]>
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> |
d532676c | 28 | #include <linux/irq.h> |
a0e39ed3 HC |
29 | |
30 | #define CREATE_TRACE_POINTS | |
ad8d75ff | 31 | #include <trace/events/irq.h> |
1da177e4 | 32 | |
1da177e4 LT |
33 | /* |
34 | - No shared variables, all the data are CPU local. | |
35 | - If a softirq needs serialization, let it serialize itself | |
36 | by its own spinlocks. | |
37 | - Even if softirq is serialized, only local cpu is marked for | |
38 | execution. Hence, we get something sort of weak cpu binding. | |
39 | Though it is still not clear, will it result in better locality | |
40 | or will not. | |
41 | ||
42 | Examples: | |
43 | - NET RX softirq. It is multithreaded and does not require | |
44 | any global serialization. | |
45 | - NET TX softirq. It kicks software netdevice queues, hence | |
46 | it is logically serialized per device, but this serialization | |
47 | is invisible to common code. | |
48 | - Tasklets: serialized wrt itself. | |
49 | */ | |
50 | ||
51 | #ifndef __ARCH_IRQ_STAT | |
52 | irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned; | |
53 | EXPORT_SYMBOL(irq_stat); | |
54 | #endif | |
55 | ||
978b0116 | 56 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 57 | |
4dd53d89 | 58 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 59 | |
ce85b4f2 | 60 | const char * const softirq_to_name[NR_SOFTIRQS] = { |
f660f606 | 61 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL", |
09223371 | 62 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
5d592b44 JB |
63 | }; |
64 | ||
1da177e4 LT |
65 | /* |
66 | * we cannot loop indefinitely here to avoid userspace starvation, | |
67 | * but we also don't want to introduce a worst case 1/HZ latency | |
68 | * to the pending events, so lets the scheduler to balance | |
69 | * the softirq load for us. | |
70 | */ | |
676cb02d | 71 | static void wakeup_softirqd(void) |
1da177e4 LT |
72 | { |
73 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 74 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 LT |
75 | |
76 | if (tsk && tsk->state != TASK_RUNNING) | |
77 | wake_up_process(tsk); | |
78 | } | |
79 | ||
4cd13c21 ED |
80 | /* |
81 | * If ksoftirqd is scheduled, we do not want to process pending softirqs | |
82 | * right now. Let ksoftirqd handle this at its own rate, to get fairness. | |
83 | */ | |
84 | static bool ksoftirqd_running(void) | |
85 | { | |
86 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); | |
87 | ||
88 | return tsk && (tsk->state == TASK_RUNNING); | |
89 | } | |
90 | ||
75e1056f VP |
91 | /* |
92 | * preempt_count and SOFTIRQ_OFFSET usage: | |
93 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving | |
94 | * softirq processing. | |
95 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
96 | * on local_bh_disable or local_bh_enable. | |
97 | * This lets us distinguish between whether we are currently processing | |
98 | * softirq and whether we just have bh disabled. | |
99 | */ | |
100 | ||
de30a2b3 IM |
101 | /* |
102 | * This one is for softirq.c-internal use, | |
103 | * where hardirqs are disabled legitimately: | |
104 | */ | |
3c829c36 | 105 | #ifdef CONFIG_TRACE_IRQFLAGS |
0bd3a173 | 106 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
107 | { |
108 | unsigned long flags; | |
109 | ||
110 | WARN_ON_ONCE(in_irq()); | |
111 | ||
112 | raw_local_irq_save(flags); | |
7e49fcce | 113 | /* |
bdb43806 | 114 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
115 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
116 | * is set and before current->softirq_enabled is cleared. | |
117 | * We must manually increment preempt_count here and manually | |
118 | * call the trace_preempt_off later. | |
119 | */ | |
bdb43806 | 120 | __preempt_count_add(cnt); |
de30a2b3 IM |
121 | /* |
122 | * Were softirqs turned off above: | |
123 | */ | |
9ea4c380 | 124 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
de30a2b3 IM |
125 | trace_softirqs_off(ip); |
126 | raw_local_irq_restore(flags); | |
7e49fcce | 127 | |
0f1ba9a2 HC |
128 | if (preempt_count() == cnt) { |
129 | #ifdef CONFIG_DEBUG_PREEMPT | |
f904f582 | 130 | current->preempt_disable_ip = get_lock_parent_ip(); |
0f1ba9a2 | 131 | #endif |
f904f582 | 132 | trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip()); |
0f1ba9a2 | 133 | } |
de30a2b3 | 134 | } |
0bd3a173 | 135 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 136 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 137 | |
75e1056f VP |
138 | static void __local_bh_enable(unsigned int cnt) |
139 | { | |
f71b74bc | 140 | lockdep_assert_irqs_disabled(); |
75e1056f | 141 | |
9ea4c380 | 142 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
d2e08473 | 143 | trace_softirqs_on(_RET_IP_); |
bdb43806 | 144 | preempt_count_sub(cnt); |
75e1056f VP |
145 | } |
146 | ||
de30a2b3 IM |
147 | /* |
148 | * Special-case - softirqs can safely be enabled in | |
149 | * cond_resched_softirq(), or by __do_softirq(), | |
150 | * without processing still-pending softirqs: | |
151 | */ | |
152 | void _local_bh_enable(void) | |
153 | { | |
5d60d3e7 | 154 | WARN_ON_ONCE(in_irq()); |
75e1056f | 155 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 156 | } |
de30a2b3 IM |
157 | EXPORT_SYMBOL(_local_bh_enable); |
158 | ||
0bd3a173 | 159 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 160 | { |
f71b74bc FW |
161 | WARN_ON_ONCE(in_irq()); |
162 | lockdep_assert_irqs_enabled(); | |
3c829c36 | 163 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 164 | local_irq_disable(); |
3c829c36 | 165 | #endif |
de30a2b3 IM |
166 | /* |
167 | * Are softirqs going to be turned on now: | |
168 | */ | |
75e1056f | 169 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 170 | trace_softirqs_on(ip); |
de30a2b3 IM |
171 | /* |
172 | * Keep preemption disabled until we are done with | |
173 | * softirq processing: | |
ce85b4f2 | 174 | */ |
0bd3a173 | 175 | preempt_count_sub(cnt - 1); |
de30a2b3 | 176 | |
0bed698a FW |
177 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
178 | /* | |
179 | * Run softirq if any pending. And do it in its own stack | |
180 | * as we may be calling this deep in a task call stack already. | |
181 | */ | |
de30a2b3 | 182 | do_softirq(); |
0bed698a | 183 | } |
de30a2b3 | 184 | |
bdb43806 | 185 | preempt_count_dec(); |
3c829c36 | 186 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 187 | local_irq_enable(); |
3c829c36 | 188 | #endif |
de30a2b3 IM |
189 | preempt_check_resched(); |
190 | } | |
0bd3a173 | 191 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 192 | |
1da177e4 | 193 | /* |
34376a50 BG |
194 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
195 | * but break the loop if need_resched() is set or after 2 ms. | |
196 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
197 | * certain cases, such as stop_machine(), jiffies may cease to | |
198 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
199 | * well to make sure we eventually return from this method. | |
1da177e4 | 200 | * |
c10d7367 | 201 | * These limits have been established via experimentation. |
1da177e4 LT |
202 | * The two things to balance is latency against fairness - |
203 | * we want to handle softirqs as soon as possible, but they | |
204 | * should not be able to lock up the box. | |
205 | */ | |
c10d7367 | 206 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 207 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 208 | |
f1a83e65 PZ |
209 | #ifdef CONFIG_TRACE_IRQFLAGS |
210 | /* | |
f1a83e65 PZ |
211 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
212 | * to keep the lockdep irq context tracking as tight as possible in order to | |
213 | * not miss-qualify lock contexts and miss possible deadlocks. | |
214 | */ | |
f1a83e65 | 215 | |
5c4853b6 | 216 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 217 | { |
5c4853b6 | 218 | bool in_hardirq = false; |
f1a83e65 | 219 | |
5c4853b6 FW |
220 | if (trace_hardirq_context(current)) { |
221 | in_hardirq = true; | |
f1a83e65 | 222 | trace_hardirq_exit(); |
5c4853b6 FW |
223 | } |
224 | ||
f1a83e65 | 225 | lockdep_softirq_enter(); |
5c4853b6 FW |
226 | |
227 | return in_hardirq; | |
f1a83e65 PZ |
228 | } |
229 | ||
5c4853b6 | 230 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
231 | { |
232 | lockdep_softirq_exit(); | |
5c4853b6 FW |
233 | |
234 | if (in_hardirq) | |
f1a83e65 | 235 | trace_hardirq_enter(); |
f1a83e65 | 236 | } |
f1a83e65 | 237 | #else |
5c4853b6 FW |
238 | static inline bool lockdep_softirq_start(void) { return false; } |
239 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
240 | #endif |
241 | ||
be7635e7 | 242 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 243 | { |
c10d7367 | 244 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 245 | unsigned long old_flags = current->flags; |
34376a50 | 246 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 247 | struct softirq_action *h; |
5c4853b6 | 248 | bool in_hardirq; |
f1a83e65 | 249 | __u32 pending; |
2e702b9f | 250 | int softirq_bit; |
907aed48 MG |
251 | |
252 | /* | |
253 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
254 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
255 | * again if the socket is related to swap | |
256 | */ | |
257 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
258 | |
259 | pending = local_softirq_pending(); | |
6a61671b | 260 | account_irq_enter_time(current); |
829035fd | 261 | |
0bd3a173 | 262 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 263 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 264 | |
1da177e4 LT |
265 | restart: |
266 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 267 | set_softirq_pending(0); |
1da177e4 | 268 | |
c70f5d66 | 269 | local_irq_enable(); |
1da177e4 LT |
270 | |
271 | h = softirq_vec; | |
272 | ||
2e702b9f JP |
273 | while ((softirq_bit = ffs(pending))) { |
274 | unsigned int vec_nr; | |
275 | int prev_count; | |
276 | ||
277 | h += softirq_bit - 1; | |
278 | ||
279 | vec_nr = h - softirq_vec; | |
280 | prev_count = preempt_count(); | |
281 | ||
282 | kstat_incr_softirqs_this_cpu(vec_nr); | |
283 | ||
284 | trace_softirq_entry(vec_nr); | |
285 | h->action(h); | |
286 | trace_softirq_exit(vec_nr); | |
287 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 288 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
289 | vec_nr, softirq_to_name[vec_nr], h->action, |
290 | prev_count, preempt_count()); | |
291 | preempt_count_set(prev_count); | |
1da177e4 LT |
292 | } |
293 | h++; | |
2e702b9f JP |
294 | pending >>= softirq_bit; |
295 | } | |
1da177e4 | 296 | |
284a8c93 | 297 | rcu_bh_qs(); |
c70f5d66 | 298 | local_irq_disable(); |
1da177e4 LT |
299 | |
300 | pending = local_softirq_pending(); | |
c10d7367 | 301 | if (pending) { |
34376a50 BG |
302 | if (time_before(jiffies, end) && !need_resched() && |
303 | --max_restart) | |
c10d7367 | 304 | goto restart; |
1da177e4 | 305 | |
1da177e4 | 306 | wakeup_softirqd(); |
c10d7367 | 307 | } |
1da177e4 | 308 | |
5c4853b6 | 309 | lockdep_softirq_end(in_hardirq); |
6a61671b | 310 | account_irq_exit_time(current); |
75e1056f | 311 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 312 | WARN_ON_ONCE(in_interrupt()); |
717a94b5 | 313 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
314 | } |
315 | ||
722a9f92 | 316 | asmlinkage __visible void do_softirq(void) |
1da177e4 LT |
317 | { |
318 | __u32 pending; | |
319 | unsigned long flags; | |
320 | ||
321 | if (in_interrupt()) | |
322 | return; | |
323 | ||
324 | local_irq_save(flags); | |
325 | ||
326 | pending = local_softirq_pending(); | |
327 | ||
4cd13c21 | 328 | if (pending && !ksoftirqd_running()) |
7d65f4a6 | 329 | do_softirq_own_stack(); |
1da177e4 LT |
330 | |
331 | local_irq_restore(flags); | |
332 | } | |
333 | ||
dde4b2b5 IM |
334 | /* |
335 | * Enter an interrupt context. | |
336 | */ | |
337 | void irq_enter(void) | |
338 | { | |
64db4cff | 339 | rcu_irq_enter(); |
0a8a2e78 | 340 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
341 | /* |
342 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
343 | * here, as softirq will be serviced on return from interrupt. | |
344 | */ | |
345 | local_bh_disable(); | |
5acac1be | 346 | tick_irq_enter(); |
d267f87f VP |
347 | _local_bh_enable(); |
348 | } | |
349 | ||
350 | __irq_enter(); | |
dde4b2b5 IM |
351 | } |
352 | ||
8d32a307 TG |
353 | static inline void invoke_softirq(void) |
354 | { | |
4cd13c21 ED |
355 | if (ksoftirqd_running()) |
356 | return; | |
357 | ||
ded79754 | 358 | if (!force_irqthreads) { |
cc1f0274 | 359 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
360 | /* |
361 | * We can safely execute softirq on the current stack if | |
362 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
363 | * at this stage. |
364 | */ | |
365 | __do_softirq(); | |
366 | #else | |
367 | /* | |
368 | * Otherwise, irq_exit() is called on the task stack that can | |
369 | * be potentially deep already. So call softirq in its own stack | |
370 | * to prevent from any overrun. | |
ded79754 | 371 | */ |
be6e1016 | 372 | do_softirq_own_stack(); |
cc1f0274 | 373 | #endif |
ded79754 | 374 | } else { |
8d32a307 | 375 | wakeup_softirqd(); |
ded79754 | 376 | } |
8d32a307 | 377 | } |
1da177e4 | 378 | |
67826eae FW |
379 | static inline void tick_irq_exit(void) |
380 | { | |
381 | #ifdef CONFIG_NO_HZ_COMMON | |
382 | int cpu = smp_processor_id(); | |
383 | ||
384 | /* Make sure that timer wheel updates are propagated */ | |
385 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
386 | if (!in_interrupt()) | |
387 | tick_nohz_irq_exit(); | |
388 | } | |
389 | #endif | |
390 | } | |
391 | ||
1da177e4 LT |
392 | /* |
393 | * Exit an interrupt context. Process softirqs if needed and possible: | |
394 | */ | |
395 | void irq_exit(void) | |
396 | { | |
74eed016 | 397 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 398 | local_irq_disable(); |
74eed016 | 399 | #else |
f71b74bc | 400 | lockdep_assert_irqs_disabled(); |
74eed016 | 401 | #endif |
6a61671b | 402 | account_irq_exit_time(current); |
bdb43806 | 403 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
404 | if (!in_interrupt() && local_softirq_pending()) |
405 | invoke_softirq(); | |
79bf2bb3 | 406 | |
67826eae | 407 | tick_irq_exit(); |
416eb33c | 408 | rcu_irq_exit(); |
f1a83e65 | 409 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
410 | } |
411 | ||
412 | /* | |
413 | * This function must run with irqs disabled! | |
414 | */ | |
7ad5b3a5 | 415 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
416 | { |
417 | __raise_softirq_irqoff(nr); | |
418 | ||
419 | /* | |
420 | * If we're in an interrupt or softirq, we're done | |
421 | * (this also catches softirq-disabled code). We will | |
422 | * actually run the softirq once we return from | |
423 | * the irq or softirq. | |
424 | * | |
425 | * Otherwise we wake up ksoftirqd to make sure we | |
426 | * schedule the softirq soon. | |
427 | */ | |
428 | if (!in_interrupt()) | |
429 | wakeup_softirqd(); | |
430 | } | |
431 | ||
7ad5b3a5 | 432 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
433 | { |
434 | unsigned long flags; | |
435 | ||
436 | local_irq_save(flags); | |
437 | raise_softirq_irqoff(nr); | |
438 | local_irq_restore(flags); | |
439 | } | |
440 | ||
f069686e SR |
441 | void __raise_softirq_irqoff(unsigned int nr) |
442 | { | |
443 | trace_softirq_raise(nr); | |
444 | or_softirq_pending(1UL << nr); | |
445 | } | |
446 | ||
962cf36c | 447 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 448 | { |
1da177e4 LT |
449 | softirq_vec[nr].action = action; |
450 | } | |
451 | ||
9ba5f005 PZ |
452 | /* |
453 | * Tasklets | |
454 | */ | |
ce85b4f2 | 455 | struct tasklet_head { |
48f20a9a OJ |
456 | struct tasklet_struct *head; |
457 | struct tasklet_struct **tail; | |
1da177e4 LT |
458 | }; |
459 | ||
4620b49f VN |
460 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
461 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 462 | |
7ad5b3a5 | 463 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
464 | { |
465 | unsigned long flags; | |
466 | ||
467 | local_irq_save(flags); | |
48f20a9a | 468 | t->next = NULL; |
909ea964 CL |
469 | *__this_cpu_read(tasklet_vec.tail) = t; |
470 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
471 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
472 | local_irq_restore(flags); | |
473 | } | |
1da177e4 LT |
474 | EXPORT_SYMBOL(__tasklet_schedule); |
475 | ||
7ad5b3a5 | 476 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
477 | { |
478 | unsigned long flags; | |
479 | ||
480 | local_irq_save(flags); | |
48f20a9a | 481 | t->next = NULL; |
909ea964 CL |
482 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
483 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
484 | raise_softirq_irqoff(HI_SOFTIRQ); |
485 | local_irq_restore(flags); | |
486 | } | |
1da177e4 LT |
487 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
488 | ||
0766f788 | 489 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 LT |
490 | { |
491 | struct tasklet_struct *list; | |
492 | ||
493 | local_irq_disable(); | |
909ea964 CL |
494 | list = __this_cpu_read(tasklet_vec.head); |
495 | __this_cpu_write(tasklet_vec.head, NULL); | |
22127e93 | 496 | __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head)); |
1da177e4 LT |
497 | local_irq_enable(); |
498 | ||
499 | while (list) { | |
500 | struct tasklet_struct *t = list; | |
501 | ||
502 | list = list->next; | |
503 | ||
504 | if (tasklet_trylock(t)) { | |
505 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
506 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
507 | &t->state)) | |
1da177e4 LT |
508 | BUG(); |
509 | t->func(t->data); | |
510 | tasklet_unlock(t); | |
511 | continue; | |
512 | } | |
513 | tasklet_unlock(t); | |
514 | } | |
515 | ||
516 | local_irq_disable(); | |
48f20a9a | 517 | t->next = NULL; |
909ea964 CL |
518 | *__this_cpu_read(tasklet_vec.tail) = t; |
519 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
520 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
521 | local_irq_enable(); | |
522 | } | |
523 | } | |
524 | ||
0766f788 | 525 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
1da177e4 LT |
526 | { |
527 | struct tasklet_struct *list; | |
528 | ||
529 | local_irq_disable(); | |
909ea964 CL |
530 | list = __this_cpu_read(tasklet_hi_vec.head); |
531 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
22127e93 | 532 | __this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head)); |
1da177e4 LT |
533 | local_irq_enable(); |
534 | ||
535 | while (list) { | |
536 | struct tasklet_struct *t = list; | |
537 | ||
538 | list = list->next; | |
539 | ||
540 | if (tasklet_trylock(t)) { | |
541 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
542 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
543 | &t->state)) | |
1da177e4 LT |
544 | BUG(); |
545 | t->func(t->data); | |
546 | tasklet_unlock(t); | |
547 | continue; | |
548 | } | |
549 | tasklet_unlock(t); | |
550 | } | |
551 | ||
552 | local_irq_disable(); | |
48f20a9a | 553 | t->next = NULL; |
909ea964 CL |
554 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
555 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
556 | __raise_softirq_irqoff(HI_SOFTIRQ); |
557 | local_irq_enable(); | |
558 | } | |
559 | } | |
560 | ||
1da177e4 LT |
561 | void tasklet_init(struct tasklet_struct *t, |
562 | void (*func)(unsigned long), unsigned long data) | |
563 | { | |
564 | t->next = NULL; | |
565 | t->state = 0; | |
566 | atomic_set(&t->count, 0); | |
567 | t->func = func; | |
568 | t->data = data; | |
569 | } | |
1da177e4 LT |
570 | EXPORT_SYMBOL(tasklet_init); |
571 | ||
572 | void tasklet_kill(struct tasklet_struct *t) | |
573 | { | |
574 | if (in_interrupt()) | |
40322764 | 575 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 LT |
576 | |
577 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 578 | do { |
1da177e4 | 579 | yield(); |
79d381c9 | 580 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
581 | } |
582 | tasklet_unlock_wait(t); | |
583 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
584 | } | |
1da177e4 LT |
585 | EXPORT_SYMBOL(tasklet_kill); |
586 | ||
9ba5f005 PZ |
587 | /* |
588 | * tasklet_hrtimer | |
589 | */ | |
590 | ||
591 | /* | |
b9c30322 PZ |
592 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
593 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
594 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
595 | */ |
596 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
597 | { | |
598 | struct tasklet_hrtimer *ttimer = | |
599 | container_of(timer, struct tasklet_hrtimer, timer); | |
600 | ||
b9c30322 PZ |
601 | tasklet_hi_schedule(&ttimer->tasklet); |
602 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
603 | } |
604 | ||
605 | /* | |
606 | * Helper function which calls the hrtimer callback from | |
607 | * tasklet/softirq context | |
608 | */ | |
609 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
610 | { | |
611 | struct tasklet_hrtimer *ttimer = (void *)data; | |
612 | enum hrtimer_restart restart; | |
613 | ||
614 | restart = ttimer->function(&ttimer->timer); | |
615 | if (restart != HRTIMER_NORESTART) | |
616 | hrtimer_restart(&ttimer->timer); | |
617 | } | |
618 | ||
619 | /** | |
620 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
621 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 622 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
623 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
624 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
625 | */ | |
626 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
627 | enum hrtimer_restart (*function)(struct hrtimer *), | |
628 | clockid_t which_clock, enum hrtimer_mode mode) | |
629 | { | |
630 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
631 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
632 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
633 | (unsigned long)ttimer); | |
634 | ttimer->function = function; | |
635 | } | |
636 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
637 | ||
1da177e4 LT |
638 | void __init softirq_init(void) |
639 | { | |
48f20a9a OJ |
640 | int cpu; |
641 | ||
642 | for_each_possible_cpu(cpu) { | |
643 | per_cpu(tasklet_vec, cpu).tail = | |
644 | &per_cpu(tasklet_vec, cpu).head; | |
645 | per_cpu(tasklet_hi_vec, cpu).tail = | |
646 | &per_cpu(tasklet_hi_vec, cpu).head; | |
647 | } | |
648 | ||
962cf36c CM |
649 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
650 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
651 | } |
652 | ||
3e339b5d | 653 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 654 | { |
3e339b5d TG |
655 | return local_softirq_pending(); |
656 | } | |
1da177e4 | 657 | |
3e339b5d TG |
658 | static void run_ksoftirqd(unsigned int cpu) |
659 | { | |
660 | local_irq_disable(); | |
661 | if (local_softirq_pending()) { | |
0bed698a FW |
662 | /* |
663 | * We can safely run softirq on inline stack, as we are not deep | |
664 | * in the task stack here. | |
665 | */ | |
3e339b5d | 666 | __do_softirq(); |
3e339b5d | 667 | local_irq_enable(); |
60479676 | 668 | cond_resched_rcu_qs(); |
3e339b5d | 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 | ||
c4544dbc | 707 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 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(); | |
c4544dbc | 730 | return 0; |
1da177e4 | 731 | } |
c4544dbc SAS |
732 | #else |
733 | #define takeover_tasklets NULL | |
1da177e4 LT |
734 | #endif /* CONFIG_HOTPLUG_CPU */ |
735 | ||
3e339b5d TG |
736 | static struct smp_hotplug_thread softirq_threads = { |
737 | .store = &ksoftirqd, | |
738 | .thread_should_run = ksoftirqd_should_run, | |
739 | .thread_fn = run_ksoftirqd, | |
740 | .thread_comm = "ksoftirqd/%u", | |
741 | }; | |
742 | ||
7babe8db | 743 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 744 | { |
c4544dbc SAS |
745 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
746 | takeover_tasklets); | |
3e339b5d TG |
747 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
748 | ||
1da177e4 LT |
749 | return 0; |
750 | } | |
7babe8db | 751 | early_initcall(spawn_ksoftirqd); |
78eef01b | 752 | |
43a25632 YL |
753 | /* |
754 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
755 | * GCC does not inline them incorrectly. ] | |
756 | */ | |
757 | ||
758 | int __init __weak early_irq_init(void) | |
759 | { | |
760 | return 0; | |
761 | } | |
762 | ||
4a046d17 YL |
763 | int __init __weak arch_probe_nr_irqs(void) |
764 | { | |
b683de2b | 765 | return NR_IRQS_LEGACY; |
4a046d17 YL |
766 | } |
767 | ||
43a25632 YL |
768 | int __init __weak arch_early_irq_init(void) |
769 | { | |
770 | return 0; | |
771 | } | |
62a08ae2 TG |
772 | |
773 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
774 | { | |
775 | return from; | |
776 | } |