]>
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 | 93 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 94 | { |
58088ad0 PT |
95 | unsigned long delta; |
96 | ktime_t soft, hard, now; | |
d0b27fa7 | 97 | |
58088ad0 PT |
98 | for (;;) { |
99 | if (hrtimer_active(period_timer)) | |
100 | break; | |
101 | ||
102 | now = hrtimer_cb_get_time(period_timer); | |
103 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 104 | |
58088ad0 PT |
105 | soft = hrtimer_get_softexpires(period_timer); |
106 | hard = hrtimer_get_expires(period_timer); | |
107 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
108 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
109 | HRTIMER_MODE_ABS_PINNED, 0); | |
110 | } | |
111 | } | |
112 | ||
029632fb PZ |
113 | DEFINE_MUTEX(sched_domains_mutex); |
114 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 115 | |
fe44d621 | 116 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 117 | |
029632fb | 118 | void update_rq_clock(struct rq *rq) |
3e51f33f | 119 | { |
fe44d621 | 120 | s64 delta; |
305e6835 | 121 | |
9edfbfed PZ |
122 | lockdep_assert_held(&rq->lock); |
123 | ||
124 | if (rq->clock_skip_update & RQCF_ACT_SKIP) | |
f26f9aff | 125 | return; |
aa483808 | 126 | |
fe44d621 | 127 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
128 | if (delta < 0) |
129 | return; | |
fe44d621 PZ |
130 | rq->clock += delta; |
131 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
132 | } |
133 | ||
bf5c91ba IM |
134 | /* |
135 | * Debugging: various feature bits | |
136 | */ | |
f00b45c1 | 137 | |
f00b45c1 PZ |
138 | #define SCHED_FEAT(name, enabled) \ |
139 | (1UL << __SCHED_FEAT_##name) * enabled | | |
140 | ||
bf5c91ba | 141 | const_debug unsigned int sysctl_sched_features = |
391e43da | 142 | #include "features.h" |
f00b45c1 PZ |
143 | 0; |
144 | ||
145 | #undef SCHED_FEAT | |
146 | ||
147 | #ifdef CONFIG_SCHED_DEBUG | |
148 | #define SCHED_FEAT(name, enabled) \ | |
149 | #name , | |
150 | ||
1292531f | 151 | static const char * const sched_feat_names[] = { |
391e43da | 152 | #include "features.h" |
f00b45c1 PZ |
153 | }; |
154 | ||
155 | #undef SCHED_FEAT | |
156 | ||
34f3a814 | 157 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 158 | { |
f00b45c1 PZ |
159 | int i; |
160 | ||
f8b6d1cc | 161 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
162 | if (!(sysctl_sched_features & (1UL << i))) |
163 | seq_puts(m, "NO_"); | |
164 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 165 | } |
34f3a814 | 166 | seq_puts(m, "\n"); |
f00b45c1 | 167 | |
34f3a814 | 168 | return 0; |
f00b45c1 PZ |
169 | } |
170 | ||
f8b6d1cc PZ |
171 | #ifdef HAVE_JUMP_LABEL |
172 | ||
c5905afb IM |
173 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
174 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
175 | |
176 | #define SCHED_FEAT(name, enabled) \ | |
177 | jump_label_key__##enabled , | |
178 | ||
c5905afb | 179 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
180 | #include "features.h" |
181 | }; | |
182 | ||
183 | #undef SCHED_FEAT | |
184 | ||
185 | static void sched_feat_disable(int i) | |
186 | { | |
c5905afb IM |
187 | if (static_key_enabled(&sched_feat_keys[i])) |
188 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
189 | } |
190 | ||
191 | static void sched_feat_enable(int i) | |
192 | { | |
c5905afb IM |
193 | if (!static_key_enabled(&sched_feat_keys[i])) |
194 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
195 | } |
196 | #else | |
197 | static void sched_feat_disable(int i) { }; | |
198 | static void sched_feat_enable(int i) { }; | |
199 | #endif /* HAVE_JUMP_LABEL */ | |
200 | ||
1a687c2e | 201 | static int sched_feat_set(char *cmp) |
f00b45c1 | 202 | { |
f00b45c1 | 203 | int i; |
1a687c2e | 204 | int neg = 0; |
f00b45c1 | 205 | |
524429c3 | 206 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
207 | neg = 1; |
208 | cmp += 3; | |
209 | } | |
210 | ||
f8b6d1cc | 211 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 212 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 213 | if (neg) { |
f00b45c1 | 214 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
215 | sched_feat_disable(i); |
216 | } else { | |
f00b45c1 | 217 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
218 | sched_feat_enable(i); |
219 | } | |
f00b45c1 PZ |
220 | break; |
221 | } | |
222 | } | |
223 | ||
1a687c2e MG |
224 | return i; |
225 | } | |
226 | ||
227 | static ssize_t | |
228 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
229 | size_t cnt, loff_t *ppos) | |
230 | { | |
231 | char buf[64]; | |
232 | char *cmp; | |
233 | int i; | |
5cd08fbf | 234 | struct inode *inode; |
1a687c2e MG |
235 | |
236 | if (cnt > 63) | |
237 | cnt = 63; | |
238 | ||
239 | if (copy_from_user(&buf, ubuf, cnt)) | |
240 | return -EFAULT; | |
241 | ||
242 | buf[cnt] = 0; | |
243 | cmp = strstrip(buf); | |
244 | ||
5cd08fbf JB |
245 | /* Ensure the static_key remains in a consistent state */ |
246 | inode = file_inode(filp); | |
247 | mutex_lock(&inode->i_mutex); | |
1a687c2e | 248 | i = sched_feat_set(cmp); |
5cd08fbf | 249 | mutex_unlock(&inode->i_mutex); |
f8b6d1cc | 250 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
251 | return -EINVAL; |
252 | ||
42994724 | 253 | *ppos += cnt; |
f00b45c1 PZ |
254 | |
255 | return cnt; | |
256 | } | |
257 | ||
34f3a814 LZ |
258 | static int sched_feat_open(struct inode *inode, struct file *filp) |
259 | { | |
260 | return single_open(filp, sched_feat_show, NULL); | |
261 | } | |
262 | ||
828c0950 | 263 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
264 | .open = sched_feat_open, |
265 | .write = sched_feat_write, | |
266 | .read = seq_read, | |
267 | .llseek = seq_lseek, | |
268 | .release = single_release, | |
f00b45c1 PZ |
269 | }; |
270 | ||
271 | static __init int sched_init_debug(void) | |
272 | { | |
f00b45c1 PZ |
273 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
274 | &sched_feat_fops); | |
275 | ||
276 | return 0; | |
277 | } | |
278 | late_initcall(sched_init_debug); | |
f8b6d1cc | 279 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 280 | |
b82d9fdd PZ |
281 | /* |
282 | * Number of tasks to iterate in a single balance run. | |
283 | * Limited because this is done with IRQs disabled. | |
284 | */ | |
285 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
286 | ||
e9e9250b PZ |
287 | /* |
288 | * period over which we average the RT time consumption, measured | |
289 | * in ms. | |
290 | * | |
291 | * default: 1s | |
292 | */ | |
293 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
294 | ||
fa85ae24 | 295 | /* |
9f0c1e56 | 296 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
297 | * default: 1s |
298 | */ | |
9f0c1e56 | 299 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 300 | |
029632fb | 301 | __read_mostly int scheduler_running; |
6892b75e | 302 | |
9f0c1e56 PZ |
303 | /* |
304 | * part of the period that we allow rt tasks to run in us. | |
305 | * default: 0.95s | |
306 | */ | |
307 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 308 | |
1da177e4 | 309 | /* |
cc2a73b5 | 310 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 311 | */ |
a9957449 | 312 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
313 | __acquires(rq->lock) |
314 | { | |
70b97a7f | 315 | struct rq *rq; |
1da177e4 LT |
316 | |
317 | local_irq_disable(); | |
318 | rq = this_rq(); | |
05fa785c | 319 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
320 | |
321 | return rq; | |
322 | } | |
323 | ||
8f4d37ec PZ |
324 | #ifdef CONFIG_SCHED_HRTICK |
325 | /* | |
326 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 327 | */ |
8f4d37ec | 328 | |
8f4d37ec PZ |
329 | static void hrtick_clear(struct rq *rq) |
330 | { | |
331 | if (hrtimer_active(&rq->hrtick_timer)) | |
332 | hrtimer_cancel(&rq->hrtick_timer); | |
333 | } | |
334 | ||
8f4d37ec PZ |
335 | /* |
336 | * High-resolution timer tick. | |
337 | * Runs from hardirq context with interrupts disabled. | |
338 | */ | |
339 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
340 | { | |
341 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
342 | ||
343 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
344 | ||
05fa785c | 345 | raw_spin_lock(&rq->lock); |
3e51f33f | 346 | update_rq_clock(rq); |
8f4d37ec | 347 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 348 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
349 | |
350 | return HRTIMER_NORESTART; | |
351 | } | |
352 | ||
95e904c7 | 353 | #ifdef CONFIG_SMP |
971ee28c PZ |
354 | |
355 | static int __hrtick_restart(struct rq *rq) | |
356 | { | |
357 | struct hrtimer *timer = &rq->hrtick_timer; | |
358 | ktime_t time = hrtimer_get_softexpires(timer); | |
359 | ||
360 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
361 | } | |
362 | ||
31656519 PZ |
363 | /* |
364 | * called from hardirq (IPI) context | |
365 | */ | |
366 | static void __hrtick_start(void *arg) | |
b328ca18 | 367 | { |
31656519 | 368 | struct rq *rq = arg; |
b328ca18 | 369 | |
05fa785c | 370 | raw_spin_lock(&rq->lock); |
971ee28c | 371 | __hrtick_restart(rq); |
31656519 | 372 | rq->hrtick_csd_pending = 0; |
05fa785c | 373 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
374 | } |
375 | ||
31656519 PZ |
376 | /* |
377 | * Called to set the hrtick timer state. | |
378 | * | |
379 | * called with rq->lock held and irqs disabled | |
380 | */ | |
029632fb | 381 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 382 | { |
31656519 | 383 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 384 | ktime_t time; |
385 | s64 delta; | |
386 | ||
387 | /* | |
388 | * Don't schedule slices shorter than 10000ns, that just | |
389 | * doesn't make sense and can cause timer DoS. | |
390 | */ | |
391 | delta = max_t(s64, delay, 10000LL); | |
392 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 393 | |
cc584b21 | 394 | hrtimer_set_expires(timer, time); |
31656519 PZ |
395 | |
396 | if (rq == this_rq()) { | |
971ee28c | 397 | __hrtick_restart(rq); |
31656519 | 398 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 399 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
400 | rq->hrtick_csd_pending = 1; |
401 | } | |
b328ca18 PZ |
402 | } |
403 | ||
404 | static int | |
405 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
406 | { | |
407 | int cpu = (int)(long)hcpu; | |
408 | ||
409 | switch (action) { | |
410 | case CPU_UP_CANCELED: | |
411 | case CPU_UP_CANCELED_FROZEN: | |
412 | case CPU_DOWN_PREPARE: | |
413 | case CPU_DOWN_PREPARE_FROZEN: | |
414 | case CPU_DEAD: | |
415 | case CPU_DEAD_FROZEN: | |
31656519 | 416 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
417 | return NOTIFY_OK; |
418 | } | |
419 | ||
420 | return NOTIFY_DONE; | |
421 | } | |
422 | ||
fa748203 | 423 | static __init void init_hrtick(void) |
b328ca18 PZ |
424 | { |
425 | hotcpu_notifier(hotplug_hrtick, 0); | |
426 | } | |
31656519 PZ |
427 | #else |
428 | /* | |
429 | * Called to set the hrtick timer state. | |
430 | * | |
431 | * called with rq->lock held and irqs disabled | |
432 | */ | |
029632fb | 433 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 434 | { |
86893335 WL |
435 | /* |
436 | * Don't schedule slices shorter than 10000ns, that just | |
437 | * doesn't make sense. Rely on vruntime for fairness. | |
438 | */ | |
439 | delay = max_t(u64, delay, 10000LL); | |
7f1e2ca9 | 440 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 441 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 442 | } |
b328ca18 | 443 | |
006c75f1 | 444 | static inline void init_hrtick(void) |
8f4d37ec | 445 | { |
8f4d37ec | 446 | } |
31656519 | 447 | #endif /* CONFIG_SMP */ |
8f4d37ec | 448 | |
31656519 | 449 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 450 | { |
31656519 PZ |
451 | #ifdef CONFIG_SMP |
452 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 453 | |
31656519 PZ |
454 | rq->hrtick_csd.flags = 0; |
455 | rq->hrtick_csd.func = __hrtick_start; | |
456 | rq->hrtick_csd.info = rq; | |
457 | #endif | |
8f4d37ec | 458 | |
31656519 PZ |
459 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
460 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 461 | } |
006c75f1 | 462 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
463 | static inline void hrtick_clear(struct rq *rq) |
464 | { | |
465 | } | |
466 | ||
8f4d37ec PZ |
467 | static inline void init_rq_hrtick(struct rq *rq) |
468 | { | |
469 | } | |
470 | ||
b328ca18 PZ |
471 | static inline void init_hrtick(void) |
472 | { | |
473 | } | |
006c75f1 | 474 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 475 | |
fd99f91a PZ |
476 | /* |
477 | * cmpxchg based fetch_or, macro so it works for different integer types | |
478 | */ | |
479 | #define fetch_or(ptr, val) \ | |
480 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
481 | for (;;) { \ | |
482 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
483 | if (__old == __val) \ | |
484 | break; \ | |
485 | __val = __old; \ | |
486 | } \ | |
487 | __old; \ | |
488 | }) | |
489 | ||
e3baac47 | 490 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
491 | /* |
492 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
493 | * this avoids any races wrt polling state changes and thereby avoids | |
494 | * spurious IPIs. | |
495 | */ | |
496 | static bool set_nr_and_not_polling(struct task_struct *p) | |
497 | { | |
498 | struct thread_info *ti = task_thread_info(p); | |
499 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
500 | } | |
e3baac47 PZ |
501 | |
502 | /* | |
503 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
504 | * | |
505 | * If this returns true, then the idle task promises to call | |
506 | * sched_ttwu_pending() and reschedule soon. | |
507 | */ | |
508 | static bool set_nr_if_polling(struct task_struct *p) | |
509 | { | |
510 | struct thread_info *ti = task_thread_info(p); | |
511 | typeof(ti->flags) old, val = ACCESS_ONCE(ti->flags); | |
512 | ||
513 | for (;;) { | |
514 | if (!(val & _TIF_POLLING_NRFLAG)) | |
515 | return false; | |
516 | if (val & _TIF_NEED_RESCHED) | |
517 | return true; | |
518 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
519 | if (old == val) | |
520 | break; | |
521 | val = old; | |
522 | } | |
523 | return true; | |
524 | } | |
525 | ||
fd99f91a PZ |
526 | #else |
527 | static bool set_nr_and_not_polling(struct task_struct *p) | |
528 | { | |
529 | set_tsk_need_resched(p); | |
530 | return true; | |
531 | } | |
e3baac47 PZ |
532 | |
533 | #ifdef CONFIG_SMP | |
534 | static bool set_nr_if_polling(struct task_struct *p) | |
535 | { | |
536 | return false; | |
537 | } | |
538 | #endif | |
fd99f91a PZ |
539 | #endif |
540 | ||
c24d20db | 541 | /* |
8875125e | 542 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
543 | * |
544 | * On UP this means the setting of the need_resched flag, on SMP it | |
545 | * might also involve a cross-CPU call to trigger the scheduler on | |
546 | * the target CPU. | |
547 | */ | |
8875125e | 548 | void resched_curr(struct rq *rq) |
c24d20db | 549 | { |
8875125e | 550 | struct task_struct *curr = rq->curr; |
c24d20db IM |
551 | int cpu; |
552 | ||
8875125e | 553 | lockdep_assert_held(&rq->lock); |
c24d20db | 554 | |
8875125e | 555 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
556 | return; |
557 | ||
8875125e | 558 | cpu = cpu_of(rq); |
fd99f91a | 559 | |
f27dde8d | 560 | if (cpu == smp_processor_id()) { |
8875125e | 561 | set_tsk_need_resched(curr); |
f27dde8d | 562 | set_preempt_need_resched(); |
c24d20db | 563 | return; |
f27dde8d | 564 | } |
c24d20db | 565 | |
8875125e | 566 | if (set_nr_and_not_polling(curr)) |
c24d20db | 567 | smp_send_reschedule(cpu); |
dfc68f29 AL |
568 | else |
569 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
570 | } |
571 | ||
029632fb | 572 | void resched_cpu(int cpu) |
c24d20db IM |
573 | { |
574 | struct rq *rq = cpu_rq(cpu); | |
575 | unsigned long flags; | |
576 | ||
05fa785c | 577 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 578 | return; |
8875125e | 579 | resched_curr(rq); |
05fa785c | 580 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 581 | } |
06d8308c | 582 | |
b021fe3e | 583 | #ifdef CONFIG_SMP |
3451d024 | 584 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
585 | /* |
586 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
587 | * from an idle cpu. This is good for power-savings. | |
588 | * | |
589 | * We don't do similar optimization for completely idle system, as | |
590 | * selecting an idle cpu will add more delays to the timers than intended | |
591 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
592 | */ | |
6201b4d6 | 593 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
594 | { |
595 | int cpu = smp_processor_id(); | |
596 | int i; | |
597 | struct sched_domain *sd; | |
598 | ||
6201b4d6 VK |
599 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
600 | return cpu; | |
601 | ||
057f3fad | 602 | rcu_read_lock(); |
83cd4fe2 | 603 | for_each_domain(cpu, sd) { |
057f3fad PZ |
604 | for_each_cpu(i, sched_domain_span(sd)) { |
605 | if (!idle_cpu(i)) { | |
606 | cpu = i; | |
607 | goto unlock; | |
608 | } | |
609 | } | |
83cd4fe2 | 610 | } |
057f3fad PZ |
611 | unlock: |
612 | rcu_read_unlock(); | |
83cd4fe2 VP |
613 | return cpu; |
614 | } | |
06d8308c TG |
615 | /* |
616 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
617 | * idle CPU then this timer might expire before the next timer event | |
618 | * which is scheduled to wake up that CPU. In case of a completely | |
619 | * idle system the next event might even be infinite time into the | |
620 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
621 | * leaves the inner idle loop so the newly added timer is taken into | |
622 | * account when the CPU goes back to idle and evaluates the timer | |
623 | * wheel for the next timer event. | |
624 | */ | |
1c20091e | 625 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
626 | { |
627 | struct rq *rq = cpu_rq(cpu); | |
628 | ||
629 | if (cpu == smp_processor_id()) | |
630 | return; | |
631 | ||
67b9ca70 | 632 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 633 | smp_send_reschedule(cpu); |
dfc68f29 AL |
634 | else |
635 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
636 | } |
637 | ||
c5bfece2 | 638 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 639 | { |
53c5fa16 FW |
640 | /* |
641 | * We just need the target to call irq_exit() and re-evaluate | |
642 | * the next tick. The nohz full kick at least implies that. | |
643 | * If needed we can still optimize that later with an | |
644 | * empty IRQ. | |
645 | */ | |
c5bfece2 | 646 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
647 | if (cpu != smp_processor_id() || |
648 | tick_nohz_tick_stopped()) | |
53c5fa16 | 649 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
650 | return true; |
651 | } | |
652 | ||
653 | return false; | |
654 | } | |
655 | ||
656 | void wake_up_nohz_cpu(int cpu) | |
657 | { | |
c5bfece2 | 658 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
659 | wake_up_idle_cpu(cpu); |
660 | } | |
661 | ||
ca38062e | 662 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 663 | { |
1c792db7 | 664 | int cpu = smp_processor_id(); |
873b4c65 VG |
665 | |
666 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
667 | return false; | |
668 | ||
669 | if (idle_cpu(cpu) && !need_resched()) | |
670 | return true; | |
671 | ||
672 | /* | |
673 | * We can't run Idle Load Balance on this CPU for this time so we | |
674 | * cancel it and clear NOHZ_BALANCE_KICK | |
675 | */ | |
676 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
677 | return false; | |
45bf76df IM |
678 | } |
679 | ||
3451d024 | 680 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 681 | |
ca38062e | 682 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 683 | { |
ca38062e | 684 | return false; |
2069dd75 PZ |
685 | } |
686 | ||
3451d024 | 687 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 688 | |
ce831b38 FW |
689 | #ifdef CONFIG_NO_HZ_FULL |
690 | bool sched_can_stop_tick(void) | |
691 | { | |
1e78cdbd RR |
692 | /* |
693 | * FIFO realtime policy runs the highest priority task. Other runnable | |
694 | * tasks are of a lower priority. The scheduler tick does nothing. | |
695 | */ | |
696 | if (current->policy == SCHED_FIFO) | |
697 | return true; | |
698 | ||
699 | /* | |
700 | * Round-robin realtime tasks time slice with other tasks at the same | |
701 | * realtime priority. Is this task the only one at this priority? | |
702 | */ | |
703 | if (current->policy == SCHED_RR) { | |
704 | struct sched_rt_entity *rt_se = ¤t->rt; | |
705 | ||
706 | return rt_se->run_list.prev == rt_se->run_list.next; | |
707 | } | |
708 | ||
3882ec64 FW |
709 | /* |
710 | * More than one running task need preemption. | |
711 | * nr_running update is assumed to be visible | |
712 | * after IPI is sent from wakers. | |
713 | */ | |
541b8264 VK |
714 | if (this_rq()->nr_running > 1) |
715 | return false; | |
ce831b38 | 716 | |
541b8264 | 717 | return true; |
ce831b38 FW |
718 | } |
719 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 720 | |
029632fb | 721 | void sched_avg_update(struct rq *rq) |
18d95a28 | 722 | { |
e9e9250b PZ |
723 | s64 period = sched_avg_period(); |
724 | ||
78becc27 | 725 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
726 | /* |
727 | * Inline assembly required to prevent the compiler | |
728 | * optimising this loop into a divmod call. | |
729 | * See __iter_div_u64_rem() for another example of this. | |
730 | */ | |
731 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
732 | rq->age_stamp += period; |
733 | rq->rt_avg /= 2; | |
734 | } | |
18d95a28 PZ |
735 | } |
736 | ||
6d6bc0ad | 737 | #endif /* CONFIG_SMP */ |
18d95a28 | 738 | |
a790de99 PT |
739 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
740 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 741 | /* |
8277434e PT |
742 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
743 | * node and @up when leaving it for the final time. | |
744 | * | |
745 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 746 | */ |
029632fb | 747 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 748 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
749 | { |
750 | struct task_group *parent, *child; | |
eb755805 | 751 | int ret; |
c09595f6 | 752 | |
8277434e PT |
753 | parent = from; |
754 | ||
c09595f6 | 755 | down: |
eb755805 PZ |
756 | ret = (*down)(parent, data); |
757 | if (ret) | |
8277434e | 758 | goto out; |
c09595f6 PZ |
759 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
760 | parent = child; | |
761 | goto down; | |
762 | ||
763 | up: | |
764 | continue; | |
765 | } | |
eb755805 | 766 | ret = (*up)(parent, data); |
8277434e PT |
767 | if (ret || parent == from) |
768 | goto out; | |
c09595f6 PZ |
769 | |
770 | child = parent; | |
771 | parent = parent->parent; | |
772 | if (parent) | |
773 | goto up; | |
8277434e | 774 | out: |
eb755805 | 775 | return ret; |
c09595f6 PZ |
776 | } |
777 | ||
029632fb | 778 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 779 | { |
e2b245f8 | 780 | return 0; |
eb755805 | 781 | } |
18d95a28 PZ |
782 | #endif |
783 | ||
45bf76df IM |
784 | static void set_load_weight(struct task_struct *p) |
785 | { | |
f05998d4 NR |
786 | int prio = p->static_prio - MAX_RT_PRIO; |
787 | struct load_weight *load = &p->se.load; | |
788 | ||
dd41f596 IM |
789 | /* |
790 | * SCHED_IDLE tasks get minimal weight: | |
791 | */ | |
792 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 793 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 794 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
795 | return; |
796 | } | |
71f8bd46 | 797 | |
c8b28116 | 798 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 799 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
800 | } |
801 | ||
371fd7e7 | 802 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 803 | { |
a64692a3 | 804 | update_rq_clock(rq); |
43148951 | 805 | sched_info_queued(rq, p); |
371fd7e7 | 806 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
807 | } |
808 | ||
371fd7e7 | 809 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 810 | { |
a64692a3 | 811 | update_rq_clock(rq); |
43148951 | 812 | sched_info_dequeued(rq, p); |
371fd7e7 | 813 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
814 | } |
815 | ||
029632fb | 816 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
817 | { |
818 | if (task_contributes_to_load(p)) | |
819 | rq->nr_uninterruptible--; | |
820 | ||
371fd7e7 | 821 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
822 | } |
823 | ||
029632fb | 824 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
825 | { |
826 | if (task_contributes_to_load(p)) | |
827 | rq->nr_uninterruptible++; | |
828 | ||
371fd7e7 | 829 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
830 | } |
831 | ||
fe44d621 | 832 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 833 | { |
095c0aa8 GC |
834 | /* |
835 | * In theory, the compile should just see 0 here, and optimize out the call | |
836 | * to sched_rt_avg_update. But I don't trust it... | |
837 | */ | |
838 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
839 | s64 steal = 0, irq_delta = 0; | |
840 | #endif | |
841 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 842 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
843 | |
844 | /* | |
845 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
846 | * this case when a previous update_rq_clock() happened inside a | |
847 | * {soft,}irq region. | |
848 | * | |
849 | * When this happens, we stop ->clock_task and only update the | |
850 | * prev_irq_time stamp to account for the part that fit, so that a next | |
851 | * update will consume the rest. This ensures ->clock_task is | |
852 | * monotonic. | |
853 | * | |
854 | * It does however cause some slight miss-attribution of {soft,}irq | |
855 | * time, a more accurate solution would be to update the irq_time using | |
856 | * the current rq->clock timestamp, except that would require using | |
857 | * atomic ops. | |
858 | */ | |
859 | if (irq_delta > delta) | |
860 | irq_delta = delta; | |
861 | ||
862 | rq->prev_irq_time += irq_delta; | |
863 | delta -= irq_delta; | |
095c0aa8 GC |
864 | #endif |
865 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 866 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
867 | steal = paravirt_steal_clock(cpu_of(rq)); |
868 | steal -= rq->prev_steal_time_rq; | |
869 | ||
870 | if (unlikely(steal > delta)) | |
871 | steal = delta; | |
872 | ||
095c0aa8 | 873 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
874 | delta -= steal; |
875 | } | |
876 | #endif | |
877 | ||
fe44d621 PZ |
878 | rq->clock_task += delta; |
879 | ||
095c0aa8 | 880 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 881 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
882 | sched_rt_avg_update(rq, irq_delta + steal); |
883 | #endif | |
aa483808 VP |
884 | } |
885 | ||
34f971f6 PZ |
886 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
887 | { | |
888 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
889 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
890 | ||
891 | if (stop) { | |
892 | /* | |
893 | * Make it appear like a SCHED_FIFO task, its something | |
894 | * userspace knows about and won't get confused about. | |
895 | * | |
896 | * Also, it will make PI more or less work without too | |
897 | * much confusion -- but then, stop work should not | |
898 | * rely on PI working anyway. | |
899 | */ | |
900 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
901 | ||
902 | stop->sched_class = &stop_sched_class; | |
903 | } | |
904 | ||
905 | cpu_rq(cpu)->stop = stop; | |
906 | ||
907 | if (old_stop) { | |
908 | /* | |
909 | * Reset it back to a normal scheduling class so that | |
910 | * it can die in pieces. | |
911 | */ | |
912 | old_stop->sched_class = &rt_sched_class; | |
913 | } | |
914 | } | |
915 | ||
14531189 | 916 | /* |
dd41f596 | 917 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 918 | */ |
14531189 IM |
919 | static inline int __normal_prio(struct task_struct *p) |
920 | { | |
dd41f596 | 921 | return p->static_prio; |
14531189 IM |
922 | } |
923 | ||
b29739f9 IM |
924 | /* |
925 | * Calculate the expected normal priority: i.e. priority | |
926 | * without taking RT-inheritance into account. Might be | |
927 | * boosted by interactivity modifiers. Changes upon fork, | |
928 | * setprio syscalls, and whenever the interactivity | |
929 | * estimator recalculates. | |
930 | */ | |
36c8b586 | 931 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
932 | { |
933 | int prio; | |
934 | ||
aab03e05 DF |
935 | if (task_has_dl_policy(p)) |
936 | prio = MAX_DL_PRIO-1; | |
937 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
938 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
939 | else | |
940 | prio = __normal_prio(p); | |
941 | return prio; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Calculate the current priority, i.e. the priority | |
946 | * taken into account by the scheduler. This value might | |
947 | * be boosted by RT tasks, or might be boosted by | |
948 | * interactivity modifiers. Will be RT if the task got | |
949 | * RT-boosted. If not then it returns p->normal_prio. | |
950 | */ | |
36c8b586 | 951 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
952 | { |
953 | p->normal_prio = normal_prio(p); | |
954 | /* | |
955 | * If we are RT tasks or we were boosted to RT priority, | |
956 | * keep the priority unchanged. Otherwise, update priority | |
957 | * to the normal priority: | |
958 | */ | |
959 | if (!rt_prio(p->prio)) | |
960 | return p->normal_prio; | |
961 | return p->prio; | |
962 | } | |
963 | ||
1da177e4 LT |
964 | /** |
965 | * task_curr - is this task currently executing on a CPU? | |
966 | * @p: the task in question. | |
e69f6186 YB |
967 | * |
968 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 969 | */ |
36c8b586 | 970 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
971 | { |
972 | return cpu_curr(task_cpu(p)) == p; | |
973 | } | |
974 | ||
67dfa1b7 KT |
975 | /* |
976 | * Can drop rq->lock because from sched_class::switched_from() methods drop it. | |
977 | */ | |
cb469845 SR |
978 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
979 | const struct sched_class *prev_class, | |
da7a735e | 980 | int oldprio) |
cb469845 SR |
981 | { |
982 | if (prev_class != p->sched_class) { | |
983 | if (prev_class->switched_from) | |
da7a735e | 984 | prev_class->switched_from(rq, p); |
67dfa1b7 | 985 | /* Possble rq->lock 'hole'. */ |
da7a735e | 986 | p->sched_class->switched_to(rq, p); |
2d3d891d | 987 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 988 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
989 | } |
990 | ||
029632fb | 991 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
992 | { |
993 | const struct sched_class *class; | |
994 | ||
995 | if (p->sched_class == rq->curr->sched_class) { | |
996 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
997 | } else { | |
998 | for_each_class(class) { | |
999 | if (class == rq->curr->sched_class) | |
1000 | break; | |
1001 | if (class == p->sched_class) { | |
8875125e | 1002 | resched_curr(rq); |
1e5a7405 PZ |
1003 | break; |
1004 | } | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * A queue event has occurred, and we're going to schedule. In | |
1010 | * this case, we can save a useless back to back clock update. | |
1011 | */ | |
da0c1e65 | 1012 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 1013 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
1014 | } |
1015 | ||
1da177e4 | 1016 | #ifdef CONFIG_SMP |
dd41f596 | 1017 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1018 | { |
e2912009 PZ |
1019 | #ifdef CONFIG_SCHED_DEBUG |
1020 | /* | |
1021 | * We should never call set_task_cpu() on a blocked task, | |
1022 | * ttwu() will sort out the placement. | |
1023 | */ | |
077614ee | 1024 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1025 | !p->on_rq); |
0122ec5b PZ |
1026 | |
1027 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1028 | /* |
1029 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1030 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1031 | * | |
1032 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1033 | * see task_group(). |
6c6c54e1 PZ |
1034 | * |
1035 | * Furthermore, all task_rq users should acquire both locks, see | |
1036 | * task_rq_lock(). | |
1037 | */ | |
0122ec5b PZ |
1038 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1039 | lockdep_is_held(&task_rq(p)->lock))); | |
1040 | #endif | |
e2912009 PZ |
1041 | #endif |
1042 | ||
de1d7286 | 1043 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1044 | |
0c69774e | 1045 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1046 | if (p->sched_class->migrate_task_rq) |
1047 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1048 | p->se.nr_migrations++; |
86038c5e | 1049 | perf_sw_event_sched(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 0); |
0c69774e | 1050 | } |
dd41f596 IM |
1051 | |
1052 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1053 | } |
1054 | ||
ac66f547 PZ |
1055 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1056 | { | |
da0c1e65 | 1057 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1058 | struct rq *src_rq, *dst_rq; |
1059 | ||
1060 | src_rq = task_rq(p); | |
1061 | dst_rq = cpu_rq(cpu); | |
1062 | ||
1063 | deactivate_task(src_rq, p, 0); | |
1064 | set_task_cpu(p, cpu); | |
1065 | activate_task(dst_rq, p, 0); | |
1066 | check_preempt_curr(dst_rq, p, 0); | |
1067 | } else { | |
1068 | /* | |
1069 | * Task isn't running anymore; make it appear like we migrated | |
1070 | * it before it went to sleep. This means on wakeup we make the | |
1071 | * previous cpu our targer instead of where it really is. | |
1072 | */ | |
1073 | p->wake_cpu = cpu; | |
1074 | } | |
1075 | } | |
1076 | ||
1077 | struct migration_swap_arg { | |
1078 | struct task_struct *src_task, *dst_task; | |
1079 | int src_cpu, dst_cpu; | |
1080 | }; | |
1081 | ||
1082 | static int migrate_swap_stop(void *data) | |
1083 | { | |
1084 | struct migration_swap_arg *arg = data; | |
1085 | struct rq *src_rq, *dst_rq; | |
1086 | int ret = -EAGAIN; | |
1087 | ||
1088 | src_rq = cpu_rq(arg->src_cpu); | |
1089 | dst_rq = cpu_rq(arg->dst_cpu); | |
1090 | ||
74602315 PZ |
1091 | double_raw_lock(&arg->src_task->pi_lock, |
1092 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1093 | double_rq_lock(src_rq, dst_rq); |
1094 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1095 | goto unlock; | |
1096 | ||
1097 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1098 | goto unlock; | |
1099 | ||
1100 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1101 | goto unlock; | |
1102 | ||
1103 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1104 | goto unlock; | |
1105 | ||
1106 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1107 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1108 | ||
1109 | ret = 0; | |
1110 | ||
1111 | unlock: | |
1112 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1113 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1114 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1115 | |
1116 | return ret; | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * Cross migrate two tasks | |
1121 | */ | |
1122 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1123 | { | |
1124 | struct migration_swap_arg arg; | |
1125 | int ret = -EINVAL; | |
1126 | ||
ac66f547 PZ |
1127 | arg = (struct migration_swap_arg){ |
1128 | .src_task = cur, | |
1129 | .src_cpu = task_cpu(cur), | |
1130 | .dst_task = p, | |
1131 | .dst_cpu = task_cpu(p), | |
1132 | }; | |
1133 | ||
1134 | if (arg.src_cpu == arg.dst_cpu) | |
1135 | goto out; | |
1136 | ||
6acce3ef PZ |
1137 | /* |
1138 | * These three tests are all lockless; this is OK since all of them | |
1139 | * will be re-checked with proper locks held further down the line. | |
1140 | */ | |
ac66f547 PZ |
1141 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1142 | goto out; | |
1143 | ||
1144 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1145 | goto out; | |
1146 | ||
1147 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1148 | goto out; | |
1149 | ||
286549dc | 1150 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1151 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1152 | ||
1153 | out: | |
ac66f547 PZ |
1154 | return ret; |
1155 | } | |
1156 | ||
969c7921 | 1157 | struct migration_arg { |
36c8b586 | 1158 | struct task_struct *task; |
1da177e4 | 1159 | int dest_cpu; |
70b97a7f | 1160 | }; |
1da177e4 | 1161 | |
969c7921 TH |
1162 | static int migration_cpu_stop(void *data); |
1163 | ||
1da177e4 LT |
1164 | /* |
1165 | * wait_task_inactive - wait for a thread to unschedule. | |
1166 | * | |
85ba2d86 RM |
1167 | * If @match_state is nonzero, it's the @p->state value just checked and |
1168 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1169 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1170 | * we return a positive number (its total switch count). If a second call | |
1171 | * a short while later returns the same number, the caller can be sure that | |
1172 | * @p has remained unscheduled the whole time. | |
1173 | * | |
1da177e4 LT |
1174 | * The caller must ensure that the task *will* unschedule sometime soon, |
1175 | * else this function might spin for a *long* time. This function can't | |
1176 | * be called with interrupts off, or it may introduce deadlock with | |
1177 | * smp_call_function() if an IPI is sent by the same process we are | |
1178 | * waiting to become inactive. | |
1179 | */ | |
85ba2d86 | 1180 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1181 | { |
1182 | unsigned long flags; | |
da0c1e65 | 1183 | int running, queued; |
85ba2d86 | 1184 | unsigned long ncsw; |
70b97a7f | 1185 | struct rq *rq; |
1da177e4 | 1186 | |
3a5c359a AK |
1187 | for (;;) { |
1188 | /* | |
1189 | * We do the initial early heuristics without holding | |
1190 | * any task-queue locks at all. We'll only try to get | |
1191 | * the runqueue lock when things look like they will | |
1192 | * work out! | |
1193 | */ | |
1194 | rq = task_rq(p); | |
fa490cfd | 1195 | |
3a5c359a AK |
1196 | /* |
1197 | * If the task is actively running on another CPU | |
1198 | * still, just relax and busy-wait without holding | |
1199 | * any locks. | |
1200 | * | |
1201 | * NOTE! Since we don't hold any locks, it's not | |
1202 | * even sure that "rq" stays as the right runqueue! | |
1203 | * But we don't care, since "task_running()" will | |
1204 | * return false if the runqueue has changed and p | |
1205 | * is actually now running somewhere else! | |
1206 | */ | |
85ba2d86 RM |
1207 | while (task_running(rq, p)) { |
1208 | if (match_state && unlikely(p->state != match_state)) | |
1209 | return 0; | |
3a5c359a | 1210 | cpu_relax(); |
85ba2d86 | 1211 | } |
fa490cfd | 1212 | |
3a5c359a AK |
1213 | /* |
1214 | * Ok, time to look more closely! We need the rq | |
1215 | * lock now, to be *sure*. If we're wrong, we'll | |
1216 | * just go back and repeat. | |
1217 | */ | |
1218 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1219 | trace_sched_wait_task(p); |
3a5c359a | 1220 | running = task_running(rq, p); |
da0c1e65 | 1221 | queued = task_on_rq_queued(p); |
85ba2d86 | 1222 | ncsw = 0; |
f31e11d8 | 1223 | if (!match_state || p->state == match_state) |
93dcf55f | 1224 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1225 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1226 | |
85ba2d86 RM |
1227 | /* |
1228 | * If it changed from the expected state, bail out now. | |
1229 | */ | |
1230 | if (unlikely(!ncsw)) | |
1231 | break; | |
1232 | ||
3a5c359a AK |
1233 | /* |
1234 | * Was it really running after all now that we | |
1235 | * checked with the proper locks actually held? | |
1236 | * | |
1237 | * Oops. Go back and try again.. | |
1238 | */ | |
1239 | if (unlikely(running)) { | |
1240 | cpu_relax(); | |
1241 | continue; | |
1242 | } | |
fa490cfd | 1243 | |
3a5c359a AK |
1244 | /* |
1245 | * It's not enough that it's not actively running, | |
1246 | * it must be off the runqueue _entirely_, and not | |
1247 | * preempted! | |
1248 | * | |
80dd99b3 | 1249 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1250 | * running right now), it's preempted, and we should |
1251 | * yield - it could be a while. | |
1252 | */ | |
da0c1e65 | 1253 | if (unlikely(queued)) { |
8eb90c30 TG |
1254 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1255 | ||
1256 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1257 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1258 | continue; |
1259 | } | |
fa490cfd | 1260 | |
3a5c359a AK |
1261 | /* |
1262 | * Ahh, all good. It wasn't running, and it wasn't | |
1263 | * runnable, which means that it will never become | |
1264 | * running in the future either. We're all done! | |
1265 | */ | |
1266 | break; | |
1267 | } | |
85ba2d86 RM |
1268 | |
1269 | return ncsw; | |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | /*** | |
1273 | * kick_process - kick a running thread to enter/exit the kernel | |
1274 | * @p: the to-be-kicked thread | |
1275 | * | |
1276 | * Cause a process which is running on another CPU to enter | |
1277 | * kernel-mode, without any delay. (to get signals handled.) | |
1278 | * | |
25985edc | 1279 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1280 | * because all it wants to ensure is that the remote task enters |
1281 | * the kernel. If the IPI races and the task has been migrated | |
1282 | * to another CPU then no harm is done and the purpose has been | |
1283 | * achieved as well. | |
1284 | */ | |
36c8b586 | 1285 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1286 | { |
1287 | int cpu; | |
1288 | ||
1289 | preempt_disable(); | |
1290 | cpu = task_cpu(p); | |
1291 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1292 | smp_send_reschedule(cpu); | |
1293 | preempt_enable(); | |
1294 | } | |
b43e3521 | 1295 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1296 | #endif /* CONFIG_SMP */ |
1da177e4 | 1297 | |
970b13ba | 1298 | #ifdef CONFIG_SMP |
30da688e | 1299 | /* |
013fdb80 | 1300 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1301 | */ |
5da9a0fb PZ |
1302 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1303 | { | |
aa00d89c TC |
1304 | int nid = cpu_to_node(cpu); |
1305 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1306 | enum { cpuset, possible, fail } state = cpuset; |
1307 | int dest_cpu; | |
5da9a0fb | 1308 | |
aa00d89c TC |
1309 | /* |
1310 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1311 | * will return -1. There is no cpu on the node, and we should | |
1312 | * select the cpu on the other node. | |
1313 | */ | |
1314 | if (nid != -1) { | |
1315 | nodemask = cpumask_of_node(nid); | |
1316 | ||
1317 | /* Look for allowed, online CPU in same node. */ | |
1318 | for_each_cpu(dest_cpu, nodemask) { | |
1319 | if (!cpu_online(dest_cpu)) | |
1320 | continue; | |
1321 | if (!cpu_active(dest_cpu)) | |
1322 | continue; | |
1323 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1324 | return dest_cpu; | |
1325 | } | |
2baab4e9 | 1326 | } |
5da9a0fb | 1327 | |
2baab4e9 PZ |
1328 | for (;;) { |
1329 | /* Any allowed, online CPU? */ | |
e3831edd | 1330 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1331 | if (!cpu_online(dest_cpu)) |
1332 | continue; | |
1333 | if (!cpu_active(dest_cpu)) | |
1334 | continue; | |
1335 | goto out; | |
1336 | } | |
5da9a0fb | 1337 | |
2baab4e9 PZ |
1338 | switch (state) { |
1339 | case cpuset: | |
1340 | /* No more Mr. Nice Guy. */ | |
1341 | cpuset_cpus_allowed_fallback(p); | |
1342 | state = possible; | |
1343 | break; | |
1344 | ||
1345 | case possible: | |
1346 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1347 | state = fail; | |
1348 | break; | |
1349 | ||
1350 | case fail: | |
1351 | BUG(); | |
1352 | break; | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | out: | |
1357 | if (state != cpuset) { | |
1358 | /* | |
1359 | * Don't tell them about moving exiting tasks or | |
1360 | * kernel threads (both mm NULL), since they never | |
1361 | * leave kernel. | |
1362 | */ | |
1363 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1364 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1365 | task_pid_nr(p), p->comm, cpu); |
1366 | } | |
5da9a0fb PZ |
1367 | } |
1368 | ||
1369 | return dest_cpu; | |
1370 | } | |
1371 | ||
e2912009 | 1372 | /* |
013fdb80 | 1373 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1374 | */ |
970b13ba | 1375 | static inline |
ac66f547 | 1376 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1377 | { |
6c1d9410 WL |
1378 | if (p->nr_cpus_allowed > 1) |
1379 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e2912009 PZ |
1380 | |
1381 | /* | |
1382 | * In order not to call set_task_cpu() on a blocking task we need | |
1383 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1384 | * cpu. | |
1385 | * | |
1386 | * Since this is common to all placement strategies, this lives here. | |
1387 | * | |
1388 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1389 | * not worry about this generic constraint ] | |
1390 | */ | |
fa17b507 | 1391 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1392 | !cpu_online(cpu))) |
5da9a0fb | 1393 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1394 | |
1395 | return cpu; | |
970b13ba | 1396 | } |
09a40af5 MG |
1397 | |
1398 | static void update_avg(u64 *avg, u64 sample) | |
1399 | { | |
1400 | s64 diff = sample - *avg; | |
1401 | *avg += diff >> 3; | |
1402 | } | |
970b13ba PZ |
1403 | #endif |
1404 | ||
d7c01d27 | 1405 | static void |
b84cb5df | 1406 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1407 | { |
d7c01d27 | 1408 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1409 | struct rq *rq = this_rq(); |
1410 | ||
d7c01d27 PZ |
1411 | #ifdef CONFIG_SMP |
1412 | int this_cpu = smp_processor_id(); | |
1413 | ||
1414 | if (cpu == this_cpu) { | |
1415 | schedstat_inc(rq, ttwu_local); | |
1416 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1417 | } else { | |
1418 | struct sched_domain *sd; | |
1419 | ||
1420 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1421 | rcu_read_lock(); |
d7c01d27 PZ |
1422 | for_each_domain(this_cpu, sd) { |
1423 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1424 | schedstat_inc(sd, ttwu_wake_remote); | |
1425 | break; | |
1426 | } | |
1427 | } | |
057f3fad | 1428 | rcu_read_unlock(); |
d7c01d27 | 1429 | } |
f339b9dc PZ |
1430 | |
1431 | if (wake_flags & WF_MIGRATED) | |
1432 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1433 | ||
d7c01d27 PZ |
1434 | #endif /* CONFIG_SMP */ |
1435 | ||
1436 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1437 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1438 | |
1439 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1440 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1441 | |
d7c01d27 PZ |
1442 | #endif /* CONFIG_SCHEDSTATS */ |
1443 | } | |
1444 | ||
1445 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1446 | { | |
9ed3811a | 1447 | activate_task(rq, p, en_flags); |
da0c1e65 | 1448 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1449 | |
1450 | /* if a worker is waking up, notify workqueue */ | |
1451 | if (p->flags & PF_WQ_WORKER) | |
1452 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1453 | } |
1454 | ||
23f41eeb PZ |
1455 | /* |
1456 | * Mark the task runnable and perform wakeup-preemption. | |
1457 | */ | |
89363381 | 1458 | static void |
23f41eeb | 1459 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1460 | { |
9ed3811a | 1461 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1462 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1463 | |
1464 | p->state = TASK_RUNNING; | |
1465 | #ifdef CONFIG_SMP | |
1466 | if (p->sched_class->task_woken) | |
1467 | p->sched_class->task_woken(rq, p); | |
1468 | ||
e69c6341 | 1469 | if (rq->idle_stamp) { |
78becc27 | 1470 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1471 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1472 | |
abfafa54 JL |
1473 | update_avg(&rq->avg_idle, delta); |
1474 | ||
1475 | if (rq->avg_idle > max) | |
9ed3811a | 1476 | rq->avg_idle = max; |
abfafa54 | 1477 | |
9ed3811a TH |
1478 | rq->idle_stamp = 0; |
1479 | } | |
1480 | #endif | |
1481 | } | |
1482 | ||
c05fbafb PZ |
1483 | static void |
1484 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1485 | { | |
1486 | #ifdef CONFIG_SMP | |
1487 | if (p->sched_contributes_to_load) | |
1488 | rq->nr_uninterruptible--; | |
1489 | #endif | |
1490 | ||
1491 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1492 | ttwu_do_wakeup(rq, p, wake_flags); | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1497 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1498 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1499 | * the task is still ->on_rq. | |
1500 | */ | |
1501 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1502 | { | |
1503 | struct rq *rq; | |
1504 | int ret = 0; | |
1505 | ||
1506 | rq = __task_rq_lock(p); | |
da0c1e65 | 1507 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1508 | /* check_preempt_curr() may use rq clock */ |
1509 | update_rq_clock(rq); | |
c05fbafb PZ |
1510 | ttwu_do_wakeup(rq, p, wake_flags); |
1511 | ret = 1; | |
1512 | } | |
1513 | __task_rq_unlock(rq); | |
1514 | ||
1515 | return ret; | |
1516 | } | |
1517 | ||
317f3941 | 1518 | #ifdef CONFIG_SMP |
e3baac47 | 1519 | void sched_ttwu_pending(void) |
317f3941 PZ |
1520 | { |
1521 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1522 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1523 | struct task_struct *p; | |
e3baac47 | 1524 | unsigned long flags; |
317f3941 | 1525 | |
e3baac47 PZ |
1526 | if (!llist) |
1527 | return; | |
1528 | ||
1529 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1530 | |
fa14ff4a PZ |
1531 | while (llist) { |
1532 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1533 | llist = llist_next(llist); | |
317f3941 PZ |
1534 | ttwu_do_activate(rq, p, 0); |
1535 | } | |
1536 | ||
e3baac47 | 1537 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1538 | } |
1539 | ||
1540 | void scheduler_ipi(void) | |
1541 | { | |
f27dde8d PZ |
1542 | /* |
1543 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1544 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1545 | * this IPI. | |
1546 | */ | |
8cb75e0c | 1547 | preempt_fold_need_resched(); |
f27dde8d | 1548 | |
fd2ac4f4 | 1549 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1550 | return; |
1551 | ||
1552 | /* | |
1553 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1554 | * traditionally all their work was done from the interrupt return | |
1555 | * path. Now that we actually do some work, we need to make sure | |
1556 | * we do call them. | |
1557 | * | |
1558 | * Some archs already do call them, luckily irq_enter/exit nest | |
1559 | * properly. | |
1560 | * | |
1561 | * Arguably we should visit all archs and update all handlers, | |
1562 | * however a fair share of IPIs are still resched only so this would | |
1563 | * somewhat pessimize the simple resched case. | |
1564 | */ | |
1565 | irq_enter(); | |
fa14ff4a | 1566 | sched_ttwu_pending(); |
ca38062e SS |
1567 | |
1568 | /* | |
1569 | * Check if someone kicked us for doing the nohz idle load balance. | |
1570 | */ | |
873b4c65 | 1571 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1572 | this_rq()->idle_balance = 1; |
ca38062e | 1573 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1574 | } |
c5d753a5 | 1575 | irq_exit(); |
317f3941 PZ |
1576 | } |
1577 | ||
1578 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1579 | { | |
e3baac47 PZ |
1580 | struct rq *rq = cpu_rq(cpu); |
1581 | ||
1582 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1583 | if (!set_nr_if_polling(rq->idle)) | |
1584 | smp_send_reschedule(cpu); | |
1585 | else | |
1586 | trace_sched_wake_idle_without_ipi(cpu); | |
1587 | } | |
317f3941 | 1588 | } |
d6aa8f85 | 1589 | |
f6be8af1 CL |
1590 | void wake_up_if_idle(int cpu) |
1591 | { | |
1592 | struct rq *rq = cpu_rq(cpu); | |
1593 | unsigned long flags; | |
1594 | ||
fd7de1e8 AL |
1595 | rcu_read_lock(); |
1596 | ||
1597 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1598 | goto out; | |
f6be8af1 CL |
1599 | |
1600 | if (set_nr_if_polling(rq->idle)) { | |
1601 | trace_sched_wake_idle_without_ipi(cpu); | |
1602 | } else { | |
1603 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1604 | if (is_idle_task(rq->curr)) | |
1605 | smp_send_reschedule(cpu); | |
1606 | /* Else cpu is not in idle, do nothing here */ | |
1607 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1608 | } | |
fd7de1e8 AL |
1609 | |
1610 | out: | |
1611 | rcu_read_unlock(); | |
f6be8af1 CL |
1612 | } |
1613 | ||
39be3501 | 1614 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1615 | { |
1616 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1617 | } | |
d6aa8f85 | 1618 | #endif /* CONFIG_SMP */ |
317f3941 | 1619 | |
c05fbafb PZ |
1620 | static void ttwu_queue(struct task_struct *p, int cpu) |
1621 | { | |
1622 | struct rq *rq = cpu_rq(cpu); | |
1623 | ||
17d9f311 | 1624 | #if defined(CONFIG_SMP) |
39be3501 | 1625 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1626 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1627 | ttwu_queue_remote(p, cpu); |
1628 | return; | |
1629 | } | |
1630 | #endif | |
1631 | ||
c05fbafb PZ |
1632 | raw_spin_lock(&rq->lock); |
1633 | ttwu_do_activate(rq, p, 0); | |
1634 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1635 | } |
1636 | ||
1637 | /** | |
1da177e4 | 1638 | * try_to_wake_up - wake up a thread |
9ed3811a | 1639 | * @p: the thread to be awakened |
1da177e4 | 1640 | * @state: the mask of task states that can be woken |
9ed3811a | 1641 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1642 | * |
1643 | * Put it on the run-queue if it's not already there. The "current" | |
1644 | * thread is always on the run-queue (except when the actual | |
1645 | * re-schedule is in progress), and as such you're allowed to do | |
1646 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1647 | * runnable without the overhead of this. | |
1648 | * | |
e69f6186 | 1649 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1650 | * or @state didn't match @p's state. |
1da177e4 | 1651 | */ |
e4a52bcb PZ |
1652 | static int |
1653 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1654 | { |
1da177e4 | 1655 | unsigned long flags; |
c05fbafb | 1656 | int cpu, success = 0; |
2398f2c6 | 1657 | |
e0acd0a6 ON |
1658 | /* |
1659 | * If we are going to wake up a thread waiting for CONDITION we | |
1660 | * need to ensure that CONDITION=1 done by the caller can not be | |
1661 | * reordered with p->state check below. This pairs with mb() in | |
1662 | * set_current_state() the waiting thread does. | |
1663 | */ | |
1664 | smp_mb__before_spinlock(); | |
013fdb80 | 1665 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1666 | if (!(p->state & state)) |
1da177e4 LT |
1667 | goto out; |
1668 | ||
c05fbafb | 1669 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1670 | cpu = task_cpu(p); |
1da177e4 | 1671 | |
c05fbafb PZ |
1672 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1673 | goto stat; | |
1da177e4 | 1674 | |
1da177e4 | 1675 | #ifdef CONFIG_SMP |
e9c84311 | 1676 | /* |
c05fbafb PZ |
1677 | * If the owning (remote) cpu is still in the middle of schedule() with |
1678 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1679 | */ |
f3e94786 | 1680 | while (p->on_cpu) |
e4a52bcb | 1681 | cpu_relax(); |
0970d299 | 1682 | /* |
e4a52bcb | 1683 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1684 | */ |
e4a52bcb | 1685 | smp_rmb(); |
1da177e4 | 1686 | |
a8e4f2ea | 1687 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1688 | p->state = TASK_WAKING; |
e7693a36 | 1689 | |
e4a52bcb | 1690 | if (p->sched_class->task_waking) |
74f8e4b2 | 1691 | p->sched_class->task_waking(p); |
efbbd05a | 1692 | |
ac66f547 | 1693 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1694 | if (task_cpu(p) != cpu) { |
1695 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1696 | set_task_cpu(p, cpu); |
f339b9dc | 1697 | } |
1da177e4 | 1698 | #endif /* CONFIG_SMP */ |
1da177e4 | 1699 | |
c05fbafb PZ |
1700 | ttwu_queue(p, cpu); |
1701 | stat: | |
b84cb5df | 1702 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1703 | out: |
013fdb80 | 1704 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1705 | |
1706 | return success; | |
1707 | } | |
1708 | ||
21aa9af0 TH |
1709 | /** |
1710 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1711 | * @p: the thread to be awakened | |
1712 | * | |
2acca55e | 1713 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1714 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1715 | * the current task. |
21aa9af0 TH |
1716 | */ |
1717 | static void try_to_wake_up_local(struct task_struct *p) | |
1718 | { | |
1719 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1720 | |
383efcd0 TH |
1721 | if (WARN_ON_ONCE(rq != this_rq()) || |
1722 | WARN_ON_ONCE(p == current)) | |
1723 | return; | |
1724 | ||
21aa9af0 TH |
1725 | lockdep_assert_held(&rq->lock); |
1726 | ||
2acca55e PZ |
1727 | if (!raw_spin_trylock(&p->pi_lock)) { |
1728 | raw_spin_unlock(&rq->lock); | |
1729 | raw_spin_lock(&p->pi_lock); | |
1730 | raw_spin_lock(&rq->lock); | |
1731 | } | |
1732 | ||
21aa9af0 | 1733 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1734 | goto out; |
21aa9af0 | 1735 | |
da0c1e65 | 1736 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
1737 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1738 | ||
23f41eeb | 1739 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1740 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1741 | out: |
1742 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1743 | } |
1744 | ||
50fa610a DH |
1745 | /** |
1746 | * wake_up_process - Wake up a specific process | |
1747 | * @p: The process to be woken up. | |
1748 | * | |
1749 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1750 | * processes. |
1751 | * | |
1752 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1753 | * |
1754 | * It may be assumed that this function implies a write memory barrier before | |
1755 | * changing the task state if and only if any tasks are woken up. | |
1756 | */ | |
7ad5b3a5 | 1757 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1758 | { |
9067ac85 ON |
1759 | WARN_ON(task_is_stopped_or_traced(p)); |
1760 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1761 | } |
1da177e4 LT |
1762 | EXPORT_SYMBOL(wake_up_process); |
1763 | ||
7ad5b3a5 | 1764 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1765 | { |
1766 | return try_to_wake_up(p, state, 0); | |
1767 | } | |
1768 | ||
a5e7be3b JL |
1769 | /* |
1770 | * This function clears the sched_dl_entity static params. | |
1771 | */ | |
1772 | void __dl_clear_params(struct task_struct *p) | |
1773 | { | |
1774 | struct sched_dl_entity *dl_se = &p->dl; | |
1775 | ||
1776 | dl_se->dl_runtime = 0; | |
1777 | dl_se->dl_deadline = 0; | |
1778 | dl_se->dl_period = 0; | |
1779 | dl_se->flags = 0; | |
1780 | dl_se->dl_bw = 0; | |
40767b0d PZ |
1781 | |
1782 | dl_se->dl_throttled = 0; | |
1783 | dl_se->dl_new = 1; | |
1784 | dl_se->dl_yielded = 0; | |
a5e7be3b JL |
1785 | } |
1786 | ||
1da177e4 LT |
1787 | /* |
1788 | * Perform scheduler related setup for a newly forked process p. | |
1789 | * p is forked by current. | |
dd41f596 IM |
1790 | * |
1791 | * __sched_fork() is basic setup used by init_idle() too: | |
1792 | */ | |
5e1576ed | 1793 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1794 | { |
fd2f4419 PZ |
1795 | p->on_rq = 0; |
1796 | ||
1797 | p->se.on_rq = 0; | |
dd41f596 IM |
1798 | p->se.exec_start = 0; |
1799 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1800 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1801 | p->se.nr_migrations = 0; |
da7a735e | 1802 | p->se.vruntime = 0; |
bb04159d KT |
1803 | #ifdef CONFIG_SMP |
1804 | p->se.avg.decay_count = 0; | |
1805 | #endif | |
fd2f4419 | 1806 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1807 | |
1808 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1809 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1810 | #endif |
476d139c | 1811 | |
aab03e05 | 1812 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 1813 | init_dl_task_timer(&p->dl); |
a5e7be3b | 1814 | __dl_clear_params(p); |
aab03e05 | 1815 | |
fa717060 | 1816 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1817 | |
e107be36 AK |
1818 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1819 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1820 | #endif | |
cbee9f88 PZ |
1821 | |
1822 | #ifdef CONFIG_NUMA_BALANCING | |
1823 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1824 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1825 | p->mm->numa_scan_seq = 0; |
1826 | } | |
1827 | ||
5e1576ed RR |
1828 | if (clone_flags & CLONE_VM) |
1829 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1830 | else | |
1831 | p->numa_preferred_nid = -1; | |
1832 | ||
cbee9f88 PZ |
1833 | p->node_stamp = 0ULL; |
1834 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1835 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1836 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 1837 | p->numa_faults = NULL; |
7e2703e6 RR |
1838 | p->last_task_numa_placement = 0; |
1839 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 1840 | |
8c8a743c | 1841 | p->numa_group = NULL; |
cbee9f88 | 1842 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1843 | } |
1844 | ||
1a687c2e | 1845 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1846 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1847 | void set_numabalancing_state(bool enabled) |
1848 | { | |
1849 | if (enabled) | |
1850 | sched_feat_set("NUMA"); | |
1851 | else | |
1852 | sched_feat_set("NO_NUMA"); | |
1853 | } | |
3105b86a MG |
1854 | #else |
1855 | __read_mostly bool numabalancing_enabled; | |
1856 | ||
1857 | void set_numabalancing_state(bool enabled) | |
1858 | { | |
1859 | numabalancing_enabled = enabled; | |
dd41f596 | 1860 | } |
3105b86a | 1861 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1862 | |
1863 | #ifdef CONFIG_PROC_SYSCTL | |
1864 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1865 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1866 | { | |
1867 | struct ctl_table t; | |
1868 | int err; | |
1869 | int state = numabalancing_enabled; | |
1870 | ||
1871 | if (write && !capable(CAP_SYS_ADMIN)) | |
1872 | return -EPERM; | |
1873 | ||
1874 | t = *table; | |
1875 | t.data = &state; | |
1876 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1877 | if (err < 0) | |
1878 | return err; | |
1879 | if (write) | |
1880 | set_numabalancing_state(state); | |
1881 | return err; | |
1882 | } | |
1883 | #endif | |
1884 | #endif | |
dd41f596 IM |
1885 | |
1886 | /* | |
1887 | * fork()/clone()-time setup: | |
1888 | */ | |
aab03e05 | 1889 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1890 | { |
0122ec5b | 1891 | unsigned long flags; |
dd41f596 IM |
1892 | int cpu = get_cpu(); |
1893 | ||
5e1576ed | 1894 | __sched_fork(clone_flags, p); |
06b83b5f | 1895 | /* |
0017d735 | 1896 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1897 | * nobody will actually run it, and a signal or other external |
1898 | * event cannot wake it up and insert it on the runqueue either. | |
1899 | */ | |
0017d735 | 1900 | p->state = TASK_RUNNING; |
dd41f596 | 1901 | |
c350a04e MG |
1902 | /* |
1903 | * Make sure we do not leak PI boosting priority to the child. | |
1904 | */ | |
1905 | p->prio = current->normal_prio; | |
1906 | ||
b9dc29e7 MG |
1907 | /* |
1908 | * Revert to default priority/policy on fork if requested. | |
1909 | */ | |
1910 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1911 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1912 | p->policy = SCHED_NORMAL; |
6c697bdf | 1913 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1914 | p->rt_priority = 0; |
1915 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1916 | p->static_prio = NICE_TO_PRIO(0); | |
1917 | ||
1918 | p->prio = p->normal_prio = __normal_prio(p); | |
1919 | set_load_weight(p); | |
6c697bdf | 1920 | |
b9dc29e7 MG |
1921 | /* |
1922 | * We don't need the reset flag anymore after the fork. It has | |
1923 | * fulfilled its duty: | |
1924 | */ | |
1925 | p->sched_reset_on_fork = 0; | |
1926 | } | |
ca94c442 | 1927 | |
aab03e05 DF |
1928 | if (dl_prio(p->prio)) { |
1929 | put_cpu(); | |
1930 | return -EAGAIN; | |
1931 | } else if (rt_prio(p->prio)) { | |
1932 | p->sched_class = &rt_sched_class; | |
1933 | } else { | |
2ddbf952 | 1934 | p->sched_class = &fair_sched_class; |
aab03e05 | 1935 | } |
b29739f9 | 1936 | |
cd29fe6f PZ |
1937 | if (p->sched_class->task_fork) |
1938 | p->sched_class->task_fork(p); | |
1939 | ||
86951599 PZ |
1940 | /* |
1941 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1942 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1943 | * is ran before sched_fork(). | |
1944 | * | |
1945 | * Silence PROVE_RCU. | |
1946 | */ | |
0122ec5b | 1947 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1948 | set_task_cpu(p, cpu); |
0122ec5b | 1949 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1950 | |
52f17b6c | 1951 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1952 | if (likely(sched_info_on())) |
52f17b6c | 1953 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1954 | #endif |
3ca7a440 PZ |
1955 | #if defined(CONFIG_SMP) |
1956 | p->on_cpu = 0; | |
4866cde0 | 1957 | #endif |
01028747 | 1958 | init_task_preempt_count(p); |
806c09a7 | 1959 | #ifdef CONFIG_SMP |
917b627d | 1960 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1961 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1962 | #endif |
917b627d | 1963 | |
476d139c | 1964 | put_cpu(); |
aab03e05 | 1965 | return 0; |
1da177e4 LT |
1966 | } |
1967 | ||
332ac17e DF |
1968 | unsigned long to_ratio(u64 period, u64 runtime) |
1969 | { | |
1970 | if (runtime == RUNTIME_INF) | |
1971 | return 1ULL << 20; | |
1972 | ||
1973 | /* | |
1974 | * Doing this here saves a lot of checks in all | |
1975 | * the calling paths, and returning zero seems | |
1976 | * safe for them anyway. | |
1977 | */ | |
1978 | if (period == 0) | |
1979 | return 0; | |
1980 | ||
1981 | return div64_u64(runtime << 20, period); | |
1982 | } | |
1983 | ||
1984 | #ifdef CONFIG_SMP | |
1985 | inline struct dl_bw *dl_bw_of(int i) | |
1986 | { | |
66339c31 KT |
1987 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
1988 | "sched RCU must be held"); | |
332ac17e DF |
1989 | return &cpu_rq(i)->rd->dl_bw; |
1990 | } | |
1991 | ||
de212f18 | 1992 | static inline int dl_bw_cpus(int i) |
332ac17e | 1993 | { |
de212f18 PZ |
1994 | struct root_domain *rd = cpu_rq(i)->rd; |
1995 | int cpus = 0; | |
1996 | ||
66339c31 KT |
1997 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
1998 | "sched RCU must be held"); | |
de212f18 PZ |
1999 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2000 | cpus++; | |
2001 | ||
2002 | return cpus; | |
332ac17e DF |
2003 | } |
2004 | #else | |
2005 | inline struct dl_bw *dl_bw_of(int i) | |
2006 | { | |
2007 | return &cpu_rq(i)->dl.dl_bw; | |
2008 | } | |
2009 | ||
de212f18 | 2010 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2011 | { |
2012 | return 1; | |
2013 | } | |
2014 | #endif | |
2015 | ||
332ac17e DF |
2016 | /* |
2017 | * We must be sure that accepting a new task (or allowing changing the | |
2018 | * parameters of an existing one) is consistent with the bandwidth | |
2019 | * constraints. If yes, this function also accordingly updates the currently | |
2020 | * allocated bandwidth to reflect the new situation. | |
2021 | * | |
2022 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2023 | * |
2024 | * XXX we should delay bw change until the task's 0-lag point, see | |
2025 | * __setparam_dl(). | |
332ac17e DF |
2026 | */ |
2027 | static int dl_overflow(struct task_struct *p, int policy, | |
2028 | const struct sched_attr *attr) | |
2029 | { | |
2030 | ||
2031 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2032 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2033 | u64 runtime = attr->sched_runtime; |
2034 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2035 | int cpus, err = -1; |
332ac17e DF |
2036 | |
2037 | if (new_bw == p->dl.dl_bw) | |
2038 | return 0; | |
2039 | ||
2040 | /* | |
2041 | * Either if a task, enters, leave, or stays -deadline but changes | |
2042 | * its parameters, we may need to update accordingly the total | |
2043 | * allocated bandwidth of the container. | |
2044 | */ | |
2045 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2046 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2047 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2048 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2049 | __dl_add(dl_b, new_bw); | |
2050 | err = 0; | |
2051 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2052 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2053 | __dl_clear(dl_b, p->dl.dl_bw); | |
2054 | __dl_add(dl_b, new_bw); | |
2055 | err = 0; | |
2056 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2057 | __dl_clear(dl_b, p->dl.dl_bw); | |
2058 | err = 0; | |
2059 | } | |
2060 | raw_spin_unlock(&dl_b->lock); | |
2061 | ||
2062 | return err; | |
2063 | } | |
2064 | ||
2065 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2066 | ||
1da177e4 LT |
2067 | /* |
2068 | * wake_up_new_task - wake up a newly created task for the first time. | |
2069 | * | |
2070 | * This function will do some initial scheduler statistics housekeeping | |
2071 | * that must be done for every newly created context, then puts the task | |
2072 | * on the runqueue and wakes it. | |
2073 | */ | |
3e51e3ed | 2074 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2075 | { |
2076 | unsigned long flags; | |
dd41f596 | 2077 | struct rq *rq; |
fabf318e | 2078 | |
ab2515c4 | 2079 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2080 | #ifdef CONFIG_SMP |
2081 | /* | |
2082 | * Fork balancing, do it here and not earlier because: | |
2083 | * - cpus_allowed can change in the fork path | |
2084 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2085 | */ |
ac66f547 | 2086 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2087 | #endif |
2088 | ||
a75cdaa9 AS |
2089 | /* Initialize new task's runnable average */ |
2090 | init_task_runnable_average(p); | |
ab2515c4 | 2091 | rq = __task_rq_lock(p); |
cd29fe6f | 2092 | activate_task(rq, p, 0); |
da0c1e65 | 2093 | p->on_rq = TASK_ON_RQ_QUEUED; |
89363381 | 2094 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2095 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2096 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2097 | if (p->sched_class->task_woken) |
2098 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2099 | #endif |
0122ec5b | 2100 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2101 | } |
2102 | ||
e107be36 AK |
2103 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2104 | ||
2105 | /** | |
80dd99b3 | 2106 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2107 | * @notifier: notifier struct to register |
e107be36 AK |
2108 | */ |
2109 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2110 | { | |
2111 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2112 | } | |
2113 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2114 | ||
2115 | /** | |
2116 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2117 | * @notifier: notifier struct to unregister |
e107be36 AK |
2118 | * |
2119 | * This is safe to call from within a preemption notifier. | |
2120 | */ | |
2121 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2122 | { | |
2123 | hlist_del(¬ifier->link); | |
2124 | } | |
2125 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2126 | ||
2127 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2128 | { | |
2129 | struct preempt_notifier *notifier; | |
e107be36 | 2130 | |
b67bfe0d | 2131 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2132 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2133 | } | |
2134 | ||
2135 | static void | |
2136 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2137 | struct task_struct *next) | |
2138 | { | |
2139 | struct preempt_notifier *notifier; | |
e107be36 | 2140 | |
b67bfe0d | 2141 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2142 | notifier->ops->sched_out(notifier, next); |
2143 | } | |
2144 | ||
6d6bc0ad | 2145 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2146 | |
2147 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2148 | { | |
2149 | } | |
2150 | ||
2151 | static void | |
2152 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2153 | struct task_struct *next) | |
2154 | { | |
2155 | } | |
2156 | ||
6d6bc0ad | 2157 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2158 | |
4866cde0 NP |
2159 | /** |
2160 | * prepare_task_switch - prepare to switch tasks | |
2161 | * @rq: the runqueue preparing to switch | |
421cee29 | 2162 | * @prev: the current task that is being switched out |
4866cde0 NP |
2163 | * @next: the task we are going to switch to. |
2164 | * | |
2165 | * This is called with the rq lock held and interrupts off. It must | |
2166 | * be paired with a subsequent finish_task_switch after the context | |
2167 | * switch. | |
2168 | * | |
2169 | * prepare_task_switch sets up locking and calls architecture specific | |
2170 | * hooks. | |
2171 | */ | |
e107be36 AK |
2172 | static inline void |
2173 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2174 | struct task_struct *next) | |
4866cde0 | 2175 | { |
895dd92c | 2176 | trace_sched_switch(prev, next); |
43148951 | 2177 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2178 | perf_event_task_sched_out(prev, next); |
e107be36 | 2179 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2180 | prepare_lock_switch(rq, next); |
2181 | prepare_arch_switch(next); | |
2182 | } | |
2183 | ||
1da177e4 LT |
2184 | /** |
2185 | * finish_task_switch - clean up after a task-switch | |
2186 | * @prev: the thread we just switched away from. | |
2187 | * | |
4866cde0 NP |
2188 | * finish_task_switch must be called after the context switch, paired |
2189 | * with a prepare_task_switch call before the context switch. | |
2190 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2191 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2192 | * |
2193 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2194 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2195 | * with the lock held can cause deadlocks; see schedule() for |
2196 | * details.) | |
dfa50b60 ON |
2197 | * |
2198 | * The context switch have flipped the stack from under us and restored the | |
2199 | * local variables which were saved when this task called schedule() in the | |
2200 | * past. prev == current is still correct but we need to recalculate this_rq | |
2201 | * because prev may have moved to another CPU. | |
1da177e4 | 2202 | */ |
dfa50b60 | 2203 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2204 | __releases(rq->lock) |
2205 | { | |
dfa50b60 | 2206 | struct rq *rq = this_rq(); |
1da177e4 | 2207 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2208 | long prev_state; |
1da177e4 LT |
2209 | |
2210 | rq->prev_mm = NULL; | |
2211 | ||
2212 | /* | |
2213 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2214 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2215 | * schedule one last time. The schedule call will never return, and |
2216 | * the scheduled task must drop that reference. | |
c394cc9f | 2217 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2218 | * still held, otherwise prev could be scheduled on another cpu, die |
2219 | * there before we look at prev->state, and then the reference would | |
2220 | * be dropped twice. | |
2221 | * Manfred Spraul <manfred@colorfullife.com> | |
2222 | */ | |
55a101f8 | 2223 | prev_state = prev->state; |
bf9fae9f | 2224 | vtime_task_switch(prev); |
4866cde0 | 2225 | finish_arch_switch(prev); |
a8d757ef | 2226 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2227 | finish_lock_switch(rq, prev); |
01f23e16 | 2228 | finish_arch_post_lock_switch(); |
e8fa1362 | 2229 | |
e107be36 | 2230 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2231 | if (mm) |
2232 | mmdrop(mm); | |
c394cc9f | 2233 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2234 | if (prev->sched_class->task_dead) |
2235 | prev->sched_class->task_dead(prev); | |
2236 | ||
c6fd91f0 | 2237 | /* |
2238 | * Remove function-return probe instances associated with this | |
2239 | * task and put them back on the free list. | |
9761eea8 | 2240 | */ |
c6fd91f0 | 2241 | kprobe_flush_task(prev); |
1da177e4 | 2242 | put_task_struct(prev); |
c6fd91f0 | 2243 | } |
99e5ada9 FW |
2244 | |
2245 | tick_nohz_task_switch(current); | |
dfa50b60 | 2246 | return rq; |
1da177e4 LT |
2247 | } |
2248 | ||
3f029d3c GH |
2249 | #ifdef CONFIG_SMP |
2250 | ||
3f029d3c GH |
2251 | /* rq->lock is NOT held, but preemption is disabled */ |
2252 | static inline void post_schedule(struct rq *rq) | |
2253 | { | |
2254 | if (rq->post_schedule) { | |
2255 | unsigned long flags; | |
2256 | ||
05fa785c | 2257 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2258 | if (rq->curr->sched_class->post_schedule) |
2259 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2260 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2261 | |
2262 | rq->post_schedule = 0; | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | #else | |
da19ab51 | 2267 | |
3f029d3c GH |
2268 | static inline void post_schedule(struct rq *rq) |
2269 | { | |
1da177e4 LT |
2270 | } |
2271 | ||
3f029d3c GH |
2272 | #endif |
2273 | ||
1da177e4 LT |
2274 | /** |
2275 | * schedule_tail - first thing a freshly forked thread must call. | |
2276 | * @prev: the thread we just switched away from. | |
2277 | */ | |
722a9f92 | 2278 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2279 | __releases(rq->lock) |
2280 | { | |
1a43a14a | 2281 | struct rq *rq; |
da19ab51 | 2282 | |
1a43a14a ON |
2283 | /* finish_task_switch() drops rq->lock and enables preemtion */ |
2284 | preempt_disable(); | |
dfa50b60 | 2285 | rq = finish_task_switch(prev); |
3f029d3c | 2286 | post_schedule(rq); |
1a43a14a | 2287 | preempt_enable(); |
70b97a7f | 2288 | |
1da177e4 | 2289 | if (current->set_child_tid) |
b488893a | 2290 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2291 | } |
2292 | ||
2293 | /* | |
dfa50b60 | 2294 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2295 | */ |
dfa50b60 | 2296 | static inline struct rq * |
70b97a7f | 2297 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2298 | struct task_struct *next) |
1da177e4 | 2299 | { |
dd41f596 | 2300 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2301 | |
e107be36 | 2302 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2303 | |
dd41f596 IM |
2304 | mm = next->mm; |
2305 | oldmm = prev->active_mm; | |
9226d125 ZA |
2306 | /* |
2307 | * For paravirt, this is coupled with an exit in switch_to to | |
2308 | * combine the page table reload and the switch backend into | |
2309 | * one hypercall. | |
2310 | */ | |
224101ed | 2311 | arch_start_context_switch(prev); |
9226d125 | 2312 | |
31915ab4 | 2313 | if (!mm) { |
1da177e4 LT |
2314 | next->active_mm = oldmm; |
2315 | atomic_inc(&oldmm->mm_count); | |
2316 | enter_lazy_tlb(oldmm, next); | |
2317 | } else | |
2318 | switch_mm(oldmm, mm, next); | |
2319 | ||
31915ab4 | 2320 | if (!prev->mm) { |
1da177e4 | 2321 | prev->active_mm = NULL; |
1da177e4 LT |
2322 | rq->prev_mm = oldmm; |
2323 | } | |
3a5f5e48 IM |
2324 | /* |
2325 | * Since the runqueue lock will be released by the next | |
2326 | * task (which is an invalid locking op but in the case | |
2327 | * of the scheduler it's an obvious special-case), so we | |
2328 | * do an early lockdep release here: | |
2329 | */ | |
8a25d5de | 2330 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 | 2331 | |
91d1aa43 | 2332 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2333 | /* Here we just switch the register state and the stack. */ |
2334 | switch_to(prev, next, prev); | |
dd41f596 | 2335 | barrier(); |
dfa50b60 ON |
2336 | |
2337 | return finish_task_switch(prev); | |
1da177e4 LT |
2338 | } |
2339 | ||
2340 | /* | |
1c3e8264 | 2341 | * nr_running and nr_context_switches: |
1da177e4 LT |
2342 | * |
2343 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2344 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2345 | */ |
2346 | unsigned long nr_running(void) | |
2347 | { | |
2348 | unsigned long i, sum = 0; | |
2349 | ||
2350 | for_each_online_cpu(i) | |
2351 | sum += cpu_rq(i)->nr_running; | |
2352 | ||
2353 | return sum; | |
f711f609 | 2354 | } |
1da177e4 | 2355 | |
2ee507c4 TC |
2356 | /* |
2357 | * Check if only the current task is running on the cpu. | |
2358 | */ | |
2359 | bool single_task_running(void) | |
2360 | { | |
2361 | if (cpu_rq(smp_processor_id())->nr_running == 1) | |
2362 | return true; | |
2363 | else | |
2364 | return false; | |
2365 | } | |
2366 | EXPORT_SYMBOL(single_task_running); | |
2367 | ||
1da177e4 | 2368 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2369 | { |
cc94abfc SR |
2370 | int i; |
2371 | unsigned long long sum = 0; | |
46cb4b7c | 2372 | |
0a945022 | 2373 | for_each_possible_cpu(i) |
1da177e4 | 2374 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2375 | |
1da177e4 LT |
2376 | return sum; |
2377 | } | |
483b4ee6 | 2378 | |
1da177e4 LT |
2379 | unsigned long nr_iowait(void) |
2380 | { | |
2381 | unsigned long i, sum = 0; | |
483b4ee6 | 2382 | |
0a945022 | 2383 | for_each_possible_cpu(i) |
1da177e4 | 2384 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2385 | |
1da177e4 LT |
2386 | return sum; |
2387 | } | |
483b4ee6 | 2388 | |
8c215bd3 | 2389 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2390 | { |
8c215bd3 | 2391 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2392 | return atomic_read(&this->nr_iowait); |
2393 | } | |
46cb4b7c | 2394 | |
372ba8cb MG |
2395 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2396 | { | |
2397 | struct rq *this = this_rq(); | |
2398 | *nr_waiters = atomic_read(&this->nr_iowait); | |
2399 | *load = this->cpu_load[0]; | |
2400 | } | |
2401 | ||
dd41f596 | 2402 | #ifdef CONFIG_SMP |
8a0be9ef | 2403 | |
46cb4b7c | 2404 | /* |
38022906 PZ |
2405 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2406 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2407 | */ |
38022906 | 2408 | void sched_exec(void) |
46cb4b7c | 2409 | { |
38022906 | 2410 | struct task_struct *p = current; |
1da177e4 | 2411 | unsigned long flags; |
0017d735 | 2412 | int dest_cpu; |
46cb4b7c | 2413 | |
8f42ced9 | 2414 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2415 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2416 | if (dest_cpu == smp_processor_id()) |
2417 | goto unlock; | |
38022906 | 2418 | |
8f42ced9 | 2419 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2420 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2421 | |
8f42ced9 PZ |
2422 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2423 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2424 | return; |
2425 | } | |
0017d735 | 2426 | unlock: |
8f42ced9 | 2427 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2428 | } |
dd41f596 | 2429 | |
1da177e4 LT |
2430 | #endif |
2431 | ||
1da177e4 | 2432 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2433 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2434 | |
2435 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2436 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2437 | |
c5f8d995 HS |
2438 | /* |
2439 | * Return accounted runtime for the task. | |
2440 | * In case the task is currently running, return the runtime plus current's | |
2441 | * pending runtime that have not been accounted yet. | |
2442 | */ | |
2443 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2444 | { | |
2445 | unsigned long flags; | |
2446 | struct rq *rq; | |
6e998916 | 2447 | u64 ns; |
c5f8d995 | 2448 | |
911b2898 PZ |
2449 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2450 | /* | |
2451 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2452 | * So we have a optimization chance when the task's delta_exec is 0. | |
2453 | * Reading ->on_cpu is racy, but this is ok. | |
2454 | * | |
2455 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2456 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2457 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2458 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2459 | * been accounted, so we're correct here as well. | |
911b2898 | 2460 | */ |
da0c1e65 | 2461 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2462 | return p->se.sum_exec_runtime; |
2463 | #endif | |
2464 | ||
c5f8d995 | 2465 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2466 | /* |
2467 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2468 | * project cycles that may never be accounted to this | |
2469 | * thread, breaking clock_gettime(). | |
2470 | */ | |
2471 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2472 | update_rq_clock(rq); | |
2473 | p->sched_class->update_curr(rq); | |
2474 | } | |
2475 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2476 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2477 | |
2478 | return ns; | |
2479 | } | |
48f24c4d | 2480 | |
7835b98b CL |
2481 | /* |
2482 | * This function gets called by the timer code, with HZ frequency. | |
2483 | * We call it with interrupts disabled. | |
7835b98b CL |
2484 | */ |
2485 | void scheduler_tick(void) | |
2486 | { | |
7835b98b CL |
2487 | int cpu = smp_processor_id(); |
2488 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2489 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2490 | |
2491 | sched_clock_tick(); | |
dd41f596 | 2492 | |
05fa785c | 2493 | raw_spin_lock(&rq->lock); |
3e51f33f | 2494 | update_rq_clock(rq); |
fa85ae24 | 2495 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2496 | update_cpu_load_active(rq); |
05fa785c | 2497 | raw_spin_unlock(&rq->lock); |
7835b98b | 2498 | |
e9d2b064 | 2499 | perf_event_task_tick(); |
e220d2dc | 2500 | |
e418e1c2 | 2501 | #ifdef CONFIG_SMP |
6eb57e0d | 2502 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2503 | trigger_load_balance(rq); |
e418e1c2 | 2504 | #endif |
265f22a9 | 2505 | rq_last_tick_reset(rq); |
1da177e4 LT |
2506 | } |
2507 | ||
265f22a9 FW |
2508 | #ifdef CONFIG_NO_HZ_FULL |
2509 | /** | |
2510 | * scheduler_tick_max_deferment | |
2511 | * | |
2512 | * Keep at least one tick per second when a single | |
2513 | * active task is running because the scheduler doesn't | |
2514 | * yet completely support full dynticks environment. | |
2515 | * | |
2516 | * This makes sure that uptime, CFS vruntime, load | |
2517 | * balancing, etc... continue to move forward, even | |
2518 | * with a very low granularity. | |
e69f6186 YB |
2519 | * |
2520 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2521 | */ |
2522 | u64 scheduler_tick_max_deferment(void) | |
2523 | { | |
2524 | struct rq *rq = this_rq(); | |
2525 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2526 | ||
2527 | next = rq->last_sched_tick + HZ; | |
2528 | ||
2529 | if (time_before_eq(next, now)) | |
2530 | return 0; | |
2531 | ||
8fe8ff09 | 2532 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2533 | } |
265f22a9 | 2534 | #endif |
1da177e4 | 2535 | |
132380a0 | 2536 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2537 | { |
2538 | if (in_lock_functions(addr)) { | |
2539 | addr = CALLER_ADDR2; | |
2540 | if (in_lock_functions(addr)) | |
2541 | addr = CALLER_ADDR3; | |
2542 | } | |
2543 | return addr; | |
2544 | } | |
1da177e4 | 2545 | |
7e49fcce SR |
2546 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2547 | defined(CONFIG_PREEMPT_TRACER)) | |
2548 | ||
edafe3a5 | 2549 | void preempt_count_add(int val) |
1da177e4 | 2550 | { |
6cd8a4bb | 2551 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2552 | /* |
2553 | * Underflow? | |
2554 | */ | |
9a11b49a IM |
2555 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2556 | return; | |
6cd8a4bb | 2557 | #endif |
bdb43806 | 2558 | __preempt_count_add(val); |
6cd8a4bb | 2559 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2560 | /* |
2561 | * Spinlock count overflowing soon? | |
2562 | */ | |
33859f7f MOS |
2563 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2564 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2565 | #endif |
8f47b187 TG |
2566 | if (preempt_count() == val) { |
2567 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2568 | #ifdef CONFIG_DEBUG_PREEMPT | |
2569 | current->preempt_disable_ip = ip; | |
2570 | #endif | |
2571 | trace_preempt_off(CALLER_ADDR0, ip); | |
2572 | } | |
1da177e4 | 2573 | } |
bdb43806 | 2574 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2575 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2576 | |
edafe3a5 | 2577 | void preempt_count_sub(int val) |
1da177e4 | 2578 | { |
6cd8a4bb | 2579 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2580 | /* |
2581 | * Underflow? | |
2582 | */ | |
01e3eb82 | 2583 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2584 | return; |
1da177e4 LT |
2585 | /* |
2586 | * Is the spinlock portion underflowing? | |
2587 | */ | |
9a11b49a IM |
2588 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2589 | !(preempt_count() & PREEMPT_MASK))) | |
2590 | return; | |
6cd8a4bb | 2591 | #endif |
9a11b49a | 2592 | |
6cd8a4bb SR |
2593 | if (preempt_count() == val) |
2594 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2595 | __preempt_count_sub(val); |
1da177e4 | 2596 | } |
bdb43806 | 2597 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2598 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2599 | |
2600 | #endif | |
2601 | ||
2602 | /* | |
dd41f596 | 2603 | * Print scheduling while atomic bug: |
1da177e4 | 2604 | */ |
dd41f596 | 2605 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2606 | { |
664dfa65 DJ |
2607 | if (oops_in_progress) |
2608 | return; | |
2609 | ||
3df0fc5b PZ |
2610 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2611 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2612 | |
dd41f596 | 2613 | debug_show_held_locks(prev); |
e21f5b15 | 2614 | print_modules(); |
dd41f596 IM |
2615 | if (irqs_disabled()) |
2616 | print_irqtrace_events(prev); | |
8f47b187 TG |
2617 | #ifdef CONFIG_DEBUG_PREEMPT |
2618 | if (in_atomic_preempt_off()) { | |
2619 | pr_err("Preemption disabled at:"); | |
2620 | print_ip_sym(current->preempt_disable_ip); | |
2621 | pr_cont("\n"); | |
2622 | } | |
2623 | #endif | |
6135fc1e | 2624 | dump_stack(); |
373d4d09 | 2625 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2626 | } |
1da177e4 | 2627 | |
dd41f596 IM |
2628 | /* |
2629 | * Various schedule()-time debugging checks and statistics: | |
2630 | */ | |
2631 | static inline void schedule_debug(struct task_struct *prev) | |
2632 | { | |
0d9e2632 AT |
2633 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
2634 | BUG_ON(unlikely(task_stack_end_corrupted(prev))); | |
2635 | #endif | |
1da177e4 | 2636 | /* |
41a2d6cf | 2637 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2638 | * schedule() atomically, we ignore that path. Otherwise whine |
2639 | * if we are scheduling when we should not. | |
1da177e4 | 2640 | */ |
192301e7 | 2641 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2642 | __schedule_bug(prev); |
b3fbab05 | 2643 | rcu_sleep_check(); |
dd41f596 | 2644 | |
1da177e4 LT |
2645 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2646 | ||
2d72376b | 2647 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2648 | } |
2649 | ||
2650 | /* | |
2651 | * Pick up the highest-prio task: | |
2652 | */ | |
2653 | static inline struct task_struct * | |
606dba2e | 2654 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2655 | { |
37e117c0 | 2656 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2657 | struct task_struct *p; |
1da177e4 LT |
2658 | |
2659 | /* | |
dd41f596 IM |
2660 | * Optimization: we know that if all tasks are in |
2661 | * the fair class we can call that function directly: | |
1da177e4 | 2662 | */ |
37e117c0 | 2663 | if (likely(prev->sched_class == class && |
38033c37 | 2664 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2665 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2666 | if (unlikely(p == RETRY_TASK)) |
2667 | goto again; | |
2668 | ||
2669 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2670 | if (unlikely(!p)) | |
2671 | p = idle_sched_class.pick_next_task(rq, prev); | |
2672 | ||
2673 | return p; | |
1da177e4 LT |
2674 | } |
2675 | ||
37e117c0 | 2676 | again: |
34f971f6 | 2677 | for_each_class(class) { |
606dba2e | 2678 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2679 | if (p) { |
2680 | if (unlikely(p == RETRY_TASK)) | |
2681 | goto again; | |
dd41f596 | 2682 | return p; |
37e117c0 | 2683 | } |
dd41f596 | 2684 | } |
34f971f6 PZ |
2685 | |
2686 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2687 | } |
1da177e4 | 2688 | |
dd41f596 | 2689 | /* |
c259e01a | 2690 | * __schedule() is the main scheduler function. |
edde96ea PE |
2691 | * |
2692 | * The main means of driving the scheduler and thus entering this function are: | |
2693 | * | |
2694 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2695 | * | |
2696 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2697 | * paths. For example, see arch/x86/entry_64.S. | |
2698 | * | |
2699 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2700 | * interrupt handler scheduler_tick(). | |
2701 | * | |
2702 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2703 | * task to the run-queue and that's it. | |
2704 | * | |
2705 | * Now, if the new task added to the run-queue preempts the current | |
2706 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2707 | * called on the nearest possible occasion: | |
2708 | * | |
2709 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2710 | * | |
2711 | * - in syscall or exception context, at the next outmost | |
2712 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2713 | * spin_unlock()!) | |
2714 | * | |
2715 | * - in IRQ context, return from interrupt-handler to | |
2716 | * preemptible context | |
2717 | * | |
2718 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2719 | * then at the next: | |
2720 | * | |
2721 | * - cond_resched() call | |
2722 | * - explicit schedule() call | |
2723 | * - return from syscall or exception to user-space | |
2724 | * - return from interrupt-handler to user-space | |
bfd9b2b5 FW |
2725 | * |
2726 | * WARNING: all callers must re-check need_resched() afterward and reschedule | |
2727 | * accordingly in case an event triggered the need for rescheduling (such as | |
2728 | * an interrupt waking up a task) while preemption was disabled in __schedule(). | |
dd41f596 | 2729 | */ |
c259e01a | 2730 | static void __sched __schedule(void) |
dd41f596 IM |
2731 | { |
2732 | struct task_struct *prev, *next; | |
67ca7bde | 2733 | unsigned long *switch_count; |
dd41f596 | 2734 | struct rq *rq; |
31656519 | 2735 | int cpu; |
dd41f596 | 2736 | |
ff743345 | 2737 | preempt_disable(); |
dd41f596 IM |
2738 | cpu = smp_processor_id(); |
2739 | rq = cpu_rq(cpu); | |
38200cf2 | 2740 | rcu_note_context_switch(); |
dd41f596 | 2741 | prev = rq->curr; |
dd41f596 | 2742 | |
dd41f596 | 2743 | schedule_debug(prev); |
1da177e4 | 2744 | |
31656519 | 2745 | if (sched_feat(HRTICK)) |
f333fdc9 | 2746 | hrtick_clear(rq); |
8f4d37ec | 2747 | |
e0acd0a6 ON |
2748 | /* |
2749 | * Make sure that signal_pending_state()->signal_pending() below | |
2750 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2751 | * done by the caller to avoid the race with signal_wake_up(). | |
2752 | */ | |
2753 | smp_mb__before_spinlock(); | |
05fa785c | 2754 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2755 | |
9edfbfed PZ |
2756 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
2757 | ||
246d86b5 | 2758 | switch_count = &prev->nivcsw; |
1da177e4 | 2759 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2760 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2761 | prev->state = TASK_RUNNING; |
21aa9af0 | 2762 | } else { |
2acca55e PZ |
2763 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2764 | prev->on_rq = 0; | |
2765 | ||
21aa9af0 | 2766 | /* |
2acca55e PZ |
2767 | * If a worker went to sleep, notify and ask workqueue |
2768 | * whether it wants to wake up a task to maintain | |
2769 | * concurrency. | |
21aa9af0 TH |
2770 | */ |
2771 | if (prev->flags & PF_WQ_WORKER) { | |
2772 | struct task_struct *to_wakeup; | |
2773 | ||
2774 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2775 | if (to_wakeup) | |
2776 | try_to_wake_up_local(to_wakeup); | |
2777 | } | |
21aa9af0 | 2778 | } |
dd41f596 | 2779 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2780 | } |
2781 | ||
9edfbfed | 2782 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
2783 | update_rq_clock(rq); |
2784 | ||
2785 | next = pick_next_task(rq, prev); | |
f26f9aff | 2786 | clear_tsk_need_resched(prev); |
f27dde8d | 2787 | clear_preempt_need_resched(); |
9edfbfed | 2788 | rq->clock_skip_update = 0; |
1da177e4 | 2789 | |
1da177e4 | 2790 | if (likely(prev != next)) { |
1da177e4 LT |
2791 | rq->nr_switches++; |
2792 | rq->curr = next; | |
2793 | ++*switch_count; | |
2794 | ||
dfa50b60 ON |
2795 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
2796 | cpu = cpu_of(rq); | |
1da177e4 | 2797 | } else |
05fa785c | 2798 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2799 | |
3f029d3c | 2800 | post_schedule(rq); |
1da177e4 | 2801 | |
ba74c144 | 2802 | sched_preempt_enable_no_resched(); |
1da177e4 | 2803 | } |
c259e01a | 2804 | |
9c40cef2 TG |
2805 | static inline void sched_submit_work(struct task_struct *tsk) |
2806 | { | |
3c7d5184 | 2807 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2808 | return; |
2809 | /* | |
2810 | * If we are going to sleep and we have plugged IO queued, | |
2811 | * make sure to submit it to avoid deadlocks. | |
2812 | */ | |
2813 | if (blk_needs_flush_plug(tsk)) | |
2814 | blk_schedule_flush_plug(tsk); | |
2815 | } | |
2816 | ||
722a9f92 | 2817 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2818 | { |
9c40cef2 TG |
2819 | struct task_struct *tsk = current; |
2820 | ||
2821 | sched_submit_work(tsk); | |
bfd9b2b5 FW |
2822 | do { |
2823 | __schedule(); | |
2824 | } while (need_resched()); | |
c259e01a | 2825 | } |
1da177e4 LT |
2826 | EXPORT_SYMBOL(schedule); |
2827 | ||
91d1aa43 | 2828 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2829 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2830 | { |
2831 | /* | |
2832 | * If we come here after a random call to set_need_resched(), | |
2833 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2834 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2835 | * we find a better solution. | |
7cc78f8f AL |
2836 | * |
2837 | * NB: There are buggy callers of this function. Ideally we | |
2838 | * should warn if prev_state != IN_USER, but that will trigger | |
2839 | * too frequently to make sense yet. | |
20ab65e3 | 2840 | */ |
7cc78f8f | 2841 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 2842 | schedule(); |
7cc78f8f | 2843 | exception_exit(prev_state); |
20ab65e3 FW |
2844 | } |
2845 | #endif | |
2846 | ||
c5491ea7 TG |
2847 | /** |
2848 | * schedule_preempt_disabled - called with preemption disabled | |
2849 | * | |
2850 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2851 | */ | |
2852 | void __sched schedule_preempt_disabled(void) | |
2853 | { | |
ba74c144 | 2854 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2855 | schedule(); |
2856 | preempt_disable(); | |
2857 | } | |
2858 | ||
06b1f808 | 2859 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
2860 | { |
2861 | do { | |
2862 | __preempt_count_add(PREEMPT_ACTIVE); | |
2863 | __schedule(); | |
2864 | __preempt_count_sub(PREEMPT_ACTIVE); | |
2865 | ||
2866 | /* | |
2867 | * Check again in case we missed a preemption opportunity | |
2868 | * between schedule and now. | |
2869 | */ | |
2870 | barrier(); | |
2871 | } while (need_resched()); | |
2872 | } | |
2873 | ||
1da177e4 LT |
2874 | #ifdef CONFIG_PREEMPT |
2875 | /* | |
2ed6e34f | 2876 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2877 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2878 | * occur there and call schedule directly. |
2879 | */ | |
722a9f92 | 2880 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2881 | { |
1da177e4 LT |
2882 | /* |
2883 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2884 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2885 | */ |
fbb00b56 | 2886 | if (likely(!preemptible())) |
1da177e4 LT |
2887 | return; |
2888 | ||
a18b5d01 | 2889 | preempt_schedule_common(); |
1da177e4 | 2890 | } |
376e2424 | 2891 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2892 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 ON |
2893 | |
2894 | #ifdef CONFIG_CONTEXT_TRACKING | |
2895 | /** | |
2896 | * preempt_schedule_context - preempt_schedule called by tracing | |
2897 | * | |
2898 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
2899 | * recursion and tracing preempt enabling caused by the tracing | |
2900 | * infrastructure itself. But as tracing can happen in areas coming | |
2901 | * from userspace or just about to enter userspace, a preempt enable | |
2902 | * can occur before user_exit() is called. This will cause the scheduler | |
2903 | * to be called when the system is still in usermode. | |
2904 | * | |
2905 | * To prevent this, the preempt_enable_notrace will use this function | |
2906 | * instead of preempt_schedule() to exit user context if needed before | |
2907 | * calling the scheduler. | |
2908 | */ | |
2909 | asmlinkage __visible void __sched notrace preempt_schedule_context(void) | |
2910 | { | |
2911 | enum ctx_state prev_ctx; | |
2912 | ||
2913 | if (likely(!preemptible())) | |
2914 | return; | |
2915 | ||
2916 | do { | |
2917 | __preempt_count_add(PREEMPT_ACTIVE); | |
2918 | /* | |
2919 | * Needs preempt disabled in case user_exit() is traced | |
2920 | * and the tracer calls preempt_enable_notrace() causing | |
2921 | * an infinite recursion. | |
2922 | */ | |
2923 | prev_ctx = exception_enter(); | |
2924 | __schedule(); | |
2925 | exception_exit(prev_ctx); | |
2926 | ||
2927 | __preempt_count_sub(PREEMPT_ACTIVE); | |
2928 | barrier(); | |
2929 | } while (need_resched()); | |
2930 | } | |
2931 | EXPORT_SYMBOL_GPL(preempt_schedule_context); | |
2932 | #endif /* CONFIG_CONTEXT_TRACKING */ | |
2933 | ||
32e475d7 | 2934 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2935 | |
2936 | /* | |
2ed6e34f | 2937 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2938 | * off of irq context. |
2939 | * Note, that this is called and return with irqs disabled. This will | |
2940 | * protect us against recursive calling from irq. | |
2941 | */ | |
722a9f92 | 2942 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2943 | { |
b22366cd | 2944 | enum ctx_state prev_state; |
6478d880 | 2945 | |
2ed6e34f | 2946 | /* Catch callers which need to be fixed */ |
f27dde8d | 2947 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2948 | |
b22366cd FW |
2949 | prev_state = exception_enter(); |
2950 | ||
3a5c359a | 2951 | do { |
bdb43806 | 2952 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2953 | local_irq_enable(); |
c259e01a | 2954 | __schedule(); |
3a5c359a | 2955 | local_irq_disable(); |
bdb43806 | 2956 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2957 | |
3a5c359a AK |
2958 | /* |
2959 | * Check again in case we missed a preemption opportunity | |
2960 | * between schedule and now. | |
2961 | */ | |
2962 | barrier(); | |
5ed0cec0 | 2963 | } while (need_resched()); |
b22366cd FW |
2964 | |
2965 | exception_exit(prev_state); | |
1da177e4 LT |
2966 | } |
2967 | ||
63859d4f | 2968 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2969 | void *key) |
1da177e4 | 2970 | { |
63859d4f | 2971 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2972 | } |
1da177e4 LT |
2973 | EXPORT_SYMBOL(default_wake_function); |
2974 | ||
b29739f9 IM |
2975 | #ifdef CONFIG_RT_MUTEXES |
2976 | ||
2977 | /* | |
2978 | * rt_mutex_setprio - set the current priority of a task | |
2979 | * @p: task | |
2980 | * @prio: prio value (kernel-internal form) | |
2981 | * | |
2982 | * This function changes the 'effective' priority of a task. It does | |
2983 | * not touch ->normal_prio like __setscheduler(). | |
2984 | * | |
c365c292 TG |
2985 | * Used by the rt_mutex code to implement priority inheritance |
2986 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2987 | */ |
36c8b586 | 2988 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2989 | { |
da0c1e65 | 2990 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 2991 | struct rq *rq; |
83ab0aa0 | 2992 | const struct sched_class *prev_class; |
b29739f9 | 2993 | |
aab03e05 | 2994 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2995 | |
0122ec5b | 2996 | rq = __task_rq_lock(p); |
b29739f9 | 2997 | |
1c4dd99b TG |
2998 | /* |
2999 | * Idle task boosting is a nono in general. There is one | |
3000 | * exception, when PREEMPT_RT and NOHZ is active: | |
3001 | * | |
3002 | * The idle task calls get_next_timer_interrupt() and holds | |
3003 | * the timer wheel base->lock on the CPU and another CPU wants | |
3004 | * to access the timer (probably to cancel it). We can safely | |
3005 | * ignore the boosting request, as the idle CPU runs this code | |
3006 | * with interrupts disabled and will complete the lock | |
3007 | * protected section without being interrupted. So there is no | |
3008 | * real need to boost. | |
3009 | */ | |
3010 | if (unlikely(p == rq->idle)) { | |
3011 | WARN_ON(p != rq->curr); | |
3012 | WARN_ON(p->pi_blocked_on); | |
3013 | goto out_unlock; | |
3014 | } | |
3015 | ||
a8027073 | 3016 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3017 | oldprio = p->prio; |
83ab0aa0 | 3018 | prev_class = p->sched_class; |
da0c1e65 | 3019 | queued = task_on_rq_queued(p); |
051a1d1a | 3020 | running = task_current(rq, p); |
da0c1e65 | 3021 | if (queued) |
69be72c1 | 3022 | dequeue_task(rq, p, 0); |
0e1f3483 | 3023 | if (running) |
f3cd1c4e | 3024 | put_prev_task(rq, p); |
dd41f596 | 3025 | |
2d3d891d DF |
3026 | /* |
3027 | * Boosting condition are: | |
3028 | * 1. -rt task is running and holds mutex A | |
3029 | * --> -dl task blocks on mutex A | |
3030 | * | |
3031 | * 2. -dl task is running and holds mutex A | |
3032 | * --> -dl task blocks on mutex A and could preempt the | |
3033 | * running task | |
3034 | */ | |
3035 | if (dl_prio(prio)) { | |
466af29b ON |
3036 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3037 | if (!dl_prio(p->normal_prio) || | |
3038 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3039 | p->dl.dl_boosted = 1; |
3040 | p->dl.dl_throttled = 0; | |
3041 | enqueue_flag = ENQUEUE_REPLENISH; | |
3042 | } else | |
3043 | p->dl.dl_boosted = 0; | |
aab03e05 | 3044 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3045 | } else if (rt_prio(prio)) { |
3046 | if (dl_prio(oldprio)) | |
3047 | p->dl.dl_boosted = 0; | |
3048 | if (oldprio < prio) | |
3049 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3050 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3051 | } else { |
3052 | if (dl_prio(oldprio)) | |
3053 | p->dl.dl_boosted = 0; | |
746db944 BS |
3054 | if (rt_prio(oldprio)) |
3055 | p->rt.timeout = 0; | |
dd41f596 | 3056 | p->sched_class = &fair_sched_class; |
2d3d891d | 3057 | } |
dd41f596 | 3058 | |
b29739f9 IM |
3059 | p->prio = prio; |
3060 | ||
0e1f3483 HS |
3061 | if (running) |
3062 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3063 | if (queued) |
2d3d891d | 3064 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3065 | |
da7a735e | 3066 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3067 | out_unlock: |
0122ec5b | 3068 | __task_rq_unlock(rq); |
b29739f9 | 3069 | } |
b29739f9 | 3070 | #endif |
d50dde5a | 3071 | |
36c8b586 | 3072 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3073 | { |
da0c1e65 | 3074 | int old_prio, delta, queued; |
1da177e4 | 3075 | unsigned long flags; |
70b97a7f | 3076 | struct rq *rq; |
1da177e4 | 3077 | |
75e45d51 | 3078 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3079 | return; |
3080 | /* | |
3081 | * We have to be careful, if called from sys_setpriority(), | |
3082 | * the task might be in the middle of scheduling on another CPU. | |
3083 | */ | |
3084 | rq = task_rq_lock(p, &flags); | |
3085 | /* | |
3086 | * The RT priorities are set via sched_setscheduler(), but we still | |
3087 | * allow the 'normal' nice value to be set - but as expected | |
3088 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3089 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3090 | */ |
aab03e05 | 3091 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3092 | p->static_prio = NICE_TO_PRIO(nice); |
3093 | goto out_unlock; | |
3094 | } | |
da0c1e65 KT |
3095 | queued = task_on_rq_queued(p); |
3096 | if (queued) | |
69be72c1 | 3097 | dequeue_task(rq, p, 0); |
1da177e4 | 3098 | |
1da177e4 | 3099 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3100 | set_load_weight(p); |
b29739f9 IM |
3101 | old_prio = p->prio; |
3102 | p->prio = effective_prio(p); | |
3103 | delta = p->prio - old_prio; | |
1da177e4 | 3104 | |
da0c1e65 | 3105 | if (queued) { |
371fd7e7 | 3106 | enqueue_task(rq, p, 0); |
1da177e4 | 3107 | /* |
d5f9f942 AM |
3108 | * If the task increased its priority or is running and |
3109 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3110 | */ |
d5f9f942 | 3111 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3112 | resched_curr(rq); |
1da177e4 LT |
3113 | } |
3114 | out_unlock: | |
0122ec5b | 3115 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3116 | } |
1da177e4 LT |
3117 | EXPORT_SYMBOL(set_user_nice); |
3118 | ||
e43379f1 MM |
3119 | /* |
3120 | * can_nice - check if a task can reduce its nice value | |
3121 | * @p: task | |
3122 | * @nice: nice value | |
3123 | */ | |
36c8b586 | 3124 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3125 | { |
024f4747 | 3126 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3127 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3128 | |
78d7d407 | 3129 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3130 | capable(CAP_SYS_NICE)); |
3131 | } | |
3132 | ||
1da177e4 LT |
3133 | #ifdef __ARCH_WANT_SYS_NICE |
3134 | ||
3135 | /* | |
3136 | * sys_nice - change the priority of the current process. | |
3137 | * @increment: priority increment | |
3138 | * | |
3139 | * sys_setpriority is a more generic, but much slower function that | |
3140 | * does similar things. | |
3141 | */ | |
5add95d4 | 3142 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3143 | { |
48f24c4d | 3144 | long nice, retval; |
1da177e4 LT |
3145 | |
3146 | /* | |
3147 | * Setpriority might change our priority at the same moment. | |
3148 | * We don't have to worry. Conceptually one call occurs first | |
3149 | * and we have a single winner. | |
3150 | */ | |
a9467fa3 | 3151 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3152 | nice = task_nice(current) + increment; |
1da177e4 | 3153 | |
a9467fa3 | 3154 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3155 | if (increment < 0 && !can_nice(current, nice)) |
3156 | return -EPERM; | |
3157 | ||
1da177e4 LT |
3158 | retval = security_task_setnice(current, nice); |
3159 | if (retval) | |
3160 | return retval; | |
3161 | ||
3162 | set_user_nice(current, nice); | |
3163 | return 0; | |
3164 | } | |
3165 | ||
3166 | #endif | |
3167 | ||
3168 | /** | |
3169 | * task_prio - return the priority value of a given task. | |
3170 | * @p: the task in question. | |
3171 | * | |
e69f6186 | 3172 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3173 | * RT tasks are offset by -200. Normal tasks are centered |
3174 | * around 0, value goes from -16 to +15. | |
3175 | */ | |
36c8b586 | 3176 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3177 | { |
3178 | return p->prio - MAX_RT_PRIO; | |
3179 | } | |
3180 | ||
1da177e4 LT |
3181 | /** |
3182 | * idle_cpu - is a given cpu idle currently? | |
3183 | * @cpu: the processor in question. | |
e69f6186 YB |
3184 | * |
3185 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3186 | */ |
3187 | int idle_cpu(int cpu) | |
3188 | { | |
908a3283 TG |
3189 | struct rq *rq = cpu_rq(cpu); |
3190 | ||
3191 | if (rq->curr != rq->idle) | |
3192 | return 0; | |
3193 | ||
3194 | if (rq->nr_running) | |
3195 | return 0; | |
3196 | ||
3197 | #ifdef CONFIG_SMP | |
3198 | if (!llist_empty(&rq->wake_list)) | |
3199 | return 0; | |
3200 | #endif | |
3201 | ||
3202 | return 1; | |
1da177e4 LT |
3203 | } |
3204 | ||
1da177e4 LT |
3205 | /** |
3206 | * idle_task - return the idle task for a given cpu. | |
3207 | * @cpu: the processor in question. | |
e69f6186 YB |
3208 | * |
3209 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3210 | */ |
36c8b586 | 3211 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3212 | { |
3213 | return cpu_rq(cpu)->idle; | |
3214 | } | |
3215 | ||
3216 | /** | |
3217 | * find_process_by_pid - find a process with a matching PID value. | |
3218 | * @pid: the pid in question. | |
e69f6186 YB |
3219 | * |
3220 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3221 | */ |
a9957449 | 3222 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3223 | { |
228ebcbe | 3224 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3225 | } |
3226 | ||
aab03e05 DF |
3227 | /* |
3228 | * This function initializes the sched_dl_entity of a newly becoming | |
3229 | * SCHED_DEADLINE task. | |
3230 | * | |
3231 | * Only the static values are considered here, the actual runtime and the | |
3232 | * absolute deadline will be properly calculated when the task is enqueued | |
3233 | * for the first time with its new policy. | |
3234 | */ | |
3235 | static void | |
3236 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3237 | { | |
3238 | struct sched_dl_entity *dl_se = &p->dl; | |
3239 | ||
aab03e05 DF |
3240 | dl_se->dl_runtime = attr->sched_runtime; |
3241 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3242 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3243 | dl_se->flags = attr->sched_flags; |
332ac17e | 3244 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3245 | |
3246 | /* | |
3247 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3248 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3249 | * | |
3250 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3251 | * point. This would include retaining the task_struct until that time | |
3252 | * and change dl_overflow() to not immediately decrement the current | |
3253 | * amount. | |
3254 | * | |
3255 | * Instead we retain the current runtime/deadline and let the new | |
3256 | * parameters take effect after the current reservation period lapses. | |
3257 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3258 | * before the current scheduling deadline. | |
3259 | * | |
3260 | * We can still have temporary overloads because we do not delay the | |
3261 | * change in bandwidth until that time; so admission control is | |
3262 | * not on the safe side. It does however guarantee tasks will never | |
3263 | * consume more than promised. | |
3264 | */ | |
aab03e05 DF |
3265 | } |
3266 | ||
c13db6b1 SR |
3267 | /* |
3268 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3269 | * it calls know not to change it. | |
3270 | */ | |
3271 | #define SETPARAM_POLICY -1 | |
3272 | ||
c365c292 TG |
3273 | static void __setscheduler_params(struct task_struct *p, |
3274 | const struct sched_attr *attr) | |
1da177e4 | 3275 | { |
d50dde5a DF |
3276 | int policy = attr->sched_policy; |
3277 | ||
c13db6b1 | 3278 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3279 | policy = p->policy; |
3280 | ||
1da177e4 | 3281 | p->policy = policy; |
d50dde5a | 3282 | |
aab03e05 DF |
3283 | if (dl_policy(policy)) |
3284 | __setparam_dl(p, attr); | |
39fd8fd2 | 3285 | else if (fair_policy(policy)) |
d50dde5a DF |
3286 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3287 | ||
39fd8fd2 PZ |
3288 | /* |
3289 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3290 | * !rt_policy. Always setting this ensures that things like | |
3291 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3292 | */ | |
3293 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3294 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3295 | set_load_weight(p); |
3296 | } | |
39fd8fd2 | 3297 | |
c365c292 TG |
3298 | /* Actually do priority change: must hold pi & rq lock. */ |
3299 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3300 | const struct sched_attr *attr) | |
3301 | { | |
3302 | __setscheduler_params(p, attr); | |
d50dde5a | 3303 | |
383afd09 SR |
3304 | /* |
3305 | * If we get here, there was no pi waiters boosting the | |
3306 | * task. It is safe to use the normal prio. | |
3307 | */ | |
3308 | p->prio = normal_prio(p); | |
3309 | ||
aab03e05 DF |
3310 | if (dl_prio(p->prio)) |
3311 | p->sched_class = &dl_sched_class; | |
3312 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3313 | p->sched_class = &rt_sched_class; |
3314 | else | |
3315 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3316 | } |
aab03e05 DF |
3317 | |
3318 | static void | |
3319 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3320 | { | |
3321 | struct sched_dl_entity *dl_se = &p->dl; | |
3322 | ||
3323 | attr->sched_priority = p->rt_priority; | |
3324 | attr->sched_runtime = dl_se->dl_runtime; | |
3325 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3326 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3327 | attr->sched_flags = dl_se->flags; |
3328 | } | |
3329 | ||
3330 | /* | |
3331 | * This function validates the new parameters of a -deadline task. | |
3332 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3333 | * than the runtime, as well as the period of being zero or |
332ac17e | 3334 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3335 | * user parameters are above the internal resolution of 1us (we |
3336 | * check sched_runtime only since it is always the smaller one) and | |
3337 | * below 2^63 ns (we have to check both sched_deadline and | |
3338 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3339 | */ |
3340 | static bool | |
3341 | __checkparam_dl(const struct sched_attr *attr) | |
3342 | { | |
b0827819 JL |
3343 | /* deadline != 0 */ |
3344 | if (attr->sched_deadline == 0) | |
3345 | return false; | |
3346 | ||
3347 | /* | |
3348 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3349 | * that big. | |
3350 | */ | |
3351 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3352 | return false; | |
3353 | ||
3354 | /* | |
3355 | * Since we use the MSB for wrap-around and sign issues, make | |
3356 | * sure it's not set (mind that period can be equal to zero). | |
3357 | */ | |
3358 | if (attr->sched_deadline & (1ULL << 63) || | |
3359 | attr->sched_period & (1ULL << 63)) | |
3360 | return false; | |
3361 | ||
3362 | /* runtime <= deadline <= period (if period != 0) */ | |
3363 | if ((attr->sched_period != 0 && | |
3364 | attr->sched_period < attr->sched_deadline) || | |
3365 | attr->sched_deadline < attr->sched_runtime) | |
3366 | return false; | |
3367 | ||
3368 | return true; | |
aab03e05 DF |
3369 | } |
3370 | ||
c69e8d9c DH |
3371 | /* |
3372 | * check the target process has a UID that matches the current process's | |
3373 | */ | |
3374 | static bool check_same_owner(struct task_struct *p) | |
3375 | { | |
3376 | const struct cred *cred = current_cred(), *pcred; | |
3377 | bool match; | |
3378 | ||
3379 | rcu_read_lock(); | |
3380 | pcred = __task_cred(p); | |
9c806aa0 EB |
3381 | match = (uid_eq(cred->euid, pcred->euid) || |
3382 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3383 | rcu_read_unlock(); |
3384 | return match; | |
3385 | } | |
3386 | ||
75381608 WL |
3387 | static bool dl_param_changed(struct task_struct *p, |
3388 | const struct sched_attr *attr) | |
3389 | { | |
3390 | struct sched_dl_entity *dl_se = &p->dl; | |
3391 | ||
3392 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3393 | dl_se->dl_deadline != attr->sched_deadline || | |
3394 | dl_se->dl_period != attr->sched_period || | |
3395 | dl_se->flags != attr->sched_flags) | |
3396 | return true; | |
3397 | ||
3398 | return false; | |
3399 | } | |
3400 | ||
d50dde5a DF |
3401 | static int __sched_setscheduler(struct task_struct *p, |
3402 | const struct sched_attr *attr, | |
3403 | bool user) | |
1da177e4 | 3404 | { |
383afd09 SR |
3405 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3406 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3407 | int retval, oldprio, oldpolicy = -1, queued, running; |
d50dde5a | 3408 | int policy = attr->sched_policy; |
1da177e4 | 3409 | unsigned long flags; |
83ab0aa0 | 3410 | const struct sched_class *prev_class; |
70b97a7f | 3411 | struct rq *rq; |
ca94c442 | 3412 | int reset_on_fork; |
1da177e4 | 3413 | |
66e5393a SR |
3414 | /* may grab non-irq protected spin_locks */ |
3415 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3416 | recheck: |
3417 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3418 | if (policy < 0) { |
3419 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3420 | policy = oldpolicy = p->policy; |
ca94c442 | 3421 | } else { |
7479f3c9 | 3422 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3423 | |
aab03e05 DF |
3424 | if (policy != SCHED_DEADLINE && |
3425 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3426 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3427 | policy != SCHED_IDLE) | |
3428 | return -EINVAL; | |
3429 | } | |
3430 | ||
7479f3c9 PZ |
3431 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3432 | return -EINVAL; | |
3433 | ||
1da177e4 LT |
3434 | /* |
3435 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3436 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3437 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3438 | */ |
0bb040a4 | 3439 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3440 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3441 | return -EINVAL; |
aab03e05 DF |
3442 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3443 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3444 | return -EINVAL; |
3445 | ||
37e4ab3f OC |
3446 | /* |
3447 | * Allow unprivileged RT tasks to decrease priority: | |
3448 | */ | |
961ccddd | 3449 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3450 | if (fair_policy(policy)) { |
d0ea0268 | 3451 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3452 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3453 | return -EPERM; |
3454 | } | |
3455 | ||
e05606d3 | 3456 | if (rt_policy(policy)) { |
a44702e8 ON |
3457 | unsigned long rlim_rtprio = |
3458 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3459 | |
3460 | /* can't set/change the rt policy */ | |
3461 | if (policy != p->policy && !rlim_rtprio) | |
3462 | return -EPERM; | |
3463 | ||
3464 | /* can't increase priority */ | |
d50dde5a DF |
3465 | if (attr->sched_priority > p->rt_priority && |
3466 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3467 | return -EPERM; |
3468 | } | |
c02aa73b | 3469 | |
d44753b8 JL |
3470 | /* |
3471 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3472 | * (safest behavior); in the future we would like to allow | |
3473 | * unprivileged DL tasks to increase their relative deadline | |
3474 | * or reduce their runtime (both ways reducing utilization) | |
3475 | */ | |
3476 | if (dl_policy(policy)) | |
3477 | return -EPERM; | |
3478 | ||
dd41f596 | 3479 | /* |
c02aa73b DH |
3480 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3481 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3482 | */ |
c02aa73b | 3483 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3484 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3485 | return -EPERM; |
3486 | } | |
5fe1d75f | 3487 | |
37e4ab3f | 3488 | /* can't change other user's priorities */ |
c69e8d9c | 3489 | if (!check_same_owner(p)) |
37e4ab3f | 3490 | return -EPERM; |
ca94c442 LP |
3491 | |
3492 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3493 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3494 | return -EPERM; | |
37e4ab3f | 3495 | } |
1da177e4 | 3496 | |
725aad24 | 3497 | if (user) { |
b0ae1981 | 3498 | retval = security_task_setscheduler(p); |
725aad24 JF |
3499 | if (retval) |
3500 | return retval; | |
3501 | } | |
3502 | ||
b29739f9 IM |
3503 | /* |
3504 | * make sure no PI-waiters arrive (or leave) while we are | |
3505 | * changing the priority of the task: | |
0122ec5b | 3506 | * |
25985edc | 3507 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3508 | * runqueue lock must be held. |
3509 | */ | |
0122ec5b | 3510 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3511 | |
34f971f6 PZ |
3512 | /* |
3513 | * Changing the policy of the stop threads its a very bad idea | |
3514 | */ | |
3515 | if (p == rq->stop) { | |
0122ec5b | 3516 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3517 | return -EINVAL; |
3518 | } | |
3519 | ||
a51e9198 | 3520 | /* |
d6b1e911 TG |
3521 | * If not changing anything there's no need to proceed further, |
3522 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3523 | */ |
d50dde5a | 3524 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3525 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3526 | goto change; |
3527 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3528 | goto change; | |
75381608 | 3529 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 3530 | goto change; |
d50dde5a | 3531 | |
d6b1e911 | 3532 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3533 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3534 | return 0; |
3535 | } | |
d50dde5a | 3536 | change: |
a51e9198 | 3537 | |
dc61b1d6 | 3538 | if (user) { |
332ac17e | 3539 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3540 | /* |
3541 | * Do not allow realtime tasks into groups that have no runtime | |
3542 | * assigned. | |
3543 | */ | |
3544 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3545 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3546 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3547 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3548 | return -EPERM; |
3549 | } | |
dc61b1d6 | 3550 | #endif |
332ac17e DF |
3551 | #ifdef CONFIG_SMP |
3552 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3553 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3554 | |
3555 | /* | |
3556 | * Don't allow tasks with an affinity mask smaller than | |
3557 | * the entire root_domain to become SCHED_DEADLINE. We | |
3558 | * will also fail if there's no bandwidth available. | |
3559 | */ | |
e4099a5e PZ |
3560 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3561 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3562 | task_rq_unlock(rq, p, &flags); |
3563 | return -EPERM; | |
3564 | } | |
3565 | } | |
3566 | #endif | |
3567 | } | |
dc61b1d6 | 3568 | |
1da177e4 LT |
3569 | /* recheck policy now with rq lock held */ |
3570 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3571 | policy = oldpolicy = -1; | |
0122ec5b | 3572 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3573 | goto recheck; |
3574 | } | |
332ac17e DF |
3575 | |
3576 | /* | |
3577 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3578 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3579 | * is available. | |
3580 | */ | |
e4099a5e | 3581 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3582 | task_rq_unlock(rq, p, &flags); |
3583 | return -EBUSY; | |
3584 | } | |
3585 | ||
c365c292 TG |
3586 | p->sched_reset_on_fork = reset_on_fork; |
3587 | oldprio = p->prio; | |
3588 | ||
3589 | /* | |
3590 | * Special case for priority boosted tasks. | |
3591 | * | |
3592 | * If the new priority is lower or equal (user space view) | |
3593 | * than the current (boosted) priority, we just store the new | |
3594 | * normal parameters and do not touch the scheduler class and | |
3595 | * the runqueue. This will be done when the task deboost | |
3596 | * itself. | |
3597 | */ | |
3598 | if (rt_mutex_check_prio(p, newprio)) { | |
3599 | __setscheduler_params(p, attr); | |
3600 | task_rq_unlock(rq, p, &flags); | |
3601 | return 0; | |
3602 | } | |
3603 | ||
da0c1e65 | 3604 | queued = task_on_rq_queued(p); |
051a1d1a | 3605 | running = task_current(rq, p); |
da0c1e65 | 3606 | if (queued) |
4ca9b72b | 3607 | dequeue_task(rq, p, 0); |
0e1f3483 | 3608 | if (running) |
f3cd1c4e | 3609 | put_prev_task(rq, p); |
f6b53205 | 3610 | |
83ab0aa0 | 3611 | prev_class = p->sched_class; |
d50dde5a | 3612 | __setscheduler(rq, p, attr); |
f6b53205 | 3613 | |
0e1f3483 HS |
3614 | if (running) |
3615 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3616 | if (queued) { |
81a44c54 TG |
3617 | /* |
3618 | * We enqueue to tail when the priority of a task is | |
3619 | * increased (user space view). | |
3620 | */ | |
3621 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3622 | } | |
cb469845 | 3623 | |
da7a735e | 3624 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3625 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3626 | |
95e02ca9 TG |
3627 | rt_mutex_adjust_pi(p); |
3628 | ||
1da177e4 LT |
3629 | return 0; |
3630 | } | |
961ccddd | 3631 | |
7479f3c9 PZ |
3632 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3633 | const struct sched_param *param, bool check) | |
3634 | { | |
3635 | struct sched_attr attr = { | |
3636 | .sched_policy = policy, | |
3637 | .sched_priority = param->sched_priority, | |
3638 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3639 | }; | |
3640 | ||
c13db6b1 SR |
3641 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3642 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3643 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3644 | policy &= ~SCHED_RESET_ON_FORK; | |
3645 | attr.sched_policy = policy; | |
3646 | } | |
3647 | ||
3648 | return __sched_setscheduler(p, &attr, check); | |
3649 | } | |
961ccddd RR |
3650 | /** |
3651 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3652 | * @p: the task in question. | |
3653 | * @policy: new policy. | |
3654 | * @param: structure containing the new RT priority. | |
3655 | * | |
e69f6186 YB |
3656 | * Return: 0 on success. An error code otherwise. |
3657 | * | |
961ccddd RR |
3658 | * NOTE that the task may be already dead. |
3659 | */ | |
3660 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3661 | const struct sched_param *param) |
961ccddd | 3662 | { |
7479f3c9 | 3663 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3664 | } |
1da177e4 LT |
3665 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3666 | ||
d50dde5a DF |
3667 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3668 | { | |
3669 | return __sched_setscheduler(p, attr, true); | |
3670 | } | |
3671 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3672 | ||
961ccddd RR |
3673 | /** |
3674 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3675 | * @p: the task in question. | |
3676 | * @policy: new policy. | |
3677 | * @param: structure containing the new RT priority. | |
3678 | * | |
3679 | * Just like sched_setscheduler, only don't bother checking if the | |
3680 | * current context has permission. For example, this is needed in | |
3681 | * stop_machine(): we create temporary high priority worker threads, | |
3682 | * but our caller might not have that capability. | |
e69f6186 YB |
3683 | * |
3684 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3685 | */ |
3686 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3687 | const struct sched_param *param) |
961ccddd | 3688 | { |
7479f3c9 | 3689 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3690 | } |
3691 | ||
95cdf3b7 IM |
3692 | static int |
3693 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3694 | { |
1da177e4 LT |
3695 | struct sched_param lparam; |
3696 | struct task_struct *p; | |
36c8b586 | 3697 | int retval; |
1da177e4 LT |
3698 | |
3699 | if (!param || pid < 0) | |
3700 | return -EINVAL; | |
3701 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3702 | return -EFAULT; | |
5fe1d75f ON |
3703 | |
3704 | rcu_read_lock(); | |
3705 | retval = -ESRCH; | |
1da177e4 | 3706 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3707 | if (p != NULL) |
3708 | retval = sched_setscheduler(p, policy, &lparam); | |
3709 | rcu_read_unlock(); | |
36c8b586 | 3710 | |
1da177e4 LT |
3711 | return retval; |
3712 | } | |
3713 | ||
d50dde5a DF |
3714 | /* |
3715 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3716 | */ | |
3717 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3718 | struct sched_attr *attr) | |
3719 | { | |
3720 | u32 size; | |
3721 | int ret; | |
3722 | ||
3723 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3724 | return -EFAULT; | |
3725 | ||
3726 | /* | |
3727 | * zero the full structure, so that a short copy will be nice. | |
3728 | */ | |
3729 | memset(attr, 0, sizeof(*attr)); | |
3730 | ||
3731 | ret = get_user(size, &uattr->size); | |
3732 | if (ret) | |
3733 | return ret; | |
3734 | ||
3735 | if (size > PAGE_SIZE) /* silly large */ | |
3736 | goto err_size; | |
3737 | ||
3738 | if (!size) /* abi compat */ | |
3739 | size = SCHED_ATTR_SIZE_VER0; | |
3740 | ||
3741 | if (size < SCHED_ATTR_SIZE_VER0) | |
3742 | goto err_size; | |
3743 | ||
3744 | /* | |
3745 | * If we're handed a bigger struct than we know of, | |
3746 | * ensure all the unknown bits are 0 - i.e. new | |
3747 | * user-space does not rely on any kernel feature | |
3748 | * extensions we dont know about yet. | |
3749 | */ | |
3750 | if (size > sizeof(*attr)) { | |
3751 | unsigned char __user *addr; | |
3752 | unsigned char __user *end; | |
3753 | unsigned char val; | |
3754 | ||
3755 | addr = (void __user *)uattr + sizeof(*attr); | |
3756 | end = (void __user *)uattr + size; | |
3757 | ||
3758 | for (; addr < end; addr++) { | |
3759 | ret = get_user(val, addr); | |
3760 | if (ret) | |
3761 | return ret; | |
3762 | if (val) | |
3763 | goto err_size; | |
3764 | } | |
3765 | size = sizeof(*attr); | |
3766 | } | |
3767 | ||
3768 | ret = copy_from_user(attr, uattr, size); | |
3769 | if (ret) | |
3770 | return -EFAULT; | |
3771 | ||
3772 | /* | |
3773 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3774 | * to be strict and return an error on out-of-bounds values? | |
3775 | */ | |
75e45d51 | 3776 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3777 | |
e78c7bca | 3778 | return 0; |
d50dde5a DF |
3779 | |
3780 | err_size: | |
3781 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3782 | return -E2BIG; |
d50dde5a DF |
3783 | } |
3784 | ||
1da177e4 LT |
3785 | /** |
3786 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3787 | * @pid: the pid in question. | |
3788 | * @policy: new policy. | |
3789 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3790 | * |
3791 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3792 | */ |
5add95d4 HC |
3793 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3794 | struct sched_param __user *, param) | |
1da177e4 | 3795 | { |
c21761f1 JB |
3796 | /* negative values for policy are not valid */ |
3797 | if (policy < 0) | |
3798 | return -EINVAL; | |
3799 | ||
1da177e4 LT |
3800 | return do_sched_setscheduler(pid, policy, param); |
3801 | } | |
3802 | ||
3803 | /** | |
3804 | * sys_sched_setparam - set/change the RT priority of a thread | |
3805 | * @pid: the pid in question. | |
3806 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3807 | * |
3808 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3809 | */ |
5add95d4 | 3810 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3811 | { |
c13db6b1 | 3812 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
3813 | } |
3814 | ||
d50dde5a DF |
3815 | /** |
3816 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3817 | * @pid: the pid in question. | |
5778fccf | 3818 | * @uattr: structure containing the extended parameters. |
db66d756 | 3819 | * @flags: for future extension. |
d50dde5a | 3820 | */ |
6d35ab48 PZ |
3821 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3822 | unsigned int, flags) | |
d50dde5a DF |
3823 | { |
3824 | struct sched_attr attr; | |
3825 | struct task_struct *p; | |
3826 | int retval; | |
3827 | ||
6d35ab48 | 3828 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3829 | return -EINVAL; |
3830 | ||
143cf23d MK |
3831 | retval = sched_copy_attr(uattr, &attr); |
3832 | if (retval) | |
3833 | return retval; | |
d50dde5a | 3834 | |
b14ed2c2 | 3835 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3836 | return -EINVAL; |
d50dde5a DF |
3837 | |
3838 | rcu_read_lock(); | |
3839 | retval = -ESRCH; | |
3840 | p = find_process_by_pid(pid); | |
3841 | if (p != NULL) | |
3842 | retval = sched_setattr(p, &attr); | |
3843 | rcu_read_unlock(); | |
3844 | ||
3845 | return retval; | |
3846 | } | |
3847 | ||
1da177e4 LT |
3848 | /** |
3849 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3850 | * @pid: the pid in question. | |
e69f6186 YB |
3851 | * |
3852 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3853 | * code. | |
1da177e4 | 3854 | */ |
5add95d4 | 3855 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3856 | { |
36c8b586 | 3857 | struct task_struct *p; |
3a5c359a | 3858 | int retval; |
1da177e4 LT |
3859 | |
3860 | if (pid < 0) | |
3a5c359a | 3861 | return -EINVAL; |
1da177e4 LT |
3862 | |
3863 | retval = -ESRCH; | |
5fe85be0 | 3864 | rcu_read_lock(); |
1da177e4 LT |
3865 | p = find_process_by_pid(pid); |
3866 | if (p) { | |
3867 | retval = security_task_getscheduler(p); | |
3868 | if (!retval) | |
ca94c442 LP |
3869 | retval = p->policy |
3870 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3871 | } |
5fe85be0 | 3872 | rcu_read_unlock(); |
1da177e4 LT |
3873 | return retval; |
3874 | } | |
3875 | ||
3876 | /** | |
ca94c442 | 3877 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3878 | * @pid: the pid in question. |
3879 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3880 | * |
3881 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3882 | * code. | |
1da177e4 | 3883 | */ |
5add95d4 | 3884 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3885 | { |
ce5f7f82 | 3886 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3887 | struct task_struct *p; |
3a5c359a | 3888 | int retval; |
1da177e4 LT |
3889 | |
3890 | if (!param || pid < 0) | |
3a5c359a | 3891 | return -EINVAL; |
1da177e4 | 3892 | |
5fe85be0 | 3893 | rcu_read_lock(); |
1da177e4 LT |
3894 | p = find_process_by_pid(pid); |
3895 | retval = -ESRCH; | |
3896 | if (!p) | |
3897 | goto out_unlock; | |
3898 | ||
3899 | retval = security_task_getscheduler(p); | |
3900 | if (retval) | |
3901 | goto out_unlock; | |
3902 | ||
ce5f7f82 PZ |
3903 | if (task_has_rt_policy(p)) |
3904 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3905 | rcu_read_unlock(); |
1da177e4 LT |
3906 | |
3907 | /* | |
3908 | * This one might sleep, we cannot do it with a spinlock held ... | |
3909 | */ | |
3910 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3911 | ||
1da177e4 LT |
3912 | return retval; |
3913 | ||
3914 | out_unlock: | |
5fe85be0 | 3915 | rcu_read_unlock(); |
1da177e4 LT |
3916 | return retval; |
3917 | } | |
3918 | ||
d50dde5a DF |
3919 | static int sched_read_attr(struct sched_attr __user *uattr, |
3920 | struct sched_attr *attr, | |
3921 | unsigned int usize) | |
3922 | { | |
3923 | int ret; | |
3924 | ||
3925 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3926 | return -EFAULT; | |
3927 | ||
3928 | /* | |
3929 | * If we're handed a smaller struct than we know of, | |
3930 | * ensure all the unknown bits are 0 - i.e. old | |
3931 | * user-space does not get uncomplete information. | |
3932 | */ | |
3933 | if (usize < sizeof(*attr)) { | |
3934 | unsigned char *addr; | |
3935 | unsigned char *end; | |
3936 | ||
3937 | addr = (void *)attr + usize; | |
3938 | end = (void *)attr + sizeof(*attr); | |
3939 | ||
3940 | for (; addr < end; addr++) { | |
3941 | if (*addr) | |
22400674 | 3942 | return -EFBIG; |
d50dde5a DF |
3943 | } |
3944 | ||
3945 | attr->size = usize; | |
3946 | } | |
3947 | ||
4efbc454 | 3948 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3949 | if (ret) |
3950 | return -EFAULT; | |
3951 | ||
22400674 | 3952 | return 0; |
d50dde5a DF |
3953 | } |
3954 | ||
3955 | /** | |
aab03e05 | 3956 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3957 | * @pid: the pid in question. |
5778fccf | 3958 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3959 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3960 | * @flags: for future extension. |
d50dde5a | 3961 | */ |
6d35ab48 PZ |
3962 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3963 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3964 | { |
3965 | struct sched_attr attr = { | |
3966 | .size = sizeof(struct sched_attr), | |
3967 | }; | |
3968 | struct task_struct *p; | |
3969 | int retval; | |
3970 | ||
3971 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3972 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3973 | return -EINVAL; |
3974 | ||
3975 | rcu_read_lock(); | |
3976 | p = find_process_by_pid(pid); | |
3977 | retval = -ESRCH; | |
3978 | if (!p) | |
3979 | goto out_unlock; | |
3980 | ||
3981 | retval = security_task_getscheduler(p); | |
3982 | if (retval) | |
3983 | goto out_unlock; | |
3984 | ||
3985 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3986 | if (p->sched_reset_on_fork) |
3987 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3988 | if (task_has_dl_policy(p)) |
3989 | __getparam_dl(p, &attr); | |
3990 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3991 | attr.sched_priority = p->rt_priority; |
3992 | else | |
d0ea0268 | 3993 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3994 | |
3995 | rcu_read_unlock(); | |
3996 | ||
3997 | retval = sched_read_attr(uattr, &attr, size); | |
3998 | return retval; | |
3999 | ||
4000 | out_unlock: | |
4001 | rcu_read_unlock(); | |
4002 | return retval; | |
4003 | } | |
4004 | ||
96f874e2 | 4005 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4006 | { |
5a16f3d3 | 4007 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4008 | struct task_struct *p; |
4009 | int retval; | |
1da177e4 | 4010 | |
23f5d142 | 4011 | rcu_read_lock(); |
1da177e4 LT |
4012 | |
4013 | p = find_process_by_pid(pid); | |
4014 | if (!p) { | |
23f5d142 | 4015 | rcu_read_unlock(); |
1da177e4 LT |
4016 | return -ESRCH; |
4017 | } | |
4018 | ||
23f5d142 | 4019 | /* Prevent p going away */ |
1da177e4 | 4020 | get_task_struct(p); |
23f5d142 | 4021 | rcu_read_unlock(); |
1da177e4 | 4022 | |
14a40ffc TH |
4023 | if (p->flags & PF_NO_SETAFFINITY) { |
4024 | retval = -EINVAL; | |
4025 | goto out_put_task; | |
4026 | } | |
5a16f3d3 RR |
4027 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4028 | retval = -ENOMEM; | |
4029 | goto out_put_task; | |
4030 | } | |
4031 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4032 | retval = -ENOMEM; | |
4033 | goto out_free_cpus_allowed; | |
4034 | } | |
1da177e4 | 4035 | retval = -EPERM; |
4c44aaaf EB |
4036 | if (!check_same_owner(p)) { |
4037 | rcu_read_lock(); | |
4038 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4039 | rcu_read_unlock(); | |
16303ab2 | 4040 | goto out_free_new_mask; |
4c44aaaf EB |
4041 | } |
4042 | rcu_read_unlock(); | |
4043 | } | |
1da177e4 | 4044 | |
b0ae1981 | 4045 | retval = security_task_setscheduler(p); |
e7834f8f | 4046 | if (retval) |
16303ab2 | 4047 | goto out_free_new_mask; |
e7834f8f | 4048 | |
e4099a5e PZ |
4049 | |
4050 | cpuset_cpus_allowed(p, cpus_allowed); | |
4051 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4052 | ||
332ac17e DF |
4053 | /* |
4054 | * Since bandwidth control happens on root_domain basis, | |
4055 | * if admission test is enabled, we only admit -deadline | |
4056 | * tasks allowed to run on all the CPUs in the task's | |
4057 | * root_domain. | |
4058 | */ | |
4059 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4060 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4061 | rcu_read_lock(); | |
4062 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4063 | retval = -EBUSY; |
f1e3a093 | 4064 | rcu_read_unlock(); |
16303ab2 | 4065 | goto out_free_new_mask; |
332ac17e | 4066 | } |
f1e3a093 | 4067 | rcu_read_unlock(); |
332ac17e DF |
4068 | } |
4069 | #endif | |
49246274 | 4070 | again: |
5a16f3d3 | 4071 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4072 | |
8707d8b8 | 4073 | if (!retval) { |
5a16f3d3 RR |
4074 | cpuset_cpus_allowed(p, cpus_allowed); |
4075 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4076 | /* |
4077 | * We must have raced with a concurrent cpuset | |
4078 | * update. Just reset the cpus_allowed to the | |
4079 | * cpuset's cpus_allowed | |
4080 | */ | |
5a16f3d3 | 4081 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4082 | goto again; |
4083 | } | |
4084 | } | |
16303ab2 | 4085 | out_free_new_mask: |
5a16f3d3 RR |
4086 | free_cpumask_var(new_mask); |
4087 | out_free_cpus_allowed: | |
4088 | free_cpumask_var(cpus_allowed); | |
4089 | out_put_task: | |
1da177e4 | 4090 | put_task_struct(p); |
1da177e4 LT |
4091 | return retval; |
4092 | } | |
4093 | ||
4094 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4095 | struct cpumask *new_mask) |
1da177e4 | 4096 | { |
96f874e2 RR |
4097 | if (len < cpumask_size()) |
4098 | cpumask_clear(new_mask); | |
4099 | else if (len > cpumask_size()) | |
4100 | len = cpumask_size(); | |
4101 | ||
1da177e4 LT |
4102 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4103 | } | |
4104 | ||
4105 | /** | |
4106 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4107 | * @pid: pid of the process | |
4108 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4109 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4110 | * |
4111 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4112 | */ |
5add95d4 HC |
4113 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4114 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4115 | { |
5a16f3d3 | 4116 | cpumask_var_t new_mask; |
1da177e4 LT |
4117 | int retval; |
4118 | ||
5a16f3d3 RR |
4119 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4120 | return -ENOMEM; | |
1da177e4 | 4121 | |
5a16f3d3 RR |
4122 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4123 | if (retval == 0) | |
4124 | retval = sched_setaffinity(pid, new_mask); | |
4125 | free_cpumask_var(new_mask); | |
4126 | return retval; | |
1da177e4 LT |
4127 | } |
4128 | ||
96f874e2 | 4129 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4130 | { |
36c8b586 | 4131 | struct task_struct *p; |
31605683 | 4132 | unsigned long flags; |
1da177e4 | 4133 | int retval; |
1da177e4 | 4134 | |
23f5d142 | 4135 | rcu_read_lock(); |
1da177e4 LT |
4136 | |
4137 | retval = -ESRCH; | |
4138 | p = find_process_by_pid(pid); | |
4139 | if (!p) | |
4140 | goto out_unlock; | |
4141 | ||
e7834f8f DQ |
4142 | retval = security_task_getscheduler(p); |
4143 | if (retval) | |
4144 | goto out_unlock; | |
4145 | ||
013fdb80 | 4146 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4147 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4148 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4149 | |
4150 | out_unlock: | |
23f5d142 | 4151 | rcu_read_unlock(); |
1da177e4 | 4152 | |
9531b62f | 4153 | return retval; |
1da177e4 LT |
4154 | } |
4155 | ||
4156 | /** | |
4157 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4158 | * @pid: pid of the process | |
4159 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4160 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4161 | * |
4162 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4163 | */ |
5add95d4 HC |
4164 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4165 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4166 | { |
4167 | int ret; | |
f17c8607 | 4168 | cpumask_var_t mask; |
1da177e4 | 4169 | |
84fba5ec | 4170 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4171 | return -EINVAL; |
4172 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4173 | return -EINVAL; |
4174 | ||
f17c8607 RR |
4175 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4176 | return -ENOMEM; | |
1da177e4 | 4177 | |
f17c8607 RR |
4178 | ret = sched_getaffinity(pid, mask); |
4179 | if (ret == 0) { | |
8bc037fb | 4180 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4181 | |
4182 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4183 | ret = -EFAULT; |
4184 | else | |
cd3d8031 | 4185 | ret = retlen; |
f17c8607 RR |
4186 | } |
4187 | free_cpumask_var(mask); | |
1da177e4 | 4188 | |
f17c8607 | 4189 | return ret; |
1da177e4 LT |
4190 | } |
4191 | ||
4192 | /** | |
4193 | * sys_sched_yield - yield the current processor to other threads. | |
4194 | * | |
dd41f596 IM |
4195 | * This function yields the current CPU to other tasks. If there are no |
4196 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4197 | * |
4198 | * Return: 0. | |
1da177e4 | 4199 | */ |
5add95d4 | 4200 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4201 | { |
70b97a7f | 4202 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4203 | |
2d72376b | 4204 | schedstat_inc(rq, yld_count); |
4530d7ab | 4205 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4206 | |
4207 | /* | |
4208 | * Since we are going to call schedule() anyway, there's | |
4209 | * no need to preempt or enable interrupts: | |
4210 | */ | |
4211 | __release(rq->lock); | |
8a25d5de | 4212 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4213 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4214 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4215 | |
4216 | schedule(); | |
4217 | ||
4218 | return 0; | |
4219 | } | |
4220 | ||
02b67cc3 | 4221 | int __sched _cond_resched(void) |
1da177e4 | 4222 | { |
d86ee480 | 4223 | if (should_resched()) { |
a18b5d01 | 4224 | preempt_schedule_common(); |
1da177e4 LT |
4225 | return 1; |
4226 | } | |
4227 | return 0; | |
4228 | } | |
02b67cc3 | 4229 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4230 | |
4231 | /* | |
613afbf8 | 4232 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4233 | * call schedule, and on return reacquire the lock. |
4234 | * | |
41a2d6cf | 4235 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4236 | * operations here to prevent schedule() from being called twice (once via |
4237 | * spin_unlock(), once by hand). | |
4238 | */ | |
613afbf8 | 4239 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4240 | { |
d86ee480 | 4241 | int resched = should_resched(); |
6df3cecb JK |
4242 | int ret = 0; |
4243 | ||
f607c668 PZ |
4244 | lockdep_assert_held(lock); |
4245 | ||
4a81e832 | 4246 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4247 | spin_unlock(lock); |
d86ee480 | 4248 | if (resched) |
a18b5d01 | 4249 | preempt_schedule_common(); |
95c354fe NP |
4250 | else |
4251 | cpu_relax(); | |
6df3cecb | 4252 | ret = 1; |
1da177e4 | 4253 | spin_lock(lock); |
1da177e4 | 4254 | } |
6df3cecb | 4255 | return ret; |
1da177e4 | 4256 | } |
613afbf8 | 4257 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4258 | |
613afbf8 | 4259 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4260 | { |
4261 | BUG_ON(!in_softirq()); | |
4262 | ||
d86ee480 | 4263 | if (should_resched()) { |
98d82567 | 4264 | local_bh_enable(); |
a18b5d01 | 4265 | preempt_schedule_common(); |
1da177e4 LT |
4266 | local_bh_disable(); |
4267 | return 1; | |
4268 | } | |
4269 | return 0; | |
4270 | } | |
613afbf8 | 4271 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4272 | |
1da177e4 LT |
4273 | /** |
4274 | * yield - yield the current processor to other threads. | |
4275 | * | |
8e3fabfd PZ |
4276 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4277 | * | |
4278 | * The scheduler is at all times free to pick the calling task as the most | |
4279 | * eligible task to run, if removing the yield() call from your code breaks | |
4280 | * it, its already broken. | |
4281 | * | |
4282 | * Typical broken usage is: | |
4283 | * | |
4284 | * while (!event) | |
4285 | * yield(); | |
4286 | * | |
4287 | * where one assumes that yield() will let 'the other' process run that will | |
4288 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4289 | * happen. Never use yield() as a progress guarantee!! | |
4290 | * | |
4291 | * If you want to use yield() to wait for something, use wait_event(). | |
4292 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4293 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4294 | */ |
4295 | void __sched yield(void) | |
4296 | { | |
4297 | set_current_state(TASK_RUNNING); | |
4298 | sys_sched_yield(); | |
4299 | } | |
1da177e4 LT |
4300 | EXPORT_SYMBOL(yield); |
4301 | ||
d95f4122 MG |
4302 | /** |
4303 | * yield_to - yield the current processor to another thread in | |
4304 | * your thread group, or accelerate that thread toward the | |
4305 | * processor it's on. | |
16addf95 RD |
4306 | * @p: target task |
4307 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4308 | * |
4309 | * It's the caller's job to ensure that the target task struct | |
4310 | * can't go away on us before we can do any checks. | |
4311 | * | |
e69f6186 | 4312 | * Return: |
7b270f60 PZ |
4313 | * true (>0) if we indeed boosted the target task. |
4314 | * false (0) if we failed to boost the target. | |
4315 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4316 | */ |
fa93384f | 4317 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4318 | { |
4319 | struct task_struct *curr = current; | |
4320 | struct rq *rq, *p_rq; | |
4321 | unsigned long flags; | |
c3c18640 | 4322 | int yielded = 0; |
d95f4122 MG |
4323 | |
4324 | local_irq_save(flags); | |
4325 | rq = this_rq(); | |
4326 | ||
4327 | again: | |
4328 | p_rq = task_rq(p); | |
7b270f60 PZ |
4329 | /* |
4330 | * If we're the only runnable task on the rq and target rq also | |
4331 | * has only one task, there's absolutely no point in yielding. | |
4332 | */ | |
4333 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4334 | yielded = -ESRCH; | |
4335 | goto out_irq; | |
4336 | } | |
4337 | ||
d95f4122 | 4338 | double_rq_lock(rq, p_rq); |
39e24d8f | 4339 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4340 | double_rq_unlock(rq, p_rq); |
4341 | goto again; | |
4342 | } | |
4343 | ||
4344 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4345 | goto out_unlock; |
d95f4122 MG |
4346 | |
4347 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4348 | goto out_unlock; |
d95f4122 MG |
4349 | |
4350 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4351 | goto out_unlock; |
d95f4122 MG |
4352 | |
4353 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4354 | if (yielded) { |
d95f4122 | 4355 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4356 | /* |
4357 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4358 | * fairness. | |
4359 | */ | |
4360 | if (preempt && rq != p_rq) | |
8875125e | 4361 | resched_curr(p_rq); |
6d1cafd8 | 4362 | } |
d95f4122 | 4363 | |
7b270f60 | 4364 | out_unlock: |
d95f4122 | 4365 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4366 | out_irq: |
d95f4122 MG |
4367 | local_irq_restore(flags); |
4368 | ||
7b270f60 | 4369 | if (yielded > 0) |
d95f4122 MG |
4370 | schedule(); |
4371 | ||
4372 | return yielded; | |
4373 | } | |
4374 | EXPORT_SYMBOL_GPL(yield_to); | |
4375 | ||
1da177e4 | 4376 | /* |
41a2d6cf | 4377 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4378 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4379 | */ |
1da177e4 LT |
4380 | long __sched io_schedule_timeout(long timeout) |
4381 | { | |
9cff8ade N |
4382 | int old_iowait = current->in_iowait; |
4383 | struct rq *rq; | |
1da177e4 LT |
4384 | long ret; |
4385 | ||
9cff8ade N |
4386 | current->in_iowait = 1; |
4387 | if (old_iowait) | |
4388 | blk_schedule_flush_plug(current); | |
4389 | else | |
4390 | blk_flush_plug(current); | |
4391 | ||
0ff92245 | 4392 | delayacct_blkio_start(); |
9cff8ade | 4393 | rq = raw_rq(); |
1da177e4 LT |
4394 | atomic_inc(&rq->nr_iowait); |
4395 | ret = schedule_timeout(timeout); | |
9cff8ade | 4396 | current->in_iowait = old_iowait; |
1da177e4 | 4397 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4398 | delayacct_blkio_end(); |
9cff8ade | 4399 | |
1da177e4 LT |
4400 | return ret; |
4401 | } | |
9cff8ade | 4402 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4403 | |
4404 | /** | |
4405 | * sys_sched_get_priority_max - return maximum RT priority. | |
4406 | * @policy: scheduling class. | |
4407 | * | |
e69f6186 YB |
4408 | * Return: On success, this syscall returns the maximum |
4409 | * rt_priority that can be used by a given scheduling class. | |
4410 | * On failure, a negative error code is returned. | |
1da177e4 | 4411 | */ |
5add95d4 | 4412 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4413 | { |
4414 | int ret = -EINVAL; | |
4415 | ||
4416 | switch (policy) { | |
4417 | case SCHED_FIFO: | |
4418 | case SCHED_RR: | |
4419 | ret = MAX_USER_RT_PRIO-1; | |
4420 | break; | |
aab03e05 | 4421 | case SCHED_DEADLINE: |
1da177e4 | 4422 | case SCHED_NORMAL: |
b0a9499c | 4423 | case SCHED_BATCH: |
dd41f596 | 4424 | case SCHED_IDLE: |
1da177e4 LT |
4425 | ret = 0; |
4426 | break; | |
4427 | } | |
4428 | return ret; | |
4429 | } | |
4430 | ||
4431 | /** | |
4432 | * sys_sched_get_priority_min - return minimum RT priority. | |
4433 | * @policy: scheduling class. | |
4434 | * | |
e69f6186 YB |
4435 | * Return: On success, this syscall returns the minimum |
4436 | * rt_priority that can be used by a given scheduling class. | |
4437 | * On failure, a negative error code is returned. | |
1da177e4 | 4438 | */ |
5add95d4 | 4439 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4440 | { |
4441 | int ret = -EINVAL; | |
4442 | ||
4443 | switch (policy) { | |
4444 | case SCHED_FIFO: | |
4445 | case SCHED_RR: | |
4446 | ret = 1; | |
4447 | break; | |
aab03e05 | 4448 | case SCHED_DEADLINE: |
1da177e4 | 4449 | case SCHED_NORMAL: |
b0a9499c | 4450 | case SCHED_BATCH: |
dd41f596 | 4451 | case SCHED_IDLE: |
1da177e4 LT |
4452 | ret = 0; |
4453 | } | |
4454 | return ret; | |
4455 | } | |
4456 | ||
4457 | /** | |
4458 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4459 | * @pid: pid of the process. | |
4460 | * @interval: userspace pointer to the timeslice value. | |
4461 | * | |
4462 | * this syscall writes the default timeslice value of a given process | |
4463 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4464 | * |
4465 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4466 | * an error code. | |
1da177e4 | 4467 | */ |
17da2bd9 | 4468 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4469 | struct timespec __user *, interval) |
1da177e4 | 4470 | { |
36c8b586 | 4471 | struct task_struct *p; |
a4ec24b4 | 4472 | unsigned int time_slice; |
dba091b9 TG |
4473 | unsigned long flags; |
4474 | struct rq *rq; | |
3a5c359a | 4475 | int retval; |
1da177e4 | 4476 | struct timespec t; |
1da177e4 LT |
4477 | |
4478 | if (pid < 0) | |
3a5c359a | 4479 | return -EINVAL; |
1da177e4 LT |
4480 | |
4481 | retval = -ESRCH; | |
1a551ae7 | 4482 | rcu_read_lock(); |
1da177e4 LT |
4483 | p = find_process_by_pid(pid); |
4484 | if (!p) | |
4485 | goto out_unlock; | |
4486 | ||
4487 | retval = security_task_getscheduler(p); | |
4488 | if (retval) | |
4489 | goto out_unlock; | |
4490 | ||
dba091b9 | 4491 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4492 | time_slice = 0; |
4493 | if (p->sched_class->get_rr_interval) | |
4494 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4495 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4496 | |
1a551ae7 | 4497 | rcu_read_unlock(); |
a4ec24b4 | 4498 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4499 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4500 | return retval; |
3a5c359a | 4501 | |
1da177e4 | 4502 | out_unlock: |
1a551ae7 | 4503 | rcu_read_unlock(); |
1da177e4 LT |
4504 | return retval; |
4505 | } | |
4506 | ||
7c731e0a | 4507 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4508 | |
82a1fcb9 | 4509 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4510 | { |
1da177e4 | 4511 | unsigned long free = 0; |
4e79752c | 4512 | int ppid; |
1f8a7633 | 4513 | unsigned long state = p->state; |
1da177e4 | 4514 | |
1f8a7633 TH |
4515 | if (state) |
4516 | state = __ffs(state) + 1; | |
28d0686c | 4517 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4518 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4519 | #if BITS_PER_LONG == 32 |
1da177e4 | 4520 | if (state == TASK_RUNNING) |
3df0fc5b | 4521 | printk(KERN_CONT " running "); |
1da177e4 | 4522 | else |
3df0fc5b | 4523 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4524 | #else |
4525 | if (state == TASK_RUNNING) | |
3df0fc5b | 4526 | printk(KERN_CONT " running task "); |
1da177e4 | 4527 | else |
3df0fc5b | 4528 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4529 | #endif |
4530 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4531 | free = stack_not_used(p); |
1da177e4 | 4532 | #endif |
a90e984c | 4533 | ppid = 0; |
4e79752c | 4534 | rcu_read_lock(); |
a90e984c ON |
4535 | if (pid_alive(p)) |
4536 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 4537 | rcu_read_unlock(); |
3df0fc5b | 4538 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4539 | task_pid_nr(p), ppid, |
aa47b7e0 | 4540 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4541 | |
3d1cb205 | 4542 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4543 | show_stack(p, NULL); |
1da177e4 LT |
4544 | } |
4545 | ||
e59e2ae2 | 4546 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4547 | { |
36c8b586 | 4548 | struct task_struct *g, *p; |
1da177e4 | 4549 | |
4bd77321 | 4550 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4551 | printk(KERN_INFO |
4552 | " task PC stack pid father\n"); | |
1da177e4 | 4553 | #else |
3df0fc5b PZ |
4554 | printk(KERN_INFO |
4555 | " task PC stack pid father\n"); | |
1da177e4 | 4556 | #endif |
510f5acc | 4557 | rcu_read_lock(); |
5d07f420 | 4558 | for_each_process_thread(g, p) { |
1da177e4 LT |
4559 | /* |
4560 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4561 | * console might take a lot of time: |
1da177e4 LT |
4562 | */ |
4563 | touch_nmi_watchdog(); | |
39bc89fd | 4564 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4565 | sched_show_task(p); |
5d07f420 | 4566 | } |
1da177e4 | 4567 | |
04c9167f JF |
4568 | touch_all_softlockup_watchdogs(); |
4569 | ||
dd41f596 IM |
4570 | #ifdef CONFIG_SCHED_DEBUG |
4571 | sysrq_sched_debug_show(); | |
4572 | #endif | |
510f5acc | 4573 | rcu_read_unlock(); |
e59e2ae2 IM |
4574 | /* |
4575 | * Only show locks if all tasks are dumped: | |
4576 | */ | |
93335a21 | 4577 | if (!state_filter) |
e59e2ae2 | 4578 | debug_show_all_locks(); |
1da177e4 LT |
4579 | } |
4580 | ||
0db0628d | 4581 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4582 | { |
dd41f596 | 4583 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4584 | } |
4585 | ||
f340c0d1 IM |
4586 | /** |
4587 | * init_idle - set up an idle thread for a given CPU | |
4588 | * @idle: task in question | |
4589 | * @cpu: cpu the idle task belongs to | |
4590 | * | |
4591 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4592 | * flag, to make booting more robust. | |
4593 | */ | |
0db0628d | 4594 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4595 | { |
70b97a7f | 4596 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4597 | unsigned long flags; |
4598 | ||
05fa785c | 4599 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4600 | |
5e1576ed | 4601 | __sched_fork(0, idle); |
06b83b5f | 4602 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4603 | idle->se.exec_start = sched_clock(); |
4604 | ||
1e1b6c51 | 4605 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4606 | /* |
4607 | * We're having a chicken and egg problem, even though we are | |
4608 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4609 | * lockdep check in task_group() will fail. | |
4610 | * | |
4611 | * Similar case to sched_fork(). / Alternatively we could | |
4612 | * use task_rq_lock() here and obtain the other rq->lock. | |
4613 | * | |
4614 | * Silence PROVE_RCU | |
4615 | */ | |
4616 | rcu_read_lock(); | |
dd41f596 | 4617 | __set_task_cpu(idle, cpu); |
6506cf6c | 4618 | rcu_read_unlock(); |
1da177e4 | 4619 | |
1da177e4 | 4620 | rq->curr = rq->idle = idle; |
da0c1e65 | 4621 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4622 | #if defined(CONFIG_SMP) |
4623 | idle->on_cpu = 1; | |
4866cde0 | 4624 | #endif |
05fa785c | 4625 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4626 | |
4627 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4628 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4629 | |
dd41f596 IM |
4630 | /* |
4631 | * The idle tasks have their own, simple scheduling class: | |
4632 | */ | |
4633 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4634 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4635 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4636 | #if defined(CONFIG_SMP) |
4637 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4638 | #endif | |
19978ca6 IM |
4639 | } |
4640 | ||
f82f8042 JL |
4641 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
4642 | const struct cpumask *trial) | |
4643 | { | |
4644 | int ret = 1, trial_cpus; | |
4645 | struct dl_bw *cur_dl_b; | |
4646 | unsigned long flags; | |
4647 | ||
bb2bc55a MG |
4648 | if (!cpumask_weight(cur)) |
4649 | return ret; | |
4650 | ||
75e23e49 | 4651 | rcu_read_lock_sched(); |
f82f8042 JL |
4652 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
4653 | trial_cpus = cpumask_weight(trial); | |
4654 | ||
4655 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
4656 | if (cur_dl_b->bw != -1 && | |
4657 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
4658 | ret = 0; | |
4659 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 4660 | rcu_read_unlock_sched(); |
f82f8042 JL |
4661 | |
4662 | return ret; | |
4663 | } | |
4664 | ||
7f51412a JL |
4665 | int task_can_attach(struct task_struct *p, |
4666 | const struct cpumask *cs_cpus_allowed) | |
4667 | { | |
4668 | int ret = 0; | |
4669 | ||
4670 | /* | |
4671 | * Kthreads which disallow setaffinity shouldn't be moved | |
4672 | * to a new cpuset; we don't want to change their cpu | |
4673 | * affinity and isolating such threads by their set of | |
4674 | * allowed nodes is unnecessary. Thus, cpusets are not | |
4675 | * applicable for such threads. This prevents checking for | |
4676 | * success of set_cpus_allowed_ptr() on all attached tasks | |
4677 | * before cpus_allowed may be changed. | |
4678 | */ | |
4679 | if (p->flags & PF_NO_SETAFFINITY) { | |
4680 | ret = -EINVAL; | |
4681 | goto out; | |
4682 | } | |
4683 | ||
4684 | #ifdef CONFIG_SMP | |
4685 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
4686 | cs_cpus_allowed)) { | |
4687 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
4688 | cs_cpus_allowed); | |
75e23e49 | 4689 | struct dl_bw *dl_b; |
7f51412a JL |
4690 | bool overflow; |
4691 | int cpus; | |
4692 | unsigned long flags; | |
4693 | ||
75e23e49 JL |
4694 | rcu_read_lock_sched(); |
4695 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
4696 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
4697 | cpus = dl_bw_cpus(dest_cpu); | |
4698 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
4699 | if (overflow) | |
4700 | ret = -EBUSY; | |
4701 | else { | |
4702 | /* | |
4703 | * We reserve space for this task in the destination | |
4704 | * root_domain, as we can't fail after this point. | |
4705 | * We will free resources in the source root_domain | |
4706 | * later on (see set_cpus_allowed_dl()). | |
4707 | */ | |
4708 | __dl_add(dl_b, p->dl.dl_bw); | |
4709 | } | |
4710 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 4711 | rcu_read_unlock_sched(); |
7f51412a JL |
4712 | |
4713 | } | |
4714 | #endif | |
4715 | out: | |
4716 | return ret; | |
4717 | } | |
4718 | ||
1da177e4 | 4719 | #ifdef CONFIG_SMP |
a15b12ac KT |
4720 | /* |
4721 | * move_queued_task - move a queued task to new rq. | |
4722 | * | |
4723 | * Returns (locked) new rq. Old rq's lock is released. | |
4724 | */ | |
4725 | static struct rq *move_queued_task(struct task_struct *p, int new_cpu) | |
4726 | { | |
4727 | struct rq *rq = task_rq(p); | |
4728 | ||
4729 | lockdep_assert_held(&rq->lock); | |
4730 | ||
4731 | dequeue_task(rq, p, 0); | |
4732 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
4733 | set_task_cpu(p, new_cpu); | |
4734 | raw_spin_unlock(&rq->lock); | |
4735 | ||
4736 | rq = cpu_rq(new_cpu); | |
4737 | ||
4738 | raw_spin_lock(&rq->lock); | |
4739 | BUG_ON(task_cpu(p) != new_cpu); | |
4740 | p->on_rq = TASK_ON_RQ_QUEUED; | |
4741 | enqueue_task(rq, p, 0); | |
4742 | check_preempt_curr(rq, p, 0); | |
4743 | ||
4744 | return rq; | |
4745 | } | |
4746 | ||
1e1b6c51 KM |
4747 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4748 | { | |
1b537c7d | 4749 | if (p->sched_class->set_cpus_allowed) |
1e1b6c51 | 4750 | p->sched_class->set_cpus_allowed(p, new_mask); |
4939602a PZ |
4751 | |
4752 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4753 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4754 | } |
4755 | ||
1da177e4 LT |
4756 | /* |
4757 | * This is how migration works: | |
4758 | * | |
969c7921 TH |
4759 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4760 | * stop_one_cpu(). | |
4761 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4762 | * off the CPU) | |
4763 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4764 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4765 | * it and puts it into the right queue. |
969c7921 TH |
4766 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4767 | * is done. | |
1da177e4 LT |
4768 | */ |
4769 | ||
4770 | /* | |
4771 | * Change a given task's CPU affinity. Migrate the thread to a | |
4772 | * proper CPU and schedule it away if the CPU it's executing on | |
4773 | * is removed from the allowed bitmask. | |
4774 | * | |
4775 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4776 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4777 | * call is not atomic; no spinlocks may be held. |
4778 | */ | |
96f874e2 | 4779 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4780 | { |
4781 | unsigned long flags; | |
70b97a7f | 4782 | struct rq *rq; |
969c7921 | 4783 | unsigned int dest_cpu; |
48f24c4d | 4784 | int ret = 0; |
1da177e4 LT |
4785 | |
4786 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4787 | |
db44fc01 YZ |
4788 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4789 | goto out; | |
4790 | ||
6ad4c188 | 4791 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4792 | ret = -EINVAL; |
4793 | goto out; | |
4794 | } | |
4795 | ||
1e1b6c51 | 4796 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4797 | |
1da177e4 | 4798 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4799 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4800 | goto out; |
4801 | ||
969c7921 | 4802 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
a15b12ac | 4803 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
969c7921 | 4804 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4805 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4806 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4807 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4808 | tlb_migrate_finish(p->mm); |
4809 | return 0; | |
a15b12ac KT |
4810 | } else if (task_on_rq_queued(p)) |
4811 | rq = move_queued_task(p, dest_cpu); | |
1da177e4 | 4812 | out: |
0122ec5b | 4813 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4814 | |
1da177e4 LT |
4815 | return ret; |
4816 | } | |
cd8ba7cd | 4817 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4818 | |
4819 | /* | |
41a2d6cf | 4820 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4821 | * this because either it can't run here any more (set_cpus_allowed() |
4822 | * away from this CPU, or CPU going down), or because we're | |
4823 | * attempting to rebalance this task on exec (sched_exec). | |
4824 | * | |
4825 | * So we race with normal scheduler movements, but that's OK, as long | |
4826 | * as the task is no longer on this CPU. | |
efc30814 KK |
4827 | * |
4828 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4829 | */ |
efc30814 | 4830 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4831 | { |
a1e01829 | 4832 | struct rq *rq; |
e2912009 | 4833 | int ret = 0; |
1da177e4 | 4834 | |
e761b772 | 4835 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4836 | return ret; |
1da177e4 | 4837 | |
a1e01829 | 4838 | rq = cpu_rq(src_cpu); |
1da177e4 | 4839 | |
0122ec5b | 4840 | raw_spin_lock(&p->pi_lock); |
a1e01829 | 4841 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
4842 | /* Already moved. */ |
4843 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4844 | goto done; |
a1e01829 | 4845 | |
1da177e4 | 4846 | /* Affinity changed (again). */ |
fa17b507 | 4847 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4848 | goto fail; |
1da177e4 | 4849 | |
e2912009 PZ |
4850 | /* |
4851 | * If we're not on a rq, the next wake-up will ensure we're | |
4852 | * placed properly. | |
4853 | */ | |
a15b12ac KT |
4854 | if (task_on_rq_queued(p)) |
4855 | rq = move_queued_task(p, dest_cpu); | |
b1e38734 | 4856 | done: |
efc30814 | 4857 | ret = 1; |
b1e38734 | 4858 | fail: |
a1e01829 | 4859 | raw_spin_unlock(&rq->lock); |
0122ec5b | 4860 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4861 | return ret; |
1da177e4 LT |
4862 | } |
4863 | ||
e6628d5b MG |
4864 | #ifdef CONFIG_NUMA_BALANCING |
4865 | /* Migrate current task p to target_cpu */ | |
4866 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4867 | { | |
4868 | struct migration_arg arg = { p, target_cpu }; | |
4869 | int curr_cpu = task_cpu(p); | |
4870 | ||
4871 | if (curr_cpu == target_cpu) | |
4872 | return 0; | |
4873 | ||
4874 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4875 | return -EINVAL; | |
4876 | ||
4877 | /* TODO: This is not properly updating schedstats */ | |
4878 | ||
286549dc | 4879 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4880 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4881 | } | |
0ec8aa00 PZ |
4882 | |
4883 | /* | |
4884 | * Requeue a task on a given node and accurately track the number of NUMA | |
4885 | * tasks on the runqueues | |
4886 | */ | |
4887 | void sched_setnuma(struct task_struct *p, int nid) | |
4888 | { | |
4889 | struct rq *rq; | |
4890 | unsigned long flags; | |
da0c1e65 | 4891 | bool queued, running; |
0ec8aa00 PZ |
4892 | |
4893 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 4894 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
4895 | running = task_current(rq, p); |
4896 | ||
da0c1e65 | 4897 | if (queued) |
0ec8aa00 PZ |
4898 | dequeue_task(rq, p, 0); |
4899 | if (running) | |
f3cd1c4e | 4900 | put_prev_task(rq, p); |
0ec8aa00 PZ |
4901 | |
4902 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4903 | |
4904 | if (running) | |
4905 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4906 | if (queued) |
0ec8aa00 PZ |
4907 | enqueue_task(rq, p, 0); |
4908 | task_rq_unlock(rq, p, &flags); | |
4909 | } | |
e6628d5b MG |
4910 | #endif |
4911 | ||
1da177e4 | 4912 | /* |
969c7921 TH |
4913 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4914 | * and performs thread migration by bumping thread off CPU then | |
4915 | * 'pushing' onto another runqueue. | |
1da177e4 | 4916 | */ |
969c7921 | 4917 | static int migration_cpu_stop(void *data) |
1da177e4 | 4918 | { |
969c7921 | 4919 | struct migration_arg *arg = data; |
f7b4cddc | 4920 | |
969c7921 TH |
4921 | /* |
4922 | * The original target cpu might have gone down and we might | |
4923 | * be on another cpu but it doesn't matter. | |
4924 | */ | |
f7b4cddc | 4925 | local_irq_disable(); |
5cd038f5 LJ |
4926 | /* |
4927 | * We need to explicitly wake pending tasks before running | |
4928 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
4929 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
4930 | */ | |
4931 | sched_ttwu_pending(); | |
969c7921 | 4932 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4933 | local_irq_enable(); |
1da177e4 | 4934 | return 0; |
f7b4cddc ON |
4935 | } |
4936 | ||
1da177e4 | 4937 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4938 | |
054b9108 | 4939 | /* |
48c5ccae PZ |
4940 | * Ensures that the idle task is using init_mm right before its cpu goes |
4941 | * offline. | |
054b9108 | 4942 | */ |
48c5ccae | 4943 | void idle_task_exit(void) |
1da177e4 | 4944 | { |
48c5ccae | 4945 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4946 | |
48c5ccae | 4947 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4948 | |
a53efe5f | 4949 | if (mm != &init_mm) { |
48c5ccae | 4950 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4951 | finish_arch_post_lock_switch(); |
4952 | } | |
48c5ccae | 4953 | mmdrop(mm); |
1da177e4 LT |
4954 | } |
4955 | ||
4956 | /* | |
5d180232 PZ |
4957 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4958 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4959 | * nr_active count is stable. | |
4960 | * | |
4961 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4962 | */ |
5d180232 | 4963 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4964 | { |
5d180232 PZ |
4965 | long delta = calc_load_fold_active(rq); |
4966 | if (delta) | |
4967 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4968 | } |
4969 | ||
3f1d2a31 PZ |
4970 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4971 | { | |
4972 | } | |
4973 | ||
4974 | static const struct sched_class fake_sched_class = { | |
4975 | .put_prev_task = put_prev_task_fake, | |
4976 | }; | |
4977 | ||
4978 | static struct task_struct fake_task = { | |
4979 | /* | |
4980 | * Avoid pull_{rt,dl}_task() | |
4981 | */ | |
4982 | .prio = MAX_PRIO + 1, | |
4983 | .sched_class = &fake_sched_class, | |
4984 | }; | |
4985 | ||
48f24c4d | 4986 | /* |
48c5ccae PZ |
4987 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4988 | * try_to_wake_up()->select_task_rq(). | |
4989 | * | |
4990 | * Called with rq->lock held even though we'er in stop_machine() and | |
4991 | * there's no concurrency possible, we hold the required locks anyway | |
4992 | * because of lock validation efforts. | |
1da177e4 | 4993 | */ |
48c5ccae | 4994 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4995 | { |
70b97a7f | 4996 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4997 | struct task_struct *next, *stop = rq->stop; |
4998 | int dest_cpu; | |
1da177e4 LT |
4999 | |
5000 | /* | |
48c5ccae PZ |
5001 | * Fudge the rq selection such that the below task selection loop |
5002 | * doesn't get stuck on the currently eligible stop task. | |
5003 | * | |
5004 | * We're currently inside stop_machine() and the rq is either stuck | |
5005 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5006 | * either way we should never end up calling schedule() until we're | |
5007 | * done here. | |
1da177e4 | 5008 | */ |
48c5ccae | 5009 | rq->stop = NULL; |
48f24c4d | 5010 | |
77bd3970 FW |
5011 | /* |
5012 | * put_prev_task() and pick_next_task() sched | |
5013 | * class method both need to have an up-to-date | |
5014 | * value of rq->clock[_task] | |
5015 | */ | |
5016 | update_rq_clock(rq); | |
5017 | ||
dd41f596 | 5018 | for ( ; ; ) { |
48c5ccae PZ |
5019 | /* |
5020 | * There's this thread running, bail when that's the only | |
5021 | * remaining thread. | |
5022 | */ | |
5023 | if (rq->nr_running == 1) | |
dd41f596 | 5024 | break; |
48c5ccae | 5025 | |
3f1d2a31 | 5026 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5027 | BUG_ON(!next); |
79c53799 | 5028 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5029 | |
48c5ccae PZ |
5030 | /* Find suitable destination for @next, with force if needed. */ |
5031 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5032 | raw_spin_unlock(&rq->lock); | |
5033 | ||
5034 | __migrate_task(next, dead_cpu, dest_cpu); | |
5035 | ||
5036 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5037 | } |
dce48a84 | 5038 | |
48c5ccae | 5039 | rq->stop = stop; |
dce48a84 | 5040 | } |
48c5ccae | 5041 | |
1da177e4 LT |
5042 | #endif /* CONFIG_HOTPLUG_CPU */ |
5043 | ||
e692ab53 NP |
5044 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5045 | ||
5046 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5047 | { |
5048 | .procname = "sched_domain", | |
c57baf1e | 5049 | .mode = 0555, |
e0361851 | 5050 | }, |
56992309 | 5051 | {} |
e692ab53 NP |
5052 | }; |
5053 | ||
5054 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5055 | { |
5056 | .procname = "kernel", | |
c57baf1e | 5057 | .mode = 0555, |
e0361851 AD |
5058 | .child = sd_ctl_dir, |
5059 | }, | |
56992309 | 5060 | {} |
e692ab53 NP |
5061 | }; |
5062 | ||
5063 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5064 | { | |
5065 | struct ctl_table *entry = | |
5cf9f062 | 5066 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5067 | |
e692ab53 NP |
5068 | return entry; |
5069 | } | |
5070 | ||
6382bc90 MM |
5071 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5072 | { | |
cd790076 | 5073 | struct ctl_table *entry; |
6382bc90 | 5074 | |
cd790076 MM |
5075 | /* |
5076 | * In the intermediate directories, both the child directory and | |
5077 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5078 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5079 | * static strings and all have proc handlers. |
5080 | */ | |
5081 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5082 | if (entry->child) |
5083 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5084 | if (entry->proc_handler == NULL) |
5085 | kfree(entry->procname); | |
5086 | } | |
6382bc90 MM |
5087 | |
5088 | kfree(*tablep); | |
5089 | *tablep = NULL; | |
5090 | } | |
5091 | ||
201c373e | 5092 | static int min_load_idx = 0; |
fd9b86d3 | 5093 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5094 | |
e692ab53 | 5095 | static void |
e0361851 | 5096 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5097 | const char *procname, void *data, int maxlen, |
201c373e NK |
5098 | umode_t mode, proc_handler *proc_handler, |
5099 | bool load_idx) | |
e692ab53 | 5100 | { |
e692ab53 NP |
5101 | entry->procname = procname; |
5102 | entry->data = data; | |
5103 | entry->maxlen = maxlen; | |
5104 | entry->mode = mode; | |
5105 | entry->proc_handler = proc_handler; | |
201c373e NK |
5106 | |
5107 | if (load_idx) { | |
5108 | entry->extra1 = &min_load_idx; | |
5109 | entry->extra2 = &max_load_idx; | |
5110 | } | |
e692ab53 NP |
5111 | } |
5112 | ||
5113 | static struct ctl_table * | |
5114 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5115 | { | |
37e6bae8 | 5116 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 5117 | |
ad1cdc1d MM |
5118 | if (table == NULL) |
5119 | return NULL; | |
5120 | ||
e0361851 | 5121 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5122 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5123 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5124 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5125 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5126 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5127 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5128 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5129 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5130 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5131 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5132 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5133 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5134 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5135 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5136 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5137 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5138 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5139 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5140 | &sd->cache_nice_tries, |
201c373e | 5141 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5142 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5143 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
5144 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5145 | &sd->max_newidle_lb_cost, | |
5146 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5147 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5148 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5149 | /* &table[13] is terminator */ |
e692ab53 NP |
5150 | |
5151 | return table; | |
5152 | } | |
5153 | ||
be7002e6 | 5154 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5155 | { |
5156 | struct ctl_table *entry, *table; | |
5157 | struct sched_domain *sd; | |
5158 | int domain_num = 0, i; | |
5159 | char buf[32]; | |
5160 | ||
5161 | for_each_domain(cpu, sd) | |
5162 | domain_num++; | |
5163 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5164 | if (table == NULL) |
5165 | return NULL; | |
e692ab53 NP |
5166 | |
5167 | i = 0; | |
5168 | for_each_domain(cpu, sd) { | |
5169 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5170 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5171 | entry->mode = 0555; |
e692ab53 NP |
5172 | entry->child = sd_alloc_ctl_domain_table(sd); |
5173 | entry++; | |
5174 | i++; | |
5175 | } | |
5176 | return table; | |
5177 | } | |
5178 | ||
5179 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5180 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5181 | { |
6ad4c188 | 5182 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5183 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5184 | char buf[32]; | |
5185 | ||
7378547f MM |
5186 | WARN_ON(sd_ctl_dir[0].child); |
5187 | sd_ctl_dir[0].child = entry; | |
5188 | ||
ad1cdc1d MM |
5189 | if (entry == NULL) |
5190 | return; | |
5191 | ||
6ad4c188 | 5192 | for_each_possible_cpu(i) { |
e692ab53 | 5193 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5194 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5195 | entry->mode = 0555; |
e692ab53 | 5196 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5197 | entry++; |
e692ab53 | 5198 | } |
7378547f MM |
5199 | |
5200 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5201 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5202 | } | |
6382bc90 | 5203 | |
7378547f | 5204 | /* may be called multiple times per register */ |
6382bc90 MM |
5205 | static void unregister_sched_domain_sysctl(void) |
5206 | { | |
7378547f MM |
5207 | if (sd_sysctl_header) |
5208 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5209 | sd_sysctl_header = NULL; |
7378547f MM |
5210 | if (sd_ctl_dir[0].child) |
5211 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5212 | } |
e692ab53 | 5213 | #else |
6382bc90 MM |
5214 | static void register_sched_domain_sysctl(void) |
5215 | { | |
5216 | } | |
5217 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5218 | { |
5219 | } | |
5220 | #endif | |
5221 | ||
1f11eb6a GH |
5222 | static void set_rq_online(struct rq *rq) |
5223 | { | |
5224 | if (!rq->online) { | |
5225 | const struct sched_class *class; | |
5226 | ||
c6c4927b | 5227 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5228 | rq->online = 1; |
5229 | ||
5230 | for_each_class(class) { | |
5231 | if (class->rq_online) | |
5232 | class->rq_online(rq); | |
5233 | } | |
5234 | } | |
5235 | } | |
5236 | ||
5237 | static void set_rq_offline(struct rq *rq) | |
5238 | { | |
5239 | if (rq->online) { | |
5240 | const struct sched_class *class; | |
5241 | ||
5242 | for_each_class(class) { | |
5243 | if (class->rq_offline) | |
5244 | class->rq_offline(rq); | |
5245 | } | |
5246 | ||
c6c4927b | 5247 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5248 | rq->online = 0; |
5249 | } | |
5250 | } | |
5251 | ||
1da177e4 LT |
5252 | /* |
5253 | * migration_call - callback that gets triggered when a CPU is added. | |
5254 | * Here we can start up the necessary migration thread for the new CPU. | |
5255 | */ | |
0db0628d | 5256 | static int |
48f24c4d | 5257 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5258 | { |
48f24c4d | 5259 | int cpu = (long)hcpu; |
1da177e4 | 5260 | unsigned long flags; |
969c7921 | 5261 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5262 | |
48c5ccae | 5263 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5264 | |
1da177e4 | 5265 | case CPU_UP_PREPARE: |
a468d389 | 5266 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5267 | break; |
48f24c4d | 5268 | |
1da177e4 | 5269 | case CPU_ONLINE: |
1f94ef59 | 5270 | /* Update our root-domain */ |
05fa785c | 5271 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5272 | if (rq->rd) { |
c6c4927b | 5273 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5274 | |
5275 | set_rq_online(rq); | |
1f94ef59 | 5276 | } |
05fa785c | 5277 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5278 | break; |
48f24c4d | 5279 | |
1da177e4 | 5280 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5281 | case CPU_DYING: |
317f3941 | 5282 | sched_ttwu_pending(); |
57d885fe | 5283 | /* Update our root-domain */ |
05fa785c | 5284 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5285 | if (rq->rd) { |
c6c4927b | 5286 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5287 | set_rq_offline(rq); |
57d885fe | 5288 | } |
48c5ccae PZ |
5289 | migrate_tasks(cpu); |
5290 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5291 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5292 | break; |
48c5ccae | 5293 | |
5d180232 | 5294 | case CPU_DEAD: |
f319da0c | 5295 | calc_load_migrate(rq); |
57d885fe | 5296 | break; |
1da177e4 LT |
5297 | #endif |
5298 | } | |
49c022e6 PZ |
5299 | |
5300 | update_max_interval(); | |
5301 | ||
1da177e4 LT |
5302 | return NOTIFY_OK; |
5303 | } | |
5304 | ||
f38b0820 PM |
5305 | /* |
5306 | * Register at high priority so that task migration (migrate_all_tasks) | |
5307 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5308 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5309 | */ |
0db0628d | 5310 | static struct notifier_block migration_notifier = { |
1da177e4 | 5311 | .notifier_call = migration_call, |
50a323b7 | 5312 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5313 | }; |
5314 | ||
a803f026 CM |
5315 | static void __cpuinit set_cpu_rq_start_time(void) |
5316 | { | |
5317 | int cpu = smp_processor_id(); | |
5318 | struct rq *rq = cpu_rq(cpu); | |
5319 | rq->age_stamp = sched_clock_cpu(cpu); | |
5320 | } | |
5321 | ||
0db0628d | 5322 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5323 | unsigned long action, void *hcpu) |
5324 | { | |
5325 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5326 | case CPU_STARTING: |
5327 | set_cpu_rq_start_time(); | |
5328 | return NOTIFY_OK; | |
3a101d05 TH |
5329 | case CPU_DOWN_FAILED: |
5330 | set_cpu_active((long)hcpu, true); | |
5331 | return NOTIFY_OK; | |
5332 | default: | |
5333 | return NOTIFY_DONE; | |
5334 | } | |
5335 | } | |
5336 | ||
0db0628d | 5337 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5338 | unsigned long action, void *hcpu) |
5339 | { | |
de212f18 PZ |
5340 | unsigned long flags; |
5341 | long cpu = (long)hcpu; | |
f10e00f4 | 5342 | struct dl_bw *dl_b; |
de212f18 | 5343 | |
3a101d05 TH |
5344 | switch (action & ~CPU_TASKS_FROZEN) { |
5345 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5346 | set_cpu_active(cpu, false); |
5347 | ||
5348 | /* explicitly allow suspend */ | |
5349 | if (!(action & CPU_TASKS_FROZEN)) { | |
de212f18 PZ |
5350 | bool overflow; |
5351 | int cpus; | |
5352 | ||
f10e00f4 KT |
5353 | rcu_read_lock_sched(); |
5354 | dl_b = dl_bw_of(cpu); | |
5355 | ||
de212f18 PZ |
5356 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5357 | cpus = dl_bw_cpus(cpu); | |
5358 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5359 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5360 | ||
f10e00f4 KT |
5361 | rcu_read_unlock_sched(); |
5362 | ||
de212f18 PZ |
5363 | if (overflow) |
5364 | return notifier_from_errno(-EBUSY); | |
5365 | } | |
3a101d05 | 5366 | return NOTIFY_OK; |
3a101d05 | 5367 | } |
de212f18 PZ |
5368 | |
5369 | return NOTIFY_DONE; | |
3a101d05 TH |
5370 | } |
5371 | ||
7babe8db | 5372 | static int __init migration_init(void) |
1da177e4 LT |
5373 | { |
5374 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5375 | int err; |
48f24c4d | 5376 | |
3a101d05 | 5377 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5378 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5379 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5380 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5381 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5382 | |
3a101d05 TH |
5383 | /* Register cpu active notifiers */ |
5384 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5385 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5386 | ||
a004cd42 | 5387 | return 0; |
1da177e4 | 5388 | } |
7babe8db | 5389 | early_initcall(migration_init); |
1da177e4 LT |
5390 | #endif |
5391 | ||
5392 | #ifdef CONFIG_SMP | |
476f3534 | 5393 | |
4cb98839 PZ |
5394 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5395 | ||
3e9830dc | 5396 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5397 | |
d039ac60 | 5398 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5399 | |
d039ac60 | 5400 | static int __init sched_debug_setup(char *str) |
f6630114 | 5401 | { |
d039ac60 | 5402 | sched_debug_enabled = 1; |
f6630114 MT |
5403 | |
5404 | return 0; | |
5405 | } | |
d039ac60 PZ |
5406 | early_param("sched_debug", sched_debug_setup); |
5407 | ||
5408 | static inline bool sched_debug(void) | |
5409 | { | |
5410 | return sched_debug_enabled; | |
5411 | } | |
f6630114 | 5412 | |
7c16ec58 | 5413 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5414 | struct cpumask *groupmask) |
1da177e4 | 5415 | { |
4dcf6aff | 5416 | struct sched_group *group = sd->groups; |
1da177e4 | 5417 | |
96f874e2 | 5418 | cpumask_clear(groupmask); |
4dcf6aff IM |
5419 | |
5420 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5421 | ||
5422 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5423 | printk("does not load-balance\n"); |
4dcf6aff | 5424 | if (sd->parent) |
3df0fc5b PZ |
5425 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5426 | " has parent"); | |
4dcf6aff | 5427 | return -1; |
41c7ce9a NP |
5428 | } |
5429 | ||
333470ee TH |
5430 | printk(KERN_CONT "span %*pbl level %s\n", |
5431 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5432 | |
758b2cdc | 5433 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5434 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5435 | "CPU%d\n", cpu); | |
4dcf6aff | 5436 | } |
758b2cdc | 5437 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5438 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5439 | " CPU%d\n", cpu); | |
4dcf6aff | 5440 | } |
1da177e4 | 5441 | |
4dcf6aff | 5442 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5443 | do { |
4dcf6aff | 5444 | if (!group) { |
3df0fc5b PZ |
5445 | printk("\n"); |
5446 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5447 | break; |
5448 | } | |
5449 | ||
c3decf0d | 5450 | /* |
63b2ca30 NP |
5451 | * Even though we initialize ->capacity to something semi-sane, |
5452 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5453 | * domain iteration is still funny without causing /0 traps. |
5454 | */ | |
63b2ca30 | 5455 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5456 | printk(KERN_CONT "\n"); |
63b2ca30 | 5457 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5458 | break; |
5459 | } | |
1da177e4 | 5460 | |
758b2cdc | 5461 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5462 | printk(KERN_CONT "\n"); |
5463 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5464 | break; |
5465 | } | |
1da177e4 | 5466 | |
cb83b629 PZ |
5467 | if (!(sd->flags & SD_OVERLAP) && |
5468 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5469 | printk(KERN_CONT "\n"); |
5470 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5471 | break; |
5472 | } | |
1da177e4 | 5473 | |
758b2cdc | 5474 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5475 | |
333470ee TH |
5476 | printk(KERN_CONT " %*pbl", |
5477 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5478 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5479 | printk(KERN_CONT " (cpu_capacity = %d)", |
5480 | group->sgc->capacity); | |
381512cf | 5481 | } |
1da177e4 | 5482 | |
4dcf6aff IM |
5483 | group = group->next; |
5484 | } while (group != sd->groups); | |
3df0fc5b | 5485 | printk(KERN_CONT "\n"); |
1da177e4 | 5486 | |
758b2cdc | 5487 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5488 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5489 | |
758b2cdc RR |
5490 | if (sd->parent && |
5491 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5492 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5493 | "of domain->span\n"); | |
4dcf6aff IM |
5494 | return 0; |
5495 | } | |
1da177e4 | 5496 | |
4dcf6aff IM |
5497 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5498 | { | |
5499 | int level = 0; | |
1da177e4 | 5500 | |
d039ac60 | 5501 | if (!sched_debug_enabled) |
f6630114 MT |
5502 | return; |
5503 | ||
4dcf6aff IM |
5504 | if (!sd) { |
5505 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5506 | return; | |
5507 | } | |
1da177e4 | 5508 | |
4dcf6aff IM |
5509 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5510 | ||
5511 | for (;;) { | |
4cb98839 | 5512 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5513 | break; |
1da177e4 LT |
5514 | level++; |
5515 | sd = sd->parent; | |
33859f7f | 5516 | if (!sd) |
4dcf6aff IM |
5517 | break; |
5518 | } | |
1da177e4 | 5519 | } |
6d6bc0ad | 5520 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5521 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5522 | static inline bool sched_debug(void) |
5523 | { | |
5524 | return false; | |
5525 | } | |
6d6bc0ad | 5526 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5527 | |
1a20ff27 | 5528 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5529 | { |
758b2cdc | 5530 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5531 | return 1; |
5532 | ||
5533 | /* Following flags need at least 2 groups */ | |
5534 | if (sd->flags & (SD_LOAD_BALANCE | | |
5535 | SD_BALANCE_NEWIDLE | | |
5536 | SD_BALANCE_FORK | | |
89c4710e | 5537 | SD_BALANCE_EXEC | |
5d4dfddd | 5538 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5539 | SD_SHARE_PKG_RESOURCES | |
5540 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5541 | if (sd->groups != sd->groups->next) |
5542 | return 0; | |
5543 | } | |
5544 | ||
5545 | /* Following flags don't use groups */ | |
c88d5910 | 5546 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5547 | return 0; |
5548 | ||
5549 | return 1; | |
5550 | } | |
5551 | ||
48f24c4d IM |
5552 | static int |
5553 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5554 | { |
5555 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5556 | ||
5557 | if (sd_degenerate(parent)) | |
5558 | return 1; | |
5559 | ||
758b2cdc | 5560 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5561 | return 0; |
5562 | ||
245af2c7 SS |
5563 | /* Flags needing groups don't count if only 1 group in parent */ |
5564 | if (parent->groups == parent->groups->next) { | |
5565 | pflags &= ~(SD_LOAD_BALANCE | | |
5566 | SD_BALANCE_NEWIDLE | | |
5567 | SD_BALANCE_FORK | | |
89c4710e | 5568 | SD_BALANCE_EXEC | |
5d4dfddd | 5569 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5570 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5571 | SD_PREFER_SIBLING | |
5572 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5573 | if (nr_node_ids == 1) |
5574 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5575 | } |
5576 | if (~cflags & pflags) | |
5577 | return 0; | |
5578 | ||
5579 | return 1; | |
5580 | } | |
5581 | ||
dce840a0 | 5582 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5583 | { |
dce840a0 | 5584 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5585 | |
68e74568 | 5586 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5587 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5588 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5589 | free_cpumask_var(rd->rto_mask); |
5590 | free_cpumask_var(rd->online); | |
5591 | free_cpumask_var(rd->span); | |
5592 | kfree(rd); | |
5593 | } | |
5594 | ||
57d885fe GH |
5595 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5596 | { | |
a0490fa3 | 5597 | struct root_domain *old_rd = NULL; |
57d885fe | 5598 | unsigned long flags; |
57d885fe | 5599 | |
05fa785c | 5600 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5601 | |
5602 | if (rq->rd) { | |
a0490fa3 | 5603 | old_rd = rq->rd; |
57d885fe | 5604 | |
c6c4927b | 5605 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5606 | set_rq_offline(rq); |
57d885fe | 5607 | |
c6c4927b | 5608 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5609 | |
a0490fa3 | 5610 | /* |
0515973f | 5611 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5612 | * set old_rd to NULL to skip the freeing later |
5613 | * in this function: | |
5614 | */ | |
5615 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5616 | old_rd = NULL; | |
57d885fe GH |
5617 | } |
5618 | ||
5619 | atomic_inc(&rd->refcount); | |
5620 | rq->rd = rd; | |
5621 | ||
c6c4927b | 5622 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5623 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5624 | set_rq_online(rq); |
57d885fe | 5625 | |
05fa785c | 5626 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5627 | |
5628 | if (old_rd) | |
dce840a0 | 5629 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5630 | } |
5631 | ||
68c38fc3 | 5632 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5633 | { |
5634 | memset(rd, 0, sizeof(*rd)); | |
5635 | ||
68c38fc3 | 5636 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5637 | goto out; |
68c38fc3 | 5638 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5639 | goto free_span; |
1baca4ce | 5640 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5641 | goto free_online; |
1baca4ce JL |
5642 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5643 | goto free_dlo_mask; | |
6e0534f2 | 5644 | |
332ac17e | 5645 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5646 | if (cpudl_init(&rd->cpudl) != 0) |
5647 | goto free_dlo_mask; | |
332ac17e | 5648 | |
68c38fc3 | 5649 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5650 | goto free_rto_mask; |
c6c4927b | 5651 | return 0; |
6e0534f2 | 5652 | |
68e74568 RR |
5653 | free_rto_mask: |
5654 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5655 | free_dlo_mask: |
5656 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5657 | free_online: |
5658 | free_cpumask_var(rd->online); | |
5659 | free_span: | |
5660 | free_cpumask_var(rd->span); | |
0c910d28 | 5661 | out: |
c6c4927b | 5662 | return -ENOMEM; |
57d885fe GH |
5663 | } |
5664 | ||
029632fb PZ |
5665 | /* |
5666 | * By default the system creates a single root-domain with all cpus as | |
5667 | * members (mimicking the global state we have today). | |
5668 | */ | |
5669 | struct root_domain def_root_domain; | |
5670 | ||
57d885fe GH |
5671 | static void init_defrootdomain(void) |
5672 | { | |
68c38fc3 | 5673 | init_rootdomain(&def_root_domain); |
c6c4927b | 5674 | |
57d885fe GH |
5675 | atomic_set(&def_root_domain.refcount, 1); |
5676 | } | |
5677 | ||
dc938520 | 5678 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5679 | { |
5680 | struct root_domain *rd; | |
5681 | ||
5682 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5683 | if (!rd) | |
5684 | return NULL; | |
5685 | ||
68c38fc3 | 5686 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5687 | kfree(rd); |
5688 | return NULL; | |
5689 | } | |
57d885fe GH |
5690 | |
5691 | return rd; | |
5692 | } | |
5693 | ||
63b2ca30 | 5694 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5695 | { |
5696 | struct sched_group *tmp, *first; | |
5697 | ||
5698 | if (!sg) | |
5699 | return; | |
5700 | ||
5701 | first = sg; | |
5702 | do { | |
5703 | tmp = sg->next; | |
5704 | ||
63b2ca30 NP |
5705 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5706 | kfree(sg->sgc); | |
e3589f6c PZ |
5707 | |
5708 | kfree(sg); | |
5709 | sg = tmp; | |
5710 | } while (sg != first); | |
5711 | } | |
5712 | ||
dce840a0 PZ |
5713 | static void free_sched_domain(struct rcu_head *rcu) |
5714 | { | |
5715 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5716 | |
5717 | /* | |
5718 | * If its an overlapping domain it has private groups, iterate and | |
5719 | * nuke them all. | |
5720 | */ | |
5721 | if (sd->flags & SD_OVERLAP) { | |
5722 | free_sched_groups(sd->groups, 1); | |
5723 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5724 | kfree(sd->groups->sgc); |
dce840a0 | 5725 | kfree(sd->groups); |
9c3f75cb | 5726 | } |
dce840a0 PZ |
5727 | kfree(sd); |
5728 | } | |
5729 | ||
5730 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5731 | { | |
5732 | call_rcu(&sd->rcu, free_sched_domain); | |
5733 | } | |
5734 | ||
5735 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5736 | { | |
5737 | for (; sd; sd = sd->parent) | |
5738 | destroy_sched_domain(sd, cpu); | |
5739 | } | |
5740 | ||
518cd623 PZ |
5741 | /* |
5742 | * Keep a special pointer to the highest sched_domain that has | |
5743 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5744 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5745 | * | |
5746 | * Also keep a unique ID per domain (we use the first cpu number in | |
5747 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5748 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5749 | */ |
5750 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5751 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5752 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5753 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5754 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5755 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5756 | |
5757 | static void update_top_cache_domain(int cpu) | |
5758 | { | |
5759 | struct sched_domain *sd; | |
5d4cf996 | 5760 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5761 | int id = cpu; |
7d9ffa89 | 5762 | int size = 1; |
518cd623 PZ |
5763 | |
5764 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5765 | if (sd) { |
518cd623 | 5766 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5767 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5768 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5769 | } |
5d4cf996 | 5770 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5771 | |
5772 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5773 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5774 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5775 | |
5776 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5777 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5778 | |
5779 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5780 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5781 | } |
5782 | ||
1da177e4 | 5783 | /* |
0eab9146 | 5784 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5785 | * hold the hotplug lock. |
5786 | */ | |
0eab9146 IM |
5787 | static void |
5788 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5789 | { |
70b97a7f | 5790 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5791 | struct sched_domain *tmp; |
5792 | ||
5793 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5794 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5795 | struct sched_domain *parent = tmp->parent; |
5796 | if (!parent) | |
5797 | break; | |
f29c9b1c | 5798 | |
1a848870 | 5799 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5800 | tmp->parent = parent->parent; |
1a848870 SS |
5801 | if (parent->parent) |
5802 | parent->parent->child = tmp; | |
10866e62 PZ |
5803 | /* |
5804 | * Transfer SD_PREFER_SIBLING down in case of a | |
5805 | * degenerate parent; the spans match for this | |
5806 | * so the property transfers. | |
5807 | */ | |
5808 | if (parent->flags & SD_PREFER_SIBLING) | |
5809 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5810 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5811 | } else |
5812 | tmp = tmp->parent; | |
245af2c7 SS |
5813 | } |
5814 | ||
1a848870 | 5815 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5816 | tmp = sd; |
245af2c7 | 5817 | sd = sd->parent; |
dce840a0 | 5818 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5819 | if (sd) |
5820 | sd->child = NULL; | |
5821 | } | |
1da177e4 | 5822 | |
4cb98839 | 5823 | sched_domain_debug(sd, cpu); |
1da177e4 | 5824 | |
57d885fe | 5825 | rq_attach_root(rq, rd); |
dce840a0 | 5826 | tmp = rq->sd; |
674311d5 | 5827 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5828 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5829 | |
5830 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5831 | } |
5832 | ||
5833 | /* cpus with isolated domains */ | |
dcc30a35 | 5834 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5835 | |
5836 | /* Setup the mask of cpus configured for isolated domains */ | |
5837 | static int __init isolated_cpu_setup(char *str) | |
5838 | { | |
bdddd296 | 5839 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5840 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5841 | return 1; |
5842 | } | |
5843 | ||
8927f494 | 5844 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5845 | |
49a02c51 | 5846 | struct s_data { |
21d42ccf | 5847 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5848 | struct root_domain *rd; |
5849 | }; | |
5850 | ||
2109b99e | 5851 | enum s_alloc { |
2109b99e | 5852 | sa_rootdomain, |
21d42ccf | 5853 | sa_sd, |
dce840a0 | 5854 | sa_sd_storage, |
2109b99e AH |
5855 | sa_none, |
5856 | }; | |
5857 | ||
c1174876 PZ |
5858 | /* |
5859 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5860 | * domain traversal. | |
5861 | * | |
5862 | * Asymmetric node setups can result in situations where the domain tree is of | |
5863 | * unequal depth, make sure to skip domains that already cover the entire | |
5864 | * range. | |
5865 | * | |
5866 | * In that case build_sched_domains() will have terminated the iteration early | |
5867 | * and our sibling sd spans will be empty. Domains should always include the | |
5868 | * cpu they're built on, so check that. | |
5869 | * | |
5870 | */ | |
5871 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5872 | { | |
5873 | const struct cpumask *span = sched_domain_span(sd); | |
5874 | struct sd_data *sdd = sd->private; | |
5875 | struct sched_domain *sibling; | |
5876 | int i; | |
5877 | ||
5878 | for_each_cpu(i, span) { | |
5879 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5880 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5881 | continue; | |
5882 | ||
5883 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5884 | } | |
5885 | } | |
5886 | ||
5887 | /* | |
5888 | * Return the canonical balance cpu for this group, this is the first cpu | |
5889 | * of this group that's also in the iteration mask. | |
5890 | */ | |
5891 | int group_balance_cpu(struct sched_group *sg) | |
5892 | { | |
5893 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5894 | } | |
5895 | ||
e3589f6c PZ |
5896 | static int |
5897 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5898 | { | |
5899 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5900 | const struct cpumask *span = sched_domain_span(sd); | |
5901 | struct cpumask *covered = sched_domains_tmpmask; | |
5902 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5903 | struct sched_domain *sibling; |
e3589f6c PZ |
5904 | int i; |
5905 | ||
5906 | cpumask_clear(covered); | |
5907 | ||
5908 | for_each_cpu(i, span) { | |
5909 | struct cpumask *sg_span; | |
5910 | ||
5911 | if (cpumask_test_cpu(i, covered)) | |
5912 | continue; | |
5913 | ||
aaecac4a | 5914 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5915 | |
5916 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5917 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5918 | continue; |
5919 | ||
e3589f6c | 5920 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5921 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5922 | |
5923 | if (!sg) | |
5924 | goto fail; | |
5925 | ||
5926 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5927 | if (sibling->child) |
5928 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5929 | else | |
e3589f6c PZ |
5930 | cpumask_set_cpu(i, sg_span); |
5931 | ||
5932 | cpumask_or(covered, covered, sg_span); | |
5933 | ||
63b2ca30 NP |
5934 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5935 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5936 | build_group_mask(sd, sg); |
5937 | ||
c3decf0d | 5938 | /* |
63b2ca30 | 5939 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5940 | * domains and no possible iteration will get us here, we won't |
5941 | * die on a /0 trap. | |
5942 | */ | |
ca8ce3d0 | 5943 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5944 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5945 | |
c1174876 PZ |
5946 | /* |
5947 | * Make sure the first group of this domain contains the | |
5948 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5949 | * breaks. See update_sg_lb_stats(). | |
5950 | */ | |
74a5ce20 | 5951 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5952 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5953 | groups = sg; |
5954 | ||
5955 | if (!first) | |
5956 | first = sg; | |
5957 | if (last) | |
5958 | last->next = sg; | |
5959 | last = sg; | |
5960 | last->next = first; | |
5961 | } | |
5962 | sd->groups = groups; | |
5963 | ||
5964 | return 0; | |
5965 | ||
5966 | fail: | |
5967 | free_sched_groups(first, 0); | |
5968 | ||
5969 | return -ENOMEM; | |
5970 | } | |
5971 | ||
dce840a0 | 5972 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5973 | { |
dce840a0 PZ |
5974 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5975 | struct sched_domain *child = sd->child; | |
1da177e4 | 5976 | |
dce840a0 PZ |
5977 | if (child) |
5978 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5979 | |
9c3f75cb | 5980 | if (sg) { |
dce840a0 | 5981 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5982 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5983 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5984 | } |
dce840a0 PZ |
5985 | |
5986 | return cpu; | |
1e9f28fa | 5987 | } |
1e9f28fa | 5988 | |
01a08546 | 5989 | /* |
dce840a0 PZ |
5990 | * build_sched_groups will build a circular linked list of the groups |
5991 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5992 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5993 | * |
5994 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5995 | */ |
e3589f6c PZ |
5996 | static int |
5997 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5998 | { |
dce840a0 PZ |
5999 | struct sched_group *first = NULL, *last = NULL; |
6000 | struct sd_data *sdd = sd->private; | |
6001 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6002 | struct cpumask *covered; |
dce840a0 | 6003 | int i; |
9c1cfda2 | 6004 | |
e3589f6c PZ |
6005 | get_group(cpu, sdd, &sd->groups); |
6006 | atomic_inc(&sd->groups->ref); | |
6007 | ||
0936629f | 6008 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6009 | return 0; |
6010 | ||
f96225fd PZ |
6011 | lockdep_assert_held(&sched_domains_mutex); |
6012 | covered = sched_domains_tmpmask; | |
6013 | ||
dce840a0 | 6014 | cpumask_clear(covered); |
6711cab4 | 6015 | |
dce840a0 PZ |
6016 | for_each_cpu(i, span) { |
6017 | struct sched_group *sg; | |
cd08e923 | 6018 | int group, j; |
6711cab4 | 6019 | |
dce840a0 PZ |
6020 | if (cpumask_test_cpu(i, covered)) |
6021 | continue; | |
6711cab4 | 6022 | |
cd08e923 | 6023 | group = get_group(i, sdd, &sg); |
c1174876 | 6024 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6025 | |
dce840a0 PZ |
6026 | for_each_cpu(j, span) { |
6027 | if (get_group(j, sdd, NULL) != group) | |
6028 | continue; | |
0601a88d | 6029 | |
dce840a0 PZ |
6030 | cpumask_set_cpu(j, covered); |
6031 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6032 | } | |
0601a88d | 6033 | |
dce840a0 PZ |
6034 | if (!first) |
6035 | first = sg; | |
6036 | if (last) | |
6037 | last->next = sg; | |
6038 | last = sg; | |
6039 | } | |
6040 | last->next = first; | |
e3589f6c PZ |
6041 | |
6042 | return 0; | |
0601a88d | 6043 | } |
51888ca2 | 6044 | |
89c4710e | 6045 | /* |
63b2ca30 | 6046 | * Initialize sched groups cpu_capacity. |
89c4710e | 6047 | * |
63b2ca30 | 6048 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6049 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6050 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6051 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6052 | * group having more cpu_capacity will pickup more load compared to the | |
6053 | * group having less cpu_capacity. | |
89c4710e | 6054 | */ |
63b2ca30 | 6055 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6056 | { |
e3589f6c | 6057 | struct sched_group *sg = sd->groups; |
89c4710e | 6058 | |
94c95ba6 | 6059 | WARN_ON(!sg); |
e3589f6c PZ |
6060 | |
6061 | do { | |
6062 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6063 | sg = sg->next; | |
6064 | } while (sg != sd->groups); | |
89c4710e | 6065 | |
c1174876 | 6066 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6067 | return; |
aae6d3dd | 6068 | |
63b2ca30 NP |
6069 | update_group_capacity(sd, cpu); |
6070 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6071 | } |
6072 | ||
7c16ec58 MT |
6073 | /* |
6074 | * Initializers for schedule domains | |
6075 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6076 | */ | |
6077 | ||
1d3504fc | 6078 | static int default_relax_domain_level = -1; |
60495e77 | 6079 | int sched_domain_level_max; |
1d3504fc HS |
6080 | |
6081 | static int __init setup_relax_domain_level(char *str) | |
6082 | { | |
a841f8ce DS |
6083 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6084 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6085 | |
1d3504fc HS |
6086 | return 1; |
6087 | } | |
6088 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6089 | ||
6090 | static void set_domain_attribute(struct sched_domain *sd, | |
6091 | struct sched_domain_attr *attr) | |
6092 | { | |
6093 | int request; | |
6094 | ||
6095 | if (!attr || attr->relax_domain_level < 0) { | |
6096 | if (default_relax_domain_level < 0) | |
6097 | return; | |
6098 | else | |
6099 | request = default_relax_domain_level; | |
6100 | } else | |
6101 | request = attr->relax_domain_level; | |
6102 | if (request < sd->level) { | |
6103 | /* turn off idle balance on this domain */ | |
c88d5910 | 6104 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6105 | } else { |
6106 | /* turn on idle balance on this domain */ | |
c88d5910 | 6107 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6108 | } |
6109 | } | |
6110 | ||
54ab4ff4 PZ |
6111 | static void __sdt_free(const struct cpumask *cpu_map); |
6112 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6113 | ||
2109b99e AH |
6114 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6115 | const struct cpumask *cpu_map) | |
6116 | { | |
6117 | switch (what) { | |
2109b99e | 6118 | case sa_rootdomain: |
822ff793 PZ |
6119 | if (!atomic_read(&d->rd->refcount)) |
6120 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6121 | case sa_sd: |
6122 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6123 | case sa_sd_storage: |
54ab4ff4 | 6124 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6125 | case sa_none: |
6126 | break; | |
6127 | } | |
6128 | } | |
3404c8d9 | 6129 | |
2109b99e AH |
6130 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6131 | const struct cpumask *cpu_map) | |
6132 | { | |
dce840a0 PZ |
6133 | memset(d, 0, sizeof(*d)); |
6134 | ||
54ab4ff4 PZ |
6135 | if (__sdt_alloc(cpu_map)) |
6136 | return sa_sd_storage; | |
dce840a0 PZ |
6137 | d->sd = alloc_percpu(struct sched_domain *); |
6138 | if (!d->sd) | |
6139 | return sa_sd_storage; | |
2109b99e | 6140 | d->rd = alloc_rootdomain(); |
dce840a0 | 6141 | if (!d->rd) |
21d42ccf | 6142 | return sa_sd; |
2109b99e AH |
6143 | return sa_rootdomain; |
6144 | } | |
57d885fe | 6145 | |
dce840a0 PZ |
6146 | /* |
6147 | * NULL the sd_data elements we've used to build the sched_domain and | |
6148 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6149 | * will not free the data we're using. | |
6150 | */ | |
6151 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6152 | { | |
6153 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6154 | |
6155 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6156 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6157 | ||
e3589f6c | 6158 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6159 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6160 | |
63b2ca30 NP |
6161 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6162 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6163 | } |
6164 | ||
cb83b629 | 6165 | #ifdef CONFIG_NUMA |
cb83b629 | 6166 | static int sched_domains_numa_levels; |
e3fe70b1 | 6167 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6168 | static int *sched_domains_numa_distance; |
9942f79b | 6169 | int sched_max_numa_distance; |
cb83b629 PZ |
6170 | static struct cpumask ***sched_domains_numa_masks; |
6171 | static int sched_domains_curr_level; | |
143e1e28 | 6172 | #endif |
cb83b629 | 6173 | |
143e1e28 VG |
6174 | /* |
6175 | * SD_flags allowed in topology descriptions. | |
6176 | * | |
5d4dfddd | 6177 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6178 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6179 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6180 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6181 | * |
6182 | * Odd one out: | |
6183 | * SD_ASYM_PACKING - describes SMT quirks | |
6184 | */ | |
6185 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6186 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6187 | SD_SHARE_PKG_RESOURCES | \ |
6188 | SD_NUMA | \ | |
d77b3ed5 VG |
6189 | SD_ASYM_PACKING | \ |
6190 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6191 | |
6192 | static struct sched_domain * | |
143e1e28 | 6193 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6194 | { |
6195 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6196 | int sd_weight, sd_flags = 0; |
6197 | ||
6198 | #ifdef CONFIG_NUMA | |
6199 | /* | |
6200 | * Ugly hack to pass state to sd_numa_mask()... | |
6201 | */ | |
6202 | sched_domains_curr_level = tl->numa_level; | |
6203 | #endif | |
6204 | ||
6205 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6206 | ||
6207 | if (tl->sd_flags) | |
6208 | sd_flags = (*tl->sd_flags)(); | |
6209 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6210 | "wrong sd_flags in topology description\n")) | |
6211 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6212 | |
6213 | *sd = (struct sched_domain){ | |
6214 | .min_interval = sd_weight, | |
6215 | .max_interval = 2*sd_weight, | |
6216 | .busy_factor = 32, | |
870a0bb5 | 6217 | .imbalance_pct = 125, |
143e1e28 VG |
6218 | |
6219 | .cache_nice_tries = 0, | |
6220 | .busy_idx = 0, | |
6221 | .idle_idx = 0, | |
cb83b629 PZ |
6222 | .newidle_idx = 0, |
6223 | .wake_idx = 0, | |
6224 | .forkexec_idx = 0, | |
6225 | ||
6226 | .flags = 1*SD_LOAD_BALANCE | |
6227 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6228 | | 1*SD_BALANCE_EXEC |
6229 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6230 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6231 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6232 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6233 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6234 | | 0*SD_SERIALIZE |
cb83b629 | 6235 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6236 | | 0*SD_NUMA |
6237 | | sd_flags | |
cb83b629 | 6238 | , |
143e1e28 | 6239 | |
cb83b629 PZ |
6240 | .last_balance = jiffies, |
6241 | .balance_interval = sd_weight, | |
143e1e28 | 6242 | .smt_gain = 0, |
2b4cfe64 JL |
6243 | .max_newidle_lb_cost = 0, |
6244 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6245 | #ifdef CONFIG_SCHED_DEBUG |
6246 | .name = tl->name, | |
6247 | #endif | |
cb83b629 | 6248 | }; |
cb83b629 PZ |
6249 | |
6250 | /* | |
143e1e28 | 6251 | * Convert topological properties into behaviour. |
cb83b629 | 6252 | */ |
143e1e28 | 6253 | |
5d4dfddd | 6254 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6255 | sd->imbalance_pct = 110; |
6256 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6257 | |
6258 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6259 | sd->imbalance_pct = 117; | |
6260 | sd->cache_nice_tries = 1; | |
6261 | sd->busy_idx = 2; | |
6262 | ||
6263 | #ifdef CONFIG_NUMA | |
6264 | } else if (sd->flags & SD_NUMA) { | |
6265 | sd->cache_nice_tries = 2; | |
6266 | sd->busy_idx = 3; | |
6267 | sd->idle_idx = 2; | |
6268 | ||
6269 | sd->flags |= SD_SERIALIZE; | |
6270 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6271 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6272 | SD_BALANCE_FORK | | |
6273 | SD_WAKE_AFFINE); | |
6274 | } | |
6275 | ||
6276 | #endif | |
6277 | } else { | |
6278 | sd->flags |= SD_PREFER_SIBLING; | |
6279 | sd->cache_nice_tries = 1; | |
6280 | sd->busy_idx = 2; | |
6281 | sd->idle_idx = 1; | |
6282 | } | |
6283 | ||
6284 | sd->private = &tl->data; | |
cb83b629 PZ |
6285 | |
6286 | return sd; | |
6287 | } | |
6288 | ||
143e1e28 VG |
6289 | /* |
6290 | * Topology list, bottom-up. | |
6291 | */ | |
6292 | static struct sched_domain_topology_level default_topology[] = { | |
6293 | #ifdef CONFIG_SCHED_SMT | |
6294 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6295 | #endif | |
6296 | #ifdef CONFIG_SCHED_MC | |
6297 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6298 | #endif |
6299 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6300 | { NULL, }, | |
6301 | }; | |
6302 | ||
6303 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6304 | ||
6305 | #define for_each_sd_topology(tl) \ | |
6306 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6307 | ||
6308 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6309 | { | |
6310 | sched_domain_topology = tl; | |
6311 | } | |
6312 | ||
6313 | #ifdef CONFIG_NUMA | |
6314 | ||
cb83b629 PZ |
6315 | static const struct cpumask *sd_numa_mask(int cpu) |
6316 | { | |
6317 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6318 | } | |
6319 | ||
d039ac60 PZ |
6320 | static void sched_numa_warn(const char *str) |
6321 | { | |
6322 | static int done = false; | |
6323 | int i,j; | |
6324 | ||
6325 | if (done) | |
6326 | return; | |
6327 | ||
6328 | done = true; | |
6329 | ||
6330 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6331 | ||
6332 | for (i = 0; i < nr_node_ids; i++) { | |
6333 | printk(KERN_WARNING " "); | |
6334 | for (j = 0; j < nr_node_ids; j++) | |
6335 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6336 | printk(KERN_CONT "\n"); | |
6337 | } | |
6338 | printk(KERN_WARNING "\n"); | |
6339 | } | |
6340 | ||
9942f79b | 6341 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6342 | { |
6343 | int i; | |
6344 | ||
6345 | if (distance == node_distance(0, 0)) | |
6346 | return true; | |
6347 | ||
6348 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6349 | if (sched_domains_numa_distance[i] == distance) | |
6350 | return true; | |
6351 | } | |
6352 | ||
6353 | return false; | |
6354 | } | |
6355 | ||
e3fe70b1 RR |
6356 | /* |
6357 | * A system can have three types of NUMA topology: | |
6358 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6359 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6360 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6361 | * | |
6362 | * The difference between a glueless mesh topology and a backplane | |
6363 | * topology lies in whether communication between not directly | |
6364 | * connected nodes goes through intermediary nodes (where programs | |
6365 | * could run), or through backplane controllers. This affects | |
6366 | * placement of programs. | |
6367 | * | |
6368 | * The type of topology can be discerned with the following tests: | |
6369 | * - If the maximum distance between any nodes is 1 hop, the system | |
6370 | * is directly connected. | |
6371 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6372 | * there is an intermediary node C, which is < N hops away from both | |
6373 | * nodes A and B, the system is a glueless mesh. | |
6374 | */ | |
6375 | static void init_numa_topology_type(void) | |
6376 | { | |
6377 | int a, b, c, n; | |
6378 | ||
6379 | n = sched_max_numa_distance; | |
6380 | ||
6381 | if (n <= 1) | |
6382 | sched_numa_topology_type = NUMA_DIRECT; | |
6383 | ||
6384 | for_each_online_node(a) { | |
6385 | for_each_online_node(b) { | |
6386 | /* Find two nodes furthest removed from each other. */ | |
6387 | if (node_distance(a, b) < n) | |
6388 | continue; | |
6389 | ||
6390 | /* Is there an intermediary node between a and b? */ | |
6391 | for_each_online_node(c) { | |
6392 | if (node_distance(a, c) < n && | |
6393 | node_distance(b, c) < n) { | |
6394 | sched_numa_topology_type = | |
6395 | NUMA_GLUELESS_MESH; | |
6396 | return; | |
6397 | } | |
6398 | } | |
6399 | ||
6400 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6401 | return; | |
6402 | } | |
6403 | } | |
6404 | } | |
6405 | ||
cb83b629 PZ |
6406 | static void sched_init_numa(void) |
6407 | { | |
6408 | int next_distance, curr_distance = node_distance(0, 0); | |
6409 | struct sched_domain_topology_level *tl; | |
6410 | int level = 0; | |
6411 | int i, j, k; | |
6412 | ||
cb83b629 PZ |
6413 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6414 | if (!sched_domains_numa_distance) | |
6415 | return; | |
6416 | ||
6417 | /* | |
6418 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6419 | * unique distances in the node_distance() table. | |
6420 | * | |
6421 | * Assumes node_distance(0,j) includes all distances in | |
6422 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6423 | */ |
6424 | next_distance = curr_distance; | |
6425 | for (i = 0; i < nr_node_ids; i++) { | |
6426 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6427 | for (k = 0; k < nr_node_ids; k++) { |
6428 | int distance = node_distance(i, k); | |
6429 | ||
6430 | if (distance > curr_distance && | |
6431 | (distance < next_distance || | |
6432 | next_distance == curr_distance)) | |
6433 | next_distance = distance; | |
6434 | ||
6435 | /* | |
6436 | * While not a strong assumption it would be nice to know | |
6437 | * about cases where if node A is connected to B, B is not | |
6438 | * equally connected to A. | |
6439 | */ | |
6440 | if (sched_debug() && node_distance(k, i) != distance) | |
6441 | sched_numa_warn("Node-distance not symmetric"); | |
6442 | ||
6443 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6444 | sched_numa_warn("Node-0 not representative"); | |
6445 | } | |
6446 | if (next_distance != curr_distance) { | |
6447 | sched_domains_numa_distance[level++] = next_distance; | |
6448 | sched_domains_numa_levels = level; | |
6449 | curr_distance = next_distance; | |
6450 | } else break; | |
cb83b629 | 6451 | } |
d039ac60 PZ |
6452 | |
6453 | /* | |
6454 | * In case of sched_debug() we verify the above assumption. | |
6455 | */ | |
6456 | if (!sched_debug()) | |
6457 | break; | |
cb83b629 | 6458 | } |
c123588b AR |
6459 | |
6460 | if (!level) | |
6461 | return; | |
6462 | ||
cb83b629 PZ |
6463 | /* |
6464 | * 'level' contains the number of unique distances, excluding the | |
6465 | * identity distance node_distance(i,i). | |
6466 | * | |
28b4a521 | 6467 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6468 | * numbers. |
6469 | */ | |
6470 | ||
5f7865f3 TC |
6471 | /* |
6472 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6473 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6474 | * the array will contain less then 'level' members. This could be | |
6475 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6476 | * in other functions. | |
6477 | * | |
6478 | * We reset it to 'level' at the end of this function. | |
6479 | */ | |
6480 | sched_domains_numa_levels = 0; | |
6481 | ||
cb83b629 PZ |
6482 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6483 | if (!sched_domains_numa_masks) | |
6484 | return; | |
6485 | ||
6486 | /* | |
6487 | * Now for each level, construct a mask per node which contains all | |
6488 | * cpus of nodes that are that many hops away from us. | |
6489 | */ | |
6490 | for (i = 0; i < level; i++) { | |
6491 | sched_domains_numa_masks[i] = | |
6492 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6493 | if (!sched_domains_numa_masks[i]) | |
6494 | return; | |
6495 | ||
6496 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6497 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6498 | if (!mask) |
6499 | return; | |
6500 | ||
6501 | sched_domains_numa_masks[i][j] = mask; | |
6502 | ||
6503 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6504 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6505 | continue; |
6506 | ||
6507 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6508 | } | |
6509 | } | |
6510 | } | |
6511 | ||
143e1e28 VG |
6512 | /* Compute default topology size */ |
6513 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6514 | ||
c515db8c | 6515 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6516 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6517 | if (!tl) | |
6518 | return; | |
6519 | ||
6520 | /* | |
6521 | * Copy the default topology bits.. | |
6522 | */ | |
143e1e28 VG |
6523 | for (i = 0; sched_domain_topology[i].mask; i++) |
6524 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6525 | |
6526 | /* | |
6527 | * .. and append 'j' levels of NUMA goodness. | |
6528 | */ | |
6529 | for (j = 0; j < level; i++, j++) { | |
6530 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6531 | .mask = sd_numa_mask, |
143e1e28 | 6532 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6533 | .flags = SDTL_OVERLAP, |
6534 | .numa_level = j, | |
143e1e28 | 6535 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6536 | }; |
6537 | } | |
6538 | ||
6539 | sched_domain_topology = tl; | |
5f7865f3 TC |
6540 | |
6541 | sched_domains_numa_levels = level; | |
9942f79b | 6542 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6543 | |
6544 | init_numa_topology_type(); | |
cb83b629 | 6545 | } |
301a5cba TC |
6546 | |
6547 | static void sched_domains_numa_masks_set(int cpu) | |
6548 | { | |
6549 | int i, j; | |
6550 | int node = cpu_to_node(cpu); | |
6551 | ||
6552 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6553 | for (j = 0; j < nr_node_ids; j++) { | |
6554 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6555 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6556 | } | |
6557 | } | |
6558 | } | |
6559 | ||
6560 | static void sched_domains_numa_masks_clear(int cpu) | |
6561 | { | |
6562 | int i, j; | |
6563 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6564 | for (j = 0; j < nr_node_ids; j++) | |
6565 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6566 | } | |
6567 | } | |
6568 | ||
6569 | /* | |
6570 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6571 | * are onlined. | |
6572 | */ | |
6573 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6574 | unsigned long action, | |
6575 | void *hcpu) | |
6576 | { | |
6577 | int cpu = (long)hcpu; | |
6578 | ||
6579 | switch (action & ~CPU_TASKS_FROZEN) { | |
6580 | case CPU_ONLINE: | |
6581 | sched_domains_numa_masks_set(cpu); | |
6582 | break; | |
6583 | ||
6584 | case CPU_DEAD: | |
6585 | sched_domains_numa_masks_clear(cpu); | |
6586 | break; | |
6587 | ||
6588 | default: | |
6589 | return NOTIFY_DONE; | |
6590 | } | |
6591 | ||
6592 | return NOTIFY_OK; | |
cb83b629 PZ |
6593 | } |
6594 | #else | |
6595 | static inline void sched_init_numa(void) | |
6596 | { | |
6597 | } | |
301a5cba TC |
6598 | |
6599 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6600 | unsigned long action, | |
6601 | void *hcpu) | |
6602 | { | |
6603 | return 0; | |
6604 | } | |
cb83b629 PZ |
6605 | #endif /* CONFIG_NUMA */ |
6606 | ||
54ab4ff4 PZ |
6607 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6608 | { | |
6609 | struct sched_domain_topology_level *tl; | |
6610 | int j; | |
6611 | ||
27723a68 | 6612 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6613 | struct sd_data *sdd = &tl->data; |
6614 | ||
6615 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6616 | if (!sdd->sd) | |
6617 | return -ENOMEM; | |
6618 | ||
6619 | sdd->sg = alloc_percpu(struct sched_group *); | |
6620 | if (!sdd->sg) | |
6621 | return -ENOMEM; | |
6622 | ||
63b2ca30 NP |
6623 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6624 | if (!sdd->sgc) | |
9c3f75cb PZ |
6625 | return -ENOMEM; |
6626 | ||
54ab4ff4 PZ |
6627 | for_each_cpu(j, cpu_map) { |
6628 | struct sched_domain *sd; | |
6629 | struct sched_group *sg; | |
63b2ca30 | 6630 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6631 | |
6632 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6633 | GFP_KERNEL, cpu_to_node(j)); | |
6634 | if (!sd) | |
6635 | return -ENOMEM; | |
6636 | ||
6637 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6638 | ||
6639 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6640 | GFP_KERNEL, cpu_to_node(j)); | |
6641 | if (!sg) | |
6642 | return -ENOMEM; | |
6643 | ||
30b4e9eb IM |
6644 | sg->next = sg; |
6645 | ||
54ab4ff4 | 6646 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6647 | |
63b2ca30 | 6648 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6649 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6650 | if (!sgc) |
9c3f75cb PZ |
6651 | return -ENOMEM; |
6652 | ||
63b2ca30 | 6653 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6654 | } |
6655 | } | |
6656 | ||
6657 | return 0; | |
6658 | } | |
6659 | ||
6660 | static void __sdt_free(const struct cpumask *cpu_map) | |
6661 | { | |
6662 | struct sched_domain_topology_level *tl; | |
6663 | int j; | |
6664 | ||
27723a68 | 6665 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6666 | struct sd_data *sdd = &tl->data; |
6667 | ||
6668 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6669 | struct sched_domain *sd; |
6670 | ||
6671 | if (sdd->sd) { | |
6672 | sd = *per_cpu_ptr(sdd->sd, j); | |
6673 | if (sd && (sd->flags & SD_OVERLAP)) | |
6674 | free_sched_groups(sd->groups, 0); | |
6675 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6676 | } | |
6677 | ||
6678 | if (sdd->sg) | |
6679 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6680 | if (sdd->sgc) |
6681 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6682 | } |
6683 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6684 | sdd->sd = NULL; |
54ab4ff4 | 6685 | free_percpu(sdd->sg); |
fb2cf2c6 | 6686 | sdd->sg = NULL; |
63b2ca30 NP |
6687 | free_percpu(sdd->sgc); |
6688 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6689 | } |
6690 | } | |
6691 | ||
2c402dc3 | 6692 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6693 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6694 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6695 | { |
143e1e28 | 6696 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6697 | if (!sd) |
d069b916 | 6698 | return child; |
2c402dc3 | 6699 | |
2c402dc3 | 6700 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6701 | if (child) { |
6702 | sd->level = child->level + 1; | |
6703 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6704 | child->parent = sd; |
c75e0128 | 6705 | sd->child = child; |
6ae72dff PZ |
6706 | |
6707 | if (!cpumask_subset(sched_domain_span(child), | |
6708 | sched_domain_span(sd))) { | |
6709 | pr_err("BUG: arch topology borken\n"); | |
6710 | #ifdef CONFIG_SCHED_DEBUG | |
6711 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6712 | child->name, sd->name); | |
6713 | #endif | |
6714 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6715 | cpumask_or(sched_domain_span(sd), | |
6716 | sched_domain_span(sd), | |
6717 | sched_domain_span(child)); | |
6718 | } | |
6719 | ||
60495e77 | 6720 | } |
a841f8ce | 6721 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6722 | |
6723 | return sd; | |
6724 | } | |
6725 | ||
2109b99e AH |
6726 | /* |
6727 | * Build sched domains for a given set of cpus and attach the sched domains | |
6728 | * to the individual cpus | |
6729 | */ | |
dce840a0 PZ |
6730 | static int build_sched_domains(const struct cpumask *cpu_map, |
6731 | struct sched_domain_attr *attr) | |
2109b99e | 6732 | { |
1c632169 | 6733 | enum s_alloc alloc_state; |
dce840a0 | 6734 | struct sched_domain *sd; |
2109b99e | 6735 | struct s_data d; |
822ff793 | 6736 | int i, ret = -ENOMEM; |
9c1cfda2 | 6737 | |
2109b99e AH |
6738 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6739 | if (alloc_state != sa_rootdomain) | |
6740 | goto error; | |
9c1cfda2 | 6741 | |
dce840a0 | 6742 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6743 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6744 | struct sched_domain_topology_level *tl; |
6745 | ||
3bd65a80 | 6746 | sd = NULL; |
27723a68 | 6747 | for_each_sd_topology(tl) { |
4a850cbe | 6748 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6749 | if (tl == sched_domain_topology) |
6750 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6751 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6752 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6753 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6754 | break; | |
e3589f6c | 6755 | } |
dce840a0 PZ |
6756 | } |
6757 | ||
6758 | /* Build the groups for the domains */ | |
6759 | for_each_cpu(i, cpu_map) { | |
6760 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6761 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6762 | if (sd->flags & SD_OVERLAP) { |
6763 | if (build_overlap_sched_groups(sd, i)) | |
6764 | goto error; | |
6765 | } else { | |
6766 | if (build_sched_groups(sd, i)) | |
6767 | goto error; | |
6768 | } | |
1cf51902 | 6769 | } |
a06dadbe | 6770 | } |
9c1cfda2 | 6771 | |
ced549fa | 6772 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6773 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6774 | if (!cpumask_test_cpu(i, cpu_map)) | |
6775 | continue; | |
9c1cfda2 | 6776 | |
dce840a0 PZ |
6777 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6778 | claim_allocations(i, sd); | |
63b2ca30 | 6779 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6780 | } |
f712c0c7 | 6781 | } |
9c1cfda2 | 6782 | |
1da177e4 | 6783 | /* Attach the domains */ |
dce840a0 | 6784 | rcu_read_lock(); |
abcd083a | 6785 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6786 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6787 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6788 | } |
dce840a0 | 6789 | rcu_read_unlock(); |
51888ca2 | 6790 | |
822ff793 | 6791 | ret = 0; |
51888ca2 | 6792 | error: |
2109b99e | 6793 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6794 | return ret; |
1da177e4 | 6795 | } |
029190c5 | 6796 | |
acc3f5d7 | 6797 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6798 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6799 | static struct sched_domain_attr *dattr_cur; |
6800 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6801 | |
6802 | /* | |
6803 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6804 | * cpumask) fails, then fallback to a single sched domain, |
6805 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6806 | */ |
4212823f | 6807 | static cpumask_var_t fallback_doms; |
029190c5 | 6808 | |
ee79d1bd HC |
6809 | /* |
6810 | * arch_update_cpu_topology lets virtualized architectures update the | |
6811 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6812 | * or 0 if it stayed the same. | |
6813 | */ | |
52f5684c | 6814 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6815 | { |
ee79d1bd | 6816 | return 0; |
22e52b07 HC |
6817 | } |
6818 | ||
acc3f5d7 RR |
6819 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6820 | { | |
6821 | int i; | |
6822 | cpumask_var_t *doms; | |
6823 | ||
6824 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6825 | if (!doms) | |
6826 | return NULL; | |
6827 | for (i = 0; i < ndoms; i++) { | |
6828 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6829 | free_sched_domains(doms, i); | |
6830 | return NULL; | |
6831 | } | |
6832 | } | |
6833 | return doms; | |
6834 | } | |
6835 | ||
6836 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6837 | { | |
6838 | unsigned int i; | |
6839 | for (i = 0; i < ndoms; i++) | |
6840 | free_cpumask_var(doms[i]); | |
6841 | kfree(doms); | |
6842 | } | |
6843 | ||
1a20ff27 | 6844 | /* |
41a2d6cf | 6845 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6846 | * For now this just excludes isolated cpus, but could be used to |
6847 | * exclude other special cases in the future. | |
1a20ff27 | 6848 | */ |
c4a8849a | 6849 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6850 | { |
7378547f MM |
6851 | int err; |
6852 | ||
22e52b07 | 6853 | arch_update_cpu_topology(); |
029190c5 | 6854 | ndoms_cur = 1; |
acc3f5d7 | 6855 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6856 | if (!doms_cur) |
acc3f5d7 RR |
6857 | doms_cur = &fallback_doms; |
6858 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6859 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6860 | register_sched_domain_sysctl(); |
7378547f MM |
6861 | |
6862 | return err; | |
1a20ff27 DG |
6863 | } |
6864 | ||
1a20ff27 DG |
6865 | /* |
6866 | * Detach sched domains from a group of cpus specified in cpu_map | |
6867 | * These cpus will now be attached to the NULL domain | |
6868 | */ | |
96f874e2 | 6869 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6870 | { |
6871 | int i; | |
6872 | ||
dce840a0 | 6873 | rcu_read_lock(); |
abcd083a | 6874 | for_each_cpu(i, cpu_map) |
57d885fe | 6875 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6876 | rcu_read_unlock(); |
1a20ff27 DG |
6877 | } |
6878 | ||
1d3504fc HS |
6879 | /* handle null as "default" */ |
6880 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6881 | struct sched_domain_attr *new, int idx_new) | |
6882 | { | |
6883 | struct sched_domain_attr tmp; | |
6884 | ||
6885 | /* fast path */ | |
6886 | if (!new && !cur) | |
6887 | return 1; | |
6888 | ||
6889 | tmp = SD_ATTR_INIT; | |
6890 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6891 | new ? (new + idx_new) : &tmp, | |
6892 | sizeof(struct sched_domain_attr)); | |
6893 | } | |
6894 | ||
029190c5 PJ |
6895 | /* |
6896 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6897 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6898 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6899 | * It destroys each deleted domain and builds each new domain. | |
6900 | * | |
acc3f5d7 | 6901 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6902 | * The masks don't intersect (don't overlap.) We should setup one |
6903 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6904 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6905 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6906 | * it as it is. | |
6907 | * | |
acc3f5d7 RR |
6908 | * The passed in 'doms_new' should be allocated using |
6909 | * alloc_sched_domains. This routine takes ownership of it and will | |
6910 | * free_sched_domains it when done with it. If the caller failed the | |
6911 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6912 | * and partition_sched_domains() will fallback to the single partition | |
6913 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6914 | * |
96f874e2 | 6915 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6916 | * ndoms_new == 0 is a special case for destroying existing domains, |
6917 | * and it will not create the default domain. | |
dfb512ec | 6918 | * |
029190c5 PJ |
6919 | * Call with hotplug lock held |
6920 | */ | |
acc3f5d7 | 6921 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6922 | struct sched_domain_attr *dattr_new) |
029190c5 | 6923 | { |
dfb512ec | 6924 | int i, j, n; |
d65bd5ec | 6925 | int new_topology; |
029190c5 | 6926 | |
712555ee | 6927 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6928 | |
7378547f MM |
6929 | /* always unregister in case we don't destroy any domains */ |
6930 | unregister_sched_domain_sysctl(); | |
6931 | ||
d65bd5ec HC |
6932 | /* Let architecture update cpu core mappings. */ |
6933 | new_topology = arch_update_cpu_topology(); | |
6934 | ||
dfb512ec | 6935 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6936 | |
6937 | /* Destroy deleted domains */ | |
6938 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6939 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6940 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6941 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6942 | goto match1; |
6943 | } | |
6944 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6945 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6946 | match1: |
6947 | ; | |
6948 | } | |
6949 | ||
c8d2d47a | 6950 | n = ndoms_cur; |
e761b772 | 6951 | if (doms_new == NULL) { |
c8d2d47a | 6952 | n = 0; |
acc3f5d7 | 6953 | doms_new = &fallback_doms; |
6ad4c188 | 6954 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6955 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6956 | } |
6957 | ||
029190c5 PJ |
6958 | /* Build new domains */ |
6959 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6960 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6961 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6962 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6963 | goto match2; |
6964 | } | |
6965 | /* no match - add a new doms_new */ | |
dce840a0 | 6966 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6967 | match2: |
6968 | ; | |
6969 | } | |
6970 | ||
6971 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6972 | if (doms_cur != &fallback_doms) |
6973 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6974 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6975 | doms_cur = doms_new; |
1d3504fc | 6976 | dattr_cur = dattr_new; |
029190c5 | 6977 | ndoms_cur = ndoms_new; |
7378547f MM |
6978 | |
6979 | register_sched_domain_sysctl(); | |
a1835615 | 6980 | |
712555ee | 6981 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6982 | } |
6983 | ||
d35be8ba SB |
6984 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6985 | ||
1da177e4 | 6986 | /* |
3a101d05 TH |
6987 | * Update cpusets according to cpu_active mask. If cpusets are |
6988 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6989 | * around partition_sched_domains(). | |
d35be8ba SB |
6990 | * |
6991 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6992 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6993 | */ |
0b2e918a TH |
6994 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6995 | void *hcpu) | |
e761b772 | 6996 | { |
d35be8ba SB |
6997 | switch (action) { |
6998 | case CPU_ONLINE_FROZEN: | |
6999 | case CPU_DOWN_FAILED_FROZEN: | |
7000 | ||
7001 | /* | |
7002 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7003 | * resume sequence. As long as this is not the last online | |
7004 | * operation in the resume sequence, just build a single sched | |
7005 | * domain, ignoring cpusets. | |
7006 | */ | |
7007 | num_cpus_frozen--; | |
7008 | if (likely(num_cpus_frozen)) { | |
7009 | partition_sched_domains(1, NULL, NULL); | |
7010 | break; | |
7011 | } | |
7012 | ||
7013 | /* | |
7014 | * This is the last CPU online operation. So fall through and | |
7015 | * restore the original sched domains by considering the | |
7016 | * cpuset configurations. | |
7017 | */ | |
7018 | ||
e761b772 | 7019 | case CPU_ONLINE: |
6ad4c188 | 7020 | case CPU_DOWN_FAILED: |
7ddf96b0 | 7021 | cpuset_update_active_cpus(true); |
d35be8ba | 7022 | break; |
3a101d05 TH |
7023 | default: |
7024 | return NOTIFY_DONE; | |
7025 | } | |
d35be8ba | 7026 | return NOTIFY_OK; |
3a101d05 | 7027 | } |
e761b772 | 7028 | |
0b2e918a TH |
7029 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7030 | void *hcpu) | |
3a101d05 | 7031 | { |
d35be8ba | 7032 | switch (action) { |
3a101d05 | 7033 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 7034 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7035 | break; |
7036 | case CPU_DOWN_PREPARE_FROZEN: | |
7037 | num_cpus_frozen++; | |
7038 | partition_sched_domains(1, NULL, NULL); | |
7039 | break; | |
e761b772 MK |
7040 | default: |
7041 | return NOTIFY_DONE; | |
7042 | } | |
d35be8ba | 7043 | return NOTIFY_OK; |
e761b772 | 7044 | } |
e761b772 | 7045 | |
1da177e4 LT |
7046 | void __init sched_init_smp(void) |
7047 | { | |
dcc30a35 RR |
7048 | cpumask_var_t non_isolated_cpus; |
7049 | ||
7050 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7051 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7052 | |
cb83b629 PZ |
7053 | sched_init_numa(); |
7054 | ||
6acce3ef PZ |
7055 | /* |
7056 | * There's no userspace yet to cause hotplug operations; hence all the | |
7057 | * cpu masks are stable and all blatant races in the below code cannot | |
7058 | * happen. | |
7059 | */ | |
712555ee | 7060 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7061 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7062 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7063 | if (cpumask_empty(non_isolated_cpus)) | |
7064 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7065 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7066 | |
301a5cba | 7067 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7068 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7069 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7070 | |
b328ca18 | 7071 | init_hrtick(); |
5c1e1767 NP |
7072 | |
7073 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7074 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7075 | BUG(); |
19978ca6 | 7076 | sched_init_granularity(); |
dcc30a35 | 7077 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7078 | |
0e3900e6 | 7079 | init_sched_rt_class(); |
1baca4ce | 7080 | init_sched_dl_class(); |
1da177e4 LT |
7081 | } |
7082 | #else | |
7083 | void __init sched_init_smp(void) | |
7084 | { | |
19978ca6 | 7085 | sched_init_granularity(); |
1da177e4 LT |
7086 | } |
7087 | #endif /* CONFIG_SMP */ | |
7088 | ||
cd1bb94b AB |
7089 | const_debug unsigned int sysctl_timer_migration = 1; |
7090 | ||
1da177e4 LT |
7091 | int in_sched_functions(unsigned long addr) |
7092 | { | |
1da177e4 LT |
7093 | return in_lock_functions(addr) || |
7094 | (addr >= (unsigned long)__sched_text_start | |
7095 | && addr < (unsigned long)__sched_text_end); | |
7096 | } | |
7097 | ||
029632fb | 7098 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7099 | /* |
7100 | * Default task group. | |
7101 | * Every task in system belongs to this group at bootup. | |
7102 | */ | |
029632fb | 7103 | struct task_group root_task_group; |
35cf4e50 | 7104 | LIST_HEAD(task_groups); |
052f1dc7 | 7105 | #endif |
6f505b16 | 7106 | |
e6252c3e | 7107 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7108 | |
1da177e4 LT |
7109 | void __init sched_init(void) |
7110 | { | |
dd41f596 | 7111 | int i, j; |
434d53b0 MT |
7112 | unsigned long alloc_size = 0, ptr; |
7113 | ||
7114 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7115 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7116 | #endif | |
7117 | #ifdef CONFIG_RT_GROUP_SCHED | |
7118 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7119 | #endif | |
434d53b0 | 7120 | if (alloc_size) { |
36b7b6d4 | 7121 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7122 | |
7123 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7124 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7125 | ptr += nr_cpu_ids * sizeof(void **); |
7126 | ||
07e06b01 | 7127 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7128 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7129 | |
6d6bc0ad | 7130 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7131 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7132 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7133 | ptr += nr_cpu_ids * sizeof(void **); |
7134 | ||
07e06b01 | 7135 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7136 | ptr += nr_cpu_ids * sizeof(void **); |
7137 | ||
6d6bc0ad | 7138 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7139 | } |
df7c8e84 | 7140 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7141 | for_each_possible_cpu(i) { |
7142 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7143 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7144 | } |
b74e6278 | 7145 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7146 | |
332ac17e DF |
7147 | init_rt_bandwidth(&def_rt_bandwidth, |
7148 | global_rt_period(), global_rt_runtime()); | |
7149 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7150 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7151 | |
57d885fe GH |
7152 | #ifdef CONFIG_SMP |
7153 | init_defrootdomain(); | |
7154 | #endif | |
7155 | ||
d0b27fa7 | 7156 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7157 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7158 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7159 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7160 | |
7c941438 | 7161 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7162 | list_add(&root_task_group.list, &task_groups); |
7163 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7164 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7165 | autogroup_init(&init_task); |
54c707e9 | 7166 | |
7c941438 | 7167 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7168 | |
0a945022 | 7169 | for_each_possible_cpu(i) { |
70b97a7f | 7170 | struct rq *rq; |
1da177e4 LT |
7171 | |
7172 | rq = cpu_rq(i); | |
05fa785c | 7173 | raw_spin_lock_init(&rq->lock); |
7897986b | 7174 | rq->nr_running = 0; |
dce48a84 TG |
7175 | rq->calc_load_active = 0; |
7176 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7177 | init_cfs_rq(&rq->cfs); |
6f505b16 | 7178 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 7179 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 7180 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7181 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7182 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7183 | /* |
07e06b01 | 7184 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7185 | * |
7186 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7187 | * gets 100% of the cpu resources in the system. This overall | |
7188 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7189 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7190 | * based on each entity's (task or task-group's) weight |
7191 | * (se->load.weight). | |
7192 | * | |
07e06b01 | 7193 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7194 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7195 | * then A0's share of the cpu resource is: | |
7196 | * | |
0d905bca | 7197 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7198 | * |
07e06b01 YZ |
7199 | * We achieve this by letting root_task_group's tasks sit |
7200 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7201 | */ |
ab84d31e | 7202 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7203 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7204 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7205 | ||
7206 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7207 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7208 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7209 | #endif |
1da177e4 | 7210 | |
dd41f596 IM |
7211 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7212 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7213 | |
7214 | rq->last_load_update_tick = jiffies; | |
7215 | ||
1da177e4 | 7216 | #ifdef CONFIG_SMP |
41c7ce9a | 7217 | rq->sd = NULL; |
57d885fe | 7218 | rq->rd = NULL; |
ca8ce3d0 | 7219 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 7220 | rq->post_schedule = 0; |
1da177e4 | 7221 | rq->active_balance = 0; |
dd41f596 | 7222 | rq->next_balance = jiffies; |
1da177e4 | 7223 | rq->push_cpu = 0; |
0a2966b4 | 7224 | rq->cpu = i; |
1f11eb6a | 7225 | rq->online = 0; |
eae0c9df MG |
7226 | rq->idle_stamp = 0; |
7227 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7228 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7229 | |
7230 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7231 | ||
dc938520 | 7232 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7233 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7234 | rq->nohz_flags = 0; |
83cd4fe2 | 7235 | #endif |
265f22a9 FW |
7236 | #ifdef CONFIG_NO_HZ_FULL |
7237 | rq->last_sched_tick = 0; | |
7238 | #endif | |
1da177e4 | 7239 | #endif |
8f4d37ec | 7240 | init_rq_hrtick(rq); |
1da177e4 | 7241 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7242 | } |
7243 | ||
2dd73a4f | 7244 | set_load_weight(&init_task); |
b50f60ce | 7245 | |
e107be36 AK |
7246 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7247 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7248 | #endif | |
7249 | ||
1da177e4 LT |
7250 | /* |
7251 | * The boot idle thread does lazy MMU switching as well: | |
7252 | */ | |
7253 | atomic_inc(&init_mm.mm_count); | |
7254 | enter_lazy_tlb(&init_mm, current); | |
7255 | ||
1b537c7d YD |
7256 | /* |
7257 | * During early bootup we pretend to be a normal task: | |
7258 | */ | |
7259 | current->sched_class = &fair_sched_class; | |
7260 | ||
1da177e4 LT |
7261 | /* |
7262 | * Make us the idle thread. Technically, schedule() should not be | |
7263 | * called from this thread, however somewhere below it might be, | |
7264 | * but because we are the idle thread, we just pick up running again | |
7265 | * when this runqueue becomes "idle". | |
7266 | */ | |
7267 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7268 | |
7269 | calc_load_update = jiffies + LOAD_FREQ; | |
7270 | ||
bf4d83f6 | 7271 | #ifdef CONFIG_SMP |
4cb98839 | 7272 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7273 | /* May be allocated at isolcpus cmdline parse time */ |
7274 | if (cpu_isolated_map == NULL) | |
7275 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7276 | idle_thread_set_boot_cpu(); |
a803f026 | 7277 | set_cpu_rq_start_time(); |
029632fb PZ |
7278 | #endif |
7279 | init_sched_fair_class(); | |
6a7b3dc3 | 7280 | |
6892b75e | 7281 | scheduler_running = 1; |
1da177e4 LT |
7282 | } |
7283 | ||
d902db1e | 7284 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7285 | static inline int preempt_count_equals(int preempt_offset) |
7286 | { | |
234da7bc | 7287 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7288 | |
4ba8216c | 7289 | return (nested == preempt_offset); |
e4aafea2 FW |
7290 | } |
7291 | ||
d894837f | 7292 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7293 | { |
8eb23b9f PZ |
7294 | /* |
7295 | * Blocking primitives will set (and therefore destroy) current->state, | |
7296 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7297 | * otherwise we will destroy state. | |
7298 | */ | |
00845eb9 | 7299 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7300 | "do not call blocking ops when !TASK_RUNNING; " |
7301 | "state=%lx set at [<%p>] %pS\n", | |
7302 | current->state, | |
7303 | (void *)current->task_state_change, | |
00845eb9 | 7304 | (void *)current->task_state_change); |
8eb23b9f | 7305 | |
3427445a PZ |
7306 | ___might_sleep(file, line, preempt_offset); |
7307 | } | |
7308 | EXPORT_SYMBOL(__might_sleep); | |
7309 | ||
7310 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7311 | { |
1da177e4 LT |
7312 | static unsigned long prev_jiffy; /* ratelimiting */ |
7313 | ||
b3fbab05 | 7314 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7315 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7316 | !is_idle_task(current)) || | |
e4aafea2 | 7317 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7318 | return; |
7319 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7320 | return; | |
7321 | prev_jiffy = jiffies; | |
7322 | ||
3df0fc5b PZ |
7323 | printk(KERN_ERR |
7324 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7325 | file, line); | |
7326 | printk(KERN_ERR | |
7327 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7328 | in_atomic(), irqs_disabled(), | |
7329 | current->pid, current->comm); | |
aef745fc | 7330 | |
a8b686b3 ES |
7331 | if (task_stack_end_corrupted(current)) |
7332 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7333 | ||
aef745fc IM |
7334 | debug_show_held_locks(current); |
7335 | if (irqs_disabled()) | |
7336 | print_irqtrace_events(current); | |
8f47b187 TG |
7337 | #ifdef CONFIG_DEBUG_PREEMPT |
7338 | if (!preempt_count_equals(preempt_offset)) { | |
7339 | pr_err("Preemption disabled at:"); | |
7340 | print_ip_sym(current->preempt_disable_ip); | |
7341 | pr_cont("\n"); | |
7342 | } | |
7343 | #endif | |
aef745fc | 7344 | dump_stack(); |
1da177e4 | 7345 | } |
3427445a | 7346 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7347 | #endif |
7348 | ||
7349 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7350 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7351 | { | |
da7a735e | 7352 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7353 | struct sched_attr attr = { |
7354 | .sched_policy = SCHED_NORMAL, | |
7355 | }; | |
da7a735e | 7356 | int old_prio = p->prio; |
da0c1e65 | 7357 | int queued; |
3e51f33f | 7358 | |
da0c1e65 KT |
7359 | queued = task_on_rq_queued(p); |
7360 | if (queued) | |
4ca9b72b | 7361 | dequeue_task(rq, p, 0); |
d50dde5a | 7362 | __setscheduler(rq, p, &attr); |
da0c1e65 | 7363 | if (queued) { |
4ca9b72b | 7364 | enqueue_task(rq, p, 0); |
8875125e | 7365 | resched_curr(rq); |
3a5e4dc1 | 7366 | } |
da7a735e PZ |
7367 | |
7368 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7369 | } |
7370 | ||
1da177e4 LT |
7371 | void normalize_rt_tasks(void) |
7372 | { | |
a0f98a1c | 7373 | struct task_struct *g, *p; |
1da177e4 | 7374 | unsigned long flags; |
70b97a7f | 7375 | struct rq *rq; |
1da177e4 | 7376 | |
3472eaa1 | 7377 | read_lock(&tasklist_lock); |
5d07f420 | 7378 | for_each_process_thread(g, p) { |
178be793 IM |
7379 | /* |
7380 | * Only normalize user tasks: | |
7381 | */ | |
3472eaa1 | 7382 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7383 | continue; |
7384 | ||
6cfb0d5d | 7385 | p->se.exec_start = 0; |
6cfb0d5d | 7386 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7387 | p->se.statistics.wait_start = 0; |
7388 | p->se.statistics.sleep_start = 0; | |
7389 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7390 | #endif |
dd41f596 | 7391 | |
aab03e05 | 7392 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7393 | /* |
7394 | * Renice negative nice level userspace | |
7395 | * tasks back to 0: | |
7396 | */ | |
3472eaa1 | 7397 | if (task_nice(p) < 0) |
dd41f596 | 7398 | set_user_nice(p, 0); |
1da177e4 | 7399 | continue; |
dd41f596 | 7400 | } |
1da177e4 | 7401 | |
3472eaa1 | 7402 | rq = task_rq_lock(p, &flags); |
178be793 | 7403 | normalize_task(rq, p); |
3472eaa1 | 7404 | task_rq_unlock(rq, p, &flags); |
5d07f420 | 7405 | } |
3472eaa1 | 7406 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7407 | } |
7408 | ||
7409 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7410 | |
67fc4e0c | 7411 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7412 | /* |
67fc4e0c | 7413 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7414 | * |
7415 | * They can only be called when the whole system has been | |
7416 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7417 | * activity can take place. Using them for anything else would | |
7418 | * be a serious bug, and as a result, they aren't even visible | |
7419 | * under any other configuration. | |
7420 | */ | |
7421 | ||
7422 | /** | |
7423 | * curr_task - return the current task for a given cpu. | |
7424 | * @cpu: the processor in question. | |
7425 | * | |
7426 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7427 | * |
7428 | * Return: The current task for @cpu. | |
1df5c10a | 7429 | */ |
36c8b586 | 7430 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7431 | { |
7432 | return cpu_curr(cpu); | |
7433 | } | |
7434 | ||
67fc4e0c JW |
7435 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7436 | ||
7437 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7438 | /** |
7439 | * set_curr_task - set the current task for a given cpu. | |
7440 | * @cpu: the processor in question. | |
7441 | * @p: the task pointer to set. | |
7442 | * | |
7443 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7444 | * are serviced on a separate stack. It allows the architecture to switch the |
7445 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7446 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7447 | * and caller must save the original value of the current task (see | |
7448 | * curr_task() above) and restore that value before reenabling interrupts and | |
7449 | * re-starting the system. | |
7450 | * | |
7451 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7452 | */ | |
36c8b586 | 7453 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7454 | { |
7455 | cpu_curr(cpu) = p; | |
7456 | } | |
7457 | ||
7458 | #endif | |
29f59db3 | 7459 | |
7c941438 | 7460 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7461 | /* task_group_lock serializes the addition/removal of task groups */ |
7462 | static DEFINE_SPINLOCK(task_group_lock); | |
7463 | ||
bccbe08a PZ |
7464 | static void free_sched_group(struct task_group *tg) |
7465 | { | |
7466 | free_fair_sched_group(tg); | |
7467 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7468 | autogroup_free(tg); |
bccbe08a PZ |
7469 | kfree(tg); |
7470 | } | |
7471 | ||
7472 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7473 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7474 | { |
7475 | struct task_group *tg; | |
bccbe08a PZ |
7476 | |
7477 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7478 | if (!tg) | |
7479 | return ERR_PTR(-ENOMEM); | |
7480 | ||
ec7dc8ac | 7481 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7482 | goto err; |
7483 | ||
ec7dc8ac | 7484 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7485 | goto err; |
7486 | ||
ace783b9 LZ |
7487 | return tg; |
7488 | ||
7489 | err: | |
7490 | free_sched_group(tg); | |
7491 | return ERR_PTR(-ENOMEM); | |
7492 | } | |
7493 | ||
7494 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7495 | { | |
7496 | unsigned long flags; | |
7497 | ||
8ed36996 | 7498 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7499 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7500 | |
7501 | WARN_ON(!parent); /* root should already exist */ | |
7502 | ||
7503 | tg->parent = parent; | |
f473aa5e | 7504 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7505 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7506 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7507 | } |
7508 | ||
9b5b7751 | 7509 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7510 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7511 | { |
29f59db3 | 7512 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7513 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7514 | } |
7515 | ||
9b5b7751 | 7516 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7517 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7518 | { |
7519 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7520 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7521 | } | |
7522 | ||
7523 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7524 | { |
8ed36996 | 7525 | unsigned long flags; |
9b5b7751 | 7526 | int i; |
29f59db3 | 7527 | |
3d4b47b4 PZ |
7528 | /* end participation in shares distribution */ |
7529 | for_each_possible_cpu(i) | |
bccbe08a | 7530 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7531 | |
7532 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7533 | list_del_rcu(&tg->list); |
f473aa5e | 7534 | list_del_rcu(&tg->siblings); |
8ed36996 | 7535 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7536 | } |
7537 | ||
9b5b7751 | 7538 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7539 | * The caller of this function should have put the task in its new group |
7540 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7541 | * reflect its new group. | |
9b5b7751 SV |
7542 | */ |
7543 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7544 | { |
8323f26c | 7545 | struct task_group *tg; |
da0c1e65 | 7546 | int queued, running; |
29f59db3 SV |
7547 | unsigned long flags; |
7548 | struct rq *rq; | |
7549 | ||
7550 | rq = task_rq_lock(tsk, &flags); | |
7551 | ||
051a1d1a | 7552 | running = task_current(rq, tsk); |
da0c1e65 | 7553 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7554 | |
da0c1e65 | 7555 | if (queued) |
29f59db3 | 7556 | dequeue_task(rq, tsk, 0); |
0e1f3483 | 7557 | if (unlikely(running)) |
f3cd1c4e | 7558 | put_prev_task(rq, tsk); |
29f59db3 | 7559 | |
f7b8a47d KT |
7560 | /* |
7561 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7562 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7563 | * to prevent lockdep warnings. | |
7564 | */ | |
7565 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7566 | struct task_group, css); |
7567 | tg = autogroup_task_group(tsk, tg); | |
7568 | tsk->sched_task_group = tg; | |
7569 | ||
810b3817 | 7570 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7571 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7572 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7573 | else |
810b3817 | 7574 | #endif |
b2b5ce02 | 7575 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7576 | |
0e1f3483 HS |
7577 | if (unlikely(running)) |
7578 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7579 | if (queued) |
371fd7e7 | 7580 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7581 | |
0122ec5b | 7582 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7583 | } |
7c941438 | 7584 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7585 | |
a790de99 PT |
7586 | #ifdef CONFIG_RT_GROUP_SCHED |
7587 | /* | |
7588 | * Ensure that the real time constraints are schedulable. | |
7589 | */ | |
7590 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7591 | |
9a7e0b18 PZ |
7592 | /* Must be called with tasklist_lock held */ |
7593 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7594 | { |
9a7e0b18 | 7595 | struct task_struct *g, *p; |
b40b2e8e | 7596 | |
1fe89e1b PZ |
7597 | /* |
7598 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7599 | */ | |
7600 | if (task_group_is_autogroup(tg)) | |
7601 | return 0; | |
7602 | ||
5d07f420 | 7603 | for_each_process_thread(g, p) { |
8651c658 | 7604 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7605 | return 1; |
5d07f420 | 7606 | } |
b40b2e8e | 7607 | |
9a7e0b18 PZ |
7608 | return 0; |
7609 | } | |
b40b2e8e | 7610 | |
9a7e0b18 PZ |
7611 | struct rt_schedulable_data { |
7612 | struct task_group *tg; | |
7613 | u64 rt_period; | |
7614 | u64 rt_runtime; | |
7615 | }; | |
b40b2e8e | 7616 | |
a790de99 | 7617 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7618 | { |
7619 | struct rt_schedulable_data *d = data; | |
7620 | struct task_group *child; | |
7621 | unsigned long total, sum = 0; | |
7622 | u64 period, runtime; | |
b40b2e8e | 7623 | |
9a7e0b18 PZ |
7624 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7625 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7626 | |
9a7e0b18 PZ |
7627 | if (tg == d->tg) { |
7628 | period = d->rt_period; | |
7629 | runtime = d->rt_runtime; | |
b40b2e8e | 7630 | } |
b40b2e8e | 7631 | |
4653f803 PZ |
7632 | /* |
7633 | * Cannot have more runtime than the period. | |
7634 | */ | |
7635 | if (runtime > period && runtime != RUNTIME_INF) | |
7636 | return -EINVAL; | |
6f505b16 | 7637 | |
4653f803 PZ |
7638 | /* |
7639 | * Ensure we don't starve existing RT tasks. | |
7640 | */ | |
9a7e0b18 PZ |
7641 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7642 | return -EBUSY; | |
6f505b16 | 7643 | |
9a7e0b18 | 7644 | total = to_ratio(period, runtime); |
6f505b16 | 7645 | |
4653f803 PZ |
7646 | /* |
7647 | * Nobody can have more than the global setting allows. | |
7648 | */ | |
7649 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7650 | return -EINVAL; | |
6f505b16 | 7651 | |
4653f803 PZ |
7652 | /* |
7653 | * The sum of our children's runtime should not exceed our own. | |
7654 | */ | |
9a7e0b18 PZ |
7655 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7656 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7657 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7658 | |
9a7e0b18 PZ |
7659 | if (child == d->tg) { |
7660 | period = d->rt_period; | |
7661 | runtime = d->rt_runtime; | |
7662 | } | |
6f505b16 | 7663 | |
9a7e0b18 | 7664 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7665 | } |
6f505b16 | 7666 | |
9a7e0b18 PZ |
7667 | if (sum > total) |
7668 | return -EINVAL; | |
7669 | ||
7670 | return 0; | |
6f505b16 PZ |
7671 | } |
7672 | ||
9a7e0b18 | 7673 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7674 | { |
8277434e PT |
7675 | int ret; |
7676 | ||
9a7e0b18 PZ |
7677 | struct rt_schedulable_data data = { |
7678 | .tg = tg, | |
7679 | .rt_period = period, | |
7680 | .rt_runtime = runtime, | |
7681 | }; | |
7682 | ||
8277434e PT |
7683 | rcu_read_lock(); |
7684 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7685 | rcu_read_unlock(); | |
7686 | ||
7687 | return ret; | |
521f1a24 DG |
7688 | } |
7689 | ||
ab84d31e | 7690 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7691 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7692 | { |
ac086bc2 | 7693 | int i, err = 0; |
9f0c1e56 | 7694 | |
2636ed5f PZ |
7695 | /* |
7696 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7697 | * kernel creating (and or operating) RT threads. | |
7698 | */ | |
7699 | if (tg == &root_task_group && rt_runtime == 0) | |
7700 | return -EINVAL; | |
7701 | ||
7702 | /* No period doesn't make any sense. */ | |
7703 | if (rt_period == 0) | |
7704 | return -EINVAL; | |
7705 | ||
9f0c1e56 | 7706 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7707 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7708 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7709 | if (err) | |
9f0c1e56 | 7710 | goto unlock; |
ac086bc2 | 7711 | |
0986b11b | 7712 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7713 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7714 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7715 | |
7716 | for_each_possible_cpu(i) { | |
7717 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7718 | ||
0986b11b | 7719 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7720 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7721 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7722 | } |
0986b11b | 7723 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7724 | unlock: |
521f1a24 | 7725 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7726 | mutex_unlock(&rt_constraints_mutex); |
7727 | ||
7728 | return err; | |
6f505b16 PZ |
7729 | } |
7730 | ||
25cc7da7 | 7731 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7732 | { |
7733 | u64 rt_runtime, rt_period; | |
7734 | ||
7735 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7736 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7737 | if (rt_runtime_us < 0) | |
7738 | rt_runtime = RUNTIME_INF; | |
7739 | ||
ab84d31e | 7740 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7741 | } |
7742 | ||
25cc7da7 | 7743 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7744 | { |
7745 | u64 rt_runtime_us; | |
7746 | ||
d0b27fa7 | 7747 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7748 | return -1; |
7749 | ||
d0b27fa7 | 7750 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7751 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7752 | return rt_runtime_us; | |
7753 | } | |
d0b27fa7 | 7754 | |
25cc7da7 | 7755 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7756 | { |
7757 | u64 rt_runtime, rt_period; | |
7758 | ||
7759 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7760 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7761 | ||
ab84d31e | 7762 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7763 | } |
7764 | ||
25cc7da7 | 7765 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7766 | { |
7767 | u64 rt_period_us; | |
7768 | ||
7769 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7770 | do_div(rt_period_us, NSEC_PER_USEC); | |
7771 | return rt_period_us; | |
7772 | } | |
332ac17e | 7773 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7774 | |
332ac17e | 7775 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7776 | static int sched_rt_global_constraints(void) |
7777 | { | |
7778 | int ret = 0; | |
7779 | ||
7780 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7781 | read_lock(&tasklist_lock); |
4653f803 | 7782 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7783 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7784 | mutex_unlock(&rt_constraints_mutex); |
7785 | ||
7786 | return ret; | |
7787 | } | |
54e99124 | 7788 | |
25cc7da7 | 7789 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7790 | { |
7791 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7792 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7793 | return 0; | |
7794 | ||
7795 | return 1; | |
7796 | } | |
7797 | ||
6d6bc0ad | 7798 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7799 | static int sched_rt_global_constraints(void) |
7800 | { | |
ac086bc2 | 7801 | unsigned long flags; |
332ac17e | 7802 | int i, ret = 0; |
ec5d4989 | 7803 | |
0986b11b | 7804 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7805 | for_each_possible_cpu(i) { |
7806 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7807 | ||
0986b11b | 7808 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7809 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7810 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7811 | } |
0986b11b | 7812 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7813 | |
332ac17e | 7814 | return ret; |
d0b27fa7 | 7815 | } |
6d6bc0ad | 7816 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7817 | |
332ac17e DF |
7818 | static int sched_dl_global_constraints(void) |
7819 | { | |
1724813d PZ |
7820 | u64 runtime = global_rt_runtime(); |
7821 | u64 period = global_rt_period(); | |
332ac17e | 7822 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7823 | struct dl_bw *dl_b; |
1724813d | 7824 | int cpu, ret = 0; |
49516342 | 7825 | unsigned long flags; |
332ac17e DF |
7826 | |
7827 | /* | |
7828 | * Here we want to check the bandwidth not being set to some | |
7829 | * value smaller than the currently allocated bandwidth in | |
7830 | * any of the root_domains. | |
7831 | * | |
7832 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7833 | * cycling on root_domains... Discussion on different/better | |
7834 | * solutions is welcome! | |
7835 | */ | |
1724813d | 7836 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7837 | rcu_read_lock_sched(); |
7838 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7839 | |
49516342 | 7840 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7841 | if (new_bw < dl_b->total_bw) |
7842 | ret = -EBUSY; | |
49516342 | 7843 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7844 | |
f10e00f4 KT |
7845 | rcu_read_unlock_sched(); |
7846 | ||
1724813d PZ |
7847 | if (ret) |
7848 | break; | |
332ac17e DF |
7849 | } |
7850 | ||
1724813d | 7851 | return ret; |
332ac17e DF |
7852 | } |
7853 | ||
1724813d | 7854 | static void sched_dl_do_global(void) |
ce0dbbbb | 7855 | { |
1724813d | 7856 | u64 new_bw = -1; |
f10e00f4 | 7857 | struct dl_bw *dl_b; |
1724813d | 7858 | int cpu; |
49516342 | 7859 | unsigned long flags; |
ce0dbbbb | 7860 | |
1724813d PZ |
7861 | def_dl_bandwidth.dl_period = global_rt_period(); |
7862 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7863 | ||
7864 | if (global_rt_runtime() != RUNTIME_INF) | |
7865 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7866 | ||
7867 | /* | |
7868 | * FIXME: As above... | |
7869 | */ | |
7870 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7871 | rcu_read_lock_sched(); |
7872 | dl_b = dl_bw_of(cpu); | |
1724813d | 7873 | |
49516342 | 7874 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7875 | dl_b->bw = new_bw; |
49516342 | 7876 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
7877 | |
7878 | rcu_read_unlock_sched(); | |
ce0dbbbb | 7879 | } |
1724813d PZ |
7880 | } |
7881 | ||
7882 | static int sched_rt_global_validate(void) | |
7883 | { | |
7884 | if (sysctl_sched_rt_period <= 0) | |
7885 | return -EINVAL; | |
7886 | ||
e9e7cb38 JL |
7887 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7888 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7889 | return -EINVAL; |
7890 | ||
7891 | return 0; | |
7892 | } | |
7893 | ||
7894 | static void sched_rt_do_global(void) | |
7895 | { | |
7896 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7897 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7898 | } |
7899 | ||
d0b27fa7 | 7900 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7901 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7902 | loff_t *ppos) |
7903 | { | |
d0b27fa7 PZ |
7904 | int old_period, old_runtime; |
7905 | static DEFINE_MUTEX(mutex); | |
1724813d | 7906 | int ret; |
d0b27fa7 PZ |
7907 | |
7908 | mutex_lock(&mutex); | |
7909 | old_period = sysctl_sched_rt_period; | |
7910 | old_runtime = sysctl_sched_rt_runtime; | |
7911 | ||
8d65af78 | 7912 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7913 | |
7914 | if (!ret && write) { | |
1724813d PZ |
7915 | ret = sched_rt_global_validate(); |
7916 | if (ret) | |
7917 | goto undo; | |
7918 | ||
d0b27fa7 | 7919 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7920 | if (ret) |
7921 | goto undo; | |
7922 | ||
7923 | ret = sched_dl_global_constraints(); | |
7924 | if (ret) | |
7925 | goto undo; | |
7926 | ||
7927 | sched_rt_do_global(); | |
7928 | sched_dl_do_global(); | |
7929 | } | |
7930 | if (0) { | |
7931 | undo: | |
7932 | sysctl_sched_rt_period = old_period; | |
7933 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7934 | } |
7935 | mutex_unlock(&mutex); | |
7936 | ||
7937 | return ret; | |
7938 | } | |
68318b8e | 7939 | |
1724813d | 7940 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7941 | void __user *buffer, size_t *lenp, |
7942 | loff_t *ppos) | |
7943 | { | |
7944 | int ret; | |
332ac17e | 7945 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7946 | |
7947 | mutex_lock(&mutex); | |
332ac17e | 7948 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7949 | /* make sure that internally we keep jiffies */ |
7950 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7951 | if (!ret && write) { |
1724813d PZ |
7952 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7953 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7954 | } |
7955 | mutex_unlock(&mutex); | |
332ac17e DF |
7956 | return ret; |
7957 | } | |
7958 | ||
052f1dc7 | 7959 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7960 | |
a7c6d554 | 7961 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7962 | { |
a7c6d554 | 7963 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7964 | } |
7965 | ||
eb95419b TH |
7966 | static struct cgroup_subsys_state * |
7967 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7968 | { |
eb95419b TH |
7969 | struct task_group *parent = css_tg(parent_css); |
7970 | struct task_group *tg; | |
68318b8e | 7971 | |
eb95419b | 7972 | if (!parent) { |
68318b8e | 7973 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7974 | return &root_task_group.css; |
68318b8e SV |
7975 | } |
7976 | ||
ec7dc8ac | 7977 | tg = sched_create_group(parent); |
68318b8e SV |
7978 | if (IS_ERR(tg)) |
7979 | return ERR_PTR(-ENOMEM); | |
7980 | ||
68318b8e SV |
7981 | return &tg->css; |
7982 | } | |
7983 | ||
eb95419b | 7984 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7985 | { |
eb95419b | 7986 | struct task_group *tg = css_tg(css); |
5c9d535b | 7987 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 7988 | |
63876986 TH |
7989 | if (parent) |
7990 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7991 | return 0; |
7992 | } | |
7993 | ||
eb95419b | 7994 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7995 | { |
eb95419b | 7996 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7997 | |
7998 | sched_destroy_group(tg); | |
7999 | } | |
8000 | ||
eb95419b | 8001 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 8002 | { |
eb95419b | 8003 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
8004 | |
8005 | sched_offline_group(tg); | |
8006 | } | |
8007 | ||
eeb61e53 KT |
8008 | static void cpu_cgroup_fork(struct task_struct *task) |
8009 | { | |
8010 | sched_move_task(task); | |
8011 | } | |
8012 | ||
eb95419b | 8013 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8014 | struct cgroup_taskset *tset) |
68318b8e | 8015 | { |
bb9d97b6 TH |
8016 | struct task_struct *task; |
8017 | ||
924f0d9a | 8018 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 8019 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8020 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8021 | return -EINVAL; |
b68aa230 | 8022 | #else |
bb9d97b6 TH |
8023 | /* We don't support RT-tasks being in separate groups */ |
8024 | if (task->sched_class != &fair_sched_class) | |
8025 | return -EINVAL; | |
b68aa230 | 8026 | #endif |
bb9d97b6 | 8027 | } |
be367d09 BB |
8028 | return 0; |
8029 | } | |
68318b8e | 8030 | |
eb95419b | 8031 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8032 | struct cgroup_taskset *tset) |
68318b8e | 8033 | { |
bb9d97b6 TH |
8034 | struct task_struct *task; |
8035 | ||
924f0d9a | 8036 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 8037 | sched_move_task(task); |
68318b8e SV |
8038 | } |
8039 | ||
eb95419b TH |
8040 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
8041 | struct cgroup_subsys_state *old_css, | |
8042 | struct task_struct *task) | |
068c5cc5 PZ |
8043 | { |
8044 | /* | |
8045 | * cgroup_exit() is called in the copy_process() failure path. | |
8046 | * Ignore this case since the task hasn't ran yet, this avoids | |
8047 | * trying to poke a half freed task state from generic code. | |
8048 | */ | |
8049 | if (!(task->flags & PF_EXITING)) | |
8050 | return; | |
8051 | ||
8052 | sched_move_task(task); | |
8053 | } | |
8054 | ||
052f1dc7 | 8055 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8056 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8057 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8058 | { |
182446d0 | 8059 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8060 | } |
8061 | ||
182446d0 TH |
8062 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8063 | struct cftype *cft) | |
68318b8e | 8064 | { |
182446d0 | 8065 | struct task_group *tg = css_tg(css); |
68318b8e | 8066 | |
c8b28116 | 8067 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8068 | } |
ab84d31e PT |
8069 | |
8070 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8071 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8072 | ||
ab84d31e PT |
8073 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8074 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8075 | ||
a790de99 PT |
8076 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8077 | ||
ab84d31e PT |
8078 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8079 | { | |
56f570e5 | 8080 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8081 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8082 | |
8083 | if (tg == &root_task_group) | |
8084 | return -EINVAL; | |
8085 | ||
8086 | /* | |
8087 | * Ensure we have at some amount of bandwidth every period. This is | |
8088 | * to prevent reaching a state of large arrears when throttled via | |
8089 | * entity_tick() resulting in prolonged exit starvation. | |
8090 | */ | |
8091 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8092 | return -EINVAL; | |
8093 | ||
8094 | /* | |
8095 | * Likewise, bound things on the otherside by preventing insane quota | |
8096 | * periods. This also allows us to normalize in computing quota | |
8097 | * feasibility. | |
8098 | */ | |
8099 | if (period > max_cfs_quota_period) | |
8100 | return -EINVAL; | |
8101 | ||
0e59bdae KT |
8102 | /* |
8103 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8104 | * unthrottle_offline_cfs_rqs(). | |
8105 | */ | |
8106 | get_online_cpus(); | |
a790de99 PT |
8107 | mutex_lock(&cfs_constraints_mutex); |
8108 | ret = __cfs_schedulable(tg, period, quota); | |
8109 | if (ret) | |
8110 | goto out_unlock; | |
8111 | ||
58088ad0 | 8112 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8113 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8114 | /* |
8115 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8116 | * before making related changes, and on->off must occur afterwards | |
8117 | */ | |
8118 | if (runtime_enabled && !runtime_was_enabled) | |
8119 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8120 | raw_spin_lock_irq(&cfs_b->lock); |
8121 | cfs_b->period = ns_to_ktime(period); | |
8122 | cfs_b->quota = quota; | |
58088ad0 | 8123 | |
a9cf55b2 | 8124 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
8125 | /* restart the period timer (if active) to handle new period expiry */ |
8126 | if (runtime_enabled && cfs_b->timer_active) { | |
8127 | /* force a reprogram */ | |
09dc4ab0 | 8128 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 8129 | } |
ab84d31e PT |
8130 | raw_spin_unlock_irq(&cfs_b->lock); |
8131 | ||
0e59bdae | 8132 | for_each_online_cpu(i) { |
ab84d31e | 8133 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8134 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8135 | |
8136 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8137 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8138 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8139 | |
029632fb | 8140 | if (cfs_rq->throttled) |
671fd9da | 8141 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8142 | raw_spin_unlock_irq(&rq->lock); |
8143 | } | |
1ee14e6c BS |
8144 | if (runtime_was_enabled && !runtime_enabled) |
8145 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8146 | out_unlock: |
8147 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8148 | put_online_cpus(); |
ab84d31e | 8149 | |
a790de99 | 8150 | return ret; |
ab84d31e PT |
8151 | } |
8152 | ||
8153 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8154 | { | |
8155 | u64 quota, period; | |
8156 | ||
029632fb | 8157 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8158 | if (cfs_quota_us < 0) |
8159 | quota = RUNTIME_INF; | |
8160 | else | |
8161 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8162 | ||
8163 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8164 | } | |
8165 | ||
8166 | long tg_get_cfs_quota(struct task_group *tg) | |
8167 | { | |
8168 | u64 quota_us; | |
8169 | ||
029632fb | 8170 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8171 | return -1; |
8172 | ||
029632fb | 8173 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8174 | do_div(quota_us, NSEC_PER_USEC); |
8175 | ||
8176 | return quota_us; | |
8177 | } | |
8178 | ||
8179 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8180 | { | |
8181 | u64 quota, period; | |
8182 | ||
8183 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8184 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8185 | |
ab84d31e PT |
8186 | return tg_set_cfs_bandwidth(tg, period, quota); |
8187 | } | |
8188 | ||
8189 | long tg_get_cfs_period(struct task_group *tg) | |
8190 | { | |
8191 | u64 cfs_period_us; | |
8192 | ||
029632fb | 8193 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8194 | do_div(cfs_period_us, NSEC_PER_USEC); |
8195 | ||
8196 | return cfs_period_us; | |
8197 | } | |
8198 | ||
182446d0 TH |
8199 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8200 | struct cftype *cft) | |
ab84d31e | 8201 | { |
182446d0 | 8202 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8203 | } |
8204 | ||
182446d0 TH |
8205 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8206 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8207 | { |
182446d0 | 8208 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8209 | } |
8210 | ||
182446d0 TH |
8211 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8212 | struct cftype *cft) | |
ab84d31e | 8213 | { |
182446d0 | 8214 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8215 | } |
8216 | ||
182446d0 TH |
8217 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8218 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8219 | { |
182446d0 | 8220 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8221 | } |
8222 | ||
a790de99 PT |
8223 | struct cfs_schedulable_data { |
8224 | struct task_group *tg; | |
8225 | u64 period, quota; | |
8226 | }; | |
8227 | ||
8228 | /* | |
8229 | * normalize group quota/period to be quota/max_period | |
8230 | * note: units are usecs | |
8231 | */ | |
8232 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8233 | struct cfs_schedulable_data *d) | |
8234 | { | |
8235 | u64 quota, period; | |
8236 | ||
8237 | if (tg == d->tg) { | |
8238 | period = d->period; | |
8239 | quota = d->quota; | |
8240 | } else { | |
8241 | period = tg_get_cfs_period(tg); | |
8242 | quota = tg_get_cfs_quota(tg); | |
8243 | } | |
8244 | ||
8245 | /* note: these should typically be equivalent */ | |
8246 | if (quota == RUNTIME_INF || quota == -1) | |
8247 | return RUNTIME_INF; | |
8248 | ||
8249 | return to_ratio(period, quota); | |
8250 | } | |
8251 | ||
8252 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8253 | { | |
8254 | struct cfs_schedulable_data *d = data; | |
029632fb | 8255 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8256 | s64 quota = 0, parent_quota = -1; |
8257 | ||
8258 | if (!tg->parent) { | |
8259 | quota = RUNTIME_INF; | |
8260 | } else { | |
029632fb | 8261 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8262 | |
8263 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8264 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8265 | |
8266 | /* | |
8267 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8268 | * limit is set | |
8269 | */ | |
8270 | if (quota == RUNTIME_INF) | |
8271 | quota = parent_quota; | |
8272 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8273 | return -EINVAL; | |
8274 | } | |
9c58c79a | 8275 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8276 | |
8277 | return 0; | |
8278 | } | |
8279 | ||
8280 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8281 | { | |
8277434e | 8282 | int ret; |
a790de99 PT |
8283 | struct cfs_schedulable_data data = { |
8284 | .tg = tg, | |
8285 | .period = period, | |
8286 | .quota = quota, | |
8287 | }; | |
8288 | ||
8289 | if (quota != RUNTIME_INF) { | |
8290 | do_div(data.period, NSEC_PER_USEC); | |
8291 | do_div(data.quota, NSEC_PER_USEC); | |
8292 | } | |
8293 | ||
8277434e PT |
8294 | rcu_read_lock(); |
8295 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8296 | rcu_read_unlock(); | |
8297 | ||
8298 | return ret; | |
a790de99 | 8299 | } |
e8da1b18 | 8300 | |
2da8ca82 | 8301 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8302 | { |
2da8ca82 | 8303 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8304 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8305 | |
44ffc75b TH |
8306 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8307 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8308 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8309 | |
8310 | return 0; | |
8311 | } | |
ab84d31e | 8312 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8313 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8314 | |
052f1dc7 | 8315 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8316 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8317 | struct cftype *cft, s64 val) | |
6f505b16 | 8318 | { |
182446d0 | 8319 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8320 | } |
8321 | ||
182446d0 TH |
8322 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8323 | struct cftype *cft) | |
6f505b16 | 8324 | { |
182446d0 | 8325 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8326 | } |
d0b27fa7 | 8327 | |
182446d0 TH |
8328 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8329 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8330 | { |
182446d0 | 8331 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8332 | } |
8333 | ||
182446d0 TH |
8334 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8335 | struct cftype *cft) | |
d0b27fa7 | 8336 | { |
182446d0 | 8337 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8338 | } |
6d6bc0ad | 8339 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8340 | |
fe5c7cc2 | 8341 | static struct cftype cpu_files[] = { |
052f1dc7 | 8342 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8343 | { |
8344 | .name = "shares", | |
f4c753b7 PM |
8345 | .read_u64 = cpu_shares_read_u64, |
8346 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8347 | }, |
052f1dc7 | 8348 | #endif |
ab84d31e PT |
8349 | #ifdef CONFIG_CFS_BANDWIDTH |
8350 | { | |
8351 | .name = "cfs_quota_us", | |
8352 | .read_s64 = cpu_cfs_quota_read_s64, | |
8353 | .write_s64 = cpu_cfs_quota_write_s64, | |
8354 | }, | |
8355 | { | |
8356 | .name = "cfs_period_us", | |
8357 | .read_u64 = cpu_cfs_period_read_u64, | |
8358 | .write_u64 = cpu_cfs_period_write_u64, | |
8359 | }, | |
e8da1b18 NR |
8360 | { |
8361 | .name = "stat", | |
2da8ca82 | 8362 | .seq_show = cpu_stats_show, |
e8da1b18 | 8363 | }, |
ab84d31e | 8364 | #endif |
052f1dc7 | 8365 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8366 | { |
9f0c1e56 | 8367 | .name = "rt_runtime_us", |
06ecb27c PM |
8368 | .read_s64 = cpu_rt_runtime_read, |
8369 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8370 | }, |
d0b27fa7 PZ |
8371 | { |
8372 | .name = "rt_period_us", | |
f4c753b7 PM |
8373 | .read_u64 = cpu_rt_period_read_uint, |
8374 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8375 | }, |
052f1dc7 | 8376 | #endif |
4baf6e33 | 8377 | { } /* terminate */ |
68318b8e SV |
8378 | }; |
8379 | ||
073219e9 | 8380 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8381 | .css_alloc = cpu_cgroup_css_alloc, |
8382 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8383 | .css_online = cpu_cgroup_css_online, |
8384 | .css_offline = cpu_cgroup_css_offline, | |
eeb61e53 | 8385 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8386 | .can_attach = cpu_cgroup_can_attach, |
8387 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8388 | .exit = cpu_cgroup_exit, |
5577964e | 8389 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8390 | .early_init = 1, |
8391 | }; | |
8392 | ||
052f1dc7 | 8393 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8394 | |
b637a328 PM |
8395 | void dump_cpu_task(int cpu) |
8396 | { | |
8397 | pr_info("Task dump for CPU %d:\n", cpu); | |
8398 | sched_show_task(cpu_curr(cpu)); | |
8399 | } |