]>
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 | ||
40322764 JP |
11 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
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 |
0bd3a173 | 94 | void __local_bh_disable_ip(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 | 101 | /* |
bdb43806 | 102 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
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 | */ | |
bdb43806 | 108 | __preempt_count_add(cnt); |
de30a2b3 IM |
109 | /* |
110 | * Were softirqs turned off above: | |
111 | */ | |
9ea4c380 | 112 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
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 | } |
0bd3a173 | 119 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 120 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 121 | |
75e1056f VP |
122 | static void __local_bh_enable(unsigned int cnt) |
123 | { | |
75e1056f VP |
124 | WARN_ON_ONCE(!irqs_disabled()); |
125 | ||
9ea4c380 | 126 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
d2e08473 | 127 | trace_softirqs_on(_RET_IP_); |
bdb43806 | 128 | preempt_count_sub(cnt); |
75e1056f VP |
129 | } |
130 | ||
de30a2b3 IM |
131 | /* |
132 | * Special-case - softirqs can safely be enabled in | |
133 | * cond_resched_softirq(), or by __do_softirq(), | |
134 | * without processing still-pending softirqs: | |
135 | */ | |
136 | void _local_bh_enable(void) | |
137 | { | |
5d60d3e7 | 138 | WARN_ON_ONCE(in_irq()); |
75e1056f | 139 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 IM |
140 | } |
141 | ||
142 | EXPORT_SYMBOL(_local_bh_enable); | |
143 | ||
0bd3a173 | 144 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 145 | { |
0f476b6d | 146 | WARN_ON_ONCE(in_irq() || irqs_disabled()); |
3c829c36 | 147 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 148 | local_irq_disable(); |
3c829c36 | 149 | #endif |
de30a2b3 IM |
150 | /* |
151 | * Are softirqs going to be turned on now: | |
152 | */ | |
75e1056f | 153 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 154 | trace_softirqs_on(ip); |
de30a2b3 IM |
155 | /* |
156 | * Keep preemption disabled until we are done with | |
157 | * softirq processing: | |
158 | */ | |
0bd3a173 | 159 | preempt_count_sub(cnt - 1); |
de30a2b3 | 160 | |
0bed698a FW |
161 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
162 | /* | |
163 | * Run softirq if any pending. And do it in its own stack | |
164 | * as we may be calling this deep in a task call stack already. | |
165 | */ | |
de30a2b3 | 166 | do_softirq(); |
0bed698a | 167 | } |
de30a2b3 | 168 | |
bdb43806 | 169 | preempt_count_dec(); |
3c829c36 | 170 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 171 | local_irq_enable(); |
3c829c36 | 172 | #endif |
de30a2b3 IM |
173 | preempt_check_resched(); |
174 | } | |
0bd3a173 | 175 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 176 | |
1da177e4 | 177 | /* |
34376a50 BG |
178 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
179 | * but break the loop if need_resched() is set or after 2 ms. | |
180 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
181 | * certain cases, such as stop_machine(), jiffies may cease to | |
182 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
183 | * well to make sure we eventually return from this method. | |
1da177e4 | 184 | * |
c10d7367 | 185 | * These limits have been established via experimentation. |
1da177e4 LT |
186 | * The two things to balance is latency against fairness - |
187 | * we want to handle softirqs as soon as possible, but they | |
188 | * should not be able to lock up the box. | |
189 | */ | |
c10d7367 | 190 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 191 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 192 | |
f1a83e65 PZ |
193 | #ifdef CONFIG_TRACE_IRQFLAGS |
194 | /* | |
f1a83e65 PZ |
195 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
196 | * to keep the lockdep irq context tracking as tight as possible in order to | |
197 | * not miss-qualify lock contexts and miss possible deadlocks. | |
198 | */ | |
f1a83e65 | 199 | |
5c4853b6 | 200 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 201 | { |
5c4853b6 | 202 | bool in_hardirq = false; |
f1a83e65 | 203 | |
5c4853b6 FW |
204 | if (trace_hardirq_context(current)) { |
205 | in_hardirq = true; | |
f1a83e65 | 206 | trace_hardirq_exit(); |
5c4853b6 FW |
207 | } |
208 | ||
f1a83e65 | 209 | lockdep_softirq_enter(); |
5c4853b6 FW |
210 | |
211 | return in_hardirq; | |
f1a83e65 PZ |
212 | } |
213 | ||
5c4853b6 | 214 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
215 | { |
216 | lockdep_softirq_exit(); | |
5c4853b6 FW |
217 | |
218 | if (in_hardirq) | |
f1a83e65 | 219 | trace_hardirq_enter(); |
f1a83e65 | 220 | } |
f1a83e65 | 221 | #else |
5c4853b6 FW |
222 | static inline bool lockdep_softirq_start(void) { return false; } |
223 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
224 | #endif |
225 | ||
1da177e4 LT |
226 | asmlinkage void __do_softirq(void) |
227 | { | |
c10d7367 | 228 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 229 | unsigned long old_flags = current->flags; |
34376a50 | 230 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 231 | struct softirq_action *h; |
5c4853b6 | 232 | bool in_hardirq; |
f1a83e65 | 233 | __u32 pending; |
2e702b9f | 234 | int softirq_bit; |
f1a83e65 | 235 | int cpu; |
907aed48 MG |
236 | |
237 | /* | |
238 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
239 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
240 | * again if the socket is related to swap | |
241 | */ | |
242 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
243 | |
244 | pending = local_softirq_pending(); | |
6a61671b | 245 | account_irq_enter_time(current); |
829035fd | 246 | |
0bd3a173 | 247 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 248 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 249 | |
1da177e4 LT |
250 | cpu = smp_processor_id(); |
251 | restart: | |
252 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 253 | set_softirq_pending(0); |
1da177e4 | 254 | |
c70f5d66 | 255 | local_irq_enable(); |
1da177e4 LT |
256 | |
257 | h = softirq_vec; | |
258 | ||
2e702b9f JP |
259 | while ((softirq_bit = ffs(pending))) { |
260 | unsigned int vec_nr; | |
261 | int prev_count; | |
262 | ||
263 | h += softirq_bit - 1; | |
264 | ||
265 | vec_nr = h - softirq_vec; | |
266 | prev_count = preempt_count(); | |
267 | ||
268 | kstat_incr_softirqs_this_cpu(vec_nr); | |
269 | ||
270 | trace_softirq_entry(vec_nr); | |
271 | h->action(h); | |
272 | trace_softirq_exit(vec_nr); | |
273 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 274 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
275 | vec_nr, softirq_to_name[vec_nr], h->action, |
276 | prev_count, preempt_count()); | |
277 | preempt_count_set(prev_count); | |
1da177e4 | 278 | } |
2e702b9f | 279 | rcu_bh_qs(cpu); |
1da177e4 | 280 | h++; |
2e702b9f JP |
281 | pending >>= softirq_bit; |
282 | } | |
1da177e4 | 283 | |
c70f5d66 | 284 | local_irq_disable(); |
1da177e4 LT |
285 | |
286 | pending = local_softirq_pending(); | |
c10d7367 | 287 | if (pending) { |
34376a50 BG |
288 | if (time_before(jiffies, end) && !need_resched() && |
289 | --max_restart) | |
c10d7367 | 290 | goto restart; |
1da177e4 | 291 | |
1da177e4 | 292 | wakeup_softirqd(); |
c10d7367 | 293 | } |
1da177e4 | 294 | |
5c4853b6 | 295 | lockdep_softirq_end(in_hardirq); |
6a61671b | 296 | account_irq_exit_time(current); |
75e1056f | 297 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 298 | WARN_ON_ONCE(in_interrupt()); |
907aed48 | 299 | tsk_restore_flags(current, old_flags, PF_MEMALLOC); |
1da177e4 LT |
300 | } |
301 | ||
1da177e4 LT |
302 | asmlinkage void do_softirq(void) |
303 | { | |
304 | __u32 pending; | |
305 | unsigned long flags; | |
306 | ||
307 | if (in_interrupt()) | |
308 | return; | |
309 | ||
310 | local_irq_save(flags); | |
311 | ||
312 | pending = local_softirq_pending(); | |
313 | ||
314 | if (pending) | |
7d65f4a6 | 315 | do_softirq_own_stack(); |
1da177e4 LT |
316 | |
317 | local_irq_restore(flags); | |
318 | } | |
319 | ||
dde4b2b5 IM |
320 | /* |
321 | * Enter an interrupt context. | |
322 | */ | |
323 | void irq_enter(void) | |
324 | { | |
64db4cff | 325 | rcu_irq_enter(); |
0a8a2e78 | 326 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
327 | /* |
328 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
329 | * here, as softirq will be serviced on return from interrupt. | |
330 | */ | |
331 | local_bh_disable(); | |
e8fcaa5c | 332 | tick_check_idle(); |
d267f87f VP |
333 | _local_bh_enable(); |
334 | } | |
335 | ||
336 | __irq_enter(); | |
dde4b2b5 IM |
337 | } |
338 | ||
8d32a307 TG |
339 | static inline void invoke_softirq(void) |
340 | { | |
ded79754 | 341 | if (!force_irqthreads) { |
cc1f0274 | 342 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
343 | /* |
344 | * We can safely execute softirq on the current stack if | |
345 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
346 | * at this stage. |
347 | */ | |
348 | __do_softirq(); | |
349 | #else | |
350 | /* | |
351 | * Otherwise, irq_exit() is called on the task stack that can | |
352 | * be potentially deep already. So call softirq in its own stack | |
353 | * to prevent from any overrun. | |
ded79754 | 354 | */ |
be6e1016 | 355 | do_softirq_own_stack(); |
cc1f0274 | 356 | #endif |
ded79754 | 357 | } else { |
8d32a307 | 358 | wakeup_softirqd(); |
ded79754 | 359 | } |
8d32a307 | 360 | } |
1da177e4 | 361 | |
67826eae FW |
362 | static inline void tick_irq_exit(void) |
363 | { | |
364 | #ifdef CONFIG_NO_HZ_COMMON | |
365 | int cpu = smp_processor_id(); | |
366 | ||
367 | /* Make sure that timer wheel updates are propagated */ | |
368 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
369 | if (!in_interrupt()) | |
370 | tick_nohz_irq_exit(); | |
371 | } | |
372 | #endif | |
373 | } | |
374 | ||
1da177e4 LT |
375 | /* |
376 | * Exit an interrupt context. Process softirqs if needed and possible: | |
377 | */ | |
378 | void irq_exit(void) | |
379 | { | |
74eed016 | 380 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 381 | local_irq_disable(); |
74eed016 TG |
382 | #else |
383 | WARN_ON_ONCE(!irqs_disabled()); | |
384 | #endif | |
385 | ||
6a61671b | 386 | account_irq_exit_time(current); |
bdb43806 | 387 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
388 | if (!in_interrupt() && local_softirq_pending()) |
389 | invoke_softirq(); | |
79bf2bb3 | 390 | |
67826eae | 391 | tick_irq_exit(); |
416eb33c | 392 | rcu_irq_exit(); |
f1a83e65 | 393 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
394 | } |
395 | ||
396 | /* | |
397 | * This function must run with irqs disabled! | |
398 | */ | |
7ad5b3a5 | 399 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
400 | { |
401 | __raise_softirq_irqoff(nr); | |
402 | ||
403 | /* | |
404 | * If we're in an interrupt or softirq, we're done | |
405 | * (this also catches softirq-disabled code). We will | |
406 | * actually run the softirq once we return from | |
407 | * the irq or softirq. | |
408 | * | |
409 | * Otherwise we wake up ksoftirqd to make sure we | |
410 | * schedule the softirq soon. | |
411 | */ | |
412 | if (!in_interrupt()) | |
413 | wakeup_softirqd(); | |
414 | } | |
415 | ||
7ad5b3a5 | 416 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
417 | { |
418 | unsigned long flags; | |
419 | ||
420 | local_irq_save(flags); | |
421 | raise_softirq_irqoff(nr); | |
422 | local_irq_restore(flags); | |
423 | } | |
424 | ||
f069686e SR |
425 | void __raise_softirq_irqoff(unsigned int nr) |
426 | { | |
427 | trace_softirq_raise(nr); | |
428 | or_softirq_pending(1UL << nr); | |
429 | } | |
430 | ||
962cf36c | 431 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 432 | { |
1da177e4 LT |
433 | softirq_vec[nr].action = action; |
434 | } | |
435 | ||
9ba5f005 PZ |
436 | /* |
437 | * Tasklets | |
438 | */ | |
1da177e4 LT |
439 | struct tasklet_head |
440 | { | |
48f20a9a OJ |
441 | struct tasklet_struct *head; |
442 | struct tasklet_struct **tail; | |
1da177e4 LT |
443 | }; |
444 | ||
4620b49f VN |
445 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
446 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 447 | |
7ad5b3a5 | 448 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
449 | { |
450 | unsigned long flags; | |
451 | ||
452 | local_irq_save(flags); | |
48f20a9a | 453 | t->next = NULL; |
909ea964 CL |
454 | *__this_cpu_read(tasklet_vec.tail) = t; |
455 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
456 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
457 | local_irq_restore(flags); | |
458 | } | |
459 | ||
460 | EXPORT_SYMBOL(__tasklet_schedule); | |
461 | ||
7ad5b3a5 | 462 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
463 | { |
464 | unsigned long flags; | |
465 | ||
466 | local_irq_save(flags); | |
48f20a9a | 467 | t->next = NULL; |
909ea964 CL |
468 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
469 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
470 | raise_softirq_irqoff(HI_SOFTIRQ); |
471 | local_irq_restore(flags); | |
472 | } | |
473 | ||
474 | EXPORT_SYMBOL(__tasklet_hi_schedule); | |
475 | ||
7c692cba VN |
476 | void __tasklet_hi_schedule_first(struct tasklet_struct *t) |
477 | { | |
478 | BUG_ON(!irqs_disabled()); | |
479 | ||
909ea964 CL |
480 | t->next = __this_cpu_read(tasklet_hi_vec.head); |
481 | __this_cpu_write(tasklet_hi_vec.head, t); | |
7c692cba VN |
482 | __raise_softirq_irqoff(HI_SOFTIRQ); |
483 | } | |
484 | ||
485 | EXPORT_SYMBOL(__tasklet_hi_schedule_first); | |
486 | ||
1da177e4 LT |
487 | static void tasklet_action(struct softirq_action *a) |
488 | { | |
489 | struct tasklet_struct *list; | |
490 | ||
491 | local_irq_disable(); | |
909ea964 CL |
492 | list = __this_cpu_read(tasklet_vec.head); |
493 | __this_cpu_write(tasklet_vec.head, NULL); | |
494 | __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head); | |
1da177e4 LT |
495 | local_irq_enable(); |
496 | ||
497 | while (list) { | |
498 | struct tasklet_struct *t = list; | |
499 | ||
500 | list = list->next; | |
501 | ||
502 | if (tasklet_trylock(t)) { | |
503 | if (!atomic_read(&t->count)) { | |
504 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
505 | BUG(); | |
506 | t->func(t->data); | |
507 | tasklet_unlock(t); | |
508 | continue; | |
509 | } | |
510 | tasklet_unlock(t); | |
511 | } | |
512 | ||
513 | local_irq_disable(); | |
48f20a9a | 514 | t->next = NULL; |
909ea964 CL |
515 | *__this_cpu_read(tasklet_vec.tail) = t; |
516 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
517 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
518 | local_irq_enable(); | |
519 | } | |
520 | } | |
521 | ||
522 | static void tasklet_hi_action(struct softirq_action *a) | |
523 | { | |
524 | struct tasklet_struct *list; | |
525 | ||
526 | local_irq_disable(); | |
909ea964 CL |
527 | list = __this_cpu_read(tasklet_hi_vec.head); |
528 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
529 | __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head); | |
1da177e4 LT |
530 | local_irq_enable(); |
531 | ||
532 | while (list) { | |
533 | struct tasklet_struct *t = list; | |
534 | ||
535 | list = list->next; | |
536 | ||
537 | if (tasklet_trylock(t)) { | |
538 | if (!atomic_read(&t->count)) { | |
539 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) | |
540 | BUG(); | |
541 | t->func(t->data); | |
542 | tasklet_unlock(t); | |
543 | continue; | |
544 | } | |
545 | tasklet_unlock(t); | |
546 | } | |
547 | ||
548 | local_irq_disable(); | |
48f20a9a | 549 | t->next = NULL; |
909ea964 CL |
550 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
551 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
552 | __raise_softirq_irqoff(HI_SOFTIRQ); |
553 | local_irq_enable(); | |
554 | } | |
555 | } | |
556 | ||
557 | ||
558 | void tasklet_init(struct tasklet_struct *t, | |
559 | void (*func)(unsigned long), unsigned long data) | |
560 | { | |
561 | t->next = NULL; | |
562 | t->state = 0; | |
563 | atomic_set(&t->count, 0); | |
564 | t->func = func; | |
565 | t->data = data; | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL(tasklet_init); | |
569 | ||
570 | void tasklet_kill(struct tasklet_struct *t) | |
571 | { | |
572 | if (in_interrupt()) | |
40322764 | 573 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 LT |
574 | |
575 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 576 | do { |
1da177e4 | 577 | yield(); |
79d381c9 | 578 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
579 | } |
580 | tasklet_unlock_wait(t); | |
581 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
582 | } | |
583 | ||
584 | EXPORT_SYMBOL(tasklet_kill); | |
585 | ||
9ba5f005 PZ |
586 | /* |
587 | * tasklet_hrtimer | |
588 | */ | |
589 | ||
590 | /* | |
b9c30322 PZ |
591 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
592 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
593 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
594 | */ |
595 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
596 | { | |
597 | struct tasklet_hrtimer *ttimer = | |
598 | container_of(timer, struct tasklet_hrtimer, timer); | |
599 | ||
b9c30322 PZ |
600 | tasklet_hi_schedule(&ttimer->tasklet); |
601 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
602 | } |
603 | ||
604 | /* | |
605 | * Helper function which calls the hrtimer callback from | |
606 | * tasklet/softirq context | |
607 | */ | |
608 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
609 | { | |
610 | struct tasklet_hrtimer *ttimer = (void *)data; | |
611 | enum hrtimer_restart restart; | |
612 | ||
613 | restart = ttimer->function(&ttimer->timer); | |
614 | if (restart != HRTIMER_NORESTART) | |
615 | hrtimer_restart(&ttimer->timer); | |
616 | } | |
617 | ||
618 | /** | |
619 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
620 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 621 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
622 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
623 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
624 | */ | |
625 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
626 | enum hrtimer_restart (*function)(struct hrtimer *), | |
627 | clockid_t which_clock, enum hrtimer_mode mode) | |
628 | { | |
629 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
630 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
631 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
632 | (unsigned long)ttimer); | |
633 | ttimer->function = function; | |
634 | } | |
635 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
636 | ||
1da177e4 LT |
637 | void __init softirq_init(void) |
638 | { | |
48f20a9a OJ |
639 | int cpu; |
640 | ||
641 | for_each_possible_cpu(cpu) { | |
642 | per_cpu(tasklet_vec, cpu).tail = | |
643 | &per_cpu(tasklet_vec, cpu).head; | |
644 | per_cpu(tasklet_hi_vec, cpu).tail = | |
645 | &per_cpu(tasklet_hi_vec, cpu).head; | |
646 | } | |
647 | ||
962cf36c CM |
648 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
649 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
650 | } |
651 | ||
3e339b5d | 652 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 653 | { |
3e339b5d TG |
654 | return local_softirq_pending(); |
655 | } | |
1da177e4 | 656 | |
3e339b5d TG |
657 | static void run_ksoftirqd(unsigned int cpu) |
658 | { | |
659 | local_irq_disable(); | |
660 | if (local_softirq_pending()) { | |
0bed698a FW |
661 | /* |
662 | * We can safely run softirq on inline stack, as we are not deep | |
663 | * in the task stack here. | |
664 | */ | |
3e339b5d TG |
665 | __do_softirq(); |
666 | rcu_note_context_switch(cpu); | |
667 | local_irq_enable(); | |
668 | cond_resched(); | |
669 | return; | |
1da177e4 | 670 | } |
3e339b5d | 671 | local_irq_enable(); |
1da177e4 LT |
672 | } |
673 | ||
674 | #ifdef CONFIG_HOTPLUG_CPU | |
675 | /* | |
676 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
677 | * scheduled for execution on @cpu. | |
678 | * | |
679 | * Unlike tasklet_kill, this function removes the tasklet | |
680 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
681 | * | |
682 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
683 | */ | |
684 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
685 | { | |
686 | struct tasklet_struct **i; | |
687 | ||
688 | BUG_ON(cpu_online(cpu)); | |
689 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
690 | ||
691 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
692 | return; | |
693 | ||
694 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 695 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
696 | if (*i == t) { |
697 | *i = t->next; | |
48f20a9a OJ |
698 | /* If this was the tail element, move the tail ptr */ |
699 | if (*i == NULL) | |
700 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
701 | return; |
702 | } | |
703 | } | |
704 | BUG(); | |
705 | } | |
706 | ||
707 | static void takeover_tasklets(unsigned int cpu) | |
708 | { | |
1da177e4 LT |
709 | /* CPU is dead, so no lock needed. */ |
710 | local_irq_disable(); | |
711 | ||
712 | /* Find end, append list for that CPU. */ | |
e5e41723 | 713 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
714 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
715 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
716 | per_cpu(tasklet_vec, cpu).head = NULL; |
717 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
718 | } | |
1da177e4 LT |
719 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
720 | ||
e5e41723 | 721 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
722 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
723 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
724 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
725 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
726 | } | |
1da177e4 LT |
727 | raise_softirq_irqoff(HI_SOFTIRQ); |
728 | ||
729 | local_irq_enable(); | |
730 | } | |
731 | #endif /* CONFIG_HOTPLUG_CPU */ | |
732 | ||
0db0628d | 733 | static int cpu_callback(struct notifier_block *nfb, |
1da177e4 LT |
734 | unsigned long action, |
735 | void *hcpu) | |
736 | { | |
1da177e4 | 737 | switch (action) { |
1da177e4 | 738 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 739 | case CPU_DEAD: |
3e339b5d TG |
740 | case CPU_DEAD_FROZEN: |
741 | takeover_tasklets((unsigned long)hcpu); | |
1da177e4 LT |
742 | break; |
743 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3e339b5d | 744 | } |
1da177e4 LT |
745 | return NOTIFY_OK; |
746 | } | |
747 | ||
0db0628d | 748 | static struct notifier_block cpu_nfb = { |
1da177e4 LT |
749 | .notifier_call = cpu_callback |
750 | }; | |
751 | ||
3e339b5d TG |
752 | static struct smp_hotplug_thread softirq_threads = { |
753 | .store = &ksoftirqd, | |
754 | .thread_should_run = ksoftirqd_should_run, | |
755 | .thread_fn = run_ksoftirqd, | |
756 | .thread_comm = "ksoftirqd/%u", | |
757 | }; | |
758 | ||
7babe8db | 759 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 760 | { |
1da177e4 | 761 | register_cpu_notifier(&cpu_nfb); |
3e339b5d TG |
762 | |
763 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); | |
764 | ||
1da177e4 LT |
765 | return 0; |
766 | } | |
7babe8db | 767 | early_initcall(spawn_ksoftirqd); |
78eef01b | 768 | |
43a25632 YL |
769 | /* |
770 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
771 | * GCC does not inline them incorrectly. ] | |
772 | */ | |
773 | ||
774 | int __init __weak early_irq_init(void) | |
775 | { | |
776 | return 0; | |
777 | } | |
778 | ||
4a046d17 YL |
779 | int __init __weak arch_probe_nr_irqs(void) |
780 | { | |
b683de2b | 781 | return NR_IRQS_LEGACY; |
4a046d17 YL |
782 | } |
783 | ||
43a25632 YL |
784 | int __init __weak arch_early_irq_init(void) |
785 | { | |
786 | return 0; | |
787 | } |