xref: /freebsd/sys/kern/kern_intr.c (revision f6a3b357e9be4c6423c85eff9a847163a0d307c8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_ddb.h"
33 #include "opt_kstack_usage_prof.h"
34 
35 #include <sys/param.h>
36 #include <sys/bus.h>
37 #include <sys/conf.h>
38 #include <sys/cpuset.h>
39 #include <sys/rtprio.h>
40 #include <sys/systm.h>
41 #include <sys/interrupt.h>
42 #include <sys/kernel.h>
43 #include <sys/kthread.h>
44 #include <sys/ktr.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mutex.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/random.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
54 #include <sys/smp.h>
55 #include <sys/sysctl.h>
56 #include <sys/syslog.h>
57 #include <sys/unistd.h>
58 #include <sys/vmmeter.h>
59 #include <machine/atomic.h>
60 #include <machine/cpu.h>
61 #include <machine/md_var.h>
62 #include <machine/stdarg.h>
63 #ifdef DDB
64 #include <ddb/ddb.h>
65 #include <ddb/db_sym.h>
66 #endif
67 
68 /*
69  * Describe an interrupt thread.  There is one of these per interrupt event.
70  */
71 struct intr_thread {
72 	struct intr_event *it_event;
73 	struct thread *it_thread;	/* Kernel thread. */
74 	int	it_flags;		/* (j) IT_* flags. */
75 	int	it_need;		/* Needs service. */
76 };
77 
78 /* Interrupt thread flags kept in it_flags */
79 #define	IT_DEAD		0x000001	/* Thread is waiting to exit. */
80 #define	IT_WAIT		0x000002	/* Thread is waiting for completion. */
81 
82 struct	intr_entropy {
83 	struct	thread *td;
84 	uintptr_t event;
85 };
86 
87 struct	intr_event *clk_intr_event;
88 struct	intr_event *tty_intr_event;
89 void	*vm_ih;
90 struct proc *intrproc;
91 
92 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
93 
94 static int intr_storm_threshold = 0;
95 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RWTUN,
96     &intr_storm_threshold, 0,
97     "Number of consecutive interrupts before storm protection is enabled");
98 static TAILQ_HEAD(, intr_event) event_list =
99     TAILQ_HEAD_INITIALIZER(event_list);
100 static struct mtx event_lock;
101 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
102 
103 static void	intr_event_update(struct intr_event *ie);
104 static int	intr_event_schedule_thread(struct intr_event *ie);
105 static struct intr_thread *ithread_create(const char *name);
106 static void	ithread_destroy(struct intr_thread *ithread);
107 static void	ithread_execute_handlers(struct proc *p,
108 		    struct intr_event *ie);
109 static void	ithread_loop(void *);
110 static void	ithread_update(struct intr_thread *ithd);
111 static void	start_softintr(void *);
112 
113 /* Map an interrupt type to an ithread priority. */
114 u_char
115 intr_priority(enum intr_type flags)
116 {
117 	u_char pri;
118 
119 	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
120 	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
121 	switch (flags) {
122 	case INTR_TYPE_TTY:
123 		pri = PI_TTY;
124 		break;
125 	case INTR_TYPE_BIO:
126 		pri = PI_DISK;
127 		break;
128 	case INTR_TYPE_NET:
129 		pri = PI_NET;
130 		break;
131 	case INTR_TYPE_CAM:
132 		pri = PI_DISK;
133 		break;
134 	case INTR_TYPE_AV:
135 		pri = PI_AV;
136 		break;
137 	case INTR_TYPE_CLK:
138 		pri = PI_REALTIME;
139 		break;
140 	case INTR_TYPE_MISC:
141 		pri = PI_DULL;          /* don't care */
142 		break;
143 	default:
144 		/* We didn't specify an interrupt level. */
145 		panic("intr_priority: no interrupt type in flags");
146 	}
147 
148 	return pri;
149 }
150 
151 /*
152  * Update an ithread based on the associated intr_event.
153  */
154 static void
155 ithread_update(struct intr_thread *ithd)
156 {
157 	struct intr_event *ie;
158 	struct thread *td;
159 	u_char pri;
160 
161 	ie = ithd->it_event;
162 	td = ithd->it_thread;
163 	mtx_assert(&ie->ie_lock, MA_OWNED);
164 
165 	/* Determine the overall priority of this event. */
166 	if (CK_SLIST_EMPTY(&ie->ie_handlers))
167 		pri = PRI_MAX_ITHD;
168 	else
169 		pri = CK_SLIST_FIRST(&ie->ie_handlers)->ih_pri;
170 
171 	/* Update name and priority. */
172 	strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
173 #ifdef KTR
174 	sched_clear_tdname(td);
175 #endif
176 	thread_lock(td);
177 	sched_prio(td, pri);
178 	thread_unlock(td);
179 }
180 
181 /*
182  * Regenerate the full name of an interrupt event and update its priority.
183  */
184 static void
185 intr_event_update(struct intr_event *ie)
186 {
187 	struct intr_handler *ih;
188 	char *last;
189 	int missed, space;
190 
191 	/* Start off with no entropy and just the name of the event. */
192 	mtx_assert(&ie->ie_lock, MA_OWNED);
193 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
194 	ie->ie_flags &= ~IE_ENTROPY;
195 	missed = 0;
196 	space = 1;
197 
198 	/* Run through all the handlers updating values. */
199 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
200 		if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
201 		    sizeof(ie->ie_fullname)) {
202 			strcat(ie->ie_fullname, " ");
203 			strcat(ie->ie_fullname, ih->ih_name);
204 			space = 0;
205 		} else
206 			missed++;
207 		if (ih->ih_flags & IH_ENTROPY)
208 			ie->ie_flags |= IE_ENTROPY;
209 	}
210 
211 	/*
212 	 * If there is only one handler and its name is too long, just copy in
213 	 * as much of the end of the name (includes the unit number) as will
214 	 * fit.  Otherwise, we have multiple handlers and not all of the names
215 	 * will fit.  Add +'s to indicate missing names.  If we run out of room
216 	 * and still have +'s to add, change the last character from a + to a *.
217 	 */
218 	if (missed == 1 && space == 1) {
219 		ih = CK_SLIST_FIRST(&ie->ie_handlers);
220 		missed = strlen(ie->ie_fullname) + strlen(ih->ih_name) + 2 -
221 		    sizeof(ie->ie_fullname);
222 		strcat(ie->ie_fullname, (missed == 0) ? " " : "-");
223 		strcat(ie->ie_fullname, &ih->ih_name[missed]);
224 		missed = 0;
225 	}
226 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
227 	while (missed-- > 0) {
228 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
229 			if (*last == '+') {
230 				*last = '*';
231 				break;
232 			} else
233 				*last = '+';
234 		} else if (space) {
235 			strcat(ie->ie_fullname, " +");
236 			space = 0;
237 		} else
238 			strcat(ie->ie_fullname, "+");
239 	}
240 
241 	/*
242 	 * If this event has an ithread, update it's priority and
243 	 * name.
244 	 */
245 	if (ie->ie_thread != NULL)
246 		ithread_update(ie->ie_thread);
247 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
248 }
249 
250 int
251 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
252     void (*pre_ithread)(void *), void (*post_ithread)(void *),
253     void (*post_filter)(void *), int (*assign_cpu)(void *, int),
254     const char *fmt, ...)
255 {
256 	struct intr_event *ie;
257 	va_list ap;
258 
259 	/* The only valid flag during creation is IE_SOFT. */
260 	if ((flags & ~IE_SOFT) != 0)
261 		return (EINVAL);
262 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
263 	ie->ie_source = source;
264 	ie->ie_pre_ithread = pre_ithread;
265 	ie->ie_post_ithread = post_ithread;
266 	ie->ie_post_filter = post_filter;
267 	ie->ie_assign_cpu = assign_cpu;
268 	ie->ie_flags = flags;
269 	ie->ie_irq = irq;
270 	ie->ie_cpu = NOCPU;
271 	CK_SLIST_INIT(&ie->ie_handlers);
272 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
273 
274 	va_start(ap, fmt);
275 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
276 	va_end(ap);
277 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
278 	mtx_lock(&event_lock);
279 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
280 	mtx_unlock(&event_lock);
281 	if (event != NULL)
282 		*event = ie;
283 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
284 	return (0);
285 }
286 
287 /*
288  * Bind an interrupt event to the specified CPU.  Note that not all
289  * platforms support binding an interrupt to a CPU.  For those
290  * platforms this request will fail.  Using a cpu id of NOCPU unbinds
291  * the interrupt event.
292  */
293 static int
294 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
295 {
296 	lwpid_t id;
297 	int error;
298 
299 	/* Need a CPU to bind to. */
300 	if (cpu != NOCPU && CPU_ABSENT(cpu))
301 		return (EINVAL);
302 
303 	if (ie->ie_assign_cpu == NULL)
304 		return (EOPNOTSUPP);
305 
306 	error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
307 	if (error)
308 		return (error);
309 
310 	/*
311 	 * If we have any ithreads try to set their mask first to verify
312 	 * permissions, etc.
313 	 */
314 	if (bindithread) {
315 		mtx_lock(&ie->ie_lock);
316 		if (ie->ie_thread != NULL) {
317 			id = ie->ie_thread->it_thread->td_tid;
318 			mtx_unlock(&ie->ie_lock);
319 			error = cpuset_setithread(id, cpu);
320 			if (error)
321 				return (error);
322 		} else
323 			mtx_unlock(&ie->ie_lock);
324 	}
325 	if (bindirq)
326 		error = ie->ie_assign_cpu(ie->ie_source, cpu);
327 	if (error) {
328 		if (bindithread) {
329 			mtx_lock(&ie->ie_lock);
330 			if (ie->ie_thread != NULL) {
331 				cpu = ie->ie_cpu;
332 				id = ie->ie_thread->it_thread->td_tid;
333 				mtx_unlock(&ie->ie_lock);
334 				(void)cpuset_setithread(id, cpu);
335 			} else
336 				mtx_unlock(&ie->ie_lock);
337 		}
338 		return (error);
339 	}
340 
341 	if (bindirq) {
342 		mtx_lock(&ie->ie_lock);
343 		ie->ie_cpu = cpu;
344 		mtx_unlock(&ie->ie_lock);
345 	}
346 
347 	return (error);
348 }
349 
350 /*
351  * Bind an interrupt event to the specified CPU.  For supported platforms, any
352  * associated ithreads as well as the primary interrupt context will be bound
353  * to the specificed CPU.
354  */
355 int
356 intr_event_bind(struct intr_event *ie, int cpu)
357 {
358 
359 	return (_intr_event_bind(ie, cpu, true, true));
360 }
361 
362 /*
363  * Bind an interrupt event to the specified CPU, but do not bind associated
364  * ithreads.
365  */
366 int
367 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
368 {
369 
370 	return (_intr_event_bind(ie, cpu, true, false));
371 }
372 
373 /*
374  * Bind an interrupt event's ithread to the specified CPU.
375  */
376 int
377 intr_event_bind_ithread(struct intr_event *ie, int cpu)
378 {
379 
380 	return (_intr_event_bind(ie, cpu, false, true));
381 }
382 
383 /*
384  * Bind an interrupt event's ithread to the specified cpuset.
385  */
386 int
387 intr_event_bind_ithread_cpuset(struct intr_event *ie, cpuset_t *cs)
388 {
389 	lwpid_t id;
390 
391 	mtx_lock(&ie->ie_lock);
392 	if (ie->ie_thread != NULL) {
393 		id = ie->ie_thread->it_thread->td_tid;
394 		mtx_unlock(&ie->ie_lock);
395 		return (cpuset_setthread(id, cs));
396 	} else {
397 		mtx_unlock(&ie->ie_lock);
398 	}
399 	return (ENODEV);
400 }
401 
402 static struct intr_event *
403 intr_lookup(int irq)
404 {
405 	struct intr_event *ie;
406 
407 	mtx_lock(&event_lock);
408 	TAILQ_FOREACH(ie, &event_list, ie_list)
409 		if (ie->ie_irq == irq &&
410 		    (ie->ie_flags & IE_SOFT) == 0 &&
411 		    CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
412 			break;
413 	mtx_unlock(&event_lock);
414 	return (ie);
415 }
416 
417 int
418 intr_setaffinity(int irq, int mode, void *m)
419 {
420 	struct intr_event *ie;
421 	cpuset_t *mask;
422 	int cpu, n;
423 
424 	mask = m;
425 	cpu = NOCPU;
426 	/*
427 	 * If we're setting all cpus we can unbind.  Otherwise make sure
428 	 * only one cpu is in the set.
429 	 */
430 	if (CPU_CMP(cpuset_root, mask)) {
431 		for (n = 0; n < CPU_SETSIZE; n++) {
432 			if (!CPU_ISSET(n, mask))
433 				continue;
434 			if (cpu != NOCPU)
435 				return (EINVAL);
436 			cpu = n;
437 		}
438 	}
439 	ie = intr_lookup(irq);
440 	if (ie == NULL)
441 		return (ESRCH);
442 	switch (mode) {
443 	case CPU_WHICH_IRQ:
444 		return (intr_event_bind(ie, cpu));
445 	case CPU_WHICH_INTRHANDLER:
446 		return (intr_event_bind_irqonly(ie, cpu));
447 	case CPU_WHICH_ITHREAD:
448 		return (intr_event_bind_ithread(ie, cpu));
449 	default:
450 		return (EINVAL);
451 	}
452 }
453 
454 int
455 intr_getaffinity(int irq, int mode, void *m)
456 {
457 	struct intr_event *ie;
458 	struct thread *td;
459 	struct proc *p;
460 	cpuset_t *mask;
461 	lwpid_t id;
462 	int error;
463 
464 	mask = m;
465 	ie = intr_lookup(irq);
466 	if (ie == NULL)
467 		return (ESRCH);
468 
469 	error = 0;
470 	CPU_ZERO(mask);
471 	switch (mode) {
472 	case CPU_WHICH_IRQ:
473 	case CPU_WHICH_INTRHANDLER:
474 		mtx_lock(&ie->ie_lock);
475 		if (ie->ie_cpu == NOCPU)
476 			CPU_COPY(cpuset_root, mask);
477 		else
478 			CPU_SET(ie->ie_cpu, mask);
479 		mtx_unlock(&ie->ie_lock);
480 		break;
481 	case CPU_WHICH_ITHREAD:
482 		mtx_lock(&ie->ie_lock);
483 		if (ie->ie_thread == NULL) {
484 			mtx_unlock(&ie->ie_lock);
485 			CPU_COPY(cpuset_root, mask);
486 		} else {
487 			id = ie->ie_thread->it_thread->td_tid;
488 			mtx_unlock(&ie->ie_lock);
489 			error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
490 			if (error != 0)
491 				return (error);
492 			CPU_COPY(&td->td_cpuset->cs_mask, mask);
493 			PROC_UNLOCK(p);
494 		}
495 	default:
496 		return (EINVAL);
497 	}
498 	return (0);
499 }
500 
501 int
502 intr_event_destroy(struct intr_event *ie)
503 {
504 
505 	mtx_lock(&event_lock);
506 	mtx_lock(&ie->ie_lock);
507 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
508 		mtx_unlock(&ie->ie_lock);
509 		mtx_unlock(&event_lock);
510 		return (EBUSY);
511 	}
512 	TAILQ_REMOVE(&event_list, ie, ie_list);
513 #ifndef notyet
514 	if (ie->ie_thread != NULL) {
515 		ithread_destroy(ie->ie_thread);
516 		ie->ie_thread = NULL;
517 	}
518 #endif
519 	mtx_unlock(&ie->ie_lock);
520 	mtx_unlock(&event_lock);
521 	mtx_destroy(&ie->ie_lock);
522 	free(ie, M_ITHREAD);
523 	return (0);
524 }
525 
526 static struct intr_thread *
527 ithread_create(const char *name)
528 {
529 	struct intr_thread *ithd;
530 	struct thread *td;
531 	int error;
532 
533 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
534 
535 	error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
536 		    &td, RFSTOPPED | RFHIGHPID,
537 		    0, "intr", "%s", name);
538 	if (error)
539 		panic("kproc_create() failed with %d", error);
540 	thread_lock(td);
541 	sched_class(td, PRI_ITHD);
542 	TD_SET_IWAIT(td);
543 	thread_unlock(td);
544 	td->td_pflags |= TDP_ITHREAD;
545 	ithd->it_thread = td;
546 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
547 	return (ithd);
548 }
549 
550 static void
551 ithread_destroy(struct intr_thread *ithread)
552 {
553 	struct thread *td;
554 
555 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
556 	td = ithread->it_thread;
557 	thread_lock(td);
558 	ithread->it_flags |= IT_DEAD;
559 	if (TD_AWAITING_INTR(td)) {
560 		TD_CLR_IWAIT(td);
561 		sched_add(td, SRQ_INTR);
562 	}
563 	thread_unlock(td);
564 }
565 
566 int
567 intr_event_add_handler(struct intr_event *ie, const char *name,
568     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
569     enum intr_type flags, void **cookiep)
570 {
571 	struct intr_handler *ih, *temp_ih;
572 	struct intr_handler **prevptr;
573 	struct intr_thread *it;
574 
575 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
576 		return (EINVAL);
577 
578 	/* Allocate and populate an interrupt handler structure. */
579 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
580 	ih->ih_filter = filter;
581 	ih->ih_handler = handler;
582 	ih->ih_argument = arg;
583 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
584 	ih->ih_event = ie;
585 	ih->ih_pri = pri;
586 	if (flags & INTR_EXCL)
587 		ih->ih_flags = IH_EXCLUSIVE;
588 	if (flags & INTR_MPSAFE)
589 		ih->ih_flags |= IH_MPSAFE;
590 	if (flags & INTR_ENTROPY)
591 		ih->ih_flags |= IH_ENTROPY;
592 
593 	/* We can only have one exclusive handler in a event. */
594 	mtx_lock(&ie->ie_lock);
595 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
596 		if ((flags & INTR_EXCL) ||
597 		    (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
598 			mtx_unlock(&ie->ie_lock);
599 			free(ih, M_ITHREAD);
600 			return (EINVAL);
601 		}
602 	}
603 
604 	/* Create a thread if we need one. */
605 	while (ie->ie_thread == NULL && handler != NULL) {
606 		if (ie->ie_flags & IE_ADDING_THREAD)
607 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
608 		else {
609 			ie->ie_flags |= IE_ADDING_THREAD;
610 			mtx_unlock(&ie->ie_lock);
611 			it = ithread_create("intr: newborn");
612 			mtx_lock(&ie->ie_lock);
613 			ie->ie_flags &= ~IE_ADDING_THREAD;
614 			ie->ie_thread = it;
615 			it->it_event = ie;
616 			ithread_update(it);
617 			wakeup(ie);
618 		}
619 	}
620 
621 	/* Add the new handler to the event in priority order. */
622 	CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
623 		if (temp_ih->ih_pri > ih->ih_pri)
624 			break;
625 	}
626 	CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
627 
628 	intr_event_update(ie);
629 
630 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
631 	    ie->ie_name);
632 	mtx_unlock(&ie->ie_lock);
633 
634 	if (cookiep != NULL)
635 		*cookiep = ih;
636 	return (0);
637 }
638 
639 /*
640  * Append a description preceded by a ':' to the name of the specified
641  * interrupt handler.
642  */
643 int
644 intr_event_describe_handler(struct intr_event *ie, void *cookie,
645     const char *descr)
646 {
647 	struct intr_handler *ih;
648 	size_t space;
649 	char *start;
650 
651 	mtx_lock(&ie->ie_lock);
652 #ifdef INVARIANTS
653 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
654 		if (ih == cookie)
655 			break;
656 	}
657 	if (ih == NULL) {
658 		mtx_unlock(&ie->ie_lock);
659 		panic("handler %p not found in interrupt event %p", cookie, ie);
660 	}
661 #endif
662 	ih = cookie;
663 
664 	/*
665 	 * Look for an existing description by checking for an
666 	 * existing ":".  This assumes device names do not include
667 	 * colons.  If one is found, prepare to insert the new
668 	 * description at that point.  If one is not found, find the
669 	 * end of the name to use as the insertion point.
670 	 */
671 	start = strchr(ih->ih_name, ':');
672 	if (start == NULL)
673 		start = strchr(ih->ih_name, 0);
674 
675 	/*
676 	 * See if there is enough remaining room in the string for the
677 	 * description + ":".  The "- 1" leaves room for the trailing
678 	 * '\0'.  The "+ 1" accounts for the colon.
679 	 */
680 	space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
681 	if (strlen(descr) + 1 > space) {
682 		mtx_unlock(&ie->ie_lock);
683 		return (ENOSPC);
684 	}
685 
686 	/* Append a colon followed by the description. */
687 	*start = ':';
688 	strcpy(start + 1, descr);
689 	intr_event_update(ie);
690 	mtx_unlock(&ie->ie_lock);
691 	return (0);
692 }
693 
694 /*
695  * Return the ie_source field from the intr_event an intr_handler is
696  * associated with.
697  */
698 void *
699 intr_handler_source(void *cookie)
700 {
701 	struct intr_handler *ih;
702 	struct intr_event *ie;
703 
704 	ih = (struct intr_handler *)cookie;
705 	if (ih == NULL)
706 		return (NULL);
707 	ie = ih->ih_event;
708 	KASSERT(ie != NULL,
709 	    ("interrupt handler \"%s\" has a NULL interrupt event",
710 	    ih->ih_name));
711 	return (ie->ie_source);
712 }
713 
714 /*
715  * If intr_event_handle() is running in the ISR context at the time of the call,
716  * then wait for it to complete.
717  */
718 static void
719 intr_event_barrier(struct intr_event *ie)
720 {
721 	int phase;
722 
723 	mtx_assert(&ie->ie_lock, MA_OWNED);
724 	phase = ie->ie_phase;
725 
726 	/*
727 	 * Switch phase to direct future interrupts to the other active counter.
728 	 * Make sure that any preceding stores are visible before the switch.
729 	 */
730 	KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
731 	atomic_store_rel_int(&ie->ie_phase, !phase);
732 
733 	/*
734 	 * This code cooperates with wait-free iteration of ie_handlers
735 	 * in intr_event_handle.
736 	 * Make sure that the removal and the phase update are not reordered
737 	 * with the active count check.
738 	 * Note that no combination of acquire and release fences can provide
739 	 * that guarantee as Store->Load sequences can always be reordered.
740 	 */
741 	atomic_thread_fence_seq_cst();
742 
743 	/*
744 	 * Now wait on the inactive phase.
745 	 * The acquire fence is needed so that that all post-barrier accesses
746 	 * are after the check.
747 	 */
748 	while (ie->ie_active[phase] > 0)
749 		cpu_spinwait();
750 	atomic_thread_fence_acq();
751 }
752 
753 static void
754 intr_handler_barrier(struct intr_handler *handler)
755 {
756 	struct intr_event *ie;
757 
758 	ie = handler->ih_event;
759 	mtx_assert(&ie->ie_lock, MA_OWNED);
760 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
761 	    ("update for a removed handler"));
762 
763 	if (ie->ie_thread == NULL) {
764 		intr_event_barrier(ie);
765 		return;
766 	}
767 	if ((handler->ih_flags & IH_CHANGED) == 0) {
768 		handler->ih_flags |= IH_CHANGED;
769 		intr_event_schedule_thread(ie);
770 	}
771 	while ((handler->ih_flags & IH_CHANGED) != 0)
772 		msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
773 }
774 
775 /*
776  * Sleep until an ithread finishes executing an interrupt handler.
777  *
778  * XXX Doesn't currently handle interrupt filters or fast interrupt
779  * handlers.  This is intended for compatibility with linux drivers
780  * only.  Do not use in BSD code.
781  */
782 void
783 _intr_drain(int irq)
784 {
785 	struct intr_event *ie;
786 	struct intr_thread *ithd;
787 	struct thread *td;
788 
789 	ie = intr_lookup(irq);
790 	if (ie == NULL)
791 		return;
792 	if (ie->ie_thread == NULL)
793 		return;
794 	ithd = ie->ie_thread;
795 	td = ithd->it_thread;
796 	/*
797 	 * We set the flag and wait for it to be cleared to avoid
798 	 * long delays with potentially busy interrupt handlers
799 	 * were we to only sample TD_AWAITING_INTR() every tick.
800 	 */
801 	thread_lock(td);
802 	if (!TD_AWAITING_INTR(td)) {
803 		ithd->it_flags |= IT_WAIT;
804 		while (ithd->it_flags & IT_WAIT) {
805 			thread_unlock(td);
806 			pause("idrain", 1);
807 			thread_lock(td);
808 		}
809 	}
810 	thread_unlock(td);
811 	return;
812 }
813 
814 int
815 intr_event_remove_handler(void *cookie)
816 {
817 	struct intr_handler *handler = (struct intr_handler *)cookie;
818 	struct intr_event *ie;
819 	struct intr_handler *ih;
820 	struct intr_handler **prevptr;
821 #ifdef notyet
822 	int dead;
823 #endif
824 
825 	if (handler == NULL)
826 		return (EINVAL);
827 	ie = handler->ih_event;
828 	KASSERT(ie != NULL,
829 	    ("interrupt handler \"%s\" has a NULL interrupt event",
830 	    handler->ih_name));
831 
832 	mtx_lock(&ie->ie_lock);
833 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
834 	    ie->ie_name);
835 	CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
836 		if (ih == handler)
837 			break;
838 	}
839 	if (ih == NULL) {
840 		panic("interrupt handler \"%s\" not found in "
841 		    "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
842 	}
843 
844 	/*
845 	 * If there is no ithread, then directly remove the handler.  Note that
846 	 * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
847 	 * care needs to be taken to keep ie_handlers consistent and to free
848 	 * the removed handler only when ie_handlers is quiescent.
849 	 */
850 	if (ie->ie_thread == NULL) {
851 		CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
852 		intr_event_barrier(ie);
853 		intr_event_update(ie);
854 		mtx_unlock(&ie->ie_lock);
855 		free(handler, M_ITHREAD);
856 		return (0);
857 	}
858 
859 	/*
860 	 * Let the interrupt thread do the job.
861 	 * The interrupt source is disabled when the interrupt thread is
862 	 * running, so it does not have to worry about interaction with
863 	 * intr_event_handle().
864 	 */
865 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
866 	    ("duplicate handle remove"));
867 	handler->ih_flags |= IH_DEAD;
868 	intr_event_schedule_thread(ie);
869 	while (handler->ih_flags & IH_DEAD)
870 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
871 	intr_event_update(ie);
872 
873 #ifdef notyet
874 	/*
875 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
876 	 * this could lead to races of stale data when servicing an
877 	 * interrupt.
878 	 */
879 	dead = 1;
880 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
881 		if (ih->ih_handler != NULL) {
882 			dead = 0;
883 			break;
884 		}
885 	}
886 	if (dead) {
887 		ithread_destroy(ie->ie_thread);
888 		ie->ie_thread = NULL;
889 	}
890 #endif
891 	mtx_unlock(&ie->ie_lock);
892 	free(handler, M_ITHREAD);
893 	return (0);
894 }
895 
896 int
897 intr_event_suspend_handler(void *cookie)
898 {
899 	struct intr_handler *handler = (struct intr_handler *)cookie;
900 	struct intr_event *ie;
901 
902 	if (handler == NULL)
903 		return (EINVAL);
904 	ie = handler->ih_event;
905 	KASSERT(ie != NULL,
906 	    ("interrupt handler \"%s\" has a NULL interrupt event",
907 	    handler->ih_name));
908 	mtx_lock(&ie->ie_lock);
909 	handler->ih_flags |= IH_SUSP;
910 	intr_handler_barrier(handler);
911 	mtx_unlock(&ie->ie_lock);
912 	return (0);
913 }
914 
915 int
916 intr_event_resume_handler(void *cookie)
917 {
918 	struct intr_handler *handler = (struct intr_handler *)cookie;
919 	struct intr_event *ie;
920 
921 	if (handler == NULL)
922 		return (EINVAL);
923 	ie = handler->ih_event;
924 	KASSERT(ie != NULL,
925 	    ("interrupt handler \"%s\" has a NULL interrupt event",
926 	    handler->ih_name));
927 
928 	/*
929 	 * intr_handler_barrier() acts not only as a barrier,
930 	 * it also allows to check for any pending interrupts.
931 	 */
932 	mtx_lock(&ie->ie_lock);
933 	handler->ih_flags &= ~IH_SUSP;
934 	intr_handler_barrier(handler);
935 	mtx_unlock(&ie->ie_lock);
936 	return (0);
937 }
938 
939 static int
940 intr_event_schedule_thread(struct intr_event *ie)
941 {
942 	struct intr_entropy entropy;
943 	struct intr_thread *it;
944 	struct thread *td;
945 	struct thread *ctd;
946 
947 	/*
948 	 * If no ithread or no handlers, then we have a stray interrupt.
949 	 */
950 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
951 	    ie->ie_thread == NULL)
952 		return (EINVAL);
953 
954 	ctd = curthread;
955 	it = ie->ie_thread;
956 	td = it->it_thread;
957 
958 	/*
959 	 * If any of the handlers for this ithread claim to be good
960 	 * sources of entropy, then gather some.
961 	 */
962 	if (ie->ie_flags & IE_ENTROPY) {
963 		entropy.event = (uintptr_t)ie;
964 		entropy.td = ctd;
965 		random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
966 	}
967 
968 	KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
969 
970 	/*
971 	 * Set it_need to tell the thread to keep running if it is already
972 	 * running.  Then, lock the thread and see if we actually need to
973 	 * put it on the runqueue.
974 	 *
975 	 * Use store_rel to arrange that the store to ih_need in
976 	 * swi_sched() is before the store to it_need and prepare for
977 	 * transfer of this order to loads in the ithread.
978 	 */
979 	atomic_store_rel_int(&it->it_need, 1);
980 	thread_lock(td);
981 	if (TD_AWAITING_INTR(td)) {
982 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
983 		    td->td_name);
984 		TD_CLR_IWAIT(td);
985 		sched_add(td, SRQ_INTR);
986 	} else {
987 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
988 		    __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
989 	}
990 	thread_unlock(td);
991 
992 	return (0);
993 }
994 
995 /*
996  * Allow interrupt event binding for software interrupt handlers -- a no-op,
997  * since interrupts are generated in software rather than being directed by
998  * a PIC.
999  */
1000 static int
1001 swi_assign_cpu(void *arg, int cpu)
1002 {
1003 
1004 	return (0);
1005 }
1006 
1007 /*
1008  * Add a software interrupt handler to a specified event.  If a given event
1009  * is not specified, then a new event is created.
1010  */
1011 int
1012 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
1013 	    void *arg, int pri, enum intr_type flags, void **cookiep)
1014 {
1015 	struct intr_event *ie;
1016 	int error;
1017 
1018 	if (flags & INTR_ENTROPY)
1019 		return (EINVAL);
1020 
1021 	ie = (eventp != NULL) ? *eventp : NULL;
1022 
1023 	if (ie != NULL) {
1024 		if (!(ie->ie_flags & IE_SOFT))
1025 			return (EINVAL);
1026 	} else {
1027 		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1028 		    NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1029 		if (error)
1030 			return (error);
1031 		if (eventp != NULL)
1032 			*eventp = ie;
1033 	}
1034 	error = intr_event_add_handler(ie, name, NULL, handler, arg,
1035 	    PI_SWI(pri), flags, cookiep);
1036 	return (error);
1037 }
1038 
1039 /*
1040  * Schedule a software interrupt thread.
1041  */
1042 void
1043 swi_sched(void *cookie, int flags)
1044 {
1045 	struct intr_handler *ih = (struct intr_handler *)cookie;
1046 	struct intr_event *ie = ih->ih_event;
1047 	struct intr_entropy entropy;
1048 	int error __unused;
1049 
1050 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1051 	    ih->ih_need);
1052 
1053 	entropy.event = (uintptr_t)ih;
1054 	entropy.td = curthread;
1055 	random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1056 
1057 	/*
1058 	 * Set ih_need for this handler so that if the ithread is already
1059 	 * running it will execute this handler on the next pass.  Otherwise,
1060 	 * it will execute it the next time it runs.
1061 	 */
1062 	ih->ih_need = 1;
1063 
1064 	if (!(flags & SWI_DELAY)) {
1065 		VM_CNT_INC(v_soft);
1066 		error = intr_event_schedule_thread(ie);
1067 		KASSERT(error == 0, ("stray software interrupt"));
1068 	}
1069 }
1070 
1071 /*
1072  * Remove a software interrupt handler.  Currently this code does not
1073  * remove the associated interrupt event if it becomes empty.  Calling code
1074  * may do so manually via intr_event_destroy(), but that's not really
1075  * an optimal interface.
1076  */
1077 int
1078 swi_remove(void *cookie)
1079 {
1080 
1081 	return (intr_event_remove_handler(cookie));
1082 }
1083 
1084 static void
1085 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1086 {
1087 	struct intr_handler *ih, *ihn, *ihp;
1088 
1089 	ihp = NULL;
1090 	CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1091 		/*
1092 		 * If this handler is marked for death, remove it from
1093 		 * the list of handlers and wake up the sleeper.
1094 		 */
1095 		if (ih->ih_flags & IH_DEAD) {
1096 			mtx_lock(&ie->ie_lock);
1097 			if (ihp == NULL)
1098 				CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1099 			else
1100 				CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1101 			ih->ih_flags &= ~IH_DEAD;
1102 			wakeup(ih);
1103 			mtx_unlock(&ie->ie_lock);
1104 			continue;
1105 		}
1106 
1107 		/*
1108 		 * Now that we know that the current element won't be removed
1109 		 * update the previous element.
1110 		 */
1111 		ihp = ih;
1112 
1113 		if ((ih->ih_flags & IH_CHANGED) != 0) {
1114 			mtx_lock(&ie->ie_lock);
1115 			ih->ih_flags &= ~IH_CHANGED;
1116 			wakeup(ih);
1117 			mtx_unlock(&ie->ie_lock);
1118 		}
1119 
1120 		/* Skip filter only handlers */
1121 		if (ih->ih_handler == NULL)
1122 			continue;
1123 
1124 		/* Skip suspended handlers */
1125 		if ((ih->ih_flags & IH_SUSP) != 0)
1126 			continue;
1127 
1128 		/*
1129 		 * For software interrupt threads, we only execute
1130 		 * handlers that have their need flag set.  Hardware
1131 		 * interrupt threads always invoke all of their handlers.
1132 		 *
1133 		 * ih_need can only be 0 or 1.  Failed cmpset below
1134 		 * means that there is no request to execute handlers,
1135 		 * so a retry of the cmpset is not needed.
1136 		 */
1137 		if ((ie->ie_flags & IE_SOFT) != 0 &&
1138 		    atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1139 			continue;
1140 
1141 		/* Execute this handler. */
1142 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1143 		    __func__, p->p_pid, (void *)ih->ih_handler,
1144 		    ih->ih_argument, ih->ih_name, ih->ih_flags);
1145 
1146 		if (!(ih->ih_flags & IH_MPSAFE))
1147 			mtx_lock(&Giant);
1148 		ih->ih_handler(ih->ih_argument);
1149 		if (!(ih->ih_flags & IH_MPSAFE))
1150 			mtx_unlock(&Giant);
1151 	}
1152 }
1153 
1154 static void
1155 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1156 {
1157 
1158 	/* Interrupt handlers should not sleep. */
1159 	if (!(ie->ie_flags & IE_SOFT))
1160 		THREAD_NO_SLEEPING();
1161 	intr_event_execute_handlers(p, ie);
1162 	if (!(ie->ie_flags & IE_SOFT))
1163 		THREAD_SLEEPING_OK();
1164 
1165 	/*
1166 	 * Interrupt storm handling:
1167 	 *
1168 	 * If this interrupt source is currently storming, then throttle
1169 	 * it to only fire the handler once  per clock tick.
1170 	 *
1171 	 * If this interrupt source is not currently storming, but the
1172 	 * number of back to back interrupts exceeds the storm threshold,
1173 	 * then enter storming mode.
1174 	 */
1175 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1176 	    !(ie->ie_flags & IE_SOFT)) {
1177 		/* Report the message only once every second. */
1178 		if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1179 			printf(
1180 	"interrupt storm detected on \"%s\"; throttling interrupt source\n",
1181 			    ie->ie_name);
1182 		}
1183 		pause("istorm", 1);
1184 	} else
1185 		ie->ie_count++;
1186 
1187 	/*
1188 	 * Now that all the handlers have had a chance to run, reenable
1189 	 * the interrupt source.
1190 	 */
1191 	if (ie->ie_post_ithread != NULL)
1192 		ie->ie_post_ithread(ie->ie_source);
1193 }
1194 
1195 /*
1196  * This is the main code for interrupt threads.
1197  */
1198 static void
1199 ithread_loop(void *arg)
1200 {
1201 	struct intr_thread *ithd;
1202 	struct intr_event *ie;
1203 	struct thread *td;
1204 	struct proc *p;
1205 	int wake;
1206 
1207 	td = curthread;
1208 	p = td->td_proc;
1209 	ithd = (struct intr_thread *)arg;
1210 	KASSERT(ithd->it_thread == td,
1211 	    ("%s: ithread and proc linkage out of sync", __func__));
1212 	ie = ithd->it_event;
1213 	ie->ie_count = 0;
1214 	wake = 0;
1215 
1216 	/*
1217 	 * As long as we have interrupts outstanding, go through the
1218 	 * list of handlers, giving each one a go at it.
1219 	 */
1220 	for (;;) {
1221 		/*
1222 		 * If we are an orphaned thread, then just die.
1223 		 */
1224 		if (ithd->it_flags & IT_DEAD) {
1225 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1226 			    p->p_pid, td->td_name);
1227 			free(ithd, M_ITHREAD);
1228 			kthread_exit();
1229 		}
1230 
1231 		/*
1232 		 * Service interrupts.  If another interrupt arrives while
1233 		 * we are running, it will set it_need to note that we
1234 		 * should make another pass.
1235 		 *
1236 		 * The load_acq part of the following cmpset ensures
1237 		 * that the load of ih_need in ithread_execute_handlers()
1238 		 * is ordered after the load of it_need here.
1239 		 */
1240 		while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0)
1241 			ithread_execute_handlers(p, ie);
1242 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1243 		mtx_assert(&Giant, MA_NOTOWNED);
1244 
1245 		/*
1246 		 * Processed all our interrupts.  Now get the sched
1247 		 * lock.  This may take a while and it_need may get
1248 		 * set again, so we have to check it again.
1249 		 */
1250 		thread_lock(td);
1251 		if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1252 		    (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1253 			TD_SET_IWAIT(td);
1254 			ie->ie_count = 0;
1255 			mi_switch(SW_VOL | SWT_IWAIT, NULL);
1256 		}
1257 		if (ithd->it_flags & IT_WAIT) {
1258 			wake = 1;
1259 			ithd->it_flags &= ~IT_WAIT;
1260 		}
1261 		thread_unlock(td);
1262 		if (wake) {
1263 			wakeup(ithd);
1264 			wake = 0;
1265 		}
1266 	}
1267 }
1268 
1269 /*
1270  * Main interrupt handling body.
1271  *
1272  * Input:
1273  * o ie:                        the event connected to this interrupt.
1274  * o frame:                     some archs (i.e. i386) pass a frame to some.
1275  *                              handlers as their main argument.
1276  * Return value:
1277  * o 0:                         everything ok.
1278  * o EINVAL:                    stray interrupt.
1279  */
1280 int
1281 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1282 {
1283 	struct intr_handler *ih;
1284 	struct trapframe *oldframe;
1285 	struct thread *td;
1286 	int phase;
1287 	int ret;
1288 	bool filter, thread;
1289 
1290 	td = curthread;
1291 
1292 #ifdef KSTACK_USAGE_PROF
1293 	intr_prof_stack_use(td, frame);
1294 #endif
1295 
1296 	/* An interrupt with no event or handlers is a stray interrupt. */
1297 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
1298 		return (EINVAL);
1299 
1300 	/*
1301 	 * Execute fast interrupt handlers directly.
1302 	 * To support clock handlers, if a handler registers
1303 	 * with a NULL argument, then we pass it a pointer to
1304 	 * a trapframe as its argument.
1305 	 */
1306 	td->td_intr_nesting_level++;
1307 	filter = false;
1308 	thread = false;
1309 	ret = 0;
1310 	critical_enter();
1311 	oldframe = td->td_intr_frame;
1312 	td->td_intr_frame = frame;
1313 
1314 	phase = ie->ie_phase;
1315 	atomic_add_int(&ie->ie_active[phase], 1);
1316 
1317 	/*
1318 	 * This fence is required to ensure that no later loads are
1319 	 * re-ordered before the ie_active store.
1320 	 */
1321 	atomic_thread_fence_seq_cst();
1322 
1323 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
1324 		if ((ih->ih_flags & IH_SUSP) != 0)
1325 			continue;
1326 		if (ih->ih_filter == NULL) {
1327 			thread = true;
1328 			continue;
1329 		}
1330 		CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1331 		    ih->ih_filter, ih->ih_argument == NULL ? frame :
1332 		    ih->ih_argument, ih->ih_name);
1333 		if (ih->ih_argument == NULL)
1334 			ret = ih->ih_filter(frame);
1335 		else
1336 			ret = ih->ih_filter(ih->ih_argument);
1337 		KASSERT(ret == FILTER_STRAY ||
1338 		    ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1339 		    (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1340 		    ("%s: incorrect return value %#x from %s", __func__, ret,
1341 		    ih->ih_name));
1342 		filter = filter || ret == FILTER_HANDLED;
1343 
1344 		/*
1345 		 * Wrapper handler special handling:
1346 		 *
1347 		 * in some particular cases (like pccard and pccbb),
1348 		 * the _real_ device handler is wrapped in a couple of
1349 		 * functions - a filter wrapper and an ithread wrapper.
1350 		 * In this case (and just in this case), the filter wrapper
1351 		 * could ask the system to schedule the ithread and mask
1352 		 * the interrupt source if the wrapped handler is composed
1353 		 * of just an ithread handler.
1354 		 *
1355 		 * TODO: write a generic wrapper to avoid people rolling
1356 		 * their own.
1357 		 */
1358 		if (!thread) {
1359 			if (ret == FILTER_SCHEDULE_THREAD)
1360 				thread = true;
1361 		}
1362 	}
1363 	atomic_add_rel_int(&ie->ie_active[phase], -1);
1364 
1365 	td->td_intr_frame = oldframe;
1366 
1367 	if (thread) {
1368 		if (ie->ie_pre_ithread != NULL)
1369 			ie->ie_pre_ithread(ie->ie_source);
1370 	} else {
1371 		if (ie->ie_post_filter != NULL)
1372 			ie->ie_post_filter(ie->ie_source);
1373 	}
1374 
1375 	/* Schedule the ithread if needed. */
1376 	if (thread) {
1377 		int error __unused;
1378 
1379 		error =  intr_event_schedule_thread(ie);
1380 		KASSERT(error == 0, ("bad stray interrupt"));
1381 	}
1382 	critical_exit();
1383 	td->td_intr_nesting_level--;
1384 #ifdef notyet
1385 	/* The interrupt is not aknowledged by any filter and has no ithread. */
1386 	if (!thread && !filter)
1387 		return (EINVAL);
1388 #endif
1389 	return (0);
1390 }
1391 
1392 #ifdef DDB
1393 /*
1394  * Dump details about an interrupt handler
1395  */
1396 static void
1397 db_dump_intrhand(struct intr_handler *ih)
1398 {
1399 	int comma;
1400 
1401 	db_printf("\t%-10s ", ih->ih_name);
1402 	switch (ih->ih_pri) {
1403 	case PI_REALTIME:
1404 		db_printf("CLK ");
1405 		break;
1406 	case PI_AV:
1407 		db_printf("AV  ");
1408 		break;
1409 	case PI_TTY:
1410 		db_printf("TTY ");
1411 		break;
1412 	case PI_NET:
1413 		db_printf("NET ");
1414 		break;
1415 	case PI_DISK:
1416 		db_printf("DISK");
1417 		break;
1418 	case PI_DULL:
1419 		db_printf("DULL");
1420 		break;
1421 	default:
1422 		if (ih->ih_pri >= PI_SOFT)
1423 			db_printf("SWI ");
1424 		else
1425 			db_printf("%4u", ih->ih_pri);
1426 		break;
1427 	}
1428 	db_printf(" ");
1429 	if (ih->ih_filter != NULL) {
1430 		db_printf("[F]");
1431 		db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
1432 	}
1433 	if (ih->ih_handler != NULL) {
1434 		if (ih->ih_filter != NULL)
1435 			db_printf(",");
1436 		db_printf("[H]");
1437 		db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1438 	}
1439 	db_printf("(%p)", ih->ih_argument);
1440 	if (ih->ih_need ||
1441 	    (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1442 	    IH_MPSAFE)) != 0) {
1443 		db_printf(" {");
1444 		comma = 0;
1445 		if (ih->ih_flags & IH_EXCLUSIVE) {
1446 			if (comma)
1447 				db_printf(", ");
1448 			db_printf("EXCL");
1449 			comma = 1;
1450 		}
1451 		if (ih->ih_flags & IH_ENTROPY) {
1452 			if (comma)
1453 				db_printf(", ");
1454 			db_printf("ENTROPY");
1455 			comma = 1;
1456 		}
1457 		if (ih->ih_flags & IH_DEAD) {
1458 			if (comma)
1459 				db_printf(", ");
1460 			db_printf("DEAD");
1461 			comma = 1;
1462 		}
1463 		if (ih->ih_flags & IH_MPSAFE) {
1464 			if (comma)
1465 				db_printf(", ");
1466 			db_printf("MPSAFE");
1467 			comma = 1;
1468 		}
1469 		if (ih->ih_need) {
1470 			if (comma)
1471 				db_printf(", ");
1472 			db_printf("NEED");
1473 		}
1474 		db_printf("}");
1475 	}
1476 	db_printf("\n");
1477 }
1478 
1479 /*
1480  * Dump details about a event.
1481  */
1482 void
1483 db_dump_intr_event(struct intr_event *ie, int handlers)
1484 {
1485 	struct intr_handler *ih;
1486 	struct intr_thread *it;
1487 	int comma;
1488 
1489 	db_printf("%s ", ie->ie_fullname);
1490 	it = ie->ie_thread;
1491 	if (it != NULL)
1492 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1493 	else
1494 		db_printf("(no thread)");
1495 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
1496 	    (it != NULL && it->it_need)) {
1497 		db_printf(" {");
1498 		comma = 0;
1499 		if (ie->ie_flags & IE_SOFT) {
1500 			db_printf("SOFT");
1501 			comma = 1;
1502 		}
1503 		if (ie->ie_flags & IE_ENTROPY) {
1504 			if (comma)
1505 				db_printf(", ");
1506 			db_printf("ENTROPY");
1507 			comma = 1;
1508 		}
1509 		if (ie->ie_flags & IE_ADDING_THREAD) {
1510 			if (comma)
1511 				db_printf(", ");
1512 			db_printf("ADDING_THREAD");
1513 			comma = 1;
1514 		}
1515 		if (it != NULL && it->it_need) {
1516 			if (comma)
1517 				db_printf(", ");
1518 			db_printf("NEED");
1519 		}
1520 		db_printf("}");
1521 	}
1522 	db_printf("\n");
1523 
1524 	if (handlers)
1525 		CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
1526 		    db_dump_intrhand(ih);
1527 }
1528 
1529 /*
1530  * Dump data about interrupt handlers
1531  */
1532 DB_SHOW_COMMAND(intr, db_show_intr)
1533 {
1534 	struct intr_event *ie;
1535 	int all, verbose;
1536 
1537 	verbose = strchr(modif, 'v') != NULL;
1538 	all = strchr(modif, 'a') != NULL;
1539 	TAILQ_FOREACH(ie, &event_list, ie_list) {
1540 		if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
1541 			continue;
1542 		db_dump_intr_event(ie, verbose);
1543 		if (db_pager_quit)
1544 			break;
1545 	}
1546 }
1547 #endif /* DDB */
1548 
1549 /*
1550  * Start standard software interrupt threads
1551  */
1552 static void
1553 start_softintr(void *dummy)
1554 {
1555 
1556 	if (swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
1557 		panic("died while creating vm swi ithread");
1558 }
1559 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1560     NULL);
1561 
1562 /*
1563  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1564  * The data for this machine dependent, and the declarations are in machine
1565  * dependent code.  The layout of intrnames and intrcnt however is machine
1566  * independent.
1567  *
1568  * We do not know the length of intrcnt and intrnames at compile time, so
1569  * calculate things at run time.
1570  */
1571 static int
1572 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1573 {
1574 	return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1575 }
1576 
1577 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1578     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1579 
1580 static int
1581 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1582 {
1583 #ifdef SCTL_MASK32
1584 	uint32_t *intrcnt32;
1585 	unsigned i;
1586 	int error;
1587 
1588 	if (req->flags & SCTL_MASK32) {
1589 		if (!req->oldptr)
1590 			return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
1591 		intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
1592 		if (intrcnt32 == NULL)
1593 			return (ENOMEM);
1594 		for (i = 0; i < sintrcnt / sizeof (u_long); i++)
1595 			intrcnt32[i] = intrcnt[i];
1596 		error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
1597 		free(intrcnt32, M_TEMP);
1598 		return (error);
1599 	}
1600 #endif
1601 	return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1602 }
1603 
1604 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1605     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1606 
1607 #ifdef DDB
1608 /*
1609  * DDB command to dump the interrupt statistics.
1610  */
1611 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
1612 {
1613 	u_long *i;
1614 	char *cp;
1615 	u_int j;
1616 
1617 	cp = intrnames;
1618 	j = 0;
1619 	for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1620 	    i++, j++) {
1621 		if (*cp == '\0')
1622 			break;
1623 		if (*i != 0)
1624 			db_printf("%s\t%lu\n", cp, *i);
1625 		cp += strlen(cp) + 1;
1626 	}
1627 }
1628 #endif
1629