]>
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 | 56 | |
4dd53d89 | 57 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 58 | |
5d592b44 | 59 | char *softirq_to_name[NR_SOFTIRQS] = { |
5dd4de58 | 60 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL", |
899039e8 | 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 | */ | |
676cb02d | 70 | static void wakeup_softirqd(void) |
1da177e4 LT |
71 | { |
72 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 73 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 LT |
74 | |
75 | if (tsk && tsk->state != TASK_RUNNING) | |
76 | wake_up_process(tsk); | |
77 | } | |
78 | ||
75e1056f VP |
79 | /* |
80 | * preempt_count and SOFTIRQ_OFFSET usage: | |
81 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving | |
82 | * softirq processing. | |
83 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
84 | * on local_bh_disable or local_bh_enable. | |
85 | * This lets us distinguish between whether we are currently processing | |
86 | * softirq and whether we just have bh disabled. | |
87 | */ | |
88 | ||
de30a2b3 IM |
89 | /* |
90 | * This one is for softirq.c-internal use, | |
91 | * where hardirqs are disabled legitimately: | |
92 | */ | |
3c829c36 | 93 | #ifdef CONFIG_TRACE_IRQFLAGS |
75e1056f | 94 | static void __local_bh_disable(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
95 | { |
96 | unsigned long flags; | |
97 | ||
98 | WARN_ON_ONCE(in_irq()); | |
99 | ||
100 | raw_local_irq_save(flags); | |
7e49fcce SR |
101 | /* |
102 | * The preempt tracer hooks into add_preempt_count and will break | |
103 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET | |
104 | * is set and before current->softirq_enabled is cleared. | |
105 | * We must manually increment preempt_count here and manually | |
106 | * call the trace_preempt_off later. | |
107 | */ | |
75e1056f | 108 | preempt_count() += cnt; |
de30a2b3 IM |
109 | /* |
110 | * Were softirqs turned off above: | |
111 | */ | |
75e1056f | 112 | if (softirq_count() == cnt) |
de30a2b3 IM |
113 | trace_softirqs_off(ip); |
114 | raw_local_irq_restore(flags); | |
7e49fcce | 115 | |
75e1056f | 116 | if (preempt_count() == cnt) |
7e49fcce | 117 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); |
de30a2b3 | 118 | } |
3c829c36 | 119 | #else /* !CONFIG_TRACE_IRQFLAGS */ |
75e1056f | 120 | static inline void __local_bh_disable(unsigned long ip, unsigned int cnt) |
3c829c36 | 121 | { |
75e1056f | 122 | add_preempt_count(cnt); |
3c829c36 TC |
123 | barrier(); |
124 | } | |
125 | #endif /* CONFIG_TRACE_IRQFLAGS */ | |
de30a2b3 IM |
126 | |
127 | void local_bh_disable(void) | |
128 | { | |
75e1056f VP |
129 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
130 | SOFTIRQ_DISABLE_OFFSET); | |
de30a2b3 IM |
131 | } |
132 | ||
133 | EXPORT_SYMBOL(local_bh_disable); | |
134 | ||
75e1056f VP |
135 | static void __local_bh_enable(unsigned int cnt) |
136 | { | |
137 | WARN_ON_ONCE(in_irq()); | |
138 | WARN_ON_ONCE(!irqs_disabled()); | |
139 | ||
140 | if (softirq_count() == cnt) | |
141 | trace_softirqs_on((unsigned long)__builtin_return_address(0)); | |
142 | sub_preempt_count(cnt); | |
143 | } | |
144 | ||
de30a2b3 IM |
145 | /* |
146 | * Special-case - softirqs can safely be enabled in | |
147 | * cond_resched_softirq(), or by __do_softirq(), | |
148 | * without processing still-pending softirqs: | |
149 | */ | |
150 | void _local_bh_enable(void) | |
151 | { | |
75e1056f | 152 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
153 | } |
154 | ||
155 | EXPORT_SYMBOL(_local_bh_enable); | |
156 | ||
0f476b6d | 157 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 158 | { |
0f476b6d | 159 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 160 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 161 | local_irq_disable(); |
3c829c36 | 162 | #endif |
de30a2b3 IM |
163 | /* |
164 | * Are softirqs going to be turned on now: | |
165 | */ | |
75e1056f | 166 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 167 | trace_softirqs_on(ip); |
de30a2b3 IM |
168 | /* |
169 | * Keep preemption disabled until we are done with | |
170 | * softirq processing: | |
171 | */ | |
75e1056f | 172 | sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1); |
de30a2b3 IM |
173 | |
174 | if (unlikely(!in_interrupt() && local_softirq_pending())) | |
175 | do_softirq(); | |
176 | ||
177 | dec_preempt_count(); | |
3c829c36 | 178 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 179 | local_irq_enable(); |
3c829c36 | 180 | #endif |
de30a2b3 IM |
181 | preempt_check_resched(); |
182 | } | |
0f476b6d JB |
183 | |
184 | void local_bh_enable(void) | |
185 | { | |
186 | _local_bh_enable_ip((unsigned long)__builtin_return_address(0)); | |
187 | } | |
de30a2b3 IM |
188 | EXPORT_SYMBOL(local_bh_enable); |
189 | ||
190 | void local_bh_enable_ip(unsigned long ip) | |
191 | { | |
0f476b6d | 192 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
193 | } |
194 | EXPORT_SYMBOL(local_bh_enable_ip); | |
195 | ||
1da177e4 LT |
196 | /* |
197 | * We restart softirq processing MAX_SOFTIRQ_RESTART times, | |
198 | * and we fall back to softirqd after that. | |
199 | * | |
200 | * This number has been established via experimentation. | |
201 | * The two things to balance is latency against fairness - | |
202 | * we want to handle softirqs as soon as possible, but they | |
203 | * should not be able to lock up the box. | |
204 | */ | |
205 | #define MAX_SOFTIRQ_RESTART 10 | |
206 | ||
207 | asmlinkage void __do_softirq(void) | |
208 | { | |
209 | struct softirq_action *h; | |
210 | __u32 pending; | |
211 | int max_restart = MAX_SOFTIRQ_RESTART; | |
212 | int cpu; | |
213 | ||
214 | pending = local_softirq_pending(); | |
829035fd PM |
215 | account_system_vtime(current); |
216 | ||
75e1056f VP |
217 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
218 | SOFTIRQ_OFFSET); | |
d820ac4c | 219 | lockdep_softirq_enter(); |
1da177e4 | 220 | |
1da177e4 LT |
221 | cpu = smp_processor_id(); |
222 | restart: | |
223 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 224 | set_softirq_pending(0); |
1da177e4 | 225 | |
c70f5d66 | 226 | local_irq_enable(); |
1da177e4 LT |
227 | |
228 | h = softirq_vec; | |
229 | ||
230 | do { | |
231 | if (pending & 1) { | |
f4bc6bb2 | 232 | unsigned int vec_nr = h - softirq_vec; |
8e85b4b5 TG |
233 | int prev_count = preempt_count(); |
234 | ||
f4bc6bb2 TG |
235 | kstat_incr_softirqs_this_cpu(vec_nr); |
236 | ||
237 | trace_softirq_entry(vec_nr); | |
1da177e4 | 238 | h->action(h); |
f4bc6bb2 | 239 | trace_softirq_exit(vec_nr); |
8e85b4b5 | 240 | if (unlikely(prev_count != preempt_count())) { |
f4bc6bb2 | 241 | printk(KERN_ERR "huh, entered softirq %u %s %p" |
8e85b4b5 | 242 | "with preempt_count %08x," |
f4bc6bb2 TG |
243 | " exited with %08x?\n", vec_nr, |
244 | softirq_to_name[vec_nr], h->action, | |
245 | prev_count, preempt_count()); | |
8e85b4b5 TG |
246 | preempt_count() = prev_count; |
247 | } | |
248 | ||
d6714c22 | 249 | rcu_bh_qs(cpu); |
1da177e4 LT |
250 | } |
251 | h++; | |
252 | pending >>= 1; | |
253 | } while (pending); | |
254 | ||
c70f5d66 | 255 | local_irq_disable(); |
1da177e4 LT |
256 | |
257 | pending = local_softirq_pending(); | |
258 | if (pending && --max_restart) | |
259 | goto restart; | |
260 | ||
261 | if (pending) | |
262 | wakeup_softirqd(); | |
263 | ||
d820ac4c | 264 | lockdep_softirq_exit(); |
829035fd PM |
265 | |
266 | account_system_vtime(current); | |
75e1056f | 267 | __local_bh_enable(SOFTIRQ_OFFSET); |
1da177e4 LT |
268 | } |
269 | ||
270 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
271 | ||
272 | asmlinkage void do_softirq(void) | |
273 | { | |
274 | __u32 pending; | |
275 | unsigned long flags; | |
276 | ||
277 | if (in_interrupt()) | |
278 | return; | |
279 | ||
280 | local_irq_save(flags); | |
281 | ||
282 | pending = local_softirq_pending(); | |
283 | ||
284 | if (pending) | |
285 | __do_softirq(); | |
286 | ||
287 | local_irq_restore(flags); | |
288 | } | |
289 | ||
1da177e4 LT |
290 | #endif |
291 | ||
dde4b2b5 IM |
292 | /* |
293 | * Enter an interrupt context. | |
294 | */ | |
295 | void irq_enter(void) | |
296 | { | |
6378ddb5 | 297 | int cpu = smp_processor_id(); |
719254fa | 298 | |
64db4cff | 299 | rcu_irq_enter(); |
ee5f80a9 | 300 | if (idle_cpu(cpu) && !in_interrupt()) { |
d267f87f VP |
301 | /* |
302 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
303 | * here, as softirq will be serviced on return from interrupt. | |
304 | */ | |
305 | local_bh_disable(); | |
719254fa | 306 | tick_check_idle(cpu); |
d267f87f VP |
307 | _local_bh_enable(); |
308 | } | |
309 | ||
310 | __irq_enter(); | |
dde4b2b5 IM |
311 | } |
312 | ||
1da177e4 | 313 | #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED |
8d32a307 TG |
314 | static inline void invoke_softirq(void) |
315 | { | |
316 | if (!force_irqthreads) | |
317 | __do_softirq(); | |
318 | else | |
319 | wakeup_softirqd(); | |
320 | } | |
1da177e4 | 321 | #else |
8d32a307 TG |
322 | static inline void invoke_softirq(void) |
323 | { | |
324 | if (!force_irqthreads) | |
325 | do_softirq(); | |
326 | else | |
327 | wakeup_softirqd(); | |
328 | } | |
1da177e4 LT |
329 | #endif |
330 | ||
331 | /* | |
332 | * Exit an interrupt context. Process softirqs if needed and possible: | |
333 | */ | |
334 | void irq_exit(void) | |
335 | { | |
336 | account_system_vtime(current); | |
de30a2b3 | 337 | trace_hardirq_exit(); |
1da177e4 LT |
338 | sub_preempt_count(IRQ_EXIT_OFFSET); |
339 | if (!in_interrupt() && local_softirq_pending()) | |
340 | invoke_softirq(); | |
79bf2bb3 | 341 | |
c5e0cb3d | 342 | rcu_irq_exit(); |
79bf2bb3 TG |
343 | #ifdef CONFIG_NO_HZ |
344 | /* Make sure that timer wheel updates are propagated */ | |
64db4cff PM |
345 | if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched()) |
346 | tick_nohz_stop_sched_tick(0); | |
79bf2bb3 | 347 | #endif |
1da177e4 LT |
348 | preempt_enable_no_resched(); |
349 | } | |
350 | ||
351 | /* | |
352 | * This function must run with irqs disabled! | |
353 | */ | |
7ad5b3a5 | 354 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
355 | { |
356 | __raise_softirq_irqoff(nr); | |
357 | ||
358 | /* | |
359 | * If we're in an interrupt or softirq, we're done | |
360 | * (this also catches softirq-disabled code). We will | |
361 | * actually run the softirq once we return from | |
362 | * the irq or softirq. | |
363 | * | |
364 | * Otherwise we wake up ksoftirqd to make sure we | |
365 | * schedule the softirq soon. | |
366 | */ | |
367 | if (!in_interrupt()) | |
368 | wakeup_softirqd(); | |
369 | } | |
370 | ||
7ad5b3a5 | 371 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
372 | { |
373 | unsigned long flags; | |
374 | ||
375 | local_irq_save(flags); | |
376 | raise_softirq_irqoff(nr); | |
377 | local_irq_restore(flags); | |
378 | } | |
379 | ||
962cf36c | 380 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 381 | { |
1da177e4 LT |
382 | softirq_vec[nr].action = action; |
383 | } | |
384 | ||
9ba5f005 PZ |
385 | /* |
386 | * Tasklets | |
387 | */ | |
1da177e4 LT |
388 | struct tasklet_head |
389 | { | |
48f20a9a OJ |
390 | struct tasklet_struct *head; |
391 | struct tasklet_struct **tail; | |
1da177e4 LT |
392 | }; |
393 | ||
4620b49f VN |
394 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
395 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 396 | |
7ad5b3a5 | 397 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
398 | { |
399 | unsigned long flags; | |
400 | ||
401 | local_irq_save(flags); | |
48f20a9a | 402 | t->next = NULL; |
909ea964 CL |
403 | *__this_cpu_read(tasklet_vec.tail) = t; |
404 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
405 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
406 | local_irq_restore(flags); | |
407 | } | |
408 | ||
409 | EXPORT_SYMBOL(__tasklet_schedule); | |
410 | ||
7ad5b3a5 | 411 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
412 | { |
413 | unsigned long flags; | |
414 | ||
415 | local_irq_save(flags); | |
48f20a9a | 416 | t->next = NULL; |
909ea964 CL |
417 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
418 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
419 | raise_softirq_irqoff(HI_SOFTIRQ); |
420 | local_irq_restore(flags); | |
421 | } | |
422 | ||
423 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
424 | ||
7c692cba VN |
425 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
426 | { | |
427 | BUG_ON(!irqs_disabled()); | |
428 | ||
909ea964 CL |
429 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
430 | __this_cpu_write(tasklet_hi_vec.head, t); | |
7c692cba VN |
431 | __raise_softirq_irqoff(HI_SOFTIRQ); |
432 | } | |
433 | ||
434 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); | |
435 | ||
1da177e4 LT |
436 | static void tasklet_action(struct softirq_action *a) |
437 | { | |
438 | struct tasklet_struct *list; | |
439 | ||
440 | local_irq_disable(); | |
909ea964 CL |
441 | list = __this_cpu_read(tasklet_vec.head); |
442 | __this_cpu_write(tasklet_vec.head, NULL); | |
443 | __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head); | |
1da177e4 LT |
444 | local_irq_enable(); |
445 | ||
446 | while (list) { | |
447 | struct tasklet_struct *t = list; | |
448 | ||
449 | list = list->next; | |
450 | ||
451 | if (tasklet_trylock(t)) { | |
452 | if (!atomic_read(&t->count)) { | |
453 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
454 | BUG(); | |
455 | t->func(t->data); | |
456 | tasklet_unlock(t); | |
457 | continue; | |
458 | } | |
459 | tasklet_unlock(t); | |
460 | } | |
461 | ||
462 | local_irq_disable(); | |
48f20a9a | 463 | t->next = NULL; |
909ea964 CL |
464 | *__this_cpu_read(tasklet_vec.tail) = t; |
465 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
466 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
467 | local_irq_enable(); | |
468 | } | |
469 | } | |
470 | ||
471 | static void tasklet_hi_action(struct softirq_action *a) | |
472 | { | |
473 | struct tasklet_struct *list; | |
474 | ||
475 | local_irq_disable(); | |
909ea964 CL |
476 | list = __this_cpu_read(tasklet_hi_vec.head); |
477 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
478 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
479 | local_irq_enable(); |
480 | ||
481 | while (list) { | |
482 | struct tasklet_struct *t = list; | |
483 | ||
484 | list = list->next; | |
485 | ||
486 | if (tasklet_trylock(t)) { | |
487 | if (!atomic_read(&t->count)) { | |
488 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
489 | BUG(); | |
490 | t->func(t->data); | |
491 | tasklet_unlock(t); | |
492 | continue; | |
493 | } | |
494 | tasklet_unlock(t); | |
495 | } | |
496 | ||
497 | local_irq_disable(); | |
48f20a9a | 498 | t->next = NULL; |
909ea964 CL |
499 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
500 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
501 | __raise_softirq_irqoff(HI_SOFTIRQ); |
502 | local_irq_enable(); | |
503 | } | |
504 | } | |
505 | ||
506 | ||
507 | void tasklet_init(struct tasklet_struct *t, | |
508 | void (*func)(unsigned long), unsigned long data) | |
509 | { | |
510 | t->next = NULL; | |
511 | t->state = 0; | |
512 | atomic_set(&t->count, 0); | |
513 | t->func = func; | |
514 | t->data = data; | |
515 | } | |
516 | ||
517 | EXPORT_SYMBOL(tasklet_init); | |
518 | ||
519 | void tasklet_kill(struct tasklet_struct *t) | |
520 | { | |
521 | if (in_interrupt()) | |
522 | printk("Attempt to kill tasklet from interrupt\n"); | |
523 | ||
524 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 525 | do { |
1da177e4 | 526 | yield(); |
79d381c9 | 527 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
528 | } |
529 | tasklet_unlock_wait(t); | |
530 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
531 | } | |
532 | ||
533 | EXPORT_SYMBOL(tasklet_kill); | |
534 | ||
9ba5f005 PZ |
535 | /* |
536 | * tasklet_hrtimer | |
537 | */ | |
538 | ||
539 | /* | |
b9c30322 PZ |
540 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
541 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
542 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
543 | */ |
544 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
545 | { | |
546 | struct tasklet_hrtimer *ttimer = | |
547 | container_of(timer, struct tasklet_hrtimer, timer); | |
548 | ||
b9c30322 PZ |
549 | tasklet_hi_schedule(&ttimer->tasklet); |
550 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
551 | } |
552 | ||
553 | /* | |
554 | * Helper function which calls the hrtimer callback from | |
555 | * tasklet/softirq context | |
556 | */ | |
557 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
558 | { | |
559 | struct tasklet_hrtimer *ttimer = (void *)data; | |
560 | enum hrtimer_restart restart; | |
561 | ||
562 | restart = ttimer->function(&ttimer->timer); | |
563 | if (restart != HRTIMER_NORESTART) | |
564 | hrtimer_restart(&ttimer->timer); | |
565 | } | |
566 | ||
567 | /** | |
568 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
569 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 570 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
571 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
572 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
573 | */ | |
574 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
575 | enum hrtimer_restart (*function)(struct hrtimer *), | |
576 | clockid_t which_clock, enum hrtimer_mode mode) | |
577 | { | |
578 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
579 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
580 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
581 | (unsigned long)ttimer); | |
582 | ttimer->function = function; | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
585 | ||
586 | /* | |
587 | * Remote softirq bits | |
588 | */ | |
589 | ||
54514a70 DM |
590 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
591 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
592 | ||
593 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
594 | { | |
595 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
596 | ||
597 | list_add_tail(&cp->list, head); | |
598 | ||
599 | /* Trigger the softirq only if the list was previously empty. */ | |
600 | if (head->next == &cp->list) | |
601 | raise_softirq_irqoff(softirq); | |
602 | } | |
603 | ||
604 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
605 | static void remote_softirq_receive(void *data) | |
606 | { | |
607 | struct call_single_data *cp = data; | |
608 | unsigned long flags; | |
609 | int softirq; | |
610 | ||
611 | softirq = cp->priv; | |
612 | ||
613 | local_irq_save(flags); | |
614 | __local_trigger(cp, softirq); | |
615 | local_irq_restore(flags); | |
616 | } | |
617 | ||
618 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
619 | { | |
620 | if (cpu_online(cpu)) { | |
621 | cp->func = remote_softirq_receive; | |
622 | cp->info = cp; | |
623 | cp->flags = 0; | |
624 | cp->priv = softirq; | |
625 | ||
6e275637 | 626 | __smp_call_function_single(cpu, cp, 0); |
54514a70 DM |
627 | return 0; |
628 | } | |
629 | return 1; | |
630 | } | |
631 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
632 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
633 | { | |
634 | return 1; | |
635 | } | |
636 | #endif | |
637 | ||
638 | /** | |
639 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
640 | * @cp: private SMP call function data area | |
641 | * @cpu: the remote cpu | |
642 | * @this_cpu: the currently executing cpu | |
643 | * @softirq: the softirq for the work | |
644 | * | |
645 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
646 | * done, the work is instead queued up on the local cpu. | |
647 | * | |
648 | * Interrupts must be disabled. | |
649 | */ | |
650 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
651 | { | |
652 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
653 | __local_trigger(cp, softirq); | |
654 | } | |
655 | EXPORT_SYMBOL(__send_remote_softirq); | |
656 | ||
657 | /** | |
658 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
659 | * @cp: private SMP call function data area | |
660 | * @cpu: the remote cpu | |
661 | * @softirq: the softirq for the work | |
662 | * | |
663 | * Like __send_remote_softirq except that disabling interrupts and | |
664 | * computing the current cpu is done for the caller. | |
665 | */ | |
666 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
667 | { | |
668 | unsigned long flags; | |
669 | int this_cpu; | |
670 | ||
671 | local_irq_save(flags); | |
672 | this_cpu = smp_processor_id(); | |
673 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
674 | local_irq_restore(flags); | |
675 | } | |
676 | EXPORT_SYMBOL(send_remote_softirq); | |
677 | ||
678 | static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self, | |
679 | unsigned long action, void *hcpu) | |
680 | { | |
681 | /* | |
682 | * If a CPU goes away, splice its entries to the current CPU | |
683 | * and trigger a run of the softirq | |
684 | */ | |
685 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
686 | int cpu = (unsigned long) hcpu; | |
687 | int i; | |
688 | ||
689 | local_irq_disable(); | |
690 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
691 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
692 | struct list_head *local_head; | |
693 | ||
694 | if (list_empty(head)) | |
695 | continue; | |
696 | ||
697 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
698 | list_splice_init(head, local_head); | |
699 | raise_softirq_irqoff(i); | |
700 | } | |
701 | local_irq_enable(); | |
702 | } | |
703 | ||
704 | return NOTIFY_OK; | |
705 | } | |
706 | ||
707 | static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = { | |
708 | .notifier_call = remote_softirq_cpu_notify, | |
709 | }; | |
710 | ||
1da177e4 LT |
711 | void __init softirq_init(void) |
712 | { | |
48f20a9a OJ |
713 | int cpu; |
714 | ||
715 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
716 | int i; |
717 | ||
48f20a9a OJ |
718 | per_cpu(tasklet_vec, cpu).tail = |
719 | &per_cpu(tasklet_vec, cpu).head; | |
720 | per_cpu(tasklet_hi_vec, cpu).tail = | |
721 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
722 | for (i = 0; i < NR_SOFTIRQS; i++) |
723 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
724 | } |
725 | ||
54514a70 DM |
726 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
727 | ||
962cf36c CM |
728 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
729 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
730 | } |
731 | ||
1871e52c | 732 | static int run_ksoftirqd(void * __bind_cpu) |
1da177e4 | 733 | { |
1da177e4 LT |
734 | set_current_state(TASK_INTERRUPTIBLE); |
735 | ||
736 | while (!kthread_should_stop()) { | |
737 | preempt_disable(); | |
738 | if (!local_softirq_pending()) { | |
739 | preempt_enable_no_resched(); | |
740 | schedule(); | |
741 | preempt_disable(); | |
742 | } | |
743 | ||
744 | __set_current_state(TASK_RUNNING); | |
745 | ||
746 | while (local_softirq_pending()) { | |
747 | /* Preempt disable stops cpu going offline. | |
748 | If already offline, we'll be on wrong CPU: | |
749 | don't process */ | |
750 | if (cpu_is_offline((long)__bind_cpu)) | |
751 | goto wait_to_die; | |
c305d524 TG |
752 | local_irq_disable(); |
753 | if (local_softirq_pending()) | |
754 | __do_softirq(); | |
755 | local_irq_enable(); | |
1da177e4 LT |
756 | preempt_enable_no_resched(); |
757 | cond_resched(); | |
758 | preempt_disable(); | |
25502a6c | 759 | rcu_note_context_switch((long)__bind_cpu); |
1da177e4 LT |
760 | } |
761 | preempt_enable(); | |
762 | set_current_state(TASK_INTERRUPTIBLE); | |
763 | } | |
764 | __set_current_state(TASK_RUNNING); | |
765 | return 0; | |
766 | ||
767 | wait_to_die: | |
768 | preempt_enable(); | |
769 | /* Wait for kthread_stop */ | |
770 | set_current_state(TASK_INTERRUPTIBLE); | |
771 | while (!kthread_should_stop()) { | |
772 | schedule(); | |
773 | set_current_state(TASK_INTERRUPTIBLE); | |
774 | } | |
775 | __set_current_state(TASK_RUNNING); | |
776 | return 0; | |
777 | } | |
778 | ||
779 | #ifdef CONFIG_HOTPLUG_CPU | |
780 | /* | |
781 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
782 | * scheduled for execution on @cpu. | |
783 | * | |
784 | * Unlike tasklet_kill, this function removes the tasklet | |
785 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
786 | * | |
787 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
788 | */ | |
789 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
790 | { | |
791 | struct tasklet_struct **i; | |
792 | ||
793 | BUG_ON(cpu_online(cpu)); | |
794 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
795 | ||
796 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
797 | return; | |
798 | ||
799 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 800 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
801 | if (*i == t) { |
802 | *i = t->next; | |
48f20a9a OJ |
803 | /* If this was the tail element, move the tail ptr */ |
804 | if (*i == NULL) | |
805 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
806 | return; |
807 | } | |
808 | } | |
809 | BUG(); | |
810 | } | |
811 | ||
812 | static void takeover_tasklets(unsigned int cpu) | |
813 | { | |
1da177e4 LT |
814 | /* CPU is dead, so no lock needed. */ |
815 | local_irq_disable(); | |
816 | ||
817 | /* Find end, append list for that CPU. */ | |
e5e41723 | 818 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
819 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
820 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
821 | per_cpu(tasklet_vec, cpu).head = NULL; |
822 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
823 | } | |
1da177e4 LT |
824 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
825 | ||
e5e41723 | 826 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
827 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
828 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
829 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
830 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
831 | } | |
1da177e4 LT |
832 | raise_softirq_irqoff(HI_SOFTIRQ); |
833 | ||
834 | local_irq_enable(); | |
835 | } | |
836 | #endif /* CONFIG_HOTPLUG_CPU */ | |
837 | ||
8c78f307 | 838 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
839 | unsigned long action, |
840 | void *hcpu) | |
841 | { | |
842 | int hotcpu = (unsigned long)hcpu; | |
843 | struct task_struct *p; | |
844 | ||
845 | switch (action) { | |
846 | case CPU_UP_PREPARE: | |
8bb78442 | 847 | case CPU_UP_PREPARE_FROZEN: |
94dcf29a ED |
848 | p = kthread_create_on_node(run_ksoftirqd, |
849 | hcpu, | |
850 | cpu_to_node(hotcpu), | |
851 | "ksoftirqd/%d", hotcpu); | |
1da177e4 LT |
852 | if (IS_ERR(p)) { |
853 | printk("ksoftirqd for %i failed\n", hotcpu); | |
80b5184c | 854 | return notifier_from_errno(PTR_ERR(p)); |
1da177e4 LT |
855 | } |
856 | kthread_bind(p, hotcpu); | |
857 | per_cpu(ksoftirqd, hotcpu) = p; | |
858 | break; | |
859 | case CPU_ONLINE: | |
8bb78442 | 860 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
861 | wake_up_process(per_cpu(ksoftirqd, hotcpu)); |
862 | break; | |
863 | #ifdef CONFIG_HOTPLUG_CPU | |
864 | case CPU_UP_CANCELED: | |
8bb78442 | 865 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
866 | if (!per_cpu(ksoftirqd, hotcpu)) |
867 | break; | |
1da177e4 | 868 | /* Unbind so it can run. Fall thru. */ |
a4c4af7c | 869 | kthread_bind(per_cpu(ksoftirqd, hotcpu), |
f1fc057c | 870 | cpumask_any(cpu_online_mask)); |
1da177e4 | 871 | case CPU_DEAD: |
1c6b4aa9 | 872 | case CPU_DEAD_FROZEN: { |
c9b5f501 | 873 | static const struct sched_param param = { |
fe7de49f KM |
874 | .sched_priority = MAX_RT_PRIO-1 |
875 | }; | |
1c6b4aa9 | 876 | |
1da177e4 LT |
877 | p = per_cpu(ksoftirqd, hotcpu); |
878 | per_cpu(ksoftirqd, hotcpu) = NULL; | |
961ccddd | 879 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
1da177e4 LT |
880 | kthread_stop(p); |
881 | takeover_tasklets(hotcpu); | |
882 | break; | |
1c6b4aa9 | 883 | } |
1da177e4 LT |
884 | #endif /* CONFIG_HOTPLUG_CPU */ |
885 | } | |
886 | return NOTIFY_OK; | |
887 | } | |
888 | ||
8c78f307 | 889 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
890 | .notifier_call = cpu_callback |
891 | }; | |
892 | ||
7babe8db | 893 | static __init int spawn_ksoftirqd(void) |
1da177e4 LT |
894 | { |
895 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 AM |
896 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
897 | ||
9e506f7a | 898 | BUG_ON(err != NOTIFY_OK); |
1da177e4 LT |
899 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
900 | register_cpu_notifier(&cpu_nfb); | |
901 | return 0; | |
902 | } | |
7babe8db | 903 | early_initcall(spawn_ksoftirqd); |
78eef01b | 904 | |
43a25632 YL |
905 | /* |
906 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
907 | * GCC does not inline them incorrectly. ] | |
908 | */ | |
909 | ||
910 | int __init __weak early_irq_init(void) | |
911 | { | |
912 | return 0; | |
913 | } | |
914 | ||
b683de2b | 915 | #ifdef CONFIG_GENERIC_HARDIRQS |
4a046d17 YL |
916 | int __init __weak arch_probe_nr_irqs(void) |
917 | { | |
b683de2b | 918 | return NR_IRQS_LEGACY; |
4a046d17 YL |
919 | } |
920 | ||
43a25632 YL |
921 | int __init __weak arch_early_irq_init(void) |
922 | { | |
923 | return 0; | |
924 | } | |
b683de2b | 925 | #endif |