xref: /freebsd/sys/kern/kern_intr.c (revision 71ceff9f0b4a3add236a6458da0ea3671fedd0a1)
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 = 1000;
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 static struct intr_event *
384 intr_lookup(int irq)
385 {
386 	struct intr_event *ie;
387 
388 	mtx_lock(&event_lock);
389 	TAILQ_FOREACH(ie, &event_list, ie_list)
390 		if (ie->ie_irq == irq &&
391 		    (ie->ie_flags & IE_SOFT) == 0 &&
392 		    CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
393 			break;
394 	mtx_unlock(&event_lock);
395 	return (ie);
396 }
397 
398 int
399 intr_setaffinity(int irq, int mode, void *m)
400 {
401 	struct intr_event *ie;
402 	cpuset_t *mask;
403 	int cpu, n;
404 
405 	mask = m;
406 	cpu = NOCPU;
407 	/*
408 	 * If we're setting all cpus we can unbind.  Otherwise make sure
409 	 * only one cpu is in the set.
410 	 */
411 	if (CPU_CMP(cpuset_root, mask)) {
412 		for (n = 0; n < CPU_SETSIZE; n++) {
413 			if (!CPU_ISSET(n, mask))
414 				continue;
415 			if (cpu != NOCPU)
416 				return (EINVAL);
417 			cpu = n;
418 		}
419 	}
420 	ie = intr_lookup(irq);
421 	if (ie == NULL)
422 		return (ESRCH);
423 	switch (mode) {
424 	case CPU_WHICH_IRQ:
425 		return (intr_event_bind(ie, cpu));
426 	case CPU_WHICH_INTRHANDLER:
427 		return (intr_event_bind_irqonly(ie, cpu));
428 	case CPU_WHICH_ITHREAD:
429 		return (intr_event_bind_ithread(ie, cpu));
430 	default:
431 		return (EINVAL);
432 	}
433 }
434 
435 int
436 intr_getaffinity(int irq, int mode, void *m)
437 {
438 	struct intr_event *ie;
439 	struct thread *td;
440 	struct proc *p;
441 	cpuset_t *mask;
442 	lwpid_t id;
443 	int error;
444 
445 	mask = m;
446 	ie = intr_lookup(irq);
447 	if (ie == NULL)
448 		return (ESRCH);
449 
450 	error = 0;
451 	CPU_ZERO(mask);
452 	switch (mode) {
453 	case CPU_WHICH_IRQ:
454 	case CPU_WHICH_INTRHANDLER:
455 		mtx_lock(&ie->ie_lock);
456 		if (ie->ie_cpu == NOCPU)
457 			CPU_COPY(cpuset_root, mask);
458 		else
459 			CPU_SET(ie->ie_cpu, mask);
460 		mtx_unlock(&ie->ie_lock);
461 		break;
462 	case CPU_WHICH_ITHREAD:
463 		mtx_lock(&ie->ie_lock);
464 		if (ie->ie_thread == NULL) {
465 			mtx_unlock(&ie->ie_lock);
466 			CPU_COPY(cpuset_root, mask);
467 		} else {
468 			id = ie->ie_thread->it_thread->td_tid;
469 			mtx_unlock(&ie->ie_lock);
470 			error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
471 			if (error != 0)
472 				return (error);
473 			CPU_COPY(&td->td_cpuset->cs_mask, mask);
474 			PROC_UNLOCK(p);
475 		}
476 	default:
477 		return (EINVAL);
478 	}
479 	return (0);
480 }
481 
482 int
483 intr_event_destroy(struct intr_event *ie)
484 {
485 
486 	mtx_lock(&event_lock);
487 	mtx_lock(&ie->ie_lock);
488 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
489 		mtx_unlock(&ie->ie_lock);
490 		mtx_unlock(&event_lock);
491 		return (EBUSY);
492 	}
493 	TAILQ_REMOVE(&event_list, ie, ie_list);
494 #ifndef notyet
495 	if (ie->ie_thread != NULL) {
496 		ithread_destroy(ie->ie_thread);
497 		ie->ie_thread = NULL;
498 	}
499 #endif
500 	mtx_unlock(&ie->ie_lock);
501 	mtx_unlock(&event_lock);
502 	mtx_destroy(&ie->ie_lock);
503 	free(ie, M_ITHREAD);
504 	return (0);
505 }
506 
507 static struct intr_thread *
508 ithread_create(const char *name)
509 {
510 	struct intr_thread *ithd;
511 	struct thread *td;
512 	int error;
513 
514 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
515 
516 	error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
517 		    &td, RFSTOPPED | RFHIGHPID,
518 		    0, "intr", "%s", name);
519 	if (error)
520 		panic("kproc_create() failed with %d", error);
521 	thread_lock(td);
522 	sched_class(td, PRI_ITHD);
523 	TD_SET_IWAIT(td);
524 	thread_unlock(td);
525 	td->td_pflags |= TDP_ITHREAD;
526 	ithd->it_thread = td;
527 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
528 	return (ithd);
529 }
530 
531 static void
532 ithread_destroy(struct intr_thread *ithread)
533 {
534 	struct thread *td;
535 
536 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
537 	td = ithread->it_thread;
538 	thread_lock(td);
539 	ithread->it_flags |= IT_DEAD;
540 	if (TD_AWAITING_INTR(td)) {
541 		TD_CLR_IWAIT(td);
542 		sched_add(td, SRQ_INTR);
543 	}
544 	thread_unlock(td);
545 }
546 
547 int
548 intr_event_add_handler(struct intr_event *ie, const char *name,
549     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
550     enum intr_type flags, void **cookiep)
551 {
552 	struct intr_handler *ih, *temp_ih;
553 	struct intr_handler **prevptr;
554 	struct intr_thread *it;
555 
556 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
557 		return (EINVAL);
558 
559 	/* Allocate and populate an interrupt handler structure. */
560 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
561 	ih->ih_filter = filter;
562 	ih->ih_handler = handler;
563 	ih->ih_argument = arg;
564 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
565 	ih->ih_event = ie;
566 	ih->ih_pri = pri;
567 	if (flags & INTR_EXCL)
568 		ih->ih_flags = IH_EXCLUSIVE;
569 	if (flags & INTR_MPSAFE)
570 		ih->ih_flags |= IH_MPSAFE;
571 	if (flags & INTR_ENTROPY)
572 		ih->ih_flags |= IH_ENTROPY;
573 
574 	/* We can only have one exclusive handler in a event. */
575 	mtx_lock(&ie->ie_lock);
576 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
577 		if ((flags & INTR_EXCL) ||
578 		    (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
579 			mtx_unlock(&ie->ie_lock);
580 			free(ih, M_ITHREAD);
581 			return (EINVAL);
582 		}
583 	}
584 
585 	/* Create a thread if we need one. */
586 	while (ie->ie_thread == NULL && handler != NULL) {
587 		if (ie->ie_flags & IE_ADDING_THREAD)
588 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
589 		else {
590 			ie->ie_flags |= IE_ADDING_THREAD;
591 			mtx_unlock(&ie->ie_lock);
592 			it = ithread_create("intr: newborn");
593 			mtx_lock(&ie->ie_lock);
594 			ie->ie_flags &= ~IE_ADDING_THREAD;
595 			ie->ie_thread = it;
596 			it->it_event = ie;
597 			ithread_update(it);
598 			wakeup(ie);
599 		}
600 	}
601 
602 	/* Add the new handler to the event in priority order. */
603 	CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
604 		if (temp_ih->ih_pri > ih->ih_pri)
605 			break;
606 	}
607 	CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
608 
609 	intr_event_update(ie);
610 
611 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
612 	    ie->ie_name);
613 	mtx_unlock(&ie->ie_lock);
614 
615 	if (cookiep != NULL)
616 		*cookiep = ih;
617 	return (0);
618 }
619 
620 /*
621  * Append a description preceded by a ':' to the name of the specified
622  * interrupt handler.
623  */
624 int
625 intr_event_describe_handler(struct intr_event *ie, void *cookie,
626     const char *descr)
627 {
628 	struct intr_handler *ih;
629 	size_t space;
630 	char *start;
631 
632 	mtx_lock(&ie->ie_lock);
633 #ifdef INVARIANTS
634 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
635 		if (ih == cookie)
636 			break;
637 	}
638 	if (ih == NULL) {
639 		mtx_unlock(&ie->ie_lock);
640 		panic("handler %p not found in interrupt event %p", cookie, ie);
641 	}
642 #endif
643 	ih = cookie;
644 
645 	/*
646 	 * Look for an existing description by checking for an
647 	 * existing ":".  This assumes device names do not include
648 	 * colons.  If one is found, prepare to insert the new
649 	 * description at that point.  If one is not found, find the
650 	 * end of the name to use as the insertion point.
651 	 */
652 	start = strchr(ih->ih_name, ':');
653 	if (start == NULL)
654 		start = strchr(ih->ih_name, 0);
655 
656 	/*
657 	 * See if there is enough remaining room in the string for the
658 	 * description + ":".  The "- 1" leaves room for the trailing
659 	 * '\0'.  The "+ 1" accounts for the colon.
660 	 */
661 	space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
662 	if (strlen(descr) + 1 > space) {
663 		mtx_unlock(&ie->ie_lock);
664 		return (ENOSPC);
665 	}
666 
667 	/* Append a colon followed by the description. */
668 	*start = ':';
669 	strcpy(start + 1, descr);
670 	intr_event_update(ie);
671 	mtx_unlock(&ie->ie_lock);
672 	return (0);
673 }
674 
675 /*
676  * Return the ie_source field from the intr_event an intr_handler is
677  * associated with.
678  */
679 void *
680 intr_handler_source(void *cookie)
681 {
682 	struct intr_handler *ih;
683 	struct intr_event *ie;
684 
685 	ih = (struct intr_handler *)cookie;
686 	if (ih == NULL)
687 		return (NULL);
688 	ie = ih->ih_event;
689 	KASSERT(ie != NULL,
690 	    ("interrupt handler \"%s\" has a NULL interrupt event",
691 	    ih->ih_name));
692 	return (ie->ie_source);
693 }
694 
695 /*
696  * If intr_event_handle() is running in the ISR context at the time of the call,
697  * then wait for it to complete.
698  */
699 static void
700 intr_event_barrier(struct intr_event *ie)
701 {
702 	int phase;
703 
704 	mtx_assert(&ie->ie_lock, MA_OWNED);
705 	phase = ie->ie_phase;
706 
707 	/*
708 	 * Switch phase to direct future interrupts to the other active counter.
709 	 * Make sure that any preceding stores are visible before the switch.
710 	 */
711 	KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
712 	atomic_store_rel_int(&ie->ie_phase, !phase);
713 
714 	/*
715 	 * This code cooperates with wait-free iteration of ie_handlers
716 	 * in intr_event_handle.
717 	 * Make sure that the removal and the phase update are not reordered
718 	 * with the active count check.
719 	 * Note that no combination of acquire and release fences can provide
720 	 * that guarantee as Store->Load sequences can always be reordered.
721 	 */
722 	atomic_thread_fence_seq_cst();
723 
724 	/*
725 	 * Now wait on the inactive phase.
726 	 * The acquire fence is needed so that that all post-barrier accesses
727 	 * are after the check.
728 	 */
729 	while (ie->ie_active[phase] > 0)
730 		cpu_spinwait();
731 	atomic_thread_fence_acq();
732 }
733 
734 static void
735 intr_handler_barrier(struct intr_handler *handler)
736 {
737 	struct intr_event *ie;
738 
739 	ie = handler->ih_event;
740 	mtx_assert(&ie->ie_lock, MA_OWNED);
741 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
742 	    ("update for a removed handler"));
743 
744 	if (ie->ie_thread == NULL) {
745 		intr_event_barrier(ie);
746 		return;
747 	}
748 	if ((handler->ih_flags & IH_CHANGED) == 0) {
749 		handler->ih_flags |= IH_CHANGED;
750 		intr_event_schedule_thread(ie);
751 	}
752 	while ((handler->ih_flags & IH_CHANGED) != 0)
753 		msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
754 }
755 
756 /*
757  * Sleep until an ithread finishes executing an interrupt handler.
758  *
759  * XXX Doesn't currently handle interrupt filters or fast interrupt
760  * handlers.  This is intended for compatibility with linux drivers
761  * only.  Do not use in BSD code.
762  */
763 void
764 _intr_drain(int irq)
765 {
766 	struct intr_event *ie;
767 	struct intr_thread *ithd;
768 	struct thread *td;
769 
770 	ie = intr_lookup(irq);
771 	if (ie == NULL)
772 		return;
773 	if (ie->ie_thread == NULL)
774 		return;
775 	ithd = ie->ie_thread;
776 	td = ithd->it_thread;
777 	/*
778 	 * We set the flag and wait for it to be cleared to avoid
779 	 * long delays with potentially busy interrupt handlers
780 	 * were we to only sample TD_AWAITING_INTR() every tick.
781 	 */
782 	thread_lock(td);
783 	if (!TD_AWAITING_INTR(td)) {
784 		ithd->it_flags |= IT_WAIT;
785 		while (ithd->it_flags & IT_WAIT) {
786 			thread_unlock(td);
787 			pause("idrain", 1);
788 			thread_lock(td);
789 		}
790 	}
791 	thread_unlock(td);
792 	return;
793 }
794 
795 int
796 intr_event_remove_handler(void *cookie)
797 {
798 	struct intr_handler *handler = (struct intr_handler *)cookie;
799 	struct intr_event *ie;
800 	struct intr_handler *ih;
801 	struct intr_handler **prevptr;
802 #ifdef notyet
803 	int dead;
804 #endif
805 
806 	if (handler == NULL)
807 		return (EINVAL);
808 	ie = handler->ih_event;
809 	KASSERT(ie != NULL,
810 	    ("interrupt handler \"%s\" has a NULL interrupt event",
811 	    handler->ih_name));
812 
813 	mtx_lock(&ie->ie_lock);
814 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
815 	    ie->ie_name);
816 	CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
817 		if (ih == handler)
818 			break;
819 	}
820 	if (ih == NULL) {
821 		panic("interrupt handler \"%s\" not found in "
822 		    "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
823 	}
824 
825 	/*
826 	 * If there is no ithread, then directly remove the handler.  Note that
827 	 * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
828 	 * care needs to be taken to keep ie_handlers consistent and to free
829 	 * the removed handler only when ie_handlers is quiescent.
830 	 */
831 	if (ie->ie_thread == NULL) {
832 		CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
833 		intr_event_barrier(ie);
834 		intr_event_update(ie);
835 		mtx_unlock(&ie->ie_lock);
836 		free(handler, M_ITHREAD);
837 		return (0);
838 	}
839 
840 	/*
841 	 * Let the interrupt thread do the job.
842 	 * The interrupt source is disabled when the interrupt thread is
843 	 * running, so it does not have to worry about interaction with
844 	 * intr_event_handle().
845 	 */
846 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
847 	    ("duplicate handle remove"));
848 	handler->ih_flags |= IH_DEAD;
849 	intr_event_schedule_thread(ie);
850 	while (handler->ih_flags & IH_DEAD)
851 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
852 	intr_event_update(ie);
853 
854 #ifdef notyet
855 	/*
856 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
857 	 * this could lead to races of stale data when servicing an
858 	 * interrupt.
859 	 */
860 	dead = 1;
861 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
862 		if (ih->ih_handler != NULL) {
863 			dead = 0;
864 			break;
865 		}
866 	}
867 	if (dead) {
868 		ithread_destroy(ie->ie_thread);
869 		ie->ie_thread = NULL;
870 	}
871 #endif
872 	mtx_unlock(&ie->ie_lock);
873 	free(handler, M_ITHREAD);
874 	return (0);
875 }
876 
877 int
878 intr_event_suspend_handler(void *cookie)
879 {
880 	struct intr_handler *handler = (struct intr_handler *)cookie;
881 	struct intr_event *ie;
882 
883 	if (handler == NULL)
884 		return (EINVAL);
885 	ie = handler->ih_event;
886 	KASSERT(ie != NULL,
887 	    ("interrupt handler \"%s\" has a NULL interrupt event",
888 	    handler->ih_name));
889 	mtx_lock(&ie->ie_lock);
890 	handler->ih_flags |= IH_SUSP;
891 	intr_handler_barrier(handler);
892 	mtx_unlock(&ie->ie_lock);
893 	return (0);
894 }
895 
896 int
897 intr_event_resume_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 
909 	/*
910 	 * intr_handler_barrier() acts not only as a barrier,
911 	 * it also allows to check for any pending interrupts.
912 	 */
913 	mtx_lock(&ie->ie_lock);
914 	handler->ih_flags &= ~IH_SUSP;
915 	intr_handler_barrier(handler);
916 	mtx_unlock(&ie->ie_lock);
917 	return (0);
918 }
919 
920 static int
921 intr_event_schedule_thread(struct intr_event *ie)
922 {
923 	struct intr_entropy entropy;
924 	struct intr_thread *it;
925 	struct thread *td;
926 	struct thread *ctd;
927 
928 	/*
929 	 * If no ithread or no handlers, then we have a stray interrupt.
930 	 */
931 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
932 	    ie->ie_thread == NULL)
933 		return (EINVAL);
934 
935 	ctd = curthread;
936 	it = ie->ie_thread;
937 	td = it->it_thread;
938 
939 	/*
940 	 * If any of the handlers for this ithread claim to be good
941 	 * sources of entropy, then gather some.
942 	 */
943 	if (ie->ie_flags & IE_ENTROPY) {
944 		entropy.event = (uintptr_t)ie;
945 		entropy.td = ctd;
946 		random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
947 	}
948 
949 	KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
950 
951 	/*
952 	 * Set it_need to tell the thread to keep running if it is already
953 	 * running.  Then, lock the thread and see if we actually need to
954 	 * put it on the runqueue.
955 	 *
956 	 * Use store_rel to arrange that the store to ih_need in
957 	 * swi_sched() is before the store to it_need and prepare for
958 	 * transfer of this order to loads in the ithread.
959 	 */
960 	atomic_store_rel_int(&it->it_need, 1);
961 	thread_lock(td);
962 	if (TD_AWAITING_INTR(td)) {
963 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
964 		    td->td_name);
965 		TD_CLR_IWAIT(td);
966 		sched_add(td, SRQ_INTR);
967 	} else {
968 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
969 		    __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
970 	}
971 	thread_unlock(td);
972 
973 	return (0);
974 }
975 
976 /*
977  * Allow interrupt event binding for software interrupt handlers -- a no-op,
978  * since interrupts are generated in software rather than being directed by
979  * a PIC.
980  */
981 static int
982 swi_assign_cpu(void *arg, int cpu)
983 {
984 
985 	return (0);
986 }
987 
988 /*
989  * Add a software interrupt handler to a specified event.  If a given event
990  * is not specified, then a new event is created.
991  */
992 int
993 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
994 	    void *arg, int pri, enum intr_type flags, void **cookiep)
995 {
996 	struct intr_event *ie;
997 	int error;
998 
999 	if (flags & INTR_ENTROPY)
1000 		return (EINVAL);
1001 
1002 	ie = (eventp != NULL) ? *eventp : NULL;
1003 
1004 	if (ie != NULL) {
1005 		if (!(ie->ie_flags & IE_SOFT))
1006 			return (EINVAL);
1007 	} else {
1008 		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1009 		    NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1010 		if (error)
1011 			return (error);
1012 		if (eventp != NULL)
1013 			*eventp = ie;
1014 	}
1015 	error = intr_event_add_handler(ie, name, NULL, handler, arg,
1016 	    PI_SWI(pri), flags, cookiep);
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Schedule a software interrupt thread.
1022  */
1023 void
1024 swi_sched(void *cookie, int flags)
1025 {
1026 	struct intr_handler *ih = (struct intr_handler *)cookie;
1027 	struct intr_event *ie = ih->ih_event;
1028 	struct intr_entropy entropy;
1029 	int error __unused;
1030 
1031 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1032 	    ih->ih_need);
1033 
1034 	entropy.event = (uintptr_t)ih;
1035 	entropy.td = curthread;
1036 	random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1037 
1038 	/*
1039 	 * Set ih_need for this handler so that if the ithread is already
1040 	 * running it will execute this handler on the next pass.  Otherwise,
1041 	 * it will execute it the next time it runs.
1042 	 */
1043 	ih->ih_need = 1;
1044 
1045 	if (!(flags & SWI_DELAY)) {
1046 		VM_CNT_INC(v_soft);
1047 		error = intr_event_schedule_thread(ie);
1048 		KASSERT(error == 0, ("stray software interrupt"));
1049 	}
1050 }
1051 
1052 /*
1053  * Remove a software interrupt handler.  Currently this code does not
1054  * remove the associated interrupt event if it becomes empty.  Calling code
1055  * may do so manually via intr_event_destroy(), but that's not really
1056  * an optimal interface.
1057  */
1058 int
1059 swi_remove(void *cookie)
1060 {
1061 
1062 	return (intr_event_remove_handler(cookie));
1063 }
1064 
1065 static void
1066 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1067 {
1068 	struct intr_handler *ih, *ihn, *ihp;
1069 
1070 	ihp = NULL;
1071 	CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1072 		/*
1073 		 * If this handler is marked for death, remove it from
1074 		 * the list of handlers and wake up the sleeper.
1075 		 */
1076 		if (ih->ih_flags & IH_DEAD) {
1077 			mtx_lock(&ie->ie_lock);
1078 			if (ihp == NULL)
1079 				CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1080 			else
1081 				CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1082 			ih->ih_flags &= ~IH_DEAD;
1083 			wakeup(ih);
1084 			mtx_unlock(&ie->ie_lock);
1085 			continue;
1086 		}
1087 
1088 		/*
1089 		 * Now that we know that the current element won't be removed
1090 		 * update the previous element.
1091 		 */
1092 		ihp = ih;
1093 
1094 		if ((ih->ih_flags & IH_CHANGED) != 0) {
1095 			mtx_lock(&ie->ie_lock);
1096 			ih->ih_flags &= ~IH_CHANGED;
1097 			wakeup(ih);
1098 			mtx_unlock(&ie->ie_lock);
1099 		}
1100 
1101 		/* Skip filter only handlers */
1102 		if (ih->ih_handler == NULL)
1103 			continue;
1104 
1105 		/* Skip suspended handlers */
1106 		if ((ih->ih_flags & IH_SUSP) != 0)
1107 			continue;
1108 
1109 		/*
1110 		 * For software interrupt threads, we only execute
1111 		 * handlers that have their need flag set.  Hardware
1112 		 * interrupt threads always invoke all of their handlers.
1113 		 *
1114 		 * ih_need can only be 0 or 1.  Failed cmpset below
1115 		 * means that there is no request to execute handlers,
1116 		 * so a retry of the cmpset is not needed.
1117 		 */
1118 		if ((ie->ie_flags & IE_SOFT) != 0 &&
1119 		    atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1120 			continue;
1121 
1122 		/* Execute this handler. */
1123 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1124 		    __func__, p->p_pid, (void *)ih->ih_handler,
1125 		    ih->ih_argument, ih->ih_name, ih->ih_flags);
1126 
1127 		if (!(ih->ih_flags & IH_MPSAFE))
1128 			mtx_lock(&Giant);
1129 		ih->ih_handler(ih->ih_argument);
1130 		if (!(ih->ih_flags & IH_MPSAFE))
1131 			mtx_unlock(&Giant);
1132 	}
1133 }
1134 
1135 static void
1136 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1137 {
1138 
1139 	/* Interrupt handlers should not sleep. */
1140 	if (!(ie->ie_flags & IE_SOFT))
1141 		THREAD_NO_SLEEPING();
1142 	intr_event_execute_handlers(p, ie);
1143 	if (!(ie->ie_flags & IE_SOFT))
1144 		THREAD_SLEEPING_OK();
1145 
1146 	/*
1147 	 * Interrupt storm handling:
1148 	 *
1149 	 * If this interrupt source is currently storming, then throttle
1150 	 * it to only fire the handler once  per clock tick.
1151 	 *
1152 	 * If this interrupt source is not currently storming, but the
1153 	 * number of back to back interrupts exceeds the storm threshold,
1154 	 * then enter storming mode.
1155 	 */
1156 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1157 	    !(ie->ie_flags & IE_SOFT)) {
1158 		/* Report the message only once every second. */
1159 		if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1160 			printf(
1161 	"interrupt storm detected on \"%s\"; throttling interrupt source\n",
1162 			    ie->ie_name);
1163 		}
1164 		pause("istorm", 1);
1165 	} else
1166 		ie->ie_count++;
1167 
1168 	/*
1169 	 * Now that all the handlers have had a chance to run, reenable
1170 	 * the interrupt source.
1171 	 */
1172 	if (ie->ie_post_ithread != NULL)
1173 		ie->ie_post_ithread(ie->ie_source);
1174 }
1175 
1176 /*
1177  * This is the main code for interrupt threads.
1178  */
1179 static void
1180 ithread_loop(void *arg)
1181 {
1182 	struct intr_thread *ithd;
1183 	struct intr_event *ie;
1184 	struct thread *td;
1185 	struct proc *p;
1186 	int wake;
1187 
1188 	td = curthread;
1189 	p = td->td_proc;
1190 	ithd = (struct intr_thread *)arg;
1191 	KASSERT(ithd->it_thread == td,
1192 	    ("%s: ithread and proc linkage out of sync", __func__));
1193 	ie = ithd->it_event;
1194 	ie->ie_count = 0;
1195 	wake = 0;
1196 
1197 	/*
1198 	 * As long as we have interrupts outstanding, go through the
1199 	 * list of handlers, giving each one a go at it.
1200 	 */
1201 	for (;;) {
1202 		/*
1203 		 * If we are an orphaned thread, then just die.
1204 		 */
1205 		if (ithd->it_flags & IT_DEAD) {
1206 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1207 			    p->p_pid, td->td_name);
1208 			free(ithd, M_ITHREAD);
1209 			kthread_exit();
1210 		}
1211 
1212 		/*
1213 		 * Service interrupts.  If another interrupt arrives while
1214 		 * we are running, it will set it_need to note that we
1215 		 * should make another pass.
1216 		 *
1217 		 * The load_acq part of the following cmpset ensures
1218 		 * that the load of ih_need in ithread_execute_handlers()
1219 		 * is ordered after the load of it_need here.
1220 		 */
1221 		while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0)
1222 			ithread_execute_handlers(p, ie);
1223 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1224 		mtx_assert(&Giant, MA_NOTOWNED);
1225 
1226 		/*
1227 		 * Processed all our interrupts.  Now get the sched
1228 		 * lock.  This may take a while and it_need may get
1229 		 * set again, so we have to check it again.
1230 		 */
1231 		thread_lock(td);
1232 		if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1233 		    (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1234 			TD_SET_IWAIT(td);
1235 			ie->ie_count = 0;
1236 			mi_switch(SW_VOL | SWT_IWAIT, NULL);
1237 		}
1238 		if (ithd->it_flags & IT_WAIT) {
1239 			wake = 1;
1240 			ithd->it_flags &= ~IT_WAIT;
1241 		}
1242 		thread_unlock(td);
1243 		if (wake) {
1244 			wakeup(ithd);
1245 			wake = 0;
1246 		}
1247 	}
1248 }
1249 
1250 /*
1251  * Main interrupt handling body.
1252  *
1253  * Input:
1254  * o ie:                        the event connected to this interrupt.
1255  * o frame:                     some archs (i.e. i386) pass a frame to some.
1256  *                              handlers as their main argument.
1257  * Return value:
1258  * o 0:                         everything ok.
1259  * o EINVAL:                    stray interrupt.
1260  */
1261 int
1262 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1263 {
1264 	struct intr_handler *ih;
1265 	struct trapframe *oldframe;
1266 	struct thread *td;
1267 	int phase;
1268 	int ret;
1269 	bool filter, thread;
1270 
1271 	td = curthread;
1272 
1273 #ifdef KSTACK_USAGE_PROF
1274 	intr_prof_stack_use(td, frame);
1275 #endif
1276 
1277 	/* An interrupt with no event or handlers is a stray interrupt. */
1278 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
1279 		return (EINVAL);
1280 
1281 	/*
1282 	 * Execute fast interrupt handlers directly.
1283 	 * To support clock handlers, if a handler registers
1284 	 * with a NULL argument, then we pass it a pointer to
1285 	 * a trapframe as its argument.
1286 	 */
1287 	td->td_intr_nesting_level++;
1288 	filter = false;
1289 	thread = false;
1290 	ret = 0;
1291 	critical_enter();
1292 	oldframe = td->td_intr_frame;
1293 	td->td_intr_frame = frame;
1294 
1295 	phase = ie->ie_phase;
1296 	atomic_add_int(&ie->ie_active[phase], 1);
1297 
1298 	/*
1299 	 * This fence is required to ensure that no later loads are
1300 	 * re-ordered before the ie_active store.
1301 	 */
1302 	atomic_thread_fence_seq_cst();
1303 
1304 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
1305 		if ((ih->ih_flags & IH_SUSP) != 0)
1306 			continue;
1307 		if (ih->ih_filter == NULL) {
1308 			thread = true;
1309 			continue;
1310 		}
1311 		CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1312 		    ih->ih_filter, ih->ih_argument == NULL ? frame :
1313 		    ih->ih_argument, ih->ih_name);
1314 		if (ih->ih_argument == NULL)
1315 			ret = ih->ih_filter(frame);
1316 		else
1317 			ret = ih->ih_filter(ih->ih_argument);
1318 		KASSERT(ret == FILTER_STRAY ||
1319 		    ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1320 		    (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1321 		    ("%s: incorrect return value %#x from %s", __func__, ret,
1322 		    ih->ih_name));
1323 		filter = filter || ret == FILTER_HANDLED;
1324 
1325 		/*
1326 		 * Wrapper handler special handling:
1327 		 *
1328 		 * in some particular cases (like pccard and pccbb),
1329 		 * the _real_ device handler is wrapped in a couple of
1330 		 * functions - a filter wrapper and an ithread wrapper.
1331 		 * In this case (and just in this case), the filter wrapper
1332 		 * could ask the system to schedule the ithread and mask
1333 		 * the interrupt source if the wrapped handler is composed
1334 		 * of just an ithread handler.
1335 		 *
1336 		 * TODO: write a generic wrapper to avoid people rolling
1337 		 * their own.
1338 		 */
1339 		if (!thread) {
1340 			if (ret == FILTER_SCHEDULE_THREAD)
1341 				thread = true;
1342 		}
1343 	}
1344 	atomic_add_rel_int(&ie->ie_active[phase], -1);
1345 
1346 	td->td_intr_frame = oldframe;
1347 
1348 	if (thread) {
1349 		if (ie->ie_pre_ithread != NULL)
1350 			ie->ie_pre_ithread(ie->ie_source);
1351 	} else {
1352 		if (ie->ie_post_filter != NULL)
1353 			ie->ie_post_filter(ie->ie_source);
1354 	}
1355 
1356 	/* Schedule the ithread if needed. */
1357 	if (thread) {
1358 		int error __unused;
1359 
1360 		error =  intr_event_schedule_thread(ie);
1361 		KASSERT(error == 0, ("bad stray interrupt"));
1362 	}
1363 	critical_exit();
1364 	td->td_intr_nesting_level--;
1365 #ifdef notyet
1366 	/* The interrupt is not aknowledged by any filter and has no ithread. */
1367 	if (!thread && !filter)
1368 		return (EINVAL);
1369 #endif
1370 	return (0);
1371 }
1372 
1373 #ifdef DDB
1374 /*
1375  * Dump details about an interrupt handler
1376  */
1377 static void
1378 db_dump_intrhand(struct intr_handler *ih)
1379 {
1380 	int comma;
1381 
1382 	db_printf("\t%-10s ", ih->ih_name);
1383 	switch (ih->ih_pri) {
1384 	case PI_REALTIME:
1385 		db_printf("CLK ");
1386 		break;
1387 	case PI_AV:
1388 		db_printf("AV  ");
1389 		break;
1390 	case PI_TTY:
1391 		db_printf("TTY ");
1392 		break;
1393 	case PI_NET:
1394 		db_printf("NET ");
1395 		break;
1396 	case PI_DISK:
1397 		db_printf("DISK");
1398 		break;
1399 	case PI_DULL:
1400 		db_printf("DULL");
1401 		break;
1402 	default:
1403 		if (ih->ih_pri >= PI_SOFT)
1404 			db_printf("SWI ");
1405 		else
1406 			db_printf("%4u", ih->ih_pri);
1407 		break;
1408 	}
1409 	db_printf(" ");
1410 	if (ih->ih_filter != NULL) {
1411 		db_printf("[F]");
1412 		db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
1413 	}
1414 	if (ih->ih_handler != NULL) {
1415 		if (ih->ih_filter != NULL)
1416 			db_printf(",");
1417 		db_printf("[H]");
1418 		db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1419 	}
1420 	db_printf("(%p)", ih->ih_argument);
1421 	if (ih->ih_need ||
1422 	    (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1423 	    IH_MPSAFE)) != 0) {
1424 		db_printf(" {");
1425 		comma = 0;
1426 		if (ih->ih_flags & IH_EXCLUSIVE) {
1427 			if (comma)
1428 				db_printf(", ");
1429 			db_printf("EXCL");
1430 			comma = 1;
1431 		}
1432 		if (ih->ih_flags & IH_ENTROPY) {
1433 			if (comma)
1434 				db_printf(", ");
1435 			db_printf("ENTROPY");
1436 			comma = 1;
1437 		}
1438 		if (ih->ih_flags & IH_DEAD) {
1439 			if (comma)
1440 				db_printf(", ");
1441 			db_printf("DEAD");
1442 			comma = 1;
1443 		}
1444 		if (ih->ih_flags & IH_MPSAFE) {
1445 			if (comma)
1446 				db_printf(", ");
1447 			db_printf("MPSAFE");
1448 			comma = 1;
1449 		}
1450 		if (ih->ih_need) {
1451 			if (comma)
1452 				db_printf(", ");
1453 			db_printf("NEED");
1454 		}
1455 		db_printf("}");
1456 	}
1457 	db_printf("\n");
1458 }
1459 
1460 /*
1461  * Dump details about a event.
1462  */
1463 void
1464 db_dump_intr_event(struct intr_event *ie, int handlers)
1465 {
1466 	struct intr_handler *ih;
1467 	struct intr_thread *it;
1468 	int comma;
1469 
1470 	db_printf("%s ", ie->ie_fullname);
1471 	it = ie->ie_thread;
1472 	if (it != NULL)
1473 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1474 	else
1475 		db_printf("(no thread)");
1476 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
1477 	    (it != NULL && it->it_need)) {
1478 		db_printf(" {");
1479 		comma = 0;
1480 		if (ie->ie_flags & IE_SOFT) {
1481 			db_printf("SOFT");
1482 			comma = 1;
1483 		}
1484 		if (ie->ie_flags & IE_ENTROPY) {
1485 			if (comma)
1486 				db_printf(", ");
1487 			db_printf("ENTROPY");
1488 			comma = 1;
1489 		}
1490 		if (ie->ie_flags & IE_ADDING_THREAD) {
1491 			if (comma)
1492 				db_printf(", ");
1493 			db_printf("ADDING_THREAD");
1494 			comma = 1;
1495 		}
1496 		if (it != NULL && it->it_need) {
1497 			if (comma)
1498 				db_printf(", ");
1499 			db_printf("NEED");
1500 		}
1501 		db_printf("}");
1502 	}
1503 	db_printf("\n");
1504 
1505 	if (handlers)
1506 		CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
1507 		    db_dump_intrhand(ih);
1508 }
1509 
1510 /*
1511  * Dump data about interrupt handlers
1512  */
1513 DB_SHOW_COMMAND(intr, db_show_intr)
1514 {
1515 	struct intr_event *ie;
1516 	int all, verbose;
1517 
1518 	verbose = strchr(modif, 'v') != NULL;
1519 	all = strchr(modif, 'a') != NULL;
1520 	TAILQ_FOREACH(ie, &event_list, ie_list) {
1521 		if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
1522 			continue;
1523 		db_dump_intr_event(ie, verbose);
1524 		if (db_pager_quit)
1525 			break;
1526 	}
1527 }
1528 #endif /* DDB */
1529 
1530 /*
1531  * Start standard software interrupt threads
1532  */
1533 static void
1534 start_softintr(void *dummy)
1535 {
1536 
1537 	if (swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
1538 		panic("died while creating vm swi ithread");
1539 }
1540 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1541     NULL);
1542 
1543 /*
1544  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1545  * The data for this machine dependent, and the declarations are in machine
1546  * dependent code.  The layout of intrnames and intrcnt however is machine
1547  * independent.
1548  *
1549  * We do not know the length of intrcnt and intrnames at compile time, so
1550  * calculate things at run time.
1551  */
1552 static int
1553 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1554 {
1555 	return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1556 }
1557 
1558 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1559     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1560 
1561 static int
1562 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1563 {
1564 #ifdef SCTL_MASK32
1565 	uint32_t *intrcnt32;
1566 	unsigned i;
1567 	int error;
1568 
1569 	if (req->flags & SCTL_MASK32) {
1570 		if (!req->oldptr)
1571 			return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
1572 		intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
1573 		if (intrcnt32 == NULL)
1574 			return (ENOMEM);
1575 		for (i = 0; i < sintrcnt / sizeof (u_long); i++)
1576 			intrcnt32[i] = intrcnt[i];
1577 		error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
1578 		free(intrcnt32, M_TEMP);
1579 		return (error);
1580 	}
1581 #endif
1582 	return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1583 }
1584 
1585 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1586     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1587 
1588 #ifdef DDB
1589 /*
1590  * DDB command to dump the interrupt statistics.
1591  */
1592 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
1593 {
1594 	u_long *i;
1595 	char *cp;
1596 	u_int j;
1597 
1598 	cp = intrnames;
1599 	j = 0;
1600 	for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1601 	    i++, j++) {
1602 		if (*cp == '\0')
1603 			break;
1604 		if (*i != 0)
1605 			db_printf("%s\t%lu\n", cp, *i);
1606 		cp += strlen(cp) + 1;
1607 	}
1608 }
1609 #endif
1610