]>
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 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
52f5684c | 76 | #include <linux/compiler.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 | ||
127 | #ifdef CONFIG_SCHED_DEBUG | |
128 | #define SCHED_FEAT(name, enabled) \ | |
129 | #name , | |
130 | ||
1292531f | 131 | static const char * const sched_feat_names[] = { |
391e43da | 132 | #include "features.h" |
f00b45c1 PZ |
133 | }; |
134 | ||
135 | #undef SCHED_FEAT | |
136 | ||
34f3a814 | 137 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 138 | { |
f00b45c1 PZ |
139 | int i; |
140 | ||
f8b6d1cc | 141 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
142 | if (!(sysctl_sched_features & (1UL << i))) |
143 | seq_puts(m, "NO_"); | |
144 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 145 | } |
34f3a814 | 146 | seq_puts(m, "\n"); |
f00b45c1 | 147 | |
34f3a814 | 148 | return 0; |
f00b45c1 PZ |
149 | } |
150 | ||
f8b6d1cc PZ |
151 | #ifdef HAVE_JUMP_LABEL |
152 | ||
c5905afb IM |
153 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
154 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
155 | |
156 | #define SCHED_FEAT(name, enabled) \ | |
157 | jump_label_key__##enabled , | |
158 | ||
c5905afb | 159 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
160 | #include "features.h" |
161 | }; | |
162 | ||
163 | #undef SCHED_FEAT | |
164 | ||
165 | static void sched_feat_disable(int i) | |
166 | { | |
e33886b3 | 167 | static_key_disable(&sched_feat_keys[i]); |
f8b6d1cc PZ |
168 | } |
169 | ||
170 | static void sched_feat_enable(int i) | |
171 | { | |
e33886b3 | 172 | static_key_enable(&sched_feat_keys[i]); |
f8b6d1cc PZ |
173 | } |
174 | #else | |
175 | static void sched_feat_disable(int i) { }; | |
176 | static void sched_feat_enable(int i) { }; | |
177 | #endif /* HAVE_JUMP_LABEL */ | |
178 | ||
1a687c2e | 179 | static int sched_feat_set(char *cmp) |
f00b45c1 | 180 | { |
f00b45c1 | 181 | int i; |
1a687c2e | 182 | int neg = 0; |
f00b45c1 | 183 | |
524429c3 | 184 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
185 | neg = 1; |
186 | cmp += 3; | |
187 | } | |
188 | ||
f8b6d1cc | 189 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 190 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 191 | if (neg) { |
f00b45c1 | 192 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
193 | sched_feat_disable(i); |
194 | } else { | |
f00b45c1 | 195 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
196 | sched_feat_enable(i); |
197 | } | |
f00b45c1 PZ |
198 | break; |
199 | } | |
200 | } | |
201 | ||
1a687c2e MG |
202 | return i; |
203 | } | |
204 | ||
205 | static ssize_t | |
206 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
207 | size_t cnt, loff_t *ppos) | |
208 | { | |
209 | char buf[64]; | |
210 | char *cmp; | |
211 | int i; | |
5cd08fbf | 212 | struct inode *inode; |
1a687c2e MG |
213 | |
214 | if (cnt > 63) | |
215 | cnt = 63; | |
216 | ||
217 | if (copy_from_user(&buf, ubuf, cnt)) | |
218 | return -EFAULT; | |
219 | ||
220 | buf[cnt] = 0; | |
221 | cmp = strstrip(buf); | |
222 | ||
5cd08fbf JB |
223 | /* Ensure the static_key remains in a consistent state */ |
224 | inode = file_inode(filp); | |
225 | mutex_lock(&inode->i_mutex); | |
1a687c2e | 226 | i = sched_feat_set(cmp); |
5cd08fbf | 227 | mutex_unlock(&inode->i_mutex); |
f8b6d1cc | 228 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
229 | return -EINVAL; |
230 | ||
42994724 | 231 | *ppos += cnt; |
f00b45c1 PZ |
232 | |
233 | return cnt; | |
234 | } | |
235 | ||
34f3a814 LZ |
236 | static int sched_feat_open(struct inode *inode, struct file *filp) |
237 | { | |
238 | return single_open(filp, sched_feat_show, NULL); | |
239 | } | |
240 | ||
828c0950 | 241 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
242 | .open = sched_feat_open, |
243 | .write = sched_feat_write, | |
244 | .read = seq_read, | |
245 | .llseek = seq_lseek, | |
246 | .release = single_release, | |
f00b45c1 PZ |
247 | }; |
248 | ||
249 | static __init int sched_init_debug(void) | |
250 | { | |
f00b45c1 PZ |
251 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
252 | &sched_feat_fops); | |
253 | ||
254 | return 0; | |
255 | } | |
256 | late_initcall(sched_init_debug); | |
f8b6d1cc | 257 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 258 | |
b82d9fdd PZ |
259 | /* |
260 | * Number of tasks to iterate in a single balance run. | |
261 | * Limited because this is done with IRQs disabled. | |
262 | */ | |
263 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
264 | ||
e9e9250b PZ |
265 | /* |
266 | * period over which we average the RT time consumption, measured | |
267 | * in ms. | |
268 | * | |
269 | * default: 1s | |
270 | */ | |
271 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
272 | ||
fa85ae24 | 273 | /* |
9f0c1e56 | 274 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
275 | * default: 1s |
276 | */ | |
9f0c1e56 | 277 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 278 | |
029632fb | 279 | __read_mostly int scheduler_running; |
6892b75e | 280 | |
9f0c1e56 PZ |
281 | /* |
282 | * part of the period that we allow rt tasks to run in us. | |
283 | * default: 0.95s | |
284 | */ | |
285 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 286 | |
3fa0818b RR |
287 | /* cpus with isolated domains */ |
288 | cpumask_var_t cpu_isolated_map; | |
289 | ||
1da177e4 | 290 | /* |
cc2a73b5 | 291 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 292 | */ |
a9957449 | 293 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
294 | __acquires(rq->lock) |
295 | { | |
70b97a7f | 296 | struct rq *rq; |
1da177e4 LT |
297 | |
298 | local_irq_disable(); | |
299 | rq = this_rq(); | |
05fa785c | 300 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
301 | |
302 | return rq; | |
303 | } | |
304 | ||
8f4d37ec PZ |
305 | #ifdef CONFIG_SCHED_HRTICK |
306 | /* | |
307 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 308 | */ |
8f4d37ec | 309 | |
8f4d37ec PZ |
310 | static void hrtick_clear(struct rq *rq) |
311 | { | |
312 | if (hrtimer_active(&rq->hrtick_timer)) | |
313 | hrtimer_cancel(&rq->hrtick_timer); | |
314 | } | |
315 | ||
8f4d37ec PZ |
316 | /* |
317 | * High-resolution timer tick. | |
318 | * Runs from hardirq context with interrupts disabled. | |
319 | */ | |
320 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
321 | { | |
322 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
323 | ||
324 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
325 | ||
05fa785c | 326 | raw_spin_lock(&rq->lock); |
3e51f33f | 327 | update_rq_clock(rq); |
8f4d37ec | 328 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 329 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
330 | |
331 | return HRTIMER_NORESTART; | |
332 | } | |
333 | ||
95e904c7 | 334 | #ifdef CONFIG_SMP |
971ee28c | 335 | |
4961b6e1 | 336 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
337 | { |
338 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 339 | |
4961b6e1 | 340 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
341 | } |
342 | ||
31656519 PZ |
343 | /* |
344 | * called from hardirq (IPI) context | |
345 | */ | |
346 | static void __hrtick_start(void *arg) | |
b328ca18 | 347 | { |
31656519 | 348 | struct rq *rq = arg; |
b328ca18 | 349 | |
05fa785c | 350 | raw_spin_lock(&rq->lock); |
971ee28c | 351 | __hrtick_restart(rq); |
31656519 | 352 | rq->hrtick_csd_pending = 0; |
05fa785c | 353 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
354 | } |
355 | ||
31656519 PZ |
356 | /* |
357 | * Called to set the hrtick timer state. | |
358 | * | |
359 | * called with rq->lock held and irqs disabled | |
360 | */ | |
029632fb | 361 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 362 | { |
31656519 | 363 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 364 | ktime_t time; |
365 | s64 delta; | |
366 | ||
367 | /* | |
368 | * Don't schedule slices shorter than 10000ns, that just | |
369 | * doesn't make sense and can cause timer DoS. | |
370 | */ | |
371 | delta = max_t(s64, delay, 10000LL); | |
372 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 373 | |
cc584b21 | 374 | hrtimer_set_expires(timer, time); |
31656519 PZ |
375 | |
376 | if (rq == this_rq()) { | |
971ee28c | 377 | __hrtick_restart(rq); |
31656519 | 378 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 379 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
380 | rq->hrtick_csd_pending = 1; |
381 | } | |
b328ca18 PZ |
382 | } |
383 | ||
384 | static int | |
385 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
386 | { | |
387 | int cpu = (int)(long)hcpu; | |
388 | ||
389 | switch (action) { | |
390 | case CPU_UP_CANCELED: | |
391 | case CPU_UP_CANCELED_FROZEN: | |
392 | case CPU_DOWN_PREPARE: | |
393 | case CPU_DOWN_PREPARE_FROZEN: | |
394 | case CPU_DEAD: | |
395 | case CPU_DEAD_FROZEN: | |
31656519 | 396 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
397 | return NOTIFY_OK; |
398 | } | |
399 | ||
400 | return NOTIFY_DONE; | |
401 | } | |
402 | ||
fa748203 | 403 | static __init void init_hrtick(void) |
b328ca18 PZ |
404 | { |
405 | hotcpu_notifier(hotplug_hrtick, 0); | |
406 | } | |
31656519 PZ |
407 | #else |
408 | /* | |
409 | * Called to set the hrtick timer state. | |
410 | * | |
411 | * called with rq->lock held and irqs disabled | |
412 | */ | |
029632fb | 413 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 414 | { |
86893335 WL |
415 | /* |
416 | * Don't schedule slices shorter than 10000ns, that just | |
417 | * doesn't make sense. Rely on vruntime for fairness. | |
418 | */ | |
419 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
420 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
421 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 422 | } |
b328ca18 | 423 | |
006c75f1 | 424 | static inline void init_hrtick(void) |
8f4d37ec | 425 | { |
8f4d37ec | 426 | } |
31656519 | 427 | #endif /* CONFIG_SMP */ |
8f4d37ec | 428 | |
31656519 | 429 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 430 | { |
31656519 PZ |
431 | #ifdef CONFIG_SMP |
432 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 433 | |
31656519 PZ |
434 | rq->hrtick_csd.flags = 0; |
435 | rq->hrtick_csd.func = __hrtick_start; | |
436 | rq->hrtick_csd.info = rq; | |
437 | #endif | |
8f4d37ec | 438 | |
31656519 PZ |
439 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
440 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 441 | } |
006c75f1 | 442 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
443 | static inline void hrtick_clear(struct rq *rq) |
444 | { | |
445 | } | |
446 | ||
8f4d37ec PZ |
447 | static inline void init_rq_hrtick(struct rq *rq) |
448 | { | |
449 | } | |
450 | ||
b328ca18 PZ |
451 | static inline void init_hrtick(void) |
452 | { | |
453 | } | |
006c75f1 | 454 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 455 | |
fd99f91a PZ |
456 | /* |
457 | * cmpxchg based fetch_or, macro so it works for different integer types | |
458 | */ | |
459 | #define fetch_or(ptr, val) \ | |
460 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
461 | for (;;) { \ | |
462 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
463 | if (__old == __val) \ | |
464 | break; \ | |
465 | __val = __old; \ | |
466 | } \ | |
467 | __old; \ | |
468 | }) | |
469 | ||
e3baac47 | 470 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
471 | /* |
472 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
473 | * this avoids any races wrt polling state changes and thereby avoids | |
474 | * spurious IPIs. | |
475 | */ | |
476 | static bool set_nr_and_not_polling(struct task_struct *p) | |
477 | { | |
478 | struct thread_info *ti = task_thread_info(p); | |
479 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
480 | } | |
e3baac47 PZ |
481 | |
482 | /* | |
483 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
484 | * | |
485 | * If this returns true, then the idle task promises to call | |
486 | * sched_ttwu_pending() and reschedule soon. | |
487 | */ | |
488 | static bool set_nr_if_polling(struct task_struct *p) | |
489 | { | |
490 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 491 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
492 | |
493 | for (;;) { | |
494 | if (!(val & _TIF_POLLING_NRFLAG)) | |
495 | return false; | |
496 | if (val & _TIF_NEED_RESCHED) | |
497 | return true; | |
498 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
499 | if (old == val) | |
500 | break; | |
501 | val = old; | |
502 | } | |
503 | return true; | |
504 | } | |
505 | ||
fd99f91a PZ |
506 | #else |
507 | static bool set_nr_and_not_polling(struct task_struct *p) | |
508 | { | |
509 | set_tsk_need_resched(p); | |
510 | return true; | |
511 | } | |
e3baac47 PZ |
512 | |
513 | #ifdef CONFIG_SMP | |
514 | static bool set_nr_if_polling(struct task_struct *p) | |
515 | { | |
516 | return false; | |
517 | } | |
518 | #endif | |
fd99f91a PZ |
519 | #endif |
520 | ||
76751049 PZ |
521 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
522 | { | |
523 | struct wake_q_node *node = &task->wake_q; | |
524 | ||
525 | /* | |
526 | * Atomically grab the task, if ->wake_q is !nil already it means | |
527 | * its already queued (either by us or someone else) and will get the | |
528 | * wakeup due to that. | |
529 | * | |
530 | * This cmpxchg() implies a full barrier, which pairs with the write | |
531 | * barrier implied by the wakeup in wake_up_list(). | |
532 | */ | |
533 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
534 | return; | |
535 | ||
536 | get_task_struct(task); | |
537 | ||
538 | /* | |
539 | * The head is context local, there can be no concurrency. | |
540 | */ | |
541 | *head->lastp = node; | |
542 | head->lastp = &node->next; | |
543 | } | |
544 | ||
545 | void wake_up_q(struct wake_q_head *head) | |
546 | { | |
547 | struct wake_q_node *node = head->first; | |
548 | ||
549 | while (node != WAKE_Q_TAIL) { | |
550 | struct task_struct *task; | |
551 | ||
552 | task = container_of(node, struct task_struct, wake_q); | |
553 | BUG_ON(!task); | |
554 | /* task can safely be re-inserted now */ | |
555 | node = node->next; | |
556 | task->wake_q.next = NULL; | |
557 | ||
558 | /* | |
559 | * wake_up_process() implies a wmb() to pair with the queueing | |
560 | * in wake_q_add() so as not to miss wakeups. | |
561 | */ | |
562 | wake_up_process(task); | |
563 | put_task_struct(task); | |
564 | } | |
565 | } | |
566 | ||
c24d20db | 567 | /* |
8875125e | 568 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
569 | * |
570 | * On UP this means the setting of the need_resched flag, on SMP it | |
571 | * might also involve a cross-CPU call to trigger the scheduler on | |
572 | * the target CPU. | |
573 | */ | |
8875125e | 574 | void resched_curr(struct rq *rq) |
c24d20db | 575 | { |
8875125e | 576 | struct task_struct *curr = rq->curr; |
c24d20db IM |
577 | int cpu; |
578 | ||
8875125e | 579 | lockdep_assert_held(&rq->lock); |
c24d20db | 580 | |
8875125e | 581 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
582 | return; |
583 | ||
8875125e | 584 | cpu = cpu_of(rq); |
fd99f91a | 585 | |
f27dde8d | 586 | if (cpu == smp_processor_id()) { |
8875125e | 587 | set_tsk_need_resched(curr); |
f27dde8d | 588 | set_preempt_need_resched(); |
c24d20db | 589 | return; |
f27dde8d | 590 | } |
c24d20db | 591 | |
8875125e | 592 | if (set_nr_and_not_polling(curr)) |
c24d20db | 593 | smp_send_reschedule(cpu); |
dfc68f29 AL |
594 | else |
595 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
596 | } |
597 | ||
029632fb | 598 | void resched_cpu(int cpu) |
c24d20db IM |
599 | { |
600 | struct rq *rq = cpu_rq(cpu); | |
601 | unsigned long flags; | |
602 | ||
05fa785c | 603 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 604 | return; |
8875125e | 605 | resched_curr(rq); |
05fa785c | 606 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 607 | } |
06d8308c | 608 | |
b021fe3e | 609 | #ifdef CONFIG_SMP |
3451d024 | 610 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
611 | /* |
612 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
613 | * from an idle cpu. This is good for power-savings. | |
614 | * | |
615 | * We don't do similar optimization for completely idle system, as | |
616 | * selecting an idle cpu will add more delays to the timers than intended | |
617 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
618 | */ | |
bc7a34b8 | 619 | int get_nohz_timer_target(void) |
83cd4fe2 | 620 | { |
bc7a34b8 | 621 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
622 | struct sched_domain *sd; |
623 | ||
bc7a34b8 | 624 | if (!idle_cpu(cpu)) |
6201b4d6 VK |
625 | return cpu; |
626 | ||
057f3fad | 627 | rcu_read_lock(); |
83cd4fe2 | 628 | for_each_domain(cpu, sd) { |
057f3fad PZ |
629 | for_each_cpu(i, sched_domain_span(sd)) { |
630 | if (!idle_cpu(i)) { | |
631 | cpu = i; | |
632 | goto unlock; | |
633 | } | |
634 | } | |
83cd4fe2 | 635 | } |
057f3fad PZ |
636 | unlock: |
637 | rcu_read_unlock(); | |
83cd4fe2 VP |
638 | return cpu; |
639 | } | |
06d8308c TG |
640 | /* |
641 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
642 | * idle CPU then this timer might expire before the next timer event | |
643 | * which is scheduled to wake up that CPU. In case of a completely | |
644 | * idle system the next event might even be infinite time into the | |
645 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
646 | * leaves the inner idle loop so the newly added timer is taken into | |
647 | * account when the CPU goes back to idle and evaluates the timer | |
648 | * wheel for the next timer event. | |
649 | */ | |
1c20091e | 650 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
651 | { |
652 | struct rq *rq = cpu_rq(cpu); | |
653 | ||
654 | if (cpu == smp_processor_id()) | |
655 | return; | |
656 | ||
67b9ca70 | 657 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 658 | smp_send_reschedule(cpu); |
dfc68f29 AL |
659 | else |
660 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
661 | } |
662 | ||
c5bfece2 | 663 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 664 | { |
53c5fa16 FW |
665 | /* |
666 | * We just need the target to call irq_exit() and re-evaluate | |
667 | * the next tick. The nohz full kick at least implies that. | |
668 | * If needed we can still optimize that later with an | |
669 | * empty IRQ. | |
670 | */ | |
c5bfece2 | 671 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
672 | if (cpu != smp_processor_id() || |
673 | tick_nohz_tick_stopped()) | |
53c5fa16 | 674 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
675 | return true; |
676 | } | |
677 | ||
678 | return false; | |
679 | } | |
680 | ||
681 | void wake_up_nohz_cpu(int cpu) | |
682 | { | |
c5bfece2 | 683 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
684 | wake_up_idle_cpu(cpu); |
685 | } | |
686 | ||
ca38062e | 687 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 688 | { |
1c792db7 | 689 | int cpu = smp_processor_id(); |
873b4c65 VG |
690 | |
691 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
692 | return false; | |
693 | ||
694 | if (idle_cpu(cpu) && !need_resched()) | |
695 | return true; | |
696 | ||
697 | /* | |
698 | * We can't run Idle Load Balance on this CPU for this time so we | |
699 | * cancel it and clear NOHZ_BALANCE_KICK | |
700 | */ | |
701 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
702 | return false; | |
45bf76df IM |
703 | } |
704 | ||
3451d024 | 705 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 706 | |
ca38062e | 707 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 708 | { |
ca38062e | 709 | return false; |
2069dd75 PZ |
710 | } |
711 | ||
3451d024 | 712 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 713 | |
ce831b38 FW |
714 | #ifdef CONFIG_NO_HZ_FULL |
715 | bool sched_can_stop_tick(void) | |
716 | { | |
1e78cdbd RR |
717 | /* |
718 | * FIFO realtime policy runs the highest priority task. Other runnable | |
719 | * tasks are of a lower priority. The scheduler tick does nothing. | |
720 | */ | |
721 | if (current->policy == SCHED_FIFO) | |
722 | return true; | |
723 | ||
724 | /* | |
725 | * Round-robin realtime tasks time slice with other tasks at the same | |
726 | * realtime priority. Is this task the only one at this priority? | |
727 | */ | |
728 | if (current->policy == SCHED_RR) { | |
729 | struct sched_rt_entity *rt_se = ¤t->rt; | |
730 | ||
731 | return rt_se->run_list.prev == rt_se->run_list.next; | |
732 | } | |
733 | ||
3882ec64 FW |
734 | /* |
735 | * More than one running task need preemption. | |
736 | * nr_running update is assumed to be visible | |
737 | * after IPI is sent from wakers. | |
738 | */ | |
541b8264 VK |
739 | if (this_rq()->nr_running > 1) |
740 | return false; | |
ce831b38 | 741 | |
541b8264 | 742 | return true; |
ce831b38 FW |
743 | } |
744 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 745 | |
029632fb | 746 | void sched_avg_update(struct rq *rq) |
18d95a28 | 747 | { |
e9e9250b PZ |
748 | s64 period = sched_avg_period(); |
749 | ||
78becc27 | 750 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
751 | /* |
752 | * Inline assembly required to prevent the compiler | |
753 | * optimising this loop into a divmod call. | |
754 | * See __iter_div_u64_rem() for another example of this. | |
755 | */ | |
756 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
757 | rq->age_stamp += period; |
758 | rq->rt_avg /= 2; | |
759 | } | |
18d95a28 PZ |
760 | } |
761 | ||
6d6bc0ad | 762 | #endif /* CONFIG_SMP */ |
18d95a28 | 763 | |
a790de99 PT |
764 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
765 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 766 | /* |
8277434e PT |
767 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
768 | * node and @up when leaving it for the final time. | |
769 | * | |
770 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 771 | */ |
029632fb | 772 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 773 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
774 | { |
775 | struct task_group *parent, *child; | |
eb755805 | 776 | int ret; |
c09595f6 | 777 | |
8277434e PT |
778 | parent = from; |
779 | ||
c09595f6 | 780 | down: |
eb755805 PZ |
781 | ret = (*down)(parent, data); |
782 | if (ret) | |
8277434e | 783 | goto out; |
c09595f6 PZ |
784 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
785 | parent = child; | |
786 | goto down; | |
787 | ||
788 | up: | |
789 | continue; | |
790 | } | |
eb755805 | 791 | ret = (*up)(parent, data); |
8277434e PT |
792 | if (ret || parent == from) |
793 | goto out; | |
c09595f6 PZ |
794 | |
795 | child = parent; | |
796 | parent = parent->parent; | |
797 | if (parent) | |
798 | goto up; | |
8277434e | 799 | out: |
eb755805 | 800 | return ret; |
c09595f6 PZ |
801 | } |
802 | ||
029632fb | 803 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 804 | { |
e2b245f8 | 805 | return 0; |
eb755805 | 806 | } |
18d95a28 PZ |
807 | #endif |
808 | ||
45bf76df IM |
809 | static void set_load_weight(struct task_struct *p) |
810 | { | |
f05998d4 NR |
811 | int prio = p->static_prio - MAX_RT_PRIO; |
812 | struct load_weight *load = &p->se.load; | |
813 | ||
dd41f596 IM |
814 | /* |
815 | * SCHED_IDLE tasks get minimal weight: | |
816 | */ | |
817 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 818 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 819 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
820 | return; |
821 | } | |
71f8bd46 | 822 | |
c8b28116 | 823 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 824 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
825 | } |
826 | ||
371fd7e7 | 827 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 828 | { |
a64692a3 | 829 | update_rq_clock(rq); |
43148951 | 830 | sched_info_queued(rq, p); |
371fd7e7 | 831 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
832 | } |
833 | ||
371fd7e7 | 834 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 835 | { |
a64692a3 | 836 | update_rq_clock(rq); |
43148951 | 837 | sched_info_dequeued(rq, p); |
371fd7e7 | 838 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
839 | } |
840 | ||
029632fb | 841 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
842 | { |
843 | if (task_contributes_to_load(p)) | |
844 | rq->nr_uninterruptible--; | |
845 | ||
371fd7e7 | 846 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
847 | } |
848 | ||
029632fb | 849 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
850 | { |
851 | if (task_contributes_to_load(p)) | |
852 | rq->nr_uninterruptible++; | |
853 | ||
371fd7e7 | 854 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
855 | } |
856 | ||
fe44d621 | 857 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 858 | { |
095c0aa8 GC |
859 | /* |
860 | * In theory, the compile should just see 0 here, and optimize out the call | |
861 | * to sched_rt_avg_update. But I don't trust it... | |
862 | */ | |
863 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
864 | s64 steal = 0, irq_delta = 0; | |
865 | #endif | |
866 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 867 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
868 | |
869 | /* | |
870 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
871 | * this case when a previous update_rq_clock() happened inside a | |
872 | * {soft,}irq region. | |
873 | * | |
874 | * When this happens, we stop ->clock_task and only update the | |
875 | * prev_irq_time stamp to account for the part that fit, so that a next | |
876 | * update will consume the rest. This ensures ->clock_task is | |
877 | * monotonic. | |
878 | * | |
879 | * It does however cause some slight miss-attribution of {soft,}irq | |
880 | * time, a more accurate solution would be to update the irq_time using | |
881 | * the current rq->clock timestamp, except that would require using | |
882 | * atomic ops. | |
883 | */ | |
884 | if (irq_delta > delta) | |
885 | irq_delta = delta; | |
886 | ||
887 | rq->prev_irq_time += irq_delta; | |
888 | delta -= irq_delta; | |
095c0aa8 GC |
889 | #endif |
890 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 891 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
892 | steal = paravirt_steal_clock(cpu_of(rq)); |
893 | steal -= rq->prev_steal_time_rq; | |
894 | ||
895 | if (unlikely(steal > delta)) | |
896 | steal = delta; | |
897 | ||
095c0aa8 | 898 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
899 | delta -= steal; |
900 | } | |
901 | #endif | |
902 | ||
fe44d621 PZ |
903 | rq->clock_task += delta; |
904 | ||
095c0aa8 | 905 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 906 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
907 | sched_rt_avg_update(rq, irq_delta + steal); |
908 | #endif | |
aa483808 VP |
909 | } |
910 | ||
34f971f6 PZ |
911 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
912 | { | |
913 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
914 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
915 | ||
916 | if (stop) { | |
917 | /* | |
918 | * Make it appear like a SCHED_FIFO task, its something | |
919 | * userspace knows about and won't get confused about. | |
920 | * | |
921 | * Also, it will make PI more or less work without too | |
922 | * much confusion -- but then, stop work should not | |
923 | * rely on PI working anyway. | |
924 | */ | |
925 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
926 | ||
927 | stop->sched_class = &stop_sched_class; | |
928 | } | |
929 | ||
930 | cpu_rq(cpu)->stop = stop; | |
931 | ||
932 | if (old_stop) { | |
933 | /* | |
934 | * Reset it back to a normal scheduling class so that | |
935 | * it can die in pieces. | |
936 | */ | |
937 | old_stop->sched_class = &rt_sched_class; | |
938 | } | |
939 | } | |
940 | ||
14531189 | 941 | /* |
dd41f596 | 942 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 943 | */ |
14531189 IM |
944 | static inline int __normal_prio(struct task_struct *p) |
945 | { | |
dd41f596 | 946 | return p->static_prio; |
14531189 IM |
947 | } |
948 | ||
b29739f9 IM |
949 | /* |
950 | * Calculate the expected normal priority: i.e. priority | |
951 | * without taking RT-inheritance into account. Might be | |
952 | * boosted by interactivity modifiers. Changes upon fork, | |
953 | * setprio syscalls, and whenever the interactivity | |
954 | * estimator recalculates. | |
955 | */ | |
36c8b586 | 956 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
957 | { |
958 | int prio; | |
959 | ||
aab03e05 DF |
960 | if (task_has_dl_policy(p)) |
961 | prio = MAX_DL_PRIO-1; | |
962 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
963 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
964 | else | |
965 | prio = __normal_prio(p); | |
966 | return prio; | |
967 | } | |
968 | ||
969 | /* | |
970 | * Calculate the current priority, i.e. the priority | |
971 | * taken into account by the scheduler. This value might | |
972 | * be boosted by RT tasks, or might be boosted by | |
973 | * interactivity modifiers. Will be RT if the task got | |
974 | * RT-boosted. If not then it returns p->normal_prio. | |
975 | */ | |
36c8b586 | 976 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
977 | { |
978 | p->normal_prio = normal_prio(p); | |
979 | /* | |
980 | * If we are RT tasks or we were boosted to RT priority, | |
981 | * keep the priority unchanged. Otherwise, update priority | |
982 | * to the normal priority: | |
983 | */ | |
984 | if (!rt_prio(p->prio)) | |
985 | return p->normal_prio; | |
986 | return p->prio; | |
987 | } | |
988 | ||
1da177e4 LT |
989 | /** |
990 | * task_curr - is this task currently executing on a CPU? | |
991 | * @p: the task in question. | |
e69f6186 YB |
992 | * |
993 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 994 | */ |
36c8b586 | 995 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
996 | { |
997 | return cpu_curr(task_cpu(p)) == p; | |
998 | } | |
999 | ||
67dfa1b7 | 1000 | /* |
4c9a4bc8 PZ |
1001 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1002 | * use the balance_callback list if you want balancing. | |
1003 | * | |
1004 | * this means any call to check_class_changed() must be followed by a call to | |
1005 | * balance_callback(). | |
67dfa1b7 | 1006 | */ |
cb469845 SR |
1007 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1008 | const struct sched_class *prev_class, | |
da7a735e | 1009 | int oldprio) |
cb469845 SR |
1010 | { |
1011 | if (prev_class != p->sched_class) { | |
1012 | if (prev_class->switched_from) | |
da7a735e | 1013 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1014 | |
da7a735e | 1015 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1016 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1017 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1018 | } |
1019 | ||
029632fb | 1020 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1021 | { |
1022 | const struct sched_class *class; | |
1023 | ||
1024 | if (p->sched_class == rq->curr->sched_class) { | |
1025 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1026 | } else { | |
1027 | for_each_class(class) { | |
1028 | if (class == rq->curr->sched_class) | |
1029 | break; | |
1030 | if (class == p->sched_class) { | |
8875125e | 1031 | resched_curr(rq); |
1e5a7405 PZ |
1032 | break; |
1033 | } | |
1034 | } | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * A queue event has occurred, and we're going to schedule. In | |
1039 | * this case, we can save a useless back to back clock update. | |
1040 | */ | |
da0c1e65 | 1041 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 1042 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
1043 | } |
1044 | ||
1da177e4 | 1045 | #ifdef CONFIG_SMP |
5cc389bc PZ |
1046 | /* |
1047 | * This is how migration works: | |
1048 | * | |
1049 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1050 | * stop_one_cpu(). | |
1051 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1052 | * off the CPU) | |
1053 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1054 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1055 | * it and puts it into the right queue. | |
1056 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1057 | * is done. | |
1058 | */ | |
1059 | ||
1060 | /* | |
1061 | * move_queued_task - move a queued task to new rq. | |
1062 | * | |
1063 | * Returns (locked) new rq. Old rq's lock is released. | |
1064 | */ | |
5e16bbc2 | 1065 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 1066 | { |
5cc389bc PZ |
1067 | lockdep_assert_held(&rq->lock); |
1068 | ||
1069 | dequeue_task(rq, p, 0); | |
1070 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
1071 | set_task_cpu(p, new_cpu); | |
1072 | raw_spin_unlock(&rq->lock); | |
1073 | ||
1074 | rq = cpu_rq(new_cpu); | |
1075 | ||
1076 | raw_spin_lock(&rq->lock); | |
1077 | BUG_ON(task_cpu(p) != new_cpu); | |
1078 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1079 | enqueue_task(rq, p, 0); | |
1080 | check_preempt_curr(rq, p, 0); | |
1081 | ||
1082 | return rq; | |
1083 | } | |
1084 | ||
1085 | struct migration_arg { | |
1086 | struct task_struct *task; | |
1087 | int dest_cpu; | |
1088 | }; | |
1089 | ||
1090 | /* | |
1091 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1092 | * this because either it can't run here any more (set_cpus_allowed() | |
1093 | * away from this CPU, or CPU going down), or because we're | |
1094 | * attempting to rebalance this task on exec (sched_exec). | |
1095 | * | |
1096 | * So we race with normal scheduler movements, but that's OK, as long | |
1097 | * as the task is no longer on this CPU. | |
5cc389bc | 1098 | */ |
5e16bbc2 | 1099 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1100 | { |
5cc389bc | 1101 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1102 | return rq; |
5cc389bc PZ |
1103 | |
1104 | /* Affinity changed (again). */ | |
1105 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1106 | return rq; |
5cc389bc | 1107 | |
5e16bbc2 PZ |
1108 | rq = move_queued_task(rq, p, dest_cpu); |
1109 | ||
1110 | return rq; | |
5cc389bc PZ |
1111 | } |
1112 | ||
1113 | /* | |
1114 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1115 | * and performs thread migration by bumping thread off CPU then | |
1116 | * 'pushing' onto another runqueue. | |
1117 | */ | |
1118 | static int migration_cpu_stop(void *data) | |
1119 | { | |
1120 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1121 | struct task_struct *p = arg->task; |
1122 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1123 | |
1124 | /* | |
1125 | * The original target cpu might have gone down and we might | |
1126 | * be on another cpu but it doesn't matter. | |
1127 | */ | |
1128 | local_irq_disable(); | |
1129 | /* | |
1130 | * We need to explicitly wake pending tasks before running | |
1131 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1132 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1133 | */ | |
1134 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1135 | |
1136 | raw_spin_lock(&p->pi_lock); | |
1137 | raw_spin_lock(&rq->lock); | |
1138 | /* | |
1139 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1140 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1141 | * we're holding p->pi_lock. | |
1142 | */ | |
1143 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1144 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1145 | raw_spin_unlock(&rq->lock); | |
1146 | raw_spin_unlock(&p->pi_lock); | |
1147 | ||
5cc389bc PZ |
1148 | local_irq_enable(); |
1149 | return 0; | |
1150 | } | |
1151 | ||
c5b28038 PZ |
1152 | /* |
1153 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1154 | * actually call this function. | |
1155 | */ | |
1156 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1157 | { |
5cc389bc PZ |
1158 | cpumask_copy(&p->cpus_allowed, new_mask); |
1159 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1160 | } | |
1161 | ||
c5b28038 PZ |
1162 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1163 | { | |
6c37067e PZ |
1164 | struct rq *rq = task_rq(p); |
1165 | bool queued, running; | |
1166 | ||
c5b28038 | 1167 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1168 | |
1169 | queued = task_on_rq_queued(p); | |
1170 | running = task_current(rq, p); | |
1171 | ||
1172 | if (queued) { | |
1173 | /* | |
1174 | * Because __kthread_bind() calls this on blocked tasks without | |
1175 | * holding rq->lock. | |
1176 | */ | |
1177 | lockdep_assert_held(&rq->lock); | |
1178 | dequeue_task(rq, p, 0); | |
1179 | } | |
1180 | if (running) | |
1181 | put_prev_task(rq, p); | |
1182 | ||
c5b28038 | 1183 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e PZ |
1184 | |
1185 | if (running) | |
1186 | p->sched_class->set_curr_task(rq); | |
1187 | if (queued) | |
1188 | enqueue_task(rq, p, 0); | |
c5b28038 PZ |
1189 | } |
1190 | ||
5cc389bc PZ |
1191 | /* |
1192 | * Change a given task's CPU affinity. Migrate the thread to a | |
1193 | * proper CPU and schedule it away if the CPU it's executing on | |
1194 | * is removed from the allowed bitmask. | |
1195 | * | |
1196 | * NOTE: the caller must have a valid reference to the task, the | |
1197 | * task must not exit() & deallocate itself prematurely. The | |
1198 | * call is not atomic; no spinlocks may be held. | |
1199 | */ | |
25834c73 PZ |
1200 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1201 | const struct cpumask *new_mask, bool check) | |
5cc389bc PZ |
1202 | { |
1203 | unsigned long flags; | |
1204 | struct rq *rq; | |
1205 | unsigned int dest_cpu; | |
1206 | int ret = 0; | |
1207 | ||
1208 | rq = task_rq_lock(p, &flags); | |
1209 | ||
25834c73 PZ |
1210 | /* |
1211 | * Must re-check here, to close a race against __kthread_bind(), | |
1212 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1213 | */ | |
1214 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1215 | ret = -EINVAL; | |
1216 | goto out; | |
1217 | } | |
1218 | ||
5cc389bc PZ |
1219 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1220 | goto out; | |
1221 | ||
1222 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { | |
1223 | ret = -EINVAL; | |
1224 | goto out; | |
1225 | } | |
1226 | ||
1227 | do_set_cpus_allowed(p, new_mask); | |
1228 | ||
1229 | /* Can the task run on the task's current CPU? If so, we're done */ | |
1230 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1231 | goto out; | |
1232 | ||
1233 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); | |
1234 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
1235 | struct migration_arg arg = { p, dest_cpu }; | |
1236 | /* Need help from migration thread: drop lock and wait. */ | |
1237 | task_rq_unlock(rq, p, &flags); | |
1238 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
1239 | tlb_migrate_finish(p->mm); | |
1240 | return 0; | |
cbce1a68 PZ |
1241 | } else if (task_on_rq_queued(p)) { |
1242 | /* | |
1243 | * OK, since we're going to drop the lock immediately | |
1244 | * afterwards anyway. | |
1245 | */ | |
1246 | lockdep_unpin_lock(&rq->lock); | |
5e16bbc2 | 1247 | rq = move_queued_task(rq, p, dest_cpu); |
cbce1a68 PZ |
1248 | lockdep_pin_lock(&rq->lock); |
1249 | } | |
5cc389bc PZ |
1250 | out: |
1251 | task_rq_unlock(rq, p, &flags); | |
1252 | ||
1253 | return ret; | |
1254 | } | |
25834c73 PZ |
1255 | |
1256 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1257 | { | |
1258 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1259 | } | |
5cc389bc PZ |
1260 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1261 | ||
dd41f596 | 1262 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1263 | { |
e2912009 PZ |
1264 | #ifdef CONFIG_SCHED_DEBUG |
1265 | /* | |
1266 | * We should never call set_task_cpu() on a blocked task, | |
1267 | * ttwu() will sort out the placement. | |
1268 | */ | |
077614ee | 1269 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1270 | !p->on_rq); |
0122ec5b PZ |
1271 | |
1272 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1273 | /* |
1274 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1275 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1276 | * | |
1277 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1278 | * see task_group(). |
6c6c54e1 PZ |
1279 | * |
1280 | * Furthermore, all task_rq users should acquire both locks, see | |
1281 | * task_rq_lock(). | |
1282 | */ | |
0122ec5b PZ |
1283 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1284 | lockdep_is_held(&task_rq(p)->lock))); | |
1285 | #endif | |
e2912009 PZ |
1286 | #endif |
1287 | ||
de1d7286 | 1288 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1289 | |
0c69774e | 1290 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1291 | if (p->sched_class->migrate_task_rq) |
1292 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1293 | p->se.nr_migrations++; |
ff303e66 | 1294 | perf_event_task_migrate(p); |
0c69774e | 1295 | } |
dd41f596 IM |
1296 | |
1297 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1298 | } |
1299 | ||
ac66f547 PZ |
1300 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1301 | { | |
da0c1e65 | 1302 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1303 | struct rq *src_rq, *dst_rq; |
1304 | ||
1305 | src_rq = task_rq(p); | |
1306 | dst_rq = cpu_rq(cpu); | |
1307 | ||
1308 | deactivate_task(src_rq, p, 0); | |
1309 | set_task_cpu(p, cpu); | |
1310 | activate_task(dst_rq, p, 0); | |
1311 | check_preempt_curr(dst_rq, p, 0); | |
1312 | } else { | |
1313 | /* | |
1314 | * Task isn't running anymore; make it appear like we migrated | |
1315 | * it before it went to sleep. This means on wakeup we make the | |
1316 | * previous cpu our targer instead of where it really is. | |
1317 | */ | |
1318 | p->wake_cpu = cpu; | |
1319 | } | |
1320 | } | |
1321 | ||
1322 | struct migration_swap_arg { | |
1323 | struct task_struct *src_task, *dst_task; | |
1324 | int src_cpu, dst_cpu; | |
1325 | }; | |
1326 | ||
1327 | static int migrate_swap_stop(void *data) | |
1328 | { | |
1329 | struct migration_swap_arg *arg = data; | |
1330 | struct rq *src_rq, *dst_rq; | |
1331 | int ret = -EAGAIN; | |
1332 | ||
1333 | src_rq = cpu_rq(arg->src_cpu); | |
1334 | dst_rq = cpu_rq(arg->dst_cpu); | |
1335 | ||
74602315 PZ |
1336 | double_raw_lock(&arg->src_task->pi_lock, |
1337 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1338 | double_rq_lock(src_rq, dst_rq); |
1339 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1340 | goto unlock; | |
1341 | ||
1342 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1343 | goto unlock; | |
1344 | ||
1345 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1346 | goto unlock; | |
1347 | ||
1348 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1349 | goto unlock; | |
1350 | ||
1351 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1352 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1353 | ||
1354 | ret = 0; | |
1355 | ||
1356 | unlock: | |
1357 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1358 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1359 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1360 | |
1361 | return ret; | |
1362 | } | |
1363 | ||
1364 | /* | |
1365 | * Cross migrate two tasks | |
1366 | */ | |
1367 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1368 | { | |
1369 | struct migration_swap_arg arg; | |
1370 | int ret = -EINVAL; | |
1371 | ||
ac66f547 PZ |
1372 | arg = (struct migration_swap_arg){ |
1373 | .src_task = cur, | |
1374 | .src_cpu = task_cpu(cur), | |
1375 | .dst_task = p, | |
1376 | .dst_cpu = task_cpu(p), | |
1377 | }; | |
1378 | ||
1379 | if (arg.src_cpu == arg.dst_cpu) | |
1380 | goto out; | |
1381 | ||
6acce3ef PZ |
1382 | /* |
1383 | * These three tests are all lockless; this is OK since all of them | |
1384 | * will be re-checked with proper locks held further down the line. | |
1385 | */ | |
ac66f547 PZ |
1386 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1387 | goto out; | |
1388 | ||
1389 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1390 | goto out; | |
1391 | ||
1392 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1393 | goto out; | |
1394 | ||
286549dc | 1395 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1396 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1397 | ||
1398 | out: | |
ac66f547 PZ |
1399 | return ret; |
1400 | } | |
1401 | ||
1da177e4 LT |
1402 | /* |
1403 | * wait_task_inactive - wait for a thread to unschedule. | |
1404 | * | |
85ba2d86 RM |
1405 | * If @match_state is nonzero, it's the @p->state value just checked and |
1406 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1407 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1408 | * we return a positive number (its total switch count). If a second call | |
1409 | * a short while later returns the same number, the caller can be sure that | |
1410 | * @p has remained unscheduled the whole time. | |
1411 | * | |
1da177e4 LT |
1412 | * The caller must ensure that the task *will* unschedule sometime soon, |
1413 | * else this function might spin for a *long* time. This function can't | |
1414 | * be called with interrupts off, or it may introduce deadlock with | |
1415 | * smp_call_function() if an IPI is sent by the same process we are | |
1416 | * waiting to become inactive. | |
1417 | */ | |
85ba2d86 | 1418 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1419 | { |
1420 | unsigned long flags; | |
da0c1e65 | 1421 | int running, queued; |
85ba2d86 | 1422 | unsigned long ncsw; |
70b97a7f | 1423 | struct rq *rq; |
1da177e4 | 1424 | |
3a5c359a AK |
1425 | for (;;) { |
1426 | /* | |
1427 | * We do the initial early heuristics without holding | |
1428 | * any task-queue locks at all. We'll only try to get | |
1429 | * the runqueue lock when things look like they will | |
1430 | * work out! | |
1431 | */ | |
1432 | rq = task_rq(p); | |
fa490cfd | 1433 | |
3a5c359a AK |
1434 | /* |
1435 | * If the task is actively running on another CPU | |
1436 | * still, just relax and busy-wait without holding | |
1437 | * any locks. | |
1438 | * | |
1439 | * NOTE! Since we don't hold any locks, it's not | |
1440 | * even sure that "rq" stays as the right runqueue! | |
1441 | * But we don't care, since "task_running()" will | |
1442 | * return false if the runqueue has changed and p | |
1443 | * is actually now running somewhere else! | |
1444 | */ | |
85ba2d86 RM |
1445 | while (task_running(rq, p)) { |
1446 | if (match_state && unlikely(p->state != match_state)) | |
1447 | return 0; | |
3a5c359a | 1448 | cpu_relax(); |
85ba2d86 | 1449 | } |
fa490cfd | 1450 | |
3a5c359a AK |
1451 | /* |
1452 | * Ok, time to look more closely! We need the rq | |
1453 | * lock now, to be *sure*. If we're wrong, we'll | |
1454 | * just go back and repeat. | |
1455 | */ | |
1456 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1457 | trace_sched_wait_task(p); |
3a5c359a | 1458 | running = task_running(rq, p); |
da0c1e65 | 1459 | queued = task_on_rq_queued(p); |
85ba2d86 | 1460 | ncsw = 0; |
f31e11d8 | 1461 | if (!match_state || p->state == match_state) |
93dcf55f | 1462 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1463 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1464 | |
85ba2d86 RM |
1465 | /* |
1466 | * If it changed from the expected state, bail out now. | |
1467 | */ | |
1468 | if (unlikely(!ncsw)) | |
1469 | break; | |
1470 | ||
3a5c359a AK |
1471 | /* |
1472 | * Was it really running after all now that we | |
1473 | * checked with the proper locks actually held? | |
1474 | * | |
1475 | * Oops. Go back and try again.. | |
1476 | */ | |
1477 | if (unlikely(running)) { | |
1478 | cpu_relax(); | |
1479 | continue; | |
1480 | } | |
fa490cfd | 1481 | |
3a5c359a AK |
1482 | /* |
1483 | * It's not enough that it's not actively running, | |
1484 | * it must be off the runqueue _entirely_, and not | |
1485 | * preempted! | |
1486 | * | |
80dd99b3 | 1487 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1488 | * running right now), it's preempted, and we should |
1489 | * yield - it could be a while. | |
1490 | */ | |
da0c1e65 | 1491 | if (unlikely(queued)) { |
8eb90c30 TG |
1492 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1493 | ||
1494 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1495 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1496 | continue; |
1497 | } | |
fa490cfd | 1498 | |
3a5c359a AK |
1499 | /* |
1500 | * Ahh, all good. It wasn't running, and it wasn't | |
1501 | * runnable, which means that it will never become | |
1502 | * running in the future either. We're all done! | |
1503 | */ | |
1504 | break; | |
1505 | } | |
85ba2d86 RM |
1506 | |
1507 | return ncsw; | |
1da177e4 LT |
1508 | } |
1509 | ||
1510 | /*** | |
1511 | * kick_process - kick a running thread to enter/exit the kernel | |
1512 | * @p: the to-be-kicked thread | |
1513 | * | |
1514 | * Cause a process which is running on another CPU to enter | |
1515 | * kernel-mode, without any delay. (to get signals handled.) | |
1516 | * | |
25985edc | 1517 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1518 | * because all it wants to ensure is that the remote task enters |
1519 | * the kernel. If the IPI races and the task has been migrated | |
1520 | * to another CPU then no harm is done and the purpose has been | |
1521 | * achieved as well. | |
1522 | */ | |
36c8b586 | 1523 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1524 | { |
1525 | int cpu; | |
1526 | ||
1527 | preempt_disable(); | |
1528 | cpu = task_cpu(p); | |
1529 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1530 | smp_send_reschedule(cpu); | |
1531 | preempt_enable(); | |
1532 | } | |
b43e3521 | 1533 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1534 | |
30da688e | 1535 | /* |
013fdb80 | 1536 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1537 | */ |
5da9a0fb PZ |
1538 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1539 | { | |
aa00d89c TC |
1540 | int nid = cpu_to_node(cpu); |
1541 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1542 | enum { cpuset, possible, fail } state = cpuset; |
1543 | int dest_cpu; | |
5da9a0fb | 1544 | |
aa00d89c TC |
1545 | /* |
1546 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1547 | * will return -1. There is no cpu on the node, and we should | |
1548 | * select the cpu on the other node. | |
1549 | */ | |
1550 | if (nid != -1) { | |
1551 | nodemask = cpumask_of_node(nid); | |
1552 | ||
1553 | /* Look for allowed, online CPU in same node. */ | |
1554 | for_each_cpu(dest_cpu, nodemask) { | |
1555 | if (!cpu_online(dest_cpu)) | |
1556 | continue; | |
1557 | if (!cpu_active(dest_cpu)) | |
1558 | continue; | |
1559 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1560 | return dest_cpu; | |
1561 | } | |
2baab4e9 | 1562 | } |
5da9a0fb | 1563 | |
2baab4e9 PZ |
1564 | for (;;) { |
1565 | /* Any allowed, online CPU? */ | |
e3831edd | 1566 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1567 | if (!cpu_online(dest_cpu)) |
1568 | continue; | |
1569 | if (!cpu_active(dest_cpu)) | |
1570 | continue; | |
1571 | goto out; | |
1572 | } | |
5da9a0fb | 1573 | |
2baab4e9 PZ |
1574 | switch (state) { |
1575 | case cpuset: | |
1576 | /* No more Mr. Nice Guy. */ | |
1577 | cpuset_cpus_allowed_fallback(p); | |
1578 | state = possible; | |
1579 | break; | |
1580 | ||
1581 | case possible: | |
1582 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1583 | state = fail; | |
1584 | break; | |
1585 | ||
1586 | case fail: | |
1587 | BUG(); | |
1588 | break; | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | out: | |
1593 | if (state != cpuset) { | |
1594 | /* | |
1595 | * Don't tell them about moving exiting tasks or | |
1596 | * kernel threads (both mm NULL), since they never | |
1597 | * leave kernel. | |
1598 | */ | |
1599 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1600 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1601 | task_pid_nr(p), p->comm, cpu); |
1602 | } | |
5da9a0fb PZ |
1603 | } |
1604 | ||
1605 | return dest_cpu; | |
1606 | } | |
1607 | ||
e2912009 | 1608 | /* |
013fdb80 | 1609 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1610 | */ |
970b13ba | 1611 | static inline |
ac66f547 | 1612 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1613 | { |
cbce1a68 PZ |
1614 | lockdep_assert_held(&p->pi_lock); |
1615 | ||
6c1d9410 WL |
1616 | if (p->nr_cpus_allowed > 1) |
1617 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e2912009 PZ |
1618 | |
1619 | /* | |
1620 | * In order not to call set_task_cpu() on a blocking task we need | |
1621 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1622 | * cpu. | |
1623 | * | |
1624 | * Since this is common to all placement strategies, this lives here. | |
1625 | * | |
1626 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1627 | * not worry about this generic constraint ] | |
1628 | */ | |
fa17b507 | 1629 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1630 | !cpu_online(cpu))) |
5da9a0fb | 1631 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1632 | |
1633 | return cpu; | |
970b13ba | 1634 | } |
09a40af5 MG |
1635 | |
1636 | static void update_avg(u64 *avg, u64 sample) | |
1637 | { | |
1638 | s64 diff = sample - *avg; | |
1639 | *avg += diff >> 3; | |
1640 | } | |
25834c73 PZ |
1641 | |
1642 | #else | |
1643 | ||
1644 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1645 | const struct cpumask *new_mask, bool check) | |
1646 | { | |
1647 | return set_cpus_allowed_ptr(p, new_mask); | |
1648 | } | |
1649 | ||
5cc389bc | 1650 | #endif /* CONFIG_SMP */ |
970b13ba | 1651 | |
d7c01d27 | 1652 | static void |
b84cb5df | 1653 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1654 | { |
d7c01d27 | 1655 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1656 | struct rq *rq = this_rq(); |
1657 | ||
d7c01d27 PZ |
1658 | #ifdef CONFIG_SMP |
1659 | int this_cpu = smp_processor_id(); | |
1660 | ||
1661 | if (cpu == this_cpu) { | |
1662 | schedstat_inc(rq, ttwu_local); | |
1663 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1664 | } else { | |
1665 | struct sched_domain *sd; | |
1666 | ||
1667 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1668 | rcu_read_lock(); |
d7c01d27 PZ |
1669 | for_each_domain(this_cpu, sd) { |
1670 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1671 | schedstat_inc(sd, ttwu_wake_remote); | |
1672 | break; | |
1673 | } | |
1674 | } | |
057f3fad | 1675 | rcu_read_unlock(); |
d7c01d27 | 1676 | } |
f339b9dc PZ |
1677 | |
1678 | if (wake_flags & WF_MIGRATED) | |
1679 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1680 | ||
d7c01d27 PZ |
1681 | #endif /* CONFIG_SMP */ |
1682 | ||
1683 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1684 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1685 | |
1686 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1687 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1688 | |
d7c01d27 PZ |
1689 | #endif /* CONFIG_SCHEDSTATS */ |
1690 | } | |
1691 | ||
1692 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1693 | { | |
9ed3811a | 1694 | activate_task(rq, p, en_flags); |
da0c1e65 | 1695 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1696 | |
1697 | /* if a worker is waking up, notify workqueue */ | |
1698 | if (p->flags & PF_WQ_WORKER) | |
1699 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1700 | } |
1701 | ||
23f41eeb PZ |
1702 | /* |
1703 | * Mark the task runnable and perform wakeup-preemption. | |
1704 | */ | |
89363381 | 1705 | static void |
23f41eeb | 1706 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1707 | { |
9ed3811a | 1708 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1709 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1710 | trace_sched_wakeup(p); |
1711 | ||
9ed3811a | 1712 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1713 | if (p->sched_class->task_woken) { |
1714 | /* | |
cbce1a68 PZ |
1715 | * Our task @p is fully woken up and running; so its safe to |
1716 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1717 | */ |
cbce1a68 | 1718 | lockdep_unpin_lock(&rq->lock); |
9ed3811a | 1719 | p->sched_class->task_woken(rq, p); |
cbce1a68 | 1720 | lockdep_pin_lock(&rq->lock); |
4c9a4bc8 | 1721 | } |
9ed3811a | 1722 | |
e69c6341 | 1723 | if (rq->idle_stamp) { |
78becc27 | 1724 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1725 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1726 | |
abfafa54 JL |
1727 | update_avg(&rq->avg_idle, delta); |
1728 | ||
1729 | if (rq->avg_idle > max) | |
9ed3811a | 1730 | rq->avg_idle = max; |
abfafa54 | 1731 | |
9ed3811a TH |
1732 | rq->idle_stamp = 0; |
1733 | } | |
1734 | #endif | |
1735 | } | |
1736 | ||
c05fbafb PZ |
1737 | static void |
1738 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1739 | { | |
cbce1a68 PZ |
1740 | lockdep_assert_held(&rq->lock); |
1741 | ||
c05fbafb PZ |
1742 | #ifdef CONFIG_SMP |
1743 | if (p->sched_contributes_to_load) | |
1744 | rq->nr_uninterruptible--; | |
1745 | #endif | |
1746 | ||
1747 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1748 | ttwu_do_wakeup(rq, p, wake_flags); | |
1749 | } | |
1750 | ||
1751 | /* | |
1752 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1753 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1754 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1755 | * the task is still ->on_rq. | |
1756 | */ | |
1757 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1758 | { | |
1759 | struct rq *rq; | |
1760 | int ret = 0; | |
1761 | ||
1762 | rq = __task_rq_lock(p); | |
da0c1e65 | 1763 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1764 | /* check_preempt_curr() may use rq clock */ |
1765 | update_rq_clock(rq); | |
c05fbafb PZ |
1766 | ttwu_do_wakeup(rq, p, wake_flags); |
1767 | ret = 1; | |
1768 | } | |
1769 | __task_rq_unlock(rq); | |
1770 | ||
1771 | return ret; | |
1772 | } | |
1773 | ||
317f3941 | 1774 | #ifdef CONFIG_SMP |
e3baac47 | 1775 | void sched_ttwu_pending(void) |
317f3941 PZ |
1776 | { |
1777 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1778 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1779 | struct task_struct *p; | |
e3baac47 | 1780 | unsigned long flags; |
317f3941 | 1781 | |
e3baac47 PZ |
1782 | if (!llist) |
1783 | return; | |
1784 | ||
1785 | raw_spin_lock_irqsave(&rq->lock, flags); | |
cbce1a68 | 1786 | lockdep_pin_lock(&rq->lock); |
317f3941 | 1787 | |
fa14ff4a PZ |
1788 | while (llist) { |
1789 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1790 | llist = llist_next(llist); | |
317f3941 PZ |
1791 | ttwu_do_activate(rq, p, 0); |
1792 | } | |
1793 | ||
cbce1a68 | 1794 | lockdep_unpin_lock(&rq->lock); |
e3baac47 | 1795 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1796 | } |
1797 | ||
1798 | void scheduler_ipi(void) | |
1799 | { | |
f27dde8d PZ |
1800 | /* |
1801 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1802 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1803 | * this IPI. | |
1804 | */ | |
8cb75e0c | 1805 | preempt_fold_need_resched(); |
f27dde8d | 1806 | |
fd2ac4f4 | 1807 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1808 | return; |
1809 | ||
1810 | /* | |
1811 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1812 | * traditionally all their work was done from the interrupt return | |
1813 | * path. Now that we actually do some work, we need to make sure | |
1814 | * we do call them. | |
1815 | * | |
1816 | * Some archs already do call them, luckily irq_enter/exit nest | |
1817 | * properly. | |
1818 | * | |
1819 | * Arguably we should visit all archs and update all handlers, | |
1820 | * however a fair share of IPIs are still resched only so this would | |
1821 | * somewhat pessimize the simple resched case. | |
1822 | */ | |
1823 | irq_enter(); | |
fa14ff4a | 1824 | sched_ttwu_pending(); |
ca38062e SS |
1825 | |
1826 | /* | |
1827 | * Check if someone kicked us for doing the nohz idle load balance. | |
1828 | */ | |
873b4c65 | 1829 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1830 | this_rq()->idle_balance = 1; |
ca38062e | 1831 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1832 | } |
c5d753a5 | 1833 | irq_exit(); |
317f3941 PZ |
1834 | } |
1835 | ||
1836 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1837 | { | |
e3baac47 PZ |
1838 | struct rq *rq = cpu_rq(cpu); |
1839 | ||
1840 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1841 | if (!set_nr_if_polling(rq->idle)) | |
1842 | smp_send_reschedule(cpu); | |
1843 | else | |
1844 | trace_sched_wake_idle_without_ipi(cpu); | |
1845 | } | |
317f3941 | 1846 | } |
d6aa8f85 | 1847 | |
f6be8af1 CL |
1848 | void wake_up_if_idle(int cpu) |
1849 | { | |
1850 | struct rq *rq = cpu_rq(cpu); | |
1851 | unsigned long flags; | |
1852 | ||
fd7de1e8 AL |
1853 | rcu_read_lock(); |
1854 | ||
1855 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1856 | goto out; | |
f6be8af1 CL |
1857 | |
1858 | if (set_nr_if_polling(rq->idle)) { | |
1859 | trace_sched_wake_idle_without_ipi(cpu); | |
1860 | } else { | |
1861 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1862 | if (is_idle_task(rq->curr)) | |
1863 | smp_send_reschedule(cpu); | |
1864 | /* Else cpu is not in idle, do nothing here */ | |
1865 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1866 | } | |
fd7de1e8 AL |
1867 | |
1868 | out: | |
1869 | rcu_read_unlock(); | |
f6be8af1 CL |
1870 | } |
1871 | ||
39be3501 | 1872 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1873 | { |
1874 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1875 | } | |
d6aa8f85 | 1876 | #endif /* CONFIG_SMP */ |
317f3941 | 1877 | |
c05fbafb PZ |
1878 | static void ttwu_queue(struct task_struct *p, int cpu) |
1879 | { | |
1880 | struct rq *rq = cpu_rq(cpu); | |
1881 | ||
17d9f311 | 1882 | #if defined(CONFIG_SMP) |
39be3501 | 1883 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1884 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1885 | ttwu_queue_remote(p, cpu); |
1886 | return; | |
1887 | } | |
1888 | #endif | |
1889 | ||
c05fbafb | 1890 | raw_spin_lock(&rq->lock); |
cbce1a68 | 1891 | lockdep_pin_lock(&rq->lock); |
c05fbafb | 1892 | ttwu_do_activate(rq, p, 0); |
cbce1a68 | 1893 | lockdep_unpin_lock(&rq->lock); |
c05fbafb | 1894 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1895 | } |
1896 | ||
1897 | /** | |
1da177e4 | 1898 | * try_to_wake_up - wake up a thread |
9ed3811a | 1899 | * @p: the thread to be awakened |
1da177e4 | 1900 | * @state: the mask of task states that can be woken |
9ed3811a | 1901 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1902 | * |
1903 | * Put it on the run-queue if it's not already there. The "current" | |
1904 | * thread is always on the run-queue (except when the actual | |
1905 | * re-schedule is in progress), and as such you're allowed to do | |
1906 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1907 | * runnable without the overhead of this. | |
1908 | * | |
e69f6186 | 1909 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1910 | * or @state didn't match @p's state. |
1da177e4 | 1911 | */ |
e4a52bcb PZ |
1912 | static int |
1913 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1914 | { |
1da177e4 | 1915 | unsigned long flags; |
c05fbafb | 1916 | int cpu, success = 0; |
2398f2c6 | 1917 | |
e0acd0a6 ON |
1918 | /* |
1919 | * If we are going to wake up a thread waiting for CONDITION we | |
1920 | * need to ensure that CONDITION=1 done by the caller can not be | |
1921 | * reordered with p->state check below. This pairs with mb() in | |
1922 | * set_current_state() the waiting thread does. | |
1923 | */ | |
1924 | smp_mb__before_spinlock(); | |
013fdb80 | 1925 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1926 | if (!(p->state & state)) |
1da177e4 LT |
1927 | goto out; |
1928 | ||
fbd705a0 PZ |
1929 | trace_sched_waking(p); |
1930 | ||
c05fbafb | 1931 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1932 | cpu = task_cpu(p); |
1da177e4 | 1933 | |
c05fbafb PZ |
1934 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1935 | goto stat; | |
1da177e4 | 1936 | |
1da177e4 | 1937 | #ifdef CONFIG_SMP |
e9c84311 | 1938 | /* |
c05fbafb PZ |
1939 | * If the owning (remote) cpu is still in the middle of schedule() with |
1940 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1941 | */ |
f3e94786 | 1942 | while (p->on_cpu) |
e4a52bcb | 1943 | cpu_relax(); |
0970d299 | 1944 | /* |
e4a52bcb | 1945 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1946 | */ |
e4a52bcb | 1947 | smp_rmb(); |
1da177e4 | 1948 | |
a8e4f2ea | 1949 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1950 | p->state = TASK_WAKING; |
e7693a36 | 1951 | |
e4a52bcb | 1952 | if (p->sched_class->task_waking) |
74f8e4b2 | 1953 | p->sched_class->task_waking(p); |
efbbd05a | 1954 | |
ac66f547 | 1955 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1956 | if (task_cpu(p) != cpu) { |
1957 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1958 | set_task_cpu(p, cpu); |
f339b9dc | 1959 | } |
1da177e4 | 1960 | #endif /* CONFIG_SMP */ |
1da177e4 | 1961 | |
c05fbafb PZ |
1962 | ttwu_queue(p, cpu); |
1963 | stat: | |
b84cb5df | 1964 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1965 | out: |
013fdb80 | 1966 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1967 | |
1968 | return success; | |
1969 | } | |
1970 | ||
21aa9af0 TH |
1971 | /** |
1972 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1973 | * @p: the thread to be awakened | |
1974 | * | |
2acca55e | 1975 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1976 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1977 | * the current task. |
21aa9af0 TH |
1978 | */ |
1979 | static void try_to_wake_up_local(struct task_struct *p) | |
1980 | { | |
1981 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1982 | |
383efcd0 TH |
1983 | if (WARN_ON_ONCE(rq != this_rq()) || |
1984 | WARN_ON_ONCE(p == current)) | |
1985 | return; | |
1986 | ||
21aa9af0 TH |
1987 | lockdep_assert_held(&rq->lock); |
1988 | ||
2acca55e | 1989 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
1990 | /* |
1991 | * This is OK, because current is on_cpu, which avoids it being | |
1992 | * picked for load-balance and preemption/IRQs are still | |
1993 | * disabled avoiding further scheduler activity on it and we've | |
1994 | * not yet picked a replacement task. | |
1995 | */ | |
1996 | lockdep_unpin_lock(&rq->lock); | |
2acca55e PZ |
1997 | raw_spin_unlock(&rq->lock); |
1998 | raw_spin_lock(&p->pi_lock); | |
1999 | raw_spin_lock(&rq->lock); | |
cbce1a68 | 2000 | lockdep_pin_lock(&rq->lock); |
2acca55e PZ |
2001 | } |
2002 | ||
21aa9af0 | 2003 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2004 | goto out; |
21aa9af0 | 2005 | |
fbd705a0 PZ |
2006 | trace_sched_waking(p); |
2007 | ||
da0c1e65 | 2008 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
2009 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2010 | ||
23f41eeb | 2011 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2012 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2013 | out: |
2014 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2015 | } |
2016 | ||
50fa610a DH |
2017 | /** |
2018 | * wake_up_process - Wake up a specific process | |
2019 | * @p: The process to be woken up. | |
2020 | * | |
2021 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2022 | * processes. |
2023 | * | |
2024 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2025 | * |
2026 | * It may be assumed that this function implies a write memory barrier before | |
2027 | * changing the task state if and only if any tasks are woken up. | |
2028 | */ | |
7ad5b3a5 | 2029 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2030 | { |
9067ac85 ON |
2031 | WARN_ON(task_is_stopped_or_traced(p)); |
2032 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 2033 | } |
1da177e4 LT |
2034 | EXPORT_SYMBOL(wake_up_process); |
2035 | ||
7ad5b3a5 | 2036 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2037 | { |
2038 | return try_to_wake_up(p, state, 0); | |
2039 | } | |
2040 | ||
a5e7be3b JL |
2041 | /* |
2042 | * This function clears the sched_dl_entity static params. | |
2043 | */ | |
2044 | void __dl_clear_params(struct task_struct *p) | |
2045 | { | |
2046 | struct sched_dl_entity *dl_se = &p->dl; | |
2047 | ||
2048 | dl_se->dl_runtime = 0; | |
2049 | dl_se->dl_deadline = 0; | |
2050 | dl_se->dl_period = 0; | |
2051 | dl_se->flags = 0; | |
2052 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2053 | |
2054 | dl_se->dl_throttled = 0; | |
2055 | dl_se->dl_new = 1; | |
2056 | dl_se->dl_yielded = 0; | |
a5e7be3b JL |
2057 | } |
2058 | ||
1da177e4 LT |
2059 | /* |
2060 | * Perform scheduler related setup for a newly forked process p. | |
2061 | * p is forked by current. | |
dd41f596 IM |
2062 | * |
2063 | * __sched_fork() is basic setup used by init_idle() too: | |
2064 | */ | |
5e1576ed | 2065 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2066 | { |
fd2f4419 PZ |
2067 | p->on_rq = 0; |
2068 | ||
2069 | p->se.on_rq = 0; | |
dd41f596 IM |
2070 | p->se.exec_start = 0; |
2071 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2072 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2073 | p->se.nr_migrations = 0; |
da7a735e | 2074 | p->se.vruntime = 0; |
fd2f4419 | 2075 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2076 | |
2077 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2078 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2079 | #endif |
476d139c | 2080 | |
aab03e05 | 2081 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2082 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2083 | __dl_clear_params(p); |
aab03e05 | 2084 | |
fa717060 | 2085 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2086 | |
e107be36 AK |
2087 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2088 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2089 | #endif | |
cbee9f88 PZ |
2090 | |
2091 | #ifdef CONFIG_NUMA_BALANCING | |
2092 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2093 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2094 | p->mm->numa_scan_seq = 0; |
2095 | } | |
2096 | ||
5e1576ed RR |
2097 | if (clone_flags & CLONE_VM) |
2098 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2099 | else | |
2100 | p->numa_preferred_nid = -1; | |
2101 | ||
cbee9f88 PZ |
2102 | p->node_stamp = 0ULL; |
2103 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2104 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2105 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2106 | p->numa_faults = NULL; |
7e2703e6 RR |
2107 | p->last_task_numa_placement = 0; |
2108 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2109 | |
8c8a743c | 2110 | p->numa_group = NULL; |
cbee9f88 | 2111 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2112 | } |
2113 | ||
1a687c2e | 2114 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 2115 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
2116 | void set_numabalancing_state(bool enabled) |
2117 | { | |
2118 | if (enabled) | |
2119 | sched_feat_set("NUMA"); | |
2120 | else | |
2121 | sched_feat_set("NO_NUMA"); | |
2122 | } | |
3105b86a MG |
2123 | #else |
2124 | __read_mostly bool numabalancing_enabled; | |
2125 | ||
2126 | void set_numabalancing_state(bool enabled) | |
2127 | { | |
2128 | numabalancing_enabled = enabled; | |
dd41f596 | 2129 | } |
3105b86a | 2130 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
2131 | |
2132 | #ifdef CONFIG_PROC_SYSCTL | |
2133 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2134 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2135 | { | |
2136 | struct ctl_table t; | |
2137 | int err; | |
2138 | int state = numabalancing_enabled; | |
2139 | ||
2140 | if (write && !capable(CAP_SYS_ADMIN)) | |
2141 | return -EPERM; | |
2142 | ||
2143 | t = *table; | |
2144 | t.data = &state; | |
2145 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2146 | if (err < 0) | |
2147 | return err; | |
2148 | if (write) | |
2149 | set_numabalancing_state(state); | |
2150 | return err; | |
2151 | } | |
2152 | #endif | |
2153 | #endif | |
dd41f596 IM |
2154 | |
2155 | /* | |
2156 | * fork()/clone()-time setup: | |
2157 | */ | |
aab03e05 | 2158 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2159 | { |
0122ec5b | 2160 | unsigned long flags; |
dd41f596 IM |
2161 | int cpu = get_cpu(); |
2162 | ||
5e1576ed | 2163 | __sched_fork(clone_flags, p); |
06b83b5f | 2164 | /* |
0017d735 | 2165 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2166 | * nobody will actually run it, and a signal or other external |
2167 | * event cannot wake it up and insert it on the runqueue either. | |
2168 | */ | |
0017d735 | 2169 | p->state = TASK_RUNNING; |
dd41f596 | 2170 | |
c350a04e MG |
2171 | /* |
2172 | * Make sure we do not leak PI boosting priority to the child. | |
2173 | */ | |
2174 | p->prio = current->normal_prio; | |
2175 | ||
b9dc29e7 MG |
2176 | /* |
2177 | * Revert to default priority/policy on fork if requested. | |
2178 | */ | |
2179 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2180 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2181 | p->policy = SCHED_NORMAL; |
6c697bdf | 2182 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2183 | p->rt_priority = 0; |
2184 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2185 | p->static_prio = NICE_TO_PRIO(0); | |
2186 | ||
2187 | p->prio = p->normal_prio = __normal_prio(p); | |
2188 | set_load_weight(p); | |
6c697bdf | 2189 | |
b9dc29e7 MG |
2190 | /* |
2191 | * We don't need the reset flag anymore after the fork. It has | |
2192 | * fulfilled its duty: | |
2193 | */ | |
2194 | p->sched_reset_on_fork = 0; | |
2195 | } | |
ca94c442 | 2196 | |
aab03e05 DF |
2197 | if (dl_prio(p->prio)) { |
2198 | put_cpu(); | |
2199 | return -EAGAIN; | |
2200 | } else if (rt_prio(p->prio)) { | |
2201 | p->sched_class = &rt_sched_class; | |
2202 | } else { | |
2ddbf952 | 2203 | p->sched_class = &fair_sched_class; |
aab03e05 | 2204 | } |
b29739f9 | 2205 | |
cd29fe6f PZ |
2206 | if (p->sched_class->task_fork) |
2207 | p->sched_class->task_fork(p); | |
2208 | ||
86951599 PZ |
2209 | /* |
2210 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2211 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2212 | * is ran before sched_fork(). | |
2213 | * | |
2214 | * Silence PROVE_RCU. | |
2215 | */ | |
0122ec5b | 2216 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2217 | set_task_cpu(p, cpu); |
0122ec5b | 2218 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2219 | |
f6db8347 | 2220 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2221 | if (likely(sched_info_on())) |
52f17b6c | 2222 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2223 | #endif |
3ca7a440 PZ |
2224 | #if defined(CONFIG_SMP) |
2225 | p->on_cpu = 0; | |
4866cde0 | 2226 | #endif |
01028747 | 2227 | init_task_preempt_count(p); |
806c09a7 | 2228 | #ifdef CONFIG_SMP |
917b627d | 2229 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2230 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2231 | #endif |
917b627d | 2232 | |
476d139c | 2233 | put_cpu(); |
aab03e05 | 2234 | return 0; |
1da177e4 LT |
2235 | } |
2236 | ||
332ac17e DF |
2237 | unsigned long to_ratio(u64 period, u64 runtime) |
2238 | { | |
2239 | if (runtime == RUNTIME_INF) | |
2240 | return 1ULL << 20; | |
2241 | ||
2242 | /* | |
2243 | * Doing this here saves a lot of checks in all | |
2244 | * the calling paths, and returning zero seems | |
2245 | * safe for them anyway. | |
2246 | */ | |
2247 | if (period == 0) | |
2248 | return 0; | |
2249 | ||
2250 | return div64_u64(runtime << 20, period); | |
2251 | } | |
2252 | ||
2253 | #ifdef CONFIG_SMP | |
2254 | inline struct dl_bw *dl_bw_of(int i) | |
2255 | { | |
f78f5b90 PM |
2256 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2257 | "sched RCU must be held"); | |
332ac17e DF |
2258 | return &cpu_rq(i)->rd->dl_bw; |
2259 | } | |
2260 | ||
de212f18 | 2261 | static inline int dl_bw_cpus(int i) |
332ac17e | 2262 | { |
de212f18 PZ |
2263 | struct root_domain *rd = cpu_rq(i)->rd; |
2264 | int cpus = 0; | |
2265 | ||
f78f5b90 PM |
2266 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2267 | "sched RCU must be held"); | |
de212f18 PZ |
2268 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2269 | cpus++; | |
2270 | ||
2271 | return cpus; | |
332ac17e DF |
2272 | } |
2273 | #else | |
2274 | inline struct dl_bw *dl_bw_of(int i) | |
2275 | { | |
2276 | return &cpu_rq(i)->dl.dl_bw; | |
2277 | } | |
2278 | ||
de212f18 | 2279 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2280 | { |
2281 | return 1; | |
2282 | } | |
2283 | #endif | |
2284 | ||
332ac17e DF |
2285 | /* |
2286 | * We must be sure that accepting a new task (or allowing changing the | |
2287 | * parameters of an existing one) is consistent with the bandwidth | |
2288 | * constraints. If yes, this function also accordingly updates the currently | |
2289 | * allocated bandwidth to reflect the new situation. | |
2290 | * | |
2291 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2292 | * |
2293 | * XXX we should delay bw change until the task's 0-lag point, see | |
2294 | * __setparam_dl(). | |
332ac17e DF |
2295 | */ |
2296 | static int dl_overflow(struct task_struct *p, int policy, | |
2297 | const struct sched_attr *attr) | |
2298 | { | |
2299 | ||
2300 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2301 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2302 | u64 runtime = attr->sched_runtime; |
2303 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2304 | int cpus, err = -1; |
332ac17e DF |
2305 | |
2306 | if (new_bw == p->dl.dl_bw) | |
2307 | return 0; | |
2308 | ||
2309 | /* | |
2310 | * Either if a task, enters, leave, or stays -deadline but changes | |
2311 | * its parameters, we may need to update accordingly the total | |
2312 | * allocated bandwidth of the container. | |
2313 | */ | |
2314 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2315 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2316 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2317 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2318 | __dl_add(dl_b, new_bw); | |
2319 | err = 0; | |
2320 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2321 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2322 | __dl_clear(dl_b, p->dl.dl_bw); | |
2323 | __dl_add(dl_b, new_bw); | |
2324 | err = 0; | |
2325 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2326 | __dl_clear(dl_b, p->dl.dl_bw); | |
2327 | err = 0; | |
2328 | } | |
2329 | raw_spin_unlock(&dl_b->lock); | |
2330 | ||
2331 | return err; | |
2332 | } | |
2333 | ||
2334 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2335 | ||
1da177e4 LT |
2336 | /* |
2337 | * wake_up_new_task - wake up a newly created task for the first time. | |
2338 | * | |
2339 | * This function will do some initial scheduler statistics housekeeping | |
2340 | * that must be done for every newly created context, then puts the task | |
2341 | * on the runqueue and wakes it. | |
2342 | */ | |
3e51e3ed | 2343 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2344 | { |
2345 | unsigned long flags; | |
dd41f596 | 2346 | struct rq *rq; |
fabf318e | 2347 | |
ab2515c4 | 2348 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2349 | #ifdef CONFIG_SMP |
2350 | /* | |
2351 | * Fork balancing, do it here and not earlier because: | |
2352 | * - cpus_allowed can change in the fork path | |
2353 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2354 | */ |
ac66f547 | 2355 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2356 | #endif |
2357 | ||
a75cdaa9 | 2358 | /* Initialize new task's runnable average */ |
540247fb | 2359 | init_entity_runnable_average(&p->se); |
ab2515c4 | 2360 | rq = __task_rq_lock(p); |
cd29fe6f | 2361 | activate_task(rq, p, 0); |
da0c1e65 | 2362 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2363 | trace_sched_wakeup_new(p); |
a7558e01 | 2364 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2365 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2366 | if (p->sched_class->task_woken) |
2367 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2368 | #endif |
0122ec5b | 2369 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2370 | } |
2371 | ||
e107be36 AK |
2372 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2373 | ||
1cde2930 PZ |
2374 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2375 | ||
2ecd9d29 PZ |
2376 | void preempt_notifier_inc(void) |
2377 | { | |
2378 | static_key_slow_inc(&preempt_notifier_key); | |
2379 | } | |
2380 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2381 | ||
2382 | void preempt_notifier_dec(void) | |
2383 | { | |
2384 | static_key_slow_dec(&preempt_notifier_key); | |
2385 | } | |
2386 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2387 | ||
e107be36 | 2388 | /** |
80dd99b3 | 2389 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2390 | * @notifier: notifier struct to register |
e107be36 AK |
2391 | */ |
2392 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2393 | { | |
2ecd9d29 PZ |
2394 | if (!static_key_false(&preempt_notifier_key)) |
2395 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2396 | ||
e107be36 AK |
2397 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2398 | } | |
2399 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2400 | ||
2401 | /** | |
2402 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2403 | * @notifier: notifier struct to unregister |
e107be36 | 2404 | * |
d84525a8 | 2405 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2406 | */ |
2407 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2408 | { | |
2409 | hlist_del(¬ifier->link); | |
2410 | } | |
2411 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2412 | ||
1cde2930 | 2413 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2414 | { |
2415 | struct preempt_notifier *notifier; | |
e107be36 | 2416 | |
b67bfe0d | 2417 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2418 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2419 | } | |
2420 | ||
1cde2930 PZ |
2421 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2422 | { | |
2423 | if (static_key_false(&preempt_notifier_key)) | |
2424 | __fire_sched_in_preempt_notifiers(curr); | |
2425 | } | |
2426 | ||
e107be36 | 2427 | static void |
1cde2930 PZ |
2428 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2429 | struct task_struct *next) | |
e107be36 AK |
2430 | { |
2431 | struct preempt_notifier *notifier; | |
e107be36 | 2432 | |
b67bfe0d | 2433 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2434 | notifier->ops->sched_out(notifier, next); |
2435 | } | |
2436 | ||
1cde2930 PZ |
2437 | static __always_inline void |
2438 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2439 | struct task_struct *next) | |
2440 | { | |
2441 | if (static_key_false(&preempt_notifier_key)) | |
2442 | __fire_sched_out_preempt_notifiers(curr, next); | |
2443 | } | |
2444 | ||
6d6bc0ad | 2445 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2446 | |
1cde2930 | 2447 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2448 | { |
2449 | } | |
2450 | ||
1cde2930 | 2451 | static inline void |
e107be36 AK |
2452 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2453 | struct task_struct *next) | |
2454 | { | |
2455 | } | |
2456 | ||
6d6bc0ad | 2457 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2458 | |
4866cde0 NP |
2459 | /** |
2460 | * prepare_task_switch - prepare to switch tasks | |
2461 | * @rq: the runqueue preparing to switch | |
421cee29 | 2462 | * @prev: the current task that is being switched out |
4866cde0 NP |
2463 | * @next: the task we are going to switch to. |
2464 | * | |
2465 | * This is called with the rq lock held and interrupts off. It must | |
2466 | * be paired with a subsequent finish_task_switch after the context | |
2467 | * switch. | |
2468 | * | |
2469 | * prepare_task_switch sets up locking and calls architecture specific | |
2470 | * hooks. | |
2471 | */ | |
e107be36 AK |
2472 | static inline void |
2473 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2474 | struct task_struct *next) | |
4866cde0 | 2475 | { |
895dd92c | 2476 | trace_sched_switch(prev, next); |
43148951 | 2477 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2478 | perf_event_task_sched_out(prev, next); |
e107be36 | 2479 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2480 | prepare_lock_switch(rq, next); |
2481 | prepare_arch_switch(next); | |
2482 | } | |
2483 | ||
1da177e4 LT |
2484 | /** |
2485 | * finish_task_switch - clean up after a task-switch | |
2486 | * @prev: the thread we just switched away from. | |
2487 | * | |
4866cde0 NP |
2488 | * finish_task_switch must be called after the context switch, paired |
2489 | * with a prepare_task_switch call before the context switch. | |
2490 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2491 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2492 | * |
2493 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2494 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2495 | * with the lock held can cause deadlocks; see schedule() for |
2496 | * details.) | |
dfa50b60 ON |
2497 | * |
2498 | * The context switch have flipped the stack from under us and restored the | |
2499 | * local variables which were saved when this task called schedule() in the | |
2500 | * past. prev == current is still correct but we need to recalculate this_rq | |
2501 | * because prev may have moved to another CPU. | |
1da177e4 | 2502 | */ |
dfa50b60 | 2503 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2504 | __releases(rq->lock) |
2505 | { | |
dfa50b60 | 2506 | struct rq *rq = this_rq(); |
1da177e4 | 2507 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2508 | long prev_state; |
1da177e4 LT |
2509 | |
2510 | rq->prev_mm = NULL; | |
2511 | ||
2512 | /* | |
2513 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2514 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2515 | * schedule one last time. The schedule call will never return, and |
2516 | * the scheduled task must drop that reference. | |
c394cc9f | 2517 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2518 | * still held, otherwise prev could be scheduled on another cpu, die |
2519 | * there before we look at prev->state, and then the reference would | |
2520 | * be dropped twice. | |
2521 | * Manfred Spraul <manfred@colorfullife.com> | |
2522 | */ | |
55a101f8 | 2523 | prev_state = prev->state; |
bf9fae9f | 2524 | vtime_task_switch(prev); |
a8d757ef | 2525 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2526 | finish_lock_switch(rq, prev); |
01f23e16 | 2527 | finish_arch_post_lock_switch(); |
e8fa1362 | 2528 | |
e107be36 | 2529 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2530 | if (mm) |
2531 | mmdrop(mm); | |
c394cc9f | 2532 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2533 | if (prev->sched_class->task_dead) |
2534 | prev->sched_class->task_dead(prev); | |
2535 | ||
c6fd91f0 | 2536 | /* |
2537 | * Remove function-return probe instances associated with this | |
2538 | * task and put them back on the free list. | |
9761eea8 | 2539 | */ |
c6fd91f0 | 2540 | kprobe_flush_task(prev); |
1da177e4 | 2541 | put_task_struct(prev); |
c6fd91f0 | 2542 | } |
99e5ada9 | 2543 | |
de734f89 | 2544 | tick_nohz_task_switch(); |
dfa50b60 | 2545 | return rq; |
1da177e4 LT |
2546 | } |
2547 | ||
3f029d3c GH |
2548 | #ifdef CONFIG_SMP |
2549 | ||
3f029d3c | 2550 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2551 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2552 | { |
e3fca9e7 PZ |
2553 | struct callback_head *head, *next; |
2554 | void (*func)(struct rq *rq); | |
2555 | unsigned long flags; | |
3f029d3c | 2556 | |
e3fca9e7 PZ |
2557 | raw_spin_lock_irqsave(&rq->lock, flags); |
2558 | head = rq->balance_callback; | |
2559 | rq->balance_callback = NULL; | |
2560 | while (head) { | |
2561 | func = (void (*)(struct rq *))head->func; | |
2562 | next = head->next; | |
2563 | head->next = NULL; | |
2564 | head = next; | |
3f029d3c | 2565 | |
e3fca9e7 | 2566 | func(rq); |
3f029d3c | 2567 | } |
e3fca9e7 PZ |
2568 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2569 | } | |
2570 | ||
2571 | static inline void balance_callback(struct rq *rq) | |
2572 | { | |
2573 | if (unlikely(rq->balance_callback)) | |
2574 | __balance_callback(rq); | |
3f029d3c GH |
2575 | } |
2576 | ||
2577 | #else | |
da19ab51 | 2578 | |
e3fca9e7 | 2579 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2580 | { |
1da177e4 LT |
2581 | } |
2582 | ||
3f029d3c GH |
2583 | #endif |
2584 | ||
1da177e4 LT |
2585 | /** |
2586 | * schedule_tail - first thing a freshly forked thread must call. | |
2587 | * @prev: the thread we just switched away from. | |
2588 | */ | |
722a9f92 | 2589 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2590 | __releases(rq->lock) |
2591 | { | |
1a43a14a | 2592 | struct rq *rq; |
da19ab51 | 2593 | |
1a43a14a ON |
2594 | /* finish_task_switch() drops rq->lock and enables preemtion */ |
2595 | preempt_disable(); | |
dfa50b60 | 2596 | rq = finish_task_switch(prev); |
e3fca9e7 | 2597 | balance_callback(rq); |
1a43a14a | 2598 | preempt_enable(); |
70b97a7f | 2599 | |
1da177e4 | 2600 | if (current->set_child_tid) |
b488893a | 2601 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2602 | } |
2603 | ||
2604 | /* | |
dfa50b60 | 2605 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2606 | */ |
dfa50b60 | 2607 | static inline struct rq * |
70b97a7f | 2608 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2609 | struct task_struct *next) |
1da177e4 | 2610 | { |
dd41f596 | 2611 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2612 | |
e107be36 | 2613 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2614 | |
dd41f596 IM |
2615 | mm = next->mm; |
2616 | oldmm = prev->active_mm; | |
9226d125 ZA |
2617 | /* |
2618 | * For paravirt, this is coupled with an exit in switch_to to | |
2619 | * combine the page table reload and the switch backend into | |
2620 | * one hypercall. | |
2621 | */ | |
224101ed | 2622 | arch_start_context_switch(prev); |
9226d125 | 2623 | |
31915ab4 | 2624 | if (!mm) { |
1da177e4 LT |
2625 | next->active_mm = oldmm; |
2626 | atomic_inc(&oldmm->mm_count); | |
2627 | enter_lazy_tlb(oldmm, next); | |
2628 | } else | |
2629 | switch_mm(oldmm, mm, next); | |
2630 | ||
31915ab4 | 2631 | if (!prev->mm) { |
1da177e4 | 2632 | prev->active_mm = NULL; |
1da177e4 LT |
2633 | rq->prev_mm = oldmm; |
2634 | } | |
3a5f5e48 IM |
2635 | /* |
2636 | * Since the runqueue lock will be released by the next | |
2637 | * task (which is an invalid locking op but in the case | |
2638 | * of the scheduler it's an obvious special-case), so we | |
2639 | * do an early lockdep release here: | |
2640 | */ | |
cbce1a68 | 2641 | lockdep_unpin_lock(&rq->lock); |
8a25d5de | 2642 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2643 | |
2644 | /* Here we just switch the register state and the stack. */ | |
2645 | switch_to(prev, next, prev); | |
dd41f596 | 2646 | barrier(); |
dfa50b60 ON |
2647 | |
2648 | return finish_task_switch(prev); | |
1da177e4 LT |
2649 | } |
2650 | ||
2651 | /* | |
1c3e8264 | 2652 | * nr_running and nr_context_switches: |
1da177e4 LT |
2653 | * |
2654 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2655 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2656 | */ |
2657 | unsigned long nr_running(void) | |
2658 | { | |
2659 | unsigned long i, sum = 0; | |
2660 | ||
2661 | for_each_online_cpu(i) | |
2662 | sum += cpu_rq(i)->nr_running; | |
2663 | ||
2664 | return sum; | |
f711f609 | 2665 | } |
1da177e4 | 2666 | |
2ee507c4 TC |
2667 | /* |
2668 | * Check if only the current task is running on the cpu. | |
2669 | */ | |
2670 | bool single_task_running(void) | |
2671 | { | |
2672 | if (cpu_rq(smp_processor_id())->nr_running == 1) | |
2673 | return true; | |
2674 | else | |
2675 | return false; | |
2676 | } | |
2677 | EXPORT_SYMBOL(single_task_running); | |
2678 | ||
1da177e4 | 2679 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2680 | { |
cc94abfc SR |
2681 | int i; |
2682 | unsigned long long sum = 0; | |
46cb4b7c | 2683 | |
0a945022 | 2684 | for_each_possible_cpu(i) |
1da177e4 | 2685 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2686 | |
1da177e4 LT |
2687 | return sum; |
2688 | } | |
483b4ee6 | 2689 | |
1da177e4 LT |
2690 | unsigned long nr_iowait(void) |
2691 | { | |
2692 | unsigned long i, sum = 0; | |
483b4ee6 | 2693 | |
0a945022 | 2694 | for_each_possible_cpu(i) |
1da177e4 | 2695 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2696 | |
1da177e4 LT |
2697 | return sum; |
2698 | } | |
483b4ee6 | 2699 | |
8c215bd3 | 2700 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2701 | { |
8c215bd3 | 2702 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2703 | return atomic_read(&this->nr_iowait); |
2704 | } | |
46cb4b7c | 2705 | |
372ba8cb MG |
2706 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2707 | { | |
3289bdb4 PZ |
2708 | struct rq *rq = this_rq(); |
2709 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2710 | *load = rq->load.weight; | |
372ba8cb MG |
2711 | } |
2712 | ||
dd41f596 | 2713 | #ifdef CONFIG_SMP |
8a0be9ef | 2714 | |
46cb4b7c | 2715 | /* |
38022906 PZ |
2716 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2717 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2718 | */ |
38022906 | 2719 | void sched_exec(void) |
46cb4b7c | 2720 | { |
38022906 | 2721 | struct task_struct *p = current; |
1da177e4 | 2722 | unsigned long flags; |
0017d735 | 2723 | int dest_cpu; |
46cb4b7c | 2724 | |
8f42ced9 | 2725 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2726 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2727 | if (dest_cpu == smp_processor_id()) |
2728 | goto unlock; | |
38022906 | 2729 | |
8f42ced9 | 2730 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2731 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2732 | |
8f42ced9 PZ |
2733 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2734 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2735 | return; |
2736 | } | |
0017d735 | 2737 | unlock: |
8f42ced9 | 2738 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2739 | } |
dd41f596 | 2740 | |
1da177e4 LT |
2741 | #endif |
2742 | ||
1da177e4 | 2743 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2744 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2745 | |
2746 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2747 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2748 | |
c5f8d995 HS |
2749 | /* |
2750 | * Return accounted runtime for the task. | |
2751 | * In case the task is currently running, return the runtime plus current's | |
2752 | * pending runtime that have not been accounted yet. | |
2753 | */ | |
2754 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2755 | { | |
2756 | unsigned long flags; | |
2757 | struct rq *rq; | |
6e998916 | 2758 | u64 ns; |
c5f8d995 | 2759 | |
911b2898 PZ |
2760 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2761 | /* | |
2762 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2763 | * So we have a optimization chance when the task's delta_exec is 0. | |
2764 | * Reading ->on_cpu is racy, but this is ok. | |
2765 | * | |
2766 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2767 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2768 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2769 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2770 | * been accounted, so we're correct here as well. | |
911b2898 | 2771 | */ |
da0c1e65 | 2772 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2773 | return p->se.sum_exec_runtime; |
2774 | #endif | |
2775 | ||
c5f8d995 | 2776 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2777 | /* |
2778 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2779 | * project cycles that may never be accounted to this | |
2780 | * thread, breaking clock_gettime(). | |
2781 | */ | |
2782 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2783 | update_rq_clock(rq); | |
2784 | p->sched_class->update_curr(rq); | |
2785 | } | |
2786 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2787 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2788 | |
2789 | return ns; | |
2790 | } | |
48f24c4d | 2791 | |
7835b98b CL |
2792 | /* |
2793 | * This function gets called by the timer code, with HZ frequency. | |
2794 | * We call it with interrupts disabled. | |
7835b98b CL |
2795 | */ |
2796 | void scheduler_tick(void) | |
2797 | { | |
7835b98b CL |
2798 | int cpu = smp_processor_id(); |
2799 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2800 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2801 | |
2802 | sched_clock_tick(); | |
dd41f596 | 2803 | |
05fa785c | 2804 | raw_spin_lock(&rq->lock); |
3e51f33f | 2805 | update_rq_clock(rq); |
fa85ae24 | 2806 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2807 | update_cpu_load_active(rq); |
3289bdb4 | 2808 | calc_global_load_tick(rq); |
05fa785c | 2809 | raw_spin_unlock(&rq->lock); |
7835b98b | 2810 | |
e9d2b064 | 2811 | perf_event_task_tick(); |
e220d2dc | 2812 | |
e418e1c2 | 2813 | #ifdef CONFIG_SMP |
6eb57e0d | 2814 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2815 | trigger_load_balance(rq); |
e418e1c2 | 2816 | #endif |
265f22a9 | 2817 | rq_last_tick_reset(rq); |
1da177e4 LT |
2818 | } |
2819 | ||
265f22a9 FW |
2820 | #ifdef CONFIG_NO_HZ_FULL |
2821 | /** | |
2822 | * scheduler_tick_max_deferment | |
2823 | * | |
2824 | * Keep at least one tick per second when a single | |
2825 | * active task is running because the scheduler doesn't | |
2826 | * yet completely support full dynticks environment. | |
2827 | * | |
2828 | * This makes sure that uptime, CFS vruntime, load | |
2829 | * balancing, etc... continue to move forward, even | |
2830 | * with a very low granularity. | |
e69f6186 YB |
2831 | * |
2832 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2833 | */ |
2834 | u64 scheduler_tick_max_deferment(void) | |
2835 | { | |
2836 | struct rq *rq = this_rq(); | |
316c1608 | 2837 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
2838 | |
2839 | next = rq->last_sched_tick + HZ; | |
2840 | ||
2841 | if (time_before_eq(next, now)) | |
2842 | return 0; | |
2843 | ||
8fe8ff09 | 2844 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2845 | } |
265f22a9 | 2846 | #endif |
1da177e4 | 2847 | |
132380a0 | 2848 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2849 | { |
2850 | if (in_lock_functions(addr)) { | |
2851 | addr = CALLER_ADDR2; | |
2852 | if (in_lock_functions(addr)) | |
2853 | addr = CALLER_ADDR3; | |
2854 | } | |
2855 | return addr; | |
2856 | } | |
1da177e4 | 2857 | |
7e49fcce SR |
2858 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2859 | defined(CONFIG_PREEMPT_TRACER)) | |
2860 | ||
edafe3a5 | 2861 | void preempt_count_add(int val) |
1da177e4 | 2862 | { |
6cd8a4bb | 2863 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2864 | /* |
2865 | * Underflow? | |
2866 | */ | |
9a11b49a IM |
2867 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2868 | return; | |
6cd8a4bb | 2869 | #endif |
bdb43806 | 2870 | __preempt_count_add(val); |
6cd8a4bb | 2871 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2872 | /* |
2873 | * Spinlock count overflowing soon? | |
2874 | */ | |
33859f7f MOS |
2875 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2876 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2877 | #endif |
8f47b187 TG |
2878 | if (preempt_count() == val) { |
2879 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2880 | #ifdef CONFIG_DEBUG_PREEMPT | |
2881 | current->preempt_disable_ip = ip; | |
2882 | #endif | |
2883 | trace_preempt_off(CALLER_ADDR0, ip); | |
2884 | } | |
1da177e4 | 2885 | } |
bdb43806 | 2886 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2887 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2888 | |
edafe3a5 | 2889 | void preempt_count_sub(int val) |
1da177e4 | 2890 | { |
6cd8a4bb | 2891 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2892 | /* |
2893 | * Underflow? | |
2894 | */ | |
01e3eb82 | 2895 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2896 | return; |
1da177e4 LT |
2897 | /* |
2898 | * Is the spinlock portion underflowing? | |
2899 | */ | |
9a11b49a IM |
2900 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2901 | !(preempt_count() & PREEMPT_MASK))) | |
2902 | return; | |
6cd8a4bb | 2903 | #endif |
9a11b49a | 2904 | |
6cd8a4bb SR |
2905 | if (preempt_count() == val) |
2906 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2907 | __preempt_count_sub(val); |
1da177e4 | 2908 | } |
bdb43806 | 2909 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2910 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2911 | |
2912 | #endif | |
2913 | ||
2914 | /* | |
dd41f596 | 2915 | * Print scheduling while atomic bug: |
1da177e4 | 2916 | */ |
dd41f596 | 2917 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2918 | { |
664dfa65 DJ |
2919 | if (oops_in_progress) |
2920 | return; | |
2921 | ||
3df0fc5b PZ |
2922 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2923 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2924 | |
dd41f596 | 2925 | debug_show_held_locks(prev); |
e21f5b15 | 2926 | print_modules(); |
dd41f596 IM |
2927 | if (irqs_disabled()) |
2928 | print_irqtrace_events(prev); | |
8f47b187 TG |
2929 | #ifdef CONFIG_DEBUG_PREEMPT |
2930 | if (in_atomic_preempt_off()) { | |
2931 | pr_err("Preemption disabled at:"); | |
2932 | print_ip_sym(current->preempt_disable_ip); | |
2933 | pr_cont("\n"); | |
2934 | } | |
2935 | #endif | |
6135fc1e | 2936 | dump_stack(); |
373d4d09 | 2937 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2938 | } |
1da177e4 | 2939 | |
dd41f596 IM |
2940 | /* |
2941 | * Various schedule()-time debugging checks and statistics: | |
2942 | */ | |
2943 | static inline void schedule_debug(struct task_struct *prev) | |
2944 | { | |
0d9e2632 AT |
2945 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
2946 | BUG_ON(unlikely(task_stack_end_corrupted(prev))); | |
2947 | #endif | |
1da177e4 | 2948 | /* |
41a2d6cf | 2949 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2950 | * schedule() atomically, we ignore that path. Otherwise whine |
2951 | * if we are scheduling when we should not. | |
1da177e4 | 2952 | */ |
192301e7 | 2953 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2954 | __schedule_bug(prev); |
b3fbab05 | 2955 | rcu_sleep_check(); |
dd41f596 | 2956 | |
1da177e4 LT |
2957 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2958 | ||
2d72376b | 2959 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2960 | } |
2961 | ||
2962 | /* | |
2963 | * Pick up the highest-prio task: | |
2964 | */ | |
2965 | static inline struct task_struct * | |
606dba2e | 2966 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2967 | { |
37e117c0 | 2968 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2969 | struct task_struct *p; |
1da177e4 LT |
2970 | |
2971 | /* | |
dd41f596 IM |
2972 | * Optimization: we know that if all tasks are in |
2973 | * the fair class we can call that function directly: | |
1da177e4 | 2974 | */ |
37e117c0 | 2975 | if (likely(prev->sched_class == class && |
38033c37 | 2976 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2977 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2978 | if (unlikely(p == RETRY_TASK)) |
2979 | goto again; | |
2980 | ||
2981 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2982 | if (unlikely(!p)) | |
2983 | p = idle_sched_class.pick_next_task(rq, prev); | |
2984 | ||
2985 | return p; | |
1da177e4 LT |
2986 | } |
2987 | ||
37e117c0 | 2988 | again: |
34f971f6 | 2989 | for_each_class(class) { |
606dba2e | 2990 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2991 | if (p) { |
2992 | if (unlikely(p == RETRY_TASK)) | |
2993 | goto again; | |
dd41f596 | 2994 | return p; |
37e117c0 | 2995 | } |
dd41f596 | 2996 | } |
34f971f6 PZ |
2997 | |
2998 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2999 | } |
1da177e4 | 3000 | |
dd41f596 | 3001 | /* |
c259e01a | 3002 | * __schedule() is the main scheduler function. |
edde96ea PE |
3003 | * |
3004 | * The main means of driving the scheduler and thus entering this function are: | |
3005 | * | |
3006 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3007 | * | |
3008 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3009 | * paths. For example, see arch/x86/entry_64.S. | |
3010 | * | |
3011 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3012 | * interrupt handler scheduler_tick(). | |
3013 | * | |
3014 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3015 | * task to the run-queue and that's it. | |
3016 | * | |
3017 | * Now, if the new task added to the run-queue preempts the current | |
3018 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3019 | * called on the nearest possible occasion: | |
3020 | * | |
3021 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3022 | * | |
3023 | * - in syscall or exception context, at the next outmost | |
3024 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3025 | * spin_unlock()!) | |
3026 | * | |
3027 | * - in IRQ context, return from interrupt-handler to | |
3028 | * preemptible context | |
3029 | * | |
3030 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3031 | * then at the next: | |
3032 | * | |
3033 | * - cond_resched() call | |
3034 | * - explicit schedule() call | |
3035 | * - return from syscall or exception to user-space | |
3036 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3037 | * |
b30f0e3f | 3038 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3039 | */ |
c259e01a | 3040 | static void __sched __schedule(void) |
dd41f596 IM |
3041 | { |
3042 | struct task_struct *prev, *next; | |
67ca7bde | 3043 | unsigned long *switch_count; |
dd41f596 | 3044 | struct rq *rq; |
31656519 | 3045 | int cpu; |
dd41f596 | 3046 | |
dd41f596 IM |
3047 | cpu = smp_processor_id(); |
3048 | rq = cpu_rq(cpu); | |
38200cf2 | 3049 | rcu_note_context_switch(); |
dd41f596 | 3050 | prev = rq->curr; |
dd41f596 | 3051 | |
dd41f596 | 3052 | schedule_debug(prev); |
1da177e4 | 3053 | |
31656519 | 3054 | if (sched_feat(HRTICK)) |
f333fdc9 | 3055 | hrtick_clear(rq); |
8f4d37ec | 3056 | |
e0acd0a6 ON |
3057 | /* |
3058 | * Make sure that signal_pending_state()->signal_pending() below | |
3059 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3060 | * done by the caller to avoid the race with signal_wake_up(). | |
3061 | */ | |
3062 | smp_mb__before_spinlock(); | |
05fa785c | 3063 | raw_spin_lock_irq(&rq->lock); |
cbce1a68 | 3064 | lockdep_pin_lock(&rq->lock); |
1da177e4 | 3065 | |
9edfbfed PZ |
3066 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3067 | ||
246d86b5 | 3068 | switch_count = &prev->nivcsw; |
1da177e4 | 3069 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3070 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3071 | prev->state = TASK_RUNNING; |
21aa9af0 | 3072 | } else { |
2acca55e PZ |
3073 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3074 | prev->on_rq = 0; | |
3075 | ||
21aa9af0 | 3076 | /* |
2acca55e PZ |
3077 | * If a worker went to sleep, notify and ask workqueue |
3078 | * whether it wants to wake up a task to maintain | |
3079 | * concurrency. | |
21aa9af0 TH |
3080 | */ |
3081 | if (prev->flags & PF_WQ_WORKER) { | |
3082 | struct task_struct *to_wakeup; | |
3083 | ||
3084 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3085 | if (to_wakeup) | |
3086 | try_to_wake_up_local(to_wakeup); | |
3087 | } | |
21aa9af0 | 3088 | } |
dd41f596 | 3089 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3090 | } |
3091 | ||
9edfbfed | 3092 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3093 | update_rq_clock(rq); |
3094 | ||
3095 | next = pick_next_task(rq, prev); | |
f26f9aff | 3096 | clear_tsk_need_resched(prev); |
f27dde8d | 3097 | clear_preempt_need_resched(); |
9edfbfed | 3098 | rq->clock_skip_update = 0; |
1da177e4 | 3099 | |
1da177e4 | 3100 | if (likely(prev != next)) { |
1da177e4 LT |
3101 | rq->nr_switches++; |
3102 | rq->curr = next; | |
3103 | ++*switch_count; | |
3104 | ||
dfa50b60 ON |
3105 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
3106 | cpu = cpu_of(rq); | |
cbce1a68 PZ |
3107 | } else { |
3108 | lockdep_unpin_lock(&rq->lock); | |
05fa785c | 3109 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3110 | } |
1da177e4 | 3111 | |
e3fca9e7 | 3112 | balance_callback(rq); |
1da177e4 | 3113 | } |
c259e01a | 3114 | |
9c40cef2 TG |
3115 | static inline void sched_submit_work(struct task_struct *tsk) |
3116 | { | |
3c7d5184 | 3117 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3118 | return; |
3119 | /* | |
3120 | * If we are going to sleep and we have plugged IO queued, | |
3121 | * make sure to submit it to avoid deadlocks. | |
3122 | */ | |
3123 | if (blk_needs_flush_plug(tsk)) | |
3124 | blk_schedule_flush_plug(tsk); | |
3125 | } | |
3126 | ||
722a9f92 | 3127 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3128 | { |
9c40cef2 TG |
3129 | struct task_struct *tsk = current; |
3130 | ||
3131 | sched_submit_work(tsk); | |
bfd9b2b5 | 3132 | do { |
b30f0e3f | 3133 | preempt_disable(); |
bfd9b2b5 | 3134 | __schedule(); |
b30f0e3f | 3135 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3136 | } while (need_resched()); |
c259e01a | 3137 | } |
1da177e4 LT |
3138 | EXPORT_SYMBOL(schedule); |
3139 | ||
91d1aa43 | 3140 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3141 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3142 | { |
3143 | /* | |
3144 | * If we come here after a random call to set_need_resched(), | |
3145 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3146 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3147 | * we find a better solution. | |
7cc78f8f AL |
3148 | * |
3149 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3150 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3151 | * too frequently to make sense yet. |
20ab65e3 | 3152 | */ |
7cc78f8f | 3153 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3154 | schedule(); |
7cc78f8f | 3155 | exception_exit(prev_state); |
20ab65e3 FW |
3156 | } |
3157 | #endif | |
3158 | ||
c5491ea7 TG |
3159 | /** |
3160 | * schedule_preempt_disabled - called with preemption disabled | |
3161 | * | |
3162 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3163 | */ | |
3164 | void __sched schedule_preempt_disabled(void) | |
3165 | { | |
ba74c144 | 3166 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3167 | schedule(); |
3168 | preempt_disable(); | |
3169 | } | |
3170 | ||
06b1f808 | 3171 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3172 | { |
3173 | do { | |
b30f0e3f | 3174 | preempt_active_enter(); |
a18b5d01 | 3175 | __schedule(); |
b30f0e3f | 3176 | preempt_active_exit(); |
a18b5d01 FW |
3177 | |
3178 | /* | |
3179 | * Check again in case we missed a preemption opportunity | |
3180 | * between schedule and now. | |
3181 | */ | |
a18b5d01 FW |
3182 | } while (need_resched()); |
3183 | } | |
3184 | ||
1da177e4 LT |
3185 | #ifdef CONFIG_PREEMPT |
3186 | /* | |
2ed6e34f | 3187 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3188 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3189 | * occur there and call schedule directly. |
3190 | */ | |
722a9f92 | 3191 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3192 | { |
1da177e4 LT |
3193 | /* |
3194 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3195 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3196 | */ |
fbb00b56 | 3197 | if (likely(!preemptible())) |
1da177e4 LT |
3198 | return; |
3199 | ||
a18b5d01 | 3200 | preempt_schedule_common(); |
1da177e4 | 3201 | } |
376e2424 | 3202 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3203 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3204 | |
009f60e2 | 3205 | /** |
4eaca0a8 | 3206 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3207 | * |
3208 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3209 | * recursion and tracing preempt enabling caused by the tracing | |
3210 | * infrastructure itself. But as tracing can happen in areas coming | |
3211 | * from userspace or just about to enter userspace, a preempt enable | |
3212 | * can occur before user_exit() is called. This will cause the scheduler | |
3213 | * to be called when the system is still in usermode. | |
3214 | * | |
3215 | * To prevent this, the preempt_enable_notrace will use this function | |
3216 | * instead of preempt_schedule() to exit user context if needed before | |
3217 | * calling the scheduler. | |
3218 | */ | |
4eaca0a8 | 3219 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3220 | { |
3221 | enum ctx_state prev_ctx; | |
3222 | ||
3223 | if (likely(!preemptible())) | |
3224 | return; | |
3225 | ||
3226 | do { | |
be690035 FW |
3227 | /* |
3228 | * Use raw __prempt_count() ops that don't call function. | |
3229 | * We can't call functions before disabling preemption which | |
3230 | * disarm preemption tracing recursions. | |
3231 | */ | |
3232 | __preempt_count_add(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); | |
3233 | barrier(); | |
009f60e2 ON |
3234 | /* |
3235 | * Needs preempt disabled in case user_exit() is traced | |
3236 | * and the tracer calls preempt_enable_notrace() causing | |
3237 | * an infinite recursion. | |
3238 | */ | |
3239 | prev_ctx = exception_enter(); | |
3240 | __schedule(); | |
3241 | exception_exit(prev_ctx); | |
3242 | ||
009f60e2 | 3243 | barrier(); |
be690035 | 3244 | __preempt_count_sub(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); |
009f60e2 ON |
3245 | } while (need_resched()); |
3246 | } | |
4eaca0a8 | 3247 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3248 | |
32e475d7 | 3249 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3250 | |
3251 | /* | |
2ed6e34f | 3252 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3253 | * off of irq context. |
3254 | * Note, that this is called and return with irqs disabled. This will | |
3255 | * protect us against recursive calling from irq. | |
3256 | */ | |
722a9f92 | 3257 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3258 | { |
b22366cd | 3259 | enum ctx_state prev_state; |
6478d880 | 3260 | |
2ed6e34f | 3261 | /* Catch callers which need to be fixed */ |
f27dde8d | 3262 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3263 | |
b22366cd FW |
3264 | prev_state = exception_enter(); |
3265 | ||
3a5c359a | 3266 | do { |
b30f0e3f | 3267 | preempt_active_enter(); |
3a5c359a | 3268 | local_irq_enable(); |
c259e01a | 3269 | __schedule(); |
3a5c359a | 3270 | local_irq_disable(); |
b30f0e3f | 3271 | preempt_active_exit(); |
5ed0cec0 | 3272 | } while (need_resched()); |
b22366cd FW |
3273 | |
3274 | exception_exit(prev_state); | |
1da177e4 LT |
3275 | } |
3276 | ||
63859d4f | 3277 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3278 | void *key) |
1da177e4 | 3279 | { |
63859d4f | 3280 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3281 | } |
1da177e4 LT |
3282 | EXPORT_SYMBOL(default_wake_function); |
3283 | ||
b29739f9 IM |
3284 | #ifdef CONFIG_RT_MUTEXES |
3285 | ||
3286 | /* | |
3287 | * rt_mutex_setprio - set the current priority of a task | |
3288 | * @p: task | |
3289 | * @prio: prio value (kernel-internal form) | |
3290 | * | |
3291 | * This function changes the 'effective' priority of a task. It does | |
3292 | * not touch ->normal_prio like __setscheduler(). | |
3293 | * | |
c365c292 TG |
3294 | * Used by the rt_mutex code to implement priority inheritance |
3295 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3296 | */ |
36c8b586 | 3297 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3298 | { |
da0c1e65 | 3299 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 3300 | struct rq *rq; |
83ab0aa0 | 3301 | const struct sched_class *prev_class; |
b29739f9 | 3302 | |
aab03e05 | 3303 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3304 | |
0122ec5b | 3305 | rq = __task_rq_lock(p); |
b29739f9 | 3306 | |
1c4dd99b TG |
3307 | /* |
3308 | * Idle task boosting is a nono in general. There is one | |
3309 | * exception, when PREEMPT_RT and NOHZ is active: | |
3310 | * | |
3311 | * The idle task calls get_next_timer_interrupt() and holds | |
3312 | * the timer wheel base->lock on the CPU and another CPU wants | |
3313 | * to access the timer (probably to cancel it). We can safely | |
3314 | * ignore the boosting request, as the idle CPU runs this code | |
3315 | * with interrupts disabled and will complete the lock | |
3316 | * protected section without being interrupted. So there is no | |
3317 | * real need to boost. | |
3318 | */ | |
3319 | if (unlikely(p == rq->idle)) { | |
3320 | WARN_ON(p != rq->curr); | |
3321 | WARN_ON(p->pi_blocked_on); | |
3322 | goto out_unlock; | |
3323 | } | |
3324 | ||
a8027073 | 3325 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3326 | oldprio = p->prio; |
83ab0aa0 | 3327 | prev_class = p->sched_class; |
da0c1e65 | 3328 | queued = task_on_rq_queued(p); |
051a1d1a | 3329 | running = task_current(rq, p); |
da0c1e65 | 3330 | if (queued) |
69be72c1 | 3331 | dequeue_task(rq, p, 0); |
0e1f3483 | 3332 | if (running) |
f3cd1c4e | 3333 | put_prev_task(rq, p); |
dd41f596 | 3334 | |
2d3d891d DF |
3335 | /* |
3336 | * Boosting condition are: | |
3337 | * 1. -rt task is running and holds mutex A | |
3338 | * --> -dl task blocks on mutex A | |
3339 | * | |
3340 | * 2. -dl task is running and holds mutex A | |
3341 | * --> -dl task blocks on mutex A and could preempt the | |
3342 | * running task | |
3343 | */ | |
3344 | if (dl_prio(prio)) { | |
466af29b ON |
3345 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3346 | if (!dl_prio(p->normal_prio) || | |
3347 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3348 | p->dl.dl_boosted = 1; |
2d3d891d DF |
3349 | enqueue_flag = ENQUEUE_REPLENISH; |
3350 | } else | |
3351 | p->dl.dl_boosted = 0; | |
aab03e05 | 3352 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3353 | } else if (rt_prio(prio)) { |
3354 | if (dl_prio(oldprio)) | |
3355 | p->dl.dl_boosted = 0; | |
3356 | if (oldprio < prio) | |
3357 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3358 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3359 | } else { |
3360 | if (dl_prio(oldprio)) | |
3361 | p->dl.dl_boosted = 0; | |
746db944 BS |
3362 | if (rt_prio(oldprio)) |
3363 | p->rt.timeout = 0; | |
dd41f596 | 3364 | p->sched_class = &fair_sched_class; |
2d3d891d | 3365 | } |
dd41f596 | 3366 | |
b29739f9 IM |
3367 | p->prio = prio; |
3368 | ||
0e1f3483 HS |
3369 | if (running) |
3370 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3371 | if (queued) |
2d3d891d | 3372 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3373 | |
da7a735e | 3374 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3375 | out_unlock: |
4c9a4bc8 | 3376 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3377 | __task_rq_unlock(rq); |
4c9a4bc8 PZ |
3378 | |
3379 | balance_callback(rq); | |
3380 | preempt_enable(); | |
b29739f9 | 3381 | } |
b29739f9 | 3382 | #endif |
d50dde5a | 3383 | |
36c8b586 | 3384 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3385 | { |
da0c1e65 | 3386 | int old_prio, delta, queued; |
1da177e4 | 3387 | unsigned long flags; |
70b97a7f | 3388 | struct rq *rq; |
1da177e4 | 3389 | |
75e45d51 | 3390 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3391 | return; |
3392 | /* | |
3393 | * We have to be careful, if called from sys_setpriority(), | |
3394 | * the task might be in the middle of scheduling on another CPU. | |
3395 | */ | |
3396 | rq = task_rq_lock(p, &flags); | |
3397 | /* | |
3398 | * The RT priorities are set via sched_setscheduler(), but we still | |
3399 | * allow the 'normal' nice value to be set - but as expected | |
3400 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3401 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3402 | */ |
aab03e05 | 3403 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3404 | p->static_prio = NICE_TO_PRIO(nice); |
3405 | goto out_unlock; | |
3406 | } | |
da0c1e65 KT |
3407 | queued = task_on_rq_queued(p); |
3408 | if (queued) | |
69be72c1 | 3409 | dequeue_task(rq, p, 0); |
1da177e4 | 3410 | |
1da177e4 | 3411 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3412 | set_load_weight(p); |
b29739f9 IM |
3413 | old_prio = p->prio; |
3414 | p->prio = effective_prio(p); | |
3415 | delta = p->prio - old_prio; | |
1da177e4 | 3416 | |
da0c1e65 | 3417 | if (queued) { |
371fd7e7 | 3418 | enqueue_task(rq, p, 0); |
1da177e4 | 3419 | /* |
d5f9f942 AM |
3420 | * If the task increased its priority or is running and |
3421 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3422 | */ |
d5f9f942 | 3423 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3424 | resched_curr(rq); |
1da177e4 LT |
3425 | } |
3426 | out_unlock: | |
0122ec5b | 3427 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3428 | } |
1da177e4 LT |
3429 | EXPORT_SYMBOL(set_user_nice); |
3430 | ||
e43379f1 MM |
3431 | /* |
3432 | * can_nice - check if a task can reduce its nice value | |
3433 | * @p: task | |
3434 | * @nice: nice value | |
3435 | */ | |
36c8b586 | 3436 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3437 | { |
024f4747 | 3438 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3439 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3440 | |
78d7d407 | 3441 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3442 | capable(CAP_SYS_NICE)); |
3443 | } | |
3444 | ||
1da177e4 LT |
3445 | #ifdef __ARCH_WANT_SYS_NICE |
3446 | ||
3447 | /* | |
3448 | * sys_nice - change the priority of the current process. | |
3449 | * @increment: priority increment | |
3450 | * | |
3451 | * sys_setpriority is a more generic, but much slower function that | |
3452 | * does similar things. | |
3453 | */ | |
5add95d4 | 3454 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3455 | { |
48f24c4d | 3456 | long nice, retval; |
1da177e4 LT |
3457 | |
3458 | /* | |
3459 | * Setpriority might change our priority at the same moment. | |
3460 | * We don't have to worry. Conceptually one call occurs first | |
3461 | * and we have a single winner. | |
3462 | */ | |
a9467fa3 | 3463 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3464 | nice = task_nice(current) + increment; |
1da177e4 | 3465 | |
a9467fa3 | 3466 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3467 | if (increment < 0 && !can_nice(current, nice)) |
3468 | return -EPERM; | |
3469 | ||
1da177e4 LT |
3470 | retval = security_task_setnice(current, nice); |
3471 | if (retval) | |
3472 | return retval; | |
3473 | ||
3474 | set_user_nice(current, nice); | |
3475 | return 0; | |
3476 | } | |
3477 | ||
3478 | #endif | |
3479 | ||
3480 | /** | |
3481 | * task_prio - return the priority value of a given task. | |
3482 | * @p: the task in question. | |
3483 | * | |
e69f6186 | 3484 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3485 | * RT tasks are offset by -200. Normal tasks are centered |
3486 | * around 0, value goes from -16 to +15. | |
3487 | */ | |
36c8b586 | 3488 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3489 | { |
3490 | return p->prio - MAX_RT_PRIO; | |
3491 | } | |
3492 | ||
1da177e4 LT |
3493 | /** |
3494 | * idle_cpu - is a given cpu idle currently? | |
3495 | * @cpu: the processor in question. | |
e69f6186 YB |
3496 | * |
3497 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3498 | */ |
3499 | int idle_cpu(int cpu) | |
3500 | { | |
908a3283 TG |
3501 | struct rq *rq = cpu_rq(cpu); |
3502 | ||
3503 | if (rq->curr != rq->idle) | |
3504 | return 0; | |
3505 | ||
3506 | if (rq->nr_running) | |
3507 | return 0; | |
3508 | ||
3509 | #ifdef CONFIG_SMP | |
3510 | if (!llist_empty(&rq->wake_list)) | |
3511 | return 0; | |
3512 | #endif | |
3513 | ||
3514 | return 1; | |
1da177e4 LT |
3515 | } |
3516 | ||
1da177e4 LT |
3517 | /** |
3518 | * idle_task - return the idle task for a given cpu. | |
3519 | * @cpu: the processor in question. | |
e69f6186 YB |
3520 | * |
3521 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3522 | */ |
36c8b586 | 3523 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3524 | { |
3525 | return cpu_rq(cpu)->idle; | |
3526 | } | |
3527 | ||
3528 | /** | |
3529 | * find_process_by_pid - find a process with a matching PID value. | |
3530 | * @pid: the pid in question. | |
e69f6186 YB |
3531 | * |
3532 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3533 | */ |
a9957449 | 3534 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3535 | { |
228ebcbe | 3536 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3537 | } |
3538 | ||
aab03e05 DF |
3539 | /* |
3540 | * This function initializes the sched_dl_entity of a newly becoming | |
3541 | * SCHED_DEADLINE task. | |
3542 | * | |
3543 | * Only the static values are considered here, the actual runtime and the | |
3544 | * absolute deadline will be properly calculated when the task is enqueued | |
3545 | * for the first time with its new policy. | |
3546 | */ | |
3547 | static void | |
3548 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3549 | { | |
3550 | struct sched_dl_entity *dl_se = &p->dl; | |
3551 | ||
aab03e05 DF |
3552 | dl_se->dl_runtime = attr->sched_runtime; |
3553 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3554 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3555 | dl_se->flags = attr->sched_flags; |
332ac17e | 3556 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3557 | |
3558 | /* | |
3559 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3560 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3561 | * | |
3562 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3563 | * point. This would include retaining the task_struct until that time | |
3564 | * and change dl_overflow() to not immediately decrement the current | |
3565 | * amount. | |
3566 | * | |
3567 | * Instead we retain the current runtime/deadline and let the new | |
3568 | * parameters take effect after the current reservation period lapses. | |
3569 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3570 | * before the current scheduling deadline. | |
3571 | * | |
3572 | * We can still have temporary overloads because we do not delay the | |
3573 | * change in bandwidth until that time; so admission control is | |
3574 | * not on the safe side. It does however guarantee tasks will never | |
3575 | * consume more than promised. | |
3576 | */ | |
aab03e05 DF |
3577 | } |
3578 | ||
c13db6b1 SR |
3579 | /* |
3580 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3581 | * it calls know not to change it. | |
3582 | */ | |
3583 | #define SETPARAM_POLICY -1 | |
3584 | ||
c365c292 TG |
3585 | static void __setscheduler_params(struct task_struct *p, |
3586 | const struct sched_attr *attr) | |
1da177e4 | 3587 | { |
d50dde5a DF |
3588 | int policy = attr->sched_policy; |
3589 | ||
c13db6b1 | 3590 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3591 | policy = p->policy; |
3592 | ||
1da177e4 | 3593 | p->policy = policy; |
d50dde5a | 3594 | |
aab03e05 DF |
3595 | if (dl_policy(policy)) |
3596 | __setparam_dl(p, attr); | |
39fd8fd2 | 3597 | else if (fair_policy(policy)) |
d50dde5a DF |
3598 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3599 | ||
39fd8fd2 PZ |
3600 | /* |
3601 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3602 | * !rt_policy. Always setting this ensures that things like | |
3603 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3604 | */ | |
3605 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3606 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3607 | set_load_weight(p); |
3608 | } | |
39fd8fd2 | 3609 | |
c365c292 TG |
3610 | /* Actually do priority change: must hold pi & rq lock. */ |
3611 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3612 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3613 | { |
3614 | __setscheduler_params(p, attr); | |
d50dde5a | 3615 | |
383afd09 | 3616 | /* |
0782e63b TG |
3617 | * Keep a potential priority boosting if called from |
3618 | * sched_setscheduler(). | |
383afd09 | 3619 | */ |
0782e63b TG |
3620 | if (keep_boost) |
3621 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3622 | else | |
3623 | p->prio = normal_prio(p); | |
383afd09 | 3624 | |
aab03e05 DF |
3625 | if (dl_prio(p->prio)) |
3626 | p->sched_class = &dl_sched_class; | |
3627 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3628 | p->sched_class = &rt_sched_class; |
3629 | else | |
3630 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3631 | } |
aab03e05 DF |
3632 | |
3633 | static void | |
3634 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3635 | { | |
3636 | struct sched_dl_entity *dl_se = &p->dl; | |
3637 | ||
3638 | attr->sched_priority = p->rt_priority; | |
3639 | attr->sched_runtime = dl_se->dl_runtime; | |
3640 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3641 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3642 | attr->sched_flags = dl_se->flags; |
3643 | } | |
3644 | ||
3645 | /* | |
3646 | * This function validates the new parameters of a -deadline task. | |
3647 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3648 | * than the runtime, as well as the period of being zero or |
332ac17e | 3649 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3650 | * user parameters are above the internal resolution of 1us (we |
3651 | * check sched_runtime only since it is always the smaller one) and | |
3652 | * below 2^63 ns (we have to check both sched_deadline and | |
3653 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3654 | */ |
3655 | static bool | |
3656 | __checkparam_dl(const struct sched_attr *attr) | |
3657 | { | |
b0827819 JL |
3658 | /* deadline != 0 */ |
3659 | if (attr->sched_deadline == 0) | |
3660 | return false; | |
3661 | ||
3662 | /* | |
3663 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3664 | * that big. | |
3665 | */ | |
3666 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3667 | return false; | |
3668 | ||
3669 | /* | |
3670 | * Since we use the MSB for wrap-around and sign issues, make | |
3671 | * sure it's not set (mind that period can be equal to zero). | |
3672 | */ | |
3673 | if (attr->sched_deadline & (1ULL << 63) || | |
3674 | attr->sched_period & (1ULL << 63)) | |
3675 | return false; | |
3676 | ||
3677 | /* runtime <= deadline <= period (if period != 0) */ | |
3678 | if ((attr->sched_period != 0 && | |
3679 | attr->sched_period < attr->sched_deadline) || | |
3680 | attr->sched_deadline < attr->sched_runtime) | |
3681 | return false; | |
3682 | ||
3683 | return true; | |
aab03e05 DF |
3684 | } |
3685 | ||
c69e8d9c DH |
3686 | /* |
3687 | * check the target process has a UID that matches the current process's | |
3688 | */ | |
3689 | static bool check_same_owner(struct task_struct *p) | |
3690 | { | |
3691 | const struct cred *cred = current_cred(), *pcred; | |
3692 | bool match; | |
3693 | ||
3694 | rcu_read_lock(); | |
3695 | pcred = __task_cred(p); | |
9c806aa0 EB |
3696 | match = (uid_eq(cred->euid, pcred->euid) || |
3697 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3698 | rcu_read_unlock(); |
3699 | return match; | |
3700 | } | |
3701 | ||
75381608 WL |
3702 | static bool dl_param_changed(struct task_struct *p, |
3703 | const struct sched_attr *attr) | |
3704 | { | |
3705 | struct sched_dl_entity *dl_se = &p->dl; | |
3706 | ||
3707 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3708 | dl_se->dl_deadline != attr->sched_deadline || | |
3709 | dl_se->dl_period != attr->sched_period || | |
3710 | dl_se->flags != attr->sched_flags) | |
3711 | return true; | |
3712 | ||
3713 | return false; | |
3714 | } | |
3715 | ||
d50dde5a DF |
3716 | static int __sched_setscheduler(struct task_struct *p, |
3717 | const struct sched_attr *attr, | |
dbc7f069 | 3718 | bool user, bool pi) |
1da177e4 | 3719 | { |
383afd09 SR |
3720 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3721 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3722 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3723 | int new_effective_prio, policy = attr->sched_policy; |
1da177e4 | 3724 | unsigned long flags; |
83ab0aa0 | 3725 | const struct sched_class *prev_class; |
70b97a7f | 3726 | struct rq *rq; |
ca94c442 | 3727 | int reset_on_fork; |
1da177e4 | 3728 | |
66e5393a SR |
3729 | /* may grab non-irq protected spin_locks */ |
3730 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3731 | recheck: |
3732 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3733 | if (policy < 0) { |
3734 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3735 | policy = oldpolicy = p->policy; |
ca94c442 | 3736 | } else { |
7479f3c9 | 3737 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3738 | |
aab03e05 DF |
3739 | if (policy != SCHED_DEADLINE && |
3740 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3741 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3742 | policy != SCHED_IDLE) | |
3743 | return -EINVAL; | |
3744 | } | |
3745 | ||
7479f3c9 PZ |
3746 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3747 | return -EINVAL; | |
3748 | ||
1da177e4 LT |
3749 | /* |
3750 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3751 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3752 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3753 | */ |
0bb040a4 | 3754 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3755 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3756 | return -EINVAL; |
aab03e05 DF |
3757 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3758 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3759 | return -EINVAL; |
3760 | ||
37e4ab3f OC |
3761 | /* |
3762 | * Allow unprivileged RT tasks to decrease priority: | |
3763 | */ | |
961ccddd | 3764 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3765 | if (fair_policy(policy)) { |
d0ea0268 | 3766 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3767 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3768 | return -EPERM; |
3769 | } | |
3770 | ||
e05606d3 | 3771 | if (rt_policy(policy)) { |
a44702e8 ON |
3772 | unsigned long rlim_rtprio = |
3773 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3774 | |
3775 | /* can't set/change the rt policy */ | |
3776 | if (policy != p->policy && !rlim_rtprio) | |
3777 | return -EPERM; | |
3778 | ||
3779 | /* can't increase priority */ | |
d50dde5a DF |
3780 | if (attr->sched_priority > p->rt_priority && |
3781 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3782 | return -EPERM; |
3783 | } | |
c02aa73b | 3784 | |
d44753b8 JL |
3785 | /* |
3786 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3787 | * (safest behavior); in the future we would like to allow | |
3788 | * unprivileged DL tasks to increase their relative deadline | |
3789 | * or reduce their runtime (both ways reducing utilization) | |
3790 | */ | |
3791 | if (dl_policy(policy)) | |
3792 | return -EPERM; | |
3793 | ||
dd41f596 | 3794 | /* |
c02aa73b DH |
3795 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3796 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3797 | */ |
c02aa73b | 3798 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3799 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3800 | return -EPERM; |
3801 | } | |
5fe1d75f | 3802 | |
37e4ab3f | 3803 | /* can't change other user's priorities */ |
c69e8d9c | 3804 | if (!check_same_owner(p)) |
37e4ab3f | 3805 | return -EPERM; |
ca94c442 LP |
3806 | |
3807 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3808 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3809 | return -EPERM; | |
37e4ab3f | 3810 | } |
1da177e4 | 3811 | |
725aad24 | 3812 | if (user) { |
b0ae1981 | 3813 | retval = security_task_setscheduler(p); |
725aad24 JF |
3814 | if (retval) |
3815 | return retval; | |
3816 | } | |
3817 | ||
b29739f9 IM |
3818 | /* |
3819 | * make sure no PI-waiters arrive (or leave) while we are | |
3820 | * changing the priority of the task: | |
0122ec5b | 3821 | * |
25985edc | 3822 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3823 | * runqueue lock must be held. |
3824 | */ | |
0122ec5b | 3825 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3826 | |
34f971f6 PZ |
3827 | /* |
3828 | * Changing the policy of the stop threads its a very bad idea | |
3829 | */ | |
3830 | if (p == rq->stop) { | |
0122ec5b | 3831 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3832 | return -EINVAL; |
3833 | } | |
3834 | ||
a51e9198 | 3835 | /* |
d6b1e911 TG |
3836 | * If not changing anything there's no need to proceed further, |
3837 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3838 | */ |
d50dde5a | 3839 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3840 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3841 | goto change; |
3842 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3843 | goto change; | |
75381608 | 3844 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 3845 | goto change; |
d50dde5a | 3846 | |
d6b1e911 | 3847 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3848 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3849 | return 0; |
3850 | } | |
d50dde5a | 3851 | change: |
a51e9198 | 3852 | |
dc61b1d6 | 3853 | if (user) { |
332ac17e | 3854 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3855 | /* |
3856 | * Do not allow realtime tasks into groups that have no runtime | |
3857 | * assigned. | |
3858 | */ | |
3859 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3860 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3861 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3862 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3863 | return -EPERM; |
3864 | } | |
dc61b1d6 | 3865 | #endif |
332ac17e DF |
3866 | #ifdef CONFIG_SMP |
3867 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3868 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3869 | |
3870 | /* | |
3871 | * Don't allow tasks with an affinity mask smaller than | |
3872 | * the entire root_domain to become SCHED_DEADLINE. We | |
3873 | * will also fail if there's no bandwidth available. | |
3874 | */ | |
e4099a5e PZ |
3875 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3876 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3877 | task_rq_unlock(rq, p, &flags); |
3878 | return -EPERM; | |
3879 | } | |
3880 | } | |
3881 | #endif | |
3882 | } | |
dc61b1d6 | 3883 | |
1da177e4 LT |
3884 | /* recheck policy now with rq lock held */ |
3885 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3886 | policy = oldpolicy = -1; | |
0122ec5b | 3887 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3888 | goto recheck; |
3889 | } | |
332ac17e DF |
3890 | |
3891 | /* | |
3892 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3893 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3894 | * is available. | |
3895 | */ | |
e4099a5e | 3896 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3897 | task_rq_unlock(rq, p, &flags); |
3898 | return -EBUSY; | |
3899 | } | |
3900 | ||
c365c292 TG |
3901 | p->sched_reset_on_fork = reset_on_fork; |
3902 | oldprio = p->prio; | |
3903 | ||
dbc7f069 PZ |
3904 | if (pi) { |
3905 | /* | |
3906 | * Take priority boosted tasks into account. If the new | |
3907 | * effective priority is unchanged, we just store the new | |
3908 | * normal parameters and do not touch the scheduler class and | |
3909 | * the runqueue. This will be done when the task deboost | |
3910 | * itself. | |
3911 | */ | |
3912 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
3913 | if (new_effective_prio == oldprio) { | |
3914 | __setscheduler_params(p, attr); | |
3915 | task_rq_unlock(rq, p, &flags); | |
3916 | return 0; | |
3917 | } | |
c365c292 TG |
3918 | } |
3919 | ||
da0c1e65 | 3920 | queued = task_on_rq_queued(p); |
051a1d1a | 3921 | running = task_current(rq, p); |
da0c1e65 | 3922 | if (queued) |
4ca9b72b | 3923 | dequeue_task(rq, p, 0); |
0e1f3483 | 3924 | if (running) |
f3cd1c4e | 3925 | put_prev_task(rq, p); |
f6b53205 | 3926 | |
83ab0aa0 | 3927 | prev_class = p->sched_class; |
dbc7f069 | 3928 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 3929 | |
0e1f3483 HS |
3930 | if (running) |
3931 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3932 | if (queued) { |
81a44c54 TG |
3933 | /* |
3934 | * We enqueue to tail when the priority of a task is | |
3935 | * increased (user space view). | |
3936 | */ | |
3937 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3938 | } | |
cb469845 | 3939 | |
da7a735e | 3940 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 3941 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3942 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3943 | |
dbc7f069 PZ |
3944 | if (pi) |
3945 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 3946 | |
4c9a4bc8 PZ |
3947 | /* |
3948 | * Run balance callbacks after we've adjusted the PI chain. | |
3949 | */ | |
3950 | balance_callback(rq); | |
3951 | preempt_enable(); | |
95e02ca9 | 3952 | |
1da177e4 LT |
3953 | return 0; |
3954 | } | |
961ccddd | 3955 | |
7479f3c9 PZ |
3956 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3957 | const struct sched_param *param, bool check) | |
3958 | { | |
3959 | struct sched_attr attr = { | |
3960 | .sched_policy = policy, | |
3961 | .sched_priority = param->sched_priority, | |
3962 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3963 | }; | |
3964 | ||
c13db6b1 SR |
3965 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3966 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3967 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3968 | policy &= ~SCHED_RESET_ON_FORK; | |
3969 | attr.sched_policy = policy; | |
3970 | } | |
3971 | ||
dbc7f069 | 3972 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 3973 | } |
961ccddd RR |
3974 | /** |
3975 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3976 | * @p: the task in question. | |
3977 | * @policy: new policy. | |
3978 | * @param: structure containing the new RT priority. | |
3979 | * | |
e69f6186 YB |
3980 | * Return: 0 on success. An error code otherwise. |
3981 | * | |
961ccddd RR |
3982 | * NOTE that the task may be already dead. |
3983 | */ | |
3984 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3985 | const struct sched_param *param) |
961ccddd | 3986 | { |
7479f3c9 | 3987 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3988 | } |
1da177e4 LT |
3989 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3990 | ||
d50dde5a DF |
3991 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3992 | { | |
dbc7f069 | 3993 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
3994 | } |
3995 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3996 | ||
961ccddd RR |
3997 | /** |
3998 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3999 | * @p: the task in question. | |
4000 | * @policy: new policy. | |
4001 | * @param: structure containing the new RT priority. | |
4002 | * | |
4003 | * Just like sched_setscheduler, only don't bother checking if the | |
4004 | * current context has permission. For example, this is needed in | |
4005 | * stop_machine(): we create temporary high priority worker threads, | |
4006 | * but our caller might not have that capability. | |
e69f6186 YB |
4007 | * |
4008 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4009 | */ |
4010 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4011 | const struct sched_param *param) |
961ccddd | 4012 | { |
7479f3c9 | 4013 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
4014 | } |
4015 | ||
95cdf3b7 IM |
4016 | static int |
4017 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4018 | { |
1da177e4 LT |
4019 | struct sched_param lparam; |
4020 | struct task_struct *p; | |
36c8b586 | 4021 | int retval; |
1da177e4 LT |
4022 | |
4023 | if (!param || pid < 0) | |
4024 | return -EINVAL; | |
4025 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4026 | return -EFAULT; | |
5fe1d75f ON |
4027 | |
4028 | rcu_read_lock(); | |
4029 | retval = -ESRCH; | |
1da177e4 | 4030 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4031 | if (p != NULL) |
4032 | retval = sched_setscheduler(p, policy, &lparam); | |
4033 | rcu_read_unlock(); | |
36c8b586 | 4034 | |
1da177e4 LT |
4035 | return retval; |
4036 | } | |
4037 | ||
d50dde5a DF |
4038 | /* |
4039 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4040 | */ | |
4041 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
4042 | struct sched_attr *attr) | |
4043 | { | |
4044 | u32 size; | |
4045 | int ret; | |
4046 | ||
4047 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4048 | return -EFAULT; | |
4049 | ||
4050 | /* | |
4051 | * zero the full structure, so that a short copy will be nice. | |
4052 | */ | |
4053 | memset(attr, 0, sizeof(*attr)); | |
4054 | ||
4055 | ret = get_user(size, &uattr->size); | |
4056 | if (ret) | |
4057 | return ret; | |
4058 | ||
4059 | if (size > PAGE_SIZE) /* silly large */ | |
4060 | goto err_size; | |
4061 | ||
4062 | if (!size) /* abi compat */ | |
4063 | size = SCHED_ATTR_SIZE_VER0; | |
4064 | ||
4065 | if (size < SCHED_ATTR_SIZE_VER0) | |
4066 | goto err_size; | |
4067 | ||
4068 | /* | |
4069 | * If we're handed a bigger struct than we know of, | |
4070 | * ensure all the unknown bits are 0 - i.e. new | |
4071 | * user-space does not rely on any kernel feature | |
4072 | * extensions we dont know about yet. | |
4073 | */ | |
4074 | if (size > sizeof(*attr)) { | |
4075 | unsigned char __user *addr; | |
4076 | unsigned char __user *end; | |
4077 | unsigned char val; | |
4078 | ||
4079 | addr = (void __user *)uattr + sizeof(*attr); | |
4080 | end = (void __user *)uattr + size; | |
4081 | ||
4082 | for (; addr < end; addr++) { | |
4083 | ret = get_user(val, addr); | |
4084 | if (ret) | |
4085 | return ret; | |
4086 | if (val) | |
4087 | goto err_size; | |
4088 | } | |
4089 | size = sizeof(*attr); | |
4090 | } | |
4091 | ||
4092 | ret = copy_from_user(attr, uattr, size); | |
4093 | if (ret) | |
4094 | return -EFAULT; | |
4095 | ||
4096 | /* | |
4097 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4098 | * to be strict and return an error on out-of-bounds values? | |
4099 | */ | |
75e45d51 | 4100 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4101 | |
e78c7bca | 4102 | return 0; |
d50dde5a DF |
4103 | |
4104 | err_size: | |
4105 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4106 | return -E2BIG; |
d50dde5a DF |
4107 | } |
4108 | ||
1da177e4 LT |
4109 | /** |
4110 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4111 | * @pid: the pid in question. | |
4112 | * @policy: new policy. | |
4113 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4114 | * |
4115 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4116 | */ |
5add95d4 HC |
4117 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4118 | struct sched_param __user *, param) | |
1da177e4 | 4119 | { |
c21761f1 JB |
4120 | /* negative values for policy are not valid */ |
4121 | if (policy < 0) | |
4122 | return -EINVAL; | |
4123 | ||
1da177e4 LT |
4124 | return do_sched_setscheduler(pid, policy, param); |
4125 | } | |
4126 | ||
4127 | /** | |
4128 | * sys_sched_setparam - set/change the RT priority of a thread | |
4129 | * @pid: the pid in question. | |
4130 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4131 | * |
4132 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4133 | */ |
5add95d4 | 4134 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4135 | { |
c13db6b1 | 4136 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4137 | } |
4138 | ||
d50dde5a DF |
4139 | /** |
4140 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4141 | * @pid: the pid in question. | |
5778fccf | 4142 | * @uattr: structure containing the extended parameters. |
db66d756 | 4143 | * @flags: for future extension. |
d50dde5a | 4144 | */ |
6d35ab48 PZ |
4145 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4146 | unsigned int, flags) | |
d50dde5a DF |
4147 | { |
4148 | struct sched_attr attr; | |
4149 | struct task_struct *p; | |
4150 | int retval; | |
4151 | ||
6d35ab48 | 4152 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4153 | return -EINVAL; |
4154 | ||
143cf23d MK |
4155 | retval = sched_copy_attr(uattr, &attr); |
4156 | if (retval) | |
4157 | return retval; | |
d50dde5a | 4158 | |
b14ed2c2 | 4159 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4160 | return -EINVAL; |
d50dde5a DF |
4161 | |
4162 | rcu_read_lock(); | |
4163 | retval = -ESRCH; | |
4164 | p = find_process_by_pid(pid); | |
4165 | if (p != NULL) | |
4166 | retval = sched_setattr(p, &attr); | |
4167 | rcu_read_unlock(); | |
4168 | ||
4169 | return retval; | |
4170 | } | |
4171 | ||
1da177e4 LT |
4172 | /** |
4173 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4174 | * @pid: the pid in question. | |
e69f6186 YB |
4175 | * |
4176 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4177 | * code. | |
1da177e4 | 4178 | */ |
5add95d4 | 4179 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4180 | { |
36c8b586 | 4181 | struct task_struct *p; |
3a5c359a | 4182 | int retval; |
1da177e4 LT |
4183 | |
4184 | if (pid < 0) | |
3a5c359a | 4185 | return -EINVAL; |
1da177e4 LT |
4186 | |
4187 | retval = -ESRCH; | |
5fe85be0 | 4188 | rcu_read_lock(); |
1da177e4 LT |
4189 | p = find_process_by_pid(pid); |
4190 | if (p) { | |
4191 | retval = security_task_getscheduler(p); | |
4192 | if (!retval) | |
ca94c442 LP |
4193 | retval = p->policy |
4194 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4195 | } |
5fe85be0 | 4196 | rcu_read_unlock(); |
1da177e4 LT |
4197 | return retval; |
4198 | } | |
4199 | ||
4200 | /** | |
ca94c442 | 4201 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4202 | * @pid: the pid in question. |
4203 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4204 | * |
4205 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4206 | * code. | |
1da177e4 | 4207 | */ |
5add95d4 | 4208 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4209 | { |
ce5f7f82 | 4210 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4211 | struct task_struct *p; |
3a5c359a | 4212 | int retval; |
1da177e4 LT |
4213 | |
4214 | if (!param || pid < 0) | |
3a5c359a | 4215 | return -EINVAL; |
1da177e4 | 4216 | |
5fe85be0 | 4217 | rcu_read_lock(); |
1da177e4 LT |
4218 | p = find_process_by_pid(pid); |
4219 | retval = -ESRCH; | |
4220 | if (!p) | |
4221 | goto out_unlock; | |
4222 | ||
4223 | retval = security_task_getscheduler(p); | |
4224 | if (retval) | |
4225 | goto out_unlock; | |
4226 | ||
ce5f7f82 PZ |
4227 | if (task_has_rt_policy(p)) |
4228 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4229 | rcu_read_unlock(); |
1da177e4 LT |
4230 | |
4231 | /* | |
4232 | * This one might sleep, we cannot do it with a spinlock held ... | |
4233 | */ | |
4234 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4235 | ||
1da177e4 LT |
4236 | return retval; |
4237 | ||
4238 | out_unlock: | |
5fe85be0 | 4239 | rcu_read_unlock(); |
1da177e4 LT |
4240 | return retval; |
4241 | } | |
4242 | ||
d50dde5a DF |
4243 | static int sched_read_attr(struct sched_attr __user *uattr, |
4244 | struct sched_attr *attr, | |
4245 | unsigned int usize) | |
4246 | { | |
4247 | int ret; | |
4248 | ||
4249 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4250 | return -EFAULT; | |
4251 | ||
4252 | /* | |
4253 | * If we're handed a smaller struct than we know of, | |
4254 | * ensure all the unknown bits are 0 - i.e. old | |
4255 | * user-space does not get uncomplete information. | |
4256 | */ | |
4257 | if (usize < sizeof(*attr)) { | |
4258 | unsigned char *addr; | |
4259 | unsigned char *end; | |
4260 | ||
4261 | addr = (void *)attr + usize; | |
4262 | end = (void *)attr + sizeof(*attr); | |
4263 | ||
4264 | for (; addr < end; addr++) { | |
4265 | if (*addr) | |
22400674 | 4266 | return -EFBIG; |
d50dde5a DF |
4267 | } |
4268 | ||
4269 | attr->size = usize; | |
4270 | } | |
4271 | ||
4efbc454 | 4272 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4273 | if (ret) |
4274 | return -EFAULT; | |
4275 | ||
22400674 | 4276 | return 0; |
d50dde5a DF |
4277 | } |
4278 | ||
4279 | /** | |
aab03e05 | 4280 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4281 | * @pid: the pid in question. |
5778fccf | 4282 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4283 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4284 | * @flags: for future extension. |
d50dde5a | 4285 | */ |
6d35ab48 PZ |
4286 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4287 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4288 | { |
4289 | struct sched_attr attr = { | |
4290 | .size = sizeof(struct sched_attr), | |
4291 | }; | |
4292 | struct task_struct *p; | |
4293 | int retval; | |
4294 | ||
4295 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4296 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4297 | return -EINVAL; |
4298 | ||
4299 | rcu_read_lock(); | |
4300 | p = find_process_by_pid(pid); | |
4301 | retval = -ESRCH; | |
4302 | if (!p) | |
4303 | goto out_unlock; | |
4304 | ||
4305 | retval = security_task_getscheduler(p); | |
4306 | if (retval) | |
4307 | goto out_unlock; | |
4308 | ||
4309 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4310 | if (p->sched_reset_on_fork) |
4311 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4312 | if (task_has_dl_policy(p)) |
4313 | __getparam_dl(p, &attr); | |
4314 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4315 | attr.sched_priority = p->rt_priority; |
4316 | else | |
d0ea0268 | 4317 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4318 | |
4319 | rcu_read_unlock(); | |
4320 | ||
4321 | retval = sched_read_attr(uattr, &attr, size); | |
4322 | return retval; | |
4323 | ||
4324 | out_unlock: | |
4325 | rcu_read_unlock(); | |
4326 | return retval; | |
4327 | } | |
4328 | ||
96f874e2 | 4329 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4330 | { |
5a16f3d3 | 4331 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4332 | struct task_struct *p; |
4333 | int retval; | |
1da177e4 | 4334 | |
23f5d142 | 4335 | rcu_read_lock(); |
1da177e4 LT |
4336 | |
4337 | p = find_process_by_pid(pid); | |
4338 | if (!p) { | |
23f5d142 | 4339 | rcu_read_unlock(); |
1da177e4 LT |
4340 | return -ESRCH; |
4341 | } | |
4342 | ||
23f5d142 | 4343 | /* Prevent p going away */ |
1da177e4 | 4344 | get_task_struct(p); |
23f5d142 | 4345 | rcu_read_unlock(); |
1da177e4 | 4346 | |
14a40ffc TH |
4347 | if (p->flags & PF_NO_SETAFFINITY) { |
4348 | retval = -EINVAL; | |
4349 | goto out_put_task; | |
4350 | } | |
5a16f3d3 RR |
4351 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4352 | retval = -ENOMEM; | |
4353 | goto out_put_task; | |
4354 | } | |
4355 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4356 | retval = -ENOMEM; | |
4357 | goto out_free_cpus_allowed; | |
4358 | } | |
1da177e4 | 4359 | retval = -EPERM; |
4c44aaaf EB |
4360 | if (!check_same_owner(p)) { |
4361 | rcu_read_lock(); | |
4362 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4363 | rcu_read_unlock(); | |
16303ab2 | 4364 | goto out_free_new_mask; |
4c44aaaf EB |
4365 | } |
4366 | rcu_read_unlock(); | |
4367 | } | |
1da177e4 | 4368 | |
b0ae1981 | 4369 | retval = security_task_setscheduler(p); |
e7834f8f | 4370 | if (retval) |
16303ab2 | 4371 | goto out_free_new_mask; |
e7834f8f | 4372 | |
e4099a5e PZ |
4373 | |
4374 | cpuset_cpus_allowed(p, cpus_allowed); | |
4375 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4376 | ||
332ac17e DF |
4377 | /* |
4378 | * Since bandwidth control happens on root_domain basis, | |
4379 | * if admission test is enabled, we only admit -deadline | |
4380 | * tasks allowed to run on all the CPUs in the task's | |
4381 | * root_domain. | |
4382 | */ | |
4383 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4384 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4385 | rcu_read_lock(); | |
4386 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4387 | retval = -EBUSY; |
f1e3a093 | 4388 | rcu_read_unlock(); |
16303ab2 | 4389 | goto out_free_new_mask; |
332ac17e | 4390 | } |
f1e3a093 | 4391 | rcu_read_unlock(); |
332ac17e DF |
4392 | } |
4393 | #endif | |
49246274 | 4394 | again: |
25834c73 | 4395 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4396 | |
8707d8b8 | 4397 | if (!retval) { |
5a16f3d3 RR |
4398 | cpuset_cpus_allowed(p, cpus_allowed); |
4399 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4400 | /* |
4401 | * We must have raced with a concurrent cpuset | |
4402 | * update. Just reset the cpus_allowed to the | |
4403 | * cpuset's cpus_allowed | |
4404 | */ | |
5a16f3d3 | 4405 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4406 | goto again; |
4407 | } | |
4408 | } | |
16303ab2 | 4409 | out_free_new_mask: |
5a16f3d3 RR |
4410 | free_cpumask_var(new_mask); |
4411 | out_free_cpus_allowed: | |
4412 | free_cpumask_var(cpus_allowed); | |
4413 | out_put_task: | |
1da177e4 | 4414 | put_task_struct(p); |
1da177e4 LT |
4415 | return retval; |
4416 | } | |
4417 | ||
4418 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4419 | struct cpumask *new_mask) |
1da177e4 | 4420 | { |
96f874e2 RR |
4421 | if (len < cpumask_size()) |
4422 | cpumask_clear(new_mask); | |
4423 | else if (len > cpumask_size()) | |
4424 | len = cpumask_size(); | |
4425 | ||
1da177e4 LT |
4426 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4427 | } | |
4428 | ||
4429 | /** | |
4430 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4431 | * @pid: pid of the process | |
4432 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4433 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4434 | * |
4435 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4436 | */ |
5add95d4 HC |
4437 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4438 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4439 | { |
5a16f3d3 | 4440 | cpumask_var_t new_mask; |
1da177e4 LT |
4441 | int retval; |
4442 | ||
5a16f3d3 RR |
4443 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4444 | return -ENOMEM; | |
1da177e4 | 4445 | |
5a16f3d3 RR |
4446 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4447 | if (retval == 0) | |
4448 | retval = sched_setaffinity(pid, new_mask); | |
4449 | free_cpumask_var(new_mask); | |
4450 | return retval; | |
1da177e4 LT |
4451 | } |
4452 | ||
96f874e2 | 4453 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4454 | { |
36c8b586 | 4455 | struct task_struct *p; |
31605683 | 4456 | unsigned long flags; |
1da177e4 | 4457 | int retval; |
1da177e4 | 4458 | |
23f5d142 | 4459 | rcu_read_lock(); |
1da177e4 LT |
4460 | |
4461 | retval = -ESRCH; | |
4462 | p = find_process_by_pid(pid); | |
4463 | if (!p) | |
4464 | goto out_unlock; | |
4465 | ||
e7834f8f DQ |
4466 | retval = security_task_getscheduler(p); |
4467 | if (retval) | |
4468 | goto out_unlock; | |
4469 | ||
013fdb80 | 4470 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4471 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4472 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4473 | |
4474 | out_unlock: | |
23f5d142 | 4475 | rcu_read_unlock(); |
1da177e4 | 4476 | |
9531b62f | 4477 | return retval; |
1da177e4 LT |
4478 | } |
4479 | ||
4480 | /** | |
4481 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4482 | * @pid: pid of the process | |
4483 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4484 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4485 | * |
4486 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4487 | */ |
5add95d4 HC |
4488 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4489 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4490 | { |
4491 | int ret; | |
f17c8607 | 4492 | cpumask_var_t mask; |
1da177e4 | 4493 | |
84fba5ec | 4494 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4495 | return -EINVAL; |
4496 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4497 | return -EINVAL; |
4498 | ||
f17c8607 RR |
4499 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4500 | return -ENOMEM; | |
1da177e4 | 4501 | |
f17c8607 RR |
4502 | ret = sched_getaffinity(pid, mask); |
4503 | if (ret == 0) { | |
8bc037fb | 4504 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4505 | |
4506 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4507 | ret = -EFAULT; |
4508 | else | |
cd3d8031 | 4509 | ret = retlen; |
f17c8607 RR |
4510 | } |
4511 | free_cpumask_var(mask); | |
1da177e4 | 4512 | |
f17c8607 | 4513 | return ret; |
1da177e4 LT |
4514 | } |
4515 | ||
4516 | /** | |
4517 | * sys_sched_yield - yield the current processor to other threads. | |
4518 | * | |
dd41f596 IM |
4519 | * This function yields the current CPU to other tasks. If there are no |
4520 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4521 | * |
4522 | * Return: 0. | |
1da177e4 | 4523 | */ |
5add95d4 | 4524 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4525 | { |
70b97a7f | 4526 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4527 | |
2d72376b | 4528 | schedstat_inc(rq, yld_count); |
4530d7ab | 4529 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4530 | |
4531 | /* | |
4532 | * Since we are going to call schedule() anyway, there's | |
4533 | * no need to preempt or enable interrupts: | |
4534 | */ | |
4535 | __release(rq->lock); | |
8a25d5de | 4536 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4537 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4538 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4539 | |
4540 | schedule(); | |
4541 | ||
4542 | return 0; | |
4543 | } | |
4544 | ||
02b67cc3 | 4545 | int __sched _cond_resched(void) |
1da177e4 | 4546 | { |
fe32d3cd | 4547 | if (should_resched(0)) { |
a18b5d01 | 4548 | preempt_schedule_common(); |
1da177e4 LT |
4549 | return 1; |
4550 | } | |
4551 | return 0; | |
4552 | } | |
02b67cc3 | 4553 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4554 | |
4555 | /* | |
613afbf8 | 4556 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4557 | * call schedule, and on return reacquire the lock. |
4558 | * | |
41a2d6cf | 4559 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4560 | * operations here to prevent schedule() from being called twice (once via |
4561 | * spin_unlock(), once by hand). | |
4562 | */ | |
613afbf8 | 4563 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4564 | { |
fe32d3cd | 4565 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4566 | int ret = 0; |
4567 | ||
f607c668 PZ |
4568 | lockdep_assert_held(lock); |
4569 | ||
4a81e832 | 4570 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4571 | spin_unlock(lock); |
d86ee480 | 4572 | if (resched) |
a18b5d01 | 4573 | preempt_schedule_common(); |
95c354fe NP |
4574 | else |
4575 | cpu_relax(); | |
6df3cecb | 4576 | ret = 1; |
1da177e4 | 4577 | spin_lock(lock); |
1da177e4 | 4578 | } |
6df3cecb | 4579 | return ret; |
1da177e4 | 4580 | } |
613afbf8 | 4581 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4582 | |
613afbf8 | 4583 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4584 | { |
4585 | BUG_ON(!in_softirq()); | |
4586 | ||
fe32d3cd | 4587 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4588 | local_bh_enable(); |
a18b5d01 | 4589 | preempt_schedule_common(); |
1da177e4 LT |
4590 | local_bh_disable(); |
4591 | return 1; | |
4592 | } | |
4593 | return 0; | |
4594 | } | |
613afbf8 | 4595 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4596 | |
1da177e4 LT |
4597 | /** |
4598 | * yield - yield the current processor to other threads. | |
4599 | * | |
8e3fabfd PZ |
4600 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4601 | * | |
4602 | * The scheduler is at all times free to pick the calling task as the most | |
4603 | * eligible task to run, if removing the yield() call from your code breaks | |
4604 | * it, its already broken. | |
4605 | * | |
4606 | * Typical broken usage is: | |
4607 | * | |
4608 | * while (!event) | |
4609 | * yield(); | |
4610 | * | |
4611 | * where one assumes that yield() will let 'the other' process run that will | |
4612 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4613 | * happen. Never use yield() as a progress guarantee!! | |
4614 | * | |
4615 | * If you want to use yield() to wait for something, use wait_event(). | |
4616 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4617 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4618 | */ |
4619 | void __sched yield(void) | |
4620 | { | |
4621 | set_current_state(TASK_RUNNING); | |
4622 | sys_sched_yield(); | |
4623 | } | |
1da177e4 LT |
4624 | EXPORT_SYMBOL(yield); |
4625 | ||
d95f4122 MG |
4626 | /** |
4627 | * yield_to - yield the current processor to another thread in | |
4628 | * your thread group, or accelerate that thread toward the | |
4629 | * processor it's on. | |
16addf95 RD |
4630 | * @p: target task |
4631 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4632 | * |
4633 | * It's the caller's job to ensure that the target task struct | |
4634 | * can't go away on us before we can do any checks. | |
4635 | * | |
e69f6186 | 4636 | * Return: |
7b270f60 PZ |
4637 | * true (>0) if we indeed boosted the target task. |
4638 | * false (0) if we failed to boost the target. | |
4639 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4640 | */ |
fa93384f | 4641 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4642 | { |
4643 | struct task_struct *curr = current; | |
4644 | struct rq *rq, *p_rq; | |
4645 | unsigned long flags; | |
c3c18640 | 4646 | int yielded = 0; |
d95f4122 MG |
4647 | |
4648 | local_irq_save(flags); | |
4649 | rq = this_rq(); | |
4650 | ||
4651 | again: | |
4652 | p_rq = task_rq(p); | |
7b270f60 PZ |
4653 | /* |
4654 | * If we're the only runnable task on the rq and target rq also | |
4655 | * has only one task, there's absolutely no point in yielding. | |
4656 | */ | |
4657 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4658 | yielded = -ESRCH; | |
4659 | goto out_irq; | |
4660 | } | |
4661 | ||
d95f4122 | 4662 | double_rq_lock(rq, p_rq); |
39e24d8f | 4663 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4664 | double_rq_unlock(rq, p_rq); |
4665 | goto again; | |
4666 | } | |
4667 | ||
4668 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4669 | goto out_unlock; |
d95f4122 MG |
4670 | |
4671 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4672 | goto out_unlock; |
d95f4122 MG |
4673 | |
4674 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4675 | goto out_unlock; |
d95f4122 MG |
4676 | |
4677 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4678 | if (yielded) { |
d95f4122 | 4679 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4680 | /* |
4681 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4682 | * fairness. | |
4683 | */ | |
4684 | if (preempt && rq != p_rq) | |
8875125e | 4685 | resched_curr(p_rq); |
6d1cafd8 | 4686 | } |
d95f4122 | 4687 | |
7b270f60 | 4688 | out_unlock: |
d95f4122 | 4689 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4690 | out_irq: |
d95f4122 MG |
4691 | local_irq_restore(flags); |
4692 | ||
7b270f60 | 4693 | if (yielded > 0) |
d95f4122 MG |
4694 | schedule(); |
4695 | ||
4696 | return yielded; | |
4697 | } | |
4698 | EXPORT_SYMBOL_GPL(yield_to); | |
4699 | ||
1da177e4 | 4700 | /* |
41a2d6cf | 4701 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4702 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4703 | */ |
1da177e4 LT |
4704 | long __sched io_schedule_timeout(long timeout) |
4705 | { | |
9cff8ade N |
4706 | int old_iowait = current->in_iowait; |
4707 | struct rq *rq; | |
1da177e4 LT |
4708 | long ret; |
4709 | ||
9cff8ade | 4710 | current->in_iowait = 1; |
10d784ea | 4711 | blk_schedule_flush_plug(current); |
9cff8ade | 4712 | |
0ff92245 | 4713 | delayacct_blkio_start(); |
9cff8ade | 4714 | rq = raw_rq(); |
1da177e4 LT |
4715 | atomic_inc(&rq->nr_iowait); |
4716 | ret = schedule_timeout(timeout); | |
9cff8ade | 4717 | current->in_iowait = old_iowait; |
1da177e4 | 4718 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4719 | delayacct_blkio_end(); |
9cff8ade | 4720 | |
1da177e4 LT |
4721 | return ret; |
4722 | } | |
9cff8ade | 4723 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4724 | |
4725 | /** | |
4726 | * sys_sched_get_priority_max - return maximum RT priority. | |
4727 | * @policy: scheduling class. | |
4728 | * | |
e69f6186 YB |
4729 | * Return: On success, this syscall returns the maximum |
4730 | * rt_priority that can be used by a given scheduling class. | |
4731 | * On failure, a negative error code is returned. | |
1da177e4 | 4732 | */ |
5add95d4 | 4733 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4734 | { |
4735 | int ret = -EINVAL; | |
4736 | ||
4737 | switch (policy) { | |
4738 | case SCHED_FIFO: | |
4739 | case SCHED_RR: | |
4740 | ret = MAX_USER_RT_PRIO-1; | |
4741 | break; | |
aab03e05 | 4742 | case SCHED_DEADLINE: |
1da177e4 | 4743 | case SCHED_NORMAL: |
b0a9499c | 4744 | case SCHED_BATCH: |
dd41f596 | 4745 | case SCHED_IDLE: |
1da177e4 LT |
4746 | ret = 0; |
4747 | break; | |
4748 | } | |
4749 | return ret; | |
4750 | } | |
4751 | ||
4752 | /** | |
4753 | * sys_sched_get_priority_min - return minimum RT priority. | |
4754 | * @policy: scheduling class. | |
4755 | * | |
e69f6186 YB |
4756 | * Return: On success, this syscall returns the minimum |
4757 | * rt_priority that can be used by a given scheduling class. | |
4758 | * On failure, a negative error code is returned. | |
1da177e4 | 4759 | */ |
5add95d4 | 4760 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4761 | { |
4762 | int ret = -EINVAL; | |
4763 | ||
4764 | switch (policy) { | |
4765 | case SCHED_FIFO: | |
4766 | case SCHED_RR: | |
4767 | ret = 1; | |
4768 | break; | |
aab03e05 | 4769 | case SCHED_DEADLINE: |
1da177e4 | 4770 | case SCHED_NORMAL: |
b0a9499c | 4771 | case SCHED_BATCH: |
dd41f596 | 4772 | case SCHED_IDLE: |
1da177e4 LT |
4773 | ret = 0; |
4774 | } | |
4775 | return ret; | |
4776 | } | |
4777 | ||
4778 | /** | |
4779 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4780 | * @pid: pid of the process. | |
4781 | * @interval: userspace pointer to the timeslice value. | |
4782 | * | |
4783 | * this syscall writes the default timeslice value of a given process | |
4784 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4785 | * |
4786 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4787 | * an error code. | |
1da177e4 | 4788 | */ |
17da2bd9 | 4789 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4790 | struct timespec __user *, interval) |
1da177e4 | 4791 | { |
36c8b586 | 4792 | struct task_struct *p; |
a4ec24b4 | 4793 | unsigned int time_slice; |
dba091b9 TG |
4794 | unsigned long flags; |
4795 | struct rq *rq; | |
3a5c359a | 4796 | int retval; |
1da177e4 | 4797 | struct timespec t; |
1da177e4 LT |
4798 | |
4799 | if (pid < 0) | |
3a5c359a | 4800 | return -EINVAL; |
1da177e4 LT |
4801 | |
4802 | retval = -ESRCH; | |
1a551ae7 | 4803 | rcu_read_lock(); |
1da177e4 LT |
4804 | p = find_process_by_pid(pid); |
4805 | if (!p) | |
4806 | goto out_unlock; | |
4807 | ||
4808 | retval = security_task_getscheduler(p); | |
4809 | if (retval) | |
4810 | goto out_unlock; | |
4811 | ||
dba091b9 | 4812 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4813 | time_slice = 0; |
4814 | if (p->sched_class->get_rr_interval) | |
4815 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4816 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4817 | |
1a551ae7 | 4818 | rcu_read_unlock(); |
a4ec24b4 | 4819 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4820 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4821 | return retval; |
3a5c359a | 4822 | |
1da177e4 | 4823 | out_unlock: |
1a551ae7 | 4824 | rcu_read_unlock(); |
1da177e4 LT |
4825 | return retval; |
4826 | } | |
4827 | ||
7c731e0a | 4828 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4829 | |
82a1fcb9 | 4830 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4831 | { |
1da177e4 | 4832 | unsigned long free = 0; |
4e79752c | 4833 | int ppid; |
1f8a7633 | 4834 | unsigned long state = p->state; |
1da177e4 | 4835 | |
1f8a7633 TH |
4836 | if (state) |
4837 | state = __ffs(state) + 1; | |
28d0686c | 4838 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4839 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4840 | #if BITS_PER_LONG == 32 |
1da177e4 | 4841 | if (state == TASK_RUNNING) |
3df0fc5b | 4842 | printk(KERN_CONT " running "); |
1da177e4 | 4843 | else |
3df0fc5b | 4844 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4845 | #else |
4846 | if (state == TASK_RUNNING) | |
3df0fc5b | 4847 | printk(KERN_CONT " running task "); |
1da177e4 | 4848 | else |
3df0fc5b | 4849 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4850 | #endif |
4851 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4852 | free = stack_not_used(p); |
1da177e4 | 4853 | #endif |
a90e984c | 4854 | ppid = 0; |
4e79752c | 4855 | rcu_read_lock(); |
a90e984c ON |
4856 | if (pid_alive(p)) |
4857 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 4858 | rcu_read_unlock(); |
3df0fc5b | 4859 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4860 | task_pid_nr(p), ppid, |
aa47b7e0 | 4861 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4862 | |
3d1cb205 | 4863 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4864 | show_stack(p, NULL); |
1da177e4 LT |
4865 | } |
4866 | ||
e59e2ae2 | 4867 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4868 | { |
36c8b586 | 4869 | struct task_struct *g, *p; |
1da177e4 | 4870 | |
4bd77321 | 4871 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4872 | printk(KERN_INFO |
4873 | " task PC stack pid father\n"); | |
1da177e4 | 4874 | #else |
3df0fc5b PZ |
4875 | printk(KERN_INFO |
4876 | " task PC stack pid father\n"); | |
1da177e4 | 4877 | #endif |
510f5acc | 4878 | rcu_read_lock(); |
5d07f420 | 4879 | for_each_process_thread(g, p) { |
1da177e4 LT |
4880 | /* |
4881 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4882 | * console might take a lot of time: |
1da177e4 LT |
4883 | */ |
4884 | touch_nmi_watchdog(); | |
39bc89fd | 4885 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4886 | sched_show_task(p); |
5d07f420 | 4887 | } |
1da177e4 | 4888 | |
04c9167f JF |
4889 | touch_all_softlockup_watchdogs(); |
4890 | ||
dd41f596 IM |
4891 | #ifdef CONFIG_SCHED_DEBUG |
4892 | sysrq_sched_debug_show(); | |
4893 | #endif | |
510f5acc | 4894 | rcu_read_unlock(); |
e59e2ae2 IM |
4895 | /* |
4896 | * Only show locks if all tasks are dumped: | |
4897 | */ | |
93335a21 | 4898 | if (!state_filter) |
e59e2ae2 | 4899 | debug_show_all_locks(); |
1da177e4 LT |
4900 | } |
4901 | ||
0db0628d | 4902 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4903 | { |
dd41f596 | 4904 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4905 | } |
4906 | ||
f340c0d1 IM |
4907 | /** |
4908 | * init_idle - set up an idle thread for a given CPU | |
4909 | * @idle: task in question | |
4910 | * @cpu: cpu the idle task belongs to | |
4911 | * | |
4912 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4913 | * flag, to make booting more robust. | |
4914 | */ | |
0db0628d | 4915 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4916 | { |
70b97a7f | 4917 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4918 | unsigned long flags; |
4919 | ||
25834c73 PZ |
4920 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
4921 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 4922 | |
5e1576ed | 4923 | __sched_fork(0, idle); |
06b83b5f | 4924 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4925 | idle->se.exec_start = sched_clock(); |
4926 | ||
1e1b6c51 | 4927 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4928 | /* |
4929 | * We're having a chicken and egg problem, even though we are | |
4930 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4931 | * lockdep check in task_group() will fail. | |
4932 | * | |
4933 | * Similar case to sched_fork(). / Alternatively we could | |
4934 | * use task_rq_lock() here and obtain the other rq->lock. | |
4935 | * | |
4936 | * Silence PROVE_RCU | |
4937 | */ | |
4938 | rcu_read_lock(); | |
dd41f596 | 4939 | __set_task_cpu(idle, cpu); |
6506cf6c | 4940 | rcu_read_unlock(); |
1da177e4 | 4941 | |
1da177e4 | 4942 | rq->curr = rq->idle = idle; |
da0c1e65 | 4943 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4944 | #if defined(CONFIG_SMP) |
4945 | idle->on_cpu = 1; | |
4866cde0 | 4946 | #endif |
25834c73 PZ |
4947 | raw_spin_unlock(&rq->lock); |
4948 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
4949 | |
4950 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4951 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4952 | |
dd41f596 IM |
4953 | /* |
4954 | * The idle tasks have their own, simple scheduling class: | |
4955 | */ | |
4956 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4957 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4958 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4959 | #if defined(CONFIG_SMP) |
4960 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4961 | #endif | |
19978ca6 IM |
4962 | } |
4963 | ||
f82f8042 JL |
4964 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
4965 | const struct cpumask *trial) | |
4966 | { | |
4967 | int ret = 1, trial_cpus; | |
4968 | struct dl_bw *cur_dl_b; | |
4969 | unsigned long flags; | |
4970 | ||
bb2bc55a MG |
4971 | if (!cpumask_weight(cur)) |
4972 | return ret; | |
4973 | ||
75e23e49 | 4974 | rcu_read_lock_sched(); |
f82f8042 JL |
4975 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
4976 | trial_cpus = cpumask_weight(trial); | |
4977 | ||
4978 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
4979 | if (cur_dl_b->bw != -1 && | |
4980 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
4981 | ret = 0; | |
4982 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 4983 | rcu_read_unlock_sched(); |
f82f8042 JL |
4984 | |
4985 | return ret; | |
4986 | } | |
4987 | ||
7f51412a JL |
4988 | int task_can_attach(struct task_struct *p, |
4989 | const struct cpumask *cs_cpus_allowed) | |
4990 | { | |
4991 | int ret = 0; | |
4992 | ||
4993 | /* | |
4994 | * Kthreads which disallow setaffinity shouldn't be moved | |
4995 | * to a new cpuset; we don't want to change their cpu | |
4996 | * affinity and isolating such threads by their set of | |
4997 | * allowed nodes is unnecessary. Thus, cpusets are not | |
4998 | * applicable for such threads. This prevents checking for | |
4999 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5000 | * before cpus_allowed may be changed. | |
5001 | */ | |
5002 | if (p->flags & PF_NO_SETAFFINITY) { | |
5003 | ret = -EINVAL; | |
5004 | goto out; | |
5005 | } | |
5006 | ||
5007 | #ifdef CONFIG_SMP | |
5008 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5009 | cs_cpus_allowed)) { | |
5010 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5011 | cs_cpus_allowed); | |
75e23e49 | 5012 | struct dl_bw *dl_b; |
7f51412a JL |
5013 | bool overflow; |
5014 | int cpus; | |
5015 | unsigned long flags; | |
5016 | ||
75e23e49 JL |
5017 | rcu_read_lock_sched(); |
5018 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5019 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5020 | cpus = dl_bw_cpus(dest_cpu); | |
5021 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5022 | if (overflow) | |
5023 | ret = -EBUSY; | |
5024 | else { | |
5025 | /* | |
5026 | * We reserve space for this task in the destination | |
5027 | * root_domain, as we can't fail after this point. | |
5028 | * We will free resources in the source root_domain | |
5029 | * later on (see set_cpus_allowed_dl()). | |
5030 | */ | |
5031 | __dl_add(dl_b, p->dl.dl_bw); | |
5032 | } | |
5033 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5034 | rcu_read_unlock_sched(); |
7f51412a JL |
5035 | |
5036 | } | |
5037 | #endif | |
5038 | out: | |
5039 | return ret; | |
5040 | } | |
5041 | ||
1da177e4 | 5042 | #ifdef CONFIG_SMP |
1da177e4 | 5043 | |
e6628d5b MG |
5044 | #ifdef CONFIG_NUMA_BALANCING |
5045 | /* Migrate current task p to target_cpu */ | |
5046 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5047 | { | |
5048 | struct migration_arg arg = { p, target_cpu }; | |
5049 | int curr_cpu = task_cpu(p); | |
5050 | ||
5051 | if (curr_cpu == target_cpu) | |
5052 | return 0; | |
5053 | ||
5054 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
5055 | return -EINVAL; | |
5056 | ||
5057 | /* TODO: This is not properly updating schedstats */ | |
5058 | ||
286549dc | 5059 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5060 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5061 | } | |
0ec8aa00 PZ |
5062 | |
5063 | /* | |
5064 | * Requeue a task on a given node and accurately track the number of NUMA | |
5065 | * tasks on the runqueues | |
5066 | */ | |
5067 | void sched_setnuma(struct task_struct *p, int nid) | |
5068 | { | |
5069 | struct rq *rq; | |
5070 | unsigned long flags; | |
da0c1e65 | 5071 | bool queued, running; |
0ec8aa00 PZ |
5072 | |
5073 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 5074 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5075 | running = task_current(rq, p); |
5076 | ||
da0c1e65 | 5077 | if (queued) |
0ec8aa00 PZ |
5078 | dequeue_task(rq, p, 0); |
5079 | if (running) | |
f3cd1c4e | 5080 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5081 | |
5082 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5083 | |
5084 | if (running) | |
5085 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5086 | if (queued) |
0ec8aa00 PZ |
5087 | enqueue_task(rq, p, 0); |
5088 | task_rq_unlock(rq, p, &flags); | |
5089 | } | |
5cc389bc | 5090 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5091 | |
1da177e4 | 5092 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5093 | /* |
48c5ccae PZ |
5094 | * Ensures that the idle task is using init_mm right before its cpu goes |
5095 | * offline. | |
054b9108 | 5096 | */ |
48c5ccae | 5097 | void idle_task_exit(void) |
1da177e4 | 5098 | { |
48c5ccae | 5099 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5100 | |
48c5ccae | 5101 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5102 | |
a53efe5f | 5103 | if (mm != &init_mm) { |
48c5ccae | 5104 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
5105 | finish_arch_post_lock_switch(); |
5106 | } | |
48c5ccae | 5107 | mmdrop(mm); |
1da177e4 LT |
5108 | } |
5109 | ||
5110 | /* | |
5d180232 PZ |
5111 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5112 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5113 | * nr_active count is stable. | |
5114 | * | |
5115 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5116 | */ |
5d180232 | 5117 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5118 | { |
5d180232 PZ |
5119 | long delta = calc_load_fold_active(rq); |
5120 | if (delta) | |
5121 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5122 | } |
5123 | ||
3f1d2a31 PZ |
5124 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5125 | { | |
5126 | } | |
5127 | ||
5128 | static const struct sched_class fake_sched_class = { | |
5129 | .put_prev_task = put_prev_task_fake, | |
5130 | }; | |
5131 | ||
5132 | static struct task_struct fake_task = { | |
5133 | /* | |
5134 | * Avoid pull_{rt,dl}_task() | |
5135 | */ | |
5136 | .prio = MAX_PRIO + 1, | |
5137 | .sched_class = &fake_sched_class, | |
5138 | }; | |
5139 | ||
48f24c4d | 5140 | /* |
48c5ccae PZ |
5141 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5142 | * try_to_wake_up()->select_task_rq(). | |
5143 | * | |
5144 | * Called with rq->lock held even though we'er in stop_machine() and | |
5145 | * there's no concurrency possible, we hold the required locks anyway | |
5146 | * because of lock validation efforts. | |
1da177e4 | 5147 | */ |
5e16bbc2 | 5148 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5149 | { |
5e16bbc2 | 5150 | struct rq *rq = dead_rq; |
48c5ccae PZ |
5151 | struct task_struct *next, *stop = rq->stop; |
5152 | int dest_cpu; | |
1da177e4 LT |
5153 | |
5154 | /* | |
48c5ccae PZ |
5155 | * Fudge the rq selection such that the below task selection loop |
5156 | * doesn't get stuck on the currently eligible stop task. | |
5157 | * | |
5158 | * We're currently inside stop_machine() and the rq is either stuck | |
5159 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5160 | * either way we should never end up calling schedule() until we're | |
5161 | * done here. | |
1da177e4 | 5162 | */ |
48c5ccae | 5163 | rq->stop = NULL; |
48f24c4d | 5164 | |
77bd3970 FW |
5165 | /* |
5166 | * put_prev_task() and pick_next_task() sched | |
5167 | * class method both need to have an up-to-date | |
5168 | * value of rq->clock[_task] | |
5169 | */ | |
5170 | update_rq_clock(rq); | |
5171 | ||
5e16bbc2 | 5172 | for (;;) { |
48c5ccae PZ |
5173 | /* |
5174 | * There's this thread running, bail when that's the only | |
5175 | * remaining thread. | |
5176 | */ | |
5177 | if (rq->nr_running == 1) | |
dd41f596 | 5178 | break; |
48c5ccae | 5179 | |
cbce1a68 PZ |
5180 | /* |
5181 | * Ensure rq->lock covers the entire task selection | |
5182 | * until the migration. | |
5183 | */ | |
5184 | lockdep_pin_lock(&rq->lock); | |
3f1d2a31 | 5185 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5186 | BUG_ON(!next); |
79c53799 | 5187 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5188 | |
48c5ccae | 5189 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5190 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5191 | |
cbce1a68 | 5192 | lockdep_unpin_lock(&rq->lock); |
5e16bbc2 PZ |
5193 | rq = __migrate_task(rq, next, dest_cpu); |
5194 | if (rq != dead_rq) { | |
5195 | raw_spin_unlock(&rq->lock); | |
5196 | rq = dead_rq; | |
5197 | raw_spin_lock(&rq->lock); | |
5198 | } | |
1da177e4 | 5199 | } |
dce48a84 | 5200 | |
48c5ccae | 5201 | rq->stop = stop; |
dce48a84 | 5202 | } |
1da177e4 LT |
5203 | #endif /* CONFIG_HOTPLUG_CPU */ |
5204 | ||
e692ab53 NP |
5205 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5206 | ||
5207 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5208 | { |
5209 | .procname = "sched_domain", | |
c57baf1e | 5210 | .mode = 0555, |
e0361851 | 5211 | }, |
56992309 | 5212 | {} |
e692ab53 NP |
5213 | }; |
5214 | ||
5215 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5216 | { |
5217 | .procname = "kernel", | |
c57baf1e | 5218 | .mode = 0555, |
e0361851 AD |
5219 | .child = sd_ctl_dir, |
5220 | }, | |
56992309 | 5221 | {} |
e692ab53 NP |
5222 | }; |
5223 | ||
5224 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5225 | { | |
5226 | struct ctl_table *entry = | |
5cf9f062 | 5227 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5228 | |
e692ab53 NP |
5229 | return entry; |
5230 | } | |
5231 | ||
6382bc90 MM |
5232 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5233 | { | |
cd790076 | 5234 | struct ctl_table *entry; |
6382bc90 | 5235 | |
cd790076 MM |
5236 | /* |
5237 | * In the intermediate directories, both the child directory and | |
5238 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5239 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5240 | * static strings and all have proc handlers. |
5241 | */ | |
5242 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5243 | if (entry->child) |
5244 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5245 | if (entry->proc_handler == NULL) |
5246 | kfree(entry->procname); | |
5247 | } | |
6382bc90 MM |
5248 | |
5249 | kfree(*tablep); | |
5250 | *tablep = NULL; | |
5251 | } | |
5252 | ||
201c373e | 5253 | static int min_load_idx = 0; |
fd9b86d3 | 5254 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5255 | |
e692ab53 | 5256 | static void |
e0361851 | 5257 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5258 | const char *procname, void *data, int maxlen, |
201c373e NK |
5259 | umode_t mode, proc_handler *proc_handler, |
5260 | bool load_idx) | |
e692ab53 | 5261 | { |
e692ab53 NP |
5262 | entry->procname = procname; |
5263 | entry->data = data; | |
5264 | entry->maxlen = maxlen; | |
5265 | entry->mode = mode; | |
5266 | entry->proc_handler = proc_handler; | |
201c373e NK |
5267 | |
5268 | if (load_idx) { | |
5269 | entry->extra1 = &min_load_idx; | |
5270 | entry->extra2 = &max_load_idx; | |
5271 | } | |
e692ab53 NP |
5272 | } |
5273 | ||
5274 | static struct ctl_table * | |
5275 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5276 | { | |
37e6bae8 | 5277 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 5278 | |
ad1cdc1d MM |
5279 | if (table == NULL) |
5280 | return NULL; | |
5281 | ||
e0361851 | 5282 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5283 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5284 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5285 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5286 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5287 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5288 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5289 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5290 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5291 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5292 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5293 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5294 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5295 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5296 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5297 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5298 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5299 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5300 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5301 | &sd->cache_nice_tries, |
201c373e | 5302 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5303 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5304 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
5305 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5306 | &sd->max_newidle_lb_cost, | |
5307 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5308 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5309 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5310 | /* &table[13] is terminator */ |
e692ab53 NP |
5311 | |
5312 | return table; | |
5313 | } | |
5314 | ||
be7002e6 | 5315 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5316 | { |
5317 | struct ctl_table *entry, *table; | |
5318 | struct sched_domain *sd; | |
5319 | int domain_num = 0, i; | |
5320 | char buf[32]; | |
5321 | ||
5322 | for_each_domain(cpu, sd) | |
5323 | domain_num++; | |
5324 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5325 | if (table == NULL) |
5326 | return NULL; | |
e692ab53 NP |
5327 | |
5328 | i = 0; | |
5329 | for_each_domain(cpu, sd) { | |
5330 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5331 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5332 | entry->mode = 0555; |
e692ab53 NP |
5333 | entry->child = sd_alloc_ctl_domain_table(sd); |
5334 | entry++; | |
5335 | i++; | |
5336 | } | |
5337 | return table; | |
5338 | } | |
5339 | ||
5340 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5341 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5342 | { |
6ad4c188 | 5343 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5344 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5345 | char buf[32]; | |
5346 | ||
7378547f MM |
5347 | WARN_ON(sd_ctl_dir[0].child); |
5348 | sd_ctl_dir[0].child = entry; | |
5349 | ||
ad1cdc1d MM |
5350 | if (entry == NULL) |
5351 | return; | |
5352 | ||
6ad4c188 | 5353 | for_each_possible_cpu(i) { |
e692ab53 | 5354 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5355 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5356 | entry->mode = 0555; |
e692ab53 | 5357 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5358 | entry++; |
e692ab53 | 5359 | } |
7378547f MM |
5360 | |
5361 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5362 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5363 | } | |
6382bc90 | 5364 | |
7378547f | 5365 | /* may be called multiple times per register */ |
6382bc90 MM |
5366 | static void unregister_sched_domain_sysctl(void) |
5367 | { | |
781b0203 | 5368 | unregister_sysctl_table(sd_sysctl_header); |
6382bc90 | 5369 | sd_sysctl_header = NULL; |
7378547f MM |
5370 | if (sd_ctl_dir[0].child) |
5371 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5372 | } |
e692ab53 | 5373 | #else |
6382bc90 MM |
5374 | static void register_sched_domain_sysctl(void) |
5375 | { | |
5376 | } | |
5377 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5378 | { |
5379 | } | |
5cc389bc | 5380 | #endif /* CONFIG_SCHED_DEBUG && CONFIG_SYSCTL */ |
e692ab53 | 5381 | |
1f11eb6a GH |
5382 | static void set_rq_online(struct rq *rq) |
5383 | { | |
5384 | if (!rq->online) { | |
5385 | const struct sched_class *class; | |
5386 | ||
c6c4927b | 5387 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5388 | rq->online = 1; |
5389 | ||
5390 | for_each_class(class) { | |
5391 | if (class->rq_online) | |
5392 | class->rq_online(rq); | |
5393 | } | |
5394 | } | |
5395 | } | |
5396 | ||
5397 | static void set_rq_offline(struct rq *rq) | |
5398 | { | |
5399 | if (rq->online) { | |
5400 | const struct sched_class *class; | |
5401 | ||
5402 | for_each_class(class) { | |
5403 | if (class->rq_offline) | |
5404 | class->rq_offline(rq); | |
5405 | } | |
5406 | ||
c6c4927b | 5407 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5408 | rq->online = 0; |
5409 | } | |
5410 | } | |
5411 | ||
1da177e4 LT |
5412 | /* |
5413 | * migration_call - callback that gets triggered when a CPU is added. | |
5414 | * Here we can start up the necessary migration thread for the new CPU. | |
5415 | */ | |
0db0628d | 5416 | static int |
48f24c4d | 5417 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5418 | { |
48f24c4d | 5419 | int cpu = (long)hcpu; |
1da177e4 | 5420 | unsigned long flags; |
969c7921 | 5421 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5422 | |
48c5ccae | 5423 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5424 | |
1da177e4 | 5425 | case CPU_UP_PREPARE: |
a468d389 | 5426 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5427 | break; |
48f24c4d | 5428 | |
1da177e4 | 5429 | case CPU_ONLINE: |
1f94ef59 | 5430 | /* Update our root-domain */ |
05fa785c | 5431 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5432 | if (rq->rd) { |
c6c4927b | 5433 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5434 | |
5435 | set_rq_online(rq); | |
1f94ef59 | 5436 | } |
05fa785c | 5437 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5438 | break; |
48f24c4d | 5439 | |
1da177e4 | 5440 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5441 | case CPU_DYING: |
317f3941 | 5442 | sched_ttwu_pending(); |
57d885fe | 5443 | /* Update our root-domain */ |
05fa785c | 5444 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5445 | if (rq->rd) { |
c6c4927b | 5446 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5447 | set_rq_offline(rq); |
57d885fe | 5448 | } |
5e16bbc2 | 5449 | migrate_tasks(rq); |
48c5ccae | 5450 | BUG_ON(rq->nr_running != 1); /* the migration thread */ |
05fa785c | 5451 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5452 | break; |
48c5ccae | 5453 | |
5d180232 | 5454 | case CPU_DEAD: |
f319da0c | 5455 | calc_load_migrate(rq); |
57d885fe | 5456 | break; |
1da177e4 LT |
5457 | #endif |
5458 | } | |
49c022e6 PZ |
5459 | |
5460 | update_max_interval(); | |
5461 | ||
1da177e4 LT |
5462 | return NOTIFY_OK; |
5463 | } | |
5464 | ||
f38b0820 PM |
5465 | /* |
5466 | * Register at high priority so that task migration (migrate_all_tasks) | |
5467 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5468 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5469 | */ |
0db0628d | 5470 | static struct notifier_block migration_notifier = { |
1da177e4 | 5471 | .notifier_call = migration_call, |
50a323b7 | 5472 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5473 | }; |
5474 | ||
6a82b60d | 5475 | static void set_cpu_rq_start_time(void) |
a803f026 CM |
5476 | { |
5477 | int cpu = smp_processor_id(); | |
5478 | struct rq *rq = cpu_rq(cpu); | |
5479 | rq->age_stamp = sched_clock_cpu(cpu); | |
5480 | } | |
5481 | ||
0db0628d | 5482 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5483 | unsigned long action, void *hcpu) |
5484 | { | |
5485 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5486 | case CPU_STARTING: |
5487 | set_cpu_rq_start_time(); | |
5488 | return NOTIFY_OK; | |
dd9d3843 JS |
5489 | case CPU_ONLINE: |
5490 | /* | |
5491 | * At this point a starting CPU has marked itself as online via | |
5492 | * set_cpu_online(). But it might not yet have marked itself | |
5493 | * as active, which is essential from here on. | |
5494 | * | |
5495 | * Thus, fall-through and help the starting CPU along. | |
5496 | */ | |
3a101d05 TH |
5497 | case CPU_DOWN_FAILED: |
5498 | set_cpu_active((long)hcpu, true); | |
5499 | return NOTIFY_OK; | |
5500 | default: | |
5501 | return NOTIFY_DONE; | |
5502 | } | |
5503 | } | |
5504 | ||
0db0628d | 5505 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5506 | unsigned long action, void *hcpu) |
5507 | { | |
5508 | switch (action & ~CPU_TASKS_FROZEN) { | |
5509 | case CPU_DOWN_PREPARE: | |
3c18d447 | 5510 | set_cpu_active((long)hcpu, false); |
3a101d05 | 5511 | return NOTIFY_OK; |
3c18d447 JL |
5512 | default: |
5513 | return NOTIFY_DONE; | |
3a101d05 TH |
5514 | } |
5515 | } | |
5516 | ||
7babe8db | 5517 | static int __init migration_init(void) |
1da177e4 LT |
5518 | { |
5519 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5520 | int err; |
48f24c4d | 5521 | |
3a101d05 | 5522 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5523 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5524 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5525 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5526 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5527 | |
3a101d05 TH |
5528 | /* Register cpu active notifiers */ |
5529 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5530 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5531 | ||
a004cd42 | 5532 | return 0; |
1da177e4 | 5533 | } |
7babe8db | 5534 | early_initcall(migration_init); |
476f3534 | 5535 | |
4cb98839 PZ |
5536 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5537 | ||
3e9830dc | 5538 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5539 | |
d039ac60 | 5540 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5541 | |
d039ac60 | 5542 | static int __init sched_debug_setup(char *str) |
f6630114 | 5543 | { |
d039ac60 | 5544 | sched_debug_enabled = 1; |
f6630114 MT |
5545 | |
5546 | return 0; | |
5547 | } | |
d039ac60 PZ |
5548 | early_param("sched_debug", sched_debug_setup); |
5549 | ||
5550 | static inline bool sched_debug(void) | |
5551 | { | |
5552 | return sched_debug_enabled; | |
5553 | } | |
f6630114 | 5554 | |
7c16ec58 | 5555 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5556 | struct cpumask *groupmask) |
1da177e4 | 5557 | { |
4dcf6aff | 5558 | struct sched_group *group = sd->groups; |
1da177e4 | 5559 | |
96f874e2 | 5560 | cpumask_clear(groupmask); |
4dcf6aff IM |
5561 | |
5562 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5563 | ||
5564 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5565 | printk("does not load-balance\n"); |
4dcf6aff | 5566 | if (sd->parent) |
3df0fc5b PZ |
5567 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5568 | " has parent"); | |
4dcf6aff | 5569 | return -1; |
41c7ce9a NP |
5570 | } |
5571 | ||
333470ee TH |
5572 | printk(KERN_CONT "span %*pbl level %s\n", |
5573 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5574 | |
758b2cdc | 5575 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5576 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5577 | "CPU%d\n", cpu); | |
4dcf6aff | 5578 | } |
758b2cdc | 5579 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5580 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5581 | " CPU%d\n", cpu); | |
4dcf6aff | 5582 | } |
1da177e4 | 5583 | |
4dcf6aff | 5584 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5585 | do { |
4dcf6aff | 5586 | if (!group) { |
3df0fc5b PZ |
5587 | printk("\n"); |
5588 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5589 | break; |
5590 | } | |
5591 | ||
758b2cdc | 5592 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5593 | printk(KERN_CONT "\n"); |
5594 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5595 | break; |
5596 | } | |
1da177e4 | 5597 | |
cb83b629 PZ |
5598 | if (!(sd->flags & SD_OVERLAP) && |
5599 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5600 | printk(KERN_CONT "\n"); |
5601 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5602 | break; |
5603 | } | |
1da177e4 | 5604 | |
758b2cdc | 5605 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5606 | |
333470ee TH |
5607 | printk(KERN_CONT " %*pbl", |
5608 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5609 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5610 | printk(KERN_CONT " (cpu_capacity = %d)", |
5611 | group->sgc->capacity); | |
381512cf | 5612 | } |
1da177e4 | 5613 | |
4dcf6aff IM |
5614 | group = group->next; |
5615 | } while (group != sd->groups); | |
3df0fc5b | 5616 | printk(KERN_CONT "\n"); |
1da177e4 | 5617 | |
758b2cdc | 5618 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5619 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5620 | |
758b2cdc RR |
5621 | if (sd->parent && |
5622 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5623 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5624 | "of domain->span\n"); | |
4dcf6aff IM |
5625 | return 0; |
5626 | } | |
1da177e4 | 5627 | |
4dcf6aff IM |
5628 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5629 | { | |
5630 | int level = 0; | |
1da177e4 | 5631 | |
d039ac60 | 5632 | if (!sched_debug_enabled) |
f6630114 MT |
5633 | return; |
5634 | ||
4dcf6aff IM |
5635 | if (!sd) { |
5636 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5637 | return; | |
5638 | } | |
1da177e4 | 5639 | |
4dcf6aff IM |
5640 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5641 | ||
5642 | for (;;) { | |
4cb98839 | 5643 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5644 | break; |
1da177e4 LT |
5645 | level++; |
5646 | sd = sd->parent; | |
33859f7f | 5647 | if (!sd) |
4dcf6aff IM |
5648 | break; |
5649 | } | |
1da177e4 | 5650 | } |
6d6bc0ad | 5651 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5652 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5653 | static inline bool sched_debug(void) |
5654 | { | |
5655 | return false; | |
5656 | } | |
6d6bc0ad | 5657 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5658 | |
1a20ff27 | 5659 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5660 | { |
758b2cdc | 5661 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5662 | return 1; |
5663 | ||
5664 | /* Following flags need at least 2 groups */ | |
5665 | if (sd->flags & (SD_LOAD_BALANCE | | |
5666 | SD_BALANCE_NEWIDLE | | |
5667 | SD_BALANCE_FORK | | |
89c4710e | 5668 | SD_BALANCE_EXEC | |
5d4dfddd | 5669 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5670 | SD_SHARE_PKG_RESOURCES | |
5671 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5672 | if (sd->groups != sd->groups->next) |
5673 | return 0; | |
5674 | } | |
5675 | ||
5676 | /* Following flags don't use groups */ | |
c88d5910 | 5677 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5678 | return 0; |
5679 | ||
5680 | return 1; | |
5681 | } | |
5682 | ||
48f24c4d IM |
5683 | static int |
5684 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5685 | { |
5686 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5687 | ||
5688 | if (sd_degenerate(parent)) | |
5689 | return 1; | |
5690 | ||
758b2cdc | 5691 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5692 | return 0; |
5693 | ||
245af2c7 SS |
5694 | /* Flags needing groups don't count if only 1 group in parent */ |
5695 | if (parent->groups == parent->groups->next) { | |
5696 | pflags &= ~(SD_LOAD_BALANCE | | |
5697 | SD_BALANCE_NEWIDLE | | |
5698 | SD_BALANCE_FORK | | |
89c4710e | 5699 | SD_BALANCE_EXEC | |
5d4dfddd | 5700 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5701 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5702 | SD_PREFER_SIBLING | |
5703 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5704 | if (nr_node_ids == 1) |
5705 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5706 | } |
5707 | if (~cflags & pflags) | |
5708 | return 0; | |
5709 | ||
5710 | return 1; | |
5711 | } | |
5712 | ||
dce840a0 | 5713 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5714 | { |
dce840a0 | 5715 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5716 | |
68e74568 | 5717 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5718 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5719 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5720 | free_cpumask_var(rd->rto_mask); |
5721 | free_cpumask_var(rd->online); | |
5722 | free_cpumask_var(rd->span); | |
5723 | kfree(rd); | |
5724 | } | |
5725 | ||
57d885fe GH |
5726 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5727 | { | |
a0490fa3 | 5728 | struct root_domain *old_rd = NULL; |
57d885fe | 5729 | unsigned long flags; |
57d885fe | 5730 | |
05fa785c | 5731 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5732 | |
5733 | if (rq->rd) { | |
a0490fa3 | 5734 | old_rd = rq->rd; |
57d885fe | 5735 | |
c6c4927b | 5736 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5737 | set_rq_offline(rq); |
57d885fe | 5738 | |
c6c4927b | 5739 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5740 | |
a0490fa3 | 5741 | /* |
0515973f | 5742 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5743 | * set old_rd to NULL to skip the freeing later |
5744 | * in this function: | |
5745 | */ | |
5746 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5747 | old_rd = NULL; | |
57d885fe GH |
5748 | } |
5749 | ||
5750 | atomic_inc(&rd->refcount); | |
5751 | rq->rd = rd; | |
5752 | ||
c6c4927b | 5753 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5754 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5755 | set_rq_online(rq); |
57d885fe | 5756 | |
05fa785c | 5757 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5758 | |
5759 | if (old_rd) | |
dce840a0 | 5760 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5761 | } |
5762 | ||
68c38fc3 | 5763 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5764 | { |
5765 | memset(rd, 0, sizeof(*rd)); | |
5766 | ||
68c38fc3 | 5767 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5768 | goto out; |
68c38fc3 | 5769 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5770 | goto free_span; |
1baca4ce | 5771 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5772 | goto free_online; |
1baca4ce JL |
5773 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5774 | goto free_dlo_mask; | |
6e0534f2 | 5775 | |
332ac17e | 5776 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5777 | if (cpudl_init(&rd->cpudl) != 0) |
5778 | goto free_dlo_mask; | |
332ac17e | 5779 | |
68c38fc3 | 5780 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5781 | goto free_rto_mask; |
c6c4927b | 5782 | return 0; |
6e0534f2 | 5783 | |
68e74568 RR |
5784 | free_rto_mask: |
5785 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5786 | free_dlo_mask: |
5787 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5788 | free_online: |
5789 | free_cpumask_var(rd->online); | |
5790 | free_span: | |
5791 | free_cpumask_var(rd->span); | |
0c910d28 | 5792 | out: |
c6c4927b | 5793 | return -ENOMEM; |
57d885fe GH |
5794 | } |
5795 | ||
029632fb PZ |
5796 | /* |
5797 | * By default the system creates a single root-domain with all cpus as | |
5798 | * members (mimicking the global state we have today). | |
5799 | */ | |
5800 | struct root_domain def_root_domain; | |
5801 | ||
57d885fe GH |
5802 | static void init_defrootdomain(void) |
5803 | { | |
68c38fc3 | 5804 | init_rootdomain(&def_root_domain); |
c6c4927b | 5805 | |
57d885fe GH |
5806 | atomic_set(&def_root_domain.refcount, 1); |
5807 | } | |
5808 | ||
dc938520 | 5809 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5810 | { |
5811 | struct root_domain *rd; | |
5812 | ||
5813 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5814 | if (!rd) | |
5815 | return NULL; | |
5816 | ||
68c38fc3 | 5817 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5818 | kfree(rd); |
5819 | return NULL; | |
5820 | } | |
57d885fe GH |
5821 | |
5822 | return rd; | |
5823 | } | |
5824 | ||
63b2ca30 | 5825 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5826 | { |
5827 | struct sched_group *tmp, *first; | |
5828 | ||
5829 | if (!sg) | |
5830 | return; | |
5831 | ||
5832 | first = sg; | |
5833 | do { | |
5834 | tmp = sg->next; | |
5835 | ||
63b2ca30 NP |
5836 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5837 | kfree(sg->sgc); | |
e3589f6c PZ |
5838 | |
5839 | kfree(sg); | |
5840 | sg = tmp; | |
5841 | } while (sg != first); | |
5842 | } | |
5843 | ||
dce840a0 PZ |
5844 | static void free_sched_domain(struct rcu_head *rcu) |
5845 | { | |
5846 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5847 | |
5848 | /* | |
5849 | * If its an overlapping domain it has private groups, iterate and | |
5850 | * nuke them all. | |
5851 | */ | |
5852 | if (sd->flags & SD_OVERLAP) { | |
5853 | free_sched_groups(sd->groups, 1); | |
5854 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5855 | kfree(sd->groups->sgc); |
dce840a0 | 5856 | kfree(sd->groups); |
9c3f75cb | 5857 | } |
dce840a0 PZ |
5858 | kfree(sd); |
5859 | } | |
5860 | ||
5861 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5862 | { | |
5863 | call_rcu(&sd->rcu, free_sched_domain); | |
5864 | } | |
5865 | ||
5866 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5867 | { | |
5868 | for (; sd; sd = sd->parent) | |
5869 | destroy_sched_domain(sd, cpu); | |
5870 | } | |
5871 | ||
518cd623 PZ |
5872 | /* |
5873 | * Keep a special pointer to the highest sched_domain that has | |
5874 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5875 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5876 | * | |
5877 | * Also keep a unique ID per domain (we use the first cpu number in | |
5878 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5879 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5880 | */ |
5881 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5882 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5883 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5884 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5885 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5886 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5887 | |
5888 | static void update_top_cache_domain(int cpu) | |
5889 | { | |
5890 | struct sched_domain *sd; | |
5d4cf996 | 5891 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5892 | int id = cpu; |
7d9ffa89 | 5893 | int size = 1; |
518cd623 PZ |
5894 | |
5895 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5896 | if (sd) { |
518cd623 | 5897 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5898 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5899 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5900 | } |
5d4cf996 | 5901 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5902 | |
5903 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5904 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5905 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5906 | |
5907 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5908 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5909 | |
5910 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5911 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5912 | } |
5913 | ||
1da177e4 | 5914 | /* |
0eab9146 | 5915 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5916 | * hold the hotplug lock. |
5917 | */ | |
0eab9146 IM |
5918 | static void |
5919 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5920 | { |
70b97a7f | 5921 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5922 | struct sched_domain *tmp; |
5923 | ||
5924 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5925 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5926 | struct sched_domain *parent = tmp->parent; |
5927 | if (!parent) | |
5928 | break; | |
f29c9b1c | 5929 | |
1a848870 | 5930 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5931 | tmp->parent = parent->parent; |
1a848870 SS |
5932 | if (parent->parent) |
5933 | parent->parent->child = tmp; | |
10866e62 PZ |
5934 | /* |
5935 | * Transfer SD_PREFER_SIBLING down in case of a | |
5936 | * degenerate parent; the spans match for this | |
5937 | * so the property transfers. | |
5938 | */ | |
5939 | if (parent->flags & SD_PREFER_SIBLING) | |
5940 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5941 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5942 | } else |
5943 | tmp = tmp->parent; | |
245af2c7 SS |
5944 | } |
5945 | ||
1a848870 | 5946 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5947 | tmp = sd; |
245af2c7 | 5948 | sd = sd->parent; |
dce840a0 | 5949 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5950 | if (sd) |
5951 | sd->child = NULL; | |
5952 | } | |
1da177e4 | 5953 | |
4cb98839 | 5954 | sched_domain_debug(sd, cpu); |
1da177e4 | 5955 | |
57d885fe | 5956 | rq_attach_root(rq, rd); |
dce840a0 | 5957 | tmp = rq->sd; |
674311d5 | 5958 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5959 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5960 | |
5961 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5962 | } |
5963 | ||
1da177e4 LT |
5964 | /* Setup the mask of cpus configured for isolated domains */ |
5965 | static int __init isolated_cpu_setup(char *str) | |
5966 | { | |
bdddd296 | 5967 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5968 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5969 | return 1; |
5970 | } | |
5971 | ||
8927f494 | 5972 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5973 | |
49a02c51 | 5974 | struct s_data { |
21d42ccf | 5975 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5976 | struct root_domain *rd; |
5977 | }; | |
5978 | ||
2109b99e | 5979 | enum s_alloc { |
2109b99e | 5980 | sa_rootdomain, |
21d42ccf | 5981 | sa_sd, |
dce840a0 | 5982 | sa_sd_storage, |
2109b99e AH |
5983 | sa_none, |
5984 | }; | |
5985 | ||
c1174876 PZ |
5986 | /* |
5987 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5988 | * domain traversal. | |
5989 | * | |
5990 | * Asymmetric node setups can result in situations where the domain tree is of | |
5991 | * unequal depth, make sure to skip domains that already cover the entire | |
5992 | * range. | |
5993 | * | |
5994 | * In that case build_sched_domains() will have terminated the iteration early | |
5995 | * and our sibling sd spans will be empty. Domains should always include the | |
5996 | * cpu they're built on, so check that. | |
5997 | * | |
5998 | */ | |
5999 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6000 | { | |
6001 | const struct cpumask *span = sched_domain_span(sd); | |
6002 | struct sd_data *sdd = sd->private; | |
6003 | struct sched_domain *sibling; | |
6004 | int i; | |
6005 | ||
6006 | for_each_cpu(i, span) { | |
6007 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6008 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6009 | continue; | |
6010 | ||
6011 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6012 | } | |
6013 | } | |
6014 | ||
6015 | /* | |
6016 | * Return the canonical balance cpu for this group, this is the first cpu | |
6017 | * of this group that's also in the iteration mask. | |
6018 | */ | |
6019 | int group_balance_cpu(struct sched_group *sg) | |
6020 | { | |
6021 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6022 | } | |
6023 | ||
e3589f6c PZ |
6024 | static int |
6025 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6026 | { | |
6027 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6028 | const struct cpumask *span = sched_domain_span(sd); | |
6029 | struct cpumask *covered = sched_domains_tmpmask; | |
6030 | struct sd_data *sdd = sd->private; | |
aaecac4a | 6031 | struct sched_domain *sibling; |
e3589f6c PZ |
6032 | int i; |
6033 | ||
6034 | cpumask_clear(covered); | |
6035 | ||
6036 | for_each_cpu(i, span) { | |
6037 | struct cpumask *sg_span; | |
6038 | ||
6039 | if (cpumask_test_cpu(i, covered)) | |
6040 | continue; | |
6041 | ||
aaecac4a | 6042 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
6043 | |
6044 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 6045 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
6046 | continue; |
6047 | ||
e3589f6c | 6048 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6049 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6050 | |
6051 | if (!sg) | |
6052 | goto fail; | |
6053 | ||
6054 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
6055 | if (sibling->child) |
6056 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
6057 | else | |
e3589f6c PZ |
6058 | cpumask_set_cpu(i, sg_span); |
6059 | ||
6060 | cpumask_or(covered, covered, sg_span); | |
6061 | ||
63b2ca30 NP |
6062 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
6063 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
6064 | build_group_mask(sd, sg); |
6065 | ||
c3decf0d | 6066 | /* |
63b2ca30 | 6067 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
6068 | * domains and no possible iteration will get us here, we won't |
6069 | * die on a /0 trap. | |
6070 | */ | |
ca8ce3d0 | 6071 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 6072 | |
c1174876 PZ |
6073 | /* |
6074 | * Make sure the first group of this domain contains the | |
6075 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6076 | * breaks. See update_sg_lb_stats(). | |
6077 | */ | |
74a5ce20 | 6078 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6079 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6080 | groups = sg; |
6081 | ||
6082 | if (!first) | |
6083 | first = sg; | |
6084 | if (last) | |
6085 | last->next = sg; | |
6086 | last = sg; | |
6087 | last->next = first; | |
6088 | } | |
6089 | sd->groups = groups; | |
6090 | ||
6091 | return 0; | |
6092 | ||
6093 | fail: | |
6094 | free_sched_groups(first, 0); | |
6095 | ||
6096 | return -ENOMEM; | |
6097 | } | |
6098 | ||
dce840a0 | 6099 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6100 | { |
dce840a0 PZ |
6101 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6102 | struct sched_domain *child = sd->child; | |
1da177e4 | 6103 | |
dce840a0 PZ |
6104 | if (child) |
6105 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6106 | |
9c3f75cb | 6107 | if (sg) { |
dce840a0 | 6108 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6109 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6110 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6111 | } |
dce840a0 PZ |
6112 | |
6113 | return cpu; | |
1e9f28fa | 6114 | } |
1e9f28fa | 6115 | |
01a08546 | 6116 | /* |
dce840a0 PZ |
6117 | * build_sched_groups will build a circular linked list of the groups |
6118 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6119 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6120 | * |
6121 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6122 | */ |
e3589f6c PZ |
6123 | static int |
6124 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6125 | { |
dce840a0 PZ |
6126 | struct sched_group *first = NULL, *last = NULL; |
6127 | struct sd_data *sdd = sd->private; | |
6128 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6129 | struct cpumask *covered; |
dce840a0 | 6130 | int i; |
9c1cfda2 | 6131 | |
e3589f6c PZ |
6132 | get_group(cpu, sdd, &sd->groups); |
6133 | atomic_inc(&sd->groups->ref); | |
6134 | ||
0936629f | 6135 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6136 | return 0; |
6137 | ||
f96225fd PZ |
6138 | lockdep_assert_held(&sched_domains_mutex); |
6139 | covered = sched_domains_tmpmask; | |
6140 | ||
dce840a0 | 6141 | cpumask_clear(covered); |
6711cab4 | 6142 | |
dce840a0 PZ |
6143 | for_each_cpu(i, span) { |
6144 | struct sched_group *sg; | |
cd08e923 | 6145 | int group, j; |
6711cab4 | 6146 | |
dce840a0 PZ |
6147 | if (cpumask_test_cpu(i, covered)) |
6148 | continue; | |
6711cab4 | 6149 | |
cd08e923 | 6150 | group = get_group(i, sdd, &sg); |
c1174876 | 6151 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6152 | |
dce840a0 PZ |
6153 | for_each_cpu(j, span) { |
6154 | if (get_group(j, sdd, NULL) != group) | |
6155 | continue; | |
0601a88d | 6156 | |
dce840a0 PZ |
6157 | cpumask_set_cpu(j, covered); |
6158 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6159 | } | |
0601a88d | 6160 | |
dce840a0 PZ |
6161 | if (!first) |
6162 | first = sg; | |
6163 | if (last) | |
6164 | last->next = sg; | |
6165 | last = sg; | |
6166 | } | |
6167 | last->next = first; | |
e3589f6c PZ |
6168 | |
6169 | return 0; | |
0601a88d | 6170 | } |
51888ca2 | 6171 | |
89c4710e | 6172 | /* |
63b2ca30 | 6173 | * Initialize sched groups cpu_capacity. |
89c4710e | 6174 | * |
63b2ca30 | 6175 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6176 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6177 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6178 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6179 | * group having more cpu_capacity will pickup more load compared to the | |
6180 | * group having less cpu_capacity. | |
89c4710e | 6181 | */ |
63b2ca30 | 6182 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6183 | { |
e3589f6c | 6184 | struct sched_group *sg = sd->groups; |
89c4710e | 6185 | |
94c95ba6 | 6186 | WARN_ON(!sg); |
e3589f6c PZ |
6187 | |
6188 | do { | |
6189 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6190 | sg = sg->next; | |
6191 | } while (sg != sd->groups); | |
89c4710e | 6192 | |
c1174876 | 6193 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6194 | return; |
aae6d3dd | 6195 | |
63b2ca30 NP |
6196 | update_group_capacity(sd, cpu); |
6197 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6198 | } |
6199 | ||
7c16ec58 MT |
6200 | /* |
6201 | * Initializers for schedule domains | |
6202 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6203 | */ | |
6204 | ||
1d3504fc | 6205 | static int default_relax_domain_level = -1; |
60495e77 | 6206 | int sched_domain_level_max; |
1d3504fc HS |
6207 | |
6208 | static int __init setup_relax_domain_level(char *str) | |
6209 | { | |
a841f8ce DS |
6210 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6211 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6212 | |
1d3504fc HS |
6213 | return 1; |
6214 | } | |
6215 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6216 | ||
6217 | static void set_domain_attribute(struct sched_domain *sd, | |
6218 | struct sched_domain_attr *attr) | |
6219 | { | |
6220 | int request; | |
6221 | ||
6222 | if (!attr || attr->relax_domain_level < 0) { | |
6223 | if (default_relax_domain_level < 0) | |
6224 | return; | |
6225 | else | |
6226 | request = default_relax_domain_level; | |
6227 | } else | |
6228 | request = attr->relax_domain_level; | |
6229 | if (request < sd->level) { | |
6230 | /* turn off idle balance on this domain */ | |
c88d5910 | 6231 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6232 | } else { |
6233 | /* turn on idle balance on this domain */ | |
c88d5910 | 6234 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6235 | } |
6236 | } | |
6237 | ||
54ab4ff4 PZ |
6238 | static void __sdt_free(const struct cpumask *cpu_map); |
6239 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6240 | ||
2109b99e AH |
6241 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6242 | const struct cpumask *cpu_map) | |
6243 | { | |
6244 | switch (what) { | |
2109b99e | 6245 | case sa_rootdomain: |
822ff793 PZ |
6246 | if (!atomic_read(&d->rd->refcount)) |
6247 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6248 | case sa_sd: |
6249 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6250 | case sa_sd_storage: |
54ab4ff4 | 6251 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6252 | case sa_none: |
6253 | break; | |
6254 | } | |
6255 | } | |
3404c8d9 | 6256 | |
2109b99e AH |
6257 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6258 | const struct cpumask *cpu_map) | |
6259 | { | |
dce840a0 PZ |
6260 | memset(d, 0, sizeof(*d)); |
6261 | ||
54ab4ff4 PZ |
6262 | if (__sdt_alloc(cpu_map)) |
6263 | return sa_sd_storage; | |
dce840a0 PZ |
6264 | d->sd = alloc_percpu(struct sched_domain *); |
6265 | if (!d->sd) | |
6266 | return sa_sd_storage; | |
2109b99e | 6267 | d->rd = alloc_rootdomain(); |
dce840a0 | 6268 | if (!d->rd) |
21d42ccf | 6269 | return sa_sd; |
2109b99e AH |
6270 | return sa_rootdomain; |
6271 | } | |
57d885fe | 6272 | |
dce840a0 PZ |
6273 | /* |
6274 | * NULL the sd_data elements we've used to build the sched_domain and | |
6275 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6276 | * will not free the data we're using. | |
6277 | */ | |
6278 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6279 | { | |
6280 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6281 | |
6282 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6283 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6284 | ||
e3589f6c | 6285 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6286 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6287 | |
63b2ca30 NP |
6288 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6289 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6290 | } |
6291 | ||
cb83b629 | 6292 | #ifdef CONFIG_NUMA |
cb83b629 | 6293 | static int sched_domains_numa_levels; |
e3fe70b1 | 6294 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6295 | static int *sched_domains_numa_distance; |
9942f79b | 6296 | int sched_max_numa_distance; |
cb83b629 PZ |
6297 | static struct cpumask ***sched_domains_numa_masks; |
6298 | static int sched_domains_curr_level; | |
143e1e28 | 6299 | #endif |
cb83b629 | 6300 | |
143e1e28 VG |
6301 | /* |
6302 | * SD_flags allowed in topology descriptions. | |
6303 | * | |
5d4dfddd | 6304 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6305 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6306 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6307 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6308 | * |
6309 | * Odd one out: | |
6310 | * SD_ASYM_PACKING - describes SMT quirks | |
6311 | */ | |
6312 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6313 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6314 | SD_SHARE_PKG_RESOURCES | \ |
6315 | SD_NUMA | \ | |
d77b3ed5 VG |
6316 | SD_ASYM_PACKING | \ |
6317 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6318 | |
6319 | static struct sched_domain * | |
143e1e28 | 6320 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6321 | { |
6322 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6323 | int sd_weight, sd_flags = 0; |
6324 | ||
6325 | #ifdef CONFIG_NUMA | |
6326 | /* | |
6327 | * Ugly hack to pass state to sd_numa_mask()... | |
6328 | */ | |
6329 | sched_domains_curr_level = tl->numa_level; | |
6330 | #endif | |
6331 | ||
6332 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6333 | ||
6334 | if (tl->sd_flags) | |
6335 | sd_flags = (*tl->sd_flags)(); | |
6336 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6337 | "wrong sd_flags in topology description\n")) | |
6338 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6339 | |
6340 | *sd = (struct sched_domain){ | |
6341 | .min_interval = sd_weight, | |
6342 | .max_interval = 2*sd_weight, | |
6343 | .busy_factor = 32, | |
870a0bb5 | 6344 | .imbalance_pct = 125, |
143e1e28 VG |
6345 | |
6346 | .cache_nice_tries = 0, | |
6347 | .busy_idx = 0, | |
6348 | .idle_idx = 0, | |
cb83b629 PZ |
6349 | .newidle_idx = 0, |
6350 | .wake_idx = 0, | |
6351 | .forkexec_idx = 0, | |
6352 | ||
6353 | .flags = 1*SD_LOAD_BALANCE | |
6354 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6355 | | 1*SD_BALANCE_EXEC |
6356 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6357 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6358 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6359 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6360 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6361 | | 0*SD_SERIALIZE |
cb83b629 | 6362 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6363 | | 0*SD_NUMA |
6364 | | sd_flags | |
cb83b629 | 6365 | , |
143e1e28 | 6366 | |
cb83b629 PZ |
6367 | .last_balance = jiffies, |
6368 | .balance_interval = sd_weight, | |
143e1e28 | 6369 | .smt_gain = 0, |
2b4cfe64 JL |
6370 | .max_newidle_lb_cost = 0, |
6371 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6372 | #ifdef CONFIG_SCHED_DEBUG |
6373 | .name = tl->name, | |
6374 | #endif | |
cb83b629 | 6375 | }; |
cb83b629 PZ |
6376 | |
6377 | /* | |
143e1e28 | 6378 | * Convert topological properties into behaviour. |
cb83b629 | 6379 | */ |
143e1e28 | 6380 | |
5d4dfddd | 6381 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6382 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6383 | sd->imbalance_pct = 110; |
6384 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6385 | |
6386 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6387 | sd->imbalance_pct = 117; | |
6388 | sd->cache_nice_tries = 1; | |
6389 | sd->busy_idx = 2; | |
6390 | ||
6391 | #ifdef CONFIG_NUMA | |
6392 | } else if (sd->flags & SD_NUMA) { | |
6393 | sd->cache_nice_tries = 2; | |
6394 | sd->busy_idx = 3; | |
6395 | sd->idle_idx = 2; | |
6396 | ||
6397 | sd->flags |= SD_SERIALIZE; | |
6398 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6399 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6400 | SD_BALANCE_FORK | | |
6401 | SD_WAKE_AFFINE); | |
6402 | } | |
6403 | ||
6404 | #endif | |
6405 | } else { | |
6406 | sd->flags |= SD_PREFER_SIBLING; | |
6407 | sd->cache_nice_tries = 1; | |
6408 | sd->busy_idx = 2; | |
6409 | sd->idle_idx = 1; | |
6410 | } | |
6411 | ||
6412 | sd->private = &tl->data; | |
cb83b629 PZ |
6413 | |
6414 | return sd; | |
6415 | } | |
6416 | ||
143e1e28 VG |
6417 | /* |
6418 | * Topology list, bottom-up. | |
6419 | */ | |
6420 | static struct sched_domain_topology_level default_topology[] = { | |
6421 | #ifdef CONFIG_SCHED_SMT | |
6422 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6423 | #endif | |
6424 | #ifdef CONFIG_SCHED_MC | |
6425 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6426 | #endif |
6427 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6428 | { NULL, }, | |
6429 | }; | |
6430 | ||
6431 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6432 | ||
6433 | #define for_each_sd_topology(tl) \ | |
6434 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6435 | ||
6436 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6437 | { | |
6438 | sched_domain_topology = tl; | |
6439 | } | |
6440 | ||
6441 | #ifdef CONFIG_NUMA | |
6442 | ||
cb83b629 PZ |
6443 | static const struct cpumask *sd_numa_mask(int cpu) |
6444 | { | |
6445 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6446 | } | |
6447 | ||
d039ac60 PZ |
6448 | static void sched_numa_warn(const char *str) |
6449 | { | |
6450 | static int done = false; | |
6451 | int i,j; | |
6452 | ||
6453 | if (done) | |
6454 | return; | |
6455 | ||
6456 | done = true; | |
6457 | ||
6458 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6459 | ||
6460 | for (i = 0; i < nr_node_ids; i++) { | |
6461 | printk(KERN_WARNING " "); | |
6462 | for (j = 0; j < nr_node_ids; j++) | |
6463 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6464 | printk(KERN_CONT "\n"); | |
6465 | } | |
6466 | printk(KERN_WARNING "\n"); | |
6467 | } | |
6468 | ||
9942f79b | 6469 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6470 | { |
6471 | int i; | |
6472 | ||
6473 | if (distance == node_distance(0, 0)) | |
6474 | return true; | |
6475 | ||
6476 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6477 | if (sched_domains_numa_distance[i] == distance) | |
6478 | return true; | |
6479 | } | |
6480 | ||
6481 | return false; | |
6482 | } | |
6483 | ||
e3fe70b1 RR |
6484 | /* |
6485 | * A system can have three types of NUMA topology: | |
6486 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6487 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6488 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6489 | * | |
6490 | * The difference between a glueless mesh topology and a backplane | |
6491 | * topology lies in whether communication between not directly | |
6492 | * connected nodes goes through intermediary nodes (where programs | |
6493 | * could run), or through backplane controllers. This affects | |
6494 | * placement of programs. | |
6495 | * | |
6496 | * The type of topology can be discerned with the following tests: | |
6497 | * - If the maximum distance between any nodes is 1 hop, the system | |
6498 | * is directly connected. | |
6499 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6500 | * there is an intermediary node C, which is < N hops away from both | |
6501 | * nodes A and B, the system is a glueless mesh. | |
6502 | */ | |
6503 | static void init_numa_topology_type(void) | |
6504 | { | |
6505 | int a, b, c, n; | |
6506 | ||
6507 | n = sched_max_numa_distance; | |
6508 | ||
e237882b | 6509 | if (sched_domains_numa_levels <= 1) { |
e3fe70b1 | 6510 | sched_numa_topology_type = NUMA_DIRECT; |
e237882b AG |
6511 | return; |
6512 | } | |
e3fe70b1 RR |
6513 | |
6514 | for_each_online_node(a) { | |
6515 | for_each_online_node(b) { | |
6516 | /* Find two nodes furthest removed from each other. */ | |
6517 | if (node_distance(a, b) < n) | |
6518 | continue; | |
6519 | ||
6520 | /* Is there an intermediary node between a and b? */ | |
6521 | for_each_online_node(c) { | |
6522 | if (node_distance(a, c) < n && | |
6523 | node_distance(b, c) < n) { | |
6524 | sched_numa_topology_type = | |
6525 | NUMA_GLUELESS_MESH; | |
6526 | return; | |
6527 | } | |
6528 | } | |
6529 | ||
6530 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6531 | return; | |
6532 | } | |
6533 | } | |
6534 | } | |
6535 | ||
cb83b629 PZ |
6536 | static void sched_init_numa(void) |
6537 | { | |
6538 | int next_distance, curr_distance = node_distance(0, 0); | |
6539 | struct sched_domain_topology_level *tl; | |
6540 | int level = 0; | |
6541 | int i, j, k; | |
6542 | ||
cb83b629 PZ |
6543 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6544 | if (!sched_domains_numa_distance) | |
6545 | return; | |
6546 | ||
6547 | /* | |
6548 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6549 | * unique distances in the node_distance() table. | |
6550 | * | |
6551 | * Assumes node_distance(0,j) includes all distances in | |
6552 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6553 | */ |
6554 | next_distance = curr_distance; | |
6555 | for (i = 0; i < nr_node_ids; i++) { | |
6556 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6557 | for (k = 0; k < nr_node_ids; k++) { |
6558 | int distance = node_distance(i, k); | |
6559 | ||
6560 | if (distance > curr_distance && | |
6561 | (distance < next_distance || | |
6562 | next_distance == curr_distance)) | |
6563 | next_distance = distance; | |
6564 | ||
6565 | /* | |
6566 | * While not a strong assumption it would be nice to know | |
6567 | * about cases where if node A is connected to B, B is not | |
6568 | * equally connected to A. | |
6569 | */ | |
6570 | if (sched_debug() && node_distance(k, i) != distance) | |
6571 | sched_numa_warn("Node-distance not symmetric"); | |
6572 | ||
6573 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6574 | sched_numa_warn("Node-0 not representative"); | |
6575 | } | |
6576 | if (next_distance != curr_distance) { | |
6577 | sched_domains_numa_distance[level++] = next_distance; | |
6578 | sched_domains_numa_levels = level; | |
6579 | curr_distance = next_distance; | |
6580 | } else break; | |
cb83b629 | 6581 | } |
d039ac60 PZ |
6582 | |
6583 | /* | |
6584 | * In case of sched_debug() we verify the above assumption. | |
6585 | */ | |
6586 | if (!sched_debug()) | |
6587 | break; | |
cb83b629 | 6588 | } |
c123588b AR |
6589 | |
6590 | if (!level) | |
6591 | return; | |
6592 | ||
cb83b629 PZ |
6593 | /* |
6594 | * 'level' contains the number of unique distances, excluding the | |
6595 | * identity distance node_distance(i,i). | |
6596 | * | |
28b4a521 | 6597 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6598 | * numbers. |
6599 | */ | |
6600 | ||
5f7865f3 TC |
6601 | /* |
6602 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6603 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6604 | * the array will contain less then 'level' members. This could be | |
6605 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6606 | * in other functions. | |
6607 | * | |
6608 | * We reset it to 'level' at the end of this function. | |
6609 | */ | |
6610 | sched_domains_numa_levels = 0; | |
6611 | ||
cb83b629 PZ |
6612 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6613 | if (!sched_domains_numa_masks) | |
6614 | return; | |
6615 | ||
6616 | /* | |
6617 | * Now for each level, construct a mask per node which contains all | |
6618 | * cpus of nodes that are that many hops away from us. | |
6619 | */ | |
6620 | for (i = 0; i < level; i++) { | |
6621 | sched_domains_numa_masks[i] = | |
6622 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6623 | if (!sched_domains_numa_masks[i]) | |
6624 | return; | |
6625 | ||
6626 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6627 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6628 | if (!mask) |
6629 | return; | |
6630 | ||
6631 | sched_domains_numa_masks[i][j] = mask; | |
6632 | ||
6633 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6634 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6635 | continue; |
6636 | ||
6637 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6638 | } | |
6639 | } | |
6640 | } | |
6641 | ||
143e1e28 VG |
6642 | /* Compute default topology size */ |
6643 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6644 | ||
c515db8c | 6645 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6646 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6647 | if (!tl) | |
6648 | return; | |
6649 | ||
6650 | /* | |
6651 | * Copy the default topology bits.. | |
6652 | */ | |
143e1e28 VG |
6653 | for (i = 0; sched_domain_topology[i].mask; i++) |
6654 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6655 | |
6656 | /* | |
6657 | * .. and append 'j' levels of NUMA goodness. | |
6658 | */ | |
6659 | for (j = 0; j < level; i++, j++) { | |
6660 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6661 | .mask = sd_numa_mask, |
143e1e28 | 6662 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6663 | .flags = SDTL_OVERLAP, |
6664 | .numa_level = j, | |
143e1e28 | 6665 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6666 | }; |
6667 | } | |
6668 | ||
6669 | sched_domain_topology = tl; | |
5f7865f3 TC |
6670 | |
6671 | sched_domains_numa_levels = level; | |
9942f79b | 6672 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6673 | |
6674 | init_numa_topology_type(); | |
cb83b629 | 6675 | } |
301a5cba TC |
6676 | |
6677 | static void sched_domains_numa_masks_set(int cpu) | |
6678 | { | |
6679 | int i, j; | |
6680 | int node = cpu_to_node(cpu); | |
6681 | ||
6682 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6683 | for (j = 0; j < nr_node_ids; j++) { | |
6684 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6685 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6686 | } | |
6687 | } | |
6688 | } | |
6689 | ||
6690 | static void sched_domains_numa_masks_clear(int cpu) | |
6691 | { | |
6692 | int i, j; | |
6693 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6694 | for (j = 0; j < nr_node_ids; j++) | |
6695 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6696 | } | |
6697 | } | |
6698 | ||
6699 | /* | |
6700 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6701 | * are onlined. | |
6702 | */ | |
6703 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6704 | unsigned long action, | |
6705 | void *hcpu) | |
6706 | { | |
6707 | int cpu = (long)hcpu; | |
6708 | ||
6709 | switch (action & ~CPU_TASKS_FROZEN) { | |
6710 | case CPU_ONLINE: | |
6711 | sched_domains_numa_masks_set(cpu); | |
6712 | break; | |
6713 | ||
6714 | case CPU_DEAD: | |
6715 | sched_domains_numa_masks_clear(cpu); | |
6716 | break; | |
6717 | ||
6718 | default: | |
6719 | return NOTIFY_DONE; | |
6720 | } | |
6721 | ||
6722 | return NOTIFY_OK; | |
cb83b629 PZ |
6723 | } |
6724 | #else | |
6725 | static inline void sched_init_numa(void) | |
6726 | { | |
6727 | } | |
301a5cba TC |
6728 | |
6729 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6730 | unsigned long action, | |
6731 | void *hcpu) | |
6732 | { | |
6733 | return 0; | |
6734 | } | |
cb83b629 PZ |
6735 | #endif /* CONFIG_NUMA */ |
6736 | ||
54ab4ff4 PZ |
6737 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6738 | { | |
6739 | struct sched_domain_topology_level *tl; | |
6740 | int j; | |
6741 | ||
27723a68 | 6742 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6743 | struct sd_data *sdd = &tl->data; |
6744 | ||
6745 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6746 | if (!sdd->sd) | |
6747 | return -ENOMEM; | |
6748 | ||
6749 | sdd->sg = alloc_percpu(struct sched_group *); | |
6750 | if (!sdd->sg) | |
6751 | return -ENOMEM; | |
6752 | ||
63b2ca30 NP |
6753 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6754 | if (!sdd->sgc) | |
9c3f75cb PZ |
6755 | return -ENOMEM; |
6756 | ||
54ab4ff4 PZ |
6757 | for_each_cpu(j, cpu_map) { |
6758 | struct sched_domain *sd; | |
6759 | struct sched_group *sg; | |
63b2ca30 | 6760 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6761 | |
5cc389bc | 6762 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6763 | GFP_KERNEL, cpu_to_node(j)); |
6764 | if (!sd) | |
6765 | return -ENOMEM; | |
6766 | ||
6767 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6768 | ||
6769 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6770 | GFP_KERNEL, cpu_to_node(j)); | |
6771 | if (!sg) | |
6772 | return -ENOMEM; | |
6773 | ||
30b4e9eb IM |
6774 | sg->next = sg; |
6775 | ||
54ab4ff4 | 6776 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6777 | |
63b2ca30 | 6778 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6779 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6780 | if (!sgc) |
9c3f75cb PZ |
6781 | return -ENOMEM; |
6782 | ||
63b2ca30 | 6783 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6784 | } |
6785 | } | |
6786 | ||
6787 | return 0; | |
6788 | } | |
6789 | ||
6790 | static void __sdt_free(const struct cpumask *cpu_map) | |
6791 | { | |
6792 | struct sched_domain_topology_level *tl; | |
6793 | int j; | |
6794 | ||
27723a68 | 6795 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6796 | struct sd_data *sdd = &tl->data; |
6797 | ||
6798 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6799 | struct sched_domain *sd; |
6800 | ||
6801 | if (sdd->sd) { | |
6802 | sd = *per_cpu_ptr(sdd->sd, j); | |
6803 | if (sd && (sd->flags & SD_OVERLAP)) | |
6804 | free_sched_groups(sd->groups, 0); | |
6805 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6806 | } | |
6807 | ||
6808 | if (sdd->sg) | |
6809 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6810 | if (sdd->sgc) |
6811 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6812 | } |
6813 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6814 | sdd->sd = NULL; |
54ab4ff4 | 6815 | free_percpu(sdd->sg); |
fb2cf2c6 | 6816 | sdd->sg = NULL; |
63b2ca30 NP |
6817 | free_percpu(sdd->sgc); |
6818 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6819 | } |
6820 | } | |
6821 | ||
2c402dc3 | 6822 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6823 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6824 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6825 | { |
143e1e28 | 6826 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6827 | if (!sd) |
d069b916 | 6828 | return child; |
2c402dc3 | 6829 | |
2c402dc3 | 6830 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6831 | if (child) { |
6832 | sd->level = child->level + 1; | |
6833 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6834 | child->parent = sd; |
c75e0128 | 6835 | sd->child = child; |
6ae72dff PZ |
6836 | |
6837 | if (!cpumask_subset(sched_domain_span(child), | |
6838 | sched_domain_span(sd))) { | |
6839 | pr_err("BUG: arch topology borken\n"); | |
6840 | #ifdef CONFIG_SCHED_DEBUG | |
6841 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6842 | child->name, sd->name); | |
6843 | #endif | |
6844 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6845 | cpumask_or(sched_domain_span(sd), | |
6846 | sched_domain_span(sd), | |
6847 | sched_domain_span(child)); | |
6848 | } | |
6849 | ||
60495e77 | 6850 | } |
a841f8ce | 6851 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6852 | |
6853 | return sd; | |
6854 | } | |
6855 | ||
2109b99e AH |
6856 | /* |
6857 | * Build sched domains for a given set of cpus and attach the sched domains | |
6858 | * to the individual cpus | |
6859 | */ | |
dce840a0 PZ |
6860 | static int build_sched_domains(const struct cpumask *cpu_map, |
6861 | struct sched_domain_attr *attr) | |
2109b99e | 6862 | { |
1c632169 | 6863 | enum s_alloc alloc_state; |
dce840a0 | 6864 | struct sched_domain *sd; |
2109b99e | 6865 | struct s_data d; |
822ff793 | 6866 | int i, ret = -ENOMEM; |
9c1cfda2 | 6867 | |
2109b99e AH |
6868 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6869 | if (alloc_state != sa_rootdomain) | |
6870 | goto error; | |
9c1cfda2 | 6871 | |
dce840a0 | 6872 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6873 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6874 | struct sched_domain_topology_level *tl; |
6875 | ||
3bd65a80 | 6876 | sd = NULL; |
27723a68 | 6877 | for_each_sd_topology(tl) { |
4a850cbe | 6878 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6879 | if (tl == sched_domain_topology) |
6880 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6881 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6882 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6883 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6884 | break; | |
e3589f6c | 6885 | } |
dce840a0 PZ |
6886 | } |
6887 | ||
6888 | /* Build the groups for the domains */ | |
6889 | for_each_cpu(i, cpu_map) { | |
6890 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6891 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6892 | if (sd->flags & SD_OVERLAP) { |
6893 | if (build_overlap_sched_groups(sd, i)) | |
6894 | goto error; | |
6895 | } else { | |
6896 | if (build_sched_groups(sd, i)) | |
6897 | goto error; | |
6898 | } | |
1cf51902 | 6899 | } |
a06dadbe | 6900 | } |
9c1cfda2 | 6901 | |
ced549fa | 6902 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6903 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6904 | if (!cpumask_test_cpu(i, cpu_map)) | |
6905 | continue; | |
9c1cfda2 | 6906 | |
dce840a0 PZ |
6907 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6908 | claim_allocations(i, sd); | |
63b2ca30 | 6909 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6910 | } |
f712c0c7 | 6911 | } |
9c1cfda2 | 6912 | |
1da177e4 | 6913 | /* Attach the domains */ |
dce840a0 | 6914 | rcu_read_lock(); |
abcd083a | 6915 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6916 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6917 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6918 | } |
dce840a0 | 6919 | rcu_read_unlock(); |
51888ca2 | 6920 | |
822ff793 | 6921 | ret = 0; |
51888ca2 | 6922 | error: |
2109b99e | 6923 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6924 | return ret; |
1da177e4 | 6925 | } |
029190c5 | 6926 | |
acc3f5d7 | 6927 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6928 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6929 | static struct sched_domain_attr *dattr_cur; |
6930 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6931 | |
6932 | /* | |
6933 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6934 | * cpumask) fails, then fallback to a single sched domain, |
6935 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6936 | */ |
4212823f | 6937 | static cpumask_var_t fallback_doms; |
029190c5 | 6938 | |
ee79d1bd HC |
6939 | /* |
6940 | * arch_update_cpu_topology lets virtualized architectures update the | |
6941 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6942 | * or 0 if it stayed the same. | |
6943 | */ | |
52f5684c | 6944 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6945 | { |
ee79d1bd | 6946 | return 0; |
22e52b07 HC |
6947 | } |
6948 | ||
acc3f5d7 RR |
6949 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6950 | { | |
6951 | int i; | |
6952 | cpumask_var_t *doms; | |
6953 | ||
6954 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6955 | if (!doms) | |
6956 | return NULL; | |
6957 | for (i = 0; i < ndoms; i++) { | |
6958 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6959 | free_sched_domains(doms, i); | |
6960 | return NULL; | |
6961 | } | |
6962 | } | |
6963 | return doms; | |
6964 | } | |
6965 | ||
6966 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6967 | { | |
6968 | unsigned int i; | |
6969 | for (i = 0; i < ndoms; i++) | |
6970 | free_cpumask_var(doms[i]); | |
6971 | kfree(doms); | |
6972 | } | |
6973 | ||
1a20ff27 | 6974 | /* |
41a2d6cf | 6975 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6976 | * For now this just excludes isolated cpus, but could be used to |
6977 | * exclude other special cases in the future. | |
1a20ff27 | 6978 | */ |
c4a8849a | 6979 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6980 | { |
7378547f MM |
6981 | int err; |
6982 | ||
22e52b07 | 6983 | arch_update_cpu_topology(); |
029190c5 | 6984 | ndoms_cur = 1; |
acc3f5d7 | 6985 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6986 | if (!doms_cur) |
acc3f5d7 RR |
6987 | doms_cur = &fallback_doms; |
6988 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6989 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6990 | register_sched_domain_sysctl(); |
7378547f MM |
6991 | |
6992 | return err; | |
1a20ff27 DG |
6993 | } |
6994 | ||
1a20ff27 DG |
6995 | /* |
6996 | * Detach sched domains from a group of cpus specified in cpu_map | |
6997 | * These cpus will now be attached to the NULL domain | |
6998 | */ | |
96f874e2 | 6999 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7000 | { |
7001 | int i; | |
7002 | ||
dce840a0 | 7003 | rcu_read_lock(); |
abcd083a | 7004 | for_each_cpu(i, cpu_map) |
57d885fe | 7005 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7006 | rcu_read_unlock(); |
1a20ff27 DG |
7007 | } |
7008 | ||
1d3504fc HS |
7009 | /* handle null as "default" */ |
7010 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7011 | struct sched_domain_attr *new, int idx_new) | |
7012 | { | |
7013 | struct sched_domain_attr tmp; | |
7014 | ||
7015 | /* fast path */ | |
7016 | if (!new && !cur) | |
7017 | return 1; | |
7018 | ||
7019 | tmp = SD_ATTR_INIT; | |
7020 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7021 | new ? (new + idx_new) : &tmp, | |
7022 | sizeof(struct sched_domain_attr)); | |
7023 | } | |
7024 | ||
029190c5 PJ |
7025 | /* |
7026 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7027 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7028 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7029 | * It destroys each deleted domain and builds each new domain. | |
7030 | * | |
acc3f5d7 | 7031 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7032 | * The masks don't intersect (don't overlap.) We should setup one |
7033 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7034 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7035 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7036 | * it as it is. | |
7037 | * | |
acc3f5d7 RR |
7038 | * The passed in 'doms_new' should be allocated using |
7039 | * alloc_sched_domains. This routine takes ownership of it and will | |
7040 | * free_sched_domains it when done with it. If the caller failed the | |
7041 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7042 | * and partition_sched_domains() will fallback to the single partition | |
7043 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7044 | * |
96f874e2 | 7045 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7046 | * ndoms_new == 0 is a special case for destroying existing domains, |
7047 | * and it will not create the default domain. | |
dfb512ec | 7048 | * |
029190c5 PJ |
7049 | * Call with hotplug lock held |
7050 | */ | |
acc3f5d7 | 7051 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7052 | struct sched_domain_attr *dattr_new) |
029190c5 | 7053 | { |
dfb512ec | 7054 | int i, j, n; |
d65bd5ec | 7055 | int new_topology; |
029190c5 | 7056 | |
712555ee | 7057 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7058 | |
7378547f MM |
7059 | /* always unregister in case we don't destroy any domains */ |
7060 | unregister_sched_domain_sysctl(); | |
7061 | ||
d65bd5ec HC |
7062 | /* Let architecture update cpu core mappings. */ |
7063 | new_topology = arch_update_cpu_topology(); | |
7064 | ||
dfb512ec | 7065 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7066 | |
7067 | /* Destroy deleted domains */ | |
7068 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7069 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7070 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7071 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7072 | goto match1; |
7073 | } | |
7074 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7075 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7076 | match1: |
7077 | ; | |
7078 | } | |
7079 | ||
c8d2d47a | 7080 | n = ndoms_cur; |
e761b772 | 7081 | if (doms_new == NULL) { |
c8d2d47a | 7082 | n = 0; |
acc3f5d7 | 7083 | doms_new = &fallback_doms; |
6ad4c188 | 7084 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7085 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7086 | } |
7087 | ||
029190c5 PJ |
7088 | /* Build new domains */ |
7089 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7090 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7091 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7092 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7093 | goto match2; |
7094 | } | |
7095 | /* no match - add a new doms_new */ | |
dce840a0 | 7096 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7097 | match2: |
7098 | ; | |
7099 | } | |
7100 | ||
7101 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7102 | if (doms_cur != &fallback_doms) |
7103 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7104 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7105 | doms_cur = doms_new; |
1d3504fc | 7106 | dattr_cur = dattr_new; |
029190c5 | 7107 | ndoms_cur = ndoms_new; |
7378547f MM |
7108 | |
7109 | register_sched_domain_sysctl(); | |
a1835615 | 7110 | |
712555ee | 7111 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7112 | } |
7113 | ||
d35be8ba SB |
7114 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7115 | ||
1da177e4 | 7116 | /* |
3a101d05 TH |
7117 | * Update cpusets according to cpu_active mask. If cpusets are |
7118 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7119 | * around partition_sched_domains(). | |
d35be8ba SB |
7120 | * |
7121 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7122 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7123 | */ |
0b2e918a TH |
7124 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7125 | void *hcpu) | |
e761b772 | 7126 | { |
d35be8ba SB |
7127 | switch (action) { |
7128 | case CPU_ONLINE_FROZEN: | |
7129 | case CPU_DOWN_FAILED_FROZEN: | |
7130 | ||
7131 | /* | |
7132 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7133 | * resume sequence. As long as this is not the last online | |
7134 | * operation in the resume sequence, just build a single sched | |
7135 | * domain, ignoring cpusets. | |
7136 | */ | |
7137 | num_cpus_frozen--; | |
7138 | if (likely(num_cpus_frozen)) { | |
7139 | partition_sched_domains(1, NULL, NULL); | |
7140 | break; | |
7141 | } | |
7142 | ||
7143 | /* | |
7144 | * This is the last CPU online operation. So fall through and | |
7145 | * restore the original sched domains by considering the | |
7146 | * cpuset configurations. | |
7147 | */ | |
7148 | ||
e761b772 | 7149 | case CPU_ONLINE: |
7ddf96b0 | 7150 | cpuset_update_active_cpus(true); |
d35be8ba | 7151 | break; |
3a101d05 TH |
7152 | default: |
7153 | return NOTIFY_DONE; | |
7154 | } | |
d35be8ba | 7155 | return NOTIFY_OK; |
3a101d05 | 7156 | } |
e761b772 | 7157 | |
0b2e918a TH |
7158 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7159 | void *hcpu) | |
3a101d05 | 7160 | { |
3c18d447 JL |
7161 | unsigned long flags; |
7162 | long cpu = (long)hcpu; | |
7163 | struct dl_bw *dl_b; | |
533445c6 OS |
7164 | bool overflow; |
7165 | int cpus; | |
3c18d447 | 7166 | |
533445c6 | 7167 | switch (action) { |
3a101d05 | 7168 | case CPU_DOWN_PREPARE: |
533445c6 OS |
7169 | rcu_read_lock_sched(); |
7170 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7171 | |
533445c6 OS |
7172 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7173 | cpus = dl_bw_cpus(cpu); | |
7174 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7175 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7176 | |
533445c6 | 7177 | rcu_read_unlock_sched(); |
3c18d447 | 7178 | |
533445c6 OS |
7179 | if (overflow) |
7180 | return notifier_from_errno(-EBUSY); | |
7ddf96b0 | 7181 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7182 | break; |
7183 | case CPU_DOWN_PREPARE_FROZEN: | |
7184 | num_cpus_frozen++; | |
7185 | partition_sched_domains(1, NULL, NULL); | |
7186 | break; | |
e761b772 MK |
7187 | default: |
7188 | return NOTIFY_DONE; | |
7189 | } | |
d35be8ba | 7190 | return NOTIFY_OK; |
e761b772 | 7191 | } |
e761b772 | 7192 | |
1da177e4 LT |
7193 | void __init sched_init_smp(void) |
7194 | { | |
dcc30a35 RR |
7195 | cpumask_var_t non_isolated_cpus; |
7196 | ||
7197 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7198 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7199 | |
8cb9764f CM |
7200 | /* nohz_full won't take effect without isolating the cpus. */ |
7201 | tick_nohz_full_add_cpus_to(cpu_isolated_map); | |
7202 | ||
cb83b629 PZ |
7203 | sched_init_numa(); |
7204 | ||
6acce3ef PZ |
7205 | /* |
7206 | * There's no userspace yet to cause hotplug operations; hence all the | |
7207 | * cpu masks are stable and all blatant races in the below code cannot | |
7208 | * happen. | |
7209 | */ | |
712555ee | 7210 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7211 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7212 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7213 | if (cpumask_empty(non_isolated_cpus)) | |
7214 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7215 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7216 | |
301a5cba | 7217 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7218 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7219 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7220 | |
b328ca18 | 7221 | init_hrtick(); |
5c1e1767 NP |
7222 | |
7223 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7224 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7225 | BUG(); |
19978ca6 | 7226 | sched_init_granularity(); |
dcc30a35 | 7227 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7228 | |
0e3900e6 | 7229 | init_sched_rt_class(); |
1baca4ce | 7230 | init_sched_dl_class(); |
1da177e4 LT |
7231 | } |
7232 | #else | |
7233 | void __init sched_init_smp(void) | |
7234 | { | |
19978ca6 | 7235 | sched_init_granularity(); |
1da177e4 LT |
7236 | } |
7237 | #endif /* CONFIG_SMP */ | |
7238 | ||
7239 | int in_sched_functions(unsigned long addr) | |
7240 | { | |
1da177e4 LT |
7241 | return in_lock_functions(addr) || |
7242 | (addr >= (unsigned long)__sched_text_start | |
7243 | && addr < (unsigned long)__sched_text_end); | |
7244 | } | |
7245 | ||
029632fb | 7246 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7247 | /* |
7248 | * Default task group. | |
7249 | * Every task in system belongs to this group at bootup. | |
7250 | */ | |
029632fb | 7251 | struct task_group root_task_group; |
35cf4e50 | 7252 | LIST_HEAD(task_groups); |
052f1dc7 | 7253 | #endif |
6f505b16 | 7254 | |
e6252c3e | 7255 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7256 | |
1da177e4 LT |
7257 | void __init sched_init(void) |
7258 | { | |
dd41f596 | 7259 | int i, j; |
434d53b0 MT |
7260 | unsigned long alloc_size = 0, ptr; |
7261 | ||
7262 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7263 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7264 | #endif | |
7265 | #ifdef CONFIG_RT_GROUP_SCHED | |
7266 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7267 | #endif | |
434d53b0 | 7268 | if (alloc_size) { |
36b7b6d4 | 7269 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7270 | |
7271 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7272 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7273 | ptr += nr_cpu_ids * sizeof(void **); |
7274 | ||
07e06b01 | 7275 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7276 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7277 | |
6d6bc0ad | 7278 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7279 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7280 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7281 | ptr += nr_cpu_ids * sizeof(void **); |
7282 | ||
07e06b01 | 7283 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7284 | ptr += nr_cpu_ids * sizeof(void **); |
7285 | ||
6d6bc0ad | 7286 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7287 | } |
df7c8e84 | 7288 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7289 | for_each_possible_cpu(i) { |
7290 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7291 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7292 | } |
b74e6278 | 7293 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7294 | |
332ac17e DF |
7295 | init_rt_bandwidth(&def_rt_bandwidth, |
7296 | global_rt_period(), global_rt_runtime()); | |
7297 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7298 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7299 | |
57d885fe GH |
7300 | #ifdef CONFIG_SMP |
7301 | init_defrootdomain(); | |
7302 | #endif | |
7303 | ||
d0b27fa7 | 7304 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7305 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7306 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7307 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7308 | |
7c941438 | 7309 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7310 | list_add(&root_task_group.list, &task_groups); |
7311 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7312 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7313 | autogroup_init(&init_task); |
54c707e9 | 7314 | |
7c941438 | 7315 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7316 | |
0a945022 | 7317 | for_each_possible_cpu(i) { |
70b97a7f | 7318 | struct rq *rq; |
1da177e4 LT |
7319 | |
7320 | rq = cpu_rq(i); | |
05fa785c | 7321 | raw_spin_lock_init(&rq->lock); |
7897986b | 7322 | rq->nr_running = 0; |
dce48a84 TG |
7323 | rq->calc_load_active = 0; |
7324 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7325 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7326 | init_rt_rq(&rq->rt); |
7327 | init_dl_rq(&rq->dl); | |
dd41f596 | 7328 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7329 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7330 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7331 | /* |
07e06b01 | 7332 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7333 | * |
7334 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7335 | * gets 100% of the cpu resources in the system. This overall | |
7336 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7337 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7338 | * based on each entity's (task or task-group's) weight |
7339 | * (se->load.weight). | |
7340 | * | |
07e06b01 | 7341 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7342 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7343 | * then A0's share of the cpu resource is: | |
7344 | * | |
0d905bca | 7345 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7346 | * |
07e06b01 YZ |
7347 | * We achieve this by letting root_task_group's tasks sit |
7348 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7349 | */ |
ab84d31e | 7350 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7351 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7352 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7353 | ||
7354 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7355 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7356 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7357 | #endif |
1da177e4 | 7358 | |
dd41f596 IM |
7359 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7360 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7361 | |
7362 | rq->last_load_update_tick = jiffies; | |
7363 | ||
1da177e4 | 7364 | #ifdef CONFIG_SMP |
41c7ce9a | 7365 | rq->sd = NULL; |
57d885fe | 7366 | rq->rd = NULL; |
ca6d75e6 | 7367 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7368 | rq->balance_callback = NULL; |
1da177e4 | 7369 | rq->active_balance = 0; |
dd41f596 | 7370 | rq->next_balance = jiffies; |
1da177e4 | 7371 | rq->push_cpu = 0; |
0a2966b4 | 7372 | rq->cpu = i; |
1f11eb6a | 7373 | rq->online = 0; |
eae0c9df MG |
7374 | rq->idle_stamp = 0; |
7375 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7376 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7377 | |
7378 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7379 | ||
dc938520 | 7380 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7381 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7382 | rq->nohz_flags = 0; |
83cd4fe2 | 7383 | #endif |
265f22a9 FW |
7384 | #ifdef CONFIG_NO_HZ_FULL |
7385 | rq->last_sched_tick = 0; | |
7386 | #endif | |
1da177e4 | 7387 | #endif |
8f4d37ec | 7388 | init_rq_hrtick(rq); |
1da177e4 | 7389 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7390 | } |
7391 | ||
2dd73a4f | 7392 | set_load_weight(&init_task); |
b50f60ce | 7393 | |
e107be36 AK |
7394 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7395 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7396 | #endif | |
7397 | ||
1da177e4 LT |
7398 | /* |
7399 | * The boot idle thread does lazy MMU switching as well: | |
7400 | */ | |
7401 | atomic_inc(&init_mm.mm_count); | |
7402 | enter_lazy_tlb(&init_mm, current); | |
7403 | ||
1b537c7d YD |
7404 | /* |
7405 | * During early bootup we pretend to be a normal task: | |
7406 | */ | |
7407 | current->sched_class = &fair_sched_class; | |
7408 | ||
1da177e4 LT |
7409 | /* |
7410 | * Make us the idle thread. Technically, schedule() should not be | |
7411 | * called from this thread, however somewhere below it might be, | |
7412 | * but because we are the idle thread, we just pick up running again | |
7413 | * when this runqueue becomes "idle". | |
7414 | */ | |
7415 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7416 | |
7417 | calc_load_update = jiffies + LOAD_FREQ; | |
7418 | ||
bf4d83f6 | 7419 | #ifdef CONFIG_SMP |
4cb98839 | 7420 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7421 | /* May be allocated at isolcpus cmdline parse time */ |
7422 | if (cpu_isolated_map == NULL) | |
7423 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7424 | idle_thread_set_boot_cpu(); |
a803f026 | 7425 | set_cpu_rq_start_time(); |
029632fb PZ |
7426 | #endif |
7427 | init_sched_fair_class(); | |
6a7b3dc3 | 7428 | |
6892b75e | 7429 | scheduler_running = 1; |
1da177e4 LT |
7430 | } |
7431 | ||
d902db1e | 7432 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7433 | static inline int preempt_count_equals(int preempt_offset) |
7434 | { | |
234da7bc | 7435 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7436 | |
4ba8216c | 7437 | return (nested == preempt_offset); |
e4aafea2 FW |
7438 | } |
7439 | ||
d894837f | 7440 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7441 | { |
8eb23b9f PZ |
7442 | /* |
7443 | * Blocking primitives will set (and therefore destroy) current->state, | |
7444 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7445 | * otherwise we will destroy state. | |
7446 | */ | |
00845eb9 | 7447 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7448 | "do not call blocking ops when !TASK_RUNNING; " |
7449 | "state=%lx set at [<%p>] %pS\n", | |
7450 | current->state, | |
7451 | (void *)current->task_state_change, | |
00845eb9 | 7452 | (void *)current->task_state_change); |
8eb23b9f | 7453 | |
3427445a PZ |
7454 | ___might_sleep(file, line, preempt_offset); |
7455 | } | |
7456 | EXPORT_SYMBOL(__might_sleep); | |
7457 | ||
7458 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7459 | { |
1da177e4 LT |
7460 | static unsigned long prev_jiffy; /* ratelimiting */ |
7461 | ||
b3fbab05 | 7462 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7463 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7464 | !is_idle_task(current)) || | |
e4aafea2 | 7465 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7466 | return; |
7467 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7468 | return; | |
7469 | prev_jiffy = jiffies; | |
7470 | ||
3df0fc5b PZ |
7471 | printk(KERN_ERR |
7472 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7473 | file, line); | |
7474 | printk(KERN_ERR | |
7475 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7476 | in_atomic(), irqs_disabled(), | |
7477 | current->pid, current->comm); | |
aef745fc | 7478 | |
a8b686b3 ES |
7479 | if (task_stack_end_corrupted(current)) |
7480 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7481 | ||
aef745fc IM |
7482 | debug_show_held_locks(current); |
7483 | if (irqs_disabled()) | |
7484 | print_irqtrace_events(current); | |
8f47b187 TG |
7485 | #ifdef CONFIG_DEBUG_PREEMPT |
7486 | if (!preempt_count_equals(preempt_offset)) { | |
7487 | pr_err("Preemption disabled at:"); | |
7488 | print_ip_sym(current->preempt_disable_ip); | |
7489 | pr_cont("\n"); | |
7490 | } | |
7491 | #endif | |
aef745fc | 7492 | dump_stack(); |
1da177e4 | 7493 | } |
3427445a | 7494 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7495 | #endif |
7496 | ||
7497 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7498 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7499 | { |
dbc7f069 | 7500 | struct task_struct *g, *p; |
d50dde5a DF |
7501 | struct sched_attr attr = { |
7502 | .sched_policy = SCHED_NORMAL, | |
7503 | }; | |
1da177e4 | 7504 | |
3472eaa1 | 7505 | read_lock(&tasklist_lock); |
5d07f420 | 7506 | for_each_process_thread(g, p) { |
178be793 IM |
7507 | /* |
7508 | * Only normalize user tasks: | |
7509 | */ | |
3472eaa1 | 7510 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7511 | continue; |
7512 | ||
6cfb0d5d | 7513 | p->se.exec_start = 0; |
6cfb0d5d | 7514 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7515 | p->se.statistics.wait_start = 0; |
7516 | p->se.statistics.sleep_start = 0; | |
7517 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7518 | #endif |
dd41f596 | 7519 | |
aab03e05 | 7520 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7521 | /* |
7522 | * Renice negative nice level userspace | |
7523 | * tasks back to 0: | |
7524 | */ | |
3472eaa1 | 7525 | if (task_nice(p) < 0) |
dd41f596 | 7526 | set_user_nice(p, 0); |
1da177e4 | 7527 | continue; |
dd41f596 | 7528 | } |
1da177e4 | 7529 | |
dbc7f069 | 7530 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7531 | } |
3472eaa1 | 7532 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7533 | } |
7534 | ||
7535 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7536 | |
67fc4e0c | 7537 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7538 | /* |
67fc4e0c | 7539 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7540 | * |
7541 | * They can only be called when the whole system has been | |
7542 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7543 | * activity can take place. Using them for anything else would | |
7544 | * be a serious bug, and as a result, they aren't even visible | |
7545 | * under any other configuration. | |
7546 | */ | |
7547 | ||
7548 | /** | |
7549 | * curr_task - return the current task for a given cpu. | |
7550 | * @cpu: the processor in question. | |
7551 | * | |
7552 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7553 | * |
7554 | * Return: The current task for @cpu. | |
1df5c10a | 7555 | */ |
36c8b586 | 7556 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7557 | { |
7558 | return cpu_curr(cpu); | |
7559 | } | |
7560 | ||
67fc4e0c JW |
7561 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7562 | ||
7563 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7564 | /** |
7565 | * set_curr_task - set the current task for a given cpu. | |
7566 | * @cpu: the processor in question. | |
7567 | * @p: the task pointer to set. | |
7568 | * | |
7569 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7570 | * are serviced on a separate stack. It allows the architecture to switch the |
7571 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7572 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7573 | * and caller must save the original value of the current task (see | |
7574 | * curr_task() above) and restore that value before reenabling interrupts and | |
7575 | * re-starting the system. | |
7576 | * | |
7577 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7578 | */ | |
36c8b586 | 7579 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7580 | { |
7581 | cpu_curr(cpu) = p; | |
7582 | } | |
7583 | ||
7584 | #endif | |
29f59db3 | 7585 | |
7c941438 | 7586 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7587 | /* task_group_lock serializes the addition/removal of task groups */ |
7588 | static DEFINE_SPINLOCK(task_group_lock); | |
7589 | ||
bccbe08a PZ |
7590 | static void free_sched_group(struct task_group *tg) |
7591 | { | |
7592 | free_fair_sched_group(tg); | |
7593 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7594 | autogroup_free(tg); |
bccbe08a PZ |
7595 | kfree(tg); |
7596 | } | |
7597 | ||
7598 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7599 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7600 | { |
7601 | struct task_group *tg; | |
bccbe08a PZ |
7602 | |
7603 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7604 | if (!tg) | |
7605 | return ERR_PTR(-ENOMEM); | |
7606 | ||
ec7dc8ac | 7607 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7608 | goto err; |
7609 | ||
ec7dc8ac | 7610 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7611 | goto err; |
7612 | ||
ace783b9 LZ |
7613 | return tg; |
7614 | ||
7615 | err: | |
7616 | free_sched_group(tg); | |
7617 | return ERR_PTR(-ENOMEM); | |
7618 | } | |
7619 | ||
7620 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7621 | { | |
7622 | unsigned long flags; | |
7623 | ||
8ed36996 | 7624 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7625 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7626 | |
7627 | WARN_ON(!parent); /* root should already exist */ | |
7628 | ||
7629 | tg->parent = parent; | |
f473aa5e | 7630 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7631 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7632 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7633 | } |
7634 | ||
9b5b7751 | 7635 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7636 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7637 | { |
29f59db3 | 7638 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7639 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7640 | } |
7641 | ||
9b5b7751 | 7642 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7643 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7644 | { |
7645 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7646 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7647 | } | |
7648 | ||
7649 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7650 | { |
8ed36996 | 7651 | unsigned long flags; |
9b5b7751 | 7652 | int i; |
29f59db3 | 7653 | |
3d4b47b4 PZ |
7654 | /* end participation in shares distribution */ |
7655 | for_each_possible_cpu(i) | |
bccbe08a | 7656 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7657 | |
7658 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7659 | list_del_rcu(&tg->list); |
f473aa5e | 7660 | list_del_rcu(&tg->siblings); |
8ed36996 | 7661 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7662 | } |
7663 | ||
9b5b7751 | 7664 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7665 | * The caller of this function should have put the task in its new group |
7666 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7667 | * reflect its new group. | |
9b5b7751 SV |
7668 | */ |
7669 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7670 | { |
8323f26c | 7671 | struct task_group *tg; |
da0c1e65 | 7672 | int queued, running; |
29f59db3 SV |
7673 | unsigned long flags; |
7674 | struct rq *rq; | |
7675 | ||
7676 | rq = task_rq_lock(tsk, &flags); | |
7677 | ||
051a1d1a | 7678 | running = task_current(rq, tsk); |
da0c1e65 | 7679 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7680 | |
da0c1e65 | 7681 | if (queued) |
29f59db3 | 7682 | dequeue_task(rq, tsk, 0); |
0e1f3483 | 7683 | if (unlikely(running)) |
f3cd1c4e | 7684 | put_prev_task(rq, tsk); |
29f59db3 | 7685 | |
f7b8a47d KT |
7686 | /* |
7687 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7688 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7689 | * to prevent lockdep warnings. | |
7690 | */ | |
7691 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7692 | struct task_group, css); |
7693 | tg = autogroup_task_group(tsk, tg); | |
7694 | tsk->sched_task_group = tg; | |
7695 | ||
810b3817 | 7696 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7697 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7698 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7699 | else |
810b3817 | 7700 | #endif |
b2b5ce02 | 7701 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7702 | |
0e1f3483 HS |
7703 | if (unlikely(running)) |
7704 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7705 | if (queued) |
371fd7e7 | 7706 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7707 | |
0122ec5b | 7708 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7709 | } |
7c941438 | 7710 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7711 | |
a790de99 PT |
7712 | #ifdef CONFIG_RT_GROUP_SCHED |
7713 | /* | |
7714 | * Ensure that the real time constraints are schedulable. | |
7715 | */ | |
7716 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7717 | |
9a7e0b18 PZ |
7718 | /* Must be called with tasklist_lock held */ |
7719 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7720 | { |
9a7e0b18 | 7721 | struct task_struct *g, *p; |
b40b2e8e | 7722 | |
1fe89e1b PZ |
7723 | /* |
7724 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7725 | */ | |
7726 | if (task_group_is_autogroup(tg)) | |
7727 | return 0; | |
7728 | ||
5d07f420 | 7729 | for_each_process_thread(g, p) { |
8651c658 | 7730 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7731 | return 1; |
5d07f420 | 7732 | } |
b40b2e8e | 7733 | |
9a7e0b18 PZ |
7734 | return 0; |
7735 | } | |
b40b2e8e | 7736 | |
9a7e0b18 PZ |
7737 | struct rt_schedulable_data { |
7738 | struct task_group *tg; | |
7739 | u64 rt_period; | |
7740 | u64 rt_runtime; | |
7741 | }; | |
b40b2e8e | 7742 | |
a790de99 | 7743 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7744 | { |
7745 | struct rt_schedulable_data *d = data; | |
7746 | struct task_group *child; | |
7747 | unsigned long total, sum = 0; | |
7748 | u64 period, runtime; | |
b40b2e8e | 7749 | |
9a7e0b18 PZ |
7750 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7751 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7752 | |
9a7e0b18 PZ |
7753 | if (tg == d->tg) { |
7754 | period = d->rt_period; | |
7755 | runtime = d->rt_runtime; | |
b40b2e8e | 7756 | } |
b40b2e8e | 7757 | |
4653f803 PZ |
7758 | /* |
7759 | * Cannot have more runtime than the period. | |
7760 | */ | |
7761 | if (runtime > period && runtime != RUNTIME_INF) | |
7762 | return -EINVAL; | |
6f505b16 | 7763 | |
4653f803 PZ |
7764 | /* |
7765 | * Ensure we don't starve existing RT tasks. | |
7766 | */ | |
9a7e0b18 PZ |
7767 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7768 | return -EBUSY; | |
6f505b16 | 7769 | |
9a7e0b18 | 7770 | total = to_ratio(period, runtime); |
6f505b16 | 7771 | |
4653f803 PZ |
7772 | /* |
7773 | * Nobody can have more than the global setting allows. | |
7774 | */ | |
7775 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7776 | return -EINVAL; | |
6f505b16 | 7777 | |
4653f803 PZ |
7778 | /* |
7779 | * The sum of our children's runtime should not exceed our own. | |
7780 | */ | |
9a7e0b18 PZ |
7781 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7782 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7783 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7784 | |
9a7e0b18 PZ |
7785 | if (child == d->tg) { |
7786 | period = d->rt_period; | |
7787 | runtime = d->rt_runtime; | |
7788 | } | |
6f505b16 | 7789 | |
9a7e0b18 | 7790 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7791 | } |
6f505b16 | 7792 | |
9a7e0b18 PZ |
7793 | if (sum > total) |
7794 | return -EINVAL; | |
7795 | ||
7796 | return 0; | |
6f505b16 PZ |
7797 | } |
7798 | ||
9a7e0b18 | 7799 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7800 | { |
8277434e PT |
7801 | int ret; |
7802 | ||
9a7e0b18 PZ |
7803 | struct rt_schedulable_data data = { |
7804 | .tg = tg, | |
7805 | .rt_period = period, | |
7806 | .rt_runtime = runtime, | |
7807 | }; | |
7808 | ||
8277434e PT |
7809 | rcu_read_lock(); |
7810 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7811 | rcu_read_unlock(); | |
7812 | ||
7813 | return ret; | |
521f1a24 DG |
7814 | } |
7815 | ||
ab84d31e | 7816 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7817 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7818 | { |
ac086bc2 | 7819 | int i, err = 0; |
9f0c1e56 | 7820 | |
2636ed5f PZ |
7821 | /* |
7822 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7823 | * kernel creating (and or operating) RT threads. | |
7824 | */ | |
7825 | if (tg == &root_task_group && rt_runtime == 0) | |
7826 | return -EINVAL; | |
7827 | ||
7828 | /* No period doesn't make any sense. */ | |
7829 | if (rt_period == 0) | |
7830 | return -EINVAL; | |
7831 | ||
9f0c1e56 | 7832 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7833 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7834 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7835 | if (err) | |
9f0c1e56 | 7836 | goto unlock; |
ac086bc2 | 7837 | |
0986b11b | 7838 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7839 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7840 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7841 | |
7842 | for_each_possible_cpu(i) { | |
7843 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7844 | ||
0986b11b | 7845 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7846 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7847 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7848 | } |
0986b11b | 7849 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7850 | unlock: |
521f1a24 | 7851 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7852 | mutex_unlock(&rt_constraints_mutex); |
7853 | ||
7854 | return err; | |
6f505b16 PZ |
7855 | } |
7856 | ||
25cc7da7 | 7857 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7858 | { |
7859 | u64 rt_runtime, rt_period; | |
7860 | ||
7861 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7862 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7863 | if (rt_runtime_us < 0) | |
7864 | rt_runtime = RUNTIME_INF; | |
7865 | ||
ab84d31e | 7866 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7867 | } |
7868 | ||
25cc7da7 | 7869 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7870 | { |
7871 | u64 rt_runtime_us; | |
7872 | ||
d0b27fa7 | 7873 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7874 | return -1; |
7875 | ||
d0b27fa7 | 7876 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7877 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7878 | return rt_runtime_us; | |
7879 | } | |
d0b27fa7 | 7880 | |
ce2f5fe4 | 7881 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7882 | { |
7883 | u64 rt_runtime, rt_period; | |
7884 | ||
ce2f5fe4 | 7885 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7886 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7887 | ||
ab84d31e | 7888 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7889 | } |
7890 | ||
25cc7da7 | 7891 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7892 | { |
7893 | u64 rt_period_us; | |
7894 | ||
7895 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7896 | do_div(rt_period_us, NSEC_PER_USEC); | |
7897 | return rt_period_us; | |
7898 | } | |
332ac17e | 7899 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7900 | |
332ac17e | 7901 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7902 | static int sched_rt_global_constraints(void) |
7903 | { | |
7904 | int ret = 0; | |
7905 | ||
7906 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7907 | read_lock(&tasklist_lock); |
4653f803 | 7908 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7909 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7910 | mutex_unlock(&rt_constraints_mutex); |
7911 | ||
7912 | return ret; | |
7913 | } | |
54e99124 | 7914 | |
25cc7da7 | 7915 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7916 | { |
7917 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7918 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7919 | return 0; | |
7920 | ||
7921 | return 1; | |
7922 | } | |
7923 | ||
6d6bc0ad | 7924 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7925 | static int sched_rt_global_constraints(void) |
7926 | { | |
ac086bc2 | 7927 | unsigned long flags; |
332ac17e | 7928 | int i, ret = 0; |
ec5d4989 | 7929 | |
0986b11b | 7930 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7931 | for_each_possible_cpu(i) { |
7932 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7933 | ||
0986b11b | 7934 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7935 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7936 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7937 | } |
0986b11b | 7938 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7939 | |
332ac17e | 7940 | return ret; |
d0b27fa7 | 7941 | } |
6d6bc0ad | 7942 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7943 | |
a1963b81 | 7944 | static int sched_dl_global_validate(void) |
332ac17e | 7945 | { |
1724813d PZ |
7946 | u64 runtime = global_rt_runtime(); |
7947 | u64 period = global_rt_period(); | |
332ac17e | 7948 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7949 | struct dl_bw *dl_b; |
1724813d | 7950 | int cpu, ret = 0; |
49516342 | 7951 | unsigned long flags; |
332ac17e DF |
7952 | |
7953 | /* | |
7954 | * Here we want to check the bandwidth not being set to some | |
7955 | * value smaller than the currently allocated bandwidth in | |
7956 | * any of the root_domains. | |
7957 | * | |
7958 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7959 | * cycling on root_domains... Discussion on different/better | |
7960 | * solutions is welcome! | |
7961 | */ | |
1724813d | 7962 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7963 | rcu_read_lock_sched(); |
7964 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7965 | |
49516342 | 7966 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7967 | if (new_bw < dl_b->total_bw) |
7968 | ret = -EBUSY; | |
49516342 | 7969 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7970 | |
f10e00f4 KT |
7971 | rcu_read_unlock_sched(); |
7972 | ||
1724813d PZ |
7973 | if (ret) |
7974 | break; | |
332ac17e DF |
7975 | } |
7976 | ||
1724813d | 7977 | return ret; |
332ac17e DF |
7978 | } |
7979 | ||
1724813d | 7980 | static void sched_dl_do_global(void) |
ce0dbbbb | 7981 | { |
1724813d | 7982 | u64 new_bw = -1; |
f10e00f4 | 7983 | struct dl_bw *dl_b; |
1724813d | 7984 | int cpu; |
49516342 | 7985 | unsigned long flags; |
ce0dbbbb | 7986 | |
1724813d PZ |
7987 | def_dl_bandwidth.dl_period = global_rt_period(); |
7988 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7989 | ||
7990 | if (global_rt_runtime() != RUNTIME_INF) | |
7991 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7992 | ||
7993 | /* | |
7994 | * FIXME: As above... | |
7995 | */ | |
7996 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7997 | rcu_read_lock_sched(); |
7998 | dl_b = dl_bw_of(cpu); | |
1724813d | 7999 | |
49516342 | 8000 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 8001 | dl_b->bw = new_bw; |
49516342 | 8002 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
8003 | |
8004 | rcu_read_unlock_sched(); | |
ce0dbbbb | 8005 | } |
1724813d PZ |
8006 | } |
8007 | ||
8008 | static int sched_rt_global_validate(void) | |
8009 | { | |
8010 | if (sysctl_sched_rt_period <= 0) | |
8011 | return -EINVAL; | |
8012 | ||
e9e7cb38 JL |
8013 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
8014 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
8015 | return -EINVAL; |
8016 | ||
8017 | return 0; | |
8018 | } | |
8019 | ||
8020 | static void sched_rt_do_global(void) | |
8021 | { | |
8022 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8023 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
8024 | } |
8025 | ||
d0b27fa7 | 8026 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 8027 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8028 | loff_t *ppos) |
8029 | { | |
d0b27fa7 PZ |
8030 | int old_period, old_runtime; |
8031 | static DEFINE_MUTEX(mutex); | |
1724813d | 8032 | int ret; |
d0b27fa7 PZ |
8033 | |
8034 | mutex_lock(&mutex); | |
8035 | old_period = sysctl_sched_rt_period; | |
8036 | old_runtime = sysctl_sched_rt_runtime; | |
8037 | ||
8d65af78 | 8038 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8039 | |
8040 | if (!ret && write) { | |
1724813d PZ |
8041 | ret = sched_rt_global_validate(); |
8042 | if (ret) | |
8043 | goto undo; | |
8044 | ||
a1963b81 | 8045 | ret = sched_dl_global_validate(); |
1724813d PZ |
8046 | if (ret) |
8047 | goto undo; | |
8048 | ||
a1963b81 | 8049 | ret = sched_rt_global_constraints(); |
1724813d PZ |
8050 | if (ret) |
8051 | goto undo; | |
8052 | ||
8053 | sched_rt_do_global(); | |
8054 | sched_dl_do_global(); | |
8055 | } | |
8056 | if (0) { | |
8057 | undo: | |
8058 | sysctl_sched_rt_period = old_period; | |
8059 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
8060 | } |
8061 | mutex_unlock(&mutex); | |
8062 | ||
8063 | return ret; | |
8064 | } | |
68318b8e | 8065 | |
1724813d | 8066 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
8067 | void __user *buffer, size_t *lenp, |
8068 | loff_t *ppos) | |
8069 | { | |
8070 | int ret; | |
332ac17e | 8071 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
8072 | |
8073 | mutex_lock(&mutex); | |
332ac17e | 8074 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
8075 | /* make sure that internally we keep jiffies */ |
8076 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8077 | if (!ret && write) { |
1724813d PZ |
8078 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8079 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8080 | } |
8081 | mutex_unlock(&mutex); | |
332ac17e DF |
8082 | return ret; |
8083 | } | |
8084 | ||
052f1dc7 | 8085 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8086 | |
a7c6d554 | 8087 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8088 | { |
a7c6d554 | 8089 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8090 | } |
8091 | ||
eb95419b TH |
8092 | static struct cgroup_subsys_state * |
8093 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8094 | { |
eb95419b TH |
8095 | struct task_group *parent = css_tg(parent_css); |
8096 | struct task_group *tg; | |
68318b8e | 8097 | |
eb95419b | 8098 | if (!parent) { |
68318b8e | 8099 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8100 | return &root_task_group.css; |
68318b8e SV |
8101 | } |
8102 | ||
ec7dc8ac | 8103 | tg = sched_create_group(parent); |
68318b8e SV |
8104 | if (IS_ERR(tg)) |
8105 | return ERR_PTR(-ENOMEM); | |
8106 | ||
68318b8e SV |
8107 | return &tg->css; |
8108 | } | |
8109 | ||
eb95419b | 8110 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 8111 | { |
eb95419b | 8112 | struct task_group *tg = css_tg(css); |
5c9d535b | 8113 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 8114 | |
63876986 TH |
8115 | if (parent) |
8116 | sched_online_group(tg, parent); | |
ace783b9 LZ |
8117 | return 0; |
8118 | } | |
8119 | ||
eb95419b | 8120 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8121 | { |
eb95419b | 8122 | struct task_group *tg = css_tg(css); |
68318b8e SV |
8123 | |
8124 | sched_destroy_group(tg); | |
8125 | } | |
8126 | ||
eb95419b | 8127 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 8128 | { |
eb95419b | 8129 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
8130 | |
8131 | sched_offline_group(tg); | |
8132 | } | |
8133 | ||
7e47682e | 8134 | static void cpu_cgroup_fork(struct task_struct *task, void *private) |
eeb61e53 KT |
8135 | { |
8136 | sched_move_task(task); | |
8137 | } | |
8138 | ||
eb95419b | 8139 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8140 | struct cgroup_taskset *tset) |
68318b8e | 8141 | { |
bb9d97b6 TH |
8142 | struct task_struct *task; |
8143 | ||
924f0d9a | 8144 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 8145 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8146 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8147 | return -EINVAL; |
b68aa230 | 8148 | #else |
bb9d97b6 TH |
8149 | /* We don't support RT-tasks being in separate groups */ |
8150 | if (task->sched_class != &fair_sched_class) | |
8151 | return -EINVAL; | |
b68aa230 | 8152 | #endif |
bb9d97b6 | 8153 | } |
be367d09 BB |
8154 | return 0; |
8155 | } | |
68318b8e | 8156 | |
eb95419b | 8157 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8158 | struct cgroup_taskset *tset) |
68318b8e | 8159 | { |
bb9d97b6 TH |
8160 | struct task_struct *task; |
8161 | ||
924f0d9a | 8162 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 8163 | sched_move_task(task); |
68318b8e SV |
8164 | } |
8165 | ||
eb95419b TH |
8166 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
8167 | struct cgroup_subsys_state *old_css, | |
8168 | struct task_struct *task) | |
068c5cc5 PZ |
8169 | { |
8170 | /* | |
8171 | * cgroup_exit() is called in the copy_process() failure path. | |
8172 | * Ignore this case since the task hasn't ran yet, this avoids | |
8173 | * trying to poke a half freed task state from generic code. | |
8174 | */ | |
8175 | if (!(task->flags & PF_EXITING)) | |
8176 | return; | |
8177 | ||
8178 | sched_move_task(task); | |
8179 | } | |
8180 | ||
052f1dc7 | 8181 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8182 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8183 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8184 | { |
182446d0 | 8185 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8186 | } |
8187 | ||
182446d0 TH |
8188 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8189 | struct cftype *cft) | |
68318b8e | 8190 | { |
182446d0 | 8191 | struct task_group *tg = css_tg(css); |
68318b8e | 8192 | |
c8b28116 | 8193 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8194 | } |
ab84d31e PT |
8195 | |
8196 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8197 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8198 | ||
ab84d31e PT |
8199 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8200 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8201 | ||
a790de99 PT |
8202 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8203 | ||
ab84d31e PT |
8204 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8205 | { | |
56f570e5 | 8206 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8207 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8208 | |
8209 | if (tg == &root_task_group) | |
8210 | return -EINVAL; | |
8211 | ||
8212 | /* | |
8213 | * Ensure we have at some amount of bandwidth every period. This is | |
8214 | * to prevent reaching a state of large arrears when throttled via | |
8215 | * entity_tick() resulting in prolonged exit starvation. | |
8216 | */ | |
8217 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8218 | return -EINVAL; | |
8219 | ||
8220 | /* | |
8221 | * Likewise, bound things on the otherside by preventing insane quota | |
8222 | * periods. This also allows us to normalize in computing quota | |
8223 | * feasibility. | |
8224 | */ | |
8225 | if (period > max_cfs_quota_period) | |
8226 | return -EINVAL; | |
8227 | ||
0e59bdae KT |
8228 | /* |
8229 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8230 | * unthrottle_offline_cfs_rqs(). | |
8231 | */ | |
8232 | get_online_cpus(); | |
a790de99 PT |
8233 | mutex_lock(&cfs_constraints_mutex); |
8234 | ret = __cfs_schedulable(tg, period, quota); | |
8235 | if (ret) | |
8236 | goto out_unlock; | |
8237 | ||
58088ad0 | 8238 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8239 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8240 | /* |
8241 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8242 | * before making related changes, and on->off must occur afterwards | |
8243 | */ | |
8244 | if (runtime_enabled && !runtime_was_enabled) | |
8245 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8246 | raw_spin_lock_irq(&cfs_b->lock); |
8247 | cfs_b->period = ns_to_ktime(period); | |
8248 | cfs_b->quota = quota; | |
58088ad0 | 8249 | |
a9cf55b2 | 8250 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8251 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8252 | if (runtime_enabled) |
8253 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8254 | raw_spin_unlock_irq(&cfs_b->lock); |
8255 | ||
0e59bdae | 8256 | for_each_online_cpu(i) { |
ab84d31e | 8257 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8258 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8259 | |
8260 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8261 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8262 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8263 | |
029632fb | 8264 | if (cfs_rq->throttled) |
671fd9da | 8265 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8266 | raw_spin_unlock_irq(&rq->lock); |
8267 | } | |
1ee14e6c BS |
8268 | if (runtime_was_enabled && !runtime_enabled) |
8269 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8270 | out_unlock: |
8271 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8272 | put_online_cpus(); |
ab84d31e | 8273 | |
a790de99 | 8274 | return ret; |
ab84d31e PT |
8275 | } |
8276 | ||
8277 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8278 | { | |
8279 | u64 quota, period; | |
8280 | ||
029632fb | 8281 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8282 | if (cfs_quota_us < 0) |
8283 | quota = RUNTIME_INF; | |
8284 | else | |
8285 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8286 | ||
8287 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8288 | } | |
8289 | ||
8290 | long tg_get_cfs_quota(struct task_group *tg) | |
8291 | { | |
8292 | u64 quota_us; | |
8293 | ||
029632fb | 8294 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8295 | return -1; |
8296 | ||
029632fb | 8297 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8298 | do_div(quota_us, NSEC_PER_USEC); |
8299 | ||
8300 | return quota_us; | |
8301 | } | |
8302 | ||
8303 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8304 | { | |
8305 | u64 quota, period; | |
8306 | ||
8307 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8308 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8309 | |
ab84d31e PT |
8310 | return tg_set_cfs_bandwidth(tg, period, quota); |
8311 | } | |
8312 | ||
8313 | long tg_get_cfs_period(struct task_group *tg) | |
8314 | { | |
8315 | u64 cfs_period_us; | |
8316 | ||
029632fb | 8317 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8318 | do_div(cfs_period_us, NSEC_PER_USEC); |
8319 | ||
8320 | return cfs_period_us; | |
8321 | } | |
8322 | ||
182446d0 TH |
8323 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8324 | struct cftype *cft) | |
ab84d31e | 8325 | { |
182446d0 | 8326 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8327 | } |
8328 | ||
182446d0 TH |
8329 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8330 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8331 | { |
182446d0 | 8332 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8333 | } |
8334 | ||
182446d0 TH |
8335 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8336 | struct cftype *cft) | |
ab84d31e | 8337 | { |
182446d0 | 8338 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8339 | } |
8340 | ||
182446d0 TH |
8341 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8342 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8343 | { |
182446d0 | 8344 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8345 | } |
8346 | ||
a790de99 PT |
8347 | struct cfs_schedulable_data { |
8348 | struct task_group *tg; | |
8349 | u64 period, quota; | |
8350 | }; | |
8351 | ||
8352 | /* | |
8353 | * normalize group quota/period to be quota/max_period | |
8354 | * note: units are usecs | |
8355 | */ | |
8356 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8357 | struct cfs_schedulable_data *d) | |
8358 | { | |
8359 | u64 quota, period; | |
8360 | ||
8361 | if (tg == d->tg) { | |
8362 | period = d->period; | |
8363 | quota = d->quota; | |
8364 | } else { | |
8365 | period = tg_get_cfs_period(tg); | |
8366 | quota = tg_get_cfs_quota(tg); | |
8367 | } | |
8368 | ||
8369 | /* note: these should typically be equivalent */ | |
8370 | if (quota == RUNTIME_INF || quota == -1) | |
8371 | return RUNTIME_INF; | |
8372 | ||
8373 | return to_ratio(period, quota); | |
8374 | } | |
8375 | ||
8376 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8377 | { | |
8378 | struct cfs_schedulable_data *d = data; | |
029632fb | 8379 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8380 | s64 quota = 0, parent_quota = -1; |
8381 | ||
8382 | if (!tg->parent) { | |
8383 | quota = RUNTIME_INF; | |
8384 | } else { | |
029632fb | 8385 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8386 | |
8387 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8388 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8389 | |
8390 | /* | |
8391 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8392 | * limit is set | |
8393 | */ | |
8394 | if (quota == RUNTIME_INF) | |
8395 | quota = parent_quota; | |
8396 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8397 | return -EINVAL; | |
8398 | } | |
9c58c79a | 8399 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8400 | |
8401 | return 0; | |
8402 | } | |
8403 | ||
8404 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8405 | { | |
8277434e | 8406 | int ret; |
a790de99 PT |
8407 | struct cfs_schedulable_data data = { |
8408 | .tg = tg, | |
8409 | .period = period, | |
8410 | .quota = quota, | |
8411 | }; | |
8412 | ||
8413 | if (quota != RUNTIME_INF) { | |
8414 | do_div(data.period, NSEC_PER_USEC); | |
8415 | do_div(data.quota, NSEC_PER_USEC); | |
8416 | } | |
8417 | ||
8277434e PT |
8418 | rcu_read_lock(); |
8419 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8420 | rcu_read_unlock(); | |
8421 | ||
8422 | return ret; | |
a790de99 | 8423 | } |
e8da1b18 | 8424 | |
2da8ca82 | 8425 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8426 | { |
2da8ca82 | 8427 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8428 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8429 | |
44ffc75b TH |
8430 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8431 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8432 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8433 | |
8434 | return 0; | |
8435 | } | |
ab84d31e | 8436 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8437 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8438 | |
052f1dc7 | 8439 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8440 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8441 | struct cftype *cft, s64 val) | |
6f505b16 | 8442 | { |
182446d0 | 8443 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8444 | } |
8445 | ||
182446d0 TH |
8446 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8447 | struct cftype *cft) | |
6f505b16 | 8448 | { |
182446d0 | 8449 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8450 | } |
d0b27fa7 | 8451 | |
182446d0 TH |
8452 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8453 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8454 | { |
182446d0 | 8455 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8456 | } |
8457 | ||
182446d0 TH |
8458 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8459 | struct cftype *cft) | |
d0b27fa7 | 8460 | { |
182446d0 | 8461 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8462 | } |
6d6bc0ad | 8463 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8464 | |
fe5c7cc2 | 8465 | static struct cftype cpu_files[] = { |
052f1dc7 | 8466 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8467 | { |
8468 | .name = "shares", | |
f4c753b7 PM |
8469 | .read_u64 = cpu_shares_read_u64, |
8470 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8471 | }, |
052f1dc7 | 8472 | #endif |
ab84d31e PT |
8473 | #ifdef CONFIG_CFS_BANDWIDTH |
8474 | { | |
8475 | .name = "cfs_quota_us", | |
8476 | .read_s64 = cpu_cfs_quota_read_s64, | |
8477 | .write_s64 = cpu_cfs_quota_write_s64, | |
8478 | }, | |
8479 | { | |
8480 | .name = "cfs_period_us", | |
8481 | .read_u64 = cpu_cfs_period_read_u64, | |
8482 | .write_u64 = cpu_cfs_period_write_u64, | |
8483 | }, | |
e8da1b18 NR |
8484 | { |
8485 | .name = "stat", | |
2da8ca82 | 8486 | .seq_show = cpu_stats_show, |
e8da1b18 | 8487 | }, |
ab84d31e | 8488 | #endif |
052f1dc7 | 8489 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8490 | { |
9f0c1e56 | 8491 | .name = "rt_runtime_us", |
06ecb27c PM |
8492 | .read_s64 = cpu_rt_runtime_read, |
8493 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8494 | }, |
d0b27fa7 PZ |
8495 | { |
8496 | .name = "rt_period_us", | |
f4c753b7 PM |
8497 | .read_u64 = cpu_rt_period_read_uint, |
8498 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8499 | }, |
052f1dc7 | 8500 | #endif |
4baf6e33 | 8501 | { } /* terminate */ |
68318b8e SV |
8502 | }; |
8503 | ||
073219e9 | 8504 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8505 | .css_alloc = cpu_cgroup_css_alloc, |
8506 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8507 | .css_online = cpu_cgroup_css_online, |
8508 | .css_offline = cpu_cgroup_css_offline, | |
eeb61e53 | 8509 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8510 | .can_attach = cpu_cgroup_can_attach, |
8511 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8512 | .exit = cpu_cgroup_exit, |
5577964e | 8513 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8514 | .early_init = 1, |
8515 | }; | |
8516 | ||
052f1dc7 | 8517 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8518 | |
b637a328 PM |
8519 | void dump_cpu_task(int cpu) |
8520 | { | |
8521 | pr_info("Task dump for CPU %d:\n", cpu); | |
8522 | sched_show_task(cpu_curr(cpu)); | |
8523 | } |