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 | |
0f6f47ba FW |
52 | DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat); |
53 | EXPORT_PER_CPU_SYMBOL(irq_stat); | |
1da177e4 LT |
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 | |
1a63dcd8 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_); |
1a63dcd8 JFG |
147 | |
148 | __preempt_count_sub(cnt); | |
75e1056f VP |
149 | } |
150 | ||
de30a2b3 | 151 | /* |
c3442697 | 152 | * Special-case - softirqs can safely be enabled by __do_softirq(), |
de30a2b3 IM |
153 | * without processing still-pending softirqs: |
154 | */ | |
155 | void _local_bh_enable(void) | |
156 | { | |
5d60d3e7 | 157 | WARN_ON_ONCE(in_irq()); |
75e1056f | 158 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 159 | } |
de30a2b3 IM |
160 | EXPORT_SYMBOL(_local_bh_enable); |
161 | ||
0bd3a173 | 162 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 163 | { |
f71b74bc FW |
164 | WARN_ON_ONCE(in_irq()); |
165 | lockdep_assert_irqs_enabled(); | |
3c829c36 | 166 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 167 | local_irq_disable(); |
3c829c36 | 168 | #endif |
de30a2b3 IM |
169 | /* |
170 | * Are softirqs going to be turned on now: | |
171 | */ | |
75e1056f | 172 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 173 | trace_softirqs_on(ip); |
de30a2b3 IM |
174 | /* |
175 | * Keep preemption disabled until we are done with | |
176 | * softirq processing: | |
ce85b4f2 | 177 | */ |
0bd3a173 | 178 | preempt_count_sub(cnt - 1); |
de30a2b3 | 179 | |
0bed698a FW |
180 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
181 | /* | |
182 | * Run softirq if any pending. And do it in its own stack | |
183 | * as we may be calling this deep in a task call stack already. | |
184 | */ | |
de30a2b3 | 185 | do_softirq(); |
0bed698a | 186 | } |
de30a2b3 | 187 | |
bdb43806 | 188 | preempt_count_dec(); |
3c829c36 | 189 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 190 | local_irq_enable(); |
3c829c36 | 191 | #endif |
de30a2b3 IM |
192 | preempt_check_resched(); |
193 | } | |
0bd3a173 | 194 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 195 | |
1da177e4 | 196 | /* |
34376a50 BG |
197 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
198 | * but break the loop if need_resched() is set or after 2 ms. | |
199 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
200 | * certain cases, such as stop_machine(), jiffies may cease to | |
201 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
202 | * well to make sure we eventually return from this method. | |
1da177e4 | 203 | * |
c10d7367 | 204 | * These limits have been established via experimentation. |
1da177e4 LT |
205 | * The two things to balance is latency against fairness - |
206 | * we want to handle softirqs as soon as possible, but they | |
207 | * should not be able to lock up the box. | |
208 | */ | |
c10d7367 | 209 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 210 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 211 | |
f1a83e65 PZ |
212 | #ifdef CONFIG_TRACE_IRQFLAGS |
213 | /* | |
f1a83e65 PZ |
214 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
215 | * to keep the lockdep irq context tracking as tight as possible in order to | |
216 | * not miss-qualify lock contexts and miss possible deadlocks. | |
217 | */ | |
f1a83e65 | 218 | |
5c4853b6 | 219 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 220 | { |
5c4853b6 | 221 | bool in_hardirq = false; |
f1a83e65 | 222 | |
5c4853b6 FW |
223 | if (trace_hardirq_context(current)) { |
224 | in_hardirq = true; | |
f1a83e65 | 225 | trace_hardirq_exit(); |
5c4853b6 FW |
226 | } |
227 | ||
f1a83e65 | 228 | lockdep_softirq_enter(); |
5c4853b6 FW |
229 | |
230 | return in_hardirq; | |
f1a83e65 PZ |
231 | } |
232 | ||
5c4853b6 | 233 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
234 | { |
235 | lockdep_softirq_exit(); | |
5c4853b6 FW |
236 | |
237 | if (in_hardirq) | |
f1a83e65 | 238 | trace_hardirq_enter(); |
f1a83e65 | 239 | } |
f1a83e65 | 240 | #else |
5c4853b6 FW |
241 | static inline bool lockdep_softirq_start(void) { return false; } |
242 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
243 | #endif |
244 | ||
be7635e7 | 245 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 246 | { |
c10d7367 | 247 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 248 | unsigned long old_flags = current->flags; |
34376a50 | 249 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 250 | struct softirq_action *h; |
5c4853b6 | 251 | bool in_hardirq; |
f1a83e65 | 252 | __u32 pending; |
2e702b9f | 253 | int softirq_bit; |
907aed48 MG |
254 | |
255 | /* | |
256 | * Mask out PF_MEMALLOC s current task context is borrowed for the | |
257 | * softirq. A softirq handled such as network RX might set PF_MEMALLOC | |
258 | * again if the socket is related to swap | |
259 | */ | |
260 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
261 | |
262 | pending = local_softirq_pending(); | |
6a61671b | 263 | account_irq_enter_time(current); |
829035fd | 264 | |
0bd3a173 | 265 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 266 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 267 | |
1da177e4 LT |
268 | restart: |
269 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 270 | set_softirq_pending(0); |
1da177e4 | 271 | |
c70f5d66 | 272 | local_irq_enable(); |
1da177e4 LT |
273 | |
274 | h = softirq_vec; | |
275 | ||
2e702b9f JP |
276 | while ((softirq_bit = ffs(pending))) { |
277 | unsigned int vec_nr; | |
278 | int prev_count; | |
279 | ||
280 | h += softirq_bit - 1; | |
281 | ||
282 | vec_nr = h - softirq_vec; | |
283 | prev_count = preempt_count(); | |
284 | ||
285 | kstat_incr_softirqs_this_cpu(vec_nr); | |
286 | ||
287 | trace_softirq_entry(vec_nr); | |
288 | h->action(h); | |
289 | trace_softirq_exit(vec_nr); | |
290 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 291 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
292 | vec_nr, softirq_to_name[vec_nr], h->action, |
293 | prev_count, preempt_count()); | |
294 | preempt_count_set(prev_count); | |
1da177e4 LT |
295 | } |
296 | h++; | |
2e702b9f JP |
297 | pending >>= softirq_bit; |
298 | } | |
1da177e4 | 299 | |
284a8c93 | 300 | rcu_bh_qs(); |
c70f5d66 | 301 | local_irq_disable(); |
1da177e4 LT |
302 | |
303 | pending = local_softirq_pending(); | |
c10d7367 | 304 | if (pending) { |
34376a50 BG |
305 | if (time_before(jiffies, end) && !need_resched() && |
306 | --max_restart) | |
c10d7367 | 307 | goto restart; |
1da177e4 | 308 | |
1da177e4 | 309 | wakeup_softirqd(); |
c10d7367 | 310 | } |
1da177e4 | 311 | |
5c4853b6 | 312 | lockdep_softirq_end(in_hardirq); |
6a61671b | 313 | account_irq_exit_time(current); |
75e1056f | 314 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 315 | WARN_ON_ONCE(in_interrupt()); |
717a94b5 | 316 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
317 | } |
318 | ||
722a9f92 | 319 | asmlinkage __visible void do_softirq(void) |
1da177e4 LT |
320 | { |
321 | __u32 pending; | |
322 | unsigned long flags; | |
323 | ||
324 | if (in_interrupt()) | |
325 | return; | |
326 | ||
327 | local_irq_save(flags); | |
328 | ||
329 | pending = local_softirq_pending(); | |
330 | ||
4cd13c21 | 331 | if (pending && !ksoftirqd_running()) |
7d65f4a6 | 332 | do_softirq_own_stack(); |
1da177e4 LT |
333 | |
334 | local_irq_restore(flags); | |
335 | } | |
336 | ||
dde4b2b5 IM |
337 | /* |
338 | * Enter an interrupt context. | |
339 | */ | |
340 | void irq_enter(void) | |
341 | { | |
64db4cff | 342 | rcu_irq_enter(); |
0a8a2e78 | 343 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
344 | /* |
345 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
346 | * here, as softirq will be serviced on return from interrupt. | |
347 | */ | |
348 | local_bh_disable(); | |
5acac1be | 349 | tick_irq_enter(); |
d267f87f VP |
350 | _local_bh_enable(); |
351 | } | |
352 | ||
353 | __irq_enter(); | |
dde4b2b5 IM |
354 | } |
355 | ||
8d32a307 TG |
356 | static inline void invoke_softirq(void) |
357 | { | |
4cd13c21 ED |
358 | if (ksoftirqd_running()) |
359 | return; | |
360 | ||
ded79754 | 361 | if (!force_irqthreads) { |
cc1f0274 | 362 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
363 | /* |
364 | * We can safely execute softirq on the current stack if | |
365 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
366 | * at this stage. |
367 | */ | |
368 | __do_softirq(); | |
369 | #else | |
370 | /* | |
371 | * Otherwise, irq_exit() is called on the task stack that can | |
372 | * be potentially deep already. So call softirq in its own stack | |
373 | * to prevent from any overrun. | |
ded79754 | 374 | */ |
be6e1016 | 375 | do_softirq_own_stack(); |
cc1f0274 | 376 | #endif |
ded79754 | 377 | } else { |
8d32a307 | 378 | wakeup_softirqd(); |
ded79754 | 379 | } |
8d32a307 | 380 | } |
1da177e4 | 381 | |
67826eae FW |
382 | static inline void tick_irq_exit(void) |
383 | { | |
384 | #ifdef CONFIG_NO_HZ_COMMON | |
385 | int cpu = smp_processor_id(); | |
386 | ||
387 | /* Make sure that timer wheel updates are propagated */ | |
388 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
389 | if (!in_interrupt()) | |
390 | tick_nohz_irq_exit(); | |
391 | } | |
392 | #endif | |
393 | } | |
394 | ||
1da177e4 LT |
395 | /* |
396 | * Exit an interrupt context. Process softirqs if needed and possible: | |
397 | */ | |
398 | void irq_exit(void) | |
399 | { | |
74eed016 | 400 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 401 | local_irq_disable(); |
74eed016 | 402 | #else |
f71b74bc | 403 | lockdep_assert_irqs_disabled(); |
74eed016 | 404 | #endif |
6a61671b | 405 | account_irq_exit_time(current); |
bdb43806 | 406 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
407 | if (!in_interrupt() && local_softirq_pending()) |
408 | invoke_softirq(); | |
79bf2bb3 | 409 | |
67826eae | 410 | tick_irq_exit(); |
416eb33c | 411 | rcu_irq_exit(); |
f1a83e65 | 412 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
413 | } |
414 | ||
415 | /* | |
416 | * This function must run with irqs disabled! | |
417 | */ | |
7ad5b3a5 | 418 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
419 | { |
420 | __raise_softirq_irqoff(nr); | |
421 | ||
422 | /* | |
423 | * If we're in an interrupt or softirq, we're done | |
424 | * (this also catches softirq-disabled code). We will | |
425 | * actually run the softirq once we return from | |
426 | * the irq or softirq. | |
427 | * | |
428 | * Otherwise we wake up ksoftirqd to make sure we | |
429 | * schedule the softirq soon. | |
430 | */ | |
431 | if (!in_interrupt()) | |
432 | wakeup_softirqd(); | |
433 | } | |
434 | ||
7ad5b3a5 | 435 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
436 | { |
437 | unsigned long flags; | |
438 | ||
439 | local_irq_save(flags); | |
440 | raise_softirq_irqoff(nr); | |
441 | local_irq_restore(flags); | |
442 | } | |
443 | ||
f069686e SR |
444 | void __raise_softirq_irqoff(unsigned int nr) |
445 | { | |
446 | trace_softirq_raise(nr); | |
447 | or_softirq_pending(1UL << nr); | |
448 | } | |
449 | ||
962cf36c | 450 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 451 | { |
1da177e4 LT |
452 | softirq_vec[nr].action = action; |
453 | } | |
454 | ||
9ba5f005 PZ |
455 | /* |
456 | * Tasklets | |
457 | */ | |
ce85b4f2 | 458 | struct tasklet_head { |
48f20a9a OJ |
459 | struct tasklet_struct *head; |
460 | struct tasklet_struct **tail; | |
1da177e4 LT |
461 | }; |
462 | ||
4620b49f VN |
463 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
464 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 465 | |
6498ddad IM |
466 | static void __tasklet_schedule_common(struct tasklet_struct *t, |
467 | struct tasklet_head __percpu *headp, | |
468 | unsigned int softirq_nr) | |
1da177e4 | 469 | { |
6498ddad | 470 | struct tasklet_head *head; |
1da177e4 LT |
471 | unsigned long flags; |
472 | ||
473 | local_irq_save(flags); | |
6498ddad | 474 | head = this_cpu_ptr(headp); |
48f20a9a | 475 | t->next = NULL; |
6498ddad IM |
476 | *head->tail = t; |
477 | head->tail = &(t->next); | |
478 | raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
479 | local_irq_restore(flags); |
480 | } | |
6498ddad IM |
481 | |
482 | void __tasklet_schedule(struct tasklet_struct *t) | |
483 | { | |
484 | __tasklet_schedule_common(t, &tasklet_vec, | |
485 | TASKLET_SOFTIRQ); | |
486 | } | |
1da177e4 LT |
487 | EXPORT_SYMBOL(__tasklet_schedule); |
488 | ||
7ad5b3a5 | 489 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 | 490 | { |
6498ddad IM |
491 | __tasklet_schedule_common(t, &tasklet_hi_vec, |
492 | HI_SOFTIRQ); | |
1da177e4 | 493 | } |
1da177e4 LT |
494 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
495 | ||
82b691be IM |
496 | static void tasklet_action_common(struct softirq_action *a, |
497 | struct tasklet_head *tl_head, | |
498 | unsigned int softirq_nr) | |
1da177e4 LT |
499 | { |
500 | struct tasklet_struct *list; | |
501 | ||
502 | local_irq_disable(); | |
82b691be IM |
503 | list = tl_head->head; |
504 | tl_head->head = NULL; | |
505 | tl_head->tail = &tl_head->head; | |
1da177e4 LT |
506 | local_irq_enable(); |
507 | ||
508 | while (list) { | |
509 | struct tasklet_struct *t = list; | |
510 | ||
511 | list = list->next; | |
512 | ||
513 | if (tasklet_trylock(t)) { | |
514 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
515 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
516 | &t->state)) | |
1da177e4 LT |
517 | BUG(); |
518 | t->func(t->data); | |
519 | tasklet_unlock(t); | |
520 | continue; | |
521 | } | |
522 | tasklet_unlock(t); | |
523 | } | |
524 | ||
525 | local_irq_disable(); | |
48f20a9a | 526 | t->next = NULL; |
82b691be IM |
527 | *tl_head->tail = t; |
528 | tl_head->tail = &t->next; | |
529 | __raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
530 | local_irq_enable(); |
531 | } | |
532 | } | |
533 | ||
82b691be | 534 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 | 535 | { |
82b691be IM |
536 | tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ); |
537 | } | |
1da177e4 | 538 | |
82b691be IM |
539 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
540 | { | |
541 | tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ); | |
1da177e4 LT |
542 | } |
543 | ||
1da177e4 LT |
544 | void tasklet_init(struct tasklet_struct *t, |
545 | void (*func)(unsigned long), unsigned long data) | |
546 | { | |
547 | t->next = NULL; | |
548 | t->state = 0; | |
549 | atomic_set(&t->count, 0); | |
550 | t->func = func; | |
551 | t->data = data; | |
552 | } | |
1da177e4 LT |
553 | EXPORT_SYMBOL(tasklet_init); |
554 | ||
555 | void tasklet_kill(struct tasklet_struct *t) | |
556 | { | |
557 | if (in_interrupt()) | |
40322764 | 558 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 LT |
559 | |
560 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 561 | do { |
1da177e4 | 562 | yield(); |
79d381c9 | 563 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
564 | } |
565 | tasklet_unlock_wait(t); | |
566 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
567 | } | |
1da177e4 LT |
568 | EXPORT_SYMBOL(tasklet_kill); |
569 | ||
9ba5f005 PZ |
570 | /* |
571 | * tasklet_hrtimer | |
572 | */ | |
573 | ||
574 | /* | |
b9c30322 PZ |
575 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
576 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
577 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
578 | */ |
579 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
580 | { | |
581 | struct tasklet_hrtimer *ttimer = | |
582 | container_of(timer, struct tasklet_hrtimer, timer); | |
583 | ||
b9c30322 PZ |
584 | tasklet_hi_schedule(&ttimer->tasklet); |
585 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
586 | } |
587 | ||
588 | /* | |
589 | * Helper function which calls the hrtimer callback from | |
590 | * tasklet/softirq context | |
591 | */ | |
592 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
593 | { | |
594 | struct tasklet_hrtimer *ttimer = (void *)data; | |
595 | enum hrtimer_restart restart; | |
596 | ||
597 | restart = ttimer->function(&ttimer->timer); | |
598 | if (restart != HRTIMER_NORESTART) | |
599 | hrtimer_restart(&ttimer->timer); | |
600 | } | |
601 | ||
602 | /** | |
603 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
604 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 605 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
606 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
607 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
608 | */ | |
609 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
610 | enum hrtimer_restart (*function)(struct hrtimer *), | |
611 | clockid_t which_clock, enum hrtimer_mode mode) | |
612 | { | |
613 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
614 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
615 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
616 | (unsigned long)ttimer); | |
617 | ttimer->function = function; | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
620 | ||
1da177e4 LT |
621 | void __init softirq_init(void) |
622 | { | |
48f20a9a OJ |
623 | int cpu; |
624 | ||
625 | for_each_possible_cpu(cpu) { | |
626 | per_cpu(tasklet_vec, cpu).tail = | |
627 | &per_cpu(tasklet_vec, cpu).head; | |
628 | per_cpu(tasklet_hi_vec, cpu).tail = | |
629 | &per_cpu(tasklet_hi_vec, cpu).head; | |
630 | } | |
631 | ||
962cf36c CM |
632 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
633 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
634 | } |
635 | ||
3e339b5d | 636 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 637 | { |
3e339b5d TG |
638 | return local_softirq_pending(); |
639 | } | |
1da177e4 | 640 | |
3e339b5d TG |
641 | static void run_ksoftirqd(unsigned int cpu) |
642 | { | |
643 | local_irq_disable(); | |
644 | if (local_softirq_pending()) { | |
0bed698a FW |
645 | /* |
646 | * We can safely run softirq on inline stack, as we are not deep | |
647 | * in the task stack here. | |
648 | */ | |
3e339b5d | 649 | __do_softirq(); |
3e339b5d | 650 | local_irq_enable(); |
edf22f4c | 651 | cond_resched(); |
3e339b5d | 652 | return; |
1da177e4 | 653 | } |
3e339b5d | 654 | local_irq_enable(); |
1da177e4 LT |
655 | } |
656 | ||
657 | #ifdef CONFIG_HOTPLUG_CPU | |
658 | /* | |
659 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
660 | * scheduled for execution on @cpu. | |
661 | * | |
662 | * Unlike tasklet_kill, this function removes the tasklet | |
663 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
664 | * | |
665 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
666 | */ | |
667 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
668 | { | |
669 | struct tasklet_struct **i; | |
670 | ||
671 | BUG_ON(cpu_online(cpu)); | |
672 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
673 | ||
674 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
675 | return; | |
676 | ||
677 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 678 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
679 | if (*i == t) { |
680 | *i = t->next; | |
48f20a9a OJ |
681 | /* If this was the tail element, move the tail ptr */ |
682 | if (*i == NULL) | |
683 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
684 | return; |
685 | } | |
686 | } | |
687 | BUG(); | |
688 | } | |
689 | ||
c4544dbc | 690 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 691 | { |
1da177e4 LT |
692 | /* CPU is dead, so no lock needed. */ |
693 | local_irq_disable(); | |
694 | ||
695 | /* Find end, append list for that CPU. */ | |
e5e41723 | 696 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
697 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
698 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
699 | per_cpu(tasklet_vec, cpu).head = NULL; |
700 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
701 | } | |
1da177e4 LT |
702 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
703 | ||
e5e41723 | 704 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
705 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
706 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
707 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
708 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
709 | } | |
1da177e4 LT |
710 | raise_softirq_irqoff(HI_SOFTIRQ); |
711 | ||
712 | local_irq_enable(); | |
c4544dbc | 713 | return 0; |
1da177e4 | 714 | } |
c4544dbc SAS |
715 | #else |
716 | #define takeover_tasklets NULL | |
1da177e4 LT |
717 | #endif /* CONFIG_HOTPLUG_CPU */ |
718 | ||
3e339b5d TG |
719 | static struct smp_hotplug_thread softirq_threads = { |
720 | .store = &ksoftirqd, | |
721 | .thread_should_run = ksoftirqd_should_run, | |
722 | .thread_fn = run_ksoftirqd, | |
723 | .thread_comm = "ksoftirqd/%u", | |
724 | }; | |
725 | ||
7babe8db | 726 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 727 | { |
c4544dbc SAS |
728 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
729 | takeover_tasklets); | |
3e339b5d TG |
730 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
731 | ||
1da177e4 LT |
732 | return 0; |
733 | } | |
7babe8db | 734 | early_initcall(spawn_ksoftirqd); |
78eef01b | 735 | |
43a25632 YL |
736 | /* |
737 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
738 | * GCC does not inline them incorrectly. ] | |
739 | */ | |
740 | ||
741 | int __init __weak early_irq_init(void) | |
742 | { | |
743 | return 0; | |
744 | } | |
745 | ||
4a046d17 YL |
746 | int __init __weak arch_probe_nr_irqs(void) |
747 | { | |
b683de2b | 748 | return NR_IRQS_LEGACY; |
4a046d17 YL |
749 | } |
750 | ||
43a25632 YL |
751 | int __init __weak arch_early_irq_init(void) |
752 | { | |
753 | return 0; | |
754 | } | |
62a08ae2 TG |
755 | |
756 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
757 | { | |
758 | return from; | |
759 | } |