
7038 | ||
7039 | static void __sdt_free(const struct cpumask *cpu_map) | |
7040 | { | |
7041 | struct sched_domain_topology_level *tl; | |
7042 | int j; | |
7043 | ||
27723a68 | 7044 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
7045 | struct sd_data *sdd = &tl->data; |
7046 | ||
7047 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 7048 | struct sched_domain *sd; |
7049 | ||
7050 | if (sdd->sd) { | |
7051 | sd = *per_cpu_ptr(sdd->sd, j); | |
7052 | if (sd && (sd->flags & SD_OVERLAP)) | |
7053 | free_sched_groups(sd->groups, 0); | |
7054 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
7055 | } | |
7056 | ||
24fc7edb PZ |
7057 | if (sdd->sds) |
7058 | kfree(*per_cpu_ptr(sdd->sds, j)); | |
fb2cf2c6 | 7059 | if (sdd->sg) |
7060 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
7061 | if (sdd->sgc) |
7062 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
7063 | } |
7064 | free_percpu(sdd->sd); | |
fb2cf2c6 | 7065 | sdd->sd = NULL; |
24fc7edb PZ |
7066 | free_percpu(sdd->sds); |
7067 | sdd->sds = NULL; | |
54ab4ff4 | 7068 | free_percpu(sdd->sg); |
fb2cf2c6 | 7069 | sdd->sg = NULL; |
63b2ca30 NP |
7070 | free_percpu(sdd->sgc); |
7071 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
7072 | } |
7073 | } | |
7074 | ||
2c402dc3 | 7075 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
7076 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
7077 | struct sched_domain *child, int cpu) | |
2c402dc3 | 7078 | { |
24fc7edb | 7079 | struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu); |
2c402dc3 | 7080 | |
60495e77 PZ |
7081 | if (child) { |
7082 | sd->level = child->level + 1; | |
7083 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 7084 | child->parent = sd; |
6ae72dff PZ |
7085 | |
7086 | if (!cpumask_subset(sched_domain_span(child), | |
7087 | sched_domain_span(sd))) { | |
7088 | pr_err("BUG: arch topology borken\n"); | |
7089 | #ifdef CONFIG_SCHED_DEBUG | |
7090 | pr_err(" the %s domain not a subset of the %s domain\n", | |
7091 | child->name, sd->name); | |
7092 | #endif | |
7093 | /* Fixup, ensure @sd has at least @child cpus. */ | |
7094 | cpumask_or(sched_domain_span(sd), | |
7095 | sched_domain_span(sd), | |
7096 | sched_domain_span(child)); | |
7097 | } | |
7098 | ||
60495e77 | 7099 | } |
a841f8ce | 7100 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
7101 | |
7102 | return sd; | |
7103 | } | |
7104 | ||
2109b99e AH |
7105 | /* |
7106 | * Build sched domains for a given set of cpus and attach the sched domains | |
7107 | * to the individual cpus | |
7108 | */ | |
dce840a0 PZ |
7109 | static int build_sched_domains(const struct cpumask *cpu_map, |
7110 | struct sched_domain_attr *attr) | |
2109b99e | 7111 | { |
1c632169 | 7112 | enum s_alloc alloc_state; |
dce840a0 | 7113 | struct sched_domain *sd; |
2109b99e | 7114 | struct s_data d; |
cd92bfd3 | 7115 | struct rq *rq = NULL; |
822ff793 | 7116 | int i, ret = -ENOMEM; |
9c1cfda2 | 7117 | |
2109b99e AH |
7118 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7119 | if (alloc_state != sa_rootdomain) | |
7120 | goto error; | |
9c1cfda2 | 7121 | |
dce840a0 | 7122 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7123 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7124 | struct sched_domain_topology_level *tl; |
7125 | ||
3bd65a80 | 7126 | sd = NULL; |
27723a68 | 7127 | for_each_sd_topology(tl) { |
4a850cbe | 7128 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
7129 | if (tl == sched_domain_topology) |
7130 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
7131 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
7132 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
7133 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
7134 | break; | |
e3589f6c | 7135 | } |
dce840a0 PZ |
7136 | } |
7137 | ||
7138 | /* Build the groups for the domains */ | |
7139 | for_each_cpu(i, cpu_map) { | |
7140 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7141 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
7142 | if (sd->flags & SD_OVERLAP) { |
7143 | if (build_overlap_sched_groups(sd, i)) | |
7144 | goto error; | |
7145 | } else { | |
7146 | if (build_sched_groups(sd, i)) | |
7147 | goto error; | |
7148 | } | |
1cf51902 | 7149 | } |
a06dadbe | 7150 | } |
9c1cfda2 | 7151 | |
ced549fa | 7152 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
7153 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7154 | if (!cpumask_test_cpu(i, cpu_map)) | |
7155 | continue; | |
9c1cfda2 | 7156 | |
dce840a0 PZ |
7157 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7158 | claim_allocations(i, sd); | |
63b2ca30 | 7159 | init_sched_groups_capacity(i, sd); |
dce840a0 | 7160 | } |
f712c0c7 | 7161 | } |
9c1cfda2 | 7162 | |
1da177e4 | 7163 | /* Attach the domains */ |
dce840a0 | 7164 | rcu_read_lock(); |
abcd083a | 7165 | for_each_cpu(i, cpu_map) { |
cd92bfd3 | 7166 | rq = cpu_rq(i); |
21d42ccf | 7167 | sd = *per_cpu_ptr(d.sd, i); |
cd92bfd3 DE |
7168 | |
7169 | /* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */ | |
7170 | if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity)) | |
7171 | WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig); | |
7172 | ||
49a02c51 | 7173 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7174 | } |
dce840a0 | 7175 | rcu_read_unlock(); |
51888ca2 | 7176 | |
a18a579e | 7177 | if (rq && sched_debug_enabled) { |
cd92bfd3 DE |
7178 | pr_info("span: %*pbl (max cpu_capacity = %lu)\n", |
7179 | cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity); | |
7180 | } | |
7181 | ||
822ff793 | 7182 | ret = 0; |
51888ca2 | 7183 | error: |
2109b99e | 7184 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7185 | return ret; |
1da177e4 | 7186 | } |
029190c5 | 7187 | |
acc3f5d7 | 7188 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7189 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7190 | static struct sched_domain_attr *dattr_cur; |
7191 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7192 | |
7193 | /* | |
7194 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7195 | * cpumask) fails, then fallback to a single sched domain, |
7196 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7197 | */ |
4212823f | 7198 | static cpumask_var_t fallback_doms; |
029190c5 | 7199 | |
ee79d1bd HC |
7200 | /* |
7201 | * arch_update_cpu_topology lets virtualized architectures update the | |
7202 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7203 | * or 0 if it stayed the same. | |
7204 | */ | |
52f5684c | 7205 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 7206 | { |
ee79d1bd | 7207 | return 0; |
22e52b07 HC |
7208 | } |
7209 | ||
acc3f5d7 RR |
7210 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7211 | { | |
7212 | int i; | |
7213 | cpumask_var_t *doms; | |
7214 | ||
7215 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7216 | if (!doms) | |
7217 | return NULL; | |
7218 | for (i = 0; i < ndoms; i++) { | |
7219 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7220 | free_sched_domains(doms, i); | |
7221 | return NULL; | |
7222 | } | |
7223 | } | |
7224 | return doms; | |
7225 | } | |
7226 | ||
7227 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7228 | { | |
7229 | unsigned int i; | |
7230 | for (i = 0; i < ndoms; i++) | |
7231 | free_cpumask_var(doms[i]); | |
7232 | kfree(doms); | |
7233 | } | |
7234 | ||
1a20ff27 | 7235 | /* |
41a2d6cf | 7236 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7237 | * For now this just excludes isolated cpus, but could be used to |
7238 | * exclude other special cases in the future. | |
1a20ff27 | 7239 | */ |
c4a8849a | 7240 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7241 | { |
7378547f MM |
7242 | int err; |
7243 | ||
22e52b07 | 7244 | arch_update_cpu_topology(); |
029190c5 | 7245 | ndoms_cur = 1; |
acc3f5d7 | 7246 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7247 | if (!doms_cur) |
acc3f5d7 RR |
7248 | doms_cur = &fallback_doms; |
7249 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 7250 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7251 | register_sched_domain_sysctl(); |
7378547f MM |
7252 | |
7253 | return err; | |
1a20ff27 DG |
7254 | } |
7255 | ||
1a20ff27 DG |
7256 | /* |
7257 | * Detach sched domains from a group of cpus specified in cpu_map | |
7258 | * These cpus will now be attached to the NULL domain | |
7259 | */ | |
96f874e2 | 7260 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7261 | { |
7262 | int i; | |
7263 | ||
dce840a0 | 7264 | rcu_read_lock(); |
abcd083a | 7265 | for_each_cpu(i, cpu_map) |
57d885fe | 7266 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7267 | rcu_read_unlock(); |
1a20ff27 DG |
7268 | } |
7269 | ||
1d3504fc HS |
7270 | /* handle null as "default" */ |
7271 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7272 | struct sched_domain_attr *new, int idx_new) | |
7273 | { | |
7274 | struct sched_domain_attr tmp; | |
7275 | ||
7276 | /* fast path */ | |
7277 | if (!new && !cur) | |
7278 | return 1; | |
7279 | ||
7280 | tmp = SD_ATTR_INIT; | |
7281 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7282 | new ? (new + idx_new) : &tmp, | |
7283 | sizeof(struct sched_domain_attr)); | |
7284 | } | |
7285 | ||
029190c5 PJ |
7286 | /* |
7287 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7288 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7289 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7290 | * It destroys each deleted domain and builds each new domain. | |
7291 | * | |
acc3f5d7 | 7292 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7293 | * The masks don't intersect (don't overlap.) We should setup one |
7294 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7295 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7296 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7297 | * it as it is. | |
7298 | * | |
acc3f5d7 RR |
7299 | * The passed in 'doms_new' should be allocated using |
7300 | * alloc_sched_domains. This routine takes ownership of it and will | |
7301 | * free_sched_domains it when done with it. If the caller failed the | |
7302 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7303 | * and partition_sched_domains() will fallback to the single partition | |
7304 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7305 | * |
96f874e2 | 7306 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7307 | * ndoms_new == 0 is a special case for destroying existing domains, |
7308 | * and it will not create the default domain. | |
dfb512ec | 7309 | * |
029190c5 PJ |
7310 | * Call with hotplug lock held |
7311 | */ | |
acc3f5d7 | 7312 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7313 | struct sched_domain_attr *dattr_new) |
029190c5 | 7314 | { |
dfb512ec | 7315 | int i, j, n; |
d65bd5ec | 7316 | int new_topology; |
029190c5 | 7317 | |
712555ee | 7318 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7319 | |
7378547f MM |
7320 | /* always unregister in case we don't destroy any domains */ |
7321 | unregister_sched_domain_sysctl(); | |
7322 | ||
d65bd5ec HC |
7323 | /* Let architecture update cpu core mappings. */ |
7324 | new_topology = arch_update_cpu_topology(); | |
7325 | ||
dfb512ec | 7326 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7327 | |
7328 | /* Destroy deleted domains */ | |
7329 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7330 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7331 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7332 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7333 | goto match1; |
7334 | } | |
7335 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7336 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7337 | match1: |
7338 | ; | |
7339 | } | |
7340 | ||
c8d2d47a | 7341 | n = ndoms_cur; |
e761b772 | 7342 | if (doms_new == NULL) { |
c8d2d47a | 7343 | n = 0; |
acc3f5d7 | 7344 | doms_new = &fallback_doms; |
6ad4c188 | 7345 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7346 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7347 | } |
7348 | ||
029190c5 PJ |
7349 | /* Build new domains */ |
7350 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7351 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7352 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7353 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7354 | goto match2; |
7355 | } | |
7356 | /* no match - add a new doms_new */ | |
dce840a0 | 7357 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7358 | match2: |
7359 | ; | |
7360 | } | |
7361 | ||
7362 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7363 | if (doms_cur != &fallback_doms) |
7364 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7365 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7366 | doms_cur = doms_new; |
1d3504fc | 7367 | dattr_cur = dattr_new; |
029190c5 | 7368 | ndoms_cur = ndoms_new; |
7378547f MM |
7369 | |
7370 | register_sched_domain_sysctl(); | |
a1835615 | 7371 | |
712555ee | 7372 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7373 | } |
7374 | ||
d35be8ba SB |
7375 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7376 | ||
1da177e4 | 7377 | /* |
3a101d05 TH |
7378 | * Update cpusets according to cpu_active mask. If cpusets are |
7379 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7380 | * around partition_sched_domains(). | |
d35be8ba SB |
7381 | * |
7382 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7383 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7384 | */ |
40190a78 | 7385 | static void cpuset_cpu_active(void) |
e761b772 | 7386 | { |
40190a78 | 7387 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7388 | /* |
7389 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7390 | * resume sequence. As long as this is not the last online | |
7391 | * operation in the resume sequence, just build a single sched | |
7392 | * domain, ignoring cpusets. | |
7393 | */ | |
7394 | num_cpus_frozen--; | |
7395 | if (likely(num_cpus_frozen)) { | |
7396 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 7397 | return; |
d35be8ba | 7398 | } |
d35be8ba SB |
7399 | /* |
7400 | * This is the last CPU online operation. So fall through and | |
7401 | * restore the original sched domains by considering the | |
7402 | * cpuset configurations. | |
7403 | */ | |
3a101d05 | 7404 | } |
135fb3e1 | 7405 | cpuset_update_active_cpus(true); |
3a101d05 | 7406 | } |
e761b772 | 7407 | |
40190a78 | 7408 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7409 | { |
3c18d447 | 7410 | unsigned long flags; |
3c18d447 | 7411 | struct dl_bw *dl_b; |
533445c6 OS |
7412 | bool overflow; |
7413 | int cpus; | |
3c18d447 | 7414 | |
40190a78 | 7415 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
7416 | rcu_read_lock_sched(); |
7417 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7418 | |
533445c6 OS |
7419 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7420 | cpus = dl_bw_cpus(cpu); | |
7421 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7422 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7423 | |
533445c6 | 7424 | rcu_read_unlock_sched(); |
3c18d447 | 7425 | |
533445c6 | 7426 | if (overflow) |
135fb3e1 | 7427 | return -EBUSY; |
7ddf96b0 | 7428 | cpuset_update_active_cpus(false); |
135fb3e1 | 7429 | } else { |
d35be8ba SB |
7430 | num_cpus_frozen++; |
7431 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7432 | } |
135fb3e1 | 7433 | return 0; |
e761b772 | 7434 | } |
e761b772 | 7435 | |
40190a78 | 7436 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7437 | { |
7d976699 TG |
7438 | struct rq *rq = cpu_rq(cpu); |
7439 | unsigned long flags; | |
7440 | ||
40190a78 | 7441 | set_cpu_active(cpu, true); |
135fb3e1 | 7442 | |
40190a78 | 7443 | if (sched_smp_initialized) { |
135fb3e1 | 7444 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7445 | cpuset_cpu_active(); |
e761b772 | 7446 | } |
7d976699 TG |
7447 | |
7448 | /* | |
7449 | * Put the rq online, if not already. This happens: | |
7450 | * | |
7451 | * 1) In the early boot process, because we build the real domains | |
7452 | * after all cpus have been brought up. | |
7453 | * | |
7454 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7455 | * domains. | |
7456 | */ | |
7457 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7458 | if (rq->rd) { | |
7459 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7460 | set_rq_online(rq); | |
7461 | } | |
7462 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7463 | ||
7464 | update_max_interval(); | |
7465 | ||
40190a78 | 7466 | return 0; |
135fb3e1 TG |
7467 | } |
7468 | ||
40190a78 | 7469 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7470 | { |
135fb3e1 TG |
7471 | int ret; |
7472 | ||
40190a78 | 7473 | set_cpu_active(cpu, false); |
b2454caa PZ |
7474 | /* |
7475 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7476 | * users of this state to go away such that all new such users will | |
7477 | * observe it. | |
7478 | * | |
7479 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
7480 | * not imply sync_sched(), so wait for both. | |
7481 | * | |
7482 | * Do sync before park smpboot threads to take care the rcu boost case. | |
7483 | */ | |
7484 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
7485 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
7486 | else | |
7487 | synchronize_rcu(); | |
40190a78 TG |
7488 | |
7489 | if (!sched_smp_initialized) | |
7490 | return 0; | |
7491 | ||
7492 | ret = cpuset_cpu_inactive(cpu); | |
7493 | if (ret) { | |
7494 | set_cpu_active(cpu, true); | |
7495 | return ret; | |
135fb3e1 | 7496 | } |
40190a78 TG |
7497 | sched_domains_numa_masks_clear(cpu); |
7498 | return 0; | |
135fb3e1 TG |
7499 | } |
7500 | ||
94baf7a5 TG |
7501 | static void sched_rq_cpu_starting(unsigned int cpu) |
7502 | { | |
7503 | struct rq *rq = cpu_rq(cpu); | |
7504 | ||
7505 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7506 | update_max_interval(); |
7507 | } | |
7508 | ||
135fb3e1 TG |
7509 | int sched_cpu_starting(unsigned int cpu) |
7510 | { | |
7511 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 7512 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 7513 | return 0; |
e761b772 | 7514 | } |
e761b772 | 7515 | |
f2785ddb TG |
7516 | #ifdef CONFIG_HOTPLUG_CPU |
7517 | int sched_cpu_dying(unsigned int cpu) | |
7518 | { | |
7519 | struct rq *rq = cpu_rq(cpu); | |
7520 | unsigned long flags; | |
7521 | ||
7522 | /* Handle pending wakeups and then migrate everything off */ | |
7523 | sched_ttwu_pending(); | |
7524 | raw_spin_lock_irqsave(&rq->lock, flags); | |
7525 | if (rq->rd) { | |
7526 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7527 | set_rq_offline(rq); | |
7528 | } | |
7529 | migrate_tasks(rq); | |
7530 | BUG_ON(rq->nr_running != 1); | |
7531 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
7532 | calc_load_migrate(rq); | |
7533 | update_max_interval(); | |
20a5c8cc | 7534 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 7535 | hrtick_clear(rq); |
f2785ddb TG |
7536 | return 0; |
7537 | } | |
7538 | #endif | |
7539 | ||
1b568f0a PZ |
7540 | #ifdef CONFIG_SCHED_SMT |
7541 | DEFINE_STATIC_KEY_FALSE(sched_smt_present); | |
7542 | ||
7543 | static void sched_init_smt(void) | |
7544 | { | |
7545 | /* | |
7546 | * We've enumerated all CPUs and will assume that if any CPU | |
7547 | * has SMT siblings, CPU0 will too. | |
7548 | */ | |
7549 | if (cpumask_weight(cpu_smt_mask(0)) > 1) | |
7550 | static_branch_enable(&sched_smt_present); | |
7551 | } | |
7552 | #else | |
7553 | static inline void sched_init_smt(void) { } | |
7554 | #endif | |
7555 | ||
1da177e4 LT |
7556 | void __init sched_init_smp(void) |
7557 | { | |
dcc30a35 RR |
7558 | cpumask_var_t non_isolated_cpus; |
7559 | ||
7560 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7561 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7562 | |
cb83b629 PZ |
7563 | sched_init_numa(); |
7564 | ||
6acce3ef PZ |
7565 | /* |
7566 | * There's no userspace yet to cause hotplug operations; hence all the | |
7567 | * cpu masks are stable and all blatant races in the below code cannot | |
7568 | * happen. | |
7569 | */ | |
712555ee | 7570 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7571 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7572 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7573 | if (cpumask_empty(non_isolated_cpus)) | |
7574 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7575 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7576 | |
5c1e1767 | 7577 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 7578 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7579 | BUG(); |
19978ca6 | 7580 | sched_init_granularity(); |
dcc30a35 | 7581 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7582 | |
0e3900e6 | 7583 | init_sched_rt_class(); |
1baca4ce | 7584 | init_sched_dl_class(); |
1b568f0a PZ |
7585 | |
7586 | sched_init_smt(); | |
9881b024 | 7587 | sched_clock_init_late(); |
1b568f0a | 7588 | |
e26fbffd | 7589 | sched_smp_initialized = true; |
1da177e4 | 7590 | } |
e26fbffd TG |
7591 | |
7592 | static int __init migration_init(void) | |
7593 | { | |
94baf7a5 | 7594 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 7595 | return 0; |
1da177e4 | 7596 | } |
e26fbffd TG |
7597 | early_initcall(migration_init); |
7598 | ||
1da177e4 LT |
7599 | #else |
7600 | void __init sched_init_smp(void) | |
7601 | { | |
19978ca6 | 7602 | sched_init_granularity(); |
9881b024 | 7603 | sched_clock_init_late(); |
1da177e4 LT |
7604 | } |
7605 | #endif /* CONFIG_SMP */ | |
7606 | ||
7607 | int in_sched_functions(unsigned long addr) | |
7608 | { | |
1da177e4 LT |
7609 | return in_lock_functions(addr) || |
7610 | (addr >= (unsigned long)__sched_text_start | |
7611 | && addr < (unsigned long)__sched_text_end); | |
7612 | } | |
7613 | ||
029632fb | 7614 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7615 | /* |
7616 | * Default task group. | |
7617 | * Every task in system belongs to this group at bootup. | |
7618 | */ | |
029632fb | 7619 | struct task_group root_task_group; |
35cf4e50 | 7620 | LIST_HEAD(task_groups); |
b0367629 WL |
7621 | |
7622 | /* Cacheline aligned slab cache for task_group */ | |
7623 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7624 | #endif |
6f505b16 | 7625 | |
e6252c3e | 7626 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7627 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7628 | |
9dcb8b68 LT |
7629 | #define WAIT_TABLE_BITS 8 |
7630 | #define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS) | |
7631 | static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned; | |
7632 | ||
7633 | wait_queue_head_t *bit_waitqueue(void *word, int bit) | |
7634 | { | |
7635 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; | |
7636 | unsigned long val = (unsigned long)word << shift | bit; | |
7637 | ||
7638 | return bit_wait_table + hash_long(val, WAIT_TABLE_BITS); | |
7639 | } | |
7640 | EXPORT_SYMBOL(bit_waitqueue); | |
7641 | ||
1da177e4 LT |
7642 | void __init sched_init(void) |
7643 | { | |
dd41f596 | 7644 | int i, j; |
434d53b0 MT |
7645 | unsigned long alloc_size = 0, ptr; |
7646 | ||
9881b024 PZ |
7647 | sched_clock_init(); |
7648 | ||
9dcb8b68 LT |
7649 | for (i = 0; i < WAIT_TABLE_SIZE; i++) |
7650 | init_waitqueue_head(bit_wait_table + i); | |
7651 | ||
434d53b0 MT |
7652 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7653 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7654 | #endif | |
7655 | #ifdef CONFIG_RT_GROUP_SCHED | |
7656 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7657 | #endif | |
434d53b0 | 7658 | if (alloc_size) { |
36b7b6d4 | 7659 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7660 | |
7661 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7662 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7663 | ptr += nr_cpu_ids * sizeof(void **); |
7664 | ||
07e06b01 | 7665 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7666 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7667 | |
6d6bc0ad | 7668 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7669 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7670 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7671 | ptr += nr_cpu_ids * sizeof(void **); |
7672 | ||
07e06b01 | 7673 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7674 | ptr += nr_cpu_ids * sizeof(void **); |
7675 | ||
6d6bc0ad | 7676 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7677 | } |
df7c8e84 | 7678 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7679 | for_each_possible_cpu(i) { |
7680 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7681 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7682 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7683 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7684 | } |
b74e6278 | 7685 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7686 | |
332ac17e DF |
7687 | init_rt_bandwidth(&def_rt_bandwidth, |
7688 | global_rt_period(), global_rt_runtime()); | |
7689 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7690 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7691 | |
57d885fe GH |
7692 | #ifdef CONFIG_SMP |
7693 | init_defrootdomain(); | |
7694 | #endif | |
7695 | ||
d0b27fa7 | 7696 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7697 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7698 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7699 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7700 | |
7c941438 | 7701 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7702 | task_group_cache = KMEM_CACHE(task_group, 0); |
7703 | ||
07e06b01 YZ |
7704 | list_add(&root_task_group.list, &task_groups); |
7705 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7706 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7707 | autogroup_init(&init_task); |
7c941438 | 7708 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7709 | |
0a945022 | 7710 | for_each_possible_cpu(i) { |
70b97a7f | 7711 | struct rq *rq; |
1da177e4 LT |
7712 | |
7713 | rq = cpu_rq(i); | |
05fa785c | 7714 | raw_spin_lock_init(&rq->lock); |
7897986b | 7715 | rq->nr_running = 0; |
dce48a84 TG |
7716 | rq->calc_load_active = 0; |
7717 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7718 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7719 | init_rt_rq(&rq->rt); |
7720 | init_dl_rq(&rq->dl); | |
dd41f596 | 7721 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7722 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7723 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7724 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7725 | /* |
07e06b01 | 7726 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7727 | * |
7728 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7729 | * gets 100% of the cpu resources in the system. This overall | |
7730 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7731 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7732 | * based on each entity's (task or task-group's) weight |
7733 | * (se->load.weight). | |
7734 | * | |
07e06b01 | 7735 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7736 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7737 | * then A0's share of the cpu resource is: | |
7738 | * | |
0d905bca | 7739 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7740 | * |
07e06b01 YZ |
7741 | * We achieve this by letting root_task_group's tasks sit |
7742 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7743 | */ |
ab84d31e | 7744 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7745 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7746 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7747 | ||
7748 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7749 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7750 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7751 | #endif |
1da177e4 | 7752 | |
dd41f596 IM |
7753 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7754 | rq->cpu_load[j] = 0; | |
fdf3e95d | 7755 | |
1da177e4 | 7756 | #ifdef CONFIG_SMP |
41c7ce9a | 7757 | rq->sd = NULL; |
57d885fe | 7758 | rq->rd = NULL; |
ca6d75e6 | 7759 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7760 | rq->balance_callback = NULL; |
1da177e4 | 7761 | rq->active_balance = 0; |
dd41f596 | 7762 | rq->next_balance = jiffies; |
1da177e4 | 7763 | rq->push_cpu = 0; |
0a2966b4 | 7764 | rq->cpu = i; |
1f11eb6a | 7765 | rq->online = 0; |
eae0c9df MG |
7766 | rq->idle_stamp = 0; |
7767 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7768 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7769 | |
7770 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7771 | ||
dc938520 | 7772 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7773 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 7774 | rq->last_load_update_tick = jiffies; |
1c792db7 | 7775 | rq->nohz_flags = 0; |
83cd4fe2 | 7776 | #endif |
265f22a9 FW |
7777 | #ifdef CONFIG_NO_HZ_FULL |
7778 | rq->last_sched_tick = 0; | |
7779 | #endif | |
9fd81dd5 | 7780 | #endif /* CONFIG_SMP */ |
8f4d37ec | 7781 | init_rq_hrtick(rq); |
1da177e4 | 7782 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7783 | } |
7784 | ||
2dd73a4f | 7785 | set_load_weight(&init_task); |
b50f60ce | 7786 | |
1da177e4 LT |
7787 | /* |
7788 | * The boot idle thread does lazy MMU switching as well: | |
7789 | */ | |
7790 | atomic_inc(&init_mm.mm_count); | |
7791 | enter_lazy_tlb(&init_mm, current); | |
7792 | ||
7793 | /* | |
7794 | * Make us the idle thread. Technically, schedule() should not be | |
7795 | * called from this thread, however somewhere below it might be, | |
7796 | * but because we are the idle thread, we just pick up running again | |
7797 | * when this runqueue becomes "idle". | |
7798 | */ | |
7799 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7800 | |
7801 | calc_load_update = jiffies + LOAD_FREQ; | |
7802 | ||
bf4d83f6 | 7803 | #ifdef CONFIG_SMP |
4cb98839 | 7804 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7805 | /* May be allocated at isolcpus cmdline parse time */ |
7806 | if (cpu_isolated_map == NULL) | |
7807 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7808 | idle_thread_set_boot_cpu(); |
9cf7243d | 7809 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
7810 | #endif |
7811 | init_sched_fair_class(); | |
6a7b3dc3 | 7812 | |
4698f88c JP |
7813 | init_schedstats(); |
7814 | ||
6892b75e | 7815 | scheduler_running = 1; |
1da177e4 LT |
7816 | } |
7817 | ||
d902db1e | 7818 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7819 | static inline int preempt_count_equals(int preempt_offset) |
7820 | { | |
da7142e2 | 7821 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7822 | |
4ba8216c | 7823 | return (nested == preempt_offset); |
e4aafea2 FW |
7824 | } |
7825 | ||
d894837f | 7826 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7827 | { |
8eb23b9f PZ |
7828 | /* |
7829 | * Blocking primitives will set (and therefore destroy) current->state, | |
7830 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7831 | * otherwise we will destroy state. | |
7832 | */ | |
00845eb9 | 7833 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7834 | "do not call blocking ops when !TASK_RUNNING; " |
7835 | "state=%lx set at [<%p>] %pS\n", | |
7836 | current->state, | |
7837 | (void *)current->task_state_change, | |
00845eb9 | 7838 | (void *)current->task_state_change); |
8eb23b9f | 7839 | |
3427445a PZ |
7840 | ___might_sleep(file, line, preempt_offset); |
7841 | } | |
7842 | EXPORT_SYMBOL(__might_sleep); | |
7843 | ||
7844 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7845 | { |
1da177e4 | 7846 | static unsigned long prev_jiffy; /* ratelimiting */ |
d1c6d149 | 7847 | unsigned long preempt_disable_ip; |
1da177e4 | 7848 | |
b3fbab05 | 7849 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7850 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7851 | !is_idle_task(current)) || | |
e4aafea2 | 7852 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7853 | return; |
7854 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7855 | return; | |
7856 | prev_jiffy = jiffies; | |
7857 | ||
d1c6d149 VN |
7858 | /* Save this before calling printk(), since that will clobber it */ |
7859 | preempt_disable_ip = get_preempt_disable_ip(current); | |
7860 | ||
3df0fc5b PZ |
7861 | printk(KERN_ERR |
7862 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7863 | file, line); | |
7864 | printk(KERN_ERR | |
7865 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7866 | in_atomic(), irqs_disabled(), | |
7867 | current->pid, current->comm); | |
aef745fc | 7868 | |
a8b686b3 ES |
7869 | if (task_stack_end_corrupted(current)) |
7870 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7871 | ||
aef745fc IM |
7872 | debug_show_held_locks(current); |
7873 | if (irqs_disabled()) | |
7874 | print_irqtrace_events(current); | |
d1c6d149 VN |
7875 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7876 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7877 | pr_err("Preemption disabled at:"); |
d1c6d149 | 7878 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
7879 | pr_cont("\n"); |
7880 | } | |
aef745fc | 7881 | dump_stack(); |
f0b22e39 | 7882 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7883 | } |
3427445a | 7884 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7885 | #endif |
7886 | ||
7887 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7888 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7889 | { |
dbc7f069 | 7890 | struct task_struct *g, *p; |
d50dde5a DF |
7891 | struct sched_attr attr = { |
7892 | .sched_policy = SCHED_NORMAL, | |
7893 | }; | |
1da177e4 | 7894 | |
3472eaa1 | 7895 | read_lock(&tasklist_lock); |
5d07f420 | 7896 | for_each_process_thread(g, p) { |
178be793 IM |
7897 | /* |
7898 | * Only normalize user tasks: | |
7899 | */ | |
3472eaa1 | 7900 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7901 | continue; |
7902 | ||
4fa8d299 JP |
7903 | p->se.exec_start = 0; |
7904 | schedstat_set(p->se.statistics.wait_start, 0); | |
7905 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7906 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7907 | |
aab03e05 | 7908 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7909 | /* |
7910 | * Renice negative nice level userspace | |
7911 | * tasks back to 0: | |
7912 | */ | |
3472eaa1 | 7913 | if (task_nice(p) < 0) |
dd41f596 | 7914 | set_user_nice(p, 0); |
1da177e4 | 7915 | continue; |
dd41f596 | 7916 | } |
1da177e4 | 7917 | |
dbc7f069 | 7918 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7919 | } |
3472eaa1 | 7920 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7921 | } |
7922 | ||
7923 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7924 | |
67fc4e0c | 7925 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7926 | /* |
67fc4e0c | 7927 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7928 | * |
7929 | * They can only be called when the whole system has been | |
7930 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7931 | * activity can take place. Using them for anything else would | |
7932 | * be a serious bug, and as a result, they aren't even visible | |
7933 | * under any other configuration. | |
7934 | */ | |
7935 | ||
7936 | /** | |
7937 | * curr_task - return the current task for a given cpu. | |
7938 | * @cpu: the processor in question. | |
7939 | * | |
7940 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7941 | * |
7942 | * Return: The current task for @cpu. | |
1df5c10a | 7943 | */ |
36c8b586 | 7944 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7945 | { |
7946 | return cpu_curr(cpu); | |
7947 | } | |
7948 | ||
67fc4e0c JW |
7949 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7950 | ||
7951 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7952 | /** |
7953 | * set_curr_task - set the current task for a given cpu. | |
7954 | * @cpu: the processor in question. | |
7955 | * @p: the task pointer to set. | |
7956 | * | |
7957 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7958 | * are serviced on a separate stack. It allows the architecture to switch the |
7959 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7960 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7961 | * and caller must save the original value of the current task (see | |
7962 | * curr_task() above) and restore that value before reenabling interrupts and | |
7963 | * re-starting the system. | |
7964 | * | |
7965 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7966 | */ | |
a458ae2e | 7967 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7968 | { |
7969 | cpu_curr(cpu) = p; | |
7970 | } | |
7971 | ||
7972 | #endif | |
29f59db3 | 7973 | |
7c941438 | 7974 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7975 | /* task_group_lock serializes the addition/removal of task groups */ |
7976 | static DEFINE_SPINLOCK(task_group_lock); | |
7977 | ||
2f5177f0 | 7978 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7979 | { |
7980 | free_fair_sched_group(tg); | |
7981 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7982 | autogroup_free(tg); |
b0367629 | 7983 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7984 | } |
7985 | ||
7986 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7987 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7988 | { |
7989 | struct task_group *tg; | |
bccbe08a | 7990 | |
b0367629 | 7991 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7992 | if (!tg) |
7993 | return ERR_PTR(-ENOMEM); | |
7994 | ||
ec7dc8ac | 7995 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7996 | goto err; |
7997 | ||
ec7dc8ac | 7998 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7999 | goto err; |
8000 | ||
ace783b9 LZ |
8001 | return tg; |
8002 | ||
8003 | err: | |
2f5177f0 | 8004 | sched_free_group(tg); |
ace783b9 LZ |
8005 | return ERR_PTR(-ENOMEM); |
8006 | } | |
8007 | ||
8008 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
8009 | { | |
8010 | unsigned long flags; | |
8011 | ||
8ed36996 | 8012 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8013 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8014 | |
8015 | WARN_ON(!parent); /* root should already exist */ | |
8016 | ||
8017 | tg->parent = parent; | |
f473aa5e | 8018 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8019 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8020 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
8021 | |
8022 | online_fair_sched_group(tg); | |
29f59db3 SV |
8023 | } |
8024 | ||
9b5b7751 | 8025 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 8026 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8027 | { |
29f59db3 | 8028 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 8029 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8030 | } |
8031 | ||
4cf86d77 | 8032 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
8033 | { |
8034 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 8035 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
8036 | } |
8037 | ||
8038 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 8039 | { |
8ed36996 | 8040 | unsigned long flags; |
29f59db3 | 8041 | |
3d4b47b4 | 8042 | /* end participation in shares distribution */ |
6fe1f348 | 8043 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
8044 | |
8045 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8046 | list_del_rcu(&tg->list); |
f473aa5e | 8047 | list_del_rcu(&tg->siblings); |
8ed36996 | 8048 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
8049 | } |
8050 | ||
ea86cb4b | 8051 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 8052 | { |
8323f26c | 8053 | struct task_group *tg; |
29f59db3 | 8054 | |
f7b8a47d KT |
8055 | /* |
8056 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
8057 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
8058 | * to prevent lockdep warnings. | |
8059 | */ | |
8060 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
8061 | struct task_group, css); |
8062 | tg = autogroup_task_group(tsk, tg); | |
8063 | tsk->sched_task_group = tg; | |
8064 | ||
810b3817 | 8065 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
8066 | if (tsk->sched_class->task_change_group) |
8067 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 8068 | else |
810b3817 | 8069 | #endif |
b2b5ce02 | 8070 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
8071 | } |
8072 | ||
8073 | /* | |
8074 | * Change task's runqueue when it moves between groups. | |
8075 | * | |
8076 | * The caller of this function should have put the task in its new group by | |
8077 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
8078 | * its new group. | |
8079 | */ | |
8080 | void sched_move_task(struct task_struct *tsk) | |
8081 | { | |
8082 | int queued, running; | |
8083 | struct rq_flags rf; | |
8084 | struct rq *rq; | |
8085 | ||
8086 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 8087 | update_rq_clock(rq); |
ea86cb4b VG |
8088 | |
8089 | running = task_current(rq, tsk); | |
8090 | queued = task_on_rq_queued(tsk); | |
8091 | ||
8092 | if (queued) | |
8093 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); | |
8094 | if (unlikely(running)) | |
8095 | put_prev_task(rq, tsk); | |
8096 | ||
8097 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 8098 | |
da0c1e65 | 8099 | if (queued) |
ff77e468 | 8100 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
a399d233 | 8101 | if (unlikely(running)) |
b2bf6c31 | 8102 | set_curr_task(rq, tsk); |
29f59db3 | 8103 | |
eb580751 | 8104 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8105 | } |
7c941438 | 8106 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8107 | |
a790de99 PT |
8108 | #ifdef CONFIG_RT_GROUP_SCHED |
8109 | /* | |
8110 | * Ensure that the real time constraints are schedulable. | |
8111 | */ | |
8112 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 8113 | |
9a7e0b18 PZ |
8114 | /* Must be called with tasklist_lock held */ |
8115 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8116 | { |
9a7e0b18 | 8117 | struct task_struct *g, *p; |
b40b2e8e | 8118 | |
1fe89e1b PZ |
8119 | /* |
8120 | * Autogroups do not have RT tasks; see autogroup_create(). | |
8121 | */ | |
8122 | if (task_group_is_autogroup(tg)) | |
8123 | return 0; | |
8124 | ||
5d07f420 | 8125 | for_each_process_thread(g, p) { |
8651c658 | 8126 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 8127 | return 1; |
5d07f420 | 8128 | } |
b40b2e8e | 8129 | |
9a7e0b18 PZ |
8130 | return 0; |
8131 | } | |
b40b2e8e | 8132 | |
9a7e0b18 PZ |
8133 | struct rt_schedulable_data { |
8134 | struct task_group *tg; | |
8135 | u64 rt_period; | |
8136 | u64 rt_runtime; | |
8137 | }; | |
b40b2e8e | 8138 | |
a790de99 | 8139 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
8140 | { |
8141 | struct rt_schedulable_data *d = data; | |
8142 | struct task_group *child; | |
8143 | unsigned long total, sum = 0; | |
8144 | u64 period, runtime; | |
b40b2e8e | 8145 | |
9a7e0b18 PZ |
8146 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8147 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8148 | |
9a7e0b18 PZ |
8149 | if (tg == d->tg) { |
8150 | period = d->rt_period; | |
8151 | runtime = d->rt_runtime; | |
b40b2e8e | 8152 | } |
b40b2e8e | 8153 | |
4653f803 PZ |
8154 | /* |
8155 | * Cannot have more runtime than the period. | |
8156 | */ | |
8157 | if (runtime > period && runtime != RUNTIME_INF) | |
8158 | return -EINVAL; | |
6f505b16 | 8159 | |
4653f803 PZ |
8160 | /* |
8161 | * Ensure we don't starve existing RT tasks. | |
8162 | */ | |
9a7e0b18 PZ |
8163 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8164 | return -EBUSY; | |
6f505b16 | 8165 | |
9a7e0b18 | 8166 | total = to_ratio(period, runtime); |
6f505b16 | 8167 | |
4653f803 PZ |
8168 | /* |
8169 | * Nobody can have more than the global setting allows. | |
8170 | */ | |
8171 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8172 | return -EINVAL; | |
6f505b16 | 8173 | |
4653f803 PZ |
8174 | /* |
8175 | * The sum of our children's runtime should not exceed our own. | |
8176 | */ | |
9a7e0b18 PZ |
8177 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8178 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8179 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8180 | |
9a7e0b18 PZ |
8181 | if (child == d->tg) { |
8182 | period = d->rt_period; | |
8183 | runtime = d->rt_runtime; | |
8184 | } | |
6f505b16 | 8185 | |
9a7e0b18 | 8186 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8187 | } |
6f505b16 | 8188 | |
9a7e0b18 PZ |
8189 | if (sum > total) |
8190 | return -EINVAL; | |
8191 | ||
8192 | return 0; | |
6f505b16 PZ |
8193 | } |
8194 | ||
9a7e0b18 | 8195 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8196 | { |
8277434e PT |
8197 | int ret; |
8198 | ||
9a7e0b18 PZ |
8199 | struct rt_schedulable_data data = { |
8200 | .tg = tg, | |
8201 | .rt_period = period, | |
8202 | .rt_runtime = runtime, | |
8203 | }; | |
8204 | ||
8277434e PT |
8205 | rcu_read_lock(); |
8206 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
8207 | rcu_read_unlock(); | |
8208 | ||
8209 | return ret; | |
521f1a24 DG |
8210 | } |
8211 | ||
ab84d31e | 8212 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 8213 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 8214 | { |
ac086bc2 | 8215 | int i, err = 0; |
9f0c1e56 | 8216 | |
2636ed5f PZ |
8217 | /* |
8218 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
8219 | * kernel creating (and or operating) RT threads. | |
8220 | */ | |
8221 | if (tg == &root_task_group && rt_runtime == 0) | |
8222 | return -EINVAL; | |
8223 | ||
8224 | /* No period doesn't make any sense. */ | |
8225 | if (rt_period == 0) | |
8226 | return -EINVAL; | |
8227 | ||
9f0c1e56 | 8228 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8229 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8230 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8231 | if (err) | |
9f0c1e56 | 8232 | goto unlock; |
ac086bc2 | 8233 | |
0986b11b | 8234 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8235 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8236 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8237 | |
8238 | for_each_possible_cpu(i) { | |
8239 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8240 | ||
0986b11b | 8241 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8242 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8243 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8244 | } |
0986b11b | 8245 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8246 | unlock: |
521f1a24 | 8247 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8248 | mutex_unlock(&rt_constraints_mutex); |
8249 | ||
8250 | return err; | |
6f505b16 PZ |
8251 | } |
8252 | ||
25cc7da7 | 8253 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
8254 | { |
8255 | u64 rt_runtime, rt_period; | |
8256 | ||
8257 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8258 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8259 | if (rt_runtime_us < 0) | |
8260 | rt_runtime = RUNTIME_INF; | |
8261 | ||
ab84d31e | 8262 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8263 | } |
8264 | ||
25cc7da7 | 8265 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
8266 | { |
8267 | u64 rt_runtime_us; | |
8268 | ||
d0b27fa7 | 8269 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8270 | return -1; |
8271 | ||
d0b27fa7 | 8272 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8273 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8274 | return rt_runtime_us; | |
8275 | } | |
d0b27fa7 | 8276 | |
ce2f5fe4 | 8277 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
8278 | { |
8279 | u64 rt_runtime, rt_period; | |
8280 | ||
ce2f5fe4 | 8281 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
8282 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
8283 | ||
ab84d31e | 8284 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
8285 | } |
8286 | ||
25cc7da7 | 8287 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
8288 | { |
8289 | u64 rt_period_us; | |
8290 | ||
8291 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8292 | do_div(rt_period_us, NSEC_PER_USEC); | |
8293 | return rt_period_us; | |
8294 | } | |
332ac17e | 8295 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8296 | |
332ac17e | 8297 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
8298 | static int sched_rt_global_constraints(void) |
8299 | { | |
8300 | int ret = 0; | |
8301 | ||
8302 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 8303 | read_lock(&tasklist_lock); |
4653f803 | 8304 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8305 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8306 | mutex_unlock(&rt_constraints_mutex); |
8307 | ||
8308 | return ret; | |
8309 | } | |
54e99124 | 8310 | |
25cc7da7 | 8311 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
8312 | { |
8313 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8314 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8315 | return 0; | |
8316 | ||
8317 | return 1; | |
8318 | } | |
8319 | ||
6d6bc0ad | 8320 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8321 | static int sched_rt_global_constraints(void) |
8322 | { | |
ac086bc2 | 8323 | unsigned long flags; |
8c5e9554 | 8324 | int i; |
ec5d4989 | 8325 | |
0986b11b | 8326 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8327 | for_each_possible_cpu(i) { |
8328 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8329 | ||
0986b11b | 8330 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8331 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8332 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8333 | } |
0986b11b | 8334 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8335 | |
8c5e9554 | 8336 | return 0; |
d0b27fa7 | 8337 | } |
6d6bc0ad | 8338 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8339 | |
a1963b81 | 8340 | static int sched_dl_global_validate(void) |
332ac17e | 8341 | { |
1724813d PZ |
8342 | u64 runtime = global_rt_runtime(); |
8343 | u64 period = global_rt_period(); | |
332ac17e | 8344 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 8345 | struct dl_bw *dl_b; |
1724813d | 8346 | int cpu, ret = 0; |
49516342 | 8347 | unsigned long flags; |
332ac17e DF |
8348 | |
8349 | /* | |
8350 | * Here we want to check the bandwidth not being set to some | |
8351 | * value smaller than the currently allocated bandwidth in | |
8352 | * any of the root_domains. | |
8353 | * | |
8354 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
8355 | * cycling on root_domains... Discussion on different/better | |
8356 | * solutions is welcome! | |
8357 | */ | |
1724813d | 8358 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
8359 | rcu_read_lock_sched(); |
8360 | dl_b = dl_bw_of(cpu); | |
332ac17e | 8361 | |
49516342 | 8362 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
8363 | if (new_bw < dl_b->total_bw) |
8364 | ret = -EBUSY; | |
49516342 | 8365 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 8366 | |
f10e00f4 KT |
8367 | rcu_read_unlock_sched(); |
8368 | ||
1724813d PZ |
8369 | if (ret) |
8370 | break; | |
332ac17e DF |
8371 | } |
8372 | ||
1724813d | 8373 | return ret; |
332ac17e DF |
8374 | } |
8375 | ||
1724813d | 8376 | static void sched_dl_do_global(void) |
ce0dbbbb | 8377 | { |
1724813d | 8378 | u64 new_bw = -1; |
f10e00f4 | 8379 | struct dl_bw *dl_b; |
1724813d | 8380 | int cpu; |
49516342 | 8381 | unsigned long flags; |
ce0dbbbb | 8382 | |
1724813d PZ |
8383 | def_dl_bandwidth.dl_period = global_rt_period(); |
8384 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
8385 | ||
8386 | if (global_rt_runtime() != RUNTIME_INF) | |
8387 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
8388 | ||
8389 | /* | |
8390 | * FIXME: As above... | |
8391 | */ | |
8392 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
8393 | rcu_read_lock_sched(); |
8394 | dl_b = dl_bw_of(cpu); | |
1724813d | 8395 | |
49516342 | 8396 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8397 | dl_b->bw = new_bw; |
49516342 | 8398 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8399 | |
8400 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8401 | } |
1724813d PZ |
8402 | } |
8403 | ||
8404 | static int sched_rt_global_validate(void) | |
8405 | { | |
8406 | if (sysctl_sched_rt_period <= 0) | |
8407 | return -EINVAL; | |
8408 | ||
e9e7cb38 JL |
8409 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8410 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8411 | return -EINVAL; |
8412 | ||
8413 | return 0; | |
8414 | } | |
8415 | ||
8416 | static void sched_rt_do_global(void) | |
8417 | { | |
8418 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8419 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8420 | } |
8421 | ||
d0b27fa7 | 8422 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8423 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8424 | loff_t *ppos) |
8425 | { | |
d0b27fa7 PZ |
8426 | int old_period, old_runtime; |
8427 | static DEFINE_MUTEX(mutex); | |
1724813d | 8428 | int ret; |
d0b27fa7 PZ |
8429 | |
8430 | mutex_lock(&mutex); | |
8431 | old_period = sysctl_sched_rt_period; | |
8432 | old_runtime = sysctl_sched_rt_runtime; | |
8433 | ||
8d65af78 | 8434 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8435 | |
8436 | if (!ret && write) { | |
1724813d PZ |
8437 | ret = sched_rt_global_validate(); |
8438 | if (ret) | |
8439 | goto undo; | |
8440 | ||
a1963b81 | 8441 | ret = sched_dl_global_validate(); |
1724813d PZ |
8442 | if (ret) |
8443 | goto undo; | |
8444 | ||
a1963b81 | 8445 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8446 | if (ret) |
8447 | goto undo; | |
8448 | ||
8449 | sched_rt_do_global(); | |
8450 | sched_dl_do_global(); | |
8451 | } | |
8452 | if (0) { | |
8453 | undo: | |
8454 | sysctl_sched_rt_period = old_period; | |
8455 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8456 | } |
8457 | mutex_unlock(&mutex); | |
8458 | ||
8459 | return ret; | |
8460 | } | |
68318b8e | 8461 | |
1724813d | 8462 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8463 | void __user *buffer, size_t *lenp, |
8464 | loff_t *ppos) | |
8465 | { | |
8466 | int ret; | |
332ac17e | 8467 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8468 | |
8469 | mutex_lock(&mutex); | |
332ac17e | 8470 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8471 | /* make sure that internally we keep jiffies */ |
8472 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8473 | if (!ret && write) { |
975e155e SZ |
8474 | sched_rr_timeslice = |
8475 | sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE : | |
8476 | msecs_to_jiffies(sysctl_sched_rr_timeslice); | |
332ac17e DF |
8477 | } |
8478 | mutex_unlock(&mutex); | |
332ac17e DF |
8479 | return ret; |
8480 | } | |
8481 | ||
052f1dc7 | 8482 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8483 | |
a7c6d554 | 8484 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8485 | { |
a7c6d554 | 8486 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8487 | } |
8488 | ||
eb95419b TH |
8489 | static struct cgroup_subsys_state * |
8490 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8491 | { |
eb95419b TH |
8492 | struct task_group *parent = css_tg(parent_css); |
8493 | struct task_group *tg; | |
68318b8e | 8494 | |
eb95419b | 8495 | if (!parent) { |
68318b8e | 8496 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8497 | return &root_task_group.css; |
68318b8e SV |
8498 | } |
8499 | ||
ec7dc8ac | 8500 | tg = sched_create_group(parent); |
68318b8e SV |
8501 | if (IS_ERR(tg)) |
8502 | return ERR_PTR(-ENOMEM); | |
8503 | ||
2f5177f0 PZ |
8504 | sched_online_group(tg, parent); |
8505 | ||
68318b8e SV |
8506 | return &tg->css; |
8507 | } | |
8508 | ||
2f5177f0 | 8509 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8510 | { |
eb95419b | 8511 | struct task_group *tg = css_tg(css); |
ace783b9 | 8512 | |
2f5177f0 | 8513 | sched_offline_group(tg); |
ace783b9 LZ |
8514 | } |
8515 | ||
eb95419b | 8516 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8517 | { |
eb95419b | 8518 | struct task_group *tg = css_tg(css); |
68318b8e | 8519 | |
2f5177f0 PZ |
8520 | /* |
8521 | * Relies on the RCU grace period between css_released() and this. | |
8522 | */ | |
8523 | sched_free_group(tg); | |
ace783b9 LZ |
8524 | } |
8525 | ||
ea86cb4b VG |
8526 | /* |
8527 | * This is called before wake_up_new_task(), therefore we really only | |
8528 | * have to set its group bits, all the other stuff does not apply. | |
8529 | */ | |
b53202e6 | 8530 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8531 | { |
ea86cb4b VG |
8532 | struct rq_flags rf; |
8533 | struct rq *rq; | |
8534 | ||
8535 | rq = task_rq_lock(task, &rf); | |
8536 | ||
80f5c1b8 | 8537 | update_rq_clock(rq); |
ea86cb4b VG |
8538 | sched_change_group(task, TASK_SET_GROUP); |
8539 | ||
8540 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8541 | } |
8542 | ||
1f7dd3e5 | 8543 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8544 | { |
bb9d97b6 | 8545 | struct task_struct *task; |
1f7dd3e5 | 8546 | struct cgroup_subsys_state *css; |
7dc603c9 | 8547 | int ret = 0; |
bb9d97b6 | 8548 | |
1f7dd3e5 | 8549 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8550 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8551 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8552 | return -EINVAL; |
b68aa230 | 8553 | #else |
bb9d97b6 TH |
8554 | /* We don't support RT-tasks being in separate groups */ |
8555 | if (task->sched_class != &fair_sched_class) | |
8556 | return -EINVAL; | |
b68aa230 | 8557 | #endif |
7dc603c9 PZ |
8558 | /* |
8559 | * Serialize against wake_up_new_task() such that if its | |
8560 | * running, we're sure to observe its full state. | |
8561 | */ | |
8562 | raw_spin_lock_irq(&task->pi_lock); | |
8563 | /* | |
8564 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8565 | * has happened. This would lead to problems with PELT, due to | |
8566 | * move wanting to detach+attach while we're not attached yet. | |
8567 | */ | |
8568 | if (task->state == TASK_NEW) | |
8569 | ret = -EINVAL; | |
8570 | raw_spin_unlock_irq(&task->pi_lock); | |
8571 | ||
8572 | if (ret) | |
8573 | break; | |
bb9d97b6 | 8574 | } |
7dc603c9 | 8575 | return ret; |
be367d09 | 8576 | } |
68318b8e | 8577 | |
1f7dd3e5 | 8578 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8579 | { |
bb9d97b6 | 8580 | struct task_struct *task; |
1f7dd3e5 | 8581 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8582 | |
1f7dd3e5 | 8583 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8584 | sched_move_task(task); |
68318b8e SV |
8585 | } |
8586 | ||
052f1dc7 | 8587 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8588 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8589 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8590 | { |
182446d0 | 8591 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8592 | } |
8593 | ||
182446d0 TH |
8594 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8595 | struct cftype *cft) | |
68318b8e | 8596 | { |
182446d0 | 8597 | struct task_group *tg = css_tg(css); |
68318b8e | 8598 | |
c8b28116 | 8599 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8600 | } |
ab84d31e PT |
8601 | |
8602 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8603 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8604 | ||
ab84d31e PT |
8605 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8606 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8607 | ||
a790de99 PT |
8608 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8609 | ||
ab84d31e PT |
8610 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8611 | { | |
56f570e5 | 8612 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8613 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8614 | |
8615 | if (tg == &root_task_group) | |
8616 | return -EINVAL; | |
8617 | ||
8618 | /* | |
8619 | * Ensure we have at some amount of bandwidth every period. This is | |
8620 | * to prevent reaching a state of large arrears when throttled via | |
8621 | * entity_tick() resulting in prolonged exit starvation. | |
8622 | */ | |
8623 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8624 | return -EINVAL; | |
8625 | ||
8626 | /* | |
8627 | * Likewise, bound things on the otherside by preventing insane quota | |
8628 | * periods. This also allows us to normalize in computing quota | |
8629 | * feasibility. | |
8630 | */ | |
8631 | if (period > max_cfs_quota_period) | |
8632 | return -EINVAL; | |
8633 | ||
0e59bdae KT |
8634 | /* |
8635 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8636 | * unthrottle_offline_cfs_rqs(). | |
8637 | */ | |
8638 | get_online_cpus(); | |
a790de99 PT |
8639 | mutex_lock(&cfs_constraints_mutex); |
8640 | ret = __cfs_schedulable(tg, period, quota); | |
8641 | if (ret) | |
8642 | goto out_unlock; | |
8643 | ||
58088ad0 | 8644 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8645 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8646 | /* |
8647 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8648 | * before making related changes, and on->off must occur afterwards | |
8649 | */ | |
8650 | if (runtime_enabled && !runtime_was_enabled) | |
8651 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8652 | raw_spin_lock_irq(&cfs_b->lock); |
8653 | cfs_b->period = ns_to_ktime(period); | |
8654 | cfs_b->quota = quota; | |
58088ad0 | 8655 | |
a9cf55b2 | 8656 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8657 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8658 | if (runtime_enabled) |
8659 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8660 | raw_spin_unlock_irq(&cfs_b->lock); |
8661 | ||
0e59bdae | 8662 | for_each_online_cpu(i) { |
ab84d31e | 8663 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8664 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8665 | |
8666 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8667 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8668 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8669 | |
029632fb | 8670 | if (cfs_rq->throttled) |
671fd9da | 8671 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8672 | raw_spin_unlock_irq(&rq->lock); |
8673 | } | |
1ee14e6c BS |
8674 | if (runtime_was_enabled && !runtime_enabled) |
8675 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8676 | out_unlock: |
8677 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8678 | put_online_cpus(); |
ab84d31e | 8679 | |
a790de99 | 8680 | return ret; |
ab84d31e PT |
8681 | } |
8682 | ||
8683 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8684 | { | |
8685 | u64 quota, period; | |
8686 | ||
029632fb | 8687 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8688 | if (cfs_quota_us < 0) |
8689 | quota = RUNTIME_INF; | |
8690 | else | |
8691 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8692 | ||
8693 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8694 | } | |
8695 | ||
8696 | long tg_get_cfs_quota(struct task_group *tg) | |
8697 | { | |
8698 | u64 quota_us; | |
8699 | ||
029632fb | 8700 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8701 | return -1; |
8702 | ||
029632fb | 8703 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8704 | do_div(quota_us, NSEC_PER_USEC); |
8705 | ||
8706 | return quota_us; | |
8707 | } | |
8708 | ||
8709 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8710 | { | |
8711 | u64 quota, period; | |
8712 | ||
8713 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8714 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8715 | |
ab84d31e PT |
8716 | return tg_set_cfs_bandwidth(tg, period, quota); |
8717 | } | |
8718 | ||
8719 | long tg_get_cfs_period(struct task_group *tg) | |
8720 | { | |
8721 | u64 cfs_period_us; | |
8722 | ||
029632fb | 8723 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8724 | do_div(cfs_period_us, NSEC_PER_USEC); |
8725 | ||
8726 | return cfs_period_us; | |
8727 | } | |
8728 | ||
182446d0 TH |
8729 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8730 | struct cftype *cft) | |
ab84d31e | 8731 | { |
182446d0 | 8732 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8733 | } |
8734 | ||
182446d0 TH |
8735 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8736 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8737 | { |
182446d0 | 8738 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8739 | } |
8740 | ||
182446d0 TH |
8741 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8742 | struct cftype *cft) | |
ab84d31e | 8743 | { |
182446d0 | 8744 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8745 | } |
8746 | ||
182446d0 TH |
8747 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8748 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8749 | { |
182446d0 | 8750 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8751 | } |
8752 | ||
a790de99 PT |
8753 | struct cfs_schedulable_data { |
8754 | struct task_group *tg; | |
8755 | u64 period, quota; | |
8756 | }; | |
8757 | ||
8758 | /* | |
8759 | * normalize group quota/period to be quota/max_period | |
8760 | * note: units are usecs | |
8761 | */ | |
8762 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8763 | struct cfs_schedulable_data *d) | |
8764 | { | |
8765 | u64 quota, period; | |
8766 | ||
8767 | if (tg == d->tg) { | |
8768 | period = d->period; | |
8769 | quota = d->quota; | |
8770 | } else { | |
8771 | period = tg_get_cfs_period(tg); | |
8772 | quota = tg_get_cfs_quota(tg); | |
8773 | } | |
8774 | ||
8775 | /* note: these should typically be equivalent */ | |
8776 | if (quota == RUNTIME_INF || quota == -1) | |
8777 | return RUNTIME_INF; | |
8778 | ||
8779 | return to_ratio(period, quota); | |
8780 | } | |
8781 | ||
8782 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8783 | { | |
8784 | struct cfs_schedulable_data *d = data; | |
029632fb | 8785 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8786 | s64 quota = 0, parent_quota = -1; |
8787 | ||
8788 | if (!tg->parent) { | |
8789 | quota = RUNTIME_INF; | |
8790 | } else { | |
029632fb | 8791 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8792 | |
8793 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8794 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8795 | |
8796 | /* | |
8797 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8798 | * limit is set | |
8799 | */ | |
8800 | if (quota == RUNTIME_INF) | |
8801 | quota = parent_quota; | |
8802 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8803 | return -EINVAL; | |
8804 | } | |
9c58c79a | 8805 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8806 | |
8807 | return 0; | |
8808 | } | |
8809 | ||
8810 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8811 | { | |
8277434e | 8812 | int ret; |
a790de99 PT |
8813 | struct cfs_schedulable_data data = { |
8814 | .tg = tg, | |
8815 | .period = period, | |
8816 | .quota = quota, | |
8817 | }; | |
8818 | ||
8819 | if (quota != RUNTIME_INF) { | |
8820 | do_div(data.period, NSEC_PER_USEC); | |
8821 | do_div(data.quota, NSEC_PER_USEC); | |
8822 | } | |
8823 | ||
8277434e PT |
8824 | rcu_read_lock(); |
8825 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8826 | rcu_read_unlock(); | |
8827 | ||
8828 | return ret; | |
a790de99 | 8829 | } |
e8da1b18 | 8830 | |
2da8ca82 | 8831 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8832 | { |
2da8ca82 | 8833 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8834 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8835 | |
44ffc75b TH |
8836 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8837 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8838 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8839 | |
8840 | return 0; | |
8841 | } | |
ab84d31e | 8842 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8843 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8844 | |
052f1dc7 | 8845 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8846 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8847 | struct cftype *cft, s64 val) | |
6f505b16 | 8848 | { |
182446d0 | 8849 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8850 | } |
8851 | ||
182446d0 TH |
8852 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8853 | struct cftype *cft) | |
6f505b16 | 8854 | { |
182446d0 | 8855 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8856 | } |
d0b27fa7 | 8857 | |
182446d0 TH |
8858 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8859 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8860 | { |
182446d0 | 8861 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8862 | } |
8863 | ||
182446d0 TH |
8864 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8865 | struct cftype *cft) | |
d0b27fa7 | 8866 | { |
182446d0 | 8867 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8868 | } |
6d6bc0ad | 8869 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8870 | |
fe5c7cc2 | 8871 | static struct cftype cpu_files[] = { |
052f1dc7 | 8872 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8873 | { |
8874 | .name = "shares", | |
f4c753b7 PM |
8875 | .read_u64 = cpu_shares_read_u64, |
8876 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8877 | }, |
052f1dc7 | 8878 | #endif |
ab84d31e PT |
8879 | #ifdef CONFIG_CFS_BANDWIDTH |
8880 | { | |
8881 | .name = "cfs_quota_us", | |
8882 | .read_s64 = cpu_cfs_quota_read_s64, | |
8883 | .write_s64 = cpu_cfs_quota_write_s64, | |
8884 | }, | |
8885 | { | |
8886 | .name = "cfs_period_us", | |
8887 | .read_u64 = cpu_cfs_period_read_u64, | |
8888 | .write_u64 = cpu_cfs_period_write_u64, | |
8889 | }, | |
e8da1b18 NR |
8890 | { |
8891 | .name = "stat", | |
2da8ca82 | 8892 | .seq_show = cpu_stats_show, |
e8da1b18 | 8893 | }, |
ab84d31e | 8894 | #endif |
052f1dc7 | 8895 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8896 | { |
9f0c1e56 | 8897 | .name = "rt_runtime_us", |
06ecb27c PM |
8898 | .read_s64 = cpu_rt_runtime_read, |
8899 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8900 | }, |
d0b27fa7 PZ |
8901 | { |
8902 | .name = "rt_period_us", | |
f4c753b7 PM |
8903 | .read_u64 = cpu_rt_period_read_uint, |
8904 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8905 | }, |
052f1dc7 | 8906 | #endif |
4baf6e33 | 8907 | { } /* terminate */ |
68318b8e SV |
8908 | }; |
8909 | ||
073219e9 | 8910 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8911 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8912 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8913 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8914 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8915 | .can_attach = cpu_cgroup_can_attach, |
8916 | .attach = cpu_cgroup_attach, | |
5577964e | 8917 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8918 | .early_init = true, |
68318b8e SV |
8919 | }; |
8920 | ||
052f1dc7 | 8921 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8922 | |
b637a328 PM |
8923 | void dump_cpu_task(int cpu) |
8924 | { | |
8925 | pr_info("Task dump for CPU %d:\n", cpu); | |
8926 | sched_show_task(cpu_curr(cpu)); | |
8927 | } | |
ed82b8a1 AK |
8928 | |
8929 | /* | |
8930 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8931 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8932 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8933 | * that remained on nice 0. | |
8934 | * | |
8935 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8936 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8937 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8938 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8939 | * the relative distance between them is ~25%.) | |
8940 | */ | |
8941 | const int sched_prio_to_weight[40] = { | |
8942 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8943 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8944 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8945 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8946 | /* 0 */ 1024, 820, 655, 526, 423, | |
8947 | /* 5 */ 335, 272, 215, 172, 137, | |
8948 | /* 10 */ 110, 87, 70, 56, 45, | |
8949 | /* 15 */ 36, 29, 23, 18, 15, | |
8950 | }; | |
8951 | ||
8952 | /* | |
8953 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8954 | * | |
8955 | * In cases where the weight does not change often, we can use the | |
8956 | * precalculated inverse to speed up arithmetics by turning divisions | |
8957 | * into multiplications: | |
8958 | */ | |
8959 | const u32 sched_prio_to_wmult[40] = { | |
8960 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8961 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8962 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8963 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8964 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8965 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8966 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8967 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8968 | }; |