]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/softirq.c | |
3 | * | |
4 | * Copyright (C) 1992 Linus Torvalds | |
5 | * | |
b10db7f0 PM |
6 | * Distribute under GPLv2. |
7 | * | |
8 | * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903) | |
54514a70 DM |
9 | * |
10 | * Remote softirq infrastructure is by Jens Axboe. | |
1da177e4 LT |
11 | */ |
12 | ||
13 | #include <linux/module.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> | |
83144186 | 21 | #include <linux/freezer.h> |
1da177e4 LT |
22 | #include <linux/kthread.h> |
23 | #include <linux/rcupdate.h> | |
7e49fcce | 24 | #include <linux/ftrace.h> |
78eef01b | 25 | #include <linux/smp.h> |
79bf2bb3 | 26 | #include <linux/tick.h> |
39842323 | 27 | #include <trace/irq.h> |
1da177e4 LT |
28 | |
29 | #include <asm/irq.h> | |
30 | /* | |
31 | - No shared variables, all the data are CPU local. | |
32 | - If a softirq needs serialization, let it serialize itself | |
33 | by its own spinlocks. | |
34 | - Even if softirq is serialized, only local cpu is marked for | |
35 | execution. Hence, we get something sort of weak cpu binding. | |
36 | Though it is still not clear, will it result in better locality | |
37 | or will not. | |
38 | ||
39 | Examples: | |
40 | - NET RX softirq. It is multithreaded and does not require | |
41 | any global serialization. | |
42 | - NET TX softirq. It kicks software netdevice queues, hence | |
43 | it is logically serialized per device, but this serialization | |
44 | is invisible to common code. | |
45 | - Tasklets: serialized wrt itself. | |
46 | */ | |
47 | ||
48 | #ifndef __ARCH_IRQ_STAT | |
49 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
50 | EXPORT_SYMBOL(irq_stat); | |
51 | #endif | |
52 | ||
978b0116 | 53 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 LT |
54 | |
55 | static DEFINE_PER_CPU(struct task_struct *, ksoftirqd); | |
56 | ||
5d592b44 | 57 | char *softirq_to_name[NR_SOFTIRQS] = { |
899039e8 SR |
58 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", |
59 | "TASKLET", "SCHED", "HRTIMER", "RCU" | |
5d592b44 JB |
60 | }; |
61 | ||
1da177e4 LT |
62 | /* |
63 | * we cannot loop indefinitely here to avoid userspace starvation, | |
64 | * but we also don't want to introduce a worst case 1/HZ latency | |
65 | * to the pending events, so lets the scheduler to balance | |
66 | * the softirq load for us. | |
67 | */ | |
7f1e2ca9 | 68 | void wakeup_softirqd(void) |
1da177e4 LT |
69 | { |
70 | /* Interrupts are disabled: no need to stop preemption */ | |
71 | struct task_struct *tsk = __get_cpu_var(ksoftirqd); | |
72 | ||
73 | if (tsk && tsk->state != TASK_RUNNING) | |
74 | wake_up_process(tsk); | |
75 | } | |
76 | ||
de30a2b3 IM |
77 | /* |
78 | * This one is for softirq.c-internal use, | |
79 | * where hardirqs are disabled legitimately: | |
80 | */ | |
3c829c36 | 81 | #ifdef CONFIG_TRACE_IRQFLAGS |
de30a2b3 IM |
82 | static void __local_bh_disable(unsigned long ip) |
83 | { | |
84 | unsigned long flags; | |
85 | ||
86 | WARN_ON_ONCE(in_irq()); | |
87 | ||
88 | raw_local_irq_save(flags); | |
7e49fcce SR |
89 | /* |
90 | * The preempt tracer hooks into add_preempt_count and will break | |
91 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET | |
92 | * is set and before current->softirq_enabled is cleared. | |
93 | * We must manually increment preempt_count here and manually | |
94 | * call the trace_preempt_off later. | |
95 | */ | |
96 | preempt_count() += SOFTIRQ_OFFSET; | |
de30a2b3 IM |
97 | /* |
98 | * Were softirqs turned off above: | |
99 | */ | |
100 | if (softirq_count() == SOFTIRQ_OFFSET) | |
101 | trace_softirqs_off(ip); | |
102 | raw_local_irq_restore(flags); | |
7e49fcce SR |
103 | |
104 | if (preempt_count() == SOFTIRQ_OFFSET) | |
105 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
de30a2b3 | 106 | } |
3c829c36 TC |
107 | #else /* !CONFIG_TRACE_IRQFLAGS */ |
108 | static inline void __local_bh_disable(unsigned long ip) | |
109 | { | |
110 | add_preempt_count(SOFTIRQ_OFFSET); | |
111 | barrier(); | |
112 | } | |
113 | #endif /* CONFIG_TRACE_IRQFLAGS */ | |
de30a2b3 IM |
114 | |
115 | void local_bh_disable(void) | |
116 | { | |
117 | __local_bh_disable((unsigned long)__builtin_return_address(0)); | |
118 | } | |
119 | ||
120 | EXPORT_SYMBOL(local_bh_disable); | |
121 | ||
de30a2b3 IM |
122 | /* |
123 | * Special-case - softirqs can safely be enabled in | |
124 | * cond_resched_softirq(), or by __do_softirq(), | |
125 | * without processing still-pending softirqs: | |
126 | */ | |
127 | void _local_bh_enable(void) | |
128 | { | |
129 | WARN_ON_ONCE(in_irq()); | |
130 | WARN_ON_ONCE(!irqs_disabled()); | |
131 | ||
132 | if (softirq_count() == SOFTIRQ_OFFSET) | |
133 | trace_softirqs_on((unsigned long)__builtin_return_address(0)); | |
134 | sub_preempt_count(SOFTIRQ_OFFSET); | |
135 | } | |
136 | ||
137 | EXPORT_SYMBOL(_local_bh_enable); | |
138 | ||
0f476b6d | 139 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 140 | { |
0f476b6d | 141 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 142 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 143 | local_irq_disable(); |
3c829c36 | 144 | #endif |
de30a2b3 IM |
145 | /* |
146 | * Are softirqs going to be turned on now: | |
147 | */ | |
148 | if (softirq_count() == SOFTIRQ_OFFSET) | |
0f476b6d | 149 | trace_softirqs_on(ip); |
de30a2b3 IM |
150 | /* |
151 | * Keep preemption disabled until we are done with | |
152 | * softirq processing: | |
153 | */ | |
154 | sub_preempt_count(SOFTIRQ_OFFSET - 1); | |
155 | ||
156 | if (unlikely(!in_interrupt() && local_softirq_pending())) | |
157 | do_softirq(); | |
158 | ||
159 | dec_preempt_count(); | |
3c829c36 | 160 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 161 | local_irq_enable(); |
3c829c36 | 162 | #endif |
de30a2b3 IM |
163 | preempt_check_resched(); |
164 | } | |
0f476b6d JB |
165 | |
166 | void local_bh_enable(void) | |
167 | { | |
168 | _local_bh_enable_ip((unsigned long)__builtin_return_address(0)); | |
169 | } | |
de30a2b3 IM |
170 | EXPORT_SYMBOL(local_bh_enable); |
171 | ||
172 | void local_bh_enable_ip(unsigned long ip) | |
173 | { | |
0f476b6d | 174 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
175 | } |
176 | EXPORT_SYMBOL(local_bh_enable_ip); | |
177 | ||
1da177e4 LT |
178 | /* |
179 | * We restart softirq processing MAX_SOFTIRQ_RESTART times, | |
180 | * and we fall back to softirqd after that. | |
181 | * | |
182 | * This number has been established via experimentation. | |
183 | * The two things to balance is latency against fairness - | |
184 | * we want to handle softirqs as soon as possible, but they | |
185 | * should not be able to lock up the box. | |
186 | */ | |
187 | #define MAX_SOFTIRQ_RESTART 10 | |
188 | ||
39842323 JB |
189 | DEFINE_TRACE(softirq_entry); |
190 | DEFINE_TRACE(softirq_exit); | |
191 | ||
1da177e4 LT |
192 | asmlinkage void __do_softirq(void) |
193 | { | |
194 | struct softirq_action *h; | |
195 | __u32 pending; | |
196 | int max_restart = MAX_SOFTIRQ_RESTART; | |
197 | int cpu; | |
198 | ||
199 | pending = local_softirq_pending(); | |
829035fd PM |
200 | account_system_vtime(current); |
201 | ||
de30a2b3 | 202 | __local_bh_disable((unsigned long)__builtin_return_address(0)); |
d820ac4c | 203 | lockdep_softirq_enter(); |
1da177e4 | 204 | |
1da177e4 LT |
205 | cpu = smp_processor_id(); |
206 | restart: | |
207 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 208 | set_softirq_pending(0); |
1da177e4 | 209 | |
c70f5d66 | 210 | local_irq_enable(); |
1da177e4 LT |
211 | |
212 | h = softirq_vec; | |
213 | ||
214 | do { | |
215 | if (pending & 1) { | |
8e85b4b5 TG |
216 | int prev_count = preempt_count(); |
217 | ||
39842323 | 218 | trace_softirq_entry(h, softirq_vec); |
1da177e4 | 219 | h->action(h); |
39842323 | 220 | trace_softirq_exit(h, softirq_vec); |
8e85b4b5 | 221 | if (unlikely(prev_count != preempt_count())) { |
5d592b44 | 222 | printk(KERN_ERR "huh, entered softirq %td %s %p" |
8e85b4b5 TG |
223 | "with preempt_count %08x," |
224 | " exited with %08x?\n", h - softirq_vec, | |
5d592b44 | 225 | softirq_to_name[h - softirq_vec], |
8e85b4b5 TG |
226 | h->action, prev_count, preempt_count()); |
227 | preempt_count() = prev_count; | |
228 | } | |
229 | ||
1da177e4 LT |
230 | rcu_bh_qsctr_inc(cpu); |
231 | } | |
232 | h++; | |
233 | pending >>= 1; | |
234 | } while (pending); | |
235 | ||
c70f5d66 | 236 | local_irq_disable(); |
1da177e4 LT |
237 | |
238 | pending = local_softirq_pending(); | |
239 | if (pending && --max_restart) | |
240 | goto restart; | |
241 | ||
242 | if (pending) | |
243 | wakeup_softirqd(); | |
244 | ||
d820ac4c | 245 | lockdep_softirq_exit(); |
829035fd PM |
246 | |
247 | account_system_vtime(current); | |
de30a2b3 | 248 | _local_bh_enable(); |
1da177e4 LT |
249 | } |
250 | ||
251 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
252 | ||
253 | asmlinkage void do_softirq(void) | |
254 | { | |
255 | __u32 pending; | |
256 | unsigned long flags; | |
257 | ||
258 | if (in_interrupt()) | |
259 | return; | |
260 | ||
261 | local_irq_save(flags); | |
262 | ||
263 | pending = local_softirq_pending(); | |
264 | ||
265 | if (pending) | |
266 | __do_softirq(); | |
267 | ||
268 | local_irq_restore(flags); | |
269 | } | |
270 | ||
1da177e4 LT |
271 | #endif |
272 | ||
dde4b2b5 IM |
273 | /* |
274 | * Enter an interrupt context. | |
275 | */ | |
276 | void irq_enter(void) | |
277 | { | |
6378ddb5 | 278 | int cpu = smp_processor_id(); |
719254fa | 279 | |
64db4cff | 280 | rcu_irq_enter(); |
ee5f80a9 TG |
281 | if (idle_cpu(cpu) && !in_interrupt()) { |
282 | __irq_enter(); | |
719254fa | 283 | tick_check_idle(cpu); |
ee5f80a9 TG |
284 | } else |
285 | __irq_enter(); | |
dde4b2b5 IM |
286 | } |
287 | ||
1da177e4 LT |
288 | #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED |
289 | # define invoke_softirq() __do_softirq() | |
290 | #else | |
291 | # define invoke_softirq() do_softirq() | |
292 | #endif | |
293 | ||
294 | /* | |
295 | * Exit an interrupt context. Process softirqs if needed and possible: | |
296 | */ | |
297 | void irq_exit(void) | |
298 | { | |
299 | account_system_vtime(current); | |
de30a2b3 | 300 | trace_hardirq_exit(); |
1da177e4 LT |
301 | sub_preempt_count(IRQ_EXIT_OFFSET); |
302 | if (!in_interrupt() && local_softirq_pending()) | |
303 | invoke_softirq(); | |
79bf2bb3 TG |
304 | |
305 | #ifdef CONFIG_NO_HZ | |
306 | /* Make sure that timer wheel updates are propagated */ | |
2232c2d8 | 307 | rcu_irq_exit(); |
64db4cff PM |
308 | if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched()) |
309 | tick_nohz_stop_sched_tick(0); | |
79bf2bb3 | 310 | #endif |
1da177e4 LT |
311 | preempt_enable_no_resched(); |
312 | } | |
313 | ||
314 | /* | |
315 | * This function must run with irqs disabled! | |
316 | */ | |
7ad5b3a5 | 317 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
318 | { |
319 | __raise_softirq_irqoff(nr); | |
320 | ||
321 | /* | |
322 | * If we're in an interrupt or softirq, we're done | |
323 | * (this also catches softirq-disabled code). We will | |
324 | * actually run the softirq once we return from | |
325 | * the irq or softirq. | |
326 | * | |
327 | * Otherwise we wake up ksoftirqd to make sure we | |
328 | * schedule the softirq soon. | |
329 | */ | |
330 | if (!in_interrupt()) | |
331 | wakeup_softirqd(); | |
332 | } | |
333 | ||
7ad5b3a5 | 334 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
335 | { |
336 | unsigned long flags; | |
337 | ||
338 | local_irq_save(flags); | |
339 | raise_softirq_irqoff(nr); | |
340 | local_irq_restore(flags); | |
341 | } | |
342 | ||
962cf36c | 343 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 344 | { |
1da177e4 LT |
345 | softirq_vec[nr].action = action; |
346 | } | |
347 | ||
1da177e4 LT |
348 | /* Tasklets */ |
349 | struct tasklet_head | |
350 | { | |
48f20a9a OJ |
351 | struct tasklet_struct *head; |
352 | struct tasklet_struct **tail; | |
1da177e4 LT |
353 | }; |
354 | ||
4620b49f VN |
355 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
356 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 357 | |
7ad5b3a5 | 358 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
359 | { |
360 | unsigned long flags; | |
361 | ||
362 | local_irq_save(flags); | |
48f20a9a OJ |
363 | t->next = NULL; |
364 | *__get_cpu_var(tasklet_vec).tail = t; | |
365 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
366 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
367 | local_irq_restore(flags); | |
368 | } | |
369 | ||
370 | EXPORT_SYMBOL(__tasklet_schedule); | |
371 | ||
7ad5b3a5 | 372 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
373 | { |
374 | unsigned long flags; | |
375 | ||
376 | local_irq_save(flags); | |
48f20a9a OJ |
377 | t->next = NULL; |
378 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
379 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
380 | raise_softirq_irqoff(HI_SOFTIRQ); |
381 | local_irq_restore(flags); | |
382 | } | |
383 | ||
384 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
385 | ||
386 | static void tasklet_action(struct softirq_action *a) | |
387 | { | |
388 | struct tasklet_struct *list; | |
389 | ||
390 | local_irq_disable(); | |
48f20a9a OJ |
391 | list = __get_cpu_var(tasklet_vec).head; |
392 | __get_cpu_var(tasklet_vec).head = NULL; | |
393 | __get_cpu_var(tasklet_vec).tail = &__get_cpu_var(tasklet_vec).head; | |
1da177e4 LT |
394 | local_irq_enable(); |
395 | ||
396 | while (list) { | |
397 | struct tasklet_struct *t = list; | |
398 | ||
399 | list = list->next; | |
400 | ||
401 | if (tasklet_trylock(t)) { | |
402 | if (!atomic_read(&t->count)) { | |
403 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
404 | BUG(); | |
405 | t->func(t->data); | |
406 | tasklet_unlock(t); | |
407 | continue; | |
408 | } | |
409 | tasklet_unlock(t); | |
410 | } | |
411 | ||
412 | local_irq_disable(); | |
48f20a9a OJ |
413 | t->next = NULL; |
414 | *__get_cpu_var(tasklet_vec).tail = t; | |
415 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
416 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
417 | local_irq_enable(); | |
418 | } | |
419 | } | |
420 | ||
421 | static void tasklet_hi_action(struct softirq_action *a) | |
422 | { | |
423 | struct tasklet_struct *list; | |
424 | ||
425 | local_irq_disable(); | |
48f20a9a OJ |
426 | list = __get_cpu_var(tasklet_hi_vec).head; |
427 | __get_cpu_var(tasklet_hi_vec).head = NULL; | |
428 | __get_cpu_var(tasklet_hi_vec).tail = &__get_cpu_var(tasklet_hi_vec).head; | |
1da177e4 LT |
429 | local_irq_enable(); |
430 | ||
431 | while (list) { | |
432 | struct tasklet_struct *t = list; | |
433 | ||
434 | list = list->next; | |
435 | ||
436 | if (tasklet_trylock(t)) { | |
437 | if (!atomic_read(&t->count)) { | |
438 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
439 | BUG(); | |
440 | t->func(t->data); | |
441 | tasklet_unlock(t); | |
442 | continue; | |
443 | } | |
444 | tasklet_unlock(t); | |
445 | } | |
446 | ||
447 | local_irq_disable(); | |
48f20a9a OJ |
448 | t->next = NULL; |
449 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
450 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
451 | __raise_softirq_irqoff(HI_SOFTIRQ); |
452 | local_irq_enable(); | |
453 | } | |
454 | } | |
455 | ||
456 | ||
457 | void tasklet_init(struct tasklet_struct *t, | |
458 | void (*func)(unsigned long), unsigned long data) | |
459 | { | |
460 | t->next = NULL; | |
461 | t->state = 0; | |
462 | atomic_set(&t->count, 0); | |
463 | t->func = func; | |
464 | t->data = data; | |
465 | } | |
466 | ||
467 | EXPORT_SYMBOL(tasklet_init); | |
468 | ||
469 | void tasklet_kill(struct tasklet_struct *t) | |
470 | { | |
471 | if (in_interrupt()) | |
472 | printk("Attempt to kill tasklet from interrupt\n"); | |
473 | ||
474 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 475 | do { |
1da177e4 | 476 | yield(); |
79d381c9 | 477 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
478 | } |
479 | tasklet_unlock_wait(t); | |
480 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
481 | } | |
482 | ||
483 | EXPORT_SYMBOL(tasklet_kill); | |
484 | ||
54514a70 DM |
485 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
486 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
487 | ||
488 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
489 | { | |
490 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
491 | ||
492 | list_add_tail(&cp->list, head); | |
493 | ||
494 | /* Trigger the softirq only if the list was previously empty. */ | |
495 | if (head->next == &cp->list) | |
496 | raise_softirq_irqoff(softirq); | |
497 | } | |
498 | ||
499 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
500 | static void remote_softirq_receive(void *data) | |
501 | { | |
502 | struct call_single_data *cp = data; | |
503 | unsigned long flags; | |
504 | int softirq; | |
505 | ||
506 | softirq = cp->priv; | |
507 | ||
508 | local_irq_save(flags); | |
509 | __local_trigger(cp, softirq); | |
510 | local_irq_restore(flags); | |
511 | } | |
512 | ||
513 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
514 | { | |
515 | if (cpu_online(cpu)) { | |
516 | cp->func = remote_softirq_receive; | |
517 | cp->info = cp; | |
518 | cp->flags = 0; | |
519 | cp->priv = softirq; | |
520 | ||
6e275637 | 521 | __smp_call_function_single(cpu, cp, 0); |
54514a70 DM |
522 | return 0; |
523 | } | |
524 | return 1; | |
525 | } | |
526 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
527 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
528 | { | |
529 | return 1; | |
530 | } | |
531 | #endif | |
532 | ||
533 | /** | |
534 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
535 | * @cp: private SMP call function data area | |
536 | * @cpu: the remote cpu | |
537 | * @this_cpu: the currently executing cpu | |
538 | * @softirq: the softirq for the work | |
539 | * | |
540 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
541 | * done, the work is instead queued up on the local cpu. | |
542 | * | |
543 | * Interrupts must be disabled. | |
544 | */ | |
545 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
546 | { | |
547 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
548 | __local_trigger(cp, softirq); | |
549 | } | |
550 | EXPORT_SYMBOL(__send_remote_softirq); | |
551 | ||
552 | /** | |
553 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
554 | * @cp: private SMP call function data area | |
555 | * @cpu: the remote cpu | |
556 | * @softirq: the softirq for the work | |
557 | * | |
558 | * Like __send_remote_softirq except that disabling interrupts and | |
559 | * computing the current cpu is done for the caller. | |
560 | */ | |
561 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
562 | { | |
563 | unsigned long flags; | |
564 | int this_cpu; | |
565 | ||
566 | local_irq_save(flags); | |
567 | this_cpu = smp_processor_id(); | |
568 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
569 | local_irq_restore(flags); | |
570 | } | |
571 | EXPORT_SYMBOL(send_remote_softirq); | |
572 | ||
573 | static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self, | |
574 | unsigned long action, void *hcpu) | |
575 | { | |
576 | /* | |
577 | * If a CPU goes away, splice its entries to the current CPU | |
578 | * and trigger a run of the softirq | |
579 | */ | |
580 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
581 | int cpu = (unsigned long) hcpu; | |
582 | int i; | |
583 | ||
584 | local_irq_disable(); | |
585 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
586 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
587 | struct list_head *local_head; | |
588 | ||
589 | if (list_empty(head)) | |
590 | continue; | |
591 | ||
592 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
593 | list_splice_init(head, local_head); | |
594 | raise_softirq_irqoff(i); | |
595 | } | |
596 | local_irq_enable(); | |
597 | } | |
598 | ||
599 | return NOTIFY_OK; | |
600 | } | |
601 | ||
602 | static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = { | |
603 | .notifier_call = remote_softirq_cpu_notify, | |
604 | }; | |
605 | ||
1da177e4 LT |
606 | void __init softirq_init(void) |
607 | { | |
48f20a9a OJ |
608 | int cpu; |
609 | ||
610 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
611 | int i; |
612 | ||
48f20a9a OJ |
613 | per_cpu(tasklet_vec, cpu).tail = |
614 | &per_cpu(tasklet_vec, cpu).head; | |
615 | per_cpu(tasklet_hi_vec, cpu).tail = | |
616 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
617 | for (i = 0; i < NR_SOFTIRQS; i++) |
618 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
619 | } |
620 | ||
54514a70 DM |
621 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
622 | ||
962cf36c CM |
623 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
624 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
625 | } |
626 | ||
627 | static int ksoftirqd(void * __bind_cpu) | |
628 | { | |
1da177e4 LT |
629 | set_current_state(TASK_INTERRUPTIBLE); |
630 | ||
631 | while (!kthread_should_stop()) { | |
632 | preempt_disable(); | |
633 | if (!local_softirq_pending()) { | |
634 | preempt_enable_no_resched(); | |
635 | schedule(); | |
636 | preempt_disable(); | |
637 | } | |
638 | ||
639 | __set_current_state(TASK_RUNNING); | |
640 | ||
641 | while (local_softirq_pending()) { | |
642 | /* Preempt disable stops cpu going offline. | |
643 | If already offline, we'll be on wrong CPU: | |
644 | don't process */ | |
645 | if (cpu_is_offline((long)__bind_cpu)) | |
646 | goto wait_to_die; | |
647 | do_softirq(); | |
648 | preempt_enable_no_resched(); | |
649 | cond_resched(); | |
650 | preempt_disable(); | |
64ca5ab9 | 651 | rcu_qsctr_inc((long)__bind_cpu); |
1da177e4 LT |
652 | } |
653 | preempt_enable(); | |
654 | set_current_state(TASK_INTERRUPTIBLE); | |
655 | } | |
656 | __set_current_state(TASK_RUNNING); | |
657 | return 0; | |
658 | ||
659 | wait_to_die: | |
660 | preempt_enable(); | |
661 | /* Wait for kthread_stop */ | |
662 | set_current_state(TASK_INTERRUPTIBLE); | |
663 | while (!kthread_should_stop()) { | |
664 | schedule(); | |
665 | set_current_state(TASK_INTERRUPTIBLE); | |
666 | } | |
667 | __set_current_state(TASK_RUNNING); | |
668 | return 0; | |
669 | } | |
670 | ||
671 | #ifdef CONFIG_HOTPLUG_CPU | |
672 | /* | |
673 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
674 | * scheduled for execution on @cpu. | |
675 | * | |
676 | * Unlike tasklet_kill, this function removes the tasklet | |
677 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
678 | * | |
679 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
680 | */ | |
681 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
682 | { | |
683 | struct tasklet_struct **i; | |
684 | ||
685 | BUG_ON(cpu_online(cpu)); | |
686 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
687 | ||
688 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
689 | return; | |
690 | ||
691 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 692 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
693 | if (*i == t) { |
694 | *i = t->next; | |
48f20a9a OJ |
695 | /* If this was the tail element, move the tail ptr */ |
696 | if (*i == NULL) | |
697 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
698 | return; |
699 | } | |
700 | } | |
701 | BUG(); | |
702 | } | |
703 | ||
704 | static void takeover_tasklets(unsigned int cpu) | |
705 | { | |
1da177e4 LT |
706 | /* CPU is dead, so no lock needed. */ |
707 | local_irq_disable(); | |
708 | ||
709 | /* Find end, append list for that CPU. */ | |
e5e41723 CB |
710 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
711 | *(__get_cpu_var(tasklet_vec).tail) = per_cpu(tasklet_vec, cpu).head; | |
712 | __get_cpu_var(tasklet_vec).tail = per_cpu(tasklet_vec, cpu).tail; | |
713 | per_cpu(tasklet_vec, cpu).head = NULL; | |
714 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
715 | } | |
1da177e4 LT |
716 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
717 | ||
e5e41723 CB |
718 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
719 | *__get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).head; | |
720 | __get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).tail; | |
721 | per_cpu(tasklet_hi_vec, cpu).head = NULL; | |
722 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
723 | } | |
1da177e4 LT |
724 | raise_softirq_irqoff(HI_SOFTIRQ); |
725 | ||
726 | local_irq_enable(); | |
727 | } | |
728 | #endif /* CONFIG_HOTPLUG_CPU */ | |
729 | ||
8c78f307 | 730 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
731 | unsigned long action, |
732 | void *hcpu) | |
733 | { | |
734 | int hotcpu = (unsigned long)hcpu; | |
735 | struct task_struct *p; | |
736 | ||
737 | switch (action) { | |
738 | case CPU_UP_PREPARE: | |
8bb78442 | 739 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
740 | p = kthread_create(ksoftirqd, hcpu, "ksoftirqd/%d", hotcpu); |
741 | if (IS_ERR(p)) { | |
742 | printk("ksoftirqd for %i failed\n", hotcpu); | |
743 | return NOTIFY_BAD; | |
744 | } | |
745 | kthread_bind(p, hotcpu); | |
746 | per_cpu(ksoftirqd, hotcpu) = p; | |
747 | break; | |
748 | case CPU_ONLINE: | |
8bb78442 | 749 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
750 | wake_up_process(per_cpu(ksoftirqd, hotcpu)); |
751 | break; | |
752 | #ifdef CONFIG_HOTPLUG_CPU | |
753 | case CPU_UP_CANCELED: | |
8bb78442 | 754 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
755 | if (!per_cpu(ksoftirqd, hotcpu)) |
756 | break; | |
1da177e4 | 757 | /* Unbind so it can run. Fall thru. */ |
a4c4af7c | 758 | kthread_bind(per_cpu(ksoftirqd, hotcpu), |
f1fc057c | 759 | cpumask_any(cpu_online_mask)); |
1da177e4 | 760 | case CPU_DEAD: |
1c6b4aa9 ST |
761 | case CPU_DEAD_FROZEN: { |
762 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; | |
763 | ||
1da177e4 LT |
764 | p = per_cpu(ksoftirqd, hotcpu); |
765 | per_cpu(ksoftirqd, hotcpu) = NULL; | |
961ccddd | 766 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
1da177e4 LT |
767 | kthread_stop(p); |
768 | takeover_tasklets(hotcpu); | |
769 | break; | |
1c6b4aa9 | 770 | } |
1da177e4 LT |
771 | #endif /* CONFIG_HOTPLUG_CPU */ |
772 | } | |
773 | return NOTIFY_OK; | |
774 | } | |
775 | ||
8c78f307 | 776 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
777 | .notifier_call = cpu_callback |
778 | }; | |
779 | ||
7babe8db | 780 | static __init int spawn_ksoftirqd(void) |
1da177e4 LT |
781 | { |
782 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 AM |
783 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
784 | ||
785 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
786 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
787 | register_cpu_notifier(&cpu_nfb); | |
788 | return 0; | |
789 | } | |
7babe8db | 790 | early_initcall(spawn_ksoftirqd); |
78eef01b AM |
791 | |
792 | #ifdef CONFIG_SMP | |
793 | /* | |
794 | * Call a function on all processors | |
795 | */ | |
15c8b6c1 | 796 | int on_each_cpu(void (*func) (void *info), void *info, int wait) |
78eef01b AM |
797 | { |
798 | int ret = 0; | |
799 | ||
800 | preempt_disable(); | |
8691e5a8 | 801 | ret = smp_call_function(func, info, wait); |
78eef01b AM |
802 | local_irq_disable(); |
803 | func(info); | |
804 | local_irq_enable(); | |
805 | preempt_enable(); | |
806 | return ret; | |
807 | } | |
808 | EXPORT_SYMBOL(on_each_cpu); | |
809 | #endif | |
43a25632 YL |
810 | |
811 | /* | |
812 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
813 | * GCC does not inline them incorrectly. ] | |
814 | */ | |
815 | ||
816 | int __init __weak early_irq_init(void) | |
817 | { | |
818 | return 0; | |
819 | } | |
820 | ||
4a046d17 YL |
821 | int __init __weak arch_probe_nr_irqs(void) |
822 | { | |
823 | return 0; | |
824 | } | |
825 | ||
43a25632 YL |
826 | int __init __weak arch_early_irq_init(void) |
827 | { | |
828 | return 0; | |
829 | } | |
830 | ||
831 | int __weak arch_init_chip_data(struct irq_desc *desc, int cpu) | |
832 | { | |
833 | return 0; | |
834 | } |