]>
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 HC |
26 | |
27 | #define CREATE_TRACE_POINTS | |
ad8d75ff | 28 | #include <trace/events/irq.h> |
1da177e4 | 29 | |
1da177e4 LT |
30 | /* |
31 | - No shared variables, all the data are CPU local. | |
32 | - If a softirq needs serialization, let it serialize itself | |
33 | by its own spinlocks. | |
34 | - Even if softirq is serialized, only local cpu is marked for | |
35 | execution. Hence, we get something sort of weak cpu binding. | |
36 | Though it is still not clear, will it result in better locality | |
37 | or will not. | |
38 | ||
39 | Examples: | |
40 | - NET RX softirq. It is multithreaded and does not require | |
41 | any global serialization. | |
42 | - NET TX softirq. It kicks software netdevice queues, hence | |
43 | it is logically serialized per device, but this serialization | |
44 | is invisible to common code. | |
45 | - Tasklets: serialized wrt itself. | |
46 | */ | |
47 | ||
48 | #ifndef __ARCH_IRQ_STAT | |
49 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
50 | EXPORT_SYMBOL(irq_stat); | |
51 | #endif | |
52 | ||
978b0116 | 53 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 54 | |
4dd53d89 | 55 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 56 | |
5d592b44 | 57 | char *softirq_to_name[NR_SOFTIRQS] = { |
5dd4de58 | 58 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL", |
09223371 | 59 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
5d592b44 JB |
60 | }; |
61 | ||
1da177e4 LT |
62 | /* |
63 | * we cannot loop indefinitely here to avoid userspace starvation, | |
64 | * but we also don't want to introduce a worst case 1/HZ latency | |
65 | * to the pending events, so lets the scheduler to balance | |
66 | * the softirq load for us. | |
67 | */ | |
676cb02d | 68 | static void wakeup_softirqd(void) |
1da177e4 LT |
69 | { |
70 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 71 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 LT |
72 | |
73 | if (tsk && tsk->state != TASK_RUNNING) | |
74 | wake_up_process(tsk); | |
75 | } | |
76 | ||
75e1056f VP |
77 | /* |
78 | * preempt_count and SOFTIRQ_OFFSET usage: | |
79 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving | |
80 | * softirq processing. | |
81 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
82 | * on local_bh_disable or local_bh_enable. | |
83 | * This lets us distinguish between whether we are currently processing | |
84 | * softirq and whether we just have bh disabled. | |
85 | */ | |
86 | ||
de30a2b3 IM |
87 | /* |
88 | * This one is for softirq.c-internal use, | |
89 | * where hardirqs are disabled legitimately: | |
90 | */ | |
3c829c36 | 91 | #ifdef CONFIG_TRACE_IRQFLAGS |
0bd3a173 | 92 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
93 | { |
94 | unsigned long flags; | |
95 | ||
96 | WARN_ON_ONCE(in_irq()); | |
97 | ||
98 | raw_local_irq_save(flags); | |
7e49fcce | 99 | /* |
bdb43806 | 100 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
101 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
102 | * is set and before current->softirq_enabled is cleared. | |
103 | * We must manually increment preempt_count here and manually | |
104 | * call the trace_preempt_off later. | |
105 | */ | |
bdb43806 | 106 | __preempt_count_add(cnt); |
de30a2b3 IM |
107 | /* |
108 | * Were softirqs turned off above: | |
109 | */ | |
9ea4c380 | 110 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
de30a2b3 IM |
111 | trace_softirqs_off(ip); |
112 | raw_local_irq_restore(flags); | |
7e49fcce | 113 | |
75e1056f | 114 | if (preempt_count() == cnt) |
7e49fcce | 115 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); |
de30a2b3 | 116 | } |
0bd3a173 | 117 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 118 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 119 | |
75e1056f VP |
120 | static void __local_bh_enable(unsigned int cnt) |
121 | { | |
75e1056f VP |
122 | WARN_ON_ONCE(!irqs_disabled()); |
123 | ||
9ea4c380 | 124 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
d2e08473 | 125 | trace_softirqs_on(_RET_IP_); |
bdb43806 | 126 | preempt_count_sub(cnt); |
75e1056f VP |
127 | } |
128 | ||
de30a2b3 IM |
129 | /* |
130 | * Special-case - softirqs can safely be enabled in | |
131 | * cond_resched_softirq(), or by __do_softirq(), | |
132 | * without processing still-pending softirqs: | |
133 | */ | |
134 | void _local_bh_enable(void) | |
135 | { | |
5d60d3e7 | 136 | WARN_ON_ONCE(in_irq()); |
75e1056f | 137 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
138 | } |
139 | ||
140 | EXPORT_SYMBOL(_local_bh_enable); | |
141 | ||
0bd3a173 | 142 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 143 | { |
0f476b6d | 144 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 145 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 146 | local_irq_disable(); |
3c829c36 | 147 | #endif |
de30a2b3 IM |
148 | /* |
149 | * Are softirqs going to be turned on now: | |
150 | */ | |
75e1056f | 151 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 152 | trace_softirqs_on(ip); |
de30a2b3 IM |
153 | /* |
154 | * Keep preemption disabled until we are done with | |
155 | * softirq processing: | |
156 | */ | |
0bd3a173 | 157 | preempt_count_sub(cnt - 1); |
de30a2b3 | 158 | |
0bed698a FW |
159 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
160 | /* | |
161 | * Run softirq if any pending. And do it in its own stack | |
162 | * as we may be calling this deep in a task call stack already. | |
163 | */ | |
de30a2b3 | 164 | do_softirq(); |
0bed698a | 165 | } |
de30a2b3 | 166 | |
bdb43806 | 167 | preempt_count_dec(); |
3c829c36 | 168 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 169 | local_irq_enable(); |
3c829c36 | 170 | #endif |
de30a2b3 IM |
171 | preempt_check_resched(); |
172 | } | |
0bd3a173 | 173 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 174 | |
1da177e4 | 175 | /* |
34376a50 BG |
176 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
177 | * but break the loop if need_resched() is set or after 2 ms. | |
178 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
179 | * certain cases, such as stop_machine(), jiffies may cease to | |
180 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
181 | * well to make sure we eventually return from this method. | |
1da177e4 | 182 | * |
c10d7367 | 183 | * These limits have been established via experimentation. |
1da177e4 LT |
184 | * The two things to balance is latency against fairness - |
185 | * we want to handle softirqs as soon as possible, but they | |
186 | * should not be able to lock up the box. | |
187 | */ | |
c10d7367 | 188 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 189 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 190 | |
f1a83e65 PZ |
191 | #ifdef CONFIG_TRACE_IRQFLAGS |
192 | /* | |
f1a83e65 PZ |
193 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
194 | * to keep the lockdep irq context tracking as tight as possible in order to | |
195 | * not miss-qualify lock contexts and miss possible deadlocks. | |
196 | */ | |
f1a83e65 | 197 | |
5c4853b6 | 198 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 199 | { |
5c4853b6 | 200 | bool in_hardirq = false; |
f1a83e65 | 201 | |
5c4853b6 FW |
202 | if (trace_hardirq_context(current)) { |
203 | in_hardirq = true; | |
f1a83e65 | 204 | trace_hardirq_exit(); |
5c4853b6 FW |
205 | } |
206 | ||
f1a83e65 | 207 | lockdep_softirq_enter(); |
5c4853b6 FW |
208 | |
209 | return in_hardirq; | |
f1a83e65 PZ |
210 | } |
211 | ||
5c4853b6 | 212 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
213 | { |
214 | lockdep_softirq_exit(); | |
5c4853b6 FW |
215 | |
216 | if (in_hardirq) | |
f1a83e65 | 217 | trace_hardirq_enter(); |
f1a83e65 | 218 | } |
f1a83e65 | 219 | #else |
5c4853b6 FW |
220 | static inline bool lockdep_softirq_start(void) { return false; } |
221 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
222 | #endif |
223 | ||
1da177e4 LT |
224 | asmlinkage void __do_softirq(void) |
225 | { | |
c10d7367 | 226 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 227 | unsigned long old_flags = current->flags; |
34376a50 | 228 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 229 | struct softirq_action *h; |
5c4853b6 | 230 | bool in_hardirq; |
f1a83e65 | 231 | __u32 pending; |
2e702b9f | 232 | int softirq_bit; |
f1a83e65 | 233 | int cpu; |
907aed48 MG |
234 | |
235 | /* | |
236 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
237 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
238 | * again if the socket is related to swap | |
239 | */ | |
240 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
241 | |
242 | pending = local_softirq_pending(); | |
6a61671b | 243 | account_irq_enter_time(current); |
829035fd | 244 | |
0bd3a173 | 245 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 246 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 247 | |
1da177e4 LT |
248 | cpu = smp_processor_id(); |
249 | restart: | |
250 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 251 | set_softirq_pending(0); |
1da177e4 | 252 | |
c70f5d66 | 253 | local_irq_enable(); |
1da177e4 LT |
254 | |
255 | h = softirq_vec; | |
256 | ||
2e702b9f JP |
257 | while ((softirq_bit = ffs(pending))) { |
258 | unsigned int vec_nr; | |
259 | int prev_count; | |
260 | ||
261 | h += softirq_bit - 1; | |
262 | ||
263 | vec_nr = h - softirq_vec; | |
264 | prev_count = preempt_count(); | |
265 | ||
266 | kstat_incr_softirqs_this_cpu(vec_nr); | |
267 | ||
268 | trace_softirq_entry(vec_nr); | |
269 | h->action(h); | |
270 | trace_softirq_exit(vec_nr); | |
271 | if (unlikely(prev_count != preempt_count())) { | |
272 | printk(KERN_ERR "huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", | |
273 | vec_nr, softirq_to_name[vec_nr], h->action, | |
274 | prev_count, preempt_count()); | |
275 | preempt_count_set(prev_count); | |
1da177e4 | 276 | } |
2e702b9f | 277 | rcu_bh_qs(cpu); |
1da177e4 | 278 | h++; |
2e702b9f JP |
279 | pending >>= softirq_bit; |
280 | } | |
1da177e4 | 281 | |
c70f5d66 | 282 | local_irq_disable(); |
1da177e4 LT |
283 | |
284 | pending = local_softirq_pending(); | |
c10d7367 | 285 | if (pending) { |
34376a50 BG |
286 | if (time_before(jiffies, end) && !need_resched() && |
287 | --max_restart) | |
c10d7367 | 288 | goto restart; |
1da177e4 | 289 | |
1da177e4 | 290 | wakeup_softirqd(); |
c10d7367 | 291 | } |
1da177e4 | 292 | |
5c4853b6 | 293 | lockdep_softirq_end(in_hardirq); |
6a61671b | 294 | account_irq_exit_time(current); |
75e1056f | 295 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 296 | WARN_ON_ONCE(in_interrupt()); |
907aed48 | 297 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
1da177e4 LT |
298 | } |
299 | ||
1da177e4 LT |
300 | asmlinkage void do_softirq(void) |
301 | { | |
302 | __u32 pending; | |
303 | unsigned long flags; | |
304 | ||
305 | if (in_interrupt()) | |
306 | return; | |
307 | ||
308 | local_irq_save(flags); | |
309 | ||
310 | pending = local_softirq_pending(); | |
311 | ||
312 | if (pending) | |
7d65f4a6 | 313 | do_softirq_own_stack(); |
1da177e4 LT |
314 | |
315 | local_irq_restore(flags); | |
316 | } | |
317 | ||
dde4b2b5 IM |
318 | /* |
319 | * Enter an interrupt context. | |
320 | */ | |
321 | void irq_enter(void) | |
322 | { | |
64db4cff | 323 | rcu_irq_enter(); |
0a8a2e78 | 324 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
325 | /* |
326 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
327 | * here, as softirq will be serviced on return from interrupt. | |
328 | */ | |
329 | local_bh_disable(); | |
e8fcaa5c | 330 | tick_check_idle(); |
d267f87f VP |
331 | _local_bh_enable(); |
332 | } | |
333 | ||
334 | __irq_enter(); | |
dde4b2b5 IM |
335 | } |
336 | ||
8d32a307 TG |
337 | static inline void invoke_softirq(void) |
338 | { | |
ded79754 | 339 | if (!force_irqthreads) { |
cc1f0274 | 340 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
341 | /* |
342 | * We can safely execute softirq on the current stack if | |
343 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
344 | * at this stage. |
345 | */ | |
346 | __do_softirq(); | |
347 | #else | |
348 | /* | |
349 | * Otherwise, irq_exit() is called on the task stack that can | |
350 | * be potentially deep already. So call softirq in its own stack | |
351 | * to prevent from any overrun. | |
ded79754 | 352 | */ |
be6e1016 | 353 | do_softirq_own_stack(); |
cc1f0274 | 354 | #endif |
ded79754 | 355 | } else { |
8d32a307 | 356 | wakeup_softirqd(); |
ded79754 | 357 | } |
8d32a307 | 358 | } |
1da177e4 | 359 | |
67826eae FW |
360 | static inline void tick_irq_exit(void) |
361 | { | |
362 | #ifdef CONFIG_NO_HZ_COMMON | |
363 | int cpu = smp_processor_id(); | |
364 | ||
365 | /* Make sure that timer wheel updates are propagated */ | |
366 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
367 | if (!in_interrupt()) | |
368 | tick_nohz_irq_exit(); | |
369 | } | |
370 | #endif | |
371 | } | |
372 | ||
1da177e4 LT |
373 | /* |
374 | * Exit an interrupt context. Process softirqs if needed and possible: | |
375 | */ | |
376 | void irq_exit(void) | |
377 | { | |
74eed016 | 378 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 379 | local_irq_disable(); |
74eed016 TG |
380 | #else |
381 | WARN_ON_ONCE(!irqs_disabled()); | |
382 | #endif | |
383 | ||
6a61671b | 384 | account_irq_exit_time(current); |
bdb43806 | 385 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
386 | if (!in_interrupt() && local_softirq_pending()) |
387 | invoke_softirq(); | |
79bf2bb3 | 388 | |
67826eae | 389 | tick_irq_exit(); |
416eb33c | 390 | rcu_irq_exit(); |
f1a83e65 | 391 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
392 | } |
393 | ||
394 | /* | |
395 | * This function must run with irqs disabled! | |
396 | */ | |
7ad5b3a5 | 397 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
398 | { |
399 | __raise_softirq_irqoff(nr); | |
400 | ||
401 | /* | |
402 | * If we're in an interrupt or softirq, we're done | |
403 | * (this also catches softirq-disabled code). We will | |
404 | * actually run the softirq once we return from | |
405 | * the irq or softirq. | |
406 | * | |
407 | * Otherwise we wake up ksoftirqd to make sure we | |
408 | * schedule the softirq soon. | |
409 | */ | |
410 | if (!in_interrupt()) | |
411 | wakeup_softirqd(); | |
412 | } | |
413 | ||
7ad5b3a5 | 414 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
415 | { |
416 | unsigned long flags; | |
417 | ||
418 | local_irq_save(flags); | |
419 | raise_softirq_irqoff(nr); | |
420 | local_irq_restore(flags); | |
421 | } | |
422 | ||
f069686e SR |
423 | void __raise_softirq_irqoff(unsigned int nr) |
424 | { | |
425 | trace_softirq_raise(nr); | |
426 | or_softirq_pending(1UL << nr); | |
427 | } | |
428 | ||
962cf36c | 429 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 430 | { |
1da177e4 LT |
431 | softirq_vec[nr].action = action; |
432 | } | |
433 | ||
9ba5f005 PZ |
434 | /* |
435 | * Tasklets | |
436 | */ | |
1da177e4 LT |
437 | struct tasklet_head |
438 | { | |
48f20a9a OJ |
439 | struct tasklet_struct *head; |
440 | struct tasklet_struct **tail; | |
1da177e4 LT |
441 | }; |
442 | ||
4620b49f VN |
443 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
444 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 445 | |
7ad5b3a5 | 446 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
447 | { |
448 | unsigned long flags; | |
449 | ||
450 | local_irq_save(flags); | |
48f20a9a | 451 | t->next = NULL; |
909ea964 CL |
452 | *__this_cpu_read(tasklet_vec.tail) = t; |
453 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
454 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
455 | local_irq_restore(flags); | |
456 | } | |
457 | ||
458 | EXPORT_SYMBOL(__tasklet_schedule); | |
459 | ||
7ad5b3a5 | 460 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
461 | { |
462 | unsigned long flags; | |
463 | ||
464 | local_irq_save(flags); | |
48f20a9a | 465 | t->next = NULL; |
909ea964 CL |
466 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
467 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
468 | raise_softirq_irqoff(HI_SOFTIRQ); |
469 | local_irq_restore(flags); | |
470 | } | |
471 | ||
472 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
473 | ||
7c692cba VN |
474 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
475 | { | |
476 | BUG_ON(!irqs_disabled()); | |
477 | ||
909ea964 CL |
478 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
479 | __this_cpu_write(tasklet_hi_vec.head, t); | |
7c692cba VN |
480 | __raise_softirq_irqoff(HI_SOFTIRQ); |
481 | } | |
482 | ||
483 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); | |
484 | ||
1da177e4 LT |
485 | static void tasklet_action(struct softirq_action *a) |
486 | { | |
487 | struct tasklet_struct *list; | |
488 | ||
489 | local_irq_disable(); | |
909ea964 CL |
490 | list = __this_cpu_read(tasklet_vec.head); |
491 | __this_cpu_write(tasklet_vec.head, NULL); | |
492 | __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head); | |
1da177e4 LT |
493 | local_irq_enable(); |
494 | ||
495 | while (list) { | |
496 | struct tasklet_struct *t = list; | |
497 | ||
498 | list = list->next; | |
499 | ||
500 | if (tasklet_trylock(t)) { | |
501 | if (!atomic_read(&t->count)) { | |
502 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
503 | BUG(); | |
504 | t->func(t->data); | |
505 | tasklet_unlock(t); | |
506 | continue; | |
507 | } | |
508 | tasklet_unlock(t); | |
509 | } | |
510 | ||
511 | local_irq_disable(); | |
48f20a9a | 512 | t->next = NULL; |
909ea964 CL |
513 | *__this_cpu_read(tasklet_vec.tail) = t; |
514 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
515 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
516 | local_irq_enable(); | |
517 | } | |
518 | } | |
519 | ||
520 | static void tasklet_hi_action(struct softirq_action *a) | |
521 | { | |
522 | struct tasklet_struct *list; | |
523 | ||
524 | local_irq_disable(); | |
909ea964 CL |
525 | list = __this_cpu_read(tasklet_hi_vec.head); |
526 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
527 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
528 | local_irq_enable(); |
529 | ||
530 | while (list) { | |
531 | struct tasklet_struct *t = list; | |
532 | ||
533 | list = list->next; | |
534 | ||
535 | if (tasklet_trylock(t)) { | |
536 | if (!atomic_read(&t->count)) { | |
537 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
538 | BUG(); | |
539 | t->func(t->data); | |
540 | tasklet_unlock(t); | |
541 | continue; | |
542 | } | |
543 | tasklet_unlock(t); | |
544 | } | |
545 | ||
546 | local_irq_disable(); | |
48f20a9a | 547 | t->next = NULL; |
909ea964 CL |
548 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
549 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
550 | __raise_softirq_irqoff(HI_SOFTIRQ); |
551 | local_irq_enable(); | |
552 | } | |
553 | } | |
554 | ||
555 | ||
556 | void tasklet_init(struct tasklet_struct *t, | |
557 | void (*func)(unsigned long), unsigned long data) | |
558 | { | |
559 | t->next = NULL; | |
560 | t->state = 0; | |
561 | atomic_set(&t->count, 0); | |
562 | t->func = func; | |
563 | t->data = data; | |
564 | } | |
565 | ||
566 | EXPORT_SYMBOL(tasklet_init); | |
567 | ||
568 | void tasklet_kill(struct tasklet_struct *t) | |
569 | { | |
570 | if (in_interrupt()) | |
571 | printk("Attempt to kill tasklet from interrupt\n"); | |
572 | ||
573 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 574 | do { |
1da177e4 | 575 | yield(); |
79d381c9 | 576 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
577 | } |
578 | tasklet_unlock_wait(t); | |
579 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
580 | } | |
581 | ||
582 | EXPORT_SYMBOL(tasklet_kill); | |
583 | ||
9ba5f005 PZ |
584 | /* |
585 | * tasklet_hrtimer | |
586 | */ | |
587 | ||
588 | /* | |
b9c30322 PZ |
589 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
590 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
591 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
592 | */ |
593 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
594 | { | |
595 | struct tasklet_hrtimer *ttimer = | |
596 | container_of(timer, struct tasklet_hrtimer, timer); | |
597 | ||
b9c30322 PZ |
598 | tasklet_hi_schedule(&ttimer->tasklet); |
599 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
600 | } |
601 | ||
602 | /* | |
603 | * Helper function which calls the hrtimer callback from | |
604 | * tasklet/softirq context | |
605 | */ | |
606 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
607 | { | |
608 | struct tasklet_hrtimer *ttimer = (void *)data; | |
609 | enum hrtimer_restart restart; | |
610 | ||
611 | restart = ttimer->function(&ttimer->timer); | |
612 | if (restart != HRTIMER_NORESTART) | |
613 | hrtimer_restart(&ttimer->timer); | |
614 | } | |
615 | ||
616 | /** | |
617 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
618 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 619 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
620 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
621 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
622 | */ | |
623 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
624 | enum hrtimer_restart (*function)(struct hrtimer *), | |
625 | clockid_t which_clock, enum hrtimer_mode mode) | |
626 | { | |
627 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
628 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
629 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
630 | (unsigned long)ttimer); | |
631 | ttimer->function = function; | |
632 | } | |
633 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
634 | ||
1da177e4 LT |
635 | void __init softirq_init(void) |
636 | { | |
48f20a9a OJ |
637 | int cpu; |
638 | ||
639 | for_each_possible_cpu(cpu) { | |
640 | per_cpu(tasklet_vec, cpu).tail = | |
641 | &per_cpu(tasklet_vec, cpu).head; | |
642 | per_cpu(tasklet_hi_vec, cpu).tail = | |
643 | &per_cpu(tasklet_hi_vec, cpu).head; | |
644 | } | |
645 | ||
962cf36c CM |
646 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
647 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
648 | } |
649 | ||
3e339b5d | 650 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 651 | { |
3e339b5d TG |
652 | return local_softirq_pending(); |
653 | } | |
1da177e4 | 654 | |
3e339b5d TG |
655 | static void run_ksoftirqd(unsigned int cpu) |
656 | { | |
657 | local_irq_disable(); | |
658 | if (local_softirq_pending()) { | |
0bed698a FW |
659 | /* |
660 | * We can safely run softirq on inline stack, as we are not deep | |
661 | * in the task stack here. | |
662 | */ | |
3e339b5d TG |
663 | __do_softirq(); |
664 | rcu_note_context_switch(cpu); | |
665 | local_irq_enable(); | |
666 | cond_resched(); | |
667 | return; | |
1da177e4 | 668 | } |
3e339b5d | 669 | local_irq_enable(); |
1da177e4 LT |
670 | } |
671 | ||
672 | #ifdef CONFIG_HOTPLUG_CPU | |
673 | /* | |
674 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
675 | * scheduled for execution on @cpu. | |
676 | * | |
677 | * Unlike tasklet_kill, this function removes the tasklet | |
678 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
679 | * | |
680 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
681 | */ | |
682 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
683 | { | |
684 | struct tasklet_struct **i; | |
685 | ||
686 | BUG_ON(cpu_online(cpu)); | |
687 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
688 | ||
689 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
690 | return; | |
691 | ||
692 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 693 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
694 | if (*i == t) { |
695 | *i = t->next; | |
48f20a9a OJ |
696 | /* If this was the tail element, move the tail ptr */ |
697 | if (*i == NULL) | |
698 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
699 | return; |
700 | } | |
701 | } | |
702 | BUG(); | |
703 | } | |
704 | ||
705 | static void takeover_tasklets(unsigned int cpu) | |
706 | { | |
1da177e4 LT |
707 | /* CPU is dead, so no lock needed. */ |
708 | local_irq_disable(); | |
709 | ||
710 | /* Find end, append list for that CPU. */ | |
e5e41723 | 711 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
712 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
713 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
714 | per_cpu(tasklet_vec, cpu).head = NULL; |
715 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
716 | } | |
1da177e4 LT |
717 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
718 | ||
e5e41723 | 719 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
720 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
721 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
722 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
723 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
724 | } | |
1da177e4 LT |
725 | raise_softirq_irqoff(HI_SOFTIRQ); |
726 | ||
727 | local_irq_enable(); | |
728 | } | |
729 | #endif /* CONFIG_HOTPLUG_CPU */ | |
730 | ||
0db0628d | 731 | static int cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
732 | unsigned long action, |
733 | void *hcpu) | |
734 | { | |
1da177e4 | 735 | switch (action) { |
1da177e4 | 736 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 737 | case CPU_DEAD: |
3e339b5d TG |
738 | case CPU_DEAD_FROZEN: |
739 | takeover_tasklets((unsigned long)hcpu); | |
1da177e4 LT |
740 | break; |
741 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3e339b5d | 742 | } |
1da177e4 LT |
743 | return NOTIFY_OK; |
744 | } | |
745 | ||
0db0628d | 746 | static struct notifier_block cpu_nfb = { |
1da177e4 LT |
747 | .notifier_call = cpu_callback |
748 | }; | |
749 | ||
3e339b5d TG |
750 | static struct smp_hotplug_thread softirq_threads = { |
751 | .store = &ksoftirqd, | |
752 | .thread_should_run = ksoftirqd_should_run, | |
753 | .thread_fn = run_ksoftirqd, | |
754 | .thread_comm = "ksoftirqd/%u", | |
755 | }; | |
756 | ||
7babe8db | 757 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 758 | { |
1da177e4 | 759 | register_cpu_notifier(&cpu_nfb); |
3e339b5d TG |
760 | |
761 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); | |
762 | ||
1da177e4 LT |
763 | return 0; |
764 | } | |
7babe8db | 765 | early_initcall(spawn_ksoftirqd); |
78eef01b | 766 | |
43a25632 YL |
767 | /* |
768 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
769 | * GCC does not inline them incorrectly. ] | |
770 | */ | |
771 | ||
772 | int __init __weak early_irq_init(void) | |
773 | { | |
774 | return 0; | |
775 | } | |
776 | ||
4a046d17 YL |
777 | int __init __weak arch_probe_nr_irqs(void) |
778 | { | |
b683de2b | 779 | return NR_IRQS_LEGACY; |
4a046d17 YL |
780 | } |
781 | ||
43a25632 YL |
782 | int __init __weak arch_early_irq_init(void) |
783 | { | |
784 | return 0; | |
785 | } |