]>
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 | |
74c7f1cd JFG |
142 | if (preempt_count() == cnt) |
143 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
144 | ||
9ea4c380 | 145 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
d2e08473 | 146 | trace_softirqs_on(_RET_IP_); |
74c7f1cd JFG |
147 | |
148 | __preempt_count_sub(cnt); | |
75e1056f VP |
149 | } |
150 | ||
de30a2b3 IM |
151 | /* |
152 | * Special-case - softirqs can safely be enabled in | |
153 | * cond_resched_softirq(), or by __do_softirq(), | |
154 | * without processing still-pending softirqs: | |
155 | */ | |
156 | void _local_bh_enable(void) | |
157 | { | |
5d60d3e7 | 158 | WARN_ON_ONCE(in_irq()); |
75e1056f | 159 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 160 | } |
de30a2b3 IM |
161 | EXPORT_SYMBOL(_local_bh_enable); |
162 | ||
0bd3a173 | 163 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 164 | { |
f71b74bc FW |
165 | WARN_ON_ONCE(in_irq()); |
166 | lockdep_assert_irqs_enabled(); | |
3c829c36 | 167 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 168 | local_irq_disable(); |
3c829c36 | 169 | #endif |
de30a2b3 IM |
170 | /* |
171 | * Are softirqs going to be turned on now: | |
172 | */ | |
75e1056f | 173 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 174 | trace_softirqs_on(ip); |
de30a2b3 IM |
175 | /* |
176 | * Keep preemption disabled until we are done with | |
177 | * softirq processing: | |
ce85b4f2 | 178 | */ |
0bd3a173 | 179 | preempt_count_sub(cnt - 1); |
de30a2b3 | 180 | |
0bed698a FW |
181 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
182 | /* | |
183 | * Run softirq if any pending. And do it in its own stack | |
184 | * as we may be calling this deep in a task call stack already. | |
185 | */ | |
de30a2b3 | 186 | do_softirq(); |
0bed698a | 187 | } |
de30a2b3 | 188 | |
bdb43806 | 189 | preempt_count_dec(); |
3c829c36 | 190 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 191 | local_irq_enable(); |
3c829c36 | 192 | #endif |
de30a2b3 IM |
193 | preempt_check_resched(); |
194 | } | |
0bd3a173 | 195 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 196 | |
1da177e4 | 197 | /* |
34376a50 BG |
198 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
199 | * but break the loop if need_resched() is set or after 2 ms. | |
200 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
201 | * certain cases, such as stop_machine(), jiffies may cease to | |
202 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
203 | * well to make sure we eventually return from this method. | |
1da177e4 | 204 | * |
c10d7367 | 205 | * These limits have been established via experimentation. |
1da177e4 LT |
206 | * The two things to balance is latency against fairness - |
207 | * we want to handle softirqs as soon as possible, but they | |
208 | * should not be able to lock up the box. | |
209 | */ | |
c10d7367 | 210 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 211 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 212 | |
f1a83e65 PZ |
213 | #ifdef CONFIG_TRACE_IRQFLAGS |
214 | /* | |
f1a83e65 PZ |
215 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
216 | * to keep the lockdep irq context tracking as tight as possible in order to | |
217 | * not miss-qualify lock contexts and miss possible deadlocks. | |
218 | */ | |
f1a83e65 | 219 | |
5c4853b6 | 220 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 221 | { |
5c4853b6 | 222 | bool in_hardirq = false; |
f1a83e65 | 223 | |
5c4853b6 FW |
224 | if (trace_hardirq_context(current)) { |
225 | in_hardirq = true; | |
f1a83e65 | 226 | trace_hardirq_exit(); |
5c4853b6 FW |
227 | } |
228 | ||
f1a83e65 | 229 | lockdep_softirq_enter(); |
5c4853b6 FW |
230 | |
231 | return in_hardirq; | |
f1a83e65 PZ |
232 | } |
233 | ||
5c4853b6 | 234 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
235 | { |
236 | lockdep_softirq_exit(); | |
5c4853b6 FW |
237 | |
238 | if (in_hardirq) | |
f1a83e65 | 239 | trace_hardirq_enter(); |
f1a83e65 | 240 | } |
f1a83e65 | 241 | #else |
5c4853b6 FW |
242 | static inline bool lockdep_softirq_start(void) { return false; } |
243 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
244 | #endif |
245 | ||
be7635e7 | 246 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 247 | { |
c10d7367 | 248 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 249 | unsigned long old_flags = current->flags; |
34376a50 | 250 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 251 | struct softirq_action *h; |
5c4853b6 | 252 | bool in_hardirq; |
f1a83e65 | 253 | __u32 pending; |
2e702b9f | 254 | int softirq_bit; |
907aed48 MG |
255 | |
256 | /* | |
257 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
258 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
259 | * again if the socket is related to swap | |
260 | */ | |
261 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
262 | |
263 | pending = local_softirq_pending(); | |
6a61671b | 264 | account_irq_enter_time(current); |
829035fd | 265 | |
0bd3a173 | 266 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 267 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 268 | |
1da177e4 LT |
269 | restart: |
270 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 271 | set_softirq_pending(0); |
1da177e4 | 272 | |
c70f5d66 | 273 | local_irq_enable(); |
1da177e4 LT |
274 | |
275 | h = softirq_vec; | |
276 | ||
2e702b9f JP |
277 | while ((softirq_bit = ffs(pending))) { |
278 | unsigned int vec_nr; | |
279 | int prev_count; | |
280 | ||
281 | h += softirq_bit - 1; | |
282 | ||
283 | vec_nr = h - softirq_vec; | |
284 | prev_count = preempt_count(); | |
285 | ||
286 | kstat_incr_softirqs_this_cpu(vec_nr); | |
287 | ||
288 | trace_softirq_entry(vec_nr); | |
289 | h->action(h); | |
290 | trace_softirq_exit(vec_nr); | |
291 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 292 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
293 | vec_nr, softirq_to_name[vec_nr], h->action, |
294 | prev_count, preempt_count()); | |
295 | preempt_count_set(prev_count); | |
1da177e4 LT |
296 | } |
297 | h++; | |
2e702b9f JP |
298 | pending >>= softirq_bit; |
299 | } | |
1da177e4 | 300 | |
284a8c93 | 301 | rcu_bh_qs(); |
c70f5d66 | 302 | local_irq_disable(); |
1da177e4 LT |
303 | |
304 | pending = local_softirq_pending(); | |
c10d7367 | 305 | if (pending) { |
34376a50 BG |
306 | if (time_before(jiffies, end) && !need_resched() && |
307 | --max_restart) | |
c10d7367 | 308 | goto restart; |
1da177e4 | 309 | |
1da177e4 | 310 | wakeup_softirqd(); |
c10d7367 | 311 | } |
1da177e4 | 312 | |
5c4853b6 | 313 | lockdep_softirq_end(in_hardirq); |
6a61671b | 314 | account_irq_exit_time(current); |
75e1056f | 315 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 316 | WARN_ON_ONCE(in_interrupt()); |
717a94b5 | 317 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
318 | } |
319 | ||
722a9f92 | 320 | asmlinkage __visible void do_softirq(void) |
1da177e4 LT |
321 | { |
322 | __u32 pending; | |
323 | unsigned long flags; | |
324 | ||
325 | if (in_interrupt()) | |
326 | return; | |
327 | ||
328 | local_irq_save(flags); | |
329 | ||
330 | pending = local_softirq_pending(); | |
331 | ||
4cd13c21 | 332 | if (pending && !ksoftirqd_running()) |
7d65f4a6 | 333 | do_softirq_own_stack(); |
1da177e4 LT |
334 | |
335 | local_irq_restore(flags); | |
336 | } | |
337 | ||
dde4b2b5 IM |
338 | /* |
339 | * Enter an interrupt context. | |
340 | */ | |
341 | void irq_enter(void) | |
342 | { | |
64db4cff | 343 | rcu_irq_enter(); |
0a8a2e78 | 344 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
345 | /* |
346 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
347 | * here, as softirq will be serviced on return from interrupt. | |
348 | */ | |
349 | local_bh_disable(); | |
5acac1be | 350 | tick_irq_enter(); |
d267f87f VP |
351 | _local_bh_enable(); |
352 | } | |
353 | ||
354 | __irq_enter(); | |
dde4b2b5 IM |
355 | } |
356 | ||
8d32a307 TG |
357 | static inline void invoke_softirq(void) |
358 | { | |
4cd13c21 ED |
359 | if (ksoftirqd_running()) |
360 | return; | |
361 | ||
ded79754 | 362 | if (!force_irqthreads) { |
cc1f0274 | 363 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
364 | /* |
365 | * We can safely execute softirq on the current stack if | |
366 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
367 | * at this stage. |
368 | */ | |
369 | __do_softirq(); | |
370 | #else | |
371 | /* | |
372 | * Otherwise, irq_exit() is called on the task stack that can | |
373 | * be potentially deep already. So call softirq in its own stack | |
374 | * to prevent from any overrun. | |
ded79754 | 375 | */ |
be6e1016 | 376 | do_softirq_own_stack(); |
cc1f0274 | 377 | #endif |
ded79754 | 378 | } else { |
8d32a307 | 379 | wakeup_softirqd(); |
ded79754 | 380 | } |
8d32a307 | 381 | } |
1da177e4 | 382 | |
67826eae FW |
383 | static inline void tick_irq_exit(void) |
384 | { | |
385 | #ifdef CONFIG_NO_HZ_COMMON | |
386 | int cpu = smp_processor_id(); | |
387 | ||
388 | /* Make sure that timer wheel updates are propagated */ | |
389 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
390 | if (!in_interrupt()) | |
391 | tick_nohz_irq_exit(); | |
392 | } | |
393 | #endif | |
394 | } | |
395 | ||
1da177e4 LT |
396 | /* |
397 | * Exit an interrupt context. Process softirqs if needed and possible: | |
398 | */ | |
399 | void irq_exit(void) | |
400 | { | |
74eed016 | 401 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 402 | local_irq_disable(); |
74eed016 | 403 | #else |
f71b74bc | 404 | lockdep_assert_irqs_disabled(); |
74eed016 | 405 | #endif |
6a61671b | 406 | account_irq_exit_time(current); |
bdb43806 | 407 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
408 | if (!in_interrupt() && local_softirq_pending()) |
409 | invoke_softirq(); | |
79bf2bb3 | 410 | |
67826eae | 411 | tick_irq_exit(); |
416eb33c | 412 | rcu_irq_exit(); |
f1a83e65 | 413 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
414 | } |
415 | ||
416 | /* | |
417 | * This function must run with irqs disabled! | |
418 | */ | |
7ad5b3a5 | 419 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
420 | { |
421 | __raise_softirq_irqoff(nr); | |
422 | ||
423 | /* | |
424 | * If we're in an interrupt or softirq, we're done | |
425 | * (this also catches softirq-disabled code). We will | |
426 | * actually run the softirq once we return from | |
427 | * the irq or softirq. | |
428 | * | |
429 | * Otherwise we wake up ksoftirqd to make sure we | |
430 | * schedule the softirq soon. | |
431 | */ | |
432 | if (!in_interrupt()) | |
433 | wakeup_softirqd(); | |
434 | } | |
435 | ||
7ad5b3a5 | 436 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
437 | { |
438 | unsigned long flags; | |
439 | ||
440 | local_irq_save(flags); | |
441 | raise_softirq_irqoff(nr); | |
442 | local_irq_restore(flags); | |
443 | } | |
444 | ||
f069686e SR |
445 | void __raise_softirq_irqoff(unsigned int nr) |
446 | { | |
447 | trace_softirq_raise(nr); | |
448 | or_softirq_pending(1UL << nr); | |
449 | } | |
450 | ||
962cf36c | 451 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 452 | { |
1da177e4 LT |
453 | softirq_vec[nr].action = action; |
454 | } | |
455 | ||
9ba5f005 PZ |
456 | /* |
457 | * Tasklets | |
458 | */ | |
ce85b4f2 | 459 | struct tasklet_head { |
48f20a9a OJ |
460 | struct tasklet_struct *head; |
461 | struct tasklet_struct **tail; | |
1da177e4 LT |
462 | }; |
463 | ||
4620b49f VN |
464 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
465 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 466 | |
7ad5b3a5 | 467 | void __tasklet_schedule(struct tasklet_struct *t) |
1da177e4 LT |
468 | { |
469 | unsigned long flags; | |
470 | ||
471 | local_irq_save(flags); | |
48f20a9a | 472 | t->next = NULL; |
909ea964 CL |
473 | *__this_cpu_read(tasklet_vec.tail) = t; |
474 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
475 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
476 | local_irq_restore(flags); | |
477 | } | |
1da177e4 LT |
478 | EXPORT_SYMBOL(__tasklet_schedule); |
479 | ||
7ad5b3a5 | 480 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 LT |
481 | { |
482 | unsigned long flags; | |
483 | ||
484 | local_irq_save(flags); | |
48f20a9a | 485 | t->next = NULL; |
909ea964 CL |
486 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
487 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
488 | raise_softirq_irqoff(HI_SOFTIRQ); |
489 | local_irq_restore(flags); | |
490 | } | |
1da177e4 LT |
491 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
492 | ||
0766f788 | 493 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 LT |
494 | { |
495 | struct tasklet_struct *list; | |
496 | ||
497 | local_irq_disable(); | |
909ea964 CL |
498 | list = __this_cpu_read(tasklet_vec.head); |
499 | __this_cpu_write(tasklet_vec.head, NULL); | |
22127e93 | 500 | __this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head)); |
1da177e4 LT |
501 | local_irq_enable(); |
502 | ||
503 | while (list) { | |
504 | struct tasklet_struct *t = list; | |
505 | ||
506 | list = list->next; | |
507 | ||
508 | if (tasklet_trylock(t)) { | |
509 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
510 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
511 | &t->state)) | |
1da177e4 LT |
512 | BUG(); |
513 | t->func(t->data); | |
514 | tasklet_unlock(t); | |
515 | continue; | |
516 | } | |
517 | tasklet_unlock(t); | |
518 | } | |
519 | ||
520 | local_irq_disable(); | |
48f20a9a | 521 | t->next = NULL; |
909ea964 CL |
522 | *__this_cpu_read(tasklet_vec.tail) = t; |
523 | __this_cpu_write(tasklet_vec.tail, &(t->next)); | |
1da177e4 LT |
524 | __raise_softirq_irqoff(TASKLET_SOFTIRQ); |
525 | local_irq_enable(); | |
526 | } | |
527 | } | |
528 | ||
0766f788 | 529 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
1da177e4 LT |
530 | { |
531 | struct tasklet_struct *list; | |
532 | ||
533 | local_irq_disable(); | |
909ea964 CL |
534 | list = __this_cpu_read(tasklet_hi_vec.head); |
535 | __this_cpu_write(tasklet_hi_vec.head, NULL); | |
22127e93 | 536 | __this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head)); |
1da177e4 LT |
537 | local_irq_enable(); |
538 | ||
539 | while (list) { | |
540 | struct tasklet_struct *t = list; | |
541 | ||
542 | list = list->next; | |
543 | ||
544 | if (tasklet_trylock(t)) { | |
545 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
546 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
547 | &t->state)) | |
1da177e4 LT |
548 | BUG(); |
549 | t->func(t->data); | |
550 | tasklet_unlock(t); | |
551 | continue; | |
552 | } | |
553 | tasklet_unlock(t); | |
554 | } | |
555 | ||
556 | local_irq_disable(); | |
48f20a9a | 557 | t->next = NULL; |
909ea964 CL |
558 | *__this_cpu_read(tasklet_hi_vec.tail) = t; |
559 | __this_cpu_write(tasklet_hi_vec.tail, &(t->next)); | |
1da177e4 LT |
560 | __raise_softirq_irqoff(HI_SOFTIRQ); |
561 | local_irq_enable(); | |
562 | } | |
563 | } | |
564 | ||
1da177e4 LT |
565 | void tasklet_init(struct tasklet_struct *t, |
566 | void (*func)(unsigned long), unsigned long data) | |
567 | { | |
568 | t->next = NULL; | |
569 | t->state = 0; | |
570 | atomic_set(&t->count, 0); | |
571 | t->func = func; | |
572 | t->data = data; | |
573 | } | |
1da177e4 LT |
574 | EXPORT_SYMBOL(tasklet_init); |
575 | ||
576 | void tasklet_kill(struct tasklet_struct *t) | |
577 | { | |
578 | if (in_interrupt()) | |
40322764 | 579 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 LT |
580 | |
581 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 582 | do { |
1da177e4 | 583 | yield(); |
79d381c9 | 584 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
585 | } |
586 | tasklet_unlock_wait(t); | |
587 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
588 | } | |
1da177e4 LT |
589 | EXPORT_SYMBOL(tasklet_kill); |
590 | ||
9ba5f005 PZ |
591 | /* |
592 | * tasklet_hrtimer | |
593 | */ | |
594 | ||
595 | /* | |
b9c30322 PZ |
596 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
597 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
598 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
599 | */ |
600 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
601 | { | |
602 | struct tasklet_hrtimer *ttimer = | |
603 | container_of(timer, struct tasklet_hrtimer, timer); | |
604 | ||
b9c30322 PZ |
605 | tasklet_hi_schedule(&ttimer->tasklet); |
606 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
607 | } |
608 | ||
609 | /* | |
610 | * Helper function which calls the hrtimer callback from | |
611 | * tasklet/softirq context | |
612 | */ | |
613 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
614 | { | |
615 | struct tasklet_hrtimer *ttimer = (void *)data; | |
616 | enum hrtimer_restart restart; | |
617 | ||
618 | restart = ttimer->function(&ttimer->timer); | |
619 | if (restart != HRTIMER_NORESTART) | |
620 | hrtimer_restart(&ttimer->timer); | |
621 | } | |
622 | ||
623 | /** | |
624 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
625 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 626 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
627 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
628 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
629 | */ | |
630 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
631 | enum hrtimer_restart (*function)(struct hrtimer *), | |
632 | clockid_t which_clock, enum hrtimer_mode mode) | |
633 | { | |
634 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
635 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
636 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
637 | (unsigned long)ttimer); | |
638 | ttimer->function = function; | |
639 | } | |
640 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
641 | ||
1da177e4 LT |
642 | void __init softirq_init(void) |
643 | { | |
48f20a9a OJ |
644 | int cpu; |
645 | ||
646 | for_each_possible_cpu(cpu) { | |
647 | per_cpu(tasklet_vec, cpu).tail = | |
648 | &per_cpu(tasklet_vec, cpu).head; | |
649 | per_cpu(tasklet_hi_vec, cpu).tail = | |
650 | &per_cpu(tasklet_hi_vec, cpu).head; | |
651 | } | |
652 | ||
962cf36c CM |
653 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
654 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
655 | } |
656 | ||
3e339b5d | 657 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 658 | { |
3e339b5d TG |
659 | return local_softirq_pending(); |
660 | } | |
1da177e4 | 661 | |
3e339b5d TG |
662 | static void run_ksoftirqd(unsigned int cpu) |
663 | { | |
664 | local_irq_disable(); | |
665 | if (local_softirq_pending()) { | |
0bed698a FW |
666 | /* |
667 | * We can safely run softirq on inline stack, as we are not deep | |
668 | * in the task stack here. | |
669 | */ | |
3e339b5d | 670 | __do_softirq(); |
3e339b5d | 671 | local_irq_enable(); |
60479676 | 672 | cond_resched_rcu_qs(); |
3e339b5d | 673 | return; |
1da177e4 | 674 | } |
3e339b5d | 675 | local_irq_enable(); |
1da177e4 LT |
676 | } |
677 | ||
678 | #ifdef CONFIG_HOTPLUG_CPU | |
679 | /* | |
680 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
681 | * scheduled for execution on @cpu. | |
682 | * | |
683 | * Unlike tasklet_kill, this function removes the tasklet | |
684 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
685 | * | |
686 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
687 | */ | |
688 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
689 | { | |
690 | struct tasklet_struct **i; | |
691 | ||
692 | BUG_ON(cpu_online(cpu)); | |
693 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
694 | ||
695 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
696 | return; | |
697 | ||
698 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 699 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
700 | if (*i == t) { |
701 | *i = t->next; | |
48f20a9a OJ |
702 | /* If this was the tail element, move the tail ptr */ |
703 | if (*i == NULL) | |
704 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
705 | return; |
706 | } | |
707 | } | |
708 | BUG(); | |
709 | } | |
710 | ||
c4544dbc | 711 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 712 | { |
1da177e4 LT |
713 | /* CPU is dead, so no lock needed. */ |
714 | local_irq_disable(); | |
715 | ||
716 | /* Find end, append list for that CPU. */ | |
e5e41723 | 717 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
718 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
719 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
720 | per_cpu(tasklet_vec, cpu).head = NULL; |
721 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
722 | } | |
1da177e4 LT |
723 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
724 | ||
e5e41723 | 725 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
726 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
727 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
728 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
729 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
730 | } | |
1da177e4 LT |
731 | raise_softirq_irqoff(HI_SOFTIRQ); |
732 | ||
733 | local_irq_enable(); | |
c4544dbc | 734 | return 0; |
1da177e4 | 735 | } |
c4544dbc SAS |
736 | #else |
737 | #define takeover_tasklets NULL | |
1da177e4 LT |
738 | #endif /* CONFIG_HOTPLUG_CPU */ |
739 | ||
3e339b5d TG |
740 | static struct smp_hotplug_thread softirq_threads = { |
741 | .store = &ksoftirqd, | |
742 | .thread_should_run = ksoftirqd_should_run, | |
743 | .thread_fn = run_ksoftirqd, | |
744 | .thread_comm = "ksoftirqd/%u", | |
745 | }; | |
746 | ||
7babe8db | 747 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 748 | { |
c4544dbc SAS |
749 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
750 | takeover_tasklets); | |
3e339b5d TG |
751 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
752 | ||
1da177e4 LT |
753 | return 0; |
754 | } | |
7babe8db | 755 | early_initcall(spawn_ksoftirqd); |
78eef01b | 756 | |
43a25632 YL |
757 | /* |
758 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
759 | * GCC does not inline them incorrectly. ] | |
760 | */ | |
761 | ||
762 | int __init __weak early_irq_init(void) | |
763 | { | |
764 | return 0; | |
765 | } | |
766 | ||
4a046d17 YL |
767 | int __init __weak arch_probe_nr_irqs(void) |
768 | { | |
b683de2b | 769 | return NR_IRQS_LEGACY; |
4a046d17 YL |
770 | } |
771 | ||
43a25632 YL |
772 | int __init __weak arch_early_irq_init(void) |
773 | { | |
774 | return 0; | |
775 | } | |
62a08ae2 TG |
776 | |
777 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
778 | { | |
779 | return from; | |
780 | } |