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) | |
1da177e4 LT |
9 | */ |
10 | ||
9984de1a | 11 | #include <linux/export.h> |
1da177e4 LT |
12 | #include <linux/kernel_stat.h> |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mm.h> | |
16 | #include <linux/notifier.h> | |
17 | #include <linux/percpu.h> | |
18 | #include <linux/cpu.h> | |
83144186 | 19 | #include <linux/freezer.h> |
1da177e4 LT |
20 | #include <linux/kthread.h> |
21 | #include <linux/rcupdate.h> | |
7e49fcce | 22 | #include <linux/ftrace.h> |
78eef01b | 23 | #include <linux/smp.h> |
3e339b5d | 24 | #include <linux/smpboot.h> |
79bf2bb3 | 25 | #include <linux/tick.h> |
a0e39ed3 | 26 | |
6fa3eb70 | 27 | #include <linux/mt_sched_mon.h> |
a0e39ed3 | 28 | #define CREATE_TRACE_POINTS |
ad8d75ff | 29 | #include <trace/events/irq.h> |
1da177e4 LT |
30 | |
31 | #include <asm/irq.h> | |
32 | /* | |
33 | - No shared variables, all the data are CPU local. | |
34 | - If a softirq needs serialization, let it serialize itself | |
35 | by its own spinlocks. | |
36 | - Even if softirq is serialized, only local cpu is marked for | |
37 | execution. Hence, we get something sort of weak cpu binding. | |
38 | Though it is still not clear, will it result in better locality | |
39 | or will not. | |
40 | ||
41 | Examples: | |
42 | - NET RX softirq. It is multithreaded and does not require | |
43 | any global serialization. | |
44 | - NET TX softirq. It kicks software netdevice queues, hence | |
45 | it is logically serialized per device, but this serialization | |
46 | is invisible to common code. | |
47 | - Tasklets: serialized wrt itself. | |
48 | */ | |
49 | ||
50 | #ifndef __ARCH_IRQ_STAT | |
51 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
52 | EXPORT_SYMBOL(irq_stat); | |
53 | #endif | |
54 | ||
978b0116 | 55 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 56 | |
4dd53d89 | 57 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 58 | |
5d592b44 | 59 | char *softirq_to_name[NR_SOFTIRQS] = { |
5dd4de58 | 60 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL", |
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 | { | |
75e1056f VP |
129 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
130 | SOFTIRQ_DISABLE_OFFSET); | |
de30a2b3 IM |
131 | } |
132 | ||
133 | EXPORT_SYMBOL(local_bh_disable); | |
134 | ||
75e1056f VP |
135 | static void __local_bh_enable(unsigned int cnt) |
136 | { | |
137 | WARN_ON_ONCE(in_irq()); | |
138 | WARN_ON_ONCE(!irqs_disabled()); | |
139 | ||
140 | if (softirq_count() == cnt) | |
141 | trace_softirqs_on((unsigned long)__builtin_return_address(0)); | |
142 | sub_preempt_count(cnt); | |
143 | } | |
144 | ||
de30a2b3 IM |
145 | /* |
146 | * Special-case - softirqs can safely be enabled in | |
147 | * cond_resched_softirq(), or by __do_softirq(), | |
148 | * without processing still-pending softirqs: | |
149 | */ | |
150 | void _local_bh_enable(void) | |
151 | { | |
75e1056f | 152 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
153 | } |
154 | ||
155 | EXPORT_SYMBOL(_local_bh_enable); | |
156 | ||
0f476b6d | 157 | static inline void _local_bh_enable_ip(unsigned long ip) |
de30a2b3 | 158 | { |
0f476b6d | 159 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 160 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 161 | local_irq_disable(); |
3c829c36 | 162 | #endif |
de30a2b3 IM |
163 | /* |
164 | * Are softirqs going to be turned on now: | |
165 | */ | |
75e1056f | 166 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 167 | trace_softirqs_on(ip); |
de30a2b3 IM |
168 | /* |
169 | * Keep preemption disabled until we are done with | |
170 | * softirq processing: | |
171 | */ | |
75e1056f | 172 | sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1); |
de30a2b3 IM |
173 | |
174 | if (unlikely(!in_interrupt() && local_softirq_pending())) | |
175 | do_softirq(); | |
176 | ||
177 | dec_preempt_count(); | |
3c829c36 | 178 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 179 | local_irq_enable(); |
3c829c36 | 180 | #endif |
de30a2b3 IM |
181 | preempt_check_resched(); |
182 | } | |
0f476b6d JB |
183 | |
184 | void local_bh_enable(void) | |
185 | { | |
186 | _local_bh_enable_ip((unsigned long)__builtin_return_address(0)); | |
187 | } | |
de30a2b3 IM |
188 | EXPORT_SYMBOL(local_bh_enable); |
189 | ||
190 | void local_bh_enable_ip(unsigned long ip) | |
191 | { | |
0f476b6d | 192 | _local_bh_enable_ip(ip); |
de30a2b3 IM |
193 | } |
194 | EXPORT_SYMBOL(local_bh_enable_ip); | |
195 | ||
1da177e4 | 196 | /* |
34376a50 BG |
197 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
198 | * but break the loop if need_resched() is set or after 2 ms. | |
199 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
200 | * certain cases, such as stop_machine(), jiffies may cease to | |
201 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
202 | * well to make sure we eventually return from this method. | |
1da177e4 | 203 | * |
c10d7367 | 204 | * These limits have been established via experimentation. |
1da177e4 LT |
205 | * The two things to balance is latency against fairness - |
206 | * we want to handle softirqs as soon as possible, but they | |
207 | * should not be able to lock up the box. | |
208 | */ | |
c10d7367 | 209 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 210 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 LT |
211 | |
212 | asmlinkage void __do_softirq(void) | |
213 | { | |
214 | struct softirq_action *h; | |
215 | __u32 pending; | |
c10d7367 | 216 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
1da177e4 | 217 | int cpu; |
907aed48 | 218 | unsigned long old_flags = current->flags; |
34376a50 | 219 | int max_restart = MAX_SOFTIRQ_RESTART; |
907aed48 MG |
220 | |
221 | /* | |
222 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
223 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
224 | * again if the socket is related to swap | |
225 | */ | |
226 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
227 | |
228 | pending = local_softirq_pending(); | |
6a61671b | 229 | account_irq_enter_time(current); |
829035fd | 230 | |
75e1056f VP |
231 | __local_bh_disable((unsigned long)__builtin_return_address(0), |
232 | SOFTIRQ_OFFSET); | |
d820ac4c | 233 | lockdep_softirq_enter(); |
1da177e4 | 234 | |
1da177e4 LT |
235 | cpu = smp_processor_id(); |
236 | restart: | |
237 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 238 | set_softirq_pending(0); |
1da177e4 | 239 | |
c70f5d66 | 240 | local_irq_enable(); |
1da177e4 LT |
241 | |
242 | h = softirq_vec; | |
243 | ||
244 | do { | |
245 | if (pending & 1) { | |
f4bc6bb2 | 246 | unsigned int vec_nr = h - softirq_vec; |
8e85b4b5 TG |
247 | int prev_count = preempt_count(); |
248 | ||
f4bc6bb2 TG |
249 | kstat_incr_softirqs_this_cpu(vec_nr); |
250 | ||
251 | trace_softirq_entry(vec_nr); | |
6fa3eb70 S |
252 | mt_trace_SoftIRQ_start(vec_nr); |
253 | h->action(h); | |
254 | mt_trace_SoftIRQ_end(vec_nr); | |
f4bc6bb2 | 255 | trace_softirq_exit(vec_nr); |
8e85b4b5 | 256 | if (unlikely(prev_count != preempt_count())) { |
f4bc6bb2 | 257 | printk(KERN_ERR "huh, entered softirq %u %s %p" |
8e85b4b5 | 258 | "with preempt_count %08x," |
f4bc6bb2 TG |
259 | " exited with %08x?\n", vec_nr, |
260 | softirq_to_name[vec_nr], h->action, | |
261 | prev_count, preempt_count()); | |
8e85b4b5 TG |
262 | preempt_count() = prev_count; |
263 | } | |
264 | ||
d6714c22 | 265 | rcu_bh_qs(cpu); |
1da177e4 LT |
266 | } |
267 | h++; | |
268 | pending >>= 1; | |
269 | } while (pending); | |
270 | ||
c70f5d66 | 271 | local_irq_disable(); |
1da177e4 LT |
272 | |
273 | pending = local_softirq_pending(); | |
c10d7367 | 274 | if (pending) { |
34376a50 BG |
275 | if (time_before(jiffies, end) && !need_resched() && |
276 | --max_restart) | |
c10d7367 | 277 | goto restart; |
1da177e4 | 278 | |
1da177e4 | 279 | wakeup_softirqd(); |
c10d7367 | 280 | } |
1da177e4 | 281 | |
d820ac4c | 282 | lockdep_softirq_exit(); |
829035fd | 283 | |
6a61671b | 284 | account_irq_exit_time(current); |
75e1056f | 285 | __local_bh_enable(SOFTIRQ_OFFSET); |
907aed48 | 286 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
1da177e4 LT |
287 | } |
288 | ||
289 | #ifndef __ARCH_HAS_DO_SOFTIRQ | |
290 | ||
291 | asmlinkage void do_softirq(void) | |
292 | { | |
293 | __u32 pending; | |
294 | unsigned long flags; | |
295 | ||
296 | if (in_interrupt()) | |
297 | return; | |
298 | ||
299 | local_irq_save(flags); | |
300 | ||
301 | pending = local_softirq_pending(); | |
302 | ||
303 | if (pending) | |
304 | __do_softirq(); | |
305 | ||
306 | local_irq_restore(flags); | |
307 | } | |
308 | ||
1da177e4 LT |
309 | #endif |
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 | { | |
6dcdd575 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 | */ | |
343 | do_softirq(); | |
344 | } else { | |
8d32a307 | 345 | wakeup_softirqd(); |
6dcdd575 | 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(); | |
6fa3eb70 | 493 | mt_trace_tasklet_start(t->func); |
1da177e4 | 494 | t->func(t->data); |
6fa3eb70 | 495 | mt_trace_tasklet_end(t->func); |
1da177e4 LT |
496 | tasklet_unlock(t); |
497 | continue; | |
498 | } | |
499 | tasklet_unlock(t); | |
500 | } | |
501 | ||
502 | local_irq_disable(); | |
48f20a9a | 503 | t->next = NULL; |
909ea964 CL |
504 | *__this_cpu_read(tasklet_vec.tail) = t; |
505 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
506 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
507 | local_irq_enable(); | |
508 | } | |
509 | } | |
510 | ||
511 | static void tasklet_hi_action(struct softirq_action *a) | |
512 | { | |
513 | struct tasklet_struct *list; | |
514 | ||
515 | local_irq_disable(); | |
909ea964 CL |
516 | list = __this_cpu_read(tasklet_hi_vec.head); |
517 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
518 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
519 | local_irq_enable(); |
520 | ||
521 | while (list) { | |
522 | struct tasklet_struct *t = list; | |
523 | ||
524 | list = list->next; | |
525 | ||
526 | if (tasklet_trylock(t)) { | |
527 | if (!atomic_read(&t->count)) { | |
528 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
529 | BUG(); | |
530 | t->func(t->data); | |
531 | tasklet_unlock(t); | |
532 | continue; | |
533 | } | |
534 | tasklet_unlock(t); | |
535 | } | |
536 | ||
537 | local_irq_disable(); | |
48f20a9a | 538 | t->next = NULL; |
909ea964 CL |
539 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
540 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
541 | __raise_softirq_irqoff(HI_SOFTIRQ); |
542 | local_irq_enable(); | |
543 | } | |
544 | } | |
545 | ||
546 | ||
547 | void tasklet_init(struct tasklet_struct *t, | |
548 | void (*func)(unsigned long), unsigned long data) | |
549 | { | |
550 | t->next = NULL; | |
551 | t->state = 0; | |
552 | atomic_set(&t->count, 0); | |
553 | t->func = func; | |
554 | t->data = data; | |
555 | } | |
556 | ||
557 | EXPORT_SYMBOL(tasklet_init); | |
558 | ||
559 | void tasklet_kill(struct tasklet_struct *t) | |
560 | { | |
561 | if (in_interrupt()) | |
562 | printk("Attempt to kill tasklet from interrupt\n"); | |
563 | ||
564 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 565 | do { |
1da177e4 | 566 | yield(); |
79d381c9 | 567 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
568 | } |
569 | tasklet_unlock_wait(t); | |
570 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
571 | } | |
572 | ||
573 | EXPORT_SYMBOL(tasklet_kill); | |
574 | ||
9ba5f005 PZ |
575 | /* |
576 | * tasklet_hrtimer | |
577 | */ | |
578 | ||
579 | /* | |
b9c30322 PZ |
580 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
581 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
582 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
583 | */ |
584 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
585 | { | |
586 | struct tasklet_hrtimer *ttimer = | |
587 | container_of(timer, struct tasklet_hrtimer, timer); | |
588 | ||
b9c30322 PZ |
589 | tasklet_hi_schedule(&ttimer->tasklet); |
590 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
591 | } |
592 | ||
593 | /* | |
594 | * Helper function which calls the hrtimer callback from | |
595 | * tasklet/softirq context | |
596 | */ | |
597 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
598 | { | |
599 | struct tasklet_hrtimer *ttimer = (void *)data; | |
600 | enum hrtimer_restart restart; | |
601 | ||
602 | restart = ttimer->function(&ttimer->timer); | |
603 | if (restart != HRTIMER_NORESTART) | |
604 | hrtimer_restart(&ttimer->timer); | |
605 | } | |
606 | ||
607 | /** | |
608 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
609 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 610 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
611 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
612 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
613 | */ | |
614 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
615 | enum hrtimer_restart (*function)(struct hrtimer *), | |
616 | clockid_t which_clock, enum hrtimer_mode mode) | |
617 | { | |
618 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
619 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
620 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
621 | (unsigned long)ttimer); | |
622 | ttimer->function = function; | |
623 | } | |
624 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
625 | ||
1da177e4 LT |
626 | void __init softirq_init(void) |
627 | { | |
48f20a9a OJ |
628 | int cpu; |
629 | ||
630 | for_each_possible_cpu(cpu) { | |
631 | per_cpu(tasklet_vec, cpu).tail = | |
632 | &per_cpu(tasklet_vec, cpu).head; | |
633 | per_cpu(tasklet_hi_vec, cpu).tail = | |
634 | &per_cpu(tasklet_hi_vec, cpu).head; | |
635 | } | |
636 | ||
962cf36c CM |
637 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
638 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
639 | } |
640 | ||
3e339b5d | 641 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 642 | { |
3e339b5d TG |
643 | return local_softirq_pending(); |
644 | } | |
1da177e4 | 645 | |
3e339b5d TG |
646 | static void run_ksoftirqd(unsigned int cpu) |
647 | { | |
648 | local_irq_disable(); | |
649 | if (local_softirq_pending()) { | |
650 | __do_softirq(); | |
3e339b5d TG |
651 | local_irq_enable(); |
652 | cond_resched(); | |
61ea92b9 CO |
653 | |
654 | preempt_disable(); | |
655 | rcu_note_context_switch(cpu); | |
656 | preempt_enable(); | |
657 | ||
3e339b5d | 658 | return; |
1da177e4 | 659 | } |
3e339b5d | 660 | local_irq_enable(); |
1da177e4 LT |
661 | } |
662 | ||
663 | #ifdef CONFIG_HOTPLUG_CPU | |
664 | /* | |
665 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
666 | * scheduled for execution on @cpu. | |
667 | * | |
668 | * Unlike tasklet_kill, this function removes the tasklet | |
669 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
670 | * | |
671 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
672 | */ | |
673 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
674 | { | |
675 | struct tasklet_struct **i; | |
676 | ||
677 | BUG_ON(cpu_online(cpu)); | |
678 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
679 | ||
680 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
681 | return; | |
682 | ||
683 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 684 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
685 | if (*i == t) { |
686 | *i = t->next; | |
48f20a9a OJ |
687 | /* If this was the tail element, move the tail ptr */ |
688 | if (*i == NULL) | |
689 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
690 | return; |
691 | } | |
692 | } | |
693 | BUG(); | |
694 | } | |
695 | ||
696 | static void takeover_tasklets(unsigned int cpu) | |
697 | { | |
1da177e4 LT |
698 | /* CPU is dead, so no lock needed. */ |
699 | local_irq_disable(); | |
700 | ||
701 | /* Find end, append list for that CPU. */ | |
e5e41723 | 702 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
703 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
704 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
705 | per_cpu(tasklet_vec, cpu).head = NULL; |
706 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
707 | } | |
1da177e4 LT |
708 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
709 | ||
e5e41723 | 710 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
711 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
712 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
713 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
714 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
715 | } | |
1da177e4 LT |
716 | raise_softirq_irqoff(HI_SOFTIRQ); |
717 | ||
718 | local_irq_enable(); | |
719 | } | |
720 | #endif /* CONFIG_HOTPLUG_CPU */ | |
721 | ||
8c78f307 | 722 | static int __cpuinit cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
723 | unsigned long action, |
724 | void *hcpu) | |
725 | { | |
1da177e4 | 726 | switch (action) { |
1da177e4 | 727 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 728 | case CPU_DEAD: |
3e339b5d TG |
729 | case CPU_DEAD_FROZEN: |
730 | takeover_tasklets((unsigned long)hcpu); | |
1da177e4 LT |
731 | break; |
732 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3e339b5d | 733 | } |
1da177e4 LT |
734 | return NOTIFY_OK; |
735 | } | |
736 | ||
8c78f307 | 737 | static struct notifier_block __cpuinitdata cpu_nfb = { |
1da177e4 LT |
738 | .notifier_call = cpu_callback |
739 | }; | |
740 | ||
3e339b5d TG |
741 | static struct smp_hotplug_thread softirq_threads = { |
742 | .store = &ksoftirqd, | |
743 | .thread_should_run = ksoftirqd_should_run, | |
744 | .thread_fn = run_ksoftirqd, | |
745 | .thread_comm = "ksoftirqd/%u", | |
746 | }; | |
747 | ||
7babe8db | 748 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 749 | { |
1da177e4 | 750 | register_cpu_notifier(&cpu_nfb); |
3e339b5d TG |
751 | |
752 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); | |
753 | ||
1da177e4 LT |
754 | return 0; |
755 | } | |
7babe8db | 756 | early_initcall(spawn_ksoftirqd); |
78eef01b | 757 | |
43a25632 YL |
758 | /* |
759 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
760 | * GCC does not inline them incorrectly. ] | |
761 | */ | |
762 | ||
763 | int __init __weak early_irq_init(void) | |
764 | { | |
765 | return 0; | |
766 | } | |
767 | ||
b683de2b | 768 | #ifdef CONFIG_GENERIC_HARDIRQS |
4a046d17 YL |
769 | int __init __weak arch_probe_nr_irqs(void) |
770 | { | |
b683de2b | 771 | return NR_IRQS_LEGACY; |
4a046d17 YL |
772 | } |
773 | ||
43a25632 YL |
774 | int __init __weak arch_early_irq_init(void) |
775 | { | |
776 | return 0; | |
777 | } | |
b683de2b | 778 | #endif |