]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 | 3 | * |
d1ccc66d | 4 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 7 | */ |
004172bd | 8 | #include <linux/sched.h> |
e6017571 | 9 | #include <linux/sched/clock.h> |
ae7e81c0 | 10 | #include <uapi/linux/sched/types.h> |
4f17722c | 11 | #include <linux/sched/loadavg.h> |
ef8bd77f | 12 | #include <linux/sched/hotplug.h> |
1da177e4 | 13 | #include <linux/cpuset.h> |
0ff92245 | 14 | #include <linux/delayacct.h> |
f1c6f1a7 | 15 | #include <linux/init_task.h> |
91d1aa43 | 16 | #include <linux/context_tracking.h> |
f9411ebe | 17 | #include <linux/rcupdate_wait.h> |
004172bd IM |
18 | |
19 | #include <linux/blkdev.h> | |
20 | #include <linux/kprobes.h> | |
21 | #include <linux/mmu_context.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/nmi.h> | |
6075620b | 24 | #include <linux/prefetch.h> |
004172bd IM |
25 | #include <linux/profile.h> |
26 | #include <linux/security.h> | |
27 | #include <linux/syscalls.h> | |
1da177e4 | 28 | |
96f951ed | 29 | #include <asm/switch_to.h> |
5517d86b | 30 | #include <asm/tlb.h> |
fd4a61e0 MB |
31 | #ifdef CONFIG_PARAVIRT |
32 | #include <asm/paravirt.h> | |
33 | #endif | |
1da177e4 | 34 | |
029632fb | 35 | #include "sched.h" |
ea138446 | 36 | #include "../workqueue_internal.h" |
29d5e047 | 37 | #include "../smpboot.h" |
6e0534f2 | 38 | |
a8d154b0 | 39 | #define CREATE_TRACE_POINTS |
ad8d75ff | 40 | #include <trace/events/sched.h> |
a8d154b0 | 41 | |
029632fb | 42 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 43 | |
bf5c91ba IM |
44 | /* |
45 | * Debugging: various feature bits | |
46 | */ | |
f00b45c1 | 47 | |
f00b45c1 PZ |
48 | #define SCHED_FEAT(name, enabled) \ |
49 | (1UL << __SCHED_FEAT_##name) * enabled | | |
50 | ||
bf5c91ba | 51 | const_debug unsigned int sysctl_sched_features = |
391e43da | 52 | #include "features.h" |
f00b45c1 PZ |
53 | 0; |
54 | ||
55 | #undef SCHED_FEAT | |
56 | ||
b82d9fdd PZ |
57 | /* |
58 | * Number of tasks to iterate in a single balance run. | |
59 | * Limited because this is done with IRQs disabled. | |
60 | */ | |
61 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
62 | ||
e9e9250b PZ |
63 | /* |
64 | * period over which we average the RT time consumption, measured | |
65 | * in ms. | |
66 | * | |
67 | * default: 1s | |
68 | */ | |
69 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
70 | ||
fa85ae24 | 71 | /* |
d1ccc66d | 72 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
73 | * default: 1s |
74 | */ | |
9f0c1e56 | 75 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 76 | |
029632fb | 77 | __read_mostly int scheduler_running; |
6892b75e | 78 | |
9f0c1e56 PZ |
79 | /* |
80 | * part of the period that we allow rt tasks to run in us. | |
81 | * default: 0.95s | |
82 | */ | |
83 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 84 | |
d1ccc66d | 85 | /* CPUs with isolated domains */ |
3fa0818b RR |
86 | cpumask_var_t cpu_isolated_map; |
87 | ||
3e71a462 PZ |
88 | /* |
89 | * __task_rq_lock - lock the rq @p resides on. | |
90 | */ | |
eb580751 | 91 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
92 | __acquires(rq->lock) |
93 | { | |
94 | struct rq *rq; | |
95 | ||
96 | lockdep_assert_held(&p->pi_lock); | |
97 | ||
98 | for (;;) { | |
99 | rq = task_rq(p); | |
100 | raw_spin_lock(&rq->lock); | |
101 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 102 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
103 | return rq; |
104 | } | |
105 | raw_spin_unlock(&rq->lock); | |
106 | ||
107 | while (unlikely(task_on_rq_migrating(p))) | |
108 | cpu_relax(); | |
109 | } | |
110 | } | |
111 | ||
112 | /* | |
113 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
114 | */ | |
eb580751 | 115 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
116 | __acquires(p->pi_lock) |
117 | __acquires(rq->lock) | |
118 | { | |
119 | struct rq *rq; | |
120 | ||
121 | for (;;) { | |
eb580751 | 122 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
123 | rq = task_rq(p); |
124 | raw_spin_lock(&rq->lock); | |
125 | /* | |
126 | * move_queued_task() task_rq_lock() | |
127 | * | |
128 | * ACQUIRE (rq->lock) | |
129 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
130 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
131 | * [S] ->cpu = new_cpu [L] task_rq() | |
132 | * [L] ->on_rq | |
133 | * RELEASE (rq->lock) | |
134 | * | |
135 | * If we observe the old cpu in task_rq_lock, the acquire of | |
136 | * the old rq->lock will fully serialize against the stores. | |
137 | * | |
d1ccc66d | 138 | * If we observe the new CPU in task_rq_lock, the acquire will |
3e71a462 PZ |
139 | * pair with the WMB to ensure we must then also see migrating. |
140 | */ | |
141 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 142 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
143 | return rq; |
144 | } | |
145 | raw_spin_unlock(&rq->lock); | |
eb580751 | 146 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
147 | |
148 | while (unlikely(task_on_rq_migrating(p))) | |
149 | cpu_relax(); | |
150 | } | |
151 | } | |
152 | ||
535b9552 IM |
153 | /* |
154 | * RQ-clock updating methods: | |
155 | */ | |
156 | ||
157 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
158 | { | |
159 | /* | |
160 | * In theory, the compile should just see 0 here, and optimize out the call | |
161 | * to sched_rt_avg_update. But I don't trust it... | |
162 | */ | |
163 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
164 | s64 steal = 0, irq_delta = 0; | |
165 | #endif | |
166 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
167 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
168 | ||
169 | /* | |
170 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
171 | * this case when a previous update_rq_clock() happened inside a | |
172 | * {soft,}irq region. | |
173 | * | |
174 | * When this happens, we stop ->clock_task and only update the | |
175 | * prev_irq_time stamp to account for the part that fit, so that a next | |
176 | * update will consume the rest. This ensures ->clock_task is | |
177 | * monotonic. | |
178 | * | |
179 | * It does however cause some slight miss-attribution of {soft,}irq | |
180 | * time, a more accurate solution would be to update the irq_time using | |
181 | * the current rq->clock timestamp, except that would require using | |
182 | * atomic ops. | |
183 | */ | |
184 | if (irq_delta > delta) | |
185 | irq_delta = delta; | |
186 | ||
187 | rq->prev_irq_time += irq_delta; | |
188 | delta -= irq_delta; | |
189 | #endif | |
190 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
191 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
192 | steal = paravirt_steal_clock(cpu_of(rq)); | |
193 | steal -= rq->prev_steal_time_rq; | |
194 | ||
195 | if (unlikely(steal > delta)) | |
196 | steal = delta; | |
197 | ||
198 | rq->prev_steal_time_rq += steal; | |
199 | delta -= steal; | |
200 | } | |
201 | #endif | |
202 | ||
203 | rq->clock_task += delta; | |
204 | ||
205 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
206 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) | |
207 | sched_rt_avg_update(rq, irq_delta + steal); | |
208 | #endif | |
209 | } | |
210 | ||
211 | void update_rq_clock(struct rq *rq) | |
212 | { | |
213 | s64 delta; | |
214 | ||
215 | lockdep_assert_held(&rq->lock); | |
216 | ||
217 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
218 | return; | |
219 | ||
220 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
221 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
222 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
223 | rq->clock_update_flags |= RQCF_UPDATED; |
224 | #endif | |
26ae58d2 | 225 | |
535b9552 IM |
226 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
227 | if (delta < 0) | |
228 | return; | |
229 | rq->clock += delta; | |
230 | update_rq_clock_task(rq, delta); | |
231 | } | |
232 | ||
233 | ||
8f4d37ec PZ |
234 | #ifdef CONFIG_SCHED_HRTICK |
235 | /* | |
236 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 237 | */ |
8f4d37ec | 238 | |
8f4d37ec PZ |
239 | static void hrtick_clear(struct rq *rq) |
240 | { | |
241 | if (hrtimer_active(&rq->hrtick_timer)) | |
242 | hrtimer_cancel(&rq->hrtick_timer); | |
243 | } | |
244 | ||
8f4d37ec PZ |
245 | /* |
246 | * High-resolution timer tick. | |
247 | * Runs from hardirq context with interrupts disabled. | |
248 | */ | |
249 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
250 | { | |
251 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 252 | struct rq_flags rf; |
8f4d37ec PZ |
253 | |
254 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
255 | ||
8a8c69c3 | 256 | rq_lock(rq, &rf); |
3e51f33f | 257 | update_rq_clock(rq); |
8f4d37ec | 258 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 259 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
260 | |
261 | return HRTIMER_NORESTART; | |
262 | } | |
263 | ||
95e904c7 | 264 | #ifdef CONFIG_SMP |
971ee28c | 265 | |
4961b6e1 | 266 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
267 | { |
268 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 269 | |
4961b6e1 | 270 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
271 | } |
272 | ||
31656519 PZ |
273 | /* |
274 | * called from hardirq (IPI) context | |
275 | */ | |
276 | static void __hrtick_start(void *arg) | |
b328ca18 | 277 | { |
31656519 | 278 | struct rq *rq = arg; |
8a8c69c3 | 279 | struct rq_flags rf; |
b328ca18 | 280 | |
8a8c69c3 | 281 | rq_lock(rq, &rf); |
971ee28c | 282 | __hrtick_restart(rq); |
31656519 | 283 | rq->hrtick_csd_pending = 0; |
8a8c69c3 | 284 | rq_unlock(rq, &rf); |
b328ca18 PZ |
285 | } |
286 | ||
31656519 PZ |
287 | /* |
288 | * Called to set the hrtick timer state. | |
289 | * | |
290 | * called with rq->lock held and irqs disabled | |
291 | */ | |
029632fb | 292 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 293 | { |
31656519 | 294 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 295 | ktime_t time; |
296 | s64 delta; | |
297 | ||
298 | /* | |
299 | * Don't schedule slices shorter than 10000ns, that just | |
300 | * doesn't make sense and can cause timer DoS. | |
301 | */ | |
302 | delta = max_t(s64, delay, 10000LL); | |
303 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 304 | |
cc584b21 | 305 | hrtimer_set_expires(timer, time); |
31656519 PZ |
306 | |
307 | if (rq == this_rq()) { | |
971ee28c | 308 | __hrtick_restart(rq); |
31656519 | 309 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 310 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
311 | rq->hrtick_csd_pending = 1; |
312 | } | |
b328ca18 PZ |
313 | } |
314 | ||
31656519 PZ |
315 | #else |
316 | /* | |
317 | * Called to set the hrtick timer state. | |
318 | * | |
319 | * called with rq->lock held and irqs disabled | |
320 | */ | |
029632fb | 321 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 322 | { |
86893335 WL |
323 | /* |
324 | * Don't schedule slices shorter than 10000ns, that just | |
325 | * doesn't make sense. Rely on vruntime for fairness. | |
326 | */ | |
327 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
328 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
329 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 330 | } |
31656519 | 331 | #endif /* CONFIG_SMP */ |
8f4d37ec | 332 | |
31656519 | 333 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 334 | { |
31656519 PZ |
335 | #ifdef CONFIG_SMP |
336 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 337 | |
31656519 PZ |
338 | rq->hrtick_csd.flags = 0; |
339 | rq->hrtick_csd.func = __hrtick_start; | |
340 | rq->hrtick_csd.info = rq; | |
341 | #endif | |
8f4d37ec | 342 | |
31656519 PZ |
343 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
344 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 345 | } |
006c75f1 | 346 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
347 | static inline void hrtick_clear(struct rq *rq) |
348 | { | |
349 | } | |
350 | ||
8f4d37ec PZ |
351 | static inline void init_rq_hrtick(struct rq *rq) |
352 | { | |
353 | } | |
006c75f1 | 354 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 355 | |
5529578a FW |
356 | /* |
357 | * cmpxchg based fetch_or, macro so it works for different integer types | |
358 | */ | |
359 | #define fetch_or(ptr, mask) \ | |
360 | ({ \ | |
361 | typeof(ptr) _ptr = (ptr); \ | |
362 | typeof(mask) _mask = (mask); \ | |
363 | typeof(*_ptr) _old, _val = *_ptr; \ | |
364 | \ | |
365 | for (;;) { \ | |
366 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
367 | if (_old == _val) \ | |
368 | break; \ | |
369 | _val = _old; \ | |
370 | } \ | |
371 | _old; \ | |
372 | }) | |
373 | ||
e3baac47 | 374 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
375 | /* |
376 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
377 | * this avoids any races wrt polling state changes and thereby avoids | |
378 | * spurious IPIs. | |
379 | */ | |
380 | static bool set_nr_and_not_polling(struct task_struct *p) | |
381 | { | |
382 | struct thread_info *ti = task_thread_info(p); | |
383 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
384 | } | |
e3baac47 PZ |
385 | |
386 | /* | |
387 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
388 | * | |
389 | * If this returns true, then the idle task promises to call | |
390 | * sched_ttwu_pending() and reschedule soon. | |
391 | */ | |
392 | static bool set_nr_if_polling(struct task_struct *p) | |
393 | { | |
394 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 395 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
396 | |
397 | for (;;) { | |
398 | if (!(val & _TIF_POLLING_NRFLAG)) | |
399 | return false; | |
400 | if (val & _TIF_NEED_RESCHED) | |
401 | return true; | |
402 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
403 | if (old == val) | |
404 | break; | |
405 | val = old; | |
406 | } | |
407 | return true; | |
408 | } | |
409 | ||
fd99f91a PZ |
410 | #else |
411 | static bool set_nr_and_not_polling(struct task_struct *p) | |
412 | { | |
413 | set_tsk_need_resched(p); | |
414 | return true; | |
415 | } | |
e3baac47 PZ |
416 | |
417 | #ifdef CONFIG_SMP | |
418 | static bool set_nr_if_polling(struct task_struct *p) | |
419 | { | |
420 | return false; | |
421 | } | |
422 | #endif | |
fd99f91a PZ |
423 | #endif |
424 | ||
76751049 PZ |
425 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
426 | { | |
427 | struct wake_q_node *node = &task->wake_q; | |
428 | ||
429 | /* | |
430 | * Atomically grab the task, if ->wake_q is !nil already it means | |
431 | * its already queued (either by us or someone else) and will get the | |
432 | * wakeup due to that. | |
433 | * | |
434 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 435 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
436 | */ |
437 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
438 | return; | |
439 | ||
440 | get_task_struct(task); | |
441 | ||
442 | /* | |
443 | * The head is context local, there can be no concurrency. | |
444 | */ | |
445 | *head->lastp = node; | |
446 | head->lastp = &node->next; | |
447 | } | |
448 | ||
449 | void wake_up_q(struct wake_q_head *head) | |
450 | { | |
451 | struct wake_q_node *node = head->first; | |
452 | ||
453 | while (node != WAKE_Q_TAIL) { | |
454 | struct task_struct *task; | |
455 | ||
456 | task = container_of(node, struct task_struct, wake_q); | |
457 | BUG_ON(!task); | |
d1ccc66d | 458 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
459 | node = node->next; |
460 | task->wake_q.next = NULL; | |
461 | ||
462 | /* | |
463 | * wake_up_process() implies a wmb() to pair with the queueing | |
464 | * in wake_q_add() so as not to miss wakeups. | |
465 | */ | |
466 | wake_up_process(task); | |
467 | put_task_struct(task); | |
468 | } | |
469 | } | |
470 | ||
c24d20db | 471 | /* |
8875125e | 472 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
473 | * |
474 | * On UP this means the setting of the need_resched flag, on SMP it | |
475 | * might also involve a cross-CPU call to trigger the scheduler on | |
476 | * the target CPU. | |
477 | */ | |
8875125e | 478 | void resched_curr(struct rq *rq) |
c24d20db | 479 | { |
8875125e | 480 | struct task_struct *curr = rq->curr; |
c24d20db IM |
481 | int cpu; |
482 | ||
8875125e | 483 | lockdep_assert_held(&rq->lock); |
c24d20db | 484 | |
8875125e | 485 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
486 | return; |
487 | ||
8875125e | 488 | cpu = cpu_of(rq); |
fd99f91a | 489 | |
f27dde8d | 490 | if (cpu == smp_processor_id()) { |
8875125e | 491 | set_tsk_need_resched(curr); |
f27dde8d | 492 | set_preempt_need_resched(); |
c24d20db | 493 | return; |
f27dde8d | 494 | } |
c24d20db | 495 | |
8875125e | 496 | if (set_nr_and_not_polling(curr)) |
c24d20db | 497 | smp_send_reschedule(cpu); |
dfc68f29 AL |
498 | else |
499 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
500 | } |
501 | ||
029632fb | 502 | void resched_cpu(int cpu) |
c24d20db IM |
503 | { |
504 | struct rq *rq = cpu_rq(cpu); | |
505 | unsigned long flags; | |
506 | ||
05fa785c | 507 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 508 | return; |
8875125e | 509 | resched_curr(rq); |
05fa785c | 510 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 511 | } |
06d8308c | 512 | |
b021fe3e | 513 | #ifdef CONFIG_SMP |
3451d024 | 514 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 515 | /* |
d1ccc66d IM |
516 | * In the semi idle case, use the nearest busy CPU for migrating timers |
517 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
518 | * |
519 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
520 | * selecting an idle CPU will add more delays to the timers than intended |
521 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 522 | */ |
bc7a34b8 | 523 | int get_nohz_timer_target(void) |
83cd4fe2 | 524 | { |
bc7a34b8 | 525 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
526 | struct sched_domain *sd; |
527 | ||
9642d18e | 528 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
529 | return cpu; |
530 | ||
057f3fad | 531 | rcu_read_lock(); |
83cd4fe2 | 532 | for_each_domain(cpu, sd) { |
057f3fad | 533 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
534 | if (cpu == i) |
535 | continue; | |
536 | ||
537 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
538 | cpu = i; |
539 | goto unlock; | |
540 | } | |
541 | } | |
83cd4fe2 | 542 | } |
9642d18e VH |
543 | |
544 | if (!is_housekeeping_cpu(cpu)) | |
545 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
546 | unlock: |
547 | rcu_read_unlock(); | |
83cd4fe2 VP |
548 | return cpu; |
549 | } | |
d1ccc66d | 550 | |
06d8308c TG |
551 | /* |
552 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
553 | * idle CPU then this timer might expire before the next timer event | |
554 | * which is scheduled to wake up that CPU. In case of a completely | |
555 | * idle system the next event might even be infinite time into the | |
556 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
557 | * leaves the inner idle loop so the newly added timer is taken into | |
558 | * account when the CPU goes back to idle and evaluates the timer | |
559 | * wheel for the next timer event. | |
560 | */ | |
1c20091e | 561 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
562 | { |
563 | struct rq *rq = cpu_rq(cpu); | |
564 | ||
565 | if (cpu == smp_processor_id()) | |
566 | return; | |
567 | ||
67b9ca70 | 568 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 569 | smp_send_reschedule(cpu); |
dfc68f29 AL |
570 | else |
571 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
572 | } |
573 | ||
c5bfece2 | 574 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 575 | { |
53c5fa16 FW |
576 | /* |
577 | * We just need the target to call irq_exit() and re-evaluate | |
578 | * the next tick. The nohz full kick at least implies that. | |
579 | * If needed we can still optimize that later with an | |
580 | * empty IRQ. | |
581 | */ | |
379d9ecb PM |
582 | if (cpu_is_offline(cpu)) |
583 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 584 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
585 | if (cpu != smp_processor_id() || |
586 | tick_nohz_tick_stopped()) | |
53c5fa16 | 587 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
588 | return true; |
589 | } | |
590 | ||
591 | return false; | |
592 | } | |
593 | ||
379d9ecb PM |
594 | /* |
595 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
596 | * caller's responsibility to deal with the lost wakeup, for example, | |
597 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
598 | */ | |
1c20091e FW |
599 | void wake_up_nohz_cpu(int cpu) |
600 | { | |
c5bfece2 | 601 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
602 | wake_up_idle_cpu(cpu); |
603 | } | |
604 | ||
ca38062e | 605 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 606 | { |
1c792db7 | 607 | int cpu = smp_processor_id(); |
873b4c65 VG |
608 | |
609 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
610 | return false; | |
611 | ||
612 | if (idle_cpu(cpu) && !need_resched()) | |
613 | return true; | |
614 | ||
615 | /* | |
616 | * We can't run Idle Load Balance on this CPU for this time so we | |
617 | * cancel it and clear NOHZ_BALANCE_KICK | |
618 | */ | |
619 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
620 | return false; | |
45bf76df IM |
621 | } |
622 | ||
3451d024 | 623 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 624 | |
ca38062e | 625 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 626 | { |
ca38062e | 627 | return false; |
2069dd75 PZ |
628 | } |
629 | ||
3451d024 | 630 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 631 | |
ce831b38 | 632 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 633 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 634 | { |
76d92ac3 FW |
635 | int fifo_nr_running; |
636 | ||
637 | /* Deadline tasks, even if single, need the tick */ | |
638 | if (rq->dl.dl_nr_running) | |
639 | return false; | |
640 | ||
1e78cdbd | 641 | /* |
2548d546 PZ |
642 | * If there are more than one RR tasks, we need the tick to effect the |
643 | * actual RR behaviour. | |
1e78cdbd | 644 | */ |
76d92ac3 FW |
645 | if (rq->rt.rr_nr_running) { |
646 | if (rq->rt.rr_nr_running == 1) | |
647 | return true; | |
648 | else | |
649 | return false; | |
1e78cdbd RR |
650 | } |
651 | ||
2548d546 PZ |
652 | /* |
653 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
654 | * forced preemption between FIFO tasks. | |
655 | */ | |
656 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
657 | if (fifo_nr_running) | |
658 | return true; | |
659 | ||
660 | /* | |
661 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
662 | * if there's more than one we need the tick for involuntary | |
663 | * preemption. | |
664 | */ | |
665 | if (rq->nr_running > 1) | |
541b8264 | 666 | return false; |
ce831b38 | 667 | |
541b8264 | 668 | return true; |
ce831b38 FW |
669 | } |
670 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 671 | |
029632fb | 672 | void sched_avg_update(struct rq *rq) |
18d95a28 | 673 | { |
e9e9250b PZ |
674 | s64 period = sched_avg_period(); |
675 | ||
78becc27 | 676 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
677 | /* |
678 | * Inline assembly required to prevent the compiler | |
679 | * optimising this loop into a divmod call. | |
680 | * See __iter_div_u64_rem() for another example of this. | |
681 | */ | |
682 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
683 | rq->age_stamp += period; |
684 | rq->rt_avg /= 2; | |
685 | } | |
18d95a28 PZ |
686 | } |
687 | ||
6d6bc0ad | 688 | #endif /* CONFIG_SMP */ |
18d95a28 | 689 | |
a790de99 PT |
690 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
691 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 692 | /* |
8277434e PT |
693 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
694 | * node and @up when leaving it for the final time. | |
695 | * | |
696 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 697 | */ |
029632fb | 698 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 699 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
700 | { |
701 | struct task_group *parent, *child; | |
eb755805 | 702 | int ret; |
c09595f6 | 703 | |
8277434e PT |
704 | parent = from; |
705 | ||
c09595f6 | 706 | down: |
eb755805 PZ |
707 | ret = (*down)(parent, data); |
708 | if (ret) | |
8277434e | 709 | goto out; |
c09595f6 PZ |
710 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
711 | parent = child; | |
712 | goto down; | |
713 | ||
714 | up: | |
715 | continue; | |
716 | } | |
eb755805 | 717 | ret = (*up)(parent, data); |
8277434e PT |
718 | if (ret || parent == from) |
719 | goto out; | |
c09595f6 PZ |
720 | |
721 | child = parent; | |
722 | parent = parent->parent; | |
723 | if (parent) | |
724 | goto up; | |
8277434e | 725 | out: |
eb755805 | 726 | return ret; |
c09595f6 PZ |
727 | } |
728 | ||
029632fb | 729 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 730 | { |
e2b245f8 | 731 | return 0; |
eb755805 | 732 | } |
18d95a28 PZ |
733 | #endif |
734 | ||
45bf76df IM |
735 | static void set_load_weight(struct task_struct *p) |
736 | { | |
f05998d4 NR |
737 | int prio = p->static_prio - MAX_RT_PRIO; |
738 | struct load_weight *load = &p->se.load; | |
739 | ||
dd41f596 IM |
740 | /* |
741 | * SCHED_IDLE tasks get minimal weight: | |
742 | */ | |
20f9cd2a | 743 | if (idle_policy(p->policy)) { |
c8b28116 | 744 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 745 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
746 | return; |
747 | } | |
71f8bd46 | 748 | |
ed82b8a1 AK |
749 | load->weight = scale_load(sched_prio_to_weight[prio]); |
750 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
751 | } |
752 | ||
1de64443 | 753 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 754 | { |
0a67d1ee PZ |
755 | if (!(flags & ENQUEUE_NOCLOCK)) |
756 | update_rq_clock(rq); | |
757 | ||
1de64443 PZ |
758 | if (!(flags & ENQUEUE_RESTORE)) |
759 | sched_info_queued(rq, p); | |
0a67d1ee | 760 | |
371fd7e7 | 761 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
762 | } |
763 | ||
1de64443 | 764 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 765 | { |
0a67d1ee PZ |
766 | if (!(flags & DEQUEUE_NOCLOCK)) |
767 | update_rq_clock(rq); | |
768 | ||
1de64443 PZ |
769 | if (!(flags & DEQUEUE_SAVE)) |
770 | sched_info_dequeued(rq, p); | |
0a67d1ee | 771 | |
371fd7e7 | 772 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
773 | } |
774 | ||
029632fb | 775 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
776 | { |
777 | if (task_contributes_to_load(p)) | |
778 | rq->nr_uninterruptible--; | |
779 | ||
371fd7e7 | 780 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
781 | } |
782 | ||
029632fb | 783 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
784 | { |
785 | if (task_contributes_to_load(p)) | |
786 | rq->nr_uninterruptible++; | |
787 | ||
371fd7e7 | 788 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
789 | } |
790 | ||
34f971f6 PZ |
791 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
792 | { | |
793 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
794 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
795 | ||
796 | if (stop) { | |
797 | /* | |
798 | * Make it appear like a SCHED_FIFO task, its something | |
799 | * userspace knows about and won't get confused about. | |
800 | * | |
801 | * Also, it will make PI more or less work without too | |
802 | * much confusion -- but then, stop work should not | |
803 | * rely on PI working anyway. | |
804 | */ | |
805 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
806 | ||
807 | stop->sched_class = &stop_sched_class; | |
808 | } | |
809 | ||
810 | cpu_rq(cpu)->stop = stop; | |
811 | ||
812 | if (old_stop) { | |
813 | /* | |
814 | * Reset it back to a normal scheduling class so that | |
815 | * it can die in pieces. | |
816 | */ | |
817 | old_stop->sched_class = &rt_sched_class; | |
818 | } | |
819 | } | |
820 | ||
14531189 | 821 | /* |
dd41f596 | 822 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 823 | */ |
14531189 IM |
824 | static inline int __normal_prio(struct task_struct *p) |
825 | { | |
dd41f596 | 826 | return p->static_prio; |
14531189 IM |
827 | } |
828 | ||
b29739f9 IM |
829 | /* |
830 | * Calculate the expected normal priority: i.e. priority | |
831 | * without taking RT-inheritance into account. Might be | |
832 | * boosted by interactivity modifiers. Changes upon fork, | |
833 | * setprio syscalls, and whenever the interactivity | |
834 | * estimator recalculates. | |
835 | */ | |
36c8b586 | 836 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
837 | { |
838 | int prio; | |
839 | ||
aab03e05 DF |
840 | if (task_has_dl_policy(p)) |
841 | prio = MAX_DL_PRIO-1; | |
842 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
843 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
844 | else | |
845 | prio = __normal_prio(p); | |
846 | return prio; | |
847 | } | |
848 | ||
849 | /* | |
850 | * Calculate the current priority, i.e. the priority | |
851 | * taken into account by the scheduler. This value might | |
852 | * be boosted by RT tasks, or might be boosted by | |
853 | * interactivity modifiers. Will be RT if the task got | |
854 | * RT-boosted. If not then it returns p->normal_prio. | |
855 | */ | |
36c8b586 | 856 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
857 | { |
858 | p->normal_prio = normal_prio(p); | |
859 | /* | |
860 | * If we are RT tasks or we were boosted to RT priority, | |
861 | * keep the priority unchanged. Otherwise, update priority | |
862 | * to the normal priority: | |
863 | */ | |
864 | if (!rt_prio(p->prio)) | |
865 | return p->normal_prio; | |
866 | return p->prio; | |
867 | } | |
868 | ||
1da177e4 LT |
869 | /** |
870 | * task_curr - is this task currently executing on a CPU? | |
871 | * @p: the task in question. | |
e69f6186 YB |
872 | * |
873 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 874 | */ |
36c8b586 | 875 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
876 | { |
877 | return cpu_curr(task_cpu(p)) == p; | |
878 | } | |
879 | ||
67dfa1b7 | 880 | /* |
4c9a4bc8 PZ |
881 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
882 | * use the balance_callback list if you want balancing. | |
883 | * | |
884 | * this means any call to check_class_changed() must be followed by a call to | |
885 | * balance_callback(). | |
67dfa1b7 | 886 | */ |
cb469845 SR |
887 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
888 | const struct sched_class *prev_class, | |
da7a735e | 889 | int oldprio) |
cb469845 SR |
890 | { |
891 | if (prev_class != p->sched_class) { | |
892 | if (prev_class->switched_from) | |
da7a735e | 893 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 894 | |
da7a735e | 895 | p->sched_class->switched_to(rq, p); |
2d3d891d | 896 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 897 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
898 | } |
899 | ||
029632fb | 900 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
901 | { |
902 | const struct sched_class *class; | |
903 | ||
904 | if (p->sched_class == rq->curr->sched_class) { | |
905 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
906 | } else { | |
907 | for_each_class(class) { | |
908 | if (class == rq->curr->sched_class) | |
909 | break; | |
910 | if (class == p->sched_class) { | |
8875125e | 911 | resched_curr(rq); |
1e5a7405 PZ |
912 | break; |
913 | } | |
914 | } | |
915 | } | |
916 | ||
917 | /* | |
918 | * A queue event has occurred, and we're going to schedule. In | |
919 | * this case, we can save a useless back to back clock update. | |
920 | */ | |
da0c1e65 | 921 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 922 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
923 | } |
924 | ||
1da177e4 | 925 | #ifdef CONFIG_SMP |
5cc389bc PZ |
926 | /* |
927 | * This is how migration works: | |
928 | * | |
929 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
930 | * stop_one_cpu(). | |
931 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
932 | * off the CPU) | |
933 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
934 | * 4) if it's in the wrong runqueue then the migration thread removes | |
935 | * it and puts it into the right queue. | |
936 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
937 | * is done. | |
938 | */ | |
939 | ||
940 | /* | |
941 | * move_queued_task - move a queued task to new rq. | |
942 | * | |
943 | * Returns (locked) new rq. Old rq's lock is released. | |
944 | */ | |
8a8c69c3 PZ |
945 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
946 | struct task_struct *p, int new_cpu) | |
5cc389bc | 947 | { |
5cc389bc PZ |
948 | lockdep_assert_held(&rq->lock); |
949 | ||
5cc389bc | 950 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 951 | dequeue_task(rq, p, 0); |
5cc389bc | 952 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 953 | rq_unlock(rq, rf); |
5cc389bc PZ |
954 | |
955 | rq = cpu_rq(new_cpu); | |
956 | ||
8a8c69c3 | 957 | rq_lock(rq, rf); |
5cc389bc | 958 | BUG_ON(task_cpu(p) != new_cpu); |
5cc389bc | 959 | enqueue_task(rq, p, 0); |
3ea94de1 | 960 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
961 | check_preempt_curr(rq, p, 0); |
962 | ||
963 | return rq; | |
964 | } | |
965 | ||
966 | struct migration_arg { | |
967 | struct task_struct *task; | |
968 | int dest_cpu; | |
969 | }; | |
970 | ||
971 | /* | |
d1ccc66d | 972 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
973 | * this because either it can't run here any more (set_cpus_allowed() |
974 | * away from this CPU, or CPU going down), or because we're | |
975 | * attempting to rebalance this task on exec (sched_exec). | |
976 | * | |
977 | * So we race with normal scheduler movements, but that's OK, as long | |
978 | * as the task is no longer on this CPU. | |
5cc389bc | 979 | */ |
8a8c69c3 PZ |
980 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
981 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 982 | { |
5cc389bc | 983 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 984 | return rq; |
5cc389bc PZ |
985 | |
986 | /* Affinity changed (again). */ | |
0c98d344 | 987 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
5e16bbc2 | 988 | return rq; |
5cc389bc | 989 | |
8a8c69c3 | 990 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
991 | |
992 | return rq; | |
5cc389bc PZ |
993 | } |
994 | ||
995 | /* | |
996 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
997 | * and performs thread migration by bumping thread off CPU then | |
998 | * 'pushing' onto another runqueue. | |
999 | */ | |
1000 | static int migration_cpu_stop(void *data) | |
1001 | { | |
1002 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1003 | struct task_struct *p = arg->task; |
1004 | struct rq *rq = this_rq(); | |
8a8c69c3 | 1005 | struct rq_flags rf; |
5cc389bc PZ |
1006 | |
1007 | /* | |
d1ccc66d IM |
1008 | * The original target CPU might have gone down and we might |
1009 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
1010 | */ |
1011 | local_irq_disable(); | |
1012 | /* | |
1013 | * We need to explicitly wake pending tasks before running | |
1014 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1015 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1016 | */ | |
1017 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1018 | |
1019 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1020 | rq_lock(rq, &rf); |
5e16bbc2 PZ |
1021 | /* |
1022 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1023 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1024 | * we're holding p->pi_lock. | |
1025 | */ | |
bf89a304 CC |
1026 | if (task_rq(p) == rq) { |
1027 | if (task_on_rq_queued(p)) | |
8a8c69c3 | 1028 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 CC |
1029 | else |
1030 | p->wake_cpu = arg->dest_cpu; | |
1031 | } | |
8a8c69c3 | 1032 | rq_unlock(rq, &rf); |
5e16bbc2 PZ |
1033 | raw_spin_unlock(&p->pi_lock); |
1034 | ||
5cc389bc PZ |
1035 | local_irq_enable(); |
1036 | return 0; | |
1037 | } | |
1038 | ||
c5b28038 PZ |
1039 | /* |
1040 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1041 | * actually call this function. | |
1042 | */ | |
1043 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1044 | { |
5cc389bc PZ |
1045 | cpumask_copy(&p->cpus_allowed, new_mask); |
1046 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1047 | } | |
1048 | ||
c5b28038 PZ |
1049 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1050 | { | |
6c37067e PZ |
1051 | struct rq *rq = task_rq(p); |
1052 | bool queued, running; | |
1053 | ||
c5b28038 | 1054 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1055 | |
1056 | queued = task_on_rq_queued(p); | |
1057 | running = task_current(rq, p); | |
1058 | ||
1059 | if (queued) { | |
1060 | /* | |
1061 | * Because __kthread_bind() calls this on blocked tasks without | |
1062 | * holding rq->lock. | |
1063 | */ | |
1064 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1065 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1066 | } |
1067 | if (running) | |
1068 | put_prev_task(rq, p); | |
1069 | ||
c5b28038 | 1070 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1071 | |
6c37067e | 1072 | if (queued) |
7134b3e9 | 1073 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 1074 | if (running) |
b2bf6c31 | 1075 | set_curr_task(rq, p); |
c5b28038 PZ |
1076 | } |
1077 | ||
5cc389bc PZ |
1078 | /* |
1079 | * Change a given task's CPU affinity. Migrate the thread to a | |
1080 | * proper CPU and schedule it away if the CPU it's executing on | |
1081 | * is removed from the allowed bitmask. | |
1082 | * | |
1083 | * NOTE: the caller must have a valid reference to the task, the | |
1084 | * task must not exit() & deallocate itself prematurely. The | |
1085 | * call is not atomic; no spinlocks may be held. | |
1086 | */ | |
25834c73 PZ |
1087 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1088 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1089 | { |
e9d867a6 | 1090 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1091 | unsigned int dest_cpu; |
eb580751 PZ |
1092 | struct rq_flags rf; |
1093 | struct rq *rq; | |
5cc389bc PZ |
1094 | int ret = 0; |
1095 | ||
eb580751 | 1096 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1097 | update_rq_clock(rq); |
5cc389bc | 1098 | |
e9d867a6 PZI |
1099 | if (p->flags & PF_KTHREAD) { |
1100 | /* | |
1101 | * Kernel threads are allowed on online && !active CPUs | |
1102 | */ | |
1103 | cpu_valid_mask = cpu_online_mask; | |
1104 | } | |
1105 | ||
25834c73 PZ |
1106 | /* |
1107 | * Must re-check here, to close a race against __kthread_bind(), | |
1108 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1109 | */ | |
1110 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1111 | ret = -EINVAL; | |
1112 | goto out; | |
1113 | } | |
1114 | ||
5cc389bc PZ |
1115 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1116 | goto out; | |
1117 | ||
e9d867a6 | 1118 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1119 | ret = -EINVAL; |
1120 | goto out; | |
1121 | } | |
1122 | ||
1123 | do_set_cpus_allowed(p, new_mask); | |
1124 | ||
e9d867a6 PZI |
1125 | if (p->flags & PF_KTHREAD) { |
1126 | /* | |
1127 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1128 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1129 | */ |
1130 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1131 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1132 | p->nr_cpus_allowed != 1); | |
1133 | } | |
1134 | ||
5cc389bc PZ |
1135 | /* Can the task run on the task's current CPU? If so, we're done */ |
1136 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1137 | goto out; | |
1138 | ||
e9d867a6 | 1139 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1140 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1141 | struct migration_arg arg = { p, dest_cpu }; | |
1142 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1143 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1144 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1145 | tlb_migrate_finish(p->mm); | |
1146 | return 0; | |
cbce1a68 PZ |
1147 | } else if (task_on_rq_queued(p)) { |
1148 | /* | |
1149 | * OK, since we're going to drop the lock immediately | |
1150 | * afterwards anyway. | |
1151 | */ | |
8a8c69c3 | 1152 | rq = move_queued_task(rq, &rf, p, dest_cpu); |
cbce1a68 | 1153 | } |
5cc389bc | 1154 | out: |
eb580751 | 1155 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1156 | |
1157 | return ret; | |
1158 | } | |
25834c73 PZ |
1159 | |
1160 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1161 | { | |
1162 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1163 | } | |
5cc389bc PZ |
1164 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1165 | ||
dd41f596 | 1166 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1167 | { |
e2912009 PZ |
1168 | #ifdef CONFIG_SCHED_DEBUG |
1169 | /* | |
1170 | * We should never call set_task_cpu() on a blocked task, | |
1171 | * ttwu() will sort out the placement. | |
1172 | */ | |
077614ee | 1173 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1174 | !p->on_rq); |
0122ec5b | 1175 | |
3ea94de1 JP |
1176 | /* |
1177 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1178 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1179 | * time relying on p->on_rq. | |
1180 | */ | |
1181 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1182 | p->sched_class == &fair_sched_class && | |
1183 | (p->on_rq && !task_on_rq_migrating(p))); | |
1184 | ||
0122ec5b | 1185 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1186 | /* |
1187 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1188 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1189 | * | |
1190 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1191 | * see task_group(). |
6c6c54e1 PZ |
1192 | * |
1193 | * Furthermore, all task_rq users should acquire both locks, see | |
1194 | * task_rq_lock(). | |
1195 | */ | |
0122ec5b PZ |
1196 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1197 | lockdep_is_held(&task_rq(p)->lock))); | |
1198 | #endif | |
e2912009 PZ |
1199 | #endif |
1200 | ||
de1d7286 | 1201 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1202 | |
0c69774e | 1203 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1204 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1205 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1206 | p->se.nr_migrations++; |
ff303e66 | 1207 | perf_event_task_migrate(p); |
0c69774e | 1208 | } |
dd41f596 IM |
1209 | |
1210 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1211 | } |
1212 | ||
ac66f547 PZ |
1213 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1214 | { | |
da0c1e65 | 1215 | if (task_on_rq_queued(p)) { |
ac66f547 | 1216 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 1217 | struct rq_flags srf, drf; |
ac66f547 PZ |
1218 | |
1219 | src_rq = task_rq(p); | |
1220 | dst_rq = cpu_rq(cpu); | |
1221 | ||
8a8c69c3 PZ |
1222 | rq_pin_lock(src_rq, &srf); |
1223 | rq_pin_lock(dst_rq, &drf); | |
1224 | ||
3ea94de1 | 1225 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1226 | deactivate_task(src_rq, p, 0); |
1227 | set_task_cpu(p, cpu); | |
1228 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1229 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 | 1230 | check_preempt_curr(dst_rq, p, 0); |
8a8c69c3 PZ |
1231 | |
1232 | rq_unpin_lock(dst_rq, &drf); | |
1233 | rq_unpin_lock(src_rq, &srf); | |
1234 | ||
ac66f547 PZ |
1235 | } else { |
1236 | /* | |
1237 | * Task isn't running anymore; make it appear like we migrated | |
1238 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 1239 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
1240 | */ |
1241 | p->wake_cpu = cpu; | |
1242 | } | |
1243 | } | |
1244 | ||
1245 | struct migration_swap_arg { | |
1246 | struct task_struct *src_task, *dst_task; | |
1247 | int src_cpu, dst_cpu; | |
1248 | }; | |
1249 | ||
1250 | static int migrate_swap_stop(void *data) | |
1251 | { | |
1252 | struct migration_swap_arg *arg = data; | |
1253 | struct rq *src_rq, *dst_rq; | |
1254 | int ret = -EAGAIN; | |
1255 | ||
62694cd5 PZ |
1256 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1257 | return -EAGAIN; | |
1258 | ||
ac66f547 PZ |
1259 | src_rq = cpu_rq(arg->src_cpu); |
1260 | dst_rq = cpu_rq(arg->dst_cpu); | |
1261 | ||
74602315 PZ |
1262 | double_raw_lock(&arg->src_task->pi_lock, |
1263 | &arg->dst_task->pi_lock); | |
ac66f547 | 1264 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1265 | |
ac66f547 PZ |
1266 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1267 | goto unlock; | |
1268 | ||
1269 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1270 | goto unlock; | |
1271 | ||
0c98d344 | 1272 | if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed)) |
ac66f547 PZ |
1273 | goto unlock; |
1274 | ||
0c98d344 | 1275 | if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed)) |
ac66f547 PZ |
1276 | goto unlock; |
1277 | ||
1278 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1279 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1280 | ||
1281 | ret = 0; | |
1282 | ||
1283 | unlock: | |
1284 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1285 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1286 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1287 | |
1288 | return ret; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Cross migrate two tasks | |
1293 | */ | |
1294 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1295 | { | |
1296 | struct migration_swap_arg arg; | |
1297 | int ret = -EINVAL; | |
1298 | ||
ac66f547 PZ |
1299 | arg = (struct migration_swap_arg){ |
1300 | .src_task = cur, | |
1301 | .src_cpu = task_cpu(cur), | |
1302 | .dst_task = p, | |
1303 | .dst_cpu = task_cpu(p), | |
1304 | }; | |
1305 | ||
1306 | if (arg.src_cpu == arg.dst_cpu) | |
1307 | goto out; | |
1308 | ||
6acce3ef PZ |
1309 | /* |
1310 | * These three tests are all lockless; this is OK since all of them | |
1311 | * will be re-checked with proper locks held further down the line. | |
1312 | */ | |
ac66f547 PZ |
1313 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1314 | goto out; | |
1315 | ||
0c98d344 | 1316 | if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed)) |
ac66f547 PZ |
1317 | goto out; |
1318 | ||
0c98d344 | 1319 | if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed)) |
ac66f547 PZ |
1320 | goto out; |
1321 | ||
286549dc | 1322 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1323 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1324 | ||
1325 | out: | |
ac66f547 PZ |
1326 | return ret; |
1327 | } | |
1328 | ||
1da177e4 LT |
1329 | /* |
1330 | * wait_task_inactive - wait for a thread to unschedule. | |
1331 | * | |
85ba2d86 RM |
1332 | * If @match_state is nonzero, it's the @p->state value just checked and |
1333 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1334 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1335 | * we return a positive number (its total switch count). If a second call | |
1336 | * a short while later returns the same number, the caller can be sure that | |
1337 | * @p has remained unscheduled the whole time. | |
1338 | * | |
1da177e4 LT |
1339 | * The caller must ensure that the task *will* unschedule sometime soon, |
1340 | * else this function might spin for a *long* time. This function can't | |
1341 | * be called with interrupts off, or it may introduce deadlock with | |
1342 | * smp_call_function() if an IPI is sent by the same process we are | |
1343 | * waiting to become inactive. | |
1344 | */ | |
85ba2d86 | 1345 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1346 | { |
da0c1e65 | 1347 | int running, queued; |
eb580751 | 1348 | struct rq_flags rf; |
85ba2d86 | 1349 | unsigned long ncsw; |
70b97a7f | 1350 | struct rq *rq; |
1da177e4 | 1351 | |
3a5c359a AK |
1352 | for (;;) { |
1353 | /* | |
1354 | * We do the initial early heuristics without holding | |
1355 | * any task-queue locks at all. We'll only try to get | |
1356 | * the runqueue lock when things look like they will | |
1357 | * work out! | |
1358 | */ | |
1359 | rq = task_rq(p); | |
fa490cfd | 1360 | |
3a5c359a AK |
1361 | /* |
1362 | * If the task is actively running on another CPU | |
1363 | * still, just relax and busy-wait without holding | |
1364 | * any locks. | |
1365 | * | |
1366 | * NOTE! Since we don't hold any locks, it's not | |
1367 | * even sure that "rq" stays as the right runqueue! | |
1368 | * But we don't care, since "task_running()" will | |
1369 | * return false if the runqueue has changed and p | |
1370 | * is actually now running somewhere else! | |
1371 | */ | |
85ba2d86 RM |
1372 | while (task_running(rq, p)) { |
1373 | if (match_state && unlikely(p->state != match_state)) | |
1374 | return 0; | |
3a5c359a | 1375 | cpu_relax(); |
85ba2d86 | 1376 | } |
fa490cfd | 1377 | |
3a5c359a AK |
1378 | /* |
1379 | * Ok, time to look more closely! We need the rq | |
1380 | * lock now, to be *sure*. If we're wrong, we'll | |
1381 | * just go back and repeat. | |
1382 | */ | |
eb580751 | 1383 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1384 | trace_sched_wait_task(p); |
3a5c359a | 1385 | running = task_running(rq, p); |
da0c1e65 | 1386 | queued = task_on_rq_queued(p); |
85ba2d86 | 1387 | ncsw = 0; |
f31e11d8 | 1388 | if (!match_state || p->state == match_state) |
93dcf55f | 1389 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1390 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1391 | |
85ba2d86 RM |
1392 | /* |
1393 | * If it changed from the expected state, bail out now. | |
1394 | */ | |
1395 | if (unlikely(!ncsw)) | |
1396 | break; | |
1397 | ||
3a5c359a AK |
1398 | /* |
1399 | * Was it really running after all now that we | |
1400 | * checked with the proper locks actually held? | |
1401 | * | |
1402 | * Oops. Go back and try again.. | |
1403 | */ | |
1404 | if (unlikely(running)) { | |
1405 | cpu_relax(); | |
1406 | continue; | |
1407 | } | |
fa490cfd | 1408 | |
3a5c359a AK |
1409 | /* |
1410 | * It's not enough that it's not actively running, | |
1411 | * it must be off the runqueue _entirely_, and not | |
1412 | * preempted! | |
1413 | * | |
80dd99b3 | 1414 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1415 | * running right now), it's preempted, and we should |
1416 | * yield - it could be a while. | |
1417 | */ | |
da0c1e65 | 1418 | if (unlikely(queued)) { |
8b0e1953 | 1419 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1420 | |
1421 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1422 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1423 | continue; |
1424 | } | |
fa490cfd | 1425 | |
3a5c359a AK |
1426 | /* |
1427 | * Ahh, all good. It wasn't running, and it wasn't | |
1428 | * runnable, which means that it will never become | |
1429 | * running in the future either. We're all done! | |
1430 | */ | |
1431 | break; | |
1432 | } | |
85ba2d86 RM |
1433 | |
1434 | return ncsw; | |
1da177e4 LT |
1435 | } |
1436 | ||
1437 | /*** | |
1438 | * kick_process - kick a running thread to enter/exit the kernel | |
1439 | * @p: the to-be-kicked thread | |
1440 | * | |
1441 | * Cause a process which is running on another CPU to enter | |
1442 | * kernel-mode, without any delay. (to get signals handled.) | |
1443 | * | |
25985edc | 1444 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1445 | * because all it wants to ensure is that the remote task enters |
1446 | * the kernel. If the IPI races and the task has been migrated | |
1447 | * to another CPU then no harm is done and the purpose has been | |
1448 | * achieved as well. | |
1449 | */ | |
36c8b586 | 1450 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1451 | { |
1452 | int cpu; | |
1453 | ||
1454 | preempt_disable(); | |
1455 | cpu = task_cpu(p); | |
1456 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1457 | smp_send_reschedule(cpu); | |
1458 | preempt_enable(); | |
1459 | } | |
b43e3521 | 1460 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1461 | |
30da688e | 1462 | /* |
013fdb80 | 1463 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1464 | * |
1465 | * A few notes on cpu_active vs cpu_online: | |
1466 | * | |
1467 | * - cpu_active must be a subset of cpu_online | |
1468 | * | |
1469 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1470 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
d1ccc66d | 1471 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
1472 | * see it. |
1473 | * | |
d1ccc66d | 1474 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 1475 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 1476 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
1477 | * off. |
1478 | * | |
1479 | * This means that fallback selection must not select !active CPUs. | |
1480 | * And can assume that any active CPU must be online. Conversely | |
1481 | * select_task_rq() below may allow selection of !active CPUs in order | |
1482 | * to satisfy the above rules. | |
30da688e | 1483 | */ |
5da9a0fb PZ |
1484 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1485 | { | |
aa00d89c TC |
1486 | int nid = cpu_to_node(cpu); |
1487 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1488 | enum { cpuset, possible, fail } state = cpuset; |
1489 | int dest_cpu; | |
5da9a0fb | 1490 | |
aa00d89c | 1491 | /* |
d1ccc66d IM |
1492 | * If the node that the CPU is on has been offlined, cpu_to_node() |
1493 | * will return -1. There is no CPU on the node, and we should | |
1494 | * select the CPU on the other node. | |
aa00d89c TC |
1495 | */ |
1496 | if (nid != -1) { | |
1497 | nodemask = cpumask_of_node(nid); | |
1498 | ||
1499 | /* Look for allowed, online CPU in same node. */ | |
1500 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1501 | if (!cpu_active(dest_cpu)) |
1502 | continue; | |
0c98d344 | 1503 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
aa00d89c TC |
1504 | return dest_cpu; |
1505 | } | |
2baab4e9 | 1506 | } |
5da9a0fb | 1507 | |
2baab4e9 PZ |
1508 | for (;;) { |
1509 | /* Any allowed, online CPU? */ | |
0c98d344 | 1510 | for_each_cpu(dest_cpu, &p->cpus_allowed) { |
feb245e3 TH |
1511 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
1512 | continue; | |
1513 | if (!cpu_online(dest_cpu)) | |
2baab4e9 PZ |
1514 | continue; |
1515 | goto out; | |
1516 | } | |
5da9a0fb | 1517 | |
e73e85f0 | 1518 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1519 | switch (state) { |
1520 | case cpuset: | |
e73e85f0 ON |
1521 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1522 | cpuset_cpus_allowed_fallback(p); | |
1523 | state = possible; | |
1524 | break; | |
1525 | } | |
d1ccc66d | 1526 | /* Fall-through */ |
2baab4e9 PZ |
1527 | case possible: |
1528 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1529 | state = fail; | |
1530 | break; | |
1531 | ||
1532 | case fail: | |
1533 | BUG(); | |
1534 | break; | |
1535 | } | |
1536 | } | |
1537 | ||
1538 | out: | |
1539 | if (state != cpuset) { | |
1540 | /* | |
1541 | * Don't tell them about moving exiting tasks or | |
1542 | * kernel threads (both mm NULL), since they never | |
1543 | * leave kernel. | |
1544 | */ | |
1545 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1546 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1547 | task_pid_nr(p), p->comm, cpu); |
1548 | } | |
5da9a0fb PZ |
1549 | } |
1550 | ||
1551 | return dest_cpu; | |
1552 | } | |
1553 | ||
e2912009 | 1554 | /* |
013fdb80 | 1555 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1556 | */ |
970b13ba | 1557 | static inline |
ac66f547 | 1558 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1559 | { |
cbce1a68 PZ |
1560 | lockdep_assert_held(&p->pi_lock); |
1561 | ||
4b53a341 | 1562 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 1563 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 1564 | else |
0c98d344 | 1565 | cpu = cpumask_any(&p->cpus_allowed); |
e2912009 PZ |
1566 | |
1567 | /* | |
1568 | * In order not to call set_task_cpu() on a blocking task we need | |
1569 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
d1ccc66d | 1570 | * CPU. |
e2912009 PZ |
1571 | * |
1572 | * Since this is common to all placement strategies, this lives here. | |
1573 | * | |
1574 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1575 | * not worry about this generic constraint ] | |
1576 | */ | |
0c98d344 | 1577 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || |
70f11205 | 1578 | !cpu_online(cpu))) |
5da9a0fb | 1579 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1580 | |
1581 | return cpu; | |
970b13ba | 1582 | } |
09a40af5 MG |
1583 | |
1584 | static void update_avg(u64 *avg, u64 sample) | |
1585 | { | |
1586 | s64 diff = sample - *avg; | |
1587 | *avg += diff >> 3; | |
1588 | } | |
25834c73 PZ |
1589 | |
1590 | #else | |
1591 | ||
1592 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1593 | const struct cpumask *new_mask, bool check) | |
1594 | { | |
1595 | return set_cpus_allowed_ptr(p, new_mask); | |
1596 | } | |
1597 | ||
5cc389bc | 1598 | #endif /* CONFIG_SMP */ |
970b13ba | 1599 | |
d7c01d27 | 1600 | static void |
b84cb5df | 1601 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1602 | { |
4fa8d299 | 1603 | struct rq *rq; |
b84cb5df | 1604 | |
4fa8d299 JP |
1605 | if (!schedstat_enabled()) |
1606 | return; | |
1607 | ||
1608 | rq = this_rq(); | |
d7c01d27 | 1609 | |
4fa8d299 JP |
1610 | #ifdef CONFIG_SMP |
1611 | if (cpu == rq->cpu) { | |
ae92882e JP |
1612 | schedstat_inc(rq->ttwu_local); |
1613 | schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1614 | } else { |
1615 | struct sched_domain *sd; | |
1616 | ||
ae92882e | 1617 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1618 | rcu_read_lock(); |
4fa8d299 | 1619 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1620 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
ae92882e | 1621 | schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1622 | break; |
1623 | } | |
1624 | } | |
057f3fad | 1625 | rcu_read_unlock(); |
d7c01d27 | 1626 | } |
f339b9dc PZ |
1627 | |
1628 | if (wake_flags & WF_MIGRATED) | |
ae92882e | 1629 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1630 | #endif /* CONFIG_SMP */ |
1631 | ||
ae92882e JP |
1632 | schedstat_inc(rq->ttwu_count); |
1633 | schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1634 | |
1635 | if (wake_flags & WF_SYNC) | |
ae92882e | 1636 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1637 | } |
1638 | ||
1de64443 | 1639 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1640 | { |
9ed3811a | 1641 | activate_task(rq, p, en_flags); |
da0c1e65 | 1642 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e | 1643 | |
d1ccc66d | 1644 | /* If a worker is waking up, notify the workqueue: */ |
c2f7115e PZ |
1645 | if (p->flags & PF_WQ_WORKER) |
1646 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1647 | } |
1648 | ||
23f41eeb PZ |
1649 | /* |
1650 | * Mark the task runnable and perform wakeup-preemption. | |
1651 | */ | |
e7904a28 | 1652 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1653 | struct rq_flags *rf) |
9ed3811a | 1654 | { |
9ed3811a | 1655 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1656 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1657 | trace_sched_wakeup(p); |
1658 | ||
9ed3811a | 1659 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1660 | if (p->sched_class->task_woken) { |
1661 | /* | |
cbce1a68 PZ |
1662 | * Our task @p is fully woken up and running; so its safe to |
1663 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1664 | */ |
d8ac8971 | 1665 | rq_unpin_lock(rq, rf); |
9ed3811a | 1666 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 1667 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 1668 | } |
9ed3811a | 1669 | |
e69c6341 | 1670 | if (rq->idle_stamp) { |
78becc27 | 1671 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1672 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1673 | |
abfafa54 JL |
1674 | update_avg(&rq->avg_idle, delta); |
1675 | ||
1676 | if (rq->avg_idle > max) | |
9ed3811a | 1677 | rq->avg_idle = max; |
abfafa54 | 1678 | |
9ed3811a TH |
1679 | rq->idle_stamp = 0; |
1680 | } | |
1681 | #endif | |
1682 | } | |
1683 | ||
c05fbafb | 1684 | static void |
e7904a28 | 1685 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1686 | struct rq_flags *rf) |
c05fbafb | 1687 | { |
77558e4d | 1688 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 1689 | |
cbce1a68 PZ |
1690 | lockdep_assert_held(&rq->lock); |
1691 | ||
c05fbafb PZ |
1692 | #ifdef CONFIG_SMP |
1693 | if (p->sched_contributes_to_load) | |
1694 | rq->nr_uninterruptible--; | |
b5179ac7 | 1695 | |
b5179ac7 | 1696 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1697 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1698 | #endif |
1699 | ||
b5179ac7 | 1700 | ttwu_activate(rq, p, en_flags); |
d8ac8971 | 1701 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
1702 | } |
1703 | ||
1704 | /* | |
1705 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1706 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1707 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1708 | * the task is still ->on_rq. | |
1709 | */ | |
1710 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1711 | { | |
eb580751 | 1712 | struct rq_flags rf; |
c05fbafb PZ |
1713 | struct rq *rq; |
1714 | int ret = 0; | |
1715 | ||
eb580751 | 1716 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1717 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1718 | /* check_preempt_curr() may use rq clock */ |
1719 | update_rq_clock(rq); | |
d8ac8971 | 1720 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
1721 | ret = 1; |
1722 | } | |
eb580751 | 1723 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1724 | |
1725 | return ret; | |
1726 | } | |
1727 | ||
317f3941 | 1728 | #ifdef CONFIG_SMP |
e3baac47 | 1729 | void sched_ttwu_pending(void) |
317f3941 PZ |
1730 | { |
1731 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1732 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1733 | struct task_struct *p; | |
d8ac8971 | 1734 | struct rq_flags rf; |
317f3941 | 1735 | |
e3baac47 PZ |
1736 | if (!llist) |
1737 | return; | |
1738 | ||
8a8c69c3 | 1739 | rq_lock_irqsave(rq, &rf); |
77558e4d | 1740 | update_rq_clock(rq); |
317f3941 | 1741 | |
fa14ff4a | 1742 | while (llist) { |
b7e7ade3 PZ |
1743 | int wake_flags = 0; |
1744 | ||
fa14ff4a PZ |
1745 | p = llist_entry(llist, struct task_struct, wake_entry); |
1746 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1747 | |
1748 | if (p->sched_remote_wakeup) | |
1749 | wake_flags = WF_MIGRATED; | |
1750 | ||
d8ac8971 | 1751 | ttwu_do_activate(rq, p, wake_flags, &rf); |
317f3941 PZ |
1752 | } |
1753 | ||
8a8c69c3 | 1754 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
1755 | } |
1756 | ||
1757 | void scheduler_ipi(void) | |
1758 | { | |
f27dde8d PZ |
1759 | /* |
1760 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1761 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1762 | * this IPI. | |
1763 | */ | |
8cb75e0c | 1764 | preempt_fold_need_resched(); |
f27dde8d | 1765 | |
fd2ac4f4 | 1766 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1767 | return; |
1768 | ||
1769 | /* | |
1770 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1771 | * traditionally all their work was done from the interrupt return | |
1772 | * path. Now that we actually do some work, we need to make sure | |
1773 | * we do call them. | |
1774 | * | |
1775 | * Some archs already do call them, luckily irq_enter/exit nest | |
1776 | * properly. | |
1777 | * | |
1778 | * Arguably we should visit all archs and update all handlers, | |
1779 | * however a fair share of IPIs are still resched only so this would | |
1780 | * somewhat pessimize the simple resched case. | |
1781 | */ | |
1782 | irq_enter(); | |
fa14ff4a | 1783 | sched_ttwu_pending(); |
ca38062e SS |
1784 | |
1785 | /* | |
1786 | * Check if someone kicked us for doing the nohz idle load balance. | |
1787 | */ | |
873b4c65 | 1788 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1789 | this_rq()->idle_balance = 1; |
ca38062e | 1790 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1791 | } |
c5d753a5 | 1792 | irq_exit(); |
317f3941 PZ |
1793 | } |
1794 | ||
b7e7ade3 | 1795 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1796 | { |
e3baac47 PZ |
1797 | struct rq *rq = cpu_rq(cpu); |
1798 | ||
b7e7ade3 PZ |
1799 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1800 | ||
e3baac47 PZ |
1801 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1802 | if (!set_nr_if_polling(rq->idle)) | |
1803 | smp_send_reschedule(cpu); | |
1804 | else | |
1805 | trace_sched_wake_idle_without_ipi(cpu); | |
1806 | } | |
317f3941 | 1807 | } |
d6aa8f85 | 1808 | |
f6be8af1 CL |
1809 | void wake_up_if_idle(int cpu) |
1810 | { | |
1811 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 1812 | struct rq_flags rf; |
f6be8af1 | 1813 | |
fd7de1e8 AL |
1814 | rcu_read_lock(); |
1815 | ||
1816 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1817 | goto out; | |
f6be8af1 CL |
1818 | |
1819 | if (set_nr_if_polling(rq->idle)) { | |
1820 | trace_sched_wake_idle_without_ipi(cpu); | |
1821 | } else { | |
8a8c69c3 | 1822 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
1823 | if (is_idle_task(rq->curr)) |
1824 | smp_send_reschedule(cpu); | |
d1ccc66d | 1825 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 1826 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 1827 | } |
fd7de1e8 AL |
1828 | |
1829 | out: | |
1830 | rcu_read_unlock(); | |
f6be8af1 CL |
1831 | } |
1832 | ||
39be3501 | 1833 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1834 | { |
1835 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1836 | } | |
d6aa8f85 | 1837 | #endif /* CONFIG_SMP */ |
317f3941 | 1838 | |
b5179ac7 | 1839 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1840 | { |
1841 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 1842 | struct rq_flags rf; |
c05fbafb | 1843 | |
17d9f311 | 1844 | #if defined(CONFIG_SMP) |
39be3501 | 1845 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
d1ccc66d | 1846 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
b7e7ade3 | 1847 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1848 | return; |
1849 | } | |
1850 | #endif | |
1851 | ||
8a8c69c3 | 1852 | rq_lock(rq, &rf); |
77558e4d | 1853 | update_rq_clock(rq); |
d8ac8971 | 1854 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 1855 | rq_unlock(rq, &rf); |
9ed3811a TH |
1856 | } |
1857 | ||
8643cda5 PZ |
1858 | /* |
1859 | * Notes on Program-Order guarantees on SMP systems. | |
1860 | * | |
1861 | * MIGRATION | |
1862 | * | |
1863 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
1864 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
1865 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
1866 | * |
1867 | * For migration (of runnable tasks) this is provided by the following means: | |
1868 | * | |
1869 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1870 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1871 | * rq(c1)->lock (if not at the same time, then in that order). | |
1872 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1873 | * | |
1874 | * Transitivity guarantees that B happens after A and C after B. | |
1875 | * Note: we only require RCpc transitivity. | |
d1ccc66d | 1876 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
1877 | * |
1878 | * Example: | |
1879 | * | |
1880 | * CPU0 CPU1 CPU2 | |
1881 | * | |
1882 | * LOCK rq(0)->lock | |
1883 | * sched-out X | |
1884 | * sched-in Y | |
1885 | * UNLOCK rq(0)->lock | |
1886 | * | |
1887 | * LOCK rq(0)->lock // orders against CPU0 | |
1888 | * dequeue X | |
1889 | * UNLOCK rq(0)->lock | |
1890 | * | |
1891 | * LOCK rq(1)->lock | |
1892 | * enqueue X | |
1893 | * UNLOCK rq(1)->lock | |
1894 | * | |
1895 | * LOCK rq(1)->lock // orders against CPU2 | |
1896 | * sched-out Z | |
1897 | * sched-in X | |
1898 | * UNLOCK rq(1)->lock | |
1899 | * | |
1900 | * | |
1901 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1902 | * | |
1903 | * For blocking we (obviously) need to provide the same guarantee as for | |
1904 | * migration. However the means are completely different as there is no lock | |
1905 | * chain to provide order. Instead we do: | |
1906 | * | |
1907 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1908 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1909 | * |
1910 | * Example: | |
1911 | * | |
1912 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1913 | * | |
1914 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1915 | * dequeue X | |
1916 | * sched-out X | |
1917 | * smp_store_release(X->on_cpu, 0); | |
1918 | * | |
1f03e8d2 | 1919 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1920 | * X->state = WAKING |
1921 | * set_task_cpu(X,2) | |
1922 | * | |
1923 | * LOCK rq(2)->lock | |
1924 | * enqueue X | |
1925 | * X->state = RUNNING | |
1926 | * UNLOCK rq(2)->lock | |
1927 | * | |
1928 | * LOCK rq(2)->lock // orders against CPU1 | |
1929 | * sched-out Z | |
1930 | * sched-in X | |
1931 | * UNLOCK rq(2)->lock | |
1932 | * | |
1933 | * UNLOCK X->pi_lock | |
1934 | * UNLOCK rq(0)->lock | |
1935 | * | |
1936 | * | |
1937 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1938 | * must observe the state that lead to our wakeup. That is, not only must our | |
1939 | * task observe its own prior state, it must also observe the stores prior to | |
1940 | * its wakeup. | |
1941 | * | |
1942 | * This means that any means of doing remote wakeups must order the CPU doing | |
1943 | * the wakeup against the CPU the task is going to end up running on. This, | |
1944 | * however, is already required for the regular Program-Order guarantee above, | |
1f03e8d2 | 1945 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
8643cda5 PZ |
1946 | * |
1947 | */ | |
1948 | ||
9ed3811a | 1949 | /** |
1da177e4 | 1950 | * try_to_wake_up - wake up a thread |
9ed3811a | 1951 | * @p: the thread to be awakened |
1da177e4 | 1952 | * @state: the mask of task states that can be woken |
9ed3811a | 1953 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 1954 | * |
a2250238 | 1955 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 1956 | * |
a2250238 PZ |
1957 | * If the task was not queued/runnable, also place it back on a runqueue. |
1958 | * | |
1959 | * Atomic against schedule() which would dequeue a task, also see | |
1960 | * set_current_state(). | |
1961 | * | |
1962 | * Return: %true if @p->state changes (an actual wakeup was done), | |
1963 | * %false otherwise. | |
1da177e4 | 1964 | */ |
e4a52bcb PZ |
1965 | static int |
1966 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1967 | { |
1da177e4 | 1968 | unsigned long flags; |
c05fbafb | 1969 | int cpu, success = 0; |
2398f2c6 | 1970 | |
e0acd0a6 ON |
1971 | /* |
1972 | * If we are going to wake up a thread waiting for CONDITION we | |
1973 | * need to ensure that CONDITION=1 done by the caller can not be | |
1974 | * reordered with p->state check below. This pairs with mb() in | |
1975 | * set_current_state() the waiting thread does. | |
1976 | */ | |
1977 | smp_mb__before_spinlock(); | |
013fdb80 | 1978 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1979 | if (!(p->state & state)) |
1da177e4 LT |
1980 | goto out; |
1981 | ||
fbd705a0 PZ |
1982 | trace_sched_waking(p); |
1983 | ||
d1ccc66d IM |
1984 | /* We're going to change ->state: */ |
1985 | success = 1; | |
1da177e4 | 1986 | cpu = task_cpu(p); |
1da177e4 | 1987 | |
135e8c92 BS |
1988 | /* |
1989 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
1990 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
1991 | * in smp_cond_load_acquire() below. | |
1992 | * | |
1993 | * sched_ttwu_pending() try_to_wake_up() | |
1994 | * [S] p->on_rq = 1; [L] P->state | |
1995 | * UNLOCK rq->lock -----. | |
1996 | * \ | |
1997 | * +--- RMB | |
1998 | * schedule() / | |
1999 | * LOCK rq->lock -----' | |
2000 | * UNLOCK rq->lock | |
2001 | * | |
2002 | * [task p] | |
2003 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq | |
2004 | * | |
2005 | * Pairs with the UNLOCK+LOCK on rq->lock from the | |
2006 | * last wakeup of our task and the schedule that got our task | |
2007 | * current. | |
2008 | */ | |
2009 | smp_rmb(); | |
c05fbafb PZ |
2010 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2011 | goto stat; | |
1da177e4 | 2012 | |
1da177e4 | 2013 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2014 | /* |
2015 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2016 | * possible to, falsely, observe p->on_cpu == 0. | |
2017 | * | |
2018 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2019 | * from the runqueue. | |
2020 | * | |
2021 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2022 | * UNLOCK rq->lock | |
2023 | * RMB | |
2024 | * LOCK rq->lock | |
2025 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2026 | * | |
2027 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2028 | * from the consecutive calls to schedule(); the first switching to our | |
2029 | * task, the second putting it to sleep. | |
2030 | */ | |
2031 | smp_rmb(); | |
2032 | ||
e9c84311 | 2033 | /* |
d1ccc66d | 2034 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2035 | * this task as prev, wait until its done referencing the task. |
b75a2253 PZ |
2036 | * |
2037 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2038 | * | |
2039 | * This ensures that tasks getting woken will be fully ordered against | |
2040 | * their previous state and preserve Program Order. | |
0970d299 | 2041 | */ |
1f03e8d2 | 2042 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2043 | |
a8e4f2ea | 2044 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2045 | p->state = TASK_WAKING; |
e7693a36 | 2046 | |
e33a9bba TH |
2047 | if (p->in_iowait) { |
2048 | delayacct_blkio_end(); | |
2049 | atomic_dec(&task_rq(p)->nr_iowait); | |
2050 | } | |
2051 | ||
ac66f547 | 2052 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2053 | if (task_cpu(p) != cpu) { |
2054 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2055 | set_task_cpu(p, cpu); |
f339b9dc | 2056 | } |
e33a9bba TH |
2057 | |
2058 | #else /* CONFIG_SMP */ | |
2059 | ||
2060 | if (p->in_iowait) { | |
2061 | delayacct_blkio_end(); | |
2062 | atomic_dec(&task_rq(p)->nr_iowait); | |
2063 | } | |
2064 | ||
1da177e4 | 2065 | #endif /* CONFIG_SMP */ |
1da177e4 | 2066 | |
b5179ac7 | 2067 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2068 | stat: |
4fa8d299 | 2069 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2070 | out: |
013fdb80 | 2071 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2072 | |
2073 | return success; | |
2074 | } | |
2075 | ||
21aa9af0 TH |
2076 | /** |
2077 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2078 | * @p: the thread to be awakened | |
9279e0d2 | 2079 | * @cookie: context's cookie for pinning |
21aa9af0 | 2080 | * |
2acca55e | 2081 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2082 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2083 | * the current task. |
21aa9af0 | 2084 | */ |
d8ac8971 | 2085 | static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf) |
21aa9af0 TH |
2086 | { |
2087 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2088 | |
383efcd0 TH |
2089 | if (WARN_ON_ONCE(rq != this_rq()) || |
2090 | WARN_ON_ONCE(p == current)) | |
2091 | return; | |
2092 | ||
21aa9af0 TH |
2093 | lockdep_assert_held(&rq->lock); |
2094 | ||
2acca55e | 2095 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2096 | /* |
2097 | * This is OK, because current is on_cpu, which avoids it being | |
2098 | * picked for load-balance and preemption/IRQs are still | |
2099 | * disabled avoiding further scheduler activity on it and we've | |
2100 | * not yet picked a replacement task. | |
2101 | */ | |
8a8c69c3 | 2102 | rq_unlock(rq, rf); |
2acca55e | 2103 | raw_spin_lock(&p->pi_lock); |
8a8c69c3 | 2104 | rq_relock(rq, rf); |
2acca55e PZ |
2105 | } |
2106 | ||
21aa9af0 | 2107 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2108 | goto out; |
21aa9af0 | 2109 | |
fbd705a0 PZ |
2110 | trace_sched_waking(p); |
2111 | ||
e33a9bba TH |
2112 | if (!task_on_rq_queued(p)) { |
2113 | if (p->in_iowait) { | |
2114 | delayacct_blkio_end(); | |
2115 | atomic_dec(&rq->nr_iowait); | |
2116 | } | |
d7c01d27 | 2117 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
e33a9bba | 2118 | } |
d7c01d27 | 2119 | |
d8ac8971 | 2120 | ttwu_do_wakeup(rq, p, 0, rf); |
4fa8d299 | 2121 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2122 | out: |
2123 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2124 | } |
2125 | ||
50fa610a DH |
2126 | /** |
2127 | * wake_up_process - Wake up a specific process | |
2128 | * @p: The process to be woken up. | |
2129 | * | |
2130 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2131 | * processes. |
2132 | * | |
2133 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2134 | * |
2135 | * It may be assumed that this function implies a write memory barrier before | |
2136 | * changing the task state if and only if any tasks are woken up. | |
2137 | */ | |
7ad5b3a5 | 2138 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2139 | { |
9067ac85 | 2140 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2141 | } |
1da177e4 LT |
2142 | EXPORT_SYMBOL(wake_up_process); |
2143 | ||
7ad5b3a5 | 2144 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2145 | { |
2146 | return try_to_wake_up(p, state, 0); | |
2147 | } | |
2148 | ||
a5e7be3b JL |
2149 | /* |
2150 | * This function clears the sched_dl_entity static params. | |
2151 | */ | |
2152 | void __dl_clear_params(struct task_struct *p) | |
2153 | { | |
2154 | struct sched_dl_entity *dl_se = &p->dl; | |
2155 | ||
2156 | dl_se->dl_runtime = 0; | |
2157 | dl_se->dl_deadline = 0; | |
2158 | dl_se->dl_period = 0; | |
2159 | dl_se->flags = 0; | |
2160 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2161 | |
2162 | dl_se->dl_throttled = 0; | |
40767b0d | 2163 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2164 | } |
2165 | ||
1da177e4 LT |
2166 | /* |
2167 | * Perform scheduler related setup for a newly forked process p. | |
2168 | * p is forked by current. | |
dd41f596 IM |
2169 | * |
2170 | * __sched_fork() is basic setup used by init_idle() too: | |
2171 | */ | |
5e1576ed | 2172 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2173 | { |
fd2f4419 PZ |
2174 | p->on_rq = 0; |
2175 | ||
2176 | p->se.on_rq = 0; | |
dd41f596 IM |
2177 | p->se.exec_start = 0; |
2178 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2179 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2180 | p->se.nr_migrations = 0; |
da7a735e | 2181 | p->se.vruntime = 0; |
fd2f4419 | 2182 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2183 | |
ad936d86 BP |
2184 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2185 | p->se.cfs_rq = NULL; | |
2186 | #endif | |
2187 | ||
6cfb0d5d | 2188 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2189 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2190 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2191 | #endif |
476d139c | 2192 | |
aab03e05 | 2193 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2194 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2195 | __dl_clear_params(p); |
aab03e05 | 2196 | |
fa717060 | 2197 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2198 | p->rt.timeout = 0; |
2199 | p->rt.time_slice = sched_rr_timeslice; | |
2200 | p->rt.on_rq = 0; | |
2201 | p->rt.on_list = 0; | |
476d139c | 2202 | |
e107be36 AK |
2203 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2204 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2205 | #endif | |
cbee9f88 PZ |
2206 | |
2207 | #ifdef CONFIG_NUMA_BALANCING | |
2208 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2209 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2210 | p->mm->numa_scan_seq = 0; |
2211 | } | |
2212 | ||
5e1576ed RR |
2213 | if (clone_flags & CLONE_VM) |
2214 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2215 | else | |
2216 | p->numa_preferred_nid = -1; | |
2217 | ||
cbee9f88 PZ |
2218 | p->node_stamp = 0ULL; |
2219 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2220 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2221 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2222 | p->numa_faults = NULL; |
7e2703e6 RR |
2223 | p->last_task_numa_placement = 0; |
2224 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2225 | |
8c8a743c | 2226 | p->numa_group = NULL; |
cbee9f88 | 2227 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2228 | } |
2229 | ||
2a595721 SD |
2230 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2231 | ||
1a687c2e | 2232 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2233 | |
1a687c2e MG |
2234 | void set_numabalancing_state(bool enabled) |
2235 | { | |
2236 | if (enabled) | |
2a595721 | 2237 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2238 | else |
2a595721 | 2239 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2240 | } |
54a43d54 AK |
2241 | |
2242 | #ifdef CONFIG_PROC_SYSCTL | |
2243 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2244 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2245 | { | |
2246 | struct ctl_table t; | |
2247 | int err; | |
2a595721 | 2248 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2249 | |
2250 | if (write && !capable(CAP_SYS_ADMIN)) | |
2251 | return -EPERM; | |
2252 | ||
2253 | t = *table; | |
2254 | t.data = &state; | |
2255 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2256 | if (err < 0) | |
2257 | return err; | |
2258 | if (write) | |
2259 | set_numabalancing_state(state); | |
2260 | return err; | |
2261 | } | |
2262 | #endif | |
2263 | #endif | |
dd41f596 | 2264 | |
4698f88c JP |
2265 | #ifdef CONFIG_SCHEDSTATS |
2266 | ||
cb251765 | 2267 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2268 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2269 | |
cb251765 MG |
2270 | static void set_schedstats(bool enabled) |
2271 | { | |
2272 | if (enabled) | |
2273 | static_branch_enable(&sched_schedstats); | |
2274 | else | |
2275 | static_branch_disable(&sched_schedstats); | |
2276 | } | |
2277 | ||
2278 | void force_schedstat_enabled(void) | |
2279 | { | |
2280 | if (!schedstat_enabled()) { | |
2281 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2282 | static_branch_enable(&sched_schedstats); | |
2283 | } | |
2284 | } | |
2285 | ||
2286 | static int __init setup_schedstats(char *str) | |
2287 | { | |
2288 | int ret = 0; | |
2289 | if (!str) | |
2290 | goto out; | |
2291 | ||
4698f88c JP |
2292 | /* |
2293 | * This code is called before jump labels have been set up, so we can't | |
2294 | * change the static branch directly just yet. Instead set a temporary | |
2295 | * variable so init_schedstats() can do it later. | |
2296 | */ | |
cb251765 | 2297 | if (!strcmp(str, "enable")) { |
4698f88c | 2298 | __sched_schedstats = true; |
cb251765 MG |
2299 | ret = 1; |
2300 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2301 | __sched_schedstats = false; |
cb251765 MG |
2302 | ret = 1; |
2303 | } | |
2304 | out: | |
2305 | if (!ret) | |
2306 | pr_warn("Unable to parse schedstats=\n"); | |
2307 | ||
2308 | return ret; | |
2309 | } | |
2310 | __setup("schedstats=", setup_schedstats); | |
2311 | ||
4698f88c JP |
2312 | static void __init init_schedstats(void) |
2313 | { | |
2314 | set_schedstats(__sched_schedstats); | |
2315 | } | |
2316 | ||
cb251765 MG |
2317 | #ifdef CONFIG_PROC_SYSCTL |
2318 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2319 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2320 | { | |
2321 | struct ctl_table t; | |
2322 | int err; | |
2323 | int state = static_branch_likely(&sched_schedstats); | |
2324 | ||
2325 | if (write && !capable(CAP_SYS_ADMIN)) | |
2326 | return -EPERM; | |
2327 | ||
2328 | t = *table; | |
2329 | t.data = &state; | |
2330 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2331 | if (err < 0) | |
2332 | return err; | |
2333 | if (write) | |
2334 | set_schedstats(state); | |
2335 | return err; | |
2336 | } | |
4698f88c JP |
2337 | #endif /* CONFIG_PROC_SYSCTL */ |
2338 | #else /* !CONFIG_SCHEDSTATS */ | |
2339 | static inline void init_schedstats(void) {} | |
2340 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2341 | |
2342 | /* | |
2343 | * fork()/clone()-time setup: | |
2344 | */ | |
aab03e05 | 2345 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2346 | { |
0122ec5b | 2347 | unsigned long flags; |
dd41f596 IM |
2348 | int cpu = get_cpu(); |
2349 | ||
5e1576ed | 2350 | __sched_fork(clone_flags, p); |
06b83b5f | 2351 | /* |
7dc603c9 | 2352 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2353 | * nobody will actually run it, and a signal or other external |
2354 | * event cannot wake it up and insert it on the runqueue either. | |
2355 | */ | |
7dc603c9 | 2356 | p->state = TASK_NEW; |
dd41f596 | 2357 | |
c350a04e MG |
2358 | /* |
2359 | * Make sure we do not leak PI boosting priority to the child. | |
2360 | */ | |
2361 | p->prio = current->normal_prio; | |
2362 | ||
b9dc29e7 MG |
2363 | /* |
2364 | * Revert to default priority/policy on fork if requested. | |
2365 | */ | |
2366 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2367 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2368 | p->policy = SCHED_NORMAL; |
6c697bdf | 2369 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2370 | p->rt_priority = 0; |
2371 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2372 | p->static_prio = NICE_TO_PRIO(0); | |
2373 | ||
2374 | p->prio = p->normal_prio = __normal_prio(p); | |
2375 | set_load_weight(p); | |
6c697bdf | 2376 | |
b9dc29e7 MG |
2377 | /* |
2378 | * We don't need the reset flag anymore after the fork. It has | |
2379 | * fulfilled its duty: | |
2380 | */ | |
2381 | p->sched_reset_on_fork = 0; | |
2382 | } | |
ca94c442 | 2383 | |
aab03e05 DF |
2384 | if (dl_prio(p->prio)) { |
2385 | put_cpu(); | |
2386 | return -EAGAIN; | |
2387 | } else if (rt_prio(p->prio)) { | |
2388 | p->sched_class = &rt_sched_class; | |
2389 | } else { | |
2ddbf952 | 2390 | p->sched_class = &fair_sched_class; |
aab03e05 | 2391 | } |
b29739f9 | 2392 | |
7dc603c9 | 2393 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2394 | |
86951599 PZ |
2395 | /* |
2396 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2397 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2398 | * is ran before sched_fork(). | |
2399 | * | |
2400 | * Silence PROVE_RCU. | |
2401 | */ | |
0122ec5b | 2402 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd | 2403 | /* |
d1ccc66d | 2404 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
2405 | * so use __set_task_cpu(). |
2406 | */ | |
2407 | __set_task_cpu(p, cpu); | |
2408 | if (p->sched_class->task_fork) | |
2409 | p->sched_class->task_fork(p); | |
0122ec5b | 2410 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2411 | |
f6db8347 | 2412 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2413 | if (likely(sched_info_on())) |
52f17b6c | 2414 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2415 | #endif |
3ca7a440 PZ |
2416 | #if defined(CONFIG_SMP) |
2417 | p->on_cpu = 0; | |
4866cde0 | 2418 | #endif |
01028747 | 2419 | init_task_preempt_count(p); |
806c09a7 | 2420 | #ifdef CONFIG_SMP |
917b627d | 2421 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2422 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2423 | #endif |
917b627d | 2424 | |
476d139c | 2425 | put_cpu(); |
aab03e05 | 2426 | return 0; |
1da177e4 LT |
2427 | } |
2428 | ||
332ac17e DF |
2429 | unsigned long to_ratio(u64 period, u64 runtime) |
2430 | { | |
2431 | if (runtime == RUNTIME_INF) | |
2432 | return 1ULL << 20; | |
2433 | ||
2434 | /* | |
2435 | * Doing this here saves a lot of checks in all | |
2436 | * the calling paths, and returning zero seems | |
2437 | * safe for them anyway. | |
2438 | */ | |
2439 | if (period == 0) | |
2440 | return 0; | |
2441 | ||
2442 | return div64_u64(runtime << 20, period); | |
2443 | } | |
2444 | ||
2445 | #ifdef CONFIG_SMP | |
2446 | inline struct dl_bw *dl_bw_of(int i) | |
2447 | { | |
f78f5b90 PM |
2448 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2449 | "sched RCU must be held"); | |
332ac17e DF |
2450 | return &cpu_rq(i)->rd->dl_bw; |
2451 | } | |
2452 | ||
de212f18 | 2453 | static inline int dl_bw_cpus(int i) |
332ac17e | 2454 | { |
de212f18 PZ |
2455 | struct root_domain *rd = cpu_rq(i)->rd; |
2456 | int cpus = 0; | |
2457 | ||
f78f5b90 PM |
2458 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2459 | "sched RCU must be held"); | |
de212f18 PZ |
2460 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2461 | cpus++; | |
2462 | ||
2463 | return cpus; | |
332ac17e DF |
2464 | } |
2465 | #else | |
2466 | inline struct dl_bw *dl_bw_of(int i) | |
2467 | { | |
2468 | return &cpu_rq(i)->dl.dl_bw; | |
2469 | } | |
2470 | ||
de212f18 | 2471 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2472 | { |
2473 | return 1; | |
2474 | } | |
2475 | #endif | |
2476 | ||
332ac17e DF |
2477 | /* |
2478 | * We must be sure that accepting a new task (or allowing changing the | |
2479 | * parameters of an existing one) is consistent with the bandwidth | |
2480 | * constraints. If yes, this function also accordingly updates the currently | |
2481 | * allocated bandwidth to reflect the new situation. | |
2482 | * | |
2483 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2484 | * |
2485 | * XXX we should delay bw change until the task's 0-lag point, see | |
2486 | * __setparam_dl(). | |
332ac17e DF |
2487 | */ |
2488 | static int dl_overflow(struct task_struct *p, int policy, | |
2489 | const struct sched_attr *attr) | |
2490 | { | |
2491 | ||
2492 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2493 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2494 | u64 runtime = attr->sched_runtime; |
2495 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2496 | int cpus, err = -1; |
332ac17e | 2497 | |
fec148c0 XP |
2498 | /* !deadline task may carry old deadline bandwidth */ |
2499 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2500 | return 0; |
2501 | ||
2502 | /* | |
2503 | * Either if a task, enters, leave, or stays -deadline but changes | |
2504 | * its parameters, we may need to update accordingly the total | |
2505 | * allocated bandwidth of the container. | |
2506 | */ | |
2507 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2508 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2509 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2510 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2511 | __dl_add(dl_b, new_bw); | |
2512 | err = 0; | |
2513 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2514 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2515 | __dl_clear(dl_b, p->dl.dl_bw); | |
2516 | __dl_add(dl_b, new_bw); | |
2517 | err = 0; | |
2518 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2519 | __dl_clear(dl_b, p->dl.dl_bw); | |
2520 | err = 0; | |
2521 | } | |
2522 | raw_spin_unlock(&dl_b->lock); | |
2523 | ||
2524 | return err; | |
2525 | } | |
2526 | ||
2527 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2528 | ||
1da177e4 LT |
2529 | /* |
2530 | * wake_up_new_task - wake up a newly created task for the first time. | |
2531 | * | |
2532 | * This function will do some initial scheduler statistics housekeeping | |
2533 | * that must be done for every newly created context, then puts the task | |
2534 | * on the runqueue and wakes it. | |
2535 | */ | |
3e51e3ed | 2536 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2537 | { |
eb580751 | 2538 | struct rq_flags rf; |
dd41f596 | 2539 | struct rq *rq; |
fabf318e | 2540 | |
eb580751 | 2541 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2542 | p->state = TASK_RUNNING; |
fabf318e PZ |
2543 | #ifdef CONFIG_SMP |
2544 | /* | |
2545 | * Fork balancing, do it here and not earlier because: | |
2546 | * - cpus_allowed can change in the fork path | |
d1ccc66d | 2547 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
2548 | * |
2549 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2550 | * as we're not fully set-up yet. | |
fabf318e | 2551 | */ |
e210bffd | 2552 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2553 | #endif |
b7fa30c9 | 2554 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 2555 | update_rq_clock(rq); |
2b8c41da | 2556 | post_init_entity_util_avg(&p->se); |
0017d735 | 2557 | |
cd29fe6f | 2558 | activate_task(rq, p, 0); |
da0c1e65 | 2559 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2560 | trace_sched_wakeup_new(p); |
a7558e01 | 2561 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2562 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2563 | if (p->sched_class->task_woken) { |
2564 | /* | |
2565 | * Nothing relies on rq->lock after this, so its fine to | |
2566 | * drop it. | |
2567 | */ | |
d8ac8971 | 2568 | rq_unpin_lock(rq, &rf); |
efbbd05a | 2569 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2570 | rq_repin_lock(rq, &rf); |
0aaafaab | 2571 | } |
9a897c5a | 2572 | #endif |
eb580751 | 2573 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2574 | } |
2575 | ||
e107be36 AK |
2576 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2577 | ||
1cde2930 PZ |
2578 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2579 | ||
2ecd9d29 PZ |
2580 | void preempt_notifier_inc(void) |
2581 | { | |
2582 | static_key_slow_inc(&preempt_notifier_key); | |
2583 | } | |
2584 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2585 | ||
2586 | void preempt_notifier_dec(void) | |
2587 | { | |
2588 | static_key_slow_dec(&preempt_notifier_key); | |
2589 | } | |
2590 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2591 | ||
e107be36 | 2592 | /** |
80dd99b3 | 2593 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2594 | * @notifier: notifier struct to register |
e107be36 AK |
2595 | */ |
2596 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2597 | { | |
2ecd9d29 PZ |
2598 | if (!static_key_false(&preempt_notifier_key)) |
2599 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2600 | ||
e107be36 AK |
2601 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2602 | } | |
2603 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2604 | ||
2605 | /** | |
2606 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2607 | * @notifier: notifier struct to unregister |
e107be36 | 2608 | * |
d84525a8 | 2609 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2610 | */ |
2611 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2612 | { | |
2613 | hlist_del(¬ifier->link); | |
2614 | } | |
2615 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2616 | ||
1cde2930 | 2617 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2618 | { |
2619 | struct preempt_notifier *notifier; | |
e107be36 | 2620 | |
b67bfe0d | 2621 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2622 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2623 | } | |
2624 | ||
1cde2930 PZ |
2625 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2626 | { | |
2627 | if (static_key_false(&preempt_notifier_key)) | |
2628 | __fire_sched_in_preempt_notifiers(curr); | |
2629 | } | |
2630 | ||
e107be36 | 2631 | static void |
1cde2930 PZ |
2632 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2633 | struct task_struct *next) | |
e107be36 AK |
2634 | { |
2635 | struct preempt_notifier *notifier; | |
e107be36 | 2636 | |
b67bfe0d | 2637 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2638 | notifier->ops->sched_out(notifier, next); |
2639 | } | |
2640 | ||
1cde2930 PZ |
2641 | static __always_inline void |
2642 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2643 | struct task_struct *next) | |
2644 | { | |
2645 | if (static_key_false(&preempt_notifier_key)) | |
2646 | __fire_sched_out_preempt_notifiers(curr, next); | |
2647 | } | |
2648 | ||
6d6bc0ad | 2649 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2650 | |
1cde2930 | 2651 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2652 | { |
2653 | } | |
2654 | ||
1cde2930 | 2655 | static inline void |
e107be36 AK |
2656 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2657 | struct task_struct *next) | |
2658 | { | |
2659 | } | |
2660 | ||
6d6bc0ad | 2661 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2662 | |
4866cde0 NP |
2663 | /** |
2664 | * prepare_task_switch - prepare to switch tasks | |
2665 | * @rq: the runqueue preparing to switch | |
421cee29 | 2666 | * @prev: the current task that is being switched out |
4866cde0 NP |
2667 | * @next: the task we are going to switch to. |
2668 | * | |
2669 | * This is called with the rq lock held and interrupts off. It must | |
2670 | * be paired with a subsequent finish_task_switch after the context | |
2671 | * switch. | |
2672 | * | |
2673 | * prepare_task_switch sets up locking and calls architecture specific | |
2674 | * hooks. | |
2675 | */ | |
e107be36 AK |
2676 | static inline void |
2677 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2678 | struct task_struct *next) | |
4866cde0 | 2679 | { |
43148951 | 2680 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2681 | perf_event_task_sched_out(prev, next); |
e107be36 | 2682 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2683 | prepare_lock_switch(rq, next); |
2684 | prepare_arch_switch(next); | |
2685 | } | |
2686 | ||
1da177e4 LT |
2687 | /** |
2688 | * finish_task_switch - clean up after a task-switch | |
2689 | * @prev: the thread we just switched away from. | |
2690 | * | |
4866cde0 NP |
2691 | * finish_task_switch must be called after the context switch, paired |
2692 | * with a prepare_task_switch call before the context switch. | |
2693 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2694 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2695 | * |
2696 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2697 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2698 | * with the lock held can cause deadlocks; see schedule() for |
2699 | * details.) | |
dfa50b60 ON |
2700 | * |
2701 | * The context switch have flipped the stack from under us and restored the | |
2702 | * local variables which were saved when this task called schedule() in the | |
2703 | * past. prev == current is still correct but we need to recalculate this_rq | |
2704 | * because prev may have moved to another CPU. | |
1da177e4 | 2705 | */ |
dfa50b60 | 2706 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2707 | __releases(rq->lock) |
2708 | { | |
dfa50b60 | 2709 | struct rq *rq = this_rq(); |
1da177e4 | 2710 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2711 | long prev_state; |
1da177e4 | 2712 | |
609ca066 PZ |
2713 | /* |
2714 | * The previous task will have left us with a preempt_count of 2 | |
2715 | * because it left us after: | |
2716 | * | |
2717 | * schedule() | |
2718 | * preempt_disable(); // 1 | |
2719 | * __schedule() | |
2720 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2721 | * | |
2722 | * Also, see FORK_PREEMPT_COUNT. | |
2723 | */ | |
e2bf1c4b PZ |
2724 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2725 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2726 | current->comm, current->pid, preempt_count())) | |
2727 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2728 | |
1da177e4 LT |
2729 | rq->prev_mm = NULL; |
2730 | ||
2731 | /* | |
2732 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2733 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2734 | * schedule one last time. The schedule call will never return, and |
2735 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2736 | * |
2737 | * We must observe prev->state before clearing prev->on_cpu (in | |
2738 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2739 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2740 | * transition, resulting in a double drop. | |
1da177e4 | 2741 | */ |
55a101f8 | 2742 | prev_state = prev->state; |
bf9fae9f | 2743 | vtime_task_switch(prev); |
a8d757ef | 2744 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2745 | finish_lock_switch(rq, prev); |
01f23e16 | 2746 | finish_arch_post_lock_switch(); |
e8fa1362 | 2747 | |
e107be36 | 2748 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2749 | if (mm) |
2750 | mmdrop(mm); | |
c394cc9f | 2751 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2752 | if (prev->sched_class->task_dead) |
2753 | prev->sched_class->task_dead(prev); | |
2754 | ||
c6fd91f0 | 2755 | /* |
2756 | * Remove function-return probe instances associated with this | |
2757 | * task and put them back on the free list. | |
9761eea8 | 2758 | */ |
c6fd91f0 | 2759 | kprobe_flush_task(prev); |
68f24b08 AL |
2760 | |
2761 | /* Task is done with its stack. */ | |
2762 | put_task_stack(prev); | |
2763 | ||
1da177e4 | 2764 | put_task_struct(prev); |
c6fd91f0 | 2765 | } |
99e5ada9 | 2766 | |
de734f89 | 2767 | tick_nohz_task_switch(); |
dfa50b60 | 2768 | return rq; |
1da177e4 LT |
2769 | } |
2770 | ||
3f029d3c GH |
2771 | #ifdef CONFIG_SMP |
2772 | ||
3f029d3c | 2773 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2774 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2775 | { |
e3fca9e7 PZ |
2776 | struct callback_head *head, *next; |
2777 | void (*func)(struct rq *rq); | |
8a8c69c3 | 2778 | struct rq_flags rf; |
3f029d3c | 2779 | |
8a8c69c3 | 2780 | rq_lock_irqsave(rq, &rf); |
e3fca9e7 PZ |
2781 | head = rq->balance_callback; |
2782 | rq->balance_callback = NULL; | |
2783 | while (head) { | |
2784 | func = (void (*)(struct rq *))head->func; | |
2785 | next = head->next; | |
2786 | head->next = NULL; | |
2787 | head = next; | |
3f029d3c | 2788 | |
e3fca9e7 | 2789 | func(rq); |
3f029d3c | 2790 | } |
8a8c69c3 | 2791 | rq_unlock_irqrestore(rq, &rf); |
e3fca9e7 PZ |
2792 | } |
2793 | ||
2794 | static inline void balance_callback(struct rq *rq) | |
2795 | { | |
2796 | if (unlikely(rq->balance_callback)) | |
2797 | __balance_callback(rq); | |
3f029d3c GH |
2798 | } |
2799 | ||
2800 | #else | |
da19ab51 | 2801 | |
e3fca9e7 | 2802 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2803 | { |
1da177e4 LT |
2804 | } |
2805 | ||
3f029d3c GH |
2806 | #endif |
2807 | ||
1da177e4 LT |
2808 | /** |
2809 | * schedule_tail - first thing a freshly forked thread must call. | |
2810 | * @prev: the thread we just switched away from. | |
2811 | */ | |
722a9f92 | 2812 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2813 | __releases(rq->lock) |
2814 | { | |
1a43a14a | 2815 | struct rq *rq; |
da19ab51 | 2816 | |
609ca066 PZ |
2817 | /* |
2818 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2819 | * finish_task_switch() for details. | |
2820 | * | |
2821 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2822 | * and the preempt_enable() will end up enabling preemption (on | |
2823 | * PREEMPT_COUNT kernels). | |
2824 | */ | |
2825 | ||
dfa50b60 | 2826 | rq = finish_task_switch(prev); |
e3fca9e7 | 2827 | balance_callback(rq); |
1a43a14a | 2828 | preempt_enable(); |
70b97a7f | 2829 | |
1da177e4 | 2830 | if (current->set_child_tid) |
b488893a | 2831 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2832 | } |
2833 | ||
2834 | /* | |
dfa50b60 | 2835 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2836 | */ |
04936948 | 2837 | static __always_inline struct rq * |
70b97a7f | 2838 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 2839 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 2840 | { |
dd41f596 | 2841 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2842 | |
e107be36 | 2843 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2844 | |
dd41f596 IM |
2845 | mm = next->mm; |
2846 | oldmm = prev->active_mm; | |
9226d125 ZA |
2847 | /* |
2848 | * For paravirt, this is coupled with an exit in switch_to to | |
2849 | * combine the page table reload and the switch backend into | |
2850 | * one hypercall. | |
2851 | */ | |
224101ed | 2852 | arch_start_context_switch(prev); |
9226d125 | 2853 | |
31915ab4 | 2854 | if (!mm) { |
1da177e4 | 2855 | next->active_mm = oldmm; |
f1f10076 | 2856 | mmgrab(oldmm); |
1da177e4 LT |
2857 | enter_lazy_tlb(oldmm, next); |
2858 | } else | |
f98db601 | 2859 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2860 | |
31915ab4 | 2861 | if (!prev->mm) { |
1da177e4 | 2862 | prev->active_mm = NULL; |
1da177e4 LT |
2863 | rq->prev_mm = oldmm; |
2864 | } | |
92509b73 | 2865 | |
cb42c9a3 | 2866 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 2867 | |
3a5f5e48 IM |
2868 | /* |
2869 | * Since the runqueue lock will be released by the next | |
2870 | * task (which is an invalid locking op but in the case | |
2871 | * of the scheduler it's an obvious special-case), so we | |
2872 | * do an early lockdep release here: | |
2873 | */ | |
d8ac8971 | 2874 | rq_unpin_lock(rq, rf); |
8a25d5de | 2875 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2876 | |
2877 | /* Here we just switch the register state and the stack. */ | |
2878 | switch_to(prev, next, prev); | |
dd41f596 | 2879 | barrier(); |
dfa50b60 ON |
2880 | |
2881 | return finish_task_switch(prev); | |
1da177e4 LT |
2882 | } |
2883 | ||
2884 | /* | |
1c3e8264 | 2885 | * nr_running and nr_context_switches: |
1da177e4 LT |
2886 | * |
2887 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2888 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2889 | */ |
2890 | unsigned long nr_running(void) | |
2891 | { | |
2892 | unsigned long i, sum = 0; | |
2893 | ||
2894 | for_each_online_cpu(i) | |
2895 | sum += cpu_rq(i)->nr_running; | |
2896 | ||
2897 | return sum; | |
f711f609 | 2898 | } |
1da177e4 | 2899 | |
2ee507c4 | 2900 | /* |
d1ccc66d | 2901 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
2902 | * |
2903 | * Caution: this function does not check that the caller has disabled | |
2904 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2905 | * race. The caller is responsible to use it correctly, for example: | |
2906 | * | |
2907 | * - from a non-preemptable section (of course) | |
2908 | * | |
2909 | * - from a thread that is bound to a single CPU | |
2910 | * | |
2911 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2912 | */ |
2913 | bool single_task_running(void) | |
2914 | { | |
00cc1633 | 2915 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2916 | } |
2917 | EXPORT_SYMBOL(single_task_running); | |
2918 | ||
1da177e4 | 2919 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2920 | { |
cc94abfc SR |
2921 | int i; |
2922 | unsigned long long sum = 0; | |
46cb4b7c | 2923 | |
0a945022 | 2924 | for_each_possible_cpu(i) |
1da177e4 | 2925 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2926 | |
1da177e4 LT |
2927 | return sum; |
2928 | } | |
483b4ee6 | 2929 | |
e33a9bba TH |
2930 | /* |
2931 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
2932 | * | |
2933 | * The idea behind IO-wait account is to account the idle time that we could | |
2934 | * have spend running if it were not for IO. That is, if we were to improve the | |
2935 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
2936 | * | |
2937 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
2938 | * idle time as IO-wait, because if the storage were faster, it could've been | |
2939 | * running and we'd not be idle. | |
2940 | * | |
2941 | * This has been extended to SMP, by doing the same for each CPU. This however | |
2942 | * is broken. | |
2943 | * | |
2944 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
2945 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
2946 | * though, if the storage were faster, both could've ran at the same time, | |
2947 | * utilising both CPUs. | |
2948 | * | |
2949 | * This means, that when looking globally, the current IO-wait accounting on | |
2950 | * SMP is a lower bound, by reason of under accounting. | |
2951 | * | |
2952 | * Worse, since the numbers are provided per CPU, they are sometimes | |
2953 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
2954 | * associated with any one particular CPU, it can wake to another CPU than it | |
2955 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
2956 | * | |
2957 | * Task CPU affinities can make all that even more 'interesting'. | |
2958 | */ | |
2959 | ||
1da177e4 LT |
2960 | unsigned long nr_iowait(void) |
2961 | { | |
2962 | unsigned long i, sum = 0; | |
483b4ee6 | 2963 | |
0a945022 | 2964 | for_each_possible_cpu(i) |
1da177e4 | 2965 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2966 | |
1da177e4 LT |
2967 | return sum; |
2968 | } | |
483b4ee6 | 2969 | |
e33a9bba TH |
2970 | /* |
2971 | * Consumers of these two interfaces, like for example the cpufreq menu | |
2972 | * governor are using nonsensical data. Boosting frequency for a CPU that has | |
2973 | * IO-wait which might not even end up running the task when it does become | |
2974 | * runnable. | |
2975 | */ | |
2976 | ||
8c215bd3 | 2977 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2978 | { |
8c215bd3 | 2979 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2980 | return atomic_read(&this->nr_iowait); |
2981 | } | |
46cb4b7c | 2982 | |
372ba8cb MG |
2983 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2984 | { | |
3289bdb4 PZ |
2985 | struct rq *rq = this_rq(); |
2986 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2987 | *load = rq->load.weight; | |
372ba8cb MG |
2988 | } |
2989 | ||
dd41f596 | 2990 | #ifdef CONFIG_SMP |
8a0be9ef | 2991 | |
46cb4b7c | 2992 | /* |
38022906 PZ |
2993 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2994 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2995 | */ |
38022906 | 2996 | void sched_exec(void) |
46cb4b7c | 2997 | { |
38022906 | 2998 | struct task_struct *p = current; |
1da177e4 | 2999 | unsigned long flags; |
0017d735 | 3000 | int dest_cpu; |
46cb4b7c | 3001 | |
8f42ced9 | 3002 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 3003 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3004 | if (dest_cpu == smp_processor_id()) |
3005 | goto unlock; | |
38022906 | 3006 | |
8f42ced9 | 3007 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3008 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3009 | |
8f42ced9 PZ |
3010 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3011 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3012 | return; |
3013 | } | |
0017d735 | 3014 | unlock: |
8f42ced9 | 3015 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3016 | } |
dd41f596 | 3017 | |
1da177e4 LT |
3018 | #endif |
3019 | ||
1da177e4 | 3020 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3021 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3022 | |
3023 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3024 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3025 | |
6075620b GG |
3026 | /* |
3027 | * The function fair_sched_class.update_curr accesses the struct curr | |
3028 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3029 | * we observe a high rate of cache misses in practice. | |
3030 | * Prefetching this data results in improved performance. | |
3031 | */ | |
3032 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3033 | { | |
3034 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3035 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3036 | #else | |
3037 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3038 | #endif | |
3039 | prefetch(curr); | |
3040 | prefetch(&curr->exec_start); | |
3041 | } | |
3042 | ||
c5f8d995 HS |
3043 | /* |
3044 | * Return accounted runtime for the task. | |
3045 | * In case the task is currently running, return the runtime plus current's | |
3046 | * pending runtime that have not been accounted yet. | |
3047 | */ | |
3048 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3049 | { | |
eb580751 | 3050 | struct rq_flags rf; |
c5f8d995 | 3051 | struct rq *rq; |
6e998916 | 3052 | u64 ns; |
c5f8d995 | 3053 | |
911b2898 PZ |
3054 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3055 | /* | |
3056 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
3057 | * So we have a optimization chance when the task's delta_exec is 0. | |
3058 | * Reading ->on_cpu is racy, but this is ok. | |
3059 | * | |
d1ccc66d IM |
3060 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3061 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3062 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3063 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3064 | * been accounted, so we're correct here as well. | |
911b2898 | 3065 | */ |
da0c1e65 | 3066 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3067 | return p->se.sum_exec_runtime; |
3068 | #endif | |
3069 | ||
eb580751 | 3070 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3071 | /* |
3072 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3073 | * project cycles that may never be accounted to this | |
3074 | * thread, breaking clock_gettime(). | |
3075 | */ | |
3076 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3077 | prefetch_curr_exec_start(p); |
6e998916 SG |
3078 | update_rq_clock(rq); |
3079 | p->sched_class->update_curr(rq); | |
3080 | } | |
3081 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3082 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3083 | |
3084 | return ns; | |
3085 | } | |
48f24c4d | 3086 | |
7835b98b CL |
3087 | /* |
3088 | * This function gets called by the timer code, with HZ frequency. | |
3089 | * We call it with interrupts disabled. | |
7835b98b CL |
3090 | */ |
3091 | void scheduler_tick(void) | |
3092 | { | |
7835b98b CL |
3093 | int cpu = smp_processor_id(); |
3094 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3095 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 3096 | struct rq_flags rf; |
3e51f33f PZ |
3097 | |
3098 | sched_clock_tick(); | |
dd41f596 | 3099 | |
8a8c69c3 PZ |
3100 | rq_lock(rq, &rf); |
3101 | ||
3e51f33f | 3102 | update_rq_clock(rq); |
fa85ae24 | 3103 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3104 | cpu_load_update_active(rq); |
3289bdb4 | 3105 | calc_global_load_tick(rq); |
8a8c69c3 PZ |
3106 | |
3107 | rq_unlock(rq, &rf); | |
7835b98b | 3108 | |
e9d2b064 | 3109 | perf_event_task_tick(); |
e220d2dc | 3110 | |
e418e1c2 | 3111 | #ifdef CONFIG_SMP |
6eb57e0d | 3112 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3113 | trigger_load_balance(rq); |
e418e1c2 | 3114 | #endif |
265f22a9 | 3115 | rq_last_tick_reset(rq); |
1da177e4 LT |
3116 | } |
3117 | ||
265f22a9 FW |
3118 | #ifdef CONFIG_NO_HZ_FULL |
3119 | /** | |
3120 | * scheduler_tick_max_deferment | |
3121 | * | |
3122 | * Keep at least one tick per second when a single | |
3123 | * active task is running because the scheduler doesn't | |
3124 | * yet completely support full dynticks environment. | |
3125 | * | |
3126 | * This makes sure that uptime, CFS vruntime, load | |
3127 | * balancing, etc... continue to move forward, even | |
3128 | * with a very low granularity. | |
e69f6186 YB |
3129 | * |
3130 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3131 | */ |
3132 | u64 scheduler_tick_max_deferment(void) | |
3133 | { | |
3134 | struct rq *rq = this_rq(); | |
316c1608 | 3135 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3136 | |
3137 | next = rq->last_sched_tick + HZ; | |
3138 | ||
3139 | if (time_before_eq(next, now)) | |
3140 | return 0; | |
3141 | ||
8fe8ff09 | 3142 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3143 | } |
265f22a9 | 3144 | #endif |
1da177e4 | 3145 | |
7e49fcce SR |
3146 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3147 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3148 | /* |
3149 | * If the value passed in is equal to the current preempt count | |
3150 | * then we just disabled preemption. Start timing the latency. | |
3151 | */ | |
3152 | static inline void preempt_latency_start(int val) | |
3153 | { | |
3154 | if (preempt_count() == val) { | |
3155 | unsigned long ip = get_lock_parent_ip(); | |
3156 | #ifdef CONFIG_DEBUG_PREEMPT | |
3157 | current->preempt_disable_ip = ip; | |
3158 | #endif | |
3159 | trace_preempt_off(CALLER_ADDR0, ip); | |
3160 | } | |
3161 | } | |
7e49fcce | 3162 | |
edafe3a5 | 3163 | void preempt_count_add(int val) |
1da177e4 | 3164 | { |
6cd8a4bb | 3165 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3166 | /* |
3167 | * Underflow? | |
3168 | */ | |
9a11b49a IM |
3169 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3170 | return; | |
6cd8a4bb | 3171 | #endif |
bdb43806 | 3172 | __preempt_count_add(val); |
6cd8a4bb | 3173 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3174 | /* |
3175 | * Spinlock count overflowing soon? | |
3176 | */ | |
33859f7f MOS |
3177 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3178 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3179 | #endif |
47252cfb | 3180 | preempt_latency_start(val); |
1da177e4 | 3181 | } |
bdb43806 | 3182 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3183 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3184 | |
47252cfb SR |
3185 | /* |
3186 | * If the value passed in equals to the current preempt count | |
3187 | * then we just enabled preemption. Stop timing the latency. | |
3188 | */ | |
3189 | static inline void preempt_latency_stop(int val) | |
3190 | { | |
3191 | if (preempt_count() == val) | |
3192 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3193 | } | |
3194 | ||
edafe3a5 | 3195 | void preempt_count_sub(int val) |
1da177e4 | 3196 | { |
6cd8a4bb | 3197 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3198 | /* |
3199 | * Underflow? | |
3200 | */ | |
01e3eb82 | 3201 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3202 | return; |
1da177e4 LT |
3203 | /* |
3204 | * Is the spinlock portion underflowing? | |
3205 | */ | |
9a11b49a IM |
3206 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3207 | !(preempt_count() & PREEMPT_MASK))) | |
3208 | return; | |
6cd8a4bb | 3209 | #endif |
9a11b49a | 3210 | |
47252cfb | 3211 | preempt_latency_stop(val); |
bdb43806 | 3212 | __preempt_count_sub(val); |
1da177e4 | 3213 | } |
bdb43806 | 3214 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3215 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3216 | |
47252cfb SR |
3217 | #else |
3218 | static inline void preempt_latency_start(int val) { } | |
3219 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3220 | #endif |
3221 | ||
59ddbcb2 IM |
3222 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
3223 | { | |
3224 | #ifdef CONFIG_DEBUG_PREEMPT | |
3225 | return p->preempt_disable_ip; | |
3226 | #else | |
3227 | return 0; | |
3228 | #endif | |
3229 | } | |
3230 | ||
1da177e4 | 3231 | /* |
dd41f596 | 3232 | * Print scheduling while atomic bug: |
1da177e4 | 3233 | */ |
dd41f596 | 3234 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3235 | { |
d1c6d149 VN |
3236 | /* Save this before calling printk(), since that will clobber it */ |
3237 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3238 | ||
664dfa65 DJ |
3239 | if (oops_in_progress) |
3240 | return; | |
3241 | ||
3df0fc5b PZ |
3242 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3243 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3244 | |
dd41f596 | 3245 | debug_show_held_locks(prev); |
e21f5b15 | 3246 | print_modules(); |
dd41f596 IM |
3247 | if (irqs_disabled()) |
3248 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3249 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3250 | && in_atomic_preempt_off()) { | |
8f47b187 | 3251 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3252 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3253 | pr_cont("\n"); |
3254 | } | |
748c7201 DBO |
3255 | if (panic_on_warn) |
3256 | panic("scheduling while atomic\n"); | |
3257 | ||
6135fc1e | 3258 | dump_stack(); |
373d4d09 | 3259 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3260 | } |
1da177e4 | 3261 | |
dd41f596 IM |
3262 | /* |
3263 | * Various schedule()-time debugging checks and statistics: | |
3264 | */ | |
3265 | static inline void schedule_debug(struct task_struct *prev) | |
3266 | { | |
0d9e2632 | 3267 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3268 | if (task_stack_end_corrupted(prev)) |
3269 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3270 | #endif |
b99def8b | 3271 | |
1dc0fffc | 3272 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3273 | __schedule_bug(prev); |
1dc0fffc PZ |
3274 | preempt_count_set(PREEMPT_DISABLED); |
3275 | } | |
b3fbab05 | 3276 | rcu_sleep_check(); |
dd41f596 | 3277 | |
1da177e4 LT |
3278 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3279 | ||
ae92882e | 3280 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3281 | } |
3282 | ||
3283 | /* | |
3284 | * Pick up the highest-prio task: | |
3285 | */ | |
3286 | static inline struct task_struct * | |
d8ac8971 | 3287 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3288 | { |
49ee5768 | 3289 | const struct sched_class *class; |
dd41f596 | 3290 | struct task_struct *p; |
1da177e4 LT |
3291 | |
3292 | /* | |
0ba87bb2 PZ |
3293 | * Optimization: we know that if all tasks are in the fair class we can |
3294 | * call that function directly, but only if the @prev task wasn't of a | |
3295 | * higher scheduling class, because otherwise those loose the | |
3296 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 3297 | */ |
0ba87bb2 PZ |
3298 | if (likely((prev->sched_class == &idle_sched_class || |
3299 | prev->sched_class == &fair_sched_class) && | |
3300 | rq->nr_running == rq->cfs.h_nr_running)) { | |
3301 | ||
d8ac8971 | 3302 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3303 | if (unlikely(p == RETRY_TASK)) |
3304 | goto again; | |
3305 | ||
d1ccc66d | 3306 | /* Assumes fair_sched_class->next == idle_sched_class */ |
6ccdc84b | 3307 | if (unlikely(!p)) |
d8ac8971 | 3308 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3309 | |
3310 | return p; | |
1da177e4 LT |
3311 | } |
3312 | ||
37e117c0 | 3313 | again: |
34f971f6 | 3314 | for_each_class(class) { |
d8ac8971 | 3315 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3316 | if (p) { |
3317 | if (unlikely(p == RETRY_TASK)) | |
3318 | goto again; | |
dd41f596 | 3319 | return p; |
37e117c0 | 3320 | } |
dd41f596 | 3321 | } |
34f971f6 | 3322 | |
d1ccc66d IM |
3323 | /* The idle class should always have a runnable task: */ |
3324 | BUG(); | |
dd41f596 | 3325 | } |
1da177e4 | 3326 | |
dd41f596 | 3327 | /* |
c259e01a | 3328 | * __schedule() is the main scheduler function. |
edde96ea PE |
3329 | * |
3330 | * The main means of driving the scheduler and thus entering this function are: | |
3331 | * | |
3332 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3333 | * | |
3334 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3335 | * paths. For example, see arch/x86/entry_64.S. | |
3336 | * | |
3337 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3338 | * interrupt handler scheduler_tick(). | |
3339 | * | |
3340 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3341 | * task to the run-queue and that's it. | |
3342 | * | |
3343 | * Now, if the new task added to the run-queue preempts the current | |
3344 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3345 | * called on the nearest possible occasion: | |
3346 | * | |
3347 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3348 | * | |
3349 | * - in syscall or exception context, at the next outmost | |
3350 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3351 | * spin_unlock()!) | |
3352 | * | |
3353 | * - in IRQ context, return from interrupt-handler to | |
3354 | * preemptible context | |
3355 | * | |
3356 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3357 | * then at the next: | |
3358 | * | |
3359 | * - cond_resched() call | |
3360 | * - explicit schedule() call | |
3361 | * - return from syscall or exception to user-space | |
3362 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3363 | * |
b30f0e3f | 3364 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3365 | */ |
499d7955 | 3366 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3367 | { |
3368 | struct task_struct *prev, *next; | |
67ca7bde | 3369 | unsigned long *switch_count; |
d8ac8971 | 3370 | struct rq_flags rf; |
dd41f596 | 3371 | struct rq *rq; |
31656519 | 3372 | int cpu; |
dd41f596 | 3373 | |
dd41f596 IM |
3374 | cpu = smp_processor_id(); |
3375 | rq = cpu_rq(cpu); | |
dd41f596 | 3376 | prev = rq->curr; |
dd41f596 | 3377 | |
dd41f596 | 3378 | schedule_debug(prev); |
1da177e4 | 3379 | |
31656519 | 3380 | if (sched_feat(HRTICK)) |
f333fdc9 | 3381 | hrtick_clear(rq); |
8f4d37ec | 3382 | |
46a5d164 PM |
3383 | local_irq_disable(); |
3384 | rcu_note_context_switch(); | |
3385 | ||
e0acd0a6 ON |
3386 | /* |
3387 | * Make sure that signal_pending_state()->signal_pending() below | |
3388 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3389 | * done by the caller to avoid the race with signal_wake_up(). | |
3390 | */ | |
3391 | smp_mb__before_spinlock(); | |
8a8c69c3 | 3392 | rq_lock(rq, &rf); |
1da177e4 | 3393 | |
d1ccc66d IM |
3394 | /* Promote REQ to ACT */ |
3395 | rq->clock_update_flags <<= 1; | |
9edfbfed | 3396 | |
246d86b5 | 3397 | switch_count = &prev->nivcsw; |
fc13aeba | 3398 | if (!preempt && prev->state) { |
21aa9af0 | 3399 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3400 | prev->state = TASK_RUNNING; |
21aa9af0 | 3401 | } else { |
2acca55e PZ |
3402 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3403 | prev->on_rq = 0; | |
3404 | ||
e33a9bba TH |
3405 | if (prev->in_iowait) { |
3406 | atomic_inc(&rq->nr_iowait); | |
3407 | delayacct_blkio_start(); | |
3408 | } | |
3409 | ||
21aa9af0 | 3410 | /* |
2acca55e PZ |
3411 | * If a worker went to sleep, notify and ask workqueue |
3412 | * whether it wants to wake up a task to maintain | |
3413 | * concurrency. | |
21aa9af0 TH |
3414 | */ |
3415 | if (prev->flags & PF_WQ_WORKER) { | |
3416 | struct task_struct *to_wakeup; | |
3417 | ||
9b7f6597 | 3418 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3419 | if (to_wakeup) |
d8ac8971 | 3420 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3421 | } |
21aa9af0 | 3422 | } |
dd41f596 | 3423 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3424 | } |
3425 | ||
9edfbfed | 3426 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3427 | update_rq_clock(rq); |
3428 | ||
d8ac8971 | 3429 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3430 | clear_tsk_need_resched(prev); |
f27dde8d | 3431 | clear_preempt_need_resched(); |
1da177e4 | 3432 | |
1da177e4 | 3433 | if (likely(prev != next)) { |
1da177e4 LT |
3434 | rq->nr_switches++; |
3435 | rq->curr = next; | |
3436 | ++*switch_count; | |
3437 | ||
c73464b1 | 3438 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
3439 | |
3440 | /* Also unlocks the rq: */ | |
3441 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 3442 | } else { |
cb42c9a3 | 3443 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
8a8c69c3 | 3444 | rq_unlock_irq(rq, &rf); |
cbce1a68 | 3445 | } |
1da177e4 | 3446 | |
e3fca9e7 | 3447 | balance_callback(rq); |
1da177e4 | 3448 | } |
c259e01a | 3449 | |
9af6528e PZ |
3450 | void __noreturn do_task_dead(void) |
3451 | { | |
3452 | /* | |
3453 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed | |
3454 | * when the following two conditions become true. | |
3455 | * - There is race condition of mmap_sem (It is acquired by | |
3456 | * exit_mm()), and | |
3457 | * - SMI occurs before setting TASK_RUNINNG. | |
3458 | * (or hypervisor of virtual machine switches to other guest) | |
3459 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD | |
3460 | * | |
3461 | * To avoid it, we have to wait for releasing tsk->pi_lock which | |
3462 | * is held by try_to_wake_up() | |
3463 | */ | |
3464 | smp_mb(); | |
3465 | raw_spin_unlock_wait(¤t->pi_lock); | |
3466 | ||
d1ccc66d | 3467 | /* Causes final put_task_struct in finish_task_switch(): */ |
9af6528e | 3468 | __set_current_state(TASK_DEAD); |
d1ccc66d IM |
3469 | |
3470 | /* Tell freezer to ignore us: */ | |
3471 | current->flags |= PF_NOFREEZE; | |
3472 | ||
9af6528e PZ |
3473 | __schedule(false); |
3474 | BUG(); | |
d1ccc66d IM |
3475 | |
3476 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 3477 | for (;;) |
d1ccc66d | 3478 | cpu_relax(); |
9af6528e PZ |
3479 | } |
3480 | ||
9c40cef2 TG |
3481 | static inline void sched_submit_work(struct task_struct *tsk) |
3482 | { | |
3c7d5184 | 3483 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3484 | return; |
3485 | /* | |
3486 | * If we are going to sleep and we have plugged IO queued, | |
3487 | * make sure to submit it to avoid deadlocks. | |
3488 | */ | |
3489 | if (blk_needs_flush_plug(tsk)) | |
3490 | blk_schedule_flush_plug(tsk); | |
3491 | } | |
3492 | ||
722a9f92 | 3493 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3494 | { |
9c40cef2 TG |
3495 | struct task_struct *tsk = current; |
3496 | ||
3497 | sched_submit_work(tsk); | |
bfd9b2b5 | 3498 | do { |
b30f0e3f | 3499 | preempt_disable(); |
fc13aeba | 3500 | __schedule(false); |
b30f0e3f | 3501 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3502 | } while (need_resched()); |
c259e01a | 3503 | } |
1da177e4 LT |
3504 | EXPORT_SYMBOL(schedule); |
3505 | ||
91d1aa43 | 3506 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3507 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3508 | { |
3509 | /* | |
3510 | * If we come here after a random call to set_need_resched(), | |
3511 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3512 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3513 | * we find a better solution. | |
7cc78f8f AL |
3514 | * |
3515 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3516 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3517 | * too frequently to make sense yet. |
20ab65e3 | 3518 | */ |
7cc78f8f | 3519 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3520 | schedule(); |
7cc78f8f | 3521 | exception_exit(prev_state); |
20ab65e3 FW |
3522 | } |
3523 | #endif | |
3524 | ||
c5491ea7 TG |
3525 | /** |
3526 | * schedule_preempt_disabled - called with preemption disabled | |
3527 | * | |
3528 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3529 | */ | |
3530 | void __sched schedule_preempt_disabled(void) | |
3531 | { | |
ba74c144 | 3532 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3533 | schedule(); |
3534 | preempt_disable(); | |
3535 | } | |
3536 | ||
06b1f808 | 3537 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3538 | { |
3539 | do { | |
47252cfb SR |
3540 | /* |
3541 | * Because the function tracer can trace preempt_count_sub() | |
3542 | * and it also uses preempt_enable/disable_notrace(), if | |
3543 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3544 | * by the function tracer will call this function again and | |
3545 | * cause infinite recursion. | |
3546 | * | |
3547 | * Preemption must be disabled here before the function | |
3548 | * tracer can trace. Break up preempt_disable() into two | |
3549 | * calls. One to disable preemption without fear of being | |
3550 | * traced. The other to still record the preemption latency, | |
3551 | * which can also be traced by the function tracer. | |
3552 | */ | |
499d7955 | 3553 | preempt_disable_notrace(); |
47252cfb | 3554 | preempt_latency_start(1); |
fc13aeba | 3555 | __schedule(true); |
47252cfb | 3556 | preempt_latency_stop(1); |
499d7955 | 3557 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3558 | |
3559 | /* | |
3560 | * Check again in case we missed a preemption opportunity | |
3561 | * between schedule and now. | |
3562 | */ | |
a18b5d01 FW |
3563 | } while (need_resched()); |
3564 | } | |
3565 | ||
1da177e4 LT |
3566 | #ifdef CONFIG_PREEMPT |
3567 | /* | |
2ed6e34f | 3568 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3569 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3570 | * occur there and call schedule directly. |
3571 | */ | |
722a9f92 | 3572 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3573 | { |
1da177e4 LT |
3574 | /* |
3575 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3576 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3577 | */ |
fbb00b56 | 3578 | if (likely(!preemptible())) |
1da177e4 LT |
3579 | return; |
3580 | ||
a18b5d01 | 3581 | preempt_schedule_common(); |
1da177e4 | 3582 | } |
376e2424 | 3583 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3584 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3585 | |
009f60e2 | 3586 | /** |
4eaca0a8 | 3587 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3588 | * |
3589 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3590 | * recursion and tracing preempt enabling caused by the tracing | |
3591 | * infrastructure itself. But as tracing can happen in areas coming | |
3592 | * from userspace or just about to enter userspace, a preempt enable | |
3593 | * can occur before user_exit() is called. This will cause the scheduler | |
3594 | * to be called when the system is still in usermode. | |
3595 | * | |
3596 | * To prevent this, the preempt_enable_notrace will use this function | |
3597 | * instead of preempt_schedule() to exit user context if needed before | |
3598 | * calling the scheduler. | |
3599 | */ | |
4eaca0a8 | 3600 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3601 | { |
3602 | enum ctx_state prev_ctx; | |
3603 | ||
3604 | if (likely(!preemptible())) | |
3605 | return; | |
3606 | ||
3607 | do { | |
47252cfb SR |
3608 | /* |
3609 | * Because the function tracer can trace preempt_count_sub() | |
3610 | * and it also uses preempt_enable/disable_notrace(), if | |
3611 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3612 | * by the function tracer will call this function again and | |
3613 | * cause infinite recursion. | |
3614 | * | |
3615 | * Preemption must be disabled here before the function | |
3616 | * tracer can trace. Break up preempt_disable() into two | |
3617 | * calls. One to disable preemption without fear of being | |
3618 | * traced. The other to still record the preemption latency, | |
3619 | * which can also be traced by the function tracer. | |
3620 | */ | |
3d8f74dd | 3621 | preempt_disable_notrace(); |
47252cfb | 3622 | preempt_latency_start(1); |
009f60e2 ON |
3623 | /* |
3624 | * Needs preempt disabled in case user_exit() is traced | |
3625 | * and the tracer calls preempt_enable_notrace() causing | |
3626 | * an infinite recursion. | |
3627 | */ | |
3628 | prev_ctx = exception_enter(); | |
fc13aeba | 3629 | __schedule(true); |
009f60e2 ON |
3630 | exception_exit(prev_ctx); |
3631 | ||
47252cfb | 3632 | preempt_latency_stop(1); |
3d8f74dd | 3633 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3634 | } while (need_resched()); |
3635 | } | |
4eaca0a8 | 3636 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3637 | |
32e475d7 | 3638 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3639 | |
3640 | /* | |
2ed6e34f | 3641 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3642 | * off of irq context. |
3643 | * Note, that this is called and return with irqs disabled. This will | |
3644 | * protect us against recursive calling from irq. | |
3645 | */ | |
722a9f92 | 3646 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3647 | { |
b22366cd | 3648 | enum ctx_state prev_state; |
6478d880 | 3649 | |
2ed6e34f | 3650 | /* Catch callers which need to be fixed */ |
f27dde8d | 3651 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3652 | |
b22366cd FW |
3653 | prev_state = exception_enter(); |
3654 | ||
3a5c359a | 3655 | do { |
3d8f74dd | 3656 | preempt_disable(); |
3a5c359a | 3657 | local_irq_enable(); |
fc13aeba | 3658 | __schedule(true); |
3a5c359a | 3659 | local_irq_disable(); |
3d8f74dd | 3660 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3661 | } while (need_resched()); |
b22366cd FW |
3662 | |
3663 | exception_exit(prev_state); | |
1da177e4 LT |
3664 | } |
3665 | ||
63859d4f | 3666 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3667 | void *key) |
1da177e4 | 3668 | { |
63859d4f | 3669 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3670 | } |
1da177e4 LT |
3671 | EXPORT_SYMBOL(default_wake_function); |
3672 | ||
b29739f9 IM |
3673 | #ifdef CONFIG_RT_MUTEXES |
3674 | ||
3675 | /* | |
3676 | * rt_mutex_setprio - set the current priority of a task | |
3677 | * @p: task | |
3678 | * @prio: prio value (kernel-internal form) | |
3679 | * | |
3680 | * This function changes the 'effective' priority of a task. It does | |
3681 | * not touch ->normal_prio like __setscheduler(). | |
3682 | * | |
c365c292 TG |
3683 | * Used by the rt_mutex code to implement priority inheritance |
3684 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3685 | */ |
36c8b586 | 3686 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3687 | { |
ff77e468 | 3688 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3689 | const struct sched_class *prev_class; |
eb580751 PZ |
3690 | struct rq_flags rf; |
3691 | struct rq *rq; | |
b29739f9 | 3692 | |
aab03e05 | 3693 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3694 | |
eb580751 | 3695 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 3696 | update_rq_clock(rq); |
b29739f9 | 3697 | |
1c4dd99b TG |
3698 | /* |
3699 | * Idle task boosting is a nono in general. There is one | |
3700 | * exception, when PREEMPT_RT and NOHZ is active: | |
3701 | * | |
3702 | * The idle task calls get_next_timer_interrupt() and holds | |
3703 | * the timer wheel base->lock on the CPU and another CPU wants | |
3704 | * to access the timer (probably to cancel it). We can safely | |
3705 | * ignore the boosting request, as the idle CPU runs this code | |
3706 | * with interrupts disabled and will complete the lock | |
3707 | * protected section without being interrupted. So there is no | |
3708 | * real need to boost. | |
3709 | */ | |
3710 | if (unlikely(p == rq->idle)) { | |
3711 | WARN_ON(p != rq->curr); | |
3712 | WARN_ON(p->pi_blocked_on); | |
3713 | goto out_unlock; | |
3714 | } | |
3715 | ||
a8027073 | 3716 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3717 | oldprio = p->prio; |
ff77e468 PZ |
3718 | |
3719 | if (oldprio == prio) | |
3720 | queue_flag &= ~DEQUEUE_MOVE; | |
3721 | ||
83ab0aa0 | 3722 | prev_class = p->sched_class; |
da0c1e65 | 3723 | queued = task_on_rq_queued(p); |
051a1d1a | 3724 | running = task_current(rq, p); |
da0c1e65 | 3725 | if (queued) |
ff77e468 | 3726 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3727 | if (running) |
f3cd1c4e | 3728 | put_prev_task(rq, p); |
dd41f596 | 3729 | |
2d3d891d DF |
3730 | /* |
3731 | * Boosting condition are: | |
3732 | * 1. -rt task is running and holds mutex A | |
3733 | * --> -dl task blocks on mutex A | |
3734 | * | |
3735 | * 2. -dl task is running and holds mutex A | |
3736 | * --> -dl task blocks on mutex A and could preempt the | |
3737 | * running task | |
3738 | */ | |
3739 | if (dl_prio(prio)) { | |
466af29b ON |
3740 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3741 | if (!dl_prio(p->normal_prio) || | |
3742 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3743 | p->dl.dl_boosted = 1; |
ff77e468 | 3744 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3745 | } else |
3746 | p->dl.dl_boosted = 0; | |
aab03e05 | 3747 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3748 | } else if (rt_prio(prio)) { |
3749 | if (dl_prio(oldprio)) | |
3750 | p->dl.dl_boosted = 0; | |
3751 | if (oldprio < prio) | |
ff77e468 | 3752 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3753 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3754 | } else { |
3755 | if (dl_prio(oldprio)) | |
3756 | p->dl.dl_boosted = 0; | |
746db944 BS |
3757 | if (rt_prio(oldprio)) |
3758 | p->rt.timeout = 0; | |
dd41f596 | 3759 | p->sched_class = &fair_sched_class; |
2d3d891d | 3760 | } |
dd41f596 | 3761 | |
b29739f9 IM |
3762 | p->prio = prio; |
3763 | ||
da0c1e65 | 3764 | if (queued) |
ff77e468 | 3765 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3766 | if (running) |
b2bf6c31 | 3767 | set_curr_task(rq, p); |
cb469845 | 3768 | |
da7a735e | 3769 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3770 | out_unlock: |
d1ccc66d IM |
3771 | /* Avoid rq from going away on us: */ |
3772 | preempt_disable(); | |
eb580751 | 3773 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3774 | |
3775 | balance_callback(rq); | |
3776 | preempt_enable(); | |
b29739f9 | 3777 | } |
b29739f9 | 3778 | #endif |
d50dde5a | 3779 | |
36c8b586 | 3780 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3781 | { |
49bd21ef PZ |
3782 | bool queued, running; |
3783 | int old_prio, delta; | |
eb580751 | 3784 | struct rq_flags rf; |
70b97a7f | 3785 | struct rq *rq; |
1da177e4 | 3786 | |
75e45d51 | 3787 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3788 | return; |
3789 | /* | |
3790 | * We have to be careful, if called from sys_setpriority(), | |
3791 | * the task might be in the middle of scheduling on another CPU. | |
3792 | */ | |
eb580751 | 3793 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
3794 | update_rq_clock(rq); |
3795 | ||
1da177e4 LT |
3796 | /* |
3797 | * The RT priorities are set via sched_setscheduler(), but we still | |
3798 | * allow the 'normal' nice value to be set - but as expected | |
3799 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3800 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3801 | */ |
aab03e05 | 3802 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3803 | p->static_prio = NICE_TO_PRIO(nice); |
3804 | goto out_unlock; | |
3805 | } | |
da0c1e65 | 3806 | queued = task_on_rq_queued(p); |
49bd21ef | 3807 | running = task_current(rq, p); |
da0c1e65 | 3808 | if (queued) |
1de64443 | 3809 | dequeue_task(rq, p, DEQUEUE_SAVE); |
49bd21ef PZ |
3810 | if (running) |
3811 | put_prev_task(rq, p); | |
1da177e4 | 3812 | |
1da177e4 | 3813 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3814 | set_load_weight(p); |
b29739f9 IM |
3815 | old_prio = p->prio; |
3816 | p->prio = effective_prio(p); | |
3817 | delta = p->prio - old_prio; | |
1da177e4 | 3818 | |
da0c1e65 | 3819 | if (queued) { |
7134b3e9 | 3820 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
1da177e4 | 3821 | /* |
d5f9f942 AM |
3822 | * If the task increased its priority or is running and |
3823 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3824 | */ |
d5f9f942 | 3825 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3826 | resched_curr(rq); |
1da177e4 | 3827 | } |
49bd21ef PZ |
3828 | if (running) |
3829 | set_curr_task(rq, p); | |
1da177e4 | 3830 | out_unlock: |
eb580751 | 3831 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3832 | } |
1da177e4 LT |
3833 | EXPORT_SYMBOL(set_user_nice); |
3834 | ||
e43379f1 MM |
3835 | /* |
3836 | * can_nice - check if a task can reduce its nice value | |
3837 | * @p: task | |
3838 | * @nice: nice value | |
3839 | */ | |
36c8b586 | 3840 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3841 | { |
d1ccc66d | 3842 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 3843 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3844 | |
78d7d407 | 3845 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3846 | capable(CAP_SYS_NICE)); |
3847 | } | |
3848 | ||
1da177e4 LT |
3849 | #ifdef __ARCH_WANT_SYS_NICE |
3850 | ||
3851 | /* | |
3852 | * sys_nice - change the priority of the current process. | |
3853 | * @increment: priority increment | |
3854 | * | |
3855 | * sys_setpriority is a more generic, but much slower function that | |
3856 | * does similar things. | |
3857 | */ | |
5add95d4 | 3858 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3859 | { |
48f24c4d | 3860 | long nice, retval; |
1da177e4 LT |
3861 | |
3862 | /* | |
3863 | * Setpriority might change our priority at the same moment. | |
3864 | * We don't have to worry. Conceptually one call occurs first | |
3865 | * and we have a single winner. | |
3866 | */ | |
a9467fa3 | 3867 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3868 | nice = task_nice(current) + increment; |
1da177e4 | 3869 | |
a9467fa3 | 3870 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3871 | if (increment < 0 && !can_nice(current, nice)) |
3872 | return -EPERM; | |
3873 | ||
1da177e4 LT |
3874 | retval = security_task_setnice(current, nice); |
3875 | if (retval) | |
3876 | return retval; | |
3877 | ||
3878 | set_user_nice(current, nice); | |
3879 | return 0; | |
3880 | } | |
3881 | ||
3882 | #endif | |
3883 | ||
3884 | /** | |
3885 | * task_prio - return the priority value of a given task. | |
3886 | * @p: the task in question. | |
3887 | * | |
e69f6186 | 3888 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3889 | * RT tasks are offset by -200. Normal tasks are centered |
3890 | * around 0, value goes from -16 to +15. | |
3891 | */ | |
36c8b586 | 3892 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3893 | { |
3894 | return p->prio - MAX_RT_PRIO; | |
3895 | } | |
3896 | ||
1da177e4 | 3897 | /** |
d1ccc66d | 3898 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 3899 | * @cpu: the processor in question. |
e69f6186 YB |
3900 | * |
3901 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3902 | */ |
3903 | int idle_cpu(int cpu) | |
3904 | { | |
908a3283 TG |
3905 | struct rq *rq = cpu_rq(cpu); |
3906 | ||
3907 | if (rq->curr != rq->idle) | |
3908 | return 0; | |
3909 | ||
3910 | if (rq->nr_running) | |
3911 | return 0; | |
3912 | ||
3913 | #ifdef CONFIG_SMP | |
3914 | if (!llist_empty(&rq->wake_list)) | |
3915 | return 0; | |
3916 | #endif | |
3917 | ||
3918 | return 1; | |
1da177e4 LT |
3919 | } |
3920 | ||
1da177e4 | 3921 | /** |
d1ccc66d | 3922 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 3923 | * @cpu: the processor in question. |
e69f6186 | 3924 | * |
d1ccc66d | 3925 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 3926 | */ |
36c8b586 | 3927 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3928 | { |
3929 | return cpu_rq(cpu)->idle; | |
3930 | } | |
3931 | ||
3932 | /** | |
3933 | * find_process_by_pid - find a process with a matching PID value. | |
3934 | * @pid: the pid in question. | |
e69f6186 YB |
3935 | * |
3936 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3937 | */ |
a9957449 | 3938 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3939 | { |
228ebcbe | 3940 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3941 | } |
3942 | ||
aab03e05 DF |
3943 | /* |
3944 | * This function initializes the sched_dl_entity of a newly becoming | |
3945 | * SCHED_DEADLINE task. | |
3946 | * | |
3947 | * Only the static values are considered here, the actual runtime and the | |
3948 | * absolute deadline will be properly calculated when the task is enqueued | |
3949 | * for the first time with its new policy. | |
3950 | */ | |
3951 | static void | |
3952 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3953 | { | |
3954 | struct sched_dl_entity *dl_se = &p->dl; | |
3955 | ||
aab03e05 DF |
3956 | dl_se->dl_runtime = attr->sched_runtime; |
3957 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3958 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3959 | dl_se->flags = attr->sched_flags; |
332ac17e | 3960 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3961 | |
3962 | /* | |
3963 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3964 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3965 | * | |
3966 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3967 | * point. This would include retaining the task_struct until that time | |
3968 | * and change dl_overflow() to not immediately decrement the current | |
3969 | * amount. | |
3970 | * | |
3971 | * Instead we retain the current runtime/deadline and let the new | |
3972 | * parameters take effect after the current reservation period lapses. | |
3973 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3974 | * before the current scheduling deadline. | |
3975 | * | |
3976 | * We can still have temporary overloads because we do not delay the | |
3977 | * change in bandwidth until that time; so admission control is | |
3978 | * not on the safe side. It does however guarantee tasks will never | |
3979 | * consume more than promised. | |
3980 | */ | |
aab03e05 DF |
3981 | } |
3982 | ||
c13db6b1 SR |
3983 | /* |
3984 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3985 | * it calls know not to change it. | |
3986 | */ | |
3987 | #define SETPARAM_POLICY -1 | |
3988 | ||
c365c292 TG |
3989 | static void __setscheduler_params(struct task_struct *p, |
3990 | const struct sched_attr *attr) | |
1da177e4 | 3991 | { |
d50dde5a DF |
3992 | int policy = attr->sched_policy; |
3993 | ||
c13db6b1 | 3994 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3995 | policy = p->policy; |
3996 | ||
1da177e4 | 3997 | p->policy = policy; |
d50dde5a | 3998 | |
aab03e05 DF |
3999 | if (dl_policy(policy)) |
4000 | __setparam_dl(p, attr); | |
39fd8fd2 | 4001 | else if (fair_policy(policy)) |
d50dde5a DF |
4002 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
4003 | ||
39fd8fd2 PZ |
4004 | /* |
4005 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
4006 | * !rt_policy. Always setting this ensures that things like | |
4007 | * getparam()/getattr() don't report silly values for !rt tasks. | |
4008 | */ | |
4009 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4010 | p->normal_prio = normal_prio(p); |
c365c292 TG |
4011 | set_load_weight(p); |
4012 | } | |
39fd8fd2 | 4013 | |
c365c292 TG |
4014 | /* Actually do priority change: must hold pi & rq lock. */ |
4015 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4016 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
4017 | { |
4018 | __setscheduler_params(p, attr); | |
d50dde5a | 4019 | |
383afd09 | 4020 | /* |
0782e63b TG |
4021 | * Keep a potential priority boosting if called from |
4022 | * sched_setscheduler(). | |
383afd09 | 4023 | */ |
0782e63b TG |
4024 | if (keep_boost) |
4025 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
4026 | else | |
4027 | p->prio = normal_prio(p); | |
383afd09 | 4028 | |
aab03e05 DF |
4029 | if (dl_prio(p->prio)) |
4030 | p->sched_class = &dl_sched_class; | |
4031 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4032 | p->sched_class = &rt_sched_class; |
4033 | else | |
4034 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4035 | } |
aab03e05 DF |
4036 | |
4037 | static void | |
4038 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
4039 | { | |
4040 | struct sched_dl_entity *dl_se = &p->dl; | |
4041 | ||
4042 | attr->sched_priority = p->rt_priority; | |
4043 | attr->sched_runtime = dl_se->dl_runtime; | |
4044 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 4045 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
4046 | attr->sched_flags = dl_se->flags; |
4047 | } | |
4048 | ||
4049 | /* | |
4050 | * This function validates the new parameters of a -deadline task. | |
4051 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 4052 | * than the runtime, as well as the period of being zero or |
332ac17e | 4053 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
4054 | * user parameters are above the internal resolution of 1us (we |
4055 | * check sched_runtime only since it is always the smaller one) and | |
4056 | * below 2^63 ns (we have to check both sched_deadline and | |
4057 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
4058 | */ |
4059 | static bool | |
4060 | __checkparam_dl(const struct sched_attr *attr) | |
4061 | { | |
b0827819 JL |
4062 | /* deadline != 0 */ |
4063 | if (attr->sched_deadline == 0) | |
4064 | return false; | |
4065 | ||
4066 | /* | |
4067 | * Since we truncate DL_SCALE bits, make sure we're at least | |
4068 | * that big. | |
4069 | */ | |
4070 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
4071 | return false; | |
4072 | ||
4073 | /* | |
4074 | * Since we use the MSB for wrap-around and sign issues, make | |
4075 | * sure it's not set (mind that period can be equal to zero). | |
4076 | */ | |
4077 | if (attr->sched_deadline & (1ULL << 63) || | |
4078 | attr->sched_period & (1ULL << 63)) | |
4079 | return false; | |
4080 | ||
4081 | /* runtime <= deadline <= period (if period != 0) */ | |
4082 | if ((attr->sched_period != 0 && | |
4083 | attr->sched_period < attr->sched_deadline) || | |
4084 | attr->sched_deadline < attr->sched_runtime) | |
4085 | return false; | |
4086 | ||
4087 | return true; | |
aab03e05 DF |
4088 | } |
4089 | ||
c69e8d9c | 4090 | /* |
d1ccc66d | 4091 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
4092 | */ |
4093 | static bool check_same_owner(struct task_struct *p) | |
4094 | { | |
4095 | const struct cred *cred = current_cred(), *pcred; | |
4096 | bool match; | |
4097 | ||
4098 | rcu_read_lock(); | |
4099 | pcred = __task_cred(p); | |
9c806aa0 EB |
4100 | match = (uid_eq(cred->euid, pcred->euid) || |
4101 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4102 | rcu_read_unlock(); |
4103 | return match; | |
4104 | } | |
4105 | ||
d1ccc66d | 4106 | static bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr) |
75381608 WL |
4107 | { |
4108 | struct sched_dl_entity *dl_se = &p->dl; | |
4109 | ||
4110 | if (dl_se->dl_runtime != attr->sched_runtime || | |
4111 | dl_se->dl_deadline != attr->sched_deadline || | |
4112 | dl_se->dl_period != attr->sched_period || | |
4113 | dl_se->flags != attr->sched_flags) | |
4114 | return true; | |
4115 | ||
4116 | return false; | |
4117 | } | |
4118 | ||
d50dde5a DF |
4119 | static int __sched_setscheduler(struct task_struct *p, |
4120 | const struct sched_attr *attr, | |
dbc7f069 | 4121 | bool user, bool pi) |
1da177e4 | 4122 | { |
383afd09 SR |
4123 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4124 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4125 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4126 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4127 | const struct sched_class *prev_class; |
eb580751 | 4128 | struct rq_flags rf; |
ca94c442 | 4129 | int reset_on_fork; |
ff77e468 | 4130 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 4131 | struct rq *rq; |
1da177e4 | 4132 | |
d1ccc66d | 4133 | /* May grab non-irq protected spin_locks: */ |
66e5393a | 4134 | BUG_ON(in_interrupt()); |
1da177e4 | 4135 | recheck: |
d1ccc66d | 4136 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
4137 | if (policy < 0) { |
4138 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4139 | policy = oldpolicy = p->policy; |
ca94c442 | 4140 | } else { |
7479f3c9 | 4141 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4142 | |
20f9cd2a | 4143 | if (!valid_policy(policy)) |
ca94c442 LP |
4144 | return -EINVAL; |
4145 | } | |
4146 | ||
7479f3c9 PZ |
4147 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
4148 | return -EINVAL; | |
4149 | ||
1da177e4 LT |
4150 | /* |
4151 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4152 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4153 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4154 | */ |
0bb040a4 | 4155 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4156 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4157 | return -EINVAL; |
aab03e05 DF |
4158 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4159 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4160 | return -EINVAL; |
4161 | ||
37e4ab3f OC |
4162 | /* |
4163 | * Allow unprivileged RT tasks to decrease priority: | |
4164 | */ | |
961ccddd | 4165 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4166 | if (fair_policy(policy)) { |
d0ea0268 | 4167 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4168 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4169 | return -EPERM; |
4170 | } | |
4171 | ||
e05606d3 | 4172 | if (rt_policy(policy)) { |
a44702e8 ON |
4173 | unsigned long rlim_rtprio = |
4174 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 4175 | |
d1ccc66d | 4176 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
4177 | if (policy != p->policy && !rlim_rtprio) |
4178 | return -EPERM; | |
4179 | ||
d1ccc66d | 4180 | /* Can't increase priority: */ |
d50dde5a DF |
4181 | if (attr->sched_priority > p->rt_priority && |
4182 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4183 | return -EPERM; |
4184 | } | |
c02aa73b | 4185 | |
d44753b8 JL |
4186 | /* |
4187 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4188 | * (safest behavior); in the future we would like to allow | |
4189 | * unprivileged DL tasks to increase their relative deadline | |
4190 | * or reduce their runtime (both ways reducing utilization) | |
4191 | */ | |
4192 | if (dl_policy(policy)) | |
4193 | return -EPERM; | |
4194 | ||
dd41f596 | 4195 | /* |
c02aa73b DH |
4196 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4197 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4198 | */ |
20f9cd2a | 4199 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4200 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4201 | return -EPERM; |
4202 | } | |
5fe1d75f | 4203 | |
d1ccc66d | 4204 | /* Can't change other user's priorities: */ |
c69e8d9c | 4205 | if (!check_same_owner(p)) |
37e4ab3f | 4206 | return -EPERM; |
ca94c442 | 4207 | |
d1ccc66d | 4208 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
4209 | if (p->sched_reset_on_fork && !reset_on_fork) |
4210 | return -EPERM; | |
37e4ab3f | 4211 | } |
1da177e4 | 4212 | |
725aad24 | 4213 | if (user) { |
b0ae1981 | 4214 | retval = security_task_setscheduler(p); |
725aad24 JF |
4215 | if (retval) |
4216 | return retval; | |
4217 | } | |
4218 | ||
b29739f9 | 4219 | /* |
d1ccc66d | 4220 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 4221 | * changing the priority of the task: |
0122ec5b | 4222 | * |
25985edc | 4223 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4224 | * runqueue lock must be held. |
4225 | */ | |
eb580751 | 4226 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4227 | update_rq_clock(rq); |
dc61b1d6 | 4228 | |
34f971f6 | 4229 | /* |
d1ccc66d | 4230 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
4231 | */ |
4232 | if (p == rq->stop) { | |
eb580751 | 4233 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4234 | return -EINVAL; |
4235 | } | |
4236 | ||
a51e9198 | 4237 | /* |
d6b1e911 TG |
4238 | * If not changing anything there's no need to proceed further, |
4239 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4240 | */ |
d50dde5a | 4241 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4242 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4243 | goto change; |
4244 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4245 | goto change; | |
75381608 | 4246 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4247 | goto change; |
d50dde5a | 4248 | |
d6b1e911 | 4249 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4250 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4251 | return 0; |
4252 | } | |
d50dde5a | 4253 | change: |
a51e9198 | 4254 | |
dc61b1d6 | 4255 | if (user) { |
332ac17e | 4256 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4257 | /* |
4258 | * Do not allow realtime tasks into groups that have no runtime | |
4259 | * assigned. | |
4260 | */ | |
4261 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4262 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4263 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4264 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4265 | return -EPERM; |
4266 | } | |
dc61b1d6 | 4267 | #endif |
332ac17e DF |
4268 | #ifdef CONFIG_SMP |
4269 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4270 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4271 | |
4272 | /* | |
4273 | * Don't allow tasks with an affinity mask smaller than | |
4274 | * the entire root_domain to become SCHED_DEADLINE. We | |
4275 | * will also fail if there's no bandwidth available. | |
4276 | */ | |
e4099a5e PZ |
4277 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4278 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4279 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4280 | return -EPERM; |
4281 | } | |
4282 | } | |
4283 | #endif | |
4284 | } | |
dc61b1d6 | 4285 | |
d1ccc66d | 4286 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
4287 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
4288 | policy = oldpolicy = -1; | |
eb580751 | 4289 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4290 | goto recheck; |
4291 | } | |
332ac17e DF |
4292 | |
4293 | /* | |
4294 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4295 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4296 | * is available. | |
4297 | */ | |
e4099a5e | 4298 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4299 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4300 | return -EBUSY; |
4301 | } | |
4302 | ||
c365c292 TG |
4303 | p->sched_reset_on_fork = reset_on_fork; |
4304 | oldprio = p->prio; | |
4305 | ||
dbc7f069 PZ |
4306 | if (pi) { |
4307 | /* | |
4308 | * Take priority boosted tasks into account. If the new | |
4309 | * effective priority is unchanged, we just store the new | |
4310 | * normal parameters and do not touch the scheduler class and | |
4311 | * the runqueue. This will be done when the task deboost | |
4312 | * itself. | |
4313 | */ | |
4314 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4315 | if (new_effective_prio == oldprio) |
4316 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4317 | } |
4318 | ||
da0c1e65 | 4319 | queued = task_on_rq_queued(p); |
051a1d1a | 4320 | running = task_current(rq, p); |
da0c1e65 | 4321 | if (queued) |
ff77e468 | 4322 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4323 | if (running) |
f3cd1c4e | 4324 | put_prev_task(rq, p); |
f6b53205 | 4325 | |
83ab0aa0 | 4326 | prev_class = p->sched_class; |
dbc7f069 | 4327 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4328 | |
da0c1e65 | 4329 | if (queued) { |
81a44c54 TG |
4330 | /* |
4331 | * We enqueue to tail when the priority of a task is | |
4332 | * increased (user space view). | |
4333 | */ | |
ff77e468 PZ |
4334 | if (oldprio < p->prio) |
4335 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4336 | |
ff77e468 | 4337 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4338 | } |
a399d233 | 4339 | if (running) |
b2bf6c31 | 4340 | set_curr_task(rq, p); |
cb469845 | 4341 | |
da7a735e | 4342 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
4343 | |
4344 | /* Avoid rq from going away on us: */ | |
4345 | preempt_disable(); | |
eb580751 | 4346 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4347 | |
dbc7f069 PZ |
4348 | if (pi) |
4349 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4350 | |
d1ccc66d | 4351 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
4352 | balance_callback(rq); |
4353 | preempt_enable(); | |
95e02ca9 | 4354 | |
1da177e4 LT |
4355 | return 0; |
4356 | } | |
961ccddd | 4357 | |
7479f3c9 PZ |
4358 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4359 | const struct sched_param *param, bool check) | |
4360 | { | |
4361 | struct sched_attr attr = { | |
4362 | .sched_policy = policy, | |
4363 | .sched_priority = param->sched_priority, | |
4364 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4365 | }; | |
4366 | ||
c13db6b1 SR |
4367 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4368 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4369 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4370 | policy &= ~SCHED_RESET_ON_FORK; | |
4371 | attr.sched_policy = policy; | |
4372 | } | |
4373 | ||
dbc7f069 | 4374 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4375 | } |
961ccddd RR |
4376 | /** |
4377 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4378 | * @p: the task in question. | |
4379 | * @policy: new policy. | |
4380 | * @param: structure containing the new RT priority. | |
4381 | * | |
e69f6186 YB |
4382 | * Return: 0 on success. An error code otherwise. |
4383 | * | |
961ccddd RR |
4384 | * NOTE that the task may be already dead. |
4385 | */ | |
4386 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4387 | const struct sched_param *param) |
961ccddd | 4388 | { |
7479f3c9 | 4389 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4390 | } |
1da177e4 LT |
4391 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4392 | ||
d50dde5a DF |
4393 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4394 | { | |
dbc7f069 | 4395 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4396 | } |
4397 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4398 | ||
961ccddd RR |
4399 | /** |
4400 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4401 | * @p: the task in question. | |
4402 | * @policy: new policy. | |
4403 | * @param: structure containing the new RT priority. | |
4404 | * | |
4405 | * Just like sched_setscheduler, only don't bother checking if the | |
4406 | * current context has permission. For example, this is needed in | |
4407 | * stop_machine(): we create temporary high priority worker threads, | |
4408 | * but our caller might not have that capability. | |
e69f6186 YB |
4409 | * |
4410 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4411 | */ |
4412 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4413 | const struct sched_param *param) |
961ccddd | 4414 | { |
7479f3c9 | 4415 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4416 | } |
84778472 | 4417 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4418 | |
95cdf3b7 IM |
4419 | static int |
4420 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4421 | { |
1da177e4 LT |
4422 | struct sched_param lparam; |
4423 | struct task_struct *p; | |
36c8b586 | 4424 | int retval; |
1da177e4 LT |
4425 | |
4426 | if (!param || pid < 0) | |
4427 | return -EINVAL; | |
4428 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4429 | return -EFAULT; | |
5fe1d75f ON |
4430 | |
4431 | rcu_read_lock(); | |
4432 | retval = -ESRCH; | |
1da177e4 | 4433 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4434 | if (p != NULL) |
4435 | retval = sched_setscheduler(p, policy, &lparam); | |
4436 | rcu_read_unlock(); | |
36c8b586 | 4437 | |
1da177e4 LT |
4438 | return retval; |
4439 | } | |
4440 | ||
d50dde5a DF |
4441 | /* |
4442 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4443 | */ | |
d1ccc66d | 4444 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
4445 | { |
4446 | u32 size; | |
4447 | int ret; | |
4448 | ||
4449 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4450 | return -EFAULT; | |
4451 | ||
d1ccc66d | 4452 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
4453 | memset(attr, 0, sizeof(*attr)); |
4454 | ||
4455 | ret = get_user(size, &uattr->size); | |
4456 | if (ret) | |
4457 | return ret; | |
4458 | ||
d1ccc66d IM |
4459 | /* Bail out on silly large: */ |
4460 | if (size > PAGE_SIZE) | |
d50dde5a DF |
4461 | goto err_size; |
4462 | ||
d1ccc66d IM |
4463 | /* ABI compatibility quirk: */ |
4464 | if (!size) | |
d50dde5a DF |
4465 | size = SCHED_ATTR_SIZE_VER0; |
4466 | ||
4467 | if (size < SCHED_ATTR_SIZE_VER0) | |
4468 | goto err_size; | |
4469 | ||
4470 | /* | |
4471 | * If we're handed a bigger struct than we know of, | |
4472 | * ensure all the unknown bits are 0 - i.e. new | |
4473 | * user-space does not rely on any kernel feature | |
4474 | * extensions we dont know about yet. | |
4475 | */ | |
4476 | if (size > sizeof(*attr)) { | |
4477 | unsigned char __user *addr; | |
4478 | unsigned char __user *end; | |
4479 | unsigned char val; | |
4480 | ||
4481 | addr = (void __user *)uattr + sizeof(*attr); | |
4482 | end = (void __user *)uattr + size; | |
4483 | ||
4484 | for (; addr < end; addr++) { | |
4485 | ret = get_user(val, addr); | |
4486 | if (ret) | |
4487 | return ret; | |
4488 | if (val) | |
4489 | goto err_size; | |
4490 | } | |
4491 | size = sizeof(*attr); | |
4492 | } | |
4493 | ||
4494 | ret = copy_from_user(attr, uattr, size); | |
4495 | if (ret) | |
4496 | return -EFAULT; | |
4497 | ||
4498 | /* | |
d1ccc66d | 4499 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
4500 | * to be strict and return an error on out-of-bounds values? |
4501 | */ | |
75e45d51 | 4502 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4503 | |
e78c7bca | 4504 | return 0; |
d50dde5a DF |
4505 | |
4506 | err_size: | |
4507 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4508 | return -E2BIG; |
d50dde5a DF |
4509 | } |
4510 | ||
1da177e4 LT |
4511 | /** |
4512 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4513 | * @pid: the pid in question. | |
4514 | * @policy: new policy. | |
4515 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4516 | * |
4517 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4518 | */ |
d1ccc66d | 4519 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 4520 | { |
c21761f1 JB |
4521 | if (policy < 0) |
4522 | return -EINVAL; | |
4523 | ||
1da177e4 LT |
4524 | return do_sched_setscheduler(pid, policy, param); |
4525 | } | |
4526 | ||
4527 | /** | |
4528 | * sys_sched_setparam - set/change the RT priority of a thread | |
4529 | * @pid: the pid in question. | |
4530 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4531 | * |
4532 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4533 | */ |
5add95d4 | 4534 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4535 | { |
c13db6b1 | 4536 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4537 | } |
4538 | ||
d50dde5a DF |
4539 | /** |
4540 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4541 | * @pid: the pid in question. | |
5778fccf | 4542 | * @uattr: structure containing the extended parameters. |
db66d756 | 4543 | * @flags: for future extension. |
d50dde5a | 4544 | */ |
6d35ab48 PZ |
4545 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4546 | unsigned int, flags) | |
d50dde5a DF |
4547 | { |
4548 | struct sched_attr attr; | |
4549 | struct task_struct *p; | |
4550 | int retval; | |
4551 | ||
6d35ab48 | 4552 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4553 | return -EINVAL; |
4554 | ||
143cf23d MK |
4555 | retval = sched_copy_attr(uattr, &attr); |
4556 | if (retval) | |
4557 | return retval; | |
d50dde5a | 4558 | |
b14ed2c2 | 4559 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4560 | return -EINVAL; |
d50dde5a DF |
4561 | |
4562 | rcu_read_lock(); | |
4563 | retval = -ESRCH; | |
4564 | p = find_process_by_pid(pid); | |
4565 | if (p != NULL) | |
4566 | retval = sched_setattr(p, &attr); | |
4567 | rcu_read_unlock(); | |
4568 | ||
4569 | return retval; | |
4570 | } | |
4571 | ||
1da177e4 LT |
4572 | /** |
4573 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4574 | * @pid: the pid in question. | |
e69f6186 YB |
4575 | * |
4576 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4577 | * code. | |
1da177e4 | 4578 | */ |
5add95d4 | 4579 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4580 | { |
36c8b586 | 4581 | struct task_struct *p; |
3a5c359a | 4582 | int retval; |
1da177e4 LT |
4583 | |
4584 | if (pid < 0) | |
3a5c359a | 4585 | return -EINVAL; |
1da177e4 LT |
4586 | |
4587 | retval = -ESRCH; | |
5fe85be0 | 4588 | rcu_read_lock(); |
1da177e4 LT |
4589 | p = find_process_by_pid(pid); |
4590 | if (p) { | |
4591 | retval = security_task_getscheduler(p); | |
4592 | if (!retval) | |
ca94c442 LP |
4593 | retval = p->policy |
4594 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4595 | } |
5fe85be0 | 4596 | rcu_read_unlock(); |
1da177e4 LT |
4597 | return retval; |
4598 | } | |
4599 | ||
4600 | /** | |
ca94c442 | 4601 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4602 | * @pid: the pid in question. |
4603 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4604 | * |
4605 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4606 | * code. | |
1da177e4 | 4607 | */ |
5add95d4 | 4608 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4609 | { |
ce5f7f82 | 4610 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4611 | struct task_struct *p; |
3a5c359a | 4612 | int retval; |
1da177e4 LT |
4613 | |
4614 | if (!param || pid < 0) | |
3a5c359a | 4615 | return -EINVAL; |
1da177e4 | 4616 | |
5fe85be0 | 4617 | rcu_read_lock(); |
1da177e4 LT |
4618 | p = find_process_by_pid(pid); |
4619 | retval = -ESRCH; | |
4620 | if (!p) | |
4621 | goto out_unlock; | |
4622 | ||
4623 | retval = security_task_getscheduler(p); | |
4624 | if (retval) | |
4625 | goto out_unlock; | |
4626 | ||
ce5f7f82 PZ |
4627 | if (task_has_rt_policy(p)) |
4628 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4629 | rcu_read_unlock(); |
1da177e4 LT |
4630 | |
4631 | /* | |
4632 | * This one might sleep, we cannot do it with a spinlock held ... | |
4633 | */ | |
4634 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4635 | ||
1da177e4 LT |
4636 | return retval; |
4637 | ||
4638 | out_unlock: | |
5fe85be0 | 4639 | rcu_read_unlock(); |
1da177e4 LT |
4640 | return retval; |
4641 | } | |
4642 | ||
d50dde5a DF |
4643 | static int sched_read_attr(struct sched_attr __user *uattr, |
4644 | struct sched_attr *attr, | |
4645 | unsigned int usize) | |
4646 | { | |
4647 | int ret; | |
4648 | ||
4649 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4650 | return -EFAULT; | |
4651 | ||
4652 | /* | |
4653 | * If we're handed a smaller struct than we know of, | |
4654 | * ensure all the unknown bits are 0 - i.e. old | |
4655 | * user-space does not get uncomplete information. | |
4656 | */ | |
4657 | if (usize < sizeof(*attr)) { | |
4658 | unsigned char *addr; | |
4659 | unsigned char *end; | |
4660 | ||
4661 | addr = (void *)attr + usize; | |
4662 | end = (void *)attr + sizeof(*attr); | |
4663 | ||
4664 | for (; addr < end; addr++) { | |
4665 | if (*addr) | |
22400674 | 4666 | return -EFBIG; |
d50dde5a DF |
4667 | } |
4668 | ||
4669 | attr->size = usize; | |
4670 | } | |
4671 | ||
4efbc454 | 4672 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4673 | if (ret) |
4674 | return -EFAULT; | |
4675 | ||
22400674 | 4676 | return 0; |
d50dde5a DF |
4677 | } |
4678 | ||
4679 | /** | |
aab03e05 | 4680 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4681 | * @pid: the pid in question. |
5778fccf | 4682 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4683 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4684 | * @flags: for future extension. |
d50dde5a | 4685 | */ |
6d35ab48 PZ |
4686 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4687 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4688 | { |
4689 | struct sched_attr attr = { | |
4690 | .size = sizeof(struct sched_attr), | |
4691 | }; | |
4692 | struct task_struct *p; | |
4693 | int retval; | |
4694 | ||
4695 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4696 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4697 | return -EINVAL; |
4698 | ||
4699 | rcu_read_lock(); | |
4700 | p = find_process_by_pid(pid); | |
4701 | retval = -ESRCH; | |
4702 | if (!p) | |
4703 | goto out_unlock; | |
4704 | ||
4705 | retval = security_task_getscheduler(p); | |
4706 | if (retval) | |
4707 | goto out_unlock; | |
4708 | ||
4709 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4710 | if (p->sched_reset_on_fork) |
4711 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4712 | if (task_has_dl_policy(p)) |
4713 | __getparam_dl(p, &attr); | |
4714 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4715 | attr.sched_priority = p->rt_priority; |
4716 | else | |
d0ea0268 | 4717 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4718 | |
4719 | rcu_read_unlock(); | |
4720 | ||
4721 | retval = sched_read_attr(uattr, &attr, size); | |
4722 | return retval; | |
4723 | ||
4724 | out_unlock: | |
4725 | rcu_read_unlock(); | |
4726 | return retval; | |
4727 | } | |
4728 | ||
96f874e2 | 4729 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4730 | { |
5a16f3d3 | 4731 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4732 | struct task_struct *p; |
4733 | int retval; | |
1da177e4 | 4734 | |
23f5d142 | 4735 | rcu_read_lock(); |
1da177e4 LT |
4736 | |
4737 | p = find_process_by_pid(pid); | |
4738 | if (!p) { | |
23f5d142 | 4739 | rcu_read_unlock(); |
1da177e4 LT |
4740 | return -ESRCH; |
4741 | } | |
4742 | ||
23f5d142 | 4743 | /* Prevent p going away */ |
1da177e4 | 4744 | get_task_struct(p); |
23f5d142 | 4745 | rcu_read_unlock(); |
1da177e4 | 4746 | |
14a40ffc TH |
4747 | if (p->flags & PF_NO_SETAFFINITY) { |
4748 | retval = -EINVAL; | |
4749 | goto out_put_task; | |
4750 | } | |
5a16f3d3 RR |
4751 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4752 | retval = -ENOMEM; | |
4753 | goto out_put_task; | |
4754 | } | |
4755 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4756 | retval = -ENOMEM; | |
4757 | goto out_free_cpus_allowed; | |
4758 | } | |
1da177e4 | 4759 | retval = -EPERM; |
4c44aaaf EB |
4760 | if (!check_same_owner(p)) { |
4761 | rcu_read_lock(); | |
4762 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4763 | rcu_read_unlock(); | |
16303ab2 | 4764 | goto out_free_new_mask; |
4c44aaaf EB |
4765 | } |
4766 | rcu_read_unlock(); | |
4767 | } | |
1da177e4 | 4768 | |
b0ae1981 | 4769 | retval = security_task_setscheduler(p); |
e7834f8f | 4770 | if (retval) |
16303ab2 | 4771 | goto out_free_new_mask; |
e7834f8f | 4772 | |
e4099a5e PZ |
4773 | |
4774 | cpuset_cpus_allowed(p, cpus_allowed); | |
4775 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4776 | ||
332ac17e DF |
4777 | /* |
4778 | * Since bandwidth control happens on root_domain basis, | |
4779 | * if admission test is enabled, we only admit -deadline | |
4780 | * tasks allowed to run on all the CPUs in the task's | |
4781 | * root_domain. | |
4782 | */ | |
4783 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4784 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4785 | rcu_read_lock(); | |
4786 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4787 | retval = -EBUSY; |
f1e3a093 | 4788 | rcu_read_unlock(); |
16303ab2 | 4789 | goto out_free_new_mask; |
332ac17e | 4790 | } |
f1e3a093 | 4791 | rcu_read_unlock(); |
332ac17e DF |
4792 | } |
4793 | #endif | |
49246274 | 4794 | again: |
25834c73 | 4795 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4796 | |
8707d8b8 | 4797 | if (!retval) { |
5a16f3d3 RR |
4798 | cpuset_cpus_allowed(p, cpus_allowed); |
4799 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4800 | /* |
4801 | * We must have raced with a concurrent cpuset | |
4802 | * update. Just reset the cpus_allowed to the | |
4803 | * cpuset's cpus_allowed | |
4804 | */ | |
5a16f3d3 | 4805 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4806 | goto again; |
4807 | } | |
4808 | } | |
16303ab2 | 4809 | out_free_new_mask: |
5a16f3d3 RR |
4810 | free_cpumask_var(new_mask); |
4811 | out_free_cpus_allowed: | |
4812 | free_cpumask_var(cpus_allowed); | |
4813 | out_put_task: | |
1da177e4 | 4814 | put_task_struct(p); |
1da177e4 LT |
4815 | return retval; |
4816 | } | |
4817 | ||
4818 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4819 | struct cpumask *new_mask) |
1da177e4 | 4820 | { |
96f874e2 RR |
4821 | if (len < cpumask_size()) |
4822 | cpumask_clear(new_mask); | |
4823 | else if (len > cpumask_size()) | |
4824 | len = cpumask_size(); | |
4825 | ||
1da177e4 LT |
4826 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4827 | } | |
4828 | ||
4829 | /** | |
d1ccc66d | 4830 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
4831 | * @pid: pid of the process |
4832 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4833 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
4834 | * |
4835 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4836 | */ |
5add95d4 HC |
4837 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4838 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4839 | { |
5a16f3d3 | 4840 | cpumask_var_t new_mask; |
1da177e4 LT |
4841 | int retval; |
4842 | ||
5a16f3d3 RR |
4843 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4844 | return -ENOMEM; | |
1da177e4 | 4845 | |
5a16f3d3 RR |
4846 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4847 | if (retval == 0) | |
4848 | retval = sched_setaffinity(pid, new_mask); | |
4849 | free_cpumask_var(new_mask); | |
4850 | return retval; | |
1da177e4 LT |
4851 | } |
4852 | ||
96f874e2 | 4853 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4854 | { |
36c8b586 | 4855 | struct task_struct *p; |
31605683 | 4856 | unsigned long flags; |
1da177e4 | 4857 | int retval; |
1da177e4 | 4858 | |
23f5d142 | 4859 | rcu_read_lock(); |
1da177e4 LT |
4860 | |
4861 | retval = -ESRCH; | |
4862 | p = find_process_by_pid(pid); | |
4863 | if (!p) | |
4864 | goto out_unlock; | |
4865 | ||
e7834f8f DQ |
4866 | retval = security_task_getscheduler(p); |
4867 | if (retval) | |
4868 | goto out_unlock; | |
4869 | ||
013fdb80 | 4870 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4871 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4872 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4873 | |
4874 | out_unlock: | |
23f5d142 | 4875 | rcu_read_unlock(); |
1da177e4 | 4876 | |
9531b62f | 4877 | return retval; |
1da177e4 LT |
4878 | } |
4879 | ||
4880 | /** | |
d1ccc66d | 4881 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
4882 | * @pid: pid of the process |
4883 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4884 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 4885 | * |
599b4840 ZW |
4886 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4887 | * error code otherwise. | |
1da177e4 | 4888 | */ |
5add95d4 HC |
4889 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4890 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4891 | { |
4892 | int ret; | |
f17c8607 | 4893 | cpumask_var_t mask; |
1da177e4 | 4894 | |
84fba5ec | 4895 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4896 | return -EINVAL; |
4897 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4898 | return -EINVAL; |
4899 | ||
f17c8607 RR |
4900 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4901 | return -ENOMEM; | |
1da177e4 | 4902 | |
f17c8607 RR |
4903 | ret = sched_getaffinity(pid, mask); |
4904 | if (ret == 0) { | |
8bc037fb | 4905 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4906 | |
4907 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4908 | ret = -EFAULT; |
4909 | else | |
cd3d8031 | 4910 | ret = retlen; |
f17c8607 RR |
4911 | } |
4912 | free_cpumask_var(mask); | |
1da177e4 | 4913 | |
f17c8607 | 4914 | return ret; |
1da177e4 LT |
4915 | } |
4916 | ||
4917 | /** | |
4918 | * sys_sched_yield - yield the current processor to other threads. | |
4919 | * | |
dd41f596 IM |
4920 | * This function yields the current CPU to other tasks. If there are no |
4921 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4922 | * |
4923 | * Return: 0. | |
1da177e4 | 4924 | */ |
5add95d4 | 4925 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4926 | { |
8a8c69c3 PZ |
4927 | struct rq_flags rf; |
4928 | struct rq *rq; | |
4929 | ||
4930 | local_irq_disable(); | |
4931 | rq = this_rq(); | |
4932 | rq_lock(rq, &rf); | |
1da177e4 | 4933 | |
ae92882e | 4934 | schedstat_inc(rq->yld_count); |
4530d7ab | 4935 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4936 | |
4937 | /* | |
4938 | * Since we are going to call schedule() anyway, there's | |
4939 | * no need to preempt or enable interrupts: | |
4940 | */ | |
8a8c69c3 PZ |
4941 | preempt_disable(); |
4942 | rq_unlock(rq, &rf); | |
ba74c144 | 4943 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4944 | |
4945 | schedule(); | |
4946 | ||
4947 | return 0; | |
4948 | } | |
4949 | ||
35a773a0 | 4950 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4951 | int __sched _cond_resched(void) |
1da177e4 | 4952 | { |
fe32d3cd | 4953 | if (should_resched(0)) { |
a18b5d01 | 4954 | preempt_schedule_common(); |
1da177e4 LT |
4955 | return 1; |
4956 | } | |
4957 | return 0; | |
4958 | } | |
02b67cc3 | 4959 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4960 | #endif |
1da177e4 LT |
4961 | |
4962 | /* | |
613afbf8 | 4963 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4964 | * call schedule, and on return reacquire the lock. |
4965 | * | |
41a2d6cf | 4966 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4967 | * operations here to prevent schedule() from being called twice (once via |
4968 | * spin_unlock(), once by hand). | |
4969 | */ | |
613afbf8 | 4970 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4971 | { |
fe32d3cd | 4972 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4973 | int ret = 0; |
4974 | ||
f607c668 PZ |
4975 | lockdep_assert_held(lock); |
4976 | ||
4a81e832 | 4977 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4978 | spin_unlock(lock); |
d86ee480 | 4979 | if (resched) |
a18b5d01 | 4980 | preempt_schedule_common(); |
95c354fe NP |
4981 | else |
4982 | cpu_relax(); | |
6df3cecb | 4983 | ret = 1; |
1da177e4 | 4984 | spin_lock(lock); |
1da177e4 | 4985 | } |
6df3cecb | 4986 | return ret; |
1da177e4 | 4987 | } |
613afbf8 | 4988 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4989 | |
613afbf8 | 4990 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4991 | { |
4992 | BUG_ON(!in_softirq()); | |
4993 | ||
fe32d3cd | 4994 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4995 | local_bh_enable(); |
a18b5d01 | 4996 | preempt_schedule_common(); |
1da177e4 LT |
4997 | local_bh_disable(); |
4998 | return 1; | |
4999 | } | |
5000 | return 0; | |
5001 | } | |
613afbf8 | 5002 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5003 | |
1da177e4 LT |
5004 | /** |
5005 | * yield - yield the current processor to other threads. | |
5006 | * | |
8e3fabfd PZ |
5007 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
5008 | * | |
5009 | * The scheduler is at all times free to pick the calling task as the most | |
5010 | * eligible task to run, if removing the yield() call from your code breaks | |
5011 | * it, its already broken. | |
5012 | * | |
5013 | * Typical broken usage is: | |
5014 | * | |
5015 | * while (!event) | |
d1ccc66d | 5016 | * yield(); |
8e3fabfd PZ |
5017 | * |
5018 | * where one assumes that yield() will let 'the other' process run that will | |
5019 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5020 | * happen. Never use yield() as a progress guarantee!! | |
5021 | * | |
5022 | * If you want to use yield() to wait for something, use wait_event(). | |
5023 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5024 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5025 | */ |
5026 | void __sched yield(void) | |
5027 | { | |
5028 | set_current_state(TASK_RUNNING); | |
5029 | sys_sched_yield(); | |
5030 | } | |
1da177e4 LT |
5031 | EXPORT_SYMBOL(yield); |
5032 | ||
d95f4122 MG |
5033 | /** |
5034 | * yield_to - yield the current processor to another thread in | |
5035 | * your thread group, or accelerate that thread toward the | |
5036 | * processor it's on. | |
16addf95 RD |
5037 | * @p: target task |
5038 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5039 | * |
5040 | * It's the caller's job to ensure that the target task struct | |
5041 | * can't go away on us before we can do any checks. | |
5042 | * | |
e69f6186 | 5043 | * Return: |
7b270f60 PZ |
5044 | * true (>0) if we indeed boosted the target task. |
5045 | * false (0) if we failed to boost the target. | |
5046 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5047 | */ |
fa93384f | 5048 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5049 | { |
5050 | struct task_struct *curr = current; | |
5051 | struct rq *rq, *p_rq; | |
5052 | unsigned long flags; | |
c3c18640 | 5053 | int yielded = 0; |
d95f4122 MG |
5054 | |
5055 | local_irq_save(flags); | |
5056 | rq = this_rq(); | |
5057 | ||
5058 | again: | |
5059 | p_rq = task_rq(p); | |
7b270f60 PZ |
5060 | /* |
5061 | * If we're the only runnable task on the rq and target rq also | |
5062 | * has only one task, there's absolutely no point in yielding. | |
5063 | */ | |
5064 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5065 | yielded = -ESRCH; | |
5066 | goto out_irq; | |
5067 | } | |
5068 | ||
d95f4122 | 5069 | double_rq_lock(rq, p_rq); |
39e24d8f | 5070 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5071 | double_rq_unlock(rq, p_rq); |
5072 | goto again; | |
5073 | } | |
5074 | ||
5075 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5076 | goto out_unlock; |
d95f4122 MG |
5077 | |
5078 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5079 | goto out_unlock; |
d95f4122 MG |
5080 | |
5081 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5082 | goto out_unlock; |
d95f4122 MG |
5083 | |
5084 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5085 | if (yielded) { |
ae92882e | 5086 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5087 | /* |
5088 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5089 | * fairness. | |
5090 | */ | |
5091 | if (preempt && rq != p_rq) | |
8875125e | 5092 | resched_curr(p_rq); |
6d1cafd8 | 5093 | } |
d95f4122 | 5094 | |
7b270f60 | 5095 | out_unlock: |
d95f4122 | 5096 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5097 | out_irq: |
d95f4122 MG |
5098 | local_irq_restore(flags); |
5099 | ||
7b270f60 | 5100 | if (yielded > 0) |
d95f4122 MG |
5101 | schedule(); |
5102 | ||
5103 | return yielded; | |
5104 | } | |
5105 | EXPORT_SYMBOL_GPL(yield_to); | |
5106 | ||
10ab5643 TH |
5107 | int io_schedule_prepare(void) |
5108 | { | |
5109 | int old_iowait = current->in_iowait; | |
5110 | ||
5111 | current->in_iowait = 1; | |
5112 | blk_schedule_flush_plug(current); | |
5113 | ||
5114 | return old_iowait; | |
5115 | } | |
5116 | ||
5117 | void io_schedule_finish(int token) | |
5118 | { | |
5119 | current->in_iowait = token; | |
5120 | } | |
5121 | ||
1da177e4 | 5122 | /* |
41a2d6cf | 5123 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5124 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5125 | */ |
1da177e4 LT |
5126 | long __sched io_schedule_timeout(long timeout) |
5127 | { | |
10ab5643 | 5128 | int token; |
1da177e4 LT |
5129 | long ret; |
5130 | ||
10ab5643 | 5131 | token = io_schedule_prepare(); |
1da177e4 | 5132 | ret = schedule_timeout(timeout); |
10ab5643 | 5133 | io_schedule_finish(token); |
9cff8ade | 5134 | |
1da177e4 LT |
5135 | return ret; |
5136 | } | |
9cff8ade | 5137 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5138 | |
10ab5643 TH |
5139 | void io_schedule(void) |
5140 | { | |
5141 | int token; | |
5142 | ||
5143 | token = io_schedule_prepare(); | |
5144 | schedule(); | |
5145 | io_schedule_finish(token); | |
5146 | } | |
5147 | EXPORT_SYMBOL(io_schedule); | |
5148 | ||
1da177e4 LT |
5149 | /** |
5150 | * sys_sched_get_priority_max - return maximum RT priority. | |
5151 | * @policy: scheduling class. | |
5152 | * | |
e69f6186 YB |
5153 | * Return: On success, this syscall returns the maximum |
5154 | * rt_priority that can be used by a given scheduling class. | |
5155 | * On failure, a negative error code is returned. | |
1da177e4 | 5156 | */ |
5add95d4 | 5157 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5158 | { |
5159 | int ret = -EINVAL; | |
5160 | ||
5161 | switch (policy) { | |
5162 | case SCHED_FIFO: | |
5163 | case SCHED_RR: | |
5164 | ret = MAX_USER_RT_PRIO-1; | |
5165 | break; | |
aab03e05 | 5166 | case SCHED_DEADLINE: |
1da177e4 | 5167 | case SCHED_NORMAL: |
b0a9499c | 5168 | case SCHED_BATCH: |
dd41f596 | 5169 | case SCHED_IDLE: |
1da177e4 LT |
5170 | ret = 0; |
5171 | break; | |
5172 | } | |
5173 | return ret; | |
5174 | } | |
5175 | ||
5176 | /** | |
5177 | * sys_sched_get_priority_min - return minimum RT priority. | |
5178 | * @policy: scheduling class. | |
5179 | * | |
e69f6186 YB |
5180 | * Return: On success, this syscall returns the minimum |
5181 | * rt_priority that can be used by a given scheduling class. | |
5182 | * On failure, a negative error code is returned. | |
1da177e4 | 5183 | */ |
5add95d4 | 5184 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5185 | { |
5186 | int ret = -EINVAL; | |
5187 | ||
5188 | switch (policy) { | |
5189 | case SCHED_FIFO: | |
5190 | case SCHED_RR: | |
5191 | ret = 1; | |
5192 | break; | |
aab03e05 | 5193 | case SCHED_DEADLINE: |
1da177e4 | 5194 | case SCHED_NORMAL: |
b0a9499c | 5195 | case SCHED_BATCH: |
dd41f596 | 5196 | case SCHED_IDLE: |
1da177e4 LT |
5197 | ret = 0; |
5198 | } | |
5199 | return ret; | |
5200 | } | |
5201 | ||
5202 | /** | |
5203 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5204 | * @pid: pid of the process. | |
5205 | * @interval: userspace pointer to the timeslice value. | |
5206 | * | |
5207 | * this syscall writes the default timeslice value of a given process | |
5208 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5209 | * |
5210 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5211 | * an error code. | |
1da177e4 | 5212 | */ |
17da2bd9 | 5213 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5214 | struct timespec __user *, interval) |
1da177e4 | 5215 | { |
36c8b586 | 5216 | struct task_struct *p; |
a4ec24b4 | 5217 | unsigned int time_slice; |
eb580751 PZ |
5218 | struct rq_flags rf; |
5219 | struct timespec t; | |
dba091b9 | 5220 | struct rq *rq; |
3a5c359a | 5221 | int retval; |
1da177e4 LT |
5222 | |
5223 | if (pid < 0) | |
3a5c359a | 5224 | return -EINVAL; |
1da177e4 LT |
5225 | |
5226 | retval = -ESRCH; | |
1a551ae7 | 5227 | rcu_read_lock(); |
1da177e4 LT |
5228 | p = find_process_by_pid(pid); |
5229 | if (!p) | |
5230 | goto out_unlock; | |
5231 | ||
5232 | retval = security_task_getscheduler(p); | |
5233 | if (retval) | |
5234 | goto out_unlock; | |
5235 | ||
eb580751 | 5236 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5237 | time_slice = 0; |
5238 | if (p->sched_class->get_rr_interval) | |
5239 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5240 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5241 | |
1a551ae7 | 5242 | rcu_read_unlock(); |
a4ec24b4 | 5243 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5244 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5245 | return retval; |
3a5c359a | 5246 | |
1da177e4 | 5247 | out_unlock: |
1a551ae7 | 5248 | rcu_read_unlock(); |
1da177e4 LT |
5249 | return retval; |
5250 | } | |
5251 | ||
7c731e0a | 5252 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5253 | |
82a1fcb9 | 5254 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5255 | { |
1da177e4 | 5256 | unsigned long free = 0; |
4e79752c | 5257 | int ppid; |
1f8a7633 | 5258 | unsigned long state = p->state; |
1da177e4 | 5259 | |
c930b2c0 IM |
5260 | /* Make sure the string lines up properly with the number of task states: */ |
5261 | BUILD_BUG_ON(sizeof(TASK_STATE_TO_CHAR_STR)-1 != ilog2(TASK_STATE_MAX)+1); | |
5262 | ||
38200502 TH |
5263 | if (!try_get_task_stack(p)) |
5264 | return; | |
1f8a7633 TH |
5265 | if (state) |
5266 | state = __ffs(state) + 1; | |
28d0686c | 5267 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5268 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
1da177e4 | 5269 | if (state == TASK_RUNNING) |
3df0fc5b | 5270 | printk(KERN_CONT " running task "); |
1da177e4 | 5271 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5272 | free = stack_not_used(p); |
1da177e4 | 5273 | #endif |
a90e984c | 5274 | ppid = 0; |
4e79752c | 5275 | rcu_read_lock(); |
a90e984c ON |
5276 | if (pid_alive(p)) |
5277 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5278 | rcu_read_unlock(); |
3df0fc5b | 5279 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5280 | task_pid_nr(p), ppid, |
aa47b7e0 | 5281 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5282 | |
3d1cb205 | 5283 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5284 | show_stack(p, NULL); |
38200502 | 5285 | put_task_stack(p); |
1da177e4 LT |
5286 | } |
5287 | ||
e59e2ae2 | 5288 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5289 | { |
36c8b586 | 5290 | struct task_struct *g, *p; |
1da177e4 | 5291 | |
4bd77321 | 5292 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5293 | printk(KERN_INFO |
5294 | " task PC stack pid father\n"); | |
1da177e4 | 5295 | #else |
3df0fc5b PZ |
5296 | printk(KERN_INFO |
5297 | " task PC stack pid father\n"); | |
1da177e4 | 5298 | #endif |
510f5acc | 5299 | rcu_read_lock(); |
5d07f420 | 5300 | for_each_process_thread(g, p) { |
1da177e4 LT |
5301 | /* |
5302 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5303 | * console might take a lot of time: |
57675cb9 AR |
5304 | * Also, reset softlockup watchdogs on all CPUs, because |
5305 | * another CPU might be blocked waiting for us to process | |
5306 | * an IPI. | |
1da177e4 LT |
5307 | */ |
5308 | touch_nmi_watchdog(); | |
57675cb9 | 5309 | touch_all_softlockup_watchdogs(); |
39bc89fd | 5310 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5311 | sched_show_task(p); |
5d07f420 | 5312 | } |
1da177e4 | 5313 | |
dd41f596 | 5314 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5315 | if (!state_filter) |
5316 | sysrq_sched_debug_show(); | |
dd41f596 | 5317 | #endif |
510f5acc | 5318 | rcu_read_unlock(); |
e59e2ae2 IM |
5319 | /* |
5320 | * Only show locks if all tasks are dumped: | |
5321 | */ | |
93335a21 | 5322 | if (!state_filter) |
e59e2ae2 | 5323 | debug_show_all_locks(); |
1da177e4 LT |
5324 | } |
5325 | ||
0db0628d | 5326 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5327 | { |
dd41f596 | 5328 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5329 | } |
5330 | ||
f340c0d1 IM |
5331 | /** |
5332 | * init_idle - set up an idle thread for a given CPU | |
5333 | * @idle: task in question | |
d1ccc66d | 5334 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
5335 | * |
5336 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5337 | * flag, to make booting more robust. | |
5338 | */ | |
0db0628d | 5339 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5340 | { |
70b97a7f | 5341 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5342 | unsigned long flags; |
5343 | ||
25834c73 PZ |
5344 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5345 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5346 | |
5e1576ed | 5347 | __sched_fork(0, idle); |
06b83b5f | 5348 | idle->state = TASK_RUNNING; |
dd41f596 | 5349 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5350 | idle->flags |= PF_IDLE; |
dd41f596 | 5351 | |
e1b77c92 MR |
5352 | kasan_unpoison_task_stack(idle); |
5353 | ||
de9b8f5d PZ |
5354 | #ifdef CONFIG_SMP |
5355 | /* | |
5356 | * Its possible that init_idle() gets called multiple times on a task, | |
5357 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5358 | * | |
5359 | * And since this is boot we can forgo the serialization. | |
5360 | */ | |
5361 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5362 | #endif | |
6506cf6c PZ |
5363 | /* |
5364 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 5365 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
5366 | * lockdep check in task_group() will fail. |
5367 | * | |
5368 | * Similar case to sched_fork(). / Alternatively we could | |
5369 | * use task_rq_lock() here and obtain the other rq->lock. | |
5370 | * | |
5371 | * Silence PROVE_RCU | |
5372 | */ | |
5373 | rcu_read_lock(); | |
dd41f596 | 5374 | __set_task_cpu(idle, cpu); |
6506cf6c | 5375 | rcu_read_unlock(); |
1da177e4 | 5376 | |
1da177e4 | 5377 | rq->curr = rq->idle = idle; |
da0c1e65 | 5378 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5379 | #ifdef CONFIG_SMP |
3ca7a440 | 5380 | idle->on_cpu = 1; |
4866cde0 | 5381 | #endif |
25834c73 PZ |
5382 | raw_spin_unlock(&rq->lock); |
5383 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5384 | |
5385 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5386 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5387 | |
dd41f596 IM |
5388 | /* |
5389 | * The idle tasks have their own, simple scheduling class: | |
5390 | */ | |
5391 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5392 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5393 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5394 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5395 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5396 | #endif | |
19978ca6 IM |
5397 | } |
5398 | ||
f82f8042 JL |
5399 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5400 | const struct cpumask *trial) | |
5401 | { | |
5402 | int ret = 1, trial_cpus; | |
5403 | struct dl_bw *cur_dl_b; | |
5404 | unsigned long flags; | |
5405 | ||
bb2bc55a MG |
5406 | if (!cpumask_weight(cur)) |
5407 | return ret; | |
5408 | ||
75e23e49 | 5409 | rcu_read_lock_sched(); |
f82f8042 JL |
5410 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5411 | trial_cpus = cpumask_weight(trial); | |
5412 | ||
5413 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5414 | if (cur_dl_b->bw != -1 && | |
5415 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5416 | ret = 0; | |
5417 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5418 | rcu_read_unlock_sched(); |
f82f8042 JL |
5419 | |
5420 | return ret; | |
5421 | } | |
5422 | ||
7f51412a JL |
5423 | int task_can_attach(struct task_struct *p, |
5424 | const struct cpumask *cs_cpus_allowed) | |
5425 | { | |
5426 | int ret = 0; | |
5427 | ||
5428 | /* | |
5429 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 5430 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
5431 | * affinity and isolating such threads by their set of |
5432 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5433 | * applicable for such threads. This prevents checking for | |
5434 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5435 | * before cpus_allowed may be changed. | |
5436 | */ | |
5437 | if (p->flags & PF_NO_SETAFFINITY) { | |
5438 | ret = -EINVAL; | |
5439 | goto out; | |
5440 | } | |
5441 | ||
5442 | #ifdef CONFIG_SMP | |
5443 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5444 | cs_cpus_allowed)) { | |
5445 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5446 | cs_cpus_allowed); | |
75e23e49 | 5447 | struct dl_bw *dl_b; |
7f51412a JL |
5448 | bool overflow; |
5449 | int cpus; | |
5450 | unsigned long flags; | |
5451 | ||
75e23e49 JL |
5452 | rcu_read_lock_sched(); |
5453 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5454 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5455 | cpus = dl_bw_cpus(dest_cpu); | |
5456 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5457 | if (overflow) | |
5458 | ret = -EBUSY; | |
5459 | else { | |
5460 | /* | |
5461 | * We reserve space for this task in the destination | |
5462 | * root_domain, as we can't fail after this point. | |
5463 | * We will free resources in the source root_domain | |
5464 | * later on (see set_cpus_allowed_dl()). | |
5465 | */ | |
5466 | __dl_add(dl_b, p->dl.dl_bw); | |
5467 | } | |
5468 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5469 | rcu_read_unlock_sched(); |
7f51412a JL |
5470 | |
5471 | } | |
5472 | #endif | |
5473 | out: | |
5474 | return ret; | |
5475 | } | |
5476 | ||
1da177e4 | 5477 | #ifdef CONFIG_SMP |
1da177e4 | 5478 | |
f2cb1360 | 5479 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 5480 | |
e6628d5b MG |
5481 | #ifdef CONFIG_NUMA_BALANCING |
5482 | /* Migrate current task p to target_cpu */ | |
5483 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5484 | { | |
5485 | struct migration_arg arg = { p, target_cpu }; | |
5486 | int curr_cpu = task_cpu(p); | |
5487 | ||
5488 | if (curr_cpu == target_cpu) | |
5489 | return 0; | |
5490 | ||
0c98d344 | 5491 | if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed)) |
e6628d5b MG |
5492 | return -EINVAL; |
5493 | ||
5494 | /* TODO: This is not properly updating schedstats */ | |
5495 | ||
286549dc | 5496 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5497 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5498 | } | |
0ec8aa00 PZ |
5499 | |
5500 | /* | |
5501 | * Requeue a task on a given node and accurately track the number of NUMA | |
5502 | * tasks on the runqueues | |
5503 | */ | |
5504 | void sched_setnuma(struct task_struct *p, int nid) | |
5505 | { | |
da0c1e65 | 5506 | bool queued, running; |
eb580751 PZ |
5507 | struct rq_flags rf; |
5508 | struct rq *rq; | |
0ec8aa00 | 5509 | |
eb580751 | 5510 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5511 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5512 | running = task_current(rq, p); |
5513 | ||
da0c1e65 | 5514 | if (queued) |
1de64443 | 5515 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5516 | if (running) |
f3cd1c4e | 5517 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5518 | |
5519 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5520 | |
da0c1e65 | 5521 | if (queued) |
7134b3e9 | 5522 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 5523 | if (running) |
b2bf6c31 | 5524 | set_curr_task(rq, p); |
eb580751 | 5525 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5526 | } |
5cc389bc | 5527 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5528 | |
1da177e4 | 5529 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5530 | /* |
d1ccc66d | 5531 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 5532 | * offline. |
054b9108 | 5533 | */ |
48c5ccae | 5534 | void idle_task_exit(void) |
1da177e4 | 5535 | { |
48c5ccae | 5536 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5537 | |
48c5ccae | 5538 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5539 | |
a53efe5f | 5540 | if (mm != &init_mm) { |
f98db601 | 5541 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5542 | finish_arch_post_lock_switch(); |
5543 | } | |
48c5ccae | 5544 | mmdrop(mm); |
1da177e4 LT |
5545 | } |
5546 | ||
5547 | /* | |
5d180232 PZ |
5548 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5549 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5550 | * nr_active count is stable. We need to take the teardown thread which |
5551 | * is calling this into account, so we hand in adjust = 1 to the load | |
5552 | * calculation. | |
5d180232 PZ |
5553 | * |
5554 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5555 | */ |
5d180232 | 5556 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5557 | { |
d60585c5 | 5558 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5559 | if (delta) |
5560 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5561 | } |
5562 | ||
3f1d2a31 PZ |
5563 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5564 | { | |
5565 | } | |
5566 | ||
5567 | static const struct sched_class fake_sched_class = { | |
5568 | .put_prev_task = put_prev_task_fake, | |
5569 | }; | |
5570 | ||
5571 | static struct task_struct fake_task = { | |
5572 | /* | |
5573 | * Avoid pull_{rt,dl}_task() | |
5574 | */ | |
5575 | .prio = MAX_PRIO + 1, | |
5576 | .sched_class = &fake_sched_class, | |
5577 | }; | |
5578 | ||
48f24c4d | 5579 | /* |
48c5ccae PZ |
5580 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5581 | * try_to_wake_up()->select_task_rq(). | |
5582 | * | |
5583 | * Called with rq->lock held even though we'er in stop_machine() and | |
5584 | * there's no concurrency possible, we hold the required locks anyway | |
5585 | * because of lock validation efforts. | |
1da177e4 | 5586 | */ |
8a8c69c3 | 5587 | static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf) |
1da177e4 | 5588 | { |
5e16bbc2 | 5589 | struct rq *rq = dead_rq; |
48c5ccae | 5590 | struct task_struct *next, *stop = rq->stop; |
8a8c69c3 | 5591 | struct rq_flags orf = *rf; |
48c5ccae | 5592 | int dest_cpu; |
1da177e4 LT |
5593 | |
5594 | /* | |
48c5ccae PZ |
5595 | * Fudge the rq selection such that the below task selection loop |
5596 | * doesn't get stuck on the currently eligible stop task. | |
5597 | * | |
5598 | * We're currently inside stop_machine() and the rq is either stuck | |
5599 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5600 | * either way we should never end up calling schedule() until we're | |
5601 | * done here. | |
1da177e4 | 5602 | */ |
48c5ccae | 5603 | rq->stop = NULL; |
48f24c4d | 5604 | |
77bd3970 FW |
5605 | /* |
5606 | * put_prev_task() and pick_next_task() sched | |
5607 | * class method both need to have an up-to-date | |
5608 | * value of rq->clock[_task] | |
5609 | */ | |
5610 | update_rq_clock(rq); | |
5611 | ||
5e16bbc2 | 5612 | for (;;) { |
48c5ccae PZ |
5613 | /* |
5614 | * There's this thread running, bail when that's the only | |
d1ccc66d | 5615 | * remaining thread: |
48c5ccae PZ |
5616 | */ |
5617 | if (rq->nr_running == 1) | |
dd41f596 | 5618 | break; |
48c5ccae | 5619 | |
cbce1a68 | 5620 | /* |
d1ccc66d | 5621 | * pick_next_task() assumes pinned rq->lock: |
cbce1a68 | 5622 | */ |
8a8c69c3 | 5623 | next = pick_next_task(rq, &fake_task, rf); |
48c5ccae | 5624 | BUG_ON(!next); |
79c53799 | 5625 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5626 | |
5473e0cc WL |
5627 | /* |
5628 | * Rules for changing task_struct::cpus_allowed are holding | |
5629 | * both pi_lock and rq->lock, such that holding either | |
5630 | * stabilizes the mask. | |
5631 | * | |
5632 | * Drop rq->lock is not quite as disastrous as it usually is | |
5633 | * because !cpu_active at this point, which means load-balance | |
5634 | * will not interfere. Also, stop-machine. | |
5635 | */ | |
8a8c69c3 | 5636 | rq_unlock(rq, rf); |
5473e0cc | 5637 | raw_spin_lock(&next->pi_lock); |
8a8c69c3 | 5638 | rq_relock(rq, rf); |
5473e0cc WL |
5639 | |
5640 | /* | |
5641 | * Since we're inside stop-machine, _nothing_ should have | |
5642 | * changed the task, WARN if weird stuff happened, because in | |
5643 | * that case the above rq->lock drop is a fail too. | |
5644 | */ | |
5645 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5646 | raw_spin_unlock(&next->pi_lock); | |
5647 | continue; | |
5648 | } | |
5649 | ||
48c5ccae | 5650 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5651 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
8a8c69c3 | 5652 | rq = __migrate_task(rq, rf, next, dest_cpu); |
5e16bbc2 | 5653 | if (rq != dead_rq) { |
8a8c69c3 | 5654 | rq_unlock(rq, rf); |
5e16bbc2 | 5655 | rq = dead_rq; |
8a8c69c3 PZ |
5656 | *rf = orf; |
5657 | rq_relock(rq, rf); | |
5e16bbc2 | 5658 | } |
5473e0cc | 5659 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5660 | } |
dce48a84 | 5661 | |
48c5ccae | 5662 | rq->stop = stop; |
dce48a84 | 5663 | } |
1da177e4 LT |
5664 | #endif /* CONFIG_HOTPLUG_CPU */ |
5665 | ||
f2cb1360 | 5666 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
5667 | { |
5668 | if (!rq->online) { | |
5669 | const struct sched_class *class; | |
5670 | ||
c6c4927b | 5671 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5672 | rq->online = 1; |
5673 | ||
5674 | for_each_class(class) { | |
5675 | if (class->rq_online) | |
5676 | class->rq_online(rq); | |
5677 | } | |
5678 | } | |
5679 | } | |
5680 | ||
f2cb1360 | 5681 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
5682 | { |
5683 | if (rq->online) { | |
5684 | const struct sched_class *class; | |
5685 | ||
5686 | for_each_class(class) { | |
5687 | if (class->rq_offline) | |
5688 | class->rq_offline(rq); | |
5689 | } | |
5690 | ||
c6c4927b | 5691 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5692 | rq->online = 0; |
5693 | } | |
5694 | } | |
5695 | ||
9cf7243d | 5696 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5697 | { |
969c7921 | 5698 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5699 | |
a803f026 CM |
5700 | rq->age_stamp = sched_clock_cpu(cpu); |
5701 | } | |
5702 | ||
d1ccc66d IM |
5703 | /* |
5704 | * used to mark begin/end of suspend/resume: | |
5705 | */ | |
5706 | static int num_cpus_frozen; | |
d35be8ba | 5707 | |
1da177e4 | 5708 | /* |
3a101d05 TH |
5709 | * Update cpusets according to cpu_active mask. If cpusets are |
5710 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
5711 | * around partition_sched_domains(). | |
d35be8ba SB |
5712 | * |
5713 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
5714 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 5715 | */ |
40190a78 | 5716 | static void cpuset_cpu_active(void) |
e761b772 | 5717 | { |
40190a78 | 5718 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
5719 | /* |
5720 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
5721 | * resume sequence. As long as this is not the last online | |
5722 | * operation in the resume sequence, just build a single sched | |
5723 | * domain, ignoring cpusets. | |
5724 | */ | |
5725 | num_cpus_frozen--; | |
5726 | if (likely(num_cpus_frozen)) { | |
5727 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 5728 | return; |
d35be8ba | 5729 | } |
d35be8ba SB |
5730 | /* |
5731 | * This is the last CPU online operation. So fall through and | |
5732 | * restore the original sched domains by considering the | |
5733 | * cpuset configurations. | |
5734 | */ | |
3a101d05 | 5735 | } |
135fb3e1 | 5736 | cpuset_update_active_cpus(true); |
3a101d05 | 5737 | } |
e761b772 | 5738 | |
40190a78 | 5739 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 5740 | { |
3c18d447 | 5741 | unsigned long flags; |
3c18d447 | 5742 | struct dl_bw *dl_b; |
533445c6 OS |
5743 | bool overflow; |
5744 | int cpus; | |
3c18d447 | 5745 | |
40190a78 | 5746 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
5747 | rcu_read_lock_sched(); |
5748 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 5749 | |
533445c6 OS |
5750 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5751 | cpus = dl_bw_cpus(cpu); | |
5752 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5753 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 5754 | |
533445c6 | 5755 | rcu_read_unlock_sched(); |
3c18d447 | 5756 | |
533445c6 | 5757 | if (overflow) |
135fb3e1 | 5758 | return -EBUSY; |
7ddf96b0 | 5759 | cpuset_update_active_cpus(false); |
135fb3e1 | 5760 | } else { |
d35be8ba SB |
5761 | num_cpus_frozen++; |
5762 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 5763 | } |
135fb3e1 | 5764 | return 0; |
e761b772 | 5765 | } |
e761b772 | 5766 | |
40190a78 | 5767 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 5768 | { |
7d976699 | 5769 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 5770 | struct rq_flags rf; |
7d976699 | 5771 | |
40190a78 | 5772 | set_cpu_active(cpu, true); |
135fb3e1 | 5773 | |
40190a78 | 5774 | if (sched_smp_initialized) { |
135fb3e1 | 5775 | sched_domains_numa_masks_set(cpu); |
40190a78 | 5776 | cpuset_cpu_active(); |
e761b772 | 5777 | } |
7d976699 TG |
5778 | |
5779 | /* | |
5780 | * Put the rq online, if not already. This happens: | |
5781 | * | |
5782 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 5783 | * after all CPUs have been brought up. |
7d976699 TG |
5784 | * |
5785 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
5786 | * domains. | |
5787 | */ | |
8a8c69c3 | 5788 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
5789 | if (rq->rd) { |
5790 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5791 | set_rq_online(rq); | |
5792 | } | |
8a8c69c3 | 5793 | rq_unlock_irqrestore(rq, &rf); |
7d976699 TG |
5794 | |
5795 | update_max_interval(); | |
5796 | ||
40190a78 | 5797 | return 0; |
135fb3e1 TG |
5798 | } |
5799 | ||
40190a78 | 5800 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 5801 | { |
135fb3e1 TG |
5802 | int ret; |
5803 | ||
40190a78 | 5804 | set_cpu_active(cpu, false); |
b2454caa PZ |
5805 | /* |
5806 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
5807 | * users of this state to go away such that all new such users will | |
5808 | * observe it. | |
5809 | * | |
5810 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
5811 | * not imply sync_sched(), so wait for both. | |
5812 | * | |
5813 | * Do sync before park smpboot threads to take care the rcu boost case. | |
5814 | */ | |
5815 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
5816 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
5817 | else | |
5818 | synchronize_rcu(); | |
40190a78 TG |
5819 | |
5820 | if (!sched_smp_initialized) | |
5821 | return 0; | |
5822 | ||
5823 | ret = cpuset_cpu_inactive(cpu); | |
5824 | if (ret) { | |
5825 | set_cpu_active(cpu, true); | |
5826 | return ret; | |
135fb3e1 | 5827 | } |
40190a78 TG |
5828 | sched_domains_numa_masks_clear(cpu); |
5829 | return 0; | |
135fb3e1 TG |
5830 | } |
5831 | ||
94baf7a5 TG |
5832 | static void sched_rq_cpu_starting(unsigned int cpu) |
5833 | { | |
5834 | struct rq *rq = cpu_rq(cpu); | |
5835 | ||
5836 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
5837 | update_max_interval(); |
5838 | } | |
5839 | ||
135fb3e1 TG |
5840 | int sched_cpu_starting(unsigned int cpu) |
5841 | { | |
5842 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 5843 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 5844 | return 0; |
e761b772 | 5845 | } |
e761b772 | 5846 | |
f2785ddb TG |
5847 | #ifdef CONFIG_HOTPLUG_CPU |
5848 | int sched_cpu_dying(unsigned int cpu) | |
5849 | { | |
5850 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 5851 | struct rq_flags rf; |
f2785ddb TG |
5852 | |
5853 | /* Handle pending wakeups and then migrate everything off */ | |
5854 | sched_ttwu_pending(); | |
8a8c69c3 PZ |
5855 | |
5856 | rq_lock_irqsave(rq, &rf); | |
f2785ddb TG |
5857 | if (rq->rd) { |
5858 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5859 | set_rq_offline(rq); | |
5860 | } | |
8a8c69c3 | 5861 | migrate_tasks(rq, &rf); |
f2785ddb | 5862 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
5863 | rq_unlock_irqrestore(rq, &rf); |
5864 | ||
f2785ddb TG |
5865 | calc_load_migrate(rq); |
5866 | update_max_interval(); | |
20a5c8cc | 5867 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 5868 | hrtick_clear(rq); |
f2785ddb TG |
5869 | return 0; |
5870 | } | |
5871 | #endif | |
5872 | ||
1b568f0a PZ |
5873 | #ifdef CONFIG_SCHED_SMT |
5874 | DEFINE_STATIC_KEY_FALSE(sched_smt_present); | |
5875 | ||
5876 | static void sched_init_smt(void) | |
5877 | { | |
5878 | /* | |
5879 | * We've enumerated all CPUs and will assume that if any CPU | |
5880 | * has SMT siblings, CPU0 will too. | |
5881 | */ | |
5882 | if (cpumask_weight(cpu_smt_mask(0)) > 1) | |
5883 | static_branch_enable(&sched_smt_present); | |
5884 | } | |
5885 | #else | |
5886 | static inline void sched_init_smt(void) { } | |
5887 | #endif | |
5888 | ||
1da177e4 LT |
5889 | void __init sched_init_smp(void) |
5890 | { | |
dcc30a35 RR |
5891 | cpumask_var_t non_isolated_cpus; |
5892 | ||
5893 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 5894 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 5895 | |
cb83b629 PZ |
5896 | sched_init_numa(); |
5897 | ||
6acce3ef PZ |
5898 | /* |
5899 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 5900 | * CPU masks are stable and all blatant races in the below code cannot |
6acce3ef PZ |
5901 | * happen. |
5902 | */ | |
712555ee | 5903 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 5904 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
5905 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
5906 | if (cpumask_empty(non_isolated_cpus)) | |
5907 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 5908 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 5909 | |
5c1e1767 | 5910 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 5911 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 5912 | BUG(); |
19978ca6 | 5913 | sched_init_granularity(); |
dcc30a35 | 5914 | free_cpumask_var(non_isolated_cpus); |
4212823f | 5915 | |
0e3900e6 | 5916 | init_sched_rt_class(); |
1baca4ce | 5917 | init_sched_dl_class(); |
1b568f0a PZ |
5918 | |
5919 | sched_init_smt(); | |
9881b024 | 5920 | sched_clock_init_late(); |
1b568f0a | 5921 | |
e26fbffd | 5922 | sched_smp_initialized = true; |
1da177e4 | 5923 | } |
e26fbffd TG |
5924 | |
5925 | static int __init migration_init(void) | |
5926 | { | |
94baf7a5 | 5927 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 5928 | return 0; |
1da177e4 | 5929 | } |
e26fbffd TG |
5930 | early_initcall(migration_init); |
5931 | ||
1da177e4 LT |
5932 | #else |
5933 | void __init sched_init_smp(void) | |
5934 | { | |
19978ca6 | 5935 | sched_init_granularity(); |
9881b024 | 5936 | sched_clock_init_late(); |
1da177e4 LT |
5937 | } |
5938 | #endif /* CONFIG_SMP */ | |
5939 | ||
5940 | int in_sched_functions(unsigned long addr) | |
5941 | { | |
1da177e4 LT |
5942 | return in_lock_functions(addr) || |
5943 | (addr >= (unsigned long)__sched_text_start | |
5944 | && addr < (unsigned long)__sched_text_end); | |
5945 | } | |
5946 | ||
029632fb | 5947 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
5948 | /* |
5949 | * Default task group. | |
5950 | * Every task in system belongs to this group at bootup. | |
5951 | */ | |
029632fb | 5952 | struct task_group root_task_group; |
35cf4e50 | 5953 | LIST_HEAD(task_groups); |
b0367629 WL |
5954 | |
5955 | /* Cacheline aligned slab cache for task_group */ | |
5956 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 5957 | #endif |
6f505b16 | 5958 | |
e6252c3e | 5959 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 5960 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 5961 | |
9dcb8b68 LT |
5962 | #define WAIT_TABLE_BITS 8 |
5963 | #define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS) | |
5964 | static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned; | |
5965 | ||
5966 | wait_queue_head_t *bit_waitqueue(void *word, int bit) | |
5967 | { | |
5968 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; | |
5969 | unsigned long val = (unsigned long)word << shift | bit; | |
5970 | ||
5971 | return bit_wait_table + hash_long(val, WAIT_TABLE_BITS); | |
5972 | } | |
5973 | EXPORT_SYMBOL(bit_waitqueue); | |
5974 | ||
1da177e4 LT |
5975 | void __init sched_init(void) |
5976 | { | |
dd41f596 | 5977 | int i, j; |
434d53b0 MT |
5978 | unsigned long alloc_size = 0, ptr; |
5979 | ||
9881b024 PZ |
5980 | sched_clock_init(); |
5981 | ||
9dcb8b68 LT |
5982 | for (i = 0; i < WAIT_TABLE_SIZE; i++) |
5983 | init_waitqueue_head(bit_wait_table + i); | |
5984 | ||
434d53b0 MT |
5985 | #ifdef CONFIG_FAIR_GROUP_SCHED |
5986 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5987 | #endif | |
5988 | #ifdef CONFIG_RT_GROUP_SCHED | |
5989 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5990 | #endif | |
434d53b0 | 5991 | if (alloc_size) { |
36b7b6d4 | 5992 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
5993 | |
5994 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 5995 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
5996 | ptr += nr_cpu_ids * sizeof(void **); |
5997 | ||
07e06b01 | 5998 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 5999 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6000 | |
6d6bc0ad | 6001 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6002 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6003 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6004 | ptr += nr_cpu_ids * sizeof(void **); |
6005 | ||
07e06b01 | 6006 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6007 | ptr += nr_cpu_ids * sizeof(void **); |
6008 | ||
6d6bc0ad | 6009 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 6010 | } |
df7c8e84 | 6011 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
6012 | for_each_possible_cpu(i) { |
6013 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
6014 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
6015 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
6016 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 6017 | } |
b74e6278 | 6018 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 6019 | |
d1ccc66d IM |
6020 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
6021 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 6022 | |
57d885fe GH |
6023 | #ifdef CONFIG_SMP |
6024 | init_defrootdomain(); | |
6025 | #endif | |
6026 | ||
d0b27fa7 | 6027 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6028 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6029 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6030 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6031 | |
7c941438 | 6032 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
6033 | task_group_cache = KMEM_CACHE(task_group, 0); |
6034 | ||
07e06b01 YZ |
6035 | list_add(&root_task_group.list, &task_groups); |
6036 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6037 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6038 | autogroup_init(&init_task); |
7c941438 | 6039 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6040 | |
0a945022 | 6041 | for_each_possible_cpu(i) { |
70b97a7f | 6042 | struct rq *rq; |
1da177e4 LT |
6043 | |
6044 | rq = cpu_rq(i); | |
05fa785c | 6045 | raw_spin_lock_init(&rq->lock); |
7897986b | 6046 | rq->nr_running = 0; |
dce48a84 TG |
6047 | rq->calc_load_active = 0; |
6048 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6049 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
6050 | init_rt_rq(&rq->rt); |
6051 | init_dl_rq(&rq->dl); | |
dd41f596 | 6052 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6053 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6054 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 6055 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 6056 | /* |
d1ccc66d | 6057 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
6058 | * |
6059 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
6060 | * gets 100% of the CPU resources in the system. This overall |
6061 | * system CPU resource is divided among the tasks of | |
07e06b01 | 6062 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6063 | * based on each entity's (task or task-group's) weight |
6064 | * (se->load.weight). | |
6065 | * | |
07e06b01 | 6066 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 6067 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 6068 | * then A0's share of the CPU resource is: |
354d60c2 | 6069 | * |
0d905bca | 6070 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6071 | * |
07e06b01 YZ |
6072 | * We achieve this by letting root_task_group's tasks sit |
6073 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6074 | */ |
ab84d31e | 6075 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6076 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6077 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6078 | ||
6079 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6080 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6081 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6082 | #endif |
1da177e4 | 6083 | |
dd41f596 IM |
6084 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6085 | rq->cpu_load[j] = 0; | |
fdf3e95d | 6086 | |
1da177e4 | 6087 | #ifdef CONFIG_SMP |
41c7ce9a | 6088 | rq->sd = NULL; |
57d885fe | 6089 | rq->rd = NULL; |
ca6d75e6 | 6090 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 6091 | rq->balance_callback = NULL; |
1da177e4 | 6092 | rq->active_balance = 0; |
dd41f596 | 6093 | rq->next_balance = jiffies; |
1da177e4 | 6094 | rq->push_cpu = 0; |
0a2966b4 | 6095 | rq->cpu = i; |
1f11eb6a | 6096 | rq->online = 0; |
eae0c9df MG |
6097 | rq->idle_stamp = 0; |
6098 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6099 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6100 | |
6101 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6102 | ||
dc938520 | 6103 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6104 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 6105 | rq->last_load_update_tick = jiffies; |
1c792db7 | 6106 | rq->nohz_flags = 0; |
83cd4fe2 | 6107 | #endif |
265f22a9 FW |
6108 | #ifdef CONFIG_NO_HZ_FULL |
6109 | rq->last_sched_tick = 0; | |
6110 | #endif | |
9fd81dd5 | 6111 | #endif /* CONFIG_SMP */ |
8f4d37ec | 6112 | init_rq_hrtick(rq); |
1da177e4 | 6113 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6114 | } |
6115 | ||
2dd73a4f | 6116 | set_load_weight(&init_task); |
b50f60ce | 6117 | |
1da177e4 LT |
6118 | /* |
6119 | * The boot idle thread does lazy MMU switching as well: | |
6120 | */ | |
f1f10076 | 6121 | mmgrab(&init_mm); |
1da177e4 LT |
6122 | enter_lazy_tlb(&init_mm, current); |
6123 | ||
6124 | /* | |
6125 | * Make us the idle thread. Technically, schedule() should not be | |
6126 | * called from this thread, however somewhere below it might be, | |
6127 | * but because we are the idle thread, we just pick up running again | |
6128 | * when this runqueue becomes "idle". | |
6129 | */ | |
6130 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6131 | |
6132 | calc_load_update = jiffies + LOAD_FREQ; | |
6133 | ||
bf4d83f6 | 6134 | #ifdef CONFIG_SMP |
4cb98839 | 6135 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6136 | /* May be allocated at isolcpus cmdline parse time */ |
6137 | if (cpu_isolated_map == NULL) | |
6138 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6139 | idle_thread_set_boot_cpu(); |
9cf7243d | 6140 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
6141 | #endif |
6142 | init_sched_fair_class(); | |
6a7b3dc3 | 6143 | |
4698f88c JP |
6144 | init_schedstats(); |
6145 | ||
6892b75e | 6146 | scheduler_running = 1; |
1da177e4 LT |
6147 | } |
6148 | ||
d902db1e | 6149 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6150 | static inline int preempt_count_equals(int preempt_offset) |
6151 | { | |
da7142e2 | 6152 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 6153 | |
4ba8216c | 6154 | return (nested == preempt_offset); |
e4aafea2 FW |
6155 | } |
6156 | ||
d894837f | 6157 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6158 | { |
8eb23b9f PZ |
6159 | /* |
6160 | * Blocking primitives will set (and therefore destroy) current->state, | |
6161 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
6162 | * otherwise we will destroy state. | |
6163 | */ | |
00845eb9 | 6164 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
6165 | "do not call blocking ops when !TASK_RUNNING; " |
6166 | "state=%lx set at [<%p>] %pS\n", | |
6167 | current->state, | |
6168 | (void *)current->task_state_change, | |
00845eb9 | 6169 | (void *)current->task_state_change); |
8eb23b9f | 6170 | |
3427445a PZ |
6171 | ___might_sleep(file, line, preempt_offset); |
6172 | } | |
6173 | EXPORT_SYMBOL(__might_sleep); | |
6174 | ||
6175 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 6176 | { |
d1ccc66d IM |
6177 | /* Ratelimiting timestamp: */ |
6178 | static unsigned long prev_jiffy; | |
6179 | ||
d1c6d149 | 6180 | unsigned long preempt_disable_ip; |
1da177e4 | 6181 | |
d1ccc66d IM |
6182 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
6183 | rcu_sleep_check(); | |
6184 | ||
db273be2 TG |
6185 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6186 | !is_idle_task(current)) || | |
e4aafea2 | 6187 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
6188 | return; |
6189 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6190 | return; | |
6191 | prev_jiffy = jiffies; | |
6192 | ||
d1ccc66d | 6193 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
6194 | preempt_disable_ip = get_preempt_disable_ip(current); |
6195 | ||
3df0fc5b PZ |
6196 | printk(KERN_ERR |
6197 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6198 | file, line); | |
6199 | printk(KERN_ERR | |
6200 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6201 | in_atomic(), irqs_disabled(), | |
6202 | current->pid, current->comm); | |
aef745fc | 6203 | |
a8b686b3 ES |
6204 | if (task_stack_end_corrupted(current)) |
6205 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
6206 | ||
aef745fc IM |
6207 | debug_show_held_locks(current); |
6208 | if (irqs_disabled()) | |
6209 | print_irqtrace_events(current); | |
d1c6d149 VN |
6210 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
6211 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 6212 | pr_err("Preemption disabled at:"); |
d1c6d149 | 6213 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
6214 | pr_cont("\n"); |
6215 | } | |
aef745fc | 6216 | dump_stack(); |
f0b22e39 | 6217 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 6218 | } |
3427445a | 6219 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
6220 | #endif |
6221 | ||
6222 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 6223 | void normalize_rt_tasks(void) |
3a5e4dc1 | 6224 | { |
dbc7f069 | 6225 | struct task_struct *g, *p; |
d50dde5a DF |
6226 | struct sched_attr attr = { |
6227 | .sched_policy = SCHED_NORMAL, | |
6228 | }; | |
1da177e4 | 6229 | |
3472eaa1 | 6230 | read_lock(&tasklist_lock); |
5d07f420 | 6231 | for_each_process_thread(g, p) { |
178be793 IM |
6232 | /* |
6233 | * Only normalize user tasks: | |
6234 | */ | |
3472eaa1 | 6235 | if (p->flags & PF_KTHREAD) |
178be793 IM |
6236 | continue; |
6237 | ||
4fa8d299 JP |
6238 | p->se.exec_start = 0; |
6239 | schedstat_set(p->se.statistics.wait_start, 0); | |
6240 | schedstat_set(p->se.statistics.sleep_start, 0); | |
6241 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 6242 | |
aab03e05 | 6243 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6244 | /* |
6245 | * Renice negative nice level userspace | |
6246 | * tasks back to 0: | |
6247 | */ | |
3472eaa1 | 6248 | if (task_nice(p) < 0) |
dd41f596 | 6249 | set_user_nice(p, 0); |
1da177e4 | 6250 | continue; |
dd41f596 | 6251 | } |
1da177e4 | 6252 | |
dbc7f069 | 6253 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 6254 | } |
3472eaa1 | 6255 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6256 | } |
6257 | ||
6258 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6259 | |
67fc4e0c | 6260 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6261 | /* |
67fc4e0c | 6262 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6263 | * |
6264 | * They can only be called when the whole system has been | |
6265 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6266 | * activity can take place. Using them for anything else would | |
6267 | * be a serious bug, and as a result, they aren't even visible | |
6268 | * under any other configuration. | |
6269 | */ | |
6270 | ||
6271 | /** | |
d1ccc66d | 6272 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
6273 | * @cpu: the processor in question. |
6274 | * | |
6275 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6276 | * |
6277 | * Return: The current task for @cpu. | |
1df5c10a | 6278 | */ |
36c8b586 | 6279 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6280 | { |
6281 | return cpu_curr(cpu); | |
6282 | } | |
6283 | ||
67fc4e0c JW |
6284 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6285 | ||
6286 | #ifdef CONFIG_IA64 | |
1df5c10a | 6287 | /** |
d1ccc66d | 6288 | * set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
6289 | * @cpu: the processor in question. |
6290 | * @p: the task pointer to set. | |
6291 | * | |
6292 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 6293 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 6294 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
6295 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6296 | * and caller must save the original value of the current task (see | |
6297 | * curr_task() above) and restore that value before reenabling interrupts and | |
6298 | * re-starting the system. | |
6299 | * | |
6300 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6301 | */ | |
a458ae2e | 6302 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6303 | { |
6304 | cpu_curr(cpu) = p; | |
6305 | } | |
6306 | ||
6307 | #endif | |
29f59db3 | 6308 | |
7c941438 | 6309 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6310 | /* task_group_lock serializes the addition/removal of task groups */ |
6311 | static DEFINE_SPINLOCK(task_group_lock); | |
6312 | ||
2f5177f0 | 6313 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
6314 | { |
6315 | free_fair_sched_group(tg); | |
6316 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6317 | autogroup_free(tg); |
b0367629 | 6318 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
6319 | } |
6320 | ||
6321 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6322 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6323 | { |
6324 | struct task_group *tg; | |
bccbe08a | 6325 | |
b0367629 | 6326 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
6327 | if (!tg) |
6328 | return ERR_PTR(-ENOMEM); | |
6329 | ||
ec7dc8ac | 6330 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6331 | goto err; |
6332 | ||
ec7dc8ac | 6333 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6334 | goto err; |
6335 | ||
ace783b9 LZ |
6336 | return tg; |
6337 | ||
6338 | err: | |
2f5177f0 | 6339 | sched_free_group(tg); |
ace783b9 LZ |
6340 | return ERR_PTR(-ENOMEM); |
6341 | } | |
6342 | ||
6343 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6344 | { | |
6345 | unsigned long flags; | |
6346 | ||
8ed36996 | 6347 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6348 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 6349 | |
d1ccc66d IM |
6350 | /* Root should already exist: */ |
6351 | WARN_ON(!parent); | |
f473aa5e PZ |
6352 | |
6353 | tg->parent = parent; | |
f473aa5e | 6354 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6355 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6356 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
6357 | |
6358 | online_fair_sched_group(tg); | |
29f59db3 SV |
6359 | } |
6360 | ||
9b5b7751 | 6361 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 6362 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6363 | { |
d1ccc66d | 6364 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 6365 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6366 | } |
6367 | ||
4cf86d77 | 6368 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 6369 | { |
d1ccc66d | 6370 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 6371 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
6372 | } |
6373 | ||
6374 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6375 | { |
8ed36996 | 6376 | unsigned long flags; |
29f59db3 | 6377 | |
d1ccc66d | 6378 | /* End participation in shares distribution: */ |
6fe1f348 | 6379 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
6380 | |
6381 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6382 | list_del_rcu(&tg->list); |
f473aa5e | 6383 | list_del_rcu(&tg->siblings); |
8ed36996 | 6384 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6385 | } |
6386 | ||
ea86cb4b | 6387 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 6388 | { |
8323f26c | 6389 | struct task_group *tg; |
29f59db3 | 6390 | |
f7b8a47d KT |
6391 | /* |
6392 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
6393 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
6394 | * to prevent lockdep warnings. | |
6395 | */ | |
6396 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
6397 | struct task_group, css); |
6398 | tg = autogroup_task_group(tsk, tg); | |
6399 | tsk->sched_task_group = tg; | |
6400 | ||
810b3817 | 6401 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
6402 | if (tsk->sched_class->task_change_group) |
6403 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 6404 | else |
810b3817 | 6405 | #endif |
b2b5ce02 | 6406 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
6407 | } |
6408 | ||
6409 | /* | |
6410 | * Change task's runqueue when it moves between groups. | |
6411 | * | |
6412 | * The caller of this function should have put the task in its new group by | |
6413 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
6414 | * its new group. | |
6415 | */ | |
6416 | void sched_move_task(struct task_struct *tsk) | |
6417 | { | |
6418 | int queued, running; | |
6419 | struct rq_flags rf; | |
6420 | struct rq *rq; | |
6421 | ||
6422 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 6423 | update_rq_clock(rq); |
ea86cb4b VG |
6424 | |
6425 | running = task_current(rq, tsk); | |
6426 | queued = task_on_rq_queued(tsk); | |
6427 | ||
6428 | if (queued) | |
6429 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); | |
bb3bac2c | 6430 | if (running) |
ea86cb4b VG |
6431 | put_prev_task(rq, tsk); |
6432 | ||
6433 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 6434 | |
da0c1e65 | 6435 | if (queued) |
7134b3e9 | 6436 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE | ENQUEUE_NOCLOCK); |
bb3bac2c | 6437 | if (running) |
b2bf6c31 | 6438 | set_curr_task(rq, tsk); |
29f59db3 | 6439 | |
eb580751 | 6440 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 6441 | } |
7c941438 | 6442 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 6443 | |
a790de99 PT |
6444 | #ifdef CONFIG_RT_GROUP_SCHED |
6445 | /* | |
6446 | * Ensure that the real time constraints are schedulable. | |
6447 | */ | |
6448 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 6449 | |
9a7e0b18 PZ |
6450 | /* Must be called with tasklist_lock held */ |
6451 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 6452 | { |
9a7e0b18 | 6453 | struct task_struct *g, *p; |
b40b2e8e | 6454 | |
1fe89e1b PZ |
6455 | /* |
6456 | * Autogroups do not have RT tasks; see autogroup_create(). | |
6457 | */ | |
6458 | if (task_group_is_autogroup(tg)) | |
6459 | return 0; | |
6460 | ||
5d07f420 | 6461 | for_each_process_thread(g, p) { |
8651c658 | 6462 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 6463 | return 1; |
5d07f420 | 6464 | } |
b40b2e8e | 6465 | |
9a7e0b18 PZ |
6466 | return 0; |
6467 | } | |
b40b2e8e | 6468 | |
9a7e0b18 PZ |
6469 | struct rt_schedulable_data { |
6470 | struct task_group *tg; | |
6471 | u64 rt_period; | |
6472 | u64 rt_runtime; | |
6473 | }; | |
b40b2e8e | 6474 | |
a790de99 | 6475 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
6476 | { |
6477 | struct rt_schedulable_data *d = data; | |
6478 | struct task_group *child; | |
6479 | unsigned long total, sum = 0; | |
6480 | u64 period, runtime; | |
b40b2e8e | 6481 | |
9a7e0b18 PZ |
6482 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
6483 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 6484 | |
9a7e0b18 PZ |
6485 | if (tg == d->tg) { |
6486 | period = d->rt_period; | |
6487 | runtime = d->rt_runtime; | |
b40b2e8e | 6488 | } |
b40b2e8e | 6489 | |
4653f803 PZ |
6490 | /* |
6491 | * Cannot have more runtime than the period. | |
6492 | */ | |
6493 | if (runtime > period && runtime != RUNTIME_INF) | |
6494 | return -EINVAL; | |
6f505b16 | 6495 | |
4653f803 PZ |
6496 | /* |
6497 | * Ensure we don't starve existing RT tasks. | |
6498 | */ | |
9a7e0b18 PZ |
6499 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
6500 | return -EBUSY; | |
6f505b16 | 6501 | |
9a7e0b18 | 6502 | total = to_ratio(period, runtime); |
6f505b16 | 6503 | |
4653f803 PZ |
6504 | /* |
6505 | * Nobody can have more than the global setting allows. | |
6506 | */ | |
6507 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
6508 | return -EINVAL; | |
6f505b16 | 6509 | |
4653f803 PZ |
6510 | /* |
6511 | * The sum of our children's runtime should not exceed our own. | |
6512 | */ | |
9a7e0b18 PZ |
6513 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
6514 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
6515 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 6516 | |
9a7e0b18 PZ |
6517 | if (child == d->tg) { |
6518 | period = d->rt_period; | |
6519 | runtime = d->rt_runtime; | |
6520 | } | |
6f505b16 | 6521 | |
9a7e0b18 | 6522 | sum += to_ratio(period, runtime); |
9f0c1e56 | 6523 | } |
6f505b16 | 6524 | |
9a7e0b18 PZ |
6525 | if (sum > total) |
6526 | return -EINVAL; | |
6527 | ||
6528 | return 0; | |
6f505b16 PZ |
6529 | } |
6530 | ||
9a7e0b18 | 6531 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 6532 | { |
8277434e PT |
6533 | int ret; |
6534 | ||
9a7e0b18 PZ |
6535 | struct rt_schedulable_data data = { |
6536 | .tg = tg, | |
6537 | .rt_period = period, | |
6538 | .rt_runtime = runtime, | |
6539 | }; | |
6540 | ||
8277434e PT |
6541 | rcu_read_lock(); |
6542 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
6543 | rcu_read_unlock(); | |
6544 | ||
6545 | return ret; | |
521f1a24 DG |
6546 | } |
6547 | ||
ab84d31e | 6548 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 6549 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 6550 | { |
ac086bc2 | 6551 | int i, err = 0; |
9f0c1e56 | 6552 | |
2636ed5f PZ |
6553 | /* |
6554 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
6555 | * kernel creating (and or operating) RT threads. | |
6556 | */ | |
6557 | if (tg == &root_task_group && rt_runtime == 0) | |
6558 | return -EINVAL; | |
6559 | ||
6560 | /* No period doesn't make any sense. */ | |
6561 | if (rt_period == 0) | |
6562 | return -EINVAL; | |
6563 | ||
9f0c1e56 | 6564 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 6565 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
6566 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
6567 | if (err) | |
9f0c1e56 | 6568 | goto unlock; |
ac086bc2 | 6569 | |
0986b11b | 6570 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
6571 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
6572 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
6573 | |
6574 | for_each_possible_cpu(i) { | |
6575 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
6576 | ||
0986b11b | 6577 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6578 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 6579 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6580 | } |
0986b11b | 6581 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 6582 | unlock: |
521f1a24 | 6583 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
6584 | mutex_unlock(&rt_constraints_mutex); |
6585 | ||
6586 | return err; | |
6f505b16 PZ |
6587 | } |
6588 | ||
25cc7da7 | 6589 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
6590 | { |
6591 | u64 rt_runtime, rt_period; | |
6592 | ||
6593 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6594 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
6595 | if (rt_runtime_us < 0) | |
6596 | rt_runtime = RUNTIME_INF; | |
6597 | ||
ab84d31e | 6598 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6599 | } |
6600 | ||
25cc7da7 | 6601 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
6602 | { |
6603 | u64 rt_runtime_us; | |
6604 | ||
d0b27fa7 | 6605 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
6606 | return -1; |
6607 | ||
d0b27fa7 | 6608 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
6609 | do_div(rt_runtime_us, NSEC_PER_USEC); |
6610 | return rt_runtime_us; | |
6611 | } | |
d0b27fa7 | 6612 | |
ce2f5fe4 | 6613 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
6614 | { |
6615 | u64 rt_runtime, rt_period; | |
6616 | ||
ce2f5fe4 | 6617 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
6618 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
6619 | ||
ab84d31e | 6620 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6621 | } |
6622 | ||
25cc7da7 | 6623 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
6624 | { |
6625 | u64 rt_period_us; | |
6626 | ||
6627 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6628 | do_div(rt_period_us, NSEC_PER_USEC); | |
6629 | return rt_period_us; | |
6630 | } | |
332ac17e | 6631 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6632 | |
332ac17e | 6633 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
6634 | static int sched_rt_global_constraints(void) |
6635 | { | |
6636 | int ret = 0; | |
6637 | ||
6638 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 6639 | read_lock(&tasklist_lock); |
4653f803 | 6640 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 6641 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
6642 | mutex_unlock(&rt_constraints_mutex); |
6643 | ||
6644 | return ret; | |
6645 | } | |
54e99124 | 6646 | |
25cc7da7 | 6647 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
6648 | { |
6649 | /* Don't accept realtime tasks when there is no way for them to run */ | |
6650 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
6651 | return 0; | |
6652 | ||
6653 | return 1; | |
6654 | } | |
6655 | ||
6d6bc0ad | 6656 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
6657 | static int sched_rt_global_constraints(void) |
6658 | { | |
ac086bc2 | 6659 | unsigned long flags; |
8c5e9554 | 6660 | int i; |
ec5d4989 | 6661 | |
0986b11b | 6662 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
6663 | for_each_possible_cpu(i) { |
6664 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
6665 | ||
0986b11b | 6666 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6667 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 6668 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6669 | } |
0986b11b | 6670 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 6671 | |
8c5e9554 | 6672 | return 0; |
d0b27fa7 | 6673 | } |
6d6bc0ad | 6674 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6675 | |
a1963b81 | 6676 | static int sched_dl_global_validate(void) |
332ac17e | 6677 | { |
1724813d PZ |
6678 | u64 runtime = global_rt_runtime(); |
6679 | u64 period = global_rt_period(); | |
332ac17e | 6680 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 6681 | struct dl_bw *dl_b; |
1724813d | 6682 | int cpu, ret = 0; |
49516342 | 6683 | unsigned long flags; |
332ac17e DF |
6684 | |
6685 | /* | |
6686 | * Here we want to check the bandwidth not being set to some | |
6687 | * value smaller than the currently allocated bandwidth in | |
6688 | * any of the root_domains. | |
6689 | * | |
6690 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
6691 | * cycling on root_domains... Discussion on different/better | |
6692 | * solutions is welcome! | |
6693 | */ | |
1724813d | 6694 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
6695 | rcu_read_lock_sched(); |
6696 | dl_b = dl_bw_of(cpu); | |
332ac17e | 6697 | |
49516342 | 6698 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
6699 | if (new_bw < dl_b->total_bw) |
6700 | ret = -EBUSY; | |
49516342 | 6701 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 6702 | |
f10e00f4 KT |
6703 | rcu_read_unlock_sched(); |
6704 | ||
1724813d PZ |
6705 | if (ret) |
6706 | break; | |
332ac17e DF |
6707 | } |
6708 | ||
1724813d | 6709 | return ret; |
332ac17e DF |
6710 | } |
6711 | ||
1724813d | 6712 | static void sched_dl_do_global(void) |
ce0dbbbb | 6713 | { |
1724813d | 6714 | u64 new_bw = -1; |
f10e00f4 | 6715 | struct dl_bw *dl_b; |
1724813d | 6716 | int cpu; |
49516342 | 6717 | unsigned long flags; |
ce0dbbbb | 6718 | |
1724813d PZ |
6719 | def_dl_bandwidth.dl_period = global_rt_period(); |
6720 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
6721 | ||
6722 | if (global_rt_runtime() != RUNTIME_INF) | |
6723 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
6724 | ||
6725 | /* | |
6726 | * FIXME: As above... | |
6727 | */ | |
6728 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
6729 | rcu_read_lock_sched(); |
6730 | dl_b = dl_bw_of(cpu); | |
1724813d | 6731 | |
49516342 | 6732 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 6733 | dl_b->bw = new_bw; |
49516342 | 6734 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
6735 | |
6736 | rcu_read_unlock_sched(); | |
ce0dbbbb | 6737 | } |
1724813d PZ |
6738 | } |
6739 | ||
6740 | static int sched_rt_global_validate(void) | |
6741 | { | |
6742 | if (sysctl_sched_rt_period <= 0) | |
6743 | return -EINVAL; | |
6744 | ||
e9e7cb38 JL |
6745 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
6746 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
6747 | return -EINVAL; |
6748 | ||
6749 | return 0; | |
6750 | } | |
6751 | ||
6752 | static void sched_rt_do_global(void) | |
6753 | { | |
6754 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
6755 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
6756 | } |
6757 | ||
d0b27fa7 | 6758 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 6759 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
6760 | loff_t *ppos) |
6761 | { | |
d0b27fa7 PZ |
6762 | int old_period, old_runtime; |
6763 | static DEFINE_MUTEX(mutex); | |
1724813d | 6764 | int ret; |
d0b27fa7 PZ |
6765 | |
6766 | mutex_lock(&mutex); | |
6767 | old_period = sysctl_sched_rt_period; | |
6768 | old_runtime = sysctl_sched_rt_runtime; | |
6769 | ||
8d65af78 | 6770 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
6771 | |
6772 | if (!ret && write) { | |
1724813d PZ |
6773 | ret = sched_rt_global_validate(); |
6774 | if (ret) | |
6775 | goto undo; | |
6776 | ||
a1963b81 | 6777 | ret = sched_dl_global_validate(); |
1724813d PZ |
6778 | if (ret) |
6779 | goto undo; | |
6780 | ||
a1963b81 | 6781 | ret = sched_rt_global_constraints(); |
1724813d PZ |
6782 | if (ret) |
6783 | goto undo; | |
6784 | ||
6785 | sched_rt_do_global(); | |
6786 | sched_dl_do_global(); | |
6787 | } | |
6788 | if (0) { | |
6789 | undo: | |
6790 | sysctl_sched_rt_period = old_period; | |
6791 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
6792 | } |
6793 | mutex_unlock(&mutex); | |
6794 | ||
6795 | return ret; | |
6796 | } | |
68318b8e | 6797 | |
1724813d | 6798 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
6799 | void __user *buffer, size_t *lenp, |
6800 | loff_t *ppos) | |
6801 | { | |
6802 | int ret; | |
332ac17e | 6803 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
6804 | |
6805 | mutex_lock(&mutex); | |
332ac17e | 6806 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d1ccc66d IM |
6807 | /* |
6808 | * Make sure that internally we keep jiffies. | |
6809 | * Also, writing zero resets the timeslice to default: | |
6810 | */ | |
332ac17e | 6811 | if (!ret && write) { |
975e155e SZ |
6812 | sched_rr_timeslice = |
6813 | sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE : | |
6814 | msecs_to_jiffies(sysctl_sched_rr_timeslice); | |
332ac17e DF |
6815 | } |
6816 | mutex_unlock(&mutex); | |
332ac17e DF |
6817 | return ret; |
6818 | } | |
6819 | ||
052f1dc7 | 6820 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 6821 | |
a7c6d554 | 6822 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 6823 | { |
a7c6d554 | 6824 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
6825 | } |
6826 | ||
eb95419b TH |
6827 | static struct cgroup_subsys_state * |
6828 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 6829 | { |
eb95419b TH |
6830 | struct task_group *parent = css_tg(parent_css); |
6831 | struct task_group *tg; | |
68318b8e | 6832 | |
eb95419b | 6833 | if (!parent) { |
68318b8e | 6834 | /* This is early initialization for the top cgroup */ |
07e06b01 | 6835 | return &root_task_group.css; |
68318b8e SV |
6836 | } |
6837 | ||
ec7dc8ac | 6838 | tg = sched_create_group(parent); |
68318b8e SV |
6839 | if (IS_ERR(tg)) |
6840 | return ERR_PTR(-ENOMEM); | |
6841 | ||
68318b8e SV |
6842 | return &tg->css; |
6843 | } | |
6844 | ||
96b77745 KK |
6845 | /* Expose task group only after completing cgroup initialization */ |
6846 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
6847 | { | |
6848 | struct task_group *tg = css_tg(css); | |
6849 | struct task_group *parent = css_tg(css->parent); | |
6850 | ||
6851 | if (parent) | |
6852 | sched_online_group(tg, parent); | |
6853 | return 0; | |
6854 | } | |
6855 | ||
2f5177f0 | 6856 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 6857 | { |
eb95419b | 6858 | struct task_group *tg = css_tg(css); |
ace783b9 | 6859 | |
2f5177f0 | 6860 | sched_offline_group(tg); |
ace783b9 LZ |
6861 | } |
6862 | ||
eb95419b | 6863 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 6864 | { |
eb95419b | 6865 | struct task_group *tg = css_tg(css); |
68318b8e | 6866 | |
2f5177f0 PZ |
6867 | /* |
6868 | * Relies on the RCU grace period between css_released() and this. | |
6869 | */ | |
6870 | sched_free_group(tg); | |
ace783b9 LZ |
6871 | } |
6872 | ||
ea86cb4b VG |
6873 | /* |
6874 | * This is called before wake_up_new_task(), therefore we really only | |
6875 | * have to set its group bits, all the other stuff does not apply. | |
6876 | */ | |
b53202e6 | 6877 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 6878 | { |
ea86cb4b VG |
6879 | struct rq_flags rf; |
6880 | struct rq *rq; | |
6881 | ||
6882 | rq = task_rq_lock(task, &rf); | |
6883 | ||
80f5c1b8 | 6884 | update_rq_clock(rq); |
ea86cb4b VG |
6885 | sched_change_group(task, TASK_SET_GROUP); |
6886 | ||
6887 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
6888 | } |
6889 | ||
1f7dd3e5 | 6890 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 6891 | { |
bb9d97b6 | 6892 | struct task_struct *task; |
1f7dd3e5 | 6893 | struct cgroup_subsys_state *css; |
7dc603c9 | 6894 | int ret = 0; |
bb9d97b6 | 6895 | |
1f7dd3e5 | 6896 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 6897 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 6898 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 6899 | return -EINVAL; |
b68aa230 | 6900 | #else |
bb9d97b6 TH |
6901 | /* We don't support RT-tasks being in separate groups */ |
6902 | if (task->sched_class != &fair_sched_class) | |
6903 | return -EINVAL; | |
b68aa230 | 6904 | #endif |
7dc603c9 PZ |
6905 | /* |
6906 | * Serialize against wake_up_new_task() such that if its | |
6907 | * running, we're sure to observe its full state. | |
6908 | */ | |
6909 | raw_spin_lock_irq(&task->pi_lock); | |
6910 | /* | |
6911 | * Avoid calling sched_move_task() before wake_up_new_task() | |
6912 | * has happened. This would lead to problems with PELT, due to | |
6913 | * move wanting to detach+attach while we're not attached yet. | |
6914 | */ | |
6915 | if (task->state == TASK_NEW) | |
6916 | ret = -EINVAL; | |
6917 | raw_spin_unlock_irq(&task->pi_lock); | |
6918 | ||
6919 | if (ret) | |
6920 | break; | |
bb9d97b6 | 6921 | } |
7dc603c9 | 6922 | return ret; |
be367d09 | 6923 | } |
68318b8e | 6924 | |
1f7dd3e5 | 6925 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 6926 | { |
bb9d97b6 | 6927 | struct task_struct *task; |
1f7dd3e5 | 6928 | struct cgroup_subsys_state *css; |
bb9d97b6 | 6929 | |
1f7dd3e5 | 6930 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 6931 | sched_move_task(task); |
68318b8e SV |
6932 | } |
6933 | ||
052f1dc7 | 6934 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
6935 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
6936 | struct cftype *cftype, u64 shareval) | |
68318b8e | 6937 | { |
182446d0 | 6938 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
6939 | } |
6940 | ||
182446d0 TH |
6941 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
6942 | struct cftype *cft) | |
68318b8e | 6943 | { |
182446d0 | 6944 | struct task_group *tg = css_tg(css); |
68318b8e | 6945 | |
c8b28116 | 6946 | return (u64) scale_load_down(tg->shares); |
68318b8e | 6947 | } |
ab84d31e PT |
6948 | |
6949 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
6950 | static DEFINE_MUTEX(cfs_constraints_mutex); |
6951 | ||
ab84d31e PT |
6952 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
6953 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
6954 | ||
a790de99 PT |
6955 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
6956 | ||
ab84d31e PT |
6957 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
6958 | { | |
56f570e5 | 6959 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 6960 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
6961 | |
6962 | if (tg == &root_task_group) | |
6963 | return -EINVAL; | |
6964 | ||
6965 | /* | |
6966 | * Ensure we have at some amount of bandwidth every period. This is | |
6967 | * to prevent reaching a state of large arrears when throttled via | |
6968 | * entity_tick() resulting in prolonged exit starvation. | |
6969 | */ | |
6970 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
6971 | return -EINVAL; | |
6972 | ||
6973 | /* | |
6974 | * Likewise, bound things on the otherside by preventing insane quota | |
6975 | * periods. This also allows us to normalize in computing quota | |
6976 | * feasibility. | |
6977 | */ | |
6978 | if (period > max_cfs_quota_period) | |
6979 | return -EINVAL; | |
6980 | ||
0e59bdae KT |
6981 | /* |
6982 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
6983 | * unthrottle_offline_cfs_rqs(). | |
6984 | */ | |
6985 | get_online_cpus(); | |
a790de99 PT |
6986 | mutex_lock(&cfs_constraints_mutex); |
6987 | ret = __cfs_schedulable(tg, period, quota); | |
6988 | if (ret) | |
6989 | goto out_unlock; | |
6990 | ||
58088ad0 | 6991 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 6992 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
6993 | /* |
6994 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
6995 | * before making related changes, and on->off must occur afterwards | |
6996 | */ | |
6997 | if (runtime_enabled && !runtime_was_enabled) | |
6998 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
6999 | raw_spin_lock_irq(&cfs_b->lock); |
7000 | cfs_b->period = ns_to_ktime(period); | |
7001 | cfs_b->quota = quota; | |
58088ad0 | 7002 | |
a9cf55b2 | 7003 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
7004 | |
7005 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
7006 | if (runtime_enabled) |
7007 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 7008 | |
ab84d31e PT |
7009 | raw_spin_unlock_irq(&cfs_b->lock); |
7010 | ||
0e59bdae | 7011 | for_each_online_cpu(i) { |
ab84d31e | 7012 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7013 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 7014 | struct rq_flags rf; |
ab84d31e | 7015 | |
8a8c69c3 | 7016 | rq_lock_irq(rq, &rf); |
58088ad0 | 7017 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7018 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7019 | |
029632fb | 7020 | if (cfs_rq->throttled) |
671fd9da | 7021 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 7022 | rq_unlock_irq(rq, &rf); |
ab84d31e | 7023 | } |
1ee14e6c BS |
7024 | if (runtime_was_enabled && !runtime_enabled) |
7025 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7026 | out_unlock: |
7027 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7028 | put_online_cpus(); |
ab84d31e | 7029 | |
a790de99 | 7030 | return ret; |
ab84d31e PT |
7031 | } |
7032 | ||
7033 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7034 | { | |
7035 | u64 quota, period; | |
7036 | ||
029632fb | 7037 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7038 | if (cfs_quota_us < 0) |
7039 | quota = RUNTIME_INF; | |
7040 | else | |
7041 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7042 | ||
7043 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7044 | } | |
7045 | ||
7046 | long tg_get_cfs_quota(struct task_group *tg) | |
7047 | { | |
7048 | u64 quota_us; | |
7049 | ||
029632fb | 7050 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7051 | return -1; |
7052 | ||
029632fb | 7053 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7054 | do_div(quota_us, NSEC_PER_USEC); |
7055 | ||
7056 | return quota_us; | |
7057 | } | |
7058 | ||
7059 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7060 | { | |
7061 | u64 quota, period; | |
7062 | ||
7063 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7064 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7065 | |
ab84d31e PT |
7066 | return tg_set_cfs_bandwidth(tg, period, quota); |
7067 | } | |
7068 | ||
7069 | long tg_get_cfs_period(struct task_group *tg) | |
7070 | { | |
7071 | u64 cfs_period_us; | |
7072 | ||
029632fb | 7073 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7074 | do_div(cfs_period_us, NSEC_PER_USEC); |
7075 | ||
7076 | return cfs_period_us; | |
7077 | } | |
7078 | ||
182446d0 TH |
7079 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7080 | struct cftype *cft) | |
ab84d31e | 7081 | { |
182446d0 | 7082 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7083 | } |
7084 | ||
182446d0 TH |
7085 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7086 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7087 | { |
182446d0 | 7088 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7089 | } |
7090 | ||
182446d0 TH |
7091 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7092 | struct cftype *cft) | |
ab84d31e | 7093 | { |
182446d0 | 7094 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7095 | } |
7096 | ||
182446d0 TH |
7097 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7098 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7099 | { |
182446d0 | 7100 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7101 | } |
7102 | ||
a790de99 PT |
7103 | struct cfs_schedulable_data { |
7104 | struct task_group *tg; | |
7105 | u64 period, quota; | |
7106 | }; | |
7107 | ||
7108 | /* | |
7109 | * normalize group quota/period to be quota/max_period | |
7110 | * note: units are usecs | |
7111 | */ | |
7112 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7113 | struct cfs_schedulable_data *d) | |
7114 | { | |
7115 | u64 quota, period; | |
7116 | ||
7117 | if (tg == d->tg) { | |
7118 | period = d->period; | |
7119 | quota = d->quota; | |
7120 | } else { | |
7121 | period = tg_get_cfs_period(tg); | |
7122 | quota = tg_get_cfs_quota(tg); | |
7123 | } | |
7124 | ||
7125 | /* note: these should typically be equivalent */ | |
7126 | if (quota == RUNTIME_INF || quota == -1) | |
7127 | return RUNTIME_INF; | |
7128 | ||
7129 | return to_ratio(period, quota); | |
7130 | } | |
7131 | ||
7132 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7133 | { | |
7134 | struct cfs_schedulable_data *d = data; | |
029632fb | 7135 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7136 | s64 quota = 0, parent_quota = -1; |
7137 | ||
7138 | if (!tg->parent) { | |
7139 | quota = RUNTIME_INF; | |
7140 | } else { | |
029632fb | 7141 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7142 | |
7143 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 7144 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
7145 | |
7146 | /* | |
d1ccc66d IM |
7147 | * Ensure max(child_quota) <= parent_quota, inherit when no |
7148 | * limit is set: | |
a790de99 PT |
7149 | */ |
7150 | if (quota == RUNTIME_INF) | |
7151 | quota = parent_quota; | |
7152 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7153 | return -EINVAL; | |
7154 | } | |
9c58c79a | 7155 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
7156 | |
7157 | return 0; | |
7158 | } | |
7159 | ||
7160 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7161 | { | |
8277434e | 7162 | int ret; |
a790de99 PT |
7163 | struct cfs_schedulable_data data = { |
7164 | .tg = tg, | |
7165 | .period = period, | |
7166 | .quota = quota, | |
7167 | }; | |
7168 | ||
7169 | if (quota != RUNTIME_INF) { | |
7170 | do_div(data.period, NSEC_PER_USEC); | |
7171 | do_div(data.quota, NSEC_PER_USEC); | |
7172 | } | |
7173 | ||
8277434e PT |
7174 | rcu_read_lock(); |
7175 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7176 | rcu_read_unlock(); | |
7177 | ||
7178 | return ret; | |
a790de99 | 7179 | } |
e8da1b18 | 7180 | |
2da8ca82 | 7181 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7182 | { |
2da8ca82 | 7183 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7184 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7185 | |
44ffc75b TH |
7186 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7187 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7188 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7189 | |
7190 | return 0; | |
7191 | } | |
ab84d31e | 7192 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7193 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7194 | |
052f1dc7 | 7195 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7196 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7197 | struct cftype *cft, s64 val) | |
6f505b16 | 7198 | { |
182446d0 | 7199 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7200 | } |
7201 | ||
182446d0 TH |
7202 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7203 | struct cftype *cft) | |
6f505b16 | 7204 | { |
182446d0 | 7205 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7206 | } |
d0b27fa7 | 7207 | |
182446d0 TH |
7208 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7209 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7210 | { |
182446d0 | 7211 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7212 | } |
7213 | ||
182446d0 TH |
7214 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7215 | struct cftype *cft) | |
d0b27fa7 | 7216 | { |
182446d0 | 7217 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7218 | } |
6d6bc0ad | 7219 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7220 | |
fe5c7cc2 | 7221 | static struct cftype cpu_files[] = { |
052f1dc7 | 7222 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7223 | { |
7224 | .name = "shares", | |
f4c753b7 PM |
7225 | .read_u64 = cpu_shares_read_u64, |
7226 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7227 | }, |
052f1dc7 | 7228 | #endif |
ab84d31e PT |
7229 | #ifdef CONFIG_CFS_BANDWIDTH |
7230 | { | |
7231 | .name = "cfs_quota_us", | |
7232 | .read_s64 = cpu_cfs_quota_read_s64, | |
7233 | .write_s64 = cpu_cfs_quota_write_s64, | |
7234 | }, | |
7235 | { | |
7236 | .name = "cfs_period_us", | |
7237 | .read_u64 = cpu_cfs_period_read_u64, | |
7238 | .write_u64 = cpu_cfs_period_write_u64, | |
7239 | }, | |
e8da1b18 NR |
7240 | { |
7241 | .name = "stat", | |
2da8ca82 | 7242 | .seq_show = cpu_stats_show, |
e8da1b18 | 7243 | }, |
ab84d31e | 7244 | #endif |
052f1dc7 | 7245 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7246 | { |
9f0c1e56 | 7247 | .name = "rt_runtime_us", |
06ecb27c PM |
7248 | .read_s64 = cpu_rt_runtime_read, |
7249 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7250 | }, |
d0b27fa7 PZ |
7251 | { |
7252 | .name = "rt_period_us", | |
f4c753b7 PM |
7253 | .read_u64 = cpu_rt_period_read_uint, |
7254 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7255 | }, |
052f1dc7 | 7256 | #endif |
d1ccc66d | 7257 | { } /* Terminate */ |
68318b8e SV |
7258 | }; |
7259 | ||
073219e9 | 7260 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 7261 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 7262 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 7263 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 7264 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 7265 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
7266 | .can_attach = cpu_cgroup_can_attach, |
7267 | .attach = cpu_cgroup_attach, | |
5577964e | 7268 | .legacy_cftypes = cpu_files, |
b38e42e9 | 7269 | .early_init = true, |
68318b8e SV |
7270 | }; |
7271 | ||
052f1dc7 | 7272 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7273 | |
b637a328 PM |
7274 | void dump_cpu_task(int cpu) |
7275 | { | |
7276 | pr_info("Task dump for CPU %d:\n", cpu); | |
7277 | sched_show_task(cpu_curr(cpu)); | |
7278 | } | |
ed82b8a1 AK |
7279 | |
7280 | /* | |
7281 | * Nice levels are multiplicative, with a gentle 10% change for every | |
7282 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
7283 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
7284 | * that remained on nice 0. | |
7285 | * | |
7286 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
7287 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
7288 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
7289 | * If a task goes up by ~10% and another task goes down by ~10% then | |
7290 | * the relative distance between them is ~25%.) | |
7291 | */ | |
7292 | const int sched_prio_to_weight[40] = { | |
7293 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
7294 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
7295 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
7296 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
7297 | /* 0 */ 1024, 820, 655, 526, 423, | |
7298 | /* 5 */ 335, 272, 215, 172, 137, | |
7299 | /* 10 */ 110, 87, 70, 56, 45, | |
7300 | /* 15 */ 36, 29, 23, 18, 15, | |
7301 | }; | |
7302 | ||
7303 | /* | |
7304 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
7305 | * | |
7306 | * In cases where the weight does not change often, we can use the | |
7307 | * precalculated inverse to speed up arithmetics by turning divisions | |
7308 | * into multiplications: | |
7309 | */ | |
7310 | const u32 sched_prio_to_wmult[40] = { | |
7311 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
7312 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
7313 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
7314 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
7315 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
7316 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
7317 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
7318 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
7319 | }; |