]>
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 | ||
9984de1a | 13 | #include <linux/export.h> |
1da177e4 LT |
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> |
3e339b5d | 26 | #include <linux/smpboot.h> |
79bf2bb3 | 27 | #include <linux/tick.h> |
a0e39ed3 HC |
28 | |
29 | #define CREATE_TRACE_POINTS | |
ad8d75ff | 30 | #include <trace/events/irq.h> |
1da177e4 | 31 | |
1da177e4 LT |
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", |
09223371 | 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 | { | |
d2e08473 | 129 | __local_bh_disable(_RET_IP_, SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
130 | } |
131 | ||
132 | EXPORT_SYMBOL(local_bh_disable); | |
133 | ||
75e1056f VP |
134 | static void __local_bh_enable(unsigned int cnt) |
135 | { | |
75e1056f VP |
136 | WARN_ON_ONCE(!irqs_disabled()); |
137 | ||
138 | if (softirq_count() == cnt) | |
d2e08473 | 139 | trace_softirqs_on(_RET_IP_); |
75e1056f VP |
140 | sub_preempt_count(cnt); |
141 | } | |
142 | ||
de30a2b3 IM |
143 | /* |
144 | * Special-case - softirqs can safely be enabled in | |
145 | * cond_resched_softirq(), or by __do_softirq(), | |
146 | * without processing still-pending softirqs: | |
147 | */ | |
148 | void _local_bh_enable(void) | |
149 | { | |
5d60d3e7 | 150 | WARN_ON_ONCE(in_irq()); |
75e1056f | 151 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
152 | } |
153 | ||
154 | EXPORT_SYMBOL(_local_bh_enable); | |
155 | ||
0f476b6d | 156 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 157 | { |
0f476b6d | 158 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 159 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 160 | local_irq_disable(); |
3c829c36 | 161 | #endif |
de30a2b3 IM |
162 | /* |
163 | * Are softirqs going to be turned on now: | |
164 | */ | |
75e1056f | 165 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 166 | trace_softirqs_on(ip); |
de30a2b3 IM |
167 | /* |
168 | * Keep preemption disabled until we are done with | |
169 | * softirq processing: | |
170 | */ | |
75e1056f | 171 | sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1); |
de30a2b3 | 172 | |
0bed698a FW |
173 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
174 | /* | |
175 | * Run softirq if any pending. And do it in its own stack | |
176 | * as we may be calling this deep in a task call stack already. | |
177 | */ | |
de30a2b3 | 178 | do_softirq(); |
0bed698a | 179 | } |
de30a2b3 IM |
180 | |
181 | dec_preempt_count(); | |
3c829c36 | 182 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 183 | local_irq_enable(); |
3c829c36 | 184 | #endif |
de30a2b3 IM |
185 | preempt_check_resched(); |
186 | } | |
0f476b6d JB |
187 | |
188 | void local_bh_enable(void) | |
189 | { | |
d2e08473 | 190 | _local_bh_enable_ip(_RET_IP_); |
0f476b6d | 191 | } |
de30a2b3 IM |
192 | EXPORT_SYMBOL(local_bh_enable); |
193 | ||
194 | void local_bh_enable_ip(unsigned long ip) | |
195 | { | |
0f476b6d | 196 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
197 | } |
198 | EXPORT_SYMBOL(local_bh_enable_ip); | |
199 | ||
1da177e4 | 200 | /* |
34376a50 BG |
201 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
202 | * but break the loop if need_resched() is set or after 2 ms. | |
203 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
204 | * certain cases, such as stop_machine(), jiffies may cease to | |
205 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
206 | * well to make sure we eventually return from this method. | |
1da177e4 | 207 | * |
c10d7367 | 208 | * These limits have been established via experimentation. |
1da177e4 LT |
209 | * The two things to balance is latency against fairness - |
210 | * we want to handle softirqs as soon as possible, but they | |
211 | * should not be able to lock up the box. | |
212 | */ | |
c10d7367 | 213 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 214 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 LT |
215 | |
216 | asmlinkage void __do_softirq(void) | |
217 | { | |
218 | struct softirq_action *h; | |
219 | __u32 pending; | |
c10d7367 | 220 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
1da177e4 | 221 | int cpu; |
907aed48 | 222 | unsigned long old_flags = current->flags; |
34376a50 | 223 | int max_restart = MAX_SOFTIRQ_RESTART; |
907aed48 MG |
224 | |
225 | /* | |
226 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
227 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
228 | * again if the socket is related to swap | |
229 | */ | |
230 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
231 | |
232 | pending = local_softirq_pending(); | |
6a61671b | 233 | account_irq_enter_time(current); |
829035fd | 234 | |
d2e08473 | 235 | __local_bh_disable(_RET_IP_, SOFTIRQ_OFFSET); |
d820ac4c | 236 | lockdep_softirq_enter(); |
1da177e4 | 237 | |
1da177e4 LT |
238 | cpu = smp_processor_id(); |
239 | restart: | |
240 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 241 | set_softirq_pending(0); |
1da177e4 | 242 | |
c70f5d66 | 243 | local_irq_enable(); |
1da177e4 LT |
244 | |
245 | h = softirq_vec; | |
246 | ||
247 | do { | |
248 | if (pending & 1) { | |
f4bc6bb2 | 249 | unsigned int vec_nr = h - softirq_vec; |
8e85b4b5 TG |
250 | int prev_count = preempt_count(); |
251 | ||
f4bc6bb2 TG |
252 | kstat_incr_softirqs_this_cpu(vec_nr); |
253 | ||
254 | trace_softirq_entry(vec_nr); | |
1da177e4 | 255 | h->action(h); |
f4bc6bb2 | 256 | trace_softirq_exit(vec_nr); |
8e85b4b5 | 257 | if (unlikely(prev_count != preempt_count())) { |
f4bc6bb2 | 258 | printk(KERN_ERR "huh, entered softirq %u %s %p" |
8e85b4b5 | 259 | "with preempt_count %08x," |
f4bc6bb2 TG |
260 | " exited with %08x?\n", vec_nr, |
261 | softirq_to_name[vec_nr], h->action, | |
262 | prev_count, preempt_count()); | |
8e85b4b5 TG |
263 | preempt_count() = prev_count; |
264 | } | |
265 | ||
d6714c22 | 266 | rcu_bh_qs(cpu); |
1da177e4 LT |
267 | } |
268 | h++; | |
269 | pending >>= 1; | |
270 | } while (pending); | |
271 | ||
c70f5d66 | 272 | local_irq_disable(); |
1da177e4 LT |
273 | |
274 | pending = local_softirq_pending(); | |
c10d7367 | 275 | if (pending) { |
34376a50 BG |
276 | if (time_before(jiffies, end) && !need_resched() && |
277 | --max_restart) | |
c10d7367 | 278 | goto restart; |
1da177e4 | 279 | |
1da177e4 | 280 | wakeup_softirqd(); |
c10d7367 | 281 | } |
1da177e4 | 282 | |
d820ac4c | 283 | lockdep_softirq_exit(); |
829035fd | 284 | |
6a61671b | 285 | account_irq_exit_time(current); |
75e1056f | 286 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 287 | WARN_ON_ONCE(in_interrupt()); |
907aed48 | 288 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
1da177e4 LT |
289 | } |
290 | ||
7d65f4a6 | 291 | |
1da177e4 LT |
292 | |
293 | asmlinkage void do_softirq(void) | |
294 | { | |
295 | __u32 pending; | |
296 | unsigned long flags; | |
297 | ||
298 | if (in_interrupt()) | |
299 | return; | |
300 | ||
301 | local_irq_save(flags); | |
302 | ||
303 | pending = local_softirq_pending(); | |
304 | ||
305 | if (pending) | |
7d65f4a6 | 306 | do_softirq_own_stack(); |
1da177e4 LT |
307 | |
308 | local_irq_restore(flags); | |
309 | } | |
310 | ||
dde4b2b5 IM |
311 | /* |
312 | * Enter an interrupt context. | |
313 | */ | |
314 | void irq_enter(void) | |
315 | { | |
6378ddb5 | 316 | int cpu = smp_processor_id(); |
719254fa | 317 | |
64db4cff | 318 | rcu_irq_enter(); |
0a8a2e78 | 319 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
320 | /* |
321 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
322 | * here, as softirq will be serviced on return from interrupt. | |
323 | */ | |
324 | local_bh_disable(); | |
719254fa | 325 | tick_check_idle(cpu); |
d267f87f VP |
326 | _local_bh_enable(); |
327 | } | |
328 | ||
329 | __irq_enter(); | |
dde4b2b5 IM |
330 | } |
331 | ||
8d32a307 TG |
332 | static inline void invoke_softirq(void) |
333 | { | |
ded79754 FW |
334 | if (!force_irqthreads) { |
335 | /* | |
336 | * We can safely execute softirq on the current stack if | |
337 | * it is the irq stack, because it should be near empty | |
338 | * at this stage. But we have no way to know if the arch | |
339 | * calls irq_exit() on the irq stack. So call softirq | |
340 | * in its own stack to prevent from any overrun on top | |
341 | * of a potentially deep task stack. | |
342 | */ | |
be6e1016 | 343 | do_softirq_own_stack(); |
ded79754 | 344 | } else { |
8d32a307 | 345 | wakeup_softirqd(); |
ded79754 | 346 | } |
8d32a307 | 347 | } |
1da177e4 | 348 | |
67826eae FW |
349 | static inline void tick_irq_exit(void) |
350 | { | |
351 | #ifdef CONFIG_NO_HZ_COMMON | |
352 | int cpu = smp_processor_id(); | |
353 | ||
354 | /* Make sure that timer wheel updates are propagated */ | |
355 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
356 | if (!in_interrupt()) | |
357 | tick_nohz_irq_exit(); | |
358 | } | |
359 | #endif | |
360 | } | |
361 | ||
1da177e4 LT |
362 | /* |
363 | * Exit an interrupt context. Process softirqs if needed and possible: | |
364 | */ | |
365 | void irq_exit(void) | |
366 | { | |
74eed016 | 367 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 368 | local_irq_disable(); |
74eed016 TG |
369 | #else |
370 | WARN_ON_ONCE(!irqs_disabled()); | |
371 | #endif | |
372 | ||
6a61671b | 373 | account_irq_exit_time(current); |
de30a2b3 | 374 | trace_hardirq_exit(); |
4d4c4e24 | 375 | sub_preempt_count(HARDIRQ_OFFSET); |
1da177e4 LT |
376 | if (!in_interrupt() && local_softirq_pending()) |
377 | invoke_softirq(); | |
79bf2bb3 | 378 | |
67826eae | 379 | tick_irq_exit(); |
416eb33c | 380 | rcu_irq_exit(); |
1da177e4 LT |
381 | } |
382 | ||
383 | /* | |
384 | * This function must run with irqs disabled! | |
385 | */ | |
7ad5b3a5 | 386 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
387 | { |
388 | __raise_softirq_irqoff(nr); | |
389 | ||
390 | /* | |
391 | * If we're in an interrupt or softirq, we're done | |
392 | * (this also catches softirq-disabled code). We will | |
393 | * actually run the softirq once we return from | |
394 | * the irq or softirq. | |
395 | * | |
396 | * Otherwise we wake up ksoftirqd to make sure we | |
397 | * schedule the softirq soon. | |
398 | */ | |
399 | if (!in_interrupt()) | |
400 | wakeup_softirqd(); | |
401 | } | |
402 | ||
7ad5b3a5 | 403 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
404 | { |
405 | unsigned long flags; | |
406 | ||
407 | local_irq_save(flags); | |
408 | raise_softirq_irqoff(nr); | |
409 | local_irq_restore(flags); | |
410 | } | |
411 | ||
f069686e SR |
412 | void __raise_softirq_irqoff(unsigned int nr) |
413 | { | |
414 | trace_softirq_raise(nr); | |
415 | or_softirq_pending(1UL << nr); | |
416 | } | |
417 | ||
962cf36c | 418 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 419 | { |
1da177e4 LT |
420 | softirq_vec[nr].action = action; |
421 | } | |
422 | ||
9ba5f005 PZ |
423 | /* |
424 | * Tasklets | |
425 | */ | |
1da177e4 LT |
426 | struct tasklet_head |
427 | { | |
48f20a9a OJ |
428 | struct tasklet_struct *head; |
429 | struct tasklet_struct **tail; | |
1da177e4 LT |
430 | }; |
431 | ||
4620b49f VN |
432 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
433 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 434 | |
7ad5b3a5 | 435 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
436 | { |
437 | unsigned long flags; | |
438 | ||
439 | local_irq_save(flags); | |
48f20a9a | 440 | t->next = NULL; |
909ea964 CL |
441 | *__this_cpu_read(tasklet_vec.tail) = t; |
442 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
443 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
444 | local_irq_restore(flags); | |
445 | } | |
446 | ||
447 | EXPORT_SYMBOL(__tasklet_schedule); | |
448 | ||
7ad5b3a5 | 449 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
450 | { |
451 | unsigned long flags; | |
452 | ||
453 | local_irq_save(flags); | |
48f20a9a | 454 | t->next = NULL; |
909ea964 CL |
455 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
456 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
457 | raise_softirq_irqoff(HI_SOFTIRQ); |
458 | local_irq_restore(flags); | |
459 | } | |
460 | ||
461 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
462 | ||
7c692cba VN |
463 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
464 | { | |
465 | BUG_ON(!irqs_disabled()); | |
466 | ||
909ea964 CL |
467 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
468 | __this_cpu_write(tasklet_hi_vec.head, t); | |
7c692cba VN |
469 | __raise_softirq_irqoff(HI_SOFTIRQ); |
470 | } | |
471 | ||
472 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); | |
473 | ||
1da177e4 LT |
474 | static void tasklet_action(struct softirq_action *a) |
475 | { | |
476 | struct tasklet_struct *list; | |
477 | ||
478 | local_irq_disable(); | |
909ea964 CL |
479 | list = __this_cpu_read(tasklet_vec.head); |
480 | __this_cpu_write(tasklet_vec.head, NULL); | |
481 | __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head); | |
1da177e4 LT |
482 | local_irq_enable(); |
483 | ||
484 | while (list) { | |
485 | struct tasklet_struct *t = list; | |
486 | ||
487 | list = list->next; | |
488 | ||
489 | if (tasklet_trylock(t)) { | |
490 | if (!atomic_read(&t->count)) { | |
491 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
492 | BUG(); | |
493 | t->func(t->data); | |
494 | tasklet_unlock(t); | |
495 | continue; | |
496 | } | |
497 | tasklet_unlock(t); | |
498 | } | |
499 | ||
500 | local_irq_disable(); | |
48f20a9a | 501 | t->next = NULL; |
909ea964 CL |
502 | *__this_cpu_read(tasklet_vec.tail) = t; |
503 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
504 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
505 | local_irq_enable(); | |
506 | } | |
507 | } | |
508 | ||
509 | static void tasklet_hi_action(struct softirq_action *a) | |
510 | { | |
511 | struct tasklet_struct *list; | |
512 | ||
513 | local_irq_disable(); | |
909ea964 CL |
514 | list = __this_cpu_read(tasklet_hi_vec.head); |
515 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
516 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
517 | local_irq_enable(); |
518 | ||
519 | while (list) { | |
520 | struct tasklet_struct *t = list; | |
521 | ||
522 | list = list->next; | |
523 | ||
524 | if (tasklet_trylock(t)) { | |
525 | if (!atomic_read(&t->count)) { | |
526 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
527 | BUG(); | |
528 | t->func(t->data); | |
529 | tasklet_unlock(t); | |
530 | continue; | |
531 | } | |
532 | tasklet_unlock(t); | |
533 | } | |
534 | ||
535 | local_irq_disable(); | |
48f20a9a | 536 | t->next = NULL; |
909ea964 CL |
537 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
538 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
539 | __raise_softirq_irqoff(HI_SOFTIRQ); |
540 | local_irq_enable(); | |
541 | } | |
542 | } | |
543 | ||
544 | ||
545 | void tasklet_init(struct tasklet_struct *t, | |
546 | void (*func)(unsigned long), unsigned long data) | |
547 | { | |
548 | t->next = NULL; | |
549 | t->state = 0; | |
550 | atomic_set(&t->count, 0); | |
551 | t->func = func; | |
552 | t->data = data; | |
553 | } | |
554 | ||
555 | EXPORT_SYMBOL(tasklet_init); | |
556 | ||
557 | void tasklet_kill(struct tasklet_struct *t) | |
558 | { | |
559 | if (in_interrupt()) | |
560 | printk("Attempt to kill tasklet from interrupt\n"); | |
561 | ||
562 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 563 | do { |
1da177e4 | 564 | yield(); |
79d381c9 | 565 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
566 | } |
567 | tasklet_unlock_wait(t); | |
568 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
569 | } | |
570 | ||
571 | EXPORT_SYMBOL(tasklet_kill); | |
572 | ||
9ba5f005 PZ |
573 | /* |
574 | * tasklet_hrtimer | |
575 | */ | |
576 | ||
577 | /* | |
b9c30322 PZ |
578 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
579 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
580 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
581 | */ |
582 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
583 | { | |
584 | struct tasklet_hrtimer *ttimer = | |
585 | container_of(timer, struct tasklet_hrtimer, timer); | |
586 | ||
b9c30322 PZ |
587 | tasklet_hi_schedule(&ttimer->tasklet); |
588 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
589 | } |
590 | ||
591 | /* | |
592 | * Helper function which calls the hrtimer callback from | |
593 | * tasklet/softirq context | |
594 | */ | |
595 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
596 | { | |
597 | struct tasklet_hrtimer *ttimer = (void *)data; | |
598 | enum hrtimer_restart restart; | |
599 | ||
600 | restart = ttimer->function(&ttimer->timer); | |
601 | if (restart != HRTIMER_NORESTART) | |
602 | hrtimer_restart(&ttimer->timer); | |
603 | } | |
604 | ||
605 | /** | |
606 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
607 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 608 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
609 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
610 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
611 | */ | |
612 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
613 | enum hrtimer_restart (*function)(struct hrtimer *), | |
614 | clockid_t which_clock, enum hrtimer_mode mode) | |
615 | { | |
616 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
617 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
618 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
619 | (unsigned long)ttimer); | |
620 | ttimer->function = function; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
623 | ||
624 | /* | |
625 | * Remote softirq bits | |
626 | */ | |
627 | ||
54514a70 DM |
628 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
629 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
630 | ||
631 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
632 | { | |
633 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
634 | ||
635 | list_add_tail(&cp->list, head); | |
636 | ||
637 | /* Trigger the softirq only if the list was previously empty. */ | |
638 | if (head->next == &cp->list) | |
639 | raise_softirq_irqoff(softirq); | |
640 | } | |
641 | ||
642 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
643 | static void remote_softirq_receive(void *data) | |
644 | { | |
645 | struct call_single_data *cp = data; | |
646 | unsigned long flags; | |
647 | int softirq; | |
648 | ||
3440a1ca | 649 | softirq = *(int *)cp->info; |
54514a70 DM |
650 | local_irq_save(flags); |
651 | __local_trigger(cp, softirq); | |
652 | local_irq_restore(flags); | |
653 | } | |
654 | ||
655 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
656 | { | |
657 | if (cpu_online(cpu)) { | |
658 | cp->func = remote_softirq_receive; | |
3440a1ca | 659 | cp->info = &softirq; |
54514a70 | 660 | cp->flags = 0; |
54514a70 | 661 | |
6e275637 | 662 | __smp_call_function_single(cpu, cp, 0); |
54514a70 DM |
663 | return 0; |
664 | } | |
665 | return 1; | |
666 | } | |
667 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
668 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
669 | { | |
670 | return 1; | |
671 | } | |
672 | #endif | |
673 | ||
674 | /** | |
675 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
676 | * @cp: private SMP call function data area | |
677 | * @cpu: the remote cpu | |
678 | * @this_cpu: the currently executing cpu | |
679 | * @softirq: the softirq for the work | |
680 | * | |
681 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
682 | * done, the work is instead queued up on the local cpu. | |
683 | * | |
684 | * Interrupts must be disabled. | |
685 | */ | |
686 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
687 | { | |
688 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
689 | __local_trigger(cp, softirq); | |
690 | } | |
691 | EXPORT_SYMBOL(__send_remote_softirq); | |
692 | ||
693 | /** | |
694 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
695 | * @cp: private SMP call function data area | |
696 | * @cpu: the remote cpu | |
697 | * @softirq: the softirq for the work | |
698 | * | |
699 | * Like __send_remote_softirq except that disabling interrupts and | |
700 | * computing the current cpu is done for the caller. | |
701 | */ | |
702 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
703 | { | |
704 | unsigned long flags; | |
705 | int this_cpu; | |
706 | ||
707 | local_irq_save(flags); | |
708 | this_cpu = smp_processor_id(); | |
709 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
710 | local_irq_restore(flags); | |
711 | } | |
712 | EXPORT_SYMBOL(send_remote_softirq); | |
713 | ||
0db0628d | 714 | static int remote_softirq_cpu_notify(struct notifier_block *self, |
54514a70 DM |
715 | unsigned long action, void *hcpu) |
716 | { | |
717 | /* | |
718 | * If a CPU goes away, splice its entries to the current CPU | |
719 | * and trigger a run of the softirq | |
720 | */ | |
721 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
722 | int cpu = (unsigned long) hcpu; | |
723 | int i; | |
724 | ||
725 | local_irq_disable(); | |
726 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
727 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
728 | struct list_head *local_head; | |
729 | ||
730 | if (list_empty(head)) | |
731 | continue; | |
732 | ||
733 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
734 | list_splice_init(head, local_head); | |
735 | raise_softirq_irqoff(i); | |
736 | } | |
737 | local_irq_enable(); | |
738 | } | |
739 | ||
740 | return NOTIFY_OK; | |
741 | } | |
742 | ||
0db0628d | 743 | static struct notifier_block remote_softirq_cpu_notifier = { |
54514a70 DM |
744 | .notifier_call = remote_softirq_cpu_notify, |
745 | }; | |
746 | ||
1da177e4 LT |
747 | void __init softirq_init(void) |
748 | { | |
48f20a9a OJ |
749 | int cpu; |
750 | ||
751 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
752 | int i; |
753 | ||
48f20a9a OJ |
754 | per_cpu(tasklet_vec, cpu).tail = |
755 | &per_cpu(tasklet_vec, cpu).head; | |
756 | per_cpu(tasklet_hi_vec, cpu).tail = | |
757 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
758 | for (i = 0; i < NR_SOFTIRQS; i++) |
759 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
760 | } |
761 | ||
54514a70 DM |
762 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
763 | ||
962cf36c CM |
764 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
765 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
766 | } |
767 | ||
3e339b5d | 768 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 769 | { |
3e339b5d TG |
770 | return local_softirq_pending(); |
771 | } | |
1da177e4 | 772 | |
3e339b5d TG |
773 | static void run_ksoftirqd(unsigned int cpu) |
774 | { | |
775 | local_irq_disable(); | |
776 | if (local_softirq_pending()) { | |
0bed698a FW |
777 | /* |
778 | * We can safely run softirq on inline stack, as we are not deep | |
779 | * in the task stack here. | |
780 | */ | |
3e339b5d TG |
781 | __do_softirq(); |
782 | rcu_note_context_switch(cpu); | |
783 | local_irq_enable(); | |
784 | cond_resched(); | |
785 | return; | |
1da177e4 | 786 | } |
3e339b5d | 787 | local_irq_enable(); |
1da177e4 LT |
788 | } |
789 | ||
790 | #ifdef CONFIG_HOTPLUG_CPU | |
791 | /* | |
792 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
793 | * scheduled for execution on @cpu. | |
794 | * | |
795 | * Unlike tasklet_kill, this function removes the tasklet | |
796 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
797 | * | |
798 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
799 | */ | |
800 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
801 | { | |
802 | struct tasklet_struct **i; | |
803 | ||
804 | BUG_ON(cpu_online(cpu)); | |
805 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
806 | ||
807 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
808 | return; | |
809 | ||
810 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 811 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
812 | if (*i == t) { |
813 | *i = t->next; | |
48f20a9a OJ |
814 | /* If this was the tail element, move the tail ptr */ |
815 | if (*i == NULL) | |
816 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
817 | return; |
818 | } | |
819 | } | |
820 | BUG(); | |
821 | } | |
822 | ||
823 | static void takeover_tasklets(unsigned int cpu) | |
824 | { | |
1da177e4 LT |
825 | /* CPU is dead, so no lock needed. */ |
826 | local_irq_disable(); | |
827 | ||
828 | /* Find end, append list for that CPU. */ | |
e5e41723 | 829 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
830 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
831 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
832 | per_cpu(tasklet_vec, cpu).head = NULL; |
833 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
834 | } | |
1da177e4 LT |
835 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
836 | ||
e5e41723 | 837 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
838 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
839 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
840 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
841 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
842 | } | |
1da177e4 LT |
843 | raise_softirq_irqoff(HI_SOFTIRQ); |
844 | ||
845 | local_irq_enable(); | |
846 | } | |
847 | #endif /* CONFIG_HOTPLUG_CPU */ | |
848 | ||
0db0628d | 849 | static int cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
850 | unsigned long action, |
851 | void *hcpu) | |
852 | { | |
1da177e4 | 853 | switch (action) { |
1da177e4 | 854 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 855 | case CPU_DEAD: |
3e339b5d TG |
856 | case CPU_DEAD_FROZEN: |
857 | takeover_tasklets((unsigned long)hcpu); | |
1da177e4 LT |
858 | break; |
859 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3e339b5d | 860 | } |
1da177e4 LT |
861 | return NOTIFY_OK; |
862 | } | |
863 | ||
0db0628d | 864 | static struct notifier_block cpu_nfb = { |
1da177e4 LT |
865 | .notifier_call = cpu_callback |
866 | }; | |
867 | ||
3e339b5d TG |
868 | static struct smp_hotplug_thread softirq_threads = { |
869 | .store = &ksoftirqd, | |
870 | .thread_should_run = ksoftirqd_should_run, | |
871 | .thread_fn = run_ksoftirqd, | |
872 | .thread_comm = "ksoftirqd/%u", | |
873 | }; | |
874 | ||
7babe8db | 875 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 876 | { |
1da177e4 | 877 | register_cpu_notifier(&cpu_nfb); |
3e339b5d TG |
878 | |
879 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); | |
880 | ||
1da177e4 LT |
881 | return 0; |
882 | } | |
7babe8db | 883 | early_initcall(spawn_ksoftirqd); |
78eef01b | 884 | |
43a25632 YL |
885 | /* |
886 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
887 | * GCC does not inline them incorrectly. ] | |
888 | */ | |
889 | ||
890 | int __init __weak early_irq_init(void) | |
891 | { | |
892 | return 0; | |
893 | } | |
894 | ||
4a046d17 YL |
895 | int __init __weak arch_probe_nr_irqs(void) |
896 | { | |
b683de2b | 897 | return NR_IRQS_LEGACY; |
4a046d17 YL |
898 | } |
899 | ||
43a25632 YL |
900 | int __init __weak arch_early_irq_init(void) |
901 | { | |
902 | return 0; | |
903 | } |