]>
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> |
a0e39ed3 | 30 | |
db1cc7ae TG |
31 | #include <asm/softirq_stack.h> |
32 | ||
a0e39ed3 | 33 | #define CREATE_TRACE_POINTS |
ad8d75ff | 34 | #include <trace/events/irq.h> |
1da177e4 | 35 | |
1da177e4 LT |
36 | /* |
37 | - No shared variables, all the data are CPU local. | |
38 | - If a softirq needs serialization, let it serialize itself | |
39 | by its own spinlocks. | |
40 | - Even if softirq is serialized, only local cpu is marked for | |
41 | execution. Hence, we get something sort of weak cpu binding. | |
42 | Though it is still not clear, will it result in better locality | |
43 | or will not. | |
44 | ||
45 | Examples: | |
46 | - NET RX softirq. It is multithreaded and does not require | |
47 | any global serialization. | |
48 | - NET TX softirq. It kicks software netdevice queues, hence | |
49 | it is logically serialized per device, but this serialization | |
50 | is invisible to common code. | |
51 | - Tasklets: serialized wrt itself. | |
52 | */ | |
53 | ||
54 | #ifndef __ARCH_IRQ_STAT | |
0f6f47ba FW |
55 | DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat); |
56 | EXPORT_PER_CPU_SYMBOL(irq_stat); | |
1da177e4 LT |
57 | #endif |
58 | ||
978b0116 | 59 | static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp; |
1da177e4 | 60 | |
4dd53d89 | 61 | DEFINE_PER_CPU(struct task_struct *, ksoftirqd); |
1da177e4 | 62 | |
ce85b4f2 | 63 | const char * const softirq_to_name[NR_SOFTIRQS] = { |
f660f606 | 64 | "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL", |
09223371 | 65 | "TASKLET", "SCHED", "HRTIMER", "RCU" |
5d592b44 JB |
66 | }; |
67 | ||
1da177e4 LT |
68 | /* |
69 | * we cannot loop indefinitely here to avoid userspace starvation, | |
70 | * but we also don't want to introduce a worst case 1/HZ latency | |
71 | * to the pending events, so lets the scheduler to balance | |
72 | * the softirq load for us. | |
73 | */ | |
676cb02d | 74 | static void wakeup_softirqd(void) |
1da177e4 LT |
75 | { |
76 | /* Interrupts are disabled: no need to stop preemption */ | |
909ea964 | 77 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); |
1da177e4 LT |
78 | |
79 | if (tsk && tsk->state != TASK_RUNNING) | |
80 | wake_up_process(tsk); | |
81 | } | |
82 | ||
4cd13c21 ED |
83 | /* |
84 | * If ksoftirqd is scheduled, we do not want to process pending softirqs | |
3c53776e LT |
85 | * right now. Let ksoftirqd handle this at its own rate, to get fairness, |
86 | * unless we're doing some of the synchronous softirqs. | |
4cd13c21 | 87 | */ |
3c53776e LT |
88 | #define SOFTIRQ_NOW_MASK ((1 << HI_SOFTIRQ) | (1 << TASKLET_SOFTIRQ)) |
89 | static bool ksoftirqd_running(unsigned long pending) | |
4cd13c21 ED |
90 | { |
91 | struct task_struct *tsk = __this_cpu_read(ksoftirqd); | |
92 | ||
3c53776e LT |
93 | if (pending & SOFTIRQ_NOW_MASK) |
94 | return false; | |
1342d808 MK |
95 | return tsk && (tsk->state == TASK_RUNNING) && |
96 | !__kthread_should_park(tsk); | |
4cd13c21 ED |
97 | } |
98 | ||
ae9ef589 TG |
99 | #ifdef CONFIG_TRACE_IRQFLAGS |
100 | DEFINE_PER_CPU(int, hardirqs_enabled); | |
101 | DEFINE_PER_CPU(int, hardirq_context); | |
102 | EXPORT_PER_CPU_SYMBOL_GPL(hardirqs_enabled); | |
103 | EXPORT_PER_CPU_SYMBOL_GPL(hardirq_context); | |
104 | #endif | |
105 | ||
75e1056f | 106 | /* |
8b1c04ac TG |
107 | * SOFTIRQ_OFFSET usage: |
108 | * | |
109 | * On !RT kernels 'count' is the preempt counter, on RT kernels this applies | |
110 | * to a per CPU counter and to task::softirqs_disabled_cnt. | |
111 | * | |
112 | * - count is changed by SOFTIRQ_OFFSET on entering or leaving softirq | |
113 | * processing. | |
114 | * | |
115 | * - count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET) | |
75e1056f | 116 | * on local_bh_disable or local_bh_enable. |
8b1c04ac | 117 | * |
75e1056f VP |
118 | * This lets us distinguish between whether we are currently processing |
119 | * softirq and whether we just have bh disabled. | |
120 | */ | |
8b1c04ac TG |
121 | #ifdef CONFIG_PREEMPT_RT |
122 | ||
123 | /* | |
124 | * RT accounts for BH disabled sections in task::softirqs_disabled_cnt and | |
125 | * also in per CPU softirq_ctrl::cnt. This is necessary to allow tasks in a | |
126 | * softirq disabled section to be preempted. | |
127 | * | |
128 | * The per task counter is used for softirq_count(), in_softirq() and | |
129 | * in_serving_softirqs() because these counts are only valid when the task | |
130 | * holding softirq_ctrl::lock is running. | |
131 | * | |
132 | * The per CPU counter prevents pointless wakeups of ksoftirqd in case that | |
133 | * the task which is in a softirq disabled section is preempted or blocks. | |
134 | */ | |
135 | struct softirq_ctrl { | |
136 | local_lock_t lock; | |
137 | int cnt; | |
138 | }; | |
139 | ||
140 | static DEFINE_PER_CPU(struct softirq_ctrl, softirq_ctrl) = { | |
141 | .lock = INIT_LOCAL_LOCK(softirq_ctrl.lock), | |
142 | }; | |
143 | ||
47c218dc TG |
144 | /** |
145 | * local_bh_blocked() - Check for idle whether BH processing is blocked | |
146 | * | |
147 | * Returns false if the per CPU softirq::cnt is 0 otherwise true. | |
148 | * | |
149 | * This is invoked from the idle task to guard against false positive | |
150 | * softirq pending warnings, which would happen when the task which holds | |
151 | * softirq_ctrl::lock was the only running task on the CPU and blocks on | |
152 | * some other lock. | |
153 | */ | |
154 | bool local_bh_blocked(void) | |
155 | { | |
156 | return __this_cpu_read(softirq_ctrl.cnt) != 0; | |
157 | } | |
158 | ||
8b1c04ac TG |
159 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
160 | { | |
161 | unsigned long flags; | |
162 | int newcnt; | |
163 | ||
164 | WARN_ON_ONCE(in_hardirq()); | |
165 | ||
166 | /* First entry of a task into a BH disabled section? */ | |
167 | if (!current->softirq_disable_cnt) { | |
168 | if (preemptible()) { | |
169 | local_lock(&softirq_ctrl.lock); | |
170 | /* Required to meet the RCU bottomhalf requirements. */ | |
171 | rcu_read_lock(); | |
172 | } else { | |
173 | DEBUG_LOCKS_WARN_ON(this_cpu_read(softirq_ctrl.cnt)); | |
174 | } | |
175 | } | |
176 | ||
177 | /* | |
178 | * Track the per CPU softirq disabled state. On RT this is per CPU | |
179 | * state to allow preemption of bottom half disabled sections. | |
180 | */ | |
181 | newcnt = __this_cpu_add_return(softirq_ctrl.cnt, cnt); | |
182 | /* | |
183 | * Reflect the result in the task state to prevent recursion on the | |
184 | * local lock and to make softirq_count() & al work. | |
185 | */ | |
186 | current->softirq_disable_cnt = newcnt; | |
187 | ||
188 | if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && newcnt == cnt) { | |
189 | raw_local_irq_save(flags); | |
190 | lockdep_softirqs_off(ip); | |
191 | raw_local_irq_restore(flags); | |
192 | } | |
193 | } | |
194 | EXPORT_SYMBOL(__local_bh_disable_ip); | |
195 | ||
196 | static void __local_bh_enable(unsigned int cnt, bool unlock) | |
197 | { | |
198 | unsigned long flags; | |
199 | int newcnt; | |
200 | ||
201 | DEBUG_LOCKS_WARN_ON(current->softirq_disable_cnt != | |
202 | this_cpu_read(softirq_ctrl.cnt)); | |
203 | ||
204 | if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && softirq_count() == cnt) { | |
205 | raw_local_irq_save(flags); | |
206 | lockdep_softirqs_on(_RET_IP_); | |
207 | raw_local_irq_restore(flags); | |
208 | } | |
209 | ||
210 | newcnt = __this_cpu_sub_return(softirq_ctrl.cnt, cnt); | |
211 | current->softirq_disable_cnt = newcnt; | |
212 | ||
213 | if (!newcnt && unlock) { | |
214 | rcu_read_unlock(); | |
215 | local_unlock(&softirq_ctrl.lock); | |
216 | } | |
217 | } | |
218 | ||
219 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) | |
220 | { | |
221 | bool preempt_on = preemptible(); | |
222 | unsigned long flags; | |
223 | u32 pending; | |
224 | int curcnt; | |
225 | ||
226 | WARN_ON_ONCE(in_irq()); | |
227 | lockdep_assert_irqs_enabled(); | |
228 | ||
229 | local_irq_save(flags); | |
230 | curcnt = __this_cpu_read(softirq_ctrl.cnt); | |
231 | ||
232 | /* | |
233 | * If this is not reenabling soft interrupts, no point in trying to | |
234 | * run pending ones. | |
235 | */ | |
236 | if (curcnt != cnt) | |
237 | goto out; | |
238 | ||
239 | pending = local_softirq_pending(); | |
240 | if (!pending || ksoftirqd_running(pending)) | |
241 | goto out; | |
242 | ||
243 | /* | |
244 | * If this was called from non preemptible context, wake up the | |
245 | * softirq daemon. | |
246 | */ | |
247 | if (!preempt_on) { | |
248 | wakeup_softirqd(); | |
249 | goto out; | |
250 | } | |
251 | ||
252 | /* | |
253 | * Adjust softirq count to SOFTIRQ_OFFSET which makes | |
254 | * in_serving_softirq() become true. | |
255 | */ | |
256 | cnt = SOFTIRQ_OFFSET; | |
257 | __local_bh_enable(cnt, false); | |
258 | __do_softirq(); | |
259 | ||
260 | out: | |
261 | __local_bh_enable(cnt, preempt_on); | |
262 | local_irq_restore(flags); | |
263 | } | |
264 | EXPORT_SYMBOL(__local_bh_enable_ip); | |
75e1056f | 265 | |
de30a2b3 | 266 | /* |
8b1c04ac TG |
267 | * Invoked from ksoftirqd_run() outside of the interrupt disabled section |
268 | * to acquire the per CPU local lock for reentrancy protection. | |
269 | */ | |
270 | static inline void ksoftirqd_run_begin(void) | |
271 | { | |
272 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); | |
273 | local_irq_disable(); | |
274 | } | |
275 | ||
276 | /* Counterpart to ksoftirqd_run_begin() */ | |
277 | static inline void ksoftirqd_run_end(void) | |
278 | { | |
279 | __local_bh_enable(SOFTIRQ_OFFSET, true); | |
280 | WARN_ON_ONCE(in_interrupt()); | |
281 | local_irq_enable(); | |
282 | } | |
283 | ||
284 | static inline void softirq_handle_begin(void) { } | |
285 | static inline void softirq_handle_end(void) { } | |
286 | ||
287 | static inline bool should_wake_ksoftirqd(void) | |
288 | { | |
289 | return !this_cpu_read(softirq_ctrl.cnt); | |
290 | } | |
291 | ||
292 | static inline void invoke_softirq(void) | |
293 | { | |
294 | if (should_wake_ksoftirqd()) | |
295 | wakeup_softirqd(); | |
296 | } | |
297 | ||
298 | #else /* CONFIG_PREEMPT_RT */ | |
299 | ||
300 | /* | |
301 | * This one is for softirq.c-internal use, where hardirqs are disabled | |
ae9ef589 | 302 | * legitimately: |
de30a2b3 | 303 | */ |
8b1c04ac | 304 | #ifdef CONFIG_TRACE_IRQFLAGS |
0bd3a173 | 305 | void __local_bh_disable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 IM |
306 | { |
307 | unsigned long flags; | |
308 | ||
309 | WARN_ON_ONCE(in_irq()); | |
310 | ||
311 | raw_local_irq_save(flags); | |
7e49fcce | 312 | /* |
bdb43806 | 313 | * The preempt tracer hooks into preempt_count_add and will break |
7e49fcce SR |
314 | * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET |
315 | * is set and before current->softirq_enabled is cleared. | |
316 | * We must manually increment preempt_count here and manually | |
317 | * call the trace_preempt_off later. | |
318 | */ | |
bdb43806 | 319 | __preempt_count_add(cnt); |
de30a2b3 IM |
320 | /* |
321 | * Were softirqs turned off above: | |
322 | */ | |
9ea4c380 | 323 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
0d38453c | 324 | lockdep_softirqs_off(ip); |
de30a2b3 | 325 | raw_local_irq_restore(flags); |
7e49fcce | 326 | |
0f1ba9a2 HC |
327 | if (preempt_count() == cnt) { |
328 | #ifdef CONFIG_DEBUG_PREEMPT | |
f904f582 | 329 | current->preempt_disable_ip = get_lock_parent_ip(); |
0f1ba9a2 | 330 | #endif |
f904f582 | 331 | trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip()); |
0f1ba9a2 | 332 | } |
de30a2b3 | 333 | } |
0bd3a173 | 334 | EXPORT_SYMBOL(__local_bh_disable_ip); |
3c829c36 | 335 | #endif /* CONFIG_TRACE_IRQFLAGS */ |
de30a2b3 | 336 | |
75e1056f VP |
337 | static void __local_bh_enable(unsigned int cnt) |
338 | { | |
f71b74bc | 339 | lockdep_assert_irqs_disabled(); |
75e1056f | 340 | |
1a63dcd8 JFG |
341 | if (preempt_count() == cnt) |
342 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
343 | ||
9ea4c380 | 344 | if (softirq_count() == (cnt & SOFTIRQ_MASK)) |
0d38453c | 345 | lockdep_softirqs_on(_RET_IP_); |
1a63dcd8 JFG |
346 | |
347 | __preempt_count_sub(cnt); | |
75e1056f VP |
348 | } |
349 | ||
de30a2b3 | 350 | /* |
c3442697 | 351 | * Special-case - softirqs can safely be enabled by __do_softirq(), |
de30a2b3 IM |
352 | * without processing still-pending softirqs: |
353 | */ | |
354 | void _local_bh_enable(void) | |
355 | { | |
5d60d3e7 | 356 | WARN_ON_ONCE(in_irq()); |
75e1056f | 357 | __local_bh_enable(SOFTIRQ_DISABLE_OFFSET); |
de30a2b3 | 358 | } |
de30a2b3 IM |
359 | EXPORT_SYMBOL(_local_bh_enable); |
360 | ||
0bd3a173 | 361 | void __local_bh_enable_ip(unsigned long ip, unsigned int cnt) |
de30a2b3 | 362 | { |
f71b74bc FW |
363 | WARN_ON_ONCE(in_irq()); |
364 | lockdep_assert_irqs_enabled(); | |
3c829c36 | 365 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 366 | local_irq_disable(); |
3c829c36 | 367 | #endif |
de30a2b3 IM |
368 | /* |
369 | * Are softirqs going to be turned on now: | |
370 | */ | |
75e1056f | 371 | if (softirq_count() == SOFTIRQ_DISABLE_OFFSET) |
0d38453c | 372 | lockdep_softirqs_on(ip); |
de30a2b3 IM |
373 | /* |
374 | * Keep preemption disabled until we are done with | |
375 | * softirq processing: | |
ce85b4f2 | 376 | */ |
91ea62d5 | 377 | __preempt_count_sub(cnt - 1); |
de30a2b3 | 378 | |
0bed698a FW |
379 | if (unlikely(!in_interrupt() && local_softirq_pending())) { |
380 | /* | |
381 | * Run softirq if any pending. And do it in its own stack | |
382 | * as we may be calling this deep in a task call stack already. | |
383 | */ | |
de30a2b3 | 384 | do_softirq(); |
0bed698a | 385 | } |
de30a2b3 | 386 | |
bdb43806 | 387 | preempt_count_dec(); |
3c829c36 | 388 | #ifdef CONFIG_TRACE_IRQFLAGS |
0f476b6d | 389 | local_irq_enable(); |
3c829c36 | 390 | #endif |
de30a2b3 IM |
391 | preempt_check_resched(); |
392 | } | |
0bd3a173 | 393 | EXPORT_SYMBOL(__local_bh_enable_ip); |
de30a2b3 | 394 | |
f02fc963 TG |
395 | static inline void softirq_handle_begin(void) |
396 | { | |
397 | __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET); | |
398 | } | |
399 | ||
400 | static inline void softirq_handle_end(void) | |
401 | { | |
402 | __local_bh_enable(SOFTIRQ_OFFSET); | |
403 | WARN_ON_ONCE(in_interrupt()); | |
404 | } | |
405 | ||
406 | static inline void ksoftirqd_run_begin(void) | |
407 | { | |
408 | local_irq_disable(); | |
409 | } | |
410 | ||
411 | static inline void ksoftirqd_run_end(void) | |
412 | { | |
413 | local_irq_enable(); | |
414 | } | |
415 | ||
416 | static inline bool should_wake_ksoftirqd(void) | |
417 | { | |
418 | return true; | |
419 | } | |
420 | ||
ae9ef589 TG |
421 | static inline void invoke_softirq(void) |
422 | { | |
423 | if (ksoftirqd_running(local_softirq_pending())) | |
424 | return; | |
425 | ||
1c0c4bc1 | 426 | if (!force_irqthreads || !__this_cpu_read(ksoftirqd)) { |
ae9ef589 TG |
427 | #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK |
428 | /* | |
429 | * We can safely execute softirq on the current stack if | |
430 | * it is the irq stack, because it should be near empty | |
431 | * at this stage. | |
432 | */ | |
433 | __do_softirq(); | |
434 | #else | |
435 | /* | |
436 | * Otherwise, irq_exit() is called on the task stack that can | |
437 | * be potentially deep already. So call softirq in its own stack | |
438 | * to prevent from any overrun. | |
439 | */ | |
440 | do_softirq_own_stack(); | |
441 | #endif | |
442 | } else { | |
443 | wakeup_softirqd(); | |
444 | } | |
445 | } | |
446 | ||
447 | asmlinkage __visible void do_softirq(void) | |
448 | { | |
449 | __u32 pending; | |
450 | unsigned long flags; | |
451 | ||
452 | if (in_interrupt()) | |
453 | return; | |
454 | ||
455 | local_irq_save(flags); | |
456 | ||
457 | pending = local_softirq_pending(); | |
458 | ||
459 | if (pending && !ksoftirqd_running(pending)) | |
460 | do_softirq_own_stack(); | |
461 | ||
462 | local_irq_restore(flags); | |
463 | } | |
464 | ||
8b1c04ac TG |
465 | #endif /* !CONFIG_PREEMPT_RT */ |
466 | ||
1da177e4 | 467 | /* |
34376a50 BG |
468 | * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times, |
469 | * but break the loop if need_resched() is set or after 2 ms. | |
470 | * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in | |
471 | * certain cases, such as stop_machine(), jiffies may cease to | |
472 | * increment and so we need the MAX_SOFTIRQ_RESTART limit as | |
473 | * well to make sure we eventually return from this method. | |
1da177e4 | 474 | * |
c10d7367 | 475 | * These limits have been established via experimentation. |
1da177e4 LT |
476 | * The two things to balance is latency against fairness - |
477 | * we want to handle softirqs as soon as possible, but they | |
478 | * should not be able to lock up the box. | |
479 | */ | |
c10d7367 | 480 | #define MAX_SOFTIRQ_TIME msecs_to_jiffies(2) |
34376a50 | 481 | #define MAX_SOFTIRQ_RESTART 10 |
1da177e4 | 482 | |
f1a83e65 PZ |
483 | #ifdef CONFIG_TRACE_IRQFLAGS |
484 | /* | |
f1a83e65 PZ |
485 | * When we run softirqs from irq_exit() and thus on the hardirq stack we need |
486 | * to keep the lockdep irq context tracking as tight as possible in order to | |
487 | * not miss-qualify lock contexts and miss possible deadlocks. | |
488 | */ | |
f1a83e65 | 489 | |
5c4853b6 | 490 | static inline bool lockdep_softirq_start(void) |
f1a83e65 | 491 | { |
5c4853b6 | 492 | bool in_hardirq = false; |
f1a83e65 | 493 | |
f9ad4a5f | 494 | if (lockdep_hardirq_context()) { |
5c4853b6 | 495 | in_hardirq = true; |
2502ec37 | 496 | lockdep_hardirq_exit(); |
5c4853b6 FW |
497 | } |
498 | ||
f1a83e65 | 499 | lockdep_softirq_enter(); |
5c4853b6 FW |
500 | |
501 | return in_hardirq; | |
f1a83e65 PZ |
502 | } |
503 | ||
5c4853b6 | 504 | static inline void lockdep_softirq_end(bool in_hardirq) |
f1a83e65 PZ |
505 | { |
506 | lockdep_softirq_exit(); | |
5c4853b6 FW |
507 | |
508 | if (in_hardirq) | |
2502ec37 | 509 | lockdep_hardirq_enter(); |
f1a83e65 | 510 | } |
f1a83e65 | 511 | #else |
5c4853b6 FW |
512 | static inline bool lockdep_softirq_start(void) { return false; } |
513 | static inline void lockdep_softirq_end(bool in_hardirq) { } | |
f1a83e65 PZ |
514 | #endif |
515 | ||
be7635e7 | 516 | asmlinkage __visible void __softirq_entry __do_softirq(void) |
1da177e4 | 517 | { |
c10d7367 | 518 | unsigned long end = jiffies + MAX_SOFTIRQ_TIME; |
907aed48 | 519 | unsigned long old_flags = current->flags; |
34376a50 | 520 | int max_restart = MAX_SOFTIRQ_RESTART; |
f1a83e65 | 521 | struct softirq_action *h; |
5c4853b6 | 522 | bool in_hardirq; |
f1a83e65 | 523 | __u32 pending; |
2e702b9f | 524 | int softirq_bit; |
907aed48 MG |
525 | |
526 | /* | |
e45506ac YL |
527 | * Mask out PF_MEMALLOC as the current task context is borrowed for the |
528 | * softirq. A softirq handled, such as network RX, might set PF_MEMALLOC | |
529 | * again if the socket is related to swapping. | |
907aed48 MG |
530 | */ |
531 | current->flags &= ~PF_MEMALLOC; | |
1da177e4 LT |
532 | |
533 | pending = local_softirq_pending(); | |
829035fd | 534 | |
f02fc963 | 535 | softirq_handle_begin(); |
5c4853b6 | 536 | in_hardirq = lockdep_softirq_start(); |
d3759e71 | 537 | account_softirq_enter(current); |
1da177e4 | 538 | |
1da177e4 LT |
539 | restart: |
540 | /* Reset the pending bitmask before enabling irqs */ | |
3f74478b | 541 | set_softirq_pending(0); |
1da177e4 | 542 | |
c70f5d66 | 543 | local_irq_enable(); |
1da177e4 LT |
544 | |
545 | h = softirq_vec; | |
546 | ||
2e702b9f JP |
547 | while ((softirq_bit = ffs(pending))) { |
548 | unsigned int vec_nr; | |
549 | int prev_count; | |
550 | ||
551 | h += softirq_bit - 1; | |
552 | ||
553 | vec_nr = h - softirq_vec; | |
554 | prev_count = preempt_count(); | |
555 | ||
556 | kstat_incr_softirqs_this_cpu(vec_nr); | |
557 | ||
558 | trace_softirq_entry(vec_nr); | |
559 | h->action(h); | |
560 | trace_softirq_exit(vec_nr); | |
561 | if (unlikely(prev_count != preempt_count())) { | |
40322764 | 562 | pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n", |
2e702b9f JP |
563 | vec_nr, softirq_to_name[vec_nr], h->action, |
564 | prev_count, preempt_count()); | |
565 | preempt_count_set(prev_count); | |
1da177e4 LT |
566 | } |
567 | h++; | |
2e702b9f JP |
568 | pending >>= softirq_bit; |
569 | } | |
1da177e4 | 570 | |
8b1c04ac TG |
571 | if (!IS_ENABLED(CONFIG_PREEMPT_RT) && |
572 | __this_cpu_read(ksoftirqd) == current) | |
d28139c4 | 573 | rcu_softirq_qs(); |
8b1c04ac | 574 | |
c70f5d66 | 575 | local_irq_disable(); |
1da177e4 LT |
576 | |
577 | pending = local_softirq_pending(); | |
c10d7367 | 578 | if (pending) { |
34376a50 BG |
579 | if (time_before(jiffies, end) && !need_resched() && |
580 | --max_restart) | |
c10d7367 | 581 | goto restart; |
1da177e4 | 582 | |
1da177e4 | 583 | wakeup_softirqd(); |
c10d7367 | 584 | } |
1da177e4 | 585 | |
d3759e71 | 586 | account_softirq_exit(current); |
5c4853b6 | 587 | lockdep_softirq_end(in_hardirq); |
f02fc963 | 588 | softirq_handle_end(); |
717a94b5 | 589 | current_restore_flags(old_flags, PF_MEMALLOC); |
1da177e4 LT |
590 | } |
591 | ||
8a6bc478 TG |
592 | /** |
593 | * irq_enter_rcu - Enter an interrupt context with RCU watching | |
dde4b2b5 | 594 | */ |
8a6bc478 | 595 | void irq_enter_rcu(void) |
dde4b2b5 | 596 | { |
d14ce74f FW |
597 | __irq_enter_raw(); |
598 | ||
599 | if (is_idle_task(current) && (irq_count() == HARDIRQ_OFFSET)) | |
5acac1be | 600 | tick_irq_enter(); |
d14ce74f FW |
601 | |
602 | account_hardirq_enter(current); | |
dde4b2b5 IM |
603 | } |
604 | ||
8a6bc478 TG |
605 | /** |
606 | * irq_enter - Enter an interrupt context including RCU update | |
607 | */ | |
608 | void irq_enter(void) | |
609 | { | |
610 | rcu_irq_enter(); | |
611 | irq_enter_rcu(); | |
612 | } | |
613 | ||
67826eae FW |
614 | static inline void tick_irq_exit(void) |
615 | { | |
616 | #ifdef CONFIG_NO_HZ_COMMON | |
617 | int cpu = smp_processor_id(); | |
618 | ||
619 | /* Make sure that timer wheel updates are propagated */ | |
620 | if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) { | |
0a0e0829 | 621 | if (!in_irq()) |
67826eae FW |
622 | tick_nohz_irq_exit(); |
623 | } | |
624 | #endif | |
625 | } | |
626 | ||
59bc300b | 627 | static inline void __irq_exit_rcu(void) |
1da177e4 | 628 | { |
74eed016 | 629 | #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED |
4cd5d111 | 630 | local_irq_disable(); |
74eed016 | 631 | #else |
f71b74bc | 632 | lockdep_assert_irqs_disabled(); |
74eed016 | 633 | #endif |
d3759e71 | 634 | account_hardirq_exit(current); |
bdb43806 | 635 | preempt_count_sub(HARDIRQ_OFFSET); |
1da177e4 LT |
636 | if (!in_interrupt() && local_softirq_pending()) |
637 | invoke_softirq(); | |
79bf2bb3 | 638 | |
67826eae | 639 | tick_irq_exit(); |
8a6bc478 TG |
640 | } |
641 | ||
59bc300b PZ |
642 | /** |
643 | * irq_exit_rcu() - Exit an interrupt context without updating RCU | |
644 | * | |
645 | * Also processes softirqs if needed and possible. | |
646 | */ | |
647 | void irq_exit_rcu(void) | |
648 | { | |
649 | __irq_exit_rcu(); | |
650 | /* must be last! */ | |
651 | lockdep_hardirq_exit(); | |
652 | } | |
653 | ||
8a6bc478 TG |
654 | /** |
655 | * irq_exit - Exit an interrupt context, update RCU and lockdep | |
656 | * | |
657 | * Also processes softirqs if needed and possible. | |
658 | */ | |
659 | void irq_exit(void) | |
660 | { | |
59bc300b | 661 | __irq_exit_rcu(); |
416eb33c | 662 | rcu_irq_exit(); |
2502ec37 TG |
663 | /* must be last! */ |
664 | lockdep_hardirq_exit(); | |
1da177e4 LT |
665 | } |
666 | ||
667 | /* | |
668 | * This function must run with irqs disabled! | |
669 | */ | |
7ad5b3a5 | 670 | inline void raise_softirq_irqoff(unsigned int nr) |
1da177e4 LT |
671 | { |
672 | __raise_softirq_irqoff(nr); | |
673 | ||
674 | /* | |
675 | * If we're in an interrupt or softirq, we're done | |
676 | * (this also catches softirq-disabled code). We will | |
677 | * actually run the softirq once we return from | |
678 | * the irq or softirq. | |
679 | * | |
680 | * Otherwise we wake up ksoftirqd to make sure we | |
681 | * schedule the softirq soon. | |
682 | */ | |
f02fc963 | 683 | if (!in_interrupt() && should_wake_ksoftirqd()) |
1da177e4 LT |
684 | wakeup_softirqd(); |
685 | } | |
686 | ||
7ad5b3a5 | 687 | void raise_softirq(unsigned int nr) |
1da177e4 LT |
688 | { |
689 | unsigned long flags; | |
690 | ||
691 | local_irq_save(flags); | |
692 | raise_softirq_irqoff(nr); | |
693 | local_irq_restore(flags); | |
694 | } | |
695 | ||
f069686e SR |
696 | void __raise_softirq_irqoff(unsigned int nr) |
697 | { | |
cdabce2e | 698 | lockdep_assert_irqs_disabled(); |
f069686e SR |
699 | trace_softirq_raise(nr); |
700 | or_softirq_pending(1UL << nr); | |
701 | } | |
702 | ||
962cf36c | 703 | void open_softirq(int nr, void (*action)(struct softirq_action *)) |
1da177e4 | 704 | { |
1da177e4 LT |
705 | softirq_vec[nr].action = action; |
706 | } | |
707 | ||
9ba5f005 PZ |
708 | /* |
709 | * Tasklets | |
710 | */ | |
ce85b4f2 | 711 | struct tasklet_head { |
48f20a9a OJ |
712 | struct tasklet_struct *head; |
713 | struct tasklet_struct **tail; | |
1da177e4 LT |
714 | }; |
715 | ||
4620b49f VN |
716 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec); |
717 | static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec); | |
1da177e4 | 718 | |
6498ddad IM |
719 | static void __tasklet_schedule_common(struct tasklet_struct *t, |
720 | struct tasklet_head __percpu *headp, | |
721 | unsigned int softirq_nr) | |
1da177e4 | 722 | { |
6498ddad | 723 | struct tasklet_head *head; |
1da177e4 LT |
724 | unsigned long flags; |
725 | ||
726 | local_irq_save(flags); | |
6498ddad | 727 | head = this_cpu_ptr(headp); |
48f20a9a | 728 | t->next = NULL; |
6498ddad IM |
729 | *head->tail = t; |
730 | head->tail = &(t->next); | |
731 | raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
732 | local_irq_restore(flags); |
733 | } | |
6498ddad IM |
734 | |
735 | void __tasklet_schedule(struct tasklet_struct *t) | |
736 | { | |
737 | __tasklet_schedule_common(t, &tasklet_vec, | |
738 | TASKLET_SOFTIRQ); | |
739 | } | |
1da177e4 LT |
740 | EXPORT_SYMBOL(__tasklet_schedule); |
741 | ||
7ad5b3a5 | 742 | void __tasklet_hi_schedule(struct tasklet_struct *t) |
1da177e4 | 743 | { |
6498ddad IM |
744 | __tasklet_schedule_common(t, &tasklet_hi_vec, |
745 | HI_SOFTIRQ); | |
1da177e4 | 746 | } |
1da177e4 LT |
747 | EXPORT_SYMBOL(__tasklet_hi_schedule); |
748 | ||
697d8c63 | 749 | static bool tasklet_clear_sched(struct tasklet_struct *t) |
6b2c339d | 750 | { |
697d8c63 PZ |
751 | if (test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) { |
752 | wake_up_var(&t->state); | |
6b2c339d | 753 | return true; |
697d8c63 | 754 | } |
6b2c339d DB |
755 | |
756 | WARN_ONCE(1, "tasklet SCHED state not set: %s %pS\n", | |
757 | t->use_callback ? "callback" : "func", | |
758 | t->use_callback ? (void *)t->callback : (void *)t->func); | |
759 | ||
760 | return false; | |
761 | } | |
762 | ||
82b691be IM |
763 | static void tasklet_action_common(struct softirq_action *a, |
764 | struct tasklet_head *tl_head, | |
765 | unsigned int softirq_nr) | |
1da177e4 LT |
766 | { |
767 | struct tasklet_struct *list; | |
768 | ||
769 | local_irq_disable(); | |
82b691be IM |
770 | list = tl_head->head; |
771 | tl_head->head = NULL; | |
772 | tl_head->tail = &tl_head->head; | |
1da177e4 LT |
773 | local_irq_enable(); |
774 | ||
775 | while (list) { | |
776 | struct tasklet_struct *t = list; | |
777 | ||
778 | list = list->next; | |
779 | ||
780 | if (tasklet_trylock(t)) { | |
781 | if (!atomic_read(&t->count)) { | |
697d8c63 | 782 | if (tasklet_clear_sched(t)) { |
6b2c339d DB |
783 | if (t->use_callback) |
784 | t->callback(t); | |
785 | else | |
786 | t->func(t->data); | |
787 | } | |
1da177e4 LT |
788 | tasklet_unlock(t); |
789 | continue; | |
790 | } | |
791 | tasklet_unlock(t); | |
792 | } | |
793 | ||
794 | local_irq_disable(); | |
48f20a9a | 795 | t->next = NULL; |
82b691be IM |
796 | *tl_head->tail = t; |
797 | tl_head->tail = &t->next; | |
798 | __raise_softirq_irqoff(softirq_nr); | |
1da177e4 LT |
799 | local_irq_enable(); |
800 | } | |
801 | } | |
802 | ||
82b691be | 803 | static __latent_entropy void tasklet_action(struct softirq_action *a) |
1da177e4 | 804 | { |
82b691be IM |
805 | tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ); |
806 | } | |
1da177e4 | 807 | |
82b691be IM |
808 | static __latent_entropy void tasklet_hi_action(struct softirq_action *a) |
809 | { | |
810 | tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ); | |
1da177e4 LT |
811 | } |
812 | ||
12cc923f RP |
813 | void tasklet_setup(struct tasklet_struct *t, |
814 | void (*callback)(struct tasklet_struct *)) | |
815 | { | |
816 | t->next = NULL; | |
817 | t->state = 0; | |
818 | atomic_set(&t->count, 0); | |
819 | t->callback = callback; | |
820 | t->use_callback = true; | |
821 | t->data = 0; | |
822 | } | |
823 | EXPORT_SYMBOL(tasklet_setup); | |
824 | ||
1da177e4 LT |
825 | void tasklet_init(struct tasklet_struct *t, |
826 | void (*func)(unsigned long), unsigned long data) | |
827 | { | |
828 | t->next = NULL; | |
829 | t->state = 0; | |
830 | atomic_set(&t->count, 0); | |
831 | t->func = func; | |
12cc923f | 832 | t->use_callback = false; |
1da177e4 LT |
833 | t->data = data; |
834 | } | |
1da177e4 LT |
835 | EXPORT_SYMBOL(tasklet_init); |
836 | ||
eb2dafbb TG |
837 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
838 | /* | |
839 | * Do not use in new code. Waiting for tasklets from atomic contexts is | |
840 | * error prone and should be avoided. | |
841 | */ | |
842 | void tasklet_unlock_spin_wait(struct tasklet_struct *t) | |
843 | { | |
844 | while (test_bit(TASKLET_STATE_RUN, &(t)->state)) { | |
845 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
846 | /* | |
847 | * Prevent a live lock when current preempted soft | |
848 | * interrupt processing or prevents ksoftirqd from | |
849 | * running. If the tasklet runs on a different CPU | |
850 | * then this has no effect other than doing the BH | |
851 | * disable/enable dance for nothing. | |
852 | */ | |
853 | local_bh_disable(); | |
854 | local_bh_enable(); | |
855 | } else { | |
856 | cpu_relax(); | |
857 | } | |
858 | } | |
859 | } | |
860 | EXPORT_SYMBOL(tasklet_unlock_spin_wait); | |
861 | #endif | |
862 | ||
1da177e4 LT |
863 | void tasklet_kill(struct tasklet_struct *t) |
864 | { | |
865 | if (in_interrupt()) | |
40322764 | 866 | pr_notice("Attempt to kill tasklet from interrupt\n"); |
1da177e4 | 867 | |
697d8c63 PZ |
868 | while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) |
869 | wait_var_event(&t->state, !test_bit(TASKLET_STATE_SCHED, &t->state)); | |
870 | ||
1da177e4 | 871 | tasklet_unlock_wait(t); |
697d8c63 | 872 | tasklet_clear_sched(t); |
1da177e4 | 873 | } |
1da177e4 LT |
874 | EXPORT_SYMBOL(tasklet_kill); |
875 | ||
eb2dafbb | 876 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
da044747 PZ |
877 | void tasklet_unlock(struct tasklet_struct *t) |
878 | { | |
879 | smp_mb__before_atomic(); | |
880 | clear_bit(TASKLET_STATE_RUN, &t->state); | |
881 | smp_mb__after_atomic(); | |
882 | wake_up_var(&t->state); | |
883 | } | |
884 | EXPORT_SYMBOL_GPL(tasklet_unlock); | |
885 | ||
886 | void tasklet_unlock_wait(struct tasklet_struct *t) | |
887 | { | |
888 | wait_var_event(&t->state, !test_bit(TASKLET_STATE_RUN, &t->state)); | |
889 | } | |
890 | EXPORT_SYMBOL_GPL(tasklet_unlock_wait); | |
891 | #endif | |
892 | ||
1da177e4 LT |
893 | void __init softirq_init(void) |
894 | { | |
48f20a9a OJ |
895 | int cpu; |
896 | ||
897 | for_each_possible_cpu(cpu) { | |
898 | per_cpu(tasklet_vec, cpu).tail = | |
899 | &per_cpu(tasklet_vec, cpu).head; | |
900 | per_cpu(tasklet_hi_vec, cpu).tail = | |
901 | &per_cpu(tasklet_hi_vec, cpu).head; | |
902 | } | |
903 | ||
962cf36c CM |
904 | open_softirq(TASKLET_SOFTIRQ, tasklet_action); |
905 | open_softirq(HI_SOFTIRQ, tasklet_hi_action); | |
1da177e4 LT |
906 | } |
907 | ||
3e339b5d | 908 | static int ksoftirqd_should_run(unsigned int cpu) |
1da177e4 | 909 | { |
3e339b5d TG |
910 | return local_softirq_pending(); |
911 | } | |
1da177e4 | 912 | |
3e339b5d TG |
913 | static void run_ksoftirqd(unsigned int cpu) |
914 | { | |
f02fc963 | 915 | ksoftirqd_run_begin(); |
3e339b5d | 916 | if (local_softirq_pending()) { |
0bed698a FW |
917 | /* |
918 | * We can safely run softirq on inline stack, as we are not deep | |
919 | * in the task stack here. | |
920 | */ | |
3e339b5d | 921 | __do_softirq(); |
f02fc963 | 922 | ksoftirqd_run_end(); |
edf22f4c | 923 | cond_resched(); |
3e339b5d | 924 | return; |
1da177e4 | 925 | } |
f02fc963 | 926 | ksoftirqd_run_end(); |
1da177e4 LT |
927 | } |
928 | ||
929 | #ifdef CONFIG_HOTPLUG_CPU | |
c4544dbc | 930 | static int takeover_tasklets(unsigned int cpu) |
1da177e4 | 931 | { |
1da177e4 LT |
932 | /* CPU is dead, so no lock needed. */ |
933 | local_irq_disable(); | |
934 | ||
935 | /* Find end, append list for that CPU. */ | |
e5e41723 | 936 | if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) { |
909ea964 | 937 | *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head; |
8afecaa6 | 938 | __this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail); |
e5e41723 CB |
939 | per_cpu(tasklet_vec, cpu).head = NULL; |
940 | per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head; | |
941 | } | |
1da177e4 LT |
942 | raise_softirq_irqoff(TASKLET_SOFTIRQ); |
943 | ||
e5e41723 | 944 | if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) { |
909ea964 CL |
945 | *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head; |
946 | __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail); | |
e5e41723 CB |
947 | per_cpu(tasklet_hi_vec, cpu).head = NULL; |
948 | per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head; | |
949 | } | |
1da177e4 LT |
950 | raise_softirq_irqoff(HI_SOFTIRQ); |
951 | ||
952 | local_irq_enable(); | |
c4544dbc | 953 | return 0; |
1da177e4 | 954 | } |
c4544dbc SAS |
955 | #else |
956 | #define takeover_tasklets NULL | |
1da177e4 LT |
957 | #endif /* CONFIG_HOTPLUG_CPU */ |
958 | ||
3e339b5d TG |
959 | static struct smp_hotplug_thread softirq_threads = { |
960 | .store = &ksoftirqd, | |
961 | .thread_should_run = ksoftirqd_should_run, | |
962 | .thread_fn = run_ksoftirqd, | |
963 | .thread_comm = "ksoftirqd/%u", | |
964 | }; | |
965 | ||
7babe8db | 966 | static __init int spawn_ksoftirqd(void) |
1da177e4 | 967 | { |
c4544dbc SAS |
968 | cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL, |
969 | takeover_tasklets); | |
3e339b5d TG |
970 | BUG_ON(smpboot_register_percpu_thread(&softirq_threads)); |
971 | ||
1da177e4 LT |
972 | return 0; |
973 | } | |
7babe8db | 974 | early_initcall(spawn_ksoftirqd); |
78eef01b | 975 | |
43a25632 YL |
976 | /* |
977 | * [ These __weak aliases are kept in a separate compilation unit, so that | |
978 | * GCC does not inline them incorrectly. ] | |
979 | */ | |
980 | ||
981 | int __init __weak early_irq_init(void) | |
982 | { | |
983 | return 0; | |
984 | } | |
985 | ||
4a046d17 YL |
986 | int __init __weak arch_probe_nr_irqs(void) |
987 | { | |
b683de2b | 988 | return NR_IRQS_LEGACY; |
4a046d17 YL |
989 | } |
990 | ||
43a25632 YL |
991 | int __init __weak arch_early_irq_init(void) |
992 | { | |
993 | return 0; | |
994 | } | |
62a08ae2 TG |
995 | |
996 | unsigned int __weak arch_dynirq_lower_bound(unsigned int from) | |
997 | { | |
998 | return from; | |
999 | } |