]>
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 | |
3fa0818b RR |
309 | /* cpus with isolated domains */ |
310 | cpumask_var_t cpu_isolated_map; | |
311 | ||
1da177e4 | 312 | /* |
cc2a73b5 | 313 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 314 | */ |
a9957449 | 315 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
316 | __acquires(rq->lock) |
317 | { | |
70b97a7f | 318 | struct rq *rq; |
1da177e4 LT |
319 | |
320 | local_irq_disable(); | |
321 | rq = this_rq(); | |
05fa785c | 322 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
323 | |
324 | return rq; | |
325 | } | |
326 | ||
8f4d37ec PZ |
327 | #ifdef CONFIG_SCHED_HRTICK |
328 | /* | |
329 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 330 | */ |
8f4d37ec | 331 | |
8f4d37ec PZ |
332 | static void hrtick_clear(struct rq *rq) |
333 | { | |
334 | if (hrtimer_active(&rq->hrtick_timer)) | |
335 | hrtimer_cancel(&rq->hrtick_timer); | |
336 | } | |
337 | ||
8f4d37ec PZ |
338 | /* |
339 | * High-resolution timer tick. | |
340 | * Runs from hardirq context with interrupts disabled. | |
341 | */ | |
342 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
343 | { | |
344 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
345 | ||
346 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
347 | ||
05fa785c | 348 | raw_spin_lock(&rq->lock); |
3e51f33f | 349 | update_rq_clock(rq); |
8f4d37ec | 350 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 351 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
352 | |
353 | return HRTIMER_NORESTART; | |
354 | } | |
355 | ||
95e904c7 | 356 | #ifdef CONFIG_SMP |
971ee28c PZ |
357 | |
358 | static int __hrtick_restart(struct rq *rq) | |
359 | { | |
360 | struct hrtimer *timer = &rq->hrtick_timer; | |
361 | ktime_t time = hrtimer_get_softexpires(timer); | |
362 | ||
363 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
364 | } | |
365 | ||
31656519 PZ |
366 | /* |
367 | * called from hardirq (IPI) context | |
368 | */ | |
369 | static void __hrtick_start(void *arg) | |
b328ca18 | 370 | { |
31656519 | 371 | struct rq *rq = arg; |
b328ca18 | 372 | |
05fa785c | 373 | raw_spin_lock(&rq->lock); |
971ee28c | 374 | __hrtick_restart(rq); |
31656519 | 375 | rq->hrtick_csd_pending = 0; |
05fa785c | 376 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
377 | } |
378 | ||
31656519 PZ |
379 | /* |
380 | * Called to set the hrtick timer state. | |
381 | * | |
382 | * called with rq->lock held and irqs disabled | |
383 | */ | |
029632fb | 384 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 385 | { |
31656519 | 386 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 387 | ktime_t time; |
388 | s64 delta; | |
389 | ||
390 | /* | |
391 | * Don't schedule slices shorter than 10000ns, that just | |
392 | * doesn't make sense and can cause timer DoS. | |
393 | */ | |
394 | delta = max_t(s64, delay, 10000LL); | |
395 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 396 | |
cc584b21 | 397 | hrtimer_set_expires(timer, time); |
31656519 PZ |
398 | |
399 | if (rq == this_rq()) { | |
971ee28c | 400 | __hrtick_restart(rq); |
31656519 | 401 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 402 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
403 | rq->hrtick_csd_pending = 1; |
404 | } | |
b328ca18 PZ |
405 | } |
406 | ||
407 | static int | |
408 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
409 | { | |
410 | int cpu = (int)(long)hcpu; | |
411 | ||
412 | switch (action) { | |
413 | case CPU_UP_CANCELED: | |
414 | case CPU_UP_CANCELED_FROZEN: | |
415 | case CPU_DOWN_PREPARE: | |
416 | case CPU_DOWN_PREPARE_FROZEN: | |
417 | case CPU_DEAD: | |
418 | case CPU_DEAD_FROZEN: | |
31656519 | 419 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
420 | return NOTIFY_OK; |
421 | } | |
422 | ||
423 | return NOTIFY_DONE; | |
424 | } | |
425 | ||
fa748203 | 426 | static __init void init_hrtick(void) |
b328ca18 PZ |
427 | { |
428 | hotcpu_notifier(hotplug_hrtick, 0); | |
429 | } | |
31656519 PZ |
430 | #else |
431 | /* | |
432 | * Called to set the hrtick timer state. | |
433 | * | |
434 | * called with rq->lock held and irqs disabled | |
435 | */ | |
029632fb | 436 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 437 | { |
86893335 WL |
438 | /* |
439 | * Don't schedule slices shorter than 10000ns, that just | |
440 | * doesn't make sense. Rely on vruntime for fairness. | |
441 | */ | |
442 | delay = max_t(u64, delay, 10000LL); | |
7f1e2ca9 | 443 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 444 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 445 | } |
b328ca18 | 446 | |
006c75f1 | 447 | static inline void init_hrtick(void) |
8f4d37ec | 448 | { |
8f4d37ec | 449 | } |
31656519 | 450 | #endif /* CONFIG_SMP */ |
8f4d37ec | 451 | |
31656519 | 452 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 453 | { |
31656519 PZ |
454 | #ifdef CONFIG_SMP |
455 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 456 | |
31656519 PZ |
457 | rq->hrtick_csd.flags = 0; |
458 | rq->hrtick_csd.func = __hrtick_start; | |
459 | rq->hrtick_csd.info = rq; | |
460 | #endif | |
8f4d37ec | 461 | |
31656519 PZ |
462 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
463 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 464 | } |
006c75f1 | 465 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
466 | static inline void hrtick_clear(struct rq *rq) |
467 | { | |
468 | } | |
469 | ||
8f4d37ec PZ |
470 | static inline void init_rq_hrtick(struct rq *rq) |
471 | { | |
472 | } | |
473 | ||
b328ca18 PZ |
474 | static inline void init_hrtick(void) |
475 | { | |
476 | } | |
006c75f1 | 477 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 478 | |
fd99f91a PZ |
479 | /* |
480 | * cmpxchg based fetch_or, macro so it works for different integer types | |
481 | */ | |
482 | #define fetch_or(ptr, val) \ | |
483 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
484 | for (;;) { \ | |
485 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
486 | if (__old == __val) \ | |
487 | break; \ | |
488 | __val = __old; \ | |
489 | } \ | |
490 | __old; \ | |
491 | }) | |
492 | ||
e3baac47 | 493 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
494 | /* |
495 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
496 | * this avoids any races wrt polling state changes and thereby avoids | |
497 | * spurious IPIs. | |
498 | */ | |
499 | static bool set_nr_and_not_polling(struct task_struct *p) | |
500 | { | |
501 | struct thread_info *ti = task_thread_info(p); | |
502 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
503 | } | |
e3baac47 PZ |
504 | |
505 | /* | |
506 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
507 | * | |
508 | * If this returns true, then the idle task promises to call | |
509 | * sched_ttwu_pending() and reschedule soon. | |
510 | */ | |
511 | static bool set_nr_if_polling(struct task_struct *p) | |
512 | { | |
513 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 514 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
515 | |
516 | for (;;) { | |
517 | if (!(val & _TIF_POLLING_NRFLAG)) | |
518 | return false; | |
519 | if (val & _TIF_NEED_RESCHED) | |
520 | return true; | |
521 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
522 | if (old == val) | |
523 | break; | |
524 | val = old; | |
525 | } | |
526 | return true; | |
527 | } | |
528 | ||
fd99f91a PZ |
529 | #else |
530 | static bool set_nr_and_not_polling(struct task_struct *p) | |
531 | { | |
532 | set_tsk_need_resched(p); | |
533 | return true; | |
534 | } | |
e3baac47 PZ |
535 | |
536 | #ifdef CONFIG_SMP | |
537 | static bool set_nr_if_polling(struct task_struct *p) | |
538 | { | |
539 | return false; | |
540 | } | |
541 | #endif | |
fd99f91a PZ |
542 | #endif |
543 | ||
76751049 PZ |
544 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
545 | { | |
546 | struct wake_q_node *node = &task->wake_q; | |
547 | ||
548 | /* | |
549 | * Atomically grab the task, if ->wake_q is !nil already it means | |
550 | * its already queued (either by us or someone else) and will get the | |
551 | * wakeup due to that. | |
552 | * | |
553 | * This cmpxchg() implies a full barrier, which pairs with the write | |
554 | * barrier implied by the wakeup in wake_up_list(). | |
555 | */ | |
556 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
557 | return; | |
558 | ||
559 | get_task_struct(task); | |
560 | ||
561 | /* | |
562 | * The head is context local, there can be no concurrency. | |
563 | */ | |
564 | *head->lastp = node; | |
565 | head->lastp = &node->next; | |
566 | } | |
567 | ||
568 | void wake_up_q(struct wake_q_head *head) | |
569 | { | |
570 | struct wake_q_node *node = head->first; | |
571 | ||
572 | while (node != WAKE_Q_TAIL) { | |
573 | struct task_struct *task; | |
574 | ||
575 | task = container_of(node, struct task_struct, wake_q); | |
576 | BUG_ON(!task); | |
577 | /* task can safely be re-inserted now */ | |
578 | node = node->next; | |
579 | task->wake_q.next = NULL; | |
580 | ||
581 | /* | |
582 | * wake_up_process() implies a wmb() to pair with the queueing | |
583 | * in wake_q_add() so as not to miss wakeups. | |
584 | */ | |
585 | wake_up_process(task); | |
586 | put_task_struct(task); | |
587 | } | |
588 | } | |
589 | ||
c24d20db | 590 | /* |
8875125e | 591 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
592 | * |
593 | * On UP this means the setting of the need_resched flag, on SMP it | |
594 | * might also involve a cross-CPU call to trigger the scheduler on | |
595 | * the target CPU. | |
596 | */ | |
8875125e | 597 | void resched_curr(struct rq *rq) |
c24d20db | 598 | { |
8875125e | 599 | struct task_struct *curr = rq->curr; |
c24d20db IM |
600 | int cpu; |
601 | ||
8875125e | 602 | lockdep_assert_held(&rq->lock); |
c24d20db | 603 | |
8875125e | 604 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
605 | return; |
606 | ||
8875125e | 607 | cpu = cpu_of(rq); |
fd99f91a | 608 | |
f27dde8d | 609 | if (cpu == smp_processor_id()) { |
8875125e | 610 | set_tsk_need_resched(curr); |
f27dde8d | 611 | set_preempt_need_resched(); |
c24d20db | 612 | return; |
f27dde8d | 613 | } |
c24d20db | 614 | |
8875125e | 615 | if (set_nr_and_not_polling(curr)) |
c24d20db | 616 | smp_send_reschedule(cpu); |
dfc68f29 AL |
617 | else |
618 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
619 | } |
620 | ||
029632fb | 621 | void resched_cpu(int cpu) |
c24d20db IM |
622 | { |
623 | struct rq *rq = cpu_rq(cpu); | |
624 | unsigned long flags; | |
625 | ||
05fa785c | 626 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 627 | return; |
8875125e | 628 | resched_curr(rq); |
05fa785c | 629 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 630 | } |
06d8308c | 631 | |
b021fe3e | 632 | #ifdef CONFIG_SMP |
3451d024 | 633 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
634 | /* |
635 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
636 | * from an idle cpu. This is good for power-savings. | |
637 | * | |
638 | * We don't do similar optimization for completely idle system, as | |
639 | * selecting an idle cpu will add more delays to the timers than intended | |
640 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
641 | */ | |
6201b4d6 | 642 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
643 | { |
644 | int cpu = smp_processor_id(); | |
645 | int i; | |
646 | struct sched_domain *sd; | |
647 | ||
6201b4d6 VK |
648 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
649 | return cpu; | |
650 | ||
057f3fad | 651 | rcu_read_lock(); |
83cd4fe2 | 652 | for_each_domain(cpu, sd) { |
057f3fad PZ |
653 | for_each_cpu(i, sched_domain_span(sd)) { |
654 | if (!idle_cpu(i)) { | |
655 | cpu = i; | |
656 | goto unlock; | |
657 | } | |
658 | } | |
83cd4fe2 | 659 | } |
057f3fad PZ |
660 | unlock: |
661 | rcu_read_unlock(); | |
83cd4fe2 VP |
662 | return cpu; |
663 | } | |
06d8308c TG |
664 | /* |
665 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
666 | * idle CPU then this timer might expire before the next timer event | |
667 | * which is scheduled to wake up that CPU. In case of a completely | |
668 | * idle system the next event might even be infinite time into the | |
669 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
670 | * leaves the inner idle loop so the newly added timer is taken into | |
671 | * account when the CPU goes back to idle and evaluates the timer | |
672 | * wheel for the next timer event. | |
673 | */ | |
1c20091e | 674 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
675 | { |
676 | struct rq *rq = cpu_rq(cpu); | |
677 | ||
678 | if (cpu == smp_processor_id()) | |
679 | return; | |
680 | ||
67b9ca70 | 681 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 682 | smp_send_reschedule(cpu); |
dfc68f29 AL |
683 | else |
684 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
685 | } |
686 | ||
c5bfece2 | 687 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 688 | { |
53c5fa16 FW |
689 | /* |
690 | * We just need the target to call irq_exit() and re-evaluate | |
691 | * the next tick. The nohz full kick at least implies that. | |
692 | * If needed we can still optimize that later with an | |
693 | * empty IRQ. | |
694 | */ | |
c5bfece2 | 695 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
696 | if (cpu != smp_processor_id() || |
697 | tick_nohz_tick_stopped()) | |
53c5fa16 | 698 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
699 | return true; |
700 | } | |
701 | ||
702 | return false; | |
703 | } | |
704 | ||
705 | void wake_up_nohz_cpu(int cpu) | |
706 | { | |
c5bfece2 | 707 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
708 | wake_up_idle_cpu(cpu); |
709 | } | |
710 | ||
ca38062e | 711 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 712 | { |
1c792db7 | 713 | int cpu = smp_processor_id(); |
873b4c65 VG |
714 | |
715 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
716 | return false; | |
717 | ||
718 | if (idle_cpu(cpu) && !need_resched()) | |
719 | return true; | |
720 | ||
721 | /* | |
722 | * We can't run Idle Load Balance on this CPU for this time so we | |
723 | * cancel it and clear NOHZ_BALANCE_KICK | |
724 | */ | |
725 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
726 | return false; | |
45bf76df IM |
727 | } |
728 | ||
3451d024 | 729 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 730 | |
ca38062e | 731 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 732 | { |
ca38062e | 733 | return false; |
2069dd75 PZ |
734 | } |
735 | ||
3451d024 | 736 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 737 | |
ce831b38 FW |
738 | #ifdef CONFIG_NO_HZ_FULL |
739 | bool sched_can_stop_tick(void) | |
740 | { | |
1e78cdbd RR |
741 | /* |
742 | * FIFO realtime policy runs the highest priority task. Other runnable | |
743 | * tasks are of a lower priority. The scheduler tick does nothing. | |
744 | */ | |
745 | if (current->policy == SCHED_FIFO) | |
746 | return true; | |
747 | ||
748 | /* | |
749 | * Round-robin realtime tasks time slice with other tasks at the same | |
750 | * realtime priority. Is this task the only one at this priority? | |
751 | */ | |
752 | if (current->policy == SCHED_RR) { | |
753 | struct sched_rt_entity *rt_se = ¤t->rt; | |
754 | ||
755 | return rt_se->run_list.prev == rt_se->run_list.next; | |
756 | } | |
757 | ||
3882ec64 FW |
758 | /* |
759 | * More than one running task need preemption. | |
760 | * nr_running update is assumed to be visible | |
761 | * after IPI is sent from wakers. | |
762 | */ | |
541b8264 VK |
763 | if (this_rq()->nr_running > 1) |
764 | return false; | |
ce831b38 | 765 | |
541b8264 | 766 | return true; |
ce831b38 FW |
767 | } |
768 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 769 | |
029632fb | 770 | void sched_avg_update(struct rq *rq) |
18d95a28 | 771 | { |
e9e9250b PZ |
772 | s64 period = sched_avg_period(); |
773 | ||
78becc27 | 774 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
775 | /* |
776 | * Inline assembly required to prevent the compiler | |
777 | * optimising this loop into a divmod call. | |
778 | * See __iter_div_u64_rem() for another example of this. | |
779 | */ | |
780 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
781 | rq->age_stamp += period; |
782 | rq->rt_avg /= 2; | |
783 | } | |
18d95a28 PZ |
784 | } |
785 | ||
6d6bc0ad | 786 | #endif /* CONFIG_SMP */ |
18d95a28 | 787 | |
a790de99 PT |
788 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
789 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 790 | /* |
8277434e PT |
791 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
792 | * node and @up when leaving it for the final time. | |
793 | * | |
794 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 795 | */ |
029632fb | 796 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 797 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
798 | { |
799 | struct task_group *parent, *child; | |
eb755805 | 800 | int ret; |
c09595f6 | 801 | |
8277434e PT |
802 | parent = from; |
803 | ||
c09595f6 | 804 | down: |
eb755805 PZ |
805 | ret = (*down)(parent, data); |
806 | if (ret) | |
8277434e | 807 | goto out; |
c09595f6 PZ |
808 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
809 | parent = child; | |
810 | goto down; | |
811 | ||
812 | up: | |
813 | continue; | |
814 | } | |
eb755805 | 815 | ret = (*up)(parent, data); |
8277434e PT |
816 | if (ret || parent == from) |
817 | goto out; | |
c09595f6 PZ |
818 | |
819 | child = parent; | |
820 | parent = parent->parent; | |
821 | if (parent) | |
822 | goto up; | |
8277434e | 823 | out: |
eb755805 | 824 | return ret; |
c09595f6 PZ |
825 | } |
826 | ||
029632fb | 827 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 828 | { |
e2b245f8 | 829 | return 0; |
eb755805 | 830 | } |
18d95a28 PZ |
831 | #endif |
832 | ||
45bf76df IM |
833 | static void set_load_weight(struct task_struct *p) |
834 | { | |
f05998d4 NR |
835 | int prio = p->static_prio - MAX_RT_PRIO; |
836 | struct load_weight *load = &p->se.load; | |
837 | ||
dd41f596 IM |
838 | /* |
839 | * SCHED_IDLE tasks get minimal weight: | |
840 | */ | |
841 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 842 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 843 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
844 | return; |
845 | } | |
71f8bd46 | 846 | |
c8b28116 | 847 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 848 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
849 | } |
850 | ||
371fd7e7 | 851 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 852 | { |
a64692a3 | 853 | update_rq_clock(rq); |
43148951 | 854 | sched_info_queued(rq, p); |
371fd7e7 | 855 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
856 | } |
857 | ||
371fd7e7 | 858 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 859 | { |
a64692a3 | 860 | update_rq_clock(rq); |
43148951 | 861 | sched_info_dequeued(rq, p); |
371fd7e7 | 862 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
863 | } |
864 | ||
029632fb | 865 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
866 | { |
867 | if (task_contributes_to_load(p)) | |
868 | rq->nr_uninterruptible--; | |
869 | ||
371fd7e7 | 870 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
871 | } |
872 | ||
029632fb | 873 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
874 | { |
875 | if (task_contributes_to_load(p)) | |
876 | rq->nr_uninterruptible++; | |
877 | ||
371fd7e7 | 878 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
879 | } |
880 | ||
fe44d621 | 881 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 882 | { |
095c0aa8 GC |
883 | /* |
884 | * In theory, the compile should just see 0 here, and optimize out the call | |
885 | * to sched_rt_avg_update. But I don't trust it... | |
886 | */ | |
887 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
888 | s64 steal = 0, irq_delta = 0; | |
889 | #endif | |
890 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 891 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
892 | |
893 | /* | |
894 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
895 | * this case when a previous update_rq_clock() happened inside a | |
896 | * {soft,}irq region. | |
897 | * | |
898 | * When this happens, we stop ->clock_task and only update the | |
899 | * prev_irq_time stamp to account for the part that fit, so that a next | |
900 | * update will consume the rest. This ensures ->clock_task is | |
901 | * monotonic. | |
902 | * | |
903 | * It does however cause some slight miss-attribution of {soft,}irq | |
904 | * time, a more accurate solution would be to update the irq_time using | |
905 | * the current rq->clock timestamp, except that would require using | |
906 | * atomic ops. | |
907 | */ | |
908 | if (irq_delta > delta) | |
909 | irq_delta = delta; | |
910 | ||
911 | rq->prev_irq_time += irq_delta; | |
912 | delta -= irq_delta; | |
095c0aa8 GC |
913 | #endif |
914 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 915 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
916 | steal = paravirt_steal_clock(cpu_of(rq)); |
917 | steal -= rq->prev_steal_time_rq; | |
918 | ||
919 | if (unlikely(steal > delta)) | |
920 | steal = delta; | |
921 | ||
095c0aa8 | 922 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
923 | delta -= steal; |
924 | } | |
925 | #endif | |
926 | ||
fe44d621 PZ |
927 | rq->clock_task += delta; |
928 | ||
095c0aa8 | 929 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 930 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
931 | sched_rt_avg_update(rq, irq_delta + steal); |
932 | #endif | |
aa483808 VP |
933 | } |
934 | ||
34f971f6 PZ |
935 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
936 | { | |
937 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
938 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
939 | ||
940 | if (stop) { | |
941 | /* | |
942 | * Make it appear like a SCHED_FIFO task, its something | |
943 | * userspace knows about and won't get confused about. | |
944 | * | |
945 | * Also, it will make PI more or less work without too | |
946 | * much confusion -- but then, stop work should not | |
947 | * rely on PI working anyway. | |
948 | */ | |
949 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
950 | ||
951 | stop->sched_class = &stop_sched_class; | |
952 | } | |
953 | ||
954 | cpu_rq(cpu)->stop = stop; | |
955 | ||
956 | if (old_stop) { | |
957 | /* | |
958 | * Reset it back to a normal scheduling class so that | |
959 | * it can die in pieces. | |
960 | */ | |
961 | old_stop->sched_class = &rt_sched_class; | |
962 | } | |
963 | } | |
964 | ||
14531189 | 965 | /* |
dd41f596 | 966 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 967 | */ |
14531189 IM |
968 | static inline int __normal_prio(struct task_struct *p) |
969 | { | |
dd41f596 | 970 | return p->static_prio; |
14531189 IM |
971 | } |
972 | ||
b29739f9 IM |
973 | /* |
974 | * Calculate the expected normal priority: i.e. priority | |
975 | * without taking RT-inheritance into account. Might be | |
976 | * boosted by interactivity modifiers. Changes upon fork, | |
977 | * setprio syscalls, and whenever the interactivity | |
978 | * estimator recalculates. | |
979 | */ | |
36c8b586 | 980 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
981 | { |
982 | int prio; | |
983 | ||
aab03e05 DF |
984 | if (task_has_dl_policy(p)) |
985 | prio = MAX_DL_PRIO-1; | |
986 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
987 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
988 | else | |
989 | prio = __normal_prio(p); | |
990 | return prio; | |
991 | } | |
992 | ||
993 | /* | |
994 | * Calculate the current priority, i.e. the priority | |
995 | * taken into account by the scheduler. This value might | |
996 | * be boosted by RT tasks, or might be boosted by | |
997 | * interactivity modifiers. Will be RT if the task got | |
998 | * RT-boosted. If not then it returns p->normal_prio. | |
999 | */ | |
36c8b586 | 1000 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1001 | { |
1002 | p->normal_prio = normal_prio(p); | |
1003 | /* | |
1004 | * If we are RT tasks or we were boosted to RT priority, | |
1005 | * keep the priority unchanged. Otherwise, update priority | |
1006 | * to the normal priority: | |
1007 | */ | |
1008 | if (!rt_prio(p->prio)) | |
1009 | return p->normal_prio; | |
1010 | return p->prio; | |
1011 | } | |
1012 | ||
1da177e4 LT |
1013 | /** |
1014 | * task_curr - is this task currently executing on a CPU? | |
1015 | * @p: the task in question. | |
e69f6186 YB |
1016 | * |
1017 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1018 | */ |
36c8b586 | 1019 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1020 | { |
1021 | return cpu_curr(task_cpu(p)) == p; | |
1022 | } | |
1023 | ||
67dfa1b7 KT |
1024 | /* |
1025 | * Can drop rq->lock because from sched_class::switched_from() methods drop it. | |
1026 | */ | |
cb469845 SR |
1027 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1028 | const struct sched_class *prev_class, | |
da7a735e | 1029 | int oldprio) |
cb469845 SR |
1030 | { |
1031 | if (prev_class != p->sched_class) { | |
1032 | if (prev_class->switched_from) | |
da7a735e | 1033 | prev_class->switched_from(rq, p); |
67dfa1b7 | 1034 | /* Possble rq->lock 'hole'. */ |
da7a735e | 1035 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1036 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1037 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1038 | } |
1039 | ||
029632fb | 1040 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1041 | { |
1042 | const struct sched_class *class; | |
1043 | ||
1044 | if (p->sched_class == rq->curr->sched_class) { | |
1045 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1046 | } else { | |
1047 | for_each_class(class) { | |
1048 | if (class == rq->curr->sched_class) | |
1049 | break; | |
1050 | if (class == p->sched_class) { | |
8875125e | 1051 | resched_curr(rq); |
1e5a7405 PZ |
1052 | break; |
1053 | } | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * A queue event has occurred, and we're going to schedule. In | |
1059 | * this case, we can save a useless back to back clock update. | |
1060 | */ | |
da0c1e65 | 1061 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 1062 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
1063 | } |
1064 | ||
1da177e4 | 1065 | #ifdef CONFIG_SMP |
dd41f596 | 1066 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1067 | { |
e2912009 PZ |
1068 | #ifdef CONFIG_SCHED_DEBUG |
1069 | /* | |
1070 | * We should never call set_task_cpu() on a blocked task, | |
1071 | * ttwu() will sort out the placement. | |
1072 | */ | |
077614ee | 1073 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1074 | !p->on_rq); |
0122ec5b PZ |
1075 | |
1076 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1077 | /* |
1078 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1079 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1080 | * | |
1081 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1082 | * see task_group(). |
6c6c54e1 PZ |
1083 | * |
1084 | * Furthermore, all task_rq users should acquire both locks, see | |
1085 | * task_rq_lock(). | |
1086 | */ | |
0122ec5b PZ |
1087 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1088 | lockdep_is_held(&task_rq(p)->lock))); | |
1089 | #endif | |
e2912009 PZ |
1090 | #endif |
1091 | ||
de1d7286 | 1092 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1093 | |
0c69774e | 1094 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1095 | if (p->sched_class->migrate_task_rq) |
1096 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1097 | p->se.nr_migrations++; |
86038c5e | 1098 | perf_sw_event_sched(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 0); |
0c69774e | 1099 | } |
dd41f596 IM |
1100 | |
1101 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1102 | } |
1103 | ||
ac66f547 PZ |
1104 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1105 | { | |
da0c1e65 | 1106 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1107 | struct rq *src_rq, *dst_rq; |
1108 | ||
1109 | src_rq = task_rq(p); | |
1110 | dst_rq = cpu_rq(cpu); | |
1111 | ||
1112 | deactivate_task(src_rq, p, 0); | |
1113 | set_task_cpu(p, cpu); | |
1114 | activate_task(dst_rq, p, 0); | |
1115 | check_preempt_curr(dst_rq, p, 0); | |
1116 | } else { | |
1117 | /* | |
1118 | * Task isn't running anymore; make it appear like we migrated | |
1119 | * it before it went to sleep. This means on wakeup we make the | |
1120 | * previous cpu our targer instead of where it really is. | |
1121 | */ | |
1122 | p->wake_cpu = cpu; | |
1123 | } | |
1124 | } | |
1125 | ||
1126 | struct migration_swap_arg { | |
1127 | struct task_struct *src_task, *dst_task; | |
1128 | int src_cpu, dst_cpu; | |
1129 | }; | |
1130 | ||
1131 | static int migrate_swap_stop(void *data) | |
1132 | { | |
1133 | struct migration_swap_arg *arg = data; | |
1134 | struct rq *src_rq, *dst_rq; | |
1135 | int ret = -EAGAIN; | |
1136 | ||
1137 | src_rq = cpu_rq(arg->src_cpu); | |
1138 | dst_rq = cpu_rq(arg->dst_cpu); | |
1139 | ||
74602315 PZ |
1140 | double_raw_lock(&arg->src_task->pi_lock, |
1141 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1142 | double_rq_lock(src_rq, dst_rq); |
1143 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1144 | goto unlock; | |
1145 | ||
1146 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1147 | goto unlock; | |
1148 | ||
1149 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1150 | goto unlock; | |
1151 | ||
1152 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1153 | goto unlock; | |
1154 | ||
1155 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1156 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1157 | ||
1158 | ret = 0; | |
1159 | ||
1160 | unlock: | |
1161 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1162 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1163 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1164 | |
1165 | return ret; | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Cross migrate two tasks | |
1170 | */ | |
1171 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1172 | { | |
1173 | struct migration_swap_arg arg; | |
1174 | int ret = -EINVAL; | |
1175 | ||
ac66f547 PZ |
1176 | arg = (struct migration_swap_arg){ |
1177 | .src_task = cur, | |
1178 | .src_cpu = task_cpu(cur), | |
1179 | .dst_task = p, | |
1180 | .dst_cpu = task_cpu(p), | |
1181 | }; | |
1182 | ||
1183 | if (arg.src_cpu == arg.dst_cpu) | |
1184 | goto out; | |
1185 | ||
6acce3ef PZ |
1186 | /* |
1187 | * These three tests are all lockless; this is OK since all of them | |
1188 | * will be re-checked with proper locks held further down the line. | |
1189 | */ | |
ac66f547 PZ |
1190 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1191 | goto out; | |
1192 | ||
1193 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1194 | goto out; | |
1195 | ||
1196 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1197 | goto out; | |
1198 | ||
286549dc | 1199 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1200 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1201 | ||
1202 | out: | |
ac66f547 PZ |
1203 | return ret; |
1204 | } | |
1205 | ||
969c7921 | 1206 | struct migration_arg { |
36c8b586 | 1207 | struct task_struct *task; |
1da177e4 | 1208 | int dest_cpu; |
70b97a7f | 1209 | }; |
1da177e4 | 1210 | |
969c7921 TH |
1211 | static int migration_cpu_stop(void *data); |
1212 | ||
1da177e4 LT |
1213 | /* |
1214 | * wait_task_inactive - wait for a thread to unschedule. | |
1215 | * | |
85ba2d86 RM |
1216 | * If @match_state is nonzero, it's the @p->state value just checked and |
1217 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1218 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1219 | * we return a positive number (its total switch count). If a second call | |
1220 | * a short while later returns the same number, the caller can be sure that | |
1221 | * @p has remained unscheduled the whole time. | |
1222 | * | |
1da177e4 LT |
1223 | * The caller must ensure that the task *will* unschedule sometime soon, |
1224 | * else this function might spin for a *long* time. This function can't | |
1225 | * be called with interrupts off, or it may introduce deadlock with | |
1226 | * smp_call_function() if an IPI is sent by the same process we are | |
1227 | * waiting to become inactive. | |
1228 | */ | |
85ba2d86 | 1229 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1230 | { |
1231 | unsigned long flags; | |
da0c1e65 | 1232 | int running, queued; |
85ba2d86 | 1233 | unsigned long ncsw; |
70b97a7f | 1234 | struct rq *rq; |
1da177e4 | 1235 | |
3a5c359a AK |
1236 | for (;;) { |
1237 | /* | |
1238 | * We do the initial early heuristics without holding | |
1239 | * any task-queue locks at all. We'll only try to get | |
1240 | * the runqueue lock when things look like they will | |
1241 | * work out! | |
1242 | */ | |
1243 | rq = task_rq(p); | |
fa490cfd | 1244 | |
3a5c359a AK |
1245 | /* |
1246 | * If the task is actively running on another CPU | |
1247 | * still, just relax and busy-wait without holding | |
1248 | * any locks. | |
1249 | * | |
1250 | * NOTE! Since we don't hold any locks, it's not | |
1251 | * even sure that "rq" stays as the right runqueue! | |
1252 | * But we don't care, since "task_running()" will | |
1253 | * return false if the runqueue has changed and p | |
1254 | * is actually now running somewhere else! | |
1255 | */ | |
85ba2d86 RM |
1256 | while (task_running(rq, p)) { |
1257 | if (match_state && unlikely(p->state != match_state)) | |
1258 | return 0; | |
3a5c359a | 1259 | cpu_relax(); |
85ba2d86 | 1260 | } |
fa490cfd | 1261 | |
3a5c359a AK |
1262 | /* |
1263 | * Ok, time to look more closely! We need the rq | |
1264 | * lock now, to be *sure*. If we're wrong, we'll | |
1265 | * just go back and repeat. | |
1266 | */ | |
1267 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1268 | trace_sched_wait_task(p); |
3a5c359a | 1269 | running = task_running(rq, p); |
da0c1e65 | 1270 | queued = task_on_rq_queued(p); |
85ba2d86 | 1271 | ncsw = 0; |
f31e11d8 | 1272 | if (!match_state || p->state == match_state) |
93dcf55f | 1273 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1274 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1275 | |
85ba2d86 RM |
1276 | /* |
1277 | * If it changed from the expected state, bail out now. | |
1278 | */ | |
1279 | if (unlikely(!ncsw)) | |
1280 | break; | |
1281 | ||
3a5c359a AK |
1282 | /* |
1283 | * Was it really running after all now that we | |
1284 | * checked with the proper locks actually held? | |
1285 | * | |
1286 | * Oops. Go back and try again.. | |
1287 | */ | |
1288 | if (unlikely(running)) { | |
1289 | cpu_relax(); | |
1290 | continue; | |
1291 | } | |
fa490cfd | 1292 | |
3a5c359a AK |
1293 | /* |
1294 | * It's not enough that it's not actively running, | |
1295 | * it must be off the runqueue _entirely_, and not | |
1296 | * preempted! | |
1297 | * | |
80dd99b3 | 1298 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1299 | * running right now), it's preempted, and we should |
1300 | * yield - it could be a while. | |
1301 | */ | |
da0c1e65 | 1302 | if (unlikely(queued)) { |
8eb90c30 TG |
1303 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1304 | ||
1305 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1306 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1307 | continue; |
1308 | } | |
fa490cfd | 1309 | |
3a5c359a AK |
1310 | /* |
1311 | * Ahh, all good. It wasn't running, and it wasn't | |
1312 | * runnable, which means that it will never become | |
1313 | * running in the future either. We're all done! | |
1314 | */ | |
1315 | break; | |
1316 | } | |
85ba2d86 RM |
1317 | |
1318 | return ncsw; | |
1da177e4 LT |
1319 | } |
1320 | ||
1321 | /*** | |
1322 | * kick_process - kick a running thread to enter/exit the kernel | |
1323 | * @p: the to-be-kicked thread | |
1324 | * | |
1325 | * Cause a process which is running on another CPU to enter | |
1326 | * kernel-mode, without any delay. (to get signals handled.) | |
1327 | * | |
25985edc | 1328 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1329 | * because all it wants to ensure is that the remote task enters |
1330 | * the kernel. If the IPI races and the task has been migrated | |
1331 | * to another CPU then no harm is done and the purpose has been | |
1332 | * achieved as well. | |
1333 | */ | |
36c8b586 | 1334 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1335 | { |
1336 | int cpu; | |
1337 | ||
1338 | preempt_disable(); | |
1339 | cpu = task_cpu(p); | |
1340 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1341 | smp_send_reschedule(cpu); | |
1342 | preempt_enable(); | |
1343 | } | |
b43e3521 | 1344 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1345 | #endif /* CONFIG_SMP */ |
1da177e4 | 1346 | |
970b13ba | 1347 | #ifdef CONFIG_SMP |
30da688e | 1348 | /* |
013fdb80 | 1349 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1350 | */ |
5da9a0fb PZ |
1351 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1352 | { | |
aa00d89c TC |
1353 | int nid = cpu_to_node(cpu); |
1354 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1355 | enum { cpuset, possible, fail } state = cpuset; |
1356 | int dest_cpu; | |
5da9a0fb | 1357 | |
aa00d89c TC |
1358 | /* |
1359 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1360 | * will return -1. There is no cpu on the node, and we should | |
1361 | * select the cpu on the other node. | |
1362 | */ | |
1363 | if (nid != -1) { | |
1364 | nodemask = cpumask_of_node(nid); | |
1365 | ||
1366 | /* Look for allowed, online CPU in same node. */ | |
1367 | for_each_cpu(dest_cpu, nodemask) { | |
1368 | if (!cpu_online(dest_cpu)) | |
1369 | continue; | |
1370 | if (!cpu_active(dest_cpu)) | |
1371 | continue; | |
1372 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1373 | return dest_cpu; | |
1374 | } | |
2baab4e9 | 1375 | } |
5da9a0fb | 1376 | |
2baab4e9 PZ |
1377 | for (;;) { |
1378 | /* Any allowed, online CPU? */ | |
e3831edd | 1379 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1380 | if (!cpu_online(dest_cpu)) |
1381 | continue; | |
1382 | if (!cpu_active(dest_cpu)) | |
1383 | continue; | |
1384 | goto out; | |
1385 | } | |
5da9a0fb | 1386 | |
2baab4e9 PZ |
1387 | switch (state) { |
1388 | case cpuset: | |
1389 | /* No more Mr. Nice Guy. */ | |
1390 | cpuset_cpus_allowed_fallback(p); | |
1391 | state = possible; | |
1392 | break; | |
1393 | ||
1394 | case possible: | |
1395 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1396 | state = fail; | |
1397 | break; | |
1398 | ||
1399 | case fail: | |
1400 | BUG(); | |
1401 | break; | |
1402 | } | |
1403 | } | |
1404 | ||
1405 | out: | |
1406 | if (state != cpuset) { | |
1407 | /* | |
1408 | * Don't tell them about moving exiting tasks or | |
1409 | * kernel threads (both mm NULL), since they never | |
1410 | * leave kernel. | |
1411 | */ | |
1412 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1413 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1414 | task_pid_nr(p), p->comm, cpu); |
1415 | } | |
5da9a0fb PZ |
1416 | } |
1417 | ||
1418 | return dest_cpu; | |
1419 | } | |
1420 | ||
e2912009 | 1421 | /* |
013fdb80 | 1422 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1423 | */ |
970b13ba | 1424 | static inline |
ac66f547 | 1425 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1426 | { |
6c1d9410 WL |
1427 | if (p->nr_cpus_allowed > 1) |
1428 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); | |
e2912009 PZ |
1429 | |
1430 | /* | |
1431 | * In order not to call set_task_cpu() on a blocking task we need | |
1432 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1433 | * cpu. | |
1434 | * | |
1435 | * Since this is common to all placement strategies, this lives here. | |
1436 | * | |
1437 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1438 | * not worry about this generic constraint ] | |
1439 | */ | |
fa17b507 | 1440 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1441 | !cpu_online(cpu))) |
5da9a0fb | 1442 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1443 | |
1444 | return cpu; | |
970b13ba | 1445 | } |
09a40af5 MG |
1446 | |
1447 | static void update_avg(u64 *avg, u64 sample) | |
1448 | { | |
1449 | s64 diff = sample - *avg; | |
1450 | *avg += diff >> 3; | |
1451 | } | |
970b13ba PZ |
1452 | #endif |
1453 | ||
d7c01d27 | 1454 | static void |
b84cb5df | 1455 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1456 | { |
d7c01d27 | 1457 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1458 | struct rq *rq = this_rq(); |
1459 | ||
d7c01d27 PZ |
1460 | #ifdef CONFIG_SMP |
1461 | int this_cpu = smp_processor_id(); | |
1462 | ||
1463 | if (cpu == this_cpu) { | |
1464 | schedstat_inc(rq, ttwu_local); | |
1465 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1466 | } else { | |
1467 | struct sched_domain *sd; | |
1468 | ||
1469 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1470 | rcu_read_lock(); |
d7c01d27 PZ |
1471 | for_each_domain(this_cpu, sd) { |
1472 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1473 | schedstat_inc(sd, ttwu_wake_remote); | |
1474 | break; | |
1475 | } | |
1476 | } | |
057f3fad | 1477 | rcu_read_unlock(); |
d7c01d27 | 1478 | } |
f339b9dc PZ |
1479 | |
1480 | if (wake_flags & WF_MIGRATED) | |
1481 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1482 | ||
d7c01d27 PZ |
1483 | #endif /* CONFIG_SMP */ |
1484 | ||
1485 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1486 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1487 | |
1488 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1489 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1490 | |
d7c01d27 PZ |
1491 | #endif /* CONFIG_SCHEDSTATS */ |
1492 | } | |
1493 | ||
1494 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1495 | { | |
9ed3811a | 1496 | activate_task(rq, p, en_flags); |
da0c1e65 | 1497 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1498 | |
1499 | /* if a worker is waking up, notify workqueue */ | |
1500 | if (p->flags & PF_WQ_WORKER) | |
1501 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1502 | } |
1503 | ||
23f41eeb PZ |
1504 | /* |
1505 | * Mark the task runnable and perform wakeup-preemption. | |
1506 | */ | |
89363381 | 1507 | static void |
23f41eeb | 1508 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1509 | { |
9ed3811a | 1510 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1511 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1512 | |
1513 | p->state = TASK_RUNNING; | |
1514 | #ifdef CONFIG_SMP | |
1515 | if (p->sched_class->task_woken) | |
1516 | p->sched_class->task_woken(rq, p); | |
1517 | ||
e69c6341 | 1518 | if (rq->idle_stamp) { |
78becc27 | 1519 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1520 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1521 | |
abfafa54 JL |
1522 | update_avg(&rq->avg_idle, delta); |
1523 | ||
1524 | if (rq->avg_idle > max) | |
9ed3811a | 1525 | rq->avg_idle = max; |
abfafa54 | 1526 | |
9ed3811a TH |
1527 | rq->idle_stamp = 0; |
1528 | } | |
1529 | #endif | |
1530 | } | |
1531 | ||
c05fbafb PZ |
1532 | static void |
1533 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1534 | { | |
1535 | #ifdef CONFIG_SMP | |
1536 | if (p->sched_contributes_to_load) | |
1537 | rq->nr_uninterruptible--; | |
1538 | #endif | |
1539 | ||
1540 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1541 | ttwu_do_wakeup(rq, p, wake_flags); | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1546 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1547 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1548 | * the task is still ->on_rq. | |
1549 | */ | |
1550 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1551 | { | |
1552 | struct rq *rq; | |
1553 | int ret = 0; | |
1554 | ||
1555 | rq = __task_rq_lock(p); | |
da0c1e65 | 1556 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1557 | /* check_preempt_curr() may use rq clock */ |
1558 | update_rq_clock(rq); | |
c05fbafb PZ |
1559 | ttwu_do_wakeup(rq, p, wake_flags); |
1560 | ret = 1; | |
1561 | } | |
1562 | __task_rq_unlock(rq); | |
1563 | ||
1564 | return ret; | |
1565 | } | |
1566 | ||
317f3941 | 1567 | #ifdef CONFIG_SMP |
e3baac47 | 1568 | void sched_ttwu_pending(void) |
317f3941 PZ |
1569 | { |
1570 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1571 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1572 | struct task_struct *p; | |
e3baac47 | 1573 | unsigned long flags; |
317f3941 | 1574 | |
e3baac47 PZ |
1575 | if (!llist) |
1576 | return; | |
1577 | ||
1578 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1579 | |
fa14ff4a PZ |
1580 | while (llist) { |
1581 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1582 | llist = llist_next(llist); | |
317f3941 PZ |
1583 | ttwu_do_activate(rq, p, 0); |
1584 | } | |
1585 | ||
e3baac47 | 1586 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1587 | } |
1588 | ||
1589 | void scheduler_ipi(void) | |
1590 | { | |
f27dde8d PZ |
1591 | /* |
1592 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1593 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1594 | * this IPI. | |
1595 | */ | |
8cb75e0c | 1596 | preempt_fold_need_resched(); |
f27dde8d | 1597 | |
fd2ac4f4 | 1598 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1599 | return; |
1600 | ||
1601 | /* | |
1602 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1603 | * traditionally all their work was done from the interrupt return | |
1604 | * path. Now that we actually do some work, we need to make sure | |
1605 | * we do call them. | |
1606 | * | |
1607 | * Some archs already do call them, luckily irq_enter/exit nest | |
1608 | * properly. | |
1609 | * | |
1610 | * Arguably we should visit all archs and update all handlers, | |
1611 | * however a fair share of IPIs are still resched only so this would | |
1612 | * somewhat pessimize the simple resched case. | |
1613 | */ | |
1614 | irq_enter(); | |
fa14ff4a | 1615 | sched_ttwu_pending(); |
ca38062e SS |
1616 | |
1617 | /* | |
1618 | * Check if someone kicked us for doing the nohz idle load balance. | |
1619 | */ | |
873b4c65 | 1620 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1621 | this_rq()->idle_balance = 1; |
ca38062e | 1622 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1623 | } |
c5d753a5 | 1624 | irq_exit(); |
317f3941 PZ |
1625 | } |
1626 | ||
1627 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1628 | { | |
e3baac47 PZ |
1629 | struct rq *rq = cpu_rq(cpu); |
1630 | ||
1631 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1632 | if (!set_nr_if_polling(rq->idle)) | |
1633 | smp_send_reschedule(cpu); | |
1634 | else | |
1635 | trace_sched_wake_idle_without_ipi(cpu); | |
1636 | } | |
317f3941 | 1637 | } |
d6aa8f85 | 1638 | |
f6be8af1 CL |
1639 | void wake_up_if_idle(int cpu) |
1640 | { | |
1641 | struct rq *rq = cpu_rq(cpu); | |
1642 | unsigned long flags; | |
1643 | ||
fd7de1e8 AL |
1644 | rcu_read_lock(); |
1645 | ||
1646 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1647 | goto out; | |
f6be8af1 CL |
1648 | |
1649 | if (set_nr_if_polling(rq->idle)) { | |
1650 | trace_sched_wake_idle_without_ipi(cpu); | |
1651 | } else { | |
1652 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1653 | if (is_idle_task(rq->curr)) | |
1654 | smp_send_reschedule(cpu); | |
1655 | /* Else cpu is not in idle, do nothing here */ | |
1656 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
1657 | } | |
fd7de1e8 AL |
1658 | |
1659 | out: | |
1660 | rcu_read_unlock(); | |
f6be8af1 CL |
1661 | } |
1662 | ||
39be3501 | 1663 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1664 | { |
1665 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1666 | } | |
d6aa8f85 | 1667 | #endif /* CONFIG_SMP */ |
317f3941 | 1668 | |
c05fbafb PZ |
1669 | static void ttwu_queue(struct task_struct *p, int cpu) |
1670 | { | |
1671 | struct rq *rq = cpu_rq(cpu); | |
1672 | ||
17d9f311 | 1673 | #if defined(CONFIG_SMP) |
39be3501 | 1674 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1675 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1676 | ttwu_queue_remote(p, cpu); |
1677 | return; | |
1678 | } | |
1679 | #endif | |
1680 | ||
c05fbafb PZ |
1681 | raw_spin_lock(&rq->lock); |
1682 | ttwu_do_activate(rq, p, 0); | |
1683 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1684 | } |
1685 | ||
1686 | /** | |
1da177e4 | 1687 | * try_to_wake_up - wake up a thread |
9ed3811a | 1688 | * @p: the thread to be awakened |
1da177e4 | 1689 | * @state: the mask of task states that can be woken |
9ed3811a | 1690 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1691 | * |
1692 | * Put it on the run-queue if it's not already there. The "current" | |
1693 | * thread is always on the run-queue (except when the actual | |
1694 | * re-schedule is in progress), and as such you're allowed to do | |
1695 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1696 | * runnable without the overhead of this. | |
1697 | * | |
e69f6186 | 1698 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1699 | * or @state didn't match @p's state. |
1da177e4 | 1700 | */ |
e4a52bcb PZ |
1701 | static int |
1702 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1703 | { |
1da177e4 | 1704 | unsigned long flags; |
c05fbafb | 1705 | int cpu, success = 0; |
2398f2c6 | 1706 | |
e0acd0a6 ON |
1707 | /* |
1708 | * If we are going to wake up a thread waiting for CONDITION we | |
1709 | * need to ensure that CONDITION=1 done by the caller can not be | |
1710 | * reordered with p->state check below. This pairs with mb() in | |
1711 | * set_current_state() the waiting thread does. | |
1712 | */ | |
1713 | smp_mb__before_spinlock(); | |
013fdb80 | 1714 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1715 | if (!(p->state & state)) |
1da177e4 LT |
1716 | goto out; |
1717 | ||
c05fbafb | 1718 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1719 | cpu = task_cpu(p); |
1da177e4 | 1720 | |
c05fbafb PZ |
1721 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1722 | goto stat; | |
1da177e4 | 1723 | |
1da177e4 | 1724 | #ifdef CONFIG_SMP |
e9c84311 | 1725 | /* |
c05fbafb PZ |
1726 | * If the owning (remote) cpu is still in the middle of schedule() with |
1727 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1728 | */ |
f3e94786 | 1729 | while (p->on_cpu) |
e4a52bcb | 1730 | cpu_relax(); |
0970d299 | 1731 | /* |
e4a52bcb | 1732 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1733 | */ |
e4a52bcb | 1734 | smp_rmb(); |
1da177e4 | 1735 | |
a8e4f2ea | 1736 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1737 | p->state = TASK_WAKING; |
e7693a36 | 1738 | |
e4a52bcb | 1739 | if (p->sched_class->task_waking) |
74f8e4b2 | 1740 | p->sched_class->task_waking(p); |
efbbd05a | 1741 | |
ac66f547 | 1742 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1743 | if (task_cpu(p) != cpu) { |
1744 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1745 | set_task_cpu(p, cpu); |
f339b9dc | 1746 | } |
1da177e4 | 1747 | #endif /* CONFIG_SMP */ |
1da177e4 | 1748 | |
c05fbafb PZ |
1749 | ttwu_queue(p, cpu); |
1750 | stat: | |
b84cb5df | 1751 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1752 | out: |
013fdb80 | 1753 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1754 | |
1755 | return success; | |
1756 | } | |
1757 | ||
21aa9af0 TH |
1758 | /** |
1759 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1760 | * @p: the thread to be awakened | |
1761 | * | |
2acca55e | 1762 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1763 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1764 | * the current task. |
21aa9af0 TH |
1765 | */ |
1766 | static void try_to_wake_up_local(struct task_struct *p) | |
1767 | { | |
1768 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1769 | |
383efcd0 TH |
1770 | if (WARN_ON_ONCE(rq != this_rq()) || |
1771 | WARN_ON_ONCE(p == current)) | |
1772 | return; | |
1773 | ||
21aa9af0 TH |
1774 | lockdep_assert_held(&rq->lock); |
1775 | ||
2acca55e PZ |
1776 | if (!raw_spin_trylock(&p->pi_lock)) { |
1777 | raw_spin_unlock(&rq->lock); | |
1778 | raw_spin_lock(&p->pi_lock); | |
1779 | raw_spin_lock(&rq->lock); | |
1780 | } | |
1781 | ||
21aa9af0 | 1782 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1783 | goto out; |
21aa9af0 | 1784 | |
da0c1e65 | 1785 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
1786 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1787 | ||
23f41eeb | 1788 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1789 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1790 | out: |
1791 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1792 | } |
1793 | ||
50fa610a DH |
1794 | /** |
1795 | * wake_up_process - Wake up a specific process | |
1796 | * @p: The process to be woken up. | |
1797 | * | |
1798 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1799 | * processes. |
1800 | * | |
1801 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1802 | * |
1803 | * It may be assumed that this function implies a write memory barrier before | |
1804 | * changing the task state if and only if any tasks are woken up. | |
1805 | */ | |
7ad5b3a5 | 1806 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1807 | { |
9067ac85 ON |
1808 | WARN_ON(task_is_stopped_or_traced(p)); |
1809 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1810 | } |
1da177e4 LT |
1811 | EXPORT_SYMBOL(wake_up_process); |
1812 | ||
7ad5b3a5 | 1813 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1814 | { |
1815 | return try_to_wake_up(p, state, 0); | |
1816 | } | |
1817 | ||
a5e7be3b JL |
1818 | /* |
1819 | * This function clears the sched_dl_entity static params. | |
1820 | */ | |
1821 | void __dl_clear_params(struct task_struct *p) | |
1822 | { | |
1823 | struct sched_dl_entity *dl_se = &p->dl; | |
1824 | ||
1825 | dl_se->dl_runtime = 0; | |
1826 | dl_se->dl_deadline = 0; | |
1827 | dl_se->dl_period = 0; | |
1828 | dl_se->flags = 0; | |
1829 | dl_se->dl_bw = 0; | |
40767b0d PZ |
1830 | |
1831 | dl_se->dl_throttled = 0; | |
1832 | dl_se->dl_new = 1; | |
1833 | dl_se->dl_yielded = 0; | |
a5e7be3b JL |
1834 | } |
1835 | ||
1da177e4 LT |
1836 | /* |
1837 | * Perform scheduler related setup for a newly forked process p. | |
1838 | * p is forked by current. | |
dd41f596 IM |
1839 | * |
1840 | * __sched_fork() is basic setup used by init_idle() too: | |
1841 | */ | |
5e1576ed | 1842 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1843 | { |
fd2f4419 PZ |
1844 | p->on_rq = 0; |
1845 | ||
1846 | p->se.on_rq = 0; | |
dd41f596 IM |
1847 | p->se.exec_start = 0; |
1848 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1849 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1850 | p->se.nr_migrations = 0; |
da7a735e | 1851 | p->se.vruntime = 0; |
bb04159d KT |
1852 | #ifdef CONFIG_SMP |
1853 | p->se.avg.decay_count = 0; | |
1854 | #endif | |
fd2f4419 | 1855 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1856 | |
1857 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1858 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1859 | #endif |
476d139c | 1860 | |
aab03e05 | 1861 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 1862 | init_dl_task_timer(&p->dl); |
a5e7be3b | 1863 | __dl_clear_params(p); |
aab03e05 | 1864 | |
fa717060 | 1865 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1866 | |
e107be36 AK |
1867 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1868 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1869 | #endif | |
cbee9f88 PZ |
1870 | |
1871 | #ifdef CONFIG_NUMA_BALANCING | |
1872 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1873 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1874 | p->mm->numa_scan_seq = 0; |
1875 | } | |
1876 | ||
5e1576ed RR |
1877 | if (clone_flags & CLONE_VM) |
1878 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1879 | else | |
1880 | p->numa_preferred_nid = -1; | |
1881 | ||
cbee9f88 PZ |
1882 | p->node_stamp = 0ULL; |
1883 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1884 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1885 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 1886 | p->numa_faults = NULL; |
7e2703e6 RR |
1887 | p->last_task_numa_placement = 0; |
1888 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 1889 | |
8c8a743c | 1890 | p->numa_group = NULL; |
cbee9f88 | 1891 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1892 | } |
1893 | ||
1a687c2e | 1894 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1895 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1896 | void set_numabalancing_state(bool enabled) |
1897 | { | |
1898 | if (enabled) | |
1899 | sched_feat_set("NUMA"); | |
1900 | else | |
1901 | sched_feat_set("NO_NUMA"); | |
1902 | } | |
3105b86a MG |
1903 | #else |
1904 | __read_mostly bool numabalancing_enabled; | |
1905 | ||
1906 | void set_numabalancing_state(bool enabled) | |
1907 | { | |
1908 | numabalancing_enabled = enabled; | |
dd41f596 | 1909 | } |
3105b86a | 1910 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1911 | |
1912 | #ifdef CONFIG_PROC_SYSCTL | |
1913 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1914 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1915 | { | |
1916 | struct ctl_table t; | |
1917 | int err; | |
1918 | int state = numabalancing_enabled; | |
1919 | ||
1920 | if (write && !capable(CAP_SYS_ADMIN)) | |
1921 | return -EPERM; | |
1922 | ||
1923 | t = *table; | |
1924 | t.data = &state; | |
1925 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1926 | if (err < 0) | |
1927 | return err; | |
1928 | if (write) | |
1929 | set_numabalancing_state(state); | |
1930 | return err; | |
1931 | } | |
1932 | #endif | |
1933 | #endif | |
dd41f596 IM |
1934 | |
1935 | /* | |
1936 | * fork()/clone()-time setup: | |
1937 | */ | |
aab03e05 | 1938 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1939 | { |
0122ec5b | 1940 | unsigned long flags; |
dd41f596 IM |
1941 | int cpu = get_cpu(); |
1942 | ||
5e1576ed | 1943 | __sched_fork(clone_flags, p); |
06b83b5f | 1944 | /* |
0017d735 | 1945 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1946 | * nobody will actually run it, and a signal or other external |
1947 | * event cannot wake it up and insert it on the runqueue either. | |
1948 | */ | |
0017d735 | 1949 | p->state = TASK_RUNNING; |
dd41f596 | 1950 | |
c350a04e MG |
1951 | /* |
1952 | * Make sure we do not leak PI boosting priority to the child. | |
1953 | */ | |
1954 | p->prio = current->normal_prio; | |
1955 | ||
b9dc29e7 MG |
1956 | /* |
1957 | * Revert to default priority/policy on fork if requested. | |
1958 | */ | |
1959 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1960 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1961 | p->policy = SCHED_NORMAL; |
6c697bdf | 1962 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1963 | p->rt_priority = 0; |
1964 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1965 | p->static_prio = NICE_TO_PRIO(0); | |
1966 | ||
1967 | p->prio = p->normal_prio = __normal_prio(p); | |
1968 | set_load_weight(p); | |
6c697bdf | 1969 | |
b9dc29e7 MG |
1970 | /* |
1971 | * We don't need the reset flag anymore after the fork. It has | |
1972 | * fulfilled its duty: | |
1973 | */ | |
1974 | p->sched_reset_on_fork = 0; | |
1975 | } | |
ca94c442 | 1976 | |
aab03e05 DF |
1977 | if (dl_prio(p->prio)) { |
1978 | put_cpu(); | |
1979 | return -EAGAIN; | |
1980 | } else if (rt_prio(p->prio)) { | |
1981 | p->sched_class = &rt_sched_class; | |
1982 | } else { | |
2ddbf952 | 1983 | p->sched_class = &fair_sched_class; |
aab03e05 | 1984 | } |
b29739f9 | 1985 | |
cd29fe6f PZ |
1986 | if (p->sched_class->task_fork) |
1987 | p->sched_class->task_fork(p); | |
1988 | ||
86951599 PZ |
1989 | /* |
1990 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1991 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1992 | * is ran before sched_fork(). | |
1993 | * | |
1994 | * Silence PROVE_RCU. | |
1995 | */ | |
0122ec5b | 1996 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1997 | set_task_cpu(p, cpu); |
0122ec5b | 1998 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1999 | |
52f17b6c | 2000 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2001 | if (likely(sched_info_on())) |
52f17b6c | 2002 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2003 | #endif |
3ca7a440 PZ |
2004 | #if defined(CONFIG_SMP) |
2005 | p->on_cpu = 0; | |
4866cde0 | 2006 | #endif |
01028747 | 2007 | init_task_preempt_count(p); |
806c09a7 | 2008 | #ifdef CONFIG_SMP |
917b627d | 2009 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2010 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2011 | #endif |
917b627d | 2012 | |
476d139c | 2013 | put_cpu(); |
aab03e05 | 2014 | return 0; |
1da177e4 LT |
2015 | } |
2016 | ||
332ac17e DF |
2017 | unsigned long to_ratio(u64 period, u64 runtime) |
2018 | { | |
2019 | if (runtime == RUNTIME_INF) | |
2020 | return 1ULL << 20; | |
2021 | ||
2022 | /* | |
2023 | * Doing this here saves a lot of checks in all | |
2024 | * the calling paths, and returning zero seems | |
2025 | * safe for them anyway. | |
2026 | */ | |
2027 | if (period == 0) | |
2028 | return 0; | |
2029 | ||
2030 | return div64_u64(runtime << 20, period); | |
2031 | } | |
2032 | ||
2033 | #ifdef CONFIG_SMP | |
2034 | inline struct dl_bw *dl_bw_of(int i) | |
2035 | { | |
66339c31 KT |
2036 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2037 | "sched RCU must be held"); | |
332ac17e DF |
2038 | return &cpu_rq(i)->rd->dl_bw; |
2039 | } | |
2040 | ||
de212f18 | 2041 | static inline int dl_bw_cpus(int i) |
332ac17e | 2042 | { |
de212f18 PZ |
2043 | struct root_domain *rd = cpu_rq(i)->rd; |
2044 | int cpus = 0; | |
2045 | ||
66339c31 KT |
2046 | rcu_lockdep_assert(rcu_read_lock_sched_held(), |
2047 | "sched RCU must be held"); | |
de212f18 PZ |
2048 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2049 | cpus++; | |
2050 | ||
2051 | return cpus; | |
332ac17e DF |
2052 | } |
2053 | #else | |
2054 | inline struct dl_bw *dl_bw_of(int i) | |
2055 | { | |
2056 | return &cpu_rq(i)->dl.dl_bw; | |
2057 | } | |
2058 | ||
de212f18 | 2059 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2060 | { |
2061 | return 1; | |
2062 | } | |
2063 | #endif | |
2064 | ||
332ac17e DF |
2065 | /* |
2066 | * We must be sure that accepting a new task (or allowing changing the | |
2067 | * parameters of an existing one) is consistent with the bandwidth | |
2068 | * constraints. If yes, this function also accordingly updates the currently | |
2069 | * allocated bandwidth to reflect the new situation. | |
2070 | * | |
2071 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2072 | * |
2073 | * XXX we should delay bw change until the task's 0-lag point, see | |
2074 | * __setparam_dl(). | |
332ac17e DF |
2075 | */ |
2076 | static int dl_overflow(struct task_struct *p, int policy, | |
2077 | const struct sched_attr *attr) | |
2078 | { | |
2079 | ||
2080 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2081 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2082 | u64 runtime = attr->sched_runtime; |
2083 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2084 | int cpus, err = -1; |
332ac17e DF |
2085 | |
2086 | if (new_bw == p->dl.dl_bw) | |
2087 | return 0; | |
2088 | ||
2089 | /* | |
2090 | * Either if a task, enters, leave, or stays -deadline but changes | |
2091 | * its parameters, we may need to update accordingly the total | |
2092 | * allocated bandwidth of the container. | |
2093 | */ | |
2094 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2095 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2096 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2097 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2098 | __dl_add(dl_b, new_bw); | |
2099 | err = 0; | |
2100 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2101 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2102 | __dl_clear(dl_b, p->dl.dl_bw); | |
2103 | __dl_add(dl_b, new_bw); | |
2104 | err = 0; | |
2105 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2106 | __dl_clear(dl_b, p->dl.dl_bw); | |
2107 | err = 0; | |
2108 | } | |
2109 | raw_spin_unlock(&dl_b->lock); | |
2110 | ||
2111 | return err; | |
2112 | } | |
2113 | ||
2114 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2115 | ||
1da177e4 LT |
2116 | /* |
2117 | * wake_up_new_task - wake up a newly created task for the first time. | |
2118 | * | |
2119 | * This function will do some initial scheduler statistics housekeeping | |
2120 | * that must be done for every newly created context, then puts the task | |
2121 | * on the runqueue and wakes it. | |
2122 | */ | |
3e51e3ed | 2123 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2124 | { |
2125 | unsigned long flags; | |
dd41f596 | 2126 | struct rq *rq; |
fabf318e | 2127 | |
ab2515c4 | 2128 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2129 | #ifdef CONFIG_SMP |
2130 | /* | |
2131 | * Fork balancing, do it here and not earlier because: | |
2132 | * - cpus_allowed can change in the fork path | |
2133 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2134 | */ |
ac66f547 | 2135 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2136 | #endif |
2137 | ||
a75cdaa9 AS |
2138 | /* Initialize new task's runnable average */ |
2139 | init_task_runnable_average(p); | |
ab2515c4 | 2140 | rq = __task_rq_lock(p); |
cd29fe6f | 2141 | activate_task(rq, p, 0); |
da0c1e65 | 2142 | p->on_rq = TASK_ON_RQ_QUEUED; |
89363381 | 2143 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2144 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2145 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2146 | if (p->sched_class->task_woken) |
2147 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2148 | #endif |
0122ec5b | 2149 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2150 | } |
2151 | ||
e107be36 AK |
2152 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2153 | ||
2154 | /** | |
80dd99b3 | 2155 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2156 | * @notifier: notifier struct to register |
e107be36 AK |
2157 | */ |
2158 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2159 | { | |
2160 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2161 | } | |
2162 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2163 | ||
2164 | /** | |
2165 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2166 | * @notifier: notifier struct to unregister |
e107be36 AK |
2167 | * |
2168 | * This is safe to call from within a preemption notifier. | |
2169 | */ | |
2170 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2171 | { | |
2172 | hlist_del(¬ifier->link); | |
2173 | } | |
2174 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2175 | ||
2176 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2177 | { | |
2178 | struct preempt_notifier *notifier; | |
e107be36 | 2179 | |
b67bfe0d | 2180 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2181 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2182 | } | |
2183 | ||
2184 | static void | |
2185 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2186 | struct task_struct *next) | |
2187 | { | |
2188 | struct preempt_notifier *notifier; | |
e107be36 | 2189 | |
b67bfe0d | 2190 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2191 | notifier->ops->sched_out(notifier, next); |
2192 | } | |
2193 | ||
6d6bc0ad | 2194 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2195 | |
2196 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2197 | { | |
2198 | } | |
2199 | ||
2200 | static void | |
2201 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2202 | struct task_struct *next) | |
2203 | { | |
2204 | } | |
2205 | ||
6d6bc0ad | 2206 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2207 | |
4866cde0 NP |
2208 | /** |
2209 | * prepare_task_switch - prepare to switch tasks | |
2210 | * @rq: the runqueue preparing to switch | |
421cee29 | 2211 | * @prev: the current task that is being switched out |
4866cde0 NP |
2212 | * @next: the task we are going to switch to. |
2213 | * | |
2214 | * This is called with the rq lock held and interrupts off. It must | |
2215 | * be paired with a subsequent finish_task_switch after the context | |
2216 | * switch. | |
2217 | * | |
2218 | * prepare_task_switch sets up locking and calls architecture specific | |
2219 | * hooks. | |
2220 | */ | |
e107be36 AK |
2221 | static inline void |
2222 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2223 | struct task_struct *next) | |
4866cde0 | 2224 | { |
895dd92c | 2225 | trace_sched_switch(prev, next); |
43148951 | 2226 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2227 | perf_event_task_sched_out(prev, next); |
e107be36 | 2228 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2229 | prepare_lock_switch(rq, next); |
2230 | prepare_arch_switch(next); | |
2231 | } | |
2232 | ||
1da177e4 LT |
2233 | /** |
2234 | * finish_task_switch - clean up after a task-switch | |
2235 | * @prev: the thread we just switched away from. | |
2236 | * | |
4866cde0 NP |
2237 | * finish_task_switch must be called after the context switch, paired |
2238 | * with a prepare_task_switch call before the context switch. | |
2239 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2240 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2241 | * |
2242 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2243 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2244 | * with the lock held can cause deadlocks; see schedule() for |
2245 | * details.) | |
dfa50b60 ON |
2246 | * |
2247 | * The context switch have flipped the stack from under us and restored the | |
2248 | * local variables which were saved when this task called schedule() in the | |
2249 | * past. prev == current is still correct but we need to recalculate this_rq | |
2250 | * because prev may have moved to another CPU. | |
1da177e4 | 2251 | */ |
dfa50b60 | 2252 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2253 | __releases(rq->lock) |
2254 | { | |
dfa50b60 | 2255 | struct rq *rq = this_rq(); |
1da177e4 | 2256 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2257 | long prev_state; |
1da177e4 LT |
2258 | |
2259 | rq->prev_mm = NULL; | |
2260 | ||
2261 | /* | |
2262 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2263 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2264 | * schedule one last time. The schedule call will never return, and |
2265 | * the scheduled task must drop that reference. | |
c394cc9f | 2266 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2267 | * still held, otherwise prev could be scheduled on another cpu, die |
2268 | * there before we look at prev->state, and then the reference would | |
2269 | * be dropped twice. | |
2270 | * Manfred Spraul <manfred@colorfullife.com> | |
2271 | */ | |
55a101f8 | 2272 | prev_state = prev->state; |
bf9fae9f | 2273 | vtime_task_switch(prev); |
4866cde0 | 2274 | finish_arch_switch(prev); |
a8d757ef | 2275 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2276 | finish_lock_switch(rq, prev); |
01f23e16 | 2277 | finish_arch_post_lock_switch(); |
e8fa1362 | 2278 | |
e107be36 | 2279 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2280 | if (mm) |
2281 | mmdrop(mm); | |
c394cc9f | 2282 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2283 | if (prev->sched_class->task_dead) |
2284 | prev->sched_class->task_dead(prev); | |
2285 | ||
c6fd91f0 | 2286 | /* |
2287 | * Remove function-return probe instances associated with this | |
2288 | * task and put them back on the free list. | |
9761eea8 | 2289 | */ |
c6fd91f0 | 2290 | kprobe_flush_task(prev); |
1da177e4 | 2291 | put_task_struct(prev); |
c6fd91f0 | 2292 | } |
99e5ada9 FW |
2293 | |
2294 | tick_nohz_task_switch(current); | |
dfa50b60 | 2295 | return rq; |
1da177e4 LT |
2296 | } |
2297 | ||
3f029d3c GH |
2298 | #ifdef CONFIG_SMP |
2299 | ||
3f029d3c GH |
2300 | /* rq->lock is NOT held, but preemption is disabled */ |
2301 | static inline void post_schedule(struct rq *rq) | |
2302 | { | |
2303 | if (rq->post_schedule) { | |
2304 | unsigned long flags; | |
2305 | ||
05fa785c | 2306 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2307 | if (rq->curr->sched_class->post_schedule) |
2308 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2309 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2310 | |
2311 | rq->post_schedule = 0; | |
2312 | } | |
2313 | } | |
2314 | ||
2315 | #else | |
da19ab51 | 2316 | |
3f029d3c GH |
2317 | static inline void post_schedule(struct rq *rq) |
2318 | { | |
1da177e4 LT |
2319 | } |
2320 | ||
3f029d3c GH |
2321 | #endif |
2322 | ||
1da177e4 LT |
2323 | /** |
2324 | * schedule_tail - first thing a freshly forked thread must call. | |
2325 | * @prev: the thread we just switched away from. | |
2326 | */ | |
722a9f92 | 2327 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2328 | __releases(rq->lock) |
2329 | { | |
1a43a14a | 2330 | struct rq *rq; |
da19ab51 | 2331 | |
1a43a14a ON |
2332 | /* finish_task_switch() drops rq->lock and enables preemtion */ |
2333 | preempt_disable(); | |
dfa50b60 | 2334 | rq = finish_task_switch(prev); |
3f029d3c | 2335 | post_schedule(rq); |
1a43a14a | 2336 | preempt_enable(); |
70b97a7f | 2337 | |
1da177e4 | 2338 | if (current->set_child_tid) |
b488893a | 2339 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2340 | } |
2341 | ||
2342 | /* | |
dfa50b60 | 2343 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2344 | */ |
dfa50b60 | 2345 | static inline struct rq * |
70b97a7f | 2346 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2347 | struct task_struct *next) |
1da177e4 | 2348 | { |
dd41f596 | 2349 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2350 | |
e107be36 | 2351 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2352 | |
dd41f596 IM |
2353 | mm = next->mm; |
2354 | oldmm = prev->active_mm; | |
9226d125 ZA |
2355 | /* |
2356 | * For paravirt, this is coupled with an exit in switch_to to | |
2357 | * combine the page table reload and the switch backend into | |
2358 | * one hypercall. | |
2359 | */ | |
224101ed | 2360 | arch_start_context_switch(prev); |
9226d125 | 2361 | |
31915ab4 | 2362 | if (!mm) { |
1da177e4 LT |
2363 | next->active_mm = oldmm; |
2364 | atomic_inc(&oldmm->mm_count); | |
2365 | enter_lazy_tlb(oldmm, next); | |
2366 | } else | |
2367 | switch_mm(oldmm, mm, next); | |
2368 | ||
31915ab4 | 2369 | if (!prev->mm) { |
1da177e4 | 2370 | prev->active_mm = NULL; |
1da177e4 LT |
2371 | rq->prev_mm = oldmm; |
2372 | } | |
3a5f5e48 IM |
2373 | /* |
2374 | * Since the runqueue lock will be released by the next | |
2375 | * task (which is an invalid locking op but in the case | |
2376 | * of the scheduler it's an obvious special-case), so we | |
2377 | * do an early lockdep release here: | |
2378 | */ | |
8a25d5de | 2379 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 | 2380 | |
91d1aa43 | 2381 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2382 | /* Here we just switch the register state and the stack. */ |
2383 | switch_to(prev, next, prev); | |
dd41f596 | 2384 | barrier(); |
dfa50b60 ON |
2385 | |
2386 | return finish_task_switch(prev); | |
1da177e4 LT |
2387 | } |
2388 | ||
2389 | /* | |
1c3e8264 | 2390 | * nr_running and nr_context_switches: |
1da177e4 LT |
2391 | * |
2392 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2393 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2394 | */ |
2395 | unsigned long nr_running(void) | |
2396 | { | |
2397 | unsigned long i, sum = 0; | |
2398 | ||
2399 | for_each_online_cpu(i) | |
2400 | sum += cpu_rq(i)->nr_running; | |
2401 | ||
2402 | return sum; | |
f711f609 | 2403 | } |
1da177e4 | 2404 | |
2ee507c4 TC |
2405 | /* |
2406 | * Check if only the current task is running on the cpu. | |
2407 | */ | |
2408 | bool single_task_running(void) | |
2409 | { | |
2410 | if (cpu_rq(smp_processor_id())->nr_running == 1) | |
2411 | return true; | |
2412 | else | |
2413 | return false; | |
2414 | } | |
2415 | EXPORT_SYMBOL(single_task_running); | |
2416 | ||
1da177e4 | 2417 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2418 | { |
cc94abfc SR |
2419 | int i; |
2420 | unsigned long long sum = 0; | |
46cb4b7c | 2421 | |
0a945022 | 2422 | for_each_possible_cpu(i) |
1da177e4 | 2423 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2424 | |
1da177e4 LT |
2425 | return sum; |
2426 | } | |
483b4ee6 | 2427 | |
1da177e4 LT |
2428 | unsigned long nr_iowait(void) |
2429 | { | |
2430 | unsigned long i, sum = 0; | |
483b4ee6 | 2431 | |
0a945022 | 2432 | for_each_possible_cpu(i) |
1da177e4 | 2433 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2434 | |
1da177e4 LT |
2435 | return sum; |
2436 | } | |
483b4ee6 | 2437 | |
8c215bd3 | 2438 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2439 | { |
8c215bd3 | 2440 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2441 | return atomic_read(&this->nr_iowait); |
2442 | } | |
46cb4b7c | 2443 | |
372ba8cb MG |
2444 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2445 | { | |
3289bdb4 PZ |
2446 | struct rq *rq = this_rq(); |
2447 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2448 | *load = rq->load.weight; | |
372ba8cb MG |
2449 | } |
2450 | ||
dd41f596 | 2451 | #ifdef CONFIG_SMP |
8a0be9ef | 2452 | |
46cb4b7c | 2453 | /* |
38022906 PZ |
2454 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2455 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2456 | */ |
38022906 | 2457 | void sched_exec(void) |
46cb4b7c | 2458 | { |
38022906 | 2459 | struct task_struct *p = current; |
1da177e4 | 2460 | unsigned long flags; |
0017d735 | 2461 | int dest_cpu; |
46cb4b7c | 2462 | |
8f42ced9 | 2463 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2464 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2465 | if (dest_cpu == smp_processor_id()) |
2466 | goto unlock; | |
38022906 | 2467 | |
8f42ced9 | 2468 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2469 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2470 | |
8f42ced9 PZ |
2471 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2472 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2473 | return; |
2474 | } | |
0017d735 | 2475 | unlock: |
8f42ced9 | 2476 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2477 | } |
dd41f596 | 2478 | |
1da177e4 LT |
2479 | #endif |
2480 | ||
1da177e4 | 2481 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2482 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2483 | |
2484 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2485 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2486 | |
c5f8d995 HS |
2487 | /* |
2488 | * Return accounted runtime for the task. | |
2489 | * In case the task is currently running, return the runtime plus current's | |
2490 | * pending runtime that have not been accounted yet. | |
2491 | */ | |
2492 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2493 | { | |
2494 | unsigned long flags; | |
2495 | struct rq *rq; | |
6e998916 | 2496 | u64 ns; |
c5f8d995 | 2497 | |
911b2898 PZ |
2498 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2499 | /* | |
2500 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2501 | * So we have a optimization chance when the task's delta_exec is 0. | |
2502 | * Reading ->on_cpu is racy, but this is ok. | |
2503 | * | |
2504 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2505 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2506 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2507 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2508 | * been accounted, so we're correct here as well. | |
911b2898 | 2509 | */ |
da0c1e65 | 2510 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2511 | return p->se.sum_exec_runtime; |
2512 | #endif | |
2513 | ||
c5f8d995 | 2514 | rq = task_rq_lock(p, &flags); |
6e998916 SG |
2515 | /* |
2516 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2517 | * project cycles that may never be accounted to this | |
2518 | * thread, breaking clock_gettime(). | |
2519 | */ | |
2520 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
2521 | update_rq_clock(rq); | |
2522 | p->sched_class->update_curr(rq); | |
2523 | } | |
2524 | ns = p->se.sum_exec_runtime; | |
0122ec5b | 2525 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2526 | |
2527 | return ns; | |
2528 | } | |
48f24c4d | 2529 | |
7835b98b CL |
2530 | /* |
2531 | * This function gets called by the timer code, with HZ frequency. | |
2532 | * We call it with interrupts disabled. | |
7835b98b CL |
2533 | */ |
2534 | void scheduler_tick(void) | |
2535 | { | |
7835b98b CL |
2536 | int cpu = smp_processor_id(); |
2537 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2538 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2539 | |
2540 | sched_clock_tick(); | |
dd41f596 | 2541 | |
05fa785c | 2542 | raw_spin_lock(&rq->lock); |
3e51f33f | 2543 | update_rq_clock(rq); |
fa85ae24 | 2544 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2545 | update_cpu_load_active(rq); |
3289bdb4 | 2546 | calc_global_load_tick(rq); |
05fa785c | 2547 | raw_spin_unlock(&rq->lock); |
7835b98b | 2548 | |
e9d2b064 | 2549 | perf_event_task_tick(); |
e220d2dc | 2550 | |
e418e1c2 | 2551 | #ifdef CONFIG_SMP |
6eb57e0d | 2552 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2553 | trigger_load_balance(rq); |
e418e1c2 | 2554 | #endif |
265f22a9 | 2555 | rq_last_tick_reset(rq); |
1da177e4 LT |
2556 | } |
2557 | ||
265f22a9 FW |
2558 | #ifdef CONFIG_NO_HZ_FULL |
2559 | /** | |
2560 | * scheduler_tick_max_deferment | |
2561 | * | |
2562 | * Keep at least one tick per second when a single | |
2563 | * active task is running because the scheduler doesn't | |
2564 | * yet completely support full dynticks environment. | |
2565 | * | |
2566 | * This makes sure that uptime, CFS vruntime, load | |
2567 | * balancing, etc... continue to move forward, even | |
2568 | * with a very low granularity. | |
e69f6186 YB |
2569 | * |
2570 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2571 | */ |
2572 | u64 scheduler_tick_max_deferment(void) | |
2573 | { | |
2574 | struct rq *rq = this_rq(); | |
316c1608 | 2575 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
2576 | |
2577 | next = rq->last_sched_tick + HZ; | |
2578 | ||
2579 | if (time_before_eq(next, now)) | |
2580 | return 0; | |
2581 | ||
8fe8ff09 | 2582 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2583 | } |
265f22a9 | 2584 | #endif |
1da177e4 | 2585 | |
132380a0 | 2586 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2587 | { |
2588 | if (in_lock_functions(addr)) { | |
2589 | addr = CALLER_ADDR2; | |
2590 | if (in_lock_functions(addr)) | |
2591 | addr = CALLER_ADDR3; | |
2592 | } | |
2593 | return addr; | |
2594 | } | |
1da177e4 | 2595 | |
7e49fcce SR |
2596 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2597 | defined(CONFIG_PREEMPT_TRACER)) | |
2598 | ||
edafe3a5 | 2599 | void preempt_count_add(int val) |
1da177e4 | 2600 | { |
6cd8a4bb | 2601 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2602 | /* |
2603 | * Underflow? | |
2604 | */ | |
9a11b49a IM |
2605 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2606 | return; | |
6cd8a4bb | 2607 | #endif |
bdb43806 | 2608 | __preempt_count_add(val); |
6cd8a4bb | 2609 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2610 | /* |
2611 | * Spinlock count overflowing soon? | |
2612 | */ | |
33859f7f MOS |
2613 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2614 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2615 | #endif |
8f47b187 TG |
2616 | if (preempt_count() == val) { |
2617 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2618 | #ifdef CONFIG_DEBUG_PREEMPT | |
2619 | current->preempt_disable_ip = ip; | |
2620 | #endif | |
2621 | trace_preempt_off(CALLER_ADDR0, ip); | |
2622 | } | |
1da177e4 | 2623 | } |
bdb43806 | 2624 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2625 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2626 | |
edafe3a5 | 2627 | void preempt_count_sub(int val) |
1da177e4 | 2628 | { |
6cd8a4bb | 2629 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2630 | /* |
2631 | * Underflow? | |
2632 | */ | |
01e3eb82 | 2633 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2634 | return; |
1da177e4 LT |
2635 | /* |
2636 | * Is the spinlock portion underflowing? | |
2637 | */ | |
9a11b49a IM |
2638 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2639 | !(preempt_count() & PREEMPT_MASK))) | |
2640 | return; | |
6cd8a4bb | 2641 | #endif |
9a11b49a | 2642 | |
6cd8a4bb SR |
2643 | if (preempt_count() == val) |
2644 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2645 | __preempt_count_sub(val); |
1da177e4 | 2646 | } |
bdb43806 | 2647 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2648 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2649 | |
2650 | #endif | |
2651 | ||
2652 | /* | |
dd41f596 | 2653 | * Print scheduling while atomic bug: |
1da177e4 | 2654 | */ |
dd41f596 | 2655 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2656 | { |
664dfa65 DJ |
2657 | if (oops_in_progress) |
2658 | return; | |
2659 | ||
3df0fc5b PZ |
2660 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2661 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2662 | |
dd41f596 | 2663 | debug_show_held_locks(prev); |
e21f5b15 | 2664 | print_modules(); |
dd41f596 IM |
2665 | if (irqs_disabled()) |
2666 | print_irqtrace_events(prev); | |
8f47b187 TG |
2667 | #ifdef CONFIG_DEBUG_PREEMPT |
2668 | if (in_atomic_preempt_off()) { | |
2669 | pr_err("Preemption disabled at:"); | |
2670 | print_ip_sym(current->preempt_disable_ip); | |
2671 | pr_cont("\n"); | |
2672 | } | |
2673 | #endif | |
6135fc1e | 2674 | dump_stack(); |
373d4d09 | 2675 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2676 | } |
1da177e4 | 2677 | |
dd41f596 IM |
2678 | /* |
2679 | * Various schedule()-time debugging checks and statistics: | |
2680 | */ | |
2681 | static inline void schedule_debug(struct task_struct *prev) | |
2682 | { | |
0d9e2632 AT |
2683 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
2684 | BUG_ON(unlikely(task_stack_end_corrupted(prev))); | |
2685 | #endif | |
1da177e4 | 2686 | /* |
41a2d6cf | 2687 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2688 | * schedule() atomically, we ignore that path. Otherwise whine |
2689 | * if we are scheduling when we should not. | |
1da177e4 | 2690 | */ |
192301e7 | 2691 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2692 | __schedule_bug(prev); |
b3fbab05 | 2693 | rcu_sleep_check(); |
dd41f596 | 2694 | |
1da177e4 LT |
2695 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2696 | ||
2d72376b | 2697 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2698 | } |
2699 | ||
2700 | /* | |
2701 | * Pick up the highest-prio task: | |
2702 | */ | |
2703 | static inline struct task_struct * | |
606dba2e | 2704 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2705 | { |
37e117c0 | 2706 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2707 | struct task_struct *p; |
1da177e4 LT |
2708 | |
2709 | /* | |
dd41f596 IM |
2710 | * Optimization: we know that if all tasks are in |
2711 | * the fair class we can call that function directly: | |
1da177e4 | 2712 | */ |
37e117c0 | 2713 | if (likely(prev->sched_class == class && |
38033c37 | 2714 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2715 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2716 | if (unlikely(p == RETRY_TASK)) |
2717 | goto again; | |
2718 | ||
2719 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2720 | if (unlikely(!p)) | |
2721 | p = idle_sched_class.pick_next_task(rq, prev); | |
2722 | ||
2723 | return p; | |
1da177e4 LT |
2724 | } |
2725 | ||
37e117c0 | 2726 | again: |
34f971f6 | 2727 | for_each_class(class) { |
606dba2e | 2728 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2729 | if (p) { |
2730 | if (unlikely(p == RETRY_TASK)) | |
2731 | goto again; | |
dd41f596 | 2732 | return p; |
37e117c0 | 2733 | } |
dd41f596 | 2734 | } |
34f971f6 PZ |
2735 | |
2736 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2737 | } |
1da177e4 | 2738 | |
dd41f596 | 2739 | /* |
c259e01a | 2740 | * __schedule() is the main scheduler function. |
edde96ea PE |
2741 | * |
2742 | * The main means of driving the scheduler and thus entering this function are: | |
2743 | * | |
2744 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2745 | * | |
2746 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2747 | * paths. For example, see arch/x86/entry_64.S. | |
2748 | * | |
2749 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2750 | * interrupt handler scheduler_tick(). | |
2751 | * | |
2752 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2753 | * task to the run-queue and that's it. | |
2754 | * | |
2755 | * Now, if the new task added to the run-queue preempts the current | |
2756 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2757 | * called on the nearest possible occasion: | |
2758 | * | |
2759 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2760 | * | |
2761 | * - in syscall or exception context, at the next outmost | |
2762 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2763 | * spin_unlock()!) | |
2764 | * | |
2765 | * - in IRQ context, return from interrupt-handler to | |
2766 | * preemptible context | |
2767 | * | |
2768 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2769 | * then at the next: | |
2770 | * | |
2771 | * - cond_resched() call | |
2772 | * - explicit schedule() call | |
2773 | * - return from syscall or exception to user-space | |
2774 | * - return from interrupt-handler to user-space | |
bfd9b2b5 FW |
2775 | * |
2776 | * WARNING: all callers must re-check need_resched() afterward and reschedule | |
2777 | * accordingly in case an event triggered the need for rescheduling (such as | |
2778 | * an interrupt waking up a task) while preemption was disabled in __schedule(). | |
dd41f596 | 2779 | */ |
c259e01a | 2780 | static void __sched __schedule(void) |
dd41f596 IM |
2781 | { |
2782 | struct task_struct *prev, *next; | |
67ca7bde | 2783 | unsigned long *switch_count; |
dd41f596 | 2784 | struct rq *rq; |
31656519 | 2785 | int cpu; |
dd41f596 | 2786 | |
ff743345 | 2787 | preempt_disable(); |
dd41f596 IM |
2788 | cpu = smp_processor_id(); |
2789 | rq = cpu_rq(cpu); | |
38200cf2 | 2790 | rcu_note_context_switch(); |
dd41f596 | 2791 | prev = rq->curr; |
dd41f596 | 2792 | |
dd41f596 | 2793 | schedule_debug(prev); |
1da177e4 | 2794 | |
31656519 | 2795 | if (sched_feat(HRTICK)) |
f333fdc9 | 2796 | hrtick_clear(rq); |
8f4d37ec | 2797 | |
e0acd0a6 ON |
2798 | /* |
2799 | * Make sure that signal_pending_state()->signal_pending() below | |
2800 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2801 | * done by the caller to avoid the race with signal_wake_up(). | |
2802 | */ | |
2803 | smp_mb__before_spinlock(); | |
05fa785c | 2804 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2805 | |
9edfbfed PZ |
2806 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
2807 | ||
246d86b5 | 2808 | switch_count = &prev->nivcsw; |
1da177e4 | 2809 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2810 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2811 | prev->state = TASK_RUNNING; |
21aa9af0 | 2812 | } else { |
2acca55e PZ |
2813 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2814 | prev->on_rq = 0; | |
2815 | ||
21aa9af0 | 2816 | /* |
2acca55e PZ |
2817 | * If a worker went to sleep, notify and ask workqueue |
2818 | * whether it wants to wake up a task to maintain | |
2819 | * concurrency. | |
21aa9af0 TH |
2820 | */ |
2821 | if (prev->flags & PF_WQ_WORKER) { | |
2822 | struct task_struct *to_wakeup; | |
2823 | ||
2824 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2825 | if (to_wakeup) | |
2826 | try_to_wake_up_local(to_wakeup); | |
2827 | } | |
21aa9af0 | 2828 | } |
dd41f596 | 2829 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2830 | } |
2831 | ||
9edfbfed | 2832 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
2833 | update_rq_clock(rq); |
2834 | ||
2835 | next = pick_next_task(rq, prev); | |
f26f9aff | 2836 | clear_tsk_need_resched(prev); |
f27dde8d | 2837 | clear_preempt_need_resched(); |
9edfbfed | 2838 | rq->clock_skip_update = 0; |
1da177e4 | 2839 | |
1da177e4 | 2840 | if (likely(prev != next)) { |
1da177e4 LT |
2841 | rq->nr_switches++; |
2842 | rq->curr = next; | |
2843 | ++*switch_count; | |
2844 | ||
dfa50b60 ON |
2845 | rq = context_switch(rq, prev, next); /* unlocks the rq */ |
2846 | cpu = cpu_of(rq); | |
1da177e4 | 2847 | } else |
05fa785c | 2848 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2849 | |
3f029d3c | 2850 | post_schedule(rq); |
1da177e4 | 2851 | |
ba74c144 | 2852 | sched_preempt_enable_no_resched(); |
1da177e4 | 2853 | } |
c259e01a | 2854 | |
9c40cef2 TG |
2855 | static inline void sched_submit_work(struct task_struct *tsk) |
2856 | { | |
3c7d5184 | 2857 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2858 | return; |
2859 | /* | |
2860 | * If we are going to sleep and we have plugged IO queued, | |
2861 | * make sure to submit it to avoid deadlocks. | |
2862 | */ | |
2863 | if (blk_needs_flush_plug(tsk)) | |
2864 | blk_schedule_flush_plug(tsk); | |
2865 | } | |
2866 | ||
722a9f92 | 2867 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2868 | { |
9c40cef2 TG |
2869 | struct task_struct *tsk = current; |
2870 | ||
2871 | sched_submit_work(tsk); | |
bfd9b2b5 FW |
2872 | do { |
2873 | __schedule(); | |
2874 | } while (need_resched()); | |
c259e01a | 2875 | } |
1da177e4 LT |
2876 | EXPORT_SYMBOL(schedule); |
2877 | ||
91d1aa43 | 2878 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2879 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2880 | { |
2881 | /* | |
2882 | * If we come here after a random call to set_need_resched(), | |
2883 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2884 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2885 | * we find a better solution. | |
7cc78f8f AL |
2886 | * |
2887 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 2888 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 2889 | * too frequently to make sense yet. |
20ab65e3 | 2890 | */ |
7cc78f8f | 2891 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 2892 | schedule(); |
7cc78f8f | 2893 | exception_exit(prev_state); |
20ab65e3 FW |
2894 | } |
2895 | #endif | |
2896 | ||
c5491ea7 TG |
2897 | /** |
2898 | * schedule_preempt_disabled - called with preemption disabled | |
2899 | * | |
2900 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2901 | */ | |
2902 | void __sched schedule_preempt_disabled(void) | |
2903 | { | |
ba74c144 | 2904 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2905 | schedule(); |
2906 | preempt_disable(); | |
2907 | } | |
2908 | ||
06b1f808 | 2909 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
2910 | { |
2911 | do { | |
2912 | __preempt_count_add(PREEMPT_ACTIVE); | |
2913 | __schedule(); | |
2914 | __preempt_count_sub(PREEMPT_ACTIVE); | |
2915 | ||
2916 | /* | |
2917 | * Check again in case we missed a preemption opportunity | |
2918 | * between schedule and now. | |
2919 | */ | |
2920 | barrier(); | |
2921 | } while (need_resched()); | |
2922 | } | |
2923 | ||
1da177e4 LT |
2924 | #ifdef CONFIG_PREEMPT |
2925 | /* | |
2ed6e34f | 2926 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2927 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2928 | * occur there and call schedule directly. |
2929 | */ | |
722a9f92 | 2930 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2931 | { |
1da177e4 LT |
2932 | /* |
2933 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2934 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2935 | */ |
fbb00b56 | 2936 | if (likely(!preemptible())) |
1da177e4 LT |
2937 | return; |
2938 | ||
a18b5d01 | 2939 | preempt_schedule_common(); |
1da177e4 | 2940 | } |
376e2424 | 2941 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2942 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 ON |
2943 | |
2944 | #ifdef CONFIG_CONTEXT_TRACKING | |
2945 | /** | |
2946 | * preempt_schedule_context - preempt_schedule called by tracing | |
2947 | * | |
2948 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
2949 | * recursion and tracing preempt enabling caused by the tracing | |
2950 | * infrastructure itself. But as tracing can happen in areas coming | |
2951 | * from userspace or just about to enter userspace, a preempt enable | |
2952 | * can occur before user_exit() is called. This will cause the scheduler | |
2953 | * to be called when the system is still in usermode. | |
2954 | * | |
2955 | * To prevent this, the preempt_enable_notrace will use this function | |
2956 | * instead of preempt_schedule() to exit user context if needed before | |
2957 | * calling the scheduler. | |
2958 | */ | |
2959 | asmlinkage __visible void __sched notrace preempt_schedule_context(void) | |
2960 | { | |
2961 | enum ctx_state prev_ctx; | |
2962 | ||
2963 | if (likely(!preemptible())) | |
2964 | return; | |
2965 | ||
2966 | do { | |
2967 | __preempt_count_add(PREEMPT_ACTIVE); | |
2968 | /* | |
2969 | * Needs preempt disabled in case user_exit() is traced | |
2970 | * and the tracer calls preempt_enable_notrace() causing | |
2971 | * an infinite recursion. | |
2972 | */ | |
2973 | prev_ctx = exception_enter(); | |
2974 | __schedule(); | |
2975 | exception_exit(prev_ctx); | |
2976 | ||
2977 | __preempt_count_sub(PREEMPT_ACTIVE); | |
2978 | barrier(); | |
2979 | } while (need_resched()); | |
2980 | } | |
2981 | EXPORT_SYMBOL_GPL(preempt_schedule_context); | |
2982 | #endif /* CONFIG_CONTEXT_TRACKING */ | |
2983 | ||
32e475d7 | 2984 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2985 | |
2986 | /* | |
2ed6e34f | 2987 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2988 | * off of irq context. |
2989 | * Note, that this is called and return with irqs disabled. This will | |
2990 | * protect us against recursive calling from irq. | |
2991 | */ | |
722a9f92 | 2992 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2993 | { |
b22366cd | 2994 | enum ctx_state prev_state; |
6478d880 | 2995 | |
2ed6e34f | 2996 | /* Catch callers which need to be fixed */ |
f27dde8d | 2997 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2998 | |
b22366cd FW |
2999 | prev_state = exception_enter(); |
3000 | ||
3a5c359a | 3001 | do { |
bdb43806 | 3002 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 3003 | local_irq_enable(); |
c259e01a | 3004 | __schedule(); |
3a5c359a | 3005 | local_irq_disable(); |
bdb43806 | 3006 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 3007 | |
3a5c359a AK |
3008 | /* |
3009 | * Check again in case we missed a preemption opportunity | |
3010 | * between schedule and now. | |
3011 | */ | |
3012 | barrier(); | |
5ed0cec0 | 3013 | } while (need_resched()); |
b22366cd FW |
3014 | |
3015 | exception_exit(prev_state); | |
1da177e4 LT |
3016 | } |
3017 | ||
63859d4f | 3018 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3019 | void *key) |
1da177e4 | 3020 | { |
63859d4f | 3021 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3022 | } |
1da177e4 LT |
3023 | EXPORT_SYMBOL(default_wake_function); |
3024 | ||
b29739f9 IM |
3025 | #ifdef CONFIG_RT_MUTEXES |
3026 | ||
3027 | /* | |
3028 | * rt_mutex_setprio - set the current priority of a task | |
3029 | * @p: task | |
3030 | * @prio: prio value (kernel-internal form) | |
3031 | * | |
3032 | * This function changes the 'effective' priority of a task. It does | |
3033 | * not touch ->normal_prio like __setscheduler(). | |
3034 | * | |
c365c292 TG |
3035 | * Used by the rt_mutex code to implement priority inheritance |
3036 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3037 | */ |
36c8b586 | 3038 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3039 | { |
da0c1e65 | 3040 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 3041 | struct rq *rq; |
83ab0aa0 | 3042 | const struct sched_class *prev_class; |
b29739f9 | 3043 | |
aab03e05 | 3044 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3045 | |
0122ec5b | 3046 | rq = __task_rq_lock(p); |
b29739f9 | 3047 | |
1c4dd99b TG |
3048 | /* |
3049 | * Idle task boosting is a nono in general. There is one | |
3050 | * exception, when PREEMPT_RT and NOHZ is active: | |
3051 | * | |
3052 | * The idle task calls get_next_timer_interrupt() and holds | |
3053 | * the timer wheel base->lock on the CPU and another CPU wants | |
3054 | * to access the timer (probably to cancel it). We can safely | |
3055 | * ignore the boosting request, as the idle CPU runs this code | |
3056 | * with interrupts disabled and will complete the lock | |
3057 | * protected section without being interrupted. So there is no | |
3058 | * real need to boost. | |
3059 | */ | |
3060 | if (unlikely(p == rq->idle)) { | |
3061 | WARN_ON(p != rq->curr); | |
3062 | WARN_ON(p->pi_blocked_on); | |
3063 | goto out_unlock; | |
3064 | } | |
3065 | ||
a8027073 | 3066 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3067 | oldprio = p->prio; |
83ab0aa0 | 3068 | prev_class = p->sched_class; |
da0c1e65 | 3069 | queued = task_on_rq_queued(p); |
051a1d1a | 3070 | running = task_current(rq, p); |
da0c1e65 | 3071 | if (queued) |
69be72c1 | 3072 | dequeue_task(rq, p, 0); |
0e1f3483 | 3073 | if (running) |
f3cd1c4e | 3074 | put_prev_task(rq, p); |
dd41f596 | 3075 | |
2d3d891d DF |
3076 | /* |
3077 | * Boosting condition are: | |
3078 | * 1. -rt task is running and holds mutex A | |
3079 | * --> -dl task blocks on mutex A | |
3080 | * | |
3081 | * 2. -dl task is running and holds mutex A | |
3082 | * --> -dl task blocks on mutex A and could preempt the | |
3083 | * running task | |
3084 | */ | |
3085 | if (dl_prio(prio)) { | |
466af29b ON |
3086 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3087 | if (!dl_prio(p->normal_prio) || | |
3088 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3089 | p->dl.dl_boosted = 1; |
3090 | p->dl.dl_throttled = 0; | |
3091 | enqueue_flag = ENQUEUE_REPLENISH; | |
3092 | } else | |
3093 | p->dl.dl_boosted = 0; | |
aab03e05 | 3094 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3095 | } else if (rt_prio(prio)) { |
3096 | if (dl_prio(oldprio)) | |
3097 | p->dl.dl_boosted = 0; | |
3098 | if (oldprio < prio) | |
3099 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3100 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3101 | } else { |
3102 | if (dl_prio(oldprio)) | |
3103 | p->dl.dl_boosted = 0; | |
746db944 BS |
3104 | if (rt_prio(oldprio)) |
3105 | p->rt.timeout = 0; | |
dd41f596 | 3106 | p->sched_class = &fair_sched_class; |
2d3d891d | 3107 | } |
dd41f596 | 3108 | |
b29739f9 IM |
3109 | p->prio = prio; |
3110 | ||
0e1f3483 HS |
3111 | if (running) |
3112 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3113 | if (queued) |
2d3d891d | 3114 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3115 | |
da7a735e | 3116 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3117 | out_unlock: |
0122ec5b | 3118 | __task_rq_unlock(rq); |
b29739f9 | 3119 | } |
b29739f9 | 3120 | #endif |
d50dde5a | 3121 | |
36c8b586 | 3122 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3123 | { |
da0c1e65 | 3124 | int old_prio, delta, queued; |
1da177e4 | 3125 | unsigned long flags; |
70b97a7f | 3126 | struct rq *rq; |
1da177e4 | 3127 | |
75e45d51 | 3128 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3129 | return; |
3130 | /* | |
3131 | * We have to be careful, if called from sys_setpriority(), | |
3132 | * the task might be in the middle of scheduling on another CPU. | |
3133 | */ | |
3134 | rq = task_rq_lock(p, &flags); | |
3135 | /* | |
3136 | * The RT priorities are set via sched_setscheduler(), but we still | |
3137 | * allow the 'normal' nice value to be set - but as expected | |
3138 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3139 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3140 | */ |
aab03e05 | 3141 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3142 | p->static_prio = NICE_TO_PRIO(nice); |
3143 | goto out_unlock; | |
3144 | } | |
da0c1e65 KT |
3145 | queued = task_on_rq_queued(p); |
3146 | if (queued) | |
69be72c1 | 3147 | dequeue_task(rq, p, 0); |
1da177e4 | 3148 | |
1da177e4 | 3149 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3150 | set_load_weight(p); |
b29739f9 IM |
3151 | old_prio = p->prio; |
3152 | p->prio = effective_prio(p); | |
3153 | delta = p->prio - old_prio; | |
1da177e4 | 3154 | |
da0c1e65 | 3155 | if (queued) { |
371fd7e7 | 3156 | enqueue_task(rq, p, 0); |
1da177e4 | 3157 | /* |
d5f9f942 AM |
3158 | * If the task increased its priority or is running and |
3159 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3160 | */ |
d5f9f942 | 3161 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3162 | resched_curr(rq); |
1da177e4 LT |
3163 | } |
3164 | out_unlock: | |
0122ec5b | 3165 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3166 | } |
1da177e4 LT |
3167 | EXPORT_SYMBOL(set_user_nice); |
3168 | ||
e43379f1 MM |
3169 | /* |
3170 | * can_nice - check if a task can reduce its nice value | |
3171 | * @p: task | |
3172 | * @nice: nice value | |
3173 | */ | |
36c8b586 | 3174 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3175 | { |
024f4747 | 3176 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3177 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3178 | |
78d7d407 | 3179 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3180 | capable(CAP_SYS_NICE)); |
3181 | } | |
3182 | ||
1da177e4 LT |
3183 | #ifdef __ARCH_WANT_SYS_NICE |
3184 | ||
3185 | /* | |
3186 | * sys_nice - change the priority of the current process. | |
3187 | * @increment: priority increment | |
3188 | * | |
3189 | * sys_setpriority is a more generic, but much slower function that | |
3190 | * does similar things. | |
3191 | */ | |
5add95d4 | 3192 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3193 | { |
48f24c4d | 3194 | long nice, retval; |
1da177e4 LT |
3195 | |
3196 | /* | |
3197 | * Setpriority might change our priority at the same moment. | |
3198 | * We don't have to worry. Conceptually one call occurs first | |
3199 | * and we have a single winner. | |
3200 | */ | |
a9467fa3 | 3201 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3202 | nice = task_nice(current) + increment; |
1da177e4 | 3203 | |
a9467fa3 | 3204 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3205 | if (increment < 0 && !can_nice(current, nice)) |
3206 | return -EPERM; | |
3207 | ||
1da177e4 LT |
3208 | retval = security_task_setnice(current, nice); |
3209 | if (retval) | |
3210 | return retval; | |
3211 | ||
3212 | set_user_nice(current, nice); | |
3213 | return 0; | |
3214 | } | |
3215 | ||
3216 | #endif | |
3217 | ||
3218 | /** | |
3219 | * task_prio - return the priority value of a given task. | |
3220 | * @p: the task in question. | |
3221 | * | |
e69f6186 | 3222 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3223 | * RT tasks are offset by -200. Normal tasks are centered |
3224 | * around 0, value goes from -16 to +15. | |
3225 | */ | |
36c8b586 | 3226 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3227 | { |
3228 | return p->prio - MAX_RT_PRIO; | |
3229 | } | |
3230 | ||
1da177e4 LT |
3231 | /** |
3232 | * idle_cpu - is a given cpu idle currently? | |
3233 | * @cpu: the processor in question. | |
e69f6186 YB |
3234 | * |
3235 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3236 | */ |
3237 | int idle_cpu(int cpu) | |
3238 | { | |
908a3283 TG |
3239 | struct rq *rq = cpu_rq(cpu); |
3240 | ||
3241 | if (rq->curr != rq->idle) | |
3242 | return 0; | |
3243 | ||
3244 | if (rq->nr_running) | |
3245 | return 0; | |
3246 | ||
3247 | #ifdef CONFIG_SMP | |
3248 | if (!llist_empty(&rq->wake_list)) | |
3249 | return 0; | |
3250 | #endif | |
3251 | ||
3252 | return 1; | |
1da177e4 LT |
3253 | } |
3254 | ||
1da177e4 LT |
3255 | /** |
3256 | * idle_task - return the idle task for a given cpu. | |
3257 | * @cpu: the processor in question. | |
e69f6186 YB |
3258 | * |
3259 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3260 | */ |
36c8b586 | 3261 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3262 | { |
3263 | return cpu_rq(cpu)->idle; | |
3264 | } | |
3265 | ||
3266 | /** | |
3267 | * find_process_by_pid - find a process with a matching PID value. | |
3268 | * @pid: the pid in question. | |
e69f6186 YB |
3269 | * |
3270 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3271 | */ |
a9957449 | 3272 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3273 | { |
228ebcbe | 3274 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3275 | } |
3276 | ||
aab03e05 DF |
3277 | /* |
3278 | * This function initializes the sched_dl_entity of a newly becoming | |
3279 | * SCHED_DEADLINE task. | |
3280 | * | |
3281 | * Only the static values are considered here, the actual runtime and the | |
3282 | * absolute deadline will be properly calculated when the task is enqueued | |
3283 | * for the first time with its new policy. | |
3284 | */ | |
3285 | static void | |
3286 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3287 | { | |
3288 | struct sched_dl_entity *dl_se = &p->dl; | |
3289 | ||
aab03e05 DF |
3290 | dl_se->dl_runtime = attr->sched_runtime; |
3291 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3292 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3293 | dl_se->flags = attr->sched_flags; |
332ac17e | 3294 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3295 | |
3296 | /* | |
3297 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3298 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3299 | * | |
3300 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3301 | * point. This would include retaining the task_struct until that time | |
3302 | * and change dl_overflow() to not immediately decrement the current | |
3303 | * amount. | |
3304 | * | |
3305 | * Instead we retain the current runtime/deadline and let the new | |
3306 | * parameters take effect after the current reservation period lapses. | |
3307 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3308 | * before the current scheduling deadline. | |
3309 | * | |
3310 | * We can still have temporary overloads because we do not delay the | |
3311 | * change in bandwidth until that time; so admission control is | |
3312 | * not on the safe side. It does however guarantee tasks will never | |
3313 | * consume more than promised. | |
3314 | */ | |
aab03e05 DF |
3315 | } |
3316 | ||
c13db6b1 SR |
3317 | /* |
3318 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3319 | * it calls know not to change it. | |
3320 | */ | |
3321 | #define SETPARAM_POLICY -1 | |
3322 | ||
c365c292 TG |
3323 | static void __setscheduler_params(struct task_struct *p, |
3324 | const struct sched_attr *attr) | |
1da177e4 | 3325 | { |
d50dde5a DF |
3326 | int policy = attr->sched_policy; |
3327 | ||
c13db6b1 | 3328 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3329 | policy = p->policy; |
3330 | ||
1da177e4 | 3331 | p->policy = policy; |
d50dde5a | 3332 | |
aab03e05 DF |
3333 | if (dl_policy(policy)) |
3334 | __setparam_dl(p, attr); | |
39fd8fd2 | 3335 | else if (fair_policy(policy)) |
d50dde5a DF |
3336 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3337 | ||
39fd8fd2 PZ |
3338 | /* |
3339 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3340 | * !rt_policy. Always setting this ensures that things like | |
3341 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3342 | */ | |
3343 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3344 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3345 | set_load_weight(p); |
3346 | } | |
39fd8fd2 | 3347 | |
c365c292 TG |
3348 | /* Actually do priority change: must hold pi & rq lock. */ |
3349 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 3350 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
3351 | { |
3352 | __setscheduler_params(p, attr); | |
d50dde5a | 3353 | |
383afd09 | 3354 | /* |
0782e63b TG |
3355 | * Keep a potential priority boosting if called from |
3356 | * sched_setscheduler(). | |
383afd09 | 3357 | */ |
0782e63b TG |
3358 | if (keep_boost) |
3359 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
3360 | else | |
3361 | p->prio = normal_prio(p); | |
383afd09 | 3362 | |
aab03e05 DF |
3363 | if (dl_prio(p->prio)) |
3364 | p->sched_class = &dl_sched_class; | |
3365 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3366 | p->sched_class = &rt_sched_class; |
3367 | else | |
3368 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3369 | } |
aab03e05 DF |
3370 | |
3371 | static void | |
3372 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3373 | { | |
3374 | struct sched_dl_entity *dl_se = &p->dl; | |
3375 | ||
3376 | attr->sched_priority = p->rt_priority; | |
3377 | attr->sched_runtime = dl_se->dl_runtime; | |
3378 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3379 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3380 | attr->sched_flags = dl_se->flags; |
3381 | } | |
3382 | ||
3383 | /* | |
3384 | * This function validates the new parameters of a -deadline task. | |
3385 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3386 | * than the runtime, as well as the period of being zero or |
332ac17e | 3387 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3388 | * user parameters are above the internal resolution of 1us (we |
3389 | * check sched_runtime only since it is always the smaller one) and | |
3390 | * below 2^63 ns (we have to check both sched_deadline and | |
3391 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3392 | */ |
3393 | static bool | |
3394 | __checkparam_dl(const struct sched_attr *attr) | |
3395 | { | |
b0827819 JL |
3396 | /* deadline != 0 */ |
3397 | if (attr->sched_deadline == 0) | |
3398 | return false; | |
3399 | ||
3400 | /* | |
3401 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3402 | * that big. | |
3403 | */ | |
3404 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3405 | return false; | |
3406 | ||
3407 | /* | |
3408 | * Since we use the MSB for wrap-around and sign issues, make | |
3409 | * sure it's not set (mind that period can be equal to zero). | |
3410 | */ | |
3411 | if (attr->sched_deadline & (1ULL << 63) || | |
3412 | attr->sched_period & (1ULL << 63)) | |
3413 | return false; | |
3414 | ||
3415 | /* runtime <= deadline <= period (if period != 0) */ | |
3416 | if ((attr->sched_period != 0 && | |
3417 | attr->sched_period < attr->sched_deadline) || | |
3418 | attr->sched_deadline < attr->sched_runtime) | |
3419 | return false; | |
3420 | ||
3421 | return true; | |
aab03e05 DF |
3422 | } |
3423 | ||
c69e8d9c DH |
3424 | /* |
3425 | * check the target process has a UID that matches the current process's | |
3426 | */ | |
3427 | static bool check_same_owner(struct task_struct *p) | |
3428 | { | |
3429 | const struct cred *cred = current_cred(), *pcred; | |
3430 | bool match; | |
3431 | ||
3432 | rcu_read_lock(); | |
3433 | pcred = __task_cred(p); | |
9c806aa0 EB |
3434 | match = (uid_eq(cred->euid, pcred->euid) || |
3435 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3436 | rcu_read_unlock(); |
3437 | return match; | |
3438 | } | |
3439 | ||
75381608 WL |
3440 | static bool dl_param_changed(struct task_struct *p, |
3441 | const struct sched_attr *attr) | |
3442 | { | |
3443 | struct sched_dl_entity *dl_se = &p->dl; | |
3444 | ||
3445 | if (dl_se->dl_runtime != attr->sched_runtime || | |
3446 | dl_se->dl_deadline != attr->sched_deadline || | |
3447 | dl_se->dl_period != attr->sched_period || | |
3448 | dl_se->flags != attr->sched_flags) | |
3449 | return true; | |
3450 | ||
3451 | return false; | |
3452 | } | |
3453 | ||
d50dde5a DF |
3454 | static int __sched_setscheduler(struct task_struct *p, |
3455 | const struct sched_attr *attr, | |
3456 | bool user) | |
1da177e4 | 3457 | { |
383afd09 SR |
3458 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3459 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3460 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 3461 | int new_effective_prio, policy = attr->sched_policy; |
1da177e4 | 3462 | unsigned long flags; |
83ab0aa0 | 3463 | const struct sched_class *prev_class; |
70b97a7f | 3464 | struct rq *rq; |
ca94c442 | 3465 | int reset_on_fork; |
1da177e4 | 3466 | |
66e5393a SR |
3467 | /* may grab non-irq protected spin_locks */ |
3468 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3469 | recheck: |
3470 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3471 | if (policy < 0) { |
3472 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3473 | policy = oldpolicy = p->policy; |
ca94c442 | 3474 | } else { |
7479f3c9 | 3475 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3476 | |
aab03e05 DF |
3477 | if (policy != SCHED_DEADLINE && |
3478 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3479 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3480 | policy != SCHED_IDLE) | |
3481 | return -EINVAL; | |
3482 | } | |
3483 | ||
7479f3c9 PZ |
3484 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3485 | return -EINVAL; | |
3486 | ||
1da177e4 LT |
3487 | /* |
3488 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3489 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3490 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3491 | */ |
0bb040a4 | 3492 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3493 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3494 | return -EINVAL; |
aab03e05 DF |
3495 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3496 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3497 | return -EINVAL; |
3498 | ||
37e4ab3f OC |
3499 | /* |
3500 | * Allow unprivileged RT tasks to decrease priority: | |
3501 | */ | |
961ccddd | 3502 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3503 | if (fair_policy(policy)) { |
d0ea0268 | 3504 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3505 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3506 | return -EPERM; |
3507 | } | |
3508 | ||
e05606d3 | 3509 | if (rt_policy(policy)) { |
a44702e8 ON |
3510 | unsigned long rlim_rtprio = |
3511 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3512 | |
3513 | /* can't set/change the rt policy */ | |
3514 | if (policy != p->policy && !rlim_rtprio) | |
3515 | return -EPERM; | |
3516 | ||
3517 | /* can't increase priority */ | |
d50dde5a DF |
3518 | if (attr->sched_priority > p->rt_priority && |
3519 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3520 | return -EPERM; |
3521 | } | |
c02aa73b | 3522 | |
d44753b8 JL |
3523 | /* |
3524 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3525 | * (safest behavior); in the future we would like to allow | |
3526 | * unprivileged DL tasks to increase their relative deadline | |
3527 | * or reduce their runtime (both ways reducing utilization) | |
3528 | */ | |
3529 | if (dl_policy(policy)) | |
3530 | return -EPERM; | |
3531 | ||
dd41f596 | 3532 | /* |
c02aa73b DH |
3533 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3534 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3535 | */ |
c02aa73b | 3536 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3537 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3538 | return -EPERM; |
3539 | } | |
5fe1d75f | 3540 | |
37e4ab3f | 3541 | /* can't change other user's priorities */ |
c69e8d9c | 3542 | if (!check_same_owner(p)) |
37e4ab3f | 3543 | return -EPERM; |
ca94c442 LP |
3544 | |
3545 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3546 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3547 | return -EPERM; | |
37e4ab3f | 3548 | } |
1da177e4 | 3549 | |
725aad24 | 3550 | if (user) { |
b0ae1981 | 3551 | retval = security_task_setscheduler(p); |
725aad24 JF |
3552 | if (retval) |
3553 | return retval; | |
3554 | } | |
3555 | ||
b29739f9 IM |
3556 | /* |
3557 | * make sure no PI-waiters arrive (or leave) while we are | |
3558 | * changing the priority of the task: | |
0122ec5b | 3559 | * |
25985edc | 3560 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3561 | * runqueue lock must be held. |
3562 | */ | |
0122ec5b | 3563 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3564 | |
34f971f6 PZ |
3565 | /* |
3566 | * Changing the policy of the stop threads its a very bad idea | |
3567 | */ | |
3568 | if (p == rq->stop) { | |
0122ec5b | 3569 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3570 | return -EINVAL; |
3571 | } | |
3572 | ||
a51e9198 | 3573 | /* |
d6b1e911 TG |
3574 | * If not changing anything there's no need to proceed further, |
3575 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3576 | */ |
d50dde5a | 3577 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3578 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3579 | goto change; |
3580 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3581 | goto change; | |
75381608 | 3582 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 3583 | goto change; |
d50dde5a | 3584 | |
d6b1e911 | 3585 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3586 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3587 | return 0; |
3588 | } | |
d50dde5a | 3589 | change: |
a51e9198 | 3590 | |
dc61b1d6 | 3591 | if (user) { |
332ac17e | 3592 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3593 | /* |
3594 | * Do not allow realtime tasks into groups that have no runtime | |
3595 | * assigned. | |
3596 | */ | |
3597 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3598 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3599 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3600 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3601 | return -EPERM; |
3602 | } | |
dc61b1d6 | 3603 | #endif |
332ac17e DF |
3604 | #ifdef CONFIG_SMP |
3605 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3606 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3607 | |
3608 | /* | |
3609 | * Don't allow tasks with an affinity mask smaller than | |
3610 | * the entire root_domain to become SCHED_DEADLINE. We | |
3611 | * will also fail if there's no bandwidth available. | |
3612 | */ | |
e4099a5e PZ |
3613 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3614 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3615 | task_rq_unlock(rq, p, &flags); |
3616 | return -EPERM; | |
3617 | } | |
3618 | } | |
3619 | #endif | |
3620 | } | |
dc61b1d6 | 3621 | |
1da177e4 LT |
3622 | /* recheck policy now with rq lock held */ |
3623 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3624 | policy = oldpolicy = -1; | |
0122ec5b | 3625 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3626 | goto recheck; |
3627 | } | |
332ac17e DF |
3628 | |
3629 | /* | |
3630 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3631 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3632 | * is available. | |
3633 | */ | |
e4099a5e | 3634 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3635 | task_rq_unlock(rq, p, &flags); |
3636 | return -EBUSY; | |
3637 | } | |
3638 | ||
c365c292 TG |
3639 | p->sched_reset_on_fork = reset_on_fork; |
3640 | oldprio = p->prio; | |
3641 | ||
3642 | /* | |
0782e63b TG |
3643 | * Take priority boosted tasks into account. If the new |
3644 | * effective priority is unchanged, we just store the new | |
c365c292 TG |
3645 | * normal parameters and do not touch the scheduler class and |
3646 | * the runqueue. This will be done when the task deboost | |
3647 | * itself. | |
3648 | */ | |
0782e63b TG |
3649 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); |
3650 | if (new_effective_prio == oldprio) { | |
c365c292 TG |
3651 | __setscheduler_params(p, attr); |
3652 | task_rq_unlock(rq, p, &flags); | |
3653 | return 0; | |
3654 | } | |
3655 | ||
da0c1e65 | 3656 | queued = task_on_rq_queued(p); |
051a1d1a | 3657 | running = task_current(rq, p); |
da0c1e65 | 3658 | if (queued) |
4ca9b72b | 3659 | dequeue_task(rq, p, 0); |
0e1f3483 | 3660 | if (running) |
f3cd1c4e | 3661 | put_prev_task(rq, p); |
f6b53205 | 3662 | |
83ab0aa0 | 3663 | prev_class = p->sched_class; |
0782e63b | 3664 | __setscheduler(rq, p, attr, true); |
f6b53205 | 3665 | |
0e1f3483 HS |
3666 | if (running) |
3667 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3668 | if (queued) { |
81a44c54 TG |
3669 | /* |
3670 | * We enqueue to tail when the priority of a task is | |
3671 | * increased (user space view). | |
3672 | */ | |
3673 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3674 | } | |
cb469845 | 3675 | |
da7a735e | 3676 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3677 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3678 | |
95e02ca9 TG |
3679 | rt_mutex_adjust_pi(p); |
3680 | ||
1da177e4 LT |
3681 | return 0; |
3682 | } | |
961ccddd | 3683 | |
7479f3c9 PZ |
3684 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3685 | const struct sched_param *param, bool check) | |
3686 | { | |
3687 | struct sched_attr attr = { | |
3688 | .sched_policy = policy, | |
3689 | .sched_priority = param->sched_priority, | |
3690 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3691 | }; | |
3692 | ||
c13db6b1 SR |
3693 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3694 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3695 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3696 | policy &= ~SCHED_RESET_ON_FORK; | |
3697 | attr.sched_policy = policy; | |
3698 | } | |
3699 | ||
3700 | return __sched_setscheduler(p, &attr, check); | |
3701 | } | |
961ccddd RR |
3702 | /** |
3703 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3704 | * @p: the task in question. | |
3705 | * @policy: new policy. | |
3706 | * @param: structure containing the new RT priority. | |
3707 | * | |
e69f6186 YB |
3708 | * Return: 0 on success. An error code otherwise. |
3709 | * | |
961ccddd RR |
3710 | * NOTE that the task may be already dead. |
3711 | */ | |
3712 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3713 | const struct sched_param *param) |
961ccddd | 3714 | { |
7479f3c9 | 3715 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3716 | } |
1da177e4 LT |
3717 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3718 | ||
d50dde5a DF |
3719 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3720 | { | |
3721 | return __sched_setscheduler(p, attr, true); | |
3722 | } | |
3723 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3724 | ||
961ccddd RR |
3725 | /** |
3726 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3727 | * @p: the task in question. | |
3728 | * @policy: new policy. | |
3729 | * @param: structure containing the new RT priority. | |
3730 | * | |
3731 | * Just like sched_setscheduler, only don't bother checking if the | |
3732 | * current context has permission. For example, this is needed in | |
3733 | * stop_machine(): we create temporary high priority worker threads, | |
3734 | * but our caller might not have that capability. | |
e69f6186 YB |
3735 | * |
3736 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3737 | */ |
3738 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3739 | const struct sched_param *param) |
961ccddd | 3740 | { |
7479f3c9 | 3741 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3742 | } |
3743 | ||
95cdf3b7 IM |
3744 | static int |
3745 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3746 | { |
1da177e4 LT |
3747 | struct sched_param lparam; |
3748 | struct task_struct *p; | |
36c8b586 | 3749 | int retval; |
1da177e4 LT |
3750 | |
3751 | if (!param || pid < 0) | |
3752 | return -EINVAL; | |
3753 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3754 | return -EFAULT; | |
5fe1d75f ON |
3755 | |
3756 | rcu_read_lock(); | |
3757 | retval = -ESRCH; | |
1da177e4 | 3758 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3759 | if (p != NULL) |
3760 | retval = sched_setscheduler(p, policy, &lparam); | |
3761 | rcu_read_unlock(); | |
36c8b586 | 3762 | |
1da177e4 LT |
3763 | return retval; |
3764 | } | |
3765 | ||
d50dde5a DF |
3766 | /* |
3767 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3768 | */ | |
3769 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3770 | struct sched_attr *attr) | |
3771 | { | |
3772 | u32 size; | |
3773 | int ret; | |
3774 | ||
3775 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3776 | return -EFAULT; | |
3777 | ||
3778 | /* | |
3779 | * zero the full structure, so that a short copy will be nice. | |
3780 | */ | |
3781 | memset(attr, 0, sizeof(*attr)); | |
3782 | ||
3783 | ret = get_user(size, &uattr->size); | |
3784 | if (ret) | |
3785 | return ret; | |
3786 | ||
3787 | if (size > PAGE_SIZE) /* silly large */ | |
3788 | goto err_size; | |
3789 | ||
3790 | if (!size) /* abi compat */ | |
3791 | size = SCHED_ATTR_SIZE_VER0; | |
3792 | ||
3793 | if (size < SCHED_ATTR_SIZE_VER0) | |
3794 | goto err_size; | |
3795 | ||
3796 | /* | |
3797 | * If we're handed a bigger struct than we know of, | |
3798 | * ensure all the unknown bits are 0 - i.e. new | |
3799 | * user-space does not rely on any kernel feature | |
3800 | * extensions we dont know about yet. | |
3801 | */ | |
3802 | if (size > sizeof(*attr)) { | |
3803 | unsigned char __user *addr; | |
3804 | unsigned char __user *end; | |
3805 | unsigned char val; | |
3806 | ||
3807 | addr = (void __user *)uattr + sizeof(*attr); | |
3808 | end = (void __user *)uattr + size; | |
3809 | ||
3810 | for (; addr < end; addr++) { | |
3811 | ret = get_user(val, addr); | |
3812 | if (ret) | |
3813 | return ret; | |
3814 | if (val) | |
3815 | goto err_size; | |
3816 | } | |
3817 | size = sizeof(*attr); | |
3818 | } | |
3819 | ||
3820 | ret = copy_from_user(attr, uattr, size); | |
3821 | if (ret) | |
3822 | return -EFAULT; | |
3823 | ||
3824 | /* | |
3825 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3826 | * to be strict and return an error on out-of-bounds values? | |
3827 | */ | |
75e45d51 | 3828 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3829 | |
e78c7bca | 3830 | return 0; |
d50dde5a DF |
3831 | |
3832 | err_size: | |
3833 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3834 | return -E2BIG; |
d50dde5a DF |
3835 | } |
3836 | ||
1da177e4 LT |
3837 | /** |
3838 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3839 | * @pid: the pid in question. | |
3840 | * @policy: new policy. | |
3841 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3842 | * |
3843 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3844 | */ |
5add95d4 HC |
3845 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3846 | struct sched_param __user *, param) | |
1da177e4 | 3847 | { |
c21761f1 JB |
3848 | /* negative values for policy are not valid */ |
3849 | if (policy < 0) | |
3850 | return -EINVAL; | |
3851 | ||
1da177e4 LT |
3852 | return do_sched_setscheduler(pid, policy, param); |
3853 | } | |
3854 | ||
3855 | /** | |
3856 | * sys_sched_setparam - set/change the RT priority of a thread | |
3857 | * @pid: the pid in question. | |
3858 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3859 | * |
3860 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3861 | */ |
5add95d4 | 3862 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3863 | { |
c13db6b1 | 3864 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
3865 | } |
3866 | ||
d50dde5a DF |
3867 | /** |
3868 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3869 | * @pid: the pid in question. | |
5778fccf | 3870 | * @uattr: structure containing the extended parameters. |
db66d756 | 3871 | * @flags: for future extension. |
d50dde5a | 3872 | */ |
6d35ab48 PZ |
3873 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3874 | unsigned int, flags) | |
d50dde5a DF |
3875 | { |
3876 | struct sched_attr attr; | |
3877 | struct task_struct *p; | |
3878 | int retval; | |
3879 | ||
6d35ab48 | 3880 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3881 | return -EINVAL; |
3882 | ||
143cf23d MK |
3883 | retval = sched_copy_attr(uattr, &attr); |
3884 | if (retval) | |
3885 | return retval; | |
d50dde5a | 3886 | |
b14ed2c2 | 3887 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3888 | return -EINVAL; |
d50dde5a DF |
3889 | |
3890 | rcu_read_lock(); | |
3891 | retval = -ESRCH; | |
3892 | p = find_process_by_pid(pid); | |
3893 | if (p != NULL) | |
3894 | retval = sched_setattr(p, &attr); | |
3895 | rcu_read_unlock(); | |
3896 | ||
3897 | return retval; | |
3898 | } | |
3899 | ||
1da177e4 LT |
3900 | /** |
3901 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3902 | * @pid: the pid in question. | |
e69f6186 YB |
3903 | * |
3904 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3905 | * code. | |
1da177e4 | 3906 | */ |
5add95d4 | 3907 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3908 | { |
36c8b586 | 3909 | struct task_struct *p; |
3a5c359a | 3910 | int retval; |
1da177e4 LT |
3911 | |
3912 | if (pid < 0) | |
3a5c359a | 3913 | return -EINVAL; |
1da177e4 LT |
3914 | |
3915 | retval = -ESRCH; | |
5fe85be0 | 3916 | rcu_read_lock(); |
1da177e4 LT |
3917 | p = find_process_by_pid(pid); |
3918 | if (p) { | |
3919 | retval = security_task_getscheduler(p); | |
3920 | if (!retval) | |
ca94c442 LP |
3921 | retval = p->policy |
3922 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3923 | } |
5fe85be0 | 3924 | rcu_read_unlock(); |
1da177e4 LT |
3925 | return retval; |
3926 | } | |
3927 | ||
3928 | /** | |
ca94c442 | 3929 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3930 | * @pid: the pid in question. |
3931 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3932 | * |
3933 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3934 | * code. | |
1da177e4 | 3935 | */ |
5add95d4 | 3936 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3937 | { |
ce5f7f82 | 3938 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3939 | struct task_struct *p; |
3a5c359a | 3940 | int retval; |
1da177e4 LT |
3941 | |
3942 | if (!param || pid < 0) | |
3a5c359a | 3943 | return -EINVAL; |
1da177e4 | 3944 | |
5fe85be0 | 3945 | rcu_read_lock(); |
1da177e4 LT |
3946 | p = find_process_by_pid(pid); |
3947 | retval = -ESRCH; | |
3948 | if (!p) | |
3949 | goto out_unlock; | |
3950 | ||
3951 | retval = security_task_getscheduler(p); | |
3952 | if (retval) | |
3953 | goto out_unlock; | |
3954 | ||
ce5f7f82 PZ |
3955 | if (task_has_rt_policy(p)) |
3956 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3957 | rcu_read_unlock(); |
1da177e4 LT |
3958 | |
3959 | /* | |
3960 | * This one might sleep, we cannot do it with a spinlock held ... | |
3961 | */ | |
3962 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3963 | ||
1da177e4 LT |
3964 | return retval; |
3965 | ||
3966 | out_unlock: | |
5fe85be0 | 3967 | rcu_read_unlock(); |
1da177e4 LT |
3968 | return retval; |
3969 | } | |
3970 | ||
d50dde5a DF |
3971 | static int sched_read_attr(struct sched_attr __user *uattr, |
3972 | struct sched_attr *attr, | |
3973 | unsigned int usize) | |
3974 | { | |
3975 | int ret; | |
3976 | ||
3977 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3978 | return -EFAULT; | |
3979 | ||
3980 | /* | |
3981 | * If we're handed a smaller struct than we know of, | |
3982 | * ensure all the unknown bits are 0 - i.e. old | |
3983 | * user-space does not get uncomplete information. | |
3984 | */ | |
3985 | if (usize < sizeof(*attr)) { | |
3986 | unsigned char *addr; | |
3987 | unsigned char *end; | |
3988 | ||
3989 | addr = (void *)attr + usize; | |
3990 | end = (void *)attr + sizeof(*attr); | |
3991 | ||
3992 | for (; addr < end; addr++) { | |
3993 | if (*addr) | |
22400674 | 3994 | return -EFBIG; |
d50dde5a DF |
3995 | } |
3996 | ||
3997 | attr->size = usize; | |
3998 | } | |
3999 | ||
4efbc454 | 4000 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4001 | if (ret) |
4002 | return -EFAULT; | |
4003 | ||
22400674 | 4004 | return 0; |
d50dde5a DF |
4005 | } |
4006 | ||
4007 | /** | |
aab03e05 | 4008 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4009 | * @pid: the pid in question. |
5778fccf | 4010 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4011 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4012 | * @flags: for future extension. |
d50dde5a | 4013 | */ |
6d35ab48 PZ |
4014 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4015 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4016 | { |
4017 | struct sched_attr attr = { | |
4018 | .size = sizeof(struct sched_attr), | |
4019 | }; | |
4020 | struct task_struct *p; | |
4021 | int retval; | |
4022 | ||
4023 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4024 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4025 | return -EINVAL; |
4026 | ||
4027 | rcu_read_lock(); | |
4028 | p = find_process_by_pid(pid); | |
4029 | retval = -ESRCH; | |
4030 | if (!p) | |
4031 | goto out_unlock; | |
4032 | ||
4033 | retval = security_task_getscheduler(p); | |
4034 | if (retval) | |
4035 | goto out_unlock; | |
4036 | ||
4037 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4038 | if (p->sched_reset_on_fork) |
4039 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4040 | if (task_has_dl_policy(p)) |
4041 | __getparam_dl(p, &attr); | |
4042 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4043 | attr.sched_priority = p->rt_priority; |
4044 | else | |
d0ea0268 | 4045 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4046 | |
4047 | rcu_read_unlock(); | |
4048 | ||
4049 | retval = sched_read_attr(uattr, &attr, size); | |
4050 | return retval; | |
4051 | ||
4052 | out_unlock: | |
4053 | rcu_read_unlock(); | |
4054 | return retval; | |
4055 | } | |
4056 | ||
96f874e2 | 4057 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4058 | { |
5a16f3d3 | 4059 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4060 | struct task_struct *p; |
4061 | int retval; | |
1da177e4 | 4062 | |
23f5d142 | 4063 | rcu_read_lock(); |
1da177e4 LT |
4064 | |
4065 | p = find_process_by_pid(pid); | |
4066 | if (!p) { | |
23f5d142 | 4067 | rcu_read_unlock(); |
1da177e4 LT |
4068 | return -ESRCH; |
4069 | } | |
4070 | ||
23f5d142 | 4071 | /* Prevent p going away */ |
1da177e4 | 4072 | get_task_struct(p); |
23f5d142 | 4073 | rcu_read_unlock(); |
1da177e4 | 4074 | |
14a40ffc TH |
4075 | if (p->flags & PF_NO_SETAFFINITY) { |
4076 | retval = -EINVAL; | |
4077 | goto out_put_task; | |
4078 | } | |
5a16f3d3 RR |
4079 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4080 | retval = -ENOMEM; | |
4081 | goto out_put_task; | |
4082 | } | |
4083 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4084 | retval = -ENOMEM; | |
4085 | goto out_free_cpus_allowed; | |
4086 | } | |
1da177e4 | 4087 | retval = -EPERM; |
4c44aaaf EB |
4088 | if (!check_same_owner(p)) { |
4089 | rcu_read_lock(); | |
4090 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4091 | rcu_read_unlock(); | |
16303ab2 | 4092 | goto out_free_new_mask; |
4c44aaaf EB |
4093 | } |
4094 | rcu_read_unlock(); | |
4095 | } | |
1da177e4 | 4096 | |
b0ae1981 | 4097 | retval = security_task_setscheduler(p); |
e7834f8f | 4098 | if (retval) |
16303ab2 | 4099 | goto out_free_new_mask; |
e7834f8f | 4100 | |
e4099a5e PZ |
4101 | |
4102 | cpuset_cpus_allowed(p, cpus_allowed); | |
4103 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4104 | ||
332ac17e DF |
4105 | /* |
4106 | * Since bandwidth control happens on root_domain basis, | |
4107 | * if admission test is enabled, we only admit -deadline | |
4108 | * tasks allowed to run on all the CPUs in the task's | |
4109 | * root_domain. | |
4110 | */ | |
4111 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4112 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4113 | rcu_read_lock(); | |
4114 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4115 | retval = -EBUSY; |
f1e3a093 | 4116 | rcu_read_unlock(); |
16303ab2 | 4117 | goto out_free_new_mask; |
332ac17e | 4118 | } |
f1e3a093 | 4119 | rcu_read_unlock(); |
332ac17e DF |
4120 | } |
4121 | #endif | |
49246274 | 4122 | again: |
5a16f3d3 | 4123 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4124 | |
8707d8b8 | 4125 | if (!retval) { |
5a16f3d3 RR |
4126 | cpuset_cpus_allowed(p, cpus_allowed); |
4127 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4128 | /* |
4129 | * We must have raced with a concurrent cpuset | |
4130 | * update. Just reset the cpus_allowed to the | |
4131 | * cpuset's cpus_allowed | |
4132 | */ | |
5a16f3d3 | 4133 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4134 | goto again; |
4135 | } | |
4136 | } | |
16303ab2 | 4137 | out_free_new_mask: |
5a16f3d3 RR |
4138 | free_cpumask_var(new_mask); |
4139 | out_free_cpus_allowed: | |
4140 | free_cpumask_var(cpus_allowed); | |
4141 | out_put_task: | |
1da177e4 | 4142 | put_task_struct(p); |
1da177e4 LT |
4143 | return retval; |
4144 | } | |
4145 | ||
4146 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4147 | struct cpumask *new_mask) |
1da177e4 | 4148 | { |
96f874e2 RR |
4149 | if (len < cpumask_size()) |
4150 | cpumask_clear(new_mask); | |
4151 | else if (len > cpumask_size()) | |
4152 | len = cpumask_size(); | |
4153 | ||
1da177e4 LT |
4154 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4155 | } | |
4156 | ||
4157 | /** | |
4158 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4159 | * @pid: pid of the process | |
4160 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4161 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4162 | * |
4163 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4164 | */ |
5add95d4 HC |
4165 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4166 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4167 | { |
5a16f3d3 | 4168 | cpumask_var_t new_mask; |
1da177e4 LT |
4169 | int retval; |
4170 | ||
5a16f3d3 RR |
4171 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4172 | return -ENOMEM; | |
1da177e4 | 4173 | |
5a16f3d3 RR |
4174 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4175 | if (retval == 0) | |
4176 | retval = sched_setaffinity(pid, new_mask); | |
4177 | free_cpumask_var(new_mask); | |
4178 | return retval; | |
1da177e4 LT |
4179 | } |
4180 | ||
96f874e2 | 4181 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4182 | { |
36c8b586 | 4183 | struct task_struct *p; |
31605683 | 4184 | unsigned long flags; |
1da177e4 | 4185 | int retval; |
1da177e4 | 4186 | |
23f5d142 | 4187 | rcu_read_lock(); |
1da177e4 LT |
4188 | |
4189 | retval = -ESRCH; | |
4190 | p = find_process_by_pid(pid); | |
4191 | if (!p) | |
4192 | goto out_unlock; | |
4193 | ||
e7834f8f DQ |
4194 | retval = security_task_getscheduler(p); |
4195 | if (retval) | |
4196 | goto out_unlock; | |
4197 | ||
013fdb80 | 4198 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4199 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4200 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4201 | |
4202 | out_unlock: | |
23f5d142 | 4203 | rcu_read_unlock(); |
1da177e4 | 4204 | |
9531b62f | 4205 | return retval; |
1da177e4 LT |
4206 | } |
4207 | ||
4208 | /** | |
4209 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4210 | * @pid: pid of the process | |
4211 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4212 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4213 | * |
4214 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4215 | */ |
5add95d4 HC |
4216 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4217 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4218 | { |
4219 | int ret; | |
f17c8607 | 4220 | cpumask_var_t mask; |
1da177e4 | 4221 | |
84fba5ec | 4222 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4223 | return -EINVAL; |
4224 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4225 | return -EINVAL; |
4226 | ||
f17c8607 RR |
4227 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4228 | return -ENOMEM; | |
1da177e4 | 4229 | |
f17c8607 RR |
4230 | ret = sched_getaffinity(pid, mask); |
4231 | if (ret == 0) { | |
8bc037fb | 4232 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4233 | |
4234 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4235 | ret = -EFAULT; |
4236 | else | |
cd3d8031 | 4237 | ret = retlen; |
f17c8607 RR |
4238 | } |
4239 | free_cpumask_var(mask); | |
1da177e4 | 4240 | |
f17c8607 | 4241 | return ret; |
1da177e4 LT |
4242 | } |
4243 | ||
4244 | /** | |
4245 | * sys_sched_yield - yield the current processor to other threads. | |
4246 | * | |
dd41f596 IM |
4247 | * This function yields the current CPU to other tasks. If there are no |
4248 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4249 | * |
4250 | * Return: 0. | |
1da177e4 | 4251 | */ |
5add95d4 | 4252 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4253 | { |
70b97a7f | 4254 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4255 | |
2d72376b | 4256 | schedstat_inc(rq, yld_count); |
4530d7ab | 4257 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4258 | |
4259 | /* | |
4260 | * Since we are going to call schedule() anyway, there's | |
4261 | * no need to preempt or enable interrupts: | |
4262 | */ | |
4263 | __release(rq->lock); | |
8a25d5de | 4264 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4265 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4266 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4267 | |
4268 | schedule(); | |
4269 | ||
4270 | return 0; | |
4271 | } | |
4272 | ||
02b67cc3 | 4273 | int __sched _cond_resched(void) |
1da177e4 | 4274 | { |
d86ee480 | 4275 | if (should_resched()) { |
a18b5d01 | 4276 | preempt_schedule_common(); |
1da177e4 LT |
4277 | return 1; |
4278 | } | |
4279 | return 0; | |
4280 | } | |
02b67cc3 | 4281 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4282 | |
4283 | /* | |
613afbf8 | 4284 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4285 | * call schedule, and on return reacquire the lock. |
4286 | * | |
41a2d6cf | 4287 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4288 | * operations here to prevent schedule() from being called twice (once via |
4289 | * spin_unlock(), once by hand). | |
4290 | */ | |
613afbf8 | 4291 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4292 | { |
d86ee480 | 4293 | int resched = should_resched(); |
6df3cecb JK |
4294 | int ret = 0; |
4295 | ||
f607c668 PZ |
4296 | lockdep_assert_held(lock); |
4297 | ||
4a81e832 | 4298 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4299 | spin_unlock(lock); |
d86ee480 | 4300 | if (resched) |
a18b5d01 | 4301 | preempt_schedule_common(); |
95c354fe NP |
4302 | else |
4303 | cpu_relax(); | |
6df3cecb | 4304 | ret = 1; |
1da177e4 | 4305 | spin_lock(lock); |
1da177e4 | 4306 | } |
6df3cecb | 4307 | return ret; |
1da177e4 | 4308 | } |
613afbf8 | 4309 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4310 | |
613afbf8 | 4311 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4312 | { |
4313 | BUG_ON(!in_softirq()); | |
4314 | ||
d86ee480 | 4315 | if (should_resched()) { |
98d82567 | 4316 | local_bh_enable(); |
a18b5d01 | 4317 | preempt_schedule_common(); |
1da177e4 LT |
4318 | local_bh_disable(); |
4319 | return 1; | |
4320 | } | |
4321 | return 0; | |
4322 | } | |
613afbf8 | 4323 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4324 | |
1da177e4 LT |
4325 | /** |
4326 | * yield - yield the current processor to other threads. | |
4327 | * | |
8e3fabfd PZ |
4328 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4329 | * | |
4330 | * The scheduler is at all times free to pick the calling task as the most | |
4331 | * eligible task to run, if removing the yield() call from your code breaks | |
4332 | * it, its already broken. | |
4333 | * | |
4334 | * Typical broken usage is: | |
4335 | * | |
4336 | * while (!event) | |
4337 | * yield(); | |
4338 | * | |
4339 | * where one assumes that yield() will let 'the other' process run that will | |
4340 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4341 | * happen. Never use yield() as a progress guarantee!! | |
4342 | * | |
4343 | * If you want to use yield() to wait for something, use wait_event(). | |
4344 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4345 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4346 | */ |
4347 | void __sched yield(void) | |
4348 | { | |
4349 | set_current_state(TASK_RUNNING); | |
4350 | sys_sched_yield(); | |
4351 | } | |
1da177e4 LT |
4352 | EXPORT_SYMBOL(yield); |
4353 | ||
d95f4122 MG |
4354 | /** |
4355 | * yield_to - yield the current processor to another thread in | |
4356 | * your thread group, or accelerate that thread toward the | |
4357 | * processor it's on. | |
16addf95 RD |
4358 | * @p: target task |
4359 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4360 | * |
4361 | * It's the caller's job to ensure that the target task struct | |
4362 | * can't go away on us before we can do any checks. | |
4363 | * | |
e69f6186 | 4364 | * Return: |
7b270f60 PZ |
4365 | * true (>0) if we indeed boosted the target task. |
4366 | * false (0) if we failed to boost the target. | |
4367 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4368 | */ |
fa93384f | 4369 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4370 | { |
4371 | struct task_struct *curr = current; | |
4372 | struct rq *rq, *p_rq; | |
4373 | unsigned long flags; | |
c3c18640 | 4374 | int yielded = 0; |
d95f4122 MG |
4375 | |
4376 | local_irq_save(flags); | |
4377 | rq = this_rq(); | |
4378 | ||
4379 | again: | |
4380 | p_rq = task_rq(p); | |
7b270f60 PZ |
4381 | /* |
4382 | * If we're the only runnable task on the rq and target rq also | |
4383 | * has only one task, there's absolutely no point in yielding. | |
4384 | */ | |
4385 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4386 | yielded = -ESRCH; | |
4387 | goto out_irq; | |
4388 | } | |
4389 | ||
d95f4122 | 4390 | double_rq_lock(rq, p_rq); |
39e24d8f | 4391 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4392 | double_rq_unlock(rq, p_rq); |
4393 | goto again; | |
4394 | } | |
4395 | ||
4396 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4397 | goto out_unlock; |
d95f4122 MG |
4398 | |
4399 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4400 | goto out_unlock; |
d95f4122 MG |
4401 | |
4402 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4403 | goto out_unlock; |
d95f4122 MG |
4404 | |
4405 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4406 | if (yielded) { |
d95f4122 | 4407 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4408 | /* |
4409 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4410 | * fairness. | |
4411 | */ | |
4412 | if (preempt && rq != p_rq) | |
8875125e | 4413 | resched_curr(p_rq); |
6d1cafd8 | 4414 | } |
d95f4122 | 4415 | |
7b270f60 | 4416 | out_unlock: |
d95f4122 | 4417 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4418 | out_irq: |
d95f4122 MG |
4419 | local_irq_restore(flags); |
4420 | ||
7b270f60 | 4421 | if (yielded > 0) |
d95f4122 MG |
4422 | schedule(); |
4423 | ||
4424 | return yielded; | |
4425 | } | |
4426 | EXPORT_SYMBOL_GPL(yield_to); | |
4427 | ||
1da177e4 | 4428 | /* |
41a2d6cf | 4429 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4430 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 4431 | */ |
1da177e4 LT |
4432 | long __sched io_schedule_timeout(long timeout) |
4433 | { | |
9cff8ade N |
4434 | int old_iowait = current->in_iowait; |
4435 | struct rq *rq; | |
1da177e4 LT |
4436 | long ret; |
4437 | ||
9cff8ade N |
4438 | current->in_iowait = 1; |
4439 | if (old_iowait) | |
4440 | blk_schedule_flush_plug(current); | |
4441 | else | |
4442 | blk_flush_plug(current); | |
4443 | ||
0ff92245 | 4444 | delayacct_blkio_start(); |
9cff8ade | 4445 | rq = raw_rq(); |
1da177e4 LT |
4446 | atomic_inc(&rq->nr_iowait); |
4447 | ret = schedule_timeout(timeout); | |
9cff8ade | 4448 | current->in_iowait = old_iowait; |
1da177e4 | 4449 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4450 | delayacct_blkio_end(); |
9cff8ade | 4451 | |
1da177e4 LT |
4452 | return ret; |
4453 | } | |
9cff8ade | 4454 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 LT |
4455 | |
4456 | /** | |
4457 | * sys_sched_get_priority_max - return maximum RT priority. | |
4458 | * @policy: scheduling class. | |
4459 | * | |
e69f6186 YB |
4460 | * Return: On success, this syscall returns the maximum |
4461 | * rt_priority that can be used by a given scheduling class. | |
4462 | * On failure, a negative error code is returned. | |
1da177e4 | 4463 | */ |
5add95d4 | 4464 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4465 | { |
4466 | int ret = -EINVAL; | |
4467 | ||
4468 | switch (policy) { | |
4469 | case SCHED_FIFO: | |
4470 | case SCHED_RR: | |
4471 | ret = MAX_USER_RT_PRIO-1; | |
4472 | break; | |
aab03e05 | 4473 | case SCHED_DEADLINE: |
1da177e4 | 4474 | case SCHED_NORMAL: |
b0a9499c | 4475 | case SCHED_BATCH: |
dd41f596 | 4476 | case SCHED_IDLE: |
1da177e4 LT |
4477 | ret = 0; |
4478 | break; | |
4479 | } | |
4480 | return ret; | |
4481 | } | |
4482 | ||
4483 | /** | |
4484 | * sys_sched_get_priority_min - return minimum RT priority. | |
4485 | * @policy: scheduling class. | |
4486 | * | |
e69f6186 YB |
4487 | * Return: On success, this syscall returns the minimum |
4488 | * rt_priority that can be used by a given scheduling class. | |
4489 | * On failure, a negative error code is returned. | |
1da177e4 | 4490 | */ |
5add95d4 | 4491 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4492 | { |
4493 | int ret = -EINVAL; | |
4494 | ||
4495 | switch (policy) { | |
4496 | case SCHED_FIFO: | |
4497 | case SCHED_RR: | |
4498 | ret = 1; | |
4499 | break; | |
aab03e05 | 4500 | case SCHED_DEADLINE: |
1da177e4 | 4501 | case SCHED_NORMAL: |
b0a9499c | 4502 | case SCHED_BATCH: |
dd41f596 | 4503 | case SCHED_IDLE: |
1da177e4 LT |
4504 | ret = 0; |
4505 | } | |
4506 | return ret; | |
4507 | } | |
4508 | ||
4509 | /** | |
4510 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4511 | * @pid: pid of the process. | |
4512 | * @interval: userspace pointer to the timeslice value. | |
4513 | * | |
4514 | * this syscall writes the default timeslice value of a given process | |
4515 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4516 | * |
4517 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4518 | * an error code. | |
1da177e4 | 4519 | */ |
17da2bd9 | 4520 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4521 | struct timespec __user *, interval) |
1da177e4 | 4522 | { |
36c8b586 | 4523 | struct task_struct *p; |
a4ec24b4 | 4524 | unsigned int time_slice; |
dba091b9 TG |
4525 | unsigned long flags; |
4526 | struct rq *rq; | |
3a5c359a | 4527 | int retval; |
1da177e4 | 4528 | struct timespec t; |
1da177e4 LT |
4529 | |
4530 | if (pid < 0) | |
3a5c359a | 4531 | return -EINVAL; |
1da177e4 LT |
4532 | |
4533 | retval = -ESRCH; | |
1a551ae7 | 4534 | rcu_read_lock(); |
1da177e4 LT |
4535 | p = find_process_by_pid(pid); |
4536 | if (!p) | |
4537 | goto out_unlock; | |
4538 | ||
4539 | retval = security_task_getscheduler(p); | |
4540 | if (retval) | |
4541 | goto out_unlock; | |
4542 | ||
dba091b9 | 4543 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4544 | time_slice = 0; |
4545 | if (p->sched_class->get_rr_interval) | |
4546 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4547 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4548 | |
1a551ae7 | 4549 | rcu_read_unlock(); |
a4ec24b4 | 4550 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4551 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4552 | return retval; |
3a5c359a | 4553 | |
1da177e4 | 4554 | out_unlock: |
1a551ae7 | 4555 | rcu_read_unlock(); |
1da177e4 LT |
4556 | return retval; |
4557 | } | |
4558 | ||
7c731e0a | 4559 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4560 | |
82a1fcb9 | 4561 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4562 | { |
1da177e4 | 4563 | unsigned long free = 0; |
4e79752c | 4564 | int ppid; |
1f8a7633 | 4565 | unsigned long state = p->state; |
1da177e4 | 4566 | |
1f8a7633 TH |
4567 | if (state) |
4568 | state = __ffs(state) + 1; | |
28d0686c | 4569 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4570 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4571 | #if BITS_PER_LONG == 32 |
1da177e4 | 4572 | if (state == TASK_RUNNING) |
3df0fc5b | 4573 | printk(KERN_CONT " running "); |
1da177e4 | 4574 | else |
3df0fc5b | 4575 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4576 | #else |
4577 | if (state == TASK_RUNNING) | |
3df0fc5b | 4578 | printk(KERN_CONT " running task "); |
1da177e4 | 4579 | else |
3df0fc5b | 4580 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4581 | #endif |
4582 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4583 | free = stack_not_used(p); |
1da177e4 | 4584 | #endif |
a90e984c | 4585 | ppid = 0; |
4e79752c | 4586 | rcu_read_lock(); |
a90e984c ON |
4587 | if (pid_alive(p)) |
4588 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 4589 | rcu_read_unlock(); |
3df0fc5b | 4590 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4591 | task_pid_nr(p), ppid, |
aa47b7e0 | 4592 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4593 | |
3d1cb205 | 4594 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4595 | show_stack(p, NULL); |
1da177e4 LT |
4596 | } |
4597 | ||
e59e2ae2 | 4598 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4599 | { |
36c8b586 | 4600 | struct task_struct *g, *p; |
1da177e4 | 4601 | |
4bd77321 | 4602 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4603 | printk(KERN_INFO |
4604 | " task PC stack pid father\n"); | |
1da177e4 | 4605 | #else |
3df0fc5b PZ |
4606 | printk(KERN_INFO |
4607 | " task PC stack pid father\n"); | |
1da177e4 | 4608 | #endif |
510f5acc | 4609 | rcu_read_lock(); |
5d07f420 | 4610 | for_each_process_thread(g, p) { |
1da177e4 LT |
4611 | /* |
4612 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4613 | * console might take a lot of time: |
1da177e4 LT |
4614 | */ |
4615 | touch_nmi_watchdog(); | |
39bc89fd | 4616 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4617 | sched_show_task(p); |
5d07f420 | 4618 | } |
1da177e4 | 4619 | |
04c9167f JF |
4620 | touch_all_softlockup_watchdogs(); |
4621 | ||
dd41f596 IM |
4622 | #ifdef CONFIG_SCHED_DEBUG |
4623 | sysrq_sched_debug_show(); | |
4624 | #endif | |
510f5acc | 4625 | rcu_read_unlock(); |
e59e2ae2 IM |
4626 | /* |
4627 | * Only show locks if all tasks are dumped: | |
4628 | */ | |
93335a21 | 4629 | if (!state_filter) |
e59e2ae2 | 4630 | debug_show_all_locks(); |
1da177e4 LT |
4631 | } |
4632 | ||
0db0628d | 4633 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4634 | { |
dd41f596 | 4635 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4636 | } |
4637 | ||
f340c0d1 IM |
4638 | /** |
4639 | * init_idle - set up an idle thread for a given CPU | |
4640 | * @idle: task in question | |
4641 | * @cpu: cpu the idle task belongs to | |
4642 | * | |
4643 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4644 | * flag, to make booting more robust. | |
4645 | */ | |
0db0628d | 4646 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4647 | { |
70b97a7f | 4648 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4649 | unsigned long flags; |
4650 | ||
05fa785c | 4651 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4652 | |
5e1576ed | 4653 | __sched_fork(0, idle); |
06b83b5f | 4654 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4655 | idle->se.exec_start = sched_clock(); |
4656 | ||
1e1b6c51 | 4657 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4658 | /* |
4659 | * We're having a chicken and egg problem, even though we are | |
4660 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4661 | * lockdep check in task_group() will fail. | |
4662 | * | |
4663 | * Similar case to sched_fork(). / Alternatively we could | |
4664 | * use task_rq_lock() here and obtain the other rq->lock. | |
4665 | * | |
4666 | * Silence PROVE_RCU | |
4667 | */ | |
4668 | rcu_read_lock(); | |
dd41f596 | 4669 | __set_task_cpu(idle, cpu); |
6506cf6c | 4670 | rcu_read_unlock(); |
1da177e4 | 4671 | |
1da177e4 | 4672 | rq->curr = rq->idle = idle; |
da0c1e65 | 4673 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4674 | #if defined(CONFIG_SMP) |
4675 | idle->on_cpu = 1; | |
4866cde0 | 4676 | #endif |
05fa785c | 4677 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4678 | |
4679 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4680 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4681 | |
dd41f596 IM |
4682 | /* |
4683 | * The idle tasks have their own, simple scheduling class: | |
4684 | */ | |
4685 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4686 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4687 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4688 | #if defined(CONFIG_SMP) |
4689 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4690 | #endif | |
19978ca6 IM |
4691 | } |
4692 | ||
f82f8042 JL |
4693 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
4694 | const struct cpumask *trial) | |
4695 | { | |
4696 | int ret = 1, trial_cpus; | |
4697 | struct dl_bw *cur_dl_b; | |
4698 | unsigned long flags; | |
4699 | ||
bb2bc55a MG |
4700 | if (!cpumask_weight(cur)) |
4701 | return ret; | |
4702 | ||
75e23e49 | 4703 | rcu_read_lock_sched(); |
f82f8042 JL |
4704 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
4705 | trial_cpus = cpumask_weight(trial); | |
4706 | ||
4707 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
4708 | if (cur_dl_b->bw != -1 && | |
4709 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
4710 | ret = 0; | |
4711 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 4712 | rcu_read_unlock_sched(); |
f82f8042 JL |
4713 | |
4714 | return ret; | |
4715 | } | |
4716 | ||
7f51412a JL |
4717 | int task_can_attach(struct task_struct *p, |
4718 | const struct cpumask *cs_cpus_allowed) | |
4719 | { | |
4720 | int ret = 0; | |
4721 | ||
4722 | /* | |
4723 | * Kthreads which disallow setaffinity shouldn't be moved | |
4724 | * to a new cpuset; we don't want to change their cpu | |
4725 | * affinity and isolating such threads by their set of | |
4726 | * allowed nodes is unnecessary. Thus, cpusets are not | |
4727 | * applicable for such threads. This prevents checking for | |
4728 | * success of set_cpus_allowed_ptr() on all attached tasks | |
4729 | * before cpus_allowed may be changed. | |
4730 | */ | |
4731 | if (p->flags & PF_NO_SETAFFINITY) { | |
4732 | ret = -EINVAL; | |
4733 | goto out; | |
4734 | } | |
4735 | ||
4736 | #ifdef CONFIG_SMP | |
4737 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
4738 | cs_cpus_allowed)) { | |
4739 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
4740 | cs_cpus_allowed); | |
75e23e49 | 4741 | struct dl_bw *dl_b; |
7f51412a JL |
4742 | bool overflow; |
4743 | int cpus; | |
4744 | unsigned long flags; | |
4745 | ||
75e23e49 JL |
4746 | rcu_read_lock_sched(); |
4747 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
4748 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
4749 | cpus = dl_bw_cpus(dest_cpu); | |
4750 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
4751 | if (overflow) | |
4752 | ret = -EBUSY; | |
4753 | else { | |
4754 | /* | |
4755 | * We reserve space for this task in the destination | |
4756 | * root_domain, as we can't fail after this point. | |
4757 | * We will free resources in the source root_domain | |
4758 | * later on (see set_cpus_allowed_dl()). | |
4759 | */ | |
4760 | __dl_add(dl_b, p->dl.dl_bw); | |
4761 | } | |
4762 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 4763 | rcu_read_unlock_sched(); |
7f51412a JL |
4764 | |
4765 | } | |
4766 | #endif | |
4767 | out: | |
4768 | return ret; | |
4769 | } | |
4770 | ||
1da177e4 | 4771 | #ifdef CONFIG_SMP |
a15b12ac KT |
4772 | /* |
4773 | * move_queued_task - move a queued task to new rq. | |
4774 | * | |
4775 | * Returns (locked) new rq. Old rq's lock is released. | |
4776 | */ | |
4777 | static struct rq *move_queued_task(struct task_struct *p, int new_cpu) | |
4778 | { | |
4779 | struct rq *rq = task_rq(p); | |
4780 | ||
4781 | lockdep_assert_held(&rq->lock); | |
4782 | ||
4783 | dequeue_task(rq, p, 0); | |
4784 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
4785 | set_task_cpu(p, new_cpu); | |
4786 | raw_spin_unlock(&rq->lock); | |
4787 | ||
4788 | rq = cpu_rq(new_cpu); | |
4789 | ||
4790 | raw_spin_lock(&rq->lock); | |
4791 | BUG_ON(task_cpu(p) != new_cpu); | |
4792 | p->on_rq = TASK_ON_RQ_QUEUED; | |
4793 | enqueue_task(rq, p, 0); | |
4794 | check_preempt_curr(rq, p, 0); | |
4795 | ||
4796 | return rq; | |
4797 | } | |
4798 | ||
1e1b6c51 KM |
4799 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4800 | { | |
1b537c7d | 4801 | if (p->sched_class->set_cpus_allowed) |
1e1b6c51 | 4802 | p->sched_class->set_cpus_allowed(p, new_mask); |
4939602a PZ |
4803 | |
4804 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4805 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4806 | } |
4807 | ||
1da177e4 LT |
4808 | /* |
4809 | * This is how migration works: | |
4810 | * | |
969c7921 TH |
4811 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4812 | * stop_one_cpu(). | |
4813 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4814 | * off the CPU) | |
4815 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4816 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4817 | * it and puts it into the right queue. |
969c7921 TH |
4818 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4819 | * is done. | |
1da177e4 LT |
4820 | */ |
4821 | ||
4822 | /* | |
4823 | * Change a given task's CPU affinity. Migrate the thread to a | |
4824 | * proper CPU and schedule it away if the CPU it's executing on | |
4825 | * is removed from the allowed bitmask. | |
4826 | * | |
4827 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4828 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4829 | * call is not atomic; no spinlocks may be held. |
4830 | */ | |
96f874e2 | 4831 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4832 | { |
4833 | unsigned long flags; | |
70b97a7f | 4834 | struct rq *rq; |
969c7921 | 4835 | unsigned int dest_cpu; |
48f24c4d | 4836 | int ret = 0; |
1da177e4 LT |
4837 | |
4838 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4839 | |
db44fc01 YZ |
4840 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4841 | goto out; | |
4842 | ||
6ad4c188 | 4843 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4844 | ret = -EINVAL; |
4845 | goto out; | |
4846 | } | |
4847 | ||
1e1b6c51 | 4848 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4849 | |
1da177e4 | 4850 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4851 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4852 | goto out; |
4853 | ||
969c7921 | 4854 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
a15b12ac | 4855 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
969c7921 | 4856 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4857 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4858 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4859 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4860 | tlb_migrate_finish(p->mm); |
4861 | return 0; | |
a15b12ac KT |
4862 | } else if (task_on_rq_queued(p)) |
4863 | rq = move_queued_task(p, dest_cpu); | |
1da177e4 | 4864 | out: |
0122ec5b | 4865 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4866 | |
1da177e4 LT |
4867 | return ret; |
4868 | } | |
cd8ba7cd | 4869 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4870 | |
4871 | /* | |
41a2d6cf | 4872 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4873 | * this because either it can't run here any more (set_cpus_allowed() |
4874 | * away from this CPU, or CPU going down), or because we're | |
4875 | * attempting to rebalance this task on exec (sched_exec). | |
4876 | * | |
4877 | * So we race with normal scheduler movements, but that's OK, as long | |
4878 | * as the task is no longer on this CPU. | |
efc30814 KK |
4879 | * |
4880 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4881 | */ |
efc30814 | 4882 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4883 | { |
a1e01829 | 4884 | struct rq *rq; |
e2912009 | 4885 | int ret = 0; |
1da177e4 | 4886 | |
e761b772 | 4887 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4888 | return ret; |
1da177e4 | 4889 | |
a1e01829 | 4890 | rq = cpu_rq(src_cpu); |
1da177e4 | 4891 | |
0122ec5b | 4892 | raw_spin_lock(&p->pi_lock); |
a1e01829 | 4893 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
4894 | /* Already moved. */ |
4895 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4896 | goto done; |
a1e01829 | 4897 | |
1da177e4 | 4898 | /* Affinity changed (again). */ |
fa17b507 | 4899 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4900 | goto fail; |
1da177e4 | 4901 | |
e2912009 PZ |
4902 | /* |
4903 | * If we're not on a rq, the next wake-up will ensure we're | |
4904 | * placed properly. | |
4905 | */ | |
a15b12ac KT |
4906 | if (task_on_rq_queued(p)) |
4907 | rq = move_queued_task(p, dest_cpu); | |
b1e38734 | 4908 | done: |
efc30814 | 4909 | ret = 1; |
b1e38734 | 4910 | fail: |
a1e01829 | 4911 | raw_spin_unlock(&rq->lock); |
0122ec5b | 4912 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4913 | return ret; |
1da177e4 LT |
4914 | } |
4915 | ||
e6628d5b MG |
4916 | #ifdef CONFIG_NUMA_BALANCING |
4917 | /* Migrate current task p to target_cpu */ | |
4918 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4919 | { | |
4920 | struct migration_arg arg = { p, target_cpu }; | |
4921 | int curr_cpu = task_cpu(p); | |
4922 | ||
4923 | if (curr_cpu == target_cpu) | |
4924 | return 0; | |
4925 | ||
4926 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4927 | return -EINVAL; | |
4928 | ||
4929 | /* TODO: This is not properly updating schedstats */ | |
4930 | ||
286549dc | 4931 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4932 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4933 | } | |
0ec8aa00 PZ |
4934 | |
4935 | /* | |
4936 | * Requeue a task on a given node and accurately track the number of NUMA | |
4937 | * tasks on the runqueues | |
4938 | */ | |
4939 | void sched_setnuma(struct task_struct *p, int nid) | |
4940 | { | |
4941 | struct rq *rq; | |
4942 | unsigned long flags; | |
da0c1e65 | 4943 | bool queued, running; |
0ec8aa00 PZ |
4944 | |
4945 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 4946 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
4947 | running = task_current(rq, p); |
4948 | ||
da0c1e65 | 4949 | if (queued) |
0ec8aa00 PZ |
4950 | dequeue_task(rq, p, 0); |
4951 | if (running) | |
f3cd1c4e | 4952 | put_prev_task(rq, p); |
0ec8aa00 PZ |
4953 | |
4954 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4955 | |
4956 | if (running) | |
4957 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4958 | if (queued) |
0ec8aa00 PZ |
4959 | enqueue_task(rq, p, 0); |
4960 | task_rq_unlock(rq, p, &flags); | |
4961 | } | |
e6628d5b MG |
4962 | #endif |
4963 | ||
1da177e4 | 4964 | /* |
969c7921 TH |
4965 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4966 | * and performs thread migration by bumping thread off CPU then | |
4967 | * 'pushing' onto another runqueue. | |
1da177e4 | 4968 | */ |
969c7921 | 4969 | static int migration_cpu_stop(void *data) |
1da177e4 | 4970 | { |
969c7921 | 4971 | struct migration_arg *arg = data; |
f7b4cddc | 4972 | |
969c7921 TH |
4973 | /* |
4974 | * The original target cpu might have gone down and we might | |
4975 | * be on another cpu but it doesn't matter. | |
4976 | */ | |
f7b4cddc | 4977 | local_irq_disable(); |
5cd038f5 LJ |
4978 | /* |
4979 | * We need to explicitly wake pending tasks before running | |
4980 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
4981 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
4982 | */ | |
4983 | sched_ttwu_pending(); | |
969c7921 | 4984 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4985 | local_irq_enable(); |
1da177e4 | 4986 | return 0; |
f7b4cddc ON |
4987 | } |
4988 | ||
1da177e4 | 4989 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4990 | |
054b9108 | 4991 | /* |
48c5ccae PZ |
4992 | * Ensures that the idle task is using init_mm right before its cpu goes |
4993 | * offline. | |
054b9108 | 4994 | */ |
48c5ccae | 4995 | void idle_task_exit(void) |
1da177e4 | 4996 | { |
48c5ccae | 4997 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4998 | |
48c5ccae | 4999 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5000 | |
a53efe5f | 5001 | if (mm != &init_mm) { |
48c5ccae | 5002 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
5003 | finish_arch_post_lock_switch(); |
5004 | } | |
48c5ccae | 5005 | mmdrop(mm); |
1da177e4 LT |
5006 | } |
5007 | ||
5008 | /* | |
5d180232 PZ |
5009 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5010 | * we might have. Assumes we're called after migrate_tasks() so that the | |
5011 | * nr_active count is stable. | |
5012 | * | |
5013 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5014 | */ |
5d180232 | 5015 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5016 | { |
5d180232 PZ |
5017 | long delta = calc_load_fold_active(rq); |
5018 | if (delta) | |
5019 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5020 | } |
5021 | ||
3f1d2a31 PZ |
5022 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5023 | { | |
5024 | } | |
5025 | ||
5026 | static const struct sched_class fake_sched_class = { | |
5027 | .put_prev_task = put_prev_task_fake, | |
5028 | }; | |
5029 | ||
5030 | static struct task_struct fake_task = { | |
5031 | /* | |
5032 | * Avoid pull_{rt,dl}_task() | |
5033 | */ | |
5034 | .prio = MAX_PRIO + 1, | |
5035 | .sched_class = &fake_sched_class, | |
5036 | }; | |
5037 | ||
48f24c4d | 5038 | /* |
48c5ccae PZ |
5039 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5040 | * try_to_wake_up()->select_task_rq(). | |
5041 | * | |
5042 | * Called with rq->lock held even though we'er in stop_machine() and | |
5043 | * there's no concurrency possible, we hold the required locks anyway | |
5044 | * because of lock validation efforts. | |
1da177e4 | 5045 | */ |
48c5ccae | 5046 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5047 | { |
70b97a7f | 5048 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5049 | struct task_struct *next, *stop = rq->stop; |
5050 | int dest_cpu; | |
1da177e4 LT |
5051 | |
5052 | /* | |
48c5ccae PZ |
5053 | * Fudge the rq selection such that the below task selection loop |
5054 | * doesn't get stuck on the currently eligible stop task. | |
5055 | * | |
5056 | * We're currently inside stop_machine() and the rq is either stuck | |
5057 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5058 | * either way we should never end up calling schedule() until we're | |
5059 | * done here. | |
1da177e4 | 5060 | */ |
48c5ccae | 5061 | rq->stop = NULL; |
48f24c4d | 5062 | |
77bd3970 FW |
5063 | /* |
5064 | * put_prev_task() and pick_next_task() sched | |
5065 | * class method both need to have an up-to-date | |
5066 | * value of rq->clock[_task] | |
5067 | */ | |
5068 | update_rq_clock(rq); | |
5069 | ||
dd41f596 | 5070 | for ( ; ; ) { |
48c5ccae PZ |
5071 | /* |
5072 | * There's this thread running, bail when that's the only | |
5073 | * remaining thread. | |
5074 | */ | |
5075 | if (rq->nr_running == 1) | |
dd41f596 | 5076 | break; |
48c5ccae | 5077 | |
3f1d2a31 | 5078 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 5079 | BUG_ON(!next); |
79c53799 | 5080 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5081 | |
48c5ccae PZ |
5082 | /* Find suitable destination for @next, with force if needed. */ |
5083 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5084 | raw_spin_unlock(&rq->lock); | |
5085 | ||
5086 | __migrate_task(next, dead_cpu, dest_cpu); | |
5087 | ||
5088 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5089 | } |
dce48a84 | 5090 | |
48c5ccae | 5091 | rq->stop = stop; |
dce48a84 | 5092 | } |
48c5ccae | 5093 | |
1da177e4 LT |
5094 | #endif /* CONFIG_HOTPLUG_CPU */ |
5095 | ||
e692ab53 NP |
5096 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5097 | ||
5098 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5099 | { |
5100 | .procname = "sched_domain", | |
c57baf1e | 5101 | .mode = 0555, |
e0361851 | 5102 | }, |
56992309 | 5103 | {} |
e692ab53 NP |
5104 | }; |
5105 | ||
5106 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5107 | { |
5108 | .procname = "kernel", | |
c57baf1e | 5109 | .mode = 0555, |
e0361851 AD |
5110 | .child = sd_ctl_dir, |
5111 | }, | |
56992309 | 5112 | {} |
e692ab53 NP |
5113 | }; |
5114 | ||
5115 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5116 | { | |
5117 | struct ctl_table *entry = | |
5cf9f062 | 5118 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5119 | |
e692ab53 NP |
5120 | return entry; |
5121 | } | |
5122 | ||
6382bc90 MM |
5123 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5124 | { | |
cd790076 | 5125 | struct ctl_table *entry; |
6382bc90 | 5126 | |
cd790076 MM |
5127 | /* |
5128 | * In the intermediate directories, both the child directory and | |
5129 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5130 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5131 | * static strings and all have proc handlers. |
5132 | */ | |
5133 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5134 | if (entry->child) |
5135 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5136 | if (entry->proc_handler == NULL) |
5137 | kfree(entry->procname); | |
5138 | } | |
6382bc90 MM |
5139 | |
5140 | kfree(*tablep); | |
5141 | *tablep = NULL; | |
5142 | } | |
5143 | ||
201c373e | 5144 | static int min_load_idx = 0; |
fd9b86d3 | 5145 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5146 | |
e692ab53 | 5147 | static void |
e0361851 | 5148 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5149 | const char *procname, void *data, int maxlen, |
201c373e NK |
5150 | umode_t mode, proc_handler *proc_handler, |
5151 | bool load_idx) | |
e692ab53 | 5152 | { |
e692ab53 NP |
5153 | entry->procname = procname; |
5154 | entry->data = data; | |
5155 | entry->maxlen = maxlen; | |
5156 | entry->mode = mode; | |
5157 | entry->proc_handler = proc_handler; | |
201c373e NK |
5158 | |
5159 | if (load_idx) { | |
5160 | entry->extra1 = &min_load_idx; | |
5161 | entry->extra2 = &max_load_idx; | |
5162 | } | |
e692ab53 NP |
5163 | } |
5164 | ||
5165 | static struct ctl_table * | |
5166 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5167 | { | |
37e6bae8 | 5168 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 5169 | |
ad1cdc1d MM |
5170 | if (table == NULL) |
5171 | return NULL; | |
5172 | ||
e0361851 | 5173 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5174 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5175 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5176 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5177 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5178 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5179 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5180 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5181 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5182 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5183 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5184 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5185 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5186 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5187 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5188 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5189 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5190 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5191 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5192 | &sd->cache_nice_tries, |
201c373e | 5193 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5194 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5195 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
5196 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5197 | &sd->max_newidle_lb_cost, | |
5198 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5199 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5200 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5201 | /* &table[13] is terminator */ |
e692ab53 NP |
5202 | |
5203 | return table; | |
5204 | } | |
5205 | ||
be7002e6 | 5206 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5207 | { |
5208 | struct ctl_table *entry, *table; | |
5209 | struct sched_domain *sd; | |
5210 | int domain_num = 0, i; | |
5211 | char buf[32]; | |
5212 | ||
5213 | for_each_domain(cpu, sd) | |
5214 | domain_num++; | |
5215 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5216 | if (table == NULL) |
5217 | return NULL; | |
e692ab53 NP |
5218 | |
5219 | i = 0; | |
5220 | for_each_domain(cpu, sd) { | |
5221 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5222 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5223 | entry->mode = 0555; |
e692ab53 NP |
5224 | entry->child = sd_alloc_ctl_domain_table(sd); |
5225 | entry++; | |
5226 | i++; | |
5227 | } | |
5228 | return table; | |
5229 | } | |
5230 | ||
5231 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5232 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5233 | { |
6ad4c188 | 5234 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5235 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5236 | char buf[32]; | |
5237 | ||
7378547f MM |
5238 | WARN_ON(sd_ctl_dir[0].child); |
5239 | sd_ctl_dir[0].child = entry; | |
5240 | ||
ad1cdc1d MM |
5241 | if (entry == NULL) |
5242 | return; | |
5243 | ||
6ad4c188 | 5244 | for_each_possible_cpu(i) { |
e692ab53 | 5245 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5246 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5247 | entry->mode = 0555; |
e692ab53 | 5248 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5249 | entry++; |
e692ab53 | 5250 | } |
7378547f MM |
5251 | |
5252 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5253 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5254 | } | |
6382bc90 | 5255 | |
7378547f | 5256 | /* may be called multiple times per register */ |
6382bc90 MM |
5257 | static void unregister_sched_domain_sysctl(void) |
5258 | { | |
7378547f MM |
5259 | if (sd_sysctl_header) |
5260 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5261 | sd_sysctl_header = NULL; |
7378547f MM |
5262 | if (sd_ctl_dir[0].child) |
5263 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5264 | } |
e692ab53 | 5265 | #else |
6382bc90 MM |
5266 | static void register_sched_domain_sysctl(void) |
5267 | { | |
5268 | } | |
5269 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5270 | { |
5271 | } | |
5272 | #endif | |
5273 | ||
1f11eb6a GH |
5274 | static void set_rq_online(struct rq *rq) |
5275 | { | |
5276 | if (!rq->online) { | |
5277 | const struct sched_class *class; | |
5278 | ||
c6c4927b | 5279 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5280 | rq->online = 1; |
5281 | ||
5282 | for_each_class(class) { | |
5283 | if (class->rq_online) | |
5284 | class->rq_online(rq); | |
5285 | } | |
5286 | } | |
5287 | } | |
5288 | ||
5289 | static void set_rq_offline(struct rq *rq) | |
5290 | { | |
5291 | if (rq->online) { | |
5292 | const struct sched_class *class; | |
5293 | ||
5294 | for_each_class(class) { | |
5295 | if (class->rq_offline) | |
5296 | class->rq_offline(rq); | |
5297 | } | |
5298 | ||
c6c4927b | 5299 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5300 | rq->online = 0; |
5301 | } | |
5302 | } | |
5303 | ||
1da177e4 LT |
5304 | /* |
5305 | * migration_call - callback that gets triggered when a CPU is added. | |
5306 | * Here we can start up the necessary migration thread for the new CPU. | |
5307 | */ | |
0db0628d | 5308 | static int |
48f24c4d | 5309 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5310 | { |
48f24c4d | 5311 | int cpu = (long)hcpu; |
1da177e4 | 5312 | unsigned long flags; |
969c7921 | 5313 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5314 | |
48c5ccae | 5315 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5316 | |
1da177e4 | 5317 | case CPU_UP_PREPARE: |
a468d389 | 5318 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5319 | break; |
48f24c4d | 5320 | |
1da177e4 | 5321 | case CPU_ONLINE: |
1f94ef59 | 5322 | /* Update our root-domain */ |
05fa785c | 5323 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5324 | if (rq->rd) { |
c6c4927b | 5325 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5326 | |
5327 | set_rq_online(rq); | |
1f94ef59 | 5328 | } |
05fa785c | 5329 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5330 | break; |
48f24c4d | 5331 | |
1da177e4 | 5332 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5333 | case CPU_DYING: |
317f3941 | 5334 | sched_ttwu_pending(); |
57d885fe | 5335 | /* Update our root-domain */ |
05fa785c | 5336 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5337 | if (rq->rd) { |
c6c4927b | 5338 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5339 | set_rq_offline(rq); |
57d885fe | 5340 | } |
48c5ccae PZ |
5341 | migrate_tasks(cpu); |
5342 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5343 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5344 | break; |
48c5ccae | 5345 | |
5d180232 | 5346 | case CPU_DEAD: |
f319da0c | 5347 | calc_load_migrate(rq); |
57d885fe | 5348 | break; |
1da177e4 LT |
5349 | #endif |
5350 | } | |
49c022e6 PZ |
5351 | |
5352 | update_max_interval(); | |
5353 | ||
1da177e4 LT |
5354 | return NOTIFY_OK; |
5355 | } | |
5356 | ||
f38b0820 PM |
5357 | /* |
5358 | * Register at high priority so that task migration (migrate_all_tasks) | |
5359 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5360 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5361 | */ |
0db0628d | 5362 | static struct notifier_block migration_notifier = { |
1da177e4 | 5363 | .notifier_call = migration_call, |
50a323b7 | 5364 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5365 | }; |
5366 | ||
6a82b60d | 5367 | static void set_cpu_rq_start_time(void) |
a803f026 CM |
5368 | { |
5369 | int cpu = smp_processor_id(); | |
5370 | struct rq *rq = cpu_rq(cpu); | |
5371 | rq->age_stamp = sched_clock_cpu(cpu); | |
5372 | } | |
5373 | ||
0db0628d | 5374 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5375 | unsigned long action, void *hcpu) |
5376 | { | |
5377 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5378 | case CPU_STARTING: |
5379 | set_cpu_rq_start_time(); | |
5380 | return NOTIFY_OK; | |
3a101d05 TH |
5381 | case CPU_DOWN_FAILED: |
5382 | set_cpu_active((long)hcpu, true); | |
5383 | return NOTIFY_OK; | |
5384 | default: | |
5385 | return NOTIFY_DONE; | |
5386 | } | |
5387 | } | |
5388 | ||
0db0628d | 5389 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5390 | unsigned long action, void *hcpu) |
5391 | { | |
5392 | switch (action & ~CPU_TASKS_FROZEN) { | |
5393 | case CPU_DOWN_PREPARE: | |
3c18d447 | 5394 | set_cpu_active((long)hcpu, false); |
3a101d05 | 5395 | return NOTIFY_OK; |
3c18d447 JL |
5396 | default: |
5397 | return NOTIFY_DONE; | |
3a101d05 TH |
5398 | } |
5399 | } | |
5400 | ||
7babe8db | 5401 | static int __init migration_init(void) |
1da177e4 LT |
5402 | { |
5403 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5404 | int err; |
48f24c4d | 5405 | |
3a101d05 | 5406 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5407 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5408 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5409 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5410 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5411 | |
3a101d05 TH |
5412 | /* Register cpu active notifiers */ |
5413 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5414 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5415 | ||
a004cd42 | 5416 | return 0; |
1da177e4 | 5417 | } |
7babe8db | 5418 | early_initcall(migration_init); |
1da177e4 LT |
5419 | #endif |
5420 | ||
5421 | #ifdef CONFIG_SMP | |
476f3534 | 5422 | |
4cb98839 PZ |
5423 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5424 | ||
3e9830dc | 5425 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5426 | |
d039ac60 | 5427 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5428 | |
d039ac60 | 5429 | static int __init sched_debug_setup(char *str) |
f6630114 | 5430 | { |
d039ac60 | 5431 | sched_debug_enabled = 1; |
f6630114 MT |
5432 | |
5433 | return 0; | |
5434 | } | |
d039ac60 PZ |
5435 | early_param("sched_debug", sched_debug_setup); |
5436 | ||
5437 | static inline bool sched_debug(void) | |
5438 | { | |
5439 | return sched_debug_enabled; | |
5440 | } | |
f6630114 | 5441 | |
7c16ec58 | 5442 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5443 | struct cpumask *groupmask) |
1da177e4 | 5444 | { |
4dcf6aff | 5445 | struct sched_group *group = sd->groups; |
1da177e4 | 5446 | |
96f874e2 | 5447 | cpumask_clear(groupmask); |
4dcf6aff IM |
5448 | |
5449 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5450 | ||
5451 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5452 | printk("does not load-balance\n"); |
4dcf6aff | 5453 | if (sd->parent) |
3df0fc5b PZ |
5454 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5455 | " has parent"); | |
4dcf6aff | 5456 | return -1; |
41c7ce9a NP |
5457 | } |
5458 | ||
333470ee TH |
5459 | printk(KERN_CONT "span %*pbl level %s\n", |
5460 | cpumask_pr_args(sched_domain_span(sd)), sd->name); | |
4dcf6aff | 5461 | |
758b2cdc | 5462 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5463 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5464 | "CPU%d\n", cpu); | |
4dcf6aff | 5465 | } |
758b2cdc | 5466 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5467 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5468 | " CPU%d\n", cpu); | |
4dcf6aff | 5469 | } |
1da177e4 | 5470 | |
4dcf6aff | 5471 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5472 | do { |
4dcf6aff | 5473 | if (!group) { |
3df0fc5b PZ |
5474 | printk("\n"); |
5475 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5476 | break; |
5477 | } | |
5478 | ||
758b2cdc | 5479 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5480 | printk(KERN_CONT "\n"); |
5481 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5482 | break; |
5483 | } | |
1da177e4 | 5484 | |
cb83b629 PZ |
5485 | if (!(sd->flags & SD_OVERLAP) && |
5486 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5487 | printk(KERN_CONT "\n"); |
5488 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5489 | break; |
5490 | } | |
1da177e4 | 5491 | |
758b2cdc | 5492 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5493 | |
333470ee TH |
5494 | printk(KERN_CONT " %*pbl", |
5495 | cpumask_pr_args(sched_group_cpus(group))); | |
ca8ce3d0 | 5496 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5497 | printk(KERN_CONT " (cpu_capacity = %d)", |
5498 | group->sgc->capacity); | |
381512cf | 5499 | } |
1da177e4 | 5500 | |
4dcf6aff IM |
5501 | group = group->next; |
5502 | } while (group != sd->groups); | |
3df0fc5b | 5503 | printk(KERN_CONT "\n"); |
1da177e4 | 5504 | |
758b2cdc | 5505 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5506 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5507 | |
758b2cdc RR |
5508 | if (sd->parent && |
5509 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5510 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5511 | "of domain->span\n"); | |
4dcf6aff IM |
5512 | return 0; |
5513 | } | |
1da177e4 | 5514 | |
4dcf6aff IM |
5515 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5516 | { | |
5517 | int level = 0; | |
1da177e4 | 5518 | |
d039ac60 | 5519 | if (!sched_debug_enabled) |
f6630114 MT |
5520 | return; |
5521 | ||
4dcf6aff IM |
5522 | if (!sd) { |
5523 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5524 | return; | |
5525 | } | |
1da177e4 | 5526 | |
4dcf6aff IM |
5527 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5528 | ||
5529 | for (;;) { | |
4cb98839 | 5530 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5531 | break; |
1da177e4 LT |
5532 | level++; |
5533 | sd = sd->parent; | |
33859f7f | 5534 | if (!sd) |
4dcf6aff IM |
5535 | break; |
5536 | } | |
1da177e4 | 5537 | } |
6d6bc0ad | 5538 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5539 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5540 | static inline bool sched_debug(void) |
5541 | { | |
5542 | return false; | |
5543 | } | |
6d6bc0ad | 5544 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5545 | |
1a20ff27 | 5546 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5547 | { |
758b2cdc | 5548 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5549 | return 1; |
5550 | ||
5551 | /* Following flags need at least 2 groups */ | |
5552 | if (sd->flags & (SD_LOAD_BALANCE | | |
5553 | SD_BALANCE_NEWIDLE | | |
5554 | SD_BALANCE_FORK | | |
89c4710e | 5555 | SD_BALANCE_EXEC | |
5d4dfddd | 5556 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5557 | SD_SHARE_PKG_RESOURCES | |
5558 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5559 | if (sd->groups != sd->groups->next) |
5560 | return 0; | |
5561 | } | |
5562 | ||
5563 | /* Following flags don't use groups */ | |
c88d5910 | 5564 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5565 | return 0; |
5566 | ||
5567 | return 1; | |
5568 | } | |
5569 | ||
48f24c4d IM |
5570 | static int |
5571 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5572 | { |
5573 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5574 | ||
5575 | if (sd_degenerate(parent)) | |
5576 | return 1; | |
5577 | ||
758b2cdc | 5578 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5579 | return 0; |
5580 | ||
245af2c7 SS |
5581 | /* Flags needing groups don't count if only 1 group in parent */ |
5582 | if (parent->groups == parent->groups->next) { | |
5583 | pflags &= ~(SD_LOAD_BALANCE | | |
5584 | SD_BALANCE_NEWIDLE | | |
5585 | SD_BALANCE_FORK | | |
89c4710e | 5586 | SD_BALANCE_EXEC | |
5d4dfddd | 5587 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5588 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5589 | SD_PREFER_SIBLING | |
5590 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5591 | if (nr_node_ids == 1) |
5592 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5593 | } |
5594 | if (~cflags & pflags) | |
5595 | return 0; | |
5596 | ||
5597 | return 1; | |
5598 | } | |
5599 | ||
dce840a0 | 5600 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5601 | { |
dce840a0 | 5602 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5603 | |
68e74568 | 5604 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5605 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5606 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5607 | free_cpumask_var(rd->rto_mask); |
5608 | free_cpumask_var(rd->online); | |
5609 | free_cpumask_var(rd->span); | |
5610 | kfree(rd); | |
5611 | } | |
5612 | ||
57d885fe GH |
5613 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5614 | { | |
a0490fa3 | 5615 | struct root_domain *old_rd = NULL; |
57d885fe | 5616 | unsigned long flags; |
57d885fe | 5617 | |
05fa785c | 5618 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5619 | |
5620 | if (rq->rd) { | |
a0490fa3 | 5621 | old_rd = rq->rd; |
57d885fe | 5622 | |
c6c4927b | 5623 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5624 | set_rq_offline(rq); |
57d885fe | 5625 | |
c6c4927b | 5626 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5627 | |
a0490fa3 | 5628 | /* |
0515973f | 5629 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5630 | * set old_rd to NULL to skip the freeing later |
5631 | * in this function: | |
5632 | */ | |
5633 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5634 | old_rd = NULL; | |
57d885fe GH |
5635 | } |
5636 | ||
5637 | atomic_inc(&rd->refcount); | |
5638 | rq->rd = rd; | |
5639 | ||
c6c4927b | 5640 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5641 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5642 | set_rq_online(rq); |
57d885fe | 5643 | |
05fa785c | 5644 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5645 | |
5646 | if (old_rd) | |
dce840a0 | 5647 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5648 | } |
5649 | ||
68c38fc3 | 5650 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5651 | { |
5652 | memset(rd, 0, sizeof(*rd)); | |
5653 | ||
68c38fc3 | 5654 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5655 | goto out; |
68c38fc3 | 5656 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5657 | goto free_span; |
1baca4ce | 5658 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5659 | goto free_online; |
1baca4ce JL |
5660 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5661 | goto free_dlo_mask; | |
6e0534f2 | 5662 | |
332ac17e | 5663 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5664 | if (cpudl_init(&rd->cpudl) != 0) |
5665 | goto free_dlo_mask; | |
332ac17e | 5666 | |
68c38fc3 | 5667 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5668 | goto free_rto_mask; |
c6c4927b | 5669 | return 0; |
6e0534f2 | 5670 | |
68e74568 RR |
5671 | free_rto_mask: |
5672 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5673 | free_dlo_mask: |
5674 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5675 | free_online: |
5676 | free_cpumask_var(rd->online); | |
5677 | free_span: | |
5678 | free_cpumask_var(rd->span); | |
0c910d28 | 5679 | out: |
c6c4927b | 5680 | return -ENOMEM; |
57d885fe GH |
5681 | } |
5682 | ||
029632fb PZ |
5683 | /* |
5684 | * By default the system creates a single root-domain with all cpus as | |
5685 | * members (mimicking the global state we have today). | |
5686 | */ | |
5687 | struct root_domain def_root_domain; | |
5688 | ||
57d885fe GH |
5689 | static void init_defrootdomain(void) |
5690 | { | |
68c38fc3 | 5691 | init_rootdomain(&def_root_domain); |
c6c4927b | 5692 | |
57d885fe GH |
5693 | atomic_set(&def_root_domain.refcount, 1); |
5694 | } | |
5695 | ||
dc938520 | 5696 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5697 | { |
5698 | struct root_domain *rd; | |
5699 | ||
5700 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5701 | if (!rd) | |
5702 | return NULL; | |
5703 | ||
68c38fc3 | 5704 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5705 | kfree(rd); |
5706 | return NULL; | |
5707 | } | |
57d885fe GH |
5708 | |
5709 | return rd; | |
5710 | } | |
5711 | ||
63b2ca30 | 5712 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5713 | { |
5714 | struct sched_group *tmp, *first; | |
5715 | ||
5716 | if (!sg) | |
5717 | return; | |
5718 | ||
5719 | first = sg; | |
5720 | do { | |
5721 | tmp = sg->next; | |
5722 | ||
63b2ca30 NP |
5723 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5724 | kfree(sg->sgc); | |
e3589f6c PZ |
5725 | |
5726 | kfree(sg); | |
5727 | sg = tmp; | |
5728 | } while (sg != first); | |
5729 | } | |
5730 | ||
dce840a0 PZ |
5731 | static void free_sched_domain(struct rcu_head *rcu) |
5732 | { | |
5733 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5734 | |
5735 | /* | |
5736 | * If its an overlapping domain it has private groups, iterate and | |
5737 | * nuke them all. | |
5738 | */ | |
5739 | if (sd->flags & SD_OVERLAP) { | |
5740 | free_sched_groups(sd->groups, 1); | |
5741 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5742 | kfree(sd->groups->sgc); |
dce840a0 | 5743 | kfree(sd->groups); |
9c3f75cb | 5744 | } |
dce840a0 PZ |
5745 | kfree(sd); |
5746 | } | |
5747 | ||
5748 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5749 | { | |
5750 | call_rcu(&sd->rcu, free_sched_domain); | |
5751 | } | |
5752 | ||
5753 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5754 | { | |
5755 | for (; sd; sd = sd->parent) | |
5756 | destroy_sched_domain(sd, cpu); | |
5757 | } | |
5758 | ||
518cd623 PZ |
5759 | /* |
5760 | * Keep a special pointer to the highest sched_domain that has | |
5761 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5762 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5763 | * | |
5764 | * Also keep a unique ID per domain (we use the first cpu number in | |
5765 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5766 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5767 | */ |
5768 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5769 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5770 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5771 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5772 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5773 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5774 | |
5775 | static void update_top_cache_domain(int cpu) | |
5776 | { | |
5777 | struct sched_domain *sd; | |
5d4cf996 | 5778 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5779 | int id = cpu; |
7d9ffa89 | 5780 | int size = 1; |
518cd623 PZ |
5781 | |
5782 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5783 | if (sd) { |
518cd623 | 5784 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5785 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5786 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5787 | } |
5d4cf996 | 5788 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5789 | |
5790 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5791 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5792 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5793 | |
5794 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5795 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5796 | |
5797 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5798 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5799 | } |
5800 | ||
1da177e4 | 5801 | /* |
0eab9146 | 5802 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5803 | * hold the hotplug lock. |
5804 | */ | |
0eab9146 IM |
5805 | static void |
5806 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5807 | { |
70b97a7f | 5808 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5809 | struct sched_domain *tmp; |
5810 | ||
5811 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5812 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5813 | struct sched_domain *parent = tmp->parent; |
5814 | if (!parent) | |
5815 | break; | |
f29c9b1c | 5816 | |
1a848870 | 5817 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5818 | tmp->parent = parent->parent; |
1a848870 SS |
5819 | if (parent->parent) |
5820 | parent->parent->child = tmp; | |
10866e62 PZ |
5821 | /* |
5822 | * Transfer SD_PREFER_SIBLING down in case of a | |
5823 | * degenerate parent; the spans match for this | |
5824 | * so the property transfers. | |
5825 | */ | |
5826 | if (parent->flags & SD_PREFER_SIBLING) | |
5827 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5828 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5829 | } else |
5830 | tmp = tmp->parent; | |
245af2c7 SS |
5831 | } |
5832 | ||
1a848870 | 5833 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5834 | tmp = sd; |
245af2c7 | 5835 | sd = sd->parent; |
dce840a0 | 5836 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5837 | if (sd) |
5838 | sd->child = NULL; | |
5839 | } | |
1da177e4 | 5840 | |
4cb98839 | 5841 | sched_domain_debug(sd, cpu); |
1da177e4 | 5842 | |
57d885fe | 5843 | rq_attach_root(rq, rd); |
dce840a0 | 5844 | tmp = rq->sd; |
674311d5 | 5845 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5846 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5847 | |
5848 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5849 | } |
5850 | ||
1da177e4 LT |
5851 | /* Setup the mask of cpus configured for isolated domains */ |
5852 | static int __init isolated_cpu_setup(char *str) | |
5853 | { | |
bdddd296 | 5854 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5855 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5856 | return 1; |
5857 | } | |
5858 | ||
8927f494 | 5859 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5860 | |
49a02c51 | 5861 | struct s_data { |
21d42ccf | 5862 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5863 | struct root_domain *rd; |
5864 | }; | |
5865 | ||
2109b99e | 5866 | enum s_alloc { |
2109b99e | 5867 | sa_rootdomain, |
21d42ccf | 5868 | sa_sd, |
dce840a0 | 5869 | sa_sd_storage, |
2109b99e AH |
5870 | sa_none, |
5871 | }; | |
5872 | ||
c1174876 PZ |
5873 | /* |
5874 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5875 | * domain traversal. | |
5876 | * | |
5877 | * Asymmetric node setups can result in situations where the domain tree is of | |
5878 | * unequal depth, make sure to skip domains that already cover the entire | |
5879 | * range. | |
5880 | * | |
5881 | * In that case build_sched_domains() will have terminated the iteration early | |
5882 | * and our sibling sd spans will be empty. Domains should always include the | |
5883 | * cpu they're built on, so check that. | |
5884 | * | |
5885 | */ | |
5886 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5887 | { | |
5888 | const struct cpumask *span = sched_domain_span(sd); | |
5889 | struct sd_data *sdd = sd->private; | |
5890 | struct sched_domain *sibling; | |
5891 | int i; | |
5892 | ||
5893 | for_each_cpu(i, span) { | |
5894 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5895 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5896 | continue; | |
5897 | ||
5898 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5899 | } | |
5900 | } | |
5901 | ||
5902 | /* | |
5903 | * Return the canonical balance cpu for this group, this is the first cpu | |
5904 | * of this group that's also in the iteration mask. | |
5905 | */ | |
5906 | int group_balance_cpu(struct sched_group *sg) | |
5907 | { | |
5908 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5909 | } | |
5910 | ||
e3589f6c PZ |
5911 | static int |
5912 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5913 | { | |
5914 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5915 | const struct cpumask *span = sched_domain_span(sd); | |
5916 | struct cpumask *covered = sched_domains_tmpmask; | |
5917 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5918 | struct sched_domain *sibling; |
e3589f6c PZ |
5919 | int i; |
5920 | ||
5921 | cpumask_clear(covered); | |
5922 | ||
5923 | for_each_cpu(i, span) { | |
5924 | struct cpumask *sg_span; | |
5925 | ||
5926 | if (cpumask_test_cpu(i, covered)) | |
5927 | continue; | |
5928 | ||
aaecac4a | 5929 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5930 | |
5931 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5932 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5933 | continue; |
5934 | ||
e3589f6c | 5935 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5936 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5937 | |
5938 | if (!sg) | |
5939 | goto fail; | |
5940 | ||
5941 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5942 | if (sibling->child) |
5943 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5944 | else | |
e3589f6c PZ |
5945 | cpumask_set_cpu(i, sg_span); |
5946 | ||
5947 | cpumask_or(covered, covered, sg_span); | |
5948 | ||
63b2ca30 NP |
5949 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5950 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5951 | build_group_mask(sd, sg); |
5952 | ||
c3decf0d | 5953 | /* |
63b2ca30 | 5954 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5955 | * domains and no possible iteration will get us here, we won't |
5956 | * die on a /0 trap. | |
5957 | */ | |
ca8ce3d0 | 5958 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
e3589f6c | 5959 | |
c1174876 PZ |
5960 | /* |
5961 | * Make sure the first group of this domain contains the | |
5962 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5963 | * breaks. See update_sg_lb_stats(). | |
5964 | */ | |
74a5ce20 | 5965 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5966 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5967 | groups = sg; |
5968 | ||
5969 | if (!first) | |
5970 | first = sg; | |
5971 | if (last) | |
5972 | last->next = sg; | |
5973 | last = sg; | |
5974 | last->next = first; | |
5975 | } | |
5976 | sd->groups = groups; | |
5977 | ||
5978 | return 0; | |
5979 | ||
5980 | fail: | |
5981 | free_sched_groups(first, 0); | |
5982 | ||
5983 | return -ENOMEM; | |
5984 | } | |
5985 | ||
dce840a0 | 5986 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5987 | { |
dce840a0 PZ |
5988 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5989 | struct sched_domain *child = sd->child; | |
1da177e4 | 5990 | |
dce840a0 PZ |
5991 | if (child) |
5992 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5993 | |
9c3f75cb | 5994 | if (sg) { |
dce840a0 | 5995 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5996 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5997 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5998 | } |
dce840a0 PZ |
5999 | |
6000 | return cpu; | |
1e9f28fa | 6001 | } |
1e9f28fa | 6002 | |
01a08546 | 6003 | /* |
dce840a0 PZ |
6004 | * build_sched_groups will build a circular linked list of the groups |
6005 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 6006 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
6007 | * |
6008 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6009 | */ |
e3589f6c PZ |
6010 | static int |
6011 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6012 | { |
dce840a0 PZ |
6013 | struct sched_group *first = NULL, *last = NULL; |
6014 | struct sd_data *sdd = sd->private; | |
6015 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6016 | struct cpumask *covered; |
dce840a0 | 6017 | int i; |
9c1cfda2 | 6018 | |
e3589f6c PZ |
6019 | get_group(cpu, sdd, &sd->groups); |
6020 | atomic_inc(&sd->groups->ref); | |
6021 | ||
0936629f | 6022 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
6023 | return 0; |
6024 | ||
f96225fd PZ |
6025 | lockdep_assert_held(&sched_domains_mutex); |
6026 | covered = sched_domains_tmpmask; | |
6027 | ||
dce840a0 | 6028 | cpumask_clear(covered); |
6711cab4 | 6029 | |
dce840a0 PZ |
6030 | for_each_cpu(i, span) { |
6031 | struct sched_group *sg; | |
cd08e923 | 6032 | int group, j; |
6711cab4 | 6033 | |
dce840a0 PZ |
6034 | if (cpumask_test_cpu(i, covered)) |
6035 | continue; | |
6711cab4 | 6036 | |
cd08e923 | 6037 | group = get_group(i, sdd, &sg); |
c1174876 | 6038 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6039 | |
dce840a0 PZ |
6040 | for_each_cpu(j, span) { |
6041 | if (get_group(j, sdd, NULL) != group) | |
6042 | continue; | |
0601a88d | 6043 | |
dce840a0 PZ |
6044 | cpumask_set_cpu(j, covered); |
6045 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6046 | } | |
0601a88d | 6047 | |
dce840a0 PZ |
6048 | if (!first) |
6049 | first = sg; | |
6050 | if (last) | |
6051 | last->next = sg; | |
6052 | last = sg; | |
6053 | } | |
6054 | last->next = first; | |
e3589f6c PZ |
6055 | |
6056 | return 0; | |
0601a88d | 6057 | } |
51888ca2 | 6058 | |
89c4710e | 6059 | /* |
63b2ca30 | 6060 | * Initialize sched groups cpu_capacity. |
89c4710e | 6061 | * |
63b2ca30 | 6062 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 6063 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
6064 | * Typically cpu_capacity for all the groups in a sched domain will be same |
6065 | * unless there are asymmetries in the topology. If there are asymmetries, | |
6066 | * group having more cpu_capacity will pickup more load compared to the | |
6067 | * group having less cpu_capacity. | |
89c4710e | 6068 | */ |
63b2ca30 | 6069 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 6070 | { |
e3589f6c | 6071 | struct sched_group *sg = sd->groups; |
89c4710e | 6072 | |
94c95ba6 | 6073 | WARN_ON(!sg); |
e3589f6c PZ |
6074 | |
6075 | do { | |
6076 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6077 | sg = sg->next; | |
6078 | } while (sg != sd->groups); | |
89c4710e | 6079 | |
c1174876 | 6080 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6081 | return; |
aae6d3dd | 6082 | |
63b2ca30 NP |
6083 | update_group_capacity(sd, cpu); |
6084 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
6085 | } |
6086 | ||
7c16ec58 MT |
6087 | /* |
6088 | * Initializers for schedule domains | |
6089 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6090 | */ | |
6091 | ||
1d3504fc | 6092 | static int default_relax_domain_level = -1; |
60495e77 | 6093 | int sched_domain_level_max; |
1d3504fc HS |
6094 | |
6095 | static int __init setup_relax_domain_level(char *str) | |
6096 | { | |
a841f8ce DS |
6097 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6098 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6099 | |
1d3504fc HS |
6100 | return 1; |
6101 | } | |
6102 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6103 | ||
6104 | static void set_domain_attribute(struct sched_domain *sd, | |
6105 | struct sched_domain_attr *attr) | |
6106 | { | |
6107 | int request; | |
6108 | ||
6109 | if (!attr || attr->relax_domain_level < 0) { | |
6110 | if (default_relax_domain_level < 0) | |
6111 | return; | |
6112 | else | |
6113 | request = default_relax_domain_level; | |
6114 | } else | |
6115 | request = attr->relax_domain_level; | |
6116 | if (request < sd->level) { | |
6117 | /* turn off idle balance on this domain */ | |
c88d5910 | 6118 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6119 | } else { |
6120 | /* turn on idle balance on this domain */ | |
c88d5910 | 6121 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6122 | } |
6123 | } | |
6124 | ||
54ab4ff4 PZ |
6125 | static void __sdt_free(const struct cpumask *cpu_map); |
6126 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6127 | ||
2109b99e AH |
6128 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6129 | const struct cpumask *cpu_map) | |
6130 | { | |
6131 | switch (what) { | |
2109b99e | 6132 | case sa_rootdomain: |
822ff793 PZ |
6133 | if (!atomic_read(&d->rd->refcount)) |
6134 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6135 | case sa_sd: |
6136 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6137 | case sa_sd_storage: |
54ab4ff4 | 6138 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6139 | case sa_none: |
6140 | break; | |
6141 | } | |
6142 | } | |
3404c8d9 | 6143 | |
2109b99e AH |
6144 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6145 | const struct cpumask *cpu_map) | |
6146 | { | |
dce840a0 PZ |
6147 | memset(d, 0, sizeof(*d)); |
6148 | ||
54ab4ff4 PZ |
6149 | if (__sdt_alloc(cpu_map)) |
6150 | return sa_sd_storage; | |
dce840a0 PZ |
6151 | d->sd = alloc_percpu(struct sched_domain *); |
6152 | if (!d->sd) | |
6153 | return sa_sd_storage; | |
2109b99e | 6154 | d->rd = alloc_rootdomain(); |
dce840a0 | 6155 | if (!d->rd) |
21d42ccf | 6156 | return sa_sd; |
2109b99e AH |
6157 | return sa_rootdomain; |
6158 | } | |
57d885fe | 6159 | |
dce840a0 PZ |
6160 | /* |
6161 | * NULL the sd_data elements we've used to build the sched_domain and | |
6162 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6163 | * will not free the data we're using. | |
6164 | */ | |
6165 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6166 | { | |
6167 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6168 | |
6169 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6170 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6171 | ||
e3589f6c | 6172 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6173 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6174 | |
63b2ca30 NP |
6175 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6176 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6177 | } |
6178 | ||
cb83b629 | 6179 | #ifdef CONFIG_NUMA |
cb83b629 | 6180 | static int sched_domains_numa_levels; |
e3fe70b1 | 6181 | enum numa_topology_type sched_numa_topology_type; |
cb83b629 | 6182 | static int *sched_domains_numa_distance; |
9942f79b | 6183 | int sched_max_numa_distance; |
cb83b629 PZ |
6184 | static struct cpumask ***sched_domains_numa_masks; |
6185 | static int sched_domains_curr_level; | |
143e1e28 | 6186 | #endif |
cb83b629 | 6187 | |
143e1e28 VG |
6188 | /* |
6189 | * SD_flags allowed in topology descriptions. | |
6190 | * | |
5d4dfddd | 6191 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6192 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6193 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6194 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6195 | * |
6196 | * Odd one out: | |
6197 | * SD_ASYM_PACKING - describes SMT quirks | |
6198 | */ | |
6199 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6200 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6201 | SD_SHARE_PKG_RESOURCES | \ |
6202 | SD_NUMA | \ | |
d77b3ed5 VG |
6203 | SD_ASYM_PACKING | \ |
6204 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6205 | |
6206 | static struct sched_domain * | |
143e1e28 | 6207 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6208 | { |
6209 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6210 | int sd_weight, sd_flags = 0; |
6211 | ||
6212 | #ifdef CONFIG_NUMA | |
6213 | /* | |
6214 | * Ugly hack to pass state to sd_numa_mask()... | |
6215 | */ | |
6216 | sched_domains_curr_level = tl->numa_level; | |
6217 | #endif | |
6218 | ||
6219 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6220 | ||
6221 | if (tl->sd_flags) | |
6222 | sd_flags = (*tl->sd_flags)(); | |
6223 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6224 | "wrong sd_flags in topology description\n")) | |
6225 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6226 | |
6227 | *sd = (struct sched_domain){ | |
6228 | .min_interval = sd_weight, | |
6229 | .max_interval = 2*sd_weight, | |
6230 | .busy_factor = 32, | |
870a0bb5 | 6231 | .imbalance_pct = 125, |
143e1e28 VG |
6232 | |
6233 | .cache_nice_tries = 0, | |
6234 | .busy_idx = 0, | |
6235 | .idle_idx = 0, | |
cb83b629 PZ |
6236 | .newidle_idx = 0, |
6237 | .wake_idx = 0, | |
6238 | .forkexec_idx = 0, | |
6239 | ||
6240 | .flags = 1*SD_LOAD_BALANCE | |
6241 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6242 | | 1*SD_BALANCE_EXEC |
6243 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6244 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6245 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6246 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6247 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6248 | | 0*SD_SERIALIZE |
cb83b629 | 6249 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6250 | | 0*SD_NUMA |
6251 | | sd_flags | |
cb83b629 | 6252 | , |
143e1e28 | 6253 | |
cb83b629 PZ |
6254 | .last_balance = jiffies, |
6255 | .balance_interval = sd_weight, | |
143e1e28 | 6256 | .smt_gain = 0, |
2b4cfe64 JL |
6257 | .max_newidle_lb_cost = 0, |
6258 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6259 | #ifdef CONFIG_SCHED_DEBUG |
6260 | .name = tl->name, | |
6261 | #endif | |
cb83b629 | 6262 | }; |
cb83b629 PZ |
6263 | |
6264 | /* | |
143e1e28 | 6265 | * Convert topological properties into behaviour. |
cb83b629 | 6266 | */ |
143e1e28 | 6267 | |
5d4dfddd | 6268 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
caff37ef | 6269 | sd->flags |= SD_PREFER_SIBLING; |
143e1e28 VG |
6270 | sd->imbalance_pct = 110; |
6271 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6272 | |
6273 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6274 | sd->imbalance_pct = 117; | |
6275 | sd->cache_nice_tries = 1; | |
6276 | sd->busy_idx = 2; | |
6277 | ||
6278 | #ifdef CONFIG_NUMA | |
6279 | } else if (sd->flags & SD_NUMA) { | |
6280 | sd->cache_nice_tries = 2; | |
6281 | sd->busy_idx = 3; | |
6282 | sd->idle_idx = 2; | |
6283 | ||
6284 | sd->flags |= SD_SERIALIZE; | |
6285 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6286 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6287 | SD_BALANCE_FORK | | |
6288 | SD_WAKE_AFFINE); | |
6289 | } | |
6290 | ||
6291 | #endif | |
6292 | } else { | |
6293 | sd->flags |= SD_PREFER_SIBLING; | |
6294 | sd->cache_nice_tries = 1; | |
6295 | sd->busy_idx = 2; | |
6296 | sd->idle_idx = 1; | |
6297 | } | |
6298 | ||
6299 | sd->private = &tl->data; | |
cb83b629 PZ |
6300 | |
6301 | return sd; | |
6302 | } | |
6303 | ||
143e1e28 VG |
6304 | /* |
6305 | * Topology list, bottom-up. | |
6306 | */ | |
6307 | static struct sched_domain_topology_level default_topology[] = { | |
6308 | #ifdef CONFIG_SCHED_SMT | |
6309 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6310 | #endif | |
6311 | #ifdef CONFIG_SCHED_MC | |
6312 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6313 | #endif |
6314 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6315 | { NULL, }, | |
6316 | }; | |
6317 | ||
6318 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6319 | ||
6320 | #define for_each_sd_topology(tl) \ | |
6321 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6322 | ||
6323 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6324 | { | |
6325 | sched_domain_topology = tl; | |
6326 | } | |
6327 | ||
6328 | #ifdef CONFIG_NUMA | |
6329 | ||
cb83b629 PZ |
6330 | static const struct cpumask *sd_numa_mask(int cpu) |
6331 | { | |
6332 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6333 | } | |
6334 | ||
d039ac60 PZ |
6335 | static void sched_numa_warn(const char *str) |
6336 | { | |
6337 | static int done = false; | |
6338 | int i,j; | |
6339 | ||
6340 | if (done) | |
6341 | return; | |
6342 | ||
6343 | done = true; | |
6344 | ||
6345 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6346 | ||
6347 | for (i = 0; i < nr_node_ids; i++) { | |
6348 | printk(KERN_WARNING " "); | |
6349 | for (j = 0; j < nr_node_ids; j++) | |
6350 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6351 | printk(KERN_CONT "\n"); | |
6352 | } | |
6353 | printk(KERN_WARNING "\n"); | |
6354 | } | |
6355 | ||
9942f79b | 6356 | bool find_numa_distance(int distance) |
d039ac60 PZ |
6357 | { |
6358 | int i; | |
6359 | ||
6360 | if (distance == node_distance(0, 0)) | |
6361 | return true; | |
6362 | ||
6363 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6364 | if (sched_domains_numa_distance[i] == distance) | |
6365 | return true; | |
6366 | } | |
6367 | ||
6368 | return false; | |
6369 | } | |
6370 | ||
e3fe70b1 RR |
6371 | /* |
6372 | * A system can have three types of NUMA topology: | |
6373 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system | |
6374 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes | |
6375 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane | |
6376 | * | |
6377 | * The difference between a glueless mesh topology and a backplane | |
6378 | * topology lies in whether communication between not directly | |
6379 | * connected nodes goes through intermediary nodes (where programs | |
6380 | * could run), or through backplane controllers. This affects | |
6381 | * placement of programs. | |
6382 | * | |
6383 | * The type of topology can be discerned with the following tests: | |
6384 | * - If the maximum distance between any nodes is 1 hop, the system | |
6385 | * is directly connected. | |
6386 | * - If for two nodes A and B, located N > 1 hops away from each other, | |
6387 | * there is an intermediary node C, which is < N hops away from both | |
6388 | * nodes A and B, the system is a glueless mesh. | |
6389 | */ | |
6390 | static void init_numa_topology_type(void) | |
6391 | { | |
6392 | int a, b, c, n; | |
6393 | ||
6394 | n = sched_max_numa_distance; | |
6395 | ||
6396 | if (n <= 1) | |
6397 | sched_numa_topology_type = NUMA_DIRECT; | |
6398 | ||
6399 | for_each_online_node(a) { | |
6400 | for_each_online_node(b) { | |
6401 | /* Find two nodes furthest removed from each other. */ | |
6402 | if (node_distance(a, b) < n) | |
6403 | continue; | |
6404 | ||
6405 | /* Is there an intermediary node between a and b? */ | |
6406 | for_each_online_node(c) { | |
6407 | if (node_distance(a, c) < n && | |
6408 | node_distance(b, c) < n) { | |
6409 | sched_numa_topology_type = | |
6410 | NUMA_GLUELESS_MESH; | |
6411 | return; | |
6412 | } | |
6413 | } | |
6414 | ||
6415 | sched_numa_topology_type = NUMA_BACKPLANE; | |
6416 | return; | |
6417 | } | |
6418 | } | |
6419 | } | |
6420 | ||
cb83b629 PZ |
6421 | static void sched_init_numa(void) |
6422 | { | |
6423 | int next_distance, curr_distance = node_distance(0, 0); | |
6424 | struct sched_domain_topology_level *tl; | |
6425 | int level = 0; | |
6426 | int i, j, k; | |
6427 | ||
cb83b629 PZ |
6428 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6429 | if (!sched_domains_numa_distance) | |
6430 | return; | |
6431 | ||
6432 | /* | |
6433 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6434 | * unique distances in the node_distance() table. | |
6435 | * | |
6436 | * Assumes node_distance(0,j) includes all distances in | |
6437 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6438 | */ |
6439 | next_distance = curr_distance; | |
6440 | for (i = 0; i < nr_node_ids; i++) { | |
6441 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6442 | for (k = 0; k < nr_node_ids; k++) { |
6443 | int distance = node_distance(i, k); | |
6444 | ||
6445 | if (distance > curr_distance && | |
6446 | (distance < next_distance || | |
6447 | next_distance == curr_distance)) | |
6448 | next_distance = distance; | |
6449 | ||
6450 | /* | |
6451 | * While not a strong assumption it would be nice to know | |
6452 | * about cases where if node A is connected to B, B is not | |
6453 | * equally connected to A. | |
6454 | */ | |
6455 | if (sched_debug() && node_distance(k, i) != distance) | |
6456 | sched_numa_warn("Node-distance not symmetric"); | |
6457 | ||
6458 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6459 | sched_numa_warn("Node-0 not representative"); | |
6460 | } | |
6461 | if (next_distance != curr_distance) { | |
6462 | sched_domains_numa_distance[level++] = next_distance; | |
6463 | sched_domains_numa_levels = level; | |
6464 | curr_distance = next_distance; | |
6465 | } else break; | |
cb83b629 | 6466 | } |
d039ac60 PZ |
6467 | |
6468 | /* | |
6469 | * In case of sched_debug() we verify the above assumption. | |
6470 | */ | |
6471 | if (!sched_debug()) | |
6472 | break; | |
cb83b629 | 6473 | } |
c123588b AR |
6474 | |
6475 | if (!level) | |
6476 | return; | |
6477 | ||
cb83b629 PZ |
6478 | /* |
6479 | * 'level' contains the number of unique distances, excluding the | |
6480 | * identity distance node_distance(i,i). | |
6481 | * | |
28b4a521 | 6482 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6483 | * numbers. |
6484 | */ | |
6485 | ||
5f7865f3 TC |
6486 | /* |
6487 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6488 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6489 | * the array will contain less then 'level' members. This could be | |
6490 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6491 | * in other functions. | |
6492 | * | |
6493 | * We reset it to 'level' at the end of this function. | |
6494 | */ | |
6495 | sched_domains_numa_levels = 0; | |
6496 | ||
cb83b629 PZ |
6497 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6498 | if (!sched_domains_numa_masks) | |
6499 | return; | |
6500 | ||
6501 | /* | |
6502 | * Now for each level, construct a mask per node which contains all | |
6503 | * cpus of nodes that are that many hops away from us. | |
6504 | */ | |
6505 | for (i = 0; i < level; i++) { | |
6506 | sched_domains_numa_masks[i] = | |
6507 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6508 | if (!sched_domains_numa_masks[i]) | |
6509 | return; | |
6510 | ||
6511 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6512 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6513 | if (!mask) |
6514 | return; | |
6515 | ||
6516 | sched_domains_numa_masks[i][j] = mask; | |
6517 | ||
6518 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6519 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6520 | continue; |
6521 | ||
6522 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6523 | } | |
6524 | } | |
6525 | } | |
6526 | ||
143e1e28 VG |
6527 | /* Compute default topology size */ |
6528 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6529 | ||
c515db8c | 6530 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6531 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6532 | if (!tl) | |
6533 | return; | |
6534 | ||
6535 | /* | |
6536 | * Copy the default topology bits.. | |
6537 | */ | |
143e1e28 VG |
6538 | for (i = 0; sched_domain_topology[i].mask; i++) |
6539 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6540 | |
6541 | /* | |
6542 | * .. and append 'j' levels of NUMA goodness. | |
6543 | */ | |
6544 | for (j = 0; j < level; i++, j++) { | |
6545 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6546 | .mask = sd_numa_mask, |
143e1e28 | 6547 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6548 | .flags = SDTL_OVERLAP, |
6549 | .numa_level = j, | |
143e1e28 | 6550 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6551 | }; |
6552 | } | |
6553 | ||
6554 | sched_domain_topology = tl; | |
5f7865f3 TC |
6555 | |
6556 | sched_domains_numa_levels = level; | |
9942f79b | 6557 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
e3fe70b1 RR |
6558 | |
6559 | init_numa_topology_type(); | |
cb83b629 | 6560 | } |
301a5cba TC |
6561 | |
6562 | static void sched_domains_numa_masks_set(int cpu) | |
6563 | { | |
6564 | int i, j; | |
6565 | int node = cpu_to_node(cpu); | |
6566 | ||
6567 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6568 | for (j = 0; j < nr_node_ids; j++) { | |
6569 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6570 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6571 | } | |
6572 | } | |
6573 | } | |
6574 | ||
6575 | static void sched_domains_numa_masks_clear(int cpu) | |
6576 | { | |
6577 | int i, j; | |
6578 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6579 | for (j = 0; j < nr_node_ids; j++) | |
6580 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6581 | } | |
6582 | } | |
6583 | ||
6584 | /* | |
6585 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6586 | * are onlined. | |
6587 | */ | |
6588 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6589 | unsigned long action, | |
6590 | void *hcpu) | |
6591 | { | |
6592 | int cpu = (long)hcpu; | |
6593 | ||
6594 | switch (action & ~CPU_TASKS_FROZEN) { | |
6595 | case CPU_ONLINE: | |
6596 | sched_domains_numa_masks_set(cpu); | |
6597 | break; | |
6598 | ||
6599 | case CPU_DEAD: | |
6600 | sched_domains_numa_masks_clear(cpu); | |
6601 | break; | |
6602 | ||
6603 | default: | |
6604 | return NOTIFY_DONE; | |
6605 | } | |
6606 | ||
6607 | return NOTIFY_OK; | |
cb83b629 PZ |
6608 | } |
6609 | #else | |
6610 | static inline void sched_init_numa(void) | |
6611 | { | |
6612 | } | |
301a5cba TC |
6613 | |
6614 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6615 | unsigned long action, | |
6616 | void *hcpu) | |
6617 | { | |
6618 | return 0; | |
6619 | } | |
cb83b629 PZ |
6620 | #endif /* CONFIG_NUMA */ |
6621 | ||
54ab4ff4 PZ |
6622 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6623 | { | |
6624 | struct sched_domain_topology_level *tl; | |
6625 | int j; | |
6626 | ||
27723a68 | 6627 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6628 | struct sd_data *sdd = &tl->data; |
6629 | ||
6630 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6631 | if (!sdd->sd) | |
6632 | return -ENOMEM; | |
6633 | ||
6634 | sdd->sg = alloc_percpu(struct sched_group *); | |
6635 | if (!sdd->sg) | |
6636 | return -ENOMEM; | |
6637 | ||
63b2ca30 NP |
6638 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6639 | if (!sdd->sgc) | |
9c3f75cb PZ |
6640 | return -ENOMEM; |
6641 | ||
54ab4ff4 PZ |
6642 | for_each_cpu(j, cpu_map) { |
6643 | struct sched_domain *sd; | |
6644 | struct sched_group *sg; | |
63b2ca30 | 6645 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6646 | |
6647 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6648 | GFP_KERNEL, cpu_to_node(j)); | |
6649 | if (!sd) | |
6650 | return -ENOMEM; | |
6651 | ||
6652 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6653 | ||
6654 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6655 | GFP_KERNEL, cpu_to_node(j)); | |
6656 | if (!sg) | |
6657 | return -ENOMEM; | |
6658 | ||
30b4e9eb IM |
6659 | sg->next = sg; |
6660 | ||
54ab4ff4 | 6661 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6662 | |
63b2ca30 | 6663 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6664 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6665 | if (!sgc) |
9c3f75cb PZ |
6666 | return -ENOMEM; |
6667 | ||
63b2ca30 | 6668 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6669 | } |
6670 | } | |
6671 | ||
6672 | return 0; | |
6673 | } | |
6674 | ||
6675 | static void __sdt_free(const struct cpumask *cpu_map) | |
6676 | { | |
6677 | struct sched_domain_topology_level *tl; | |
6678 | int j; | |
6679 | ||
27723a68 | 6680 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6681 | struct sd_data *sdd = &tl->data; |
6682 | ||
6683 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6684 | struct sched_domain *sd; |
6685 | ||
6686 | if (sdd->sd) { | |
6687 | sd = *per_cpu_ptr(sdd->sd, j); | |
6688 | if (sd && (sd->flags & SD_OVERLAP)) | |
6689 | free_sched_groups(sd->groups, 0); | |
6690 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6691 | } | |
6692 | ||
6693 | if (sdd->sg) | |
6694 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6695 | if (sdd->sgc) |
6696 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6697 | } |
6698 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6699 | sdd->sd = NULL; |
54ab4ff4 | 6700 | free_percpu(sdd->sg); |
fb2cf2c6 | 6701 | sdd->sg = NULL; |
63b2ca30 NP |
6702 | free_percpu(sdd->sgc); |
6703 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6704 | } |
6705 | } | |
6706 | ||
2c402dc3 | 6707 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6708 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6709 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6710 | { |
143e1e28 | 6711 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6712 | if (!sd) |
d069b916 | 6713 | return child; |
2c402dc3 | 6714 | |
2c402dc3 | 6715 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6716 | if (child) { |
6717 | sd->level = child->level + 1; | |
6718 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6719 | child->parent = sd; |
c75e0128 | 6720 | sd->child = child; |
6ae72dff PZ |
6721 | |
6722 | if (!cpumask_subset(sched_domain_span(child), | |
6723 | sched_domain_span(sd))) { | |
6724 | pr_err("BUG: arch topology borken\n"); | |
6725 | #ifdef CONFIG_SCHED_DEBUG | |
6726 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6727 | child->name, sd->name); | |
6728 | #endif | |
6729 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6730 | cpumask_or(sched_domain_span(sd), | |
6731 | sched_domain_span(sd), | |
6732 | sched_domain_span(child)); | |
6733 | } | |
6734 | ||
60495e77 | 6735 | } |
a841f8ce | 6736 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6737 | |
6738 | return sd; | |
6739 | } | |
6740 | ||
2109b99e AH |
6741 | /* |
6742 | * Build sched domains for a given set of cpus and attach the sched domains | |
6743 | * to the individual cpus | |
6744 | */ | |
dce840a0 PZ |
6745 | static int build_sched_domains(const struct cpumask *cpu_map, |
6746 | struct sched_domain_attr *attr) | |
2109b99e | 6747 | { |
1c632169 | 6748 | enum s_alloc alloc_state; |
dce840a0 | 6749 | struct sched_domain *sd; |
2109b99e | 6750 | struct s_data d; |
822ff793 | 6751 | int i, ret = -ENOMEM; |
9c1cfda2 | 6752 | |
2109b99e AH |
6753 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6754 | if (alloc_state != sa_rootdomain) | |
6755 | goto error; | |
9c1cfda2 | 6756 | |
dce840a0 | 6757 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6758 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6759 | struct sched_domain_topology_level *tl; |
6760 | ||
3bd65a80 | 6761 | sd = NULL; |
27723a68 | 6762 | for_each_sd_topology(tl) { |
4a850cbe | 6763 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6764 | if (tl == sched_domain_topology) |
6765 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6766 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6767 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6768 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6769 | break; | |
e3589f6c | 6770 | } |
dce840a0 PZ |
6771 | } |
6772 | ||
6773 | /* Build the groups for the domains */ | |
6774 | for_each_cpu(i, cpu_map) { | |
6775 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6776 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6777 | if (sd->flags & SD_OVERLAP) { |
6778 | if (build_overlap_sched_groups(sd, i)) | |
6779 | goto error; | |
6780 | } else { | |
6781 | if (build_sched_groups(sd, i)) | |
6782 | goto error; | |
6783 | } | |
1cf51902 | 6784 | } |
a06dadbe | 6785 | } |
9c1cfda2 | 6786 | |
ced549fa | 6787 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6788 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6789 | if (!cpumask_test_cpu(i, cpu_map)) | |
6790 | continue; | |
9c1cfda2 | 6791 | |
dce840a0 PZ |
6792 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6793 | claim_allocations(i, sd); | |
63b2ca30 | 6794 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6795 | } |
f712c0c7 | 6796 | } |
9c1cfda2 | 6797 | |
1da177e4 | 6798 | /* Attach the domains */ |
dce840a0 | 6799 | rcu_read_lock(); |
abcd083a | 6800 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6801 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6802 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6803 | } |
dce840a0 | 6804 | rcu_read_unlock(); |
51888ca2 | 6805 | |
822ff793 | 6806 | ret = 0; |
51888ca2 | 6807 | error: |
2109b99e | 6808 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6809 | return ret; |
1da177e4 | 6810 | } |
029190c5 | 6811 | |
acc3f5d7 | 6812 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6813 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6814 | static struct sched_domain_attr *dattr_cur; |
6815 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6816 | |
6817 | /* | |
6818 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6819 | * cpumask) fails, then fallback to a single sched domain, |
6820 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6821 | */ |
4212823f | 6822 | static cpumask_var_t fallback_doms; |
029190c5 | 6823 | |
ee79d1bd HC |
6824 | /* |
6825 | * arch_update_cpu_topology lets virtualized architectures update the | |
6826 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6827 | * or 0 if it stayed the same. | |
6828 | */ | |
52f5684c | 6829 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6830 | { |
ee79d1bd | 6831 | return 0; |
22e52b07 HC |
6832 | } |
6833 | ||
acc3f5d7 RR |
6834 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6835 | { | |
6836 | int i; | |
6837 | cpumask_var_t *doms; | |
6838 | ||
6839 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6840 | if (!doms) | |
6841 | return NULL; | |
6842 | for (i = 0; i < ndoms; i++) { | |
6843 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6844 | free_sched_domains(doms, i); | |
6845 | return NULL; | |
6846 | } | |
6847 | } | |
6848 | return doms; | |
6849 | } | |
6850 | ||
6851 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6852 | { | |
6853 | unsigned int i; | |
6854 | for (i = 0; i < ndoms; i++) | |
6855 | free_cpumask_var(doms[i]); | |
6856 | kfree(doms); | |
6857 | } | |
6858 | ||
1a20ff27 | 6859 | /* |
41a2d6cf | 6860 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6861 | * For now this just excludes isolated cpus, but could be used to |
6862 | * exclude other special cases in the future. | |
1a20ff27 | 6863 | */ |
c4a8849a | 6864 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6865 | { |
7378547f MM |
6866 | int err; |
6867 | ||
22e52b07 | 6868 | arch_update_cpu_topology(); |
029190c5 | 6869 | ndoms_cur = 1; |
acc3f5d7 | 6870 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6871 | if (!doms_cur) |
acc3f5d7 RR |
6872 | doms_cur = &fallback_doms; |
6873 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6874 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6875 | register_sched_domain_sysctl(); |
7378547f MM |
6876 | |
6877 | return err; | |
1a20ff27 DG |
6878 | } |
6879 | ||
1a20ff27 DG |
6880 | /* |
6881 | * Detach sched domains from a group of cpus specified in cpu_map | |
6882 | * These cpus will now be attached to the NULL domain | |
6883 | */ | |
96f874e2 | 6884 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6885 | { |
6886 | int i; | |
6887 | ||
dce840a0 | 6888 | rcu_read_lock(); |
abcd083a | 6889 | for_each_cpu(i, cpu_map) |
57d885fe | 6890 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6891 | rcu_read_unlock(); |
1a20ff27 DG |
6892 | } |
6893 | ||
1d3504fc HS |
6894 | /* handle null as "default" */ |
6895 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6896 | struct sched_domain_attr *new, int idx_new) | |
6897 | { | |
6898 | struct sched_domain_attr tmp; | |
6899 | ||
6900 | /* fast path */ | |
6901 | if (!new && !cur) | |
6902 | return 1; | |
6903 | ||
6904 | tmp = SD_ATTR_INIT; | |
6905 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6906 | new ? (new + idx_new) : &tmp, | |
6907 | sizeof(struct sched_domain_attr)); | |
6908 | } | |
6909 | ||
029190c5 PJ |
6910 | /* |
6911 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6912 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6913 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6914 | * It destroys each deleted domain and builds each new domain. | |
6915 | * | |
acc3f5d7 | 6916 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6917 | * The masks don't intersect (don't overlap.) We should setup one |
6918 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6919 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6920 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6921 | * it as it is. | |
6922 | * | |
acc3f5d7 RR |
6923 | * The passed in 'doms_new' should be allocated using |
6924 | * alloc_sched_domains. This routine takes ownership of it and will | |
6925 | * free_sched_domains it when done with it. If the caller failed the | |
6926 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6927 | * and partition_sched_domains() will fallback to the single partition | |
6928 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6929 | * |
96f874e2 | 6930 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6931 | * ndoms_new == 0 is a special case for destroying existing domains, |
6932 | * and it will not create the default domain. | |
dfb512ec | 6933 | * |
029190c5 PJ |
6934 | * Call with hotplug lock held |
6935 | */ | |
acc3f5d7 | 6936 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6937 | struct sched_domain_attr *dattr_new) |
029190c5 | 6938 | { |
dfb512ec | 6939 | int i, j, n; |
d65bd5ec | 6940 | int new_topology; |
029190c5 | 6941 | |
712555ee | 6942 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6943 | |
7378547f MM |
6944 | /* always unregister in case we don't destroy any domains */ |
6945 | unregister_sched_domain_sysctl(); | |
6946 | ||
d65bd5ec HC |
6947 | /* Let architecture update cpu core mappings. */ |
6948 | new_topology = arch_update_cpu_topology(); | |
6949 | ||
dfb512ec | 6950 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6951 | |
6952 | /* Destroy deleted domains */ | |
6953 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6954 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6955 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6956 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6957 | goto match1; |
6958 | } | |
6959 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6960 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6961 | match1: |
6962 | ; | |
6963 | } | |
6964 | ||
c8d2d47a | 6965 | n = ndoms_cur; |
e761b772 | 6966 | if (doms_new == NULL) { |
c8d2d47a | 6967 | n = 0; |
acc3f5d7 | 6968 | doms_new = &fallback_doms; |
6ad4c188 | 6969 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6970 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6971 | } |
6972 | ||
029190c5 PJ |
6973 | /* Build new domains */ |
6974 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6975 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6976 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6977 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6978 | goto match2; |
6979 | } | |
6980 | /* no match - add a new doms_new */ | |
dce840a0 | 6981 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6982 | match2: |
6983 | ; | |
6984 | } | |
6985 | ||
6986 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6987 | if (doms_cur != &fallback_doms) |
6988 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6989 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6990 | doms_cur = doms_new; |
1d3504fc | 6991 | dattr_cur = dattr_new; |
029190c5 | 6992 | ndoms_cur = ndoms_new; |
7378547f MM |
6993 | |
6994 | register_sched_domain_sysctl(); | |
a1835615 | 6995 | |
712555ee | 6996 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6997 | } |
6998 | ||
d35be8ba SB |
6999 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7000 | ||
1da177e4 | 7001 | /* |
3a101d05 TH |
7002 | * Update cpusets according to cpu_active mask. If cpusets are |
7003 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7004 | * around partition_sched_domains(). | |
d35be8ba SB |
7005 | * |
7006 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7007 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7008 | */ |
0b2e918a TH |
7009 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7010 | void *hcpu) | |
e761b772 | 7011 | { |
d35be8ba SB |
7012 | switch (action) { |
7013 | case CPU_ONLINE_FROZEN: | |
7014 | case CPU_DOWN_FAILED_FROZEN: | |
7015 | ||
7016 | /* | |
7017 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7018 | * resume sequence. As long as this is not the last online | |
7019 | * operation in the resume sequence, just build a single sched | |
7020 | * domain, ignoring cpusets. | |
7021 | */ | |
7022 | num_cpus_frozen--; | |
7023 | if (likely(num_cpus_frozen)) { | |
7024 | partition_sched_domains(1, NULL, NULL); | |
7025 | break; | |
7026 | } | |
7027 | ||
7028 | /* | |
7029 | * This is the last CPU online operation. So fall through and | |
7030 | * restore the original sched domains by considering the | |
7031 | * cpuset configurations. | |
7032 | */ | |
7033 | ||
e761b772 | 7034 | case CPU_ONLINE: |
7ddf96b0 | 7035 | cpuset_update_active_cpus(true); |
d35be8ba | 7036 | break; |
3a101d05 TH |
7037 | default: |
7038 | return NOTIFY_DONE; | |
7039 | } | |
d35be8ba | 7040 | return NOTIFY_OK; |
3a101d05 | 7041 | } |
e761b772 | 7042 | |
0b2e918a TH |
7043 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7044 | void *hcpu) | |
3a101d05 | 7045 | { |
3c18d447 JL |
7046 | unsigned long flags; |
7047 | long cpu = (long)hcpu; | |
7048 | struct dl_bw *dl_b; | |
533445c6 OS |
7049 | bool overflow; |
7050 | int cpus; | |
3c18d447 | 7051 | |
533445c6 | 7052 | switch (action) { |
3a101d05 | 7053 | case CPU_DOWN_PREPARE: |
533445c6 OS |
7054 | rcu_read_lock_sched(); |
7055 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 7056 | |
533445c6 OS |
7057 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
7058 | cpus = dl_bw_cpus(cpu); | |
7059 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
7060 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 7061 | |
533445c6 | 7062 | rcu_read_unlock_sched(); |
3c18d447 | 7063 | |
533445c6 OS |
7064 | if (overflow) |
7065 | return notifier_from_errno(-EBUSY); | |
7ddf96b0 | 7066 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7067 | break; |
7068 | case CPU_DOWN_PREPARE_FROZEN: | |
7069 | num_cpus_frozen++; | |
7070 | partition_sched_domains(1, NULL, NULL); | |
7071 | break; | |
e761b772 MK |
7072 | default: |
7073 | return NOTIFY_DONE; | |
7074 | } | |
d35be8ba | 7075 | return NOTIFY_OK; |
e761b772 | 7076 | } |
e761b772 | 7077 | |
1da177e4 LT |
7078 | void __init sched_init_smp(void) |
7079 | { | |
dcc30a35 RR |
7080 | cpumask_var_t non_isolated_cpus; |
7081 | ||
7082 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7083 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7084 | |
cb83b629 PZ |
7085 | sched_init_numa(); |
7086 | ||
6acce3ef PZ |
7087 | /* |
7088 | * There's no userspace yet to cause hotplug operations; hence all the | |
7089 | * cpu masks are stable and all blatant races in the below code cannot | |
7090 | * happen. | |
7091 | */ | |
712555ee | 7092 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7093 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7094 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7095 | if (cpumask_empty(non_isolated_cpus)) | |
7096 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7097 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7098 | |
301a5cba | 7099 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
7100 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7101 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 7102 | |
b328ca18 | 7103 | init_hrtick(); |
5c1e1767 NP |
7104 | |
7105 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7106 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7107 | BUG(); |
19978ca6 | 7108 | sched_init_granularity(); |
dcc30a35 | 7109 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7110 | |
0e3900e6 | 7111 | init_sched_rt_class(); |
1baca4ce | 7112 | init_sched_dl_class(); |
1da177e4 LT |
7113 | } |
7114 | #else | |
7115 | void __init sched_init_smp(void) | |
7116 | { | |
19978ca6 | 7117 | sched_init_granularity(); |
1da177e4 LT |
7118 | } |
7119 | #endif /* CONFIG_SMP */ | |
7120 | ||
cd1bb94b AB |
7121 | const_debug unsigned int sysctl_timer_migration = 1; |
7122 | ||
1da177e4 LT |
7123 | int in_sched_functions(unsigned long addr) |
7124 | { | |
1da177e4 LT |
7125 | return in_lock_functions(addr) || |
7126 | (addr >= (unsigned long)__sched_text_start | |
7127 | && addr < (unsigned long)__sched_text_end); | |
7128 | } | |
7129 | ||
029632fb | 7130 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7131 | /* |
7132 | * Default task group. | |
7133 | * Every task in system belongs to this group at bootup. | |
7134 | */ | |
029632fb | 7135 | struct task_group root_task_group; |
35cf4e50 | 7136 | LIST_HEAD(task_groups); |
052f1dc7 | 7137 | #endif |
6f505b16 | 7138 | |
e6252c3e | 7139 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 7140 | |
1da177e4 LT |
7141 | void __init sched_init(void) |
7142 | { | |
dd41f596 | 7143 | int i, j; |
434d53b0 MT |
7144 | unsigned long alloc_size = 0, ptr; |
7145 | ||
7146 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7147 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7148 | #endif | |
7149 | #ifdef CONFIG_RT_GROUP_SCHED | |
7150 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7151 | #endif | |
434d53b0 | 7152 | if (alloc_size) { |
36b7b6d4 | 7153 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7154 | |
7155 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7156 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7157 | ptr += nr_cpu_ids * sizeof(void **); |
7158 | ||
07e06b01 | 7159 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7160 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7161 | |
6d6bc0ad | 7162 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7163 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7164 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7165 | ptr += nr_cpu_ids * sizeof(void **); |
7166 | ||
07e06b01 | 7167 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7168 | ptr += nr_cpu_ids * sizeof(void **); |
7169 | ||
6d6bc0ad | 7170 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7171 | } |
df7c8e84 | 7172 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7173 | for_each_possible_cpu(i) { |
7174 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7175 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7176 | } |
b74e6278 | 7177 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7178 | |
332ac17e DF |
7179 | init_rt_bandwidth(&def_rt_bandwidth, |
7180 | global_rt_period(), global_rt_runtime()); | |
7181 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 7182 | global_rt_period(), global_rt_runtime()); |
332ac17e | 7183 | |
57d885fe GH |
7184 | #ifdef CONFIG_SMP |
7185 | init_defrootdomain(); | |
7186 | #endif | |
7187 | ||
d0b27fa7 | 7188 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7189 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7190 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7191 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7192 | |
7c941438 | 7193 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7194 | list_add(&root_task_group.list, &task_groups); |
7195 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7196 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7197 | autogroup_init(&init_task); |
54c707e9 | 7198 | |
7c941438 | 7199 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7200 | |
0a945022 | 7201 | for_each_possible_cpu(i) { |
70b97a7f | 7202 | struct rq *rq; |
1da177e4 LT |
7203 | |
7204 | rq = cpu_rq(i); | |
05fa785c | 7205 | raw_spin_lock_init(&rq->lock); |
7897986b | 7206 | rq->nr_running = 0; |
dce48a84 TG |
7207 | rq->calc_load_active = 0; |
7208 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7209 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7210 | init_rt_rq(&rq->rt); |
7211 | init_dl_rq(&rq->dl); | |
dd41f596 | 7212 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7213 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7214 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7215 | /* |
07e06b01 | 7216 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7217 | * |
7218 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7219 | * gets 100% of the cpu resources in the system. This overall | |
7220 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7221 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7222 | * based on each entity's (task or task-group's) weight |
7223 | * (se->load.weight). | |
7224 | * | |
07e06b01 | 7225 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7226 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7227 | * then A0's share of the cpu resource is: | |
7228 | * | |
0d905bca | 7229 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7230 | * |
07e06b01 YZ |
7231 | * We achieve this by letting root_task_group's tasks sit |
7232 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7233 | */ |
ab84d31e | 7234 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7235 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7236 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7237 | ||
7238 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7239 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7240 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7241 | #endif |
1da177e4 | 7242 | |
dd41f596 IM |
7243 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7244 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7245 | |
7246 | rq->last_load_update_tick = jiffies; | |
7247 | ||
1da177e4 | 7248 | #ifdef CONFIG_SMP |
41c7ce9a | 7249 | rq->sd = NULL; |
57d885fe | 7250 | rq->rd = NULL; |
ca6d75e6 | 7251 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
3f029d3c | 7252 | rq->post_schedule = 0; |
1da177e4 | 7253 | rq->active_balance = 0; |
dd41f596 | 7254 | rq->next_balance = jiffies; |
1da177e4 | 7255 | rq->push_cpu = 0; |
0a2966b4 | 7256 | rq->cpu = i; |
1f11eb6a | 7257 | rq->online = 0; |
eae0c9df MG |
7258 | rq->idle_stamp = 0; |
7259 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7260 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7261 | |
7262 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7263 | ||
dc938520 | 7264 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7265 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7266 | rq->nohz_flags = 0; |
83cd4fe2 | 7267 | #endif |
265f22a9 FW |
7268 | #ifdef CONFIG_NO_HZ_FULL |
7269 | rq->last_sched_tick = 0; | |
7270 | #endif | |
1da177e4 | 7271 | #endif |
8f4d37ec | 7272 | init_rq_hrtick(rq); |
1da177e4 | 7273 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7274 | } |
7275 | ||
2dd73a4f | 7276 | set_load_weight(&init_task); |
b50f60ce | 7277 | |
e107be36 AK |
7278 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7279 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7280 | #endif | |
7281 | ||
1da177e4 LT |
7282 | /* |
7283 | * The boot idle thread does lazy MMU switching as well: | |
7284 | */ | |
7285 | atomic_inc(&init_mm.mm_count); | |
7286 | enter_lazy_tlb(&init_mm, current); | |
7287 | ||
1b537c7d YD |
7288 | /* |
7289 | * During early bootup we pretend to be a normal task: | |
7290 | */ | |
7291 | current->sched_class = &fair_sched_class; | |
7292 | ||
1da177e4 LT |
7293 | /* |
7294 | * Make us the idle thread. Technically, schedule() should not be | |
7295 | * called from this thread, however somewhere below it might be, | |
7296 | * but because we are the idle thread, we just pick up running again | |
7297 | * when this runqueue becomes "idle". | |
7298 | */ | |
7299 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7300 | |
7301 | calc_load_update = jiffies + LOAD_FREQ; | |
7302 | ||
bf4d83f6 | 7303 | #ifdef CONFIG_SMP |
4cb98839 | 7304 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7305 | /* May be allocated at isolcpus cmdline parse time */ |
7306 | if (cpu_isolated_map == NULL) | |
7307 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7308 | idle_thread_set_boot_cpu(); |
a803f026 | 7309 | set_cpu_rq_start_time(); |
029632fb PZ |
7310 | #endif |
7311 | init_sched_fair_class(); | |
6a7b3dc3 | 7312 | |
6892b75e | 7313 | scheduler_running = 1; |
1da177e4 LT |
7314 | } |
7315 | ||
d902db1e | 7316 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7317 | static inline int preempt_count_equals(int preempt_offset) |
7318 | { | |
234da7bc | 7319 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7320 | |
4ba8216c | 7321 | return (nested == preempt_offset); |
e4aafea2 FW |
7322 | } |
7323 | ||
d894837f | 7324 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7325 | { |
8eb23b9f PZ |
7326 | /* |
7327 | * Blocking primitives will set (and therefore destroy) current->state, | |
7328 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7329 | * otherwise we will destroy state. | |
7330 | */ | |
00845eb9 | 7331 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7332 | "do not call blocking ops when !TASK_RUNNING; " |
7333 | "state=%lx set at [<%p>] %pS\n", | |
7334 | current->state, | |
7335 | (void *)current->task_state_change, | |
00845eb9 | 7336 | (void *)current->task_state_change); |
8eb23b9f | 7337 | |
3427445a PZ |
7338 | ___might_sleep(file, line, preempt_offset); |
7339 | } | |
7340 | EXPORT_SYMBOL(__might_sleep); | |
7341 | ||
7342 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7343 | { |
1da177e4 LT |
7344 | static unsigned long prev_jiffy; /* ratelimiting */ |
7345 | ||
b3fbab05 | 7346 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7347 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7348 | !is_idle_task(current)) || | |
e4aafea2 | 7349 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7350 | return; |
7351 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7352 | return; | |
7353 | prev_jiffy = jiffies; | |
7354 | ||
3df0fc5b PZ |
7355 | printk(KERN_ERR |
7356 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7357 | file, line); | |
7358 | printk(KERN_ERR | |
7359 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7360 | in_atomic(), irqs_disabled(), | |
7361 | current->pid, current->comm); | |
aef745fc | 7362 | |
a8b686b3 ES |
7363 | if (task_stack_end_corrupted(current)) |
7364 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7365 | ||
aef745fc IM |
7366 | debug_show_held_locks(current); |
7367 | if (irqs_disabled()) | |
7368 | print_irqtrace_events(current); | |
8f47b187 TG |
7369 | #ifdef CONFIG_DEBUG_PREEMPT |
7370 | if (!preempt_count_equals(preempt_offset)) { | |
7371 | pr_err("Preemption disabled at:"); | |
7372 | print_ip_sym(current->preempt_disable_ip); | |
7373 | pr_cont("\n"); | |
7374 | } | |
7375 | #endif | |
aef745fc | 7376 | dump_stack(); |
1da177e4 | 7377 | } |
3427445a | 7378 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
7379 | #endif |
7380 | ||
7381 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7382 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7383 | { | |
da7a735e | 7384 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7385 | struct sched_attr attr = { |
7386 | .sched_policy = SCHED_NORMAL, | |
7387 | }; | |
da7a735e | 7388 | int old_prio = p->prio; |
da0c1e65 | 7389 | int queued; |
3e51f33f | 7390 | |
da0c1e65 KT |
7391 | queued = task_on_rq_queued(p); |
7392 | if (queued) | |
4ca9b72b | 7393 | dequeue_task(rq, p, 0); |
0782e63b | 7394 | __setscheduler(rq, p, &attr, false); |
da0c1e65 | 7395 | if (queued) { |
4ca9b72b | 7396 | enqueue_task(rq, p, 0); |
8875125e | 7397 | resched_curr(rq); |
3a5e4dc1 | 7398 | } |
da7a735e PZ |
7399 | |
7400 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7401 | } |
7402 | ||
1da177e4 LT |
7403 | void normalize_rt_tasks(void) |
7404 | { | |
a0f98a1c | 7405 | struct task_struct *g, *p; |
1da177e4 | 7406 | unsigned long flags; |
70b97a7f | 7407 | struct rq *rq; |
1da177e4 | 7408 | |
3472eaa1 | 7409 | read_lock(&tasklist_lock); |
5d07f420 | 7410 | for_each_process_thread(g, p) { |
178be793 IM |
7411 | /* |
7412 | * Only normalize user tasks: | |
7413 | */ | |
3472eaa1 | 7414 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7415 | continue; |
7416 | ||
6cfb0d5d | 7417 | p->se.exec_start = 0; |
6cfb0d5d | 7418 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7419 | p->se.statistics.wait_start = 0; |
7420 | p->se.statistics.sleep_start = 0; | |
7421 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7422 | #endif |
dd41f596 | 7423 | |
aab03e05 | 7424 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7425 | /* |
7426 | * Renice negative nice level userspace | |
7427 | * tasks back to 0: | |
7428 | */ | |
3472eaa1 | 7429 | if (task_nice(p) < 0) |
dd41f596 | 7430 | set_user_nice(p, 0); |
1da177e4 | 7431 | continue; |
dd41f596 | 7432 | } |
1da177e4 | 7433 | |
3472eaa1 | 7434 | rq = task_rq_lock(p, &flags); |
178be793 | 7435 | normalize_task(rq, p); |
3472eaa1 | 7436 | task_rq_unlock(rq, p, &flags); |
5d07f420 | 7437 | } |
3472eaa1 | 7438 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7439 | } |
7440 | ||
7441 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7442 | |
67fc4e0c | 7443 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7444 | /* |
67fc4e0c | 7445 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7446 | * |
7447 | * They can only be called when the whole system has been | |
7448 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7449 | * activity can take place. Using them for anything else would | |
7450 | * be a serious bug, and as a result, they aren't even visible | |
7451 | * under any other configuration. | |
7452 | */ | |
7453 | ||
7454 | /** | |
7455 | * curr_task - return the current task for a given cpu. | |
7456 | * @cpu: the processor in question. | |
7457 | * | |
7458 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7459 | * |
7460 | * Return: The current task for @cpu. | |
1df5c10a | 7461 | */ |
36c8b586 | 7462 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7463 | { |
7464 | return cpu_curr(cpu); | |
7465 | } | |
7466 | ||
67fc4e0c JW |
7467 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7468 | ||
7469 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7470 | /** |
7471 | * set_curr_task - set the current task for a given cpu. | |
7472 | * @cpu: the processor in question. | |
7473 | * @p: the task pointer to set. | |
7474 | * | |
7475 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7476 | * are serviced on a separate stack. It allows the architecture to switch the |
7477 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7478 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7479 | * and caller must save the original value of the current task (see | |
7480 | * curr_task() above) and restore that value before reenabling interrupts and | |
7481 | * re-starting the system. | |
7482 | * | |
7483 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7484 | */ | |
36c8b586 | 7485 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7486 | { |
7487 | cpu_curr(cpu) = p; | |
7488 | } | |
7489 | ||
7490 | #endif | |
29f59db3 | 7491 | |
7c941438 | 7492 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7493 | /* task_group_lock serializes the addition/removal of task groups */ |
7494 | static DEFINE_SPINLOCK(task_group_lock); | |
7495 | ||
bccbe08a PZ |
7496 | static void free_sched_group(struct task_group *tg) |
7497 | { | |
7498 | free_fair_sched_group(tg); | |
7499 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7500 | autogroup_free(tg); |
bccbe08a PZ |
7501 | kfree(tg); |
7502 | } | |
7503 | ||
7504 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7505 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7506 | { |
7507 | struct task_group *tg; | |
bccbe08a PZ |
7508 | |
7509 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7510 | if (!tg) | |
7511 | return ERR_PTR(-ENOMEM); | |
7512 | ||
ec7dc8ac | 7513 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7514 | goto err; |
7515 | ||
ec7dc8ac | 7516 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7517 | goto err; |
7518 | ||
ace783b9 LZ |
7519 | return tg; |
7520 | ||
7521 | err: | |
7522 | free_sched_group(tg); | |
7523 | return ERR_PTR(-ENOMEM); | |
7524 | } | |
7525 | ||
7526 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7527 | { | |
7528 | unsigned long flags; | |
7529 | ||
8ed36996 | 7530 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7531 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7532 | |
7533 | WARN_ON(!parent); /* root should already exist */ | |
7534 | ||
7535 | tg->parent = parent; | |
f473aa5e | 7536 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7537 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7538 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7539 | } |
7540 | ||
9b5b7751 | 7541 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7542 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7543 | { |
29f59db3 | 7544 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7545 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7546 | } |
7547 | ||
9b5b7751 | 7548 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7549 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7550 | { |
7551 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7552 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7553 | } | |
7554 | ||
7555 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7556 | { |
8ed36996 | 7557 | unsigned long flags; |
9b5b7751 | 7558 | int i; |
29f59db3 | 7559 | |
3d4b47b4 PZ |
7560 | /* end participation in shares distribution */ |
7561 | for_each_possible_cpu(i) | |
bccbe08a | 7562 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7563 | |
7564 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7565 | list_del_rcu(&tg->list); |
f473aa5e | 7566 | list_del_rcu(&tg->siblings); |
8ed36996 | 7567 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7568 | } |
7569 | ||
9b5b7751 | 7570 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7571 | * The caller of this function should have put the task in its new group |
7572 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7573 | * reflect its new group. | |
9b5b7751 SV |
7574 | */ |
7575 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7576 | { |
8323f26c | 7577 | struct task_group *tg; |
da0c1e65 | 7578 | int queued, running; |
29f59db3 SV |
7579 | unsigned long flags; |
7580 | struct rq *rq; | |
7581 | ||
7582 | rq = task_rq_lock(tsk, &flags); | |
7583 | ||
051a1d1a | 7584 | running = task_current(rq, tsk); |
da0c1e65 | 7585 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7586 | |
da0c1e65 | 7587 | if (queued) |
29f59db3 | 7588 | dequeue_task(rq, tsk, 0); |
0e1f3483 | 7589 | if (unlikely(running)) |
f3cd1c4e | 7590 | put_prev_task(rq, tsk); |
29f59db3 | 7591 | |
f7b8a47d KT |
7592 | /* |
7593 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7594 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7595 | * to prevent lockdep warnings. | |
7596 | */ | |
7597 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7598 | struct task_group, css); |
7599 | tg = autogroup_task_group(tsk, tg); | |
7600 | tsk->sched_task_group = tg; | |
7601 | ||
810b3817 | 7602 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7603 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7604 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7605 | else |
810b3817 | 7606 | #endif |
b2b5ce02 | 7607 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7608 | |
0e1f3483 HS |
7609 | if (unlikely(running)) |
7610 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7611 | if (queued) |
371fd7e7 | 7612 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7613 | |
0122ec5b | 7614 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7615 | } |
7c941438 | 7616 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7617 | |
a790de99 PT |
7618 | #ifdef CONFIG_RT_GROUP_SCHED |
7619 | /* | |
7620 | * Ensure that the real time constraints are schedulable. | |
7621 | */ | |
7622 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7623 | |
9a7e0b18 PZ |
7624 | /* Must be called with tasklist_lock held */ |
7625 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7626 | { |
9a7e0b18 | 7627 | struct task_struct *g, *p; |
b40b2e8e | 7628 | |
1fe89e1b PZ |
7629 | /* |
7630 | * Autogroups do not have RT tasks; see autogroup_create(). | |
7631 | */ | |
7632 | if (task_group_is_autogroup(tg)) | |
7633 | return 0; | |
7634 | ||
5d07f420 | 7635 | for_each_process_thread(g, p) { |
8651c658 | 7636 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 7637 | return 1; |
5d07f420 | 7638 | } |
b40b2e8e | 7639 | |
9a7e0b18 PZ |
7640 | return 0; |
7641 | } | |
b40b2e8e | 7642 | |
9a7e0b18 PZ |
7643 | struct rt_schedulable_data { |
7644 | struct task_group *tg; | |
7645 | u64 rt_period; | |
7646 | u64 rt_runtime; | |
7647 | }; | |
b40b2e8e | 7648 | |
a790de99 | 7649 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7650 | { |
7651 | struct rt_schedulable_data *d = data; | |
7652 | struct task_group *child; | |
7653 | unsigned long total, sum = 0; | |
7654 | u64 period, runtime; | |
b40b2e8e | 7655 | |
9a7e0b18 PZ |
7656 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7657 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7658 | |
9a7e0b18 PZ |
7659 | if (tg == d->tg) { |
7660 | period = d->rt_period; | |
7661 | runtime = d->rt_runtime; | |
b40b2e8e | 7662 | } |
b40b2e8e | 7663 | |
4653f803 PZ |
7664 | /* |
7665 | * Cannot have more runtime than the period. | |
7666 | */ | |
7667 | if (runtime > period && runtime != RUNTIME_INF) | |
7668 | return -EINVAL; | |
6f505b16 | 7669 | |
4653f803 PZ |
7670 | /* |
7671 | * Ensure we don't starve existing RT tasks. | |
7672 | */ | |
9a7e0b18 PZ |
7673 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7674 | return -EBUSY; | |
6f505b16 | 7675 | |
9a7e0b18 | 7676 | total = to_ratio(period, runtime); |
6f505b16 | 7677 | |
4653f803 PZ |
7678 | /* |
7679 | * Nobody can have more than the global setting allows. | |
7680 | */ | |
7681 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7682 | return -EINVAL; | |
6f505b16 | 7683 | |
4653f803 PZ |
7684 | /* |
7685 | * The sum of our children's runtime should not exceed our own. | |
7686 | */ | |
9a7e0b18 PZ |
7687 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7688 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7689 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7690 | |
9a7e0b18 PZ |
7691 | if (child == d->tg) { |
7692 | period = d->rt_period; | |
7693 | runtime = d->rt_runtime; | |
7694 | } | |
6f505b16 | 7695 | |
9a7e0b18 | 7696 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7697 | } |
6f505b16 | 7698 | |
9a7e0b18 PZ |
7699 | if (sum > total) |
7700 | return -EINVAL; | |
7701 | ||
7702 | return 0; | |
6f505b16 PZ |
7703 | } |
7704 | ||
9a7e0b18 | 7705 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7706 | { |
8277434e PT |
7707 | int ret; |
7708 | ||
9a7e0b18 PZ |
7709 | struct rt_schedulable_data data = { |
7710 | .tg = tg, | |
7711 | .rt_period = period, | |
7712 | .rt_runtime = runtime, | |
7713 | }; | |
7714 | ||
8277434e PT |
7715 | rcu_read_lock(); |
7716 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7717 | rcu_read_unlock(); | |
7718 | ||
7719 | return ret; | |
521f1a24 DG |
7720 | } |
7721 | ||
ab84d31e | 7722 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7723 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7724 | { |
ac086bc2 | 7725 | int i, err = 0; |
9f0c1e56 | 7726 | |
2636ed5f PZ |
7727 | /* |
7728 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
7729 | * kernel creating (and or operating) RT threads. | |
7730 | */ | |
7731 | if (tg == &root_task_group && rt_runtime == 0) | |
7732 | return -EINVAL; | |
7733 | ||
7734 | /* No period doesn't make any sense. */ | |
7735 | if (rt_period == 0) | |
7736 | return -EINVAL; | |
7737 | ||
9f0c1e56 | 7738 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7739 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7740 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7741 | if (err) | |
9f0c1e56 | 7742 | goto unlock; |
ac086bc2 | 7743 | |
0986b11b | 7744 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7745 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7746 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7747 | |
7748 | for_each_possible_cpu(i) { | |
7749 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7750 | ||
0986b11b | 7751 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7752 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7753 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7754 | } |
0986b11b | 7755 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7756 | unlock: |
521f1a24 | 7757 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7758 | mutex_unlock(&rt_constraints_mutex); |
7759 | ||
7760 | return err; | |
6f505b16 PZ |
7761 | } |
7762 | ||
25cc7da7 | 7763 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7764 | { |
7765 | u64 rt_runtime, rt_period; | |
7766 | ||
7767 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7768 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7769 | if (rt_runtime_us < 0) | |
7770 | rt_runtime = RUNTIME_INF; | |
7771 | ||
ab84d31e | 7772 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7773 | } |
7774 | ||
25cc7da7 | 7775 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7776 | { |
7777 | u64 rt_runtime_us; | |
7778 | ||
d0b27fa7 | 7779 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7780 | return -1; |
7781 | ||
d0b27fa7 | 7782 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7783 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7784 | return rt_runtime_us; | |
7785 | } | |
d0b27fa7 | 7786 | |
ce2f5fe4 | 7787 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
7788 | { |
7789 | u64 rt_runtime, rt_period; | |
7790 | ||
ce2f5fe4 | 7791 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
7792 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
7793 | ||
ab84d31e | 7794 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7795 | } |
7796 | ||
25cc7da7 | 7797 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7798 | { |
7799 | u64 rt_period_us; | |
7800 | ||
7801 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7802 | do_div(rt_period_us, NSEC_PER_USEC); | |
7803 | return rt_period_us; | |
7804 | } | |
332ac17e | 7805 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7806 | |
332ac17e | 7807 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7808 | static int sched_rt_global_constraints(void) |
7809 | { | |
7810 | int ret = 0; | |
7811 | ||
7812 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7813 | read_lock(&tasklist_lock); |
4653f803 | 7814 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7815 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7816 | mutex_unlock(&rt_constraints_mutex); |
7817 | ||
7818 | return ret; | |
7819 | } | |
54e99124 | 7820 | |
25cc7da7 | 7821 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7822 | { |
7823 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7824 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7825 | return 0; | |
7826 | ||
7827 | return 1; | |
7828 | } | |
7829 | ||
6d6bc0ad | 7830 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7831 | static int sched_rt_global_constraints(void) |
7832 | { | |
ac086bc2 | 7833 | unsigned long flags; |
332ac17e | 7834 | int i, ret = 0; |
ec5d4989 | 7835 | |
0986b11b | 7836 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7837 | for_each_possible_cpu(i) { |
7838 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7839 | ||
0986b11b | 7840 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7841 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7842 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7843 | } |
0986b11b | 7844 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7845 | |
332ac17e | 7846 | return ret; |
d0b27fa7 | 7847 | } |
6d6bc0ad | 7848 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7849 | |
a1963b81 | 7850 | static int sched_dl_global_validate(void) |
332ac17e | 7851 | { |
1724813d PZ |
7852 | u64 runtime = global_rt_runtime(); |
7853 | u64 period = global_rt_period(); | |
332ac17e | 7854 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 7855 | struct dl_bw *dl_b; |
1724813d | 7856 | int cpu, ret = 0; |
49516342 | 7857 | unsigned long flags; |
332ac17e DF |
7858 | |
7859 | /* | |
7860 | * Here we want to check the bandwidth not being set to some | |
7861 | * value smaller than the currently allocated bandwidth in | |
7862 | * any of the root_domains. | |
7863 | * | |
7864 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7865 | * cycling on root_domains... Discussion on different/better | |
7866 | * solutions is welcome! | |
7867 | */ | |
1724813d | 7868 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
7869 | rcu_read_lock_sched(); |
7870 | dl_b = dl_bw_of(cpu); | |
332ac17e | 7871 | |
49516342 | 7872 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7873 | if (new_bw < dl_b->total_bw) |
7874 | ret = -EBUSY; | |
49516342 | 7875 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 7876 | |
f10e00f4 KT |
7877 | rcu_read_unlock_sched(); |
7878 | ||
1724813d PZ |
7879 | if (ret) |
7880 | break; | |
332ac17e DF |
7881 | } |
7882 | ||
1724813d | 7883 | return ret; |
332ac17e DF |
7884 | } |
7885 | ||
1724813d | 7886 | static void sched_dl_do_global(void) |
ce0dbbbb | 7887 | { |
1724813d | 7888 | u64 new_bw = -1; |
f10e00f4 | 7889 | struct dl_bw *dl_b; |
1724813d | 7890 | int cpu; |
49516342 | 7891 | unsigned long flags; |
ce0dbbbb | 7892 | |
1724813d PZ |
7893 | def_dl_bandwidth.dl_period = global_rt_period(); |
7894 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7895 | ||
7896 | if (global_rt_runtime() != RUNTIME_INF) | |
7897 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7898 | ||
7899 | /* | |
7900 | * FIXME: As above... | |
7901 | */ | |
7902 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
7903 | rcu_read_lock_sched(); |
7904 | dl_b = dl_bw_of(cpu); | |
1724813d | 7905 | |
49516342 | 7906 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7907 | dl_b->bw = new_bw; |
49516342 | 7908 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
7909 | |
7910 | rcu_read_unlock_sched(); | |
ce0dbbbb | 7911 | } |
1724813d PZ |
7912 | } |
7913 | ||
7914 | static int sched_rt_global_validate(void) | |
7915 | { | |
7916 | if (sysctl_sched_rt_period <= 0) | |
7917 | return -EINVAL; | |
7918 | ||
e9e7cb38 JL |
7919 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7920 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7921 | return -EINVAL; |
7922 | ||
7923 | return 0; | |
7924 | } | |
7925 | ||
7926 | static void sched_rt_do_global(void) | |
7927 | { | |
7928 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7929 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7930 | } |
7931 | ||
d0b27fa7 | 7932 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7933 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7934 | loff_t *ppos) |
7935 | { | |
d0b27fa7 PZ |
7936 | int old_period, old_runtime; |
7937 | static DEFINE_MUTEX(mutex); | |
1724813d | 7938 | int ret; |
d0b27fa7 PZ |
7939 | |
7940 | mutex_lock(&mutex); | |
7941 | old_period = sysctl_sched_rt_period; | |
7942 | old_runtime = sysctl_sched_rt_runtime; | |
7943 | ||
8d65af78 | 7944 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7945 | |
7946 | if (!ret && write) { | |
1724813d PZ |
7947 | ret = sched_rt_global_validate(); |
7948 | if (ret) | |
7949 | goto undo; | |
7950 | ||
a1963b81 | 7951 | ret = sched_dl_global_validate(); |
1724813d PZ |
7952 | if (ret) |
7953 | goto undo; | |
7954 | ||
a1963b81 | 7955 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7956 | if (ret) |
7957 | goto undo; | |
7958 | ||
7959 | sched_rt_do_global(); | |
7960 | sched_dl_do_global(); | |
7961 | } | |
7962 | if (0) { | |
7963 | undo: | |
7964 | sysctl_sched_rt_period = old_period; | |
7965 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7966 | } |
7967 | mutex_unlock(&mutex); | |
7968 | ||
7969 | return ret; | |
7970 | } | |
68318b8e | 7971 | |
1724813d | 7972 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7973 | void __user *buffer, size_t *lenp, |
7974 | loff_t *ppos) | |
7975 | { | |
7976 | int ret; | |
332ac17e | 7977 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7978 | |
7979 | mutex_lock(&mutex); | |
332ac17e | 7980 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7981 | /* make sure that internally we keep jiffies */ |
7982 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7983 | if (!ret && write) { |
1724813d PZ |
7984 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7985 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7986 | } |
7987 | mutex_unlock(&mutex); | |
332ac17e DF |
7988 | return ret; |
7989 | } | |
7990 | ||
052f1dc7 | 7991 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7992 | |
a7c6d554 | 7993 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7994 | { |
a7c6d554 | 7995 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7996 | } |
7997 | ||
eb95419b TH |
7998 | static struct cgroup_subsys_state * |
7999 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8000 | { |
eb95419b TH |
8001 | struct task_group *parent = css_tg(parent_css); |
8002 | struct task_group *tg; | |
68318b8e | 8003 | |
eb95419b | 8004 | if (!parent) { |
68318b8e | 8005 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8006 | return &root_task_group.css; |
68318b8e SV |
8007 | } |
8008 | ||
ec7dc8ac | 8009 | tg = sched_create_group(parent); |
68318b8e SV |
8010 | if (IS_ERR(tg)) |
8011 | return ERR_PTR(-ENOMEM); | |
8012 | ||
68318b8e SV |
8013 | return &tg->css; |
8014 | } | |
8015 | ||
eb95419b | 8016 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 8017 | { |
eb95419b | 8018 | struct task_group *tg = css_tg(css); |
5c9d535b | 8019 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 8020 | |
63876986 TH |
8021 | if (parent) |
8022 | sched_online_group(tg, parent); | |
ace783b9 LZ |
8023 | return 0; |
8024 | } | |
8025 | ||
eb95419b | 8026 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8027 | { |
eb95419b | 8028 | struct task_group *tg = css_tg(css); |
68318b8e SV |
8029 | |
8030 | sched_destroy_group(tg); | |
8031 | } | |
8032 | ||
eb95419b | 8033 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 8034 | { |
eb95419b | 8035 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
8036 | |
8037 | sched_offline_group(tg); | |
8038 | } | |
8039 | ||
eeb61e53 KT |
8040 | static void cpu_cgroup_fork(struct task_struct *task) |
8041 | { | |
8042 | sched_move_task(task); | |
8043 | } | |
8044 | ||
eb95419b | 8045 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8046 | struct cgroup_taskset *tset) |
68318b8e | 8047 | { |
bb9d97b6 TH |
8048 | struct task_struct *task; |
8049 | ||
924f0d9a | 8050 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 8051 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8052 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8053 | return -EINVAL; |
b68aa230 | 8054 | #else |
bb9d97b6 TH |
8055 | /* We don't support RT-tasks being in separate groups */ |
8056 | if (task->sched_class != &fair_sched_class) | |
8057 | return -EINVAL; | |
b68aa230 | 8058 | #endif |
bb9d97b6 | 8059 | } |
be367d09 BB |
8060 | return 0; |
8061 | } | |
68318b8e | 8062 | |
eb95419b | 8063 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 8064 | struct cgroup_taskset *tset) |
68318b8e | 8065 | { |
bb9d97b6 TH |
8066 | struct task_struct *task; |
8067 | ||
924f0d9a | 8068 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 8069 | sched_move_task(task); |
68318b8e SV |
8070 | } |
8071 | ||
eb95419b TH |
8072 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
8073 | struct cgroup_subsys_state *old_css, | |
8074 | struct task_struct *task) | |
068c5cc5 PZ |
8075 | { |
8076 | /* | |
8077 | * cgroup_exit() is called in the copy_process() failure path. | |
8078 | * Ignore this case since the task hasn't ran yet, this avoids | |
8079 | * trying to poke a half freed task state from generic code. | |
8080 | */ | |
8081 | if (!(task->flags & PF_EXITING)) | |
8082 | return; | |
8083 | ||
8084 | sched_move_task(task); | |
8085 | } | |
8086 | ||
052f1dc7 | 8087 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8088 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8089 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8090 | { |
182446d0 | 8091 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8092 | } |
8093 | ||
182446d0 TH |
8094 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8095 | struct cftype *cft) | |
68318b8e | 8096 | { |
182446d0 | 8097 | struct task_group *tg = css_tg(css); |
68318b8e | 8098 | |
c8b28116 | 8099 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8100 | } |
ab84d31e PT |
8101 | |
8102 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8103 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8104 | ||
ab84d31e PT |
8105 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8106 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8107 | ||
a790de99 PT |
8108 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8109 | ||
ab84d31e PT |
8110 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8111 | { | |
56f570e5 | 8112 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8113 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8114 | |
8115 | if (tg == &root_task_group) | |
8116 | return -EINVAL; | |
8117 | ||
8118 | /* | |
8119 | * Ensure we have at some amount of bandwidth every period. This is | |
8120 | * to prevent reaching a state of large arrears when throttled via | |
8121 | * entity_tick() resulting in prolonged exit starvation. | |
8122 | */ | |
8123 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8124 | return -EINVAL; | |
8125 | ||
8126 | /* | |
8127 | * Likewise, bound things on the otherside by preventing insane quota | |
8128 | * periods. This also allows us to normalize in computing quota | |
8129 | * feasibility. | |
8130 | */ | |
8131 | if (period > max_cfs_quota_period) | |
8132 | return -EINVAL; | |
8133 | ||
0e59bdae KT |
8134 | /* |
8135 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8136 | * unthrottle_offline_cfs_rqs(). | |
8137 | */ | |
8138 | get_online_cpus(); | |
a790de99 PT |
8139 | mutex_lock(&cfs_constraints_mutex); |
8140 | ret = __cfs_schedulable(tg, period, quota); | |
8141 | if (ret) | |
8142 | goto out_unlock; | |
8143 | ||
58088ad0 | 8144 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8145 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8146 | /* |
8147 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8148 | * before making related changes, and on->off must occur afterwards | |
8149 | */ | |
8150 | if (runtime_enabled && !runtime_was_enabled) | |
8151 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8152 | raw_spin_lock_irq(&cfs_b->lock); |
8153 | cfs_b->period = ns_to_ktime(period); | |
8154 | cfs_b->quota = quota; | |
58088ad0 | 8155 | |
a9cf55b2 | 8156 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
8157 | /* restart the period timer (if active) to handle new period expiry */ |
8158 | if (runtime_enabled && cfs_b->timer_active) { | |
8159 | /* force a reprogram */ | |
09dc4ab0 | 8160 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 8161 | } |
ab84d31e PT |
8162 | raw_spin_unlock_irq(&cfs_b->lock); |
8163 | ||
0e59bdae | 8164 | for_each_online_cpu(i) { |
ab84d31e | 8165 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8166 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8167 | |
8168 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8169 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8170 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8171 | |
029632fb | 8172 | if (cfs_rq->throttled) |
671fd9da | 8173 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8174 | raw_spin_unlock_irq(&rq->lock); |
8175 | } | |
1ee14e6c BS |
8176 | if (runtime_was_enabled && !runtime_enabled) |
8177 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8178 | out_unlock: |
8179 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8180 | put_online_cpus(); |
ab84d31e | 8181 | |
a790de99 | 8182 | return ret; |
ab84d31e PT |
8183 | } |
8184 | ||
8185 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8186 | { | |
8187 | u64 quota, period; | |
8188 | ||
029632fb | 8189 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8190 | if (cfs_quota_us < 0) |
8191 | quota = RUNTIME_INF; | |
8192 | else | |
8193 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8194 | ||
8195 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8196 | } | |
8197 | ||
8198 | long tg_get_cfs_quota(struct task_group *tg) | |
8199 | { | |
8200 | u64 quota_us; | |
8201 | ||
029632fb | 8202 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8203 | return -1; |
8204 | ||
029632fb | 8205 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8206 | do_div(quota_us, NSEC_PER_USEC); |
8207 | ||
8208 | return quota_us; | |
8209 | } | |
8210 | ||
8211 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8212 | { | |
8213 | u64 quota, period; | |
8214 | ||
8215 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8216 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8217 | |
ab84d31e PT |
8218 | return tg_set_cfs_bandwidth(tg, period, quota); |
8219 | } | |
8220 | ||
8221 | long tg_get_cfs_period(struct task_group *tg) | |
8222 | { | |
8223 | u64 cfs_period_us; | |
8224 | ||
029632fb | 8225 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8226 | do_div(cfs_period_us, NSEC_PER_USEC); |
8227 | ||
8228 | return cfs_period_us; | |
8229 | } | |
8230 | ||
182446d0 TH |
8231 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8232 | struct cftype *cft) | |
ab84d31e | 8233 | { |
182446d0 | 8234 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8235 | } |
8236 | ||
182446d0 TH |
8237 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8238 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8239 | { |
182446d0 | 8240 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8241 | } |
8242 | ||
182446d0 TH |
8243 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8244 | struct cftype *cft) | |
ab84d31e | 8245 | { |
182446d0 | 8246 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8247 | } |
8248 | ||
182446d0 TH |
8249 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8250 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8251 | { |
182446d0 | 8252 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8253 | } |
8254 | ||
a790de99 PT |
8255 | struct cfs_schedulable_data { |
8256 | struct task_group *tg; | |
8257 | u64 period, quota; | |
8258 | }; | |
8259 | ||
8260 | /* | |
8261 | * normalize group quota/period to be quota/max_period | |
8262 | * note: units are usecs | |
8263 | */ | |
8264 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8265 | struct cfs_schedulable_data *d) | |
8266 | { | |
8267 | u64 quota, period; | |
8268 | ||
8269 | if (tg == d->tg) { | |
8270 | period = d->period; | |
8271 | quota = d->quota; | |
8272 | } else { | |
8273 | period = tg_get_cfs_period(tg); | |
8274 | quota = tg_get_cfs_quota(tg); | |
8275 | } | |
8276 | ||
8277 | /* note: these should typically be equivalent */ | |
8278 | if (quota == RUNTIME_INF || quota == -1) | |
8279 | return RUNTIME_INF; | |
8280 | ||
8281 | return to_ratio(period, quota); | |
8282 | } | |
8283 | ||
8284 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8285 | { | |
8286 | struct cfs_schedulable_data *d = data; | |
029632fb | 8287 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8288 | s64 quota = 0, parent_quota = -1; |
8289 | ||
8290 | if (!tg->parent) { | |
8291 | quota = RUNTIME_INF; | |
8292 | } else { | |
029632fb | 8293 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8294 | |
8295 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8296 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8297 | |
8298 | /* | |
8299 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8300 | * limit is set | |
8301 | */ | |
8302 | if (quota == RUNTIME_INF) | |
8303 | quota = parent_quota; | |
8304 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8305 | return -EINVAL; | |
8306 | } | |
9c58c79a | 8307 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8308 | |
8309 | return 0; | |
8310 | } | |
8311 | ||
8312 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8313 | { | |
8277434e | 8314 | int ret; |
a790de99 PT |
8315 | struct cfs_schedulable_data data = { |
8316 | .tg = tg, | |
8317 | .period = period, | |
8318 | .quota = quota, | |
8319 | }; | |
8320 | ||
8321 | if (quota != RUNTIME_INF) { | |
8322 | do_div(data.period, NSEC_PER_USEC); | |
8323 | do_div(data.quota, NSEC_PER_USEC); | |
8324 | } | |
8325 | ||
8277434e PT |
8326 | rcu_read_lock(); |
8327 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8328 | rcu_read_unlock(); | |
8329 | ||
8330 | return ret; | |
a790de99 | 8331 | } |
e8da1b18 | 8332 | |
2da8ca82 | 8333 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8334 | { |
2da8ca82 | 8335 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8336 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8337 | |
44ffc75b TH |
8338 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8339 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8340 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8341 | |
8342 | return 0; | |
8343 | } | |
ab84d31e | 8344 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8345 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8346 | |
052f1dc7 | 8347 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8348 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8349 | struct cftype *cft, s64 val) | |
6f505b16 | 8350 | { |
182446d0 | 8351 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8352 | } |
8353 | ||
182446d0 TH |
8354 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8355 | struct cftype *cft) | |
6f505b16 | 8356 | { |
182446d0 | 8357 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8358 | } |
d0b27fa7 | 8359 | |
182446d0 TH |
8360 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8361 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8362 | { |
182446d0 | 8363 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8364 | } |
8365 | ||
182446d0 TH |
8366 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8367 | struct cftype *cft) | |
d0b27fa7 | 8368 | { |
182446d0 | 8369 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8370 | } |
6d6bc0ad | 8371 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8372 | |
fe5c7cc2 | 8373 | static struct cftype cpu_files[] = { |
052f1dc7 | 8374 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8375 | { |
8376 | .name = "shares", | |
f4c753b7 PM |
8377 | .read_u64 = cpu_shares_read_u64, |
8378 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8379 | }, |
052f1dc7 | 8380 | #endif |
ab84d31e PT |
8381 | #ifdef CONFIG_CFS_BANDWIDTH |
8382 | { | |
8383 | .name = "cfs_quota_us", | |
8384 | .read_s64 = cpu_cfs_quota_read_s64, | |
8385 | .write_s64 = cpu_cfs_quota_write_s64, | |
8386 | }, | |
8387 | { | |
8388 | .name = "cfs_period_us", | |
8389 | .read_u64 = cpu_cfs_period_read_u64, | |
8390 | .write_u64 = cpu_cfs_period_write_u64, | |
8391 | }, | |
e8da1b18 NR |
8392 | { |
8393 | .name = "stat", | |
2da8ca82 | 8394 | .seq_show = cpu_stats_show, |
e8da1b18 | 8395 | }, |
ab84d31e | 8396 | #endif |
052f1dc7 | 8397 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8398 | { |
9f0c1e56 | 8399 | .name = "rt_runtime_us", |
06ecb27c PM |
8400 | .read_s64 = cpu_rt_runtime_read, |
8401 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8402 | }, |
d0b27fa7 PZ |
8403 | { |
8404 | .name = "rt_period_us", | |
f4c753b7 PM |
8405 | .read_u64 = cpu_rt_period_read_uint, |
8406 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8407 | }, |
052f1dc7 | 8408 | #endif |
4baf6e33 | 8409 | { } /* terminate */ |
68318b8e SV |
8410 | }; |
8411 | ||
073219e9 | 8412 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8413 | .css_alloc = cpu_cgroup_css_alloc, |
8414 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8415 | .css_online = cpu_cgroup_css_online, |
8416 | .css_offline = cpu_cgroup_css_offline, | |
eeb61e53 | 8417 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8418 | .can_attach = cpu_cgroup_can_attach, |
8419 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8420 | .exit = cpu_cgroup_exit, |
5577964e | 8421 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8422 | .early_init = 1, |
8423 | }; | |
8424 | ||
052f1dc7 | 8425 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8426 | |
b637a328 PM |
8427 | void dump_cpu_task(int cpu) |
8428 | { | |
8429 | pr_info("Task dump for CPU %d:\n", cpu); | |
8430 | sched_show_task(cpu_curr(cpu)); | |
8431 | } |