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 LT |
31 | |
32 | #include <asm/irq.h> | |
33 | /* | |
34 | - No shared variables, all the data are CPU local. | |
35 | - If a softirq needs serialization, let it serialize itself | |
36 | by its own spinlocks. | |
37 | - Even if softirq is serialized, only local cpu is marked for | |
38 | execution. Hence, we get something sort of weak cpu binding. | |
39 | Though it is still not clear, will it result in better locality | |
40 | or will not. | |
41 | ||
42 | Examples: | |
43 | - NET RX softirq. It is multithreaded and does not require | |
44 | any global serialization. | |
45 | - NET TX softirq. It kicks software netdevice queues, hence | |
46 | it is logically serialized per device, but this serialization | |
47 | is invisible to common code. | |
48 | - Tasklets: serialized wrt itself. | |
49 | */ | |
50 | ||
51 | #ifndef __ARCH_IRQ_STAT | |
52 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
53 | EXPORT_SYMBOL(irq_stat); | |
54 | #endif | |
55 | ||
978b0116 | 56 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 57 | |
4dd53d89 | 58 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 59 | |
5d592b44 | 60 | char *softirq_to_name[NR_SOFTIRQS] = { |
5dd4de58 | 61 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL", |
09223371 | 62 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
5d592b44 JB |
63 | }; |
64 | ||
1da177e4 LT |
65 | /* |
66 | * we cannot loop indefinitely here to avoid userspace starvation, | |
67 | * but we also don't want to introduce a worst case 1/HZ latency | |
68 | * to the pending events, so lets the scheduler to balance | |
69 | * the softirq load for us. | |
70 | */ | |
676cb02d | 71 | static void wakeup_softirqd(void) |
1da177e4 LT |
72 | { |
73 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 74 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 LT |
75 | |
76 | if (tsk && tsk->state != TASK_RUNNING) | |
77 | wake_up_process(tsk); | |
78 | } | |
79 | ||
75e1056f VP |
80 | /* |
81 | * preempt_count and SOFTIRQ_OFFSET usage: | |
82 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving | |
83 | * softirq processing. | |
84 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
85 | * on local_bh_disable or local_bh_enable. | |
86 | * This lets us distinguish between whether we are currently processing | |
87 | * softirq and whether we just have bh disabled. | |
88 | */ | |
89 | ||
de30a2b3 IM |
90 | /* |
91 | * This one is for softirq.c-internal use, | |
92 | * where hardirqs are disabled legitimately: | |
93 | */ | |
3c829c36 | 94 | #ifdef CONFIG_TRACE_IRQFLAGS |
75e1056f | 95 | static void __local_bh_disable(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
96 | { |
97 | unsigned long flags; | |
98 | ||
99 | WARN_ON_ONCE(in_irq()); | |
100 | ||
101 | raw_local_irq_save(flags); | |
7e49fcce SR |
102 | /* |
103 | * The preempt tracer hooks into add_preempt_count and will break | |
104 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET | |
105 | * is set and before current->softirq_enabled is cleared. | |
106 | * We must manually increment preempt_count here and manually | |
107 | * call the trace_preempt_off later. | |
108 | */ | |
75e1056f | 109 | preempt_count() += cnt; |
de30a2b3 IM |
110 | /* |
111 | * Were softirqs turned off above: | |
112 | */ | |
75e1056f | 113 | if (softirq_count() == cnt) |
de30a2b3 IM |
114 | trace_softirqs_off(ip); |
115 | raw_local_irq_restore(flags); | |
7e49fcce | 116 | |
75e1056f | 117 | if (preempt_count() == cnt) |
7e49fcce | 118 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); |
de30a2b3 | 119 | } |
3c829c36 | 120 | #else /* !CONFIG_TRACE_IRQFLAGS */ |
75e1056f | 121 | static inline void __local_bh_disable(unsigned long ip, unsigned int cnt) |
3c829c36 | 122 | { |
75e1056f | 123 | add_preempt_count(cnt); |
3c829c36 TC |
124 | barrier(); |
125 | } | |
126 | #endif /* CONFIG_TRACE_IRQFLAGS */ | |
de30a2b3 IM |
127 | |
128 | void local_bh_disable(void) | |
129 | { | |
75e1056f VP |
130 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
131 | SOFTIRQ_DISABLE_OFFSET); | |
de30a2b3 IM |
132 | } |
133 | ||
134 | EXPORT_SYMBOL(local_bh_disable); | |
135 | ||
75e1056f VP |
136 | static void __local_bh_enable(unsigned int cnt) |
137 | { | |
138 | WARN_ON_ONCE(in_irq()); | |
139 | WARN_ON_ONCE(!irqs_disabled()); | |
140 | ||
141 | if (softirq_count() == cnt) | |
142 | trace_softirqs_on((unsigned long)__builtin_return_address(0)); | |
143 | sub_preempt_count(cnt); | |
144 | } | |
145 | ||
de30a2b3 IM |
146 | /* |
147 | * Special-case - softirqs can safely be enabled in | |
148 | * cond_resched_softirq(), or by __do_softirq(), | |
149 | * without processing still-pending softirqs: | |
150 | */ | |
151 | void _local_bh_enable(void) | |
152 | { | |
75e1056f | 153 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
154 | } |
155 | ||
156 | EXPORT_SYMBOL(_local_bh_enable); | |
157 | ||
0f476b6d | 158 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 159 | { |
0f476b6d | 160 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 161 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 162 | local_irq_disable(); |
3c829c36 | 163 | #endif |
de30a2b3 IM |
164 | /* |
165 | * Are softirqs going to be turned on now: | |
166 | */ | |
75e1056f | 167 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 168 | trace_softirqs_on(ip); |
de30a2b3 IM |
169 | /* |
170 | * Keep preemption disabled until we are done with | |
171 | * softirq processing: | |
172 | */ | |
75e1056f | 173 | sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1); |
de30a2b3 IM |
174 | |
175 | if (unlikely(!in_interrupt() && local_softirq_pending())) | |
176 | do_softirq(); | |
177 | ||
178 | dec_preempt_count(); | |
3c829c36 | 179 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 180 | local_irq_enable(); |
3c829c36 | 181 | #endif |
de30a2b3 IM |
182 | preempt_check_resched(); |
183 | } | |
0f476b6d JB |
184 | |
185 | void local_bh_enable(void) | |
186 | { | |
187 | _local_bh_enable_ip((unsigned long)__builtin_return_address(0)); | |
188 | } | |
de30a2b3 IM |
189 | EXPORT_SYMBOL(local_bh_enable); |
190 | ||
191 | void local_bh_enable_ip(unsigned long ip) | |
192 | { | |
0f476b6d | 193 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
194 | } |
195 | EXPORT_SYMBOL(local_bh_enable_ip); | |
196 | ||
1da177e4 | 197 | /* |
34376a50 BG |
198 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
199 | * but break the loop if need_resched() is set or after 2 ms. | |
200 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
201 | * certain cases, such as stop_machine(), jiffies may cease to | |
202 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
203 | * well to make sure we eventually return from this method. | |
1da177e4 | 204 | * |
c10d7367 | 205 | * These limits have been established via experimentation. |
1da177e4 LT |
206 | * The two things to balance is latency against fairness - |
207 | * we want to handle softirqs as soon as possible, but they | |
208 | * should not be able to lock up the box. | |
209 | */ | |
c10d7367 | 210 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 211 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 LT |
212 | |
213 | asmlinkage void __do_softirq(void) | |
214 | { | |
215 | struct softirq_action *h; | |
216 | __u32 pending; | |
c10d7367 | 217 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
1da177e4 | 218 | int cpu; |
907aed48 | 219 | unsigned long old_flags = current->flags; |
34376a50 | 220 | int max_restart = MAX_SOFTIRQ_RESTART; |
907aed48 MG |
221 | |
222 | /* | |
223 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
224 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
225 | * again if the socket is related to swap | |
226 | */ | |
227 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
228 | |
229 | pending = local_softirq_pending(); | |
6a61671b | 230 | account_irq_enter_time(current); |
829035fd | 231 | |
75e1056f VP |
232 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
233 | SOFTIRQ_OFFSET); | |
d820ac4c | 234 | lockdep_softirq_enter(); |
1da177e4 | 235 | |
1da177e4 LT |
236 | cpu = smp_processor_id(); |
237 | restart: | |
238 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 239 | set_softirq_pending(0); |
1da177e4 | 240 | |
c70f5d66 | 241 | local_irq_enable(); |
1da177e4 LT |
242 | |
243 | h = softirq_vec; | |
244 | ||
245 | do { | |
246 | if (pending & 1) { | |
f4bc6bb2 | 247 | unsigned int vec_nr = h - softirq_vec; |
8e85b4b5 TG |
248 | int prev_count = preempt_count(); |
249 | ||
f4bc6bb2 TG |
250 | kstat_incr_softirqs_this_cpu(vec_nr); |
251 | ||
252 | trace_softirq_entry(vec_nr); | |
1da177e4 | 253 | h->action(h); |
f4bc6bb2 | 254 | trace_softirq_exit(vec_nr); |
8e85b4b5 | 255 | if (unlikely(prev_count != preempt_count())) { |
f4bc6bb2 | 256 | printk(KERN_ERR "huh, entered softirq %u %s %p" |
8e85b4b5 | 257 | "with preempt_count %08x," |
f4bc6bb2 TG |
258 | " exited with %08x?\n", vec_nr, |
259 | softirq_to_name[vec_nr], h->action, | |
260 | prev_count, preempt_count()); | |
8e85b4b5 TG |
261 | preempt_count() = prev_count; |
262 | } | |
263 | ||
d6714c22 | 264 | rcu_bh_qs(cpu); |
1da177e4 LT |
265 | } |
266 | h++; | |
267 | pending >>= 1; | |
268 | } while (pending); | |
269 | ||
c70f5d66 | 270 | local_irq_disable(); |
1da177e4 LT |
271 | |
272 | pending = local_softirq_pending(); | |
c10d7367 | 273 | if (pending) { |
34376a50 BG |
274 | if (time_before(jiffies, end) && !need_resched() && |
275 | --max_restart) | |
c10d7367 | 276 | goto restart; |
1da177e4 | 277 | |
1da177e4 | 278 | wakeup_softirqd(); |
c10d7367 | 279 | } |
1da177e4 | 280 | |
d820ac4c | 281 | lockdep_softirq_exit(); |
829035fd | 282 | |
6a61671b | 283 | account_irq_exit_time(current); |
75e1056f | 284 | __local_bh_enable(SOFTIRQ_OFFSET); |
907aed48 | 285 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
1da177e4 LT |
286 | } |
287 | ||
288 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
289 | ||
290 | asmlinkage void do_softirq(void) | |
291 | { | |
292 | __u32 pending; | |
293 | unsigned long flags; | |
294 | ||
295 | if (in_interrupt()) | |
296 | return; | |
297 | ||
298 | local_irq_save(flags); | |
299 | ||
300 | pending = local_softirq_pending(); | |
301 | ||
302 | if (pending) | |
303 | __do_softirq(); | |
304 | ||
305 | local_irq_restore(flags); | |
306 | } | |
307 | ||
1da177e4 LT |
308 | #endif |
309 | ||
dde4b2b5 IM |
310 | /* |
311 | * Enter an interrupt context. | |
312 | */ | |
313 | void irq_enter(void) | |
314 | { | |
6378ddb5 | 315 | int cpu = smp_processor_id(); |
719254fa | 316 | |
64db4cff | 317 | rcu_irq_enter(); |
0a8a2e78 | 318 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
319 | /* |
320 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
321 | * here, as softirq will be serviced on return from interrupt. | |
322 | */ | |
323 | local_bh_disable(); | |
719254fa | 324 | tick_check_idle(cpu); |
d267f87f VP |
325 | _local_bh_enable(); |
326 | } | |
327 | ||
328 | __irq_enter(); | |
dde4b2b5 IM |
329 | } |
330 | ||
8d32a307 TG |
331 | static inline void invoke_softirq(void) |
332 | { | |
facd8b80 | 333 | if (!force_irqthreads) |
8d32a307 | 334 | __do_softirq(); |
facd8b80 | 335 | else |
8d32a307 TG |
336 | wakeup_softirqd(); |
337 | } | |
1da177e4 | 338 | |
67826eae FW |
339 | static inline void tick_irq_exit(void) |
340 | { | |
341 | #ifdef CONFIG_NO_HZ_COMMON | |
342 | int cpu = smp_processor_id(); | |
343 | ||
344 | /* Make sure that timer wheel updates are propagated */ | |
345 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
346 | if (!in_interrupt()) | |
347 | tick_nohz_irq_exit(); | |
348 | } | |
349 | #endif | |
350 | } | |
351 | ||
1da177e4 LT |
352 | /* |
353 | * Exit an interrupt context. Process softirqs if needed and possible: | |
354 | */ | |
355 | void irq_exit(void) | |
356 | { | |
74eed016 | 357 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 358 | local_irq_disable(); |
74eed016 TG |
359 | #else |
360 | WARN_ON_ONCE(!irqs_disabled()); | |
361 | #endif | |
362 | ||
6a61671b | 363 | account_irq_exit_time(current); |
de30a2b3 | 364 | trace_hardirq_exit(); |
4d4c4e24 | 365 | sub_preempt_count(HARDIRQ_OFFSET); |
1da177e4 LT |
366 | if (!in_interrupt() && local_softirq_pending()) |
367 | invoke_softirq(); | |
79bf2bb3 | 368 | |
67826eae | 369 | tick_irq_exit(); |
416eb33c | 370 | rcu_irq_exit(); |
1da177e4 LT |
371 | } |
372 | ||
373 | /* | |
374 | * This function must run with irqs disabled! | |
375 | */ | |
7ad5b3a5 | 376 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
377 | { |
378 | __raise_softirq_irqoff(nr); | |
379 | ||
380 | /* | |
381 | * If we're in an interrupt or softirq, we're done | |
382 | * (this also catches softirq-disabled code). We will | |
383 | * actually run the softirq once we return from | |
384 | * the irq or softirq. | |
385 | * | |
386 | * Otherwise we wake up ksoftirqd to make sure we | |
387 | * schedule the softirq soon. | |
388 | */ | |
389 | if (!in_interrupt()) | |
390 | wakeup_softirqd(); | |
391 | } | |
392 | ||
7ad5b3a5 | 393 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
394 | { |
395 | unsigned long flags; | |
396 | ||
397 | local_irq_save(flags); | |
398 | raise_softirq_irqoff(nr); | |
399 | local_irq_restore(flags); | |
400 | } | |
401 | ||
f069686e SR |
402 | void __raise_softirq_irqoff(unsigned int nr) |
403 | { | |
404 | trace_softirq_raise(nr); | |
405 | or_softirq_pending(1UL << nr); | |
406 | } | |
407 | ||
962cf36c | 408 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 409 | { |
1da177e4 LT |
410 | softirq_vec[nr].action = action; |
411 | } | |
412 | ||
9ba5f005 PZ |
413 | /* |
414 | * Tasklets | |
415 | */ | |
1da177e4 LT |
416 | struct tasklet_head |
417 | { | |
48f20a9a OJ |
418 | struct tasklet_struct *head; |
419 | struct tasklet_struct **tail; | |
1da177e4 LT |
420 | }; |
421 | ||
4620b49f VN |
422 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
423 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 424 | |
7ad5b3a5 | 425 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
426 | { |
427 | unsigned long flags; | |
428 | ||
429 | local_irq_save(flags); | |
48f20a9a | 430 | t->next = NULL; |
909ea964 CL |
431 | *__this_cpu_read(tasklet_vec.tail) = t; |
432 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
433 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
434 | local_irq_restore(flags); | |
435 | } | |
436 | ||
437 | EXPORT_SYMBOL(__tasklet_schedule); | |
438 | ||
7ad5b3a5 | 439 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
440 | { |
441 | unsigned long flags; | |
442 | ||
443 | local_irq_save(flags); | |
48f20a9a | 444 | t->next = NULL; |
909ea964 CL |
445 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
446 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
447 | raise_softirq_irqoff(HI_SOFTIRQ); |
448 | local_irq_restore(flags); | |
449 | } | |
450 | ||
451 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
452 | ||
7c692cba VN |
453 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
454 | { | |
455 | BUG_ON(!irqs_disabled()); | |
456 | ||
909ea964 CL |
457 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
458 | __this_cpu_write(tasklet_hi_vec.head, t); | |
7c692cba VN |
459 | __raise_softirq_irqoff(HI_SOFTIRQ); |
460 | } | |
461 | ||
462 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); | |
463 | ||
1da177e4 LT |
464 | static void tasklet_action(struct softirq_action *a) |
465 | { | |
466 | struct tasklet_struct *list; | |
467 | ||
468 | local_irq_disable(); | |
909ea964 CL |
469 | list = __this_cpu_read(tasklet_vec.head); |
470 | __this_cpu_write(tasklet_vec.head, NULL); | |
471 | __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head); | |
1da177e4 LT |
472 | local_irq_enable(); |
473 | ||
474 | while (list) { | |
475 | struct tasklet_struct *t = list; | |
476 | ||
477 | list = list->next; | |
478 | ||
479 | if (tasklet_trylock(t)) { | |
480 | if (!atomic_read(&t->count)) { | |
481 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
482 | BUG(); | |
483 | t->func(t->data); | |
484 | tasklet_unlock(t); | |
485 | continue; | |
486 | } | |
487 | tasklet_unlock(t); | |
488 | } | |
489 | ||
490 | local_irq_disable(); | |
48f20a9a | 491 | t->next = NULL; |
909ea964 CL |
492 | *__this_cpu_read(tasklet_vec.tail) = t; |
493 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
494 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
495 | local_irq_enable(); | |
496 | } | |
497 | } | |
498 | ||
499 | static void tasklet_hi_action(struct softirq_action *a) | |
500 | { | |
501 | struct tasklet_struct *list; | |
502 | ||
503 | local_irq_disable(); | |
909ea964 CL |
504 | list = __this_cpu_read(tasklet_hi_vec.head); |
505 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
506 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
507 | local_irq_enable(); |
508 | ||
509 | while (list) { | |
510 | struct tasklet_struct *t = list; | |
511 | ||
512 | list = list->next; | |
513 | ||
514 | if (tasklet_trylock(t)) { | |
515 | if (!atomic_read(&t->count)) { | |
516 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
517 | BUG(); | |
518 | t->func(t->data); | |
519 | tasklet_unlock(t); | |
520 | continue; | |
521 | } | |
522 | tasklet_unlock(t); | |
523 | } | |
524 | ||
525 | local_irq_disable(); | |
48f20a9a | 526 | t->next = NULL; |
909ea964 CL |
527 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
528 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
529 | __raise_softirq_irqoff(HI_SOFTIRQ); |
530 | local_irq_enable(); | |
531 | } | |
532 | } | |
533 | ||
534 | ||
535 | void tasklet_init(struct tasklet_struct *t, | |
536 | void (*func)(unsigned long), unsigned long data) | |
537 | { | |
538 | t->next = NULL; | |
539 | t->state = 0; | |
540 | atomic_set(&t->count, 0); | |
541 | t->func = func; | |
542 | t->data = data; | |
543 | } | |
544 | ||
545 | EXPORT_SYMBOL(tasklet_init); | |
546 | ||
547 | void tasklet_kill(struct tasklet_struct *t) | |
548 | { | |
549 | if (in_interrupt()) | |
550 | printk("Attempt to kill tasklet from interrupt\n"); | |
551 | ||
552 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 553 | do { |
1da177e4 | 554 | yield(); |
79d381c9 | 555 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
556 | } |
557 | tasklet_unlock_wait(t); | |
558 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
559 | } | |
560 | ||
561 | EXPORT_SYMBOL(tasklet_kill); | |
562 | ||
9ba5f005 PZ |
563 | /* |
564 | * tasklet_hrtimer | |
565 | */ | |
566 | ||
567 | /* | |
b9c30322 PZ |
568 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
569 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
570 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
571 | */ |
572 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
573 | { | |
574 | struct tasklet_hrtimer *ttimer = | |
575 | container_of(timer, struct tasklet_hrtimer, timer); | |
576 | ||
b9c30322 PZ |
577 | tasklet_hi_schedule(&ttimer->tasklet); |
578 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
579 | } |
580 | ||
581 | /* | |
582 | * Helper function which calls the hrtimer callback from | |
583 | * tasklet/softirq context | |
584 | */ | |
585 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
586 | { | |
587 | struct tasklet_hrtimer *ttimer = (void *)data; | |
588 | enum hrtimer_restart restart; | |
589 | ||
590 | restart = ttimer->function(&ttimer->timer); | |
591 | if (restart != HRTIMER_NORESTART) | |
592 | hrtimer_restart(&ttimer->timer); | |
593 | } | |
594 | ||
595 | /** | |
596 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
597 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 598 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
599 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
600 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
601 | */ | |
602 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
603 | enum hrtimer_restart (*function)(struct hrtimer *), | |
604 | clockid_t which_clock, enum hrtimer_mode mode) | |
605 | { | |
606 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
607 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
608 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
609 | (unsigned long)ttimer); | |
610 | ttimer->function = function; | |
611 | } | |
612 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
613 | ||
614 | /* | |
615 | * Remote softirq bits | |
616 | */ | |
617 | ||
54514a70 DM |
618 | DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list); |
619 | EXPORT_PER_CPU_SYMBOL(softirq_work_list); | |
620 | ||
621 | static void __local_trigger(struct call_single_data *cp, int softirq) | |
622 | { | |
623 | struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]); | |
624 | ||
625 | list_add_tail(&cp->list, head); | |
626 | ||
627 | /* Trigger the softirq only if the list was previously empty. */ | |
628 | if (head->next == &cp->list) | |
629 | raise_softirq_irqoff(softirq); | |
630 | } | |
631 | ||
632 | #ifdef CONFIG_USE_GENERIC_SMP_HELPERS | |
633 | static void remote_softirq_receive(void *data) | |
634 | { | |
635 | struct call_single_data *cp = data; | |
636 | unsigned long flags; | |
637 | int softirq; | |
638 | ||
3440a1ca | 639 | softirq = *(int *)cp->info; |
54514a70 DM |
640 | local_irq_save(flags); |
641 | __local_trigger(cp, softirq); | |
642 | local_irq_restore(flags); | |
643 | } | |
644 | ||
645 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
646 | { | |
647 | if (cpu_online(cpu)) { | |
648 | cp->func = remote_softirq_receive; | |
3440a1ca | 649 | cp->info = &softirq; |
54514a70 | 650 | cp->flags = 0; |
54514a70 | 651 | |
6e275637 | 652 | __smp_call_function_single(cpu, cp, 0); |
54514a70 DM |
653 | return 0; |
654 | } | |
655 | return 1; | |
656 | } | |
657 | #else /* CONFIG_USE_GENERIC_SMP_HELPERS */ | |
658 | static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
659 | { | |
660 | return 1; | |
661 | } | |
662 | #endif | |
663 | ||
664 | /** | |
665 | * __send_remote_softirq - try to schedule softirq work on a remote cpu | |
666 | * @cp: private SMP call function data area | |
667 | * @cpu: the remote cpu | |
668 | * @this_cpu: the currently executing cpu | |
669 | * @softirq: the softirq for the work | |
670 | * | |
671 | * Attempt to schedule softirq work on a remote cpu. If this cannot be | |
672 | * done, the work is instead queued up on the local cpu. | |
673 | * | |
674 | * Interrupts must be disabled. | |
675 | */ | |
676 | void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq) | |
677 | { | |
678 | if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq)) | |
679 | __local_trigger(cp, softirq); | |
680 | } | |
681 | EXPORT_SYMBOL(__send_remote_softirq); | |
682 | ||
683 | /** | |
684 | * send_remote_softirq - try to schedule softirq work on a remote cpu | |
685 | * @cp: private SMP call function data area | |
686 | * @cpu: the remote cpu | |
687 | * @softirq: the softirq for the work | |
688 | * | |
689 | * Like __send_remote_softirq except that disabling interrupts and | |
690 | * computing the current cpu is done for the caller. | |
691 | */ | |
692 | void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq) | |
693 | { | |
694 | unsigned long flags; | |
695 | int this_cpu; | |
696 | ||
697 | local_irq_save(flags); | |
698 | this_cpu = smp_processor_id(); | |
699 | __send_remote_softirq(cp, cpu, this_cpu, softirq); | |
700 | local_irq_restore(flags); | |
701 | } | |
702 | EXPORT_SYMBOL(send_remote_softirq); | |
703 | ||
704 | static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self, | |
705 | unsigned long action, void *hcpu) | |
706 | { | |
707 | /* | |
708 | * If a CPU goes away, splice its entries to the current CPU | |
709 | * and trigger a run of the softirq | |
710 | */ | |
711 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
712 | int cpu = (unsigned long) hcpu; | |
713 | int i; | |
714 | ||
715 | local_irq_disable(); | |
716 | for (i = 0; i < NR_SOFTIRQS; i++) { | |
717 | struct list_head *head = &per_cpu(softirq_work_list[i], cpu); | |
718 | struct list_head *local_head; | |
719 | ||
720 | if (list_empty(head)) | |
721 | continue; | |
722 | ||
723 | local_head = &__get_cpu_var(softirq_work_list[i]); | |
724 | list_splice_init(head, local_head); | |
725 | raise_softirq_irqoff(i); | |
726 | } | |
727 | local_irq_enable(); | |
728 | } | |
729 | ||
730 | return NOTIFY_OK; | |
731 | } | |
732 | ||
733 | static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = { | |
734 | .notifier_call = remote_softirq_cpu_notify, | |
735 | }; | |
736 | ||
1da177e4 LT |
737 | void __init softirq_init(void) |
738 | { | |
48f20a9a OJ |
739 | int cpu; |
740 | ||
741 | for_each_possible_cpu(cpu) { | |
54514a70 DM |
742 | int i; |
743 | ||
48f20a9a OJ |
744 | per_cpu(tasklet_vec, cpu).tail = |
745 | &per_cpu(tasklet_vec, cpu).head; | |
746 | per_cpu(tasklet_hi_vec, cpu).tail = | |
747 | &per_cpu(tasklet_hi_vec, cpu).head; | |
54514a70 DM |
748 | for (i = 0; i < NR_SOFTIRQS; i++) |
749 | INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu)); | |
48f20a9a OJ |
750 | } |
751 | ||
54514a70 DM |
752 | register_hotcpu_notifier(&remote_softirq_cpu_notifier); |
753 | ||
962cf36c CM |
754 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
755 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
756 | } |
757 | ||
3e339b5d | 758 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 759 | { |
3e339b5d TG |
760 | return local_softirq_pending(); |
761 | } | |
1da177e4 | 762 | |
3e339b5d TG |
763 | static void run_ksoftirqd(unsigned int cpu) |
764 | { | |
765 | local_irq_disable(); | |
766 | if (local_softirq_pending()) { | |
767 | __do_softirq(); | |
768 | rcu_note_context_switch(cpu); | |
769 | local_irq_enable(); | |
770 | cond_resched(); | |
771 | return; | |
1da177e4 | 772 | } |
3e339b5d | 773 | local_irq_enable(); |
1da177e4 LT |
774 | } |
775 | ||
776 | #ifdef CONFIG_HOTPLUG_CPU | |
777 | /* | |
778 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
779 | * scheduled for execution on @cpu. | |
780 | * | |
781 | * Unlike tasklet_kill, this function removes the tasklet | |
782 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
783 | * | |
784 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
785 | */ | |
786 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
787 | { | |
788 | struct tasklet_struct **i; | |
789 | ||
790 | BUG_ON(cpu_online(cpu)); | |
791 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
792 | ||
793 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
794 | return; | |
795 | ||
796 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 797 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
798 | if (*i == t) { |
799 | *i = t->next; | |
48f20a9a OJ |
800 | /* If this was the tail element, move the tail ptr */ |
801 | if (*i == NULL) | |
802 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
803 | return; |
804 | } | |
805 | } | |
806 | BUG(); | |
807 | } | |
808 | ||
809 | static void takeover_tasklets(unsigned int cpu) | |
810 | { | |
1da177e4 LT |
811 | /* CPU is dead, so no lock needed. */ |
812 | local_irq_disable(); | |
813 | ||
814 | /* Find end, append list for that CPU. */ | |
e5e41723 | 815 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
816 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
817 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
818 | per_cpu(tasklet_vec, cpu).head = NULL; |
819 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
820 | } | |
1da177e4 LT |
821 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
822 | ||
e5e41723 | 823 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
824 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
825 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
826 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
827 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
828 | } | |
1da177e4 LT |
829 | raise_softirq_irqoff(HI_SOFTIRQ); |
830 | ||
831 | local_irq_enable(); | |
832 | } | |
833 | #endif /* CONFIG_HOTPLUG_CPU */ | |
834 | ||
8c78f307 | 835 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
836 | unsigned long action, |
837 | void *hcpu) | |
838 | { | |
1da177e4 | 839 | switch (action) { |
1da177e4 | 840 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 841 | case CPU_DEAD: |
3e339b5d TG |
842 | case CPU_DEAD_FROZEN: |
843 | takeover_tasklets((unsigned long)hcpu); | |
1da177e4 LT |
844 | break; |
845 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3e339b5d | 846 | } |
1da177e4 LT |
847 | return NOTIFY_OK; |
848 | } | |
849 | ||
8c78f307 | 850 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
851 | .notifier_call = cpu_callback |
852 | }; | |
853 | ||
3e339b5d TG |
854 | static struct smp_hotplug_thread softirq_threads = { |
855 | .store = &ksoftirqd, | |
856 | .thread_should_run = ksoftirqd_should_run, | |
857 | .thread_fn = run_ksoftirqd, | |
858 | .thread_comm = "ksoftirqd/%u", | |
859 | }; | |
860 | ||
7babe8db | 861 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 862 | { |
1da177e4 | 863 | register_cpu_notifier(&cpu_nfb); |
3e339b5d TG |
864 | |
865 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); | |
866 | ||
1da177e4 LT |
867 | return 0; |
868 | } | |
7babe8db | 869 | early_initcall(spawn_ksoftirqd); |
78eef01b | 870 | |
43a25632 YL |
871 | /* |
872 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
873 | * GCC does not inline them incorrectly. ] | |
874 | */ | |
875 | ||
876 | int __init __weak early_irq_init(void) | |
877 | { | |
878 | return 0; | |
879 | } | |
880 | ||
b683de2b | 881 | #ifdef CONFIG_GENERIC_HARDIRQS |
4a046d17 YL |
882 | int __init __weak arch_probe_nr_irqs(void) |
883 | { | |
b683de2b | 884 | return NR_IRQS_LEGACY; |
4a046d17 YL |
885 | } |
886 | ||
43a25632 YL |
887 | int __init __weak arch_early_irq_init(void) |
888 | { | |
889 | return 0; | |
890 | } | |
b683de2b | 891 | #endif |