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