]>
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 | |
3c53776e LT |
82 | * right now. Let ksoftirqd handle this at its own rate, to get fairness, |
83 | * unless we're doing some of the synchronous softirqs. | |
4cd13c21 | 84 | */ |
3c53776e LT |
85 | #define SOFTIRQ_NOW_MASK ((1 << HI_SOFTIRQ) | (1 << TASKLET_SOFTIRQ)) |
86 | static bool ksoftirqd_running(unsigned long pending) | |
4cd13c21 ED |
87 | { |
88 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); | |
89 | ||
3c53776e LT |
90 | if (pending & SOFTIRQ_NOW_MASK) |
91 | return false; | |
1342d808 MK |
92 | return tsk && (tsk->state == TASK_RUNNING) && |
93 | !__kthread_should_park(tsk); | |
4cd13c21 ED |
94 | } |
95 | ||
75e1056f VP |
96 | /* |
97 | * preempt_count and SOFTIRQ_OFFSET usage: | |
98 | * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving | |
99 | * softirq processing. | |
100 | * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
101 | * on local_bh_disable or local_bh_enable. | |
102 | * This lets us distinguish between whether we are currently processing | |
103 | * softirq and whether we just have bh disabled. | |
104 | */ | |
105 | ||
de30a2b3 IM |
106 | /* |
107 | * This one is for softirq.c-internal use, | |
108 | * where hardirqs are disabled legitimately: | |
109 | */ | |
3c829c36 | 110 | #ifdef CONFIG_TRACE_IRQFLAGS |
0bd3a173 | 111 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
112 | { |
113 | unsigned long flags; | |
114 | ||
115 | WARN_ON_ONCE(in_irq()); | |
116 | ||
117 | raw_local_irq_save(flags); | |
7e49fcce | 118 | /* |
bdb43806 | 119 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
120 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
121 | * is set and before current->softirq_enabled is cleared. | |
122 | * We must manually increment preempt_count here and manually | |
123 | * call the trace_preempt_off later. | |
124 | */ | |
bdb43806 | 125 | __preempt_count_add(cnt); |
de30a2b3 IM |
126 | /* |
127 | * Were softirqs turned off above: | |
128 | */ | |
9ea4c380 | 129 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
de30a2b3 IM |
130 | trace_softirqs_off(ip); |
131 | raw_local_irq_restore(flags); | |
7e49fcce | 132 | |
0f1ba9a2 HC |
133 | if (preempt_count() == cnt) { |
134 | #ifdef CONFIG_DEBUG_PREEMPT | |
f904f582 | 135 | current->preempt_disable_ip = get_lock_parent_ip(); |
0f1ba9a2 | 136 | #endif |
f904f582 | 137 | trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip()); |
0f1ba9a2 | 138 | } |
de30a2b3 | 139 | } |
0bd3a173 | 140 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 141 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 142 | |
75e1056f VP |
143 | static void __local_bh_enable(unsigned int cnt) |
144 | { | |
f71b74bc | 145 | lockdep_assert_irqs_disabled(); |
75e1056f | 146 | |
1a63dcd8 JFG |
147 | if (preempt_count() == cnt) |
148 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
149 | ||
9ea4c380 | 150 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
d2e08473 | 151 | trace_softirqs_on(_RET_IP_); |
1a63dcd8 JFG |
152 | |
153 | __preempt_count_sub(cnt); | |
75e1056f VP |
154 | } |
155 | ||
de30a2b3 | 156 | /* |
c3442697 | 157 | * Special-case - softirqs can safely be enabled by __do_softirq(), |
de30a2b3 IM |
158 | * without processing still-pending softirqs: |
159 | */ | |
160 | void _local_bh_enable(void) | |
161 | { | |
5d60d3e7 | 162 | WARN_ON_ONCE(in_irq()); |
75e1056f | 163 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 164 | } |
de30a2b3 IM |
165 | EXPORT_SYMBOL(_local_bh_enable); |
166 | ||
0bd3a173 | 167 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 168 | { |
f71b74bc FW |
169 | WARN_ON_ONCE(in_irq()); |
170 | lockdep_assert_irqs_enabled(); | |
3c829c36 | 171 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 172 | local_irq_disable(); |
3c829c36 | 173 | #endif |
de30a2b3 IM |
174 | /* |
175 | * Are softirqs going to be turned on now: | |
176 | */ | |
75e1056f | 177 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0f476b6d | 178 | trace_softirqs_on(ip); |
de30a2b3 IM |
179 | /* |
180 | * Keep preemption disabled until we are done with | |
181 | * softirq processing: | |
ce85b4f2 | 182 | */ |
0bd3a173 | 183 | preempt_count_sub(cnt - 1); |
de30a2b3 | 184 | |
0bed698a FW |
185 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
186 | /* | |
187 | * Run softirq if any pending. And do it in its own stack | |
188 | * as we may be calling this deep in a task call stack already. | |
189 | */ | |
de30a2b3 | 190 | do_softirq(); |
0bed698a | 191 | } |
de30a2b3 | 192 | |
bdb43806 | 193 | preempt_count_dec(); |
3c829c36 | 194 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 195 | local_irq_enable(); |
3c829c36 | 196 | #endif |
de30a2b3 IM |
197 | preempt_check_resched(); |
198 | } | |
0bd3a173 | 199 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 200 | |
1da177e4 | 201 | /* |
34376a50 BG |
202 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
203 | * but break the loop if need_resched() is set or after 2 ms. | |
204 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
205 | * certain cases, such as stop_machine(), jiffies may cease to | |
206 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
207 | * well to make sure we eventually return from this method. | |
1da177e4 | 208 | * |
c10d7367 | 209 | * These limits have been established via experimentation. |
1da177e4 LT |
210 | * The two things to balance is latency against fairness - |
211 | * we want to handle softirqs as soon as possible, but they | |
212 | * should not be able to lock up the box. | |
213 | */ | |
c10d7367 | 214 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 215 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 216 | |
f1a83e65 PZ |
217 | #ifdef CONFIG_TRACE_IRQFLAGS |
218 | /* | |
f1a83e65 PZ |
219 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
220 | * to keep the lockdep irq context tracking as tight as possible in order to | |
221 | * not miss-qualify lock contexts and miss possible deadlocks. | |
222 | */ | |
f1a83e65 | 223 | |
5c4853b6 | 224 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 225 | { |
5c4853b6 | 226 | bool in_hardirq = false; |
f1a83e65 | 227 | |
5c4853b6 FW |
228 | if (trace_hardirq_context(current)) { |
229 | in_hardirq = true; | |
f1a83e65 | 230 | trace_hardirq_exit(); |
5c4853b6 FW |
231 | } |
232 | ||
f1a83e65 | 233 | lockdep_softirq_enter(); |
5c4853b6 FW |
234 | |
235 | return in_hardirq; | |
f1a83e65 PZ |
236 | } |
237 | ||
5c4853b6 | 238 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
239 | { |
240 | lockdep_softirq_exit(); | |
5c4853b6 FW |
241 | |
242 | if (in_hardirq) | |
f1a83e65 | 243 | trace_hardirq_enter(); |
f1a83e65 | 244 | } |
f1a83e65 | 245 | #else |
5c4853b6 FW |
246 | static inline bool lockdep_softirq_start(void) { return false; } |
247 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
248 | #endif |
249 | ||
be7635e7 | 250 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 251 | { |
c10d7367 | 252 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 253 | unsigned long old_flags = current->flags; |
34376a50 | 254 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 255 | struct softirq_action *h; |
5c4853b6 | 256 | bool in_hardirq; |
f1a83e65 | 257 | __u32 pending; |
2e702b9f | 258 | int softirq_bit; |
907aed48 MG |
259 | |
260 | /* | |
e45506ac YL |
261 | * Mask out PF_MEMALLOC as the current task context is borrowed for the |
262 | * softirq. A softirq handled, such as network RX, might set PF_MEMALLOC | |
263 | * again if the socket is related to swapping. | |
907aed48 MG |
264 | */ |
265 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
266 | |
267 | pending = local_softirq_pending(); | |
6a61671b | 268 | account_irq_enter_time(current); |
829035fd | 269 | |
0bd3a173 | 270 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); |
5c4853b6 | 271 | in_hardirq = lockdep_softirq_start(); |
1da177e4 | 272 | |
1da177e4 LT |
273 | restart: |
274 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 275 | set_softirq_pending(0); |
1da177e4 | 276 | |
c70f5d66 | 277 | local_irq_enable(); |
1da177e4 LT |
278 | |
279 | h = softirq_vec; | |
280 | ||
2e702b9f JP |
281 | while ((softirq_bit = ffs(pending))) { |
282 | unsigned int vec_nr; | |
283 | int prev_count; | |
284 | ||
285 | h += softirq_bit - 1; | |
286 | ||
287 | vec_nr = h - softirq_vec; | |
288 | prev_count = preempt_count(); | |
289 | ||
290 | kstat_incr_softirqs_this_cpu(vec_nr); | |
291 | ||
292 | trace_softirq_entry(vec_nr); | |
293 | h->action(h); | |
294 | trace_softirq_exit(vec_nr); | |
295 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 296 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
297 | vec_nr, softirq_to_name[vec_nr], h->action, |
298 | prev_count, preempt_count()); | |
299 | preempt_count_set(prev_count); | |
1da177e4 LT |
300 | } |
301 | h++; | |
2e702b9f JP |
302 | pending >>= softirq_bit; |
303 | } | |
1da177e4 | 304 | |
d28139c4 PM |
305 | if (__this_cpu_read(ksoftirqd) == current) |
306 | rcu_softirq_qs(); | |
c70f5d66 | 307 | local_irq_disable(); |
1da177e4 LT |
308 | |
309 | pending = local_softirq_pending(); | |
c10d7367 | 310 | if (pending) { |
34376a50 BG |
311 | if (time_before(jiffies, end) && !need_resched() && |
312 | --max_restart) | |
c10d7367 | 313 | goto restart; |
1da177e4 | 314 | |
1da177e4 | 315 | wakeup_softirqd(); |
c10d7367 | 316 | } |
1da177e4 | 317 | |
5c4853b6 | 318 | lockdep_softirq_end(in_hardirq); |
6a61671b | 319 | account_irq_exit_time(current); |
75e1056f | 320 | __local_bh_enable(SOFTIRQ_OFFSET); |
5d60d3e7 | 321 | WARN_ON_ONCE(in_interrupt()); |
717a94b5 | 322 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
323 | } |
324 | ||
722a9f92 | 325 | asmlinkage __visible void do_softirq(void) |
1da177e4 LT |
326 | { |
327 | __u32 pending; | |
328 | unsigned long flags; | |
329 | ||
330 | if (in_interrupt()) | |
331 | return; | |
332 | ||
333 | local_irq_save(flags); | |
334 | ||
335 | pending = local_softirq_pending(); | |
336 | ||
3c53776e | 337 | if (pending && !ksoftirqd_running(pending)) |
7d65f4a6 | 338 | do_softirq_own_stack(); |
1da177e4 LT |
339 | |
340 | local_irq_restore(flags); | |
341 | } | |
342 | ||
dde4b2b5 IM |
343 | /* |
344 | * Enter an interrupt context. | |
345 | */ | |
346 | void irq_enter(void) | |
347 | { | |
64db4cff | 348 | rcu_irq_enter(); |
0a8a2e78 | 349 | if (is_idle_task(current) && !in_interrupt()) { |
d267f87f VP |
350 | /* |
351 | * Prevent raise_softirq from needlessly waking up ksoftirqd | |
352 | * here, as softirq will be serviced on return from interrupt. | |
353 | */ | |
354 | local_bh_disable(); | |
5acac1be | 355 | tick_irq_enter(); |
d267f87f VP |
356 | _local_bh_enable(); |
357 | } | |
358 | ||
359 | __irq_enter(); | |
dde4b2b5 IM |
360 | } |
361 | ||
8d32a307 TG |
362 | static inline void invoke_softirq(void) |
363 | { | |
3c53776e | 364 | if (ksoftirqd_running(local_softirq_pending())) |
4cd13c21 ED |
365 | return; |
366 | ||
ded79754 | 367 | if (!force_irqthreads) { |
cc1f0274 | 368 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
ded79754 FW |
369 | /* |
370 | * We can safely execute softirq on the current stack if | |
371 | * it is the irq stack, because it should be near empty | |
cc1f0274 FW |
372 | * at this stage. |
373 | */ | |
374 | __do_softirq(); | |
375 | #else | |
376 | /* | |
377 | * Otherwise, irq_exit() is called on the task stack that can | |
378 | * be potentially deep already. So call softirq in its own stack | |
379 | * to prevent from any overrun. | |
ded79754 | 380 | */ |
be6e1016 | 381 | do_softirq_own_stack(); |
cc1f0274 | 382 | #endif |
ded79754 | 383 | } else { |
8d32a307 | 384 | wakeup_softirqd(); |
ded79754 | 385 | } |
8d32a307 | 386 | } |
1da177e4 | 387 | |
67826eae FW |
388 | static inline void tick_irq_exit(void) |
389 | { | |
390 | #ifdef CONFIG_NO_HZ_COMMON | |
391 | int cpu = smp_processor_id(); | |
392 | ||
393 | /* Make sure that timer wheel updates are propagated */ | |
394 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
0a0e0829 | 395 | if (!in_irq()) |
67826eae FW |
396 | tick_nohz_irq_exit(); |
397 | } | |
398 | #endif | |
399 | } | |
400 | ||
1da177e4 LT |
401 | /* |
402 | * Exit an interrupt context. Process softirqs if needed and possible: | |
403 | */ | |
404 | void irq_exit(void) | |
405 | { | |
74eed016 | 406 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 407 | local_irq_disable(); |
74eed016 | 408 | #else |
f71b74bc | 409 | lockdep_assert_irqs_disabled(); |
74eed016 | 410 | #endif |
6a61671b | 411 | account_irq_exit_time(current); |
bdb43806 | 412 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
413 | if (!in_interrupt() && local_softirq_pending()) |
414 | invoke_softirq(); | |
79bf2bb3 | 415 | |
67826eae | 416 | tick_irq_exit(); |
416eb33c | 417 | rcu_irq_exit(); |
f1a83e65 | 418 | trace_hardirq_exit(); /* must be last! */ |
1da177e4 LT |
419 | } |
420 | ||
421 | /* | |
422 | * This function must run with irqs disabled! | |
423 | */ | |
7ad5b3a5 | 424 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
425 | { |
426 | __raise_softirq_irqoff(nr); | |
427 | ||
428 | /* | |
429 | * If we're in an interrupt or softirq, we're done | |
430 | * (this also catches softirq-disabled code). We will | |
431 | * actually run the softirq once we return from | |
432 | * the irq or softirq. | |
433 | * | |
434 | * Otherwise we wake up ksoftirqd to make sure we | |
435 | * schedule the softirq soon. | |
436 | */ | |
437 | if (!in_interrupt()) | |
438 | wakeup_softirqd(); | |
439 | } | |
440 | ||
7ad5b3a5 | 441 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
442 | { |
443 | unsigned long flags; | |
444 | ||
445 | local_irq_save(flags); | |
446 | raise_softirq_irqoff(nr); | |
447 | local_irq_restore(flags); | |
448 | } | |
449 | ||
f069686e SR |
450 | void __raise_softirq_irqoff(unsigned int nr) |
451 | { | |
452 | trace_softirq_raise(nr); | |
453 | or_softirq_pending(1UL << nr); | |
454 | } | |
455 | ||
962cf36c | 456 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 457 | { |
1da177e4 LT |
458 | softirq_vec[nr].action = action; |
459 | } | |
460 | ||
9ba5f005 PZ |
461 | /* |
462 | * Tasklets | |
463 | */ | |
ce85b4f2 | 464 | struct tasklet_head { |
48f20a9a OJ |
465 | struct tasklet_struct *head; |
466 | struct tasklet_struct **tail; | |
1da177e4 LT |
467 | }; |
468 | ||
4620b49f VN |
469 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
470 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 471 | |
6498ddad IM |
472 | static void __tasklet_schedule_common(struct tasklet_struct *t, |
473 | struct tasklet_head __percpu *headp, | |
474 | unsigned int softirq_nr) | |
1da177e4 | 475 | { |
6498ddad | 476 | struct tasklet_head *head; |
1da177e4 LT |
477 | unsigned long flags; |
478 | ||
479 | local_irq_save(flags); | |
6498ddad | 480 | head = this_cpu_ptr(headp); |
48f20a9a | 481 | t->next = NULL; |
6498ddad IM |
482 | *head->tail = t; |
483 | head->tail = &(t->next); | |
484 | raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
485 | local_irq_restore(flags); |
486 | } | |
6498ddad IM |
487 | |
488 | void __tasklet_schedule(struct tasklet_struct *t) | |
489 | { | |
490 | __tasklet_schedule_common(t, &tasklet_vec, | |
491 | TASKLET_SOFTIRQ); | |
492 | } | |
1da177e4 LT |
493 | EXPORT_SYMBOL(__tasklet_schedule); |
494 | ||
7ad5b3a5 | 495 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 | 496 | { |
6498ddad IM |
497 | __tasklet_schedule_common(t, &tasklet_hi_vec, |
498 | HI_SOFTIRQ); | |
1da177e4 | 499 | } |
1da177e4 LT |
500 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
501 | ||
82b691be IM |
502 | static void tasklet_action_common(struct softirq_action *a, |
503 | struct tasklet_head *tl_head, | |
504 | unsigned int softirq_nr) | |
1da177e4 LT |
505 | { |
506 | struct tasklet_struct *list; | |
507 | ||
508 | local_irq_disable(); | |
82b691be IM |
509 | list = tl_head->head; |
510 | tl_head->head = NULL; | |
511 | tl_head->tail = &tl_head->head; | |
1da177e4 LT |
512 | local_irq_enable(); |
513 | ||
514 | while (list) { | |
515 | struct tasklet_struct *t = list; | |
516 | ||
517 | list = list->next; | |
518 | ||
519 | if (tasklet_trylock(t)) { | |
520 | if (!atomic_read(&t->count)) { | |
ce85b4f2 JP |
521 | if (!test_and_clear_bit(TASKLET_STATE_SCHED, |
522 | &t->state)) | |
1da177e4 LT |
523 | BUG(); |
524 | t->func(t->data); | |
525 | tasklet_unlock(t); | |
526 | continue; | |
527 | } | |
528 | tasklet_unlock(t); | |
529 | } | |
530 | ||
531 | local_irq_disable(); | |
48f20a9a | 532 | t->next = NULL; |
82b691be IM |
533 | *tl_head->tail = t; |
534 | tl_head->tail = &t->next; | |
535 | __raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
536 | local_irq_enable(); |
537 | } | |
538 | } | |
539 | ||
82b691be | 540 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 | 541 | { |
82b691be IM |
542 | tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ); |
543 | } | |
1da177e4 | 544 | |
82b691be IM |
545 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
546 | { | |
547 | tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ); | |
1da177e4 LT |
548 | } |
549 | ||
1da177e4 LT |
550 | void tasklet_init(struct tasklet_struct *t, |
551 | void (*func)(unsigned long), unsigned long data) | |
552 | { | |
553 | t->next = NULL; | |
554 | t->state = 0; | |
555 | atomic_set(&t->count, 0); | |
556 | t->func = func; | |
557 | t->data = data; | |
558 | } | |
1da177e4 LT |
559 | EXPORT_SYMBOL(tasklet_init); |
560 | ||
561 | void tasklet_kill(struct tasklet_struct *t) | |
562 | { | |
563 | if (in_interrupt()) | |
40322764 | 564 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 LT |
565 | |
566 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { | |
79d381c9 | 567 | do { |
1da177e4 | 568 | yield(); |
79d381c9 | 569 | } while (test_bit(TASKLET_STATE_SCHED, &t->state)); |
1da177e4 LT |
570 | } |
571 | tasklet_unlock_wait(t); | |
572 | clear_bit(TASKLET_STATE_SCHED, &t->state); | |
573 | } | |
1da177e4 LT |
574 | EXPORT_SYMBOL(tasklet_kill); |
575 | ||
9ba5f005 PZ |
576 | /* |
577 | * tasklet_hrtimer | |
578 | */ | |
579 | ||
580 | /* | |
b9c30322 PZ |
581 | * The trampoline is called when the hrtimer expires. It schedules a tasklet |
582 | * to run __tasklet_hrtimer_trampoline() which in turn will call the intended | |
583 | * hrtimer callback, but from softirq context. | |
9ba5f005 PZ |
584 | */ |
585 | static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer) | |
586 | { | |
587 | struct tasklet_hrtimer *ttimer = | |
588 | container_of(timer, struct tasklet_hrtimer, timer); | |
589 | ||
b9c30322 PZ |
590 | tasklet_hi_schedule(&ttimer->tasklet); |
591 | return HRTIMER_NORESTART; | |
9ba5f005 PZ |
592 | } |
593 | ||
594 | /* | |
595 | * Helper function which calls the hrtimer callback from | |
596 | * tasklet/softirq context | |
597 | */ | |
598 | static void __tasklet_hrtimer_trampoline(unsigned long data) | |
599 | { | |
600 | struct tasklet_hrtimer *ttimer = (void *)data; | |
601 | enum hrtimer_restart restart; | |
602 | ||
603 | restart = ttimer->function(&ttimer->timer); | |
604 | if (restart != HRTIMER_NORESTART) | |
605 | hrtimer_restart(&ttimer->timer); | |
606 | } | |
607 | ||
608 | /** | |
609 | * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks | |
610 | * @ttimer: tasklet_hrtimer which is initialized | |
25985edc | 611 | * @function: hrtimer callback function which gets called from softirq context |
9ba5f005 PZ |
612 | * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME) |
613 | * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL) | |
614 | */ | |
615 | void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer, | |
616 | enum hrtimer_restart (*function)(struct hrtimer *), | |
617 | clockid_t which_clock, enum hrtimer_mode mode) | |
618 | { | |
619 | hrtimer_init(&ttimer->timer, which_clock, mode); | |
620 | ttimer->timer.function = __hrtimer_tasklet_trampoline; | |
621 | tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline, | |
622 | (unsigned long)ttimer); | |
623 | ttimer->function = function; | |
624 | } | |
625 | EXPORT_SYMBOL_GPL(tasklet_hrtimer_init); | |
626 | ||
1da177e4 LT |
627 | void __init softirq_init(void) |
628 | { | |
48f20a9a OJ |
629 | int cpu; |
630 | ||
631 | for_each_possible_cpu(cpu) { | |
632 | per_cpu(tasklet_vec, cpu).tail = | |
633 | &per_cpu(tasklet_vec, cpu).head; | |
634 | per_cpu(tasklet_hi_vec, cpu).tail = | |
635 | &per_cpu(tasklet_hi_vec, cpu).head; | |
636 | } | |
637 | ||
962cf36c CM |
638 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
639 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
640 | } |
641 | ||
3e339b5d | 642 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 643 | { |
3e339b5d TG |
644 | return local_softirq_pending(); |
645 | } | |
1da177e4 | 646 | |
3e339b5d TG |
647 | static void run_ksoftirqd(unsigned int cpu) |
648 | { | |
649 | local_irq_disable(); | |
650 | if (local_softirq_pending()) { | |
0bed698a FW |
651 | /* |
652 | * We can safely run softirq on inline stack, as we are not deep | |
653 | * in the task stack here. | |
654 | */ | |
3e339b5d | 655 | __do_softirq(); |
3e339b5d | 656 | local_irq_enable(); |
edf22f4c | 657 | cond_resched(); |
3e339b5d | 658 | return; |
1da177e4 | 659 | } |
3e339b5d | 660 | local_irq_enable(); |
1da177e4 LT |
661 | } |
662 | ||
663 | #ifdef CONFIG_HOTPLUG_CPU | |
664 | /* | |
665 | * tasklet_kill_immediate is called to remove a tasklet which can already be | |
666 | * scheduled for execution on @cpu. | |
667 | * | |
668 | * Unlike tasklet_kill, this function removes the tasklet | |
669 | * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state. | |
670 | * | |
671 | * When this function is called, @cpu must be in the CPU_DEAD state. | |
672 | */ | |
673 | void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu) | |
674 | { | |
675 | struct tasklet_struct **i; | |
676 | ||
677 | BUG_ON(cpu_online(cpu)); | |
678 | BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state)); | |
679 | ||
680 | if (!test_bit(TASKLET_STATE_SCHED, &t->state)) | |
681 | return; | |
682 | ||
683 | /* CPU is dead, so no lock needed. */ | |
48f20a9a | 684 | for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) { |
1da177e4 LT |
685 | if (*i == t) { |
686 | *i = t->next; | |
48f20a9a OJ |
687 | /* If this was the tail element, move the tail ptr */ |
688 | if (*i == NULL) | |
689 | per_cpu(tasklet_vec, cpu).tail = i; | |
1da177e4 LT |
690 | return; |
691 | } | |
692 | } | |
693 | BUG(); | |
694 | } | |
695 | ||
c4544dbc | 696 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 697 | { |
1da177e4 LT |
698 | /* CPU is dead, so no lock needed. */ |
699 | local_irq_disable(); | |
700 | ||
701 | /* Find end, append list for that CPU. */ | |
e5e41723 | 702 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 CL |
703 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
704 | this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); | |
e5e41723 CB |
705 | per_cpu(tasklet_vec, cpu).head = NULL; |
706 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
707 | } | |
1da177e4 LT |
708 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
709 | ||
e5e41723 | 710 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
711 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
712 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
713 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
714 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
715 | } | |
1da177e4 LT |
716 | raise_softirq_irqoff(HI_SOFTIRQ); |
717 | ||
718 | local_irq_enable(); | |
c4544dbc | 719 | return 0; |
1da177e4 | 720 | } |
c4544dbc SAS |
721 | #else |
722 | #define takeover_tasklets NULL | |
1da177e4 LT |
723 | #endif /* CONFIG_HOTPLUG_CPU */ |
724 | ||
3e339b5d TG |
725 | static struct smp_hotplug_thread softirq_threads = { |
726 | .store = &ksoftirqd, | |
727 | .thread_should_run = ksoftirqd_should_run, | |
728 | .thread_fn = run_ksoftirqd, | |
729 | .thread_comm = "ksoftirqd/%u", | |
730 | }; | |
731 | ||
7babe8db | 732 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 733 | { |
c4544dbc SAS |
734 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
735 | takeover_tasklets); | |
3e339b5d TG |
736 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
737 | ||
1da177e4 LT |
738 | return 0; |
739 | } | |
7babe8db | 740 | early_initcall(spawn_ksoftirqd); |
78eef01b | 741 | |
43a25632 YL |
742 | /* |
743 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
744 | * GCC does not inline them incorrectly. ] | |
745 | */ | |
746 | ||
747 | int __init __weak early_irq_init(void) | |
748 | { | |
749 | return 0; | |
750 | } | |
751 | ||
4a046d17 YL |
752 | int __init __weak arch_probe_nr_irqs(void) |
753 | { | |
b683de2b | 754 | return NR_IRQS_LEGACY; |
4a046d17 YL |
755 | } |
756 | ||
43a25632 YL |
757 | int __init __weak arch_early_irq_init(void) |
758 | { | |
759 | return 0; | |
760 | } | |
62a08ae2 TG |
761 | |
762 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
763 | { | |
764 | return from; | |
765 | } |