]>
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> |
a0e39ed3 HC |
27 | |
28 | #define CREATE_TRACE_POINTS | |
ad8d75ff | 29 | #include <trace/events/irq.h> |
1da177e4 LT |
30 | |
31 | #include <asm/irq.h> | |
32 | /* | |
33 | - No shared variables, all the data are CPU local. | |
34 | - If a softirq needs serialization, let it serialize itself | |
35 | by its own spinlocks. | |
36 | - Even if softirq is serialized, only local cpu is marked for | |
37 | execution. Hence, we get something sort of weak cpu binding. | |
38 | Though it is still not clear, will it result in better locality | |
39 | or will not. | |
40 | ||
41 | Examples: | |
42 | - NET RX softirq. It is multithreaded and does not require | |
43 | any global serialization. | |
44 | - NET TX softirq. It kicks software netdevice queues, hence | |
45 | it is logically serialized per device, but this serialization | |
46 | is invisible to common code. | |
47 | - Tasklets: serialized wrt itself. | |
48 | */ | |
49 | ||
50 | #ifndef __ARCH_IRQ_STAT | |
51 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
52 | EXPORT_SYMBOL(irq_stat); | |
53 | #endif | |
54 | ||
978b0116 | 55 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 LT |
56 | |
57 | static DEFINE_PER_CPU(struct task_struct *, ksoftirqd); | |
58 | ||
5d592b44 | 59 | char *softirq_to_name[NR_SOFTIRQS] = { |
899039e8 SR |
60 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", |
61 | "TASKLET", "SCHED", "HRTIMER", "RCU" | |
5d592b44 JB |
62 | }; |
63 | ||
1da177e4 LT |
64 | /* |
65 | * we cannot loop indefinitely here to avoid userspace starvation, | |
66 | * but we also don't want to introduce a worst case 1/HZ latency | |
67 | * to the pending events, so lets the scheduler to balance | |
68 | * the softirq load for us. | |
69 | */ | |
7f1e2ca9 | 70 | void wakeup_softirqd(void) |
1da177e4 LT |
71 | { |
72 | /* Interrupts are disabled: no need to stop preemption */ | |
73 | struct task_struct *tsk = __get_cpu_var(ksoftirqd); | |
74 | ||
75 | if (tsk && tsk->state != TASK_RUNNING) | |
76 | wake_up_process(tsk); | |
77 | } | |
78 | ||
de30a2b3 IM |
79 | /* |
80 | * This one is for softirq.c-internal use, | |
81 | * where hardirqs are disabled legitimately: | |
82 | */ | |
3c829c36 | 83 | #ifdef CONFIG_TRACE_IRQFLAGS |
de30a2b3 IM |
84 | static void __local_bh_disable(unsigned long ip) |
85 | { | |
86 | unsigned long flags; | |
87 | ||
88 | WARN_ON_ONCE(in_irq()); | |
89 | ||
90 | raw_local_irq_save(flags); | |
7e49fcce SR |
91 | /* |
92 | * The preempt tracer hooks into add_preempt_count and will break | |
93 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET | |
94 | * is set and before current->softirq_enabled is cleared. | |
95 | * We must manually increment preempt_count here and manually | |
96 | * call the trace_preempt_off later. | |
97 | */ | |
98 | preempt_count() += SOFTIRQ_OFFSET; | |
de30a2b3 IM |
99 | /* |
100 | * Were softirqs turned off above: | |
101 | */ | |
102 | if (softirq_count() == SOFTIRQ_OFFSET) | |
103 | trace_softirqs_off(ip); | |
104 | raw_local_irq_restore(flags); | |
7e49fcce SR |
105 | |
106 | if (preempt_count() == SOFTIRQ_OFFSET) | |
107 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
de30a2b3 | 108 | } |
3c829c36 TC |
109 | #else /* !CONFIG_TRACE_IRQFLAGS */ |
110 | static inline void __local_bh_disable(unsigned long ip) | |
111 | { | |
112 | add_preempt_count(SOFTIRQ_OFFSET); | |
113 | barrier(); | |
114 | } | |
115 | #endif /* CONFIG_TRACE_IRQFLAGS */ | |
de30a2b3 IM |
116 | |
117 | void local_bh_disable(void) | |
118 | { | |
119 | __local_bh_disable((unsigned long)__builtin_return_address(0)); | |
120 | } | |
121 | ||
122 | EXPORT_SYMBOL(local_bh_disable); | |
123 | ||
de30a2b3 IM |
124 | /* |
125 | * Special-case - softirqs can safely be enabled in | |
126 | * cond_resched_softirq(), or by __do_softirq(), | |
127 | * without processing still-pending softirqs: | |
128 | */ | |
129 | void _local_bh_enable(void) | |
130 | { | |
131 | WARN_ON_ONCE(in_irq()); | |
132 | WARN_ON_ONCE(!irqs_disabled()); | |
133 | ||
134 | if (softirq_count() == SOFTIRQ_OFFSET) | |
135 | trace_softirqs_on((unsigned long)__builtin_return_address(0)); | |
136 | sub_preempt_count(SOFTIRQ_OFFSET); | |
137 | } | |
138 | ||
139 | EXPORT_SYMBOL(_local_bh_enable); | |
140 | ||
0f476b6d | 141 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 142 | { |
0f476b6d | 143 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 144 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 145 | local_irq_disable(); |
3c829c36 | 146 | #endif |
de30a2b3 IM |
147 | /* |
148 | * Are softirqs going to be turned on now: | |
149 | */ | |
150 | if (softirq_count() == SOFTIRQ_OFFSET) | |
0f476b6d | 151 | trace_softirqs_on(ip); |
de30a2b3 IM |
152 | /* |
153 | * Keep preemption disabled until we are done with | |
154 | * softirq processing: | |
155 | */ | |
156 | sub_preempt_count(SOFTIRQ_OFFSET - 1); | |
157 | ||
158 | if (unlikely(!in_interrupt() && local_softirq_pending())) | |
159 | do_softirq(); | |
160 | ||
161 | dec_preempt_count(); | |
3c829c36 | 162 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 163 | local_irq_enable(); |
3c829c36 | 164 | #endif |
de30a2b3 IM |
165 | preempt_check_resched(); |
166 | } | |
0f476b6d JB |
167 | |
168 | void local_bh_enable(void) | |
169 | { | |
170 | _local_bh_enable_ip((unsigned long)__builtin_return_address(0)); | |
171 | } | |
de30a2b3 IM |
172 | EXPORT_SYMBOL(local_bh_enable); |
173 | ||
174 | void local_bh_enable_ip(unsigned long ip) | |
175 | { | |
0f476b6d | 176 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
177 | } |
178 | EXPORT_SYMBOL(local_bh_enable_ip); | |
179 | ||
1da177e4 LT |
180 | /* |
181 | * We restart softirq processing MAX_SOFTIRQ_RESTART times, | |
182 | * and we fall back to softirqd after that. | |
183 | * | |
184 | * This number has been established via experimentation. | |
185 | * The two things to balance is latency against fairness - | |
186 | * we want to handle softirqs as soon as possible, but they | |
187 | * should not be able to lock up the box. | |
188 | */ | |
189 | #define MAX_SOFTIRQ_RESTART 10 | |
190 | ||
191 | asmlinkage void __do_softirq(void) | |
192 | { | |
193 | struct softirq_action *h; | |
194 | __u32 pending; | |
195 | int max_restart = MAX_SOFTIRQ_RESTART; | |
196 | int cpu; | |
197 | ||
198 | pending = local_softirq_pending(); | |
829035fd PM |
199 | account_system_vtime(current); |
200 | ||
de30a2b3 | 201 | __local_bh_disable((unsigned long)__builtin_return_address(0)); |
d820ac4c | 202 | lockdep_softirq_enter(); |
1da177e4 | 203 | |
1da177e4 LT |
204 | cpu = smp_processor_id(); |
205 | restart: | |
206 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 207 | set_softirq_pending(0); |
1da177e4 | 208 | |
c70f5d66 | 209 | local_irq_enable(); |
1da177e4 LT |
210 | |
211 | h = softirq_vec; | |
212 | ||
213 | do { | |
214 | if (pending & 1) { | |
8e85b4b5 TG |
215 | int prev_count = preempt_count(); |
216 | ||
39842323 | 217 | trace_softirq_entry(h, softirq_vec); |
1da177e4 | 218 | h->action(h); |
39842323 | 219 | trace_softirq_exit(h, softirq_vec); |
8e85b4b5 | 220 | if (unlikely(prev_count != preempt_count())) { |
5d592b44 | 221 | printk(KERN_ERR "huh, entered softirq %td %s %p" |
8e85b4b5 TG |
222 | "with preempt_count %08x," |
223 | " exited with %08x?\n", h - softirq_vec, | |
5d592b44 | 224 | softirq_to_name[h - softirq_vec], |
8e85b4b5 TG |
225 | h->action, prev_count, preempt_count()); |
226 | preempt_count() = prev_count; | |
227 | } | |
228 | ||
1da177e4 LT |
229 | rcu_bh_qsctr_inc(cpu); |
230 | } | |
231 | h++; | |
232 | pending >>= 1; | |
233 | } while (pending); | |
234 | ||
c70f5d66 | 235 | local_irq_disable(); |
1da177e4 LT |
236 | |
237 | pending = local_softirq_pending(); | |
238 | if (pending && --max_restart) | |
239 | goto restart; | |
240 | ||
241 | if (pending) | |
242 | wakeup_softirqd(); | |
243 | ||
d820ac4c | 244 | lockdep_softirq_exit(); |
829035fd PM |
245 | |
246 | account_system_vtime(current); | |
de30a2b3 | 247 | _local_bh_enable(); |
1da177e4 LT |
248 | } |
249 | ||
250 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
251 | ||
252 | asmlinkage void do_softirq(void) | |
253 | { | |
254 | __u32 pending; | |
255 | unsigned long flags; | |
256 | ||
257 | if (in_interrupt()) | |
258 | return; | |
259 | ||
260 | local_irq_save(flags); | |
261 | ||
262 | pending = local_softirq_pending(); | |
263 | ||
264 | if (pending) | |
265 | __do_softirq(); | |
266 | ||
267 | local_irq_restore(flags); | |
268 | } | |
269 | ||
1da177e4 LT |
270 | #endif |
271 | ||
dde4b2b5 IM |
272 | /* |
273 | * Enter an interrupt context. | |
274 | */ | |
275 | void irq_enter(void) | |
276 | { | |
6378ddb5 | 277 | int cpu = smp_processor_id(); |
719254fa | 278 | |
64db4cff | 279 | rcu_irq_enter(); |
ee5f80a9 TG |
280 | if (idle_cpu(cpu) && !in_interrupt()) { |
281 | __irq_enter(); | |
719254fa | 282 | tick_check_idle(cpu); |
ee5f80a9 TG |
283 | } else |
284 | __irq_enter(); | |
dde4b2b5 IM |
285 | } |
286 | ||
1da177e4 LT |
287 | #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED |
288 | # define invoke_softirq() __do_softirq() | |
289 | #else | |
290 | # define invoke_softirq() do_softirq() | |
291 | #endif | |
292 | ||
293 | /* | |
294 | * Exit an interrupt context. Process softirqs if needed and possible: | |
295 | */ | |
296 | void irq_exit(void) | |
297 | { | |
298 | account_system_vtime(current); | |
de30a2b3 | 299 | trace_hardirq_exit(); |
1da177e4 LT |
300 | sub_preempt_count(IRQ_EXIT_OFFSET); |
301 | if (!in_interrupt() && local_softirq_pending()) | |
302 | invoke_softirq(); | |
79bf2bb3 TG |
303 | |
304 | #ifdef CONFIG_NO_HZ | |
305 | /* Make sure that timer wheel updates are propagated */ | |
2232c2d8 | 306 | rcu_irq_exit(); |
64db4cff PM |
307 | if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched()) |
308 | tick_nohz_stop_sched_tick(0); | |
79bf2bb3 | 309 | #endif |
1da177e4 LT |
310 | preempt_enable_no_resched(); |
311 | } | |
312 | ||
313 | /* | |
314 | * This function must run with irqs disabled! | |
315 | */ | |
7ad5b3a5 | 316 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
317 | { |
318 | __raise_softirq_irqoff(nr); | |
319 | ||
320 | /* | |
321 | * If we're in an interrupt or softirq, we're done | |
322 | * (this also catches softirq-disabled code). We will | |
323 | * actually run the softirq once we return from | |
324 | * the irq or softirq. | |
325 | * | |
326 | * Otherwise we wake up ksoftirqd to make sure we | |
327 | * schedule the softirq soon. | |
328 | */ | |
329 | if (!in_interrupt()) | |
330 | wakeup_softirqd(); | |
331 | } | |
332 | ||
7ad5b3a5 | 333 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
334 | { |
335 | unsigned long flags; | |
336 | ||
337 | local_irq_save(flags); | |
338 | raise_softirq_irqoff(nr); | |
339 | local_irq_restore(flags); | |
340 | } | |
341 | ||
962cf36c | 342 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 343 | { |
1da177e4 LT |
344 | softirq_vec[nr].action = action; |
345 | } | |
346 | ||
1da177e4 LT |
347 | /* Tasklets */ |
348 | struct tasklet_head | |
349 | { | |
48f20a9a OJ |
350 | struct tasklet_struct *head; |
351 | struct tasklet_struct **tail; | |
1da177e4 LT |
352 | }; |
353 | ||
4620b49f VN |
354 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
355 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 356 | |
7ad5b3a5 | 357 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
358 | { |
359 | unsigned long flags; | |
360 | ||
361 | local_irq_save(flags); | |
48f20a9a OJ |
362 | t->next = NULL; |
363 | *__get_cpu_var(tasklet_vec).tail = t; | |
364 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
365 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
366 | local_irq_restore(flags); | |
367 | } | |
368 | ||
369 | EXPORT_SYMBOL(__tasklet_schedule); | |
370 | ||
7ad5b3a5 | 371 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
372 | { |
373 | unsigned long flags; | |
374 | ||
375 | local_irq_save(flags); | |
48f20a9a OJ |
376 | t->next = NULL; |
377 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
378 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
379 | raise_softirq_irqoff(HI_SOFTIRQ); |
380 | local_irq_restore(flags); | |
381 | } | |
382 | ||
383 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
384 | ||
385 | static void tasklet_action(struct softirq_action *a) | |
386 | { | |
387 | struct tasklet_struct *list; | |
388 | ||
389 | local_irq_disable(); | |
48f20a9a OJ |
390 | list = __get_cpu_var(tasklet_vec).head; |
391 | __get_cpu_var(tasklet_vec).head = NULL; | |
392 | __get_cpu_var(tasklet_vec).tail = &__get_cpu_var(tasklet_vec).head; | |
1da177e4 LT |
393 | local_irq_enable(); |
394 | ||
395 | while (list) { | |
396 | struct tasklet_struct *t = list; | |
397 | ||
398 | list = list->next; | |
399 | ||
400 | if (tasklet_trylock(t)) { | |
401 | if (!atomic_read(&t->count)) { | |
402 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
403 | BUG(); | |
404 | t->func(t->data); | |
405 | tasklet_unlock(t); | |
406 | continue; | |
407 | } | |
408 | tasklet_unlock(t); | |
409 | } | |
410 | ||
411 | local_irq_disable(); | |
48f20a9a OJ |
412 | t->next = NULL; |
413 | *__get_cpu_var(tasklet_vec).tail = t; | |
414 | __get_cpu_var(tasklet_vec).tail = &(t->next); | |
1da177e4 LT |
415 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
416 | local_irq_enable(); | |
417 | } | |
418 | } | |
419 | ||
420 | static void tasklet_hi_action(struct softirq_action *a) | |
421 | { | |
422 | struct tasklet_struct *list; | |
423 | ||
424 | local_irq_disable(); | |
48f20a9a OJ |
425 | list = __get_cpu_var(tasklet_hi_vec).head; |
426 | __get_cpu_var(tasklet_hi_vec).head = NULL; | |
427 | __get_cpu_var(tasklet_hi_vec).tail = &__get_cpu_var(tasklet_hi_vec).head; | |
1da177e4 LT |
428 | local_irq_enable(); |
429 | ||
430 | while (list) { | |
431 | struct tasklet_struct *t = list; | |
432 | ||
433 | list = list->next; | |
434 | ||
435 | if (tasklet_trylock(t)) { | |
436 | if (!atomic_read(&t->count)) { | |
437 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
438 | BUG(); | |
439 | t->func(t->data); | |
440 | tasklet_unlock(t); | |
441 | continue; | |
442 | } | |
443 | tasklet_unlock(t); | |
444 | } | |
445 | ||
446 | local_irq_disable(); | |
48f20a9a OJ |
447 | t->next = NULL; |
448 | *__get_cpu_var(tasklet_hi_vec).tail = t; | |
449 | __get_cpu_var(tasklet_hi_vec).tail = &(t->next); | |
1da177e4 LT |
450 | __raise_softirq_irqoff(HI_SOFTIRQ); |
451 | local_irq_enable(); | |
452 | } | |
453 | } | |
454 | ||
455 | ||
456 | void tasklet_init(struct tasklet_struct *t, | |
457 | void (*func)(unsigned long), unsigned long data) | |
458 | { | |
459 | t->next = NULL; | |
460 | t->state = 0; | |
461 | atomic_set(&t->count, 0); | |
462 | t->func = func; | |
463 | t->data = data; | |
464 | } | |
465 | ||
466 | EXPORT_SYMBOL(tasklet_init); | |
467 | ||
468 | void tasklet_kill(struct tasklet_struct *t) | |
469 | { | |
470 | if (in_interrupt()) | |
471 | printk("Attempt to kill tasklet from interrupt\n"); | |
472 | ||
473 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 474 | do { |
1da177e4 | 475 | yield(); |
79d381c9 | 476 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
477 | } |
478 | tasklet_unlock_wait(t); | |
479 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
480 | } | |
481 | ||
482 | EXPORT_SYMBOL(tasklet_kill); | |
483 | ||
54514a70 DM |
484 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
485 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
486 | ||
487 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
488 | { | |
489 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
490 | ||
491 | list_add_tail(&cp->list, head); | |
492 | ||
493 | /* Trigger the softirq only if the list was previously empty. */ | |
494 | if (head->next == &cp->list) | |
495 | raise_softirq_irqoff(softirq); | |
496 | } | |
497 | ||
498 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
499 | static void remote_softirq_receive(void *data) | |
500 | { | |
501 | struct call_single_data *cp = data; | |
502 | unsigned long flags; | |
503 | int softirq; | |
504 | ||
505 | softirq = cp->priv; | |
506 | ||
507 | local_irq_save(flags); | |
508 | __local_trigger(cp, softirq); | |
509 | local_irq_restore(flags); | |
510 | } | |
511 | ||
512 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
513 | { | |
514 | if (cpu_online(cpu)) { | |
515 | cp->func = remote_softirq_receive; | |
516 | cp->info = cp; | |
517 | cp->flags = 0; | |
518 | cp->priv = softirq; | |
519 | ||
6e275637 | 520 | __smp_call_function_single(cpu, cp, 0); |
54514a70 DM |
521 | return 0; |
522 | } | |
523 | return 1; | |
524 | } | |
525 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
526 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
527 | { | |
528 | return 1; | |
529 | } | |
530 | #endif | |
531 | ||
532 | /** | |
533 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
534 | * @cp: private SMP call function data area | |
535 | * @cpu: the remote cpu | |
536 | * @this_cpu: the currently executing cpu | |
537 | * @softirq: the softirq for the work | |
538 | * | |
539 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
540 | * done, the work is instead queued up on the local cpu. | |
541 | * | |
542 | * Interrupts must be disabled. | |
543 | */ | |
544 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
545 | { | |
546 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
547 | __local_trigger(cp, softirq); | |
548 | } | |
549 | EXPORT_SYMBOL(__send_remote_softirq); | |
550 | ||
551 | /** | |
552 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
553 | * @cp: private SMP call function data area | |
554 | * @cpu: the remote cpu | |
555 | * @softirq: the softirq for the work | |
556 | * | |
557 | * Like __send_remote_softirq except that disabling interrupts and | |
558 | * computing the current cpu is done for the caller. | |
559 | */ | |
560 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
561 | { | |
562 | unsigned long flags; | |
563 | int this_cpu; | |
564 | ||
565 | local_irq_save(flags); | |
566 | this_cpu = smp_processor_id(); | |
567 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
568 | local_irq_restore(flags); | |
569 | } | |
570 | EXPORT_SYMBOL(send_remote_softirq); | |
571 | ||
572 | static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self, | |
573 | unsigned long action, void *hcpu) | |
574 | { | |
575 | /* | |
576 | * If a CPU goes away, splice its entries to the current CPU | |
577 | * and trigger a run of the softirq | |
578 | */ | |
579 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
580 | int cpu = (unsigned long) hcpu; | |
581 | int i; | |
582 | ||
583 | local_irq_disable(); | |
584 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
585 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
586 | struct list_head *local_head; | |
587 | ||
588 | if (list_empty(head)) | |
589 | continue; | |
590 | ||
591 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
592 | list_splice_init(head, local_head); | |
593 | raise_softirq_irqoff(i); | |
594 | } | |
595 | local_irq_enable(); | |
596 | } | |
597 | ||
598 | return NOTIFY_OK; | |
599 | } | |
600 | ||
601 | static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = { | |
602 | .notifier_call = remote_softirq_cpu_notify, | |
603 | }; | |
604 | ||
1da177e4 LT |
605 | void __init softirq_init(void) |
606 | { | |
48f20a9a OJ |
607 | int cpu; |
608 | ||
609 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
610 | int i; |
611 | ||
48f20a9a OJ |
612 | per_cpu(tasklet_vec, cpu).tail = |
613 | &per_cpu(tasklet_vec, cpu).head; | |
614 | per_cpu(tasklet_hi_vec, cpu).tail = | |
615 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
616 | for (i = 0; i < NR_SOFTIRQS; i++) |
617 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
618 | } |
619 | ||
54514a70 DM |
620 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
621 | ||
962cf36c CM |
622 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
623 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
624 | } |
625 | ||
626 | static int ksoftirqd(void * __bind_cpu) | |
627 | { | |
1da177e4 LT |
628 | set_current_state(TASK_INTERRUPTIBLE); |
629 | ||
630 | while (!kthread_should_stop()) { | |
631 | preempt_disable(); | |
632 | if (!local_softirq_pending()) { | |
633 | preempt_enable_no_resched(); | |
634 | schedule(); | |
635 | preempt_disable(); | |
636 | } | |
637 | ||
638 | __set_current_state(TASK_RUNNING); | |
639 | ||
640 | while (local_softirq_pending()) { | |
641 | /* Preempt disable stops cpu going offline. | |
642 | If already offline, we'll be on wrong CPU: | |
643 | don't process */ | |
644 | if (cpu_is_offline((long)__bind_cpu)) | |
645 | goto wait_to_die; | |
646 | do_softirq(); | |
647 | preempt_enable_no_resched(); | |
648 | cond_resched(); | |
649 | preempt_disable(); | |
64ca5ab9 | 650 | rcu_qsctr_inc((long)__bind_cpu); |
1da177e4 LT |
651 | } |
652 | preempt_enable(); | |
653 | set_current_state(TASK_INTERRUPTIBLE); | |
654 | } | |
655 | __set_current_state(TASK_RUNNING); | |
656 | return 0; | |
657 | ||
658 | wait_to_die: | |
659 | preempt_enable(); | |
660 | /* Wait for kthread_stop */ | |
661 | set_current_state(TASK_INTERRUPTIBLE); | |
662 | while (!kthread_should_stop()) { | |
663 | schedule(); | |
664 | set_current_state(TASK_INTERRUPTIBLE); | |
665 | } | |
666 | __set_current_state(TASK_RUNNING); | |
667 | return 0; | |
668 | } | |
669 | ||
670 | #ifdef CONFIG_HOTPLUG_CPU | |
671 | /* | |
672 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
673 | * scheduled for execution on @cpu. | |
674 | * | |
675 | * Unlike tasklet_kill, this function removes the tasklet | |
676 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
677 | * | |
678 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
679 | */ | |
680 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
681 | { | |
682 | struct tasklet_struct **i; | |
683 | ||
684 | BUG_ON(cpu_online(cpu)); | |
685 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
686 | ||
687 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
688 | return; | |
689 | ||
690 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 691 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
692 | if (*i == t) { |
693 | *i = t->next; | |
48f20a9a OJ |
694 | /* If this was the tail element, move the tail ptr */ |
695 | if (*i == NULL) | |
696 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
697 | return; |
698 | } | |
699 | } | |
700 | BUG(); | |
701 | } | |
702 | ||
703 | static void takeover_tasklets(unsigned int cpu) | |
704 | { | |
1da177e4 LT |
705 | /* CPU is dead, so no lock needed. */ |
706 | local_irq_disable(); | |
707 | ||
708 | /* Find end, append list for that CPU. */ | |
e5e41723 CB |
709 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
710 | *(__get_cpu_var(tasklet_vec).tail) = per_cpu(tasklet_vec, cpu).head; | |
711 | __get_cpu_var(tasklet_vec).tail = per_cpu(tasklet_vec, cpu).tail; | |
712 | per_cpu(tasklet_vec, cpu).head = NULL; | |
713 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
714 | } | |
1da177e4 LT |
715 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
716 | ||
e5e41723 CB |
717 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
718 | *__get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).head; | |
719 | __get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).tail; | |
720 | per_cpu(tasklet_hi_vec, cpu).head = NULL; | |
721 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
722 | } | |
1da177e4 LT |
723 | raise_softirq_irqoff(HI_SOFTIRQ); |
724 | ||
725 | local_irq_enable(); | |
726 | } | |
727 | #endif /* CONFIG_HOTPLUG_CPU */ | |
728 | ||
8c78f307 | 729 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
730 | unsigned long action, |
731 | void *hcpu) | |
732 | { | |
733 | int hotcpu = (unsigned long)hcpu; | |
734 | struct task_struct *p; | |
735 | ||
736 | switch (action) { | |
737 | case CPU_UP_PREPARE: | |
8bb78442 | 738 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
739 | p = kthread_create(ksoftirqd, hcpu, "ksoftirqd/%d", hotcpu); |
740 | if (IS_ERR(p)) { | |
741 | printk("ksoftirqd for %i failed\n", hotcpu); | |
742 | return NOTIFY_BAD; | |
743 | } | |
744 | kthread_bind(p, hotcpu); | |
745 | per_cpu(ksoftirqd, hotcpu) = p; | |
746 | break; | |
747 | case CPU_ONLINE: | |
8bb78442 | 748 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
749 | wake_up_process(per_cpu(ksoftirqd, hotcpu)); |
750 | break; | |
751 | #ifdef CONFIG_HOTPLUG_CPU | |
752 | case CPU_UP_CANCELED: | |
8bb78442 | 753 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
754 | if (!per_cpu(ksoftirqd, hotcpu)) |
755 | break; | |
1da177e4 | 756 | /* Unbind so it can run. Fall thru. */ |
a4c4af7c | 757 | kthread_bind(per_cpu(ksoftirqd, hotcpu), |
f1fc057c | 758 | cpumask_any(cpu_online_mask)); |
1da177e4 | 759 | case CPU_DEAD: |
1c6b4aa9 ST |
760 | case CPU_DEAD_FROZEN: { |
761 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; | |
762 | ||
1da177e4 LT |
763 | p = per_cpu(ksoftirqd, hotcpu); |
764 | per_cpu(ksoftirqd, hotcpu) = NULL; | |
961ccddd | 765 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
1da177e4 LT |
766 | kthread_stop(p); |
767 | takeover_tasklets(hotcpu); | |
768 | break; | |
1c6b4aa9 | 769 | } |
1da177e4 LT |
770 | #endif /* CONFIG_HOTPLUG_CPU */ |
771 | } | |
772 | return NOTIFY_OK; | |
773 | } | |
774 | ||
8c78f307 | 775 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
776 | .notifier_call = cpu_callback |
777 | }; | |
778 | ||
7babe8db | 779 | static __init int spawn_ksoftirqd(void) |
1da177e4 LT |
780 | { |
781 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 AM |
782 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
783 | ||
784 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
785 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
786 | register_cpu_notifier(&cpu_nfb); | |
787 | return 0; | |
788 | } | |
7babe8db | 789 | early_initcall(spawn_ksoftirqd); |
78eef01b AM |
790 | |
791 | #ifdef CONFIG_SMP | |
792 | /* | |
793 | * Call a function on all processors | |
794 | */ | |
15c8b6c1 | 795 | int on_each_cpu(void (*func) (void *info), void *info, int wait) |
78eef01b AM |
796 | { |
797 | int ret = 0; | |
798 | ||
799 | preempt_disable(); | |
8691e5a8 | 800 | ret = smp_call_function(func, info, wait); |
78eef01b AM |
801 | local_irq_disable(); |
802 | func(info); | |
803 | local_irq_enable(); | |
804 | preempt_enable(); | |
805 | return ret; | |
806 | } | |
807 | EXPORT_SYMBOL(on_each_cpu); | |
808 | #endif | |
43a25632 YL |
809 | |
810 | /* | |
811 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
812 | * GCC does not inline them incorrectly. ] | |
813 | */ | |
814 | ||
815 | int __init __weak early_irq_init(void) | |
816 | { | |
817 | return 0; | |
818 | } | |
819 | ||
4a046d17 YL |
820 | int __init __weak arch_probe_nr_irqs(void) |
821 | { | |
822 | return 0; | |
823 | } | |
824 | ||
43a25632 YL |
825 | int __init __weak arch_early_irq_init(void) |
826 | { | |
827 | return 0; | |
828 | } | |
829 | ||
85ac16d0 | 830 | int __weak arch_init_chip_data(struct irq_desc *desc, int node) |
43a25632 YL |
831 | { |
832 | return 0; | |
833 | } |