xref: /freebsd/sys/kern/kern_intr.c (revision d8ffc21c5ca6f7d4f2d9a65dc6308699af0b6a01)
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 *tty_intr_event;
88 void	*vm_ih;
89 struct proc *intrproc;
90 
91 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
92 
93 static int intr_storm_threshold = 0;
94 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RWTUN,
95     &intr_storm_threshold, 0,
96     "Number of consecutive interrupts before storm protection is enabled");
97 static TAILQ_HEAD(, intr_event) event_list =
98     TAILQ_HEAD_INITIALIZER(event_list);
99 static struct mtx event_lock;
100 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
101 
102 static void	intr_event_update(struct intr_event *ie);
103 static int	intr_event_schedule_thread(struct intr_event *ie);
104 static struct intr_thread *ithread_create(const char *name);
105 static void	ithread_destroy(struct intr_thread *ithread);
106 static void	ithread_execute_handlers(struct proc *p,
107 		    struct intr_event *ie);
108 static void	ithread_loop(void *);
109 static void	ithread_update(struct intr_thread *ithd);
110 static void	start_softintr(void *);
111 
112 /* Map an interrupt type to an ithread priority. */
113 u_char
114 intr_priority(enum intr_type flags)
115 {
116 	u_char pri;
117 
118 	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
119 	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
120 	switch (flags) {
121 	case INTR_TYPE_TTY:
122 		pri = PI_TTY;
123 		break;
124 	case INTR_TYPE_BIO:
125 		pri = PI_DISK;
126 		break;
127 	case INTR_TYPE_NET:
128 		pri = PI_NET;
129 		break;
130 	case INTR_TYPE_CAM:
131 		pri = PI_DISK;
132 		break;
133 	case INTR_TYPE_AV:
134 		pri = PI_AV;
135 		break;
136 	case INTR_TYPE_CLK:
137 		pri = PI_REALTIME;
138 		break;
139 	case INTR_TYPE_MISC:
140 		pri = PI_DULL;          /* don't care */
141 		break;
142 	default:
143 		/* We didn't specify an interrupt level. */
144 		panic("intr_priority: no interrupt type in flags");
145 	}
146 
147 	return pri;
148 }
149 
150 /*
151  * Update an ithread based on the associated intr_event.
152  */
153 static void
154 ithread_update(struct intr_thread *ithd)
155 {
156 	struct intr_event *ie;
157 	struct thread *td;
158 	u_char pri;
159 
160 	ie = ithd->it_event;
161 	td = ithd->it_thread;
162 	mtx_assert(&ie->ie_lock, MA_OWNED);
163 
164 	/* Determine the overall priority of this event. */
165 	if (CK_SLIST_EMPTY(&ie->ie_handlers))
166 		pri = PRI_MAX_ITHD;
167 	else
168 		pri = CK_SLIST_FIRST(&ie->ie_handlers)->ih_pri;
169 
170 	/* Update name and priority. */
171 	strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
172 #ifdef KTR
173 	sched_clear_tdname(td);
174 #endif
175 	thread_lock(td);
176 	sched_prio(td, pri);
177 	thread_unlock(td);
178 }
179 
180 /*
181  * Regenerate the full name of an interrupt event and update its priority.
182  */
183 static void
184 intr_event_update(struct intr_event *ie)
185 {
186 	struct intr_handler *ih;
187 	char *last;
188 	int missed, space;
189 
190 	/* Start off with no entropy and just the name of the event. */
191 	mtx_assert(&ie->ie_lock, MA_OWNED);
192 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
193 	ie->ie_flags &= ~IE_ENTROPY;
194 	missed = 0;
195 	space = 1;
196 
197 	/* Run through all the handlers updating values. */
198 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
199 		if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
200 		    sizeof(ie->ie_fullname)) {
201 			strcat(ie->ie_fullname, " ");
202 			strcat(ie->ie_fullname, ih->ih_name);
203 			space = 0;
204 		} else
205 			missed++;
206 		if (ih->ih_flags & IH_ENTROPY)
207 			ie->ie_flags |= IE_ENTROPY;
208 	}
209 
210 	/*
211 	 * If there is only one handler and its name is too long, just copy in
212 	 * as much of the end of the name (includes the unit number) as will
213 	 * fit.  Otherwise, we have multiple handlers and not all of the names
214 	 * will fit.  Add +'s to indicate missing names.  If we run out of room
215 	 * and still have +'s to add, change the last character from a + to a *.
216 	 */
217 	if (missed == 1 && space == 1) {
218 		ih = CK_SLIST_FIRST(&ie->ie_handlers);
219 		missed = strlen(ie->ie_fullname) + strlen(ih->ih_name) + 2 -
220 		    sizeof(ie->ie_fullname);
221 		strcat(ie->ie_fullname, (missed == 0) ? " " : "-");
222 		strcat(ie->ie_fullname, &ih->ih_name[missed]);
223 		missed = 0;
224 	}
225 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
226 	while (missed-- > 0) {
227 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
228 			if (*last == '+') {
229 				*last = '*';
230 				break;
231 			} else
232 				*last = '+';
233 		} else if (space) {
234 			strcat(ie->ie_fullname, " +");
235 			space = 0;
236 		} else
237 			strcat(ie->ie_fullname, "+");
238 	}
239 
240 	/*
241 	 * If this event has an ithread, update it's priority and
242 	 * name.
243 	 */
244 	if (ie->ie_thread != NULL)
245 		ithread_update(ie->ie_thread);
246 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
247 }
248 
249 int
250 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
251     void (*pre_ithread)(void *), void (*post_ithread)(void *),
252     void (*post_filter)(void *), int (*assign_cpu)(void *, int),
253     const char *fmt, ...)
254 {
255 	struct intr_event *ie;
256 	va_list ap;
257 
258 	/* The only valid flag during creation is IE_SOFT. */
259 	if ((flags & ~IE_SOFT) != 0)
260 		return (EINVAL);
261 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
262 	ie->ie_source = source;
263 	ie->ie_pre_ithread = pre_ithread;
264 	ie->ie_post_ithread = post_ithread;
265 	ie->ie_post_filter = post_filter;
266 	ie->ie_assign_cpu = assign_cpu;
267 	ie->ie_flags = flags;
268 	ie->ie_irq = irq;
269 	ie->ie_cpu = NOCPU;
270 	CK_SLIST_INIT(&ie->ie_handlers);
271 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
272 
273 	va_start(ap, fmt);
274 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
275 	va_end(ap);
276 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
277 	mtx_lock(&event_lock);
278 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
279 	mtx_unlock(&event_lock);
280 	if (event != NULL)
281 		*event = ie;
282 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
283 	return (0);
284 }
285 
286 /*
287  * Bind an interrupt event to the specified CPU.  Note that not all
288  * platforms support binding an interrupt to a CPU.  For those
289  * platforms this request will fail.  Using a cpu id of NOCPU unbinds
290  * the interrupt event.
291  */
292 static int
293 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
294 {
295 	lwpid_t id;
296 	int error;
297 
298 	/* Need a CPU to bind to. */
299 	if (cpu != NOCPU && CPU_ABSENT(cpu))
300 		return (EINVAL);
301 
302 	if (ie->ie_assign_cpu == NULL)
303 		return (EOPNOTSUPP);
304 
305 	error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
306 	if (error)
307 		return (error);
308 
309 	/*
310 	 * If we have any ithreads try to set their mask first to verify
311 	 * permissions, etc.
312 	 */
313 	if (bindithread) {
314 		mtx_lock(&ie->ie_lock);
315 		if (ie->ie_thread != NULL) {
316 			id = ie->ie_thread->it_thread->td_tid;
317 			mtx_unlock(&ie->ie_lock);
318 			error = cpuset_setithread(id, cpu);
319 			if (error)
320 				return (error);
321 		} else
322 			mtx_unlock(&ie->ie_lock);
323 	}
324 	if (bindirq)
325 		error = ie->ie_assign_cpu(ie->ie_source, cpu);
326 	if (error) {
327 		if (bindithread) {
328 			mtx_lock(&ie->ie_lock);
329 			if (ie->ie_thread != NULL) {
330 				cpu = ie->ie_cpu;
331 				id = ie->ie_thread->it_thread->td_tid;
332 				mtx_unlock(&ie->ie_lock);
333 				(void)cpuset_setithread(id, cpu);
334 			} else
335 				mtx_unlock(&ie->ie_lock);
336 		}
337 		return (error);
338 	}
339 
340 	if (bindirq) {
341 		mtx_lock(&ie->ie_lock);
342 		ie->ie_cpu = cpu;
343 		mtx_unlock(&ie->ie_lock);
344 	}
345 
346 	return (error);
347 }
348 
349 /*
350  * Bind an interrupt event to the specified CPU.  For supported platforms, any
351  * associated ithreads as well as the primary interrupt context will be bound
352  * to the specificed CPU.
353  */
354 int
355 intr_event_bind(struct intr_event *ie, int cpu)
356 {
357 
358 	return (_intr_event_bind(ie, cpu, true, true));
359 }
360 
361 /*
362  * Bind an interrupt event to the specified CPU, but do not bind associated
363  * ithreads.
364  */
365 int
366 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
367 {
368 
369 	return (_intr_event_bind(ie, cpu, true, false));
370 }
371 
372 /*
373  * Bind an interrupt event's ithread to the specified CPU.
374  */
375 int
376 intr_event_bind_ithread(struct intr_event *ie, int cpu)
377 {
378 
379 	return (_intr_event_bind(ie, cpu, false, true));
380 }
381 
382 /*
383  * Bind an interrupt event's ithread to the specified cpuset.
384  */
385 int
386 intr_event_bind_ithread_cpuset(struct intr_event *ie, cpuset_t *cs)
387 {
388 	lwpid_t id;
389 
390 	mtx_lock(&ie->ie_lock);
391 	if (ie->ie_thread != NULL) {
392 		id = ie->ie_thread->it_thread->td_tid;
393 		mtx_unlock(&ie->ie_lock);
394 		return (cpuset_setthread(id, cs));
395 	} else {
396 		mtx_unlock(&ie->ie_lock);
397 	}
398 	return (ENODEV);
399 }
400 
401 static struct intr_event *
402 intr_lookup(int irq)
403 {
404 	struct intr_event *ie;
405 
406 	mtx_lock(&event_lock);
407 	TAILQ_FOREACH(ie, &event_list, ie_list)
408 		if (ie->ie_irq == irq &&
409 		    (ie->ie_flags & IE_SOFT) == 0 &&
410 		    CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
411 			break;
412 	mtx_unlock(&event_lock);
413 	return (ie);
414 }
415 
416 int
417 intr_setaffinity(int irq, int mode, void *m)
418 {
419 	struct intr_event *ie;
420 	cpuset_t *mask;
421 	int cpu, n;
422 
423 	mask = m;
424 	cpu = NOCPU;
425 	/*
426 	 * If we're setting all cpus we can unbind.  Otherwise make sure
427 	 * only one cpu is in the set.
428 	 */
429 	if (CPU_CMP(cpuset_root, mask)) {
430 		for (n = 0; n < CPU_SETSIZE; n++) {
431 			if (!CPU_ISSET(n, mask))
432 				continue;
433 			if (cpu != NOCPU)
434 				return (EINVAL);
435 			cpu = n;
436 		}
437 	}
438 	ie = intr_lookup(irq);
439 	if (ie == NULL)
440 		return (ESRCH);
441 	switch (mode) {
442 	case CPU_WHICH_IRQ:
443 		return (intr_event_bind(ie, cpu));
444 	case CPU_WHICH_INTRHANDLER:
445 		return (intr_event_bind_irqonly(ie, cpu));
446 	case CPU_WHICH_ITHREAD:
447 		return (intr_event_bind_ithread(ie, cpu));
448 	default:
449 		return (EINVAL);
450 	}
451 }
452 
453 int
454 intr_getaffinity(int irq, int mode, void *m)
455 {
456 	struct intr_event *ie;
457 	struct thread *td;
458 	struct proc *p;
459 	cpuset_t *mask;
460 	lwpid_t id;
461 	int error;
462 
463 	mask = m;
464 	ie = intr_lookup(irq);
465 	if (ie == NULL)
466 		return (ESRCH);
467 
468 	error = 0;
469 	CPU_ZERO(mask);
470 	switch (mode) {
471 	case CPU_WHICH_IRQ:
472 	case CPU_WHICH_INTRHANDLER:
473 		mtx_lock(&ie->ie_lock);
474 		if (ie->ie_cpu == NOCPU)
475 			CPU_COPY(cpuset_root, mask);
476 		else
477 			CPU_SET(ie->ie_cpu, mask);
478 		mtx_unlock(&ie->ie_lock);
479 		break;
480 	case CPU_WHICH_ITHREAD:
481 		mtx_lock(&ie->ie_lock);
482 		if (ie->ie_thread == NULL) {
483 			mtx_unlock(&ie->ie_lock);
484 			CPU_COPY(cpuset_root, mask);
485 		} else {
486 			id = ie->ie_thread->it_thread->td_tid;
487 			mtx_unlock(&ie->ie_lock);
488 			error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
489 			if (error != 0)
490 				return (error);
491 			CPU_COPY(&td->td_cpuset->cs_mask, mask);
492 			PROC_UNLOCK(p);
493 		}
494 	default:
495 		return (EINVAL);
496 	}
497 	return (0);
498 }
499 
500 int
501 intr_event_destroy(struct intr_event *ie)
502 {
503 
504 	mtx_lock(&event_lock);
505 	mtx_lock(&ie->ie_lock);
506 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
507 		mtx_unlock(&ie->ie_lock);
508 		mtx_unlock(&event_lock);
509 		return (EBUSY);
510 	}
511 	TAILQ_REMOVE(&event_list, ie, ie_list);
512 #ifndef notyet
513 	if (ie->ie_thread != NULL) {
514 		ithread_destroy(ie->ie_thread);
515 		ie->ie_thread = NULL;
516 	}
517 #endif
518 	mtx_unlock(&ie->ie_lock);
519 	mtx_unlock(&event_lock);
520 	mtx_destroy(&ie->ie_lock);
521 	free(ie, M_ITHREAD);
522 	return (0);
523 }
524 
525 static struct intr_thread *
526 ithread_create(const char *name)
527 {
528 	struct intr_thread *ithd;
529 	struct thread *td;
530 	int error;
531 
532 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
533 
534 	error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
535 		    &td, RFSTOPPED | RFHIGHPID,
536 		    0, "intr", "%s", name);
537 	if (error)
538 		panic("kproc_create() failed with %d", error);
539 	thread_lock(td);
540 	sched_class(td, PRI_ITHD);
541 	TD_SET_IWAIT(td);
542 	thread_unlock(td);
543 	td->td_pflags |= TDP_ITHREAD;
544 	ithd->it_thread = td;
545 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
546 	return (ithd);
547 }
548 
549 static void
550 ithread_destroy(struct intr_thread *ithread)
551 {
552 	struct thread *td;
553 
554 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
555 	td = ithread->it_thread;
556 	thread_lock(td);
557 	ithread->it_flags |= IT_DEAD;
558 	if (TD_AWAITING_INTR(td)) {
559 		TD_CLR_IWAIT(td);
560 		sched_add(td, SRQ_INTR);
561 	} else
562 		thread_unlock(td);
563 }
564 
565 int
566 intr_event_add_handler(struct intr_event *ie, const char *name,
567     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
568     enum intr_type flags, void **cookiep)
569 {
570 	struct intr_handler *ih, *temp_ih;
571 	struct intr_handler **prevptr;
572 	struct intr_thread *it;
573 
574 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
575 		return (EINVAL);
576 
577 	/* Allocate and populate an interrupt handler structure. */
578 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
579 	ih->ih_filter = filter;
580 	ih->ih_handler = handler;
581 	ih->ih_argument = arg;
582 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
583 	ih->ih_event = ie;
584 	ih->ih_pri = pri;
585 	if (flags & INTR_EXCL)
586 		ih->ih_flags = IH_EXCLUSIVE;
587 	if (flags & INTR_MPSAFE)
588 		ih->ih_flags |= IH_MPSAFE;
589 	if (flags & INTR_ENTROPY)
590 		ih->ih_flags |= IH_ENTROPY;
591 
592 	/* We can only have one exclusive handler in a event. */
593 	mtx_lock(&ie->ie_lock);
594 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
595 		if ((flags & INTR_EXCL) ||
596 		    (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
597 			mtx_unlock(&ie->ie_lock);
598 			free(ih, M_ITHREAD);
599 			return (EINVAL);
600 		}
601 	}
602 
603 	/* Create a thread if we need one. */
604 	while (ie->ie_thread == NULL && handler != NULL) {
605 		if (ie->ie_flags & IE_ADDING_THREAD)
606 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
607 		else {
608 			ie->ie_flags |= IE_ADDING_THREAD;
609 			mtx_unlock(&ie->ie_lock);
610 			it = ithread_create("intr: newborn");
611 			mtx_lock(&ie->ie_lock);
612 			ie->ie_flags &= ~IE_ADDING_THREAD;
613 			ie->ie_thread = it;
614 			it->it_event = ie;
615 			ithread_update(it);
616 			wakeup(ie);
617 		}
618 	}
619 
620 	/* Add the new handler to the event in priority order. */
621 	CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
622 		if (temp_ih->ih_pri > ih->ih_pri)
623 			break;
624 	}
625 	CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
626 
627 	intr_event_update(ie);
628 
629 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
630 	    ie->ie_name);
631 	mtx_unlock(&ie->ie_lock);
632 
633 	if (cookiep != NULL)
634 		*cookiep = ih;
635 	return (0);
636 }
637 
638 /*
639  * Append a description preceded by a ':' to the name of the specified
640  * interrupt handler.
641  */
642 int
643 intr_event_describe_handler(struct intr_event *ie, void *cookie,
644     const char *descr)
645 {
646 	struct intr_handler *ih;
647 	size_t space;
648 	char *start;
649 
650 	mtx_lock(&ie->ie_lock);
651 #ifdef INVARIANTS
652 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
653 		if (ih == cookie)
654 			break;
655 	}
656 	if (ih == NULL) {
657 		mtx_unlock(&ie->ie_lock);
658 		panic("handler %p not found in interrupt event %p", cookie, ie);
659 	}
660 #endif
661 	ih = cookie;
662 
663 	/*
664 	 * Look for an existing description by checking for an
665 	 * existing ":".  This assumes device names do not include
666 	 * colons.  If one is found, prepare to insert the new
667 	 * description at that point.  If one is not found, find the
668 	 * end of the name to use as the insertion point.
669 	 */
670 	start = strchr(ih->ih_name, ':');
671 	if (start == NULL)
672 		start = strchr(ih->ih_name, 0);
673 
674 	/*
675 	 * See if there is enough remaining room in the string for the
676 	 * description + ":".  The "- 1" leaves room for the trailing
677 	 * '\0'.  The "+ 1" accounts for the colon.
678 	 */
679 	space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
680 	if (strlen(descr) + 1 > space) {
681 		mtx_unlock(&ie->ie_lock);
682 		return (ENOSPC);
683 	}
684 
685 	/* Append a colon followed by the description. */
686 	*start = ':';
687 	strcpy(start + 1, descr);
688 	intr_event_update(ie);
689 	mtx_unlock(&ie->ie_lock);
690 	return (0);
691 }
692 
693 /*
694  * Return the ie_source field from the intr_event an intr_handler is
695  * associated with.
696  */
697 void *
698 intr_handler_source(void *cookie)
699 {
700 	struct intr_handler *ih;
701 	struct intr_event *ie;
702 
703 	ih = (struct intr_handler *)cookie;
704 	if (ih == NULL)
705 		return (NULL);
706 	ie = ih->ih_event;
707 	KASSERT(ie != NULL,
708 	    ("interrupt handler \"%s\" has a NULL interrupt event",
709 	    ih->ih_name));
710 	return (ie->ie_source);
711 }
712 
713 /*
714  * If intr_event_handle() is running in the ISR context at the time of the call,
715  * then wait for it to complete.
716  */
717 static void
718 intr_event_barrier(struct intr_event *ie)
719 {
720 	int phase;
721 
722 	mtx_assert(&ie->ie_lock, MA_OWNED);
723 	phase = ie->ie_phase;
724 
725 	/*
726 	 * Switch phase to direct future interrupts to the other active counter.
727 	 * Make sure that any preceding stores are visible before the switch.
728 	 */
729 	KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
730 	atomic_store_rel_int(&ie->ie_phase, !phase);
731 
732 	/*
733 	 * This code cooperates with wait-free iteration of ie_handlers
734 	 * in intr_event_handle.
735 	 * Make sure that the removal and the phase update are not reordered
736 	 * with the active count check.
737 	 * Note that no combination of acquire and release fences can provide
738 	 * that guarantee as Store->Load sequences can always be reordered.
739 	 */
740 	atomic_thread_fence_seq_cst();
741 
742 	/*
743 	 * Now wait on the inactive phase.
744 	 * The acquire fence is needed so that that all post-barrier accesses
745 	 * are after the check.
746 	 */
747 	while (ie->ie_active[phase] > 0)
748 		cpu_spinwait();
749 	atomic_thread_fence_acq();
750 }
751 
752 static void
753 intr_handler_barrier(struct intr_handler *handler)
754 {
755 	struct intr_event *ie;
756 
757 	ie = handler->ih_event;
758 	mtx_assert(&ie->ie_lock, MA_OWNED);
759 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
760 	    ("update for a removed handler"));
761 
762 	if (ie->ie_thread == NULL) {
763 		intr_event_barrier(ie);
764 		return;
765 	}
766 	if ((handler->ih_flags & IH_CHANGED) == 0) {
767 		handler->ih_flags |= IH_CHANGED;
768 		intr_event_schedule_thread(ie);
769 	}
770 	while ((handler->ih_flags & IH_CHANGED) != 0)
771 		msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
772 }
773 
774 /*
775  * Sleep until an ithread finishes executing an interrupt handler.
776  *
777  * XXX Doesn't currently handle interrupt filters or fast interrupt
778  * handlers.  This is intended for compatibility with linux drivers
779  * only.  Do not use in BSD code.
780  */
781 void
782 _intr_drain(int irq)
783 {
784 	struct intr_event *ie;
785 	struct intr_thread *ithd;
786 	struct thread *td;
787 
788 	ie = intr_lookup(irq);
789 	if (ie == NULL)
790 		return;
791 	if (ie->ie_thread == NULL)
792 		return;
793 	ithd = ie->ie_thread;
794 	td = ithd->it_thread;
795 	/*
796 	 * We set the flag and wait for it to be cleared to avoid
797 	 * long delays with potentially busy interrupt handlers
798 	 * were we to only sample TD_AWAITING_INTR() every tick.
799 	 */
800 	thread_lock(td);
801 	if (!TD_AWAITING_INTR(td)) {
802 		ithd->it_flags |= IT_WAIT;
803 		while (ithd->it_flags & IT_WAIT) {
804 			thread_unlock(td);
805 			pause("idrain", 1);
806 			thread_lock(td);
807 		}
808 	}
809 	thread_unlock(td);
810 	return;
811 }
812 
813 int
814 intr_event_remove_handler(void *cookie)
815 {
816 	struct intr_handler *handler = (struct intr_handler *)cookie;
817 	struct intr_event *ie;
818 	struct intr_handler *ih;
819 	struct intr_handler **prevptr;
820 #ifdef notyet
821 	int dead;
822 #endif
823 
824 	if (handler == NULL)
825 		return (EINVAL);
826 	ie = handler->ih_event;
827 	KASSERT(ie != NULL,
828 	    ("interrupt handler \"%s\" has a NULL interrupt event",
829 	    handler->ih_name));
830 
831 	mtx_lock(&ie->ie_lock);
832 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
833 	    ie->ie_name);
834 	CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
835 		if (ih == handler)
836 			break;
837 	}
838 	if (ih == NULL) {
839 		panic("interrupt handler \"%s\" not found in "
840 		    "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
841 	}
842 
843 	/*
844 	 * If there is no ithread, then directly remove the handler.  Note that
845 	 * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
846 	 * care needs to be taken to keep ie_handlers consistent and to free
847 	 * the removed handler only when ie_handlers is quiescent.
848 	 */
849 	if (ie->ie_thread == NULL) {
850 		CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
851 		intr_event_barrier(ie);
852 		intr_event_update(ie);
853 		mtx_unlock(&ie->ie_lock);
854 		free(handler, M_ITHREAD);
855 		return (0);
856 	}
857 
858 	/*
859 	 * Let the interrupt thread do the job.
860 	 * The interrupt source is disabled when the interrupt thread is
861 	 * running, so it does not have to worry about interaction with
862 	 * intr_event_handle().
863 	 */
864 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
865 	    ("duplicate handle remove"));
866 	handler->ih_flags |= IH_DEAD;
867 	intr_event_schedule_thread(ie);
868 	while (handler->ih_flags & IH_DEAD)
869 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
870 	intr_event_update(ie);
871 
872 #ifdef notyet
873 	/*
874 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
875 	 * this could lead to races of stale data when servicing an
876 	 * interrupt.
877 	 */
878 	dead = 1;
879 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
880 		if (ih->ih_handler != NULL) {
881 			dead = 0;
882 			break;
883 		}
884 	}
885 	if (dead) {
886 		ithread_destroy(ie->ie_thread);
887 		ie->ie_thread = NULL;
888 	}
889 #endif
890 	mtx_unlock(&ie->ie_lock);
891 	free(handler, M_ITHREAD);
892 	return (0);
893 }
894 
895 int
896 intr_event_suspend_handler(void *cookie)
897 {
898 	struct intr_handler *handler = (struct intr_handler *)cookie;
899 	struct intr_event *ie;
900 
901 	if (handler == NULL)
902 		return (EINVAL);
903 	ie = handler->ih_event;
904 	KASSERT(ie != NULL,
905 	    ("interrupt handler \"%s\" has a NULL interrupt event",
906 	    handler->ih_name));
907 	mtx_lock(&ie->ie_lock);
908 	handler->ih_flags |= IH_SUSP;
909 	intr_handler_barrier(handler);
910 	mtx_unlock(&ie->ie_lock);
911 	return (0);
912 }
913 
914 int
915 intr_event_resume_handler(void *cookie)
916 {
917 	struct intr_handler *handler = (struct intr_handler *)cookie;
918 	struct intr_event *ie;
919 
920 	if (handler == NULL)
921 		return (EINVAL);
922 	ie = handler->ih_event;
923 	KASSERT(ie != NULL,
924 	    ("interrupt handler \"%s\" has a NULL interrupt event",
925 	    handler->ih_name));
926 
927 	/*
928 	 * intr_handler_barrier() acts not only as a barrier,
929 	 * it also allows to check for any pending interrupts.
930 	 */
931 	mtx_lock(&ie->ie_lock);
932 	handler->ih_flags &= ~IH_SUSP;
933 	intr_handler_barrier(handler);
934 	mtx_unlock(&ie->ie_lock);
935 	return (0);
936 }
937 
938 static int
939 intr_event_schedule_thread(struct intr_event *ie)
940 {
941 	struct intr_entropy entropy;
942 	struct intr_thread *it;
943 	struct thread *td;
944 	struct thread *ctd;
945 
946 	/*
947 	 * If no ithread or no handlers, then we have a stray interrupt.
948 	 */
949 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
950 	    ie->ie_thread == NULL)
951 		return (EINVAL);
952 
953 	ctd = curthread;
954 	it = ie->ie_thread;
955 	td = it->it_thread;
956 
957 	/*
958 	 * If any of the handlers for this ithread claim to be good
959 	 * sources of entropy, then gather some.
960 	 */
961 	if (ie->ie_flags & IE_ENTROPY) {
962 		entropy.event = (uintptr_t)ie;
963 		entropy.td = ctd;
964 		random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
965 	}
966 
967 	KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
968 
969 	/*
970 	 * Set it_need to tell the thread to keep running if it is already
971 	 * running.  Then, lock the thread and see if we actually need to
972 	 * put it on the runqueue.
973 	 *
974 	 * Use store_rel to arrange that the store to ih_need in
975 	 * swi_sched() is before the store to it_need and prepare for
976 	 * transfer of this order to loads in the ithread.
977 	 */
978 	atomic_store_rel_int(&it->it_need, 1);
979 	thread_lock(td);
980 	if (TD_AWAITING_INTR(td)) {
981 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
982 		    td->td_name);
983 		TD_CLR_IWAIT(td);
984 		sched_add(td, SRQ_INTR);
985 	} else {
986 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
987 		    __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
988 		thread_unlock(td);
989 	}
990 
991 	return (0);
992 }
993 
994 /*
995  * Allow interrupt event binding for software interrupt handlers -- a no-op,
996  * since interrupts are generated in software rather than being directed by
997  * a PIC.
998  */
999 static int
1000 swi_assign_cpu(void *arg, int cpu)
1001 {
1002 
1003 	return (0);
1004 }
1005 
1006 /*
1007  * Add a software interrupt handler to a specified event.  If a given event
1008  * is not specified, then a new event is created.
1009  */
1010 int
1011 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
1012 	    void *arg, int pri, enum intr_type flags, void **cookiep)
1013 {
1014 	struct intr_event *ie;
1015 	int error;
1016 
1017 	if (flags & INTR_ENTROPY)
1018 		return (EINVAL);
1019 
1020 	ie = (eventp != NULL) ? *eventp : NULL;
1021 
1022 	if (ie != NULL) {
1023 		if (!(ie->ie_flags & IE_SOFT))
1024 			return (EINVAL);
1025 	} else {
1026 		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1027 		    NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1028 		if (error)
1029 			return (error);
1030 		if (eventp != NULL)
1031 			*eventp = ie;
1032 	}
1033 	error = intr_event_add_handler(ie, name, NULL, handler, arg,
1034 	    PI_SWI(pri), flags, cookiep);
1035 	return (error);
1036 }
1037 
1038 /*
1039  * Schedule a software interrupt thread.
1040  */
1041 void
1042 swi_sched(void *cookie, int flags)
1043 {
1044 	struct intr_handler *ih = (struct intr_handler *)cookie;
1045 	struct intr_event *ie = ih->ih_event;
1046 	struct intr_entropy entropy;
1047 	int error __unused;
1048 
1049 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1050 	    ih->ih_need);
1051 
1052 	entropy.event = (uintptr_t)ih;
1053 	entropy.td = curthread;
1054 	random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1055 
1056 	/*
1057 	 * Set ih_need for this handler so that if the ithread is already
1058 	 * running it will execute this handler on the next pass.  Otherwise,
1059 	 * it will execute it the next time it runs.
1060 	 */
1061 	ih->ih_need = 1;
1062 
1063 	if (!(flags & SWI_DELAY)) {
1064 		VM_CNT_INC(v_soft);
1065 		error = intr_event_schedule_thread(ie);
1066 		KASSERT(error == 0, ("stray software interrupt"));
1067 	}
1068 }
1069 
1070 /*
1071  * Remove a software interrupt handler.  Currently this code does not
1072  * remove the associated interrupt event if it becomes empty.  Calling code
1073  * may do so manually via intr_event_destroy(), but that's not really
1074  * an optimal interface.
1075  */
1076 int
1077 swi_remove(void *cookie)
1078 {
1079 
1080 	return (intr_event_remove_handler(cookie));
1081 }
1082 
1083 static void
1084 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1085 {
1086 	struct intr_handler *ih, *ihn, *ihp;
1087 
1088 	ihp = NULL;
1089 	CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1090 		/*
1091 		 * If this handler is marked for death, remove it from
1092 		 * the list of handlers and wake up the sleeper.
1093 		 */
1094 		if (ih->ih_flags & IH_DEAD) {
1095 			mtx_lock(&ie->ie_lock);
1096 			if (ihp == NULL)
1097 				CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1098 			else
1099 				CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1100 			ih->ih_flags &= ~IH_DEAD;
1101 			wakeup(ih);
1102 			mtx_unlock(&ie->ie_lock);
1103 			continue;
1104 		}
1105 
1106 		/*
1107 		 * Now that we know that the current element won't be removed
1108 		 * update the previous element.
1109 		 */
1110 		ihp = ih;
1111 
1112 		if ((ih->ih_flags & IH_CHANGED) != 0) {
1113 			mtx_lock(&ie->ie_lock);
1114 			ih->ih_flags &= ~IH_CHANGED;
1115 			wakeup(ih);
1116 			mtx_unlock(&ie->ie_lock);
1117 		}
1118 
1119 		/* Skip filter only handlers */
1120 		if (ih->ih_handler == NULL)
1121 			continue;
1122 
1123 		/* Skip suspended handlers */
1124 		if ((ih->ih_flags & IH_SUSP) != 0)
1125 			continue;
1126 
1127 		/*
1128 		 * For software interrupt threads, we only execute
1129 		 * handlers that have their need flag set.  Hardware
1130 		 * interrupt threads always invoke all of their handlers.
1131 		 *
1132 		 * ih_need can only be 0 or 1.  Failed cmpset below
1133 		 * means that there is no request to execute handlers,
1134 		 * so a retry of the cmpset is not needed.
1135 		 */
1136 		if ((ie->ie_flags & IE_SOFT) != 0 &&
1137 		    atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1138 			continue;
1139 
1140 		/* Execute this handler. */
1141 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1142 		    __func__, p->p_pid, (void *)ih->ih_handler,
1143 		    ih->ih_argument, ih->ih_name, ih->ih_flags);
1144 
1145 		if (!(ih->ih_flags & IH_MPSAFE))
1146 			mtx_lock(&Giant);
1147 		ih->ih_handler(ih->ih_argument);
1148 		if (!(ih->ih_flags & IH_MPSAFE))
1149 			mtx_unlock(&Giant);
1150 	}
1151 }
1152 
1153 static void
1154 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1155 {
1156 
1157 	/* Interrupt handlers should not sleep. */
1158 	if (!(ie->ie_flags & IE_SOFT))
1159 		THREAD_NO_SLEEPING();
1160 	intr_event_execute_handlers(p, ie);
1161 	if (!(ie->ie_flags & IE_SOFT))
1162 		THREAD_SLEEPING_OK();
1163 
1164 	/*
1165 	 * Interrupt storm handling:
1166 	 *
1167 	 * If this interrupt source is currently storming, then throttle
1168 	 * it to only fire the handler once  per clock tick.
1169 	 *
1170 	 * If this interrupt source is not currently storming, but the
1171 	 * number of back to back interrupts exceeds the storm threshold,
1172 	 * then enter storming mode.
1173 	 */
1174 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1175 	    !(ie->ie_flags & IE_SOFT)) {
1176 		/* Report the message only once every second. */
1177 		if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1178 			printf(
1179 	"interrupt storm detected on \"%s\"; throttling interrupt source\n",
1180 			    ie->ie_name);
1181 		}
1182 		pause("istorm", 1);
1183 	} else
1184 		ie->ie_count++;
1185 
1186 	/*
1187 	 * Now that all the handlers have had a chance to run, reenable
1188 	 * the interrupt source.
1189 	 */
1190 	if (ie->ie_post_ithread != NULL)
1191 		ie->ie_post_ithread(ie->ie_source);
1192 }
1193 
1194 /*
1195  * This is the main code for interrupt threads.
1196  */
1197 static void
1198 ithread_loop(void *arg)
1199 {
1200 	struct intr_thread *ithd;
1201 	struct intr_event *ie;
1202 	struct thread *td;
1203 	struct proc *p;
1204 	int wake;
1205 
1206 	td = curthread;
1207 	p = td->td_proc;
1208 	ithd = (struct intr_thread *)arg;
1209 	KASSERT(ithd->it_thread == td,
1210 	    ("%s: ithread and proc linkage out of sync", __func__));
1211 	ie = ithd->it_event;
1212 	ie->ie_count = 0;
1213 	wake = 0;
1214 
1215 	/*
1216 	 * As long as we have interrupts outstanding, go through the
1217 	 * list of handlers, giving each one a go at it.
1218 	 */
1219 	for (;;) {
1220 		/*
1221 		 * If we are an orphaned thread, then just die.
1222 		 */
1223 		if (ithd->it_flags & IT_DEAD) {
1224 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1225 			    p->p_pid, td->td_name);
1226 			free(ithd, M_ITHREAD);
1227 			kthread_exit();
1228 		}
1229 
1230 		/*
1231 		 * Service interrupts.  If another interrupt arrives while
1232 		 * we are running, it will set it_need to note that we
1233 		 * should make another pass.
1234 		 *
1235 		 * The load_acq part of the following cmpset ensures
1236 		 * that the load of ih_need in ithread_execute_handlers()
1237 		 * is ordered after the load of it_need here.
1238 		 */
1239 		while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0)
1240 			ithread_execute_handlers(p, ie);
1241 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1242 		mtx_assert(&Giant, MA_NOTOWNED);
1243 
1244 		/*
1245 		 * Processed all our interrupts.  Now get the sched
1246 		 * lock.  This may take a while and it_need may get
1247 		 * set again, so we have to check it again.
1248 		 */
1249 		thread_lock(td);
1250 		if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1251 		    (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1252 			TD_SET_IWAIT(td);
1253 			ie->ie_count = 0;
1254 			mi_switch(SW_VOL | SWT_IWAIT);
1255 		} else {
1256 			if (ithd->it_flags & IT_WAIT) {
1257 				wake = 1;
1258 				ithd->it_flags &= ~IT_WAIT;
1259 			}
1260 			thread_unlock(td);
1261 		}
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