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> |
1da177e4 LT |
27 | |
28 | #include <asm/irq.h> | |
29 | /* | |
30 | - No shared variables, all the data are CPU local. | |
31 | - If a softirq needs serialization, let it serialize itself | |
32 | by its own spinlocks. | |
33 | - Even if softirq is serialized, only local cpu is marked for | |
34 | execution. Hence, we get something sort of weak cpu binding. | |
35 | Though it is still not clear, will it result in better locality | |
36 | or will not. | |
37 | ||
38 | Examples: | |
39 | - NET RX softirq. It is multithreaded and does not require | |
40 | any global serialization. | |
41 | - NET TX softirq. It kicks software netdevice queues, hence | |
42 | it is logically serialized per device, but this serialization | |
43 | is invisible to common code. | |
44 | - Tasklets: serialized wrt itself. | |
45 | */ | |
46 | ||
47 | #ifndef __ARCH_IRQ_STAT | |
48 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
49 | EXPORT_SYMBOL(irq_stat); | |
50 | #endif | |
51 | ||
978b0116 | 52 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 LT |
53 | |
54 | static DEFINE_PER_CPU(struct task_struct *, ksoftirqd); | |
55 | ||
5d592b44 JB |
56 | char *softirq_to_name[NR_SOFTIRQS] = { |
57 | "HI_SOFTIRQ", "TIMER_SOFTIRQ", "NET_TX_SOFTIRQ", "NET_RX_SOFTIRQ", | |
58 | "BLOCK_SOFTIRQ", "TASKLET_SOFTIRQ", "SCHED_SOFTIRQ", "HRTIMER_SOFTIRQ", | |
59 | "RCU_SOFTIRQ" | |
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 | */ | |
68 | static inline void wakeup_softirqd(void) | |
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 | ||
189 | asmlinkage void __do_softirq(void) | |
190 | { | |
191 | struct softirq_action *h; | |
192 | __u32 pending; | |
193 | int max_restart = MAX_SOFTIRQ_RESTART; | |
194 | int cpu; | |
195 | ||
196 | pending = local_softirq_pending(); | |
829035fd PM |
197 | account_system_vtime(current); |
198 | ||
de30a2b3 | 199 | __local_bh_disable((unsigned long)__builtin_return_address(0)); |
d820ac4c | 200 | lockdep_softirq_enter(); |
1da177e4 | 201 | |
1da177e4 LT |
202 | cpu = smp_processor_id(); |
203 | restart: | |
204 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 205 | set_softirq_pending(0); |
1da177e4 | 206 | |
c70f5d66 | 207 | local_irq_enable(); |
1da177e4 LT |
208 | |
209 | h = softirq_vec; | |
210 | ||
211 | do { | |
212 | if (pending & 1) { | |
8e85b4b5 TG |
213 | int prev_count = preempt_count(); |
214 | ||
1da177e4 | 215 | h->action(h); |
8e85b4b5 TG |
216 | |
217 | if (unlikely(prev_count != preempt_count())) { | |
5d592b44 | 218 | printk(KERN_ERR "huh, entered softirq %td %s %p" |
8e85b4b5 TG |
219 | "with preempt_count %08x," |
220 | " exited with %08x?\n", h - softirq_vec, | |
5d592b44 | 221 | softirq_to_name[h - softirq_vec], |
8e85b4b5 TG |
222 | h->action, prev_count, preempt_count()); |
223 | preempt_count() = prev_count; | |
224 | } | |
225 | ||
1da177e4 LT |
226 | rcu_bh_qsctr_inc(cpu); |
227 | } | |
228 | h++; | |
229 | pending >>= 1; | |
230 | } while (pending); | |
231 | ||
c70f5d66 | 232 | local_irq_disable(); |
1da177e4 LT |
233 | |
234 | pending = local_softirq_pending(); | |
235 | if (pending && --max_restart) | |
236 | goto restart; | |
237 | ||
238 | if (pending) | |
239 | wakeup_softirqd(); | |
240 | ||
d820ac4c | 241 | lockdep_softirq_exit(); |
829035fd PM |
242 | |
243 | account_system_vtime(current); | |
de30a2b3 | 244 | _local_bh_enable(); |
1da177e4 LT |
245 | } |
246 | ||
247 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
248 | ||
249 | asmlinkage void do_softirq(void) | |
250 | { | |
251 | __u32 pending; | |
252 | unsigned long flags; | |
253 | ||
254 | if (in_interrupt()) | |
255 | return; | |
256 | ||
257 | local_irq_save(flags); | |
258 | ||
259 | pending = local_softirq_pending(); | |
260 | ||
261 | if (pending) | |
262 | __do_softirq(); | |
263 | ||
264 | local_irq_restore(flags); | |
265 | } | |
266 | ||
1da177e4 LT |
267 | #endif |
268 | ||
dde4b2b5 IM |
269 | /* |
270 | * Enter an interrupt context. | |
271 | */ | |
272 | void irq_enter(void) | |
273 | { | |
6378ddb5 | 274 | int cpu = smp_processor_id(); |
719254fa | 275 | |
64db4cff | 276 | rcu_irq_enter(); |
ee5f80a9 TG |
277 | if (idle_cpu(cpu) && !in_interrupt()) { |
278 | __irq_enter(); | |
719254fa | 279 | tick_check_idle(cpu); |
ee5f80a9 TG |
280 | } else |
281 | __irq_enter(); | |
dde4b2b5 IM |
282 | } |
283 | ||
1da177e4 LT |
284 | #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED |
285 | # define invoke_softirq() __do_softirq() | |
286 | #else | |
287 | # define invoke_softirq() do_softirq() | |
288 | #endif | |
289 | ||
290 | /* | |
291 | * Exit an interrupt context. Process softirqs if needed and possible: | |
292 | */ | |
293 | void irq_exit(void) | |
294 | { | |
295 | account_system_vtime(current); | |
de30a2b3 | 296 | trace_hardirq_exit(); |
1da177e4 LT |
297 | sub_preempt_count(IRQ_EXIT_OFFSET); |
298 | if (!in_interrupt() && local_softirq_pending()) | |
299 | invoke_softirq(); | |
79bf2bb3 TG |
300 | |
301 | #ifdef CONFIG_NO_HZ | |
302 | /* Make sure that timer wheel updates are propagated */ | |
2232c2d8 | 303 | rcu_irq_exit(); |
64db4cff PM |
304 | if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched()) |
305 | tick_nohz_stop_sched_tick(0); | |
79bf2bb3 | 306 | #endif |
1da177e4 LT |
307 | preempt_enable_no_resched(); |
308 | } | |
309 | ||
310 | /* | |
311 | * This function must run with irqs disabled! | |
312 | */ | |
7ad5b3a5 | 313 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
314 | { |
315 | __raise_softirq_irqoff(nr); | |
316 | ||
317 | /* | |
318 | * If we're in an interrupt or softirq, we're done | |
319 | * (this also catches softirq-disabled code). We will | |
320 | * actually run the softirq once we return from | |
321 | * the irq or softirq. | |
322 | * | |
323 | * Otherwise we wake up ksoftirqd to make sure we | |
324 | * schedule the softirq soon. | |
325 | */ | |
326 | if (!in_interrupt()) | |
327 | wakeup_softirqd(); | |
328 | } | |
329 | ||
7ad5b3a5 | 330 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
331 | { |
332 | unsigned long flags; | |
333 | ||
334 | local_irq_save(flags); | |
335 | raise_softirq_irqoff(nr); | |
336 | local_irq_restore(flags); | |
337 | } | |
338 | ||
962cf36c | 339 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 340 | { |
1da177e4 LT |
341 | softirq_vec[nr].action = action; |
342 | } | |
343 | ||
1da177e4 LT |
344 | /* Tasklets */ |
345 | struct tasklet_head | |
346 | { | |
48f20a9a OJ |
347 | struct tasklet_struct *head; |
348 | struct tasklet_struct **tail; | |
1da177e4 LT |
349 | }; |
350 | ||
4620b49f VN |
351 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
352 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 353 | |
7ad5b3a5 | 354 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
355 | { |
356 | unsigned long flags; | |
357 | ||
358 | local_irq_save(flags); | |
48f20a9a OJ |
359 | t->next = NULL; |
360 | *__get_cpu_var(tasklet_vec).tail = t; | |
361 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
362 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
363 | local_irq_restore(flags); | |
364 | } | |
365 | ||
366 | EXPORT_SYMBOL(__tasklet_schedule); | |
367 | ||
7ad5b3a5 | 368 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
369 | { |
370 | unsigned long flags; | |
371 | ||
372 | local_irq_save(flags); | |
48f20a9a OJ |
373 | t->next = NULL; |
374 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
375 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
376 | raise_softirq_irqoff(HI_SOFTIRQ); |
377 | local_irq_restore(flags); | |
378 | } | |
379 | ||
380 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
381 | ||
382 | static void tasklet_action(struct softirq_action *a) | |
383 | { | |
384 | struct tasklet_struct *list; | |
385 | ||
386 | local_irq_disable(); | |
48f20a9a OJ |
387 | list = __get_cpu_var(tasklet_vec).head; |
388 | __get_cpu_var(tasklet_vec).head = NULL; | |
389 | __get_cpu_var(tasklet_vec).tail = &__get_cpu_var(tasklet_vec).head; | |
1da177e4 LT |
390 | local_irq_enable(); |
391 | ||
392 | while (list) { | |
393 | struct tasklet_struct *t = list; | |
394 | ||
395 | list = list->next; | |
396 | ||
397 | if (tasklet_trylock(t)) { | |
398 | if (!atomic_read(&t->count)) { | |
399 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
400 | BUG(); | |
401 | t->func(t->data); | |
402 | tasklet_unlock(t); | |
403 | continue; | |
404 | } | |
405 | tasklet_unlock(t); | |
406 | } | |
407 | ||
408 | local_irq_disable(); | |
48f20a9a OJ |
409 | t->next = NULL; |
410 | *__get_cpu_var(tasklet_vec).tail = t; | |
411 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
412 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
413 | local_irq_enable(); | |
414 | } | |
415 | } | |
416 | ||
417 | static void tasklet_hi_action(struct softirq_action *a) | |
418 | { | |
419 | struct tasklet_struct *list; | |
420 | ||
421 | local_irq_disable(); | |
48f20a9a OJ |
422 | list = __get_cpu_var(tasklet_hi_vec).head; |
423 | __get_cpu_var(tasklet_hi_vec).head = NULL; | |
424 | __get_cpu_var(tasklet_hi_vec).tail = &__get_cpu_var(tasklet_hi_vec).head; | |
1da177e4 LT |
425 | local_irq_enable(); |
426 | ||
427 | while (list) { | |
428 | struct tasklet_struct *t = list; | |
429 | ||
430 | list = list->next; | |
431 | ||
432 | if (tasklet_trylock(t)) { | |
433 | if (!atomic_read(&t->count)) { | |
434 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
435 | BUG(); | |
436 | t->func(t->data); | |
437 | tasklet_unlock(t); | |
438 | continue; | |
439 | } | |
440 | tasklet_unlock(t); | |
441 | } | |
442 | ||
443 | local_irq_disable(); | |
48f20a9a OJ |
444 | t->next = NULL; |
445 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
446 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
447 | __raise_softirq_irqoff(HI_SOFTIRQ); |
448 | local_irq_enable(); | |
449 | } | |
450 | } | |
451 | ||
452 | ||
453 | void tasklet_init(struct tasklet_struct *t, | |
454 | void (*func)(unsigned long), unsigned long data) | |
455 | { | |
456 | t->next = NULL; | |
457 | t->state = 0; | |
458 | atomic_set(&t->count, 0); | |
459 | t->func = func; | |
460 | t->data = data; | |
461 | } | |
462 | ||
463 | EXPORT_SYMBOL(tasklet_init); | |
464 | ||
465 | void tasklet_kill(struct tasklet_struct *t) | |
466 | { | |
467 | if (in_interrupt()) | |
468 | printk("Attempt to kill tasklet from interrupt\n"); | |
469 | ||
470 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
471 | do | |
472 | yield(); | |
473 | while (test_bit(TASKLET_STATE_SCHED, &t->state)); | |
474 | } | |
475 | tasklet_unlock_wait(t); | |
476 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
477 | } | |
478 | ||
479 | EXPORT_SYMBOL(tasklet_kill); | |
480 | ||
54514a70 DM |
481 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
482 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
483 | ||
484 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
485 | { | |
486 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
487 | ||
488 | list_add_tail(&cp->list, head); | |
489 | ||
490 | /* Trigger the softirq only if the list was previously empty. */ | |
491 | if (head->next == &cp->list) | |
492 | raise_softirq_irqoff(softirq); | |
493 | } | |
494 | ||
495 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
496 | static void remote_softirq_receive(void *data) | |
497 | { | |
498 | struct call_single_data *cp = data; | |
499 | unsigned long flags; | |
500 | int softirq; | |
501 | ||
502 | softirq = cp->priv; | |
503 | ||
504 | local_irq_save(flags); | |
505 | __local_trigger(cp, softirq); | |
506 | local_irq_restore(flags); | |
507 | } | |
508 | ||
509 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
510 | { | |
511 | if (cpu_online(cpu)) { | |
512 | cp->func = remote_softirq_receive; | |
513 | cp->info = cp; | |
514 | cp->flags = 0; | |
515 | cp->priv = softirq; | |
516 | ||
517 | __smp_call_function_single(cpu, cp); | |
518 | return 0; | |
519 | } | |
520 | return 1; | |
521 | } | |
522 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
523 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
524 | { | |
525 | return 1; | |
526 | } | |
527 | #endif | |
528 | ||
529 | /** | |
530 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
531 | * @cp: private SMP call function data area | |
532 | * @cpu: the remote cpu | |
533 | * @this_cpu: the currently executing cpu | |
534 | * @softirq: the softirq for the work | |
535 | * | |
536 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
537 | * done, the work is instead queued up on the local cpu. | |
538 | * | |
539 | * Interrupts must be disabled. | |
540 | */ | |
541 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
542 | { | |
543 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
544 | __local_trigger(cp, softirq); | |
545 | } | |
546 | EXPORT_SYMBOL(__send_remote_softirq); | |
547 | ||
548 | /** | |
549 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
550 | * @cp: private SMP call function data area | |
551 | * @cpu: the remote cpu | |
552 | * @softirq: the softirq for the work | |
553 | * | |
554 | * Like __send_remote_softirq except that disabling interrupts and | |
555 | * computing the current cpu is done for the caller. | |
556 | */ | |
557 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
558 | { | |
559 | unsigned long flags; | |
560 | int this_cpu; | |
561 | ||
562 | local_irq_save(flags); | |
563 | this_cpu = smp_processor_id(); | |
564 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
565 | local_irq_restore(flags); | |
566 | } | |
567 | EXPORT_SYMBOL(send_remote_softirq); | |
568 | ||
569 | static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self, | |
570 | unsigned long action, void *hcpu) | |
571 | { | |
572 | /* | |
573 | * If a CPU goes away, splice its entries to the current CPU | |
574 | * and trigger a run of the softirq | |
575 | */ | |
576 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
577 | int cpu = (unsigned long) hcpu; | |
578 | int i; | |
579 | ||
580 | local_irq_disable(); | |
581 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
582 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
583 | struct list_head *local_head; | |
584 | ||
585 | if (list_empty(head)) | |
586 | continue; | |
587 | ||
588 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
589 | list_splice_init(head, local_head); | |
590 | raise_softirq_irqoff(i); | |
591 | } | |
592 | local_irq_enable(); | |
593 | } | |
594 | ||
595 | return NOTIFY_OK; | |
596 | } | |
597 | ||
598 | static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = { | |
599 | .notifier_call = remote_softirq_cpu_notify, | |
600 | }; | |
601 | ||
1da177e4 LT |
602 | void __init softirq_init(void) |
603 | { | |
48f20a9a OJ |
604 | int cpu; |
605 | ||
606 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
607 | int i; |
608 | ||
48f20a9a OJ |
609 | per_cpu(tasklet_vec, cpu).tail = |
610 | &per_cpu(tasklet_vec, cpu).head; | |
611 | per_cpu(tasklet_hi_vec, cpu).tail = | |
612 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
613 | for (i = 0; i < NR_SOFTIRQS; i++) |
614 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
615 | } |
616 | ||
54514a70 DM |
617 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
618 | ||
962cf36c CM |
619 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
620 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
621 | } |
622 | ||
623 | static int ksoftirqd(void * __bind_cpu) | |
624 | { | |
1da177e4 LT |
625 | set_current_state(TASK_INTERRUPTIBLE); |
626 | ||
627 | while (!kthread_should_stop()) { | |
628 | preempt_disable(); | |
629 | if (!local_softirq_pending()) { | |
630 | preempt_enable_no_resched(); | |
631 | schedule(); | |
632 | preempt_disable(); | |
633 | } | |
634 | ||
635 | __set_current_state(TASK_RUNNING); | |
636 | ||
637 | while (local_softirq_pending()) { | |
638 | /* Preempt disable stops cpu going offline. | |
639 | If already offline, we'll be on wrong CPU: | |
640 | don't process */ | |
641 | if (cpu_is_offline((long)__bind_cpu)) | |
642 | goto wait_to_die; | |
643 | do_softirq(); | |
644 | preempt_enable_no_resched(); | |
645 | cond_resched(); | |
646 | preempt_disable(); | |
64ca5ab9 | 647 | rcu_qsctr_inc((long)__bind_cpu); |
1da177e4 LT |
648 | } |
649 | preempt_enable(); | |
650 | set_current_state(TASK_INTERRUPTIBLE); | |
651 | } | |
652 | __set_current_state(TASK_RUNNING); | |
653 | return 0; | |
654 | ||
655 | wait_to_die: | |
656 | preempt_enable(); | |
657 | /* Wait for kthread_stop */ | |
658 | set_current_state(TASK_INTERRUPTIBLE); | |
659 | while (!kthread_should_stop()) { | |
660 | schedule(); | |
661 | set_current_state(TASK_INTERRUPTIBLE); | |
662 | } | |
663 | __set_current_state(TASK_RUNNING); | |
664 | return 0; | |
665 | } | |
666 | ||
667 | #ifdef CONFIG_HOTPLUG_CPU | |
668 | /* | |
669 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
670 | * scheduled for execution on @cpu. | |
671 | * | |
672 | * Unlike tasklet_kill, this function removes the tasklet | |
673 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
674 | * | |
675 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
676 | */ | |
677 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
678 | { | |
679 | struct tasklet_struct **i; | |
680 | ||
681 | BUG_ON(cpu_online(cpu)); | |
682 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
683 | ||
684 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
685 | return; | |
686 | ||
687 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 688 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
689 | if (*i == t) { |
690 | *i = t->next; | |
48f20a9a OJ |
691 | /* If this was the tail element, move the tail ptr */ |
692 | if (*i == NULL) | |
693 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
694 | return; |
695 | } | |
696 | } | |
697 | BUG(); | |
698 | } | |
699 | ||
700 | static void takeover_tasklets(unsigned int cpu) | |
701 | { | |
1da177e4 LT |
702 | /* CPU is dead, so no lock needed. */ |
703 | local_irq_disable(); | |
704 | ||
705 | /* Find end, append list for that CPU. */ | |
e5e41723 CB |
706 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
707 | *(__get_cpu_var(tasklet_vec).tail) = per_cpu(tasklet_vec, cpu).head; | |
708 | __get_cpu_var(tasklet_vec).tail = per_cpu(tasklet_vec, cpu).tail; | |
709 | per_cpu(tasklet_vec, cpu).head = NULL; | |
710 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
711 | } | |
1da177e4 LT |
712 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
713 | ||
e5e41723 CB |
714 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
715 | *__get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).head; | |
716 | __get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).tail; | |
717 | per_cpu(tasklet_hi_vec, cpu).head = NULL; | |
718 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
719 | } | |
1da177e4 LT |
720 | raise_softirq_irqoff(HI_SOFTIRQ); |
721 | ||
722 | local_irq_enable(); | |
723 | } | |
724 | #endif /* CONFIG_HOTPLUG_CPU */ | |
725 | ||
8c78f307 | 726 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
727 | unsigned long action, |
728 | void *hcpu) | |
729 | { | |
730 | int hotcpu = (unsigned long)hcpu; | |
731 | struct task_struct *p; | |
732 | ||
733 | switch (action) { | |
734 | case CPU_UP_PREPARE: | |
8bb78442 | 735 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
736 | p = kthread_create(ksoftirqd, hcpu, "ksoftirqd/%d", hotcpu); |
737 | if (IS_ERR(p)) { | |
738 | printk("ksoftirqd for %i failed\n", hotcpu); | |
739 | return NOTIFY_BAD; | |
740 | } | |
741 | kthread_bind(p, hotcpu); | |
742 | per_cpu(ksoftirqd, hotcpu) = p; | |
743 | break; | |
744 | case CPU_ONLINE: | |
8bb78442 | 745 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
746 | wake_up_process(per_cpu(ksoftirqd, hotcpu)); |
747 | break; | |
748 | #ifdef CONFIG_HOTPLUG_CPU | |
749 | case CPU_UP_CANCELED: | |
8bb78442 | 750 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
751 | if (!per_cpu(ksoftirqd, hotcpu)) |
752 | break; | |
1da177e4 | 753 | /* Unbind so it can run. Fall thru. */ |
a4c4af7c | 754 | kthread_bind(per_cpu(ksoftirqd, hotcpu), |
f1fc057c | 755 | cpumask_any(cpu_online_mask)); |
1da177e4 | 756 | case CPU_DEAD: |
1c6b4aa9 ST |
757 | case CPU_DEAD_FROZEN: { |
758 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; | |
759 | ||
1da177e4 LT |
760 | p = per_cpu(ksoftirqd, hotcpu); |
761 | per_cpu(ksoftirqd, hotcpu) = NULL; | |
961ccddd | 762 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
1da177e4 LT |
763 | kthread_stop(p); |
764 | takeover_tasklets(hotcpu); | |
765 | break; | |
1c6b4aa9 | 766 | } |
1da177e4 LT |
767 | #endif /* CONFIG_HOTPLUG_CPU */ |
768 | } | |
769 | return NOTIFY_OK; | |
770 | } | |
771 | ||
8c78f307 | 772 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
773 | .notifier_call = cpu_callback |
774 | }; | |
775 | ||
7babe8db | 776 | static __init int spawn_ksoftirqd(void) |
1da177e4 LT |
777 | { |
778 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 AM |
779 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
780 | ||
781 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
782 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
783 | register_cpu_notifier(&cpu_nfb); | |
784 | return 0; | |
785 | } | |
7babe8db | 786 | early_initcall(spawn_ksoftirqd); |
78eef01b AM |
787 | |
788 | #ifdef CONFIG_SMP | |
789 | /* | |
790 | * Call a function on all processors | |
791 | */ | |
15c8b6c1 | 792 | int on_each_cpu(void (*func) (void *info), void *info, int wait) |
78eef01b AM |
793 | { |
794 | int ret = 0; | |
795 | ||
796 | preempt_disable(); | |
8691e5a8 | 797 | ret = smp_call_function(func, info, wait); |
78eef01b AM |
798 | local_irq_disable(); |
799 | func(info); | |
800 | local_irq_enable(); | |
801 | preempt_enable(); | |
802 | return ret; | |
803 | } | |
804 | EXPORT_SYMBOL(on_each_cpu); | |
805 | #endif | |
43a25632 YL |
806 | |
807 | /* | |
808 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
809 | * GCC does not inline them incorrectly. ] | |
810 | */ | |
811 | ||
812 | int __init __weak early_irq_init(void) | |
813 | { | |
814 | return 0; | |
815 | } | |
816 | ||
4a046d17 YL |
817 | int __init __weak arch_probe_nr_irqs(void) |
818 | { | |
819 | return 0; | |
820 | } | |
821 | ||
43a25632 YL |
822 | int __init __weak arch_early_irq_init(void) |
823 | { | |
824 | return 0; | |
825 | } | |
826 | ||
827 | int __weak arch_init_chip_data(struct irq_desc *desc, int cpu) | |
828 | { | |
829 | return 0; | |
830 | } |