blob: f2b26a6387806d0a1e9bd1dc0ad1b80a0b5a654b
1 | /* |
2 | * linux/kernel/softirq.c |
3 | * |
4 | * Copyright (C) 1992 Linus Torvalds |
5 | * |
6 | * Distribute under GPLv2. |
7 | * |
8 | * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903) |
9 | */ |
10 | |
11 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
12 | |
13 | #include <linux/export.h> |
14 | #include <linux/kernel_stat.h> |
15 | #include <linux/interrupt.h> |
16 | #include <linux/init.h> |
17 | #include <linux/mm.h> |
18 | #include <linux/notifier.h> |
19 | #include <linux/percpu.h> |
20 | #include <linux/cpu.h> |
21 | #include <linux/freezer.h> |
22 | #include <linux/kthread.h> |
23 | #include <linux/rcupdate.h> |
24 | #include <linux/ftrace.h> |
25 | #include <linux/smp.h> |
26 | #include <linux/smpboot.h> |
27 | #include <linux/tick.h> |
28 | #include <linux/irq.h> |
29 | |
30 | #define CREATE_TRACE_POINTS |
31 | #include <trace/events/irq.h> |
32 | |
33 | /* |
34 | - No shared variables, all the data are CPU local. |
35 | - If a softirq needs serialization, let it serialize itself |
36 | by its own spinlocks. |
37 | - Even if softirq is serialized, only local cpu is marked for |
38 | execution. Hence, we get something sort of weak cpu binding. |
39 | Though it is still not clear, will it result in better locality |
40 | or will not. |
41 | |
42 | Examples: |
43 | - NET RX softirq. It is multithreaded and does not require |
44 | any global serialization. |
45 | - NET TX softirq. It kicks software netdevice queues, hence |
46 | it is logically serialized per device, but this serialization |
47 | is invisible to common code. |
48 | - Tasklets: serialized wrt itself. |
49 | */ |
50 | |
51 | #ifndef __ARCH_IRQ_STAT |
52 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; |
53 | EXPORT_SYMBOL(irq_stat); |
54 | #endif |
55 | |
56 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
57 | |
58 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
59 | |
60 | const char * const softirq_to_name[NR_SOFTIRQS] = { |
61 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL", |
62 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
63 | }; |
64 | |
65 | /* |
66 | * we cannot loop indefinitely here to avoid userspace starvation, |
67 | * but we also don't want to introduce a worst case 1/HZ latency |
68 | * to the pending events, so lets the scheduler to balance |
69 | * the softirq load for us. |
70 | */ |
71 | static void wakeup_softirqd(void) |
72 | { |
73 | /* Interrupts are disabled: no need to stop preemption */ |
74 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
75 | |
76 | if (tsk && tsk->state != TASK_RUNNING) |
77 | wake_up_process(tsk); |
78 | } |
79 | |
80 | /* |
81 | * If ksoftirqd is scheduled, we do not want to process pending softirqs |
82 | * right now. Let ksoftirqd handle this at its own rate, to get fairness, |
83 | * unless we're doing some of the synchronous softirqs. |
84 | */ |
85 | #define SOFTIRQ_NOW_MASK ((1 << HI_SOFTIRQ) | (1 << TASKLET_SOFTIRQ)) |
86 | static bool ksoftirqd_running(unsigned long pending) |
87 | { |
88 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
89 | |
90 | if (pending & SOFTIRQ_NOW_MASK) |
91 | return false; |
92 | return tsk && (tsk->state == TASK_RUNNING); |
93 | } |
94 | |
95 | /* |
96 | * preempt_count and SOFTIRQ_OFFSET usage: |
97 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving |
98 | * softirq processing. |
99 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) |
100 | * on local_bh_disable or local_bh_enable. |
101 | * This lets us distinguish between whether we are currently processing |
102 | * softirq and whether we just have bh disabled. |
103 | */ |
104 | |
105 | /* |
106 | * This one is for softirq.c-internal use, |
107 | * where hardirqs are disabled legitimately: |
108 | */ |
109 | #ifdef CONFIG_TRACE_IRQFLAGS |
110 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
111 | { |
112 | unsigned long flags; |
113 | |
114 | WARN_ON_ONCE(in_irq()); |
115 | |
116 | raw_local_irq_save(flags); |
117 | /* |
118 | * The preempt tracer hooks into preempt_count_add and will break |
119 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
120 | * is set and before current->softirq_enabled is cleared. |
121 | * We must manually increment preempt_count here and manually |
122 | * call the trace_preempt_off later. |
123 | */ |
124 | __preempt_count_add(cnt); |
125 | /* |
126 | * Were softirqs turned off above: |
127 | */ |
128 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
129 | trace_softirqs_off(ip); |
130 | raw_local_irq_restore(flags); |
131 | |
132 | if (preempt_count() == cnt) { |
133 | #ifdef CONFIG_DEBUG_PREEMPT |
134 | current->preempt_disable_ip = get_lock_parent_ip(); |
135 | #endif |
136 | trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip()); |
137 | } |
138 | } |
139 | EXPORT_SYMBOL(__local_bh_disable_ip); |
140 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
141 | |
142 | static void __local_bh_enable(unsigned int cnt) |
143 | { |
144 | WARN_ON_ONCE(!irqs_disabled()); |
145 | |
146 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
147 | trace_softirqs_on(_RET_IP_); |
148 | preempt_count_sub(cnt); |
149 | } |
150 | |
151 | /* |
152 | * Special-case - softirqs can safely be enabled in |
153 | * cond_resched_softirq(), or by __do_softirq(), |
154 | * without processing still-pending softirqs: |
155 | */ |
156 | void _local_bh_enable(void) |
157 | { |
158 | WARN_ON_ONCE(in_irq()); |
159 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
160 | } |
161 | EXPORT_SYMBOL(_local_bh_enable); |
162 | |
163 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
164 | { |
165 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
166 | #ifdef CONFIG_TRACE_IRQFLAGS |
167 | local_irq_disable(); |
168 | #endif |
169 | /* |
170 | * Are softirqs going to be turned on now: |
171 | */ |
172 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
173 | trace_softirqs_on(ip); |
174 | /* |
175 | * Keep preemption disabled until we are done with |
176 | * softirq processing: |
177 | */ |
178 | preempt_count_sub(cnt - 1); |
179 | |
180 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
181 | /* |
182 | * Run softirq if any pending. And do it in its own stack |
183 | * as we may be calling this deep in a task call stack already. |
184 | */ |
185 | do_softirq(); |
186 | } |
187 | |
188 | preempt_count_dec(); |
189 | #ifdef CONFIG_TRACE_IRQFLAGS |
190 | local_irq_enable(); |
191 | #endif |
192 | preempt_check_resched(); |
193 | } |
194 | EXPORT_SYMBOL(__local_bh_enable_ip); |
195 | |
196 | /* |
197 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
198 | * but break the loop if need_resched() is set or after 2 ms. |
199 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in |
200 | * certain cases, such as stop_machine(), jiffies may cease to |
201 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as |
202 | * well to make sure we eventually return from this method. |
203 | * |
204 | * These limits have been established via experimentation. |
205 | * The two things to balance is latency against fairness - |
206 | * we want to handle softirqs as soon as possible, but they |
207 | * should not be able to lock up the box. |
208 | */ |
209 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
210 | #define MAX_SOFTIRQ_RESTART 10 |
211 | |
212 | #ifdef CONFIG_TRACE_IRQFLAGS |
213 | /* |
214 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
215 | * to keep the lockdep irq context tracking as tight as possible in order to |
216 | * not miss-qualify lock contexts and miss possible deadlocks. |
217 | */ |
218 | |
219 | static inline bool lockdep_softirq_start(void) |
220 | { |
221 | bool in_hardirq = false; |
222 | |
223 | if (trace_hardirq_context(current)) { |
224 | in_hardirq = true; |
225 | trace_hardirq_exit(); |
226 | } |
227 | |
228 | lockdep_softirq_enter(); |
229 | |
230 | return in_hardirq; |
231 | } |
232 | |
233 | static inline void lockdep_softirq_end(bool in_hardirq) |
234 | { |
235 | lockdep_softirq_exit(); |
236 | |
237 | if (in_hardirq) |
238 | trace_hardirq_enter(); |
239 | } |
240 | #else |
241 | static inline bool lockdep_softirq_start(void) { return false; } |
242 | static inline void lockdep_softirq_end(bool in_hardirq) { } |
243 | #endif |
244 | |
245 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
246 | { |
247 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
248 | unsigned long old_flags = current->flags; |
249 | int max_restart = MAX_SOFTIRQ_RESTART; |
250 | struct softirq_action *h; |
251 | bool in_hardirq; |
252 | __u32 pending; |
253 | int softirq_bit; |
254 | #ifdef CONFIG_AMLOGIC_DEBUG_LOCKUP |
255 | int cpu; |
256 | unsigned long long tin; |
257 | #endif |
258 | |
259 | /* |
260 | * Mask out PF_MEMALLOC s current task context is borrowed for the |
261 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC |
262 | * again if the socket is related to swap |
263 | */ |
264 | current->flags &= ~PF_MEMALLOC; |
265 | |
266 | pending = local_softirq_pending(); |
267 | account_irq_enter_time(current); |
268 | |
269 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
270 | in_hardirq = lockdep_softirq_start(); |
271 | |
272 | restart: |
273 | /* Reset the pending bitmask before enabling irqs */ |
274 | set_softirq_pending(0); |
275 | |
276 | local_irq_enable(); |
277 | |
278 | h = softirq_vec; |
279 | |
280 | while ((softirq_bit = ffs(pending))) { |
281 | unsigned int vec_nr; |
282 | int prev_count; |
283 | |
284 | h += softirq_bit - 1; |
285 | |
286 | vec_nr = h - softirq_vec; |
287 | prev_count = preempt_count(); |
288 | |
289 | kstat_incr_softirqs_this_cpu(vec_nr); |
290 | |
291 | trace_softirq_entry(vec_nr); |
292 | #ifdef CONFIG_AMLOGIC_DEBUG_LOCKUP |
293 | cpu = smp_processor_id(); |
294 | sirq_in_hook(cpu, &tin, (void *)h->action); |
295 | #endif |
296 | h->action(h); |
297 | #ifdef CONFIG_AMLOGIC_DEBUG_LOCKUP |
298 | sirq_out_hook(cpu, tin, (void *)h->action); |
299 | #endif |
300 | trace_softirq_exit(vec_nr); |
301 | if (unlikely(prev_count != preempt_count())) { |
302 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
303 | vec_nr, softirq_to_name[vec_nr], h->action, |
304 | prev_count, preempt_count()); |
305 | preempt_count_set(prev_count); |
306 | } |
307 | h++; |
308 | pending >>= softirq_bit; |
309 | } |
310 | |
311 | rcu_bh_qs(); |
312 | local_irq_disable(); |
313 | |
314 | pending = local_softirq_pending(); |
315 | if (pending) { |
316 | if (time_before(jiffies, end) && !need_resched() && |
317 | --max_restart) |
318 | goto restart; |
319 | |
320 | wakeup_softirqd(); |
321 | } |
322 | |
323 | lockdep_softirq_end(in_hardirq); |
324 | account_irq_exit_time(current); |
325 | __local_bh_enable(SOFTIRQ_OFFSET); |
326 | WARN_ON_ONCE(in_interrupt()); |
327 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
328 | } |
329 | |
330 | asmlinkage __visible void do_softirq(void) |
331 | { |
332 | __u32 pending; |
333 | unsigned long flags; |
334 | |
335 | if (in_interrupt()) |
336 | return; |
337 | |
338 | local_irq_save(flags); |
339 | |
340 | pending = local_softirq_pending(); |
341 | |
342 | if (pending && !ksoftirqd_running(pending)) |
343 | do_softirq_own_stack(); |
344 | |
345 | local_irq_restore(flags); |
346 | } |
347 | |
348 | /* |
349 | * Enter an interrupt context. |
350 | */ |
351 | void irq_enter(void) |
352 | { |
353 | rcu_irq_enter(); |
354 | if (is_idle_task(current) && !in_interrupt()) { |
355 | /* |
356 | * Prevent raise_softirq from needlessly waking up ksoftirqd |
357 | * here, as softirq will be serviced on return from interrupt. |
358 | */ |
359 | local_bh_disable(); |
360 | tick_irq_enter(); |
361 | _local_bh_enable(); |
362 | } |
363 | |
364 | __irq_enter(); |
365 | } |
366 | |
367 | static inline void invoke_softirq(void) |
368 | { |
369 | if (ksoftirqd_running(local_softirq_pending())) |
370 | return; |
371 | |
372 | if (!force_irqthreads) { |
373 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
374 | /* |
375 | * We can safely execute softirq on the current stack if |
376 | * it is the irq stack, because it should be near empty |
377 | * at this stage. |
378 | */ |
379 | __do_softirq(); |
380 | #else |
381 | /* |
382 | * Otherwise, irq_exit() is called on the task stack that can |
383 | * be potentially deep already. So call softirq in its own stack |
384 | * to prevent from any overrun. |
385 | */ |
386 | do_softirq_own_stack(); |
387 | #endif |
388 | } else { |
389 | wakeup_softirqd(); |
390 | } |
391 | } |
392 | |
393 | static inline void tick_irq_exit(void) |
394 | { |
395 | #ifdef CONFIG_NO_HZ_COMMON |
396 | int cpu = smp_processor_id(); |
397 | |
398 | /* Make sure that timer wheel updates are propagated */ |
399 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { |
400 | if (!in_interrupt()) |
401 | tick_nohz_irq_exit(); |
402 | } |
403 | #endif |
404 | } |
405 | |
406 | /* |
407 | * Exit an interrupt context. Process softirqs if needed and possible: |
408 | */ |
409 | void irq_exit(void) |
410 | { |
411 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
412 | local_irq_disable(); |
413 | #else |
414 | WARN_ON_ONCE(!irqs_disabled()); |
415 | #endif |
416 | |
417 | account_irq_exit_time(current); |
418 | preempt_count_sub(HARDIRQ_OFFSET); |
419 | if (!in_interrupt() && local_softirq_pending()) |
420 | invoke_softirq(); |
421 | |
422 | tick_irq_exit(); |
423 | rcu_irq_exit(); |
424 | trace_hardirq_exit(); /* must be last! */ |
425 | } |
426 | |
427 | /* |
428 | * This function must run with irqs disabled! |
429 | */ |
430 | inline void raise_softirq_irqoff(unsigned int nr) |
431 | { |
432 | __raise_softirq_irqoff(nr); |
433 | |
434 | /* |
435 | * If we're in an interrupt or softirq, we're done |
436 | * (this also catches softirq-disabled code). We will |
437 | * actually run the softirq once we return from |
438 | * the irq or softirq. |
439 | * |
440 | * Otherwise we wake up ksoftirqd to make sure we |
441 | * schedule the softirq soon. |
442 | */ |
443 | if (!in_interrupt()) |
444 | wakeup_softirqd(); |
445 | } |
446 | |
447 | void raise_softirq(unsigned int nr) |
448 | { |
449 | unsigned long flags; |
450 | |
451 | local_irq_save(flags); |
452 | raise_softirq_irqoff(nr); |
453 | local_irq_restore(flags); |
454 | } |
455 | |
456 | void __raise_softirq_irqoff(unsigned int nr) |
457 | { |
458 | trace_softirq_raise(nr); |
459 | or_softirq_pending(1UL << nr); |
460 | } |
461 | |
462 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
463 | { |
464 | softirq_vec[nr].action = action; |
465 | } |
466 | |
467 | /* |
468 | * Tasklets |
469 | */ |
470 | struct tasklet_head { |
471 | struct tasklet_struct *head; |
472 | struct tasklet_struct **tail; |
473 | }; |
474 | |
475 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
476 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); |
477 | |
478 | void __tasklet_schedule(struct tasklet_struct *t) |
479 | { |
480 | unsigned long flags; |
481 | |
482 | local_irq_save(flags); |
483 | t->next = NULL; |
484 | *__this_cpu_read(tasklet_vec.tail) = t; |
485 | __this_cpu_write(tasklet_vec.tail, &(t->next)); |
486 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
487 | local_irq_restore(flags); |
488 | } |
489 | EXPORT_SYMBOL(__tasklet_schedule); |
490 | |
491 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
492 | { |
493 | unsigned long flags; |
494 | |
495 | local_irq_save(flags); |
496 | t->next = NULL; |
497 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
498 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); |
499 | raise_softirq_irqoff(HI_SOFTIRQ); |
500 | local_irq_restore(flags); |
501 | } |
502 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
503 | |
504 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
505 | { |
506 | BUG_ON(!irqs_disabled()); |
507 | |
508 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
509 | __this_cpu_write(tasklet_hi_vec.head, t); |
510 | __raise_softirq_irqoff(HI_SOFTIRQ); |
511 | } |
512 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); |
513 | |
514 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
515 | { |
516 | struct tasklet_struct *list; |
517 | |
518 | local_irq_disable(); |
519 | list = __this_cpu_read(tasklet_vec.head); |
520 | __this_cpu_write(tasklet_vec.head, NULL); |
521 | __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head)); |
522 | local_irq_enable(); |
523 | |
524 | while (list) { |
525 | struct tasklet_struct *t = list; |
526 | |
527 | list = list->next; |
528 | |
529 | if (tasklet_trylock(t)) { |
530 | if (!atomic_read(&t->count)) { |
531 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
532 | &t->state)) |
533 | BUG(); |
534 | t->func(t->data); |
535 | tasklet_unlock(t); |
536 | continue; |
537 | } |
538 | tasklet_unlock(t); |
539 | } |
540 | |
541 | local_irq_disable(); |
542 | t->next = NULL; |
543 | *__this_cpu_read(tasklet_vec.tail) = t; |
544 | __this_cpu_write(tasklet_vec.tail, &(t->next)); |
545 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
546 | local_irq_enable(); |
547 | } |
548 | } |
549 | |
550 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
551 | { |
552 | struct tasklet_struct *list; |
553 | |
554 | local_irq_disable(); |
555 | list = __this_cpu_read(tasklet_hi_vec.head); |
556 | __this_cpu_write(tasklet_hi_vec.head, NULL); |
557 | __this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head)); |
558 | local_irq_enable(); |
559 | |
560 | while (list) { |
561 | struct tasklet_struct *t = list; |
562 | |
563 | list = list->next; |
564 | |
565 | if (tasklet_trylock(t)) { |
566 | if (!atomic_read(&t->count)) { |
567 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
568 | &t->state)) |
569 | BUG(); |
570 | t->func(t->data); |
571 | tasklet_unlock(t); |
572 | continue; |
573 | } |
574 | tasklet_unlock(t); |
575 | } |
576 | |
577 | local_irq_disable(); |
578 | t->next = NULL; |
579 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
580 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); |
581 | __raise_softirq_irqoff(HI_SOFTIRQ); |
582 | local_irq_enable(); |
583 | } |
584 | } |
585 | |
586 | void tasklet_init(struct tasklet_struct *t, |
587 | void (*func)(unsigned long), unsigned long data) |
588 | { |
589 | t->next = NULL; |
590 | t->state = 0; |
591 | atomic_set(&t->count, 0); |
592 | t->func = func; |
593 | t->data = data; |
594 | } |
595 | EXPORT_SYMBOL(tasklet_init); |
596 | |
597 | void tasklet_kill(struct tasklet_struct *t) |
598 | { |
599 | if (in_interrupt()) |
600 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
601 | |
602 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { |
603 | do { |
604 | yield(); |
605 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
606 | } |
607 | tasklet_unlock_wait(t); |
608 | clear_bit(TASKLET_STATE_SCHED, &t->state); |
609 | } |
610 | EXPORT_SYMBOL(tasklet_kill); |
611 | |
612 | /* |
613 | * tasklet_hrtimer |
614 | */ |
615 | |
616 | /* |
617 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
618 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended |
619 | * hrtimer callback, but from softirq context. |
620 | */ |
621 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) |
622 | { |
623 | struct tasklet_hrtimer *ttimer = |
624 | container_of(timer, struct tasklet_hrtimer, timer); |
625 | |
626 | tasklet_hi_schedule(&ttimer->tasklet); |
627 | return HRTIMER_NORESTART; |
628 | } |
629 | |
630 | /* |
631 | * Helper function which calls the hrtimer callback from |
632 | * tasklet/softirq context |
633 | */ |
634 | static void __tasklet_hrtimer_trampoline(unsigned long data) |
635 | { |
636 | struct tasklet_hrtimer *ttimer = (void *)data; |
637 | enum hrtimer_restart restart; |
638 | |
639 | restart = ttimer->function(&ttimer->timer); |
640 | if (restart != HRTIMER_NORESTART) |
641 | hrtimer_restart(&ttimer->timer); |
642 | } |
643 | |
644 | /** |
645 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks |
646 | * @ttimer: tasklet_hrtimer which is initialized |
647 | * @function: hrtimer callback function which gets called from softirq context |
648 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
649 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) |
650 | */ |
651 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, |
652 | enum hrtimer_restart (*function)(struct hrtimer *), |
653 | clockid_t which_clock, enum hrtimer_mode mode) |
654 | { |
655 | hrtimer_init(&ttimer->timer, which_clock, mode); |
656 | ttimer->timer.function = __hrtimer_tasklet_trampoline; |
657 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, |
658 | (unsigned long)ttimer); |
659 | ttimer->function = function; |
660 | } |
661 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); |
662 | |
663 | void __init softirq_init(void) |
664 | { |
665 | int cpu; |
666 | |
667 | for_each_possible_cpu(cpu) { |
668 | per_cpu(tasklet_vec, cpu).tail = |
669 | &per_cpu(tasklet_vec, cpu).head; |
670 | per_cpu(tasklet_hi_vec, cpu).tail = |
671 | &per_cpu(tasklet_hi_vec, cpu).head; |
672 | } |
673 | |
674 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
675 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); |
676 | } |
677 | |
678 | static int ksoftirqd_should_run(unsigned int cpu) |
679 | { |
680 | return local_softirq_pending(); |
681 | } |
682 | |
683 | static void run_ksoftirqd(unsigned int cpu) |
684 | { |
685 | local_irq_disable(); |
686 | if (local_softirq_pending()) { |
687 | /* |
688 | * We can safely run softirq on inline stack, as we are not deep |
689 | * in the task stack here. |
690 | */ |
691 | __do_softirq(); |
692 | local_irq_enable(); |
693 | cond_resched_rcu_qs(); |
694 | return; |
695 | } |
696 | local_irq_enable(); |
697 | } |
698 | |
699 | #ifdef CONFIG_HOTPLUG_CPU |
700 | /* |
701 | * tasklet_kill_immediate is called to remove a tasklet which can already be |
702 | * scheduled for execution on @cpu. |
703 | * |
704 | * Unlike tasklet_kill, this function removes the tasklet |
705 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. |
706 | * |
707 | * When this function is called, @cpu must be in the CPU_DEAD state. |
708 | */ |
709 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) |
710 | { |
711 | struct tasklet_struct **i; |
712 | |
713 | BUG_ON(cpu_online(cpu)); |
714 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); |
715 | |
716 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) |
717 | return; |
718 | |
719 | /* CPU is dead, so no lock needed. */ |
720 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
721 | if (*i == t) { |
722 | *i = t->next; |
723 | /* If this was the tail element, move the tail ptr */ |
724 | if (*i == NULL) |
725 | per_cpu(tasklet_vec, cpu).tail = i; |
726 | return; |
727 | } |
728 | } |
729 | BUG(); |
730 | } |
731 | |
732 | static int takeover_tasklets(unsigned int cpu) |
733 | { |
734 | /* CPU is dead, so no lock needed. */ |
735 | local_irq_disable(); |
736 | |
737 | /* Find end, append list for that CPU. */ |
738 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
739 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
740 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); |
741 | per_cpu(tasklet_vec, cpu).head = NULL; |
742 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; |
743 | } |
744 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
745 | |
746 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
747 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
748 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); |
749 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
750 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; |
751 | } |
752 | raise_softirq_irqoff(HI_SOFTIRQ); |
753 | |
754 | local_irq_enable(); |
755 | return 0; |
756 | } |
757 | #else |
758 | #define takeover_tasklets NULL |
759 | #endif /* CONFIG_HOTPLUG_CPU */ |
760 | |
761 | static struct smp_hotplug_thread softirq_threads = { |
762 | .store = &ksoftirqd, |
763 | .thread_should_run = ksoftirqd_should_run, |
764 | .thread_fn = run_ksoftirqd, |
765 | .thread_comm = "ksoftirqd/%u", |
766 | }; |
767 | |
768 | static __init int spawn_ksoftirqd(void) |
769 | { |
770 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
771 | takeover_tasklets); |
772 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
773 | |
774 | return 0; |
775 | } |
776 | early_initcall(spawn_ksoftirqd); |
777 | |
778 | /* |
779 | * [ These __weak aliases are kept in a separate compilation unit, so that |
780 | * GCC does not inline them incorrectly. ] |
781 | */ |
782 | |
783 | int __init __weak early_irq_init(void) |
784 | { |
785 | return 0; |
786 | } |
787 | |
788 | int __init __weak arch_probe_nr_irqs(void) |
789 | { |
790 | return NR_IRQS_LEGACY; |
791 | } |
792 | |
793 | int __init __weak arch_early_irq_init(void) |
794 | { |
795 | return 0; |
796 | } |
797 | |
798 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) |
799 | { |
800 | return from; |
801 | } |
802 |