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