Commit | Line | Data |
---|---|---|
767a67b0 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/kernel/softirq.c | |
4 | * | |
5 | * Copyright (C) 1992 Linus Torvalds | |
6 | * | |
b10db7f0 | 7 | * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903) |
1da177e4 LT |
8 | */ |
9 | ||
40322764 JP |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
9984de1a | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/kernel_stat.h> |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/init.h> | |
8b1c04ac | 16 | #include <linux/local_lock.h> |
1da177e4 LT |
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> |
da044747 | 29 | #include <linux/wait_bit.h> |
4cb1ef64 | 30 | #include <linux/workqueue.h> |
a0e39ed3 | 31 | |
db1cc7ae TG |
32 | #include <asm/softirq_stack.h> |
33 | ||
a0e39ed3 | 34 | #define CREATE_TRACE_POINTS |
ad8d75ff | 35 | #include <trace/events/irq.h> |
1da177e4 | 36 | |
1da177e4 LT |
37 | /* |
38 | - No shared variables, all the data are CPU local. | |
39 | - If a softirq needs serialization, let it serialize itself | |
40 | by its own spinlocks. | |
41 | - Even if softirq is serialized, only local cpu is marked for | |
42 | execution. Hence, we get something sort of weak cpu binding. | |
43 | Though it is still not clear, will it result in better locality | |
44 | or will not. | |
45 | ||
46 | Examples: | |
47 | - NET RX softirq. It is multithreaded and does not require | |
48 | any global serialization. | |
49 | - NET TX softirq. It kicks software netdevice queues, hence | |
50 | it is logically serialized per device, but this serialization | |
51 | is invisible to common code. | |
52 | - Tasklets: serialized wrt itself. | |
53 | */ | |
54 | ||
55 | #ifndef __ARCH_IRQ_STAT | |
0f6f47ba FW |
56 | DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat); |
57 | EXPORT_PER_CPU_SYMBOL(irq_stat); | |
1da177e4 LT |
58 | #endif |
59 | ||
978b0116 | 60 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 61 | |
4dd53d89 | 62 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 63 | |
ce85b4f2 | 64 | const char * const softirq_to_name[NR_SOFTIRQS] = { |
f660f606 | 65 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL", |
09223371 | 66 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
5d592b44 JB |
67 | }; |
68 | ||
1da177e4 LT |
69 | /* |
70 | * we cannot loop indefinitely here to avoid userspace starvation, | |
71 | * but we also don't want to introduce a worst case 1/HZ latency | |
72 | * to the pending events, so lets the scheduler to balance | |
73 | * the softirq load for us. | |
74 | */ | |
676cb02d | 75 | static void wakeup_softirqd(void) |
1da177e4 LT |
76 | { |
77 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 78 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 | 79 | |
37aadc68 | 80 | if (tsk) |
1da177e4 LT |
81 | wake_up_process(tsk); |
82 | } | |
83 | ||
ae9ef589 TG |
84 | #ifdef CONFIG_TRACE_IRQFLAGS |
85 | DEFINE_PER_CPU(int, hardirqs_enabled); | |
86 | DEFINE_PER_CPU(int, hardirq_context); | |
87 | EXPORT_PER_CPU_SYMBOL_GPL(hardirqs_enabled); | |
88 | EXPORT_PER_CPU_SYMBOL_GPL(hardirq_context); | |
89 | #endif | |
90 | ||
75e1056f | 91 | /* |
8b1c04ac TG |
92 | * SOFTIRQ_OFFSET usage: |
93 | * | |
94 | * On !RT kernels 'count' is the preempt counter, on RT kernels this applies | |
95 | * to a per CPU counter and to task::softirqs_disabled_cnt. | |
96 | * | |
97 | * - count is changed by SOFTIRQ_OFFSET on entering or leaving softirq | |
98 | * processing. | |
99 | * | |
100 | * - count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
75e1056f | 101 | * on local_bh_disable or local_bh_enable. |
8b1c04ac | 102 | * |
75e1056f VP |
103 | * This lets us distinguish between whether we are currently processing |
104 | * softirq and whether we just have bh disabled. | |
105 | */ | |
8b1c04ac TG |
106 | #ifdef CONFIG_PREEMPT_RT |
107 | ||
108 | /* | |
109 | * RT accounts for BH disabled sections in task::softirqs_disabled_cnt and | |
110 | * also in per CPU softirq_ctrl::cnt. This is necessary to allow tasks in a | |
111 | * softirq disabled section to be preempted. | |
112 | * | |
113 | * The per task counter is used for softirq_count(), in_softirq() and | |
114 | * in_serving_softirqs() because these counts are only valid when the task | |
115 | * holding softirq_ctrl::lock is running. | |
116 | * | |
117 | * The per CPU counter prevents pointless wakeups of ksoftirqd in case that | |
118 | * the task which is in a softirq disabled section is preempted or blocks. | |
119 | */ | |
120 | struct softirq_ctrl { | |
121 | local_lock_t lock; | |
122 | int cnt; | |
123 | }; | |
124 | ||
125 | static DEFINE_PER_CPU(struct softirq_ctrl, softirq_ctrl) = { | |
126 | .lock = INIT_LOCAL_LOCK(softirq_ctrl.lock), | |
127 | }; | |
128 | ||
47c218dc TG |
129 | /** |
130 | * local_bh_blocked() - Check for idle whether BH processing is blocked | |
131 | * | |
132 | * Returns false if the per CPU softirq::cnt is 0 otherwise true. | |
133 | * | |
134 | * This is invoked from the idle task to guard against false positive | |
135 | * softirq pending warnings, which would happen when the task which holds | |
136 | * softirq_ctrl::lock was the only running task on the CPU and blocks on | |
137 | * some other lock. | |
138 | */ | |
139 | bool local_bh_blocked(void) | |
140 | { | |
141 | return __this_cpu_read(softirq_ctrl.cnt) != 0; | |
142 | } | |
143 | ||
8b1c04ac TG |
144 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
145 | { | |
146 | unsigned long flags; | |
147 | int newcnt; | |
148 | ||
149 | WARN_ON_ONCE(in_hardirq()); | |
150 | ||
151 | /* First entry of a task into a BH disabled section? */ | |
152 | if (!current->softirq_disable_cnt) { | |
153 | if (preemptible()) { | |
154 | local_lock(&softirq_ctrl.lock); | |
155 | /* Required to meet the RCU bottomhalf requirements. */ | |
156 | rcu_read_lock(); | |
157 | } else { | |
158 | DEBUG_LOCKS_WARN_ON(this_cpu_read(softirq_ctrl.cnt)); | |
159 | } | |
160 | } | |
161 | ||
162 | /* | |
163 | * Track the per CPU softirq disabled state. On RT this is per CPU | |
164 | * state to allow preemption of bottom half disabled sections. | |
165 | */ | |
166 | newcnt = __this_cpu_add_return(softirq_ctrl.cnt, cnt); | |
167 | /* | |
168 | * Reflect the result in the task state to prevent recursion on the | |
169 | * local lock and to make softirq_count() & al work. | |
170 | */ | |
171 | current->softirq_disable_cnt = newcnt; | |
172 | ||
173 | if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && newcnt == cnt) { | |
174 | raw_local_irq_save(flags); | |
175 | lockdep_softirqs_off(ip); | |
176 | raw_local_irq_restore(flags); | |
177 | } | |
178 | } | |
179 | EXPORT_SYMBOL(__local_bh_disable_ip); | |
180 | ||
181 | static void __local_bh_enable(unsigned int cnt, bool unlock) | |
182 | { | |
183 | unsigned long flags; | |
184 | int newcnt; | |
185 | ||
186 | DEBUG_LOCKS_WARN_ON(current->softirq_disable_cnt != | |
187 | this_cpu_read(softirq_ctrl.cnt)); | |
188 | ||
189 | if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && softirq_count() == cnt) { | |
190 | raw_local_irq_save(flags); | |
191 | lockdep_softirqs_on(_RET_IP_); | |
192 | raw_local_irq_restore(flags); | |
193 | } | |
194 | ||
195 | newcnt = __this_cpu_sub_return(softirq_ctrl.cnt, cnt); | |
196 | current->softirq_disable_cnt = newcnt; | |
197 | ||
198 | if (!newcnt && unlock) { | |
199 | rcu_read_unlock(); | |
200 | local_unlock(&softirq_ctrl.lock); | |
201 | } | |
202 | } | |
203 | ||
204 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) | |
205 | { | |
206 | bool preempt_on = preemptible(); | |
207 | unsigned long flags; | |
208 | u32 pending; | |
209 | int curcnt; | |
210 | ||
fe13889c | 211 | WARN_ON_ONCE(in_hardirq()); |
8b1c04ac TG |
212 | lockdep_assert_irqs_enabled(); |
213 | ||
214 | local_irq_save(flags); | |
215 | curcnt = __this_cpu_read(softirq_ctrl.cnt); | |
216 | ||
217 | /* | |
218 | * If this is not reenabling soft interrupts, no point in trying to | |
219 | * run pending ones. | |
220 | */ | |
221 | if (curcnt != cnt) | |
222 | goto out; | |
223 | ||
224 | pending = local_softirq_pending(); | |
d15121be | 225 | if (!pending) |
8b1c04ac TG |
226 | goto out; |
227 | ||
228 | /* | |
229 | * If this was called from non preemptible context, wake up the | |
230 | * softirq daemon. | |
231 | */ | |
232 | if (!preempt_on) { | |
233 | wakeup_softirqd(); | |
234 | goto out; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Adjust softirq count to SOFTIRQ_OFFSET which makes | |
239 | * in_serving_softirq() become true. | |
240 | */ | |
241 | cnt = SOFTIRQ_OFFSET; | |
242 | __local_bh_enable(cnt, false); | |
243 | __do_softirq(); | |
244 | ||
245 | out: | |
246 | __local_bh_enable(cnt, preempt_on); | |
247 | local_irq_restore(flags); | |
248 | } | |
249 | EXPORT_SYMBOL(__local_bh_enable_ip); | |
75e1056f | 250 | |
de30a2b3 | 251 | /* |
8b1c04ac TG |
252 | * Invoked from ksoftirqd_run() outside of the interrupt disabled section |
253 | * to acquire the per CPU local lock for reentrancy protection. | |
254 | */ | |
255 | static inline void ksoftirqd_run_begin(void) | |
256 | { | |
257 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); | |
258 | local_irq_disable(); | |
259 | } | |
260 | ||
261 | /* Counterpart to ksoftirqd_run_begin() */ | |
262 | static inline void ksoftirqd_run_end(void) | |
263 | { | |
264 | __local_bh_enable(SOFTIRQ_OFFSET, true); | |
265 | WARN_ON_ONCE(in_interrupt()); | |
266 | local_irq_enable(); | |
267 | } | |
268 | ||
269 | static inline void softirq_handle_begin(void) { } | |
270 | static inline void softirq_handle_end(void) { } | |
271 | ||
272 | static inline bool should_wake_ksoftirqd(void) | |
273 | { | |
274 | return !this_cpu_read(softirq_ctrl.cnt); | |
275 | } | |
276 | ||
277 | static inline void invoke_softirq(void) | |
278 | { | |
279 | if (should_wake_ksoftirqd()) | |
280 | wakeup_softirqd(); | |
281 | } | |
282 | ||
1a90bfd2 SAS |
283 | /* |
284 | * flush_smp_call_function_queue() can raise a soft interrupt in a function | |
285 | * call. On RT kernels this is undesired and the only known functionality | |
286 | * in the block layer which does this is disabled on RT. If soft interrupts | |
287 | * get raised which haven't been raised before the flush, warn so it can be | |
288 | * investigated. | |
289 | */ | |
290 | void do_softirq_post_smp_call_flush(unsigned int was_pending) | |
291 | { | |
292 | if (WARN_ON_ONCE(was_pending != local_softirq_pending())) | |
293 | invoke_softirq(); | |
294 | } | |
295 | ||
8b1c04ac TG |
296 | #else /* CONFIG_PREEMPT_RT */ |
297 | ||
298 | /* | |
299 | * This one is for softirq.c-internal use, where hardirqs are disabled | |
ae9ef589 | 300 | * legitimately: |
de30a2b3 | 301 | */ |
8b1c04ac | 302 | #ifdef CONFIG_TRACE_IRQFLAGS |
0bd3a173 | 303 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
304 | { |
305 | unsigned long flags; | |
306 | ||
fe13889c | 307 | WARN_ON_ONCE(in_hardirq()); |
de30a2b3 IM |
308 | |
309 | raw_local_irq_save(flags); | |
7e49fcce | 310 | /* |
bdb43806 | 311 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
312 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
313 | * is set and before current->softirq_enabled is cleared. | |
314 | * We must manually increment preempt_count here and manually | |
315 | * call the trace_preempt_off later. | |
316 | */ | |
bdb43806 | 317 | __preempt_count_add(cnt); |
de30a2b3 IM |
318 | /* |
319 | * Were softirqs turned off above: | |
320 | */ | |
9ea4c380 | 321 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
0d38453c | 322 | lockdep_softirqs_off(ip); |
de30a2b3 | 323 | raw_local_irq_restore(flags); |
7e49fcce | 324 | |
0f1ba9a2 HC |
325 | if (preempt_count() == cnt) { |
326 | #ifdef CONFIG_DEBUG_PREEMPT | |
f904f582 | 327 | current->preempt_disable_ip = get_lock_parent_ip(); |
0f1ba9a2 | 328 | #endif |
f904f582 | 329 | trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip()); |
0f1ba9a2 | 330 | } |
de30a2b3 | 331 | } |
0bd3a173 | 332 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 333 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 334 | |
75e1056f VP |
335 | static void __local_bh_enable(unsigned int cnt) |
336 | { | |
f71b74bc | 337 | lockdep_assert_irqs_disabled(); |
75e1056f | 338 | |
1a63dcd8 JFG |
339 | if (preempt_count() == cnt) |
340 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
341 | ||
9ea4c380 | 342 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
0d38453c | 343 | lockdep_softirqs_on(_RET_IP_); |
1a63dcd8 JFG |
344 | |
345 | __preempt_count_sub(cnt); | |
75e1056f VP |
346 | } |
347 | ||
de30a2b3 | 348 | /* |
c3442697 | 349 | * Special-case - softirqs can safely be enabled by __do_softirq(), |
de30a2b3 IM |
350 | * without processing still-pending softirqs: |
351 | */ | |
352 | void _local_bh_enable(void) | |
353 | { | |
fe13889c | 354 | WARN_ON_ONCE(in_hardirq()); |
75e1056f | 355 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 356 | } |
de30a2b3 IM |
357 | EXPORT_SYMBOL(_local_bh_enable); |
358 | ||
0bd3a173 | 359 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 360 | { |
fe13889c | 361 | WARN_ON_ONCE(in_hardirq()); |
f71b74bc | 362 | lockdep_assert_irqs_enabled(); |
3c829c36 | 363 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 364 | local_irq_disable(); |
3c829c36 | 365 | #endif |
de30a2b3 IM |
366 | /* |
367 | * Are softirqs going to be turned on now: | |
368 | */ | |
75e1056f | 369 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0d38453c | 370 | lockdep_softirqs_on(ip); |
de30a2b3 IM |
371 | /* |
372 | * Keep preemption disabled until we are done with | |
373 | * softirq processing: | |
ce85b4f2 | 374 | */ |
91ea62d5 | 375 | __preempt_count_sub(cnt - 1); |
de30a2b3 | 376 | |
0bed698a FW |
377 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
378 | /* | |
379 | * Run softirq if any pending. And do it in its own stack | |
380 | * as we may be calling this deep in a task call stack already. | |
381 | */ | |
de30a2b3 | 382 | do_softirq(); |
0bed698a | 383 | } |
de30a2b3 | 384 | |
bdb43806 | 385 | preempt_count_dec(); |
3c829c36 | 386 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 387 | local_irq_enable(); |
3c829c36 | 388 | #endif |
de30a2b3 IM |
389 | preempt_check_resched(); |
390 | } | |
0bd3a173 | 391 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 392 | |
f02fc963 TG |
393 | static inline void softirq_handle_begin(void) |
394 | { | |
395 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); | |
396 | } | |
397 | ||
398 | static inline void softirq_handle_end(void) | |
399 | { | |
400 | __local_bh_enable(SOFTIRQ_OFFSET); | |
401 | WARN_ON_ONCE(in_interrupt()); | |
402 | } | |
403 | ||
404 | static inline void ksoftirqd_run_begin(void) | |
405 | { | |
406 | local_irq_disable(); | |
407 | } | |
408 | ||
409 | static inline void ksoftirqd_run_end(void) | |
410 | { | |
411 | local_irq_enable(); | |
412 | } | |
413 | ||
414 | static inline bool should_wake_ksoftirqd(void) | |
415 | { | |
416 | return true; | |
417 | } | |
418 | ||
ae9ef589 TG |
419 | static inline void invoke_softirq(void) |
420 | { | |
91cc470e | 421 | if (!force_irqthreads() || !__this_cpu_read(ksoftirqd)) { |
ae9ef589 TG |
422 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
423 | /* | |
424 | * We can safely execute softirq on the current stack if | |
425 | * it is the irq stack, because it should be near empty | |
426 | * at this stage. | |
427 | */ | |
428 | __do_softirq(); | |
429 | #else | |
430 | /* | |
431 | * Otherwise, irq_exit() is called on the task stack that can | |
432 | * be potentially deep already. So call softirq in its own stack | |
433 | * to prevent from any overrun. | |
434 | */ | |
435 | do_softirq_own_stack(); | |
436 | #endif | |
437 | } else { | |
438 | wakeup_softirqd(); | |
439 | } | |
440 | } | |
441 | ||
442 | asmlinkage __visible void do_softirq(void) | |
443 | { | |
444 | __u32 pending; | |
445 | unsigned long flags; | |
446 | ||
447 | if (in_interrupt()) | |
448 | return; | |
449 | ||
450 | local_irq_save(flags); | |
451 | ||
452 | pending = local_softirq_pending(); | |
453 | ||
d15121be | 454 | if (pending) |
ae9ef589 TG |
455 | do_softirq_own_stack(); |
456 | ||
457 | local_irq_restore(flags); | |
458 | } | |
459 | ||
8b1c04ac TG |
460 | #endif /* !CONFIG_PREEMPT_RT */ |
461 | ||
1da177e4 | 462 | /* |
34376a50 BG |
463 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
464 | * but break the loop if need_resched() is set or after 2 ms. | |
465 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
466 | * certain cases, such as stop_machine(), jiffies may cease to | |
467 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
468 | * well to make sure we eventually return from this method. | |
1da177e4 | 469 | * |
c10d7367 | 470 | * These limits have been established via experimentation. |
1da177e4 LT |
471 | * The two things to balance is latency against fairness - |
472 | * we want to handle softirqs as soon as possible, but they | |
473 | * should not be able to lock up the box. | |
474 | */ | |
c10d7367 | 475 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 476 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 477 | |
f1a83e65 PZ |
478 | #ifdef CONFIG_TRACE_IRQFLAGS |
479 | /* | |
f1a83e65 PZ |
480 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
481 | * to keep the lockdep irq context tracking as tight as possible in order to | |
482 | * not miss-qualify lock contexts and miss possible deadlocks. | |
483 | */ | |
f1a83e65 | 484 | |
5c4853b6 | 485 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 486 | { |
5c4853b6 | 487 | bool in_hardirq = false; |
f1a83e65 | 488 | |
f9ad4a5f | 489 | if (lockdep_hardirq_context()) { |
5c4853b6 | 490 | in_hardirq = true; |
2502ec37 | 491 | lockdep_hardirq_exit(); |
5c4853b6 FW |
492 | } |
493 | ||
f1a83e65 | 494 | lockdep_softirq_enter(); |
5c4853b6 FW |
495 | |
496 | return in_hardirq; | |
f1a83e65 PZ |
497 | } |
498 | ||
5c4853b6 | 499 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
500 | { |
501 | lockdep_softirq_exit(); | |
5c4853b6 FW |
502 | |
503 | if (in_hardirq) | |
2502ec37 | 504 | lockdep_hardirq_enter(); |
f1a83e65 | 505 | } |
f1a83e65 | 506 | #else |
5c4853b6 FW |
507 | static inline bool lockdep_softirq_start(void) { return false; } |
508 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
509 | #endif |
510 | ||
be7635e7 | 511 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 512 | { |
c10d7367 | 513 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 514 | unsigned long old_flags = current->flags; |
34376a50 | 515 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 516 | struct softirq_action *h; |
5c4853b6 | 517 | bool in_hardirq; |
f1a83e65 | 518 | __u32 pending; |
2e702b9f | 519 | int softirq_bit; |
907aed48 MG |
520 | |
521 | /* | |
e45506ac YL |
522 | * Mask out PF_MEMALLOC as the current task context is borrowed for the |
523 | * softirq. A softirq handled, such as network RX, might set PF_MEMALLOC | |
524 | * again if the socket is related to swapping. | |
907aed48 MG |
525 | */ |
526 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
527 | |
528 | pending = local_softirq_pending(); | |
829035fd | 529 | |
f02fc963 | 530 | softirq_handle_begin(); |
5c4853b6 | 531 | in_hardirq = lockdep_softirq_start(); |
d3759e71 | 532 | account_softirq_enter(current); |
1da177e4 | 533 | |
1da177e4 LT |
534 | restart: |
535 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 536 | set_softirq_pending(0); |
1da177e4 | 537 | |
c70f5d66 | 538 | local_irq_enable(); |
1da177e4 LT |
539 | |
540 | h = softirq_vec; | |
541 | ||
2e702b9f JP |
542 | while ((softirq_bit = ffs(pending))) { |
543 | unsigned int vec_nr; | |
544 | int prev_count; | |
545 | ||
546 | h += softirq_bit - 1; | |
547 | ||
548 | vec_nr = h - softirq_vec; | |
549 | prev_count = preempt_count(); | |
550 | ||
551 | kstat_incr_softirqs_this_cpu(vec_nr); | |
552 | ||
553 | trace_softirq_entry(vec_nr); | |
554 | h->action(h); | |
555 | trace_softirq_exit(vec_nr); | |
556 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 557 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
558 | vec_nr, softirq_to_name[vec_nr], h->action, |
559 | prev_count, preempt_count()); | |
560 | preempt_count_set(prev_count); | |
1da177e4 LT |
561 | } |
562 | h++; | |
2e702b9f JP |
563 | pending >>= softirq_bit; |
564 | } | |
1da177e4 | 565 | |
8b1c04ac TG |
566 | if (!IS_ENABLED(CONFIG_PREEMPT_RT) && |
567 | __this_cpu_read(ksoftirqd) == current) | |
d28139c4 | 568 | rcu_softirq_qs(); |
8b1c04ac | 569 | |
c70f5d66 | 570 | local_irq_disable(); |
1da177e4 LT |
571 | |
572 | pending = local_softirq_pending(); | |
c10d7367 | 573 | if (pending) { |
34376a50 BG |
574 | if (time_before(jiffies, end) && !need_resched() && |
575 | --max_restart) | |
c10d7367 | 576 | goto restart; |
1da177e4 | 577 | |
1da177e4 | 578 | wakeup_softirqd(); |
c10d7367 | 579 | } |
1da177e4 | 580 | |
d3759e71 | 581 | account_softirq_exit(current); |
5c4853b6 | 582 | lockdep_softirq_end(in_hardirq); |
f02fc963 | 583 | softirq_handle_end(); |
717a94b5 | 584 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
585 | } |
586 | ||
8a6bc478 TG |
587 | /** |
588 | * irq_enter_rcu - Enter an interrupt context with RCU watching | |
dde4b2b5 | 589 | */ |
8a6bc478 | 590 | void irq_enter_rcu(void) |
dde4b2b5 | 591 | { |
d14ce74f FW |
592 | __irq_enter_raw(); |
593 | ||
53e87e3c FW |
594 | if (tick_nohz_full_cpu(smp_processor_id()) || |
595 | (is_idle_task(current) && (irq_count() == HARDIRQ_OFFSET))) | |
5acac1be | 596 | tick_irq_enter(); |
d14ce74f FW |
597 | |
598 | account_hardirq_enter(current); | |
dde4b2b5 IM |
599 | } |
600 | ||
8a6bc478 TG |
601 | /** |
602 | * irq_enter - Enter an interrupt context including RCU update | |
603 | */ | |
604 | void irq_enter(void) | |
605 | { | |
6f0e6c15 | 606 | ct_irq_enter(); |
8a6bc478 TG |
607 | irq_enter_rcu(); |
608 | } | |
609 | ||
67826eae FW |
610 | static inline void tick_irq_exit(void) |
611 | { | |
612 | #ifdef CONFIG_NO_HZ_COMMON | |
613 | int cpu = smp_processor_id(); | |
614 | ||
615 | /* Make sure that timer wheel updates are propagated */ | |
548796e2 | 616 | if ((sched_core_idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { |
fe13889c | 617 | if (!in_hardirq()) |
67826eae FW |
618 | tick_nohz_irq_exit(); |
619 | } | |
620 | #endif | |
621 | } | |
622 | ||
59bc300b | 623 | static inline void __irq_exit_rcu(void) |
1da177e4 | 624 | { |
74eed016 | 625 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 626 | local_irq_disable(); |
74eed016 | 627 | #else |
f71b74bc | 628 | lockdep_assert_irqs_disabled(); |
74eed016 | 629 | #endif |
d3759e71 | 630 | account_hardirq_exit(current); |
bdb43806 | 631 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
632 | if (!in_interrupt() && local_softirq_pending()) |
633 | invoke_softirq(); | |
79bf2bb3 | 634 | |
67826eae | 635 | tick_irq_exit(); |
8a6bc478 TG |
636 | } |
637 | ||
59bc300b PZ |
638 | /** |
639 | * irq_exit_rcu() - Exit an interrupt context without updating RCU | |
640 | * | |
641 | * Also processes softirqs if needed and possible. | |
642 | */ | |
643 | void irq_exit_rcu(void) | |
644 | { | |
645 | __irq_exit_rcu(); | |
646 | /* must be last! */ | |
647 | lockdep_hardirq_exit(); | |
648 | } | |
649 | ||
8a6bc478 TG |
650 | /** |
651 | * irq_exit - Exit an interrupt context, update RCU and lockdep | |
652 | * | |
653 | * Also processes softirqs if needed and possible. | |
654 | */ | |
655 | void irq_exit(void) | |
656 | { | |
59bc300b | 657 | __irq_exit_rcu(); |
6f0e6c15 | 658 | ct_irq_exit(); |
2502ec37 TG |
659 | /* must be last! */ |
660 | lockdep_hardirq_exit(); | |
1da177e4 LT |
661 | } |
662 | ||
663 | /* | |
664 | * This function must run with irqs disabled! | |
665 | */ | |
7ad5b3a5 | 666 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
667 | { |
668 | __raise_softirq_irqoff(nr); | |
669 | ||
670 | /* | |
671 | * If we're in an interrupt or softirq, we're done | |
672 | * (this also catches softirq-disabled code). We will | |
673 | * actually run the softirq once we return from | |
674 | * the irq or softirq. | |
675 | * | |
676 | * Otherwise we wake up ksoftirqd to make sure we | |
677 | * schedule the softirq soon. | |
678 | */ | |
f02fc963 | 679 | if (!in_interrupt() && should_wake_ksoftirqd()) |
1da177e4 LT |
680 | wakeup_softirqd(); |
681 | } | |
682 | ||
7ad5b3a5 | 683 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
684 | { |
685 | unsigned long flags; | |
686 | ||
687 | local_irq_save(flags); | |
688 | raise_softirq_irqoff(nr); | |
689 | local_irq_restore(flags); | |
690 | } | |
691 | ||
f069686e SR |
692 | void __raise_softirq_irqoff(unsigned int nr) |
693 | { | |
cdabce2e | 694 | lockdep_assert_irqs_disabled(); |
f069686e SR |
695 | trace_softirq_raise(nr); |
696 | or_softirq_pending(1UL << nr); | |
697 | } | |
698 | ||
962cf36c | 699 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 700 | { |
1da177e4 LT |
701 | softirq_vec[nr].action = action; |
702 | } | |
703 | ||
9ba5f005 PZ |
704 | /* |
705 | * Tasklets | |
706 | */ | |
ce85b4f2 | 707 | struct tasklet_head { |
48f20a9a OJ |
708 | struct tasklet_struct *head; |
709 | struct tasklet_struct **tail; | |
1da177e4 LT |
710 | }; |
711 | ||
4620b49f VN |
712 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
713 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 714 | |
6498ddad IM |
715 | static void __tasklet_schedule_common(struct tasklet_struct *t, |
716 | struct tasklet_head __percpu *headp, | |
717 | unsigned int softirq_nr) | |
1da177e4 | 718 | { |
6498ddad | 719 | struct tasklet_head *head; |
1da177e4 LT |
720 | unsigned long flags; |
721 | ||
722 | local_irq_save(flags); | |
6498ddad | 723 | head = this_cpu_ptr(headp); |
48f20a9a | 724 | t->next = NULL; |
6498ddad IM |
725 | *head->tail = t; |
726 | head->tail = &(t->next); | |
727 | raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
728 | local_irq_restore(flags); |
729 | } | |
6498ddad IM |
730 | |
731 | void __tasklet_schedule(struct tasklet_struct *t) | |
732 | { | |
733 | __tasklet_schedule_common(t, &tasklet_vec, | |
734 | TASKLET_SOFTIRQ); | |
735 | } | |
1da177e4 LT |
736 | EXPORT_SYMBOL(__tasklet_schedule); |
737 | ||
7ad5b3a5 | 738 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 | 739 | { |
6498ddad IM |
740 | __tasklet_schedule_common(t, &tasklet_hi_vec, |
741 | HI_SOFTIRQ); | |
1da177e4 | 742 | } |
1da177e4 LT |
743 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
744 | ||
697d8c63 | 745 | static bool tasklet_clear_sched(struct tasklet_struct *t) |
6b2c339d | 746 | { |
697d8c63 PZ |
747 | if (test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) { |
748 | wake_up_var(&t->state); | |
6b2c339d | 749 | return true; |
697d8c63 | 750 | } |
6b2c339d DB |
751 | |
752 | WARN_ONCE(1, "tasklet SCHED state not set: %s %pS\n", | |
753 | t->use_callback ? "callback" : "func", | |
754 | t->use_callback ? (void *)t->callback : (void *)t->func); | |
755 | ||
756 | return false; | |
757 | } | |
758 | ||
82b691be IM |
759 | static void tasklet_action_common(struct softirq_action *a, |
760 | struct tasklet_head *tl_head, | |
761 | unsigned int softirq_nr) | |
1da177e4 LT |
762 | { |
763 | struct tasklet_struct *list; | |
764 | ||
765 | local_irq_disable(); | |
82b691be IM |
766 | list = tl_head->head; |
767 | tl_head->head = NULL; | |
768 | tl_head->tail = &tl_head->head; | |
1da177e4 LT |
769 | local_irq_enable(); |
770 | ||
771 | while (list) { | |
772 | struct tasklet_struct *t = list; | |
773 | ||
774 | list = list->next; | |
775 | ||
776 | if (tasklet_trylock(t)) { | |
777 | if (!atomic_read(&t->count)) { | |
697d8c63 | 778 | if (tasklet_clear_sched(t)) { |
f4bf3ca2 LC |
779 | if (t->use_callback) { |
780 | trace_tasklet_entry(t, t->callback); | |
6b2c339d | 781 | t->callback(t); |
f4bf3ca2 LC |
782 | trace_tasklet_exit(t, t->callback); |
783 | } else { | |
784 | trace_tasklet_entry(t, t->func); | |
6b2c339d | 785 | t->func(t->data); |
f4bf3ca2 LC |
786 | trace_tasklet_exit(t, t->func); |
787 | } | |
6b2c339d | 788 | } |
1da177e4 LT |
789 | tasklet_unlock(t); |
790 | continue; | |
791 | } | |
792 | tasklet_unlock(t); | |
793 | } | |
794 | ||
795 | local_irq_disable(); | |
48f20a9a | 796 | t->next = NULL; |
82b691be IM |
797 | *tl_head->tail = t; |
798 | tl_head->tail = &t->next; | |
799 | __raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
800 | local_irq_enable(); |
801 | } | |
802 | } | |
803 | ||
82b691be | 804 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 | 805 | { |
4cb1ef64 | 806 | workqueue_softirq_action(false); |
82b691be IM |
807 | tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ); |
808 | } | |
1da177e4 | 809 | |
82b691be IM |
810 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
811 | { | |
4cb1ef64 | 812 | workqueue_softirq_action(true); |
82b691be | 813 | tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ); |
1da177e4 LT |
814 | } |
815 | ||
12cc923f RP |
816 | void tasklet_setup(struct tasklet_struct *t, |
817 | void (*callback)(struct tasklet_struct *)) | |
818 | { | |
819 | t->next = NULL; | |
820 | t->state = 0; | |
821 | atomic_set(&t->count, 0); | |
822 | t->callback = callback; | |
823 | t->use_callback = true; | |
824 | t->data = 0; | |
825 | } | |
826 | EXPORT_SYMBOL(tasklet_setup); | |
827 | ||
1da177e4 LT |
828 | void tasklet_init(struct tasklet_struct *t, |
829 | void (*func)(unsigned long), unsigned long data) | |
830 | { | |
831 | t->next = NULL; | |
832 | t->state = 0; | |
833 | atomic_set(&t->count, 0); | |
834 | t->func = func; | |
12cc923f | 835 | t->use_callback = false; |
1da177e4 LT |
836 | t->data = data; |
837 | } | |
1da177e4 LT |
838 | EXPORT_SYMBOL(tasklet_init); |
839 | ||
eb2dafbb TG |
840 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
841 | /* | |
842 | * Do not use in new code. Waiting for tasklets from atomic contexts is | |
843 | * error prone and should be avoided. | |
844 | */ | |
845 | void tasklet_unlock_spin_wait(struct tasklet_struct *t) | |
846 | { | |
847 | while (test_bit(TASKLET_STATE_RUN, &(t)->state)) { | |
848 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
849 | /* | |
850 | * Prevent a live lock when current preempted soft | |
851 | * interrupt processing or prevents ksoftirqd from | |
852 | * running. If the tasklet runs on a different CPU | |
853 | * then this has no effect other than doing the BH | |
854 | * disable/enable dance for nothing. | |
855 | */ | |
856 | local_bh_disable(); | |
857 | local_bh_enable(); | |
858 | } else { | |
859 | cpu_relax(); | |
860 | } | |
861 | } | |
862 | } | |
863 | EXPORT_SYMBOL(tasklet_unlock_spin_wait); | |
864 | #endif | |
865 | ||
1da177e4 LT |
866 | void tasklet_kill(struct tasklet_struct *t) |
867 | { | |
868 | if (in_interrupt()) | |
40322764 | 869 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 | 870 | |
697d8c63 PZ |
871 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) |
872 | wait_var_event(&t->state, !test_bit(TASKLET_STATE_SCHED, &t->state)); | |
873 | ||
1da177e4 | 874 | tasklet_unlock_wait(t); |
697d8c63 | 875 | tasklet_clear_sched(t); |
1da177e4 | 876 | } |
1da177e4 LT |
877 | EXPORT_SYMBOL(tasklet_kill); |
878 | ||
eb2dafbb | 879 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
da044747 PZ |
880 | void tasklet_unlock(struct tasklet_struct *t) |
881 | { | |
882 | smp_mb__before_atomic(); | |
883 | clear_bit(TASKLET_STATE_RUN, &t->state); | |
884 | smp_mb__after_atomic(); | |
885 | wake_up_var(&t->state); | |
886 | } | |
887 | EXPORT_SYMBOL_GPL(tasklet_unlock); | |
888 | ||
889 | void tasklet_unlock_wait(struct tasklet_struct *t) | |
890 | { | |
891 | wait_var_event(&t->state, !test_bit(TASKLET_STATE_RUN, &t->state)); | |
892 | } | |
893 | EXPORT_SYMBOL_GPL(tasklet_unlock_wait); | |
894 | #endif | |
895 | ||
1da177e4 LT |
896 | void __init softirq_init(void) |
897 | { | |
48f20a9a OJ |
898 | int cpu; |
899 | ||
900 | for_each_possible_cpu(cpu) { | |
901 | per_cpu(tasklet_vec, cpu).tail = | |
902 | &per_cpu(tasklet_vec, cpu).head; | |
903 | per_cpu(tasklet_hi_vec, cpu).tail = | |
904 | &per_cpu(tasklet_hi_vec, cpu).head; | |
905 | } | |
906 | ||
962cf36c CM |
907 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
908 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
909 | } |
910 | ||
3e339b5d | 911 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 912 | { |
3e339b5d TG |
913 | return local_softirq_pending(); |
914 | } | |
1da177e4 | 915 | |
3e339b5d TG |
916 | static void run_ksoftirqd(unsigned int cpu) |
917 | { | |
f02fc963 | 918 | ksoftirqd_run_begin(); |
3e339b5d | 919 | if (local_softirq_pending()) { |
0bed698a FW |
920 | /* |
921 | * We can safely run softirq on inline stack, as we are not deep | |
922 | * in the task stack here. | |
923 | */ | |
3e339b5d | 924 | __do_softirq(); |
f02fc963 | 925 | ksoftirqd_run_end(); |
edf22f4c | 926 | cond_resched(); |
3e339b5d | 927 | return; |
1da177e4 | 928 | } |
f02fc963 | 929 | ksoftirqd_run_end(); |
1da177e4 LT |
930 | } |
931 | ||
932 | #ifdef CONFIG_HOTPLUG_CPU | |
c4544dbc | 933 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 934 | { |
1acd92d9 TH |
935 | workqueue_softirq_dead(cpu); |
936 | ||
1da177e4 LT |
937 | /* CPU is dead, so no lock needed. */ |
938 | local_irq_disable(); | |
939 | ||
940 | /* Find end, append list for that CPU. */ | |
e5e41723 | 941 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 | 942 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
8afecaa6 | 943 | __this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); |
e5e41723 CB |
944 | per_cpu(tasklet_vec, cpu).head = NULL; |
945 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
946 | } | |
1da177e4 LT |
947 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
948 | ||
e5e41723 | 949 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
950 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
951 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
952 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
953 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
954 | } | |
1da177e4 LT |
955 | raise_softirq_irqoff(HI_SOFTIRQ); |
956 | ||
957 | local_irq_enable(); | |
c4544dbc | 958 | return 0; |
1da177e4 | 959 | } |
c4544dbc SAS |
960 | #else |
961 | #define takeover_tasklets NULL | |
1da177e4 LT |
962 | #endif /* CONFIG_HOTPLUG_CPU */ |
963 | ||
3e339b5d TG |
964 | static struct smp_hotplug_thread softirq_threads = { |
965 | .store = &ksoftirqd, | |
966 | .thread_should_run = ksoftirqd_should_run, | |
967 | .thread_fn = run_ksoftirqd, | |
968 | .thread_comm = "ksoftirqd/%u", | |
969 | }; | |
970 | ||
7babe8db | 971 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 972 | { |
c4544dbc SAS |
973 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
974 | takeover_tasklets); | |
3e339b5d TG |
975 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
976 | ||
1da177e4 LT |
977 | return 0; |
978 | } | |
7babe8db | 979 | early_initcall(spawn_ksoftirqd); |
78eef01b | 980 | |
43a25632 YL |
981 | /* |
982 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
983 | * GCC does not inline them incorrectly. ] | |
984 | */ | |
985 | ||
986 | int __init __weak early_irq_init(void) | |
987 | { | |
988 | return 0; | |
989 | } | |
990 | ||
4a046d17 YL |
991 | int __init __weak arch_probe_nr_irqs(void) |
992 | { | |
b683de2b | 993 | return NR_IRQS_LEGACY; |
4a046d17 YL |
994 | } |
995 | ||
43a25632 YL |
996 | int __init __weak arch_early_irq_init(void) |
997 | { | |
998 | return 0; | |
999 | } | |
62a08ae2 TG |
1000 | |
1001 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
1002 | { | |
1003 | return from; | |
1004 | } |