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 | { | |
c5905afb IM |
167 | if (static_key_enabled(&sched_feat_keys[i])) |
168 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
169 | } |
170 | ||
171 | static void sched_feat_enable(int i) | |
172 | { | |
c5905afb IM |
173 | if (!static_key_enabled(&sched_feat_keys[i])) |
174 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
175 | } |
176 | #else | |
177 | static void sched_feat_disable(int i) { }; | |
178 | static void sched_feat_enable(int i) { }; | |
179 | #endif /* HAVE_JUMP_LABEL */ | |
180 | ||
1a687c2e | 181 | static int sched_feat_set(char *cmp) |
f00b45c1 | 182 | { |
f00b45c1 | 183 | int i; |
1a687c2e | 184 | int neg = 0; |
f00b45c1 | 185 | |
524429c3 | 186 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
187 | neg = 1; |
188 | cmp += 3; | |
189 | } | |
190 | ||
f8b6d1cc | 191 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 192 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 193 | if (neg) { |
f00b45c1 | 194 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
195 | sched_feat_disable(i); |
196 | } else { | |
f00b45c1 | 197 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
198 | sched_feat_enable(i); |
199 | } | |
f00b45c1 PZ |
200 | break; |
201 | } | |
202 | } | |
203 | ||
1a687c2e MG |
204 | return i; |
205 | } | |
206 | ||
207 | static ssize_t | |
208 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
209 | size_t cnt, loff_t *ppos) | |
210 | { | |
211 | char buf[64]; | |
212 | char *cmp; | |
213 | int i; | |
5cd08fbf | 214 | struct inode *inode; |
1a687c2e MG |
215 | |
216 | if (cnt > 63) | |
217 | cnt = 63; | |
218 | ||
219 | if (copy_from_user(&buf, ubuf, cnt)) | |
220 | return -EFAULT; | |
221 | ||
222 | buf[cnt] = 0; | |
223 | cmp = strstrip(buf); | |
224 | ||
5cd08fbf JB |
225 | /* Ensure the static_key remains in a consistent state */ |
226 | inode = file_inode(filp); | |
227 | mutex_lock(&inode->i_mutex); | |
1a687c2e | 228 | i = sched_feat_set(cmp); |
5cd08fbf | 229 | mutex_unlock(&inode->i_mutex); |
f8b6d1cc | 230 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
231 | return -EINVAL; |
232 | ||
42994724 | 233 | *ppos += cnt; |
f00b45c1 PZ |
234 | |
235 | return cnt; | |
236 | } | |
237 | ||
34f3a814 LZ |
238 | static int sched_feat_open(struct inode *inode, struct file *filp) |
239 | { | |
240 | return single_open(filp, sched_feat_show, NULL); | |
241 | } | |
242 | ||
828c0950 | 243 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
244 | .open = sched_feat_open, |
245 | .write = sched_feat_write, | |
246 | .read = seq_read, | |
247 | .llseek = seq_lseek, | |
248 | .release = single_release, | |
f00b45c1 PZ |
249 | }; |
250 | ||
251 | static __init int sched_init_debug(void) | |
252 | { | |
f00b45c1 PZ |
253 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
254 | &sched_feat_fops); | |
255 | ||
256 | return 0; | |
257 | } | |
258 | late_initcall(sched_init_debug); | |
f8b6d1cc | 259 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 260 | |
b82d9fdd PZ |
261 | /* |
262 | * Number of tasks to iterate in a single balance run. | |
263 | * Limited because this is done with IRQs disabled. | |
264 | */ | |
265 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
266 | ||
e9e9250b PZ |
267 | /* |
268 | * period over which we average the RT time consumption, measured | |
269 | * in ms. | |
270 | * | |
271 | * default: 1s | |
272 | */ | |
273 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
274 | ||
fa85ae24 | 275 | /* |
9f0c1e56 | 276 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
277 | * default: 1s |
278 | */ | |
9f0c1e56 | 279 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 280 | |
029632fb | 281 | __read_mostly int scheduler_running; |
6892b75e | 282 | |
9f0c1e56 PZ |
283 | /* |
284 | * part of the period that we allow rt tasks to run in us. | |
285 | * default: 0.95s | |
286 | */ | |
287 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 288 | |
3fa0818b RR |
289 | /* cpus with isolated domains */ |
290 | cpumask_var_t cpu_isolated_map; | |
291 | ||
1da177e4 | 292 | /* |
cc2a73b5 | 293 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 294 | */ |
a9957449 | 295 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
296 | __acquires(rq->lock) |
297 | { | |
70b97a7f | 298 | struct rq *rq; |
1da177e4 LT |
299 | |
300 | local_irq_disable(); | |
301 | rq = this_rq(); | |
05fa785c | 302 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
303 | |
304 | return rq; | |
305 | } | |
306 | ||
8f4d37ec PZ |
307 | #ifdef CONFIG_SCHED_HRTICK |
308 | /* | |
309 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 310 | */ |
8f4d37ec | 311 | |
8f4d37ec PZ |
312 | static void hrtick_clear(struct rq *rq) |
313 | { | |
314 | if (hrtimer_active(&rq->hrtick_timer)) | |
315 | hrtimer_cancel(&rq->hrtick_timer); | |
316 | } | |
317 | ||
8f4d37ec PZ |
318 | /* |
319 | * High-resolution timer tick. | |
320 | * Runs from hardirq context with interrupts disabled. | |
321 | */ | |
322 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
323 | { | |
324 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
325 | ||
326 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
327 | ||
05fa785c | 328 | raw_spin_lock(&rq->lock); |
3e51f33f | 329 | update_rq_clock(rq); |
8f4d37ec | 330 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 331 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
332 | |
333 | return HRTIMER_NORESTART; | |
334 | } | |
335 | ||
95e904c7 | 336 | #ifdef CONFIG_SMP |
971ee28c | 337 | |
4961b6e1 | 338 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
339 | { |
340 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 341 | |
4961b6e1 | 342 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
343 | } |
344 | ||
31656519 PZ |
345 | /* |
346 | * called from hardirq (IPI) context | |
347 | */ | |
348 | static void __hrtick_start(void *arg) | |
b328ca18 | 349 | { |
31656519 | 350 | struct rq *rq = arg; |
b328ca18 | 351 | |
05fa785c | 352 | raw_spin_lock(&rq->lock); |
971ee28c | 353 | __hrtick_restart(rq); |
31656519 | 354 | rq->hrtick_csd_pending = 0; |
05fa785c | 355 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
356 | } |
357 | ||
31656519 PZ |
358 | /* |
359 | * Called to set the hrtick timer state. | |
360 | * | |
361 | * called with rq->lock held and irqs disabled | |
362 | */ | |
029632fb | 363 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 364 | { |
31656519 | 365 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 366 | ktime_t time; |
367 | s64 delta; | |
368 | ||
369 | /* | |
370 | * Don't schedule slices shorter than 10000ns, that just | |
371 | * doesn't make sense and can cause timer DoS. | |
372 | */ | |
373 | delta = max_t(s64, delay, 10000LL); | |
374 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 375 | |
cc584b21 | 376 | hrtimer_set_expires(timer, time); |
31656519 PZ |
377 | |
378 | if (rq == this_rq()) { | |
971ee28c | 379 | __hrtick_restart(rq); |
31656519 | 380 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 381 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
382 | rq->hrtick_csd_pending = 1; |
383 | } | |
b328ca18 PZ |
384 | } |
385 | ||
386 | static int | |
387 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
388 | { | |
389 | int cpu = (int)(long)hcpu; | |
390 | ||
391 | switch (action) { | |
392 | case CPU_UP_CANCELED: | |
393 | case CPU_UP_CANCELED_FROZEN: | |
394 | case CPU_DOWN_PREPARE: | |
395 | case CPU_DOWN_PREPARE_FROZEN: | |
396 | case CPU_DEAD: | |
397 | case CPU_DEAD_FROZEN: | |
31656519 | 398 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
399 | return NOTIFY_OK; |
400 | } | |
401 | ||
402 | return NOTIFY_DONE; | |
403 | } | |
404 | ||
fa748203 | 405 | static __init void init_hrtick(void) |
b328ca18 PZ |
406 | { |
407 | hotcpu_notifier(hotplug_hrtick, 0); | |
408 | } | |
31656519 PZ |
409 | #else |
410 | /* | |
411 | * Called to set the hrtick timer state. | |
412 | * | |
413 | * called with rq->lock held and irqs disabled | |
414 | */ | |
029632fb | 415 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 416 | { |
86893335 WL |
417 | /* |
418 | * Don't schedule slices shorter than 10000ns, that just | |
419 | * doesn't make sense. Rely on vruntime for fairness. | |
420 | */ | |
421 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
422 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
423 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 424 | } |
b328ca18 | 425 | |
006c75f1 | 426 | static inline void init_hrtick(void) |
8f4d37ec | 427 | { |
8f4d37ec | 428 | } |
31656519 | 429 | #endif /* CONFIG_SMP */ |
8f4d37ec | 430 | |
31656519 | 431 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 432 | { |
31656519 PZ |
433 | #ifdef CONFIG_SMP |
434 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 435 | |
31656519 PZ |
436 | rq->hrtick_csd.flags = 0; |
437 | rq->hrtick_csd.func = __hrtick_start; | |
438 | rq->hrtick_csd.info = rq; | |
439 | #endif | |
8f4d37ec | 440 | |
31656519 PZ |
441 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
442 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 443 | } |
006c75f1 | 444 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
445 | static inline void hrtick_clear(struct rq *rq) |
446 | { | |
447 | } | |
448 | ||
8f4d37ec PZ |
449 | static inline void init_rq_hrtick(struct rq *rq) |
450 | { | |
451 | } | |
452 | ||
b328ca18 PZ |
453 | static inline void init_hrtick(void) |
454 | { | |
455 | } | |
006c75f1 | 456 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 457 | |
fd99f91a PZ |
458 | /* |
459 | * cmpxchg based fetch_or, macro so it works for different integer types | |
460 | */ | |
461 | #define fetch_or(ptr, val) \ | |
462 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
463 | for (;;) { \ | |
464 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
465 | if (__old == __val) \ | |
466 | break; \ | |
467 | __val = __old; \ | |
468 | } \ | |
469 | __old; \ | |
470 | }) | |
471 | ||
e3baac47 | 472 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
473 | /* |
474 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
475 | * this avoids any races wrt polling state changes and thereby avoids | |
476 | * spurious IPIs. | |
477 | */ | |
478 | static bool set_nr_and_not_polling(struct task_struct *p) | |
479 | { | |
480 | struct thread_info *ti = task_thread_info(p); | |
481 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
482 | } | |
e3baac47 PZ |
483 | |
484 | /* | |
485 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
486 | * | |
487 | * If this returns true, then the idle task promises to call | |
488 | * sched_ttwu_pending() and reschedule soon. | |
489 | */ | |
490 | static bool set_nr_if_polling(struct task_struct *p) | |
491 | { | |
492 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 493 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
494 | |
495 | for (;;) { | |
496 | if (!(val & _TIF_POLLING_NRFLAG)) | |
497 | return false; | |
498 | if (val & _TIF_NEED_RESCHED) | |
499 | return true; | |
500 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
501 | if (old == val) | |
502 | break; | |
503 | val = old; | |
504 | } | |
505 | return true; | |
506 | } | |
507 | ||
fd99f91a PZ |
508 | #else |
509 | static bool set_nr_and_not_polling(struct task_struct *p) | |
510 | { | |
511 | set_tsk_need_resched(p); | |
512 | return true; | |
513 | } | |
e3baac47 PZ |
514 | |
515 | #ifdef CONFIG_SMP | |
516 | static bool set_nr_if_polling(struct task_struct *p) | |
517 | { | |
518 | return false; | |
519 | } | |
520 | #endif | |
fd99f91a PZ |
521 | #endif |
522 | ||
76751049 PZ |
523 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
524 | { | |
525 | struct wake_q_node *node = &task->wake_q; | |
526 | ||
527 | /* | |
528 | * Atomically grab the task, if ->wake_q is !nil already it means | |
529 | * its already queued (either by us or someone else) and will get the | |
530 | * wakeup due to that. | |
531 | * | |
532 | * This cmpxchg() implies a full barrier, which pairs with the write | |
533 | * barrier implied by the wakeup in wake_up_list(). | |
534 | */ | |
535 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
536 | return; | |
537 | ||
538 | get_task_struct(task); | |
539 | ||
540 | /* | |
541 | * The head is context local, there can be no concurrency. | |
542 | */ | |
543 | *head->lastp = node; | |
544 | head->lastp = &node->next; | |
545 | } | |
546 | ||
547 | void wake_up_q(struct wake_q_head *head) | |
548 | { | |
549 | struct wake_q_node *node = head->first; | |
550 | ||
551 | while (node != WAKE_Q_TAIL) { | |
552 | struct task_struct *task; | |
553 | ||
554 | task = container_of(node, struct task_struct, wake_q); | |
555 | BUG_ON(!task); | |
556 | /* task can safely be re-inserted now */ | |
557 | node = node->next; | |
558 | task->wake_q.next = NULL; | |
559 | ||
560 | /* | |
561 | * wake_up_process() implies a wmb() to pair with the queueing | |
562 | * in wake_q_add() so as not to miss wakeups. | |
563 | */ | |
564 | wake_up_process(task); | |
565 | put_task_struct(task); | |
566 | } | |
567 | } | |
568 | ||
c24d20db | 569 | /* |
8875125e | 570 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
571 | * |
572 | * On UP this means the setting of the need_resched flag, on SMP it | |
573 | * might also involve a cross-CPU call to trigger the scheduler on | |
574 | * the target CPU. | |
575 | */ | |
8875125e | 576 | void resched_curr(struct rq *rq) |
c24d20db | 577 | { |
8875125e | 578 | struct task_struct *curr = rq->curr; |
c24d20db IM |
579 | int cpu; |
580 | ||
8875125e | 581 | lockdep_assert_held(&rq->lock); |
c24d20db | 582 | |
8875125e | 583 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
584 | return; |
585 | ||
8875125e | 586 | cpu = cpu_of(rq); |
fd99f91a | 587 | |
f27dde8d | 588 | if (cpu == smp_processor_id()) { |
8875125e | 589 | set_tsk_need_resched(curr); |
f27dde8d | 590 | set_preempt_need_resched(); |
c24d20db | 591 | return; |
f27dde8d | 592 | } |
c24d20db | 593 | |
8875125e | 594 | if (set_nr_and_not_polling(curr)) |
c24d20db | 595 | smp_send_reschedule(cpu); |
dfc68f29 AL |
596 | else |
597 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
598 | } |
599 | ||
029632fb | 600 | void resched_cpu(int cpu) |
c24d20db IM |
601 | { |
602 | struct rq *rq = cpu_rq(cpu); | |
603 | unsigned long flags; | |
604 | ||
05fa785c | 605 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 606 | return; |
8875125e | 607 | resched_curr(rq); |
05fa785c | 608 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 609 | } |
06d8308c | 610 | |
b021fe3e | 611 | #ifdef CONFIG_SMP |
3451d024 | 612 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
613 | /* |
614 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
615 | * from an idle cpu. This is good for power-savings. | |
616 | * | |
617 | * We don't do similar optimization for completely idle system, as | |
618 | * selecting an idle cpu will add more delays to the timers than intended | |
619 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
620 | */ | |
bc7a34b8 | 621 | int get_nohz_timer_target(void) |
83cd4fe2 | 622 | { |
bc7a34b8 | 623 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
624 | struct sched_domain *sd; |
625 | ||
bc7a34b8 | 626 | if (!idle_cpu(cpu)) |
6201b4d6 VK |
627 | return cpu; |
628 | ||
057f3fad | 629 | rcu_read_lock(); |
83cd4fe2 | 630 | for_each_domain(cpu, sd) { |
057f3fad PZ |
631 | for_each_cpu(i, sched_domain_span(sd)) { |
632 | if (!idle_cpu(i)) { | |
633 | cpu = i; | |
634 | goto unlock; | |
635 | } | |
636 | } | |
83cd4fe2 | 637 | } |
057f3fad PZ |
638 | unlock: |
639 | rcu_read_unlock(); | |
83cd4fe2 VP |
640 | return cpu; |
641 | } | |
06d8308c TG |
642 | /* |
643 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
644 | * idle CPU then this timer might expire before the next timer event | |
645 | * which is scheduled to wake up that CPU. In case of a completely | |
646 | * idle system the next event might even be infinite time into the | |
647 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
648 | * leaves the inner idle loop so the newly added timer is taken into | |
649 | * account when the CPU goes back to idle and evaluates the timer | |
650 | * wheel for the next timer event. | |
651 | */ | |
1c20091e | 652 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
653 | { |
654 | struct rq *rq = cpu_rq(cpu); | |
655 | ||
656 | if (cpu == smp_processor_id()) | |
657 | return; | |
658 | ||
67b9ca70 | 659 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 660 | smp_send_reschedule(cpu); |
dfc68f29 AL |
661 | else |
662 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
663 | } |
664 | ||
c5bfece2 | 665 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 666 | { |
53c5fa16 FW |
667 | /* |
668 | * We just need the target to call irq_exit() and re-evaluate | |
669 | * the next tick. The nohz full kick at least implies that. | |
670 | * If needed we can still optimize that later with an | |
671 | * empty IRQ. | |
672 | */ | |
c5bfece2 | 673 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
674 | if (cpu != smp_processor_id() || |
675 | tick_nohz_tick_stopped()) | |
53c5fa16 | 676 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
677 | return true; |
678 | } | |
679 | ||
680 | return false; | |
681 | } | |
682 | ||
683 | void wake_up_nohz_cpu(int cpu) | |
684 | { | |
c5bfece2 | 685 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
686 | wake_up_idle_cpu(cpu); |
687 | } | |
688 | ||
ca38062e | 689 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 690 | { |
1c792db7 | 691 | int cpu = smp_processor_id(); |
873b4c65 VG |
692 | |
693 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
694 | return false; | |
695 | ||
696 | if (idle_cpu(cpu) && !need_resched()) | |
697 | return true; | |
698 | ||
699 | /* | |
700 | * We can't run Idle Load Balance on this CPU for this time so we | |
701 | * cancel it and clear NOHZ_BALANCE_KICK | |
702 | */ | |
703 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
704 | return false; | |
45bf76df IM |
705 | } |
706 | ||
3451d024 | 707 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 708 | |
ca38062e | 709 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 710 | { |
ca38062e | 711 | return false; |
2069dd75 PZ |
712 | } |
713 | ||
3451d024 | 714 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 715 | |
ce831b38 FW |
716 | #ifdef CONFIG_NO_HZ_FULL |
717 | bool sched_can_stop_tick(void) | |
718 | { | |
1e78cdbd RR |
719 | /* |
720 | * FIFO realtime policy runs the highest priority task. Other runnable | |
721 | * tasks are of a lower priority. The scheduler tick does nothing. | |
722 | */ | |
723 | if (current->policy == SCHED_FIFO) | |
724 | return true; | |
725 | ||
726 | /* | |
727 | * Round-robin realtime tasks time slice with other tasks at the same | |
728 | * realtime priority. Is this task the only one at this priority? | |
729 | */ | |
730 | if (current->policy == SCHED_RR) { | |
731 | struct sched_rt_entity *rt_se = ¤t->rt; | |
732 | ||
733 | return rt_se->run_list.prev == rt_se->run_list.next; | |
734 | } | |
735 | ||
3882ec64 FW |
736 | /* |
737 | * More than one running task need preemption. | |
738 | * nr_running update is assumed to be visible | |
739 | * after IPI is sent from wakers. | |
740 | */ | |
541b8264 VK |
741 | if (this_rq()->nr_running > 1) |
742 | return false; | |
ce831b38 | 743 | |
541b8264 | 744 | return true; |
ce831b38 FW |
745 | } |
746 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 747 | |
029632fb | 748 | void sched_avg_update(struct rq *rq) |
18d95a28 | 749 | { |
e9e9250b PZ |
750 | s64 period = sched_avg_period(); |
751 | ||
78becc27 | 752 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
753 | /* |
754 | * Inline assembly required to prevent the compiler | |
755 | * optimising this loop into a divmod call. | |
756 | * See __iter_div_u64_rem() for another example of this. | |
757 | */ | |
758 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
759 | rq->age_stamp += period; |
760 | rq->rt_avg /= 2; | |
761 | } | |
18d95a28 PZ |
762 | } |
763 | ||
6d6bc0ad | 764 | #endif /* CONFIG_SMP */ |
18d95a28 | 765 | |
a790de99 PT |
766 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
767 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 768 | /* |
8277434e PT |
769 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
770 | * node and @up when leaving it for the final time. | |
771 | * | |
772 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 773 | */ |
029632fb | 774 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 775 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
776 | { |
777 | struct task_group *parent, *child; | |
eb755805 | 778 | int ret; |
c09595f6 | 779 | |
8277434e PT |
780 | parent = from; |
781 | ||
c09595f6 | 782 | down: |
eb755805 PZ |
783 | ret = (*down)(parent, data); |
784 | if (ret) | |
8277434e | 785 | goto out; |
c09595f6 PZ |
786 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
787 | parent = child; | |
788 | goto down; | |
789 | ||
790 | up: | |
791 | continue; | |
792 | } | |
eb755805 | 793 | ret = (*up)(parent, data); |
8277434e PT |
794 | if (ret || parent == from) |
795 | goto out; | |
c09595f6 PZ |
796 | |
797 | child = parent; | |
798 | parent = parent->parent; | |
799 | if (parent) | |
800 | goto up; | |
8277434e | 801 | out: |
eb755805 | 802 | return ret; |
c09595f6 PZ |
803 | } |
804 | ||
029632fb | 805 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 806 | { |
e2b245f8 | 807 | return 0; |
eb755805 | 808 | } |
18d95a28 PZ |
809 | #endif |
810 | ||
45bf76df IM |
811 | static void set_load_weight(struct task_struct *p) |
812 | { | |
f05998d4 NR |
813 | int prio = p->static_prio - MAX_RT_PRIO; |
814 | struct load_weight *load = &p->se.load; | |
815 | ||
dd41f596 IM |
816 | /* |
817 | * SCHED_IDLE tasks get minimal weight: | |
818 | */ | |
819 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 820 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 821 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
822 | return; |
823 | } | |
71f8bd46 | 824 | |
c8b28116 | 825 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 826 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
827 | } |
828 | ||
371fd7e7 | 829 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 830 | { |
a64692a3 | 831 | update_rq_clock(rq); |
43148951 | 832 | sched_info_queued(rq, p); |
371fd7e7 | 833 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
834 | } |
835 | ||
371fd7e7 | 836 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 837 | { |
a64692a3 | 838 | update_rq_clock(rq); |
43148951 | 839 | sched_info_dequeued(rq, p); |
371fd7e7 | 840 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
841 | } |
842 | ||
029632fb | 843 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
844 | { |
845 | if (task_contributes_to_load(p)) | |
846 | rq->nr_uninterruptible--; | |
847 | ||
371fd7e7 | 848 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
849 | } |
850 | ||
029632fb | 851 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
852 | { |
853 | if (task_contributes_to_load(p)) | |
854 | rq->nr_uninterruptible++; | |
855 | ||
371fd7e7 | 856 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
857 | } |
858 | ||
fe44d621 | 859 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 860 | { |
095c0aa8 GC |
861 | /* |
862 | * In theory, the compile should just see 0 here, and optimize out the call | |
863 | * to sched_rt_avg_update. But I don't trust it... | |
864 | */ | |
865 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
866 | s64 steal = 0, irq_delta = 0; | |
867 | #endif | |
868 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 869 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
870 | |
871 | /* | |
872 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
873 | * this case when a previous update_rq_clock() happened inside a | |
874 | * {soft,}irq region. | |
875 | * | |
876 | * When this happens, we stop ->clock_task and only update the | |
877 | * prev_irq_time stamp to account for the part that fit, so that a next | |
878 | * update will consume the rest. This ensures ->clock_task is | |
879 | * monotonic. | |
880 | * | |
881 | * It does however cause some slight miss-attribution of {soft,}irq | |
882 | * time, a more accurate solution would be to update the irq_time using | |
883 | * the current rq->clock timestamp, except that would require using | |
884 | * atomic ops. | |
885 | */ | |
886 | if (irq_delta > delta) | |
887 | irq_delta = delta; | |
888 | ||
889 | rq->prev_irq_time += irq_delta; | |
890 | delta -= irq_delta; | |
095c0aa8 GC |
891 | #endif |
892 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 893 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
894 | steal = paravirt_steal_clock(cpu_of(rq)); |
895 | steal -= rq->prev_steal_time_rq; | |
896 | ||
897 | if (unlikely(steal > delta)) | |
898 | steal = delta; | |
899 | ||
095c0aa8 | 900 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
901 | delta -= steal; |
902 | } | |
903 | #endif | |
904 | ||
fe44d621 PZ |
905 | rq->clock_task += delta; |
906 | ||
095c0aa8 | 907 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 908 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
909 | sched_rt_avg_update(rq, irq_delta + steal); |
910 | #endif | |
aa483808 VP |
911 | } |
912 | ||
34f971f6 PZ |
913 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
914 | { | |
915 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
916 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
917 | ||
918 | if (stop) { | |
919 | /* | |
920 | * Make it appear like a SCHED_FIFO task, its something | |
921 | * userspace knows about and won't get confused about. | |
922 | * | |
923 | * Also, it will make PI more or less work without too | |
924 | * much confusion -- but then, stop work should not | |
925 | * rely on PI working anyway. | |
926 | */ | |
927 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
928 | ||
929 | stop->sched_class = &stop_sched_class; | |
930 | } | |
931 | ||
932 | cpu_rq(cpu)->stop = stop; | |
933 | ||
934 | if (old_stop) { | |
935 | /* | |
936 | * Reset it back to a normal scheduling class so that | |
937 | * it can die in pieces. | |
938 | */ | |
939 | old_stop->sched_class = &rt_sched_class; | |
940 | } | |
941 | } | |
942 | ||
14531189 | 943 | /* |
dd41f596 | 944 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 945 | */ |
14531189 IM |
946 | static inline int __normal_prio(struct task_struct *p) |
947 | { | |
dd41f596 | 948 | return p->static_prio; |
14531189 IM |
949 | } |
950 | ||
b29739f9 IM |
951 | /* |
952 | * Calculate the expected normal priority: i.e. priority | |
953 | * without taking RT-inheritance into account. Might be | |
954 | * boosted by interactivity modifiers. Changes upon fork, | |
955 | * setprio syscalls, and whenever the interactivity | |
956 | * estimator recalculates. | |
957 | */ | |
36c8b586 | 958 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
959 | { |
960 | int prio; | |
961 | ||
aab03e05 DF |
962 | if (task_has_dl_policy(p)) |
963 | prio = MAX_DL_PRIO-1; | |
964 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
965 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
966 | else | |
967 | prio = __normal_prio(p); | |
968 | return prio; | |
969 | } | |
970 | ||
971 | /* | |
972 | * Calculate the current priority, i.e. the priority | |
973 | * taken into account by the scheduler. This value might | |
974 | * be boosted by RT tasks, or might be boosted by | |
975 | * interactivity modifiers. Will be RT if the task got | |
976 | * RT-boosted. If not then it returns p->normal_prio. | |
977 | */ | |
36c8b586 | 978 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
979 | { |
980 | p->normal_prio = normal_prio(p); | |
981 | /* | |
982 | * If we are RT tasks or we were boosted to RT priority, | |
983 | * keep the priority unchanged. Otherwise, update priority | |
984 | * to the normal priority: | |
985 | */ | |
986 | if (!rt_prio(p->prio)) | |
987 | return p->normal_prio; | |
988 | return p->prio; | |
989 | } | |
990 | ||
1da177e4 LT |
991 | /** |
992 | * task_curr - is this task currently executing on a CPU? | |
993 | * @p: the task in question. | |
e69f6186 YB |
994 | * |
995 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 996 | */ |
36c8b586 | 997 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
998 | { |
999 | return cpu_curr(task_cpu(p)) == p; | |
1000 | } | |
1001 | ||
67dfa1b7 | 1002 | /* |
4c9a4bc8 PZ |
1003 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1004 | * use the balance_callback list if you want balancing. | |
1005 | * | |
1006 | * this means any call to check_class_changed() must be followed by a call to | |
1007 | * balance_callback(). | |
67dfa1b7 | 1008 | */ |
cb469845 SR |
1009 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1010 | const struct sched_class *prev_class, | |
da7a735e | 1011 | int oldprio) |
cb469845 SR |
1012 | { |
1013 | if (prev_class != p->sched_class) { | |
1014 | if (prev_class->switched_from) | |
da7a735e | 1015 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1016 | |
da7a735e | 1017 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1018 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1019 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1020 | } |
1021 | ||
029632fb | 1022 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1023 | { |
1024 | const struct sched_class *class; | |
1025 | ||
1026 | if (p->sched_class == rq->curr->sched_class) { | |
1027 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1028 | } else { | |
1029 | for_each_class(class) { | |
1030 | if (class == rq->curr->sched_class) | |
1031 | break; | |
1032 | if (class == p->sched_class) { | |
8875125e | 1033 | resched_curr(rq); |
1e5a7405 PZ |
1034 | break; |
1035 | } | |
1036 | } | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * A queue event has occurred, and we're going to schedule. In | |
1041 | * this case, we can save a useless back to back clock update. | |
1042 | */ | |
da0c1e65 | 1043 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 1044 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
1045 | } |
1046 | ||
1da177e4 | 1047 | #ifdef CONFIG_SMP |
5cc389bc PZ |
1048 | /* |
1049 | * This is how migration works: | |
1050 | * | |
1051 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1052 | * stop_one_cpu(). | |
1053 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1054 | * off the CPU) | |
1055 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1056 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1057 | * it and puts it into the right queue. | |
1058 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1059 | * is done. | |
1060 | */ | |
1061 | ||
1062 | /* | |
1063 | * move_queued_task - move a queued task to new rq. | |
1064 | * | |
1065 | * Returns (locked) new rq. Old rq's lock is released. | |
1066 | */ | |
5e16bbc2 | 1067 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 1068 | { |
5cc389bc PZ |
1069 | lockdep_assert_held(&rq->lock); |
1070 | ||
1071 | dequeue_task(rq, p, 0); | |
1072 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
1073 | set_task_cpu(p, new_cpu); | |
1074 | raw_spin_unlock(&rq->lock); | |
1075 | ||
1076 | rq = cpu_rq(new_cpu); | |
1077 | ||
1078 | raw_spin_lock(&rq->lock); | |
1079 | BUG_ON(task_cpu(p) != new_cpu); | |
1080 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1081 | enqueue_task(rq, p, 0); | |
1082 | check_preempt_curr(rq, p, 0); | |
1083 | ||
1084 | return rq; | |
1085 | } | |
1086 | ||
1087 | struct migration_arg { | |
1088 | struct task_struct *task; | |
1089 | int dest_cpu; | |
1090 | }; | |
1091 | ||
1092 | /* | |
1093 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
1094 | * this because either it can't run here any more (set_cpus_allowed() | |
1095 | * away from this CPU, or CPU going down), or because we're | |
1096 | * attempting to rebalance this task on exec (sched_exec). | |
1097 | * | |
1098 | * So we race with normal scheduler movements, but that's OK, as long | |
1099 | * as the task is no longer on this CPU. | |
5cc389bc | 1100 | */ |
5e16bbc2 | 1101 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 1102 | { |
5cc389bc | 1103 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 1104 | return rq; |
5cc389bc PZ |
1105 | |
1106 | /* Affinity changed (again). */ | |
1107 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
5e16bbc2 | 1108 | return rq; |
5cc389bc | 1109 | |
5e16bbc2 PZ |
1110 | rq = move_queued_task(rq, p, dest_cpu); |
1111 | ||
1112 | return rq; | |
5cc389bc PZ |
1113 | } |
1114 | ||
1115 | /* | |
1116 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1117 | * and performs thread migration by bumping thread off CPU then | |
1118 | * 'pushing' onto another runqueue. | |
1119 | */ | |
1120 | static int migration_cpu_stop(void *data) | |
1121 | { | |
1122 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1123 | struct task_struct *p = arg->task; |
1124 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1125 | |
1126 | /* | |
1127 | * The original target cpu might have gone down and we might | |
1128 | * be on another cpu but it doesn't matter. | |
1129 | */ | |
1130 | local_irq_disable(); | |
1131 | /* | |
1132 | * We need to explicitly wake pending tasks before running | |
1133 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1134 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1135 | */ | |
1136 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1137 | |
1138 | raw_spin_lock(&p->pi_lock); | |
1139 | raw_spin_lock(&rq->lock); | |
1140 | /* | |
1141 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1142 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1143 | * we're holding p->pi_lock. | |
1144 | */ | |
1145 | if (task_rq(p) == rq && task_on_rq_queued(p)) | |
1146 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1147 | raw_spin_unlock(&rq->lock); | |
1148 | raw_spin_unlock(&p->pi_lock); | |
1149 | ||
5cc389bc PZ |
1150 | local_irq_enable(); |
1151 | return 0; | |
1152 | } | |
1153 | ||
1154 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) | |
1155 | { | |
1156 | if (p->sched_class->set_cpus_allowed) | |
1157 | p->sched_class->set_cpus_allowed(p, new_mask); | |
1158 | ||
1159 | cpumask_copy(&p->cpus_allowed, new_mask); | |
1160 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * Change a given task's CPU affinity. Migrate the thread to a | |
1165 | * proper CPU and schedule it away if the CPU it's executing on | |
1166 | * is removed from the allowed bitmask. | |
1167 | * | |
1168 | * NOTE: the caller must have a valid reference to the task, the | |
1169 | * task must not exit() & deallocate itself prematurely. The | |
1170 | * call is not atomic; no spinlocks may be held. | |
1171 | */ | |
1172 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1173 | { | |
1174 | unsigned long flags; | |
1175 | struct rq *rq; | |
1176 | unsigned int dest_cpu; | |
1177 | int ret = 0; | |
1178 | ||
1179 | rq = task_rq_lock(p, &flags); | |
1180 | ||
1181 | if (cpumask_equal(&p->cpus_allowed, new_mask)) | |
1182 | goto out; | |
1183 | ||
1184 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { | |
1185 | ret = -EINVAL; | |
1186 | goto out; | |
1187 | } | |
1188 | ||
1189 | do_set_cpus_allowed(p, new_mask); | |
1190 | ||
1191 | /* Can the task run on the task's current CPU? If so, we're done */ | |
1192 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1193 | goto out; | |
1194 | ||
1195 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); | |
1196 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
1197 | struct migration_arg arg = { p, dest_cpu }; | |
1198 | /* Need help from migration thread: drop lock and wait. */ | |
1199 | task_rq_unlock(rq, p, &flags); | |
1200 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
1201 | tlb_migrate_finish(p->mm); | |
1202 | return 0; | |
cbce1a68 PZ |
1203 | } else if (task_on_rq_queued(p)) { |
1204 | /* | |
1205 | * OK, since we're going to drop the lock immediately | |
1206 | * afterwards anyway. | |
1207 | */ | |
1208 | lockdep_unpin_lock(&rq->lock); | |
5e16bbc2 | 1209 | rq = move_queued_task(rq, p, dest_cpu); |
cbce1a68 PZ |
1210 | lockdep_pin_lock(&rq->lock); |
1211 | } | |
5cc389bc PZ |
1212 | out: |
1213 | task_rq_unlock(rq, p, &flags); | |
1214 | ||
1215 | return ret; | |
1216 | } | |
1217 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); | |
1218 | ||
dd41f596 | 1219 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1220 | { |
e2912009 PZ |
1221 | #ifdef CONFIG_SCHED_DEBUG |
1222 | /* | |
1223 | * We should never call set_task_cpu() on a blocked task, | |
1224 | * ttwu() will sort out the placement. | |
1225 | */ | |
077614ee | 1226 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1227 | !p->on_rq); |
0122ec5b PZ |
1228 | |
1229 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1230 | /* |
1231 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1232 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1233 | * | |
1234 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1235 | * see task_group(). |
6c6c54e1 PZ |
1236 | * |
1237 | * Furthermore, all task_rq users should acquire both locks, see | |
1238 | * task_rq_lock(). | |
1239 | */ | |
0122ec5b PZ |
1240 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1241 | lockdep_is_held(&task_rq(p)->lock))); | |
1242 | #endif | |
e2912009 PZ |
1243 | #endif |
1244 | ||
de1d7286 | 1245 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1246 | |
0c69774e | 1247 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1248 | if (p->sched_class->migrate_task_rq) |
1249 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1250 | p->se.nr_migrations++; |
ff303e66 | 1251 | perf_event_task_migrate(p); |
0c69774e | 1252 | } |
dd41f596 IM |
1253 | |
1254 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1255 | } |
1256 | ||
ac66f547 PZ |
1257 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1258 | { | |
da0c1e65 | 1259 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1260 | struct rq *src_rq, *dst_rq; |
1261 | ||
1262 | src_rq = task_rq(p); | |
1263 | dst_rq = cpu_rq(cpu); | |
1264 | ||
1265 | deactivate_task(src_rq, p, 0); | |
1266 | set_task_cpu(p, cpu); | |
1267 | activate_task(dst_rq, p, 0); | |
1268 | check_preempt_curr(dst_rq, p, 0); | |
1269 | } else { | |
1270 | /* | |
1271 | * Task isn't running anymore; make it appear like we migrated | |
1272 | * it before it went to sleep. This means on wakeup we make the | |
1273 | * previous cpu our targer instead of where it really is. | |
1274 | */ | |
1275 | p->wake_cpu = cpu; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | struct migration_swap_arg { | |
1280 | struct task_struct *src_task, *dst_task; | |
1281 | int src_cpu, dst_cpu; | |
1282 | }; | |
1283 | ||
1284 | static int migrate_swap_stop(void *data) | |
1285 | { | |
1286 | struct migration_swap_arg *arg = data; | |
1287 | struct rq *src_rq, *dst_rq; | |
1288 | int ret = -EAGAIN; | |
1289 | ||
1290 | src_rq = cpu_rq(arg->src_cpu); | |
1291 | dst_rq = cpu_rq(arg->dst_cpu); | |
1292 | ||
74602315 PZ |
1293 | double_raw_lock(&arg->src_task->pi_lock, |
1294 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1295 | double_rq_lock(src_rq, dst_rq); |
1296 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1297 | goto unlock; | |
1298 | ||
1299 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1300 | goto unlock; | |
1301 | ||
1302 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1303 | goto unlock; | |
1304 | ||
1305 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1306 | goto unlock; | |
1307 | ||
1308 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1309 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1310 | ||
1311 | ret = 0; | |
1312 | ||
1313 | unlock: | |
1314 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1315 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1316 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1317 | |
1318 | return ret; | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Cross migrate two tasks | |
1323 | */ | |
1324 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1325 | { | |
1326 | struct migration_swap_arg arg; | |
1327 | int ret = -EINVAL; | |
1328 | ||
ac66f547 PZ |
1329 | arg = (struct migration_swap_arg){ |
1330 | .src_task = cur, | |
1331 | .src_cpu = task_cpu(cur), | |
1332 | .dst_task = p, | |
1333 | .dst_cpu = task_cpu(p), | |
1334 | }; | |
1335 | ||
1336 | if (arg.src_cpu == arg.dst_cpu) | |
1337 | goto out; | |
1338 | ||
6acce3ef PZ |
1339 | /* |
1340 | * These three tests are all lockless; this is OK since all of them | |
1341 | * will be re-checked with proper locks held further down the line. | |
1342 | */ | |
ac66f547 PZ |
1343 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1344 | goto out; | |
1345 | ||
1346 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1347 | goto out; | |
1348 | ||
1349 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1350 | goto out; | |
1351 | ||
286549dc | 1352 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1353 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1354 | ||
1355 | out: | |
ac66f547 PZ |
1356 | return ret; |
1357 | } | |
1358 | ||
1da177e4 LT |
1359 | /* |
1360 | * wait_task_inactive - wait for a thread to unschedule. | |
1361 | * | |
85ba2d86 RM |
1362 | * If @match_state is nonzero, it's the @p->state value just checked and |
1363 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1364 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1365 | * we return a positive number (its total switch count). If a second call | |
1366 | * a short while later returns the same number, the caller can be sure that | |
1367 | * @p has remained unscheduled the whole time. | |
1368 | * | |
1da177e4 LT |
1369 | * The caller must ensure that the task *will* unschedule sometime soon, |
1370 | * else this function might spin for a *long* time. This function can't | |
1371 | * be called with interrupts off, or it may introduce deadlock with | |
1372 | * smp_call_function() if an IPI is sent by the same process we are | |
1373 | * waiting to become inactive. | |
1374 | */ | |
85ba2d86 | 1375 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1376 | { |
1377 | unsigned long flags; | |
da0c1e65 | 1378 | int running, queued; |
85ba2d86 | 1379 | unsigned long ncsw; |
70b97a7f | 1380 | struct rq *rq; |
1da177e4 | 1381 | |
3a5c359a AK |
1382 | for (;;) { |
1383 | /* | |
1384 | * We do the initial early heuristics without holding | |
1385 | * any task-queue locks at all. We'll only try to get | |
1386 | * the runqueue lock when things look like they will | |
1387 | * work out! | |
1388 | */ | |
1389 | rq = task_rq(p); | |
fa490cfd | 1390 | |
3a5c359a AK |
1391 | /* |
1392 | * If the task is actively running on another CPU | |
1393 | * still, just relax and busy-wait without holding | |
1394 | * any locks. | |
1395 | * | |
1396 | * NOTE! Since we don't hold any locks, it's not | |
1397 | * even sure that "rq" stays as the right runqueue! | |
1398 | * But we don't care, since "task_running()" will | |
1399 | * return false if the runqueue has changed and p | |
1400 | * is actually now running somewhere else! | |
1401 | */ | |
85ba2d86 RM |
1402 | while (task_running(rq, p)) { |
1403 | if (match_state && unlikely(p->state != match_state)) | |
1404 | return 0; | |
3a5c359a | 1405 | cpu_relax(); |
85ba2d86 | 1406 | } |
fa490cfd | 1407 | |
3a5c359a AK |
1408 | /* |
1409 | * Ok, time to look more closely! We need the rq | |
1410 | * lock now, to be *sure*. If we're wrong, we'll | |
1411 | * just go back and repeat. | |
1412 | */ | |
1413 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1414 | trace_sched_wait_task(p); |
3a5c359a | 1415 | running = task_running(rq, p); |
da0c1e65 | 1416 | queued = task_on_rq_queued(p); |
85ba2d86 | 1417 | ncsw = 0; |
f31e11d8 | 1418 | if (!match_state || p->state == match_state) |
93dcf55f | 1419 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1420 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1421 | |
85ba2d86 RM |
1422 | /* |
1423 | * If it changed from the expected state, bail out now. | |
1424 | */ | |
1425 | if (unlikely(!ncsw)) | |
1426 | break; | |
1427 | ||
3a5c359a AK |
1428 | /* |
1429 | * Was it really running after all now that we | |
1430 | * checked with the proper locks actually held? | |
1431 | * | |
1432 | * Oops. Go back and try again.. | |
1433 | */ | |
1434 | if (unlikely(running)) { | |
1435 | cpu_relax(); | |
1436 | continue; | |
1437 | } | |
fa490cfd | 1438 | |
3a5c359a AK |
1439 | /* |
1440 | * It's not enough that it's not actively running, | |
1441 | * it must be off the runqueue _entirely_, and not | |
1442 | * preempted! | |
1443 | * | |
80dd99b3 | 1444 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1445 | * running right now), it's preempted, and we should |
1446 | * yield - it could be a while. | |
1447 | */ | |
da0c1e65 | 1448 | if (unlikely(queued)) { |
8eb90c30 TG |
1449 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1450 | ||
1451 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1452 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1453 | continue; |
1454 | } | |
fa490cfd | 1455 | |
3a5c359a AK |
1456 | /* |
1457 | * Ahh, all good. It wasn't running, and it wasn't | |
1458 | * runnable, which means that it will never become | |
1459 | * running in the future either. We're all done! | |
1460 | */ | |
1461 | break; | |
1462 | } | |
85ba2d86 RM |
1463 | |
1464 | return ncsw; | |
1da177e4 LT |
1465 | } |
1466 | ||
1467 | /*** | |
1468 | * kick_process - kick a running thread to enter/exit the kernel | |
1469 | * @p: the to-be-kicked thread | |
1470 | * | |
1471 | * Cause a process which is running on another CPU to enter | |
1472 | * kernel-mode, without any delay. (to get signals handled.) | |
1473 | * | |
25985edc | 1474 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1475 | * because all it wants to ensure is that the remote task enters |
1476 | * the kernel. If the IPI races and the task has been migrated | |
1477 | * to another CPU then no harm is done and the purpose has been | |
1478 | * achieved as well. | |
1479 | */ | |
36c8b586 | 1480 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1481 | { |
1482 | int cpu; | |
1483 | ||
1484 | preempt_disable(); | |
1485 | cpu = task_cpu(p); | |
1486 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1487 | smp_send_reschedule(cpu); | |
1488 | preempt_enable(); | |
1489 | } | |
b43e3521 | 1490 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1491 | |
30da688e | 1492 | /* |
013fdb80 | 1493 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1494 | */ |
5da9a0fb PZ |
1495 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1496 | { | |
aa00d89c TC |
1497 | int nid = cpu_to_node(cpu); |
1498 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1499 | enum { cpuset, possible, fail } state = cpuset; |
1500 | int dest_cpu; | |
5da9a0fb | 1501 | |
aa00d89c TC |
1502 | /* |
1503 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1504 | * will return -1. There is no cpu on the node, and we should | |
1505 | * select the cpu on the other node. | |
1506 | */ | |
1507 | if (nid != -1) { | |
1508 | nodemask = cpumask_of_node(nid); | |
1509 | ||
1510 | /* Look for allowed, online CPU in same node. */ | |
1511 | for_each_cpu(dest_cpu, nodemask) { | |
1512 | if (!cpu_online(dest_cpu)) | |
1513 | continue; | |
1514 | if (!cpu_active(dest_cpu)) | |
1515 | continue; | |
1516 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1517 | return dest_cpu; | |
1518 | } | |
2baab4e9 | 1519 | } |
5da9a0fb | 1520 | |
2baab4e9 PZ |
1521 | for (;;) { |
1522 | /* Any allowed, online CPU? */ | |
e3831edd | 1523 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1524 | if (!cpu_online(dest_cpu)) |
1525 | continue; | |
1526 | if (!cpu_active(dest_cpu)) | |
1527 | continue; | |
1528 | goto out; | |
1529 | } | |
5da9a0fb | 1530 | |
2baab4e9 PZ |
1531 | switch (state) { |
1532 | case cpuset: | |
1533 | /* No more Mr. Nice Guy. */ | |
1534 | cpuset_cpus_allowed_fallback(p); | |
1535 | state = possible; | |
1536 | break; | |
1537 | ||
1538 | case possible: | |
1539 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1540 | state = fail; | |
1541 | break; | |
1542 | ||
1543 | case fail: | |
1544 | BUG(); | |
1545 | break; | |
1546 | } | |
1547 | } | |
1548 | ||
1549 | out: | |
1550 | if (state != cpuset) { | |
1551 | /* | |
1552 | * Don't tell them about moving exiting tasks or | |
1553 | * kernel threads (both mm NULL), since they never | |
1554 | * leave kernel. | |
1555 | */ | |
1556 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1557 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1558 | task_pid_nr(p), p->comm, cpu); |
1559 | } | |
5da9a0fb PZ |
1560 | } |
1561 | ||
1562 | return dest_cpu; | |
1563 | } | |
1564 | ||
e2912009 | 1565 | /* |
013fdb80 | 1566 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1567 | */ |
970b13ba | 1568 | static inline |
ac66f547 | 1569 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1570 | { |
cbce1a68 PZ |
1571 | lockdep_assert_held(&p->pi_lock); |
1572 | ||
6c1d9410 WL |
1573 | if (p->nr_cpus_allowed > 1) |
1574 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e2912009 PZ |
1575 | |
1576 | /* | |
1577 | * In order not to call set_task_cpu() on a blocking task we need | |
1578 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1579 | * cpu. | |
1580 | * | |
1581 | * Since this is common to all placement strategies, this lives here. | |
1582 | * | |
1583 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1584 | * not worry about this generic constraint ] | |
1585 | */ | |
fa17b507 | 1586 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1587 | !cpu_online(cpu))) |
5da9a0fb | 1588 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1589 | |
1590 | return cpu; | |
970b13ba | 1591 | } |
09a40af5 MG |
1592 | |
1593 | static void update_avg(u64 *avg, u64 sample) | |
1594 | { | |
1595 | s64 diff = sample - *avg; | |
1596 | *avg += diff >> 3; | |
1597 | } | |
5cc389bc | 1598 | #endif /* CONFIG_SMP */ |
970b13ba | 1599 | |
d7c01d27 | 1600 | static void |
b84cb5df | 1601 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1602 | { |
d7c01d27 | 1603 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1604 | struct rq *rq = this_rq(); |
1605 | ||
d7c01d27 PZ |
1606 | #ifdef CONFIG_SMP |
1607 | int this_cpu = smp_processor_id(); | |
1608 | ||
1609 | if (cpu == this_cpu) { | |
1610 | schedstat_inc(rq, ttwu_local); | |
1611 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1612 | } else { | |
1613 | struct sched_domain *sd; | |
1614 | ||
1615 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1616 | rcu_read_lock(); |
d7c01d27 PZ |
1617 | for_each_domain(this_cpu, sd) { |
1618 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1619 | schedstat_inc(sd, ttwu_wake_remote); | |
1620 | break; | |
1621 | } | |
1622 | } | |
057f3fad | 1623 | rcu_read_unlock(); |
d7c01d27 | 1624 | } |
f339b9dc PZ |
1625 | |
1626 | if (wake_flags & WF_MIGRATED) | |
1627 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1628 | ||
d7c01d27 PZ |
1629 | #endif /* CONFIG_SMP */ |
1630 | ||
1631 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1632 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1633 | |
1634 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1635 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1636 | |
d7c01d27 PZ |
1637 | #endif /* CONFIG_SCHEDSTATS */ |
1638 | } | |
1639 | ||
1640 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1641 | { | |
9ed3811a | 1642 | activate_task(rq, p, en_flags); |
da0c1e65 | 1643 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1644 | |
1645 | /* if a worker is waking up, notify workqueue */ | |
1646 | if (p->flags & PF_WQ_WORKER) | |
1647 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1648 | } |
1649 | ||
23f41eeb PZ |
1650 | /* |
1651 | * Mark the task runnable and perform wakeup-preemption. | |
1652 | */ | |
89363381 | 1653 | static void |
23f41eeb | 1654 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1655 | { |
9ed3811a | 1656 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1657 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1658 | |
1659 | p->state = TASK_RUNNING; | |
1660 | #ifdef CONFIG_SMP | |
4c9a4bc8 PZ |
1661 | if (p->sched_class->task_woken) { |
1662 | /* | |
cbce1a68 PZ |
1663 | * Our task @p is fully woken up and running; so its safe to |
1664 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1665 | */ |
cbce1a68 | 1666 | lockdep_unpin_lock(&rq->lock); |
9ed3811a | 1667 | p->sched_class->task_woken(rq, p); |
cbce1a68 | 1668 | lockdep_pin_lock(&rq->lock); |
4c9a4bc8 | 1669 | } |
9ed3811a | 1670 | |
e69c6341 | 1671 | if (rq->idle_stamp) { |
78becc27 | 1672 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1673 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1674 | |
abfafa54 JL |
1675 | update_avg(&rq->avg_idle, delta); |
1676 | ||
1677 | if (rq->avg_idle > max) | |
9ed3811a | 1678 | rq->avg_idle = max; |
abfafa54 | 1679 | |
9ed3811a TH |
1680 | rq->idle_stamp = 0; |
1681 | } | |
1682 | #endif | |
1683 | } | |
1684 | ||
c05fbafb PZ |
1685 | static void |
1686 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1687 | { | |
cbce1a68 PZ |
1688 | lockdep_assert_held(&rq->lock); |
1689 | ||
c05fbafb PZ |
1690 | #ifdef CONFIG_SMP |
1691 | if (p->sched_contributes_to_load) | |
1692 | rq->nr_uninterruptible--; | |
1693 | #endif | |
1694 | ||
1695 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1696 | ttwu_do_wakeup(rq, p, wake_flags); | |
1697 | } | |
1698 | ||
1699 | /* | |
1700 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1701 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1702 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1703 | * the task is still ->on_rq. | |
1704 | */ | |
1705 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1706 | { | |
1707 | struct rq *rq; | |
1708 | int ret = 0; | |
1709 | ||
1710 | rq = __task_rq_lock(p); | |
da0c1e65 | 1711 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1712 | /* check_preempt_curr() may use rq clock */ |
1713 | update_rq_clock(rq); | |
c05fbafb PZ |
1714 | ttwu_do_wakeup(rq, p, wake_flags); |
1715 | ret = 1; | |
1716 | } | |
1717 | __task_rq_unlock(rq); | |
1718 | ||
1719 | return ret; | |
1720 | } | |
1721 | ||
317f3941 | 1722 | #ifdef CONFIG_SMP |
e3baac47 | 1723 | void sched_ttwu_pending(void) |
317f3941 PZ |
1724 | { |
1725 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1726 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1727 | struct task_struct *p; | |
e3baac47 | 1728 | unsigned long flags; |
317f3941 | 1729 | |
e3baac47 PZ |
1730 | if (!llist) |
1731 | return; | |
1732 | ||
1733 | raw_spin_lock_irqsave(&rq->lock, flags); | |
cbce1a68 | 1734 | lockdep_pin_lock(&rq->lock); |
317f3941 | 1735 | |
fa14ff4a PZ |
1736 | while (llist) { |
1737 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1738 | llist = llist_next(llist); | |
317f3941 PZ |
1739 | ttwu_do_activate(rq, p, 0); |
1740 | } | |
1741 | ||
cbce1a68 | 1742 | lockdep_unpin_lock(&rq->lock); |
e3baac47 | 1743 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1744 | } |
1745 | ||
1746 | void scheduler_ipi(void) | |
1747 | { | |
f27dde8d PZ |
1748 | /* |
1749 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1750 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1751 | * this IPI. | |
1752 | */ | |
8cb75e0c | 1753 | preempt_fold_need_resched(); |
f27dde8d | 1754 | |
fd2ac4f4 | 1755 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1756 | return; |
1757 | ||
1758 | /* | |
1759 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1760 | * traditionally all their work was done from the interrupt return | |
1761 | * path. Now that we actually do some work, we need to make sure | |
1762 | * we do call them. | |
1763 | * | |
1764 | * Some archs already do call them, luckily irq_enter/exit nest | |
1765 | * properly. | |
1766 | * | |
1767 | * Arguably we should visit all archs and update all handlers, | |
1768 | * however a fair share of IPIs are still resched only so this would | |
1769 | * somewhat pessimize the simple resched case. | |
1770 | */ | |
1771 | irq_enter(); | |
fa14ff4a | 1772 | sched_ttwu_pending(); |
ca38062e SS |
1773 | |
1774 | /* | |
1775 | * Check if someone kicked us for doing the nohz idle load balance. | |
1776 | */ | |
873b4c65 | 1777 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1778 | this_rq()->idle_balance = 1; |
ca38062e | 1779 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1780 | } |
c5d753a5 | 1781 | irq_exit(); |
317f3941 PZ |
1782 | } |
1783 | ||
1784 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1785 | { | |
e3baac47 PZ |
1786 | struct rq *rq = cpu_rq(cpu); |
1787 | ||
1788 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1789 | if (!set_nr_if_polling(rq->idle)) | |
1790 | smp_send_reschedule(cpu); | |
1791 | else | |
1792 | trace_sched_wake_idle_without_ipi(cpu); | |
1793 | } | |
317f3941 | 1794 | } |
d6aa8f85 | 1795 | |
f6be8af1 CL |
1796 | void wake_up_if_idle(int cpu) |
1797 | { | |
1798 | struct rq *rq = cpu_rq(cpu); | |
1799 | unsigned long flags; | |
1800 | ||
fd7de1e8 AL |
1801 | rcu_read_lock(); |
1802 | ||
1803 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1804 | goto out; | |
f6be8af1 CL |
1805 | |
1806 | if (set_nr_if_polling(rq->idle)) { | |
1807 | trace_sched_wake_idle_without_ipi(cpu); | |
1808 | } else { | |
1809 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1810 | if (is_idle_task(rq->curr)) | |
1811 | smp_send_reschedule(cpu); | |
1812 | /* Else cpu is not in idle, do nothing here */ | |
1813 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1814 | } | |
fd7de1e8 AL |
1815 | |
1816 | out: | |
1817 | rcu_read_unlock(); | |
f6be8af1 CL |
1818 | } |
1819 | ||
39be3501 | 1820 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1821 | { |
1822 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1823 | } | |
d6aa8f85 | 1824 | #endif /* CONFIG_SMP */ |
317f3941 | 1825 | |
c05fbafb PZ |
1826 | static void ttwu_queue(struct task_struct *p, int cpu) |
1827 | { | |
1828 | struct rq *rq = cpu_rq(cpu); | |
1829 | ||
17d9f311 | 1830 | #if defined(CONFIG_SMP) |
39be3501 | 1831 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1832 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1833 | ttwu_queue_remote(p, cpu); |
1834 | return; | |
1835 | } | |
1836 | #endif | |
1837 | ||
c05fbafb | 1838 | raw_spin_lock(&rq->lock); |
cbce1a68 | 1839 | lockdep_pin_lock(&rq->lock); |
c05fbafb | 1840 | ttwu_do_activate(rq, p, 0); |
cbce1a68 | 1841 | lockdep_unpin_lock(&rq->lock); |
c05fbafb | 1842 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1843 | } |
1844 | ||
1845 | /** | |
1da177e4 | 1846 | * try_to_wake_up - wake up a thread |
9ed3811a | 1847 | * @p: the thread to be awakened |
1da177e4 | 1848 | * @state: the mask of task states that can be woken |
9ed3811a | 1849 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1850 | * |
1851 | * Put it on the run-queue if it's not already there. The "current" | |
1852 | * thread is always on the run-queue (except when the actual | |
1853 | * re-schedule is in progress), and as such you're allowed to do | |
1854 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1855 | * runnable without the overhead of this. | |
1856 | * | |
e69f6186 | 1857 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1858 | * or @state didn't match @p's state. |
1da177e4 | 1859 | */ |
e4a52bcb PZ |
1860 | static int |
1861 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1862 | { |
1da177e4 | 1863 | unsigned long flags; |
c05fbafb | 1864 | int cpu, success = 0; |
2398f2c6 | 1865 | |
e0acd0a6 ON |
1866 | /* |
1867 | * If we are going to wake up a thread waiting for CONDITION we | |
1868 | * need to ensure that CONDITION=1 done by the caller can not be | |
1869 | * reordered with p->state check below. This pairs with mb() in | |
1870 | * set_current_state() the waiting thread does. | |
1871 | */ | |
1872 | smp_mb__before_spinlock(); | |
013fdb80 | 1873 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1874 | if (!(p->state & state)) |
1da177e4 LT |
1875 | goto out; |
1876 | ||
c05fbafb | 1877 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1878 | cpu = task_cpu(p); |
1da177e4 | 1879 | |
c05fbafb PZ |
1880 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1881 | goto stat; | |
1da177e4 | 1882 | |
1da177e4 | 1883 | #ifdef CONFIG_SMP |
e9c84311 | 1884 | /* |
c05fbafb PZ |
1885 | * If the owning (remote) cpu is still in the middle of schedule() with |
1886 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1887 | */ |
f3e94786 | 1888 | while (p->on_cpu) |
e4a52bcb | 1889 | cpu_relax(); |
0970d299 | 1890 | /* |
e4a52bcb | 1891 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1892 | */ |
e4a52bcb | 1893 | smp_rmb(); |
1da177e4 | 1894 | |
a8e4f2ea | 1895 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1896 | p->state = TASK_WAKING; |
e7693a36 | 1897 | |
e4a52bcb | 1898 | if (p->sched_class->task_waking) |
74f8e4b2 | 1899 | p->sched_class->task_waking(p); |
efbbd05a | 1900 | |
ac66f547 | 1901 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1902 | if (task_cpu(p) != cpu) { |
1903 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1904 | set_task_cpu(p, cpu); |
f339b9dc | 1905 | } |
1da177e4 | 1906 | #endif /* CONFIG_SMP */ |
1da177e4 | 1907 | |
c05fbafb PZ |
1908 | ttwu_queue(p, cpu); |
1909 | stat: | |
b84cb5df | 1910 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1911 | out: |
013fdb80 | 1912 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1913 | |
1914 | return success; | |
1915 | } | |
1916 | ||
21aa9af0 TH |
1917 | /** |
1918 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1919 | * @p: the thread to be awakened | |
1920 | * | |
2acca55e | 1921 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1922 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1923 | * the current task. |
21aa9af0 TH |
1924 | */ |
1925 | static void try_to_wake_up_local(struct task_struct *p) | |
1926 | { | |
1927 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1928 | |
383efcd0 TH |
1929 | if (WARN_ON_ONCE(rq != this_rq()) || |
1930 | WARN_ON_ONCE(p == current)) | |
1931 | return; | |
1932 | ||
21aa9af0 TH |
1933 | lockdep_assert_held(&rq->lock); |
1934 | ||
2acca55e | 1935 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
1936 | /* |
1937 | * This is OK, because current is on_cpu, which avoids it being | |
1938 | * picked for load-balance and preemption/IRQs are still | |
1939 | * disabled avoiding further scheduler activity on it and we've | |
1940 | * not yet picked a replacement task. | |
1941 | */ | |
1942 | lockdep_unpin_lock(&rq->lock); | |
2acca55e PZ |
1943 | raw_spin_unlock(&rq->lock); |
1944 | raw_spin_lock(&p->pi_lock); | |
1945 | raw_spin_lock(&rq->lock); | |
cbce1a68 | 1946 | lockdep_pin_lock(&rq->lock); |
2acca55e PZ |
1947 | } |
1948 | ||
21aa9af0 | 1949 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1950 | goto out; |
21aa9af0 | 1951 | |
da0c1e65 | 1952 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
1953 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1954 | ||
23f41eeb | 1955 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1956 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1957 | out: |
1958 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1959 | } |
1960 | ||
50fa610a DH |
1961 | /** |
1962 | * wake_up_process - Wake up a specific process | |
1963 | * @p: The process to be woken up. | |
1964 | * | |
1965 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1966 | * processes. |
1967 | * | |
1968 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1969 | * |
1970 | * It may be assumed that this function implies a write memory barrier before | |
1971 | * changing the task state if and only if any tasks are woken up. | |
1972 | */ | |
7ad5b3a5 | 1973 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1974 | { |
9067ac85 ON |
1975 | WARN_ON(task_is_stopped_or_traced(p)); |
1976 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1977 | } |
1da177e4 LT |
1978 | EXPORT_SYMBOL(wake_up_process); |
1979 | ||
7ad5b3a5 | 1980 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1981 | { |
1982 | return try_to_wake_up(p, state, 0); | |
1983 | } | |
1984 | ||
a5e7be3b JL |
1985 | /* |
1986 | * This function clears the sched_dl_entity static params. | |
1987 | */ | |
1988 | void __dl_clear_params(struct task_struct *p) | |
1989 | { | |
1990 | struct sched_dl_entity *dl_se = &p->dl; | |
1991 | ||
1992 | dl_se->dl_runtime = 0; | |
1993 | dl_se->dl_deadline = 0; | |
1994 | dl_se->dl_period = 0; | |
1995 | dl_se->flags = 0; | |
1996 | dl_se->dl_bw = 0; | |
40767b0d PZ |
1997 | |
1998 | dl_se->dl_throttled = 0; | |
1999 | dl_se->dl_new = 1; | |
2000 | dl_se->dl_yielded = 0; | |
a5e7be3b JL |
2001 | } |
2002 | ||
1da177e4 LT |
2003 | /* |
2004 | * Perform scheduler related setup for a newly forked process p. | |
2005 | * p is forked by current. | |
dd41f596 IM |
2006 | * |
2007 | * __sched_fork() is basic setup used by init_idle() too: | |
2008 | */ | |
5e1576ed | 2009 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2010 | { |
fd2f4419 PZ |
2011 | p->on_rq = 0; |
2012 | ||
2013 | p->se.on_rq = 0; | |
dd41f596 IM |
2014 | p->se.exec_start = 0; |
2015 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2016 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2017 | p->se.nr_migrations = 0; |
da7a735e | 2018 | p->se.vruntime = 0; |
bb04159d KT |
2019 | #ifdef CONFIG_SMP |
2020 | p->se.avg.decay_count = 0; | |
2021 | #endif | |
fd2f4419 | 2022 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2023 | |
2024 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2025 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2026 | #endif |
476d139c | 2027 | |
aab03e05 | 2028 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2029 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2030 | __dl_clear_params(p); |
aab03e05 | 2031 | |
fa717060 | 2032 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2033 | |
e107be36 AK |
2034 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2035 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2036 | #endif | |
cbee9f88 PZ |
2037 | |
2038 | #ifdef CONFIG_NUMA_BALANCING | |
2039 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2040 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2041 | p->mm->numa_scan_seq = 0; |
2042 | } | |
2043 | ||
5e1576ed RR |
2044 | if (clone_flags & CLONE_VM) |
2045 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2046 | else | |
2047 | p->numa_preferred_nid = -1; | |
2048 | ||
cbee9f88 PZ |
2049 | p->node_stamp = 0ULL; |
2050 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2051 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2052 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2053 | p->numa_faults = NULL; |
7e2703e6 RR |
2054 | p->last_task_numa_placement = 0; |
2055 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2056 | |
8c8a743c | 2057 | p->numa_group = NULL; |
cbee9f88 | 2058 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2059 | } |
2060 | ||
1a687c2e | 2061 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 2062 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
2063 | void set_numabalancing_state(bool enabled) |
2064 | { | |
2065 | if (enabled) | |
2066 | sched_feat_set("NUMA"); | |
2067 | else | |
2068 | sched_feat_set("NO_NUMA"); | |
2069 | } | |
3105b86a MG |
2070 | #else |
2071 | __read_mostly bool numabalancing_enabled; | |
2072 | ||
2073 | void set_numabalancing_state(bool enabled) | |
2074 | { | |
2075 | numabalancing_enabled = enabled; | |
dd41f596 | 2076 | } |
3105b86a | 2077 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
2078 | |
2079 | #ifdef CONFIG_PROC_SYSCTL | |
2080 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2081 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2082 | { | |
2083 | struct ctl_table t; | |
2084 | int err; | |
2085 | int state = numabalancing_enabled; | |
2086 | ||
2087 | if (write && !capable(CAP_SYS_ADMIN)) | |
2088 | return -EPERM; | |
2089 | ||
2090 | t = *table; | |
2091 | t.data = &state; | |
2092 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2093 | if (err < 0) | |
2094 | return err; | |
2095 | if (write) | |
2096 | set_numabalancing_state(state); | |
2097 | return err; | |
2098 | } | |
2099 | #endif | |
2100 | #endif | |
dd41f596 IM |
2101 | |
2102 | /* | |
2103 | * fork()/clone()-time setup: | |
2104 | */ | |
aab03e05 | 2105 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2106 | { |
0122ec5b | 2107 | unsigned long flags; |
dd41f596 IM |
2108 | int cpu = get_cpu(); |
2109 | ||
5e1576ed | 2110 | __sched_fork(clone_flags, p); |
06b83b5f | 2111 | /* |
0017d735 | 2112 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2113 | * nobody will actually run it, and a signal or other external |
2114 | * event cannot wake it up and insert it on the runqueue either. | |
2115 | */ | |
0017d735 | 2116 | p->state = TASK_RUNNING; |
dd41f596 | 2117 | |
c350a04e MG |
2118 | /* |
2119 | * Make sure we do not leak PI boosting priority to the child. | |
2120 | */ | |
2121 | p->prio = current->normal_prio; | |
2122 | ||
b9dc29e7 MG |
2123 | /* |
2124 | * Revert to default priority/policy on fork if requested. | |
2125 | */ | |
2126 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2127 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2128 | p->policy = SCHED_NORMAL; |
6c697bdf | 2129 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2130 | p->rt_priority = 0; |
2131 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2132 | p->static_prio = NICE_TO_PRIO(0); | |
2133 | ||
2134 | p->prio = p->normal_prio = __normal_prio(p); | |
2135 | set_load_weight(p); | |
6c697bdf | 2136 | |
b9dc29e7 MG |
2137 | /* |
2138 | * We don't need the reset flag anymore after the fork. It has | |
2139 | * fulfilled its duty: | |
2140 | */ | |
2141 | p->sched_reset_on_fork = 0; | |
2142 | } | |
ca94c442 | 2143 | |
aab03e05 DF |
2144 | if (dl_prio(p->prio)) { |
2145 | put_cpu(); | |
2146 | return -EAGAIN; | |
2147 | } else if (rt_prio(p->prio)) { | |
2148 | p->sched_class = &rt_sched_class; | |
2149 | } else { | |
2ddbf952 | 2150 | p->sched_class = &fair_sched_class; |
aab03e05 | 2151 | } |
b29739f9 | 2152 | |
cd29fe6f PZ |
2153 | if (p->sched_class->task_fork) |
2154 | p->sched_class->task_fork(p); | |
2155 | ||
86951599 PZ |
2156 | /* |
2157 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2158 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2159 | * is ran before sched_fork(). | |
2160 | * | |
2161 | * Silence PROVE_RCU. | |
2162 | */ | |
0122ec5b | 2163 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2164 | set_task_cpu(p, cpu); |
0122ec5b | 2165 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2166 | |
52f17b6c | 2167 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2168 | if (likely(sched_info_on())) |
52f17b6c | 2169 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2170 | #endif |
3ca7a440 PZ |
2171 | #if defined(CONFIG_SMP) |
2172 | p->on_cpu = 0; | |
4866cde0 | 2173 | #endif |
01028747 | 2174 | init_task_preempt_count(p); |
806c09a7 | 2175 | #ifdef CONFIG_SMP |
917b627d | 2176 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2177 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2178 | #endif |
917b627d | 2179 | |
476d139c | 2180 | put_cpu(); |
aab03e05 | 2181 | return 0; |
1da177e4 LT |
2182 | } |
2183 | ||
332ac17e DF |
2184 | unsigned long to_ratio(u64 period, u64 runtime) |
2185 | { | |
2186 | if (runtime == RUNTIME_INF) | |
2187 | return 1ULL << 20; | |
2188 | ||
2189 | /* | |
2190 | * Doing this here saves a lot of checks in all | |
2191 | * the calling paths, and returning zero seems | |
2192 | * safe for them anyway. | |
2193 | */ | |
2194 | if (period == 0) | |
2195 | return 0; | |
2196 | ||
2197 | return div64_u64(runtime << 20, period); | |
2198 | } | |
2199 | ||
2200 | #ifdef CONFIG_SMP | |
2201 | inline struct dl_bw *dl_bw_of(int i) | |
2202 | { | |
66339c31 KT |
2203 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2204 | "sched RCU must be held"); | |
332ac17e DF |
2205 | return &cpu_rq(i)->rd->dl_bw; |
2206 | } | |
2207 | ||
de212f18 | 2208 | static inline int dl_bw_cpus(int i) |
332ac17e | 2209 | { |
de212f18 PZ |
2210 | struct root_domain *rd = cpu_rq(i)->rd; |
2211 | int cpus = 0; | |
2212 | ||
66339c31 KT |
2213 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2214 | "sched RCU must be held"); | |
de212f18 PZ |
2215 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2216 | cpus++; | |
2217 | ||
2218 | return cpus; | |
332ac17e DF |
2219 | } |
2220 | #else | |
2221 | inline struct dl_bw *dl_bw_of(int i) | |
2222 | { | |
2223 | return &cpu_rq(i)->dl.dl_bw; | |
2224 | } | |
2225 | ||
de212f18 | 2226 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2227 | { |
2228 | return 1; | |
2229 | } | |
2230 | #endif | |
2231 | ||
332ac17e DF |
2232 | /* |
2233 | * We must be sure that accepting a new task (or allowing changing the | |
2234 | * parameters of an existing one) is consistent with the bandwidth | |
2235 | * constraints. If yes, this function also accordingly updates the currently | |
2236 | * allocated bandwidth to reflect the new situation. | |
2237 | * | |
2238 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2239 | * |
2240 | * XXX we should delay bw change until the task's 0-lag point, see | |
2241 | * __setparam_dl(). | |
332ac17e DF |
2242 | */ |
2243 | static int dl_overflow(struct task_struct *p, int policy, | |
2244 | const struct sched_attr *attr) | |
2245 | { | |
2246 | ||
2247 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2248 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2249 | u64 runtime = attr->sched_runtime; |
2250 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2251 | int cpus, err = -1; |
332ac17e DF |
2252 | |
2253 | if (new_bw == p->dl.dl_bw) | |
2254 | return 0; | |
2255 | ||
2256 | /* | |
2257 | * Either if a task, enters, leave, or stays -deadline but changes | |
2258 | * its parameters, we may need to update accordingly the total | |
2259 | * allocated bandwidth of the container. | |
2260 | */ | |
2261 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2262 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2263 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2264 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2265 | __dl_add(dl_b, new_bw); | |
2266 | err = 0; | |
2267 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2268 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2269 | __dl_clear(dl_b, p->dl.dl_bw); | |
2270 | __dl_add(dl_b, new_bw); | |
2271 | err = 0; | |
2272 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2273 | __dl_clear(dl_b, p->dl.dl_bw); | |
2274 | err = 0; | |
2275 | } | |
2276 | raw_spin_unlock(&dl_b->lock); | |
2277 | ||
2278 | return err; | |
2279 | } | |
2280 | ||
2281 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2282 | ||
1da177e4 LT |
2283 | /* |
2284 | * wake_up_new_task - wake up a newly created task for the first time. | |
2285 | * | |
2286 | * This function will do some initial scheduler statistics housekeeping | |
2287 | * that must be done for every newly created context, then puts the task | |
2288 | * on the runqueue and wakes it. | |
2289 | */ | |
3e51e3ed | 2290 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2291 | { |
2292 | unsigned long flags; | |
dd41f596 | 2293 | struct rq *rq; |
fabf318e | 2294 | |
ab2515c4 | 2295 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2296 | #ifdef CONFIG_SMP |
2297 | /* | |
2298 | * Fork balancing, do it here and not earlier because: | |
2299 | * - cpus_allowed can change in the fork path | |
2300 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2301 | */ |
ac66f547 | 2302 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2303 | #endif |
2304 | ||
a75cdaa9 AS |
2305 | /* Initialize new task's runnable average */ |
2306 | init_task_runnable_average(p); | |
ab2515c4 | 2307 | rq = __task_rq_lock(p); |
cd29fe6f | 2308 | activate_task(rq, p, 0); |
da0c1e65 | 2309 | p->on_rq = TASK_ON_RQ_QUEUED; |
89363381 | 2310 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2311 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2312 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2313 | if (p->sched_class->task_woken) |
2314 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2315 | #endif |
0122ec5b | 2316 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2317 | } |
2318 | ||
e107be36 AK |
2319 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2320 | ||
1cde2930 PZ |
2321 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2322 | ||
e107be36 | 2323 | /** |
80dd99b3 | 2324 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2325 | * @notifier: notifier struct to register |
e107be36 AK |
2326 | */ |
2327 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2328 | { | |
1cde2930 | 2329 | static_key_slow_inc(&preempt_notifier_key); |
e107be36 AK |
2330 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2331 | } | |
2332 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2333 | ||
2334 | /** | |
2335 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2336 | * @notifier: notifier struct to unregister |
e107be36 | 2337 | * |
d84525a8 | 2338 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2339 | */ |
2340 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2341 | { | |
2342 | hlist_del(¬ifier->link); | |
1cde2930 | 2343 | static_key_slow_dec(&preempt_notifier_key); |
e107be36 AK |
2344 | } |
2345 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2346 | ||
1cde2930 | 2347 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2348 | { |
2349 | struct preempt_notifier *notifier; | |
e107be36 | 2350 | |
b67bfe0d | 2351 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2352 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2353 | } | |
2354 | ||
1cde2930 PZ |
2355 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2356 | { | |
2357 | if (static_key_false(&preempt_notifier_key)) | |
2358 | __fire_sched_in_preempt_notifiers(curr); | |
2359 | } | |
2360 | ||
e107be36 | 2361 | static void |
1cde2930 PZ |
2362 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2363 | struct task_struct *next) | |
e107be36 AK |
2364 | { |
2365 | struct preempt_notifier *notifier; | |
e107be36 | 2366 | |
b67bfe0d | 2367 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2368 | notifier->ops->sched_out(notifier, next); |
2369 | } | |
2370 | ||
1cde2930 PZ |
2371 | static __always_inline void |
2372 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2373 | struct task_struct *next) | |
2374 | { | |
2375 | if (static_key_false(&preempt_notifier_key)) | |
2376 | __fire_sched_out_preempt_notifiers(curr, next); | |
2377 | } | |
2378 | ||
6d6bc0ad | 2379 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2380 | |
1cde2930 | 2381 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2382 | { |
2383 | } | |
2384 | ||
1cde2930 | 2385 | static inline void |
e107be36 AK |
2386 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2387 | struct task_struct *next) | |
2388 | { | |
2389 | } | |
2390 | ||
6d6bc0ad | 2391 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2392 | |
4866cde0 NP |
2393 | /** |
2394 | * prepare_task_switch - prepare to switch tasks | |
2395 | * @rq: the runqueue preparing to switch | |
421cee29 | 2396 | * @prev: the current task that is being switched out |
4866cde0 NP |
2397 | * @next: the task we are going to switch to. |
2398 | * | |
2399 | * This is called with the rq lock held and interrupts off. It must | |
2400 | * be paired with a subsequent finish_task_switch after the context | |
2401 | * switch. | |
2402 | * | |
2403 | * prepare_task_switch sets up locking and calls architecture specific | |
2404 | * hooks. | |
2405 | */ | |
e107be36 AK |
2406 | static inline void |
2407 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2408 | struct task_struct *next) | |
4866cde0 | 2409 | { |
895dd92c | 2410 | trace_sched_switch(prev, next); |
43148951 | 2411 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2412 | perf_event_task_sched_out(prev, next); |
e107be36 | 2413 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2414 | prepare_lock_switch(rq, next); |
2415 | prepare_arch_switch(next); | |
2416 | } | |
2417 | ||
1da177e4 LT |
2418 | /** |
2419 | * finish_task_switch - clean up after a task-switch | |
2420 | * @prev: the thread we just switched away from. | |
2421 | * | |
4866cde0 NP |
2422 | * finish_task_switch must be called after the context switch, paired |
2423 | * with a prepare_task_switch call before the context switch. | |
2424 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2425 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2426 | * |
2427 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2428 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2429 | * with the lock held can cause deadlocks; see schedule() for |
2430 | * details.) | |
dfa50b60 ON |
2431 | * |
2432 | * The context switch have flipped the stack from under us and restored the | |
2433 | * local variables which were saved when this task called schedule() in the | |
2434 | * past. prev == current is still correct but we need to recalculate this_rq | |
2435 | * because prev may have moved to another CPU. | |
1da177e4 | 2436 | */ |
dfa50b60 | 2437 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2438 | __releases(rq->lock) |
2439 | { | |
dfa50b60 | 2440 | struct rq *rq = this_rq(); |
1da177e4 | 2441 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2442 | long prev_state; |
1da177e4 LT |
2443 | |
2444 | rq->prev_mm = NULL; | |
2445 | ||
2446 | /* | |
2447 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2448 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2449 | * schedule one last time. The schedule call will never return, and |
2450 | * the scheduled task must drop that reference. | |
c394cc9f | 2451 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2452 | * still held, otherwise prev could be scheduled on another cpu, die |
2453 | * there before we look at prev->state, and then the reference would | |
2454 | * be dropped twice. | |
2455 | * Manfred Spraul <manfred@colorfullife.com> | |
2456 | */ | |
55a101f8 | 2457 | prev_state = prev->state; |
bf9fae9f | 2458 | vtime_task_switch(prev); |
4866cde0 | 2459 | finish_arch_switch(prev); |
a8d757ef | 2460 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2461 | finish_lock_switch(rq, prev); |
01f23e16 | 2462 | finish_arch_post_lock_switch(); |
e8fa1362 | 2463 | |
e107be36 | 2464 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2465 | if (mm) |
2466 | mmdrop(mm); | |
c394cc9f | 2467 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2468 | if (prev->sched_class->task_dead) |
2469 | prev->sched_class->task_dead(prev); | |
2470 | ||
c6fd91f0 | 2471 | /* |
2472 | * Remove function-return probe instances associated with this | |
2473 | * task and put them back on the free list. | |
9761eea8 | 2474 | */ |
c6fd91f0 | 2475 | kprobe_flush_task(prev); |
1da177e4 | 2476 | put_task_struct(prev); |
c6fd91f0 | 2477 | } |
99e5ada9 FW |
2478 | |
2479 | tick_nohz_task_switch(current); | |
dfa50b60 | 2480 | return rq; |
1da177e4 LT |
2481 | } |
2482 | ||
3f029d3c GH |
2483 | #ifdef CONFIG_SMP |
2484 | ||
3f029d3c | 2485 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2486 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2487 | { |
e3fca9e7 PZ |
2488 | struct callback_head *head, *next; |
2489 | void (*func)(struct rq *rq); | |
2490 | unsigned long flags; | |
3f029d3c | 2491 | |
e3fca9e7 PZ |
2492 | raw_spin_lock_irqsave(&rq->lock, flags); |
2493 | head = rq->balance_callback; | |
2494 | rq->balance_callback = NULL; | |
2495 | while (head) { | |
2496 | func = (void (*)(struct rq *))head->func; | |
2497 | next = head->next; | |
2498 | head->next = NULL; | |
2499 | head = next; | |
3f029d3c | 2500 | |
e3fca9e7 | 2501 | func(rq); |
3f029d3c | 2502 | } |
e3fca9e7 PZ |
2503 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2504 | } | |
2505 | ||
2506 | static inline void balance_callback(struct rq *rq) | |
2507 | { | |
2508 | if (unlikely(rq->balance_callback)) | |
2509 | __balance_callback(rq); | |
3f029d3c GH |
2510 | } |
2511 | ||
2512 | #else | |
da19ab51 | 2513 | |
e3fca9e7 | 2514 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2515 | { |
1da177e4 LT |
2516 | } |
2517 | ||
3f029d3c GH |
2518 | #endif |
2519 | ||
1da177e4 LT |
2520 | /** |
2521 | * schedule_tail - first thing a freshly forked thread must call. | |
2522 | * @prev: the thread we just switched away from. | |
2523 | */ | |
722a9f92 | 2524 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2525 | __releases(rq->lock) |
2526 | { | |
1a43a14a | 2527 | struct rq *rq; |
da19ab51 | 2528 | |
1a43a14a ON |
2529 | /* finish_task_switch() drops rq->lock and enables preemtion */ |
2530 | preempt_disable(); | |
dfa50b60 | 2531 | rq = finish_task_switch(prev); |
e3fca9e7 | 2532 | balance_callback(rq); |
1a43a14a | 2533 | preempt_enable(); |
70b97a7f | 2534 | |
1da177e4 | 2535 | if (current->set_child_tid) |
b488893a | 2536 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2537 | } |
2538 | ||
2539 | /* | |
dfa50b60 | 2540 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2541 | */ |
dfa50b60 | 2542 | static inline struct rq * |
70b97a7f | 2543 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2544 | struct task_struct *next) |
1da177e4 | 2545 | { |
dd41f596 | 2546 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2547 | |
e107be36 | 2548 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2549 | |
dd41f596 IM |
2550 | mm = next->mm; |
2551 | oldmm = prev->active_mm; | |
9226d125 ZA |
2552 | /* |
2553 | * For paravirt, this is coupled with an exit in switch_to to | |
2554 | * combine the page table reload and the switch backend into | |
2555 | * one hypercall. | |
2556 | */ | |
224101ed | 2557 | arch_start_context_switch(prev); |
9226d125 | 2558 | |
31915ab4 | 2559 | if (!mm) { |
1da177e4 LT |
2560 | next->active_mm = oldmm; |
2561 | atomic_inc(&oldmm->mm_count); | |
2562 | enter_lazy_tlb(oldmm, next); | |
2563 | } else | |
2564 | switch_mm(oldmm, mm, next); | |
2565 | ||
31915ab4 | 2566 | if (!prev->mm) { |
1da177e4 | 2567 | prev->active_mm = NULL; |
1da177e4 LT |
2568 | rq->prev_mm = oldmm; |
2569 | } | |
3a5f5e48 IM |
2570 | /* |
2571 | * Since the runqueue lock will be released by the next | |
2572 | * task (which is an invalid locking op but in the case | |
2573 | * of the scheduler it's an obvious special-case), so we | |
2574 | * do an early lockdep release here: | |
2575 | */ | |
cbce1a68 | 2576 | lockdep_unpin_lock(&rq->lock); |
8a25d5de | 2577 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2578 | |
2579 | /* Here we just switch the register state and the stack. */ | |
2580 | switch_to(prev, next, prev); | |
dd41f596 | 2581 | barrier(); |
dfa50b60 ON |
2582 | |
2583 | return finish_task_switch(prev); | |
1da177e4 LT |
2584 | } |
2585 | ||
2586 | /* | |
1c3e8264 | 2587 | * nr_running and nr_context_switches: |
1da177e4 LT |
2588 | * |
2589 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2590 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2591 | */ |
2592 | unsigned long nr_running(void) | |
2593 | { | |
2594 | unsigned long i, sum = 0; | |
2595 | ||
2596 | for_each_online_cpu(i) | |
2597 | sum += cpu_rq(i)->nr_running; | |
2598 | ||
2599 | return sum; | |
f711f609 | 2600 | } |
1da177e4 | 2601 | |
2ee507c4 TC |
2602 | /* |
2603 | * Check if only the current task is running on the cpu. | |
2604 | */ | |
2605 | bool single_task_running(void) | |
2606 | { | |
2607 | if (cpu_rq(smp_processor_id())->nr_running == 1) | |
2608 | return true; | |
2609 | else | |
2610 | return false; | |
2611 | } | |
2612 | EXPORT_SYMBOL(single_task_running); | |
2613 | ||
1da177e4 | 2614 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2615 | { |
cc94abfc SR |
2616 | int i; |
2617 | unsigned long long sum = 0; | |
46cb4b7c | 2618 | |
0a945022 | 2619 | for_each_possible_cpu(i) |
1da177e4 | 2620 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2621 | |
1da177e4 LT |
2622 | return sum; |
2623 | } | |
483b4ee6 | 2624 | |
1da177e4 LT |
2625 | unsigned long nr_iowait(void) |
2626 | { | |
2627 | unsigned long i, sum = 0; | |
483b4ee6 | 2628 | |
0a945022 | 2629 | for_each_possible_cpu(i) |
1da177e4 | 2630 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2631 | |
1da177e4 LT |
2632 | return sum; |
2633 | } | |
483b4ee6 | 2634 | |
8c215bd3 | 2635 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2636 | { |
8c215bd3 | 2637 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2638 | return atomic_read(&this->nr_iowait); |
2639 | } | |
46cb4b7c | 2640 | |
372ba8cb MG |
2641 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2642 | { | |
3289bdb4 PZ |
2643 | struct rq *rq = this_rq(); |
2644 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2645 | *load = rq->load.weight; | |
372ba8cb MG |
2646 | } |
2647 | ||
dd41f596 | 2648 | #ifdef CONFIG_SMP |
8a0be9ef | 2649 | |
46cb4b7c | 2650 | /* |
38022906 PZ |
2651 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2652 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2653 | */ |
38022906 | 2654 | void sched_exec(void) |
46cb4b7c | 2655 | { |
38022906 | 2656 | struct task_struct *p = current; |
1da177e4 | 2657 | unsigned long flags; |
0017d735 | 2658 | int dest_cpu; |
46cb4b7c | 2659 | |
8f42ced9 | 2660 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2661 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2662 | if (dest_cpu == smp_processor_id()) |
2663 | goto unlock; | |
38022906 | 2664 | |
8f42ced9 | 2665 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2666 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2667 | |
8f42ced9 PZ |
2668 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2669 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2670 | return; |
2671 | } | |
0017d735 | 2672 | unlock: |
8f42ced9 | 2673 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2674 | } |
dd41f596 | 2675 | |
1da177e4 LT |
2676 | #endif |
2677 | ||
1da177e4 | 2678 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2679 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2680 | |
2681 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2682 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2683 | |
c5f8d995 HS |
2684 | /* |
2685 | * Return accounted runtime for the task. | |
2686 | * In case the task is currently running, return the runtime plus current's | |
2687 | * pending runtime that have not been accounted yet. | |
2688 | */ | |
2689 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2690 | { | |
2691 | unsigned long flags; | |
2692 | struct rq *rq; | |
6e998916 | 2693 | u64 ns; |
c5f8d995 | 2694 | |
911b2898 PZ |
2695 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2696 | /* | |
2697 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2698 | * So we have a optimization chance when the task's delta_exec is 0. | |
2699 | * Reading ->on_cpu is racy, but this is ok. | |
2700 | * | |
2701 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2702 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2703 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2704 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2705 | * been accounted, so we're correct here as well. | |
911b2898 | 2706 | */ |
da0c1e65 | 2707 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2708 | return p->se.sum_exec_runtime; |
2709 | #endif | |
2710 | ||
c5f8d995 | 2711 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2712 | /* |
2713 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2714 | * project cycles that may never be accounted to this | |
2715 | * thread, breaking clock_gettime(). | |
2716 | */ | |
2717 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2718 | update_rq_clock(rq); | |
2719 | p->sched_class->update_curr(rq); | |
2720 | } | |
2721 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2722 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2723 | |
2724 | return ns; | |
2725 | } | |
48f24c4d | 2726 | |
7835b98b CL |
2727 | /* |
2728 | * This function gets called by the timer code, with HZ frequency. | |
2729 | * We call it with interrupts disabled. | |
7835b98b CL |
2730 | */ |
2731 | void scheduler_tick(void) | |
2732 | { | |
7835b98b CL |
2733 | int cpu = smp_processor_id(); |
2734 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2735 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2736 | |
2737 | sched_clock_tick(); | |
dd41f596 | 2738 | |
05fa785c | 2739 | raw_spin_lock(&rq->lock); |
3e51f33f | 2740 | update_rq_clock(rq); |
fa85ae24 | 2741 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2742 | update_cpu_load_active(rq); |
3289bdb4 | 2743 | calc_global_load_tick(rq); |
05fa785c | 2744 | raw_spin_unlock(&rq->lock); |
7835b98b | 2745 | |
e9d2b064 | 2746 | perf_event_task_tick(); |
e220d2dc | 2747 | |
e418e1c2 | 2748 | #ifdef CONFIG_SMP |
6eb57e0d | 2749 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2750 | trigger_load_balance(rq); |
e418e1c2 | 2751 | #endif |
265f22a9 | 2752 | rq_last_tick_reset(rq); |
1da177e4 LT |
2753 | } |
2754 | ||
265f22a9 FW |
2755 | #ifdef CONFIG_NO_HZ_FULL |
2756 | /** | |
2757 | * scheduler_tick_max_deferment | |
2758 | * | |
2759 | * Keep at least one tick per second when a single | |
2760 | * active task is running because the scheduler doesn't | |
2761 | * yet completely support full dynticks environment. | |
2762 | * | |
2763 | * This makes sure that uptime, CFS vruntime, load | |
2764 | * balancing, etc... continue to move forward, even | |
2765 | * with a very low granularity. | |
e69f6186 YB |
2766 | * |
2767 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2768 | */ |
2769 | u64 scheduler_tick_max_deferment(void) | |
2770 | { | |
2771 | struct rq *rq = this_rq(); | |
316c1608 | 2772 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
2773 | |
2774 | next = rq->last_sched_tick + HZ; | |
2775 | ||
2776 | if (time_before_eq(next, now)) | |
2777 | return 0; | |
2778 | ||
8fe8ff09 | 2779 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2780 | } |
265f22a9 | 2781 | #endif |
1da177e4 | 2782 | |
132380a0 | 2783 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2784 | { |
2785 | if (in_lock_functions(addr)) { | |
2786 | addr = CALLER_ADDR2; | |
2787 | if (in_lock_functions(addr)) | |
2788 | addr = CALLER_ADDR3; | |
2789 | } | |
2790 | return addr; | |
2791 | } | |
1da177e4 | 2792 | |
7e49fcce SR |
2793 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2794 | defined(CONFIG_PREEMPT_TRACER)) | |
2795 | ||
edafe3a5 | 2796 | void preempt_count_add(int val) |
1da177e4 | 2797 | { |
6cd8a4bb | 2798 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2799 | /* |
2800 | * Underflow? | |
2801 | */ | |
9a11b49a IM |
2802 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2803 | return; | |
6cd8a4bb | 2804 | #endif |
bdb43806 | 2805 | __preempt_count_add(val); |
6cd8a4bb | 2806 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2807 | /* |
2808 | * Spinlock count overflowing soon? | |
2809 | */ | |
33859f7f MOS |
2810 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2811 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2812 | #endif |
8f47b187 TG |
2813 | if (preempt_count() == val) { |
2814 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2815 | #ifdef CONFIG_DEBUG_PREEMPT | |
2816 | current->preempt_disable_ip = ip; | |
2817 | #endif | |
2818 | trace_preempt_off(CALLER_ADDR0, ip); | |
2819 | } | |
1da177e4 | 2820 | } |
bdb43806 | 2821 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2822 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2823 | |
edafe3a5 | 2824 | void preempt_count_sub(int val) |
1da177e4 | 2825 | { |
6cd8a4bb | 2826 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2827 | /* |
2828 | * Underflow? | |
2829 | */ | |
01e3eb82 | 2830 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2831 | return; |
1da177e4 LT |
2832 | /* |
2833 | * Is the spinlock portion underflowing? | |
2834 | */ | |
9a11b49a IM |
2835 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2836 | !(preempt_count() & PREEMPT_MASK))) | |
2837 | return; | |
6cd8a4bb | 2838 | #endif |
9a11b49a | 2839 | |
6cd8a4bb SR |
2840 | if (preempt_count() == val) |
2841 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2842 | __preempt_count_sub(val); |
1da177e4 | 2843 | } |
bdb43806 | 2844 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2845 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2846 | |
2847 | #endif | |
2848 | ||
2849 | /* | |
dd41f596 | 2850 | * Print scheduling while atomic bug: |
1da177e4 | 2851 | */ |
dd41f596 | 2852 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2853 | { |
664dfa65 DJ |
2854 | if (oops_in_progress) |
2855 | return; | |
2856 | ||
3df0fc5b PZ |
2857 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2858 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2859 | |
dd41f596 | 2860 | debug_show_held_locks(prev); |
e21f5b15 | 2861 | print_modules(); |
dd41f596 IM |
2862 | if (irqs_disabled()) |
2863 | print_irqtrace_events(prev); | |
8f47b187 TG |
2864 | #ifdef CONFIG_DEBUG_PREEMPT |
2865 | if (in_atomic_preempt_off()) { | |
2866 | pr_err("Preemption disabled at:"); | |
2867 | print_ip_sym(current->preempt_disable_ip); | |
2868 | pr_cont("\n"); | |
2869 | } | |
2870 | #endif | |
6135fc1e | 2871 | dump_stack(); |
373d4d09 | 2872 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2873 | } |
1da177e4 | 2874 | |
dd41f596 IM |
2875 | /* |
2876 | * Various schedule()-time debugging checks and statistics: | |
2877 | */ | |
2878 | static inline void schedule_debug(struct task_struct *prev) | |
2879 | { | |
0d9e2632 AT |
2880 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
2881 | BUG_ON(unlikely(task_stack_end_corrupted(prev))); | |
2882 | #endif | |
1da177e4 | 2883 | /* |
41a2d6cf | 2884 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2885 | * schedule() atomically, we ignore that path. Otherwise whine |
2886 | * if we are scheduling when we should not. | |
1da177e4 | 2887 | */ |
192301e7 | 2888 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2889 | __schedule_bug(prev); |
b3fbab05 | 2890 | rcu_sleep_check(); |
dd41f596 | 2891 | |
1da177e4 LT |
2892 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2893 | ||
2d72376b | 2894 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2895 | } |
2896 | ||
2897 | /* | |
2898 | * Pick up the highest-prio task: | |
2899 | */ | |
2900 | static inline struct task_struct * | |
606dba2e | 2901 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2902 | { |
37e117c0 | 2903 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2904 | struct task_struct *p; |
1da177e4 LT |
2905 | |
2906 | /* | |
dd41f596 IM |
2907 | * Optimization: we know that if all tasks are in |
2908 | * the fair class we can call that function directly: | |
1da177e4 | 2909 | */ |
37e117c0 | 2910 | if (likely(prev->sched_class == class && |
38033c37 | 2911 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2912 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2913 | if (unlikely(p == RETRY_TASK)) |
2914 | goto again; | |
2915 | ||
2916 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2917 | if (unlikely(!p)) | |
2918 | p = idle_sched_class.pick_next_task(rq, prev); | |
2919 | ||
2920 | return p; | |
1da177e4 LT |
2921 | } |
2922 | ||
37e117c0 | 2923 | again: |
34f971f6 | 2924 | for_each_class(class) { |
606dba2e | 2925 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2926 | if (p) { |
2927 | if (unlikely(p == RETRY_TASK)) | |
2928 | goto again; | |
dd41f596 | 2929 | return p; |
37e117c0 | 2930 | } |
dd41f596 | 2931 | } |
34f971f6 PZ |
2932 | |
2933 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2934 | } |
1da177e4 | 2935 | |
dd41f596 | 2936 | /* |
c259e01a | 2937 | * __schedule() is the main scheduler function. |
edde96ea PE |
2938 | * |
2939 | * The main means of driving the scheduler and thus entering this function are: | |
2940 | * | |
2941 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2942 | * | |
2943 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2944 | * paths. For example, see arch/x86/entry_64.S. | |
2945 | * | |
2946 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2947 | * interrupt handler scheduler_tick(). | |
2948 | * | |
2949 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2950 | * task to the run-queue and that's it. | |
2951 | * | |
2952 | * Now, if the new task added to the run-queue preempts the current | |
2953 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2954 | * called on the nearest possible occasion: | |
2955 | * | |
2956 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2957 | * | |
2958 | * - in syscall or exception context, at the next outmost | |
2959 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2960 | * spin_unlock()!) | |
2961 | * | |
2962 | * - in IRQ context, return from interrupt-handler to | |
2963 | * preemptible context | |
2964 | * | |
2965 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2966 | * then at the next: | |
2967 | * | |
2968 | * - cond_resched() call | |
2969 | * - explicit schedule() call | |
2970 | * - return from syscall or exception to user-space | |
2971 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 2972 | * |
b30f0e3f | 2973 | * WARNING: must be called with preemption disabled! |
dd41f596 | 2974 | */ |
c259e01a | 2975 | static void __sched __schedule(void) |
dd41f596 IM |
2976 | { |
2977 | struct task_struct *prev, *next; | |
67ca7bde | 2978 | unsigned long *switch_count; |
dd41f596 | 2979 | struct rq *rq; |
31656519 | 2980 | int cpu; |
dd41f596 | 2981 | |
dd41f596 IM |
2982 | cpu = smp_processor_id(); |
2983 | rq = cpu_rq(cpu); | |
38200cf2 | 2984 | rcu_note_context_switch(); |
dd41f596 | 2985 | prev = rq->curr; |
dd41f596 | 2986 | |
dd41f596 | 2987 | schedule_debug(prev); |
1da177e4 | 2988 | |
31656519 | 2989 | if (sched_feat(HRTICK)) |
f333fdc9 | 2990 | hrtick_clear(rq); |
8f4d37ec | 2991 | |
e0acd0a6 ON |
2992 | /* |
2993 | * Make sure that signal_pending_state()->signal_pending() below | |
2994 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2995 | * done by the caller to avoid the race with signal_wake_up(). | |
2996 | */ | |
2997 | smp_mb__before_spinlock(); | |
05fa785c | 2998 | raw_spin_lock_irq(&rq->lock); |
cbce1a68 | 2999 | lockdep_pin_lock(&rq->lock); |
1da177e4 | 3000 | |
9edfbfed PZ |
3001 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
3002 | ||
246d86b5 | 3003 | switch_count = &prev->nivcsw; |
1da177e4 | 3004 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3005 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3006 | prev->state = TASK_RUNNING; |
21aa9af0 | 3007 | } else { |
2acca55e PZ |
3008 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3009 | prev->on_rq = 0; | |
3010 | ||
21aa9af0 | 3011 | /* |
2acca55e PZ |
3012 | * If a worker went to sleep, notify and ask workqueue |
3013 | * whether it wants to wake up a task to maintain | |
3014 | * concurrency. | |
21aa9af0 TH |
3015 | */ |
3016 | if (prev->flags & PF_WQ_WORKER) { | |
3017 | struct task_struct *to_wakeup; | |
3018 | ||
3019 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3020 | if (to_wakeup) | |
3021 | try_to_wake_up_local(to_wakeup); | |
3022 | } | |
21aa9af0 | 3023 | } |
dd41f596 | 3024 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3025 | } |
3026 | ||
9edfbfed | 3027 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3028 | update_rq_clock(rq); |
3029 | ||
3030 | next = pick_next_task(rq, prev); | |
f26f9aff | 3031 | clear_tsk_need_resched(prev); |
f27dde8d | 3032 | clear_preempt_need_resched(); |
9edfbfed | 3033 | rq->clock_skip_update = 0; |
1da177e4 | 3034 | |
1da177e4 | 3035 | if (likely(prev != next)) { |
1da177e4 LT |
3036 | rq->nr_switches++; |
3037 | rq->curr = next; | |
3038 | ++*switch_count; | |
3039 | ||
dfa50b60 ON |
3040 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
3041 | cpu = cpu_of(rq); | |
cbce1a68 PZ |
3042 | } else { |
3043 | lockdep_unpin_lock(&rq->lock); | |
05fa785c | 3044 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3045 | } |
1da177e4 | 3046 | |
e3fca9e7 | 3047 | balance_callback(rq); |
1da177e4 | 3048 | } |
c259e01a | 3049 | |
9c40cef2 TG |
3050 | static inline void sched_submit_work(struct task_struct *tsk) |
3051 | { | |
3c7d5184 | 3052 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3053 | return; |
3054 | /* | |
3055 | * If we are going to sleep and we have plugged IO queued, | |
3056 | * make sure to submit it to avoid deadlocks. | |
3057 | */ | |
3058 | if (blk_needs_flush_plug(tsk)) | |
3059 | blk_schedule_flush_plug(tsk); | |
3060 | } | |
3061 | ||
722a9f92 | 3062 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3063 | { |
9c40cef2 TG |
3064 | struct task_struct *tsk = current; |
3065 | ||
3066 | sched_submit_work(tsk); | |
bfd9b2b5 | 3067 | do { |
b30f0e3f | 3068 | preempt_disable(); |
bfd9b2b5 | 3069 | __schedule(); |
b30f0e3f | 3070 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3071 | } while (need_resched()); |
c259e01a | 3072 | } |
1da177e4 LT |
3073 | EXPORT_SYMBOL(schedule); |
3074 | ||
91d1aa43 | 3075 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3076 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3077 | { |
3078 | /* | |
3079 | * If we come here after a random call to set_need_resched(), | |
3080 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3081 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3082 | * we find a better solution. | |
7cc78f8f AL |
3083 | * |
3084 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3085 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3086 | * too frequently to make sense yet. |
20ab65e3 | 3087 | */ |
7cc78f8f | 3088 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3089 | schedule(); |
7cc78f8f | 3090 | exception_exit(prev_state); |
20ab65e3 FW |
3091 | } |
3092 | #endif | |
3093 | ||
c5491ea7 TG |
3094 | /** |
3095 | * schedule_preempt_disabled - called with preemption disabled | |
3096 | * | |
3097 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3098 | */ | |
3099 | void __sched schedule_preempt_disabled(void) | |
3100 | { | |
ba74c144 | 3101 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3102 | schedule(); |
3103 | preempt_disable(); | |
3104 | } | |
3105 | ||
06b1f808 | 3106 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3107 | { |
3108 | do { | |
b30f0e3f | 3109 | preempt_active_enter(); |
a18b5d01 | 3110 | __schedule(); |
b30f0e3f | 3111 | preempt_active_exit(); |
a18b5d01 FW |
3112 | |
3113 | /* | |
3114 | * Check again in case we missed a preemption opportunity | |
3115 | * between schedule and now. | |
3116 | */ | |
a18b5d01 FW |
3117 | } while (need_resched()); |
3118 | } | |
3119 | ||
1da177e4 LT |
3120 | #ifdef CONFIG_PREEMPT |
3121 | /* | |
2ed6e34f | 3122 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3123 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3124 | * occur there and call schedule directly. |
3125 | */ | |
722a9f92 | 3126 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3127 | { |
1da177e4 LT |
3128 | /* |
3129 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3130 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3131 | */ |
fbb00b56 | 3132 | if (likely(!preemptible())) |
1da177e4 LT |
3133 | return; |
3134 | ||
a18b5d01 | 3135 | preempt_schedule_common(); |
1da177e4 | 3136 | } |
376e2424 | 3137 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3138 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3139 | |
009f60e2 | 3140 | /** |
4eaca0a8 | 3141 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3142 | * |
3143 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3144 | * recursion and tracing preempt enabling caused by the tracing | |
3145 | * infrastructure itself. But as tracing can happen in areas coming | |
3146 | * from userspace or just about to enter userspace, a preempt enable | |
3147 | * can occur before user_exit() is called. This will cause the scheduler | |
3148 | * to be called when the system is still in usermode. | |
3149 | * | |
3150 | * To prevent this, the preempt_enable_notrace will use this function | |
3151 | * instead of preempt_schedule() to exit user context if needed before | |
3152 | * calling the scheduler. | |
3153 | */ | |
4eaca0a8 | 3154 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3155 | { |
3156 | enum ctx_state prev_ctx; | |
3157 | ||
3158 | if (likely(!preemptible())) | |
3159 | return; | |
3160 | ||
3161 | do { | |
be690035 FW |
3162 | /* |
3163 | * Use raw __prempt_count() ops that don't call function. | |
3164 | * We can't call functions before disabling preemption which | |
3165 | * disarm preemption tracing recursions. | |
3166 | */ | |
3167 | __preempt_count_add(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); | |
3168 | barrier(); | |
009f60e2 ON |
3169 | /* |
3170 | * Needs preempt disabled in case user_exit() is traced | |
3171 | * and the tracer calls preempt_enable_notrace() causing | |
3172 | * an infinite recursion. | |
3173 | */ | |
3174 | prev_ctx = exception_enter(); | |
3175 | __schedule(); | |
3176 | exception_exit(prev_ctx); | |
3177 | ||
009f60e2 | 3178 | barrier(); |
be690035 | 3179 | __preempt_count_sub(PREEMPT_ACTIVE + PREEMPT_DISABLE_OFFSET); |
009f60e2 ON |
3180 | } while (need_resched()); |
3181 | } | |
4eaca0a8 | 3182 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3183 | |
32e475d7 | 3184 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3185 | |
3186 | /* | |
2ed6e34f | 3187 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3188 | * off of irq context. |
3189 | * Note, that this is called and return with irqs disabled. This will | |
3190 | * protect us against recursive calling from irq. | |
3191 | */ | |
722a9f92 | 3192 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3193 | { |
b22366cd | 3194 | enum ctx_state prev_state; |
6478d880 | 3195 | |
2ed6e34f | 3196 | /* Catch callers which need to be fixed */ |
f27dde8d | 3197 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3198 | |
b22366cd FW |
3199 | prev_state = exception_enter(); |
3200 | ||
3a5c359a | 3201 | do { |
b30f0e3f | 3202 | preempt_active_enter(); |
3a5c359a | 3203 | local_irq_enable(); |
c259e01a | 3204 | __schedule(); |
3a5c359a | 3205 | local_irq_disable(); |
b30f0e3f | 3206 | preempt_active_exit(); |
5ed0cec0 | 3207 | } while (need_resched()); |
b22366cd FW |
3208 | |
3209 | exception_exit(prev_state); | |
1da177e4 LT |
3210 | } |
3211 | ||
63859d4f | 3212 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3213 | void *key) |
1da177e4 | 3214 | { |
63859d4f | 3215 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3216 | } |
1da177e4 LT |
3217 | EXPORT_SYMBOL(default_wake_function); |
3218 | ||
b29739f9 IM |
3219 | #ifdef CONFIG_RT_MUTEXES |
3220 | ||
3221 | /* | |
3222 | * rt_mutex_setprio - set the current priority of a task | |
3223 | * @p: task | |
3224 | * @prio: prio value (kernel-internal form) | |
3225 | * | |
3226 | * This function changes the 'effective' priority of a task. It does | |
3227 | * not touch ->normal_prio like __setscheduler(). | |
3228 | * | |
c365c292 TG |
3229 | * Used by the rt_mutex code to implement priority inheritance |
3230 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3231 | */ |
36c8b586 | 3232 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3233 | { |
da0c1e65 | 3234 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 3235 | struct rq *rq; |
83ab0aa0 | 3236 | const struct sched_class *prev_class; |
b29739f9 | 3237 | |
aab03e05 | 3238 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3239 | |
0122ec5b | 3240 | rq = __task_rq_lock(p); |
b29739f9 | 3241 | |
1c4dd99b TG |
3242 | /* |
3243 | * Idle task boosting is a nono in general. There is one | |
3244 | * exception, when PREEMPT_RT and NOHZ is active: | |
3245 | * | |
3246 | * The idle task calls get_next_timer_interrupt() and holds | |
3247 | * the timer wheel base->lock on the CPU and another CPU wants | |
3248 | * to access the timer (probably to cancel it). We can safely | |
3249 | * ignore the boosting request, as the idle CPU runs this code | |
3250 | * with interrupts disabled and will complete the lock | |
3251 | * protected section without being interrupted. So there is no | |
3252 | * real need to boost. | |
3253 | */ | |
3254 | if (unlikely(p == rq->idle)) { | |
3255 | WARN_ON(p != rq->curr); | |
3256 | WARN_ON(p->pi_blocked_on); | |
3257 | goto out_unlock; | |
3258 | } | |
3259 | ||
a8027073 | 3260 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3261 | oldprio = p->prio; |
83ab0aa0 | 3262 | prev_class = p->sched_class; |
da0c1e65 | 3263 | queued = task_on_rq_queued(p); |
051a1d1a | 3264 | running = task_current(rq, p); |
da0c1e65 | 3265 | if (queued) |
69be72c1 | 3266 | dequeue_task(rq, p, 0); |
0e1f3483 | 3267 | if (running) |
f3cd1c4e | 3268 | put_prev_task(rq, p); |
dd41f596 | 3269 | |
2d3d891d DF |
3270 | /* |
3271 | * Boosting condition are: | |
3272 | * 1. -rt task is running and holds mutex A | |
3273 | * --> -dl task blocks on mutex A | |
3274 | * | |
3275 | * 2. -dl task is running and holds mutex A | |
3276 | * --> -dl task blocks on mutex A and could preempt the | |
3277 | * running task | |
3278 | */ | |
3279 | if (dl_prio(prio)) { | |
466af29b ON |
3280 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3281 | if (!dl_prio(p->normal_prio) || | |
3282 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3283 | p->dl.dl_boosted = 1; |
2d3d891d DF |
3284 | enqueue_flag = ENQUEUE_REPLENISH; |
3285 | } else | |
3286 | p->dl.dl_boosted = 0; | |
aab03e05 | 3287 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3288 | } else if (rt_prio(prio)) { |
3289 | if (dl_prio(oldprio)) | |
3290 | p->dl.dl_boosted = 0; | |
3291 | if (oldprio < prio) | |
3292 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3293 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3294 | } else { |
3295 | if (dl_prio(oldprio)) | |
3296 | p->dl.dl_boosted = 0; | |
746db944 BS |
3297 | if (rt_prio(oldprio)) |
3298 | p->rt.timeout = 0; | |
dd41f596 | 3299 | p->sched_class = &fair_sched_class; |
2d3d891d | 3300 | } |
dd41f596 | 3301 | |
b29739f9 IM |
3302 | p->prio = prio; |
3303 | ||
0e1f3483 HS |
3304 | if (running) |
3305 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3306 | if (queued) |
2d3d891d | 3307 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3308 | |
da7a735e | 3309 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3310 | out_unlock: |
4c9a4bc8 | 3311 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3312 | __task_rq_unlock(rq); |
4c9a4bc8 PZ |
3313 | |
3314 | balance_callback(rq); | |
3315 | preempt_enable(); | |
b29739f9 | 3316 | } |
b29739f9 | 3317 | #endif |
d50dde5a | 3318 | |
36c8b586 | 3319 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3320 | { |
da0c1e65 | 3321 | int old_prio, delta, queued; |
1da177e4 | 3322 | unsigned long flags; |
70b97a7f | 3323 | struct rq *rq; |
1da177e4 | 3324 | |
75e45d51 | 3325 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3326 | return; |
3327 | /* | |
3328 | * We have to be careful, if called from sys_setpriority(), | |
3329 | * the task might be in the middle of scheduling on another CPU. | |
3330 | */ | |
3331 | rq = task_rq_lock(p, &flags); | |
3332 | /* | |
3333 | * The RT priorities are set via sched_setscheduler(), but we still | |
3334 | * allow the 'normal' nice value to be set - but as expected | |
3335 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3336 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3337 | */ |
aab03e05 | 3338 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3339 | p->static_prio = NICE_TO_PRIO(nice); |
3340 | goto out_unlock; | |
3341 | } | |
da0c1e65 KT |
3342 | queued = task_on_rq_queued(p); |
3343 | if (queued) | |
69be72c1 | 3344 | dequeue_task(rq, p, 0); |
1da177e4 | 3345 | |
1da177e4 | 3346 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3347 | set_load_weight(p); |
b29739f9 IM |
3348 | old_prio = p->prio; |
3349 | p->prio = effective_prio(p); | |
3350 | delta = p->prio - old_prio; | |
1da177e4 | 3351 | |
da0c1e65 | 3352 | if (queued) { |
371fd7e7 | 3353 | enqueue_task(rq, p, 0); |
1da177e4 | 3354 | /* |
d5f9f942 AM |
3355 | * If the task increased its priority or is running and |
3356 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3357 | */ |
d5f9f942 | 3358 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3359 | resched_curr(rq); |
1da177e4 LT |
3360 | } |
3361 | out_unlock: | |
0122ec5b | 3362 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3363 | } |
1da177e4 LT |
3364 | EXPORT_SYMBOL(set_user_nice); |
3365 | ||
e43379f1 MM |
3366 | /* |
3367 | * can_nice - check if a task can reduce its nice value | |
3368 | * @p: task | |
3369 | * @nice: nice value | |
3370 | */ | |
36c8b586 | 3371 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3372 | { |
024f4747 | 3373 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3374 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3375 | |
78d7d407 | 3376 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3377 | capable(CAP_SYS_NICE)); |
3378 | } | |
3379 | ||
1da177e4 LT |
3380 | #ifdef __ARCH_WANT_SYS_NICE |
3381 | ||
3382 | /* | |
3383 | * sys_nice - change the priority of the current process. | |
3384 | * @increment: priority increment | |
3385 | * | |
3386 | * sys_setpriority is a more generic, but much slower function that | |
3387 | * does similar things. | |
3388 | */ | |
5add95d4 | 3389 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3390 | { |
48f24c4d | 3391 | long nice, retval; |
1da177e4 LT |
3392 | |
3393 | /* | |
3394 | * Setpriority might change our priority at the same moment. | |
3395 | * We don't have to worry. Conceptually one call occurs first | |
3396 | * and we have a single winner. | |
3397 | */ | |
a9467fa3 | 3398 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3399 | nice = task_nice(current) + increment; |
1da177e4 | 3400 | |
a9467fa3 | 3401 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3402 | if (increment < 0 && !can_nice(current, nice)) |
3403 | return -EPERM; | |
3404 | ||
1da177e4 LT |
3405 | retval = security_task_setnice(current, nice); |
3406 | if (retval) | |
3407 | return retval; | |
3408 | ||
3409 | set_user_nice(current, nice); | |
3410 | return 0; | |
3411 | } | |
3412 | ||
3413 | #endif | |
3414 | ||
3415 | /** | |
3416 | * task_prio - return the priority value of a given task. | |
3417 | * @p: the task in question. | |
3418 | * | |
e69f6186 | 3419 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3420 | * RT tasks are offset by -200. Normal tasks are centered |
3421 | * around 0, value goes from -16 to +15. | |
3422 | */ | |
36c8b586 | 3423 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3424 | { |
3425 | return p->prio - MAX_RT_PRIO; | |
3426 | } | |
3427 | ||
1da177e4 LT |
3428 | /** |
3429 | * idle_cpu - is a given cpu idle currently? | |
3430 | * @cpu: the processor in question. | |
e69f6186 YB |
3431 | * |
3432 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3433 | */ |
3434 | int idle_cpu(int cpu) | |
3435 | { | |
908a3283 TG |
3436 | struct rq *rq = cpu_rq(cpu); |
3437 | ||
3438 | if (rq->curr != rq->idle) | |
3439 | return 0; | |
3440 | ||
3441 | if (rq->nr_running) | |
3442 | return 0; | |
3443 | ||
3444 | #ifdef CONFIG_SMP | |
3445 | if (!llist_empty(&rq->wake_list)) | |
3446 | return 0; | |
3447 | #endif | |
3448 | ||
3449 | return 1; | |
1da177e4 LT |
3450 | } |
3451 | ||
1da177e4 LT |
3452 | /** |
3453 | * idle_task - return the idle task for a given cpu. | |
3454 | * @cpu: the processor in question. | |
e69f6186 YB |
3455 | * |
3456 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3457 | */ |
36c8b586 | 3458 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3459 | { |
3460 | return cpu_rq(cpu)->idle; | |
3461 | } | |
3462 | ||
3463 | /** | |
3464 | * find_process_by_pid - find a process with a matching PID value. | |
3465 | * @pid: the pid in question. | |
e69f6186 YB |
3466 | * |
3467 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3468 | */ |
a9957449 | 3469 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3470 | { |
228ebcbe | 3471 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3472 | } |
3473 | ||
aab03e05 DF |
3474 | /* |
3475 | * This function initializes the sched_dl_entity of a newly becoming | |
3476 | * SCHED_DEADLINE task. | |
3477 | * | |
3478 | * Only the static values are considered here, the actual runtime and the | |
3479 | * absolute deadline will be properly calculated when the task is enqueued | |
3480 | * for the first time with its new policy. | |
3481 | */ | |
3482 | static void | |
3483 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3484 | { | |
3485 | struct sched_dl_entity *dl_se = &p->dl; | |
3486 | ||
aab03e05 DF |
3487 | dl_se->dl_runtime = attr->sched_runtime; |
3488 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3489 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3490 | dl_se->flags = attr->sched_flags; |
332ac17e | 3491 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3492 | |
3493 | /* | |
3494 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3495 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3496 | * | |
3497 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3498 | * point. This would include retaining the task_struct until that time | |
3499 | * and change dl_overflow() to not immediately decrement the current | |
3500 | * amount. | |
3501 | * | |
3502 | * Instead we retain the current runtime/deadline and let the new | |
3503 | * parameters take effect after the current reservation period lapses. | |
3504 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3505 | * before the current scheduling deadline. | |
3506 | * | |
3507 | * We can still have temporary overloads because we do not delay the | |
3508 | * change in bandwidth until that time; so admission control is | |
3509 | * not on the safe side. It does however guarantee tasks will never | |
3510 | * consume more than promised. | |
3511 | */ | |
aab03e05 DF |
3512 | } |
3513 | ||
c13db6b1 SR |
3514 | /* |
3515 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3516 | * it calls know not to change it. | |
3517 | */ | |
3518 | #define SETPARAM_POLICY -1 | |
3519 | ||
c365c292 TG |
3520 | static void __setscheduler_params(struct task_struct *p, |
3521 | const struct sched_attr *attr) | |
1da177e4 | 3522 | { |
d50dde5a DF |
3523 | int policy = attr->sched_policy; |
3524 | ||
c13db6b1 | 3525 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3526 | policy = p->policy; |
3527 | ||
1da177e4 | 3528 | p->policy = policy; |
d50dde5a | 3529 | |
aab03e05 DF |
3530 | if (dl_policy(policy)) |
3531 | __setparam_dl(p, attr); | |
39fd8fd2 | 3532 | else if (fair_policy(policy)) |
d50dde5a DF |
3533 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3534 | ||
39fd8fd2 PZ |
3535 | /* |
3536 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3537 | * !rt_policy. Always setting this ensures that things like | |
3538 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3539 | */ | |
3540 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3541 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3542 | set_load_weight(p); |
3543 | } | |
39fd8fd2 | 3544 | |
c365c292 TG |
3545 | /* Actually do priority change: must hold pi & rq lock. */ |
3546 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3547 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3548 | { |
3549 | __setscheduler_params(p, attr); | |
d50dde5a | 3550 | |
383afd09 | 3551 | /* |
0782e63b TG |
3552 | * Keep a potential priority boosting if called from |
3553 | * sched_setscheduler(). | |
383afd09 | 3554 | */ |
0782e63b TG |
3555 | if (keep_boost) |
3556 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3557 | else | |
3558 | p->prio = normal_prio(p); | |
383afd09 | 3559 | |
aab03e05 DF |
3560 | if (dl_prio(p->prio)) |
3561 | p->sched_class = &dl_sched_class; | |
3562 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3563 | p->sched_class = &rt_sched_class; |
3564 | else | |
3565 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3566 | } |
aab03e05 DF |
3567 | |
3568 | static void | |
3569 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3570 | { | |
3571 | struct sched_dl_entity *dl_se = &p->dl; | |
3572 | ||
3573 | attr->sched_priority = p->rt_priority; | |
3574 | attr->sched_runtime = dl_se->dl_runtime; | |
3575 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3576 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3577 | attr->sched_flags = dl_se->flags; |
3578 | } | |
3579 | ||
3580 | /* | |
3581 | * This function validates the new parameters of a -deadline task. | |
3582 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3583 | * than the runtime, as well as the period of being zero or |
332ac17e | 3584 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3585 | * user parameters are above the internal resolution of 1us (we |
3586 | * check sched_runtime only since it is always the smaller one) and | |
3587 | * below 2^63 ns (we have to check both sched_deadline and | |
3588 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3589 | */ |
3590 | static bool | |
3591 | __checkparam_dl(const struct sched_attr *attr) | |
3592 | { | |
b0827819 JL |
3593 | /* deadline != 0 */ |
3594 | if (attr->sched_deadline == 0) | |
3595 | return false; | |
3596 | ||
3597 | /* | |
3598 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3599 | * that big. | |
3600 | */ | |
3601 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3602 | return false; | |
3603 | ||
3604 | /* | |
3605 | * Since we use the MSB for wrap-around and sign issues, make | |
3606 | * sure it's not set (mind that period can be equal to zero). | |
3607 | */ | |
3608 | if (attr->sched_deadline & (1ULL << 63) || | |
3609 | attr->sched_period & (1ULL << 63)) | |
3610 | return false; | |
3611 | ||
3612 | /* runtime <= deadline <= period (if period != 0) */ | |
3613 | if ((attr->sched_period != 0 && | |
3614 | attr->sched_period < attr->sched_deadline) || | |
3615 | attr->sched_deadline < attr->sched_runtime) | |
3616 | return false; | |
3617 | ||
3618 | return true; | |
aab03e05 DF |
3619 | } |
3620 | ||
c69e8d9c DH |
3621 | /* |
3622 | * check the target process has a UID that matches the current process's | |
3623 | */ | |
3624 | static bool check_same_owner(struct task_struct *p) | |
3625 | { | |
3626 | const struct cred *cred = current_cred(), *pcred; | |
3627 | bool match; | |
3628 | ||
3629 | rcu_read_lock(); | |
3630 | pcred = __task_cred(p); | |
9c806aa0 EB |
3631 | match = (uid_eq(cred->euid, pcred->euid) || |
3632 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3633 | rcu_read_unlock(); |
3634 | return match; | |
3635 | } | |
3636 | ||
75381608 WL |
3637 | static bool dl_param_changed(struct task_struct *p, |
3638 | const struct sched_attr *attr) | |
3639 | { | |
3640 | struct sched_dl_entity *dl_se = &p->dl; | |
3641 | ||
3642 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3643 | dl_se->dl_deadline != attr->sched_deadline || | |
3644 | dl_se->dl_period != attr->sched_period || | |
3645 | dl_se->flags != attr->sched_flags) | |
3646 | return true; | |
3647 | ||
3648 | return false; | |
3649 | } | |
3650 | ||
d50dde5a DF |
3651 | static int __sched_setscheduler(struct task_struct *p, |
3652 | const struct sched_attr *attr, | |
dbc7f069 | 3653 | bool user, bool pi) |
1da177e4 | 3654 | { |
383afd09 SR |
3655 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3656 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3657 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3658 | int new_effective_prio, policy = attr->sched_policy; |
1da177e4 | 3659 | unsigned long flags; |
83ab0aa0 | 3660 | const struct sched_class *prev_class; |
70b97a7f | 3661 | struct rq *rq; |
ca94c442 | 3662 | int reset_on_fork; |
1da177e4 | 3663 | |
66e5393a SR |
3664 | /* may grab non-irq protected spin_locks */ |
3665 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3666 | recheck: |
3667 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3668 | if (policy < 0) { |
3669 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3670 | policy = oldpolicy = p->policy; |
ca94c442 | 3671 | } else { |
7479f3c9 | 3672 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3673 | |
aab03e05 DF |
3674 | if (policy != SCHED_DEADLINE && |
3675 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3676 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3677 | policy != SCHED_IDLE) | |
3678 | return -EINVAL; | |
3679 | } | |
3680 | ||
7479f3c9 PZ |
3681 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3682 | return -EINVAL; | |
3683 | ||
1da177e4 LT |
3684 | /* |
3685 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3686 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3687 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3688 | */ |
0bb040a4 | 3689 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3690 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3691 | return -EINVAL; |
aab03e05 DF |
3692 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3693 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3694 | return -EINVAL; |
3695 | ||
37e4ab3f OC |
3696 | /* |
3697 | * Allow unprivileged RT tasks to decrease priority: | |
3698 | */ | |
961ccddd | 3699 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3700 | if (fair_policy(policy)) { |
d0ea0268 | 3701 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3702 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3703 | return -EPERM; |
3704 | } | |
3705 | ||
e05606d3 | 3706 | if (rt_policy(policy)) { |
a44702e8 ON |
3707 | unsigned long rlim_rtprio = |
3708 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3709 | |
3710 | /* can't set/change the rt policy */ | |
3711 | if (policy != p->policy && !rlim_rtprio) | |
3712 | return -EPERM; | |
3713 | ||
3714 | /* can't increase priority */ | |
d50dde5a DF |
3715 | if (attr->sched_priority > p->rt_priority && |
3716 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3717 | return -EPERM; |
3718 | } | |
c02aa73b | 3719 | |
d44753b8 JL |
3720 | /* |
3721 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3722 | * (safest behavior); in the future we would like to allow | |
3723 | * unprivileged DL tasks to increase their relative deadline | |
3724 | * or reduce their runtime (both ways reducing utilization) | |
3725 | */ | |
3726 | if (dl_policy(policy)) | |
3727 | return -EPERM; | |
3728 | ||
dd41f596 | 3729 | /* |
c02aa73b DH |
3730 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3731 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3732 | */ |
c02aa73b | 3733 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3734 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3735 | return -EPERM; |
3736 | } | |
5fe1d75f | 3737 | |
37e4ab3f | 3738 | /* can't change other user's priorities */ |
c69e8d9c | 3739 | if (!check_same_owner(p)) |
37e4ab3f | 3740 | return -EPERM; |
ca94c442 LP |
3741 | |
3742 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3743 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3744 | return -EPERM; | |
37e4ab3f | 3745 | } |
1da177e4 | 3746 | |
725aad24 | 3747 | if (user) { |
b0ae1981 | 3748 | retval = security_task_setscheduler(p); |
725aad24 JF |
3749 | if (retval) |
3750 | return retval; | |
3751 | } | |
3752 | ||
b29739f9 IM |
3753 | /* |
3754 | * make sure no PI-waiters arrive (or leave) while we are | |
3755 | * changing the priority of the task: | |
0122ec5b | 3756 | * |
25985edc | 3757 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3758 | * runqueue lock must be held. |
3759 | */ | |
0122ec5b | 3760 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3761 | |
34f971f6 PZ |
3762 | /* |
3763 | * Changing the policy of the stop threads its a very bad idea | |
3764 | */ | |
3765 | if (p == rq->stop) { | |
0122ec5b | 3766 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3767 | return -EINVAL; |
3768 | } | |
3769 | ||
a51e9198 | 3770 | /* |
d6b1e911 TG |
3771 | * If not changing anything there's no need to proceed further, |
3772 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3773 | */ |
d50dde5a | 3774 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3775 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3776 | goto change; |
3777 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3778 | goto change; | |
75381608 | 3779 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 3780 | goto change; |
d50dde5a | 3781 | |
d6b1e911 | 3782 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3783 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3784 | return 0; |
3785 | } | |
d50dde5a | 3786 | change: |
a51e9198 | 3787 | |
dc61b1d6 | 3788 | if (user) { |
332ac17e | 3789 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3790 | /* |
3791 | * Do not allow realtime tasks into groups that have no runtime | |
3792 | * assigned. | |
3793 | */ | |
3794 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3795 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3796 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3797 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3798 | return -EPERM; |
3799 | } | |
dc61b1d6 | 3800 | #endif |
332ac17e DF |
3801 | #ifdef CONFIG_SMP |
3802 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3803 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3804 | |
3805 | /* | |
3806 | * Don't allow tasks with an affinity mask smaller than | |
3807 | * the entire root_domain to become SCHED_DEADLINE. We | |
3808 | * will also fail if there's no bandwidth available. | |
3809 | */ | |
e4099a5e PZ |
3810 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3811 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3812 | task_rq_unlock(rq, p, &flags); |
3813 | return -EPERM; | |
3814 | } | |
3815 | } | |
3816 | #endif | |
3817 | } | |
dc61b1d6 | 3818 | |
1da177e4 LT |
3819 | /* recheck policy now with rq lock held */ |
3820 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3821 | policy = oldpolicy = -1; | |
0122ec5b | 3822 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3823 | goto recheck; |
3824 | } | |
332ac17e DF |
3825 | |
3826 | /* | |
3827 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3828 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3829 | * is available. | |
3830 | */ | |
e4099a5e | 3831 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3832 | task_rq_unlock(rq, p, &flags); |
3833 | return -EBUSY; | |
3834 | } | |
3835 | ||
c365c292 TG |
3836 | p->sched_reset_on_fork = reset_on_fork; |
3837 | oldprio = p->prio; | |
3838 | ||
dbc7f069 PZ |
3839 | if (pi) { |
3840 | /* | |
3841 | * Take priority boosted tasks into account. If the new | |
3842 | * effective priority is unchanged, we just store the new | |
3843 | * normal parameters and do not touch the scheduler class and | |
3844 | * the runqueue. This will be done when the task deboost | |
3845 | * itself. | |
3846 | */ | |
3847 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
3848 | if (new_effective_prio == oldprio) { | |
3849 | __setscheduler_params(p, attr); | |
3850 | task_rq_unlock(rq, p, &flags); | |
3851 | return 0; | |
3852 | } | |
c365c292 TG |
3853 | } |
3854 | ||
da0c1e65 | 3855 | queued = task_on_rq_queued(p); |
051a1d1a | 3856 | running = task_current(rq, p); |
da0c1e65 | 3857 | if (queued) |
4ca9b72b | 3858 | dequeue_task(rq, p, 0); |
0e1f3483 | 3859 | if (running) |
f3cd1c4e | 3860 | put_prev_task(rq, p); |
f6b53205 | 3861 | |
83ab0aa0 | 3862 | prev_class = p->sched_class; |
dbc7f069 | 3863 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 3864 | |
0e1f3483 HS |
3865 | if (running) |
3866 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3867 | if (queued) { |
81a44c54 TG |
3868 | /* |
3869 | * We enqueue to tail when the priority of a task is | |
3870 | * increased (user space view). | |
3871 | */ | |
3872 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3873 | } | |
cb469845 | 3874 | |
da7a735e | 3875 | check_class_changed(rq, p, prev_class, oldprio); |
4c9a4bc8 | 3876 | preempt_disable(); /* avoid rq from going away on us */ |
0122ec5b | 3877 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3878 | |
dbc7f069 PZ |
3879 | if (pi) |
3880 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 3881 | |
4c9a4bc8 PZ |
3882 | /* |
3883 | * Run balance callbacks after we've adjusted the PI chain. | |
3884 | */ | |
3885 | balance_callback(rq); | |
3886 | preempt_enable(); | |
95e02ca9 | 3887 | |
1da177e4 LT |
3888 | return 0; |
3889 | } | |
961ccddd | 3890 | |
7479f3c9 PZ |
3891 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3892 | const struct sched_param *param, bool check) | |
3893 | { | |
3894 | struct sched_attr attr = { | |
3895 | .sched_policy = policy, | |
3896 | .sched_priority = param->sched_priority, | |
3897 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3898 | }; | |
3899 | ||
c13db6b1 SR |
3900 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3901 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3902 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3903 | policy &= ~SCHED_RESET_ON_FORK; | |
3904 | attr.sched_policy = policy; | |
3905 | } | |
3906 | ||
dbc7f069 | 3907 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 3908 | } |
961ccddd RR |
3909 | /** |
3910 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3911 | * @p: the task in question. | |
3912 | * @policy: new policy. | |
3913 | * @param: structure containing the new RT priority. | |
3914 | * | |
e69f6186 YB |
3915 | * Return: 0 on success. An error code otherwise. |
3916 | * | |
961ccddd RR |
3917 | * NOTE that the task may be already dead. |
3918 | */ | |
3919 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3920 | const struct sched_param *param) |
961ccddd | 3921 | { |
7479f3c9 | 3922 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3923 | } |
1da177e4 LT |
3924 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3925 | ||
d50dde5a DF |
3926 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3927 | { | |
dbc7f069 | 3928 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
3929 | } |
3930 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3931 | ||
961ccddd RR |
3932 | /** |
3933 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3934 | * @p: the task in question. | |
3935 | * @policy: new policy. | |
3936 | * @param: structure containing the new RT priority. | |
3937 | * | |
3938 | * Just like sched_setscheduler, only don't bother checking if the | |
3939 | * current context has permission. For example, this is needed in | |
3940 | * stop_machine(): we create temporary high priority worker threads, | |
3941 | * but our caller might not have that capability. | |
e69f6186 YB |
3942 | * |
3943 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3944 | */ |
3945 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3946 | const struct sched_param *param) |
961ccddd | 3947 | { |
7479f3c9 | 3948 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3949 | } |
3950 | ||
95cdf3b7 IM |
3951 | static int |
3952 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3953 | { |
1da177e4 LT |
3954 | struct sched_param lparam; |
3955 | struct task_struct *p; | |
36c8b586 | 3956 | int retval; |
1da177e4 LT |
3957 | |
3958 | if (!param || pid < 0) | |
3959 | return -EINVAL; | |
3960 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3961 | return -EFAULT; | |
5fe1d75f ON |
3962 | |
3963 | rcu_read_lock(); | |
3964 | retval = -ESRCH; | |
1da177e4 | 3965 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3966 | if (p != NULL) |
3967 | retval = sched_setscheduler(p, policy, &lparam); | |
3968 | rcu_read_unlock(); | |
36c8b586 | 3969 | |
1da177e4 LT |
3970 | return retval; |
3971 | } | |
3972 | ||
d50dde5a DF |
3973 | /* |
3974 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3975 | */ | |
3976 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3977 | struct sched_attr *attr) | |
3978 | { | |
3979 | u32 size; | |
3980 | int ret; | |
3981 | ||
3982 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3983 | return -EFAULT; | |
3984 | ||
3985 | /* | |
3986 | * zero the full structure, so that a short copy will be nice. | |
3987 | */ | |
3988 | memset(attr, 0, sizeof(*attr)); | |
3989 | ||
3990 | ret = get_user(size, &uattr->size); | |
3991 | if (ret) | |
3992 | return ret; | |
3993 | ||
3994 | if (size > PAGE_SIZE) /* silly large */ | |
3995 | goto err_size; | |
3996 | ||
3997 | if (!size) /* abi compat */ | |
3998 | size = SCHED_ATTR_SIZE_VER0; | |
3999 | ||
4000 | if (size < SCHED_ATTR_SIZE_VER0) | |
4001 | goto err_size; | |
4002 | ||
4003 | /* | |
4004 | * If we're handed a bigger struct than we know of, | |
4005 | * ensure all the unknown bits are 0 - i.e. new | |
4006 | * user-space does not rely on any kernel feature | |
4007 | * extensions we dont know about yet. | |
4008 | */ | |
4009 | if (size > sizeof(*attr)) { | |
4010 | unsigned char __user *addr; | |
4011 | unsigned char __user *end; | |
4012 | unsigned char val; | |
4013 | ||
4014 | addr = (void __user *)uattr + sizeof(*attr); | |
4015 | end = (void __user *)uattr + size; | |
4016 | ||
4017 | for (; addr < end; addr++) { | |
4018 | ret = get_user(val, addr); | |
4019 | if (ret) | |
4020 | return ret; | |
4021 | if (val) | |
4022 | goto err_size; | |
4023 | } | |
4024 | size = sizeof(*attr); | |
4025 | } | |
4026 | ||
4027 | ret = copy_from_user(attr, uattr, size); | |
4028 | if (ret) | |
4029 | return -EFAULT; | |
4030 | ||
4031 | /* | |
4032 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
4033 | * to be strict and return an error on out-of-bounds values? | |
4034 | */ | |
75e45d51 | 4035 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4036 | |
e78c7bca | 4037 | return 0; |
d50dde5a DF |
4038 | |
4039 | err_size: | |
4040 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4041 | return -E2BIG; |
d50dde5a DF |
4042 | } |
4043 | ||
1da177e4 LT |
4044 | /** |
4045 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4046 | * @pid: the pid in question. | |
4047 | * @policy: new policy. | |
4048 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4049 | * |
4050 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4051 | */ |
5add95d4 HC |
4052 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4053 | struct sched_param __user *, param) | |
1da177e4 | 4054 | { |
c21761f1 JB |
4055 | /* negative values for policy are not valid */ |
4056 | if (policy < 0) | |
4057 | return -EINVAL; | |
4058 | ||
1da177e4 LT |
4059 | return do_sched_setscheduler(pid, policy, param); |
4060 | } | |
4061 | ||
4062 | /** | |
4063 | * sys_sched_setparam - set/change the RT priority of a thread | |
4064 | * @pid: the pid in question. | |
4065 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4066 | * |
4067 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4068 | */ |
5add95d4 | 4069 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4070 | { |
c13db6b1 | 4071 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4072 | } |
4073 | ||
d50dde5a DF |
4074 | /** |
4075 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4076 | * @pid: the pid in question. | |
5778fccf | 4077 | * @uattr: structure containing the extended parameters. |
db66d756 | 4078 | * @flags: for future extension. |
d50dde5a | 4079 | */ |
6d35ab48 PZ |
4080 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4081 | unsigned int, flags) | |
d50dde5a DF |
4082 | { |
4083 | struct sched_attr attr; | |
4084 | struct task_struct *p; | |
4085 | int retval; | |
4086 | ||
6d35ab48 | 4087 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4088 | return -EINVAL; |
4089 | ||
143cf23d MK |
4090 | retval = sched_copy_attr(uattr, &attr); |
4091 | if (retval) | |
4092 | return retval; | |
d50dde5a | 4093 | |
b14ed2c2 | 4094 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4095 | return -EINVAL; |
d50dde5a DF |
4096 | |
4097 | rcu_read_lock(); | |
4098 | retval = -ESRCH; | |
4099 | p = find_process_by_pid(pid); | |
4100 | if (p != NULL) | |
4101 | retval = sched_setattr(p, &attr); | |
4102 | rcu_read_unlock(); | |
4103 | ||
4104 | return retval; | |
4105 | } | |
4106 | ||
1da177e4 LT |
4107 | /** |
4108 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4109 | * @pid: the pid in question. | |
e69f6186 YB |
4110 | * |
4111 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4112 | * code. | |
1da177e4 | 4113 | */ |
5add95d4 | 4114 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4115 | { |
36c8b586 | 4116 | struct task_struct *p; |
3a5c359a | 4117 | int retval; |
1da177e4 LT |
4118 | |
4119 | if (pid < 0) | |
3a5c359a | 4120 | return -EINVAL; |
1da177e4 LT |
4121 | |
4122 | retval = -ESRCH; | |
5fe85be0 | 4123 | rcu_read_lock(); |
1da177e4 LT |
4124 | p = find_process_by_pid(pid); |
4125 | if (p) { | |
4126 | retval = security_task_getscheduler(p); | |
4127 | if (!retval) | |
ca94c442 LP |
4128 | retval = p->policy |
4129 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4130 | } |
5fe85be0 | 4131 | rcu_read_unlock(); |
1da177e4 LT |
4132 | return retval; |
4133 | } | |
4134 | ||
4135 | /** | |
ca94c442 | 4136 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4137 | * @pid: the pid in question. |
4138 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4139 | * |
4140 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4141 | * code. | |
1da177e4 | 4142 | */ |
5add95d4 | 4143 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4144 | { |
ce5f7f82 | 4145 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4146 | struct task_struct *p; |
3a5c359a | 4147 | int retval; |
1da177e4 LT |
4148 | |
4149 | if (!param || pid < 0) | |
3a5c359a | 4150 | return -EINVAL; |
1da177e4 | 4151 | |
5fe85be0 | 4152 | rcu_read_lock(); |
1da177e4 LT |
4153 | p = find_process_by_pid(pid); |
4154 | retval = -ESRCH; | |
4155 | if (!p) | |
4156 | goto out_unlock; | |
4157 | ||
4158 | retval = security_task_getscheduler(p); | |
4159 | if (retval) | |
4160 | goto out_unlock; | |
4161 | ||
ce5f7f82 PZ |
4162 | if (task_has_rt_policy(p)) |
4163 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4164 | rcu_read_unlock(); |
1da177e4 LT |
4165 | |
4166 | /* | |
4167 | * This one might sleep, we cannot do it with a spinlock held ... | |
4168 | */ | |
4169 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4170 | ||
1da177e4 LT |
4171 | return retval; |
4172 | ||
4173 | out_unlock: | |
5fe85be0 | 4174 | rcu_read_unlock(); |
1da177e4 LT |
4175 | return retval; |
4176 | } | |
4177 | ||
d50dde5a DF |
4178 | static int sched_read_attr(struct sched_attr __user *uattr, |
4179 | struct sched_attr *attr, | |
4180 | unsigned int usize) | |
4181 | { | |
4182 | int ret; | |
4183 | ||
4184 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4185 | return -EFAULT; | |
4186 | ||
4187 | /* | |
4188 | * If we're handed a smaller struct than we know of, | |
4189 | * ensure all the unknown bits are 0 - i.e. old | |
4190 | * user-space does not get uncomplete information. | |
4191 | */ | |
4192 | if (usize < sizeof(*attr)) { | |
4193 | unsigned char *addr; | |
4194 | unsigned char *end; | |
4195 | ||
4196 | addr = (void *)attr + usize; | |
4197 | end = (void *)attr + sizeof(*attr); | |
4198 | ||
4199 | for (; addr < end; addr++) { | |
4200 | if (*addr) | |
22400674 | 4201 | return -EFBIG; |
d50dde5a DF |
4202 | } |
4203 | ||
4204 | attr->size = usize; | |
4205 | } | |
4206 | ||
4efbc454 | 4207 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4208 | if (ret) |
4209 | return -EFAULT; | |
4210 | ||
22400674 | 4211 | return 0; |
d50dde5a DF |
4212 | } |
4213 | ||
4214 | /** | |
aab03e05 | 4215 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4216 | * @pid: the pid in question. |
5778fccf | 4217 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4218 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4219 | * @flags: for future extension. |
d50dde5a | 4220 | */ |
6d35ab48 PZ |
4221 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4222 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4223 | { |
4224 | struct sched_attr attr = { | |
4225 | .size = sizeof(struct sched_attr), | |
4226 | }; | |
4227 | struct task_struct *p; | |
4228 | int retval; | |
4229 | ||
4230 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4231 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4232 | return -EINVAL; |
4233 | ||
4234 | rcu_read_lock(); | |
4235 | p = find_process_by_pid(pid); | |
4236 | retval = -ESRCH; | |
4237 | if (!p) | |
4238 | goto out_unlock; | |
4239 | ||
4240 | retval = security_task_getscheduler(p); | |
4241 | if (retval) | |
4242 | goto out_unlock; | |
4243 | ||
4244 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4245 | if (p->sched_reset_on_fork) |
4246 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4247 | if (task_has_dl_policy(p)) |
4248 | __getparam_dl(p, &attr); | |
4249 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4250 | attr.sched_priority = p->rt_priority; |
4251 | else | |
d0ea0268 | 4252 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4253 | |
4254 | rcu_read_unlock(); | |
4255 | ||
4256 | retval = sched_read_attr(uattr, &attr, size); | |
4257 | return retval; | |
4258 | ||
4259 | out_unlock: | |
4260 | rcu_read_unlock(); | |
4261 | return retval; | |
4262 | } | |
4263 | ||
96f874e2 | 4264 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4265 | { |
5a16f3d3 | 4266 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4267 | struct task_struct *p; |
4268 | int retval; | |
1da177e4 | 4269 | |
23f5d142 | 4270 | rcu_read_lock(); |
1da177e4 LT |
4271 | |
4272 | p = find_process_by_pid(pid); | |
4273 | if (!p) { | |
23f5d142 | 4274 | rcu_read_unlock(); |
1da177e4 LT |
4275 | return -ESRCH; |
4276 | } | |
4277 | ||
23f5d142 | 4278 | /* Prevent p going away */ |
1da177e4 | 4279 | get_task_struct(p); |
23f5d142 | 4280 | rcu_read_unlock(); |
1da177e4 | 4281 | |
14a40ffc TH |
4282 | if (p->flags & PF_NO_SETAFFINITY) { |
4283 | retval = -EINVAL; | |
4284 | goto out_put_task; | |
4285 | } | |
5a16f3d3 RR |
4286 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4287 | retval = -ENOMEM; | |
4288 | goto out_put_task; | |
4289 | } | |
4290 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4291 | retval = -ENOMEM; | |
4292 | goto out_free_cpus_allowed; | |
4293 | } | |
1da177e4 | 4294 | retval = -EPERM; |
4c44aaaf EB |
4295 | if (!check_same_owner(p)) { |
4296 | rcu_read_lock(); | |
4297 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4298 | rcu_read_unlock(); | |
16303ab2 | 4299 | goto out_free_new_mask; |
4c44aaaf EB |
4300 | } |
4301 | rcu_read_unlock(); | |
4302 | } | |
1da177e4 | 4303 | |
b0ae1981 | 4304 | retval = security_task_setscheduler(p); |
e7834f8f | 4305 | if (retval) |
16303ab2 | 4306 | goto out_free_new_mask; |
e7834f8f | 4307 | |
e4099a5e PZ |
4308 | |
4309 | cpuset_cpus_allowed(p, cpus_allowed); | |
4310 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4311 | ||
332ac17e DF |
4312 | /* |
4313 | * Since bandwidth control happens on root_domain basis, | |
4314 | * if admission test is enabled, we only admit -deadline | |
4315 | * tasks allowed to run on all the CPUs in the task's | |
4316 | * root_domain. | |
4317 | */ | |
4318 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4319 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4320 | rcu_read_lock(); | |
4321 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4322 | retval = -EBUSY; |
f1e3a093 | 4323 | rcu_read_unlock(); |
16303ab2 | 4324 | goto out_free_new_mask; |
332ac17e | 4325 | } |
f1e3a093 | 4326 | rcu_read_unlock(); |
332ac17e DF |
4327 | } |
4328 | #endif | |
49246274 | 4329 | again: |
5a16f3d3 | 4330 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4331 | |
8707d8b8 | 4332 | if (!retval) { |
5a16f3d3 RR |
4333 | cpuset_cpus_allowed(p, cpus_allowed); |
4334 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4335 | /* |
4336 | * We must have raced with a concurrent cpuset | |
4337 | * update. Just reset the cpus_allowed to the | |
4338 | * cpuset's cpus_allowed | |
4339 | */ | |
5a16f3d3 | 4340 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4341 | goto again; |
4342 | } | |
4343 | } | |
16303ab2 | 4344 | out_free_new_mask: |
5a16f3d3 RR |
4345 | free_cpumask_var(new_mask); |
4346 | out_free_cpus_allowed: | |
4347 | free_cpumask_var(cpus_allowed); | |
4348 | out_put_task: | |
1da177e4 | 4349 | put_task_struct(p); |
1da177e4 LT |
4350 | return retval; |
4351 | } | |
4352 | ||
4353 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4354 | struct cpumask *new_mask) |
1da177e4 | 4355 | { |
96f874e2 RR |
4356 | if (len < cpumask_size()) |
4357 | cpumask_clear(new_mask); | |
4358 | else if (len > cpumask_size()) | |
4359 | len = cpumask_size(); | |
4360 | ||
1da177e4 LT |
4361 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4362 | } | |
4363 | ||
4364 | /** | |
4365 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4366 | * @pid: pid of the process | |
4367 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4368 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4369 | * |
4370 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4371 | */ |
5add95d4 HC |
4372 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4373 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4374 | { |
5a16f3d3 | 4375 | cpumask_var_t new_mask; |
1da177e4 LT |
4376 | int retval; |
4377 | ||
5a16f3d3 RR |
4378 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4379 | return -ENOMEM; | |
1da177e4 | 4380 | |
5a16f3d3 RR |
4381 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4382 | if (retval == 0) | |
4383 | retval = sched_setaffinity(pid, new_mask); | |
4384 | free_cpumask_var(new_mask); | |
4385 | return retval; | |
1da177e4 LT |
4386 | } |
4387 | ||
96f874e2 | 4388 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4389 | { |
36c8b586 | 4390 | struct task_struct *p; |
31605683 | 4391 | unsigned long flags; |
1da177e4 | 4392 | int retval; |
1da177e4 | 4393 | |
23f5d142 | 4394 | rcu_read_lock(); |
1da177e4 LT |
4395 | |
4396 | retval = -ESRCH; | |
4397 | p = find_process_by_pid(pid); | |
4398 | if (!p) | |
4399 | goto out_unlock; | |
4400 | ||
e7834f8f DQ |
4401 | retval = security_task_getscheduler(p); |
4402 | if (retval) | |
4403 | goto out_unlock; | |
4404 | ||
013fdb80 | 4405 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4406 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4407 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4408 | |
4409 | out_unlock: | |
23f5d142 | 4410 | rcu_read_unlock(); |
1da177e4 | 4411 | |
9531b62f | 4412 | return retval; |
1da177e4 LT |
4413 | } |
4414 | ||
4415 | /** | |
4416 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4417 | * @pid: pid of the process | |
4418 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4419 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4420 | * |
4421 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4422 | */ |
5add95d4 HC |
4423 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4424 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4425 | { |
4426 | int ret; | |
f17c8607 | 4427 | cpumask_var_t mask; |
1da177e4 | 4428 | |
84fba5ec | 4429 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4430 | return -EINVAL; |
4431 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4432 | return -EINVAL; |
4433 | ||
f17c8607 RR |
4434 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4435 | return -ENOMEM; | |
1da177e4 | 4436 | |
f17c8607 RR |
4437 | ret = sched_getaffinity(pid, mask); |
4438 | if (ret == 0) { | |
8bc037fb | 4439 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4440 | |
4441 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4442 | ret = -EFAULT; |
4443 | else | |
cd3d8031 | 4444 | ret = retlen; |
f17c8607 RR |
4445 | } |
4446 | free_cpumask_var(mask); | |
1da177e4 | 4447 | |
f17c8607 | 4448 | return ret; |
1da177e4 LT |
4449 | } |
4450 | ||
4451 | /** | |
4452 | * sys_sched_yield - yield the current processor to other threads. | |
4453 | * | |
dd41f596 IM |
4454 | * This function yields the current CPU to other tasks. If there are no |
4455 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4456 | * |
4457 | * Return: 0. | |
1da177e4 | 4458 | */ |
5add95d4 | 4459 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4460 | { |
70b97a7f | 4461 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4462 | |
2d72376b | 4463 | schedstat_inc(rq, yld_count); |
4530d7ab | 4464 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4465 | |
4466 | /* | |
4467 | * Since we are going to call schedule() anyway, there's | |
4468 | * no need to preempt or enable interrupts: | |
4469 | */ | |
4470 | __release(rq->lock); | |
8a25d5de | 4471 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4472 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4473 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4474 | |
4475 | schedule(); | |
4476 | ||
4477 | return 0; | |
4478 | } | |
4479 | ||
02b67cc3 | 4480 | int __sched _cond_resched(void) |
1da177e4 | 4481 | { |
d86ee480 | 4482 | if (should_resched()) { |
a18b5d01 | 4483 | preempt_schedule_common(); |
1da177e4 LT |
4484 | return 1; |
4485 | } | |
4486 | return 0; | |
4487 | } | |
02b67cc3 | 4488 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4489 | |
4490 | /* | |
613afbf8 | 4491 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4492 | * call schedule, and on return reacquire the lock. |
4493 | * | |
41a2d6cf | 4494 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4495 | * operations here to prevent schedule() from being called twice (once via |
4496 | * spin_unlock(), once by hand). | |
4497 | */ | |
613afbf8 | 4498 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4499 | { |
d86ee480 | 4500 | int resched = should_resched(); |
6df3cecb JK |
4501 | int ret = 0; |
4502 | ||
f607c668 PZ |
4503 | lockdep_assert_held(lock); |
4504 | ||
4a81e832 | 4505 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4506 | spin_unlock(lock); |
d86ee480 | 4507 | if (resched) |
a18b5d01 | 4508 | preempt_schedule_common(); |
95c354fe NP |
4509 | else |
4510 | cpu_relax(); | |
6df3cecb | 4511 | ret = 1; |
1da177e4 | 4512 | spin_lock(lock); |
1da177e4 | 4513 | } |
6df3cecb | 4514 | return ret; |
1da177e4 | 4515 | } |
613afbf8 | 4516 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4517 | |
613afbf8 | 4518 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4519 | { |
4520 | BUG_ON(!in_softirq()); | |
4521 | ||
d86ee480 | 4522 | if (should_resched()) { |
98d82567 | 4523 | local_bh_enable(); |
a18b5d01 | 4524 | preempt_schedule_common(); |
1da177e4 LT |
4525 | local_bh_disable(); |
4526 | return 1; | |
4527 | } | |
4528 | return 0; | |
4529 | } | |
613afbf8 | 4530 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4531 | |
1da177e4 LT |
4532 | /** |
4533 | * yield - yield the current processor to other threads. | |
4534 | * | |
8e3fabfd PZ |
4535 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4536 | * | |
4537 | * The scheduler is at all times free to pick the calling task as the most | |
4538 | * eligible task to run, if removing the yield() call from your code breaks | |
4539 | * it, its already broken. | |
4540 | * | |
4541 | * Typical broken usage is: | |
4542 | * | |
4543 | * while (!event) | |
4544 | * yield(); | |
4545 | * | |
4546 | * where one assumes that yield() will let 'the other' process run that will | |
4547 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4548 | * happen. Never use yield() as a progress guarantee!! | |
4549 | * | |
4550 | * If you want to use yield() to wait for something, use wait_event(). | |
4551 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4552 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4553 | */ |
4554 | void __sched yield(void) | |
4555 | { | |
4556 | set_current_state(TASK_RUNNING); | |
4557 | sys_sched_yield(); | |
4558 | } | |
1da177e4 LT |
4559 | EXPORT_SYMBOL(yield); |
4560 | ||
d95f4122 MG |
4561 | /** |
4562 | * yield_to - yield the current processor to another thread in | |
4563 | * your thread group, or accelerate that thread toward the | |
4564 | * processor it's on. | |
16addf95 RD |
4565 | * @p: target task |
4566 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4567 | * |
4568 | * It's the caller's job to ensure that the target task struct | |
4569 | * can't go away on us before we can do any checks. | |
4570 | * | |
e69f6186 | 4571 | * Return: |
7b270f60 PZ |
4572 | * true (>0) if we indeed boosted the target task. |
4573 | * false (0) if we failed to boost the target. | |
4574 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4575 | */ |
fa93384f | 4576 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4577 | { |
4578 | struct task_struct *curr = current; | |
4579 | struct rq *rq, *p_rq; | |
4580 | unsigned long flags; | |
c3c18640 | 4581 | int yielded = 0; |
d95f4122 MG |
4582 | |
4583 | local_irq_save(flags); | |
4584 | rq = this_rq(); | |
4585 | ||
4586 | again: | |
4587 | p_rq = task_rq(p); | |
7b270f60 PZ |
4588 | /* |
4589 | * If we're the only runnable task on the rq and target rq also | |
4590 | * has only one task, there's absolutely no point in yielding. | |
4591 | */ | |
4592 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4593 | yielded = -ESRCH; | |
4594 | goto out_irq; | |
4595 | } | |
4596 | ||
d95f4122 | 4597 | double_rq_lock(rq, p_rq); |
39e24d8f | 4598 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4599 | double_rq_unlock(rq, p_rq); |
4600 | goto again; | |
4601 | } | |
4602 | ||
4603 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4604 | goto out_unlock; |
d95f4122 MG |
4605 | |
4606 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4607 | goto out_unlock; |
d95f4122 MG |
4608 | |
4609 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4610 | goto out_unlock; |
d95f4122 MG |
4611 | |
4612 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4613 | if (yielded) { |
d95f4122 | 4614 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4615 | /* |
4616 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4617 | * fairness. | |
4618 | */ | |
4619 | if (preempt && rq != p_rq) | |
8875125e | 4620 | resched_curr(p_rq); |
6d1cafd8 | 4621 | } |
d95f4122 | 4622 | |
7b270f60 | 4623 | out_unlock: |
d95f4122 | 4624 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4625 | out_irq: |
d95f4122 MG |
4626 | local_irq_restore(flags); |
4627 | ||
7b270f60 | 4628 | if (yielded > 0) |
d95f4122 MG |
4629 | schedule(); |
4630 | ||
4631 | return yielded; | |
4632 | } | |
4633 | EXPORT_SYMBOL_GPL(yield_to); | |
4634 | ||
1da177e4 | 4635 | /* |
41a2d6cf | 4636 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4637 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4638 | */ |
1da177e4 LT |
4639 | long __sched io_schedule_timeout(long timeout) |
4640 | { | |
9cff8ade N |
4641 | int old_iowait = current->in_iowait; |
4642 | struct rq *rq; | |
1da177e4 LT |
4643 | long ret; |
4644 | ||
9cff8ade | 4645 | current->in_iowait = 1; |
10d784ea | 4646 | blk_schedule_flush_plug(current); |
9cff8ade | 4647 | |
0ff92245 | 4648 | delayacct_blkio_start(); |
9cff8ade | 4649 | rq = raw_rq(); |
1da177e4 LT |
4650 | atomic_inc(&rq->nr_iowait); |
4651 | ret = schedule_timeout(timeout); | |
9cff8ade | 4652 | current->in_iowait = old_iowait; |
1da177e4 | 4653 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4654 | delayacct_blkio_end(); |
9cff8ade | 4655 | |
1da177e4 LT |
4656 | return ret; |
4657 | } | |
9cff8ade | 4658 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4659 | |
4660 | /** | |
4661 | * sys_sched_get_priority_max - return maximum RT priority. | |
4662 | * @policy: scheduling class. | |
4663 | * | |
e69f6186 YB |
4664 | * Return: On success, this syscall returns the maximum |
4665 | * rt_priority that can be used by a given scheduling class. | |
4666 | * On failure, a negative error code is returned. | |
1da177e4 | 4667 | */ |
5add95d4 | 4668 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4669 | { |
4670 | int ret = -EINVAL; | |
4671 | ||
4672 | switch (policy) { | |
4673 | case SCHED_FIFO: | |
4674 | case SCHED_RR: | |
4675 | ret = MAX_USER_RT_PRIO-1; | |
4676 | break; | |
aab03e05 | 4677 | case SCHED_DEADLINE: |
1da177e4 | 4678 | case SCHED_NORMAL: |
b0a9499c | 4679 | case SCHED_BATCH: |
dd41f596 | 4680 | case SCHED_IDLE: |
1da177e4 LT |
4681 | ret = 0; |
4682 | break; | |
4683 | } | |
4684 | return ret; | |
4685 | } | |
4686 | ||
4687 | /** | |
4688 | * sys_sched_get_priority_min - return minimum RT priority. | |
4689 | * @policy: scheduling class. | |
4690 | * | |
e69f6186 YB |
4691 | * Return: On success, this syscall returns the minimum |
4692 | * rt_priority that can be used by a given scheduling class. | |
4693 | * On failure, a negative error code is returned. | |
1da177e4 | 4694 | */ |
5add95d4 | 4695 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4696 | { |
4697 | int ret = -EINVAL; | |
4698 | ||
4699 | switch (policy) { | |
4700 | case SCHED_FIFO: | |
4701 | case SCHED_RR: | |
4702 | ret = 1; | |
4703 | break; | |
aab03e05 | 4704 | case SCHED_DEADLINE: |
1da177e4 | 4705 | case SCHED_NORMAL: |
b0a9499c | 4706 | case SCHED_BATCH: |
dd41f596 | 4707 | case SCHED_IDLE: |
1da177e4 LT |
4708 | ret = 0; |
4709 | } | |
4710 | return ret; | |
4711 | } | |
4712 | ||
4713 | /** | |
4714 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4715 | * @pid: pid of the process. | |
4716 | * @interval: userspace pointer to the timeslice value. | |
4717 | * | |
4718 | * this syscall writes the default timeslice value of a given process | |
4719 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4720 | * |
4721 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4722 | * an error code. | |
1da177e4 | 4723 | */ |
17da2bd9 | 4724 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4725 | struct timespec __user *, interval) |
1da177e4 | 4726 | { |
36c8b586 | 4727 | struct task_struct *p; |
a4ec24b4 | 4728 | unsigned int time_slice; |
dba091b9 TG |
4729 | unsigned long flags; |
4730 | struct rq *rq; | |
3a5c359a | 4731 | int retval; |
1da177e4 | 4732 | struct timespec t; |
1da177e4 LT |
4733 | |
4734 | if (pid < 0) | |
3a5c359a | 4735 | return -EINVAL; |
1da177e4 LT |
4736 | |
4737 | retval = -ESRCH; | |
1a551ae7 | 4738 | rcu_read_lock(); |
1da177e4 LT |
4739 | p = find_process_by_pid(pid); |
4740 | if (!p) | |
4741 | goto out_unlock; | |
4742 | ||
4743 | retval = security_task_getscheduler(p); | |
4744 | if (retval) | |
4745 | goto out_unlock; | |
4746 | ||
dba091b9 | 4747 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4748 | time_slice = 0; |
4749 | if (p->sched_class->get_rr_interval) | |
4750 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4751 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4752 | |
1a551ae7 | 4753 | rcu_read_unlock(); |
a4ec24b4 | 4754 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4755 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4756 | return retval; |
3a5c359a | 4757 | |
1da177e4 | 4758 | out_unlock: |
1a551ae7 | 4759 | rcu_read_unlock(); |
1da177e4 LT |
4760 | return retval; |
4761 | } | |
4762 | ||
7c731e0a | 4763 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4764 | |
82a1fcb9 | 4765 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4766 | { |
1da177e4 | 4767 | unsigned long free = 0; |
4e79752c | 4768 | int ppid; |
1f8a7633 | 4769 | unsigned long state = p->state; |
1da177e4 | 4770 | |
1f8a7633 TH |
4771 | if (state) |
4772 | state = __ffs(state) + 1; | |
28d0686c | 4773 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4774 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4775 | #if BITS_PER_LONG == 32 |
1da177e4 | 4776 | if (state == TASK_RUNNING) |
3df0fc5b | 4777 | printk(KERN_CONT " running "); |
1da177e4 | 4778 | else |
3df0fc5b | 4779 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4780 | #else |
4781 | if (state == TASK_RUNNING) | |
3df0fc5b | 4782 | printk(KERN_CONT " running task "); |
1da177e4 | 4783 | else |
3df0fc5b | 4784 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4785 | #endif |
4786 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4787 | free = stack_not_used(p); |
1da177e4 | 4788 | #endif |
a90e984c | 4789 | ppid = 0; |
4e79752c | 4790 | rcu_read_lock(); |
a90e984c ON |
4791 | if (pid_alive(p)) |
4792 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 4793 | rcu_read_unlock(); |
3df0fc5b | 4794 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4795 | task_pid_nr(p), ppid, |
aa47b7e0 | 4796 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4797 | |
3d1cb205 | 4798 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4799 | show_stack(p, NULL); |
1da177e4 LT |
4800 | } |
4801 | ||
e59e2ae2 | 4802 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4803 | { |
36c8b586 | 4804 | struct task_struct *g, *p; |
1da177e4 | 4805 | |
4bd77321 | 4806 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4807 | printk(KERN_INFO |
4808 | " task PC stack pid father\n"); | |
1da177e4 | 4809 | #else |
3df0fc5b PZ |
4810 | printk(KERN_INFO |
4811 | " task PC stack pid father\n"); | |
1da177e4 | 4812 | #endif |
510f5acc | 4813 | rcu_read_lock(); |
5d07f420 | 4814 | for_each_process_thread(g, p) { |
1da177e4 LT |
4815 | /* |
4816 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4817 | * console might take a lot of time: |
1da177e4 LT |
4818 | */ |
4819 | touch_nmi_watchdog(); | |
39bc89fd | 4820 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4821 | sched_show_task(p); |
5d07f420 | 4822 | } |
1da177e4 | 4823 | |
04c9167f JF |
4824 | touch_all_softlockup_watchdogs(); |
4825 | ||
dd41f596 IM |
4826 | #ifdef CONFIG_SCHED_DEBUG |
4827 | sysrq_sched_debug_show(); | |
4828 | #endif | |
510f5acc | 4829 | rcu_read_unlock(); |
e59e2ae2 IM |
4830 | /* |
4831 | * Only show locks if all tasks are dumped: | |
4832 | */ | |
93335a21 | 4833 | if (!state_filter) |
e59e2ae2 | 4834 | debug_show_all_locks(); |
1da177e4 LT |
4835 | } |
4836 | ||
0db0628d | 4837 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4838 | { |
dd41f596 | 4839 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4840 | } |
4841 | ||
f340c0d1 IM |
4842 | /** |
4843 | * init_idle - set up an idle thread for a given CPU | |
4844 | * @idle: task in question | |
4845 | * @cpu: cpu the idle task belongs to | |
4846 | * | |
4847 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4848 | * flag, to make booting more robust. | |
4849 | */ | |
0db0628d | 4850 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4851 | { |
70b97a7f | 4852 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4853 | unsigned long flags; |
4854 | ||
05fa785c | 4855 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4856 | |
5e1576ed | 4857 | __sched_fork(0, idle); |
06b83b5f | 4858 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4859 | idle->se.exec_start = sched_clock(); |
4860 | ||
1e1b6c51 | 4861 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4862 | /* |
4863 | * We're having a chicken and egg problem, even though we are | |
4864 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4865 | * lockdep check in task_group() will fail. | |
4866 | * | |
4867 | * Similar case to sched_fork(). / Alternatively we could | |
4868 | * use task_rq_lock() here and obtain the other rq->lock. | |
4869 | * | |
4870 | * Silence PROVE_RCU | |
4871 | */ | |
4872 | rcu_read_lock(); | |
dd41f596 | 4873 | __set_task_cpu(idle, cpu); |
6506cf6c | 4874 | rcu_read_unlock(); |
1da177e4 | 4875 | |
1da177e4 | 4876 | rq->curr = rq->idle = idle; |
da0c1e65 | 4877 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4878 | #if defined(CONFIG_SMP) |
4879 | idle->on_cpu = 1; | |
4866cde0 | 4880 | #endif |
05fa785c | 4881 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4882 | |
4883 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4884 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4885 | |
dd41f596 IM |
4886 | /* |
4887 | * The idle tasks have their own, simple scheduling class: | |
4888 | */ | |
4889 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4890 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4891 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4892 | #if defined(CONFIG_SMP) |
4893 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4894 | #endif | |
19978ca6 IM |
4895 | } |
4896 | ||
f82f8042 JL |
4897 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
4898 | const struct cpumask *trial) | |
4899 | { | |
4900 | int ret = 1, trial_cpus; | |
4901 | struct dl_bw *cur_dl_b; | |
4902 | unsigned long flags; | |
4903 | ||
bb2bc55a MG |
4904 | if (!cpumask_weight(cur)) |
4905 | return ret; | |
4906 | ||
75e23e49 | 4907 | rcu_read_lock_sched(); |
f82f8042 JL |
4908 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
4909 | trial_cpus = cpumask_weight(trial); | |
4910 | ||
4911 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
4912 | if (cur_dl_b->bw != -1 && | |
4913 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
4914 | ret = 0; | |
4915 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 4916 | rcu_read_unlock_sched(); |
f82f8042 JL |
4917 | |
4918 | return ret; | |
4919 | } | |
4920 | ||
7f51412a JL |
4921 | int task_can_attach(struct task_struct *p, |
4922 | const struct cpumask *cs_cpus_allowed) | |
4923 | { | |
4924 | int ret = 0; | |
4925 | ||
4926 | /* | |
4927 | * Kthreads which disallow setaffinity shouldn't be moved | |
4928 | * to a new cpuset; we don't want to change their cpu | |
4929 | * affinity and isolating such threads by their set of | |
4930 | * allowed nodes is unnecessary. Thus, cpusets are not | |
4931 | * applicable for such threads. This prevents checking for | |
4932 | * success of set_cpus_allowed_ptr() on all attached tasks | |
4933 | * before cpus_allowed may be changed. | |
4934 | */ | |
4935 | if (p->flags & PF_NO_SETAFFINITY) { | |
4936 | ret = -EINVAL; | |
4937 | goto out; | |
4938 | } | |
4939 | ||
4940 | #ifdef CONFIG_SMP | |
4941 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
4942 | cs_cpus_allowed)) { | |
4943 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
4944 | cs_cpus_allowed); | |
75e23e49 | 4945 | struct dl_bw *dl_b; |
7f51412a JL |
4946 | bool overflow; |
4947 | int cpus; | |
4948 | unsigned long flags; | |
4949 | ||
75e23e49 JL |
4950 | rcu_read_lock_sched(); |
4951 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
4952 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
4953 | cpus = dl_bw_cpus(dest_cpu); | |
4954 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
4955 | if (overflow) | |
4956 | ret = -EBUSY; | |
4957 | else { | |
4958 | /* | |
4959 | * We reserve space for this task in the destination | |
4960 | * root_domain, as we can't fail after this point. | |
4961 | * We will free resources in the source root_domain | |
4962 | * later on (see set_cpus_allowed_dl()). | |
4963 | */ | |
4964 | __dl_add(dl_b, p->dl.dl_bw); | |
4965 | } | |
4966 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 4967 | rcu_read_unlock_sched(); |
7f51412a JL |
4968 | |
4969 | } | |
4970 | #endif | |
4971 | out: | |
4972 | return ret; | |
4973 | } | |
4974 | ||
1da177e4 | 4975 | #ifdef CONFIG_SMP |
1da177e4 | 4976 | |
e6628d5b MG |
4977 | #ifdef CONFIG_NUMA_BALANCING |
4978 | /* Migrate current task p to target_cpu */ | |
4979 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4980 | { | |
4981 | struct migration_arg arg = { p, target_cpu }; | |
4982 | int curr_cpu = task_cpu(p); | |
4983 | ||
4984 | if (curr_cpu == target_cpu) | |
4985 | return 0; | |
4986 | ||
4987 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4988 | return -EINVAL; | |
4989 | ||
4990 | /* TODO: This is not properly updating schedstats */ | |
4991 | ||
286549dc | 4992 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4993 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4994 | } | |
0ec8aa00 PZ |
4995 | |
4996 | /* | |
4997 | * Requeue a task on a given node and accurately track the number of NUMA | |
4998 | * tasks on the runqueues | |
4999 | */ | |
5000 | void sched_setnuma(struct task_struct *p, int nid) | |
5001 | { | |
5002 | struct rq *rq; | |
5003 | unsigned long flags; | |
da0c1e65 | 5004 | bool queued, running; |
0ec8aa00 PZ |
5005 | |
5006 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 5007 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5008 | running = task_current(rq, p); |
5009 | ||
da0c1e65 | 5010 | if (queued) |
0ec8aa00 PZ |
5011 | dequeue_task(rq, p, 0); |
5012 | if (running) | |
f3cd1c4e | 5013 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5014 | |
5015 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
5016 | |
5017 | if (running) | |
5018 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 5019 | if (queued) |
0ec8aa00 PZ |
5020 | enqueue_task(rq, p, 0); |
5021 | task_rq_unlock(rq, p, &flags); | |
5022 | } | |
5cc389bc | 5023 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5024 | |
1da177e4 | 5025 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5026 | /* |
48c5ccae PZ |
5027 | * Ensures that the idle task is using init_mm right before its cpu goes |
5028 | * offline. | |
054b9108 | 5029 | */ |
48c5ccae | 5030 | void idle_task_exit(void) |
1da177e4 | 5031 | { |
48c5ccae | 5032 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5033 | |
48c5ccae | 5034 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5035 | |
a53efe5f | 5036 | if (mm != &init_mm) { |
48c5ccae | 5037 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
5038 | finish_arch_post_lock_switch(); |
5039 | } | |
48c5ccae | 5040 | mmdrop(mm); |
1da177e4 LT |
5041 | } |
5042 | ||
5043 | /* | |
5d180232 PZ |
5044 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5045 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5046 | * nr_active count is stable. | |
5047 | * | |
5048 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5049 | */ |
5d180232 | 5050 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5051 | { |
5d180232 PZ |
5052 | long delta = calc_load_fold_active(rq); |
5053 | if (delta) | |
5054 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5055 | } |
5056 | ||
3f1d2a31 PZ |
5057 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5058 | { | |
5059 | } | |
5060 | ||
5061 | static const struct sched_class fake_sched_class = { | |
5062 | .put_prev_task = put_prev_task_fake, | |
5063 | }; | |
5064 | ||
5065 | static struct task_struct fake_task = { | |
5066 | /* | |
5067 | * Avoid pull_{rt,dl}_task() | |
5068 | */ | |
5069 | .prio = MAX_PRIO + 1, | |
5070 | .sched_class = &fake_sched_class, | |
5071 | }; | |
5072 | ||
48f24c4d | 5073 | /* |
48c5ccae PZ |
5074 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5075 | * try_to_wake_up()->select_task_rq(). | |
5076 | * | |
5077 | * Called with rq->lock held even though we'er in stop_machine() and | |
5078 | * there's no concurrency possible, we hold the required locks anyway | |
5079 | * because of lock validation efforts. | |
1da177e4 | 5080 | */ |
5e16bbc2 | 5081 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5082 | { |
5e16bbc2 | 5083 | struct rq *rq = dead_rq; |
48c5ccae PZ |
5084 | struct task_struct *next, *stop = rq->stop; |
5085 | int dest_cpu; | |
1da177e4 LT |
5086 | |
5087 | /* | |
48c5ccae PZ |
5088 | * Fudge the rq selection such that the below task selection loop |
5089 | * doesn't get stuck on the currently eligible stop task. | |
5090 | * | |
5091 | * We're currently inside stop_machine() and the rq is either stuck | |
5092 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5093 | * either way we should never end up calling schedule() until we're | |
5094 | * done here. | |
1da177e4 | 5095 | */ |
48c5ccae | 5096 | rq->stop = NULL; |
48f24c4d | 5097 | |
77bd3970 FW |
5098 | /* |
5099 | * put_prev_task() and pick_next_task() sched | |
5100 | * class method both need to have an up-to-date | |
5101 | * value of rq->clock[_task] | |
5102 | */ | |
5103 | update_rq_clock(rq); | |
5104 | ||
5e16bbc2 | 5105 | for (;;) { |
48c5ccae PZ |
5106 | /* |
5107 | * There's this thread running, bail when that's the only | |
5108 | * remaining thread. | |
5109 | */ | |
5110 | if (rq->nr_running == 1) | |
dd41f596 | 5111 | break; |
48c5ccae | 5112 | |
cbce1a68 PZ |
5113 | /* |
5114 | * Ensure rq->lock covers the entire task selection | |
5115 | * until the migration. | |
5116 | */ | |
5117 | lockdep_pin_lock(&rq->lock); | |
3f1d2a31 | 5118 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5119 | BUG_ON(!next); |
79c53799 | 5120 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5121 | |
48c5ccae | 5122 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5123 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5124 | |
cbce1a68 | 5125 | lockdep_unpin_lock(&rq->lock); |
5e16bbc2 PZ |
5126 | rq = __migrate_task(rq, next, dest_cpu); |
5127 | if (rq != dead_rq) { | |
5128 | raw_spin_unlock(&rq->lock); | |
5129 | rq = dead_rq; | |
5130 | raw_spin_lock(&rq->lock); | |
5131 | } | |
1da177e4 | 5132 | } |
dce48a84 | 5133 | |
48c5ccae | 5134 | rq->stop = stop; |
dce48a84 | 5135 | } |
1da177e4 LT |
5136 | #endif /* CONFIG_HOTPLUG_CPU */ |
5137 | ||
e692ab53 NP |
5138 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5139 | ||
5140 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5141 | { |
5142 | .procname = "sched_domain", | |
c57baf1e | 5143 | .mode = 0555, |
e0361851 | 5144 | }, |
56992309 | 5145 | {} |
e692ab53 NP |
5146 | }; |
5147 | ||
5148 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5149 | { |
5150 | .procname = "kernel", | |
c57baf1e | 5151 | .mode = 0555, |
e0361851 AD |
5152 | .child = sd_ctl_dir, |
5153 | }, | |
56992309 | 5154 | {} |
e692ab53 NP |
5155 | }; |
5156 | ||
5157 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5158 | { | |
5159 | struct ctl_table *entry = | |
5cf9f062 | 5160 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5161 | |
e692ab53 NP |
5162 | return entry; |
5163 | } | |
5164 | ||
6382bc90 MM |
5165 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5166 | { | |
cd790076 | 5167 | struct ctl_table *entry; |
6382bc90 | 5168 | |
cd790076 MM |
5169 | /* |
5170 | * In the intermediate directories, both the child directory and | |
5171 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5172 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5173 | * static strings and all have proc handlers. |
5174 | */ | |
5175 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5176 | if (entry->child) |
5177 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5178 | if (entry->proc_handler == NULL) |
5179 | kfree(entry->procname); | |
5180 | } | |
6382bc90 MM |
5181 | |
5182 | kfree(*tablep); | |
5183 | *tablep = NULL; | |
5184 | } | |
5185 | ||
201c373e | 5186 | static int min_load_idx = 0; |
fd9b86d3 | 5187 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5188 | |
e692ab53 | 5189 | static void |
e0361851 | 5190 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5191 | const char *procname, void *data, int maxlen, |
201c373e NK |
5192 | umode_t mode, proc_handler *proc_handler, |
5193 | bool load_idx) | |
e692ab53 | 5194 | { |
e692ab53 NP |
5195 | entry->procname = procname; |
5196 | entry->data = data; | |
5197 | entry->maxlen = maxlen; | |
5198 | entry->mode = mode; | |
5199 | entry->proc_handler = proc_handler; | |
201c373e NK |
5200 | |
5201 | if (load_idx) { | |
5202 | entry->extra1 = &min_load_idx; | |
5203 | entry->extra2 = &max_load_idx; | |
5204 | } | |
e692ab53 NP |
5205 | } |
5206 | ||
5207 | static struct ctl_table * | |
5208 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5209 | { | |
37e6bae8 | 5210 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 5211 | |
ad1cdc1d MM |
5212 | if (table == NULL) |
5213 | return NULL; | |
5214 | ||
e0361851 | 5215 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5216 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5217 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5218 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5219 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5220 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5221 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5222 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5223 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5224 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5225 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5226 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5227 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5228 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5229 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5230 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5231 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5232 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5233 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5234 | &sd->cache_nice_tries, |
201c373e | 5235 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5236 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5237 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
5238 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5239 | &sd->max_newidle_lb_cost, | |
5240 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5241 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5242 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5243 | /* &table[13] is terminator */ |
e692ab53 NP |
5244 | |
5245 | return table; | |
5246 | } | |
5247 | ||
be7002e6 | 5248 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5249 | { |
5250 | struct ctl_table *entry, *table; | |
5251 | struct sched_domain *sd; | |
5252 | int domain_num = 0, i; | |
5253 | char buf[32]; | |
5254 | ||
5255 | for_each_domain(cpu, sd) | |
5256 | domain_num++; | |
5257 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5258 | if (table == NULL) |
5259 | return NULL; | |
e692ab53 NP |
5260 | |
5261 | i = 0; | |
5262 | for_each_domain(cpu, sd) { | |
5263 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5264 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5265 | entry->mode = 0555; |
e692ab53 NP |
5266 | entry->child = sd_alloc_ctl_domain_table(sd); |
5267 | entry++; | |
5268 | i++; | |
5269 | } | |
5270 | return table; | |
5271 | } | |
5272 | ||
5273 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5274 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5275 | { |
6ad4c188 | 5276 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5277 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5278 | char buf[32]; | |
5279 | ||
7378547f MM |
5280 | WARN_ON(sd_ctl_dir[0].child); |
5281 | sd_ctl_dir[0].child = entry; | |
5282 | ||
ad1cdc1d MM |
5283 | if (entry == NULL) |
5284 | return; | |
5285 | ||
6ad4c188 | 5286 | for_each_possible_cpu(i) { |
e692ab53 | 5287 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5288 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5289 | entry->mode = 0555; |
e692ab53 | 5290 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5291 | entry++; |
e692ab53 | 5292 | } |
7378547f MM |
5293 | |
5294 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5295 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5296 | } | |
6382bc90 | 5297 | |
7378547f | 5298 | /* may be called multiple times per register */ |
6382bc90 MM |
5299 | static void unregister_sched_domain_sysctl(void) |
5300 | { | |
7378547f MM |
5301 | if (sd_sysctl_header) |
5302 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5303 | sd_sysctl_header = NULL; |
7378547f MM |
5304 | if (sd_ctl_dir[0].child) |
5305 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5306 | } |
e692ab53 | 5307 | #else |
6382bc90 MM |
5308 | static void register_sched_domain_sysctl(void) |
5309 | { | |
5310 | } | |
5311 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5312 | { |
5313 | } | |
5cc389bc | 5314 | #endif /* CONFIG_SCHED_DEBUG && CONFIG_SYSCTL */ |
e692ab53 | 5315 | |
1f11eb6a GH |
5316 | static void set_rq_online(struct rq *rq) |
5317 | { | |
5318 | if (!rq->online) { | |
5319 | const struct sched_class *class; | |
5320 | ||
c6c4927b | 5321 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5322 | rq->online = 1; |
5323 | ||
5324 | for_each_class(class) { | |
5325 | if (class->rq_online) | |
5326 | class->rq_online(rq); | |
5327 | } | |
5328 | } | |
5329 | } | |
5330 | ||
5331 | static void set_rq_offline(struct rq *rq) | |
5332 | { | |
5333 | if (rq->online) { | |
5334 | const struct sched_class *class; | |
5335 | ||
5336 | for_each_class(class) { | |
5337 | if (class->rq_offline) | |
5338 | class->rq_offline(rq); | |
5339 | } | |
5340 | ||
c6c4927b | 5341 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5342 | rq->online = 0; |
5343 | } | |
5344 | } | |
5345 | ||
1da177e4 LT |
5346 | /* |
5347 | * migration_call - callback that gets triggered when a CPU is added. | |
5348 | * Here we can start up the necessary migration thread for the new CPU. | |
5349 | */ | |
0db0628d | 5350 | static int |
48f24c4d | 5351 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5352 | { |
48f24c4d | 5353 | int cpu = (long)hcpu; |
1da177e4 | 5354 | unsigned long flags; |
969c7921 | 5355 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5356 | |
48c5ccae | 5357 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5358 | |
1da177e4 | 5359 | case CPU_UP_PREPARE: |
a468d389 | 5360 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5361 | break; |
48f24c4d | 5362 | |
1da177e4 | 5363 | case CPU_ONLINE: |
1f94ef59 | 5364 | /* Update our root-domain */ |
05fa785c | 5365 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5366 | if (rq->rd) { |
c6c4927b | 5367 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5368 | |
5369 | set_rq_online(rq); | |
1f94ef59 | 5370 | } |
05fa785c | 5371 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5372 | break; |
48f24c4d | 5373 | |
1da177e4 | 5374 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5375 | case CPU_DYING: |
317f3941 | 5376 | sched_ttwu_pending(); |
57d885fe | 5377 | /* Update our root-domain */ |
05fa785c | 5378 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5379 | if (rq->rd) { |
c6c4927b | 5380 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5381 | set_rq_offline(rq); |
57d885fe | 5382 | } |
5e16bbc2 | 5383 | migrate_tasks(rq); |
48c5ccae | 5384 | BUG_ON(rq->nr_running != 1); /* the migration thread */ |
05fa785c | 5385 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5386 | break; |
48c5ccae | 5387 | |
5d180232 | 5388 | case CPU_DEAD: |
f319da0c | 5389 | calc_load_migrate(rq); |
57d885fe | 5390 | break; |
1da177e4 LT |
5391 | #endif |
5392 | } | |
49c022e6 PZ |
5393 | |
5394 | update_max_interval(); | |
5395 | ||
1da177e4 LT |
5396 | return NOTIFY_OK; |
5397 | } | |
5398 | ||
f38b0820 PM |
5399 | /* |
5400 | * Register at high priority so that task migration (migrate_all_tasks) | |
5401 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5402 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5403 | */ |
0db0628d | 5404 | static struct notifier_block migration_notifier = { |
1da177e4 | 5405 | .notifier_call = migration_call, |
50a323b7 | 5406 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5407 | }; |
5408 | ||
6a82b60d | 5409 | static void set_cpu_rq_start_time(void) |
a803f026 CM |
5410 | { |
5411 | int cpu = smp_processor_id(); | |
5412 | struct rq *rq = cpu_rq(cpu); | |
5413 | rq->age_stamp = sched_clock_cpu(cpu); | |
5414 | } | |
5415 | ||
0db0628d | 5416 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5417 | unsigned long action, void *hcpu) |
5418 | { | |
5419 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5420 | case CPU_STARTING: |
5421 | set_cpu_rq_start_time(); | |
5422 | return NOTIFY_OK; | |
3a101d05 TH |
5423 | case CPU_DOWN_FAILED: |
5424 | set_cpu_active((long)hcpu, true); | |
5425 | return NOTIFY_OK; | |
5426 | default: | |
5427 | return NOTIFY_DONE; | |
5428 | } | |
5429 | } | |
5430 | ||
0db0628d | 5431 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5432 | unsigned long action, void *hcpu) |
5433 | { | |
5434 | switch (action & ~CPU_TASKS_FROZEN) { | |
5435 | case CPU_DOWN_PREPARE: | |
3c18d447 | 5436 | set_cpu_active((long)hcpu, false); |
3a101d05 | 5437 | return NOTIFY_OK; |
3c18d447 JL |
5438 | default: |
5439 | return NOTIFY_DONE; | |
3a101d05 TH |
5440 | } |
5441 | } | |
5442 | ||
7babe8db | 5443 | static int __init migration_init(void) |
1da177e4 LT |
5444 | { |
5445 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5446 | int err; |
48f24c4d | 5447 | |
3a101d05 | 5448 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5449 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5450 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5451 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5452 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5453 | |
3a101d05 TH |
5454 | /* Register cpu active notifiers */ |
5455 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5456 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5457 | ||
a004cd42 | 5458 | return 0; |
1da177e4 | 5459 | } |
7babe8db | 5460 | early_initcall(migration_init); |
476f3534 | 5461 | |
4cb98839 PZ |
5462 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5463 | ||
3e9830dc | 5464 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5465 | |
d039ac60 | 5466 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5467 | |
d039ac60 | 5468 | static int __init sched_debug_setup(char *str) |
f6630114 | 5469 | { |
d039ac60 | 5470 | sched_debug_enabled = 1; |
f6630114 MT |
5471 | |
5472 | return 0; | |
5473 | } | |
d039ac60 PZ |
5474 | early_param("sched_debug", sched_debug_setup); |
5475 | ||
5476 | static inline bool sched_debug(void) | |
5477 | { | |
5478 | return sched_debug_enabled; | |
5479 | } | |
f6630114 | 5480 | |
7c16ec58 | 5481 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5482 | struct cpumask *groupmask) |
1da177e4 | 5483 | { |
4dcf6aff | 5484 | struct sched_group *group = sd->groups; |
1da177e4 | 5485 | |
96f874e2 | 5486 | cpumask_clear(groupmask); |
4dcf6aff IM |
5487 | |
5488 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5489 | ||
5490 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5491 | printk("does not load-balance\n"); |
4dcf6aff | 5492 | if (sd->parent) |
3df0fc5b PZ |
5493 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5494 | " has parent"); | |
4dcf6aff | 5495 | return -1; |
41c7ce9a NP |
5496 | } |
5497 | ||
333470ee TH |
5498 | printk(KERN_CONT "span %*pbl level %s\n", |
5499 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5500 | |
758b2cdc | 5501 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5502 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5503 | "CPU%d\n", cpu); | |
4dcf6aff | 5504 | } |
758b2cdc | 5505 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5506 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5507 | " CPU%d\n", cpu); | |
4dcf6aff | 5508 | } |
1da177e4 | 5509 | |
4dcf6aff | 5510 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5511 | do { |
4dcf6aff | 5512 | if (!group) { |
3df0fc5b PZ |
5513 | printk("\n"); |
5514 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5515 | break; |
5516 | } | |
5517 | ||
758b2cdc | 5518 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5519 | printk(KERN_CONT "\n"); |
5520 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5521 | break; |
5522 | } | |
1da177e4 | 5523 | |
cb83b629 PZ |
5524 | if (!(sd->flags & SD_OVERLAP) && |
5525 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5526 | printk(KERN_CONT "\n"); |
5527 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5528 | break; |
5529 | } | |
1da177e4 | 5530 | |
758b2cdc | 5531 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5532 | |
333470ee TH |
5533 | printk(KERN_CONT " %*pbl", |
5534 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5535 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5536 | printk(KERN_CONT " (cpu_capacity = %d)", |
5537 | group->sgc->capacity); | |
381512cf | 5538 | } |
1da177e4 | 5539 | |
4dcf6aff IM |
5540 | group = group->next; |
5541 | } while (group != sd->groups); | |
3df0fc5b | 5542 | printk(KERN_CONT "\n"); |
1da177e4 | 5543 | |
758b2cdc | 5544 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5545 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5546 | |
758b2cdc RR |
5547 | if (sd->parent && |
5548 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5549 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5550 | "of domain->span\n"); | |
4dcf6aff IM |
5551 | return 0; |
5552 | } | |
1da177e4 | 5553 | |
4dcf6aff IM |
5554 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5555 | { | |
5556 | int level = 0; | |
1da177e4 | 5557 | |
d039ac60 | 5558 | if (!sched_debug_enabled) |
f6630114 MT |
5559 | return; |
5560 | ||
4dcf6aff IM |
5561 | if (!sd) { |
5562 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5563 | return; | |
5564 | } | |
1da177e4 | 5565 | |
4dcf6aff IM |
5566 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5567 | ||
5568 | for (;;) { | |
4cb98839 | 5569 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5570 | break; |
1da177e4 LT |
5571 | level++; |
5572 | sd = sd->parent; | |
33859f7f | 5573 | if (!sd) |
4dcf6aff IM |
5574 | break; |
5575 | } | |
1da177e4 | 5576 | } |
6d6bc0ad | 5577 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5578 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5579 | static inline bool sched_debug(void) |
5580 | { | |
5581 | return false; | |
5582 | } | |
6d6bc0ad | 5583 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5584 | |
1a20ff27 | 5585 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5586 | { |
758b2cdc | 5587 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5588 | return 1; |
5589 | ||
5590 | /* Following flags need at least 2 groups */ | |
5591 | if (sd->flags & (SD_LOAD_BALANCE | | |
5592 | SD_BALANCE_NEWIDLE | | |
5593 | SD_BALANCE_FORK | | |
89c4710e | 5594 | SD_BALANCE_EXEC | |
5d4dfddd | 5595 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5596 | SD_SHARE_PKG_RESOURCES | |
5597 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5598 | if (sd->groups != sd->groups->next) |
5599 | return 0; | |
5600 | } | |
5601 | ||
5602 | /* Following flags don't use groups */ | |
c88d5910 | 5603 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5604 | return 0; |
5605 | ||
5606 | return 1; | |
5607 | } | |
5608 | ||
48f24c4d IM |
5609 | static int |
5610 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5611 | { |
5612 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5613 | ||
5614 | if (sd_degenerate(parent)) | |
5615 | return 1; | |
5616 | ||
758b2cdc | 5617 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5618 | return 0; |
5619 | ||
245af2c7 SS |
5620 | /* Flags needing groups don't count if only 1 group in parent */ |
5621 | if (parent->groups == parent->groups->next) { | |
5622 | pflags &= ~(SD_LOAD_BALANCE | | |
5623 | SD_BALANCE_NEWIDLE | | |
5624 | SD_BALANCE_FORK | | |
89c4710e | 5625 | SD_BALANCE_EXEC | |
5d4dfddd | 5626 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5627 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5628 | SD_PREFER_SIBLING | |
5629 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5630 | if (nr_node_ids == 1) |
5631 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5632 | } |
5633 | if (~cflags & pflags) | |
5634 | return 0; | |
5635 | ||
5636 | return 1; | |
5637 | } | |
5638 | ||
dce840a0 | 5639 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5640 | { |
dce840a0 | 5641 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5642 | |
68e74568 | 5643 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5644 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5645 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5646 | free_cpumask_var(rd->rto_mask); |
5647 | free_cpumask_var(rd->online); | |
5648 | free_cpumask_var(rd->span); | |
5649 | kfree(rd); | |
5650 | } | |
5651 | ||
57d885fe GH |
5652 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5653 | { | |
a0490fa3 | 5654 | struct root_domain *old_rd = NULL; |
57d885fe | 5655 | unsigned long flags; |
57d885fe | 5656 | |
05fa785c | 5657 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5658 | |
5659 | if (rq->rd) { | |
a0490fa3 | 5660 | old_rd = rq->rd; |
57d885fe | 5661 | |
c6c4927b | 5662 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5663 | set_rq_offline(rq); |
57d885fe | 5664 | |
c6c4927b | 5665 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5666 | |
a0490fa3 | 5667 | /* |
0515973f | 5668 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5669 | * set old_rd to NULL to skip the freeing later |
5670 | * in this function: | |
5671 | */ | |
5672 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5673 | old_rd = NULL; | |
57d885fe GH |
5674 | } |
5675 | ||
5676 | atomic_inc(&rd->refcount); | |
5677 | rq->rd = rd; | |
5678 | ||
c6c4927b | 5679 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5680 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5681 | set_rq_online(rq); |
57d885fe | 5682 | |
05fa785c | 5683 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5684 | |
5685 | if (old_rd) | |
dce840a0 | 5686 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5687 | } |
5688 | ||
68c38fc3 | 5689 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5690 | { |
5691 | memset(rd, 0, sizeof(*rd)); | |
5692 | ||
68c38fc3 | 5693 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5694 | goto out; |
68c38fc3 | 5695 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5696 | goto free_span; |
1baca4ce | 5697 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5698 | goto free_online; |
1baca4ce JL |
5699 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5700 | goto free_dlo_mask; | |
6e0534f2 | 5701 | |
332ac17e | 5702 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5703 | if (cpudl_init(&rd->cpudl) != 0) |
5704 | goto free_dlo_mask; | |
332ac17e | 5705 | |
68c38fc3 | 5706 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5707 | goto free_rto_mask; |
c6c4927b | 5708 | return 0; |
6e0534f2 | 5709 | |
68e74568 RR |
5710 | free_rto_mask: |
5711 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5712 | free_dlo_mask: |
5713 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5714 | free_online: |
5715 | free_cpumask_var(rd->online); | |
5716 | free_span: | |
5717 | free_cpumask_var(rd->span); | |
0c910d28 | 5718 | out: |
c6c4927b | 5719 | return -ENOMEM; |
57d885fe GH |
5720 | } |
5721 | ||
029632fb PZ |
5722 | /* |
5723 | * By default the system creates a single root-domain with all cpus as | |
5724 | * members (mimicking the global state we have today). | |
5725 | */ | |
5726 | struct root_domain def_root_domain; | |
5727 | ||
57d885fe GH |
5728 | static void init_defrootdomain(void) |
5729 | { | |
68c38fc3 | 5730 | init_rootdomain(&def_root_domain); |
c6c4927b | 5731 | |
57d885fe GH |
5732 | atomic_set(&def_root_domain.refcount, 1); |
5733 | } | |
5734 | ||
dc938520 | 5735 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5736 | { |
5737 | struct root_domain *rd; | |
5738 | ||
5739 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5740 | if (!rd) | |
5741 | return NULL; | |
5742 | ||
68c38fc3 | 5743 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5744 | kfree(rd); |
5745 | return NULL; | |
5746 | } | |
57d885fe GH |
5747 | |
5748 | return rd; | |
5749 | } | |
5750 | ||
63b2ca30 | 5751 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5752 | { |
5753 | struct sched_group *tmp, *first; | |
5754 | ||
5755 | if (!sg) | |
5756 | return; | |
5757 | ||
5758 | first = sg; | |
5759 | do { | |
5760 | tmp = sg->next; | |
5761 | ||
63b2ca30 NP |
5762 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5763 | kfree(sg->sgc); | |
e3589f6c PZ |
5764 | |
5765 | kfree(sg); | |
5766 | sg = tmp; | |
5767 | } while (sg != first); | |
5768 | } | |
5769 | ||
dce840a0 PZ |
5770 | static void free_sched_domain(struct rcu_head *rcu) |
5771 | { | |
5772 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5773 | |
5774 | /* | |
5775 | * If its an overlapping domain it has private groups, iterate and | |
5776 | * nuke them all. | |
5777 | */ | |
5778 | if (sd->flags & SD_OVERLAP) { | |
5779 | free_sched_groups(sd->groups, 1); | |
5780 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5781 | kfree(sd->groups->sgc); |
dce840a0 | 5782 | kfree(sd->groups); |
9c3f75cb | 5783 | } |
dce840a0 PZ |
5784 | kfree(sd); |
5785 | } | |
5786 | ||
5787 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5788 | { | |
5789 | call_rcu(&sd->rcu, free_sched_domain); | |
5790 | } | |
5791 | ||
5792 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5793 | { | |
5794 | for (; sd; sd = sd->parent) | |
5795 | destroy_sched_domain(sd, cpu); | |
5796 | } | |
5797 | ||
518cd623 PZ |
5798 | /* |
5799 | * Keep a special pointer to the highest sched_domain that has | |
5800 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5801 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5802 | * | |
5803 | * Also keep a unique ID per domain (we use the first cpu number in | |
5804 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5805 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5806 | */ |
5807 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5808 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5809 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5810 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5811 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5812 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5813 | |
5814 | static void update_top_cache_domain(int cpu) | |
5815 | { | |
5816 | struct sched_domain *sd; | |
5d4cf996 | 5817 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5818 | int id = cpu; |
7d9ffa89 | 5819 | int size = 1; |
518cd623 PZ |
5820 | |
5821 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5822 | if (sd) { |
518cd623 | 5823 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5824 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5825 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5826 | } |
5d4cf996 | 5827 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5828 | |
5829 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5830 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5831 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5832 | |
5833 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5834 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5835 | |
5836 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5837 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5838 | } |
5839 | ||
1da177e4 | 5840 | /* |
0eab9146 | 5841 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5842 | * hold the hotplug lock. |
5843 | */ | |
0eab9146 IM |
5844 | static void |
5845 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5846 | { |
70b97a7f | 5847 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5848 | struct sched_domain *tmp; |
5849 | ||
5850 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5851 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5852 | struct sched_domain *parent = tmp->parent; |
5853 | if (!parent) | |
5854 | break; | |
f29c9b1c | 5855 | |
1a848870 | 5856 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5857 | tmp->parent = parent->parent; |
1a848870 SS |
5858 | if (parent->parent) |
5859 | parent->parent->child = tmp; | |
10866e62 PZ |
5860 | /* |
5861 | * Transfer SD_PREFER_SIBLING down in case of a | |
5862 | * degenerate parent; the spans match for this | |
5863 | * so the property transfers. | |
5864 | */ | |
5865 | if (parent->flags & SD_PREFER_SIBLING) | |
5866 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5867 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5868 | } else |
5869 | tmp = tmp->parent; | |
245af2c7 SS |
5870 | } |
5871 | ||
1a848870 | 5872 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5873 | tmp = sd; |
245af2c7 | 5874 | sd = sd->parent; |
dce840a0 | 5875 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5876 | if (sd) |
5877 | sd->child = NULL; | |
5878 | } | |
1da177e4 | 5879 | |
4cb98839 | 5880 | sched_domain_debug(sd, cpu); |
1da177e4 | 5881 | |
57d885fe | 5882 | rq_attach_root(rq, rd); |
dce840a0 | 5883 | tmp = rq->sd; |
674311d5 | 5884 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5885 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5886 | |
5887 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5888 | } |
5889 | ||
1da177e4 LT |
5890 | /* Setup the mask of cpus configured for isolated domains */ |
5891 | static int __init isolated_cpu_setup(char *str) | |
5892 | { | |
bdddd296 | 5893 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5894 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5895 | return 1; |
5896 | } | |
5897 | ||
8927f494 | 5898 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5899 | |
49a02c51 | 5900 | struct s_data { |
21d42ccf | 5901 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5902 | struct root_domain *rd; |
5903 | }; | |
5904 | ||
2109b99e | 5905 | enum s_alloc { |
2109b99e | 5906 | sa_rootdomain, |
21d42ccf | 5907 | sa_sd, |
dce840a0 | 5908 | sa_sd_storage, |
2109b99e AH |
5909 | sa_none, |
5910 | }; | |
5911 | ||
c1174876 PZ |
5912 | /* |
5913 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5914 | * domain traversal. | |
5915 | * | |
5916 | * Asymmetric node setups can result in situations where the domain tree is of | |
5917 | * unequal depth, make sure to skip domains that already cover the entire | |
5918 | * range. | |
5919 | * | |
5920 | * In that case build_sched_domains() will have terminated the iteration early | |
5921 | * and our sibling sd spans will be empty. Domains should always include the | |
5922 | * cpu they're built on, so check that. | |
5923 | * | |
5924 | */ | |
5925 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5926 | { | |
5927 | const struct cpumask *span = sched_domain_span(sd); | |
5928 | struct sd_data *sdd = sd->private; | |
5929 | struct sched_domain *sibling; | |
5930 | int i; | |
5931 | ||
5932 | for_each_cpu(i, span) { | |
5933 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5934 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5935 | continue; | |
5936 | ||
5937 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5938 | } | |
5939 | } | |
5940 | ||
5941 | /* | |
5942 | * Return the canonical balance cpu for this group, this is the first cpu | |
5943 | * of this group that's also in the iteration mask. | |
5944 | */ | |
5945 | int group_balance_cpu(struct sched_group *sg) | |
5946 | { | |
5947 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5948 | } | |
5949 | ||
e3589f6c PZ |
5950 | static int |
5951 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5952 | { | |
5953 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5954 | const struct cpumask *span = sched_domain_span(sd); | |
5955 | struct cpumask *covered = sched_domains_tmpmask; | |
5956 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5957 | struct sched_domain *sibling; |
e3589f6c PZ |
5958 | int i; |
5959 | ||
5960 | cpumask_clear(covered); | |
5961 | ||
5962 | for_each_cpu(i, span) { | |
5963 | struct cpumask *sg_span; | |
5964 | ||
5965 | if (cpumask_test_cpu(i, covered)) | |
5966 | continue; | |
5967 | ||
aaecac4a | 5968 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5969 | |
5970 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5971 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5972 | continue; |
5973 | ||
e3589f6c | 5974 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5975 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5976 | |
5977 | if (!sg) | |
5978 | goto fail; | |
5979 | ||
5980 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5981 | if (sibling->child) |
5982 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5983 | else | |
e3589f6c PZ |
5984 | cpumask_set_cpu(i, sg_span); |
5985 | ||
5986 | cpumask_or(covered, covered, sg_span); | |
5987 | ||
63b2ca30 NP |
5988 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5989 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5990 | build_group_mask(sd, sg); |
5991 | ||
c3decf0d | 5992 | /* |
63b2ca30 | 5993 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5994 | * domains and no possible iteration will get us here, we won't |
5995 | * die on a /0 trap. | |
5996 | */ | |
ca8ce3d0 | 5997 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 5998 | |
c1174876 PZ |
5999 | /* |
6000 | * Make sure the first group of this domain contains the | |
6001 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6002 | * breaks. See update_sg_lb_stats(). | |
6003 | */ | |
74a5ce20 | 6004 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6005 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6006 | groups = sg; |
6007 | ||
6008 | if (!first) | |
6009 | first = sg; | |
6010 | if (last) | |
6011 | last->next = sg; | |
6012 | last = sg; | |
6013 | last->next = first; | |
6014 | } | |
6015 | sd->groups = groups; | |
6016 | ||
6017 | return 0; | |
6018 | ||
6019 | fail: | |
6020 | free_sched_groups(first, 0); | |
6021 | ||
6022 | return -ENOMEM; | |
6023 | } | |
6024 | ||
dce840a0 | 6025 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6026 | { |
dce840a0 PZ |
6027 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6028 | struct sched_domain *child = sd->child; | |
1da177e4 | 6029 | |
dce840a0 PZ |
6030 | if (child) |
6031 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6032 | |
9c3f75cb | 6033 | if (sg) { |
dce840a0 | 6034 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
6035 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
6036 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 6037 | } |
dce840a0 PZ |
6038 | |
6039 | return cpu; | |
1e9f28fa | 6040 | } |
1e9f28fa | 6041 | |
01a08546 | 6042 | /* |
dce840a0 PZ |
6043 | * build_sched_groups will build a circular linked list of the groups |
6044 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6045 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6046 | * |
6047 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6048 | */ |
e3589f6c PZ |
6049 | static int |
6050 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6051 | { |
dce840a0 PZ |
6052 | struct sched_group *first = NULL, *last = NULL; |
6053 | struct sd_data *sdd = sd->private; | |
6054 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6055 | struct cpumask *covered; |
dce840a0 | 6056 | int i; |
9c1cfda2 | 6057 | |
e3589f6c PZ |
6058 | get_group(cpu, sdd, &sd->groups); |
6059 | atomic_inc(&sd->groups->ref); | |
6060 | ||
0936629f | 6061 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6062 | return 0; |
6063 | ||
f96225fd PZ |
6064 | lockdep_assert_held(&sched_domains_mutex); |
6065 | covered = sched_domains_tmpmask; | |
6066 | ||
dce840a0 | 6067 | cpumask_clear(covered); |
6711cab4 | 6068 | |
dce840a0 PZ |
6069 | for_each_cpu(i, span) { |
6070 | struct sched_group *sg; | |
cd08e923 | 6071 | int group, j; |
6711cab4 | 6072 | |
dce840a0 PZ |
6073 | if (cpumask_test_cpu(i, covered)) |
6074 | continue; | |
6711cab4 | 6075 | |
cd08e923 | 6076 | group = get_group(i, sdd, &sg); |
c1174876 | 6077 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6078 | |
dce840a0 PZ |
6079 | for_each_cpu(j, span) { |
6080 | if (get_group(j, sdd, NULL) != group) | |
6081 | continue; | |
0601a88d | 6082 | |
dce840a0 PZ |
6083 | cpumask_set_cpu(j, covered); |
6084 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6085 | } | |
0601a88d | 6086 | |
dce840a0 PZ |
6087 | if (!first) |
6088 | first = sg; | |
6089 | if (last) | |
6090 | last->next = sg; | |
6091 | last = sg; | |
6092 | } | |
6093 | last->next = first; | |
e3589f6c PZ |
6094 | |
6095 | return 0; | |
0601a88d | 6096 | } |
51888ca2 | 6097 | |
89c4710e | 6098 | /* |
63b2ca30 | 6099 | * Initialize sched groups cpu_capacity. |
89c4710e | 6100 | * |
63b2ca30 | 6101 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6102 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6103 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6104 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6105 | * group having more cpu_capacity will pickup more load compared to the | |
6106 | * group having less cpu_capacity. | |
89c4710e | 6107 | */ |
63b2ca30 | 6108 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6109 | { |
e3589f6c | 6110 | struct sched_group *sg = sd->groups; |
89c4710e | 6111 | |
94c95ba6 | 6112 | WARN_ON(!sg); |
e3589f6c PZ |
6113 | |
6114 | do { | |
6115 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6116 | sg = sg->next; | |
6117 | } while (sg != sd->groups); | |
89c4710e | 6118 | |
c1174876 | 6119 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6120 | return; |
aae6d3dd | 6121 | |
63b2ca30 NP |
6122 | update_group_capacity(sd, cpu); |
6123 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6124 | } |
6125 | ||
7c16ec58 MT |
6126 | /* |
6127 | * Initializers for schedule domains | |
6128 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6129 | */ | |
6130 | ||
1d3504fc | 6131 | static int default_relax_domain_level = -1; |
60495e77 | 6132 | int sched_domain_level_max; |
1d3504fc HS |
6133 | |
6134 | static int __init setup_relax_domain_level(char *str) | |
6135 | { | |
a841f8ce DS |
6136 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6137 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6138 | |
1d3504fc HS |
6139 | return 1; |
6140 | } | |
6141 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6142 | ||
6143 | static void set_domain_attribute(struct sched_domain *sd, | |
6144 | struct sched_domain_attr *attr) | |
6145 | { | |
6146 | int request; | |
6147 | ||
6148 | if (!attr || attr->relax_domain_level < 0) { | |
6149 | if (default_relax_domain_level < 0) | |
6150 | return; | |
6151 | else | |
6152 | request = default_relax_domain_level; | |
6153 | } else | |
6154 | request = attr->relax_domain_level; | |
6155 | if (request < sd->level) { | |
6156 | /* turn off idle balance on this domain */ | |
c88d5910 | 6157 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6158 | } else { |
6159 | /* turn on idle balance on this domain */ | |
c88d5910 | 6160 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6161 | } |
6162 | } | |
6163 | ||
54ab4ff4 PZ |
6164 | static void __sdt_free(const struct cpumask *cpu_map); |
6165 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6166 | ||
2109b99e AH |
6167 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6168 | const struct cpumask *cpu_map) | |
6169 | { | |
6170 | switch (what) { | |
2109b99e | 6171 | case sa_rootdomain: |
822ff793 PZ |
6172 | if (!atomic_read(&d->rd->refcount)) |
6173 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6174 | case sa_sd: |
6175 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6176 | case sa_sd_storage: |
54ab4ff4 | 6177 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6178 | case sa_none: |
6179 | break; | |
6180 | } | |
6181 | } | |
3404c8d9 | 6182 | |
2109b99e AH |
6183 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6184 | const struct cpumask *cpu_map) | |
6185 | { | |
dce840a0 PZ |
6186 | memset(d, 0, sizeof(*d)); |
6187 | ||
54ab4ff4 PZ |
6188 | if (__sdt_alloc(cpu_map)) |
6189 | return sa_sd_storage; | |
dce840a0 PZ |
6190 | d->sd = alloc_percpu(struct sched_domain *); |
6191 | if (!d->sd) | |
6192 | return sa_sd_storage; | |
2109b99e | 6193 | d->rd = alloc_rootdomain(); |
dce840a0 | 6194 | if (!d->rd) |
21d42ccf | 6195 | return sa_sd; |
2109b99e AH |
6196 | return sa_rootdomain; |
6197 | } | |
57d885fe | 6198 | |
dce840a0 PZ |
6199 | /* |
6200 | * NULL the sd_data elements we've used to build the sched_domain and | |
6201 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6202 | * will not free the data we're using. | |
6203 | */ | |
6204 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6205 | { | |
6206 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6207 | |
6208 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6209 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6210 | ||
e3589f6c | 6211 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6212 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6213 | |
63b2ca30 NP |
6214 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6215 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6216 | } |
6217 | ||
cb83b629 | 6218 | #ifdef CONFIG_NUMA |
cb83b629 | 6219 | static int sched_domains_numa_levels; |
e3fe70b1 | 6220 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6221 | static int *sched_domains_numa_distance; |
9942f79b | 6222 | int sched_max_numa_distance; |
cb83b629 PZ |
6223 | static struct cpumask ***sched_domains_numa_masks; |
6224 | static int sched_domains_curr_level; | |
143e1e28 | 6225 | #endif |
cb83b629 | 6226 | |
143e1e28 VG |
6227 | /* |
6228 | * SD_flags allowed in topology descriptions. | |
6229 | * | |
5d4dfddd | 6230 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6231 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6232 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6233 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6234 | * |
6235 | * Odd one out: | |
6236 | * SD_ASYM_PACKING - describes SMT quirks | |
6237 | */ | |
6238 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6239 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6240 | SD_SHARE_PKG_RESOURCES | \ |
6241 | SD_NUMA | \ | |
d77b3ed5 VG |
6242 | SD_ASYM_PACKING | \ |
6243 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6244 | |
6245 | static struct sched_domain * | |
143e1e28 | 6246 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6247 | { |
6248 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6249 | int sd_weight, sd_flags = 0; |
6250 | ||
6251 | #ifdef CONFIG_NUMA | |
6252 | /* | |
6253 | * Ugly hack to pass state to sd_numa_mask()... | |
6254 | */ | |
6255 | sched_domains_curr_level = tl->numa_level; | |
6256 | #endif | |
6257 | ||
6258 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6259 | ||
6260 | if (tl->sd_flags) | |
6261 | sd_flags = (*tl->sd_flags)(); | |
6262 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6263 | "wrong sd_flags in topology description\n")) | |
6264 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6265 | |
6266 | *sd = (struct sched_domain){ | |
6267 | .min_interval = sd_weight, | |
6268 | .max_interval = 2*sd_weight, | |
6269 | .busy_factor = 32, | |
870a0bb5 | 6270 | .imbalance_pct = 125, |
143e1e28 VG |
6271 | |
6272 | .cache_nice_tries = 0, | |
6273 | .busy_idx = 0, | |
6274 | .idle_idx = 0, | |
cb83b629 PZ |
6275 | .newidle_idx = 0, |
6276 | .wake_idx = 0, | |
6277 | .forkexec_idx = 0, | |
6278 | ||
6279 | .flags = 1*SD_LOAD_BALANCE | |
6280 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6281 | | 1*SD_BALANCE_EXEC |
6282 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6283 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6284 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6285 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6286 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6287 | | 0*SD_SERIALIZE |
cb83b629 | 6288 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6289 | | 0*SD_NUMA |
6290 | | sd_flags | |
cb83b629 | 6291 | , |
143e1e28 | 6292 | |
cb83b629 PZ |
6293 | .last_balance = jiffies, |
6294 | .balance_interval = sd_weight, | |
143e1e28 | 6295 | .smt_gain = 0, |
2b4cfe64 JL |
6296 | .max_newidle_lb_cost = 0, |
6297 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6298 | #ifdef CONFIG_SCHED_DEBUG |
6299 | .name = tl->name, | |
6300 | #endif | |
cb83b629 | 6301 | }; |
cb83b629 PZ |
6302 | |
6303 | /* | |
143e1e28 | 6304 | * Convert topological properties into behaviour. |
cb83b629 | 6305 | */ |
143e1e28 | 6306 | |
5d4dfddd | 6307 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6308 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6309 | sd->imbalance_pct = 110; |
6310 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6311 | |
6312 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6313 | sd->imbalance_pct = 117; | |
6314 | sd->cache_nice_tries = 1; | |
6315 | sd->busy_idx = 2; | |
6316 | ||
6317 | #ifdef CONFIG_NUMA | |
6318 | } else if (sd->flags & SD_NUMA) { | |
6319 | sd->cache_nice_tries = 2; | |
6320 | sd->busy_idx = 3; | |
6321 | sd->idle_idx = 2; | |
6322 | ||
6323 | sd->flags |= SD_SERIALIZE; | |
6324 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6325 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6326 | SD_BALANCE_FORK | | |
6327 | SD_WAKE_AFFINE); | |
6328 | } | |
6329 | ||
6330 | #endif | |
6331 | } else { | |
6332 | sd->flags |= SD_PREFER_SIBLING; | |
6333 | sd->cache_nice_tries = 1; | |
6334 | sd->busy_idx = 2; | |
6335 | sd->idle_idx = 1; | |
6336 | } | |
6337 | ||
6338 | sd->private = &tl->data; | |
cb83b629 PZ |
6339 | |
6340 | return sd; | |
6341 | } | |
6342 | ||
143e1e28 VG |
6343 | /* |
6344 | * Topology list, bottom-up. | |
6345 | */ | |
6346 | static struct sched_domain_topology_level default_topology[] = { | |
6347 | #ifdef CONFIG_SCHED_SMT | |
6348 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6349 | #endif | |
6350 | #ifdef CONFIG_SCHED_MC | |
6351 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6352 | #endif |
6353 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6354 | { NULL, }, | |
6355 | }; | |
6356 | ||
6357 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6358 | ||
6359 | #define for_each_sd_topology(tl) \ | |
6360 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6361 | ||
6362 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6363 | { | |
6364 | sched_domain_topology = tl; | |
6365 | } | |
6366 | ||
6367 | #ifdef CONFIG_NUMA | |
6368 | ||
cb83b629 PZ |
6369 | static const struct cpumask *sd_numa_mask(int cpu) |
6370 | { | |
6371 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6372 | } | |
6373 | ||
d039ac60 PZ |
6374 | static void sched_numa_warn(const char *str) |
6375 | { | |
6376 | static int done = false; | |
6377 | int i,j; | |
6378 | ||
6379 | if (done) | |
6380 | return; | |
6381 | ||
6382 | done = true; | |
6383 | ||
6384 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6385 | ||
6386 | for (i = 0; i < nr_node_ids; i++) { | |
6387 | printk(KERN_WARNING " "); | |
6388 | for (j = 0; j < nr_node_ids; j++) | |
6389 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6390 | printk(KERN_CONT "\n"); | |
6391 | } | |
6392 | printk(KERN_WARNING "\n"); | |
6393 | } | |
6394 | ||
9942f79b | 6395 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6396 | { |
6397 | int i; | |
6398 | ||
6399 | if (distance == node_distance(0, 0)) | |
6400 | return true; | |
6401 | ||
6402 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6403 | if (sched_domains_numa_distance[i] == distance) | |
6404 | return true; | |
6405 | } | |
6406 | ||
6407 | return false; | |
6408 | } | |
6409 | ||
e3fe70b1 RR |
6410 | /* |
6411 | * A system can have three types of NUMA topology: | |
6412 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6413 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6414 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6415 | * | |
6416 | * The difference between a glueless mesh topology and a backplane | |
6417 | * topology lies in whether communication between not directly | |
6418 | * connected nodes goes through intermediary nodes (where programs | |
6419 | * could run), or through backplane controllers. This affects | |
6420 | * placement of programs. | |
6421 | * | |
6422 | * The type of topology can be discerned with the following tests: | |
6423 | * - If the maximum distance between any nodes is 1 hop, the system | |
6424 | * is directly connected. | |
6425 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6426 | * there is an intermediary node C, which is < N hops away from both | |
6427 | * nodes A and B, the system is a glueless mesh. | |
6428 | */ | |
6429 | static void init_numa_topology_type(void) | |
6430 | { | |
6431 | int a, b, c, n; | |
6432 | ||
6433 | n = sched_max_numa_distance; | |
6434 | ||
6435 | if (n <= 1) | |
6436 | sched_numa_topology_type = NUMA_DIRECT; | |
6437 | ||
6438 | for_each_online_node(a) { | |
6439 | for_each_online_node(b) { | |
6440 | /* Find two nodes furthest removed from each other. */ | |
6441 | if (node_distance(a, b) < n) | |
6442 | continue; | |
6443 | ||
6444 | /* Is there an intermediary node between a and b? */ | |
6445 | for_each_online_node(c) { | |
6446 | if (node_distance(a, c) < n && | |
6447 | node_distance(b, c) < n) { | |
6448 | sched_numa_topology_type = | |
6449 | NUMA_GLUELESS_MESH; | |
6450 | return; | |
6451 | } | |
6452 | } | |
6453 | ||
6454 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6455 | return; | |
6456 | } | |
6457 | } | |
6458 | } | |
6459 | ||
cb83b629 PZ |
6460 | static void sched_init_numa(void) |
6461 | { | |
6462 | int next_distance, curr_distance = node_distance(0, 0); | |
6463 | struct sched_domain_topology_level *tl; | |
6464 | int level = 0; | |
6465 | int i, j, k; | |
6466 | ||
cb83b629 PZ |
6467 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6468 | if (!sched_domains_numa_distance) | |
6469 | return; | |
6470 | ||
6471 | /* | |
6472 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6473 | * unique distances in the node_distance() table. | |
6474 | * | |
6475 | * Assumes node_distance(0,j) includes all distances in | |
6476 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6477 | */ |
6478 | next_distance = curr_distance; | |
6479 | for (i = 0; i < nr_node_ids; i++) { | |
6480 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6481 | for (k = 0; k < nr_node_ids; k++) { |
6482 | int distance = node_distance(i, k); | |
6483 | ||
6484 | if (distance > curr_distance && | |
6485 | (distance < next_distance || | |
6486 | next_distance == curr_distance)) | |
6487 | next_distance = distance; | |
6488 | ||
6489 | /* | |
6490 | * While not a strong assumption it would be nice to know | |
6491 | * about cases where if node A is connected to B, B is not | |
6492 | * equally connected to A. | |
6493 | */ | |
6494 | if (sched_debug() && node_distance(k, i) != distance) | |
6495 | sched_numa_warn("Node-distance not symmetric"); | |
6496 | ||
6497 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6498 | sched_numa_warn("Node-0 not representative"); | |
6499 | } | |
6500 | if (next_distance != curr_distance) { | |
6501 | sched_domains_numa_distance[level++] = next_distance; | |
6502 | sched_domains_numa_levels = level; | |
6503 | curr_distance = next_distance; | |
6504 | } else break; | |
cb83b629 | 6505 | } |
d039ac60 PZ |
6506 | |
6507 | /* | |
6508 | * In case of sched_debug() we verify the above assumption. | |
6509 | */ | |
6510 | if (!sched_debug()) | |
6511 | break; | |
cb83b629 | 6512 | } |
c123588b AR |
6513 | |
6514 | if (!level) | |
6515 | return; | |
6516 | ||
cb83b629 PZ |
6517 | /* |
6518 | * 'level' contains the number of unique distances, excluding the | |
6519 | * identity distance node_distance(i,i). | |
6520 | * | |
28b4a521 | 6521 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6522 | * numbers. |
6523 | */ | |
6524 | ||
5f7865f3 TC |
6525 | /* |
6526 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6527 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6528 | * the array will contain less then 'level' members. This could be | |
6529 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6530 | * in other functions. | |
6531 | * | |
6532 | * We reset it to 'level' at the end of this function. | |
6533 | */ | |
6534 | sched_domains_numa_levels = 0; | |
6535 | ||
cb83b629 PZ |
6536 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6537 | if (!sched_domains_numa_masks) | |
6538 | return; | |
6539 | ||
6540 | /* | |
6541 | * Now for each level, construct a mask per node which contains all | |
6542 | * cpus of nodes that are that many hops away from us. | |
6543 | */ | |
6544 | for (i = 0; i < level; i++) { | |
6545 | sched_domains_numa_masks[i] = | |
6546 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6547 | if (!sched_domains_numa_masks[i]) | |
6548 | return; | |
6549 | ||
6550 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6551 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6552 | if (!mask) |
6553 | return; | |
6554 | ||
6555 | sched_domains_numa_masks[i][j] = mask; | |
6556 | ||
6557 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6558 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6559 | continue; |
6560 | ||
6561 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6562 | } | |
6563 | } | |
6564 | } | |
6565 | ||
143e1e28 VG |
6566 | /* Compute default topology size */ |
6567 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6568 | ||
c515db8c | 6569 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6570 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6571 | if (!tl) | |
6572 | return; | |
6573 | ||
6574 | /* | |
6575 | * Copy the default topology bits.. | |
6576 | */ | |
143e1e28 VG |
6577 | for (i = 0; sched_domain_topology[i].mask; i++) |
6578 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6579 | |
6580 | /* | |
6581 | * .. and append 'j' levels of NUMA goodness. | |
6582 | */ | |
6583 | for (j = 0; j < level; i++, j++) { | |
6584 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6585 | .mask = sd_numa_mask, |
143e1e28 | 6586 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6587 | .flags = SDTL_OVERLAP, |
6588 | .numa_level = j, | |
143e1e28 | 6589 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6590 | }; |
6591 | } | |
6592 | ||
6593 | sched_domain_topology = tl; | |
5f7865f3 TC |
6594 | |
6595 | sched_domains_numa_levels = level; | |
9942f79b | 6596 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6597 | |
6598 | init_numa_topology_type(); | |
cb83b629 | 6599 | } |
301a5cba TC |
6600 | |
6601 | static void sched_domains_numa_masks_set(int cpu) | |
6602 | { | |
6603 | int i, j; | |
6604 | int node = cpu_to_node(cpu); | |
6605 | ||
6606 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6607 | for (j = 0; j < nr_node_ids; j++) { | |
6608 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6609 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6610 | } | |
6611 | } | |
6612 | } | |
6613 | ||
6614 | static void sched_domains_numa_masks_clear(int cpu) | |
6615 | { | |
6616 | int i, j; | |
6617 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6618 | for (j = 0; j < nr_node_ids; j++) | |
6619 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6620 | } | |
6621 | } | |
6622 | ||
6623 | /* | |
6624 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6625 | * are onlined. | |
6626 | */ | |
6627 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6628 | unsigned long action, | |
6629 | void *hcpu) | |
6630 | { | |
6631 | int cpu = (long)hcpu; | |
6632 | ||
6633 | switch (action & ~CPU_TASKS_FROZEN) { | |
6634 | case CPU_ONLINE: | |
6635 | sched_domains_numa_masks_set(cpu); | |
6636 | break; | |
6637 | ||
6638 | case CPU_DEAD: | |
6639 | sched_domains_numa_masks_clear(cpu); | |
6640 | break; | |
6641 | ||
6642 | default: | |
6643 | return NOTIFY_DONE; | |
6644 | } | |
6645 | ||
6646 | return NOTIFY_OK; | |
cb83b629 PZ |
6647 | } |
6648 | #else | |
6649 | static inline void sched_init_numa(void) | |
6650 | { | |
6651 | } | |
301a5cba TC |
6652 | |
6653 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6654 | unsigned long action, | |
6655 | void *hcpu) | |
6656 | { | |
6657 | return 0; | |
6658 | } | |
cb83b629 PZ |
6659 | #endif /* CONFIG_NUMA */ |
6660 | ||
54ab4ff4 PZ |
6661 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6662 | { | |
6663 | struct sched_domain_topology_level *tl; | |
6664 | int j; | |
6665 | ||
27723a68 | 6666 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6667 | struct sd_data *sdd = &tl->data; |
6668 | ||
6669 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6670 | if (!sdd->sd) | |
6671 | return -ENOMEM; | |
6672 | ||
6673 | sdd->sg = alloc_percpu(struct sched_group *); | |
6674 | if (!sdd->sg) | |
6675 | return -ENOMEM; | |
6676 | ||
63b2ca30 NP |
6677 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6678 | if (!sdd->sgc) | |
9c3f75cb PZ |
6679 | return -ENOMEM; |
6680 | ||
54ab4ff4 PZ |
6681 | for_each_cpu(j, cpu_map) { |
6682 | struct sched_domain *sd; | |
6683 | struct sched_group *sg; | |
63b2ca30 | 6684 | struct sched_group_capacity *sgc; |
54ab4ff4 | 6685 | |
5cc389bc | 6686 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
54ab4ff4 PZ |
6687 | GFP_KERNEL, cpu_to_node(j)); |
6688 | if (!sd) | |
6689 | return -ENOMEM; | |
6690 | ||
6691 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6692 | ||
6693 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6694 | GFP_KERNEL, cpu_to_node(j)); | |
6695 | if (!sg) | |
6696 | return -ENOMEM; | |
6697 | ||
30b4e9eb IM |
6698 | sg->next = sg; |
6699 | ||
54ab4ff4 | 6700 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6701 | |
63b2ca30 | 6702 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6703 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6704 | if (!sgc) |
9c3f75cb PZ |
6705 | return -ENOMEM; |
6706 | ||
63b2ca30 | 6707 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6708 | } |
6709 | } | |
6710 | ||
6711 | return 0; | |
6712 | } | |
6713 | ||
6714 | static void __sdt_free(const struct cpumask *cpu_map) | |
6715 | { | |
6716 | struct sched_domain_topology_level *tl; | |
6717 | int j; | |
6718 | ||
27723a68 | 6719 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6720 | struct sd_data *sdd = &tl->data; |
6721 | ||
6722 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6723 | struct sched_domain *sd; |
6724 | ||
6725 | if (sdd->sd) { | |
6726 | sd = *per_cpu_ptr(sdd->sd, j); | |
6727 | if (sd && (sd->flags & SD_OVERLAP)) | |
6728 | free_sched_groups(sd->groups, 0); | |
6729 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6730 | } | |
6731 | ||
6732 | if (sdd->sg) | |
6733 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6734 | if (sdd->sgc) |
6735 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6736 | } |
6737 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6738 | sdd->sd = NULL; |
54ab4ff4 | 6739 | free_percpu(sdd->sg); |
fb2cf2c6 | 6740 | sdd->sg = NULL; |
63b2ca30 NP |
6741 | free_percpu(sdd->sgc); |
6742 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6743 | } |
6744 | } | |
6745 | ||
2c402dc3 | 6746 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6747 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6748 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6749 | { |
143e1e28 | 6750 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6751 | if (!sd) |
d069b916 | 6752 | return child; |
2c402dc3 | 6753 | |
2c402dc3 | 6754 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6755 | if (child) { |
6756 | sd->level = child->level + 1; | |
6757 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6758 | child->parent = sd; |
c75e0128 | 6759 | sd->child = child; |
6ae72dff PZ |
6760 | |
6761 | if (!cpumask_subset(sched_domain_span(child), | |
6762 | sched_domain_span(sd))) { | |
6763 | pr_err("BUG: arch topology borken\n"); | |
6764 | #ifdef CONFIG_SCHED_DEBUG | |
6765 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6766 | child->name, sd->name); | |
6767 | #endif | |
6768 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6769 | cpumask_or(sched_domain_span(sd), | |
6770 | sched_domain_span(sd), | |
6771 | sched_domain_span(child)); | |
6772 | } | |
6773 | ||
60495e77 | 6774 | } |
a841f8ce | 6775 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6776 | |
6777 | return sd; | |
6778 | } | |
6779 | ||
2109b99e AH |
6780 | /* |
6781 | * Build sched domains for a given set of cpus and attach the sched domains | |
6782 | * to the individual cpus | |
6783 | */ | |
dce840a0 PZ |
6784 | static int build_sched_domains(const struct cpumask *cpu_map, |
6785 | struct sched_domain_attr *attr) | |
2109b99e | 6786 | { |
1c632169 | 6787 | enum s_alloc alloc_state; |
dce840a0 | 6788 | struct sched_domain *sd; |
2109b99e | 6789 | struct s_data d; |
822ff793 | 6790 | int i, ret = -ENOMEM; |
9c1cfda2 | 6791 | |
2109b99e AH |
6792 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6793 | if (alloc_state != sa_rootdomain) | |
6794 | goto error; | |
9c1cfda2 | 6795 | |
dce840a0 | 6796 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6797 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6798 | struct sched_domain_topology_level *tl; |
6799 | ||
3bd65a80 | 6800 | sd = NULL; |
27723a68 | 6801 | for_each_sd_topology(tl) { |
4a850cbe | 6802 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6803 | if (tl == sched_domain_topology) |
6804 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6805 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6806 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6807 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6808 | break; | |
e3589f6c | 6809 | } |
dce840a0 PZ |
6810 | } |
6811 | ||
6812 | /* Build the groups for the domains */ | |
6813 | for_each_cpu(i, cpu_map) { | |
6814 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6815 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6816 | if (sd->flags & SD_OVERLAP) { |
6817 | if (build_overlap_sched_groups(sd, i)) | |
6818 | goto error; | |
6819 | } else { | |
6820 | if (build_sched_groups(sd, i)) | |
6821 | goto error; | |
6822 | } | |
1cf51902 | 6823 | } |
a06dadbe | 6824 | } |
9c1cfda2 | 6825 | |
ced549fa | 6826 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6827 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6828 | if (!cpumask_test_cpu(i, cpu_map)) | |
6829 | continue; | |
9c1cfda2 | 6830 | |
dce840a0 PZ |
6831 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6832 | claim_allocations(i, sd); | |
63b2ca30 | 6833 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6834 | } |
f712c0c7 | 6835 | } |
9c1cfda2 | 6836 | |
1da177e4 | 6837 | /* Attach the domains */ |
dce840a0 | 6838 | rcu_read_lock(); |
abcd083a | 6839 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6840 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6841 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6842 | } |
dce840a0 | 6843 | rcu_read_unlock(); |
51888ca2 | 6844 | |
822ff793 | 6845 | ret = 0; |
51888ca2 | 6846 | error: |
2109b99e | 6847 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6848 | return ret; |
1da177e4 | 6849 | } |
029190c5 | 6850 | |
acc3f5d7 | 6851 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6852 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6853 | static struct sched_domain_attr *dattr_cur; |
6854 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6855 | |
6856 | /* | |
6857 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6858 | * cpumask) fails, then fallback to a single sched domain, |
6859 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6860 | */ |
4212823f | 6861 | static cpumask_var_t fallback_doms; |
029190c5 | 6862 | |
ee79d1bd HC |
6863 | /* |
6864 | * arch_update_cpu_topology lets virtualized architectures update the | |
6865 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6866 | * or 0 if it stayed the same. | |
6867 | */ | |
52f5684c | 6868 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6869 | { |
ee79d1bd | 6870 | return 0; |
22e52b07 HC |
6871 | } |
6872 | ||
acc3f5d7 RR |
6873 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6874 | { | |
6875 | int i; | |
6876 | cpumask_var_t *doms; | |
6877 | ||
6878 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6879 | if (!doms) | |
6880 | return NULL; | |
6881 | for (i = 0; i < ndoms; i++) { | |
6882 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6883 | free_sched_domains(doms, i); | |
6884 | return NULL; | |
6885 | } | |
6886 | } | |
6887 | return doms; | |
6888 | } | |
6889 | ||
6890 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6891 | { | |
6892 | unsigned int i; | |
6893 | for (i = 0; i < ndoms; i++) | |
6894 | free_cpumask_var(doms[i]); | |
6895 | kfree(doms); | |
6896 | } | |
6897 | ||
1a20ff27 | 6898 | /* |
41a2d6cf | 6899 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6900 | * For now this just excludes isolated cpus, but could be used to |
6901 | * exclude other special cases in the future. | |
1a20ff27 | 6902 | */ |
c4a8849a | 6903 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6904 | { |
7378547f MM |
6905 | int err; |
6906 | ||
22e52b07 | 6907 | arch_update_cpu_topology(); |
029190c5 | 6908 | ndoms_cur = 1; |
acc3f5d7 | 6909 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6910 | if (!doms_cur) |
acc3f5d7 RR |
6911 | doms_cur = &fallback_doms; |
6912 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6913 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6914 | register_sched_domain_sysctl(); |
7378547f MM |
6915 | |
6916 | return err; | |
1a20ff27 DG |
6917 | } |
6918 | ||
1a20ff27 DG |
6919 | /* |
6920 | * Detach sched domains from a group of cpus specified in cpu_map | |
6921 | * These cpus will now be attached to the NULL domain | |
6922 | */ | |
96f874e2 | 6923 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6924 | { |
6925 | int i; | |
6926 | ||
dce840a0 | 6927 | rcu_read_lock(); |
abcd083a | 6928 | for_each_cpu(i, cpu_map) |
57d885fe | 6929 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6930 | rcu_read_unlock(); |
1a20ff27 DG |
6931 | } |
6932 | ||
1d3504fc HS |
6933 | /* handle null as "default" */ |
6934 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6935 | struct sched_domain_attr *new, int idx_new) | |
6936 | { | |
6937 | struct sched_domain_attr tmp; | |
6938 | ||
6939 | /* fast path */ | |
6940 | if (!new && !cur) | |
6941 | return 1; | |
6942 | ||
6943 | tmp = SD_ATTR_INIT; | |
6944 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6945 | new ? (new + idx_new) : &tmp, | |
6946 | sizeof(struct sched_domain_attr)); | |
6947 | } | |
6948 | ||
029190c5 PJ |
6949 | /* |
6950 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6951 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6952 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6953 | * It destroys each deleted domain and builds each new domain. | |
6954 | * | |
acc3f5d7 | 6955 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6956 | * The masks don't intersect (don't overlap.) We should setup one |
6957 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6958 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6959 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6960 | * it as it is. | |
6961 | * | |
acc3f5d7 RR |
6962 | * The passed in 'doms_new' should be allocated using |
6963 | * alloc_sched_domains. This routine takes ownership of it and will | |
6964 | * free_sched_domains it when done with it. If the caller failed the | |
6965 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6966 | * and partition_sched_domains() will fallback to the single partition | |
6967 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6968 | * |
96f874e2 | 6969 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6970 | * ndoms_new == 0 is a special case for destroying existing domains, |
6971 | * and it will not create the default domain. | |
dfb512ec | 6972 | * |
029190c5 PJ |
6973 | * Call with hotplug lock held |
6974 | */ | |
acc3f5d7 | 6975 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6976 | struct sched_domain_attr *dattr_new) |
029190c5 | 6977 | { |
dfb512ec | 6978 | int i, j, n; |
d65bd5ec | 6979 | int new_topology; |
029190c5 | 6980 | |
712555ee | 6981 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6982 | |
7378547f MM |
6983 | /* always unregister in case we don't destroy any domains */ |
6984 | unregister_sched_domain_sysctl(); | |
6985 | ||
d65bd5ec HC |
6986 | /* Let architecture update cpu core mappings. */ |
6987 | new_topology = arch_update_cpu_topology(); | |
6988 | ||
dfb512ec | 6989 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6990 | |
6991 | /* Destroy deleted domains */ | |
6992 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6993 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6994 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6995 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6996 | goto match1; |
6997 | } | |
6998 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6999 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7000 | match1: |
7001 | ; | |
7002 | } | |
7003 | ||
c8d2d47a | 7004 | n = ndoms_cur; |
e761b772 | 7005 | if (doms_new == NULL) { |
c8d2d47a | 7006 | n = 0; |
acc3f5d7 | 7007 | doms_new = &fallback_doms; |
6ad4c188 | 7008 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7009 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7010 | } |
7011 | ||
029190c5 PJ |
7012 | /* Build new domains */ |
7013 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 7014 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7015 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7016 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7017 | goto match2; |
7018 | } | |
7019 | /* no match - add a new doms_new */ | |
dce840a0 | 7020 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7021 | match2: |
7022 | ; | |
7023 | } | |
7024 | ||
7025 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7026 | if (doms_cur != &fallback_doms) |
7027 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7028 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7029 | doms_cur = doms_new; |
1d3504fc | 7030 | dattr_cur = dattr_new; |
029190c5 | 7031 | ndoms_cur = ndoms_new; |
7378547f MM |
7032 | |
7033 | register_sched_domain_sysctl(); | |
a1835615 | 7034 | |
712555ee | 7035 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7036 | } |
7037 | ||
d35be8ba SB |
7038 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7039 | ||
1da177e4 | 7040 | /* |
3a101d05 TH |
7041 | * Update cpusets according to cpu_active mask. If cpusets are |
7042 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7043 | * around partition_sched_domains(). | |
d35be8ba SB |
7044 | * |
7045 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7046 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7047 | */ |
0b2e918a TH |
7048 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7049 | void *hcpu) | |
e761b772 | 7050 | { |
d35be8ba SB |
7051 | switch (action) { |
7052 | case CPU_ONLINE_FROZEN: | |
7053 | case CPU_DOWN_FAILED_FROZEN: | |
7054 | ||
7055 | /* | |
7056 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7057 | * resume sequence. As long as this is not the last online | |
7058 | * operation in the resume sequence, just build a single sched | |
7059 | * domain, ignoring cpusets. | |
7060 | */ | |
7061 | num_cpus_frozen--; | |
7062 | if (likely(num_cpus_frozen)) { | |
7063 | partition_sched_domains(1, NULL, NULL); | |
7064 | break; | |
7065 | } | |
7066 | ||
7067 | /* | |
7068 | * This is the last CPU online operation. So fall through and | |
7069 | * restore the original sched domains by considering the | |
7070 | * cpuset configurations. | |
7071 | */ | |
7072 | ||
e761b772 | 7073 | case CPU_ONLINE: |
7ddf96b0 | 7074 | cpuset_update_active_cpus(true); |
d35be8ba | 7075 | break; |
3a101d05 TH |
7076 | default: |
7077 | return NOTIFY_DONE; | |
7078 | } | |
d35be8ba | 7079 | return NOTIFY_OK; |
3a101d05 | 7080 | } |
e761b772 | 7081 | |
0b2e918a TH |
7082 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7083 | void *hcpu) | |
3a101d05 | 7084 | { |
3c18d447 JL |
7085 | unsigned long flags; |
7086 | long cpu = (long)hcpu; | |
7087 | struct dl_bw *dl_b; | |
533445c6 OS |
7088 | bool overflow; |
7089 | int cpus; | |
3c18d447 | 7090 | |
533445c6 | 7091 | switch (action) { |
3a101d05 | 7092 | case CPU_DOWN_PREPARE: |
533445c6 OS |
7093 | rcu_read_lock_sched(); |
7094 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7095 | |
533445c6 OS |
7096 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7097 | cpus = dl_bw_cpus(cpu); | |
7098 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7099 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7100 | |
533445c6 | 7101 | rcu_read_unlock_sched(); |
3c18d447 | 7102 | |
533445c6 OS |
7103 | if (overflow) |
7104 | return notifier_from_errno(-EBUSY); | |
7ddf96b0 | 7105 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7106 | break; |
7107 | case CPU_DOWN_PREPARE_FROZEN: | |
7108 | num_cpus_frozen++; | |
7109 | partition_sched_domains(1, NULL, NULL); | |
7110 | break; | |
e761b772 MK |
7111 | default: |
7112 | return NOTIFY_DONE; | |
7113 | } | |
d35be8ba | 7114 | return NOTIFY_OK; |
e761b772 | 7115 | } |
e761b772 | 7116 | |
1da177e4 LT |
7117 | void __init sched_init_smp(void) |
7118 | { | |
dcc30a35 RR |
7119 | cpumask_var_t non_isolated_cpus; |
7120 | ||
7121 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7122 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7123 | |
8cb9764f CM |
7124 | /* nohz_full won't take effect without isolating the cpus. */ |
7125 | tick_nohz_full_add_cpus_to(cpu_isolated_map); | |
7126 | ||
cb83b629 PZ |
7127 | sched_init_numa(); |
7128 | ||
6acce3ef PZ |
7129 | /* |
7130 | * There's no userspace yet to cause hotplug operations; hence all the | |
7131 | * cpu masks are stable and all blatant races in the below code cannot | |
7132 | * happen. | |
7133 | */ | |
712555ee | 7134 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7135 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7136 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7137 | if (cpumask_empty(non_isolated_cpus)) | |
7138 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7139 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7140 | |
301a5cba | 7141 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7142 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7143 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7144 | |
b328ca18 | 7145 | init_hrtick(); |
5c1e1767 NP |
7146 | |
7147 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7148 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7149 | BUG(); |
19978ca6 | 7150 | sched_init_granularity(); |
dcc30a35 | 7151 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7152 | |
0e3900e6 | 7153 | init_sched_rt_class(); |
1baca4ce | 7154 | init_sched_dl_class(); |
1da177e4 LT |
7155 | } |
7156 | #else | |
7157 | void __init sched_init_smp(void) | |
7158 | { | |
19978ca6 | 7159 | sched_init_granularity(); |
1da177e4 LT |
7160 | } |
7161 | #endif /* CONFIG_SMP */ | |
7162 | ||
7163 | int in_sched_functions(unsigned long addr) | |
7164 | { | |
1da177e4 LT |
7165 | return in_lock_functions(addr) || |
7166 | (addr >= (unsigned long)__sched_text_start | |
7167 | && addr < (unsigned long)__sched_text_end); | |
7168 | } | |
7169 | ||
029632fb | 7170 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7171 | /* |
7172 | * Default task group. | |
7173 | * Every task in system belongs to this group at bootup. | |
7174 | */ | |
029632fb | 7175 | struct task_group root_task_group; |
35cf4e50 | 7176 | LIST_HEAD(task_groups); |
052f1dc7 | 7177 | #endif |
6f505b16 | 7178 | |
e6252c3e | 7179 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7180 | |
1da177e4 LT |
7181 | void __init sched_init(void) |
7182 | { | |
dd41f596 | 7183 | int i, j; |
434d53b0 MT |
7184 | unsigned long alloc_size = 0, ptr; |
7185 | ||
7186 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7187 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7188 | #endif | |
7189 | #ifdef CONFIG_RT_GROUP_SCHED | |
7190 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7191 | #endif | |
434d53b0 | 7192 | if (alloc_size) { |
36b7b6d4 | 7193 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7194 | |
7195 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7196 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7197 | ptr += nr_cpu_ids * sizeof(void **); |
7198 | ||
07e06b01 | 7199 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7200 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7201 | |
6d6bc0ad | 7202 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7203 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7204 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7205 | ptr += nr_cpu_ids * sizeof(void **); |
7206 | ||
07e06b01 | 7207 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7208 | ptr += nr_cpu_ids * sizeof(void **); |
7209 | ||
6d6bc0ad | 7210 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7211 | } |
df7c8e84 | 7212 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7213 | for_each_possible_cpu(i) { |
7214 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7215 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7216 | } |
b74e6278 | 7217 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7218 | |
332ac17e DF |
7219 | init_rt_bandwidth(&def_rt_bandwidth, |
7220 | global_rt_period(), global_rt_runtime()); | |
7221 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7222 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7223 | |
57d885fe GH |
7224 | #ifdef CONFIG_SMP |
7225 | init_defrootdomain(); | |
7226 | #endif | |
7227 | ||
d0b27fa7 | 7228 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7229 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7230 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7231 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7232 | |
7c941438 | 7233 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7234 | list_add(&root_task_group.list, &task_groups); |
7235 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7236 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7237 | autogroup_init(&init_task); |
54c707e9 | 7238 | |
7c941438 | 7239 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7240 | |
0a945022 | 7241 | for_each_possible_cpu(i) { |
70b97a7f | 7242 | struct rq *rq; |
1da177e4 LT |
7243 | |
7244 | rq = cpu_rq(i); | |
05fa785c | 7245 | raw_spin_lock_init(&rq->lock); |
7897986b | 7246 | rq->nr_running = 0; |
dce48a84 TG |
7247 | rq->calc_load_active = 0; |
7248 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7249 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7250 | init_rt_rq(&rq->rt); |
7251 | init_dl_rq(&rq->dl); | |
dd41f596 | 7252 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7253 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7254 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7255 | /* |
07e06b01 | 7256 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7257 | * |
7258 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7259 | * gets 100% of the cpu resources in the system. This overall | |
7260 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7261 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7262 | * based on each entity's (task or task-group's) weight |
7263 | * (se->load.weight). | |
7264 | * | |
07e06b01 | 7265 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7266 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7267 | * then A0's share of the cpu resource is: | |
7268 | * | |
0d905bca | 7269 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7270 | * |
07e06b01 YZ |
7271 | * We achieve this by letting root_task_group's tasks sit |
7272 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7273 | */ |
ab84d31e | 7274 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7275 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7276 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7277 | ||
7278 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7279 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7280 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7281 | #endif |
1da177e4 | 7282 | |
dd41f596 IM |
7283 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7284 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7285 | |
7286 | rq->last_load_update_tick = jiffies; | |
7287 | ||
1da177e4 | 7288 | #ifdef CONFIG_SMP |
41c7ce9a | 7289 | rq->sd = NULL; |
57d885fe | 7290 | rq->rd = NULL; |
ca6d75e6 | 7291 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7292 | rq->balance_callback = NULL; |
1da177e4 | 7293 | rq->active_balance = 0; |
dd41f596 | 7294 | rq->next_balance = jiffies; |
1da177e4 | 7295 | rq->push_cpu = 0; |
0a2966b4 | 7296 | rq->cpu = i; |
1f11eb6a | 7297 | rq->online = 0; |
eae0c9df MG |
7298 | rq->idle_stamp = 0; |
7299 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7300 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7301 | |
7302 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7303 | ||
dc938520 | 7304 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7305 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7306 | rq->nohz_flags = 0; |
83cd4fe2 | 7307 | #endif |
265f22a9 FW |
7308 | #ifdef CONFIG_NO_HZ_FULL |
7309 | rq->last_sched_tick = 0; | |
7310 | #endif | |
1da177e4 | 7311 | #endif |
8f4d37ec | 7312 | init_rq_hrtick(rq); |
1da177e4 | 7313 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7314 | } |
7315 | ||
2dd73a4f | 7316 | set_load_weight(&init_task); |
b50f60ce | 7317 | |
e107be36 AK |
7318 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7319 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7320 | #endif | |
7321 | ||
1da177e4 LT |
7322 | /* |
7323 | * The boot idle thread does lazy MMU switching as well: | |
7324 | */ | |
7325 | atomic_inc(&init_mm.mm_count); | |
7326 | enter_lazy_tlb(&init_mm, current); | |
7327 | ||
1b537c7d YD |
7328 | /* |
7329 | * During early bootup we pretend to be a normal task: | |
7330 | */ | |
7331 | current->sched_class = &fair_sched_class; | |
7332 | ||
1da177e4 LT |
7333 | /* |
7334 | * Make us the idle thread. Technically, schedule() should not be | |
7335 | * called from this thread, however somewhere below it might be, | |
7336 | * but because we are the idle thread, we just pick up running again | |
7337 | * when this runqueue becomes "idle". | |
7338 | */ | |
7339 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7340 | |
7341 | calc_load_update = jiffies + LOAD_FREQ; | |
7342 | ||
bf4d83f6 | 7343 | #ifdef CONFIG_SMP |
4cb98839 | 7344 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7345 | /* May be allocated at isolcpus cmdline parse time */ |
7346 | if (cpu_isolated_map == NULL) | |
7347 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7348 | idle_thread_set_boot_cpu(); |
a803f026 | 7349 | set_cpu_rq_start_time(); |
029632fb PZ |
7350 | #endif |
7351 | init_sched_fair_class(); | |
6a7b3dc3 | 7352 | |
6892b75e | 7353 | scheduler_running = 1; |
1da177e4 LT |
7354 | } |
7355 | ||
d902db1e | 7356 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7357 | static inline int preempt_count_equals(int preempt_offset) |
7358 | { | |
234da7bc | 7359 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7360 | |
4ba8216c | 7361 | return (nested == preempt_offset); |
e4aafea2 FW |
7362 | } |
7363 | ||
d894837f | 7364 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7365 | { |
8eb23b9f PZ |
7366 | /* |
7367 | * Blocking primitives will set (and therefore destroy) current->state, | |
7368 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7369 | * otherwise we will destroy state. | |
7370 | */ | |
00845eb9 | 7371 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7372 | "do not call blocking ops when !TASK_RUNNING; " |
7373 | "state=%lx set at [<%p>] %pS\n", | |
7374 | current->state, | |
7375 | (void *)current->task_state_change, | |
00845eb9 | 7376 | (void *)current->task_state_change); |
8eb23b9f | 7377 | |
3427445a PZ |
7378 | ___might_sleep(file, line, preempt_offset); |
7379 | } | |
7380 | EXPORT_SYMBOL(__might_sleep); | |
7381 | ||
7382 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7383 | { |
1da177e4 LT |
7384 | static unsigned long prev_jiffy; /* ratelimiting */ |
7385 | ||
b3fbab05 | 7386 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7387 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7388 | !is_idle_task(current)) || | |
e4aafea2 | 7389 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7390 | return; |
7391 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7392 | return; | |
7393 | prev_jiffy = jiffies; | |
7394 | ||
3df0fc5b PZ |
7395 | printk(KERN_ERR |
7396 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7397 | file, line); | |
7398 | printk(KERN_ERR | |
7399 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7400 | in_atomic(), irqs_disabled(), | |
7401 | current->pid, current->comm); | |
aef745fc | 7402 | |
a8b686b3 ES |
7403 | if (task_stack_end_corrupted(current)) |
7404 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7405 | ||
aef745fc IM |
7406 | debug_show_held_locks(current); |
7407 | if (irqs_disabled()) | |
7408 | print_irqtrace_events(current); | |
8f47b187 TG |
7409 | #ifdef CONFIG_DEBUG_PREEMPT |
7410 | if (!preempt_count_equals(preempt_offset)) { | |
7411 | pr_err("Preemption disabled at:"); | |
7412 | print_ip_sym(current->preempt_disable_ip); | |
7413 | pr_cont("\n"); | |
7414 | } | |
7415 | #endif | |
aef745fc | 7416 | dump_stack(); |
1da177e4 | 7417 | } |
3427445a | 7418 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7419 | #endif |
7420 | ||
7421 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7422 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7423 | { |
dbc7f069 | 7424 | struct task_struct *g, *p; |
d50dde5a DF |
7425 | struct sched_attr attr = { |
7426 | .sched_policy = SCHED_NORMAL, | |
7427 | }; | |
1da177e4 | 7428 | |
3472eaa1 | 7429 | read_lock(&tasklist_lock); |
5d07f420 | 7430 | for_each_process_thread(g, p) { |
178be793 IM |
7431 | /* |
7432 | * Only normalize user tasks: | |
7433 | */ | |
3472eaa1 | 7434 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7435 | continue; |
7436 | ||
6cfb0d5d | 7437 | p->se.exec_start = 0; |
6cfb0d5d | 7438 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7439 | p->se.statistics.wait_start = 0; |
7440 | p->se.statistics.sleep_start = 0; | |
7441 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7442 | #endif |
dd41f596 | 7443 | |
aab03e05 | 7444 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7445 | /* |
7446 | * Renice negative nice level userspace | |
7447 | * tasks back to 0: | |
7448 | */ | |
3472eaa1 | 7449 | if (task_nice(p) < 0) |
dd41f596 | 7450 | set_user_nice(p, 0); |
1da177e4 | 7451 | continue; |
dd41f596 | 7452 | } |
1da177e4 | 7453 | |
dbc7f069 | 7454 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7455 | } |
3472eaa1 | 7456 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7457 | } |
7458 | ||
7459 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7460 | |
67fc4e0c | 7461 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7462 | /* |
67fc4e0c | 7463 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7464 | * |
7465 | * They can only be called when the whole system has been | |
7466 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7467 | * activity can take place. Using them for anything else would | |
7468 | * be a serious bug, and as a result, they aren't even visible | |
7469 | * under any other configuration. | |
7470 | */ | |
7471 | ||
7472 | /** | |
7473 | * curr_task - return the current task for a given cpu. | |
7474 | * @cpu: the processor in question. | |
7475 | * | |
7476 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7477 | * |
7478 | * Return: The current task for @cpu. | |
1df5c10a | 7479 | */ |
36c8b586 | 7480 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7481 | { |
7482 | return cpu_curr(cpu); | |
7483 | } | |
7484 | ||
67fc4e0c JW |
7485 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7486 | ||
7487 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7488 | /** |
7489 | * set_curr_task - set the current task for a given cpu. | |
7490 | * @cpu: the processor in question. | |
7491 | * @p: the task pointer to set. | |
7492 | * | |
7493 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7494 | * are serviced on a separate stack. It allows the architecture to switch the |
7495 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7496 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7497 | * and caller must save the original value of the current task (see | |
7498 | * curr_task() above) and restore that value before reenabling interrupts and | |
7499 | * re-starting the system. | |
7500 | * | |
7501 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7502 | */ | |
36c8b586 | 7503 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7504 | { |
7505 | cpu_curr(cpu) = p; | |
7506 | } | |
7507 | ||
7508 | #endif | |
29f59db3 | 7509 | |
7c941438 | 7510 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7511 | /* task_group_lock serializes the addition/removal of task groups */ |
7512 | static DEFINE_SPINLOCK(task_group_lock); | |
7513 | ||
bccbe08a PZ |
7514 | static void free_sched_group(struct task_group *tg) |
7515 | { | |
7516 | free_fair_sched_group(tg); | |
7517 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7518 | autogroup_free(tg); |
bccbe08a PZ |
7519 | kfree(tg); |
7520 | } | |
7521 | ||
7522 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7523 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7524 | { |
7525 | struct task_group *tg; | |
bccbe08a PZ |
7526 | |
7527 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7528 | if (!tg) | |
7529 | return ERR_PTR(-ENOMEM); | |
7530 | ||
ec7dc8ac | 7531 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7532 | goto err; |
7533 | ||
ec7dc8ac | 7534 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7535 | goto err; |
7536 | ||
ace783b9 LZ |
7537 | return tg; |
7538 | ||
7539 | err: | |
7540 | free_sched_group(tg); | |
7541 | return ERR_PTR(-ENOMEM); | |
7542 | } | |
7543 | ||
7544 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7545 | { | |
7546 | unsigned long flags; | |
7547 | ||
8ed36996 | 7548 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7549 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7550 | |
7551 | WARN_ON(!parent); /* root should already exist */ | |
7552 | ||
7553 | tg->parent = parent; | |
f473aa5e | 7554 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7555 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7556 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7557 | } |
7558 | ||
9b5b7751 | 7559 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7560 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7561 | { |
29f59db3 | 7562 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7563 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7564 | } |
7565 | ||
9b5b7751 | 7566 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7567 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7568 | { |
7569 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7570 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7571 | } | |
7572 | ||
7573 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7574 | { |
8ed36996 | 7575 | unsigned long flags; |
9b5b7751 | 7576 | int i; |
29f59db3 | 7577 | |
3d4b47b4 PZ |
7578 | /* end participation in shares distribution */ |
7579 | for_each_possible_cpu(i) | |
bccbe08a | 7580 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7581 | |
7582 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7583 | list_del_rcu(&tg->list); |
f473aa5e | 7584 | list_del_rcu(&tg->siblings); |
8ed36996 | 7585 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7586 | } |
7587 | ||
9b5b7751 | 7588 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7589 | * The caller of this function should have put the task in its new group |
7590 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7591 | * reflect its new group. | |
9b5b7751 SV |
7592 | */ |
7593 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7594 | { |
8323f26c | 7595 | struct task_group *tg; |
da0c1e65 | 7596 | int queued, running; |
29f59db3 SV |
7597 | unsigned long flags; |
7598 | struct rq *rq; | |
7599 | ||
7600 | rq = task_rq_lock(tsk, &flags); | |
7601 | ||
051a1d1a | 7602 | running = task_current(rq, tsk); |
da0c1e65 | 7603 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7604 | |
da0c1e65 | 7605 | if (queued) |
29f59db3 | 7606 | dequeue_task(rq, tsk, 0); |
0e1f3483 | 7607 | if (unlikely(running)) |
f3cd1c4e | 7608 | put_prev_task(rq, tsk); |
29f59db3 | 7609 | |
f7b8a47d KT |
7610 | /* |
7611 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7612 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7613 | * to prevent lockdep warnings. | |
7614 | */ | |
7615 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7616 | struct task_group, css); |
7617 | tg = autogroup_task_group(tsk, tg); | |
7618 | tsk->sched_task_group = tg; | |
7619 | ||
810b3817 | 7620 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7621 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7622 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7623 | else |
810b3817 | 7624 | #endif |
b2b5ce02 | 7625 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7626 | |
0e1f3483 HS |
7627 | if (unlikely(running)) |
7628 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7629 | if (queued) |
371fd7e7 | 7630 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7631 | |
0122ec5b | 7632 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7633 | } |
7c941438 | 7634 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7635 | |
a790de99 PT |
7636 | #ifdef CONFIG_RT_GROUP_SCHED |
7637 | /* | |
7638 | * Ensure that the real time constraints are schedulable. | |
7639 | */ | |
7640 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7641 | |
9a7e0b18 PZ |
7642 | /* Must be called with tasklist_lock held */ |
7643 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7644 | { |
9a7e0b18 | 7645 | struct task_struct *g, *p; |
b40b2e8e | 7646 | |
1fe89e1b PZ |
7647 | /* |
7648 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7649 | */ | |
7650 | if (task_group_is_autogroup(tg)) | |
7651 | return 0; | |
7652 | ||
5d07f420 | 7653 | for_each_process_thread(g, p) { |
8651c658 | 7654 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7655 | return 1; |
5d07f420 | 7656 | } |
b40b2e8e | 7657 | |
9a7e0b18 PZ |
7658 | return 0; |
7659 | } | |
b40b2e8e | 7660 | |
9a7e0b18 PZ |
7661 | struct rt_schedulable_data { |
7662 | struct task_group *tg; | |
7663 | u64 rt_period; | |
7664 | u64 rt_runtime; | |
7665 | }; | |
b40b2e8e | 7666 | |
a790de99 | 7667 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7668 | { |
7669 | struct rt_schedulable_data *d = data; | |
7670 | struct task_group *child; | |
7671 | unsigned long total, sum = 0; | |
7672 | u64 period, runtime; | |
b40b2e8e | 7673 | |
9a7e0b18 PZ |
7674 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7675 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7676 | |
9a7e0b18 PZ |
7677 | if (tg == d->tg) { |
7678 | period = d->rt_period; | |
7679 | runtime = d->rt_runtime; | |
b40b2e8e | 7680 | } |
b40b2e8e | 7681 | |
4653f803 PZ |
7682 | /* |
7683 | * Cannot have more runtime than the period. | |
7684 | */ | |
7685 | if (runtime > period && runtime != RUNTIME_INF) | |
7686 | return -EINVAL; | |
6f505b16 | 7687 | |
4653f803 PZ |
7688 | /* |
7689 | * Ensure we don't starve existing RT tasks. | |
7690 | */ | |
9a7e0b18 PZ |
7691 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7692 | return -EBUSY; | |
6f505b16 | 7693 | |
9a7e0b18 | 7694 | total = to_ratio(period, runtime); |
6f505b16 | 7695 | |
4653f803 PZ |
7696 | /* |
7697 | * Nobody can have more than the global setting allows. | |
7698 | */ | |
7699 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7700 | return -EINVAL; | |
6f505b16 | 7701 | |
4653f803 PZ |
7702 | /* |
7703 | * The sum of our children's runtime should not exceed our own. | |
7704 | */ | |
9a7e0b18 PZ |
7705 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7706 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7707 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7708 | |
9a7e0b18 PZ |
7709 | if (child == d->tg) { |
7710 | period = d->rt_period; | |
7711 | runtime = d->rt_runtime; | |
7712 | } | |
6f505b16 | 7713 | |
9a7e0b18 | 7714 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7715 | } |
6f505b16 | 7716 | |
9a7e0b18 PZ |
7717 | if (sum > total) |
7718 | return -EINVAL; | |
7719 | ||
7720 | return 0; | |
6f505b16 PZ |
7721 | } |
7722 | ||
9a7e0b18 | 7723 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7724 | { |
8277434e PT |
7725 | int ret; |
7726 | ||
9a7e0b18 PZ |
7727 | struct rt_schedulable_data data = { |
7728 | .tg = tg, | |
7729 | .rt_period = period, | |
7730 | .rt_runtime = runtime, | |
7731 | }; | |
7732 | ||
8277434e PT |
7733 | rcu_read_lock(); |
7734 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7735 | rcu_read_unlock(); | |
7736 | ||
7737 | return ret; | |
521f1a24 DG |
7738 | } |
7739 | ||
ab84d31e | 7740 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7741 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7742 | { |
ac086bc2 | 7743 | int i, err = 0; |
9f0c1e56 | 7744 | |
2636ed5f PZ |
7745 | /* |
7746 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7747 | * kernel creating (and or operating) RT threads. | |
7748 | */ | |
7749 | if (tg == &root_task_group && rt_runtime == 0) | |
7750 | return -EINVAL; | |
7751 | ||
7752 | /* No period doesn't make any sense. */ | |
7753 | if (rt_period == 0) | |
7754 | return -EINVAL; | |
7755 | ||
9f0c1e56 | 7756 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7757 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7758 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7759 | if (err) | |
9f0c1e56 | 7760 | goto unlock; |
ac086bc2 | 7761 | |
0986b11b | 7762 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7763 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7764 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7765 | |
7766 | for_each_possible_cpu(i) { | |
7767 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7768 | ||
0986b11b | 7769 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7770 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7771 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7772 | } |
0986b11b | 7773 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7774 | unlock: |
521f1a24 | 7775 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7776 | mutex_unlock(&rt_constraints_mutex); |
7777 | ||
7778 | return err; | |
6f505b16 PZ |
7779 | } |
7780 | ||
25cc7da7 | 7781 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7782 | { |
7783 | u64 rt_runtime, rt_period; | |
7784 | ||
7785 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7786 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7787 | if (rt_runtime_us < 0) | |
7788 | rt_runtime = RUNTIME_INF; | |
7789 | ||
ab84d31e | 7790 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7791 | } |
7792 | ||
25cc7da7 | 7793 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7794 | { |
7795 | u64 rt_runtime_us; | |
7796 | ||
d0b27fa7 | 7797 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7798 | return -1; |
7799 | ||
d0b27fa7 | 7800 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7801 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7802 | return rt_runtime_us; | |
7803 | } | |
d0b27fa7 | 7804 | |
ce2f5fe4 | 7805 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7806 | { |
7807 | u64 rt_runtime, rt_period; | |
7808 | ||
ce2f5fe4 | 7809 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7810 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7811 | ||
ab84d31e | 7812 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7813 | } |
7814 | ||
25cc7da7 | 7815 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7816 | { |
7817 | u64 rt_period_us; | |
7818 | ||
7819 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7820 | do_div(rt_period_us, NSEC_PER_USEC); | |
7821 | return rt_period_us; | |
7822 | } | |
332ac17e | 7823 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7824 | |
332ac17e | 7825 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7826 | static int sched_rt_global_constraints(void) |
7827 | { | |
7828 | int ret = 0; | |
7829 | ||
7830 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7831 | read_lock(&tasklist_lock); |
4653f803 | 7832 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7833 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7834 | mutex_unlock(&rt_constraints_mutex); |
7835 | ||
7836 | return ret; | |
7837 | } | |
54e99124 | 7838 | |
25cc7da7 | 7839 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7840 | { |
7841 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7842 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7843 | return 0; | |
7844 | ||
7845 | return 1; | |
7846 | } | |
7847 | ||
6d6bc0ad | 7848 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7849 | static int sched_rt_global_constraints(void) |
7850 | { | |
ac086bc2 | 7851 | unsigned long flags; |
332ac17e | 7852 | int i, ret = 0; |
ec5d4989 | 7853 | |
0986b11b | 7854 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7855 | for_each_possible_cpu(i) { |
7856 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7857 | ||
0986b11b | 7858 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7859 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7860 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7861 | } |
0986b11b | 7862 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7863 | |
332ac17e | 7864 | return ret; |
d0b27fa7 | 7865 | } |
6d6bc0ad | 7866 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7867 | |
a1963b81 | 7868 | static int sched_dl_global_validate(void) |
332ac17e | 7869 | { |
1724813d PZ |
7870 | u64 runtime = global_rt_runtime(); |
7871 | u64 period = global_rt_period(); | |
332ac17e | 7872 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7873 | struct dl_bw *dl_b; |
1724813d | 7874 | int cpu, ret = 0; |
49516342 | 7875 | unsigned long flags; |
332ac17e DF |
7876 | |
7877 | /* | |
7878 | * Here we want to check the bandwidth not being set to some | |
7879 | * value smaller than the currently allocated bandwidth in | |
7880 | * any of the root_domains. | |
7881 | * | |
7882 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7883 | * cycling on root_domains... Discussion on different/better | |
7884 | * solutions is welcome! | |
7885 | */ | |
1724813d | 7886 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7887 | rcu_read_lock_sched(); |
7888 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7889 | |
49516342 | 7890 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7891 | if (new_bw < dl_b->total_bw) |
7892 | ret = -EBUSY; | |
49516342 | 7893 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7894 | |
f10e00f4 KT |
7895 | rcu_read_unlock_sched(); |
7896 | ||
1724813d PZ |
7897 | if (ret) |
7898 | break; | |
332ac17e DF |
7899 | } |
7900 | ||
1724813d | 7901 | return ret; |
332ac17e DF |
7902 | } |
7903 | ||
1724813d | 7904 | static void sched_dl_do_global(void) |
ce0dbbbb | 7905 | { |
1724813d | 7906 | u64 new_bw = -1; |
f10e00f4 | 7907 | struct dl_bw *dl_b; |
1724813d | 7908 | int cpu; |
49516342 | 7909 | unsigned long flags; |
ce0dbbbb | 7910 | |
1724813d PZ |
7911 | def_dl_bandwidth.dl_period = global_rt_period(); |
7912 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7913 | ||
7914 | if (global_rt_runtime() != RUNTIME_INF) | |
7915 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7916 | ||
7917 | /* | |
7918 | * FIXME: As above... | |
7919 | */ | |
7920 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7921 | rcu_read_lock_sched(); |
7922 | dl_b = dl_bw_of(cpu); | |
1724813d | 7923 | |
49516342 | 7924 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7925 | dl_b->bw = new_bw; |
49516342 | 7926 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
7927 | |
7928 | rcu_read_unlock_sched(); | |
ce0dbbbb | 7929 | } |
1724813d PZ |
7930 | } |
7931 | ||
7932 | static int sched_rt_global_validate(void) | |
7933 | { | |
7934 | if (sysctl_sched_rt_period <= 0) | |
7935 | return -EINVAL; | |
7936 | ||
e9e7cb38 JL |
7937 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7938 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7939 | return -EINVAL; |
7940 | ||
7941 | return 0; | |
7942 | } | |
7943 | ||
7944 | static void sched_rt_do_global(void) | |
7945 | { | |
7946 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7947 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7948 | } |
7949 | ||
d0b27fa7 | 7950 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7951 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7952 | loff_t *ppos) |
7953 | { | |
d0b27fa7 PZ |
7954 | int old_period, old_runtime; |
7955 | static DEFINE_MUTEX(mutex); | |
1724813d | 7956 | int ret; |
d0b27fa7 PZ |
7957 | |
7958 | mutex_lock(&mutex); | |
7959 | old_period = sysctl_sched_rt_period; | |
7960 | old_runtime = sysctl_sched_rt_runtime; | |
7961 | ||
8d65af78 | 7962 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7963 | |
7964 | if (!ret && write) { | |
1724813d PZ |
7965 | ret = sched_rt_global_validate(); |
7966 | if (ret) | |
7967 | goto undo; | |
7968 | ||
a1963b81 | 7969 | ret = sched_dl_global_validate(); |
1724813d PZ |
7970 | if (ret) |
7971 | goto undo; | |
7972 | ||
a1963b81 | 7973 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7974 | if (ret) |
7975 | goto undo; | |
7976 | ||
7977 | sched_rt_do_global(); | |
7978 | sched_dl_do_global(); | |
7979 | } | |
7980 | if (0) { | |
7981 | undo: | |
7982 | sysctl_sched_rt_period = old_period; | |
7983 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7984 | } |
7985 | mutex_unlock(&mutex); | |
7986 | ||
7987 | return ret; | |
7988 | } | |
68318b8e | 7989 | |
1724813d | 7990 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7991 | void __user *buffer, size_t *lenp, |
7992 | loff_t *ppos) | |
7993 | { | |
7994 | int ret; | |
332ac17e | 7995 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7996 | |
7997 | mutex_lock(&mutex); | |
332ac17e | 7998 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7999 | /* make sure that internally we keep jiffies */ |
8000 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 8001 | if (!ret && write) { |
1724813d PZ |
8002 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
8003 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
8004 | } |
8005 | mutex_unlock(&mutex); | |
332ac17e DF |
8006 | return ret; |
8007 | } | |
8008 | ||
052f1dc7 | 8009 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 8010 | |
a7c6d554 | 8011 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8012 | { |
a7c6d554 | 8013 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8014 | } |
8015 | ||
eb95419b TH |
8016 | static struct cgroup_subsys_state * |
8017 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8018 | { |
eb95419b TH |
8019 | struct task_group *parent = css_tg(parent_css); |
8020 | struct task_group *tg; | |
68318b8e | 8021 | |
eb95419b | 8022 | if (!parent) { |
68318b8e | 8023 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8024 | return &root_task_group.css; |
68318b8e SV |
8025 | } |
8026 | ||
ec7dc8ac | 8027 | tg = sched_create_group(parent); |
68318b8e SV |
8028 | if (IS_ERR(tg)) |
8029 | return ERR_PTR(-ENOMEM); | |
8030 | ||
68318b8e SV |
8031 | return &tg->css; |
8032 | } | |
8033 | ||
eb95419b | 8034 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 8035 | { |
eb95419b | 8036 | struct task_group *tg = css_tg(css); |
5c9d535b | 8037 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 8038 | |
63876986 TH |
8039 | if (parent) |
8040 | sched_online_group(tg, parent); | |
ace783b9 LZ |
8041 | return 0; |
8042 | } | |
8043 | ||
eb95419b | 8044 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8045 | { |
eb95419b | 8046 | struct task_group *tg = css_tg(css); |
68318b8e SV |
8047 | |
8048 | sched_destroy_group(tg); | |
8049 | } | |
8050 | ||
eb95419b | 8051 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 8052 | { |
eb95419b | 8053 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
8054 | |
8055 | sched_offline_group(tg); | |
8056 | } | |
8057 | ||
eeb61e53 KT |
8058 | static void cpu_cgroup_fork(struct task_struct *task) |
8059 | { | |
8060 | sched_move_task(task); | |
8061 | } | |
8062 | ||
eb95419b | 8063 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8064 | struct cgroup_taskset *tset) |
68318b8e | 8065 | { |
bb9d97b6 TH |
8066 | struct task_struct *task; |
8067 | ||
924f0d9a | 8068 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 8069 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8070 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8071 | return -EINVAL; |
b68aa230 | 8072 | #else |
bb9d97b6 TH |
8073 | /* We don't support RT-tasks being in separate groups */ |
8074 | if (task->sched_class != &fair_sched_class) | |
8075 | return -EINVAL; | |
b68aa230 | 8076 | #endif |
bb9d97b6 | 8077 | } |
be367d09 BB |
8078 | return 0; |
8079 | } | |
68318b8e | 8080 | |
eb95419b | 8081 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8082 | struct cgroup_taskset *tset) |
68318b8e | 8083 | { |
bb9d97b6 TH |
8084 | struct task_struct *task; |
8085 | ||
924f0d9a | 8086 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 8087 | sched_move_task(task); |
68318b8e SV |
8088 | } |
8089 | ||
eb95419b TH |
8090 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
8091 | struct cgroup_subsys_state *old_css, | |
8092 | struct task_struct *task) | |
068c5cc5 PZ |
8093 | { |
8094 | /* | |
8095 | * cgroup_exit() is called in the copy_process() failure path. | |
8096 | * Ignore this case since the task hasn't ran yet, this avoids | |
8097 | * trying to poke a half freed task state from generic code. | |
8098 | */ | |
8099 | if (!(task->flags & PF_EXITING)) | |
8100 | return; | |
8101 | ||
8102 | sched_move_task(task); | |
8103 | } | |
8104 | ||
052f1dc7 | 8105 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8106 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8107 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8108 | { |
182446d0 | 8109 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8110 | } |
8111 | ||
182446d0 TH |
8112 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8113 | struct cftype *cft) | |
68318b8e | 8114 | { |
182446d0 | 8115 | struct task_group *tg = css_tg(css); |
68318b8e | 8116 | |
c8b28116 | 8117 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8118 | } |
ab84d31e PT |
8119 | |
8120 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8121 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8122 | ||
ab84d31e PT |
8123 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8124 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8125 | ||
a790de99 PT |
8126 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8127 | ||
ab84d31e PT |
8128 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8129 | { | |
56f570e5 | 8130 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8131 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8132 | |
8133 | if (tg == &root_task_group) | |
8134 | return -EINVAL; | |
8135 | ||
8136 | /* | |
8137 | * Ensure we have at some amount of bandwidth every period. This is | |
8138 | * to prevent reaching a state of large arrears when throttled via | |
8139 | * entity_tick() resulting in prolonged exit starvation. | |
8140 | */ | |
8141 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8142 | return -EINVAL; | |
8143 | ||
8144 | /* | |
8145 | * Likewise, bound things on the otherside by preventing insane quota | |
8146 | * periods. This also allows us to normalize in computing quota | |
8147 | * feasibility. | |
8148 | */ | |
8149 | if (period > max_cfs_quota_period) | |
8150 | return -EINVAL; | |
8151 | ||
0e59bdae KT |
8152 | /* |
8153 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8154 | * unthrottle_offline_cfs_rqs(). | |
8155 | */ | |
8156 | get_online_cpus(); | |
a790de99 PT |
8157 | mutex_lock(&cfs_constraints_mutex); |
8158 | ret = __cfs_schedulable(tg, period, quota); | |
8159 | if (ret) | |
8160 | goto out_unlock; | |
8161 | ||
58088ad0 | 8162 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8163 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8164 | /* |
8165 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8166 | * before making related changes, and on->off must occur afterwards | |
8167 | */ | |
8168 | if (runtime_enabled && !runtime_was_enabled) | |
8169 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8170 | raw_spin_lock_irq(&cfs_b->lock); |
8171 | cfs_b->period = ns_to_ktime(period); | |
8172 | cfs_b->quota = quota; | |
58088ad0 | 8173 | |
a9cf55b2 | 8174 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 | 8175 | /* restart the period timer (if active) to handle new period expiry */ |
77a4d1a1 PZ |
8176 | if (runtime_enabled) |
8177 | start_cfs_bandwidth(cfs_b); | |
ab84d31e PT |
8178 | raw_spin_unlock_irq(&cfs_b->lock); |
8179 | ||
0e59bdae | 8180 | for_each_online_cpu(i) { |
ab84d31e | 8181 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8182 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8183 | |
8184 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8185 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8186 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8187 | |
029632fb | 8188 | if (cfs_rq->throttled) |
671fd9da | 8189 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8190 | raw_spin_unlock_irq(&rq->lock); |
8191 | } | |
1ee14e6c BS |
8192 | if (runtime_was_enabled && !runtime_enabled) |
8193 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8194 | out_unlock: |
8195 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8196 | put_online_cpus(); |
ab84d31e | 8197 | |
a790de99 | 8198 | return ret; |
ab84d31e PT |
8199 | } |
8200 | ||
8201 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8202 | { | |
8203 | u64 quota, period; | |
8204 | ||
029632fb | 8205 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8206 | if (cfs_quota_us < 0) |
8207 | quota = RUNTIME_INF; | |
8208 | else | |
8209 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8210 | ||
8211 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8212 | } | |
8213 | ||
8214 | long tg_get_cfs_quota(struct task_group *tg) | |
8215 | { | |
8216 | u64 quota_us; | |
8217 | ||
029632fb | 8218 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8219 | return -1; |
8220 | ||
029632fb | 8221 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8222 | do_div(quota_us, NSEC_PER_USEC); |
8223 | ||
8224 | return quota_us; | |
8225 | } | |
8226 | ||
8227 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8228 | { | |
8229 | u64 quota, period; | |
8230 | ||
8231 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8232 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8233 | |
ab84d31e PT |
8234 | return tg_set_cfs_bandwidth(tg, period, quota); |
8235 | } | |
8236 | ||
8237 | long tg_get_cfs_period(struct task_group *tg) | |
8238 | { | |
8239 | u64 cfs_period_us; | |
8240 | ||
029632fb | 8241 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8242 | do_div(cfs_period_us, NSEC_PER_USEC); |
8243 | ||
8244 | return cfs_period_us; | |
8245 | } | |
8246 | ||
182446d0 TH |
8247 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8248 | struct cftype *cft) | |
ab84d31e | 8249 | { |
182446d0 | 8250 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8251 | } |
8252 | ||
182446d0 TH |
8253 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8254 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8255 | { |
182446d0 | 8256 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8257 | } |
8258 | ||
182446d0 TH |
8259 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8260 | struct cftype *cft) | |
ab84d31e | 8261 | { |
182446d0 | 8262 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8263 | } |
8264 | ||
182446d0 TH |
8265 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8266 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8267 | { |
182446d0 | 8268 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8269 | } |
8270 | ||
a790de99 PT |
8271 | struct cfs_schedulable_data { |
8272 | struct task_group *tg; | |
8273 | u64 period, quota; | |
8274 | }; | |
8275 | ||
8276 | /* | |
8277 | * normalize group quota/period to be quota/max_period | |
8278 | * note: units are usecs | |
8279 | */ | |
8280 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8281 | struct cfs_schedulable_data *d) | |
8282 | { | |
8283 | u64 quota, period; | |
8284 | ||
8285 | if (tg == d->tg) { | |
8286 | period = d->period; | |
8287 | quota = d->quota; | |
8288 | } else { | |
8289 | period = tg_get_cfs_period(tg); | |
8290 | quota = tg_get_cfs_quota(tg); | |
8291 | } | |
8292 | ||
8293 | /* note: these should typically be equivalent */ | |
8294 | if (quota == RUNTIME_INF || quota == -1) | |
8295 | return RUNTIME_INF; | |
8296 | ||
8297 | return to_ratio(period, quota); | |
8298 | } | |
8299 | ||
8300 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8301 | { | |
8302 | struct cfs_schedulable_data *d = data; | |
029632fb | 8303 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8304 | s64 quota = 0, parent_quota = -1; |
8305 | ||
8306 | if (!tg->parent) { | |
8307 | quota = RUNTIME_INF; | |
8308 | } else { | |
029632fb | 8309 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8310 | |
8311 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8312 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8313 | |
8314 | /* | |
8315 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8316 | * limit is set | |
8317 | */ | |
8318 | if (quota == RUNTIME_INF) | |
8319 | quota = parent_quota; | |
8320 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8321 | return -EINVAL; | |
8322 | } | |
9c58c79a | 8323 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8324 | |
8325 | return 0; | |
8326 | } | |
8327 | ||
8328 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8329 | { | |
8277434e | 8330 | int ret; |
a790de99 PT |
8331 | struct cfs_schedulable_data data = { |
8332 | .tg = tg, | |
8333 | .period = period, | |
8334 | .quota = quota, | |
8335 | }; | |
8336 | ||
8337 | if (quota != RUNTIME_INF) { | |
8338 | do_div(data.period, NSEC_PER_USEC); | |
8339 | do_div(data.quota, NSEC_PER_USEC); | |
8340 | } | |
8341 | ||
8277434e PT |
8342 | rcu_read_lock(); |
8343 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8344 | rcu_read_unlock(); | |
8345 | ||
8346 | return ret; | |
a790de99 | 8347 | } |
e8da1b18 | 8348 | |
2da8ca82 | 8349 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8350 | { |
2da8ca82 | 8351 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8352 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8353 | |
44ffc75b TH |
8354 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8355 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8356 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8357 | |
8358 | return 0; | |
8359 | } | |
ab84d31e | 8360 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8361 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8362 | |
052f1dc7 | 8363 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8364 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8365 | struct cftype *cft, s64 val) | |
6f505b16 | 8366 | { |
182446d0 | 8367 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8368 | } |
8369 | ||
182446d0 TH |
8370 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8371 | struct cftype *cft) | |
6f505b16 | 8372 | { |
182446d0 | 8373 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8374 | } |
d0b27fa7 | 8375 | |
182446d0 TH |
8376 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8377 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8378 | { |
182446d0 | 8379 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8380 | } |
8381 | ||
182446d0 TH |
8382 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8383 | struct cftype *cft) | |
d0b27fa7 | 8384 | { |
182446d0 | 8385 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8386 | } |
6d6bc0ad | 8387 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8388 | |
fe5c7cc2 | 8389 | static struct cftype cpu_files[] = { |
052f1dc7 | 8390 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8391 | { |
8392 | .name = "shares", | |
f4c753b7 PM |
8393 | .read_u64 = cpu_shares_read_u64, |
8394 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8395 | }, |
052f1dc7 | 8396 | #endif |
ab84d31e PT |
8397 | #ifdef CONFIG_CFS_BANDWIDTH |
8398 | { | |
8399 | .name = "cfs_quota_us", | |
8400 | .read_s64 = cpu_cfs_quota_read_s64, | |
8401 | .write_s64 = cpu_cfs_quota_write_s64, | |
8402 | }, | |
8403 | { | |
8404 | .name = "cfs_period_us", | |
8405 | .read_u64 = cpu_cfs_period_read_u64, | |
8406 | .write_u64 = cpu_cfs_period_write_u64, | |
8407 | }, | |
e8da1b18 NR |
8408 | { |
8409 | .name = "stat", | |
2da8ca82 | 8410 | .seq_show = cpu_stats_show, |
e8da1b18 | 8411 | }, |
ab84d31e | 8412 | #endif |
052f1dc7 | 8413 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8414 | { |
9f0c1e56 | 8415 | .name = "rt_runtime_us", |
06ecb27c PM |
8416 | .read_s64 = cpu_rt_runtime_read, |
8417 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8418 | }, |
d0b27fa7 PZ |
8419 | { |
8420 | .name = "rt_period_us", | |
f4c753b7 PM |
8421 | .read_u64 = cpu_rt_period_read_uint, |
8422 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8423 | }, |
052f1dc7 | 8424 | #endif |
4baf6e33 | 8425 | { } /* terminate */ |
68318b8e SV |
8426 | }; |
8427 | ||
073219e9 | 8428 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8429 | .css_alloc = cpu_cgroup_css_alloc, |
8430 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8431 | .css_online = cpu_cgroup_css_online, |
8432 | .css_offline = cpu_cgroup_css_offline, | |
eeb61e53 | 8433 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8434 | .can_attach = cpu_cgroup_can_attach, |
8435 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8436 | .exit = cpu_cgroup_exit, |
5577964e | 8437 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8438 | .early_init = 1, |
8439 | }; | |
8440 | ||
052f1dc7 | 8441 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8442 | |
b637a328 PM |
8443 | void dump_cpu_task(int cpu) |
8444 | { | |
8445 | pr_info("Task dump for CPU %d:\n", cpu); | |
8446 | sched_show_task(cpu_curr(cpu)); | |
8447 | } |