]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
e1b77c92 | 29 | #include <linux/kasan.h> |
1da177e4 LT |
30 | #include <linux/mm.h> |
31 | #include <linux/module.h> | |
32 | #include <linux/nmi.h> | |
33 | #include <linux/init.h> | |
dff06c15 | 34 | #include <linux/uaccess.h> |
1da177e4 | 35 | #include <linux/highmem.h> |
f98db601 | 36 | #include <linux/mmu_context.h> |
1da177e4 | 37 | #include <linux/interrupt.h> |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 | 70 | #include <linux/ctype.h> |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
91d1aa43 | 74 | #include <linux/context_tracking.h> |
52f5684c | 75 | #include <linux/compiler.h> |
8e05e96a | 76 | #include <linux/frame.h> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
029632fb PZ |
93 | DEFINE_MUTEX(sched_domains_mutex); |
94 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 95 | |
fe44d621 | 96 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 97 | |
029632fb | 98 | void update_rq_clock(struct rq *rq) |
3e51f33f | 99 | { |
fe44d621 | 100 | s64 delta; |
305e6835 | 101 | |
9edfbfed PZ |
102 | lockdep_assert_held(&rq->lock); |
103 | ||
104 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 105 | return; |
aa483808 | 106 | |
fe44d621 | 107 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
108 | if (delta < 0) |
109 | return; | |
fe44d621 PZ |
110 | rq->clock += delta; |
111 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
112 | } |
113 | ||
bf5c91ba IM |
114 | /* |
115 | * Debugging: various feature bits | |
116 | */ | |
f00b45c1 | 117 | |
f00b45c1 PZ |
118 | #define SCHED_FEAT(name, enabled) \ |
119 | (1UL << __SCHED_FEAT_##name) * enabled | | |
120 | ||
bf5c91ba | 121 | const_debug unsigned int sysctl_sched_features = |
391e43da | 122 | #include "features.h" |
f00b45c1 PZ |
123 | 0; |
124 | ||
125 | #undef SCHED_FEAT | |
126 | ||
b82d9fdd PZ |
127 | /* |
128 | * Number of tasks to iterate in a single balance run. | |
129 | * Limited because this is done with IRQs disabled. | |
130 | */ | |
131 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
132 | ||
e9e9250b PZ |
133 | /* |
134 | * period over which we average the RT time consumption, measured | |
135 | * in ms. | |
136 | * | |
137 | * default: 1s | |
138 | */ | |
139 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
140 | ||
fa85ae24 | 141 | /* |
9f0c1e56 | 142 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
143 | * default: 1s |
144 | */ | |
9f0c1e56 | 145 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 146 | |
029632fb | 147 | __read_mostly int scheduler_running; |
6892b75e | 148 | |
9f0c1e56 PZ |
149 | /* |
150 | * part of the period that we allow rt tasks to run in us. | |
151 | * default: 0.95s | |
152 | */ | |
153 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 154 | |
3fa0818b RR |
155 | /* cpus with isolated domains */ |
156 | cpumask_var_t cpu_isolated_map; | |
157 | ||
1da177e4 | 158 | /* |
cc2a73b5 | 159 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 160 | */ |
a9957449 | 161 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
162 | __acquires(rq->lock) |
163 | { | |
70b97a7f | 164 | struct rq *rq; |
1da177e4 LT |
165 | |
166 | local_irq_disable(); | |
167 | rq = this_rq(); | |
05fa785c | 168 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
169 | |
170 | return rq; | |
171 | } | |
172 | ||
8f4d37ec PZ |
173 | #ifdef CONFIG_SCHED_HRTICK |
174 | /* | |
175 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 176 | */ |
8f4d37ec | 177 | |
8f4d37ec PZ |
178 | static void hrtick_clear(struct rq *rq) |
179 | { | |
180 | if (hrtimer_active(&rq->hrtick_timer)) | |
181 | hrtimer_cancel(&rq->hrtick_timer); | |
182 | } | |
183 | ||
8f4d37ec PZ |
184 | /* |
185 | * High-resolution timer tick. | |
186 | * Runs from hardirq context with interrupts disabled. | |
187 | */ | |
188 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
189 | { | |
190 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
191 | ||
192 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
193 | ||
05fa785c | 194 | raw_spin_lock(&rq->lock); |
3e51f33f | 195 | update_rq_clock(rq); |
8f4d37ec | 196 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 197 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
198 | |
199 | return HRTIMER_NORESTART; | |
200 | } | |
201 | ||
95e904c7 | 202 | #ifdef CONFIG_SMP |
971ee28c | 203 | |
4961b6e1 | 204 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
205 | { |
206 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 207 | |
4961b6e1 | 208 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
209 | } |
210 | ||
31656519 PZ |
211 | /* |
212 | * called from hardirq (IPI) context | |
213 | */ | |
214 | static void __hrtick_start(void *arg) | |
b328ca18 | 215 | { |
31656519 | 216 | struct rq *rq = arg; |
b328ca18 | 217 | |
05fa785c | 218 | raw_spin_lock(&rq->lock); |
971ee28c | 219 | __hrtick_restart(rq); |
31656519 | 220 | rq->hrtick_csd_pending = 0; |
05fa785c | 221 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
222 | } |
223 | ||
31656519 PZ |
224 | /* |
225 | * Called to set the hrtick timer state. | |
226 | * | |
227 | * called with rq->lock held and irqs disabled | |
228 | */ | |
029632fb | 229 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 230 | { |
31656519 | 231 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 232 | ktime_t time; |
233 | s64 delta; | |
234 | ||
235 | /* | |
236 | * Don't schedule slices shorter than 10000ns, that just | |
237 | * doesn't make sense and can cause timer DoS. | |
238 | */ | |
239 | delta = max_t(s64, delay, 10000LL); | |
240 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 241 | |
cc584b21 | 242 | hrtimer_set_expires(timer, time); |
31656519 PZ |
243 | |
244 | if (rq == this_rq()) { | |
971ee28c | 245 | __hrtick_restart(rq); |
31656519 | 246 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 247 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
248 | rq->hrtick_csd_pending = 1; |
249 | } | |
b328ca18 PZ |
250 | } |
251 | ||
252 | static int | |
253 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
254 | { | |
255 | int cpu = (int)(long)hcpu; | |
256 | ||
257 | switch (action) { | |
258 | case CPU_UP_CANCELED: | |
259 | case CPU_UP_CANCELED_FROZEN: | |
260 | case CPU_DOWN_PREPARE: | |
261 | case CPU_DOWN_PREPARE_FROZEN: | |
262 | case CPU_DEAD: | |
263 | case CPU_DEAD_FROZEN: | |
31656519 | 264 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
265 | return NOTIFY_OK; |
266 | } | |
267 | ||
268 | return NOTIFY_DONE; | |
269 | } | |
270 | ||
fa748203 | 271 | static __init void init_hrtick(void) |
b328ca18 PZ |
272 | { |
273 | hotcpu_notifier(hotplug_hrtick, 0); | |
274 | } | |
31656519 PZ |
275 | #else |
276 | /* | |
277 | * Called to set the hrtick timer state. | |
278 | * | |
279 | * called with rq->lock held and irqs disabled | |
280 | */ | |
029632fb | 281 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 282 | { |
86893335 WL |
283 | /* |
284 | * Don't schedule slices shorter than 10000ns, that just | |
285 | * doesn't make sense. Rely on vruntime for fairness. | |
286 | */ | |
287 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
288 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
289 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 290 | } |
b328ca18 | 291 | |
006c75f1 | 292 | static inline void init_hrtick(void) |
8f4d37ec | 293 | { |
8f4d37ec | 294 | } |
31656519 | 295 | #endif /* CONFIG_SMP */ |
8f4d37ec | 296 | |
31656519 | 297 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 298 | { |
31656519 PZ |
299 | #ifdef CONFIG_SMP |
300 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 301 | |
31656519 PZ |
302 | rq->hrtick_csd.flags = 0; |
303 | rq->hrtick_csd.func = __hrtick_start; | |
304 | rq->hrtick_csd.info = rq; | |
305 | #endif | |
8f4d37ec | 306 | |
31656519 PZ |
307 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
308 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 309 | } |
006c75f1 | 310 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
311 | static inline void hrtick_clear(struct rq *rq) |
312 | { | |
313 | } | |
314 | ||
8f4d37ec PZ |
315 | static inline void init_rq_hrtick(struct rq *rq) |
316 | { | |
317 | } | |
318 | ||
b328ca18 PZ |
319 | static inline void init_hrtick(void) |
320 | { | |
321 | } | |
006c75f1 | 322 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 323 | |
5529578a FW |
324 | /* |
325 | * cmpxchg based fetch_or, macro so it works for different integer types | |
326 | */ | |
327 | #define fetch_or(ptr, mask) \ | |
328 | ({ \ | |
329 | typeof(ptr) _ptr = (ptr); \ | |
330 | typeof(mask) _mask = (mask); \ | |
331 | typeof(*_ptr) _old, _val = *_ptr; \ | |
332 | \ | |
333 | for (;;) { \ | |
334 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
335 | if (_old == _val) \ | |
336 | break; \ | |
337 | _val = _old; \ | |
338 | } \ | |
339 | _old; \ | |
340 | }) | |
341 | ||
e3baac47 | 342 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
343 | /* |
344 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
345 | * this avoids any races wrt polling state changes and thereby avoids | |
346 | * spurious IPIs. | |
347 | */ | |
348 | static bool set_nr_and_not_polling(struct task_struct *p) | |
349 | { | |
350 | struct thread_info *ti = task_thread_info(p); | |
351 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
352 | } | |
e3baac47 PZ |
353 | |
354 | /* | |
355 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
356 | * | |
357 | * If this returns true, then the idle task promises to call | |
358 | * sched_ttwu_pending() and reschedule soon. | |
359 | */ | |
360 | static bool set_nr_if_polling(struct task_struct *p) | |
361 | { | |
362 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 363 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
364 | |
365 | for (;;) { | |
366 | if (!(val & _TIF_POLLING_NRFLAG)) | |
367 | return false; | |
368 | if (val & _TIF_NEED_RESCHED) | |
369 | return true; | |
370 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
371 | if (old == val) | |
372 | break; | |
373 | val = old; | |
374 | } | |
375 | return true; | |
376 | } | |
377 | ||
fd99f91a PZ |
378 | #else |
379 | static bool set_nr_and_not_polling(struct task_struct *p) | |
380 | { | |
381 | set_tsk_need_resched(p); | |
382 | return true; | |
383 | } | |
e3baac47 PZ |
384 | |
385 | #ifdef CONFIG_SMP | |
386 | static bool set_nr_if_polling(struct task_struct *p) | |
387 | { | |
388 | return false; | |
389 | } | |
390 | #endif | |
fd99f91a PZ |
391 | #endif |
392 | ||
76751049 PZ |
393 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
394 | { | |
395 | struct wake_q_node *node = &task->wake_q; | |
396 | ||
397 | /* | |
398 | * Atomically grab the task, if ->wake_q is !nil already it means | |
399 | * its already queued (either by us or someone else) and will get the | |
400 | * wakeup due to that. | |
401 | * | |
402 | * This cmpxchg() implies a full barrier, which pairs with the write | |
403 | * barrier implied by the wakeup in wake_up_list(). | |
404 | */ | |
405 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
406 | return; | |
407 | ||
408 | get_task_struct(task); | |
409 | ||
410 | /* | |
411 | * The head is context local, there can be no concurrency. | |
412 | */ | |
413 | *head->lastp = node; | |
414 | head->lastp = &node->next; | |
415 | } | |
416 | ||
417 | void wake_up_q(struct wake_q_head *head) | |
418 | { | |
419 | struct wake_q_node *node = head->first; | |
420 | ||
421 | while (node != WAKE_Q_TAIL) { | |
422 | struct task_struct *task; | |
423 | ||
424 | task = container_of(node, struct task_struct, wake_q); | |
425 | BUG_ON(!task); | |
426 | /* task can safely be re-inserted now */ | |
427 | node = node->next; | |
428 | task->wake_q.next = NULL; | |
429 | ||
430 | /* | |
431 | * wake_up_process() implies a wmb() to pair with the queueing | |
432 | * in wake_q_add() so as not to miss wakeups. | |
433 | */ | |
434 | wake_up_process(task); | |
435 | put_task_struct(task); | |
436 | } | |
437 | } | |
438 | ||
c24d20db | 439 | /* |
8875125e | 440 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
441 | * |
442 | * On UP this means the setting of the need_resched flag, on SMP it | |
443 | * might also involve a cross-CPU call to trigger the scheduler on | |
444 | * the target CPU. | |
445 | */ | |
8875125e | 446 | void resched_curr(struct rq *rq) |
c24d20db | 447 | { |
8875125e | 448 | struct task_struct *curr = rq->curr; |
c24d20db IM |
449 | int cpu; |
450 | ||
8875125e | 451 | lockdep_assert_held(&rq->lock); |
c24d20db | 452 | |
8875125e | 453 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
454 | return; |
455 | ||
8875125e | 456 | cpu = cpu_of(rq); |
fd99f91a | 457 | |
f27dde8d | 458 | if (cpu == smp_processor_id()) { |
8875125e | 459 | set_tsk_need_resched(curr); |
f27dde8d | 460 | set_preempt_need_resched(); |
c24d20db | 461 | return; |
f27dde8d | 462 | } |
c24d20db | 463 | |
8875125e | 464 | if (set_nr_and_not_polling(curr)) |
c24d20db | 465 | smp_send_reschedule(cpu); |
dfc68f29 AL |
466 | else |
467 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
468 | } |
469 | ||
029632fb | 470 | void resched_cpu(int cpu) |
c24d20db IM |
471 | { |
472 | struct rq *rq = cpu_rq(cpu); | |
473 | unsigned long flags; | |
474 | ||
05fa785c | 475 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 476 | return; |
8875125e | 477 | resched_curr(rq); |
05fa785c | 478 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 479 | } |
06d8308c | 480 | |
b021fe3e | 481 | #ifdef CONFIG_SMP |
3451d024 | 482 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
483 | /* |
484 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
485 | * from an idle cpu. This is good for power-savings. | |
486 | * | |
487 | * We don't do similar optimization for completely idle system, as | |
488 | * selecting an idle cpu will add more delays to the timers than intended | |
489 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
490 | */ | |
bc7a34b8 | 491 | int get_nohz_timer_target(void) |
83cd4fe2 | 492 | { |
bc7a34b8 | 493 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
494 | struct sched_domain *sd; |
495 | ||
9642d18e | 496 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
497 | return cpu; |
498 | ||
057f3fad | 499 | rcu_read_lock(); |
83cd4fe2 | 500 | for_each_domain(cpu, sd) { |
057f3fad | 501 | for_each_cpu(i, sched_domain_span(sd)) { |
9642d18e | 502 | if (!idle_cpu(i) && is_housekeeping_cpu(cpu)) { |
057f3fad PZ |
503 | cpu = i; |
504 | goto unlock; | |
505 | } | |
506 | } | |
83cd4fe2 | 507 | } |
9642d18e VH |
508 | |
509 | if (!is_housekeeping_cpu(cpu)) | |
510 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
511 | unlock: |
512 | rcu_read_unlock(); | |
83cd4fe2 VP |
513 | return cpu; |
514 | } | |
06d8308c TG |
515 | /* |
516 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
517 | * idle CPU then this timer might expire before the next timer event | |
518 | * which is scheduled to wake up that CPU. In case of a completely | |
519 | * idle system the next event might even be infinite time into the | |
520 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
521 | * leaves the inner idle loop so the newly added timer is taken into | |
522 | * account when the CPU goes back to idle and evaluates the timer | |
523 | * wheel for the next timer event. | |
524 | */ | |
1c20091e | 525 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
526 | { |
527 | struct rq *rq = cpu_rq(cpu); | |
528 | ||
529 | if (cpu == smp_processor_id()) | |
530 | return; | |
531 | ||
67b9ca70 | 532 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 533 | smp_send_reschedule(cpu); |
dfc68f29 AL |
534 | else |
535 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
536 | } |
537 | ||
c5bfece2 | 538 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 539 | { |
53c5fa16 FW |
540 | /* |
541 | * We just need the target to call irq_exit() and re-evaluate | |
542 | * the next tick. The nohz full kick at least implies that. | |
543 | * If needed we can still optimize that later with an | |
544 | * empty IRQ. | |
545 | */ | |
c5bfece2 | 546 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
547 | if (cpu != smp_processor_id() || |
548 | tick_nohz_tick_stopped()) | |
53c5fa16 | 549 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
550 | return true; |
551 | } | |
552 | ||
553 | return false; | |
554 | } | |
555 | ||
556 | void wake_up_nohz_cpu(int cpu) | |
557 | { | |
c5bfece2 | 558 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
559 | wake_up_idle_cpu(cpu); |
560 | } | |
561 | ||
ca38062e | 562 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 563 | { |
1c792db7 | 564 | int cpu = smp_processor_id(); |
873b4c65 VG |
565 | |
566 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
567 | return false; | |
568 | ||
569 | if (idle_cpu(cpu) && !need_resched()) | |
570 | return true; | |
571 | ||
572 | /* | |
573 | * We can't run Idle Load Balance on this CPU for this time so we | |
574 | * cancel it and clear NOHZ_BALANCE_KICK | |
575 | */ | |
576 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
577 | return false; | |
45bf76df IM |
578 | } |
579 | ||
3451d024 | 580 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 581 | |
ca38062e | 582 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 583 | { |
ca38062e | 584 | return false; |
2069dd75 PZ |
585 | } |
586 | ||
3451d024 | 587 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 588 | |
ce831b38 | 589 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 590 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 591 | { |
76d92ac3 FW |
592 | int fifo_nr_running; |
593 | ||
594 | /* Deadline tasks, even if single, need the tick */ | |
595 | if (rq->dl.dl_nr_running) | |
596 | return false; | |
597 | ||
1e78cdbd | 598 | /* |
2548d546 PZ |
599 | * If there are more than one RR tasks, we need the tick to effect the |
600 | * actual RR behaviour. | |
1e78cdbd | 601 | */ |
76d92ac3 FW |
602 | if (rq->rt.rr_nr_running) { |
603 | if (rq->rt.rr_nr_running == 1) | |
604 | return true; | |
605 | else | |
606 | return false; | |
1e78cdbd RR |
607 | } |
608 | ||
2548d546 PZ |
609 | /* |
610 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
611 | * forced preemption between FIFO tasks. | |
612 | */ | |
613 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
614 | if (fifo_nr_running) | |
615 | return true; | |
616 | ||
617 | /* | |
618 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
619 | * if there's more than one we need the tick for involuntary | |
620 | * preemption. | |
621 | */ | |
622 | if (rq->nr_running > 1) | |
541b8264 | 623 | return false; |
ce831b38 | 624 | |
541b8264 | 625 | return true; |
ce831b38 FW |
626 | } |
627 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 628 | |
029632fb | 629 | void sched_avg_update(struct rq *rq) |
18d95a28 | 630 | { |
e9e9250b PZ |
631 | s64 period = sched_avg_period(); |
632 | ||
78becc27 | 633 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
634 | /* |
635 | * Inline assembly required to prevent the compiler | |
636 | * optimising this loop into a divmod call. | |
637 | * See __iter_div_u64_rem() for another example of this. | |
638 | */ | |
639 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
640 | rq->age_stamp += period; |
641 | rq->rt_avg /= 2; | |
642 | } | |
18d95a28 PZ |
643 | } |
644 | ||
6d6bc0ad | 645 | #endif /* CONFIG_SMP */ |
18d95a28 | 646 | |
a790de99 PT |
647 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
648 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 649 | /* |
8277434e PT |
650 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
651 | * node and @up when leaving it for the final time. | |
652 | * | |
653 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 654 | */ |
029632fb | 655 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 656 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
657 | { |
658 | struct task_group *parent, *child; | |
eb755805 | 659 | int ret; |
c09595f6 | 660 | |
8277434e PT |
661 | parent = from; |
662 | ||
c09595f6 | 663 | down: |
eb755805 PZ |
664 | ret = (*down)(parent, data); |
665 | if (ret) | |
8277434e | 666 | goto out; |
c09595f6 PZ |
667 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
668 | parent = child; | |
669 | goto down; | |
670 | ||
671 | up: | |
672 | continue; | |
673 | } | |
eb755805 | 674 | ret = (*up)(parent, data); |
8277434e PT |
675 | if (ret || parent == from) |
676 | goto out; | |
c09595f6 PZ |
677 | |
678 | child = parent; | |
679 | parent = parent->parent; | |
680 | if (parent) | |
681 | goto up; | |
8277434e | 682 | out: |
eb755805 | 683 | return ret; |
c09595f6 PZ |
684 | } |
685 | ||
029632fb | 686 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 687 | { |
e2b245f8 | 688 | return 0; |
eb755805 | 689 | } |
18d95a28 PZ |
690 | #endif |
691 | ||
45bf76df IM |
692 | static void set_load_weight(struct task_struct *p) |
693 | { | |
f05998d4 NR |
694 | int prio = p->static_prio - MAX_RT_PRIO; |
695 | struct load_weight *load = &p->se.load; | |
696 | ||
dd41f596 IM |
697 | /* |
698 | * SCHED_IDLE tasks get minimal weight: | |
699 | */ | |
20f9cd2a | 700 | if (idle_policy(p->policy)) { |
c8b28116 | 701 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 702 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
703 | return; |
704 | } | |
71f8bd46 | 705 | |
ed82b8a1 AK |
706 | load->weight = scale_load(sched_prio_to_weight[prio]); |
707 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
708 | } |
709 | ||
1de64443 | 710 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 711 | { |
a64692a3 | 712 | update_rq_clock(rq); |
1de64443 PZ |
713 | if (!(flags & ENQUEUE_RESTORE)) |
714 | sched_info_queued(rq, p); | |
371fd7e7 | 715 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
716 | } |
717 | ||
1de64443 | 718 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 719 | { |
a64692a3 | 720 | update_rq_clock(rq); |
1de64443 PZ |
721 | if (!(flags & DEQUEUE_SAVE)) |
722 | sched_info_dequeued(rq, p); | |
371fd7e7 | 723 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
724 | } |
725 | ||
029632fb | 726 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
727 | { |
728 | if (task_contributes_to_load(p)) | |
729 | rq->nr_uninterruptible--; | |
730 | ||
371fd7e7 | 731 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
732 | } |
733 | ||
029632fb | 734 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
735 | { |
736 | if (task_contributes_to_load(p)) | |
737 | rq->nr_uninterruptible++; | |
738 | ||
371fd7e7 | 739 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
740 | } |
741 | ||
fe44d621 | 742 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 743 | { |
095c0aa8 GC |
744 | /* |
745 | * In theory, the compile should just see 0 here, and optimize out the call | |
746 | * to sched_rt_avg_update. But I don't trust it... | |
747 | */ | |
748 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
749 | s64 steal = 0, irq_delta = 0; | |
750 | #endif | |
751 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 752 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
753 | |
754 | /* | |
755 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
756 | * this case when a previous update_rq_clock() happened inside a | |
757 | * {soft,}irq region. | |
758 | * | |
759 | * When this happens, we stop ->clock_task and only update the | |
760 | * prev_irq_time stamp to account for the part that fit, so that a next | |
761 | * update will consume the rest. This ensures ->clock_task is | |
762 | * monotonic. | |
763 | * | |
764 | * It does however cause some slight miss-attribution of {soft,}irq | |
765 | * time, a more accurate solution would be to update the irq_time using | |
766 | * the current rq->clock timestamp, except that would require using | |
767 | * atomic ops. | |
768 | */ | |
769 | if (irq_delta > delta) | |
770 | irq_delta = delta; | |
771 | ||
772 | rq->prev_irq_time += irq_delta; | |
773 | delta -= irq_delta; | |
095c0aa8 GC |
774 | #endif |
775 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 776 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
777 | steal = paravirt_steal_clock(cpu_of(rq)); |
778 | steal -= rq->prev_steal_time_rq; | |
779 | ||
780 | if (unlikely(steal > delta)) | |
781 | steal = delta; | |
782 | ||
095c0aa8 | 783 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
784 | delta -= steal; |
785 | } | |
786 | #endif | |
787 | ||
fe44d621 PZ |
788 | rq->clock_task += delta; |
789 | ||
095c0aa8 | 790 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 791 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
792 | sched_rt_avg_update(rq, irq_delta + steal); |
793 | #endif | |
aa483808 VP |
794 | } |
795 | ||
34f971f6 PZ |
796 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
797 | { | |
798 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
799 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
800 | ||
801 | if (stop) { | |
802 | /* | |
803 | * Make it appear like a SCHED_FIFO task, its something | |
804 | * userspace knows about and won't get confused about. | |
805 | * | |
806 | * Also, it will make PI more or less work without too | |
807 | * much confusion -- but then, stop work should not | |
808 | * rely on PI working anyway. | |
809 | */ | |
810 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
811 | ||
812 | stop->sched_class = &stop_sched_class; | |
813 | } | |
814 | ||
815 | cpu_rq(cpu)->stop = stop; | |
816 | ||
817 | if (old_stop) { | |
818 | /* | |
819 | * Reset it back to a normal scheduling class so that | |
820 | * it can die in pieces. | |
821 | */ | |
822 | old_stop->sched_class = &rt_sched_class; | |
823 | } | |
824 | } | |
825 | ||
14531189 | 826 | /* |
dd41f596 | 827 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 828 | */ |
14531189 IM |
829 | static inline int __normal_prio(struct task_struct *p) |
830 | { | |
dd41f596 | 831 | return p->static_prio; |
14531189 IM |
832 | } |
833 | ||
b29739f9 IM |
834 | /* |
835 | * Calculate the expected normal priority: i.e. priority | |
836 | * without taking RT-inheritance into account. Might be | |
837 | * boosted by interactivity modifiers. Changes upon fork, | |
838 | * setprio syscalls, and whenever the interactivity | |
839 | * estimator recalculates. | |
840 | */ | |
36c8b586 | 841 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
842 | { |
843 | int prio; | |
844 | ||
aab03e05 DF |
845 | if (task_has_dl_policy(p)) |
846 | prio = MAX_DL_PRIO-1; | |
847 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
848 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
849 | else | |
850 | prio = __normal_prio(p); | |
851 | return prio; | |
852 | } | |
853 | ||
854 | /* | |
855 | * Calculate the current priority, i.e. the priority | |
856 | * taken into account by the scheduler. This value might | |
857 | * be boosted by RT tasks, or might be boosted by | |
858 | * interactivity modifiers. Will be RT if the task got | |
859 | * RT-boosted. If not then it returns p->normal_prio. | |
860 | */ | |
36c8b586 | 861 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
862 | { |
863 | p->normal_prio = normal_prio(p); | |
864 | /* | |
865 | * If we are RT tasks or we were boosted to RT priority, | |
866 | * keep the priority unchanged. Otherwise, update priority | |
867 | * to the normal priority: | |
868 | */ | |
869 | if (!rt_prio(p->prio)) | |
870 | return p->normal_prio; | |
871 | return p->prio; | |
872 | } | |
873 | ||
1da177e4 LT |
874 | /** |
875 | * task_curr - is this task currently executing on a CPU? | |
876 | * @p: the task in question. | |
e69f6186 YB |
877 | * |
878 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 879 | */ |
36c8b586 | 880 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
881 | { |
882 | return cpu_curr(task_cpu(p)) == p; | |
883 | } | |
884 | ||
67dfa1b7 | 885 | /* |
4c9a4bc8 PZ |
886 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
887 | * use the balance_callback list if you want balancing. | |
888 | * | |
889 | * this means any call to check_class_changed() must be followed by a call to | |
890 | * balance_callback(). | |
67dfa1b7 | 891 | */ |
cb469845 SR |
892 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
893 | const struct sched_class *prev_class, | |
da7a735e | 894 | int oldprio) |
cb469845 SR |
895 | { |
896 | if (prev_class != p->sched_class) { | |
897 | if (prev_class->switched_from) | |
da7a735e | 898 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 899 | |
da7a735e | 900 | p->sched_class->switched_to(rq, p); |
2d3d891d | 901 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 902 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
903 | } |
904 | ||
029632fb | 905 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
906 | { |
907 | const struct sched_class *class; | |
908 | ||
909 | if (p->sched_class == rq->curr->sched_class) { | |
910 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
911 | } else { | |
912 | for_each_class(class) { | |
913 | if (class == rq->curr->sched_class) | |
914 | break; | |
915 | if (class == p->sched_class) { | |
8875125e | 916 | resched_curr(rq); |
1e5a7405 PZ |
917 | break; |
918 | } | |
919 | } | |
920 | } | |
921 | ||
922 | /* | |
923 | * A queue event has occurred, and we're going to schedule. In | |
924 | * this case, we can save a useless back to back clock update. | |
925 | */ | |
da0c1e65 | 926 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 927 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
928 | } |
929 | ||
1da177e4 | 930 | #ifdef CONFIG_SMP |
5cc389bc PZ |
931 | /* |
932 | * This is how migration works: | |
933 | * | |
934 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
935 | * stop_one_cpu(). | |
936 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
937 | * off the CPU) | |
938 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
939 | * 4) if it's in the wrong runqueue then the migration thread removes | |
940 | * it and puts it into the right queue. | |
941 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
942 | * is done. | |
943 | */ | |
944 | ||
945 | /* | |
946 | * move_queued_task - move a queued task to new rq. | |
947 | * | |
948 | * Returns (locked) new rq. Old rq's lock is released. | |
949 | */ | |
5e16bbc2 | 950 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 951 | { |
5cc389bc PZ |
952 | lockdep_assert_held(&rq->lock); |
953 | ||
5cc389bc | 954 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 955 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
956 | set_task_cpu(p, new_cpu); |
957 | raw_spin_unlock(&rq->lock); | |
958 | ||
959 | rq = cpu_rq(new_cpu); | |
960 | ||
961 | raw_spin_lock(&rq->lock); | |
962 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 963 | enqueue_task(rq, p, 0); |
3ea94de1 | 964 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
965 | check_preempt_curr(rq, p, 0); |
966 | ||
967 | return rq; | |
968 | } | |
969 | ||
970 | struct migration_arg { | |
971 | struct task_struct *task; | |
972 | int dest_cpu; | |
973 | }; | |
974 | ||
975 | /* | |
976 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
977 | * this because either it can't run here any more (set_cpus_allowed() | |
978 | * away from this CPU, or CPU going down), or because we're | |
979 | * attempting to rebalance this task on exec (sched_exec). | |
980 | * | |
981 | * So we race with normal scheduler movements, but that's OK, as long | |
982 | * as the task is no longer on this CPU. | |
5cc389bc | 983 | */ |
5e16bbc2 | 984 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 985 | { |
5cc389bc | 986 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 987 | return rq; |
5cc389bc PZ |
988 | |
989 | /* Affinity changed (again). */ | |
990 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 991 | return rq; |
5cc389bc | 992 | |
5e16bbc2 PZ |
993 | rq = move_queued_task(rq, p, dest_cpu); |
994 | ||
995 | return rq; | |
5cc389bc PZ |
996 | } |
997 | ||
998 | /* | |
999 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1000 | * and performs thread migration by bumping thread off CPU then | |
1001 | * 'pushing' onto another runqueue. | |
1002 | */ | |
1003 | static int migration_cpu_stop(void *data) | |
1004 | { | |
1005 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1006 | struct task_struct *p = arg->task; |
1007 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1008 | |
1009 | /* | |
1010 | * The original target cpu might have gone down and we might | |
1011 | * be on another cpu but it doesn't matter. | |
1012 | */ | |
1013 | local_irq_disable(); | |
1014 | /* | |
1015 | * We need to explicitly wake pending tasks before running | |
1016 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1017 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1018 | */ | |
1019 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1020 | |
1021 | raw_spin_lock(&p->pi_lock); | |
1022 | raw_spin_lock(&rq->lock); | |
1023 | /* | |
1024 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1025 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1026 | * we're holding p->pi_lock. | |
1027 | */ | |
1028 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1029 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1030 | raw_spin_unlock(&rq->lock); | |
1031 | raw_spin_unlock(&p->pi_lock); | |
1032 | ||
5cc389bc PZ |
1033 | local_irq_enable(); |
1034 | return 0; | |
1035 | } | |
1036 | ||
c5b28038 PZ |
1037 | /* |
1038 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1039 | * actually call this function. | |
1040 | */ | |
1041 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1042 | { |
5cc389bc PZ |
1043 | cpumask_copy(&p->cpus_allowed, new_mask); |
1044 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1045 | } | |
1046 | ||
c5b28038 PZ |
1047 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1048 | { | |
6c37067e PZ |
1049 | struct rq *rq = task_rq(p); |
1050 | bool queued, running; | |
1051 | ||
c5b28038 | 1052 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1053 | |
1054 | queued = task_on_rq_queued(p); | |
1055 | running = task_current(rq, p); | |
1056 | ||
1057 | if (queued) { | |
1058 | /* | |
1059 | * Because __kthread_bind() calls this on blocked tasks without | |
1060 | * holding rq->lock. | |
1061 | */ | |
1062 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1063 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1064 | } |
1065 | if (running) | |
1066 | put_prev_task(rq, p); | |
1067 | ||
c5b28038 | 1068 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e PZ |
1069 | |
1070 | if (running) | |
1071 | p->sched_class->set_curr_task(rq); | |
1072 | if (queued) | |
1de64443 | 1073 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
c5b28038 PZ |
1074 | } |
1075 | ||
5cc389bc PZ |
1076 | /* |
1077 | * Change a given task's CPU affinity. Migrate the thread to a | |
1078 | * proper CPU and schedule it away if the CPU it's executing on | |
1079 | * is removed from the allowed bitmask. | |
1080 | * | |
1081 | * NOTE: the caller must have a valid reference to the task, the | |
1082 | * task must not exit() & deallocate itself prematurely. The | |
1083 | * call is not atomic; no spinlocks may be held. | |
1084 | */ | |
25834c73 PZ |
1085 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1086 | const struct cpumask *new_mask, bool check) | |
5cc389bc PZ |
1087 | { |
1088 | unsigned long flags; | |
1089 | struct rq *rq; | |
1090 | unsigned int dest_cpu; | |
1091 | int ret = 0; | |
1092 | ||
1093 | rq = task_rq_lock(p, &flags); | |
1094 | ||
25834c73 PZ |
1095 | /* |
1096 | * Must re-check here, to close a race against __kthread_bind(), | |
1097 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1098 | */ | |
1099 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1100 | ret = -EINVAL; | |
1101 | goto out; | |
1102 | } | |
1103 | ||
5cc389bc PZ |
1104 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1105 | goto out; | |
1106 | ||
1107 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { | |
1108 | ret = -EINVAL; | |
1109 | goto out; | |
1110 | } | |
1111 | ||
1112 | do_set_cpus_allowed(p, new_mask); | |
1113 | ||
1114 | /* Can the task run on the task's current CPU? If so, we're done */ | |
1115 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1116 | goto out; | |
1117 | ||
1118 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); | |
1119 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
1120 | struct migration_arg arg = { p, dest_cpu }; | |
1121 | /* Need help from migration thread: drop lock and wait. */ | |
1122 | task_rq_unlock(rq, p, &flags); | |
1123 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
1124 | tlb_migrate_finish(p->mm); | |
1125 | return 0; | |
cbce1a68 PZ |
1126 | } else if (task_on_rq_queued(p)) { |
1127 | /* | |
1128 | * OK, since we're going to drop the lock immediately | |
1129 | * afterwards anyway. | |
1130 | */ | |
1131 | lockdep_unpin_lock(&rq->lock); | |
5e16bbc2 | 1132 | rq = move_queued_task(rq, p, dest_cpu); |
cbce1a68 PZ |
1133 | lockdep_pin_lock(&rq->lock); |
1134 | } | |
5cc389bc PZ |
1135 | out: |
1136 | task_rq_unlock(rq, p, &flags); | |
1137 | ||
1138 | return ret; | |
1139 | } | |
25834c73 PZ |
1140 | |
1141 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1142 | { | |
1143 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1144 | } | |
5cc389bc PZ |
1145 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1146 | ||
dd41f596 | 1147 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1148 | { |
e2912009 PZ |
1149 | #ifdef CONFIG_SCHED_DEBUG |
1150 | /* | |
1151 | * We should never call set_task_cpu() on a blocked task, | |
1152 | * ttwu() will sort out the placement. | |
1153 | */ | |
077614ee | 1154 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1155 | !p->on_rq); |
0122ec5b | 1156 | |
3ea94de1 JP |
1157 | /* |
1158 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1159 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1160 | * time relying on p->on_rq. | |
1161 | */ | |
1162 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1163 | p->sched_class == &fair_sched_class && | |
1164 | (p->on_rq && !task_on_rq_migrating(p))); | |
1165 | ||
0122ec5b | 1166 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1167 | /* |
1168 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1169 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1170 | * | |
1171 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1172 | * see task_group(). |
6c6c54e1 PZ |
1173 | * |
1174 | * Furthermore, all task_rq users should acquire both locks, see | |
1175 | * task_rq_lock(). | |
1176 | */ | |
0122ec5b PZ |
1177 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1178 | lockdep_is_held(&task_rq(p)->lock))); | |
1179 | #endif | |
e2912009 PZ |
1180 | #endif |
1181 | ||
de1d7286 | 1182 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1183 | |
0c69774e | 1184 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1185 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1186 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1187 | p->se.nr_migrations++; |
ff303e66 | 1188 | perf_event_task_migrate(p); |
0c69774e | 1189 | } |
dd41f596 IM |
1190 | |
1191 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1192 | } |
1193 | ||
ac66f547 PZ |
1194 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1195 | { | |
da0c1e65 | 1196 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1197 | struct rq *src_rq, *dst_rq; |
1198 | ||
1199 | src_rq = task_rq(p); | |
1200 | dst_rq = cpu_rq(cpu); | |
1201 | ||
3ea94de1 | 1202 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1203 | deactivate_task(src_rq, p, 0); |
1204 | set_task_cpu(p, cpu); | |
1205 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1206 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1207 | check_preempt_curr(dst_rq, p, 0); |
1208 | } else { | |
1209 | /* | |
1210 | * Task isn't running anymore; make it appear like we migrated | |
1211 | * it before it went to sleep. This means on wakeup we make the | |
1212 | * previous cpu our targer instead of where it really is. | |
1213 | */ | |
1214 | p->wake_cpu = cpu; | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | struct migration_swap_arg { | |
1219 | struct task_struct *src_task, *dst_task; | |
1220 | int src_cpu, dst_cpu; | |
1221 | }; | |
1222 | ||
1223 | static int migrate_swap_stop(void *data) | |
1224 | { | |
1225 | struct migration_swap_arg *arg = data; | |
1226 | struct rq *src_rq, *dst_rq; | |
1227 | int ret = -EAGAIN; | |
1228 | ||
62694cd5 PZ |
1229 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1230 | return -EAGAIN; | |
1231 | ||
ac66f547 PZ |
1232 | src_rq = cpu_rq(arg->src_cpu); |
1233 | dst_rq = cpu_rq(arg->dst_cpu); | |
1234 | ||
74602315 PZ |
1235 | double_raw_lock(&arg->src_task->pi_lock, |
1236 | &arg->dst_task->pi_lock); | |
ac66f547 | 1237 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1238 | |
ac66f547 PZ |
1239 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1240 | goto unlock; | |
1241 | ||
1242 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1243 | goto unlock; | |
1244 | ||
1245 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1246 | goto unlock; | |
1247 | ||
1248 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1249 | goto unlock; | |
1250 | ||
1251 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1252 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1253 | ||
1254 | ret = 0; | |
1255 | ||
1256 | unlock: | |
1257 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1258 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1259 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1260 | |
1261 | return ret; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * Cross migrate two tasks | |
1266 | */ | |
1267 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1268 | { | |
1269 | struct migration_swap_arg arg; | |
1270 | int ret = -EINVAL; | |
1271 | ||
ac66f547 PZ |
1272 | arg = (struct migration_swap_arg){ |
1273 | .src_task = cur, | |
1274 | .src_cpu = task_cpu(cur), | |
1275 | .dst_task = p, | |
1276 | .dst_cpu = task_cpu(p), | |
1277 | }; | |
1278 | ||
1279 | if (arg.src_cpu == arg.dst_cpu) | |
1280 | goto out; | |
1281 | ||
6acce3ef PZ |
1282 | /* |
1283 | * These three tests are all lockless; this is OK since all of them | |
1284 | * will be re-checked with proper locks held further down the line. | |
1285 | */ | |
ac66f547 PZ |
1286 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1287 | goto out; | |
1288 | ||
1289 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1290 | goto out; | |
1291 | ||
1292 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1293 | goto out; | |
1294 | ||
286549dc | 1295 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1296 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1297 | ||
1298 | out: | |
ac66f547 PZ |
1299 | return ret; |
1300 | } | |
1301 | ||
1da177e4 LT |
1302 | /* |
1303 | * wait_task_inactive - wait for a thread to unschedule. | |
1304 | * | |
85ba2d86 RM |
1305 | * If @match_state is nonzero, it's the @p->state value just checked and |
1306 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1307 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1308 | * we return a positive number (its total switch count). If a second call | |
1309 | * a short while later returns the same number, the caller can be sure that | |
1310 | * @p has remained unscheduled the whole time. | |
1311 | * | |
1da177e4 LT |
1312 | * The caller must ensure that the task *will* unschedule sometime soon, |
1313 | * else this function might spin for a *long* time. This function can't | |
1314 | * be called with interrupts off, or it may introduce deadlock with | |
1315 | * smp_call_function() if an IPI is sent by the same process we are | |
1316 | * waiting to become inactive. | |
1317 | */ | |
85ba2d86 | 1318 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1319 | { |
1320 | unsigned long flags; | |
da0c1e65 | 1321 | int running, queued; |
85ba2d86 | 1322 | unsigned long ncsw; |
70b97a7f | 1323 | struct rq *rq; |
1da177e4 | 1324 | |
3a5c359a AK |
1325 | for (;;) { |
1326 | /* | |
1327 | * We do the initial early heuristics without holding | |
1328 | * any task-queue locks at all. We'll only try to get | |
1329 | * the runqueue lock when things look like they will | |
1330 | * work out! | |
1331 | */ | |
1332 | rq = task_rq(p); | |
fa490cfd | 1333 | |
3a5c359a AK |
1334 | /* |
1335 | * If the task is actively running on another CPU | |
1336 | * still, just relax and busy-wait without holding | |
1337 | * any locks. | |
1338 | * | |
1339 | * NOTE! Since we don't hold any locks, it's not | |
1340 | * even sure that "rq" stays as the right runqueue! | |
1341 | * But we don't care, since "task_running()" will | |
1342 | * return false if the runqueue has changed and p | |
1343 | * is actually now running somewhere else! | |
1344 | */ | |
85ba2d86 RM |
1345 | while (task_running(rq, p)) { |
1346 | if (match_state && unlikely(p->state != match_state)) | |
1347 | return 0; | |
3a5c359a | 1348 | cpu_relax(); |
85ba2d86 | 1349 | } |
fa490cfd | 1350 | |
3a5c359a AK |
1351 | /* |
1352 | * Ok, time to look more closely! We need the rq | |
1353 | * lock now, to be *sure*. If we're wrong, we'll | |
1354 | * just go back and repeat. | |
1355 | */ | |
1356 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1357 | trace_sched_wait_task(p); |
3a5c359a | 1358 | running = task_running(rq, p); |
da0c1e65 | 1359 | queued = task_on_rq_queued(p); |
85ba2d86 | 1360 | ncsw = 0; |
f31e11d8 | 1361 | if (!match_state || p->state == match_state) |
93dcf55f | 1362 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1363 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1364 | |
85ba2d86 RM |
1365 | /* |
1366 | * If it changed from the expected state, bail out now. | |
1367 | */ | |
1368 | if (unlikely(!ncsw)) | |
1369 | break; | |
1370 | ||
3a5c359a AK |
1371 | /* |
1372 | * Was it really running after all now that we | |
1373 | * checked with the proper locks actually held? | |
1374 | * | |
1375 | * Oops. Go back and try again.. | |
1376 | */ | |
1377 | if (unlikely(running)) { | |
1378 | cpu_relax(); | |
1379 | continue; | |
1380 | } | |
fa490cfd | 1381 | |
3a5c359a AK |
1382 | /* |
1383 | * It's not enough that it's not actively running, | |
1384 | * it must be off the runqueue _entirely_, and not | |
1385 | * preempted! | |
1386 | * | |
80dd99b3 | 1387 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1388 | * running right now), it's preempted, and we should |
1389 | * yield - it could be a while. | |
1390 | */ | |
da0c1e65 | 1391 | if (unlikely(queued)) { |
8eb90c30 TG |
1392 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1393 | ||
1394 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1395 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1396 | continue; |
1397 | } | |
fa490cfd | 1398 | |
3a5c359a AK |
1399 | /* |
1400 | * Ahh, all good. It wasn't running, and it wasn't | |
1401 | * runnable, which means that it will never become | |
1402 | * running in the future either. We're all done! | |
1403 | */ | |
1404 | break; | |
1405 | } | |
85ba2d86 RM |
1406 | |
1407 | return ncsw; | |
1da177e4 LT |
1408 | } |
1409 | ||
1410 | /*** | |
1411 | * kick_process - kick a running thread to enter/exit the kernel | |
1412 | * @p: the to-be-kicked thread | |
1413 | * | |
1414 | * Cause a process which is running on another CPU to enter | |
1415 | * kernel-mode, without any delay. (to get signals handled.) | |
1416 | * | |
25985edc | 1417 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1418 | * because all it wants to ensure is that the remote task enters |
1419 | * the kernel. If the IPI races and the task has been migrated | |
1420 | * to another CPU then no harm is done and the purpose has been | |
1421 | * achieved as well. | |
1422 | */ | |
36c8b586 | 1423 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1424 | { |
1425 | int cpu; | |
1426 | ||
1427 | preempt_disable(); | |
1428 | cpu = task_cpu(p); | |
1429 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1430 | smp_send_reschedule(cpu); | |
1431 | preempt_enable(); | |
1432 | } | |
b43e3521 | 1433 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1434 | |
30da688e | 1435 | /* |
013fdb80 | 1436 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1437 | */ |
5da9a0fb PZ |
1438 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1439 | { | |
aa00d89c TC |
1440 | int nid = cpu_to_node(cpu); |
1441 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1442 | enum { cpuset, possible, fail } state = cpuset; |
1443 | int dest_cpu; | |
5da9a0fb | 1444 | |
aa00d89c TC |
1445 | /* |
1446 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1447 | * will return -1. There is no cpu on the node, and we should | |
1448 | * select the cpu on the other node. | |
1449 | */ | |
1450 | if (nid != -1) { | |
1451 | nodemask = cpumask_of_node(nid); | |
1452 | ||
1453 | /* Look for allowed, online CPU in same node. */ | |
1454 | for_each_cpu(dest_cpu, nodemask) { | |
1455 | if (!cpu_online(dest_cpu)) | |
1456 | continue; | |
1457 | if (!cpu_active(dest_cpu)) | |
1458 | continue; | |
1459 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1460 | return dest_cpu; | |
1461 | } | |
2baab4e9 | 1462 | } |
5da9a0fb | 1463 | |
2baab4e9 PZ |
1464 | for (;;) { |
1465 | /* Any allowed, online CPU? */ | |
e3831edd | 1466 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1467 | if (!cpu_online(dest_cpu)) |
1468 | continue; | |
1469 | if (!cpu_active(dest_cpu)) | |
1470 | continue; | |
1471 | goto out; | |
1472 | } | |
5da9a0fb | 1473 | |
e73e85f0 | 1474 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1475 | switch (state) { |
1476 | case cpuset: | |
e73e85f0 ON |
1477 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1478 | cpuset_cpus_allowed_fallback(p); | |
1479 | state = possible; | |
1480 | break; | |
1481 | } | |
1482 | /* fall-through */ | |
2baab4e9 PZ |
1483 | case possible: |
1484 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1485 | state = fail; | |
1486 | break; | |
1487 | ||
1488 | case fail: | |
1489 | BUG(); | |
1490 | break; | |
1491 | } | |
1492 | } | |
1493 | ||
1494 | out: | |
1495 | if (state != cpuset) { | |
1496 | /* | |
1497 | * Don't tell them about moving exiting tasks or | |
1498 | * kernel threads (both mm NULL), since they never | |
1499 | * leave kernel. | |
1500 | */ | |
1501 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1502 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1503 | task_pid_nr(p), p->comm, cpu); |
1504 | } | |
5da9a0fb PZ |
1505 | } |
1506 | ||
1507 | return dest_cpu; | |
1508 | } | |
1509 | ||
e2912009 | 1510 | /* |
013fdb80 | 1511 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1512 | */ |
970b13ba | 1513 | static inline |
ac66f547 | 1514 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1515 | { |
cbce1a68 PZ |
1516 | lockdep_assert_held(&p->pi_lock); |
1517 | ||
6c1d9410 WL |
1518 | if (p->nr_cpus_allowed > 1) |
1519 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e2912009 PZ |
1520 | |
1521 | /* | |
1522 | * In order not to call set_task_cpu() on a blocking task we need | |
1523 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1524 | * cpu. | |
1525 | * | |
1526 | * Since this is common to all placement strategies, this lives here. | |
1527 | * | |
1528 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1529 | * not worry about this generic constraint ] | |
1530 | */ | |
fa17b507 | 1531 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1532 | !cpu_online(cpu))) |
5da9a0fb | 1533 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1534 | |
1535 | return cpu; | |
970b13ba | 1536 | } |
09a40af5 MG |
1537 | |
1538 | static void update_avg(u64 *avg, u64 sample) | |
1539 | { | |
1540 | s64 diff = sample - *avg; | |
1541 | *avg += diff >> 3; | |
1542 | } | |
25834c73 PZ |
1543 | |
1544 | #else | |
1545 | ||
1546 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1547 | const struct cpumask *new_mask, bool check) | |
1548 | { | |
1549 | return set_cpus_allowed_ptr(p, new_mask); | |
1550 | } | |
1551 | ||
5cc389bc | 1552 | #endif /* CONFIG_SMP */ |
970b13ba | 1553 | |
d7c01d27 | 1554 | static void |
b84cb5df | 1555 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1556 | { |
d7c01d27 | 1557 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1558 | struct rq *rq = this_rq(); |
1559 | ||
d7c01d27 PZ |
1560 | #ifdef CONFIG_SMP |
1561 | int this_cpu = smp_processor_id(); | |
1562 | ||
1563 | if (cpu == this_cpu) { | |
1564 | schedstat_inc(rq, ttwu_local); | |
1565 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1566 | } else { | |
1567 | struct sched_domain *sd; | |
1568 | ||
1569 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1570 | rcu_read_lock(); |
d7c01d27 PZ |
1571 | for_each_domain(this_cpu, sd) { |
1572 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1573 | schedstat_inc(sd, ttwu_wake_remote); | |
1574 | break; | |
1575 | } | |
1576 | } | |
057f3fad | 1577 | rcu_read_unlock(); |
d7c01d27 | 1578 | } |
f339b9dc PZ |
1579 | |
1580 | if (wake_flags & WF_MIGRATED) | |
1581 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1582 | ||
d7c01d27 PZ |
1583 | #endif /* CONFIG_SMP */ |
1584 | ||
1585 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1586 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1587 | |
1588 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1589 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1590 | |
d7c01d27 PZ |
1591 | #endif /* CONFIG_SCHEDSTATS */ |
1592 | } | |
1593 | ||
1de64443 | 1594 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1595 | { |
9ed3811a | 1596 | activate_task(rq, p, en_flags); |
da0c1e65 | 1597 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1598 | |
1599 | /* if a worker is waking up, notify workqueue */ | |
1600 | if (p->flags & PF_WQ_WORKER) | |
1601 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1602 | } |
1603 | ||
23f41eeb PZ |
1604 | /* |
1605 | * Mark the task runnable and perform wakeup-preemption. | |
1606 | */ | |
89363381 | 1607 | static void |
23f41eeb | 1608 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1609 | { |
9ed3811a | 1610 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1611 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1612 | trace_sched_wakeup(p); |
1613 | ||
9ed3811a | 1614 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1615 | if (p->sched_class->task_woken) { |
1616 | /* | |
cbce1a68 PZ |
1617 | * Our task @p is fully woken up and running; so its safe to |
1618 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1619 | */ |
cbce1a68 | 1620 | lockdep_unpin_lock(&rq->lock); |
9ed3811a | 1621 | p->sched_class->task_woken(rq, p); |
cbce1a68 | 1622 | lockdep_pin_lock(&rq->lock); |
4c9a4bc8 | 1623 | } |
9ed3811a | 1624 | |
e69c6341 | 1625 | if (rq->idle_stamp) { |
78becc27 | 1626 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1627 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1628 | |
abfafa54 JL |
1629 | update_avg(&rq->avg_idle, delta); |
1630 | ||
1631 | if (rq->avg_idle > max) | |
9ed3811a | 1632 | rq->avg_idle = max; |
abfafa54 | 1633 | |
9ed3811a TH |
1634 | rq->idle_stamp = 0; |
1635 | } | |
1636 | #endif | |
1637 | } | |
1638 | ||
c05fbafb PZ |
1639 | static void |
1640 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1641 | { | |
cbce1a68 PZ |
1642 | lockdep_assert_held(&rq->lock); |
1643 | ||
c05fbafb PZ |
1644 | #ifdef CONFIG_SMP |
1645 | if (p->sched_contributes_to_load) | |
1646 | rq->nr_uninterruptible--; | |
1647 | #endif | |
1648 | ||
1649 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1650 | ttwu_do_wakeup(rq, p, wake_flags); | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1655 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1656 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1657 | * the task is still ->on_rq. | |
1658 | */ | |
1659 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1660 | { | |
1661 | struct rq *rq; | |
1662 | int ret = 0; | |
1663 | ||
1664 | rq = __task_rq_lock(p); | |
da0c1e65 | 1665 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1666 | /* check_preempt_curr() may use rq clock */ |
1667 | update_rq_clock(rq); | |
c05fbafb PZ |
1668 | ttwu_do_wakeup(rq, p, wake_flags); |
1669 | ret = 1; | |
1670 | } | |
1671 | __task_rq_unlock(rq); | |
1672 | ||
1673 | return ret; | |
1674 | } | |
1675 | ||
317f3941 | 1676 | #ifdef CONFIG_SMP |
e3baac47 | 1677 | void sched_ttwu_pending(void) |
317f3941 PZ |
1678 | { |
1679 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1680 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1681 | struct task_struct *p; | |
e3baac47 | 1682 | unsigned long flags; |
317f3941 | 1683 | |
e3baac47 PZ |
1684 | if (!llist) |
1685 | return; | |
1686 | ||
1687 | raw_spin_lock_irqsave(&rq->lock, flags); | |
cbce1a68 | 1688 | lockdep_pin_lock(&rq->lock); |
317f3941 | 1689 | |
fa14ff4a PZ |
1690 | while (llist) { |
1691 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1692 | llist = llist_next(llist); | |
317f3941 PZ |
1693 | ttwu_do_activate(rq, p, 0); |
1694 | } | |
1695 | ||
cbce1a68 | 1696 | lockdep_unpin_lock(&rq->lock); |
e3baac47 | 1697 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1698 | } |
1699 | ||
1700 | void scheduler_ipi(void) | |
1701 | { | |
f27dde8d PZ |
1702 | /* |
1703 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1704 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1705 | * this IPI. | |
1706 | */ | |
8cb75e0c | 1707 | preempt_fold_need_resched(); |
f27dde8d | 1708 | |
fd2ac4f4 | 1709 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1710 | return; |
1711 | ||
1712 | /* | |
1713 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1714 | * traditionally all their work was done from the interrupt return | |
1715 | * path. Now that we actually do some work, we need to make sure | |
1716 | * we do call them. | |
1717 | * | |
1718 | * Some archs already do call them, luckily irq_enter/exit nest | |
1719 | * properly. | |
1720 | * | |
1721 | * Arguably we should visit all archs and update all handlers, | |
1722 | * however a fair share of IPIs are still resched only so this would | |
1723 | * somewhat pessimize the simple resched case. | |
1724 | */ | |
1725 | irq_enter(); | |
fa14ff4a | 1726 | sched_ttwu_pending(); |
ca38062e SS |
1727 | |
1728 | /* | |
1729 | * Check if someone kicked us for doing the nohz idle load balance. | |
1730 | */ | |
873b4c65 | 1731 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1732 | this_rq()->idle_balance = 1; |
ca38062e | 1733 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1734 | } |
c5d753a5 | 1735 | irq_exit(); |
317f3941 PZ |
1736 | } |
1737 | ||
1738 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1739 | { | |
e3baac47 PZ |
1740 | struct rq *rq = cpu_rq(cpu); |
1741 | ||
1742 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1743 | if (!set_nr_if_polling(rq->idle)) | |
1744 | smp_send_reschedule(cpu); | |
1745 | else | |
1746 | trace_sched_wake_idle_without_ipi(cpu); | |
1747 | } | |
317f3941 | 1748 | } |
d6aa8f85 | 1749 | |
f6be8af1 CL |
1750 | void wake_up_if_idle(int cpu) |
1751 | { | |
1752 | struct rq *rq = cpu_rq(cpu); | |
1753 | unsigned long flags; | |
1754 | ||
fd7de1e8 AL |
1755 | rcu_read_lock(); |
1756 | ||
1757 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1758 | goto out; | |
f6be8af1 CL |
1759 | |
1760 | if (set_nr_if_polling(rq->idle)) { | |
1761 | trace_sched_wake_idle_without_ipi(cpu); | |
1762 | } else { | |
1763 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1764 | if (is_idle_task(rq->curr)) | |
1765 | smp_send_reschedule(cpu); | |
1766 | /* Else cpu is not in idle, do nothing here */ | |
1767 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1768 | } | |
fd7de1e8 AL |
1769 | |
1770 | out: | |
1771 | rcu_read_unlock(); | |
f6be8af1 CL |
1772 | } |
1773 | ||
39be3501 | 1774 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1775 | { |
1776 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1777 | } | |
d6aa8f85 | 1778 | #endif /* CONFIG_SMP */ |
317f3941 | 1779 | |
c05fbafb PZ |
1780 | static void ttwu_queue(struct task_struct *p, int cpu) |
1781 | { | |
1782 | struct rq *rq = cpu_rq(cpu); | |
1783 | ||
17d9f311 | 1784 | #if defined(CONFIG_SMP) |
39be3501 | 1785 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1786 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1787 | ttwu_queue_remote(p, cpu); |
1788 | return; | |
1789 | } | |
1790 | #endif | |
1791 | ||
c05fbafb | 1792 | raw_spin_lock(&rq->lock); |
cbce1a68 | 1793 | lockdep_pin_lock(&rq->lock); |
c05fbafb | 1794 | ttwu_do_activate(rq, p, 0); |
cbce1a68 | 1795 | lockdep_unpin_lock(&rq->lock); |
c05fbafb | 1796 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1797 | } |
1798 | ||
8643cda5 PZ |
1799 | /* |
1800 | * Notes on Program-Order guarantees on SMP systems. | |
1801 | * | |
1802 | * MIGRATION | |
1803 | * | |
1804 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
1805 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent | |
1806 | * execution on its new cpu [c1]. | |
1807 | * | |
1808 | * For migration (of runnable tasks) this is provided by the following means: | |
1809 | * | |
1810 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1811 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1812 | * rq(c1)->lock (if not at the same time, then in that order). | |
1813 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1814 | * | |
1815 | * Transitivity guarantees that B happens after A and C after B. | |
1816 | * Note: we only require RCpc transitivity. | |
1817 | * Note: the cpu doing B need not be c0 or c1 | |
1818 | * | |
1819 | * Example: | |
1820 | * | |
1821 | * CPU0 CPU1 CPU2 | |
1822 | * | |
1823 | * LOCK rq(0)->lock | |
1824 | * sched-out X | |
1825 | * sched-in Y | |
1826 | * UNLOCK rq(0)->lock | |
1827 | * | |
1828 | * LOCK rq(0)->lock // orders against CPU0 | |
1829 | * dequeue X | |
1830 | * UNLOCK rq(0)->lock | |
1831 | * | |
1832 | * LOCK rq(1)->lock | |
1833 | * enqueue X | |
1834 | * UNLOCK rq(1)->lock | |
1835 | * | |
1836 | * LOCK rq(1)->lock // orders against CPU2 | |
1837 | * sched-out Z | |
1838 | * sched-in X | |
1839 | * UNLOCK rq(1)->lock | |
1840 | * | |
1841 | * | |
1842 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1843 | * | |
1844 | * For blocking we (obviously) need to provide the same guarantee as for | |
1845 | * migration. However the means are completely different as there is no lock | |
1846 | * chain to provide order. Instead we do: | |
1847 | * | |
1848 | * 1) smp_store_release(X->on_cpu, 0) | |
1849 | * 2) smp_cond_acquire(!X->on_cpu) | |
1850 | * | |
1851 | * Example: | |
1852 | * | |
1853 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1854 | * | |
1855 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1856 | * dequeue X | |
1857 | * sched-out X | |
1858 | * smp_store_release(X->on_cpu, 0); | |
1859 | * | |
1860 | * smp_cond_acquire(!X->on_cpu); | |
1861 | * X->state = WAKING | |
1862 | * set_task_cpu(X,2) | |
1863 | * | |
1864 | * LOCK rq(2)->lock | |
1865 | * enqueue X | |
1866 | * X->state = RUNNING | |
1867 | * UNLOCK rq(2)->lock | |
1868 | * | |
1869 | * LOCK rq(2)->lock // orders against CPU1 | |
1870 | * sched-out Z | |
1871 | * sched-in X | |
1872 | * UNLOCK rq(2)->lock | |
1873 | * | |
1874 | * UNLOCK X->pi_lock | |
1875 | * UNLOCK rq(0)->lock | |
1876 | * | |
1877 | * | |
1878 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1879 | * must observe the state that lead to our wakeup. That is, not only must our | |
1880 | * task observe its own prior state, it must also observe the stores prior to | |
1881 | * its wakeup. | |
1882 | * | |
1883 | * This means that any means of doing remote wakeups must order the CPU doing | |
1884 | * the wakeup against the CPU the task is going to end up running on. This, | |
1885 | * however, is already required for the regular Program-Order guarantee above, | |
1886 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_acquire). | |
1887 | * | |
1888 | */ | |
1889 | ||
9ed3811a | 1890 | /** |
1da177e4 | 1891 | * try_to_wake_up - wake up a thread |
9ed3811a | 1892 | * @p: the thread to be awakened |
1da177e4 | 1893 | * @state: the mask of task states that can be woken |
9ed3811a | 1894 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1895 | * |
1896 | * Put it on the run-queue if it's not already there. The "current" | |
1897 | * thread is always on the run-queue (except when the actual | |
1898 | * re-schedule is in progress), and as such you're allowed to do | |
1899 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1900 | * runnable without the overhead of this. | |
1901 | * | |
e69f6186 | 1902 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1903 | * or @state didn't match @p's state. |
1da177e4 | 1904 | */ |
e4a52bcb PZ |
1905 | static int |
1906 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1907 | { |
1da177e4 | 1908 | unsigned long flags; |
c05fbafb | 1909 | int cpu, success = 0; |
2398f2c6 | 1910 | |
e0acd0a6 ON |
1911 | /* |
1912 | * If we are going to wake up a thread waiting for CONDITION we | |
1913 | * need to ensure that CONDITION=1 done by the caller can not be | |
1914 | * reordered with p->state check below. This pairs with mb() in | |
1915 | * set_current_state() the waiting thread does. | |
1916 | */ | |
1917 | smp_mb__before_spinlock(); | |
013fdb80 | 1918 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1919 | if (!(p->state & state)) |
1da177e4 LT |
1920 | goto out; |
1921 | ||
fbd705a0 PZ |
1922 | trace_sched_waking(p); |
1923 | ||
c05fbafb | 1924 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1925 | cpu = task_cpu(p); |
1da177e4 | 1926 | |
c05fbafb PZ |
1927 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1928 | goto stat; | |
1da177e4 | 1929 | |
1da177e4 | 1930 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
1931 | /* |
1932 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
1933 | * possible to, falsely, observe p->on_cpu == 0. | |
1934 | * | |
1935 | * One must be running (->on_cpu == 1) in order to remove oneself | |
1936 | * from the runqueue. | |
1937 | * | |
1938 | * [S] ->on_cpu = 1; [L] ->on_rq | |
1939 | * UNLOCK rq->lock | |
1940 | * RMB | |
1941 | * LOCK rq->lock | |
1942 | * [S] ->on_rq = 0; [L] ->on_cpu | |
1943 | * | |
1944 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
1945 | * from the consecutive calls to schedule(); the first switching to our | |
1946 | * task, the second putting it to sleep. | |
1947 | */ | |
1948 | smp_rmb(); | |
1949 | ||
e9c84311 | 1950 | /* |
c05fbafb PZ |
1951 | * If the owning (remote) cpu is still in the middle of schedule() with |
1952 | * this task as prev, wait until its done referencing the task. | |
b75a2253 PZ |
1953 | * |
1954 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
1955 | * | |
1956 | * This ensures that tasks getting woken will be fully ordered against | |
1957 | * their previous state and preserve Program Order. | |
0970d299 | 1958 | */ |
b3e0b1b6 | 1959 | smp_cond_acquire(!p->on_cpu); |
1da177e4 | 1960 | |
a8e4f2ea | 1961 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1962 | p->state = TASK_WAKING; |
e7693a36 | 1963 | |
e4a52bcb | 1964 | if (p->sched_class->task_waking) |
74f8e4b2 | 1965 | p->sched_class->task_waking(p); |
efbbd05a | 1966 | |
ac66f547 | 1967 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1968 | if (task_cpu(p) != cpu) { |
1969 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1970 | set_task_cpu(p, cpu); |
f339b9dc | 1971 | } |
1da177e4 | 1972 | #endif /* CONFIG_SMP */ |
1da177e4 | 1973 | |
c05fbafb PZ |
1974 | ttwu_queue(p, cpu); |
1975 | stat: | |
cb251765 MG |
1976 | if (schedstat_enabled()) |
1977 | ttwu_stat(p, cpu, wake_flags); | |
1da177e4 | 1978 | out: |
013fdb80 | 1979 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1980 | |
1981 | return success; | |
1982 | } | |
1983 | ||
21aa9af0 TH |
1984 | /** |
1985 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1986 | * @p: the thread to be awakened | |
1987 | * | |
2acca55e | 1988 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1989 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1990 | * the current task. |
21aa9af0 TH |
1991 | */ |
1992 | static void try_to_wake_up_local(struct task_struct *p) | |
1993 | { | |
1994 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1995 | |
383efcd0 TH |
1996 | if (WARN_ON_ONCE(rq != this_rq()) || |
1997 | WARN_ON_ONCE(p == current)) | |
1998 | return; | |
1999 | ||
21aa9af0 TH |
2000 | lockdep_assert_held(&rq->lock); |
2001 | ||
2acca55e | 2002 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2003 | /* |
2004 | * This is OK, because current is on_cpu, which avoids it being | |
2005 | * picked for load-balance and preemption/IRQs are still | |
2006 | * disabled avoiding further scheduler activity on it and we've | |
2007 | * not yet picked a replacement task. | |
2008 | */ | |
2009 | lockdep_unpin_lock(&rq->lock); | |
2acca55e PZ |
2010 | raw_spin_unlock(&rq->lock); |
2011 | raw_spin_lock(&p->pi_lock); | |
2012 | raw_spin_lock(&rq->lock); | |
cbce1a68 | 2013 | lockdep_pin_lock(&rq->lock); |
2acca55e PZ |
2014 | } |
2015 | ||
21aa9af0 | 2016 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2017 | goto out; |
21aa9af0 | 2018 | |
fbd705a0 PZ |
2019 | trace_sched_waking(p); |
2020 | ||
da0c1e65 | 2021 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2022 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2023 | ||
23f41eeb | 2024 | ttwu_do_wakeup(rq, p, 0); |
cb251765 MG |
2025 | if (schedstat_enabled()) |
2026 | ttwu_stat(p, smp_processor_id(), 0); | |
2acca55e PZ |
2027 | out: |
2028 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2029 | } |
2030 | ||
50fa610a DH |
2031 | /** |
2032 | * wake_up_process - Wake up a specific process | |
2033 | * @p: The process to be woken up. | |
2034 | * | |
2035 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2036 | * processes. |
2037 | * | |
2038 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2039 | * |
2040 | * It may be assumed that this function implies a write memory barrier before | |
2041 | * changing the task state if and only if any tasks are woken up. | |
2042 | */ | |
7ad5b3a5 | 2043 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2044 | { |
9067ac85 | 2045 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2046 | } |
1da177e4 LT |
2047 | EXPORT_SYMBOL(wake_up_process); |
2048 | ||
7ad5b3a5 | 2049 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2050 | { |
2051 | return try_to_wake_up(p, state, 0); | |
2052 | } | |
2053 | ||
a5e7be3b JL |
2054 | /* |
2055 | * This function clears the sched_dl_entity static params. | |
2056 | */ | |
2057 | void __dl_clear_params(struct task_struct *p) | |
2058 | { | |
2059 | struct sched_dl_entity *dl_se = &p->dl; | |
2060 | ||
2061 | dl_se->dl_runtime = 0; | |
2062 | dl_se->dl_deadline = 0; | |
2063 | dl_se->dl_period = 0; | |
2064 | dl_se->flags = 0; | |
2065 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2066 | |
2067 | dl_se->dl_throttled = 0; | |
40767b0d | 2068 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2069 | } |
2070 | ||
1da177e4 LT |
2071 | /* |
2072 | * Perform scheduler related setup for a newly forked process p. | |
2073 | * p is forked by current. | |
dd41f596 IM |
2074 | * |
2075 | * __sched_fork() is basic setup used by init_idle() too: | |
2076 | */ | |
5e1576ed | 2077 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2078 | { |
fd2f4419 PZ |
2079 | p->on_rq = 0; |
2080 | ||
2081 | p->se.on_rq = 0; | |
dd41f596 IM |
2082 | p->se.exec_start = 0; |
2083 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2084 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2085 | p->se.nr_migrations = 0; |
da7a735e | 2086 | p->se.vruntime = 0; |
fd2f4419 | 2087 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2088 | |
ad936d86 BP |
2089 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2090 | p->se.cfs_rq = NULL; | |
2091 | #endif | |
2092 | ||
6cfb0d5d | 2093 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2094 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2095 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2096 | #endif |
476d139c | 2097 | |
aab03e05 | 2098 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2099 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2100 | __dl_clear_params(p); |
aab03e05 | 2101 | |
fa717060 | 2102 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2103 | p->rt.timeout = 0; |
2104 | p->rt.time_slice = sched_rr_timeslice; | |
2105 | p->rt.on_rq = 0; | |
2106 | p->rt.on_list = 0; | |
476d139c | 2107 | |
e107be36 AK |
2108 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2109 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2110 | #endif | |
cbee9f88 PZ |
2111 | |
2112 | #ifdef CONFIG_NUMA_BALANCING | |
2113 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2114 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2115 | p->mm->numa_scan_seq = 0; |
2116 | } | |
2117 | ||
5e1576ed RR |
2118 | if (clone_flags & CLONE_VM) |
2119 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2120 | else | |
2121 | p->numa_preferred_nid = -1; | |
2122 | ||
cbee9f88 PZ |
2123 | p->node_stamp = 0ULL; |
2124 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2125 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2126 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2127 | p->numa_faults = NULL; |
7e2703e6 RR |
2128 | p->last_task_numa_placement = 0; |
2129 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2130 | |
8c8a743c | 2131 | p->numa_group = NULL; |
cbee9f88 | 2132 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2133 | } |
2134 | ||
2a595721 SD |
2135 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2136 | ||
1a687c2e | 2137 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2138 | |
1a687c2e MG |
2139 | void set_numabalancing_state(bool enabled) |
2140 | { | |
2141 | if (enabled) | |
2a595721 | 2142 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2143 | else |
2a595721 | 2144 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2145 | } |
54a43d54 AK |
2146 | |
2147 | #ifdef CONFIG_PROC_SYSCTL | |
2148 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2149 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2150 | { | |
2151 | struct ctl_table t; | |
2152 | int err; | |
2a595721 | 2153 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2154 | |
2155 | if (write && !capable(CAP_SYS_ADMIN)) | |
2156 | return -EPERM; | |
2157 | ||
2158 | t = *table; | |
2159 | t.data = &state; | |
2160 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2161 | if (err < 0) | |
2162 | return err; | |
2163 | if (write) | |
2164 | set_numabalancing_state(state); | |
2165 | return err; | |
2166 | } | |
2167 | #endif | |
2168 | #endif | |
dd41f596 | 2169 | |
cb251765 MG |
2170 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
2171 | ||
2172 | #ifdef CONFIG_SCHEDSTATS | |
2173 | static void set_schedstats(bool enabled) | |
2174 | { | |
2175 | if (enabled) | |
2176 | static_branch_enable(&sched_schedstats); | |
2177 | else | |
2178 | static_branch_disable(&sched_schedstats); | |
2179 | } | |
2180 | ||
2181 | void force_schedstat_enabled(void) | |
2182 | { | |
2183 | if (!schedstat_enabled()) { | |
2184 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2185 | static_branch_enable(&sched_schedstats); | |
2186 | } | |
2187 | } | |
2188 | ||
2189 | static int __init setup_schedstats(char *str) | |
2190 | { | |
2191 | int ret = 0; | |
2192 | if (!str) | |
2193 | goto out; | |
2194 | ||
2195 | if (!strcmp(str, "enable")) { | |
2196 | set_schedstats(true); | |
2197 | ret = 1; | |
2198 | } else if (!strcmp(str, "disable")) { | |
2199 | set_schedstats(false); | |
2200 | ret = 1; | |
2201 | } | |
2202 | out: | |
2203 | if (!ret) | |
2204 | pr_warn("Unable to parse schedstats=\n"); | |
2205 | ||
2206 | return ret; | |
2207 | } | |
2208 | __setup("schedstats=", setup_schedstats); | |
2209 | ||
2210 | #ifdef CONFIG_PROC_SYSCTL | |
2211 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2212 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2213 | { | |
2214 | struct ctl_table t; | |
2215 | int err; | |
2216 | int state = static_branch_likely(&sched_schedstats); | |
2217 | ||
2218 | if (write && !capable(CAP_SYS_ADMIN)) | |
2219 | return -EPERM; | |
2220 | ||
2221 | t = *table; | |
2222 | t.data = &state; | |
2223 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2224 | if (err < 0) | |
2225 | return err; | |
2226 | if (write) | |
2227 | set_schedstats(state); | |
2228 | return err; | |
2229 | } | |
2230 | #endif | |
2231 | #endif | |
dd41f596 IM |
2232 | |
2233 | /* | |
2234 | * fork()/clone()-time setup: | |
2235 | */ | |
aab03e05 | 2236 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2237 | { |
0122ec5b | 2238 | unsigned long flags; |
dd41f596 IM |
2239 | int cpu = get_cpu(); |
2240 | ||
5e1576ed | 2241 | __sched_fork(clone_flags, p); |
06b83b5f | 2242 | /* |
0017d735 | 2243 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2244 | * nobody will actually run it, and a signal or other external |
2245 | * event cannot wake it up and insert it on the runqueue either. | |
2246 | */ | |
0017d735 | 2247 | p->state = TASK_RUNNING; |
dd41f596 | 2248 | |
c350a04e MG |
2249 | /* |
2250 | * Make sure we do not leak PI boosting priority to the child. | |
2251 | */ | |
2252 | p->prio = current->normal_prio; | |
2253 | ||
b9dc29e7 MG |
2254 | /* |
2255 | * Revert to default priority/policy on fork if requested. | |
2256 | */ | |
2257 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2258 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2259 | p->policy = SCHED_NORMAL; |
6c697bdf | 2260 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2261 | p->rt_priority = 0; |
2262 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2263 | p->static_prio = NICE_TO_PRIO(0); | |
2264 | ||
2265 | p->prio = p->normal_prio = __normal_prio(p); | |
2266 | set_load_weight(p); | |
6c697bdf | 2267 | |
b9dc29e7 MG |
2268 | /* |
2269 | * We don't need the reset flag anymore after the fork. It has | |
2270 | * fulfilled its duty: | |
2271 | */ | |
2272 | p->sched_reset_on_fork = 0; | |
2273 | } | |
ca94c442 | 2274 | |
aab03e05 DF |
2275 | if (dl_prio(p->prio)) { |
2276 | put_cpu(); | |
2277 | return -EAGAIN; | |
2278 | } else if (rt_prio(p->prio)) { | |
2279 | p->sched_class = &rt_sched_class; | |
2280 | } else { | |
2ddbf952 | 2281 | p->sched_class = &fair_sched_class; |
aab03e05 | 2282 | } |
b29739f9 | 2283 | |
cd29fe6f PZ |
2284 | if (p->sched_class->task_fork) |
2285 | p->sched_class->task_fork(p); | |
2286 | ||
86951599 PZ |
2287 | /* |
2288 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2289 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2290 | * is ran before sched_fork(). | |
2291 | * | |
2292 | * Silence PROVE_RCU. | |
2293 | */ | |
0122ec5b | 2294 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2295 | set_task_cpu(p, cpu); |
0122ec5b | 2296 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2297 | |
f6db8347 | 2298 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2299 | if (likely(sched_info_on())) |
52f17b6c | 2300 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2301 | #endif |
3ca7a440 PZ |
2302 | #if defined(CONFIG_SMP) |
2303 | p->on_cpu = 0; | |
4866cde0 | 2304 | #endif |
01028747 | 2305 | init_task_preempt_count(p); |
806c09a7 | 2306 | #ifdef CONFIG_SMP |
917b627d | 2307 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2308 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2309 | #endif |
917b627d | 2310 | |
476d139c | 2311 | put_cpu(); |
aab03e05 | 2312 | return 0; |
1da177e4 LT |
2313 | } |
2314 | ||
332ac17e DF |
2315 | unsigned long to_ratio(u64 period, u64 runtime) |
2316 | { | |
2317 | if (runtime == RUNTIME_INF) | |
2318 | return 1ULL << 20; | |
2319 | ||
2320 | /* | |
2321 | * Doing this here saves a lot of checks in all | |
2322 | * the calling paths, and returning zero seems | |
2323 | * safe for them anyway. | |
2324 | */ | |
2325 | if (period == 0) | |
2326 | return 0; | |
2327 | ||
2328 | return div64_u64(runtime << 20, period); | |
2329 | } | |
2330 | ||
2331 | #ifdef CONFIG_SMP | |
2332 | inline struct dl_bw *dl_bw_of(int i) | |
2333 | { | |
f78f5b90 PM |
2334 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2335 | "sched RCU must be held"); | |
332ac17e DF |
2336 | return &cpu_rq(i)->rd->dl_bw; |
2337 | } | |
2338 | ||
de212f18 | 2339 | static inline int dl_bw_cpus(int i) |
332ac17e | 2340 | { |
de212f18 PZ |
2341 | struct root_domain *rd = cpu_rq(i)->rd; |
2342 | int cpus = 0; | |
2343 | ||
f78f5b90 PM |
2344 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2345 | "sched RCU must be held"); | |
de212f18 PZ |
2346 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2347 | cpus++; | |
2348 | ||
2349 | return cpus; | |
332ac17e DF |
2350 | } |
2351 | #else | |
2352 | inline struct dl_bw *dl_bw_of(int i) | |
2353 | { | |
2354 | return &cpu_rq(i)->dl.dl_bw; | |
2355 | } | |
2356 | ||
de212f18 | 2357 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2358 | { |
2359 | return 1; | |
2360 | } | |
2361 | #endif | |
2362 | ||
332ac17e DF |
2363 | /* |
2364 | * We must be sure that accepting a new task (or allowing changing the | |
2365 | * parameters of an existing one) is consistent with the bandwidth | |
2366 | * constraints. If yes, this function also accordingly updates the currently | |
2367 | * allocated bandwidth to reflect the new situation. | |
2368 | * | |
2369 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2370 | * |
2371 | * XXX we should delay bw change until the task's 0-lag point, see | |
2372 | * __setparam_dl(). | |
332ac17e DF |
2373 | */ |
2374 | static int dl_overflow(struct task_struct *p, int policy, | |
2375 | const struct sched_attr *attr) | |
2376 | { | |
2377 | ||
2378 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2379 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2380 | u64 runtime = attr->sched_runtime; |
2381 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2382 | int cpus, err = -1; |
332ac17e | 2383 | |
fec148c0 XP |
2384 | /* !deadline task may carry old deadline bandwidth */ |
2385 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2386 | return 0; |
2387 | ||
2388 | /* | |
2389 | * Either if a task, enters, leave, or stays -deadline but changes | |
2390 | * its parameters, we may need to update accordingly the total | |
2391 | * allocated bandwidth of the container. | |
2392 | */ | |
2393 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2394 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2395 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2396 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2397 | __dl_add(dl_b, new_bw); | |
2398 | err = 0; | |
2399 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2400 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2401 | __dl_clear(dl_b, p->dl.dl_bw); | |
2402 | __dl_add(dl_b, new_bw); | |
2403 | err = 0; | |
2404 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2405 | __dl_clear(dl_b, p->dl.dl_bw); | |
2406 | err = 0; | |
2407 | } | |
2408 | raw_spin_unlock(&dl_b->lock); | |
2409 | ||
2410 | return err; | |
2411 | } | |
2412 | ||
2413 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2414 | ||
1da177e4 LT |
2415 | /* |
2416 | * wake_up_new_task - wake up a newly created task for the first time. | |
2417 | * | |
2418 | * This function will do some initial scheduler statistics housekeeping | |
2419 | * that must be done for every newly created context, then puts the task | |
2420 | * on the runqueue and wakes it. | |
2421 | */ | |
3e51e3ed | 2422 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2423 | { |
2424 | unsigned long flags; | |
dd41f596 | 2425 | struct rq *rq; |
fabf318e | 2426 | |
ab2515c4 | 2427 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
98d8fd81 MR |
2428 | /* Initialize new task's runnable average */ |
2429 | init_entity_runnable_average(&p->se); | |
fabf318e PZ |
2430 | #ifdef CONFIG_SMP |
2431 | /* | |
2432 | * Fork balancing, do it here and not earlier because: | |
2433 | * - cpus_allowed can change in the fork path | |
2434 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2435 | */ |
ac66f547 | 2436 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2437 | #endif |
2b8c41da YD |
2438 | /* Post initialize new task's util average when its cfs_rq is set */ |
2439 | post_init_entity_util_avg(&p->se); | |
0017d735 | 2440 | |
ab2515c4 | 2441 | rq = __task_rq_lock(p); |
cd29fe6f | 2442 | activate_task(rq, p, 0); |
da0c1e65 | 2443 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2444 | trace_sched_wakeup_new(p); |
a7558e01 | 2445 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2446 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2447 | if (p->sched_class->task_woken) { |
2448 | /* | |
2449 | * Nothing relies on rq->lock after this, so its fine to | |
2450 | * drop it. | |
2451 | */ | |
2452 | lockdep_unpin_lock(&rq->lock); | |
efbbd05a | 2453 | p->sched_class->task_woken(rq, p); |
0aaafaab PZ |
2454 | lockdep_pin_lock(&rq->lock); |
2455 | } | |
9a897c5a | 2456 | #endif |
0122ec5b | 2457 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2458 | } |
2459 | ||
e107be36 AK |
2460 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2461 | ||
1cde2930 PZ |
2462 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2463 | ||
2ecd9d29 PZ |
2464 | void preempt_notifier_inc(void) |
2465 | { | |
2466 | static_key_slow_inc(&preempt_notifier_key); | |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2469 | ||
2470 | void preempt_notifier_dec(void) | |
2471 | { | |
2472 | static_key_slow_dec(&preempt_notifier_key); | |
2473 | } | |
2474 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2475 | ||
e107be36 | 2476 | /** |
80dd99b3 | 2477 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2478 | * @notifier: notifier struct to register |
e107be36 AK |
2479 | */ |
2480 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2481 | { | |
2ecd9d29 PZ |
2482 | if (!static_key_false(&preempt_notifier_key)) |
2483 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2484 | ||
e107be36 AK |
2485 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2486 | } | |
2487 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2488 | ||
2489 | /** | |
2490 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2491 | * @notifier: notifier struct to unregister |
e107be36 | 2492 | * |
d84525a8 | 2493 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2494 | */ |
2495 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2496 | { | |
2497 | hlist_del(¬ifier->link); | |
2498 | } | |
2499 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2500 | ||
1cde2930 | 2501 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2502 | { |
2503 | struct preempt_notifier *notifier; | |
e107be36 | 2504 | |
b67bfe0d | 2505 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2506 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2507 | } | |
2508 | ||
1cde2930 PZ |
2509 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2510 | { | |
2511 | if (static_key_false(&preempt_notifier_key)) | |
2512 | __fire_sched_in_preempt_notifiers(curr); | |
2513 | } | |
2514 | ||
e107be36 | 2515 | static void |
1cde2930 PZ |
2516 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2517 | struct task_struct *next) | |
e107be36 AK |
2518 | { |
2519 | struct preempt_notifier *notifier; | |
e107be36 | 2520 | |
b67bfe0d | 2521 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2522 | notifier->ops->sched_out(notifier, next); |
2523 | } | |
2524 | ||
1cde2930 PZ |
2525 | static __always_inline void |
2526 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2527 | struct task_struct *next) | |
2528 | { | |
2529 | if (static_key_false(&preempt_notifier_key)) | |
2530 | __fire_sched_out_preempt_notifiers(curr, next); | |
2531 | } | |
2532 | ||
6d6bc0ad | 2533 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2534 | |
1cde2930 | 2535 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2536 | { |
2537 | } | |
2538 | ||
1cde2930 | 2539 | static inline void |
e107be36 AK |
2540 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2541 | struct task_struct *next) | |
2542 | { | |
2543 | } | |
2544 | ||
6d6bc0ad | 2545 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2546 | |
4866cde0 NP |
2547 | /** |
2548 | * prepare_task_switch - prepare to switch tasks | |
2549 | * @rq: the runqueue preparing to switch | |
421cee29 | 2550 | * @prev: the current task that is being switched out |
4866cde0 NP |
2551 | * @next: the task we are going to switch to. |
2552 | * | |
2553 | * This is called with the rq lock held and interrupts off. It must | |
2554 | * be paired with a subsequent finish_task_switch after the context | |
2555 | * switch. | |
2556 | * | |
2557 | * prepare_task_switch sets up locking and calls architecture specific | |
2558 | * hooks. | |
2559 | */ | |
e107be36 AK |
2560 | static inline void |
2561 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2562 | struct task_struct *next) | |
4866cde0 | 2563 | { |
43148951 | 2564 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2565 | perf_event_task_sched_out(prev, next); |
e107be36 | 2566 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2567 | prepare_lock_switch(rq, next); |
2568 | prepare_arch_switch(next); | |
2569 | } | |
2570 | ||
1da177e4 LT |
2571 | /** |
2572 | * finish_task_switch - clean up after a task-switch | |
2573 | * @prev: the thread we just switched away from. | |
2574 | * | |
4866cde0 NP |
2575 | * finish_task_switch must be called after the context switch, paired |
2576 | * with a prepare_task_switch call before the context switch. | |
2577 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2578 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2579 | * |
2580 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2581 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2582 | * with the lock held can cause deadlocks; see schedule() for |
2583 | * details.) | |
dfa50b60 ON |
2584 | * |
2585 | * The context switch have flipped the stack from under us and restored the | |
2586 | * local variables which were saved when this task called schedule() in the | |
2587 | * past. prev == current is still correct but we need to recalculate this_rq | |
2588 | * because prev may have moved to another CPU. | |
1da177e4 | 2589 | */ |
dfa50b60 | 2590 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2591 | __releases(rq->lock) |
2592 | { | |
dfa50b60 | 2593 | struct rq *rq = this_rq(); |
1da177e4 | 2594 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2595 | long prev_state; |
1da177e4 | 2596 | |
609ca066 PZ |
2597 | /* |
2598 | * The previous task will have left us with a preempt_count of 2 | |
2599 | * because it left us after: | |
2600 | * | |
2601 | * schedule() | |
2602 | * preempt_disable(); // 1 | |
2603 | * __schedule() | |
2604 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2605 | * | |
2606 | * Also, see FORK_PREEMPT_COUNT. | |
2607 | */ | |
e2bf1c4b PZ |
2608 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2609 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2610 | current->comm, current->pid, preempt_count())) | |
2611 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2612 | |
1da177e4 LT |
2613 | rq->prev_mm = NULL; |
2614 | ||
2615 | /* | |
2616 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2617 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2618 | * schedule one last time. The schedule call will never return, and |
2619 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2620 | * |
2621 | * We must observe prev->state before clearing prev->on_cpu (in | |
2622 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2623 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2624 | * transition, resulting in a double drop. | |
1da177e4 | 2625 | */ |
55a101f8 | 2626 | prev_state = prev->state; |
bf9fae9f | 2627 | vtime_task_switch(prev); |
a8d757ef | 2628 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2629 | finish_lock_switch(rq, prev); |
01f23e16 | 2630 | finish_arch_post_lock_switch(); |
e8fa1362 | 2631 | |
e107be36 | 2632 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2633 | if (mm) |
2634 | mmdrop(mm); | |
c394cc9f | 2635 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2636 | if (prev->sched_class->task_dead) |
2637 | prev->sched_class->task_dead(prev); | |
2638 | ||
c6fd91f0 | 2639 | /* |
2640 | * Remove function-return probe instances associated with this | |
2641 | * task and put them back on the free list. | |
9761eea8 | 2642 | */ |
c6fd91f0 | 2643 | kprobe_flush_task(prev); |
1da177e4 | 2644 | put_task_struct(prev); |
c6fd91f0 | 2645 | } |
99e5ada9 | 2646 | |
de734f89 | 2647 | tick_nohz_task_switch(); |
dfa50b60 | 2648 | return rq; |
1da177e4 LT |
2649 | } |
2650 | ||
3f029d3c GH |
2651 | #ifdef CONFIG_SMP |
2652 | ||
3f029d3c | 2653 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2654 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2655 | { |
e3fca9e7 PZ |
2656 | struct callback_head *head, *next; |
2657 | void (*func)(struct rq *rq); | |
2658 | unsigned long flags; | |
3f029d3c | 2659 | |
e3fca9e7 PZ |
2660 | raw_spin_lock_irqsave(&rq->lock, flags); |
2661 | head = rq->balance_callback; | |
2662 | rq->balance_callback = NULL; | |
2663 | while (head) { | |
2664 | func = (void (*)(struct rq *))head->func; | |
2665 | next = head->next; | |
2666 | head->next = NULL; | |
2667 | head = next; | |
3f029d3c | 2668 | |
e3fca9e7 | 2669 | func(rq); |
3f029d3c | 2670 | } |
e3fca9e7 PZ |
2671 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2672 | } | |
2673 | ||
2674 | static inline void balance_callback(struct rq *rq) | |
2675 | { | |
2676 | if (unlikely(rq->balance_callback)) | |
2677 | __balance_callback(rq); | |
3f029d3c GH |
2678 | } |
2679 | ||
2680 | #else | |
da19ab51 | 2681 | |
e3fca9e7 | 2682 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2683 | { |
1da177e4 LT |
2684 | } |
2685 | ||
3f029d3c GH |
2686 | #endif |
2687 | ||
1da177e4 LT |
2688 | /** |
2689 | * schedule_tail - first thing a freshly forked thread must call. | |
2690 | * @prev: the thread we just switched away from. | |
2691 | */ | |
722a9f92 | 2692 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2693 | __releases(rq->lock) |
2694 | { | |
1a43a14a | 2695 | struct rq *rq; |
da19ab51 | 2696 | |
609ca066 PZ |
2697 | /* |
2698 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2699 | * finish_task_switch() for details. | |
2700 | * | |
2701 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2702 | * and the preempt_enable() will end up enabling preemption (on | |
2703 | * PREEMPT_COUNT kernels). | |
2704 | */ | |
2705 | ||
dfa50b60 | 2706 | rq = finish_task_switch(prev); |
e3fca9e7 | 2707 | balance_callback(rq); |
1a43a14a | 2708 | preempt_enable(); |
70b97a7f | 2709 | |
1da177e4 | 2710 | if (current->set_child_tid) |
b488893a | 2711 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2712 | } |
2713 | ||
2714 | /* | |
dfa50b60 | 2715 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2716 | */ |
04936948 | 2717 | static __always_inline struct rq * |
70b97a7f | 2718 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2719 | struct task_struct *next) |
1da177e4 | 2720 | { |
dd41f596 | 2721 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2722 | |
e107be36 | 2723 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2724 | |
dd41f596 IM |
2725 | mm = next->mm; |
2726 | oldmm = prev->active_mm; | |
9226d125 ZA |
2727 | /* |
2728 | * For paravirt, this is coupled with an exit in switch_to to | |
2729 | * combine the page table reload and the switch backend into | |
2730 | * one hypercall. | |
2731 | */ | |
224101ed | 2732 | arch_start_context_switch(prev); |
9226d125 | 2733 | |
31915ab4 | 2734 | if (!mm) { |
1da177e4 LT |
2735 | next->active_mm = oldmm; |
2736 | atomic_inc(&oldmm->mm_count); | |
2737 | enter_lazy_tlb(oldmm, next); | |
2738 | } else | |
f98db601 | 2739 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2740 | |
31915ab4 | 2741 | if (!prev->mm) { |
1da177e4 | 2742 | prev->active_mm = NULL; |
1da177e4 LT |
2743 | rq->prev_mm = oldmm; |
2744 | } | |
3a5f5e48 IM |
2745 | /* |
2746 | * Since the runqueue lock will be released by the next | |
2747 | * task (which is an invalid locking op but in the case | |
2748 | * of the scheduler it's an obvious special-case), so we | |
2749 | * do an early lockdep release here: | |
2750 | */ | |
cbce1a68 | 2751 | lockdep_unpin_lock(&rq->lock); |
8a25d5de | 2752 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2753 | |
2754 | /* Here we just switch the register state and the stack. */ | |
2755 | switch_to(prev, next, prev); | |
dd41f596 | 2756 | barrier(); |
dfa50b60 ON |
2757 | |
2758 | return finish_task_switch(prev); | |
1da177e4 LT |
2759 | } |
2760 | ||
2761 | /* | |
1c3e8264 | 2762 | * nr_running and nr_context_switches: |
1da177e4 LT |
2763 | * |
2764 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2765 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2766 | */ |
2767 | unsigned long nr_running(void) | |
2768 | { | |
2769 | unsigned long i, sum = 0; | |
2770 | ||
2771 | for_each_online_cpu(i) | |
2772 | sum += cpu_rq(i)->nr_running; | |
2773 | ||
2774 | return sum; | |
f711f609 | 2775 | } |
1da177e4 | 2776 | |
2ee507c4 TC |
2777 | /* |
2778 | * Check if only the current task is running on the cpu. | |
00cc1633 DD |
2779 | * |
2780 | * Caution: this function does not check that the caller has disabled | |
2781 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2782 | * race. The caller is responsible to use it correctly, for example: | |
2783 | * | |
2784 | * - from a non-preemptable section (of course) | |
2785 | * | |
2786 | * - from a thread that is bound to a single CPU | |
2787 | * | |
2788 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2789 | */ |
2790 | bool single_task_running(void) | |
2791 | { | |
00cc1633 | 2792 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2793 | } |
2794 | EXPORT_SYMBOL(single_task_running); | |
2795 | ||
1da177e4 | 2796 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2797 | { |
cc94abfc SR |
2798 | int i; |
2799 | unsigned long long sum = 0; | |
46cb4b7c | 2800 | |
0a945022 | 2801 | for_each_possible_cpu(i) |
1da177e4 | 2802 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2803 | |
1da177e4 LT |
2804 | return sum; |
2805 | } | |
483b4ee6 | 2806 | |
1da177e4 LT |
2807 | unsigned long nr_iowait(void) |
2808 | { | |
2809 | unsigned long i, sum = 0; | |
483b4ee6 | 2810 | |
0a945022 | 2811 | for_each_possible_cpu(i) |
1da177e4 | 2812 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2813 | |
1da177e4 LT |
2814 | return sum; |
2815 | } | |
483b4ee6 | 2816 | |
8c215bd3 | 2817 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2818 | { |
8c215bd3 | 2819 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2820 | return atomic_read(&this->nr_iowait); |
2821 | } | |
46cb4b7c | 2822 | |
372ba8cb MG |
2823 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2824 | { | |
3289bdb4 PZ |
2825 | struct rq *rq = this_rq(); |
2826 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2827 | *load = rq->load.weight; | |
372ba8cb MG |
2828 | } |
2829 | ||
dd41f596 | 2830 | #ifdef CONFIG_SMP |
8a0be9ef | 2831 | |
46cb4b7c | 2832 | /* |
38022906 PZ |
2833 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2834 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2835 | */ |
38022906 | 2836 | void sched_exec(void) |
46cb4b7c | 2837 | { |
38022906 | 2838 | struct task_struct *p = current; |
1da177e4 | 2839 | unsigned long flags; |
0017d735 | 2840 | int dest_cpu; |
46cb4b7c | 2841 | |
8f42ced9 | 2842 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2843 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2844 | if (dest_cpu == smp_processor_id()) |
2845 | goto unlock; | |
38022906 | 2846 | |
8f42ced9 | 2847 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2848 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2849 | |
8f42ced9 PZ |
2850 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2851 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2852 | return; |
2853 | } | |
0017d735 | 2854 | unlock: |
8f42ced9 | 2855 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2856 | } |
dd41f596 | 2857 | |
1da177e4 LT |
2858 | #endif |
2859 | ||
1da177e4 | 2860 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2861 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2862 | |
2863 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2864 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2865 | |
c5f8d995 HS |
2866 | /* |
2867 | * Return accounted runtime for the task. | |
2868 | * In case the task is currently running, return the runtime plus current's | |
2869 | * pending runtime that have not been accounted yet. | |
2870 | */ | |
2871 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2872 | { | |
2873 | unsigned long flags; | |
2874 | struct rq *rq; | |
6e998916 | 2875 | u64 ns; |
c5f8d995 | 2876 | |
911b2898 PZ |
2877 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2878 | /* | |
2879 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2880 | * So we have a optimization chance when the task's delta_exec is 0. | |
2881 | * Reading ->on_cpu is racy, but this is ok. | |
2882 | * | |
2883 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2884 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2885 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2886 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2887 | * been accounted, so we're correct here as well. | |
911b2898 | 2888 | */ |
da0c1e65 | 2889 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2890 | return p->se.sum_exec_runtime; |
2891 | #endif | |
2892 | ||
c5f8d995 | 2893 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2894 | /* |
2895 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2896 | * project cycles that may never be accounted to this | |
2897 | * thread, breaking clock_gettime(). | |
2898 | */ | |
2899 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2900 | update_rq_clock(rq); | |
2901 | p->sched_class->update_curr(rq); | |
2902 | } | |
2903 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2904 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2905 | |
2906 | return ns; | |
2907 | } | |
48f24c4d | 2908 | |
7835b98b CL |
2909 | /* |
2910 | * This function gets called by the timer code, with HZ frequency. | |
2911 | * We call it with interrupts disabled. | |
7835b98b CL |
2912 | */ |
2913 | void scheduler_tick(void) | |
2914 | { | |
7835b98b CL |
2915 | int cpu = smp_processor_id(); |
2916 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2917 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2918 | |
2919 | sched_clock_tick(); | |
dd41f596 | 2920 | |
05fa785c | 2921 | raw_spin_lock(&rq->lock); |
3e51f33f | 2922 | update_rq_clock(rq); |
fa85ae24 | 2923 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 2924 | cpu_load_update_active(rq); |
3289bdb4 | 2925 | calc_global_load_tick(rq); |
05fa785c | 2926 | raw_spin_unlock(&rq->lock); |
7835b98b | 2927 | |
e9d2b064 | 2928 | perf_event_task_tick(); |
e220d2dc | 2929 | |
e418e1c2 | 2930 | #ifdef CONFIG_SMP |
6eb57e0d | 2931 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2932 | trigger_load_balance(rq); |
e418e1c2 | 2933 | #endif |
265f22a9 | 2934 | rq_last_tick_reset(rq); |
1da177e4 LT |
2935 | } |
2936 | ||
265f22a9 FW |
2937 | #ifdef CONFIG_NO_HZ_FULL |
2938 | /** | |
2939 | * scheduler_tick_max_deferment | |
2940 | * | |
2941 | * Keep at least one tick per second when a single | |
2942 | * active task is running because the scheduler doesn't | |
2943 | * yet completely support full dynticks environment. | |
2944 | * | |
2945 | * This makes sure that uptime, CFS vruntime, load | |
2946 | * balancing, etc... continue to move forward, even | |
2947 | * with a very low granularity. | |
e69f6186 YB |
2948 | * |
2949 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2950 | */ |
2951 | u64 scheduler_tick_max_deferment(void) | |
2952 | { | |
2953 | struct rq *rq = this_rq(); | |
316c1608 | 2954 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
2955 | |
2956 | next = rq->last_sched_tick + HZ; | |
2957 | ||
2958 | if (time_before_eq(next, now)) | |
2959 | return 0; | |
2960 | ||
8fe8ff09 | 2961 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2962 | } |
265f22a9 | 2963 | #endif |
1da177e4 | 2964 | |
7e49fcce SR |
2965 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2966 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
2967 | /* |
2968 | * If the value passed in is equal to the current preempt count | |
2969 | * then we just disabled preemption. Start timing the latency. | |
2970 | */ | |
2971 | static inline void preempt_latency_start(int val) | |
2972 | { | |
2973 | if (preempt_count() == val) { | |
2974 | unsigned long ip = get_lock_parent_ip(); | |
2975 | #ifdef CONFIG_DEBUG_PREEMPT | |
2976 | current->preempt_disable_ip = ip; | |
2977 | #endif | |
2978 | trace_preempt_off(CALLER_ADDR0, ip); | |
2979 | } | |
2980 | } | |
7e49fcce | 2981 | |
edafe3a5 | 2982 | void preempt_count_add(int val) |
1da177e4 | 2983 | { |
6cd8a4bb | 2984 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2985 | /* |
2986 | * Underflow? | |
2987 | */ | |
9a11b49a IM |
2988 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2989 | return; | |
6cd8a4bb | 2990 | #endif |
bdb43806 | 2991 | __preempt_count_add(val); |
6cd8a4bb | 2992 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2993 | /* |
2994 | * Spinlock count overflowing soon? | |
2995 | */ | |
33859f7f MOS |
2996 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2997 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2998 | #endif |
47252cfb | 2999 | preempt_latency_start(val); |
1da177e4 | 3000 | } |
bdb43806 | 3001 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3002 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3003 | |
47252cfb SR |
3004 | /* |
3005 | * If the value passed in equals to the current preempt count | |
3006 | * then we just enabled preemption. Stop timing the latency. | |
3007 | */ | |
3008 | static inline void preempt_latency_stop(int val) | |
3009 | { | |
3010 | if (preempt_count() == val) | |
3011 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3012 | } | |
3013 | ||
edafe3a5 | 3014 | void preempt_count_sub(int val) |
1da177e4 | 3015 | { |
6cd8a4bb | 3016 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3017 | /* |
3018 | * Underflow? | |
3019 | */ | |
01e3eb82 | 3020 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3021 | return; |
1da177e4 LT |
3022 | /* |
3023 | * Is the spinlock portion underflowing? | |
3024 | */ | |
9a11b49a IM |
3025 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3026 | !(preempt_count() & PREEMPT_MASK))) | |
3027 | return; | |
6cd8a4bb | 3028 | #endif |
9a11b49a | 3029 | |
47252cfb | 3030 | preempt_latency_stop(val); |
bdb43806 | 3031 | __preempt_count_sub(val); |
1da177e4 | 3032 | } |
bdb43806 | 3033 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3034 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3035 | |
47252cfb SR |
3036 | #else |
3037 | static inline void preempt_latency_start(int val) { } | |
3038 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3039 | #endif |
3040 | ||
3041 | /* | |
dd41f596 | 3042 | * Print scheduling while atomic bug: |
1da177e4 | 3043 | */ |
dd41f596 | 3044 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3045 | { |
664dfa65 DJ |
3046 | if (oops_in_progress) |
3047 | return; | |
3048 | ||
3df0fc5b PZ |
3049 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3050 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3051 | |
dd41f596 | 3052 | debug_show_held_locks(prev); |
e21f5b15 | 3053 | print_modules(); |
dd41f596 IM |
3054 | if (irqs_disabled()) |
3055 | print_irqtrace_events(prev); | |
8f47b187 TG |
3056 | #ifdef CONFIG_DEBUG_PREEMPT |
3057 | if (in_atomic_preempt_off()) { | |
3058 | pr_err("Preemption disabled at:"); | |
3059 | print_ip_sym(current->preempt_disable_ip); | |
3060 | pr_cont("\n"); | |
3061 | } | |
3062 | #endif | |
6135fc1e | 3063 | dump_stack(); |
373d4d09 | 3064 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3065 | } |
1da177e4 | 3066 | |
dd41f596 IM |
3067 | /* |
3068 | * Various schedule()-time debugging checks and statistics: | |
3069 | */ | |
3070 | static inline void schedule_debug(struct task_struct *prev) | |
3071 | { | |
0d9e2632 | 3072 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
ce03e413 | 3073 | BUG_ON(task_stack_end_corrupted(prev)); |
0d9e2632 | 3074 | #endif |
b99def8b | 3075 | |
1dc0fffc | 3076 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3077 | __schedule_bug(prev); |
1dc0fffc PZ |
3078 | preempt_count_set(PREEMPT_DISABLED); |
3079 | } | |
b3fbab05 | 3080 | rcu_sleep_check(); |
dd41f596 | 3081 | |
1da177e4 LT |
3082 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3083 | ||
2d72376b | 3084 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3085 | } |
3086 | ||
3087 | /* | |
3088 | * Pick up the highest-prio task: | |
3089 | */ | |
3090 | static inline struct task_struct * | |
606dba2e | 3091 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3092 | { |
37e117c0 | 3093 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 3094 | struct task_struct *p; |
1da177e4 LT |
3095 | |
3096 | /* | |
dd41f596 IM |
3097 | * Optimization: we know that if all tasks are in |
3098 | * the fair class we can call that function directly: | |
1da177e4 | 3099 | */ |
37e117c0 | 3100 | if (likely(prev->sched_class == class && |
38033c37 | 3101 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 3102 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
3103 | if (unlikely(p == RETRY_TASK)) |
3104 | goto again; | |
3105 | ||
3106 | /* assumes fair_sched_class->next == idle_sched_class */ | |
3107 | if (unlikely(!p)) | |
3108 | p = idle_sched_class.pick_next_task(rq, prev); | |
3109 | ||
3110 | return p; | |
1da177e4 LT |
3111 | } |
3112 | ||
37e117c0 | 3113 | again: |
34f971f6 | 3114 | for_each_class(class) { |
606dba2e | 3115 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
3116 | if (p) { |
3117 | if (unlikely(p == RETRY_TASK)) | |
3118 | goto again; | |
dd41f596 | 3119 | return p; |
37e117c0 | 3120 | } |
dd41f596 | 3121 | } |
34f971f6 PZ |
3122 | |
3123 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3124 | } |
1da177e4 | 3125 | |
dd41f596 | 3126 | /* |
c259e01a | 3127 | * __schedule() is the main scheduler function. |
edde96ea PE |
3128 | * |
3129 | * The main means of driving the scheduler and thus entering this function are: | |
3130 | * | |
3131 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3132 | * | |
3133 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3134 | * paths. For example, see arch/x86/entry_64.S. | |
3135 | * | |
3136 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3137 | * interrupt handler scheduler_tick(). | |
3138 | * | |
3139 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3140 | * task to the run-queue and that's it. | |
3141 | * | |
3142 | * Now, if the new task added to the run-queue preempts the current | |
3143 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3144 | * called on the nearest possible occasion: | |
3145 | * | |
3146 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3147 | * | |
3148 | * - in syscall or exception context, at the next outmost | |
3149 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3150 | * spin_unlock()!) | |
3151 | * | |
3152 | * - in IRQ context, return from interrupt-handler to | |
3153 | * preemptible context | |
3154 | * | |
3155 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3156 | * then at the next: | |
3157 | * | |
3158 | * - cond_resched() call | |
3159 | * - explicit schedule() call | |
3160 | * - return from syscall or exception to user-space | |
3161 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3162 | * |
b30f0e3f | 3163 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3164 | */ |
499d7955 | 3165 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3166 | { |
3167 | struct task_struct *prev, *next; | |
67ca7bde | 3168 | unsigned long *switch_count; |
dd41f596 | 3169 | struct rq *rq; |
31656519 | 3170 | int cpu; |
dd41f596 | 3171 | |
dd41f596 IM |
3172 | cpu = smp_processor_id(); |
3173 | rq = cpu_rq(cpu); | |
dd41f596 | 3174 | prev = rq->curr; |
dd41f596 | 3175 | |
b99def8b PZ |
3176 | /* |
3177 | * do_exit() calls schedule() with preemption disabled as an exception; | |
3178 | * however we must fix that up, otherwise the next task will see an | |
3179 | * inconsistent (higher) preempt count. | |
3180 | * | |
3181 | * It also avoids the below schedule_debug() test from complaining | |
3182 | * about this. | |
3183 | */ | |
3184 | if (unlikely(prev->state == TASK_DEAD)) | |
3185 | preempt_enable_no_resched_notrace(); | |
3186 | ||
dd41f596 | 3187 | schedule_debug(prev); |
1da177e4 | 3188 | |
31656519 | 3189 | if (sched_feat(HRTICK)) |
f333fdc9 | 3190 | hrtick_clear(rq); |
8f4d37ec | 3191 | |
46a5d164 PM |
3192 | local_irq_disable(); |
3193 | rcu_note_context_switch(); | |
3194 | ||
e0acd0a6 ON |
3195 | /* |
3196 | * Make sure that signal_pending_state()->signal_pending() below | |
3197 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3198 | * done by the caller to avoid the race with signal_wake_up(). | |
3199 | */ | |
3200 | smp_mb__before_spinlock(); | |
46a5d164 | 3201 | raw_spin_lock(&rq->lock); |
cbce1a68 | 3202 | lockdep_pin_lock(&rq->lock); |
1da177e4 | 3203 | |
9edfbfed PZ |
3204 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3205 | ||
246d86b5 | 3206 | switch_count = &prev->nivcsw; |
fc13aeba | 3207 | if (!preempt && prev->state) { |
21aa9af0 | 3208 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3209 | prev->state = TASK_RUNNING; |
21aa9af0 | 3210 | } else { |
2acca55e PZ |
3211 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3212 | prev->on_rq = 0; | |
3213 | ||
21aa9af0 | 3214 | /* |
2acca55e PZ |
3215 | * If a worker went to sleep, notify and ask workqueue |
3216 | * whether it wants to wake up a task to maintain | |
3217 | * concurrency. | |
21aa9af0 TH |
3218 | */ |
3219 | if (prev->flags & PF_WQ_WORKER) { | |
3220 | struct task_struct *to_wakeup; | |
3221 | ||
9b7f6597 | 3222 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 TH |
3223 | if (to_wakeup) |
3224 | try_to_wake_up_local(to_wakeup); | |
3225 | } | |
21aa9af0 | 3226 | } |
dd41f596 | 3227 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3228 | } |
3229 | ||
9edfbfed | 3230 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3231 | update_rq_clock(rq); |
3232 | ||
3233 | next = pick_next_task(rq, prev); | |
f26f9aff | 3234 | clear_tsk_need_resched(prev); |
f27dde8d | 3235 | clear_preempt_need_resched(); |
9edfbfed | 3236 | rq->clock_skip_update = 0; |
1da177e4 | 3237 | |
1da177e4 | 3238 | if (likely(prev != next)) { |
1da177e4 LT |
3239 | rq->nr_switches++; |
3240 | rq->curr = next; | |
3241 | ++*switch_count; | |
3242 | ||
c73464b1 | 3243 | trace_sched_switch(preempt, prev, next); |
dfa50b60 | 3244 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
cbce1a68 PZ |
3245 | } else { |
3246 | lockdep_unpin_lock(&rq->lock); | |
05fa785c | 3247 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3248 | } |
1da177e4 | 3249 | |
e3fca9e7 | 3250 | balance_callback(rq); |
1da177e4 | 3251 | } |
8e05e96a | 3252 | STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */ |
c259e01a | 3253 | |
9c40cef2 TG |
3254 | static inline void sched_submit_work(struct task_struct *tsk) |
3255 | { | |
3c7d5184 | 3256 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3257 | return; |
3258 | /* | |
3259 | * If we are going to sleep and we have plugged IO queued, | |
3260 | * make sure to submit it to avoid deadlocks. | |
3261 | */ | |
3262 | if (blk_needs_flush_plug(tsk)) | |
3263 | blk_schedule_flush_plug(tsk); | |
3264 | } | |
3265 | ||
722a9f92 | 3266 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3267 | { |
9c40cef2 TG |
3268 | struct task_struct *tsk = current; |
3269 | ||
3270 | sched_submit_work(tsk); | |
bfd9b2b5 | 3271 | do { |
b30f0e3f | 3272 | preempt_disable(); |
fc13aeba | 3273 | __schedule(false); |
b30f0e3f | 3274 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3275 | } while (need_resched()); |
c259e01a | 3276 | } |
1da177e4 LT |
3277 | EXPORT_SYMBOL(schedule); |
3278 | ||
91d1aa43 | 3279 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3280 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3281 | { |
3282 | /* | |
3283 | * If we come here after a random call to set_need_resched(), | |
3284 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3285 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3286 | * we find a better solution. | |
7cc78f8f AL |
3287 | * |
3288 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3289 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3290 | * too frequently to make sense yet. |
20ab65e3 | 3291 | */ |
7cc78f8f | 3292 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3293 | schedule(); |
7cc78f8f | 3294 | exception_exit(prev_state); |
20ab65e3 FW |
3295 | } |
3296 | #endif | |
3297 | ||
c5491ea7 TG |
3298 | /** |
3299 | * schedule_preempt_disabled - called with preemption disabled | |
3300 | * | |
3301 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3302 | */ | |
3303 | void __sched schedule_preempt_disabled(void) | |
3304 | { | |
ba74c144 | 3305 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3306 | schedule(); |
3307 | preempt_disable(); | |
3308 | } | |
3309 | ||
06b1f808 | 3310 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3311 | { |
3312 | do { | |
47252cfb SR |
3313 | /* |
3314 | * Because the function tracer can trace preempt_count_sub() | |
3315 | * and it also uses preempt_enable/disable_notrace(), if | |
3316 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3317 | * by the function tracer will call this function again and | |
3318 | * cause infinite recursion. | |
3319 | * | |
3320 | * Preemption must be disabled here before the function | |
3321 | * tracer can trace. Break up preempt_disable() into two | |
3322 | * calls. One to disable preemption without fear of being | |
3323 | * traced. The other to still record the preemption latency, | |
3324 | * which can also be traced by the function tracer. | |
3325 | */ | |
499d7955 | 3326 | preempt_disable_notrace(); |
47252cfb | 3327 | preempt_latency_start(1); |
fc13aeba | 3328 | __schedule(true); |
47252cfb | 3329 | preempt_latency_stop(1); |
499d7955 | 3330 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3331 | |
3332 | /* | |
3333 | * Check again in case we missed a preemption opportunity | |
3334 | * between schedule and now. | |
3335 | */ | |
a18b5d01 FW |
3336 | } while (need_resched()); |
3337 | } | |
3338 | ||
1da177e4 LT |
3339 | #ifdef CONFIG_PREEMPT |
3340 | /* | |
2ed6e34f | 3341 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3342 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3343 | * occur there and call schedule directly. |
3344 | */ | |
722a9f92 | 3345 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3346 | { |
1da177e4 LT |
3347 | /* |
3348 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3349 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3350 | */ |
fbb00b56 | 3351 | if (likely(!preemptible())) |
1da177e4 LT |
3352 | return; |
3353 | ||
a18b5d01 | 3354 | preempt_schedule_common(); |
1da177e4 | 3355 | } |
376e2424 | 3356 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3357 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3358 | |
009f60e2 | 3359 | /** |
4eaca0a8 | 3360 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3361 | * |
3362 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3363 | * recursion and tracing preempt enabling caused by the tracing | |
3364 | * infrastructure itself. But as tracing can happen in areas coming | |
3365 | * from userspace or just about to enter userspace, a preempt enable | |
3366 | * can occur before user_exit() is called. This will cause the scheduler | |
3367 | * to be called when the system is still in usermode. | |
3368 | * | |
3369 | * To prevent this, the preempt_enable_notrace will use this function | |
3370 | * instead of preempt_schedule() to exit user context if needed before | |
3371 | * calling the scheduler. | |
3372 | */ | |
4eaca0a8 | 3373 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3374 | { |
3375 | enum ctx_state prev_ctx; | |
3376 | ||
3377 | if (likely(!preemptible())) | |
3378 | return; | |
3379 | ||
3380 | do { | |
47252cfb SR |
3381 | /* |
3382 | * Because the function tracer can trace preempt_count_sub() | |
3383 | * and it also uses preempt_enable/disable_notrace(), if | |
3384 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3385 | * by the function tracer will call this function again and | |
3386 | * cause infinite recursion. | |
3387 | * | |
3388 | * Preemption must be disabled here before the function | |
3389 | * tracer can trace. Break up preempt_disable() into two | |
3390 | * calls. One to disable preemption without fear of being | |
3391 | * traced. The other to still record the preemption latency, | |
3392 | * which can also be traced by the function tracer. | |
3393 | */ | |
3d8f74dd | 3394 | preempt_disable_notrace(); |
47252cfb | 3395 | preempt_latency_start(1); |
009f60e2 ON |
3396 | /* |
3397 | * Needs preempt disabled in case user_exit() is traced | |
3398 | * and the tracer calls preempt_enable_notrace() causing | |
3399 | * an infinite recursion. | |
3400 | */ | |
3401 | prev_ctx = exception_enter(); | |
fc13aeba | 3402 | __schedule(true); |
009f60e2 ON |
3403 | exception_exit(prev_ctx); |
3404 | ||
47252cfb | 3405 | preempt_latency_stop(1); |
3d8f74dd | 3406 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3407 | } while (need_resched()); |
3408 | } | |
4eaca0a8 | 3409 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3410 | |
32e475d7 | 3411 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3412 | |
3413 | /* | |
2ed6e34f | 3414 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3415 | * off of irq context. |
3416 | * Note, that this is called and return with irqs disabled. This will | |
3417 | * protect us against recursive calling from irq. | |
3418 | */ | |
722a9f92 | 3419 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3420 | { |
b22366cd | 3421 | enum ctx_state prev_state; |
6478d880 | 3422 | |
2ed6e34f | 3423 | /* Catch callers which need to be fixed */ |
f27dde8d | 3424 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3425 | |
b22366cd FW |
3426 | prev_state = exception_enter(); |
3427 | ||
3a5c359a | 3428 | do { |
3d8f74dd | 3429 | preempt_disable(); |
3a5c359a | 3430 | local_irq_enable(); |
fc13aeba | 3431 | __schedule(true); |
3a5c359a | 3432 | local_irq_disable(); |
3d8f74dd | 3433 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3434 | } while (need_resched()); |
b22366cd FW |
3435 | |
3436 | exception_exit(prev_state); | |
1da177e4 LT |
3437 | } |
3438 | ||
63859d4f | 3439 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3440 | void *key) |
1da177e4 | 3441 | { |
63859d4f | 3442 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3443 | } |
1da177e4 LT |
3444 | EXPORT_SYMBOL(default_wake_function); |
3445 | ||
b29739f9 IM |
3446 | #ifdef CONFIG_RT_MUTEXES |
3447 | ||
3448 | /* | |
3449 | * rt_mutex_setprio - set the current priority of a task | |
3450 | * @p: task | |
3451 | * @prio: prio value (kernel-internal form) | |
3452 | * | |
3453 | * This function changes the 'effective' priority of a task. It does | |
3454 | * not touch ->normal_prio like __setscheduler(). | |
3455 | * | |
c365c292 TG |
3456 | * Used by the rt_mutex code to implement priority inheritance |
3457 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3458 | */ |
36c8b586 | 3459 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3460 | { |
ff77e468 | 3461 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
70b97a7f | 3462 | struct rq *rq; |
83ab0aa0 | 3463 | const struct sched_class *prev_class; |
b29739f9 | 3464 | |
aab03e05 | 3465 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3466 | |
0122ec5b | 3467 | rq = __task_rq_lock(p); |
b29739f9 | 3468 | |
1c4dd99b TG |
3469 | /* |
3470 | * Idle task boosting is a nono in general. There is one | |
3471 | * exception, when PREEMPT_RT and NOHZ is active: | |
3472 | * | |
3473 | * The idle task calls get_next_timer_interrupt() and holds | |
3474 | * the timer wheel base->lock on the CPU and another CPU wants | |
3475 | * to access the timer (probably to cancel it). We can safely | |
3476 | * ignore the boosting request, as the idle CPU runs this code | |
3477 | * with interrupts disabled and will complete the lock | |
3478 | * protected section without being interrupted. So there is no | |
3479 | * real need to boost. | |
3480 | */ | |
3481 | if (unlikely(p == rq->idle)) { | |
3482 | WARN_ON(p != rq->curr); | |
3483 | WARN_ON(p->pi_blocked_on); | |
3484 | goto out_unlock; | |
3485 | } | |
3486 | ||
a8027073 | 3487 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3488 | oldprio = p->prio; |
ff77e468 PZ |
3489 | |
3490 | if (oldprio == prio) | |
3491 | queue_flag &= ~DEQUEUE_MOVE; | |
3492 | ||
83ab0aa0 | 3493 | prev_class = p->sched_class; |
da0c1e65 | 3494 | queued = task_on_rq_queued(p); |
051a1d1a | 3495 | running = task_current(rq, p); |
da0c1e65 | 3496 | if (queued) |
ff77e468 | 3497 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3498 | if (running) |
f3cd1c4e | 3499 | put_prev_task(rq, p); |
dd41f596 | 3500 | |
2d3d891d DF |
3501 | /* |
3502 | * Boosting condition are: | |
3503 | * 1. -rt task is running and holds mutex A | |
3504 | * --> -dl task blocks on mutex A | |
3505 | * | |
3506 | * 2. -dl task is running and holds mutex A | |
3507 | * --> -dl task blocks on mutex A and could preempt the | |
3508 | * running task | |
3509 | */ | |
3510 | if (dl_prio(prio)) { | |
466af29b ON |
3511 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3512 | if (!dl_prio(p->normal_prio) || | |
3513 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3514 | p->dl.dl_boosted = 1; |
ff77e468 | 3515 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3516 | } else |
3517 | p->dl.dl_boosted = 0; | |
aab03e05 | 3518 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3519 | } else if (rt_prio(prio)) { |
3520 | if (dl_prio(oldprio)) | |
3521 | p->dl.dl_boosted = 0; | |
3522 | if (oldprio < prio) | |
ff77e468 | 3523 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3524 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3525 | } else { |
3526 | if (dl_prio(oldprio)) | |
3527 | p->dl.dl_boosted = 0; | |
746db944 BS |
3528 | if (rt_prio(oldprio)) |
3529 | p->rt.timeout = 0; | |
dd41f596 | 3530 | p->sched_class = &fair_sched_class; |
2d3d891d | 3531 | } |
dd41f596 | 3532 | |
b29739f9 IM |
3533 | p->prio = prio; |
3534 | ||
0e1f3483 HS |
3535 | if (running) |
3536 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3537 | if (queued) |
ff77e468 | 3538 | enqueue_task(rq, p, queue_flag); |
cb469845 | 3539 | |
da7a735e | 3540 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3541 | out_unlock: |
4c9a4bc8 | 3542 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3543 | __task_rq_unlock(rq); |
4c9a4bc8 PZ |
3544 | |
3545 | balance_callback(rq); | |
3546 | preempt_enable(); | |
b29739f9 | 3547 | } |
b29739f9 | 3548 | #endif |
d50dde5a | 3549 | |
36c8b586 | 3550 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3551 | { |
da0c1e65 | 3552 | int old_prio, delta, queued; |
1da177e4 | 3553 | unsigned long flags; |
70b97a7f | 3554 | struct rq *rq; |
1da177e4 | 3555 | |
75e45d51 | 3556 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3557 | return; |
3558 | /* | |
3559 | * We have to be careful, if called from sys_setpriority(), | |
3560 | * the task might be in the middle of scheduling on another CPU. | |
3561 | */ | |
3562 | rq = task_rq_lock(p, &flags); | |
3563 | /* | |
3564 | * The RT priorities are set via sched_setscheduler(), but we still | |
3565 | * allow the 'normal' nice value to be set - but as expected | |
3566 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3567 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3568 | */ |
aab03e05 | 3569 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3570 | p->static_prio = NICE_TO_PRIO(nice); |
3571 | goto out_unlock; | |
3572 | } | |
da0c1e65 KT |
3573 | queued = task_on_rq_queued(p); |
3574 | if (queued) | |
1de64443 | 3575 | dequeue_task(rq, p, DEQUEUE_SAVE); |
1da177e4 | 3576 | |
1da177e4 | 3577 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3578 | set_load_weight(p); |
b29739f9 IM |
3579 | old_prio = p->prio; |
3580 | p->prio = effective_prio(p); | |
3581 | delta = p->prio - old_prio; | |
1da177e4 | 3582 | |
da0c1e65 | 3583 | if (queued) { |
1de64443 | 3584 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3585 | /* |
d5f9f942 AM |
3586 | * If the task increased its priority or is running and |
3587 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3588 | */ |
d5f9f942 | 3589 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3590 | resched_curr(rq); |
1da177e4 LT |
3591 | } |
3592 | out_unlock: | |
0122ec5b | 3593 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3594 | } |
1da177e4 LT |
3595 | EXPORT_SYMBOL(set_user_nice); |
3596 | ||
e43379f1 MM |
3597 | /* |
3598 | * can_nice - check if a task can reduce its nice value | |
3599 | * @p: task | |
3600 | * @nice: nice value | |
3601 | */ | |
36c8b586 | 3602 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3603 | { |
024f4747 | 3604 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3605 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3606 | |
78d7d407 | 3607 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3608 | capable(CAP_SYS_NICE)); |
3609 | } | |
3610 | ||
1da177e4 LT |
3611 | #ifdef __ARCH_WANT_SYS_NICE |
3612 | ||
3613 | /* | |
3614 | * sys_nice - change the priority of the current process. | |
3615 | * @increment: priority increment | |
3616 | * | |
3617 | * sys_setpriority is a more generic, but much slower function that | |
3618 | * does similar things. | |
3619 | */ | |
5add95d4 | 3620 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3621 | { |
48f24c4d | 3622 | long nice, retval; |
1da177e4 LT |
3623 | |
3624 | /* | |
3625 | * Setpriority might change our priority at the same moment. | |
3626 | * We don't have to worry. Conceptually one call occurs first | |
3627 | * and we have a single winner. | |
3628 | */ | |
a9467fa3 | 3629 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3630 | nice = task_nice(current) + increment; |
1da177e4 | 3631 | |
a9467fa3 | 3632 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3633 | if (increment < 0 && !can_nice(current, nice)) |
3634 | return -EPERM; | |
3635 | ||
1da177e4 LT |
3636 | retval = security_task_setnice(current, nice); |
3637 | if (retval) | |
3638 | return retval; | |
3639 | ||
3640 | set_user_nice(current, nice); | |
3641 | return 0; | |
3642 | } | |
3643 | ||
3644 | #endif | |
3645 | ||
3646 | /** | |
3647 | * task_prio - return the priority value of a given task. | |
3648 | * @p: the task in question. | |
3649 | * | |
e69f6186 | 3650 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3651 | * RT tasks are offset by -200. Normal tasks are centered |
3652 | * around 0, value goes from -16 to +15. | |
3653 | */ | |
36c8b586 | 3654 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3655 | { |
3656 | return p->prio - MAX_RT_PRIO; | |
3657 | } | |
3658 | ||
1da177e4 LT |
3659 | /** |
3660 | * idle_cpu - is a given cpu idle currently? | |
3661 | * @cpu: the processor in question. | |
e69f6186 YB |
3662 | * |
3663 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3664 | */ |
3665 | int idle_cpu(int cpu) | |
3666 | { | |
908a3283 TG |
3667 | struct rq *rq = cpu_rq(cpu); |
3668 | ||
3669 | if (rq->curr != rq->idle) | |
3670 | return 0; | |
3671 | ||
3672 | if (rq->nr_running) | |
3673 | return 0; | |
3674 | ||
3675 | #ifdef CONFIG_SMP | |
3676 | if (!llist_empty(&rq->wake_list)) | |
3677 | return 0; | |
3678 | #endif | |
3679 | ||
3680 | return 1; | |
1da177e4 LT |
3681 | } |
3682 | ||
1da177e4 LT |
3683 | /** |
3684 | * idle_task - return the idle task for a given cpu. | |
3685 | * @cpu: the processor in question. | |
e69f6186 YB |
3686 | * |
3687 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3688 | */ |
36c8b586 | 3689 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3690 | { |
3691 | return cpu_rq(cpu)->idle; | |
3692 | } | |
3693 | ||
3694 | /** | |
3695 | * find_process_by_pid - find a process with a matching PID value. | |
3696 | * @pid: the pid in question. | |
e69f6186 YB |
3697 | * |
3698 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3699 | */ |
a9957449 | 3700 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3701 | { |
228ebcbe | 3702 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3703 | } |
3704 | ||
aab03e05 DF |
3705 | /* |
3706 | * This function initializes the sched_dl_entity of a newly becoming | |
3707 | * SCHED_DEADLINE task. | |
3708 | * | |
3709 | * Only the static values are considered here, the actual runtime and the | |
3710 | * absolute deadline will be properly calculated when the task is enqueued | |
3711 | * for the first time with its new policy. | |
3712 | */ | |
3713 | static void | |
3714 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3715 | { | |
3716 | struct sched_dl_entity *dl_se = &p->dl; | |
3717 | ||
aab03e05 DF |
3718 | dl_se->dl_runtime = attr->sched_runtime; |
3719 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3720 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3721 | dl_se->flags = attr->sched_flags; |
332ac17e | 3722 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3723 | |
3724 | /* | |
3725 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3726 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3727 | * | |
3728 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3729 | * point. This would include retaining the task_struct until that time | |
3730 | * and change dl_overflow() to not immediately decrement the current | |
3731 | * amount. | |
3732 | * | |
3733 | * Instead we retain the current runtime/deadline and let the new | |
3734 | * parameters take effect after the current reservation period lapses. | |
3735 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3736 | * before the current scheduling deadline. | |
3737 | * | |
3738 | * We can still have temporary overloads because we do not delay the | |
3739 | * change in bandwidth until that time; so admission control is | |
3740 | * not on the safe side. It does however guarantee tasks will never | |
3741 | * consume more than promised. | |
3742 | */ | |
aab03e05 DF |
3743 | } |
3744 | ||
c13db6b1 SR |
3745 | /* |
3746 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3747 | * it calls know not to change it. | |
3748 | */ | |
3749 | #define SETPARAM_POLICY -1 | |
3750 | ||
c365c292 TG |
3751 | static void __setscheduler_params(struct task_struct *p, |
3752 | const struct sched_attr *attr) | |
1da177e4 | 3753 | { |
d50dde5a DF |
3754 | int policy = attr->sched_policy; |
3755 | ||
c13db6b1 | 3756 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3757 | policy = p->policy; |
3758 | ||
1da177e4 | 3759 | p->policy = policy; |
d50dde5a | 3760 | |
aab03e05 DF |
3761 | if (dl_policy(policy)) |
3762 | __setparam_dl(p, attr); | |
39fd8fd2 | 3763 | else if (fair_policy(policy)) |
d50dde5a DF |
3764 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3765 | ||
39fd8fd2 PZ |
3766 | /* |
3767 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3768 | * !rt_policy. Always setting this ensures that things like | |
3769 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3770 | */ | |
3771 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3772 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3773 | set_load_weight(p); |
3774 | } | |
39fd8fd2 | 3775 | |
c365c292 TG |
3776 | /* Actually do priority change: must hold pi & rq lock. */ |
3777 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3778 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3779 | { |
3780 | __setscheduler_params(p, attr); | |
d50dde5a | 3781 | |
383afd09 | 3782 | /* |
0782e63b TG |
3783 | * Keep a potential priority boosting if called from |
3784 | * sched_setscheduler(). | |
383afd09 | 3785 | */ |
0782e63b TG |
3786 | if (keep_boost) |
3787 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3788 | else | |
3789 | p->prio = normal_prio(p); | |
383afd09 | 3790 | |
aab03e05 DF |
3791 | if (dl_prio(p->prio)) |
3792 | p->sched_class = &dl_sched_class; | |
3793 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3794 | p->sched_class = &rt_sched_class; |
3795 | else | |
3796 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3797 | } |
aab03e05 DF |
3798 | |
3799 | static void | |
3800 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3801 | { | |
3802 | struct sched_dl_entity *dl_se = &p->dl; | |
3803 | ||
3804 | attr->sched_priority = p->rt_priority; | |
3805 | attr->sched_runtime = dl_se->dl_runtime; | |
3806 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3807 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3808 | attr->sched_flags = dl_se->flags; |
3809 | } | |
3810 | ||
3811 | /* | |
3812 | * This function validates the new parameters of a -deadline task. | |
3813 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3814 | * than the runtime, as well as the period of being zero or |
332ac17e | 3815 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3816 | * user parameters are above the internal resolution of 1us (we |
3817 | * check sched_runtime only since it is always the smaller one) and | |
3818 | * below 2^63 ns (we have to check both sched_deadline and | |
3819 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3820 | */ |
3821 | static bool | |
3822 | __checkparam_dl(const struct sched_attr *attr) | |
3823 | { | |
b0827819 JL |
3824 | /* deadline != 0 */ |
3825 | if (attr->sched_deadline == 0) | |
3826 | return false; | |
3827 | ||
3828 | /* | |
3829 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3830 | * that big. | |
3831 | */ | |
3832 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3833 | return false; | |
3834 | ||
3835 | /* | |
3836 | * Since we use the MSB for wrap-around and sign issues, make | |
3837 | * sure it's not set (mind that period can be equal to zero). | |
3838 | */ | |
3839 | if (attr->sched_deadline & (1ULL << 63) || | |
3840 | attr->sched_period & (1ULL << 63)) | |
3841 | return false; | |
3842 | ||
3843 | /* runtime <= deadline <= period (if period != 0) */ | |
3844 | if ((attr->sched_period != 0 && | |
3845 | attr->sched_period < attr->sched_deadline) || | |
3846 | attr->sched_deadline < attr->sched_runtime) | |
3847 | return false; | |
3848 | ||
3849 | return true; | |
aab03e05 DF |
3850 | } |
3851 | ||
c69e8d9c DH |
3852 | /* |
3853 | * check the target process has a UID that matches the current process's | |
3854 | */ | |
3855 | static bool check_same_owner(struct task_struct *p) | |
3856 | { | |
3857 | const struct cred *cred = current_cred(), *pcred; | |
3858 | bool match; | |
3859 | ||
3860 | rcu_read_lock(); | |
3861 | pcred = __task_cred(p); | |
9c806aa0 EB |
3862 | match = (uid_eq(cred->euid, pcred->euid) || |
3863 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3864 | rcu_read_unlock(); |
3865 | return match; | |
3866 | } | |
3867 | ||
75381608 WL |
3868 | static bool dl_param_changed(struct task_struct *p, |
3869 | const struct sched_attr *attr) | |
3870 | { | |
3871 | struct sched_dl_entity *dl_se = &p->dl; | |
3872 | ||
3873 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3874 | dl_se->dl_deadline != attr->sched_deadline || | |
3875 | dl_se->dl_period != attr->sched_period || | |
3876 | dl_se->flags != attr->sched_flags) | |
3877 | return true; | |
3878 | ||
3879 | return false; | |
3880 | } | |
3881 | ||
d50dde5a DF |
3882 | static int __sched_setscheduler(struct task_struct *p, |
3883 | const struct sched_attr *attr, | |
dbc7f069 | 3884 | bool user, bool pi) |
1da177e4 | 3885 | { |
383afd09 SR |
3886 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3887 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3888 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3889 | int new_effective_prio, policy = attr->sched_policy; |
1da177e4 | 3890 | unsigned long flags; |
83ab0aa0 | 3891 | const struct sched_class *prev_class; |
70b97a7f | 3892 | struct rq *rq; |
ca94c442 | 3893 | int reset_on_fork; |
ff77e468 | 3894 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
1da177e4 | 3895 | |
66e5393a SR |
3896 | /* may grab non-irq protected spin_locks */ |
3897 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3898 | recheck: |
3899 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3900 | if (policy < 0) { |
3901 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3902 | policy = oldpolicy = p->policy; |
ca94c442 | 3903 | } else { |
7479f3c9 | 3904 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3905 | |
20f9cd2a | 3906 | if (!valid_policy(policy)) |
ca94c442 LP |
3907 | return -EINVAL; |
3908 | } | |
3909 | ||
7479f3c9 PZ |
3910 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3911 | return -EINVAL; | |
3912 | ||
1da177e4 LT |
3913 | /* |
3914 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3915 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3916 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3917 | */ |
0bb040a4 | 3918 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3919 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3920 | return -EINVAL; |
aab03e05 DF |
3921 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3922 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3923 | return -EINVAL; |
3924 | ||
37e4ab3f OC |
3925 | /* |
3926 | * Allow unprivileged RT tasks to decrease priority: | |
3927 | */ | |
961ccddd | 3928 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3929 | if (fair_policy(policy)) { |
d0ea0268 | 3930 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3931 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3932 | return -EPERM; |
3933 | } | |
3934 | ||
e05606d3 | 3935 | if (rt_policy(policy)) { |
a44702e8 ON |
3936 | unsigned long rlim_rtprio = |
3937 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3938 | |
3939 | /* can't set/change the rt policy */ | |
3940 | if (policy != p->policy && !rlim_rtprio) | |
3941 | return -EPERM; | |
3942 | ||
3943 | /* can't increase priority */ | |
d50dde5a DF |
3944 | if (attr->sched_priority > p->rt_priority && |
3945 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3946 | return -EPERM; |
3947 | } | |
c02aa73b | 3948 | |
d44753b8 JL |
3949 | /* |
3950 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3951 | * (safest behavior); in the future we would like to allow | |
3952 | * unprivileged DL tasks to increase their relative deadline | |
3953 | * or reduce their runtime (both ways reducing utilization) | |
3954 | */ | |
3955 | if (dl_policy(policy)) | |
3956 | return -EPERM; | |
3957 | ||
dd41f596 | 3958 | /* |
c02aa73b DH |
3959 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3960 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3961 | */ |
20f9cd2a | 3962 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 3963 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3964 | return -EPERM; |
3965 | } | |
5fe1d75f | 3966 | |
37e4ab3f | 3967 | /* can't change other user's priorities */ |
c69e8d9c | 3968 | if (!check_same_owner(p)) |
37e4ab3f | 3969 | return -EPERM; |
ca94c442 LP |
3970 | |
3971 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3972 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3973 | return -EPERM; | |
37e4ab3f | 3974 | } |
1da177e4 | 3975 | |
725aad24 | 3976 | if (user) { |
b0ae1981 | 3977 | retval = security_task_setscheduler(p); |
725aad24 JF |
3978 | if (retval) |
3979 | return retval; | |
3980 | } | |
3981 | ||
b29739f9 IM |
3982 | /* |
3983 | * make sure no PI-waiters arrive (or leave) while we are | |
3984 | * changing the priority of the task: | |
0122ec5b | 3985 | * |
25985edc | 3986 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3987 | * runqueue lock must be held. |
3988 | */ | |
0122ec5b | 3989 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3990 | |
34f971f6 PZ |
3991 | /* |
3992 | * Changing the policy of the stop threads its a very bad idea | |
3993 | */ | |
3994 | if (p == rq->stop) { | |
0122ec5b | 3995 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3996 | return -EINVAL; |
3997 | } | |
3998 | ||
a51e9198 | 3999 | /* |
d6b1e911 TG |
4000 | * If not changing anything there's no need to proceed further, |
4001 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4002 | */ |
d50dde5a | 4003 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4004 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4005 | goto change; |
4006 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4007 | goto change; | |
75381608 | 4008 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4009 | goto change; |
d50dde5a | 4010 | |
d6b1e911 | 4011 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 4012 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
4013 | return 0; |
4014 | } | |
d50dde5a | 4015 | change: |
a51e9198 | 4016 | |
dc61b1d6 | 4017 | if (user) { |
332ac17e | 4018 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4019 | /* |
4020 | * Do not allow realtime tasks into groups that have no runtime | |
4021 | * assigned. | |
4022 | */ | |
4023 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4024 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4025 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4026 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4027 | return -EPERM; |
4028 | } | |
dc61b1d6 | 4029 | #endif |
332ac17e DF |
4030 | #ifdef CONFIG_SMP |
4031 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4032 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4033 | |
4034 | /* | |
4035 | * Don't allow tasks with an affinity mask smaller than | |
4036 | * the entire root_domain to become SCHED_DEADLINE. We | |
4037 | * will also fail if there's no bandwidth available. | |
4038 | */ | |
e4099a5e PZ |
4039 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4040 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
4041 | task_rq_unlock(rq, p, &flags); |
4042 | return -EPERM; | |
4043 | } | |
4044 | } | |
4045 | #endif | |
4046 | } | |
dc61b1d6 | 4047 | |
1da177e4 LT |
4048 | /* recheck policy now with rq lock held */ |
4049 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4050 | policy = oldpolicy = -1; | |
0122ec5b | 4051 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4052 | goto recheck; |
4053 | } | |
332ac17e DF |
4054 | |
4055 | /* | |
4056 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4057 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4058 | * is available. | |
4059 | */ | |
e4099a5e | 4060 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
4061 | task_rq_unlock(rq, p, &flags); |
4062 | return -EBUSY; | |
4063 | } | |
4064 | ||
c365c292 TG |
4065 | p->sched_reset_on_fork = reset_on_fork; |
4066 | oldprio = p->prio; | |
4067 | ||
dbc7f069 PZ |
4068 | if (pi) { |
4069 | /* | |
4070 | * Take priority boosted tasks into account. If the new | |
4071 | * effective priority is unchanged, we just store the new | |
4072 | * normal parameters and do not touch the scheduler class and | |
4073 | * the runqueue. This will be done when the task deboost | |
4074 | * itself. | |
4075 | */ | |
4076 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4077 | if (new_effective_prio == oldprio) |
4078 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4079 | } |
4080 | ||
da0c1e65 | 4081 | queued = task_on_rq_queued(p); |
051a1d1a | 4082 | running = task_current(rq, p); |
da0c1e65 | 4083 | if (queued) |
ff77e468 | 4084 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4085 | if (running) |
f3cd1c4e | 4086 | put_prev_task(rq, p); |
f6b53205 | 4087 | |
83ab0aa0 | 4088 | prev_class = p->sched_class; |
dbc7f069 | 4089 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4090 | |
0e1f3483 HS |
4091 | if (running) |
4092 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4093 | if (queued) { |
81a44c54 TG |
4094 | /* |
4095 | * We enqueue to tail when the priority of a task is | |
4096 | * increased (user space view). | |
4097 | */ | |
ff77e468 PZ |
4098 | if (oldprio < p->prio) |
4099 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4100 | |
ff77e468 | 4101 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4102 | } |
cb469845 | 4103 | |
da7a735e | 4104 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 4105 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 4106 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4107 | |
dbc7f069 PZ |
4108 | if (pi) |
4109 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4110 | |
4c9a4bc8 PZ |
4111 | /* |
4112 | * Run balance callbacks after we've adjusted the PI chain. | |
4113 | */ | |
4114 | balance_callback(rq); | |
4115 | preempt_enable(); | |
95e02ca9 | 4116 | |
1da177e4 LT |
4117 | return 0; |
4118 | } | |
961ccddd | 4119 | |
7479f3c9 PZ |
4120 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4121 | const struct sched_param *param, bool check) | |
4122 | { | |
4123 | struct sched_attr attr = { | |
4124 | .sched_policy = policy, | |
4125 | .sched_priority = param->sched_priority, | |
4126 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4127 | }; | |
4128 | ||
c13db6b1 SR |
4129 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4130 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4131 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4132 | policy &= ~SCHED_RESET_ON_FORK; | |
4133 | attr.sched_policy = policy; | |
4134 | } | |
4135 | ||
dbc7f069 | 4136 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4137 | } |
961ccddd RR |
4138 | /** |
4139 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4140 | * @p: the task in question. | |
4141 | * @policy: new policy. | |
4142 | * @param: structure containing the new RT priority. | |
4143 | * | |
e69f6186 YB |
4144 | * Return: 0 on success. An error code otherwise. |
4145 | * | |
961ccddd RR |
4146 | * NOTE that the task may be already dead. |
4147 | */ | |
4148 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4149 | const struct sched_param *param) |
961ccddd | 4150 | { |
7479f3c9 | 4151 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4152 | } |
1da177e4 LT |
4153 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4154 | ||
d50dde5a DF |
4155 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4156 | { | |
dbc7f069 | 4157 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4158 | } |
4159 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4160 | ||
961ccddd RR |
4161 | /** |
4162 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4163 | * @p: the task in question. | |
4164 | * @policy: new policy. | |
4165 | * @param: structure containing the new RT priority. | |
4166 | * | |
4167 | * Just like sched_setscheduler, only don't bother checking if the | |
4168 | * current context has permission. For example, this is needed in | |
4169 | * stop_machine(): we create temporary high priority worker threads, | |
4170 | * but our caller might not have that capability. | |
e69f6186 YB |
4171 | * |
4172 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4173 | */ |
4174 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4175 | const struct sched_param *param) |
961ccddd | 4176 | { |
7479f3c9 | 4177 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4178 | } |
84778472 | 4179 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4180 | |
95cdf3b7 IM |
4181 | static int |
4182 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4183 | { |
1da177e4 LT |
4184 | struct sched_param lparam; |
4185 | struct task_struct *p; | |
36c8b586 | 4186 | int retval; |
1da177e4 LT |
4187 | |
4188 | if (!param || pid < 0) | |
4189 | return -EINVAL; | |
4190 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4191 | return -EFAULT; | |
5fe1d75f ON |
4192 | |
4193 | rcu_read_lock(); | |
4194 | retval = -ESRCH; | |
1da177e4 | 4195 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4196 | if (p != NULL) |
4197 | retval = sched_setscheduler(p, policy, &lparam); | |
4198 | rcu_read_unlock(); | |
36c8b586 | 4199 | |
1da177e4 LT |
4200 | return retval; |
4201 | } | |
4202 | ||
d50dde5a DF |
4203 | /* |
4204 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4205 | */ | |
4206 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4207 | struct sched_attr *attr) | |
4208 | { | |
4209 | u32 size; | |
4210 | int ret; | |
4211 | ||
4212 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4213 | return -EFAULT; | |
4214 | ||
4215 | /* | |
4216 | * zero the full structure, so that a short copy will be nice. | |
4217 | */ | |
4218 | memset(attr, 0, sizeof(*attr)); | |
4219 | ||
4220 | ret = get_user(size, &uattr->size); | |
4221 | if (ret) | |
4222 | return ret; | |
4223 | ||
4224 | if (size > PAGE_SIZE) /* silly large */ | |
4225 | goto err_size; | |
4226 | ||
4227 | if (!size) /* abi compat */ | |
4228 | size = SCHED_ATTR_SIZE_VER0; | |
4229 | ||
4230 | if (size < SCHED_ATTR_SIZE_VER0) | |
4231 | goto err_size; | |
4232 | ||
4233 | /* | |
4234 | * If we're handed a bigger struct than we know of, | |
4235 | * ensure all the unknown bits are 0 - i.e. new | |
4236 | * user-space does not rely on any kernel feature | |
4237 | * extensions we dont know about yet. | |
4238 | */ | |
4239 | if (size > sizeof(*attr)) { | |
4240 | unsigned char __user *addr; | |
4241 | unsigned char __user *end; | |
4242 | unsigned char val; | |
4243 | ||
4244 | addr = (void __user *)uattr + sizeof(*attr); | |
4245 | end = (void __user *)uattr + size; | |
4246 | ||
4247 | for (; addr < end; addr++) { | |
4248 | ret = get_user(val, addr); | |
4249 | if (ret) | |
4250 | return ret; | |
4251 | if (val) | |
4252 | goto err_size; | |
4253 | } | |
4254 | size = sizeof(*attr); | |
4255 | } | |
4256 | ||
4257 | ret = copy_from_user(attr, uattr, size); | |
4258 | if (ret) | |
4259 | return -EFAULT; | |
4260 | ||
4261 | /* | |
4262 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4263 | * to be strict and return an error on out-of-bounds values? | |
4264 | */ | |
75e45d51 | 4265 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4266 | |
e78c7bca | 4267 | return 0; |
d50dde5a DF |
4268 | |
4269 | err_size: | |
4270 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4271 | return -E2BIG; |
d50dde5a DF |
4272 | } |
4273 | ||
1da177e4 LT |
4274 | /** |
4275 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4276 | * @pid: the pid in question. | |
4277 | * @policy: new policy. | |
4278 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4279 | * |
4280 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4281 | */ |
5add95d4 HC |
4282 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4283 | struct sched_param __user *, param) | |
1da177e4 | 4284 | { |
c21761f1 JB |
4285 | /* negative values for policy are not valid */ |
4286 | if (policy < 0) | |
4287 | return -EINVAL; | |
4288 | ||
1da177e4 LT |
4289 | return do_sched_setscheduler(pid, policy, param); |
4290 | } | |
4291 | ||
4292 | /** | |
4293 | * sys_sched_setparam - set/change the RT priority of a thread | |
4294 | * @pid: the pid in question. | |
4295 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4296 | * |
4297 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4298 | */ |
5add95d4 | 4299 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4300 | { |
c13db6b1 | 4301 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4302 | } |
4303 | ||
d50dde5a DF |
4304 | /** |
4305 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4306 | * @pid: the pid in question. | |
5778fccf | 4307 | * @uattr: structure containing the extended parameters. |
db66d756 | 4308 | * @flags: for future extension. |
d50dde5a | 4309 | */ |
6d35ab48 PZ |
4310 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4311 | unsigned int, flags) | |
d50dde5a DF |
4312 | { |
4313 | struct sched_attr attr; | |
4314 | struct task_struct *p; | |
4315 | int retval; | |
4316 | ||
6d35ab48 | 4317 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4318 | return -EINVAL; |
4319 | ||
143cf23d MK |
4320 | retval = sched_copy_attr(uattr, &attr); |
4321 | if (retval) | |
4322 | return retval; | |
d50dde5a | 4323 | |
b14ed2c2 | 4324 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4325 | return -EINVAL; |
d50dde5a DF |
4326 | |
4327 | rcu_read_lock(); | |
4328 | retval = -ESRCH; | |
4329 | p = find_process_by_pid(pid); | |
4330 | if (p != NULL) | |
4331 | retval = sched_setattr(p, &attr); | |
4332 | rcu_read_unlock(); | |
4333 | ||
4334 | return retval; | |
4335 | } | |
4336 | ||
1da177e4 LT |
4337 | /** |
4338 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4339 | * @pid: the pid in question. | |
e69f6186 YB |
4340 | * |
4341 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4342 | * code. | |
1da177e4 | 4343 | */ |
5add95d4 | 4344 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4345 | { |
36c8b586 | 4346 | struct task_struct *p; |
3a5c359a | 4347 | int retval; |
1da177e4 LT |
4348 | |
4349 | if (pid < 0) | |
3a5c359a | 4350 | return -EINVAL; |
1da177e4 LT |
4351 | |
4352 | retval = -ESRCH; | |
5fe85be0 | 4353 | rcu_read_lock(); |
1da177e4 LT |
4354 | p = find_process_by_pid(pid); |
4355 | if (p) { | |
4356 | retval = security_task_getscheduler(p); | |
4357 | if (!retval) | |
ca94c442 LP |
4358 | retval = p->policy |
4359 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4360 | } |
5fe85be0 | 4361 | rcu_read_unlock(); |
1da177e4 LT |
4362 | return retval; |
4363 | } | |
4364 | ||
4365 | /** | |
ca94c442 | 4366 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4367 | * @pid: the pid in question. |
4368 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4369 | * |
4370 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4371 | * code. | |
1da177e4 | 4372 | */ |
5add95d4 | 4373 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4374 | { |
ce5f7f82 | 4375 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4376 | struct task_struct *p; |
3a5c359a | 4377 | int retval; |
1da177e4 LT |
4378 | |
4379 | if (!param || pid < 0) | |
3a5c359a | 4380 | return -EINVAL; |
1da177e4 | 4381 | |
5fe85be0 | 4382 | rcu_read_lock(); |
1da177e4 LT |
4383 | p = find_process_by_pid(pid); |
4384 | retval = -ESRCH; | |
4385 | if (!p) | |
4386 | goto out_unlock; | |
4387 | ||
4388 | retval = security_task_getscheduler(p); | |
4389 | if (retval) | |
4390 | goto out_unlock; | |
4391 | ||
ce5f7f82 PZ |
4392 | if (task_has_rt_policy(p)) |
4393 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4394 | rcu_read_unlock(); |
1da177e4 LT |
4395 | |
4396 | /* | |
4397 | * This one might sleep, we cannot do it with a spinlock held ... | |
4398 | */ | |
4399 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4400 | ||
1da177e4 LT |
4401 | return retval; |
4402 | ||
4403 | out_unlock: | |
5fe85be0 | 4404 | rcu_read_unlock(); |
1da177e4 LT |
4405 | return retval; |
4406 | } | |
4407 | ||
d50dde5a DF |
4408 | static int sched_read_attr(struct sched_attr __user *uattr, |
4409 | struct sched_attr *attr, | |
4410 | unsigned int usize) | |
4411 | { | |
4412 | int ret; | |
4413 | ||
4414 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4415 | return -EFAULT; | |
4416 | ||
4417 | /* | |
4418 | * If we're handed a smaller struct than we know of, | |
4419 | * ensure all the unknown bits are 0 - i.e. old | |
4420 | * user-space does not get uncomplete information. | |
4421 | */ | |
4422 | if (usize < sizeof(*attr)) { | |
4423 | unsigned char *addr; | |
4424 | unsigned char *end; | |
4425 | ||
4426 | addr = (void *)attr + usize; | |
4427 | end = (void *)attr + sizeof(*attr); | |
4428 | ||
4429 | for (; addr < end; addr++) { | |
4430 | if (*addr) | |
22400674 | 4431 | return -EFBIG; |
d50dde5a DF |
4432 | } |
4433 | ||
4434 | attr->size = usize; | |
4435 | } | |
4436 | ||
4efbc454 | 4437 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4438 | if (ret) |
4439 | return -EFAULT; | |
4440 | ||
22400674 | 4441 | return 0; |
d50dde5a DF |
4442 | } |
4443 | ||
4444 | /** | |
aab03e05 | 4445 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4446 | * @pid: the pid in question. |
5778fccf | 4447 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4448 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4449 | * @flags: for future extension. |
d50dde5a | 4450 | */ |
6d35ab48 PZ |
4451 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4452 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4453 | { |
4454 | struct sched_attr attr = { | |
4455 | .size = sizeof(struct sched_attr), | |
4456 | }; | |
4457 | struct task_struct *p; | |
4458 | int retval; | |
4459 | ||
4460 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4461 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4462 | return -EINVAL; |
4463 | ||
4464 | rcu_read_lock(); | |
4465 | p = find_process_by_pid(pid); | |
4466 | retval = -ESRCH; | |
4467 | if (!p) | |
4468 | goto out_unlock; | |
4469 | ||
4470 | retval = security_task_getscheduler(p); | |
4471 | if (retval) | |
4472 | goto out_unlock; | |
4473 | ||
4474 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4475 | if (p->sched_reset_on_fork) |
4476 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4477 | if (task_has_dl_policy(p)) |
4478 | __getparam_dl(p, &attr); | |
4479 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4480 | attr.sched_priority = p->rt_priority; |
4481 | else | |
d0ea0268 | 4482 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4483 | |
4484 | rcu_read_unlock(); | |
4485 | ||
4486 | retval = sched_read_attr(uattr, &attr, size); | |
4487 | return retval; | |
4488 | ||
4489 | out_unlock: | |
4490 | rcu_read_unlock(); | |
4491 | return retval; | |
4492 | } | |
4493 | ||
96f874e2 | 4494 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4495 | { |
5a16f3d3 | 4496 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4497 | struct task_struct *p; |
4498 | int retval; | |
1da177e4 | 4499 | |
23f5d142 | 4500 | rcu_read_lock(); |
1da177e4 LT |
4501 | |
4502 | p = find_process_by_pid(pid); | |
4503 | if (!p) { | |
23f5d142 | 4504 | rcu_read_unlock(); |
1da177e4 LT |
4505 | return -ESRCH; |
4506 | } | |
4507 | ||
23f5d142 | 4508 | /* Prevent p going away */ |
1da177e4 | 4509 | get_task_struct(p); |
23f5d142 | 4510 | rcu_read_unlock(); |
1da177e4 | 4511 | |
14a40ffc TH |
4512 | if (p->flags & PF_NO_SETAFFINITY) { |
4513 | retval = -EINVAL; | |
4514 | goto out_put_task; | |
4515 | } | |
5a16f3d3 RR |
4516 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4517 | retval = -ENOMEM; | |
4518 | goto out_put_task; | |
4519 | } | |
4520 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4521 | retval = -ENOMEM; | |
4522 | goto out_free_cpus_allowed; | |
4523 | } | |
1da177e4 | 4524 | retval = -EPERM; |
4c44aaaf EB |
4525 | if (!check_same_owner(p)) { |
4526 | rcu_read_lock(); | |
4527 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4528 | rcu_read_unlock(); | |
16303ab2 | 4529 | goto out_free_new_mask; |
4c44aaaf EB |
4530 | } |
4531 | rcu_read_unlock(); | |
4532 | } | |
1da177e4 | 4533 | |
b0ae1981 | 4534 | retval = security_task_setscheduler(p); |
e7834f8f | 4535 | if (retval) |
16303ab2 | 4536 | goto out_free_new_mask; |
e7834f8f | 4537 | |
e4099a5e PZ |
4538 | |
4539 | cpuset_cpus_allowed(p, cpus_allowed); | |
4540 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4541 | ||
332ac17e DF |
4542 | /* |
4543 | * Since bandwidth control happens on root_domain basis, | |
4544 | * if admission test is enabled, we only admit -deadline | |
4545 | * tasks allowed to run on all the CPUs in the task's | |
4546 | * root_domain. | |
4547 | */ | |
4548 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4549 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4550 | rcu_read_lock(); | |
4551 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4552 | retval = -EBUSY; |
f1e3a093 | 4553 | rcu_read_unlock(); |
16303ab2 | 4554 | goto out_free_new_mask; |
332ac17e | 4555 | } |
f1e3a093 | 4556 | rcu_read_unlock(); |
332ac17e DF |
4557 | } |
4558 | #endif | |
49246274 | 4559 | again: |
25834c73 | 4560 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4561 | |
8707d8b8 | 4562 | if (!retval) { |
5a16f3d3 RR |
4563 | cpuset_cpus_allowed(p, cpus_allowed); |
4564 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4565 | /* |
4566 | * We must have raced with a concurrent cpuset | |
4567 | * update. Just reset the cpus_allowed to the | |
4568 | * cpuset's cpus_allowed | |
4569 | */ | |
5a16f3d3 | 4570 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4571 | goto again; |
4572 | } | |
4573 | } | |
16303ab2 | 4574 | out_free_new_mask: |
5a16f3d3 RR |
4575 | free_cpumask_var(new_mask); |
4576 | out_free_cpus_allowed: | |
4577 | free_cpumask_var(cpus_allowed); | |
4578 | out_put_task: | |
1da177e4 | 4579 | put_task_struct(p); |
1da177e4 LT |
4580 | return retval; |
4581 | } | |
4582 | ||
4583 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4584 | struct cpumask *new_mask) |
1da177e4 | 4585 | { |
96f874e2 RR |
4586 | if (len < cpumask_size()) |
4587 | cpumask_clear(new_mask); | |
4588 | else if (len > cpumask_size()) | |
4589 | len = cpumask_size(); | |
4590 | ||
1da177e4 LT |
4591 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4592 | } | |
4593 | ||
4594 | /** | |
4595 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4596 | * @pid: pid of the process | |
4597 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4598 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4599 | * |
4600 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4601 | */ |
5add95d4 HC |
4602 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4603 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4604 | { |
5a16f3d3 | 4605 | cpumask_var_t new_mask; |
1da177e4 LT |
4606 | int retval; |
4607 | ||
5a16f3d3 RR |
4608 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4609 | return -ENOMEM; | |
1da177e4 | 4610 | |
5a16f3d3 RR |
4611 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4612 | if (retval == 0) | |
4613 | retval = sched_setaffinity(pid, new_mask); | |
4614 | free_cpumask_var(new_mask); | |
4615 | return retval; | |
1da177e4 LT |
4616 | } |
4617 | ||
96f874e2 | 4618 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4619 | { |
36c8b586 | 4620 | struct task_struct *p; |
31605683 | 4621 | unsigned long flags; |
1da177e4 | 4622 | int retval; |
1da177e4 | 4623 | |
23f5d142 | 4624 | rcu_read_lock(); |
1da177e4 LT |
4625 | |
4626 | retval = -ESRCH; | |
4627 | p = find_process_by_pid(pid); | |
4628 | if (!p) | |
4629 | goto out_unlock; | |
4630 | ||
e7834f8f DQ |
4631 | retval = security_task_getscheduler(p); |
4632 | if (retval) | |
4633 | goto out_unlock; | |
4634 | ||
013fdb80 | 4635 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4636 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4637 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4638 | |
4639 | out_unlock: | |
23f5d142 | 4640 | rcu_read_unlock(); |
1da177e4 | 4641 | |
9531b62f | 4642 | return retval; |
1da177e4 LT |
4643 | } |
4644 | ||
4645 | /** | |
4646 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4647 | * @pid: pid of the process | |
4648 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4649 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4650 | * |
4651 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4652 | */ |
5add95d4 HC |
4653 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4654 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4655 | { |
4656 | int ret; | |
f17c8607 | 4657 | cpumask_var_t mask; |
1da177e4 | 4658 | |
84fba5ec | 4659 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4660 | return -EINVAL; |
4661 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4662 | return -EINVAL; |
4663 | ||
f17c8607 RR |
4664 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4665 | return -ENOMEM; | |
1da177e4 | 4666 | |
f17c8607 RR |
4667 | ret = sched_getaffinity(pid, mask); |
4668 | if (ret == 0) { | |
8bc037fb | 4669 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4670 | |
4671 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4672 | ret = -EFAULT; |
4673 | else | |
cd3d8031 | 4674 | ret = retlen; |
f17c8607 RR |
4675 | } |
4676 | free_cpumask_var(mask); | |
1da177e4 | 4677 | |
f17c8607 | 4678 | return ret; |
1da177e4 LT |
4679 | } |
4680 | ||
4681 | /** | |
4682 | * sys_sched_yield - yield the current processor to other threads. | |
4683 | * | |
dd41f596 IM |
4684 | * This function yields the current CPU to other tasks. If there are no |
4685 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4686 | * |
4687 | * Return: 0. | |
1da177e4 | 4688 | */ |
5add95d4 | 4689 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4690 | { |
70b97a7f | 4691 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4692 | |
2d72376b | 4693 | schedstat_inc(rq, yld_count); |
4530d7ab | 4694 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4695 | |
4696 | /* | |
4697 | * Since we are going to call schedule() anyway, there's | |
4698 | * no need to preempt or enable interrupts: | |
4699 | */ | |
4700 | __release(rq->lock); | |
8a25d5de | 4701 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4702 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4703 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4704 | |
4705 | schedule(); | |
4706 | ||
4707 | return 0; | |
4708 | } | |
4709 | ||
02b67cc3 | 4710 | int __sched _cond_resched(void) |
1da177e4 | 4711 | { |
fe32d3cd | 4712 | if (should_resched(0)) { |
a18b5d01 | 4713 | preempt_schedule_common(); |
1da177e4 LT |
4714 | return 1; |
4715 | } | |
4716 | return 0; | |
4717 | } | |
02b67cc3 | 4718 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4719 | |
4720 | /* | |
613afbf8 | 4721 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4722 | * call schedule, and on return reacquire the lock. |
4723 | * | |
41a2d6cf | 4724 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4725 | * operations here to prevent schedule() from being called twice (once via |
4726 | * spin_unlock(), once by hand). | |
4727 | */ | |
613afbf8 | 4728 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4729 | { |
fe32d3cd | 4730 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4731 | int ret = 0; |
4732 | ||
f607c668 PZ |
4733 | lockdep_assert_held(lock); |
4734 | ||
4a81e832 | 4735 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4736 | spin_unlock(lock); |
d86ee480 | 4737 | if (resched) |
a18b5d01 | 4738 | preempt_schedule_common(); |
95c354fe NP |
4739 | else |
4740 | cpu_relax(); | |
6df3cecb | 4741 | ret = 1; |
1da177e4 | 4742 | spin_lock(lock); |
1da177e4 | 4743 | } |
6df3cecb | 4744 | return ret; |
1da177e4 | 4745 | } |
613afbf8 | 4746 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4747 | |
613afbf8 | 4748 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4749 | { |
4750 | BUG_ON(!in_softirq()); | |
4751 | ||
fe32d3cd | 4752 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4753 | local_bh_enable(); |
a18b5d01 | 4754 | preempt_schedule_common(); |
1da177e4 LT |
4755 | local_bh_disable(); |
4756 | return 1; | |
4757 | } | |
4758 | return 0; | |
4759 | } | |
613afbf8 | 4760 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4761 | |
1da177e4 LT |
4762 | /** |
4763 | * yield - yield the current processor to other threads. | |
4764 | * | |
8e3fabfd PZ |
4765 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4766 | * | |
4767 | * The scheduler is at all times free to pick the calling task as the most | |
4768 | * eligible task to run, if removing the yield() call from your code breaks | |
4769 | * it, its already broken. | |
4770 | * | |
4771 | * Typical broken usage is: | |
4772 | * | |
4773 | * while (!event) | |
4774 | * yield(); | |
4775 | * | |
4776 | * where one assumes that yield() will let 'the other' process run that will | |
4777 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4778 | * happen. Never use yield() as a progress guarantee!! | |
4779 | * | |
4780 | * If you want to use yield() to wait for something, use wait_event(). | |
4781 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4782 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4783 | */ |
4784 | void __sched yield(void) | |
4785 | { | |
4786 | set_current_state(TASK_RUNNING); | |
4787 | sys_sched_yield(); | |
4788 | } | |
1da177e4 LT |
4789 | EXPORT_SYMBOL(yield); |
4790 | ||
d95f4122 MG |
4791 | /** |
4792 | * yield_to - yield the current processor to another thread in | |
4793 | * your thread group, or accelerate that thread toward the | |
4794 | * processor it's on. | |
16addf95 RD |
4795 | * @p: target task |
4796 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4797 | * |
4798 | * It's the caller's job to ensure that the target task struct | |
4799 | * can't go away on us before we can do any checks. | |
4800 | * | |
e69f6186 | 4801 | * Return: |
7b270f60 PZ |
4802 | * true (>0) if we indeed boosted the target task. |
4803 | * false (0) if we failed to boost the target. | |
4804 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4805 | */ |
fa93384f | 4806 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4807 | { |
4808 | struct task_struct *curr = current; | |
4809 | struct rq *rq, *p_rq; | |
4810 | unsigned long flags; | |
c3c18640 | 4811 | int yielded = 0; |
d95f4122 MG |
4812 | |
4813 | local_irq_save(flags); | |
4814 | rq = this_rq(); | |
4815 | ||
4816 | again: | |
4817 | p_rq = task_rq(p); | |
7b270f60 PZ |
4818 | /* |
4819 | * If we're the only runnable task on the rq and target rq also | |
4820 | * has only one task, there's absolutely no point in yielding. | |
4821 | */ | |
4822 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4823 | yielded = -ESRCH; | |
4824 | goto out_irq; | |
4825 | } | |
4826 | ||
d95f4122 | 4827 | double_rq_lock(rq, p_rq); |
39e24d8f | 4828 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4829 | double_rq_unlock(rq, p_rq); |
4830 | goto again; | |
4831 | } | |
4832 | ||
4833 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4834 | goto out_unlock; |
d95f4122 MG |
4835 | |
4836 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4837 | goto out_unlock; |
d95f4122 MG |
4838 | |
4839 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4840 | goto out_unlock; |
d95f4122 MG |
4841 | |
4842 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4843 | if (yielded) { |
d95f4122 | 4844 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4845 | /* |
4846 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4847 | * fairness. | |
4848 | */ | |
4849 | if (preempt && rq != p_rq) | |
8875125e | 4850 | resched_curr(p_rq); |
6d1cafd8 | 4851 | } |
d95f4122 | 4852 | |
7b270f60 | 4853 | out_unlock: |
d95f4122 | 4854 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4855 | out_irq: |
d95f4122 MG |
4856 | local_irq_restore(flags); |
4857 | ||
7b270f60 | 4858 | if (yielded > 0) |
d95f4122 MG |
4859 | schedule(); |
4860 | ||
4861 | return yielded; | |
4862 | } | |
4863 | EXPORT_SYMBOL_GPL(yield_to); | |
4864 | ||
1da177e4 | 4865 | /* |
41a2d6cf | 4866 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4867 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4868 | */ |
1da177e4 LT |
4869 | long __sched io_schedule_timeout(long timeout) |
4870 | { | |
9cff8ade N |
4871 | int old_iowait = current->in_iowait; |
4872 | struct rq *rq; | |
1da177e4 LT |
4873 | long ret; |
4874 | ||
9cff8ade | 4875 | current->in_iowait = 1; |
10d784ea | 4876 | blk_schedule_flush_plug(current); |
9cff8ade | 4877 | |
0ff92245 | 4878 | delayacct_blkio_start(); |
9cff8ade | 4879 | rq = raw_rq(); |
1da177e4 LT |
4880 | atomic_inc(&rq->nr_iowait); |
4881 | ret = schedule_timeout(timeout); | |
9cff8ade | 4882 | current->in_iowait = old_iowait; |
1da177e4 | 4883 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4884 | delayacct_blkio_end(); |
9cff8ade | 4885 | |
1da177e4 LT |
4886 | return ret; |
4887 | } | |
9cff8ade | 4888 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4889 | |
4890 | /** | |
4891 | * sys_sched_get_priority_max - return maximum RT priority. | |
4892 | * @policy: scheduling class. | |
4893 | * | |
e69f6186 YB |
4894 | * Return: On success, this syscall returns the maximum |
4895 | * rt_priority that can be used by a given scheduling class. | |
4896 | * On failure, a negative error code is returned. | |
1da177e4 | 4897 | */ |
5add95d4 | 4898 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4899 | { |
4900 | int ret = -EINVAL; | |
4901 | ||
4902 | switch (policy) { | |
4903 | case SCHED_FIFO: | |
4904 | case SCHED_RR: | |
4905 | ret = MAX_USER_RT_PRIO-1; | |
4906 | break; | |
aab03e05 | 4907 | case SCHED_DEADLINE: |
1da177e4 | 4908 | case SCHED_NORMAL: |
b0a9499c | 4909 | case SCHED_BATCH: |
dd41f596 | 4910 | case SCHED_IDLE: |
1da177e4 LT |
4911 | ret = 0; |
4912 | break; | |
4913 | } | |
4914 | return ret; | |
4915 | } | |
4916 | ||
4917 | /** | |
4918 | * sys_sched_get_priority_min - return minimum RT priority. | |
4919 | * @policy: scheduling class. | |
4920 | * | |
e69f6186 YB |
4921 | * Return: On success, this syscall returns the minimum |
4922 | * rt_priority that can be used by a given scheduling class. | |
4923 | * On failure, a negative error code is returned. | |
1da177e4 | 4924 | */ |
5add95d4 | 4925 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4926 | { |
4927 | int ret = -EINVAL; | |
4928 | ||
4929 | switch (policy) { | |
4930 | case SCHED_FIFO: | |
4931 | case SCHED_RR: | |
4932 | ret = 1; | |
4933 | break; | |
aab03e05 | 4934 | case SCHED_DEADLINE: |
1da177e4 | 4935 | case SCHED_NORMAL: |
b0a9499c | 4936 | case SCHED_BATCH: |
dd41f596 | 4937 | case SCHED_IDLE: |
1da177e4 LT |
4938 | ret = 0; |
4939 | } | |
4940 | return ret; | |
4941 | } | |
4942 | ||
4943 | /** | |
4944 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4945 | * @pid: pid of the process. | |
4946 | * @interval: userspace pointer to the timeslice value. | |
4947 | * | |
4948 | * this syscall writes the default timeslice value of a given process | |
4949 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4950 | * |
4951 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4952 | * an error code. | |
1da177e4 | 4953 | */ |
17da2bd9 | 4954 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4955 | struct timespec __user *, interval) |
1da177e4 | 4956 | { |
36c8b586 | 4957 | struct task_struct *p; |
a4ec24b4 | 4958 | unsigned int time_slice; |
dba091b9 TG |
4959 | unsigned long flags; |
4960 | struct rq *rq; | |
3a5c359a | 4961 | int retval; |
1da177e4 | 4962 | struct timespec t; |
1da177e4 LT |
4963 | |
4964 | if (pid < 0) | |
3a5c359a | 4965 | return -EINVAL; |
1da177e4 LT |
4966 | |
4967 | retval = -ESRCH; | |
1a551ae7 | 4968 | rcu_read_lock(); |
1da177e4 LT |
4969 | p = find_process_by_pid(pid); |
4970 | if (!p) | |
4971 | goto out_unlock; | |
4972 | ||
4973 | retval = security_task_getscheduler(p); | |
4974 | if (retval) | |
4975 | goto out_unlock; | |
4976 | ||
dba091b9 | 4977 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4978 | time_slice = 0; |
4979 | if (p->sched_class->get_rr_interval) | |
4980 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4981 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4982 | |
1a551ae7 | 4983 | rcu_read_unlock(); |
a4ec24b4 | 4984 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4985 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4986 | return retval; |
3a5c359a | 4987 | |
1da177e4 | 4988 | out_unlock: |
1a551ae7 | 4989 | rcu_read_unlock(); |
1da177e4 LT |
4990 | return retval; |
4991 | } | |
4992 | ||
7c731e0a | 4993 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4994 | |
82a1fcb9 | 4995 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4996 | { |
1da177e4 | 4997 | unsigned long free = 0; |
4e79752c | 4998 | int ppid; |
1f8a7633 | 4999 | unsigned long state = p->state; |
1da177e4 | 5000 | |
1f8a7633 TH |
5001 | if (state) |
5002 | state = __ffs(state) + 1; | |
28d0686c | 5003 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5004 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5005 | #if BITS_PER_LONG == 32 |
1da177e4 | 5006 | if (state == TASK_RUNNING) |
3df0fc5b | 5007 | printk(KERN_CONT " running "); |
1da177e4 | 5008 | else |
3df0fc5b | 5009 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5010 | #else |
5011 | if (state == TASK_RUNNING) | |
3df0fc5b | 5012 | printk(KERN_CONT " running task "); |
1da177e4 | 5013 | else |
3df0fc5b | 5014 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5015 | #endif |
5016 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5017 | free = stack_not_used(p); |
1da177e4 | 5018 | #endif |
a90e984c | 5019 | ppid = 0; |
4e79752c | 5020 | rcu_read_lock(); |
a90e984c ON |
5021 | if (pid_alive(p)) |
5022 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5023 | rcu_read_unlock(); |
3df0fc5b | 5024 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5025 | task_pid_nr(p), ppid, |
aa47b7e0 | 5026 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5027 | |
3d1cb205 | 5028 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5029 | show_stack(p, NULL); |
1da177e4 LT |
5030 | } |
5031 | ||
e59e2ae2 | 5032 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5033 | { |
36c8b586 | 5034 | struct task_struct *g, *p; |
1da177e4 | 5035 | |
4bd77321 | 5036 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5037 | printk(KERN_INFO |
5038 | " task PC stack pid father\n"); | |
1da177e4 | 5039 | #else |
3df0fc5b PZ |
5040 | printk(KERN_INFO |
5041 | " task PC stack pid father\n"); | |
1da177e4 | 5042 | #endif |
510f5acc | 5043 | rcu_read_lock(); |
5d07f420 | 5044 | for_each_process_thread(g, p) { |
1da177e4 LT |
5045 | /* |
5046 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5047 | * console might take a lot of time: |
1da177e4 LT |
5048 | */ |
5049 | touch_nmi_watchdog(); | |
39bc89fd | 5050 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5051 | sched_show_task(p); |
5d07f420 | 5052 | } |
1da177e4 | 5053 | |
04c9167f JF |
5054 | touch_all_softlockup_watchdogs(); |
5055 | ||
dd41f596 | 5056 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5057 | if (!state_filter) |
5058 | sysrq_sched_debug_show(); | |
dd41f596 | 5059 | #endif |
510f5acc | 5060 | rcu_read_unlock(); |
e59e2ae2 IM |
5061 | /* |
5062 | * Only show locks if all tasks are dumped: | |
5063 | */ | |
93335a21 | 5064 | if (!state_filter) |
e59e2ae2 | 5065 | debug_show_all_locks(); |
1da177e4 LT |
5066 | } |
5067 | ||
0db0628d | 5068 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5069 | { |
dd41f596 | 5070 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5071 | } |
5072 | ||
f340c0d1 IM |
5073 | /** |
5074 | * init_idle - set up an idle thread for a given CPU | |
5075 | * @idle: task in question | |
5076 | * @cpu: cpu the idle task belongs to | |
5077 | * | |
5078 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5079 | * flag, to make booting more robust. | |
5080 | */ | |
0db0628d | 5081 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5082 | { |
70b97a7f | 5083 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5084 | unsigned long flags; |
5085 | ||
25834c73 PZ |
5086 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5087 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5088 | |
5e1576ed | 5089 | __sched_fork(0, idle); |
06b83b5f | 5090 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5091 | idle->se.exec_start = sched_clock(); |
5092 | ||
e1b77c92 MR |
5093 | kasan_unpoison_task_stack(idle); |
5094 | ||
de9b8f5d PZ |
5095 | #ifdef CONFIG_SMP |
5096 | /* | |
5097 | * Its possible that init_idle() gets called multiple times on a task, | |
5098 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5099 | * | |
5100 | * And since this is boot we can forgo the serialization. | |
5101 | */ | |
5102 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5103 | #endif | |
6506cf6c PZ |
5104 | /* |
5105 | * We're having a chicken and egg problem, even though we are | |
5106 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5107 | * lockdep check in task_group() will fail. | |
5108 | * | |
5109 | * Similar case to sched_fork(). / Alternatively we could | |
5110 | * use task_rq_lock() here and obtain the other rq->lock. | |
5111 | * | |
5112 | * Silence PROVE_RCU | |
5113 | */ | |
5114 | rcu_read_lock(); | |
dd41f596 | 5115 | __set_task_cpu(idle, cpu); |
6506cf6c | 5116 | rcu_read_unlock(); |
1da177e4 | 5117 | |
1da177e4 | 5118 | rq->curr = rq->idle = idle; |
da0c1e65 | 5119 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5120 | #ifdef CONFIG_SMP |
3ca7a440 | 5121 | idle->on_cpu = 1; |
4866cde0 | 5122 | #endif |
25834c73 PZ |
5123 | raw_spin_unlock(&rq->lock); |
5124 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5125 | |
5126 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5127 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5128 | |
dd41f596 IM |
5129 | /* |
5130 | * The idle tasks have their own, simple scheduling class: | |
5131 | */ | |
5132 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5133 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5134 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5135 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5136 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5137 | #endif | |
19978ca6 IM |
5138 | } |
5139 | ||
f82f8042 JL |
5140 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5141 | const struct cpumask *trial) | |
5142 | { | |
5143 | int ret = 1, trial_cpus; | |
5144 | struct dl_bw *cur_dl_b; | |
5145 | unsigned long flags; | |
5146 | ||
bb2bc55a MG |
5147 | if (!cpumask_weight(cur)) |
5148 | return ret; | |
5149 | ||
75e23e49 | 5150 | rcu_read_lock_sched(); |
f82f8042 JL |
5151 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5152 | trial_cpus = cpumask_weight(trial); | |
5153 | ||
5154 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5155 | if (cur_dl_b->bw != -1 && | |
5156 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5157 | ret = 0; | |
5158 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5159 | rcu_read_unlock_sched(); |
f82f8042 JL |
5160 | |
5161 | return ret; | |
5162 | } | |
5163 | ||
7f51412a JL |
5164 | int task_can_attach(struct task_struct *p, |
5165 | const struct cpumask *cs_cpus_allowed) | |
5166 | { | |
5167 | int ret = 0; | |
5168 | ||
5169 | /* | |
5170 | * Kthreads which disallow setaffinity shouldn't be moved | |
5171 | * to a new cpuset; we don't want to change their cpu | |
5172 | * affinity and isolating such threads by their set of | |
5173 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5174 | * applicable for such threads. This prevents checking for | |
5175 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5176 | * before cpus_allowed may be changed. | |
5177 | */ | |
5178 | if (p->flags & PF_NO_SETAFFINITY) { | |
5179 | ret = -EINVAL; | |
5180 | goto out; | |
5181 | } | |
5182 | ||
5183 | #ifdef CONFIG_SMP | |
5184 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5185 | cs_cpus_allowed)) { | |
5186 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5187 | cs_cpus_allowed); | |
75e23e49 | 5188 | struct dl_bw *dl_b; |
7f51412a JL |
5189 | bool overflow; |
5190 | int cpus; | |
5191 | unsigned long flags; | |
5192 | ||
75e23e49 JL |
5193 | rcu_read_lock_sched(); |
5194 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5195 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5196 | cpus = dl_bw_cpus(dest_cpu); | |
5197 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5198 | if (overflow) | |
5199 | ret = -EBUSY; | |
5200 | else { | |
5201 | /* | |
5202 | * We reserve space for this task in the destination | |
5203 | * root_domain, as we can't fail after this point. | |
5204 | * We will free resources in the source root_domain | |
5205 | * later on (see set_cpus_allowed_dl()). | |
5206 | */ | |
5207 | __dl_add(dl_b, p->dl.dl_bw); | |
5208 | } | |
5209 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5210 | rcu_read_unlock_sched(); |
7f51412a JL |
5211 | |
5212 | } | |
5213 | #endif | |
5214 | out: | |
5215 | return ret; | |
5216 | } | |
5217 | ||
1da177e4 | 5218 | #ifdef CONFIG_SMP |
1da177e4 | 5219 | |
e6628d5b MG |
5220 | #ifdef CONFIG_NUMA_BALANCING |
5221 | /* Migrate current task p to target_cpu */ | |
5222 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5223 | { | |
5224 | struct migration_arg arg = { p, target_cpu }; | |
5225 | int curr_cpu = task_cpu(p); | |
5226 | ||
5227 | if (curr_cpu == target_cpu) | |
5228 | return 0; | |
5229 | ||
5230 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5231 | return -EINVAL; | |
5232 | ||
5233 | /* TODO: This is not properly updating schedstats */ | |
5234 | ||
286549dc | 5235 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5236 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5237 | } | |
0ec8aa00 PZ |
5238 | |
5239 | /* | |
5240 | * Requeue a task on a given node and accurately track the number of NUMA | |
5241 | * tasks on the runqueues | |
5242 | */ | |
5243 | void sched_setnuma(struct task_struct *p, int nid) | |
5244 | { | |
5245 | struct rq *rq; | |
5246 | unsigned long flags; | |
da0c1e65 | 5247 | bool queued, running; |
0ec8aa00 PZ |
5248 | |
5249 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 5250 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5251 | running = task_current(rq, p); |
5252 | ||
da0c1e65 | 5253 | if (queued) |
1de64443 | 5254 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5255 | if (running) |
f3cd1c4e | 5256 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5257 | |
5258 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5259 | |
5260 | if (running) | |
5261 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5262 | if (queued) |
1de64443 | 5263 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
0ec8aa00 PZ |
5264 | task_rq_unlock(rq, p, &flags); |
5265 | } | |
5cc389bc | 5266 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5267 | |
1da177e4 | 5268 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5269 | /* |
48c5ccae PZ |
5270 | * Ensures that the idle task is using init_mm right before its cpu goes |
5271 | * offline. | |
054b9108 | 5272 | */ |
48c5ccae | 5273 | void idle_task_exit(void) |
1da177e4 | 5274 | { |
48c5ccae | 5275 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5276 | |
48c5ccae | 5277 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5278 | |
a53efe5f | 5279 | if (mm != &init_mm) { |
f98db601 | 5280 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5281 | finish_arch_post_lock_switch(); |
5282 | } | |
48c5ccae | 5283 | mmdrop(mm); |
1da177e4 LT |
5284 | } |
5285 | ||
5286 | /* | |
5d180232 PZ |
5287 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5288 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5289 | * nr_active count is stable. | |
5290 | * | |
5291 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5292 | */ |
5d180232 | 5293 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5294 | { |
5d180232 PZ |
5295 | long delta = calc_load_fold_active(rq); |
5296 | if (delta) | |
5297 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5298 | } |
5299 | ||
3f1d2a31 PZ |
5300 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5301 | { | |
5302 | } | |
5303 | ||
5304 | static const struct sched_class fake_sched_class = { | |
5305 | .put_prev_task = put_prev_task_fake, | |
5306 | }; | |
5307 | ||
5308 | static struct task_struct fake_task = { | |
5309 | /* | |
5310 | * Avoid pull_{rt,dl}_task() | |
5311 | */ | |
5312 | .prio = MAX_PRIO + 1, | |
5313 | .sched_class = &fake_sched_class, | |
5314 | }; | |
5315 | ||
48f24c4d | 5316 | /* |
48c5ccae PZ |
5317 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5318 | * try_to_wake_up()->select_task_rq(). | |
5319 | * | |
5320 | * Called with rq->lock held even though we'er in stop_machine() and | |
5321 | * there's no concurrency possible, we hold the required locks anyway | |
5322 | * because of lock validation efforts. | |
1da177e4 | 5323 | */ |
5e16bbc2 | 5324 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5325 | { |
5e16bbc2 | 5326 | struct rq *rq = dead_rq; |
48c5ccae PZ |
5327 | struct task_struct *next, *stop = rq->stop; |
5328 | int dest_cpu; | |
1da177e4 LT |
5329 | |
5330 | /* | |
48c5ccae PZ |
5331 | * Fudge the rq selection such that the below task selection loop |
5332 | * doesn't get stuck on the currently eligible stop task. | |
5333 | * | |
5334 | * We're currently inside stop_machine() and the rq is either stuck | |
5335 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5336 | * either way we should never end up calling schedule() until we're | |
5337 | * done here. | |
1da177e4 | 5338 | */ |
48c5ccae | 5339 | rq->stop = NULL; |
48f24c4d | 5340 | |
77bd3970 FW |
5341 | /* |
5342 | * put_prev_task() and pick_next_task() sched | |
5343 | * class method both need to have an up-to-date | |
5344 | * value of rq->clock[_task] | |
5345 | */ | |
5346 | update_rq_clock(rq); | |
5347 | ||
5e16bbc2 | 5348 | for (;;) { |
48c5ccae PZ |
5349 | /* |
5350 | * There's this thread running, bail when that's the only | |
5351 | * remaining thread. | |
5352 | */ | |
5353 | if (rq->nr_running == 1) | |
dd41f596 | 5354 | break; |
48c5ccae | 5355 | |
cbce1a68 | 5356 | /* |
5473e0cc | 5357 | * pick_next_task assumes pinned rq->lock. |
cbce1a68 PZ |
5358 | */ |
5359 | lockdep_pin_lock(&rq->lock); | |
3f1d2a31 | 5360 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5361 | BUG_ON(!next); |
79c53799 | 5362 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5363 | |
5473e0cc WL |
5364 | /* |
5365 | * Rules for changing task_struct::cpus_allowed are holding | |
5366 | * both pi_lock and rq->lock, such that holding either | |
5367 | * stabilizes the mask. | |
5368 | * | |
5369 | * Drop rq->lock is not quite as disastrous as it usually is | |
5370 | * because !cpu_active at this point, which means load-balance | |
5371 | * will not interfere. Also, stop-machine. | |
5372 | */ | |
5373 | lockdep_unpin_lock(&rq->lock); | |
5374 | raw_spin_unlock(&rq->lock); | |
5375 | raw_spin_lock(&next->pi_lock); | |
5376 | raw_spin_lock(&rq->lock); | |
5377 | ||
5378 | /* | |
5379 | * Since we're inside stop-machine, _nothing_ should have | |
5380 | * changed the task, WARN if weird stuff happened, because in | |
5381 | * that case the above rq->lock drop is a fail too. | |
5382 | */ | |
5383 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5384 | raw_spin_unlock(&next->pi_lock); | |
5385 | continue; | |
5386 | } | |
5387 | ||
48c5ccae | 5388 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5389 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5390 | |
5e16bbc2 PZ |
5391 | rq = __migrate_task(rq, next, dest_cpu); |
5392 | if (rq != dead_rq) { | |
5393 | raw_spin_unlock(&rq->lock); | |
5394 | rq = dead_rq; | |
5395 | raw_spin_lock(&rq->lock); | |
5396 | } | |
5473e0cc | 5397 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5398 | } |
dce48a84 | 5399 | |
48c5ccae | 5400 | rq->stop = stop; |
dce48a84 | 5401 | } |
1da177e4 LT |
5402 | #endif /* CONFIG_HOTPLUG_CPU */ |
5403 | ||
1f11eb6a GH |
5404 | static void set_rq_online(struct rq *rq) |
5405 | { | |
5406 | if (!rq->online) { | |
5407 | const struct sched_class *class; | |
5408 | ||
c6c4927b | 5409 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5410 | rq->online = 1; |
5411 | ||
5412 | for_each_class(class) { | |
5413 | if (class->rq_online) | |
5414 | class->rq_online(rq); | |
5415 | } | |
5416 | } | |
5417 | } | |
5418 | ||
5419 | static void set_rq_offline(struct rq *rq) | |
5420 | { | |
5421 | if (rq->online) { | |
5422 | const struct sched_class *class; | |
5423 | ||
5424 | for_each_class(class) { | |
5425 | if (class->rq_offline) | |
5426 | class->rq_offline(rq); | |
5427 | } | |
5428 | ||
c6c4927b | 5429 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5430 | rq->online = 0; |
5431 | } | |
5432 | } | |
5433 | ||
1da177e4 LT |
5434 | /* |
5435 | * migration_call - callback that gets triggered when a CPU is added. | |
5436 | * Here we can start up the necessary migration thread for the new CPU. | |
5437 | */ | |
0db0628d | 5438 | static int |
48f24c4d | 5439 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5440 | { |
48f24c4d | 5441 | int cpu = (long)hcpu; |
1da177e4 | 5442 | unsigned long flags; |
969c7921 | 5443 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5444 | |
48c5ccae | 5445 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5446 | |
1da177e4 | 5447 | case CPU_UP_PREPARE: |
a468d389 | 5448 | rq->calc_load_update = calc_load_update; |
e9532e69 | 5449 | account_reset_rq(rq); |
1da177e4 | 5450 | break; |
48f24c4d | 5451 | |
1da177e4 | 5452 | case CPU_ONLINE: |
1f94ef59 | 5453 | /* Update our root-domain */ |
05fa785c | 5454 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5455 | if (rq->rd) { |
c6c4927b | 5456 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5457 | |
5458 | set_rq_online(rq); | |
1f94ef59 | 5459 | } |
05fa785c | 5460 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5461 | break; |
48f24c4d | 5462 | |
1da177e4 | 5463 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5464 | case CPU_DYING: |
317f3941 | 5465 | sched_ttwu_pending(); |
57d885fe | 5466 | /* Update our root-domain */ |
05fa785c | 5467 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5468 | if (rq->rd) { |
c6c4927b | 5469 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5470 | set_rq_offline(rq); |
57d885fe | 5471 | } |
5e16bbc2 | 5472 | migrate_tasks(rq); |
48c5ccae | 5473 | BUG_ON(rq->nr_running != 1); /* the migration thread */ |
05fa785c | 5474 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5475 | break; |
48c5ccae | 5476 | |
5d180232 | 5477 | case CPU_DEAD: |
f319da0c | 5478 | calc_load_migrate(rq); |
57d885fe | 5479 | break; |
1da177e4 LT |
5480 | #endif |
5481 | } | |
49c022e6 PZ |
5482 | |
5483 | update_max_interval(); | |
5484 | ||
1da177e4 LT |
5485 | return NOTIFY_OK; |
5486 | } | |
5487 | ||
f38b0820 PM |
5488 | /* |
5489 | * Register at high priority so that task migration (migrate_all_tasks) | |
5490 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5491 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5492 | */ |
0db0628d | 5493 | static struct notifier_block migration_notifier = { |
1da177e4 | 5494 | .notifier_call = migration_call, |
50a323b7 | 5495 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5496 | }; |
5497 | ||
6a82b60d | 5498 | static void set_cpu_rq_start_time(void) |
a803f026 CM |
5499 | { |
5500 | int cpu = smp_processor_id(); | |
5501 | struct rq *rq = cpu_rq(cpu); | |
5502 | rq->age_stamp = sched_clock_cpu(cpu); | |
5503 | } | |
5504 | ||
0db0628d | 5505 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5506 | unsigned long action, void *hcpu) |
5507 | { | |
07f06cb3 PZ |
5508 | int cpu = (long)hcpu; |
5509 | ||
3a101d05 | 5510 | switch (action & ~CPU_TASKS_FROZEN) { |
a803f026 CM |
5511 | case CPU_STARTING: |
5512 | set_cpu_rq_start_time(); | |
5513 | return NOTIFY_OK; | |
07f06cb3 | 5514 | |
3a101d05 | 5515 | case CPU_DOWN_FAILED: |
07f06cb3 | 5516 | set_cpu_active(cpu, true); |
3a101d05 | 5517 | return NOTIFY_OK; |
07f06cb3 | 5518 | |
3a101d05 TH |
5519 | default: |
5520 | return NOTIFY_DONE; | |
5521 | } | |
5522 | } | |
5523 | ||
0db0628d | 5524 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5525 | unsigned long action, void *hcpu) |
5526 | { | |
5527 | switch (action & ~CPU_TASKS_FROZEN) { | |
5528 | case CPU_DOWN_PREPARE: | |
3c18d447 | 5529 | set_cpu_active((long)hcpu, false); |
3a101d05 | 5530 | return NOTIFY_OK; |
3c18d447 JL |
5531 | default: |
5532 | return NOTIFY_DONE; | |
3a101d05 TH |
5533 | } |
5534 | } | |
5535 | ||
7babe8db | 5536 | static int __init migration_init(void) |
1da177e4 LT |
5537 | { |
5538 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5539 | int err; |
48f24c4d | 5540 | |
3a101d05 | 5541 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5542 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5543 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5544 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5545 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5546 | |
3a101d05 TH |
5547 | /* Register cpu active notifiers */ |
5548 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5549 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5550 | ||
a004cd42 | 5551 | return 0; |
1da177e4 | 5552 | } |
7babe8db | 5553 | early_initcall(migration_init); |
476f3534 | 5554 | |
4cb98839 PZ |
5555 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5556 | ||
3e9830dc | 5557 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5558 | |
d039ac60 | 5559 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5560 | |
d039ac60 | 5561 | static int __init sched_debug_setup(char *str) |
f6630114 | 5562 | { |
d039ac60 | 5563 | sched_debug_enabled = 1; |
f6630114 MT |
5564 | |
5565 | return 0; | |
5566 | } | |
d039ac60 PZ |
5567 | early_param("sched_debug", sched_debug_setup); |
5568 | ||
5569 | static inline bool sched_debug(void) | |
5570 | { | |
5571 | return sched_debug_enabled; | |
5572 | } | |
f6630114 | 5573 | |
7c16ec58 | 5574 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5575 | struct cpumask *groupmask) |
1da177e4 | 5576 | { |
4dcf6aff | 5577 | struct sched_group *group = sd->groups; |
1da177e4 | 5578 | |
96f874e2 | 5579 | cpumask_clear(groupmask); |
4dcf6aff IM |
5580 | |
5581 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5582 | ||
5583 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5584 | printk("does not load-balance\n"); |
4dcf6aff | 5585 | if (sd->parent) |
3df0fc5b PZ |
5586 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5587 | " has parent"); | |
4dcf6aff | 5588 | return -1; |
41c7ce9a NP |
5589 | } |
5590 | ||
333470ee TH |
5591 | printk(KERN_CONT "span %*pbl level %s\n", |
5592 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5593 | |
758b2cdc | 5594 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5595 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5596 | "CPU%d\n", cpu); | |
4dcf6aff | 5597 | } |
758b2cdc | 5598 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5599 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5600 | " CPU%d\n", cpu); | |
4dcf6aff | 5601 | } |
1da177e4 | 5602 | |
4dcf6aff | 5603 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5604 | do { |
4dcf6aff | 5605 | if (!group) { |
3df0fc5b PZ |
5606 | printk("\n"); |
5607 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5608 | break; |
5609 | } | |
5610 | ||
758b2cdc | 5611 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5612 | printk(KERN_CONT "\n"); |
5613 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5614 | break; |
5615 | } | |
1da177e4 | 5616 | |
cb83b629 PZ |
5617 | if (!(sd->flags & SD_OVERLAP) && |
5618 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5619 | printk(KERN_CONT "\n"); |
5620 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5621 | break; |
5622 | } | |
1da177e4 | 5623 | |
758b2cdc | 5624 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5625 | |
333470ee TH |
5626 | printk(KERN_CONT " %*pbl", |
5627 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5628 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5629 | printk(KERN_CONT " (cpu_capacity = %d)", |
5630 | group->sgc->capacity); | |
381512cf | 5631 | } |
1da177e4 | 5632 | |
4dcf6aff IM |
5633 | group = group->next; |
5634 | } while (group != sd->groups); | |
3df0fc5b | 5635 | printk(KERN_CONT "\n"); |
1da177e4 | 5636 | |
758b2cdc | 5637 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5638 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5639 | |
758b2cdc RR |
5640 | if (sd->parent && |
5641 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5642 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5643 | "of domain->span\n"); | |
4dcf6aff IM |
5644 | return 0; |
5645 | } | |
1da177e4 | 5646 | |
4dcf6aff IM |
5647 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5648 | { | |
5649 | int level = 0; | |
1da177e4 | 5650 | |
d039ac60 | 5651 | if (!sched_debug_enabled) |
f6630114 MT |
5652 | return; |
5653 | ||
4dcf6aff IM |
5654 | if (!sd) { |
5655 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5656 | return; | |
5657 | } | |
1da177e4 | 5658 | |
4dcf6aff IM |
5659 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5660 | ||
5661 | for (;;) { | |
4cb98839 | 5662 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5663 | break; |
1da177e4 LT |
5664 | level++; |
5665 | sd = sd->parent; | |
33859f7f | 5666 | if (!sd) |
4dcf6aff IM |
5667 | break; |
5668 | } | |
1da177e4 | 5669 | } |
6d6bc0ad | 5670 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5671 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5672 | static inline bool sched_debug(void) |
5673 | { | |
5674 | return false; | |
5675 | } | |
6d6bc0ad | 5676 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5677 | |
1a20ff27 | 5678 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5679 | { |
758b2cdc | 5680 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5681 | return 1; |
5682 | ||
5683 | /* Following flags need at least 2 groups */ | |
5684 | if (sd->flags & (SD_LOAD_BALANCE | | |
5685 | SD_BALANCE_NEWIDLE | | |
5686 | SD_BALANCE_FORK | | |
89c4710e | 5687 | SD_BALANCE_EXEC | |
5d4dfddd | 5688 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5689 | SD_SHARE_PKG_RESOURCES | |
5690 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5691 | if (sd->groups != sd->groups->next) |
5692 | return 0; | |
5693 | } | |
5694 | ||
5695 | /* Following flags don't use groups */ | |
c88d5910 | 5696 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5697 | return 0; |
5698 | ||
5699 | return 1; | |
5700 | } | |
5701 | ||
48f24c4d IM |
5702 | static int |
5703 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5704 | { |
5705 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5706 | ||
5707 | if (sd_degenerate(parent)) | |
5708 | return 1; | |
5709 | ||
758b2cdc | 5710 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5711 | return 0; |
5712 | ||
245af2c7 SS |
5713 | /* Flags needing groups don't count if only 1 group in parent */ |
5714 | if (parent->groups == parent->groups->next) { | |
5715 | pflags &= ~(SD_LOAD_BALANCE | | |
5716 | SD_BALANCE_NEWIDLE | | |
5717 | SD_BALANCE_FORK | | |
89c4710e | 5718 | SD_BALANCE_EXEC | |
5d4dfddd | 5719 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5720 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5721 | SD_PREFER_SIBLING | |
5722 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5723 | if (nr_node_ids == 1) |
5724 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5725 | } |
5726 | if (~cflags & pflags) | |
5727 | return 0; | |
5728 | ||
5729 | return 1; | |
5730 | } | |
5731 | ||
dce840a0 | 5732 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5733 | { |
dce840a0 | 5734 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5735 | |
68e74568 | 5736 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5737 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5738 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5739 | free_cpumask_var(rd->rto_mask); |
5740 | free_cpumask_var(rd->online); | |
5741 | free_cpumask_var(rd->span); | |
5742 | kfree(rd); | |
5743 | } | |
5744 | ||
57d885fe GH |
5745 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5746 | { | |
a0490fa3 | 5747 | struct root_domain *old_rd = NULL; |
57d885fe | 5748 | unsigned long flags; |
57d885fe | 5749 | |
05fa785c | 5750 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5751 | |
5752 | if (rq->rd) { | |
a0490fa3 | 5753 | old_rd = rq->rd; |
57d885fe | 5754 | |
c6c4927b | 5755 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5756 | set_rq_offline(rq); |
57d885fe | 5757 | |
c6c4927b | 5758 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5759 | |
a0490fa3 | 5760 | /* |
0515973f | 5761 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5762 | * set old_rd to NULL to skip the freeing later |
5763 | * in this function: | |
5764 | */ | |
5765 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5766 | old_rd = NULL; | |
57d885fe GH |
5767 | } |
5768 | ||
5769 | atomic_inc(&rd->refcount); | |
5770 | rq->rd = rd; | |
5771 | ||
c6c4927b | 5772 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5773 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5774 | set_rq_online(rq); |
57d885fe | 5775 | |
05fa785c | 5776 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5777 | |
5778 | if (old_rd) | |
dce840a0 | 5779 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5780 | } |
5781 | ||
68c38fc3 | 5782 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5783 | { |
5784 | memset(rd, 0, sizeof(*rd)); | |
5785 | ||
8295c699 | 5786 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5787 | goto out; |
8295c699 | 5788 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5789 | goto free_span; |
8295c699 | 5790 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5791 | goto free_online; |
8295c699 | 5792 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
1baca4ce | 5793 | goto free_dlo_mask; |
6e0534f2 | 5794 | |
332ac17e | 5795 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5796 | if (cpudl_init(&rd->cpudl) != 0) |
5797 | goto free_dlo_mask; | |
332ac17e | 5798 | |
68c38fc3 | 5799 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5800 | goto free_rto_mask; |
c6c4927b | 5801 | return 0; |
6e0534f2 | 5802 | |
68e74568 RR |
5803 | free_rto_mask: |
5804 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5805 | free_dlo_mask: |
5806 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5807 | free_online: |
5808 | free_cpumask_var(rd->online); | |
5809 | free_span: | |
5810 | free_cpumask_var(rd->span); | |
0c910d28 | 5811 | out: |
c6c4927b | 5812 | return -ENOMEM; |
57d885fe GH |
5813 | } |
5814 | ||
029632fb PZ |
5815 | /* |
5816 | * By default the system creates a single root-domain with all cpus as | |
5817 | * members (mimicking the global state we have today). | |
5818 | */ | |
5819 | struct root_domain def_root_domain; | |
5820 | ||
57d885fe GH |
5821 | static void init_defrootdomain(void) |
5822 | { | |
68c38fc3 | 5823 | init_rootdomain(&def_root_domain); |
c6c4927b | 5824 | |
57d885fe GH |
5825 | atomic_set(&def_root_domain.refcount, 1); |
5826 | } | |
5827 | ||
dc938520 | 5828 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5829 | { |
5830 | struct root_domain *rd; | |
5831 | ||
5832 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5833 | if (!rd) | |
5834 | return NULL; | |
5835 | ||
68c38fc3 | 5836 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5837 | kfree(rd); |
5838 | return NULL; | |
5839 | } | |
57d885fe GH |
5840 | |
5841 | return rd; | |
5842 | } | |
5843 | ||
63b2ca30 | 5844 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5845 | { |
5846 | struct sched_group *tmp, *first; | |
5847 | ||
5848 | if (!sg) | |
5849 | return; | |
5850 | ||
5851 | first = sg; | |
5852 | do { | |
5853 | tmp = sg->next; | |
5854 | ||
63b2ca30 NP |
5855 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5856 | kfree(sg->sgc); | |
e3589f6c PZ |
5857 | |
5858 | kfree(sg); | |
5859 | sg = tmp; | |
5860 | } while (sg != first); | |
5861 | } | |
5862 | ||
dce840a0 PZ |
5863 | static void free_sched_domain(struct rcu_head *rcu) |
5864 | { | |
5865 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5866 | |
5867 | /* | |
5868 | * If its an overlapping domain it has private groups, iterate and | |
5869 | * nuke them all. | |
5870 | */ | |
5871 | if (sd->flags & SD_OVERLAP) { | |
5872 | free_sched_groups(sd->groups, 1); | |
5873 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5874 | kfree(sd->groups->sgc); |
dce840a0 | 5875 | kfree(sd->groups); |
9c3f75cb | 5876 | } |
dce840a0 PZ |
5877 | kfree(sd); |
5878 | } | |
5879 | ||
5880 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5881 | { | |
5882 | call_rcu(&sd->rcu, free_sched_domain); | |
5883 | } | |
5884 | ||
5885 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5886 | { | |
5887 | for (; sd; sd = sd->parent) | |
5888 | destroy_sched_domain(sd, cpu); | |
5889 | } | |
5890 | ||
518cd623 PZ |
5891 | /* |
5892 | * Keep a special pointer to the highest sched_domain that has | |
5893 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5894 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5895 | * | |
5896 | * Also keep a unique ID per domain (we use the first cpu number in | |
5897 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5898 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5899 | */ |
5900 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5901 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5902 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5903 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5904 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5905 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5906 | |
5907 | static void update_top_cache_domain(int cpu) | |
5908 | { | |
5909 | struct sched_domain *sd; | |
5d4cf996 | 5910 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5911 | int id = cpu; |
7d9ffa89 | 5912 | int size = 1; |
518cd623 PZ |
5913 | |
5914 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5915 | if (sd) { |
518cd623 | 5916 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5917 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5918 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5919 | } |
5d4cf996 | 5920 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5921 | |
5922 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5923 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5924 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5925 | |
5926 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5927 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5928 | |
5929 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5930 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5931 | } |
5932 | ||
1da177e4 | 5933 | /* |
0eab9146 | 5934 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5935 | * hold the hotplug lock. |
5936 | */ | |
0eab9146 IM |
5937 | static void |
5938 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5939 | { |
70b97a7f | 5940 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5941 | struct sched_domain *tmp; |
5942 | ||
5943 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5944 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5945 | struct sched_domain *parent = tmp->parent; |
5946 | if (!parent) | |
5947 | break; | |
f29c9b1c | 5948 | |
1a848870 | 5949 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5950 | tmp->parent = parent->parent; |
1a848870 SS |
5951 | if (parent->parent) |
5952 | parent->parent->child = tmp; | |
10866e62 PZ |
5953 | /* |
5954 | * Transfer SD_PREFER_SIBLING down in case of a | |
5955 | * degenerate parent; the spans match for this | |
5956 | * so the property transfers. | |
5957 | */ | |
5958 | if (parent->flags & SD_PREFER_SIBLING) | |
5959 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5960 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5961 | } else |
5962 | tmp = tmp->parent; | |
245af2c7 SS |
5963 | } |
5964 | ||
1a848870 | 5965 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5966 | tmp = sd; |
245af2c7 | 5967 | sd = sd->parent; |
dce840a0 | 5968 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5969 | if (sd) |
5970 | sd->child = NULL; | |
5971 | } | |
1da177e4 | 5972 | |
4cb98839 | 5973 | sched_domain_debug(sd, cpu); |
1da177e4 | 5974 | |
57d885fe | 5975 | rq_attach_root(rq, rd); |
dce840a0 | 5976 | tmp = rq->sd; |
674311d5 | 5977 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5978 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5979 | |
5980 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5981 | } |
5982 | ||
1da177e4 LT |
5983 | /* Setup the mask of cpus configured for isolated domains */ |
5984 | static int __init isolated_cpu_setup(char *str) | |
5985 | { | |
a6e4491c PB |
5986 | int ret; |
5987 | ||
bdddd296 | 5988 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
a6e4491c PB |
5989 | ret = cpulist_parse(str, cpu_isolated_map); |
5990 | if (ret) { | |
5991 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); | |
5992 | return 0; | |
5993 | } | |
1da177e4 LT |
5994 | return 1; |
5995 | } | |
8927f494 | 5996 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5997 | |
49a02c51 | 5998 | struct s_data { |
21d42ccf | 5999 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6000 | struct root_domain *rd; |
6001 | }; | |
6002 | ||
2109b99e | 6003 | enum s_alloc { |
2109b99e | 6004 | sa_rootdomain, |
21d42ccf | 6005 | sa_sd, |
dce840a0 | 6006 | sa_sd_storage, |
2109b99e AH |
6007 | sa_none, |
6008 | }; | |
6009 | ||
c1174876 PZ |
6010 | /* |
6011 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
6012 | * domain traversal. | |
6013 | * | |
6014 | * Asymmetric node setups can result in situations where the domain tree is of | |
6015 | * unequal depth, make sure to skip domains that already cover the entire | |
6016 | * range. | |
6017 | * | |
6018 | * In that case build_sched_domains() will have terminated the iteration early | |
6019 | * and our sibling sd spans will be empty. Domains should always include the | |
6020 | * cpu they're built on, so check that. | |
6021 | * | |
6022 | */ | |
6023 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6024 | { | |
6025 | const struct cpumask *span = sched_domain_span(sd); | |
6026 | struct sd_data *sdd = sd->private; | |
6027 | struct sched_domain *sibling; | |
6028 | int i; | |
6029 | ||
6030 | for_each_cpu(i, span) { | |
6031 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6032 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6033 | continue; | |
6034 | ||
6035 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6036 | } | |
6037 | } | |
6038 | ||
6039 | /* | |
6040 | * Return the canonical balance cpu for this group, this is the first cpu | |
6041 | * of this group that's also in the iteration mask. | |
6042 | */ | |
6043 | int group_balance_cpu(struct sched_group *sg) | |
6044 | { | |
6045 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6046 | } | |
6047 | ||
e3589f6c PZ |
6048 | static int |
6049 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6050 | { | |
6051 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6052 | const struct cpumask *span = sched_domain_span(sd); | |
6053 | struct cpumask *covered = sched_domains_tmpmask; | |
6054 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6055 | struct sched_domain *sibling; |
e3589f6c PZ |
6056 | int i; |
6057 | ||
6058 | cpumask_clear(covered); | |
6059 | ||
6060 | for_each_cpu(i, span) { | |
6061 | struct cpumask *sg_span; | |
6062 | ||
6063 | if (cpumask_test_cpu(i, covered)) | |
6064 | continue; | |
6065 | ||
aaecac4a | 6066 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6067 | |
6068 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6069 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6070 | continue; |
6071 | ||
e3589f6c | 6072 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6073 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6074 | |
6075 | if (!sg) | |
6076 | goto fail; | |
6077 | ||
6078 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6079 | if (sibling->child) |
6080 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6081 | else | |
e3589f6c PZ |
6082 | cpumask_set_cpu(i, sg_span); |
6083 | ||
6084 | cpumask_or(covered, covered, sg_span); | |
6085 | ||
63b2ca30 NP |
6086 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6087 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6088 | build_group_mask(sd, sg); |
6089 | ||
c3decf0d | 6090 | /* |
63b2ca30 | 6091 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6092 | * domains and no possible iteration will get us here, we won't |
6093 | * die on a /0 trap. | |
6094 | */ | |
ca8ce3d0 | 6095 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 6096 | |
c1174876 PZ |
6097 | /* |
6098 | * Make sure the first group of this domain contains the | |
6099 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6100 | * breaks. See update_sg_lb_stats(). | |
6101 | */ | |
74a5ce20 | 6102 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6103 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6104 | groups = sg; |
6105 | ||
6106 | if (!first) | |
6107 | first = sg; | |
6108 | if (last) | |
6109 | last->next = sg; | |
6110 | last = sg; | |
6111 | last->next = first; | |
6112 | } | |
6113 | sd->groups = groups; | |
6114 | ||
6115 | return 0; | |
6116 | ||
6117 | fail: | |
6118 | free_sched_groups(first, 0); | |
6119 | ||
6120 | return -ENOMEM; | |
6121 | } | |
6122 | ||
dce840a0 | 6123 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6124 | { |
dce840a0 PZ |
6125 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6126 | struct sched_domain *child = sd->child; | |
1da177e4 | 6127 | |
dce840a0 PZ |
6128 | if (child) |
6129 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6130 | |
9c3f75cb | 6131 | if (sg) { |
dce840a0 | 6132 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6133 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6134 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6135 | } |
dce840a0 PZ |
6136 | |
6137 | return cpu; | |
1e9f28fa | 6138 | } |
1e9f28fa | 6139 | |
01a08546 | 6140 | /* |
dce840a0 PZ |
6141 | * build_sched_groups will build a circular linked list of the groups |
6142 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6143 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6144 | * |
6145 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6146 | */ |
e3589f6c PZ |
6147 | static int |
6148 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6149 | { |
dce840a0 PZ |
6150 | struct sched_group *first = NULL, *last = NULL; |
6151 | struct sd_data *sdd = sd->private; | |
6152 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6153 | struct cpumask *covered; |
dce840a0 | 6154 | int i; |
9c1cfda2 | 6155 | |
e3589f6c PZ |
6156 | get_group(cpu, sdd, &sd->groups); |
6157 | atomic_inc(&sd->groups->ref); | |
6158 | ||
0936629f | 6159 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6160 | return 0; |
6161 | ||
f96225fd PZ |
6162 | lockdep_assert_held(&sched_domains_mutex); |
6163 | covered = sched_domains_tmpmask; | |
6164 | ||
dce840a0 | 6165 | cpumask_clear(covered); |
6711cab4 | 6166 | |
dce840a0 PZ |
6167 | for_each_cpu(i, span) { |
6168 | struct sched_group *sg; | |
cd08e923 | 6169 | int group, j; |
6711cab4 | 6170 | |
dce840a0 PZ |
6171 | if (cpumask_test_cpu(i, covered)) |
6172 | continue; | |
6711cab4 | 6173 | |
cd08e923 | 6174 | group = get_group(i, sdd, &sg); |
c1174876 | 6175 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6176 | |
dce840a0 PZ |
6177 | for_each_cpu(j, span) { |
6178 | if (get_group(j, sdd, NULL) != group) | |
6179 | continue; | |
0601a88d | 6180 | |
dce840a0 PZ |
6181 | cpumask_set_cpu(j, covered); |
6182 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6183 | } | |
0601a88d | 6184 | |
dce840a0 PZ |
6185 | if (!first) |
6186 | first = sg; | |
6187 | if (last) | |
6188 | last->next = sg; | |
6189 | last = sg; | |
6190 | } | |
6191 | last->next = first; | |
e3589f6c PZ |
6192 | |
6193 | return 0; | |
0601a88d | 6194 | } |
51888ca2 | 6195 | |
89c4710e | 6196 | /* |
63b2ca30 | 6197 | * Initialize sched groups cpu_capacity. |
89c4710e | 6198 | * |
63b2ca30 | 6199 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6200 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6201 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6202 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6203 | * group having more cpu_capacity will pickup more load compared to the | |
6204 | * group having less cpu_capacity. | |
89c4710e | 6205 | */ |
63b2ca30 | 6206 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6207 | { |
e3589f6c | 6208 | struct sched_group *sg = sd->groups; |
89c4710e | 6209 | |
94c95ba6 | 6210 | WARN_ON(!sg); |
e3589f6c PZ |
6211 | |
6212 | do { | |
6213 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6214 | sg = sg->next; | |
6215 | } while (sg != sd->groups); | |
89c4710e | 6216 | |
c1174876 | 6217 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6218 | return; |
aae6d3dd | 6219 | |
63b2ca30 NP |
6220 | update_group_capacity(sd, cpu); |
6221 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6222 | } |
6223 | ||
7c16ec58 MT |
6224 | /* |
6225 | * Initializers for schedule domains | |
6226 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6227 | */ | |
6228 | ||
1d3504fc | 6229 | static int default_relax_domain_level = -1; |
60495e77 | 6230 | int sched_domain_level_max; |
1d3504fc HS |
6231 | |
6232 | static int __init setup_relax_domain_level(char *str) | |
6233 | { | |
a841f8ce DS |
6234 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6235 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6236 | |
1d3504fc HS |
6237 | return 1; |
6238 | } | |
6239 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6240 | ||
6241 | static void set_domain_attribute(struct sched_domain *sd, | |
6242 | struct sched_domain_attr *attr) | |
6243 | { | |
6244 | int request; | |
6245 | ||
6246 | if (!attr || attr->relax_domain_level < 0) { | |
6247 | if (default_relax_domain_level < 0) | |
6248 | return; | |
6249 | else | |
6250 | request = default_relax_domain_level; | |
6251 | } else | |
6252 | request = attr->relax_domain_level; | |
6253 | if (request < sd->level) { | |
6254 | /* turn off idle balance on this domain */ | |
c88d5910 | 6255 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6256 | } else { |
6257 | /* turn on idle balance on this domain */ | |
c88d5910 | 6258 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6259 | } |
6260 | } | |
6261 | ||
54ab4ff4 PZ |
6262 | static void __sdt_free(const struct cpumask *cpu_map); |
6263 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6264 | ||
2109b99e AH |
6265 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6266 | const struct cpumask *cpu_map) | |
6267 | { | |
6268 | switch (what) { | |
2109b99e | 6269 | case sa_rootdomain: |
822ff793 PZ |
6270 | if (!atomic_read(&d->rd->refcount)) |
6271 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6272 | case sa_sd: |
6273 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6274 | case sa_sd_storage: |
54ab4ff4 | 6275 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6276 | case sa_none: |
6277 | break; | |
6278 | } | |
6279 | } | |
3404c8d9 | 6280 | |
2109b99e AH |
6281 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6282 | const struct cpumask *cpu_map) | |
6283 | { | |
dce840a0 PZ |
6284 | memset(d, 0, sizeof(*d)); |
6285 | ||
54ab4ff4 PZ |
6286 | if (__sdt_alloc(cpu_map)) |
6287 | return sa_sd_storage; | |
dce840a0 PZ |
6288 | d->sd = alloc_percpu(struct sched_domain *); |
6289 | if (!d->sd) | |
6290 | return sa_sd_storage; | |
2109b99e | 6291 | d->rd = alloc_rootdomain(); |
dce840a0 | 6292 | if (!d->rd) |
21d42ccf | 6293 | return sa_sd; |
2109b99e AH |
6294 | return sa_rootdomain; |
6295 | } | |
57d885fe | 6296 | |
dce840a0 PZ |
6297 | /* |
6298 | * NULL the sd_data elements we've used to build the sched_domain and | |
6299 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6300 | * will not free the data we're using. | |
6301 | */ | |
6302 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6303 | { | |
6304 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6305 | |
6306 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6307 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6308 | ||
e3589f6c | 6309 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6310 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6311 | |
63b2ca30 NP |
6312 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6313 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6314 | } |
6315 | ||
cb83b629 | 6316 | #ifdef CONFIG_NUMA |
cb83b629 | 6317 | static int sched_domains_numa_levels; |
e3fe70b1 | 6318 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6319 | static int *sched_domains_numa_distance; |
9942f79b | 6320 | int sched_max_numa_distance; |
cb83b629 PZ |
6321 | static struct cpumask ***sched_domains_numa_masks; |
6322 | static int sched_domains_curr_level; | |
143e1e28 | 6323 | #endif |
cb83b629 | 6324 | |
143e1e28 VG |
6325 | /* |
6326 | * SD_flags allowed in topology descriptions. | |
6327 | * | |
5d4dfddd | 6328 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6329 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6330 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6331 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6332 | * |
6333 | * Odd one out: | |
6334 | * SD_ASYM_PACKING - describes SMT quirks | |
6335 | */ | |
6336 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6337 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6338 | SD_SHARE_PKG_RESOURCES | \ |
6339 | SD_NUMA | \ | |
d77b3ed5 VG |
6340 | SD_ASYM_PACKING | \ |
6341 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6342 | |
6343 | static struct sched_domain * | |
143e1e28 | 6344 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6345 | { |
6346 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6347 | int sd_weight, sd_flags = 0; |
6348 | ||
6349 | #ifdef CONFIG_NUMA | |
6350 | /* | |
6351 | * Ugly hack to pass state to sd_numa_mask()... | |
6352 | */ | |
6353 | sched_domains_curr_level = tl->numa_level; | |
6354 | #endif | |
6355 | ||
6356 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6357 | ||
6358 | if (tl->sd_flags) | |
6359 | sd_flags = (*tl->sd_flags)(); | |
6360 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6361 | "wrong sd_flags in topology description\n")) | |
6362 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6363 | |
6364 | *sd = (struct sched_domain){ | |
6365 | .min_interval = sd_weight, | |
6366 | .max_interval = 2*sd_weight, | |
6367 | .busy_factor = 32, | |
870a0bb5 | 6368 | .imbalance_pct = 125, |
143e1e28 VG |
6369 | |
6370 | .cache_nice_tries = 0, | |
6371 | .busy_idx = 0, | |
6372 | .idle_idx = 0, | |
cb83b629 PZ |
6373 | .newidle_idx = 0, |
6374 | .wake_idx = 0, | |
6375 | .forkexec_idx = 0, | |
6376 | ||
6377 | .flags = 1*SD_LOAD_BALANCE | |
6378 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6379 | | 1*SD_BALANCE_EXEC |
6380 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6381 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6382 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6383 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6384 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6385 | | 0*SD_SERIALIZE |
cb83b629 | 6386 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6387 | | 0*SD_NUMA |
6388 | | sd_flags | |
cb83b629 | 6389 | , |
143e1e28 | 6390 | |
cb83b629 PZ |
6391 | .last_balance = jiffies, |
6392 | .balance_interval = sd_weight, | |
143e1e28 | 6393 | .smt_gain = 0, |
2b4cfe64 JL |
6394 | .max_newidle_lb_cost = 0, |
6395 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6396 | #ifdef CONFIG_SCHED_DEBUG |
6397 | .name = tl->name, | |
6398 | #endif | |
cb83b629 | 6399 | }; |
cb83b629 PZ |
6400 | |
6401 | /* | |
143e1e28 | 6402 | * Convert topological properties into behaviour. |
cb83b629 | 6403 | */ |
143e1e28 | 6404 | |
5d4dfddd | 6405 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6406 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6407 | sd->imbalance_pct = 110; |
6408 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6409 | |
6410 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6411 | sd->imbalance_pct = 117; | |
6412 | sd->cache_nice_tries = 1; | |
6413 | sd->busy_idx = 2; | |
6414 | ||
6415 | #ifdef CONFIG_NUMA | |
6416 | } else if (sd->flags & SD_NUMA) { | |
6417 | sd->cache_nice_tries = 2; | |
6418 | sd->busy_idx = 3; | |
6419 | sd->idle_idx = 2; | |
6420 | ||
6421 | sd->flags |= SD_SERIALIZE; | |
6422 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6423 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6424 | SD_BALANCE_FORK | | |
6425 | SD_WAKE_AFFINE); | |
6426 | } | |
6427 | ||
6428 | #endif | |
6429 | } else { | |
6430 | sd->flags |= SD_PREFER_SIBLING; | |
6431 | sd->cache_nice_tries = 1; | |
6432 | sd->busy_idx = 2; | |
6433 | sd->idle_idx = 1; | |
6434 | } | |
6435 | ||
6436 | sd->private = &tl->data; | |
cb83b629 PZ |
6437 | |
6438 | return sd; | |
6439 | } | |
6440 | ||
143e1e28 VG |
6441 | /* |
6442 | * Topology list, bottom-up. | |
6443 | */ | |
6444 | static struct sched_domain_topology_level default_topology[] = { | |
6445 | #ifdef CONFIG_SCHED_SMT | |
6446 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6447 | #endif | |
6448 | #ifdef CONFIG_SCHED_MC | |
6449 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6450 | #endif |
6451 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6452 | { NULL, }, | |
6453 | }; | |
6454 | ||
c6e1e7b5 JG |
6455 | static struct sched_domain_topology_level *sched_domain_topology = |
6456 | default_topology; | |
143e1e28 VG |
6457 | |
6458 | #define for_each_sd_topology(tl) \ | |
6459 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6460 | ||
6461 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6462 | { | |
6463 | sched_domain_topology = tl; | |
6464 | } | |
6465 | ||
6466 | #ifdef CONFIG_NUMA | |
6467 | ||
cb83b629 PZ |
6468 | static const struct cpumask *sd_numa_mask(int cpu) |
6469 | { | |
6470 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6471 | } | |
6472 | ||
d039ac60 PZ |
6473 | static void sched_numa_warn(const char *str) |
6474 | { | |
6475 | static int done = false; | |
6476 | int i,j; | |
6477 | ||
6478 | if (done) | |
6479 | return; | |
6480 | ||
6481 | done = true; | |
6482 | ||
6483 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6484 | ||
6485 | for (i = 0; i < nr_node_ids; i++) { | |
6486 | printk(KERN_WARNING " "); | |
6487 | for (j = 0; j < nr_node_ids; j++) | |
6488 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6489 | printk(KERN_CONT "\n"); | |
6490 | } | |
6491 | printk(KERN_WARNING "\n"); | |
6492 | } | |
6493 | ||
9942f79b | 6494 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6495 | { |
6496 | int i; | |
6497 | ||
6498 | if (distance == node_distance(0, 0)) | |
6499 | return true; | |
6500 | ||
6501 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6502 | if (sched_domains_numa_distance[i] == distance) | |
6503 | return true; | |
6504 | } | |
6505 | ||
6506 | return false; | |
6507 | } | |
6508 | ||
e3fe70b1 RR |
6509 | /* |
6510 | * A system can have three types of NUMA topology: | |
6511 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6512 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6513 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6514 | * | |
6515 | * The difference between a glueless mesh topology and a backplane | |
6516 | * topology lies in whether communication between not directly | |
6517 | * connected nodes goes through intermediary nodes (where programs | |
6518 | * could run), or through backplane controllers. This affects | |
6519 | * placement of programs. | |
6520 | * | |
6521 | * The type of topology can be discerned with the following tests: | |
6522 | * - If the maximum distance between any nodes is 1 hop, the system | |
6523 | * is directly connected. | |
6524 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6525 | * there is an intermediary node C, which is < N hops away from both | |
6526 | * nodes A and B, the system is a glueless mesh. | |
6527 | */ | |
6528 | static void init_numa_topology_type(void) | |
6529 | { | |
6530 | int a, b, c, n; | |
6531 | ||
6532 | n = sched_max_numa_distance; | |
6533 | ||
e237882b | 6534 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6535 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6536 | return; |
6537 | } | |
e3fe70b1 RR |
6538 | |
6539 | for_each_online_node(a) { | |
6540 | for_each_online_node(b) { | |
6541 | /* Find two nodes furthest removed from each other. */ | |
6542 | if (node_distance(a, b) < n) | |
6543 | continue; | |
6544 | ||
6545 | /* Is there an intermediary node between a and b? */ | |
6546 | for_each_online_node(c) { | |
6547 | if (node_distance(a, c) < n && | |
6548 | node_distance(b, c) < n) { | |
6549 | sched_numa_topology_type = | |
6550 | NUMA_GLUELESS_MESH; | |
6551 | return; | |
6552 | } | |
6553 | } | |
6554 | ||
6555 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6556 | return; | |
6557 | } | |
6558 | } | |
6559 | } | |
6560 | ||
cb83b629 PZ |
6561 | static void sched_init_numa(void) |
6562 | { | |
6563 | int next_distance, curr_distance = node_distance(0, 0); | |
6564 | struct sched_domain_topology_level *tl; | |
6565 | int level = 0; | |
6566 | int i, j, k; | |
6567 | ||
cb83b629 PZ |
6568 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6569 | if (!sched_domains_numa_distance) | |
6570 | return; | |
6571 | ||
6572 | /* | |
6573 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6574 | * unique distances in the node_distance() table. | |
6575 | * | |
6576 | * Assumes node_distance(0,j) includes all distances in | |
6577 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6578 | */ |
6579 | next_distance = curr_distance; | |
6580 | for (i = 0; i < nr_node_ids; i++) { | |
6581 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6582 | for (k = 0; k < nr_node_ids; k++) { |
6583 | int distance = node_distance(i, k); | |
6584 | ||
6585 | if (distance > curr_distance && | |
6586 | (distance < next_distance || | |
6587 | next_distance == curr_distance)) | |
6588 | next_distance = distance; | |
6589 | ||
6590 | /* | |
6591 | * While not a strong assumption it would be nice to know | |
6592 | * about cases where if node A is connected to B, B is not | |
6593 | * equally connected to A. | |
6594 | */ | |
6595 | if (sched_debug() && node_distance(k, i) != distance) | |
6596 | sched_numa_warn("Node-distance not symmetric"); | |
6597 | ||
6598 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6599 | sched_numa_warn("Node-0 not representative"); | |
6600 | } | |
6601 | if (next_distance != curr_distance) { | |
6602 | sched_domains_numa_distance[level++] = next_distance; | |
6603 | sched_domains_numa_levels = level; | |
6604 | curr_distance = next_distance; | |
6605 | } else break; | |
cb83b629 | 6606 | } |
d039ac60 PZ |
6607 | |
6608 | /* | |
6609 | * In case of sched_debug() we verify the above assumption. | |
6610 | */ | |
6611 | if (!sched_debug()) | |
6612 | break; | |
cb83b629 | 6613 | } |
c123588b AR |
6614 | |
6615 | if (!level) | |
6616 | return; | |
6617 | ||
cb83b629 PZ |
6618 | /* |
6619 | * 'level' contains the number of unique distances, excluding the | |
6620 | * identity distance node_distance(i,i). | |
6621 | * | |
28b4a521 | 6622 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6623 | * numbers. |
6624 | */ | |
6625 | ||
5f7865f3 TC |
6626 | /* |
6627 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6628 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6629 | * the array will contain less then 'level' members. This could be | |
6630 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6631 | * in other functions. | |
6632 | * | |
6633 | * We reset it to 'level' at the end of this function. | |
6634 | */ | |
6635 | sched_domains_numa_levels = 0; | |
6636 | ||
cb83b629 PZ |
6637 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6638 | if (!sched_domains_numa_masks) | |
6639 | return; | |
6640 | ||
6641 | /* | |
6642 | * Now for each level, construct a mask per node which contains all | |
6643 | * cpus of nodes that are that many hops away from us. | |
6644 | */ | |
6645 | for (i = 0; i < level; i++) { | |
6646 | sched_domains_numa_masks[i] = | |
6647 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6648 | if (!sched_domains_numa_masks[i]) | |
6649 | return; | |
6650 | ||
6651 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6652 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6653 | if (!mask) |
6654 | return; | |
6655 | ||
6656 | sched_domains_numa_masks[i][j] = mask; | |
6657 | ||
9c03ee14 | 6658 | for_each_node(k) { |
dd7d8634 | 6659 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6660 | continue; |
6661 | ||
6662 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6663 | } | |
6664 | } | |
6665 | } | |
6666 | ||
143e1e28 VG |
6667 | /* Compute default topology size */ |
6668 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6669 | ||
c515db8c | 6670 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6671 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6672 | if (!tl) | |
6673 | return; | |
6674 | ||
6675 | /* | |
6676 | * Copy the default topology bits.. | |
6677 | */ | |
143e1e28 VG |
6678 | for (i = 0; sched_domain_topology[i].mask; i++) |
6679 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6680 | |
6681 | /* | |
6682 | * .. and append 'j' levels of NUMA goodness. | |
6683 | */ | |
6684 | for (j = 0; j < level; i++, j++) { | |
6685 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6686 | .mask = sd_numa_mask, |
143e1e28 | 6687 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6688 | .flags = SDTL_OVERLAP, |
6689 | .numa_level = j, | |
143e1e28 | 6690 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6691 | }; |
6692 | } | |
6693 | ||
6694 | sched_domain_topology = tl; | |
5f7865f3 TC |
6695 | |
6696 | sched_domains_numa_levels = level; | |
9942f79b | 6697 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6698 | |
6699 | init_numa_topology_type(); | |
cb83b629 | 6700 | } |
301a5cba TC |
6701 | |
6702 | static void sched_domains_numa_masks_set(int cpu) | |
6703 | { | |
6704 | int i, j; | |
6705 | int node = cpu_to_node(cpu); | |
6706 | ||
6707 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6708 | for (j = 0; j < nr_node_ids; j++) { | |
6709 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6710 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6711 | } | |
6712 | } | |
6713 | } | |
6714 | ||
6715 | static void sched_domains_numa_masks_clear(int cpu) | |
6716 | { | |
6717 | int i, j; | |
6718 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6719 | for (j = 0; j < nr_node_ids; j++) | |
6720 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6721 | } | |
6722 | } | |
6723 | ||
6724 | /* | |
6725 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6726 | * are onlined. | |
6727 | */ | |
6728 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6729 | unsigned long action, | |
6730 | void *hcpu) | |
6731 | { | |
6732 | int cpu = (long)hcpu; | |
6733 | ||
6734 | switch (action & ~CPU_TASKS_FROZEN) { | |
6735 | case CPU_ONLINE: | |
6736 | sched_domains_numa_masks_set(cpu); | |
6737 | break; | |
6738 | ||
6739 | case CPU_DEAD: | |
6740 | sched_domains_numa_masks_clear(cpu); | |
6741 | break; | |
6742 | ||
6743 | default: | |
6744 | return NOTIFY_DONE; | |
6745 | } | |
6746 | ||
6747 | return NOTIFY_OK; | |
cb83b629 PZ |
6748 | } |
6749 | #else | |
6750 | static inline void sched_init_numa(void) | |
6751 | { | |
6752 | } | |
301a5cba TC |
6753 | |
6754 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6755 | unsigned long action, | |
6756 | void *hcpu) | |
6757 | { | |
6758 | return 0; | |
6759 | } | |
cb83b629 PZ |
6760 | #endif /* CONFIG_NUMA */ |
6761 | ||
54ab4ff4 PZ |
6762 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6763 | { | |
6764 | struct sched_domain_topology_level *tl; | |
6765 | int j; | |
6766 | ||
27723a68 | 6767 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6768 | struct sd_data *sdd = &tl->data; |
6769 | ||
6770 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6771 | if (!sdd->sd) | |
6772 | return -ENOMEM; | |
6773 | ||
6774 | sdd->sg = alloc_percpu(struct sched_group *); | |
6775 | if (!sdd->sg) | |
6776 | return -ENOMEM; | |
6777 | ||
63b2ca30 NP |
6778 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6779 | if (!sdd->sgc) | |
9c3f75cb PZ |
6780 | return -ENOMEM; |
6781 | ||
54ab4ff4 PZ |
6782 | for_each_cpu(j, cpu_map) { |
6783 | struct sched_domain *sd; | |
6784 | struct sched_group *sg; | |
63b2ca30 | 6785 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6786 | |
5cc389bc | 6787 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6788 | GFP_KERNEL, cpu_to_node(j)); |
6789 | if (!sd) | |
6790 | return -ENOMEM; | |
6791 | ||
6792 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6793 | ||
6794 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6795 | GFP_KERNEL, cpu_to_node(j)); | |
6796 | if (!sg) | |
6797 | return -ENOMEM; | |
6798 | ||
30b4e9eb IM |
6799 | sg->next = sg; |
6800 | ||
54ab4ff4 | 6801 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6802 | |
63b2ca30 | 6803 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6804 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6805 | if (!sgc) |
9c3f75cb PZ |
6806 | return -ENOMEM; |
6807 | ||
63b2ca30 | 6808 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6809 | } |
6810 | } | |
6811 | ||
6812 | return 0; | |
6813 | } | |
6814 | ||
6815 | static void __sdt_free(const struct cpumask *cpu_map) | |
6816 | { | |
6817 | struct sched_domain_topology_level *tl; | |
6818 | int j; | |
6819 | ||
27723a68 | 6820 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6821 | struct sd_data *sdd = &tl->data; |
6822 | ||
6823 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6824 | struct sched_domain *sd; |
6825 | ||
6826 | if (sdd->sd) { | |
6827 | sd = *per_cpu_ptr(sdd->sd, j); | |
6828 | if (sd && (sd->flags & SD_OVERLAP)) | |
6829 | free_sched_groups(sd->groups, 0); | |
6830 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6831 | } | |
6832 | ||
6833 | if (sdd->sg) | |
6834 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6835 | if (sdd->sgc) |
6836 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6837 | } |
6838 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6839 | sdd->sd = NULL; |
54ab4ff4 | 6840 | free_percpu(sdd->sg); |
fb2cf2c6 | 6841 | sdd->sg = NULL; |
63b2ca30 NP |
6842 | free_percpu(sdd->sgc); |
6843 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6844 | } |
6845 | } | |
6846 | ||
2c402dc3 | 6847 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6848 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6849 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6850 | { |
143e1e28 | 6851 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6852 | if (!sd) |
d069b916 | 6853 | return child; |
2c402dc3 | 6854 | |
2c402dc3 | 6855 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6856 | if (child) { |
6857 | sd->level = child->level + 1; | |
6858 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6859 | child->parent = sd; |
c75e0128 | 6860 | sd->child = child; |
6ae72dff PZ |
6861 | |
6862 | if (!cpumask_subset(sched_domain_span(child), | |
6863 | sched_domain_span(sd))) { | |
6864 | pr_err("BUG: arch topology borken\n"); | |
6865 | #ifdef CONFIG_SCHED_DEBUG | |
6866 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6867 | child->name, sd->name); | |
6868 | #endif | |
6869 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6870 | cpumask_or(sched_domain_span(sd), | |
6871 | sched_domain_span(sd), | |
6872 | sched_domain_span(child)); | |
6873 | } | |
6874 | ||
60495e77 | 6875 | } |
a841f8ce | 6876 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6877 | |
6878 | return sd; | |
6879 | } | |
6880 | ||
2109b99e AH |
6881 | /* |
6882 | * Build sched domains for a given set of cpus and attach the sched domains | |
6883 | * to the individual cpus | |
6884 | */ | |
dce840a0 PZ |
6885 | static int build_sched_domains(const struct cpumask *cpu_map, |
6886 | struct sched_domain_attr *attr) | |
2109b99e | 6887 | { |
1c632169 | 6888 | enum s_alloc alloc_state; |
dce840a0 | 6889 | struct sched_domain *sd; |
2109b99e | 6890 | struct s_data d; |
822ff793 | 6891 | int i, ret = -ENOMEM; |
9c1cfda2 | 6892 | |
2109b99e AH |
6893 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6894 | if (alloc_state != sa_rootdomain) | |
6895 | goto error; | |
9c1cfda2 | 6896 | |
dce840a0 | 6897 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6898 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6899 | struct sched_domain_topology_level *tl; |
6900 | ||
3bd65a80 | 6901 | sd = NULL; |
27723a68 | 6902 | for_each_sd_topology(tl) { |
4a850cbe | 6903 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6904 | if (tl == sched_domain_topology) |
6905 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6906 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6907 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6908 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6909 | break; | |
e3589f6c | 6910 | } |
dce840a0 PZ |
6911 | } |
6912 | ||
6913 | /* Build the groups for the domains */ | |
6914 | for_each_cpu(i, cpu_map) { | |
6915 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6916 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6917 | if (sd->flags & SD_OVERLAP) { |
6918 | if (build_overlap_sched_groups(sd, i)) | |
6919 | goto error; | |
6920 | } else { | |
6921 | if (build_sched_groups(sd, i)) | |
6922 | goto error; | |
6923 | } | |
1cf51902 | 6924 | } |
a06dadbe | 6925 | } |
9c1cfda2 | 6926 | |
ced549fa | 6927 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6928 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6929 | if (!cpumask_test_cpu(i, cpu_map)) | |
6930 | continue; | |
9c1cfda2 | 6931 | |
dce840a0 PZ |
6932 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6933 | claim_allocations(i, sd); | |
63b2ca30 | 6934 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6935 | } |
f712c0c7 | 6936 | } |
9c1cfda2 | 6937 | |
1da177e4 | 6938 | /* Attach the domains */ |
dce840a0 | 6939 | rcu_read_lock(); |
abcd083a | 6940 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6941 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6942 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6943 | } |
dce840a0 | 6944 | rcu_read_unlock(); |
51888ca2 | 6945 | |
822ff793 | 6946 | ret = 0; |
51888ca2 | 6947 | error: |
2109b99e | 6948 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6949 | return ret; |
1da177e4 | 6950 | } |
029190c5 | 6951 | |
acc3f5d7 | 6952 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6953 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6954 | static struct sched_domain_attr *dattr_cur; |
6955 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6956 | |
6957 | /* | |
6958 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6959 | * cpumask) fails, then fallback to a single sched domain, |
6960 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6961 | */ |
4212823f | 6962 | static cpumask_var_t fallback_doms; |
029190c5 | 6963 | |
ee79d1bd HC |
6964 | /* |
6965 | * arch_update_cpu_topology lets virtualized architectures update the | |
6966 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6967 | * or 0 if it stayed the same. | |
6968 | */ | |
52f5684c | 6969 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6970 | { |
ee79d1bd | 6971 | return 0; |
22e52b07 HC |
6972 | } |
6973 | ||
acc3f5d7 RR |
6974 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6975 | { | |
6976 | int i; | |
6977 | cpumask_var_t *doms; | |
6978 | ||
6979 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6980 | if (!doms) | |
6981 | return NULL; | |
6982 | for (i = 0; i < ndoms; i++) { | |
6983 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6984 | free_sched_domains(doms, i); | |
6985 | return NULL; | |
6986 | } | |
6987 | } | |
6988 | return doms; | |
6989 | } | |
6990 | ||
6991 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6992 | { | |
6993 | unsigned int i; | |
6994 | for (i = 0; i < ndoms; i++) | |
6995 | free_cpumask_var(doms[i]); | |
6996 | kfree(doms); | |
6997 | } | |
6998 | ||
1a20ff27 | 6999 | /* |
41a2d6cf | 7000 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7001 | * For now this just excludes isolated cpus, but could be used to |
7002 | * exclude other special cases in the future. | |
1a20ff27 | 7003 | */ |
c4a8849a | 7004 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7005 | { |
7378547f MM |
7006 | int err; |
7007 | ||
22e52b07 | 7008 | arch_update_cpu_topology(); |
029190c5 | 7009 | ndoms_cur = 1; |
acc3f5d7 | 7010 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7011 | if (!doms_cur) |
acc3f5d7 RR |
7012 | doms_cur = &fallback_doms; |
7013 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 7014 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7015 | register_sched_domain_sysctl(); |
7378547f MM |
7016 | |
7017 | return err; | |
1a20ff27 DG |
7018 | } |
7019 | ||
1a20ff27 DG |
7020 | /* |
7021 | * Detach sched domains from a group of cpus specified in cpu_map | |
7022 | * These cpus will now be attached to the NULL domain | |
7023 | */ | |
96f874e2 | 7024 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7025 | { |
7026 | int i; | |
7027 | ||
dce840a0 | 7028 | rcu_read_lock(); |
abcd083a | 7029 | for_each_cpu(i, cpu_map) |
57d885fe | 7030 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7031 | rcu_read_unlock(); |
1a20ff27 DG |
7032 | } |
7033 | ||
1d3504fc HS |
7034 | /* handle null as "default" */ |
7035 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7036 | struct sched_domain_attr *new, int idx_new) | |
7037 | { | |
7038 | struct sched_domain_attr tmp; | |
7039 | ||
7040 | /* fast path */ | |
7041 | if (!new && !cur) | |
7042 | return 1; | |
7043 | ||
7044 | tmp = SD_ATTR_INIT; | |
7045 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7046 | new ? (new + idx_new) : &tmp, | |
7047 | sizeof(struct sched_domain_attr)); | |
7048 | } | |
7049 | ||
029190c5 PJ |
7050 | /* |
7051 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7052 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7053 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7054 | * It destroys each deleted domain and builds each new domain. | |
7055 | * | |
acc3f5d7 | 7056 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7057 | * The masks don't intersect (don't overlap.) We should setup one |
7058 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7059 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7060 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7061 | * it as it is. | |
7062 | * | |
acc3f5d7 RR |
7063 | * The passed in 'doms_new' should be allocated using |
7064 | * alloc_sched_domains. This routine takes ownership of it and will | |
7065 | * free_sched_domains it when done with it. If the caller failed the | |
7066 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7067 | * and partition_sched_domains() will fallback to the single partition | |
7068 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7069 | * |
96f874e2 | 7070 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7071 | * ndoms_new == 0 is a special case for destroying existing domains, |
7072 | * and it will not create the default domain. | |
dfb512ec | 7073 | * |
029190c5 PJ |
7074 | * Call with hotplug lock held |
7075 | */ | |
acc3f5d7 | 7076 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7077 | struct sched_domain_attr *dattr_new) |
029190c5 | 7078 | { |
dfb512ec | 7079 | int i, j, n; |
d65bd5ec | 7080 | int new_topology; |
029190c5 | 7081 | |
712555ee | 7082 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7083 | |
7378547f MM |
7084 | /* always unregister in case we don't destroy any domains */ |
7085 | unregister_sched_domain_sysctl(); | |
7086 | ||
d65bd5ec HC |
7087 | /* Let architecture update cpu core mappings. */ |
7088 | new_topology = arch_update_cpu_topology(); | |
7089 | ||
dfb512ec | 7090 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7091 | |
7092 | /* Destroy deleted domains */ | |
7093 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7094 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7095 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7096 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7097 | goto match1; |
7098 | } | |
7099 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7100 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7101 | match1: |
7102 | ; | |
7103 | } | |
7104 | ||
c8d2d47a | 7105 | n = ndoms_cur; |
e761b772 | 7106 | if (doms_new == NULL) { |
c8d2d47a | 7107 | n = 0; |
acc3f5d7 | 7108 | doms_new = &fallback_doms; |
6ad4c188 | 7109 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7110 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7111 | } |
7112 | ||
029190c5 PJ |
7113 | /* Build new domains */ |
7114 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7115 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7116 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7117 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7118 | goto match2; |
7119 | } | |
7120 | /* no match - add a new doms_new */ | |
dce840a0 | 7121 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7122 | match2: |
7123 | ; | |
7124 | } | |
7125 | ||
7126 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7127 | if (doms_cur != &fallback_doms) |
7128 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7129 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7130 | doms_cur = doms_new; |
1d3504fc | 7131 | dattr_cur = dattr_new; |
029190c5 | 7132 | ndoms_cur = ndoms_new; |
7378547f MM |
7133 | |
7134 | register_sched_domain_sysctl(); | |
a1835615 | 7135 | |
712555ee | 7136 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7137 | } |
7138 | ||
d35be8ba SB |
7139 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7140 | ||
1da177e4 | 7141 | /* |
3a101d05 TH |
7142 | * Update cpusets according to cpu_active mask. If cpusets are |
7143 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7144 | * around partition_sched_domains(). | |
d35be8ba SB |
7145 | * |
7146 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7147 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7148 | */ |
0b2e918a TH |
7149 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7150 | void *hcpu) | |
e761b772 | 7151 | { |
d35be8ba SB |
7152 | switch (action) { |
7153 | case CPU_ONLINE_FROZEN: | |
7154 | case CPU_DOWN_FAILED_FROZEN: | |
7155 | ||
7156 | /* | |
7157 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7158 | * resume sequence. As long as this is not the last online | |
7159 | * operation in the resume sequence, just build a single sched | |
7160 | * domain, ignoring cpusets. | |
7161 | */ | |
7162 | num_cpus_frozen--; | |
7163 | if (likely(num_cpus_frozen)) { | |
7164 | partition_sched_domains(1, NULL, NULL); | |
7165 | break; | |
7166 | } | |
7167 | ||
7168 | /* | |
7169 | * This is the last CPU online operation. So fall through and | |
7170 | * restore the original sched domains by considering the | |
7171 | * cpuset configurations. | |
7172 | */ | |
7173 | ||
e761b772 | 7174 | case CPU_ONLINE: |
7ddf96b0 | 7175 | cpuset_update_active_cpus(true); |
d35be8ba | 7176 | break; |
3a101d05 TH |
7177 | default: |
7178 | return NOTIFY_DONE; | |
7179 | } | |
d35be8ba | 7180 | return NOTIFY_OK; |
3a101d05 | 7181 | } |
e761b772 | 7182 | |
0b2e918a TH |
7183 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7184 | void *hcpu) | |
3a101d05 | 7185 | { |
3c18d447 JL |
7186 | unsigned long flags; |
7187 | long cpu = (long)hcpu; | |
7188 | struct dl_bw *dl_b; | |
533445c6 OS |
7189 | bool overflow; |
7190 | int cpus; | |
3c18d447 | 7191 | |
533445c6 | 7192 | switch (action) { |
3a101d05 | 7193 | case CPU_DOWN_PREPARE: |
533445c6 OS |
7194 | rcu_read_lock_sched(); |
7195 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7196 | |
533445c6 OS |
7197 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7198 | cpus = dl_bw_cpus(cpu); | |
7199 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7200 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7201 | |
533445c6 | 7202 | rcu_read_unlock_sched(); |
3c18d447 | 7203 | |
533445c6 OS |
7204 | if (overflow) |
7205 | return notifier_from_errno(-EBUSY); | |
7ddf96b0 | 7206 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7207 | break; |
7208 | case CPU_DOWN_PREPARE_FROZEN: | |
7209 | num_cpus_frozen++; | |
7210 | partition_sched_domains(1, NULL, NULL); | |
7211 | break; | |
e761b772 MK |
7212 | default: |
7213 | return NOTIFY_DONE; | |
7214 | } | |
d35be8ba | 7215 | return NOTIFY_OK; |
e761b772 | 7216 | } |
e761b772 | 7217 | |
1da177e4 LT |
7218 | void __init sched_init_smp(void) |
7219 | { | |
dcc30a35 RR |
7220 | cpumask_var_t non_isolated_cpus; |
7221 | ||
7222 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7223 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7224 | |
cb83b629 PZ |
7225 | sched_init_numa(); |
7226 | ||
6acce3ef PZ |
7227 | /* |
7228 | * There's no userspace yet to cause hotplug operations; hence all the | |
7229 | * cpu masks are stable and all blatant races in the below code cannot | |
7230 | * happen. | |
7231 | */ | |
712555ee | 7232 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7233 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7234 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7235 | if (cpumask_empty(non_isolated_cpus)) | |
7236 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7237 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7238 | |
301a5cba | 7239 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7240 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7241 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7242 | |
b328ca18 | 7243 | init_hrtick(); |
5c1e1767 NP |
7244 | |
7245 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7246 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7247 | BUG(); |
19978ca6 | 7248 | sched_init_granularity(); |
dcc30a35 | 7249 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7250 | |
0e3900e6 | 7251 | init_sched_rt_class(); |
1baca4ce | 7252 | init_sched_dl_class(); |
1da177e4 LT |
7253 | } |
7254 | #else | |
7255 | void __init sched_init_smp(void) | |
7256 | { | |
19978ca6 | 7257 | sched_init_granularity(); |
1da177e4 LT |
7258 | } |
7259 | #endif /* CONFIG_SMP */ | |
7260 | ||
7261 | int in_sched_functions(unsigned long addr) | |
7262 | { | |
1da177e4 LT |
7263 | return in_lock_functions(addr) || |
7264 | (addr >= (unsigned long)__sched_text_start | |
7265 | && addr < (unsigned long)__sched_text_end); | |
7266 | } | |
7267 | ||
029632fb | 7268 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7269 | /* |
7270 | * Default task group. | |
7271 | * Every task in system belongs to this group at bootup. | |
7272 | */ | |
029632fb | 7273 | struct task_group root_task_group; |
35cf4e50 | 7274 | LIST_HEAD(task_groups); |
b0367629 WL |
7275 | |
7276 | /* Cacheline aligned slab cache for task_group */ | |
7277 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7278 | #endif |
6f505b16 | 7279 | |
e6252c3e | 7280 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7281 | |
1da177e4 LT |
7282 | void __init sched_init(void) |
7283 | { | |
dd41f596 | 7284 | int i, j; |
434d53b0 MT |
7285 | unsigned long alloc_size = 0, ptr; |
7286 | ||
7287 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7288 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7289 | #endif | |
7290 | #ifdef CONFIG_RT_GROUP_SCHED | |
7291 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7292 | #endif | |
434d53b0 | 7293 | if (alloc_size) { |
36b7b6d4 | 7294 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7295 | |
7296 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7297 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7298 | ptr += nr_cpu_ids * sizeof(void **); |
7299 | ||
07e06b01 | 7300 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7301 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7302 | |
6d6bc0ad | 7303 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7304 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7305 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7306 | ptr += nr_cpu_ids * sizeof(void **); |
7307 | ||
07e06b01 | 7308 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7309 | ptr += nr_cpu_ids * sizeof(void **); |
7310 | ||
6d6bc0ad | 7311 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7312 | } |
df7c8e84 | 7313 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7314 | for_each_possible_cpu(i) { |
7315 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7316 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7317 | } |
b74e6278 | 7318 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7319 | |
332ac17e DF |
7320 | init_rt_bandwidth(&def_rt_bandwidth, |
7321 | global_rt_period(), global_rt_runtime()); | |
7322 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7323 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7324 | |
57d885fe GH |
7325 | #ifdef CONFIG_SMP |
7326 | init_defrootdomain(); | |
7327 | #endif | |
7328 | ||
d0b27fa7 | 7329 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7330 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7331 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7332 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7333 | |
7c941438 | 7334 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7335 | task_group_cache = KMEM_CACHE(task_group, 0); |
7336 | ||
07e06b01 YZ |
7337 | list_add(&root_task_group.list, &task_groups); |
7338 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7339 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7340 | autogroup_init(&init_task); |
7c941438 | 7341 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7342 | |
0a945022 | 7343 | for_each_possible_cpu(i) { |
70b97a7f | 7344 | struct rq *rq; |
1da177e4 LT |
7345 | |
7346 | rq = cpu_rq(i); | |
05fa785c | 7347 | raw_spin_lock_init(&rq->lock); |
7897986b | 7348 | rq->nr_running = 0; |
dce48a84 TG |
7349 | rq->calc_load_active = 0; |
7350 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7351 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7352 | init_rt_rq(&rq->rt); |
7353 | init_dl_rq(&rq->dl); | |
dd41f596 | 7354 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7355 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7356 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7357 | /* |
07e06b01 | 7358 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7359 | * |
7360 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7361 | * gets 100% of the cpu resources in the system. This overall | |
7362 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7363 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7364 | * based on each entity's (task or task-group's) weight |
7365 | * (se->load.weight). | |
7366 | * | |
07e06b01 | 7367 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7368 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7369 | * then A0's share of the cpu resource is: | |
7370 | * | |
0d905bca | 7371 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7372 | * |
07e06b01 YZ |
7373 | * We achieve this by letting root_task_group's tasks sit |
7374 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7375 | */ |
ab84d31e | 7376 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7377 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7378 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7379 | ||
7380 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7381 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7382 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7383 | #endif |
1da177e4 | 7384 | |
dd41f596 IM |
7385 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7386 | rq->cpu_load[j] = 0; | |
fdf3e95d | 7387 | |
1da177e4 | 7388 | #ifdef CONFIG_SMP |
41c7ce9a | 7389 | rq->sd = NULL; |
57d885fe | 7390 | rq->rd = NULL; |
ca6d75e6 | 7391 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7392 | rq->balance_callback = NULL; |
1da177e4 | 7393 | rq->active_balance = 0; |
dd41f596 | 7394 | rq->next_balance = jiffies; |
1da177e4 | 7395 | rq->push_cpu = 0; |
0a2966b4 | 7396 | rq->cpu = i; |
1f11eb6a | 7397 | rq->online = 0; |
eae0c9df MG |
7398 | rq->idle_stamp = 0; |
7399 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7400 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7401 | |
7402 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7403 | ||
dc938520 | 7404 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7405 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 7406 | rq->last_load_update_tick = jiffies; |
1c792db7 | 7407 | rq->nohz_flags = 0; |
83cd4fe2 | 7408 | #endif |
265f22a9 FW |
7409 | #ifdef CONFIG_NO_HZ_FULL |
7410 | rq->last_sched_tick = 0; | |
7411 | #endif | |
9fd81dd5 | 7412 | #endif /* CONFIG_SMP */ |
8f4d37ec | 7413 | init_rq_hrtick(rq); |
1da177e4 | 7414 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7415 | } |
7416 | ||
2dd73a4f | 7417 | set_load_weight(&init_task); |
b50f60ce | 7418 | |
e107be36 AK |
7419 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7420 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7421 | #endif | |
7422 | ||
1da177e4 LT |
7423 | /* |
7424 | * The boot idle thread does lazy MMU switching as well: | |
7425 | */ | |
7426 | atomic_inc(&init_mm.mm_count); | |
7427 | enter_lazy_tlb(&init_mm, current); | |
7428 | ||
1b537c7d YD |
7429 | /* |
7430 | * During early bootup we pretend to be a normal task: | |
7431 | */ | |
7432 | current->sched_class = &fair_sched_class; | |
7433 | ||
1da177e4 LT |
7434 | /* |
7435 | * Make us the idle thread. Technically, schedule() should not be | |
7436 | * called from this thread, however somewhere below it might be, | |
7437 | * but because we are the idle thread, we just pick up running again | |
7438 | * when this runqueue becomes "idle". | |
7439 | */ | |
7440 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7441 | |
7442 | calc_load_update = jiffies + LOAD_FREQ; | |
7443 | ||
bf4d83f6 | 7444 | #ifdef CONFIG_SMP |
4cb98839 | 7445 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7446 | /* May be allocated at isolcpus cmdline parse time */ |
7447 | if (cpu_isolated_map == NULL) | |
7448 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7449 | idle_thread_set_boot_cpu(); |
a803f026 | 7450 | set_cpu_rq_start_time(); |
029632fb PZ |
7451 | #endif |
7452 | init_sched_fair_class(); | |
6a7b3dc3 | 7453 | |
6892b75e | 7454 | scheduler_running = 1; |
1da177e4 LT |
7455 | } |
7456 | ||
d902db1e | 7457 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7458 | static inline int preempt_count_equals(int preempt_offset) |
7459 | { | |
da7142e2 | 7460 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7461 | |
4ba8216c | 7462 | return (nested == preempt_offset); |
e4aafea2 FW |
7463 | } |
7464 | ||
d894837f | 7465 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7466 | { |
8eb23b9f PZ |
7467 | /* |
7468 | * Blocking primitives will set (and therefore destroy) current->state, | |
7469 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7470 | * otherwise we will destroy state. | |
7471 | */ | |
00845eb9 | 7472 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7473 | "do not call blocking ops when !TASK_RUNNING; " |
7474 | "state=%lx set at [<%p>] %pS\n", | |
7475 | current->state, | |
7476 | (void *)current->task_state_change, | |
00845eb9 | 7477 | (void *)current->task_state_change); |
8eb23b9f | 7478 | |
3427445a PZ |
7479 | ___might_sleep(file, line, preempt_offset); |
7480 | } | |
7481 | EXPORT_SYMBOL(__might_sleep); | |
7482 | ||
7483 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7484 | { |
1da177e4 LT |
7485 | static unsigned long prev_jiffy; /* ratelimiting */ |
7486 | ||
b3fbab05 | 7487 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7488 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7489 | !is_idle_task(current)) || | |
e4aafea2 | 7490 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7491 | return; |
7492 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7493 | return; | |
7494 | prev_jiffy = jiffies; | |
7495 | ||
3df0fc5b PZ |
7496 | printk(KERN_ERR |
7497 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7498 | file, line); | |
7499 | printk(KERN_ERR | |
7500 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7501 | in_atomic(), irqs_disabled(), | |
7502 | current->pid, current->comm); | |
aef745fc | 7503 | |
a8b686b3 ES |
7504 | if (task_stack_end_corrupted(current)) |
7505 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7506 | ||
aef745fc IM |
7507 | debug_show_held_locks(current); |
7508 | if (irqs_disabled()) | |
7509 | print_irqtrace_events(current); | |
8f47b187 TG |
7510 | #ifdef CONFIG_DEBUG_PREEMPT |
7511 | if (!preempt_count_equals(preempt_offset)) { | |
7512 | pr_err("Preemption disabled at:"); | |
7513 | print_ip_sym(current->preempt_disable_ip); | |
7514 | pr_cont("\n"); | |
7515 | } | |
7516 | #endif | |
aef745fc | 7517 | dump_stack(); |
1da177e4 | 7518 | } |
3427445a | 7519 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7520 | #endif |
7521 | ||
7522 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7523 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7524 | { |
dbc7f069 | 7525 | struct task_struct *g, *p; |
d50dde5a DF |
7526 | struct sched_attr attr = { |
7527 | .sched_policy = SCHED_NORMAL, | |
7528 | }; | |
1da177e4 | 7529 | |
3472eaa1 | 7530 | read_lock(&tasklist_lock); |
5d07f420 | 7531 | for_each_process_thread(g, p) { |
178be793 IM |
7532 | /* |
7533 | * Only normalize user tasks: | |
7534 | */ | |
3472eaa1 | 7535 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7536 | continue; |
7537 | ||
6cfb0d5d | 7538 | p->se.exec_start = 0; |
6cfb0d5d | 7539 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7540 | p->se.statistics.wait_start = 0; |
7541 | p->se.statistics.sleep_start = 0; | |
7542 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7543 | #endif |
dd41f596 | 7544 | |
aab03e05 | 7545 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7546 | /* |
7547 | * Renice negative nice level userspace | |
7548 | * tasks back to 0: | |
7549 | */ | |
3472eaa1 | 7550 | if (task_nice(p) < 0) |
dd41f596 | 7551 | set_user_nice(p, 0); |
1da177e4 | 7552 | continue; |
dd41f596 | 7553 | } |
1da177e4 | 7554 | |
dbc7f069 | 7555 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7556 | } |
3472eaa1 | 7557 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7558 | } |
7559 | ||
7560 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7561 | |
67fc4e0c | 7562 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7563 | /* |
67fc4e0c | 7564 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7565 | * |
7566 | * They can only be called when the whole system has been | |
7567 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7568 | * activity can take place. Using them for anything else would | |
7569 | * be a serious bug, and as a result, they aren't even visible | |
7570 | * under any other configuration. | |
7571 | */ | |
7572 | ||
7573 | /** | |
7574 | * curr_task - return the current task for a given cpu. | |
7575 | * @cpu: the processor in question. | |
7576 | * | |
7577 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7578 | * |
7579 | * Return: The current task for @cpu. | |
1df5c10a | 7580 | */ |
36c8b586 | 7581 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7582 | { |
7583 | return cpu_curr(cpu); | |
7584 | } | |
7585 | ||
67fc4e0c JW |
7586 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7587 | ||
7588 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7589 | /** |
7590 | * set_curr_task - set the current task for a given cpu. | |
7591 | * @cpu: the processor in question. | |
7592 | * @p: the task pointer to set. | |
7593 | * | |
7594 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7595 | * are serviced on a separate stack. It allows the architecture to switch the |
7596 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7597 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7598 | * and caller must save the original value of the current task (see | |
7599 | * curr_task() above) and restore that value before reenabling interrupts and | |
7600 | * re-starting the system. | |
7601 | * | |
7602 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7603 | */ | |
36c8b586 | 7604 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7605 | { |
7606 | cpu_curr(cpu) = p; | |
7607 | } | |
7608 | ||
7609 | #endif | |
29f59db3 | 7610 | |
7c941438 | 7611 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7612 | /* task_group_lock serializes the addition/removal of task groups */ |
7613 | static DEFINE_SPINLOCK(task_group_lock); | |
7614 | ||
2f5177f0 | 7615 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7616 | { |
7617 | free_fair_sched_group(tg); | |
7618 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7619 | autogroup_free(tg); |
b0367629 | 7620 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7621 | } |
7622 | ||
7623 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7624 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7625 | { |
7626 | struct task_group *tg; | |
bccbe08a | 7627 | |
b0367629 | 7628 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7629 | if (!tg) |
7630 | return ERR_PTR(-ENOMEM); | |
7631 | ||
ec7dc8ac | 7632 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7633 | goto err; |
7634 | ||
ec7dc8ac | 7635 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7636 | goto err; |
7637 | ||
ace783b9 LZ |
7638 | return tg; |
7639 | ||
7640 | err: | |
2f5177f0 | 7641 | sched_free_group(tg); |
ace783b9 LZ |
7642 | return ERR_PTR(-ENOMEM); |
7643 | } | |
7644 | ||
7645 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7646 | { | |
7647 | unsigned long flags; | |
7648 | ||
8ed36996 | 7649 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7650 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7651 | |
7652 | WARN_ON(!parent); /* root should already exist */ | |
7653 | ||
7654 | tg->parent = parent; | |
f473aa5e | 7655 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7656 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7657 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7658 | } |
7659 | ||
9b5b7751 | 7660 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7661 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7662 | { |
29f59db3 | 7663 | /* now it should be safe to free those cfs_rqs */ |
2f5177f0 | 7664 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7665 | } |
7666 | ||
4cf86d77 | 7667 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7668 | { |
7669 | /* wait for possible concurrent references to cfs_rqs complete */ | |
2f5177f0 | 7670 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7671 | } |
7672 | ||
7673 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7674 | { |
8ed36996 | 7675 | unsigned long flags; |
29f59db3 | 7676 | |
3d4b47b4 | 7677 | /* end participation in shares distribution */ |
6fe1f348 | 7678 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7679 | |
7680 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7681 | list_del_rcu(&tg->list); |
f473aa5e | 7682 | list_del_rcu(&tg->siblings); |
8ed36996 | 7683 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7684 | } |
7685 | ||
9b5b7751 | 7686 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7687 | * The caller of this function should have put the task in its new group |
7688 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7689 | * reflect its new group. | |
9b5b7751 SV |
7690 | */ |
7691 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7692 | { |
8323f26c | 7693 | struct task_group *tg; |
da0c1e65 | 7694 | int queued, running; |
29f59db3 SV |
7695 | unsigned long flags; |
7696 | struct rq *rq; | |
7697 | ||
7698 | rq = task_rq_lock(tsk, &flags); | |
7699 | ||
051a1d1a | 7700 | running = task_current(rq, tsk); |
da0c1e65 | 7701 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7702 | |
da0c1e65 | 7703 | if (queued) |
ff77e468 | 7704 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); |
0e1f3483 | 7705 | if (unlikely(running)) |
f3cd1c4e | 7706 | put_prev_task(rq, tsk); |
29f59db3 | 7707 | |
f7b8a47d KT |
7708 | /* |
7709 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7710 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7711 | * to prevent lockdep warnings. | |
7712 | */ | |
7713 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7714 | struct task_group, css); |
7715 | tg = autogroup_task_group(tsk, tg); | |
7716 | tsk->sched_task_group = tg; | |
7717 | ||
810b3817 | 7718 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7719 | if (tsk->sched_class->task_move_group) |
bc54da21 | 7720 | tsk->sched_class->task_move_group(tsk); |
b2b5ce02 | 7721 | else |
810b3817 | 7722 | #endif |
b2b5ce02 | 7723 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7724 | |
0e1f3483 HS |
7725 | if (unlikely(running)) |
7726 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7727 | if (queued) |
ff77e468 | 7728 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
29f59db3 | 7729 | |
0122ec5b | 7730 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7731 | } |
7c941438 | 7732 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7733 | |
a790de99 PT |
7734 | #ifdef CONFIG_RT_GROUP_SCHED |
7735 | /* | |
7736 | * Ensure that the real time constraints are schedulable. | |
7737 | */ | |
7738 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7739 | |
9a7e0b18 PZ |
7740 | /* Must be called with tasklist_lock held */ |
7741 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7742 | { |
9a7e0b18 | 7743 | struct task_struct *g, *p; |
b40b2e8e | 7744 | |
1fe89e1b PZ |
7745 | /* |
7746 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7747 | */ | |
7748 | if (task_group_is_autogroup(tg)) | |
7749 | return 0; | |
7750 | ||
5d07f420 | 7751 | for_each_process_thread(g, p) { |
8651c658 | 7752 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7753 | return 1; |
5d07f420 | 7754 | } |
b40b2e8e | 7755 | |
9a7e0b18 PZ |
7756 | return 0; |
7757 | } | |
b40b2e8e | 7758 | |
9a7e0b18 PZ |
7759 | struct rt_schedulable_data { |
7760 | struct task_group *tg; | |
7761 | u64 rt_period; | |
7762 | u64 rt_runtime; | |
7763 | }; | |
b40b2e8e | 7764 | |
a790de99 | 7765 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7766 | { |
7767 | struct rt_schedulable_data *d = data; | |
7768 | struct task_group *child; | |
7769 | unsigned long total, sum = 0; | |
7770 | u64 period, runtime; | |
b40b2e8e | 7771 | |
9a7e0b18 PZ |
7772 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7773 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7774 | |
9a7e0b18 PZ |
7775 | if (tg == d->tg) { |
7776 | period = d->rt_period; | |
7777 | runtime = d->rt_runtime; | |
b40b2e8e | 7778 | } |
b40b2e8e | 7779 | |
4653f803 PZ |
7780 | /* |
7781 | * Cannot have more runtime than the period. | |
7782 | */ | |
7783 | if (runtime > period && runtime != RUNTIME_INF) | |
7784 | return -EINVAL; | |
6f505b16 | 7785 | |
4653f803 PZ |
7786 | /* |
7787 | * Ensure we don't starve existing RT tasks. | |
7788 | */ | |
9a7e0b18 PZ |
7789 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7790 | return -EBUSY; | |
6f505b16 | 7791 | |
9a7e0b18 | 7792 | total = to_ratio(period, runtime); |
6f505b16 | 7793 | |
4653f803 PZ |
7794 | /* |
7795 | * Nobody can have more than the global setting allows. | |
7796 | */ | |
7797 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7798 | return -EINVAL; | |
6f505b16 | 7799 | |
4653f803 PZ |
7800 | /* |
7801 | * The sum of our children's runtime should not exceed our own. | |
7802 | */ | |
9a7e0b18 PZ |
7803 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7804 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7805 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7806 | |
9a7e0b18 PZ |
7807 | if (child == d->tg) { |
7808 | period = d->rt_period; | |
7809 | runtime = d->rt_runtime; | |
7810 | } | |
6f505b16 | 7811 | |
9a7e0b18 | 7812 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7813 | } |
6f505b16 | 7814 | |
9a7e0b18 PZ |
7815 | if (sum > total) |
7816 | return -EINVAL; | |
7817 | ||
7818 | return 0; | |
6f505b16 PZ |
7819 | } |
7820 | ||
9a7e0b18 | 7821 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7822 | { |
8277434e PT |
7823 | int ret; |
7824 | ||
9a7e0b18 PZ |
7825 | struct rt_schedulable_data data = { |
7826 | .tg = tg, | |
7827 | .rt_period = period, | |
7828 | .rt_runtime = runtime, | |
7829 | }; | |
7830 | ||
8277434e PT |
7831 | rcu_read_lock(); |
7832 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7833 | rcu_read_unlock(); | |
7834 | ||
7835 | return ret; | |
521f1a24 DG |
7836 | } |
7837 | ||
ab84d31e | 7838 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7839 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7840 | { |
ac086bc2 | 7841 | int i, err = 0; |
9f0c1e56 | 7842 | |
2636ed5f PZ |
7843 | /* |
7844 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7845 | * kernel creating (and or operating) RT threads. | |
7846 | */ | |
7847 | if (tg == &root_task_group && rt_runtime == 0) | |
7848 | return -EINVAL; | |
7849 | ||
7850 | /* No period doesn't make any sense. */ | |
7851 | if (rt_period == 0) | |
7852 | return -EINVAL; | |
7853 | ||
9f0c1e56 | 7854 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7855 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7856 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7857 | if (err) | |
9f0c1e56 | 7858 | goto unlock; |
ac086bc2 | 7859 | |
0986b11b | 7860 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7861 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7862 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7863 | |
7864 | for_each_possible_cpu(i) { | |
7865 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7866 | ||
0986b11b | 7867 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7868 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7869 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7870 | } |
0986b11b | 7871 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7872 | unlock: |
521f1a24 | 7873 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7874 | mutex_unlock(&rt_constraints_mutex); |
7875 | ||
7876 | return err; | |
6f505b16 PZ |
7877 | } |
7878 | ||
25cc7da7 | 7879 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7880 | { |
7881 | u64 rt_runtime, rt_period; | |
7882 | ||
7883 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7884 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7885 | if (rt_runtime_us < 0) | |
7886 | rt_runtime = RUNTIME_INF; | |
7887 | ||
ab84d31e | 7888 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7889 | } |
7890 | ||
25cc7da7 | 7891 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7892 | { |
7893 | u64 rt_runtime_us; | |
7894 | ||
d0b27fa7 | 7895 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7896 | return -1; |
7897 | ||
d0b27fa7 | 7898 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7899 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7900 | return rt_runtime_us; | |
7901 | } | |
d0b27fa7 | 7902 | |
ce2f5fe4 | 7903 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7904 | { |
7905 | u64 rt_runtime, rt_period; | |
7906 | ||
ce2f5fe4 | 7907 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7908 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7909 | ||
ab84d31e | 7910 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7911 | } |
7912 | ||
25cc7da7 | 7913 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7914 | { |
7915 | u64 rt_period_us; | |
7916 | ||
7917 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7918 | do_div(rt_period_us, NSEC_PER_USEC); | |
7919 | return rt_period_us; | |
7920 | } | |
332ac17e | 7921 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7922 | |
332ac17e | 7923 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7924 | static int sched_rt_global_constraints(void) |
7925 | { | |
7926 | int ret = 0; | |
7927 | ||
7928 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7929 | read_lock(&tasklist_lock); |
4653f803 | 7930 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7931 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7932 | mutex_unlock(&rt_constraints_mutex); |
7933 | ||
7934 | return ret; | |
7935 | } | |
54e99124 | 7936 | |
25cc7da7 | 7937 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7938 | { |
7939 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7940 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7941 | return 0; | |
7942 | ||
7943 | return 1; | |
7944 | } | |
7945 | ||
6d6bc0ad | 7946 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7947 | static int sched_rt_global_constraints(void) |
7948 | { | |
ac086bc2 | 7949 | unsigned long flags; |
332ac17e | 7950 | int i, ret = 0; |
ec5d4989 | 7951 | |
0986b11b | 7952 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7953 | for_each_possible_cpu(i) { |
7954 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7955 | ||
0986b11b | 7956 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7957 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7958 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7959 | } |
0986b11b | 7960 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7961 | |
332ac17e | 7962 | return ret; |
d0b27fa7 | 7963 | } |
6d6bc0ad | 7964 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7965 | |
a1963b81 | 7966 | static int sched_dl_global_validate(void) |
332ac17e | 7967 | { |
1724813d PZ |
7968 | u64 runtime = global_rt_runtime(); |
7969 | u64 period = global_rt_period(); | |
332ac17e | 7970 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7971 | struct dl_bw *dl_b; |
1724813d | 7972 | int cpu, ret = 0; |
49516342 | 7973 | unsigned long flags; |
332ac17e DF |
7974 | |
7975 | /* | |
7976 | * Here we want to check the bandwidth not being set to some | |
7977 | * value smaller than the currently allocated bandwidth in | |
7978 | * any of the root_domains. | |
7979 | * | |
7980 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7981 | * cycling on root_domains... Discussion on different/better | |
7982 | * solutions is welcome! | |
7983 | */ | |
1724813d | 7984 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7985 | rcu_read_lock_sched(); |
7986 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7987 | |
49516342 | 7988 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7989 | if (new_bw < dl_b->total_bw) |
7990 | ret = -EBUSY; | |
49516342 | 7991 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7992 | |
f10e00f4 KT |
7993 | rcu_read_unlock_sched(); |
7994 | ||
1724813d PZ |
7995 | if (ret) |
7996 | break; | |
332ac17e DF |
7997 | } |
7998 | ||
1724813d | 7999 | return ret; |
332ac17e DF |
8000 | } |
8001 | ||
1724813d | 8002 | static void sched_dl_do_global(void) |
ce0dbbbb | 8003 | { |
1724813d | 8004 | u64 new_bw = -1; |
f10e00f4 | 8005 | struct dl_bw *dl_b; |
1724813d | 8006 | int cpu; |
49516342 | 8007 | unsigned long flags; |
ce0dbbbb | 8008 | |
1724813d PZ |
8009 | def_dl_bandwidth.dl_period = global_rt_period(); |
8010 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
8011 | ||
8012 | if (global_rt_runtime() != RUNTIME_INF) | |
8013 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
8014 | ||
8015 | /* | |
8016 | * FIXME: As above... | |
8017 | */ | |
8018 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
8019 | rcu_read_lock_sched(); |
8020 | dl_b = dl_bw_of(cpu); | |
1724813d | 8021 | |
49516342 | 8022 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8023 | dl_b->bw = new_bw; |
49516342 | 8024 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8025 | |
8026 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8027 | } |
1724813d PZ |
8028 | } |
8029 | ||
8030 | static int sched_rt_global_validate(void) | |
8031 | { | |
8032 | if (sysctl_sched_rt_period <= 0) | |
8033 | return -EINVAL; | |
8034 | ||
e9e7cb38 JL |
8035 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8036 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8037 | return -EINVAL; |
8038 | ||
8039 | return 0; | |
8040 | } | |
8041 | ||
8042 | static void sched_rt_do_global(void) | |
8043 | { | |
8044 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8045 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8046 | } |
8047 | ||
d0b27fa7 | 8048 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8049 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8050 | loff_t *ppos) |
8051 | { | |
d0b27fa7 PZ |
8052 | int old_period, old_runtime; |
8053 | static DEFINE_MUTEX(mutex); | |
1724813d | 8054 | int ret; |
d0b27fa7 PZ |
8055 | |
8056 | mutex_lock(&mutex); | |
8057 | old_period = sysctl_sched_rt_period; | |
8058 | old_runtime = sysctl_sched_rt_runtime; | |
8059 | ||
8d65af78 | 8060 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8061 | |
8062 | if (!ret && write) { | |
1724813d PZ |
8063 | ret = sched_rt_global_validate(); |
8064 | if (ret) | |
8065 | goto undo; | |
8066 | ||
a1963b81 | 8067 | ret = sched_dl_global_validate(); |
1724813d PZ |
8068 | if (ret) |
8069 | goto undo; | |
8070 | ||
a1963b81 | 8071 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8072 | if (ret) |
8073 | goto undo; | |
8074 | ||
8075 | sched_rt_do_global(); | |
8076 | sched_dl_do_global(); | |
8077 | } | |
8078 | if (0) { | |
8079 | undo: | |
8080 | sysctl_sched_rt_period = old_period; | |
8081 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8082 | } |
8083 | mutex_unlock(&mutex); | |
8084 | ||
8085 | return ret; | |
8086 | } | |
68318b8e | 8087 | |
1724813d | 8088 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8089 | void __user *buffer, size_t *lenp, |
8090 | loff_t *ppos) | |
8091 | { | |
8092 | int ret; | |
332ac17e | 8093 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8094 | |
8095 | mutex_lock(&mutex); | |
332ac17e | 8096 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8097 | /* make sure that internally we keep jiffies */ |
8098 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8099 | if (!ret && write) { |
1724813d PZ |
8100 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8101 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8102 | } |
8103 | mutex_unlock(&mutex); | |
332ac17e DF |
8104 | return ret; |
8105 | } | |
8106 | ||
052f1dc7 | 8107 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8108 | |
a7c6d554 | 8109 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8110 | { |
a7c6d554 | 8111 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8112 | } |
8113 | ||
eb95419b TH |
8114 | static struct cgroup_subsys_state * |
8115 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8116 | { |
eb95419b TH |
8117 | struct task_group *parent = css_tg(parent_css); |
8118 | struct task_group *tg; | |
68318b8e | 8119 | |
eb95419b | 8120 | if (!parent) { |
68318b8e | 8121 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8122 | return &root_task_group.css; |
68318b8e SV |
8123 | } |
8124 | ||
ec7dc8ac | 8125 | tg = sched_create_group(parent); |
68318b8e SV |
8126 | if (IS_ERR(tg)) |
8127 | return ERR_PTR(-ENOMEM); | |
8128 | ||
2f5177f0 PZ |
8129 | sched_online_group(tg, parent); |
8130 | ||
68318b8e SV |
8131 | return &tg->css; |
8132 | } | |
8133 | ||
2f5177f0 | 8134 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8135 | { |
eb95419b | 8136 | struct task_group *tg = css_tg(css); |
ace783b9 | 8137 | |
2f5177f0 | 8138 | sched_offline_group(tg); |
ace783b9 LZ |
8139 | } |
8140 | ||
eb95419b | 8141 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8142 | { |
eb95419b | 8143 | struct task_group *tg = css_tg(css); |
68318b8e | 8144 | |
2f5177f0 PZ |
8145 | /* |
8146 | * Relies on the RCU grace period between css_released() and this. | |
8147 | */ | |
8148 | sched_free_group(tg); | |
ace783b9 LZ |
8149 | } |
8150 | ||
b53202e6 | 8151 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 KT |
8152 | { |
8153 | sched_move_task(task); | |
8154 | } | |
8155 | ||
1f7dd3e5 | 8156 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8157 | { |
bb9d97b6 | 8158 | struct task_struct *task; |
1f7dd3e5 | 8159 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8160 | |
1f7dd3e5 | 8161 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8162 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8163 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8164 | return -EINVAL; |
b68aa230 | 8165 | #else |
bb9d97b6 TH |
8166 | /* We don't support RT-tasks being in separate groups */ |
8167 | if (task->sched_class != &fair_sched_class) | |
8168 | return -EINVAL; | |
b68aa230 | 8169 | #endif |
bb9d97b6 | 8170 | } |
be367d09 BB |
8171 | return 0; |
8172 | } | |
68318b8e | 8173 | |
1f7dd3e5 | 8174 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8175 | { |
bb9d97b6 | 8176 | struct task_struct *task; |
1f7dd3e5 | 8177 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8178 | |
1f7dd3e5 | 8179 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8180 | sched_move_task(task); |
68318b8e SV |
8181 | } |
8182 | ||
052f1dc7 | 8183 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8184 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8185 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8186 | { |
182446d0 | 8187 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8188 | } |
8189 | ||
182446d0 TH |
8190 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8191 | struct cftype *cft) | |
68318b8e | 8192 | { |
182446d0 | 8193 | struct task_group *tg = css_tg(css); |
68318b8e | 8194 | |
c8b28116 | 8195 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8196 | } |
ab84d31e PT |
8197 | |
8198 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8199 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8200 | ||
ab84d31e PT |
8201 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8202 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8203 | ||
a790de99 PT |
8204 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8205 | ||
ab84d31e PT |
8206 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8207 | { | |
56f570e5 | 8208 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8209 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8210 | |
8211 | if (tg == &root_task_group) | |
8212 | return -EINVAL; | |
8213 | ||
8214 | /* | |
8215 | * Ensure we have at some amount of bandwidth every period. This is | |
8216 | * to prevent reaching a state of large arrears when throttled via | |
8217 | * entity_tick() resulting in prolonged exit starvation. | |
8218 | */ | |
8219 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8220 | return -EINVAL; | |
8221 | ||
8222 | /* | |
8223 | * Likewise, bound things on the otherside by preventing insane quota | |
8224 | * periods. This also allows us to normalize in computing quota | |
8225 | * feasibility. | |
8226 | */ | |
8227 | if (period > max_cfs_quota_period) | |
8228 | return -EINVAL; | |
8229 | ||
0e59bdae KT |
8230 | /* |
8231 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8232 | * unthrottle_offline_cfs_rqs(). | |
8233 | */ | |
8234 | get_online_cpus(); | |
a790de99 PT |
8235 | mutex_lock(&cfs_constraints_mutex); |
8236 | ret = __cfs_schedulable(tg, period, quota); | |
8237 | if (ret) | |
8238 | goto out_unlock; | |
8239 | ||
58088ad0 | 8240 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8241 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8242 | /* |
8243 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8244 | * before making related changes, and on->off must occur afterwards | |
8245 | */ | |
8246 | if (runtime_enabled && !runtime_was_enabled) | |
8247 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8248 | raw_spin_lock_irq(&cfs_b->lock); |
8249 | cfs_b->period = ns_to_ktime(period); | |
8250 | cfs_b->quota = quota; | |
58088ad0 | 8251 | |
a9cf55b2 | 8252 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8253 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8254 | if (runtime_enabled) |
8255 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8256 | raw_spin_unlock_irq(&cfs_b->lock); |
8257 | ||
0e59bdae | 8258 | for_each_online_cpu(i) { |
ab84d31e | 8259 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8260 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8261 | |
8262 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8263 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8264 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8265 | |
029632fb | 8266 | if (cfs_rq->throttled) |
671fd9da | 8267 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8268 | raw_spin_unlock_irq(&rq->lock); |
8269 | } | |
1ee14e6c BS |
8270 | if (runtime_was_enabled && !runtime_enabled) |
8271 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8272 | out_unlock: |
8273 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8274 | put_online_cpus(); |
ab84d31e | 8275 | |
a790de99 | 8276 | return ret; |
ab84d31e PT |
8277 | } |
8278 | ||
8279 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8280 | { | |
8281 | u64 quota, period; | |
8282 | ||
029632fb | 8283 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8284 | if (cfs_quota_us < 0) |
8285 | quota = RUNTIME_INF; | |
8286 | else | |
8287 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8288 | ||
8289 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8290 | } | |
8291 | ||
8292 | long tg_get_cfs_quota(struct task_group *tg) | |
8293 | { | |
8294 | u64 quota_us; | |
8295 | ||
029632fb | 8296 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8297 | return -1; |
8298 | ||
029632fb | 8299 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8300 | do_div(quota_us, NSEC_PER_USEC); |
8301 | ||
8302 | return quota_us; | |
8303 | } | |
8304 | ||
8305 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8306 | { | |
8307 | u64 quota, period; | |
8308 | ||
8309 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8310 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8311 | |
ab84d31e PT |
8312 | return tg_set_cfs_bandwidth(tg, period, quota); |
8313 | } | |
8314 | ||
8315 | long tg_get_cfs_period(struct task_group *tg) | |
8316 | { | |
8317 | u64 cfs_period_us; | |
8318 | ||
029632fb | 8319 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8320 | do_div(cfs_period_us, NSEC_PER_USEC); |
8321 | ||
8322 | return cfs_period_us; | |
8323 | } | |
8324 | ||
182446d0 TH |
8325 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8326 | struct cftype *cft) | |
ab84d31e | 8327 | { |
182446d0 | 8328 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8329 | } |
8330 | ||
182446d0 TH |
8331 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8332 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8333 | { |
182446d0 | 8334 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8335 | } |
8336 | ||
182446d0 TH |
8337 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8338 | struct cftype *cft) | |
ab84d31e | 8339 | { |
182446d0 | 8340 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8341 | } |
8342 | ||
182446d0 TH |
8343 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8344 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8345 | { |
182446d0 | 8346 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8347 | } |
8348 | ||
a790de99 PT |
8349 | struct cfs_schedulable_data { |
8350 | struct task_group *tg; | |
8351 | u64 period, quota; | |
8352 | }; | |
8353 | ||
8354 | /* | |
8355 | * normalize group quota/period to be quota/max_period | |
8356 | * note: units are usecs | |
8357 | */ | |
8358 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8359 | struct cfs_schedulable_data *d) | |
8360 | { | |
8361 | u64 quota, period; | |
8362 | ||
8363 | if (tg == d->tg) { | |
8364 | period = d->period; | |
8365 | quota = d->quota; | |
8366 | } else { | |
8367 | period = tg_get_cfs_period(tg); | |
8368 | quota = tg_get_cfs_quota(tg); | |
8369 | } | |
8370 | ||
8371 | /* note: these should typically be equivalent */ | |
8372 | if (quota == RUNTIME_INF || quota == -1) | |
8373 | return RUNTIME_INF; | |
8374 | ||
8375 | return to_ratio(period, quota); | |
8376 | } | |
8377 | ||
8378 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8379 | { | |
8380 | struct cfs_schedulable_data *d = data; | |
029632fb | 8381 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8382 | s64 quota = 0, parent_quota = -1; |
8383 | ||
8384 | if (!tg->parent) { | |
8385 | quota = RUNTIME_INF; | |
8386 | } else { | |
029632fb | 8387 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8388 | |
8389 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8390 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8391 | |
8392 | /* | |
8393 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8394 | * limit is set | |
8395 | */ | |
8396 | if (quota == RUNTIME_INF) | |
8397 | quota = parent_quota; | |
8398 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8399 | return -EINVAL; | |
8400 | } | |
9c58c79a | 8401 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8402 | |
8403 | return 0; | |
8404 | } | |
8405 | ||
8406 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8407 | { | |
8277434e | 8408 | int ret; |
a790de99 PT |
8409 | struct cfs_schedulable_data data = { |
8410 | .tg = tg, | |
8411 | .period = period, | |
8412 | .quota = quota, | |
8413 | }; | |
8414 | ||
8415 | if (quota != RUNTIME_INF) { | |
8416 | do_div(data.period, NSEC_PER_USEC); | |
8417 | do_div(data.quota, NSEC_PER_USEC); | |
8418 | } | |
8419 | ||
8277434e PT |
8420 | rcu_read_lock(); |
8421 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8422 | rcu_read_unlock(); | |
8423 | ||
8424 | return ret; | |
a790de99 | 8425 | } |
e8da1b18 | 8426 | |
2da8ca82 | 8427 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8428 | { |
2da8ca82 | 8429 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8430 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8431 | |
44ffc75b TH |
8432 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8433 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8434 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8435 | |
8436 | return 0; | |
8437 | } | |
ab84d31e | 8438 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8439 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8440 | |
052f1dc7 | 8441 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8442 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8443 | struct cftype *cft, s64 val) | |
6f505b16 | 8444 | { |
182446d0 | 8445 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8446 | } |
8447 | ||
182446d0 TH |
8448 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8449 | struct cftype *cft) | |
6f505b16 | 8450 | { |
182446d0 | 8451 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8452 | } |
d0b27fa7 | 8453 | |
182446d0 TH |
8454 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8455 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8456 | { |
182446d0 | 8457 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8458 | } |
8459 | ||
182446d0 TH |
8460 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8461 | struct cftype *cft) | |
d0b27fa7 | 8462 | { |
182446d0 | 8463 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8464 | } |
6d6bc0ad | 8465 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8466 | |
fe5c7cc2 | 8467 | static struct cftype cpu_files[] = { |
052f1dc7 | 8468 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8469 | { |
8470 | .name = "shares", | |
f4c753b7 PM |
8471 | .read_u64 = cpu_shares_read_u64, |
8472 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8473 | }, |
052f1dc7 | 8474 | #endif |
ab84d31e PT |
8475 | #ifdef CONFIG_CFS_BANDWIDTH |
8476 | { | |
8477 | .name = "cfs_quota_us", | |
8478 | .read_s64 = cpu_cfs_quota_read_s64, | |
8479 | .write_s64 = cpu_cfs_quota_write_s64, | |
8480 | }, | |
8481 | { | |
8482 | .name = "cfs_period_us", | |
8483 | .read_u64 = cpu_cfs_period_read_u64, | |
8484 | .write_u64 = cpu_cfs_period_write_u64, | |
8485 | }, | |
e8da1b18 NR |
8486 | { |
8487 | .name = "stat", | |
2da8ca82 | 8488 | .seq_show = cpu_stats_show, |
e8da1b18 | 8489 | }, |
ab84d31e | 8490 | #endif |
052f1dc7 | 8491 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8492 | { |
9f0c1e56 | 8493 | .name = "rt_runtime_us", |
06ecb27c PM |
8494 | .read_s64 = cpu_rt_runtime_read, |
8495 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8496 | }, |
d0b27fa7 PZ |
8497 | { |
8498 | .name = "rt_period_us", | |
f4c753b7 PM |
8499 | .read_u64 = cpu_rt_period_read_uint, |
8500 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8501 | }, |
052f1dc7 | 8502 | #endif |
4baf6e33 | 8503 | { } /* terminate */ |
68318b8e SV |
8504 | }; |
8505 | ||
073219e9 | 8506 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8507 | .css_alloc = cpu_cgroup_css_alloc, |
2f5177f0 | 8508 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8509 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 8510 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8511 | .can_attach = cpu_cgroup_can_attach, |
8512 | .attach = cpu_cgroup_attach, | |
5577964e | 8513 | .legacy_cftypes = cpu_files, |
b38e42e9 | 8514 | .early_init = true, |
68318b8e SV |
8515 | }; |
8516 | ||
052f1dc7 | 8517 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8518 | |
b637a328 PM |
8519 | void dump_cpu_task(int cpu) |
8520 | { | |
8521 | pr_info("Task dump for CPU %d:\n", cpu); | |
8522 | sched_show_task(cpu_curr(cpu)); | |
8523 | } | |
ed82b8a1 AK |
8524 | |
8525 | /* | |
8526 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8527 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8528 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8529 | * that remained on nice 0. | |
8530 | * | |
8531 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8532 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8533 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8534 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8535 | * the relative distance between them is ~25%.) | |
8536 | */ | |
8537 | const int sched_prio_to_weight[40] = { | |
8538 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8539 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8540 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8541 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8542 | /* 0 */ 1024, 820, 655, 526, 423, | |
8543 | /* 5 */ 335, 272, 215, 172, 137, | |
8544 | /* 10 */ 110, 87, 70, 56, 45, | |
8545 | /* 15 */ 36, 29, 23, 18, 15, | |
8546 | }; | |
8547 | ||
8548 | /* | |
8549 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8550 | * | |
8551 | * In cases where the weight does not change often, we can use the | |
8552 | * precalculated inverse to speed up arithmetics by turning divisions | |
8553 | * into multiplications: | |
8554 | */ | |
8555 | const u32 sched_prio_to_wmult[40] = { | |
8556 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8557 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8558 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8559 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8560 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8561 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8562 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8563 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8564 | }; |