xref: /freebsd/sys/kern/kern_intr.c (revision 0957b409)
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 the handler names were too long, add +'s to indicate missing
213 	 * names. If we run out of room and still have +'s to add, change
214 	 * the last character from a + to a *.
215 	 */
216 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
217 	while (missed-- > 0) {
218 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
219 			if (*last == '+') {
220 				*last = '*';
221 				break;
222 			} else
223 				*last = '+';
224 		} else if (space) {
225 			strcat(ie->ie_fullname, " +");
226 			space = 0;
227 		} else
228 			strcat(ie->ie_fullname, "+");
229 	}
230 
231 	/*
232 	 * If this event has an ithread, update it's priority and
233 	 * name.
234 	 */
235 	if (ie->ie_thread != NULL)
236 		ithread_update(ie->ie_thread);
237 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
238 }
239 
240 int
241 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
242     void (*pre_ithread)(void *), void (*post_ithread)(void *),
243     void (*post_filter)(void *), int (*assign_cpu)(void *, int),
244     const char *fmt, ...)
245 {
246 	struct intr_event *ie;
247 	va_list ap;
248 
249 	/* The only valid flag during creation is IE_SOFT. */
250 	if ((flags & ~IE_SOFT) != 0)
251 		return (EINVAL);
252 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
253 	ie->ie_source = source;
254 	ie->ie_pre_ithread = pre_ithread;
255 	ie->ie_post_ithread = post_ithread;
256 	ie->ie_post_filter = post_filter;
257 	ie->ie_assign_cpu = assign_cpu;
258 	ie->ie_flags = flags;
259 	ie->ie_irq = irq;
260 	ie->ie_cpu = NOCPU;
261 	CK_SLIST_INIT(&ie->ie_handlers);
262 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
263 
264 	va_start(ap, fmt);
265 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
266 	va_end(ap);
267 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
268 	mtx_lock(&event_lock);
269 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
270 	mtx_unlock(&event_lock);
271 	if (event != NULL)
272 		*event = ie;
273 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
274 	return (0);
275 }
276 
277 /*
278  * Bind an interrupt event to the specified CPU.  Note that not all
279  * platforms support binding an interrupt to a CPU.  For those
280  * platforms this request will fail.  Using a cpu id of NOCPU unbinds
281  * the interrupt event.
282  */
283 static int
284 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
285 {
286 	lwpid_t id;
287 	int error;
288 
289 	/* Need a CPU to bind to. */
290 	if (cpu != NOCPU && CPU_ABSENT(cpu))
291 		return (EINVAL);
292 
293 	if (ie->ie_assign_cpu == NULL)
294 		return (EOPNOTSUPP);
295 
296 	error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
297 	if (error)
298 		return (error);
299 
300 	/*
301 	 * If we have any ithreads try to set their mask first to verify
302 	 * permissions, etc.
303 	 */
304 	if (bindithread) {
305 		mtx_lock(&ie->ie_lock);
306 		if (ie->ie_thread != NULL) {
307 			id = ie->ie_thread->it_thread->td_tid;
308 			mtx_unlock(&ie->ie_lock);
309 			error = cpuset_setithread(id, cpu);
310 			if (error)
311 				return (error);
312 		} else
313 			mtx_unlock(&ie->ie_lock);
314 	}
315 	if (bindirq)
316 		error = ie->ie_assign_cpu(ie->ie_source, cpu);
317 	if (error) {
318 		if (bindithread) {
319 			mtx_lock(&ie->ie_lock);
320 			if (ie->ie_thread != NULL) {
321 				cpu = ie->ie_cpu;
322 				id = ie->ie_thread->it_thread->td_tid;
323 				mtx_unlock(&ie->ie_lock);
324 				(void)cpuset_setithread(id, cpu);
325 			} else
326 				mtx_unlock(&ie->ie_lock);
327 		}
328 		return (error);
329 	}
330 
331 	if (bindirq) {
332 		mtx_lock(&ie->ie_lock);
333 		ie->ie_cpu = cpu;
334 		mtx_unlock(&ie->ie_lock);
335 	}
336 
337 	return (error);
338 }
339 
340 /*
341  * Bind an interrupt event to the specified CPU.  For supported platforms, any
342  * associated ithreads as well as the primary interrupt context will be bound
343  * to the specificed CPU.
344  */
345 int
346 intr_event_bind(struct intr_event *ie, int cpu)
347 {
348 
349 	return (_intr_event_bind(ie, cpu, true, true));
350 }
351 
352 /*
353  * Bind an interrupt event to the specified CPU, but do not bind associated
354  * ithreads.
355  */
356 int
357 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
358 {
359 
360 	return (_intr_event_bind(ie, cpu, true, false));
361 }
362 
363 /*
364  * Bind an interrupt event's ithread to the specified CPU.
365  */
366 int
367 intr_event_bind_ithread(struct intr_event *ie, int cpu)
368 {
369 
370 	return (_intr_event_bind(ie, cpu, false, true));
371 }
372 
373 static struct intr_event *
374 intr_lookup(int irq)
375 {
376 	struct intr_event *ie;
377 
378 	mtx_lock(&event_lock);
379 	TAILQ_FOREACH(ie, &event_list, ie_list)
380 		if (ie->ie_irq == irq &&
381 		    (ie->ie_flags & IE_SOFT) == 0 &&
382 		    CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
383 			break;
384 	mtx_unlock(&event_lock);
385 	return (ie);
386 }
387 
388 int
389 intr_setaffinity(int irq, int mode, void *m)
390 {
391 	struct intr_event *ie;
392 	cpuset_t *mask;
393 	int cpu, n;
394 
395 	mask = m;
396 	cpu = NOCPU;
397 	/*
398 	 * If we're setting all cpus we can unbind.  Otherwise make sure
399 	 * only one cpu is in the set.
400 	 */
401 	if (CPU_CMP(cpuset_root, mask)) {
402 		for (n = 0; n < CPU_SETSIZE; n++) {
403 			if (!CPU_ISSET(n, mask))
404 				continue;
405 			if (cpu != NOCPU)
406 				return (EINVAL);
407 			cpu = n;
408 		}
409 	}
410 	ie = intr_lookup(irq);
411 	if (ie == NULL)
412 		return (ESRCH);
413 	switch (mode) {
414 	case CPU_WHICH_IRQ:
415 		return (intr_event_bind(ie, cpu));
416 	case CPU_WHICH_INTRHANDLER:
417 		return (intr_event_bind_irqonly(ie, cpu));
418 	case CPU_WHICH_ITHREAD:
419 		return (intr_event_bind_ithread(ie, cpu));
420 	default:
421 		return (EINVAL);
422 	}
423 }
424 
425 int
426 intr_getaffinity(int irq, int mode, void *m)
427 {
428 	struct intr_event *ie;
429 	struct thread *td;
430 	struct proc *p;
431 	cpuset_t *mask;
432 	lwpid_t id;
433 	int error;
434 
435 	mask = m;
436 	ie = intr_lookup(irq);
437 	if (ie == NULL)
438 		return (ESRCH);
439 
440 	error = 0;
441 	CPU_ZERO(mask);
442 	switch (mode) {
443 	case CPU_WHICH_IRQ:
444 	case CPU_WHICH_INTRHANDLER:
445 		mtx_lock(&ie->ie_lock);
446 		if (ie->ie_cpu == NOCPU)
447 			CPU_COPY(cpuset_root, mask);
448 		else
449 			CPU_SET(ie->ie_cpu, mask);
450 		mtx_unlock(&ie->ie_lock);
451 		break;
452 	case CPU_WHICH_ITHREAD:
453 		mtx_lock(&ie->ie_lock);
454 		if (ie->ie_thread == NULL) {
455 			mtx_unlock(&ie->ie_lock);
456 			CPU_COPY(cpuset_root, mask);
457 		} else {
458 			id = ie->ie_thread->it_thread->td_tid;
459 			mtx_unlock(&ie->ie_lock);
460 			error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
461 			if (error != 0)
462 				return (error);
463 			CPU_COPY(&td->td_cpuset->cs_mask, mask);
464 			PROC_UNLOCK(p);
465 		}
466 	default:
467 		return (EINVAL);
468 	}
469 	return (0);
470 }
471 
472 int
473 intr_event_destroy(struct intr_event *ie)
474 {
475 
476 	mtx_lock(&event_lock);
477 	mtx_lock(&ie->ie_lock);
478 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
479 		mtx_unlock(&ie->ie_lock);
480 		mtx_unlock(&event_lock);
481 		return (EBUSY);
482 	}
483 	TAILQ_REMOVE(&event_list, ie, ie_list);
484 #ifndef notyet
485 	if (ie->ie_thread != NULL) {
486 		ithread_destroy(ie->ie_thread);
487 		ie->ie_thread = NULL;
488 	}
489 #endif
490 	mtx_unlock(&ie->ie_lock);
491 	mtx_unlock(&event_lock);
492 	mtx_destroy(&ie->ie_lock);
493 	free(ie, M_ITHREAD);
494 	return (0);
495 }
496 
497 static struct intr_thread *
498 ithread_create(const char *name)
499 {
500 	struct intr_thread *ithd;
501 	struct thread *td;
502 	int error;
503 
504 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
505 
506 	error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
507 		    &td, RFSTOPPED | RFHIGHPID,
508 		    0, "intr", "%s", name);
509 	if (error)
510 		panic("kproc_create() failed with %d", error);
511 	thread_lock(td);
512 	sched_class(td, PRI_ITHD);
513 	TD_SET_IWAIT(td);
514 	thread_unlock(td);
515 	td->td_pflags |= TDP_ITHREAD;
516 	ithd->it_thread = td;
517 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
518 	return (ithd);
519 }
520 
521 static void
522 ithread_destroy(struct intr_thread *ithread)
523 {
524 	struct thread *td;
525 
526 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
527 	td = ithread->it_thread;
528 	thread_lock(td);
529 	ithread->it_flags |= IT_DEAD;
530 	if (TD_AWAITING_INTR(td)) {
531 		TD_CLR_IWAIT(td);
532 		sched_add(td, SRQ_INTR);
533 	}
534 	thread_unlock(td);
535 }
536 
537 int
538 intr_event_add_handler(struct intr_event *ie, const char *name,
539     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
540     enum intr_type flags, void **cookiep)
541 {
542 	struct intr_handler *ih, *temp_ih;
543 	struct intr_handler **prevptr;
544 	struct intr_thread *it;
545 
546 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
547 		return (EINVAL);
548 
549 	/* Allocate and populate an interrupt handler structure. */
550 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
551 	ih->ih_filter = filter;
552 	ih->ih_handler = handler;
553 	ih->ih_argument = arg;
554 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
555 	ih->ih_event = ie;
556 	ih->ih_pri = pri;
557 	if (flags & INTR_EXCL)
558 		ih->ih_flags = IH_EXCLUSIVE;
559 	if (flags & INTR_MPSAFE)
560 		ih->ih_flags |= IH_MPSAFE;
561 	if (flags & INTR_ENTROPY)
562 		ih->ih_flags |= IH_ENTROPY;
563 
564 	/* We can only have one exclusive handler in a event. */
565 	mtx_lock(&ie->ie_lock);
566 	if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
567 		if ((flags & INTR_EXCL) ||
568 		    (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
569 			mtx_unlock(&ie->ie_lock);
570 			free(ih, M_ITHREAD);
571 			return (EINVAL);
572 		}
573 	}
574 
575 	/* Create a thread if we need one. */
576 	while (ie->ie_thread == NULL && handler != NULL) {
577 		if (ie->ie_flags & IE_ADDING_THREAD)
578 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
579 		else {
580 			ie->ie_flags |= IE_ADDING_THREAD;
581 			mtx_unlock(&ie->ie_lock);
582 			it = ithread_create("intr: newborn");
583 			mtx_lock(&ie->ie_lock);
584 			ie->ie_flags &= ~IE_ADDING_THREAD;
585 			ie->ie_thread = it;
586 			it->it_event = ie;
587 			ithread_update(it);
588 			wakeup(ie);
589 		}
590 	}
591 
592 	/* Add the new handler to the event in priority order. */
593 	CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
594 		if (temp_ih->ih_pri > ih->ih_pri)
595 			break;
596 	}
597 	CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
598 
599 	intr_event_update(ie);
600 
601 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
602 	    ie->ie_name);
603 	mtx_unlock(&ie->ie_lock);
604 
605 	if (cookiep != NULL)
606 		*cookiep = ih;
607 	return (0);
608 }
609 
610 /*
611  * Append a description preceded by a ':' to the name of the specified
612  * interrupt handler.
613  */
614 int
615 intr_event_describe_handler(struct intr_event *ie, void *cookie,
616     const char *descr)
617 {
618 	struct intr_handler *ih;
619 	size_t space;
620 	char *start;
621 
622 	mtx_lock(&ie->ie_lock);
623 #ifdef INVARIANTS
624 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
625 		if (ih == cookie)
626 			break;
627 	}
628 	if (ih == NULL) {
629 		mtx_unlock(&ie->ie_lock);
630 		panic("handler %p not found in interrupt event %p", cookie, ie);
631 	}
632 #endif
633 	ih = cookie;
634 
635 	/*
636 	 * Look for an existing description by checking for an
637 	 * existing ":".  This assumes device names do not include
638 	 * colons.  If one is found, prepare to insert the new
639 	 * description at that point.  If one is not found, find the
640 	 * end of the name to use as the insertion point.
641 	 */
642 	start = strchr(ih->ih_name, ':');
643 	if (start == NULL)
644 		start = strchr(ih->ih_name, 0);
645 
646 	/*
647 	 * See if there is enough remaining room in the string for the
648 	 * description + ":".  The "- 1" leaves room for the trailing
649 	 * '\0'.  The "+ 1" accounts for the colon.
650 	 */
651 	space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
652 	if (strlen(descr) + 1 > space) {
653 		mtx_unlock(&ie->ie_lock);
654 		return (ENOSPC);
655 	}
656 
657 	/* Append a colon followed by the description. */
658 	*start = ':';
659 	strcpy(start + 1, descr);
660 	intr_event_update(ie);
661 	mtx_unlock(&ie->ie_lock);
662 	return (0);
663 }
664 
665 /*
666  * Return the ie_source field from the intr_event an intr_handler is
667  * associated with.
668  */
669 void *
670 intr_handler_source(void *cookie)
671 {
672 	struct intr_handler *ih;
673 	struct intr_event *ie;
674 
675 	ih = (struct intr_handler *)cookie;
676 	if (ih == NULL)
677 		return (NULL);
678 	ie = ih->ih_event;
679 	KASSERT(ie != NULL,
680 	    ("interrupt handler \"%s\" has a NULL interrupt event",
681 	    ih->ih_name));
682 	return (ie->ie_source);
683 }
684 
685 /*
686  * If intr_event_handle() is running in the ISR context at the time of the call,
687  * then wait for it to complete.
688  */
689 static void
690 intr_event_barrier(struct intr_event *ie)
691 {
692 	int phase;
693 
694 	mtx_assert(&ie->ie_lock, MA_OWNED);
695 	phase = ie->ie_phase;
696 
697 	/*
698 	 * Switch phase to direct future interrupts to the other active counter.
699 	 * Make sure that any preceding stores are visible before the switch.
700 	 */
701 	KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
702 	atomic_store_rel_int(&ie->ie_phase, !phase);
703 
704 	/*
705 	 * This code cooperates with wait-free iteration of ie_handlers
706 	 * in intr_event_handle.
707 	 * Make sure that the removal and the phase update are not reordered
708 	 * with the active count check.
709 	 * Note that no combination of acquire and release fences can provide
710 	 * that guarantee as Store->Load sequences can always be reordered.
711 	 */
712 	atomic_thread_fence_seq_cst();
713 
714 	/*
715 	 * Now wait on the inactive phase.
716 	 * The acquire fence is needed so that that all post-barrier accesses
717 	 * are after the check.
718 	 */
719 	while (ie->ie_active[phase] > 0)
720 		cpu_spinwait();
721 	atomic_thread_fence_acq();
722 }
723 
724 static void
725 intr_handler_barrier(struct intr_handler *handler)
726 {
727 	struct intr_event *ie;
728 
729 	ie = handler->ih_event;
730 	mtx_assert(&ie->ie_lock, MA_OWNED);
731 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
732 	    ("update for a removed handler"));
733 
734 	if (ie->ie_thread == NULL) {
735 		intr_event_barrier(ie);
736 		return;
737 	}
738 	if ((handler->ih_flags & IH_CHANGED) == 0) {
739 		handler->ih_flags |= IH_CHANGED;
740 		intr_event_schedule_thread(ie);
741 	}
742 	while ((handler->ih_flags & IH_CHANGED) != 0)
743 		msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
744 }
745 
746 /*
747  * Sleep until an ithread finishes executing an interrupt handler.
748  *
749  * XXX Doesn't currently handle interrupt filters or fast interrupt
750  * handlers.  This is intended for compatibility with linux drivers
751  * only.  Do not use in BSD code.
752  */
753 void
754 _intr_drain(int irq)
755 {
756 	struct intr_event *ie;
757 	struct intr_thread *ithd;
758 	struct thread *td;
759 
760 	ie = intr_lookup(irq);
761 	if (ie == NULL)
762 		return;
763 	if (ie->ie_thread == NULL)
764 		return;
765 	ithd = ie->ie_thread;
766 	td = ithd->it_thread;
767 	/*
768 	 * We set the flag and wait for it to be cleared to avoid
769 	 * long delays with potentially busy interrupt handlers
770 	 * were we to only sample TD_AWAITING_INTR() every tick.
771 	 */
772 	thread_lock(td);
773 	if (!TD_AWAITING_INTR(td)) {
774 		ithd->it_flags |= IT_WAIT;
775 		while (ithd->it_flags & IT_WAIT) {
776 			thread_unlock(td);
777 			pause("idrain", 1);
778 			thread_lock(td);
779 		}
780 	}
781 	thread_unlock(td);
782 	return;
783 }
784 
785 int
786 intr_event_remove_handler(void *cookie)
787 {
788 	struct intr_handler *handler = (struct intr_handler *)cookie;
789 	struct intr_event *ie;
790 	struct intr_handler *ih;
791 	struct intr_handler **prevptr;
792 #ifdef notyet
793 	int dead;
794 #endif
795 
796 	if (handler == NULL)
797 		return (EINVAL);
798 	ie = handler->ih_event;
799 	KASSERT(ie != NULL,
800 	    ("interrupt handler \"%s\" has a NULL interrupt event",
801 	    handler->ih_name));
802 
803 	mtx_lock(&ie->ie_lock);
804 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
805 	    ie->ie_name);
806 	CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
807 		if (ih == handler)
808 			break;
809 	}
810 	if (ih == NULL) {
811 		panic("interrupt handler \"%s\" not found in "
812 		    "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
813 	}
814 
815 	/*
816 	 * If there is no ithread, then directly remove the handler.  Note that
817 	 * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
818 	 * care needs to be taken to keep ie_handlers consistent and to free
819 	 * the removed handler only when ie_handlers is quiescent.
820 	 */
821 	if (ie->ie_thread == NULL) {
822 		CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
823 		intr_event_barrier(ie);
824 		intr_event_update(ie);
825 		mtx_unlock(&ie->ie_lock);
826 		free(handler, M_ITHREAD);
827 		return (0);
828 	}
829 
830 	/*
831 	 * Let the interrupt thread do the job.
832 	 * The interrupt source is disabled when the interrupt thread is
833 	 * running, so it does not have to worry about interaction with
834 	 * intr_event_handle().
835 	 */
836 	KASSERT((handler->ih_flags & IH_DEAD) == 0,
837 	    ("duplicate handle remove"));
838 	handler->ih_flags |= IH_DEAD;
839 	intr_event_schedule_thread(ie);
840 	while (handler->ih_flags & IH_DEAD)
841 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
842 	intr_event_update(ie);
843 
844 #ifdef notyet
845 	/*
846 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
847 	 * this could lead to races of stale data when servicing an
848 	 * interrupt.
849 	 */
850 	dead = 1;
851 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
852 		if (ih->ih_handler != NULL) {
853 			dead = 0;
854 			break;
855 		}
856 	}
857 	if (dead) {
858 		ithread_destroy(ie->ie_thread);
859 		ie->ie_thread = NULL;
860 	}
861 #endif
862 	mtx_unlock(&ie->ie_lock);
863 	free(handler, M_ITHREAD);
864 	return (0);
865 }
866 
867 int
868 intr_event_suspend_handler(void *cookie)
869 {
870 	struct intr_handler *handler = (struct intr_handler *)cookie;
871 	struct intr_event *ie;
872 
873 	if (handler == NULL)
874 		return (EINVAL);
875 	ie = handler->ih_event;
876 	KASSERT(ie != NULL,
877 	    ("interrupt handler \"%s\" has a NULL interrupt event",
878 	    handler->ih_name));
879 	mtx_lock(&ie->ie_lock);
880 	handler->ih_flags |= IH_SUSP;
881 	intr_handler_barrier(handler);
882 	mtx_unlock(&ie->ie_lock);
883 	return (0);
884 }
885 
886 int
887 intr_event_resume_handler(void *cookie)
888 {
889 	struct intr_handler *handler = (struct intr_handler *)cookie;
890 	struct intr_event *ie;
891 
892 	if (handler == NULL)
893 		return (EINVAL);
894 	ie = handler->ih_event;
895 	KASSERT(ie != NULL,
896 	    ("interrupt handler \"%s\" has a NULL interrupt event",
897 	    handler->ih_name));
898 
899 	/*
900 	 * intr_handler_barrier() acts not only as a barrier,
901 	 * it also allows to check for any pending interrupts.
902 	 */
903 	mtx_lock(&ie->ie_lock);
904 	handler->ih_flags &= ~IH_SUSP;
905 	intr_handler_barrier(handler);
906 	mtx_unlock(&ie->ie_lock);
907 	return (0);
908 }
909 
910 static int
911 intr_event_schedule_thread(struct intr_event *ie)
912 {
913 	struct intr_entropy entropy;
914 	struct intr_thread *it;
915 	struct thread *td;
916 	struct thread *ctd;
917 
918 	/*
919 	 * If no ithread or no handlers, then we have a stray interrupt.
920 	 */
921 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
922 	    ie->ie_thread == NULL)
923 		return (EINVAL);
924 
925 	ctd = curthread;
926 	it = ie->ie_thread;
927 	td = it->it_thread;
928 
929 	/*
930 	 * If any of the handlers for this ithread claim to be good
931 	 * sources of entropy, then gather some.
932 	 */
933 	if (ie->ie_flags & IE_ENTROPY) {
934 		entropy.event = (uintptr_t)ie;
935 		entropy.td = ctd;
936 		random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
937 	}
938 
939 	KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
940 
941 	/*
942 	 * Set it_need to tell the thread to keep running if it is already
943 	 * running.  Then, lock the thread and see if we actually need to
944 	 * put it on the runqueue.
945 	 *
946 	 * Use store_rel to arrange that the store to ih_need in
947 	 * swi_sched() is before the store to it_need and prepare for
948 	 * transfer of this order to loads in the ithread.
949 	 */
950 	atomic_store_rel_int(&it->it_need, 1);
951 	thread_lock(td);
952 	if (TD_AWAITING_INTR(td)) {
953 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
954 		    td->td_name);
955 		TD_CLR_IWAIT(td);
956 		sched_add(td, SRQ_INTR);
957 	} else {
958 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
959 		    __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
960 	}
961 	thread_unlock(td);
962 
963 	return (0);
964 }
965 
966 /*
967  * Allow interrupt event binding for software interrupt handlers -- a no-op,
968  * since interrupts are generated in software rather than being directed by
969  * a PIC.
970  */
971 static int
972 swi_assign_cpu(void *arg, int cpu)
973 {
974 
975 	return (0);
976 }
977 
978 /*
979  * Add a software interrupt handler to a specified event.  If a given event
980  * is not specified, then a new event is created.
981  */
982 int
983 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
984 	    void *arg, int pri, enum intr_type flags, void **cookiep)
985 {
986 	struct intr_event *ie;
987 	int error;
988 
989 	if (flags & INTR_ENTROPY)
990 		return (EINVAL);
991 
992 	ie = (eventp != NULL) ? *eventp : NULL;
993 
994 	if (ie != NULL) {
995 		if (!(ie->ie_flags & IE_SOFT))
996 			return (EINVAL);
997 	} else {
998 		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
999 		    NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1000 		if (error)
1001 			return (error);
1002 		if (eventp != NULL)
1003 			*eventp = ie;
1004 	}
1005 	error = intr_event_add_handler(ie, name, NULL, handler, arg,
1006 	    PI_SWI(pri), flags, cookiep);
1007 	return (error);
1008 }
1009 
1010 /*
1011  * Schedule a software interrupt thread.
1012  */
1013 void
1014 swi_sched(void *cookie, int flags)
1015 {
1016 	struct intr_handler *ih = (struct intr_handler *)cookie;
1017 	struct intr_event *ie = ih->ih_event;
1018 	struct intr_entropy entropy;
1019 	int error __unused;
1020 
1021 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1022 	    ih->ih_need);
1023 
1024 	entropy.event = (uintptr_t)ih;
1025 	entropy.td = curthread;
1026 	random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1027 
1028 	/*
1029 	 * Set ih_need for this handler so that if the ithread is already
1030 	 * running it will execute this handler on the next pass.  Otherwise,
1031 	 * it will execute it the next time it runs.
1032 	 */
1033 	ih->ih_need = 1;
1034 
1035 	if (!(flags & SWI_DELAY)) {
1036 		VM_CNT_INC(v_soft);
1037 		error = intr_event_schedule_thread(ie);
1038 		KASSERT(error == 0, ("stray software interrupt"));
1039 	}
1040 }
1041 
1042 /*
1043  * Remove a software interrupt handler.  Currently this code does not
1044  * remove the associated interrupt event if it becomes empty.  Calling code
1045  * may do so manually via intr_event_destroy(), but that's not really
1046  * an optimal interface.
1047  */
1048 int
1049 swi_remove(void *cookie)
1050 {
1051 
1052 	return (intr_event_remove_handler(cookie));
1053 }
1054 
1055 static void
1056 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1057 {
1058 	struct intr_handler *ih, *ihn, *ihp;
1059 
1060 	ihp = NULL;
1061 	CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1062 		/*
1063 		 * If this handler is marked for death, remove it from
1064 		 * the list of handlers and wake up the sleeper.
1065 		 */
1066 		if (ih->ih_flags & IH_DEAD) {
1067 			mtx_lock(&ie->ie_lock);
1068 			if (ihp == NULL)
1069 				CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1070 			else
1071 				CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1072 			ih->ih_flags &= ~IH_DEAD;
1073 			wakeup(ih);
1074 			mtx_unlock(&ie->ie_lock);
1075 			continue;
1076 		}
1077 
1078 		/*
1079 		 * Now that we know that the current element won't be removed
1080 		 * update the previous element.
1081 		 */
1082 		ihp = ih;
1083 
1084 		if ((ih->ih_flags & IH_CHANGED) != 0) {
1085 			mtx_lock(&ie->ie_lock);
1086 			ih->ih_flags &= ~IH_CHANGED;
1087 			wakeup(ih);
1088 			mtx_unlock(&ie->ie_lock);
1089 		}
1090 
1091 		/* Skip filter only handlers */
1092 		if (ih->ih_handler == NULL)
1093 			continue;
1094 
1095 		/* Skip suspended handlers */
1096 		if ((ih->ih_flags & IH_SUSP) != 0)
1097 			continue;
1098 
1099 		/*
1100 		 * For software interrupt threads, we only execute
1101 		 * handlers that have their need flag set.  Hardware
1102 		 * interrupt threads always invoke all of their handlers.
1103 		 *
1104 		 * ih_need can only be 0 or 1.  Failed cmpset below
1105 		 * means that there is no request to execute handlers,
1106 		 * so a retry of the cmpset is not needed.
1107 		 */
1108 		if ((ie->ie_flags & IE_SOFT) != 0 &&
1109 		    atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1110 			continue;
1111 
1112 		/* Execute this handler. */
1113 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1114 		    __func__, p->p_pid, (void *)ih->ih_handler,
1115 		    ih->ih_argument, ih->ih_name, ih->ih_flags);
1116 
1117 		if (!(ih->ih_flags & IH_MPSAFE))
1118 			mtx_lock(&Giant);
1119 		ih->ih_handler(ih->ih_argument);
1120 		if (!(ih->ih_flags & IH_MPSAFE))
1121 			mtx_unlock(&Giant);
1122 	}
1123 }
1124 
1125 static void
1126 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1127 {
1128 
1129 	/* Interrupt handlers should not sleep. */
1130 	if (!(ie->ie_flags & IE_SOFT))
1131 		THREAD_NO_SLEEPING();
1132 	intr_event_execute_handlers(p, ie);
1133 	if (!(ie->ie_flags & IE_SOFT))
1134 		THREAD_SLEEPING_OK();
1135 
1136 	/*
1137 	 * Interrupt storm handling:
1138 	 *
1139 	 * If this interrupt source is currently storming, then throttle
1140 	 * it to only fire the handler once  per clock tick.
1141 	 *
1142 	 * If this interrupt source is not currently storming, but the
1143 	 * number of back to back interrupts exceeds the storm threshold,
1144 	 * then enter storming mode.
1145 	 */
1146 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1147 	    !(ie->ie_flags & IE_SOFT)) {
1148 		/* Report the message only once every second. */
1149 		if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1150 			printf(
1151 	"interrupt storm detected on \"%s\"; throttling interrupt source\n",
1152 			    ie->ie_name);
1153 		}
1154 		pause("istorm", 1);
1155 	} else
1156 		ie->ie_count++;
1157 
1158 	/*
1159 	 * Now that all the handlers have had a chance to run, reenable
1160 	 * the interrupt source.
1161 	 */
1162 	if (ie->ie_post_ithread != NULL)
1163 		ie->ie_post_ithread(ie->ie_source);
1164 }
1165 
1166 /*
1167  * This is the main code for interrupt threads.
1168  */
1169 static void
1170 ithread_loop(void *arg)
1171 {
1172 	struct intr_thread *ithd;
1173 	struct intr_event *ie;
1174 	struct thread *td;
1175 	struct proc *p;
1176 	int wake;
1177 
1178 	td = curthread;
1179 	p = td->td_proc;
1180 	ithd = (struct intr_thread *)arg;
1181 	KASSERT(ithd->it_thread == td,
1182 	    ("%s: ithread and proc linkage out of sync", __func__));
1183 	ie = ithd->it_event;
1184 	ie->ie_count = 0;
1185 	wake = 0;
1186 
1187 	/*
1188 	 * As long as we have interrupts outstanding, go through the
1189 	 * list of handlers, giving each one a go at it.
1190 	 */
1191 	for (;;) {
1192 		/*
1193 		 * If we are an orphaned thread, then just die.
1194 		 */
1195 		if (ithd->it_flags & IT_DEAD) {
1196 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1197 			    p->p_pid, td->td_name);
1198 			free(ithd, M_ITHREAD);
1199 			kthread_exit();
1200 		}
1201 
1202 		/*
1203 		 * Service interrupts.  If another interrupt arrives while
1204 		 * we are running, it will set it_need to note that we
1205 		 * should make another pass.
1206 		 *
1207 		 * The load_acq part of the following cmpset ensures
1208 		 * that the load of ih_need in ithread_execute_handlers()
1209 		 * is ordered after the load of it_need here.
1210 		 */
1211 		while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0)
1212 			ithread_execute_handlers(p, ie);
1213 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1214 		mtx_assert(&Giant, MA_NOTOWNED);
1215 
1216 		/*
1217 		 * Processed all our interrupts.  Now get the sched
1218 		 * lock.  This may take a while and it_need may get
1219 		 * set again, so we have to check it again.
1220 		 */
1221 		thread_lock(td);
1222 		if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1223 		    (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1224 			TD_SET_IWAIT(td);
1225 			ie->ie_count = 0;
1226 			mi_switch(SW_VOL | SWT_IWAIT, NULL);
1227 		}
1228 		if (ithd->it_flags & IT_WAIT) {
1229 			wake = 1;
1230 			ithd->it_flags &= ~IT_WAIT;
1231 		}
1232 		thread_unlock(td);
1233 		if (wake) {
1234 			wakeup(ithd);
1235 			wake = 0;
1236 		}
1237 	}
1238 }
1239 
1240 /*
1241  * Main interrupt handling body.
1242  *
1243  * Input:
1244  * o ie:                        the event connected to this interrupt.
1245  * o frame:                     some archs (i.e. i386) pass a frame to some.
1246  *                              handlers as their main argument.
1247  * Return value:
1248  * o 0:                         everything ok.
1249  * o EINVAL:                    stray interrupt.
1250  */
1251 int
1252 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1253 {
1254 	struct intr_handler *ih;
1255 	struct trapframe *oldframe;
1256 	struct thread *td;
1257 	int phase;
1258 	int ret;
1259 	bool filter, thread;
1260 
1261 	td = curthread;
1262 
1263 #ifdef KSTACK_USAGE_PROF
1264 	intr_prof_stack_use(td, frame);
1265 #endif
1266 
1267 	/* An interrupt with no event or handlers is a stray interrupt. */
1268 	if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
1269 		return (EINVAL);
1270 
1271 	/*
1272 	 * Execute fast interrupt handlers directly.
1273 	 * To support clock handlers, if a handler registers
1274 	 * with a NULL argument, then we pass it a pointer to
1275 	 * a trapframe as its argument.
1276 	 */
1277 	td->td_intr_nesting_level++;
1278 	filter = false;
1279 	thread = false;
1280 	ret = 0;
1281 	critical_enter();
1282 	oldframe = td->td_intr_frame;
1283 	td->td_intr_frame = frame;
1284 
1285 	phase = ie->ie_phase;
1286 	atomic_add_int(&ie->ie_active[phase], 1);
1287 
1288 	/*
1289 	 * This fence is required to ensure that no later loads are
1290 	 * re-ordered before the ie_active store.
1291 	 */
1292 	atomic_thread_fence_seq_cst();
1293 
1294 	CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
1295 		if ((ih->ih_flags & IH_SUSP) != 0)
1296 			continue;
1297 		if (ih->ih_filter == NULL) {
1298 			thread = true;
1299 			continue;
1300 		}
1301 		CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1302 		    ih->ih_filter, ih->ih_argument == NULL ? frame :
1303 		    ih->ih_argument, ih->ih_name);
1304 		if (ih->ih_argument == NULL)
1305 			ret = ih->ih_filter(frame);
1306 		else
1307 			ret = ih->ih_filter(ih->ih_argument);
1308 		KASSERT(ret == FILTER_STRAY ||
1309 		    ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1310 		    (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1311 		    ("%s: incorrect return value %#x from %s", __func__, ret,
1312 		    ih->ih_name));
1313 		filter = filter || ret == FILTER_HANDLED;
1314 
1315 		/*
1316 		 * Wrapper handler special handling:
1317 		 *
1318 		 * in some particular cases (like pccard and pccbb),
1319 		 * the _real_ device handler is wrapped in a couple of
1320 		 * functions - a filter wrapper and an ithread wrapper.
1321 		 * In this case (and just in this case), the filter wrapper
1322 		 * could ask the system to schedule the ithread and mask
1323 		 * the interrupt source if the wrapped handler is composed
1324 		 * of just an ithread handler.
1325 		 *
1326 		 * TODO: write a generic wrapper to avoid people rolling
1327 		 * their own.
1328 		 */
1329 		if (!thread) {
1330 			if (ret == FILTER_SCHEDULE_THREAD)
1331 				thread = true;
1332 		}
1333 	}
1334 	atomic_add_rel_int(&ie->ie_active[phase], -1);
1335 
1336 	td->td_intr_frame = oldframe;
1337 
1338 	if (thread) {
1339 		if (ie->ie_pre_ithread != NULL)
1340 			ie->ie_pre_ithread(ie->ie_source);
1341 	} else {
1342 		if (ie->ie_post_filter != NULL)
1343 			ie->ie_post_filter(ie->ie_source);
1344 	}
1345 
1346 	/* Schedule the ithread if needed. */
1347 	if (thread) {
1348 		int error __unused;
1349 
1350 		error =  intr_event_schedule_thread(ie);
1351 		KASSERT(error == 0, ("bad stray interrupt"));
1352 	}
1353 	critical_exit();
1354 	td->td_intr_nesting_level--;
1355 #ifdef notyet
1356 	/* The interrupt is not aknowledged by any filter and has no ithread. */
1357 	if (!thread && !filter)
1358 		return (EINVAL);
1359 #endif
1360 	return (0);
1361 }
1362 
1363 #ifdef DDB
1364 /*
1365  * Dump details about an interrupt handler
1366  */
1367 static void
1368 db_dump_intrhand(struct intr_handler *ih)
1369 {
1370 	int comma;
1371 
1372 	db_printf("\t%-10s ", ih->ih_name);
1373 	switch (ih->ih_pri) {
1374 	case PI_REALTIME:
1375 		db_printf("CLK ");
1376 		break;
1377 	case PI_AV:
1378 		db_printf("AV  ");
1379 		break;
1380 	case PI_TTY:
1381 		db_printf("TTY ");
1382 		break;
1383 	case PI_NET:
1384 		db_printf("NET ");
1385 		break;
1386 	case PI_DISK:
1387 		db_printf("DISK");
1388 		break;
1389 	case PI_DULL:
1390 		db_printf("DULL");
1391 		break;
1392 	default:
1393 		if (ih->ih_pri >= PI_SOFT)
1394 			db_printf("SWI ");
1395 		else
1396 			db_printf("%4u", ih->ih_pri);
1397 		break;
1398 	}
1399 	db_printf(" ");
1400 	if (ih->ih_filter != NULL) {
1401 		db_printf("[F]");
1402 		db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
1403 	}
1404 	if (ih->ih_handler != NULL) {
1405 		if (ih->ih_filter != NULL)
1406 			db_printf(",");
1407 		db_printf("[H]");
1408 		db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1409 	}
1410 	db_printf("(%p)", ih->ih_argument);
1411 	if (ih->ih_need ||
1412 	    (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1413 	    IH_MPSAFE)) != 0) {
1414 		db_printf(" {");
1415 		comma = 0;
1416 		if (ih->ih_flags & IH_EXCLUSIVE) {
1417 			if (comma)
1418 				db_printf(", ");
1419 			db_printf("EXCL");
1420 			comma = 1;
1421 		}
1422 		if (ih->ih_flags & IH_ENTROPY) {
1423 			if (comma)
1424 				db_printf(", ");
1425 			db_printf("ENTROPY");
1426 			comma = 1;
1427 		}
1428 		if (ih->ih_flags & IH_DEAD) {
1429 			if (comma)
1430 				db_printf(", ");
1431 			db_printf("DEAD");
1432 			comma = 1;
1433 		}
1434 		if (ih->ih_flags & IH_MPSAFE) {
1435 			if (comma)
1436 				db_printf(", ");
1437 			db_printf("MPSAFE");
1438 			comma = 1;
1439 		}
1440 		if (ih->ih_need) {
1441 			if (comma)
1442 				db_printf(", ");
1443 			db_printf("NEED");
1444 		}
1445 		db_printf("}");
1446 	}
1447 	db_printf("\n");
1448 }
1449 
1450 /*
1451  * Dump details about a event.
1452  */
1453 void
1454 db_dump_intr_event(struct intr_event *ie, int handlers)
1455 {
1456 	struct intr_handler *ih;
1457 	struct intr_thread *it;
1458 	int comma;
1459 
1460 	db_printf("%s ", ie->ie_fullname);
1461 	it = ie->ie_thread;
1462 	if (it != NULL)
1463 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1464 	else
1465 		db_printf("(no thread)");
1466 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
1467 	    (it != NULL && it->it_need)) {
1468 		db_printf(" {");
1469 		comma = 0;
1470 		if (ie->ie_flags & IE_SOFT) {
1471 			db_printf("SOFT");
1472 			comma = 1;
1473 		}
1474 		if (ie->ie_flags & IE_ENTROPY) {
1475 			if (comma)
1476 				db_printf(", ");
1477 			db_printf("ENTROPY");
1478 			comma = 1;
1479 		}
1480 		if (ie->ie_flags & IE_ADDING_THREAD) {
1481 			if (comma)
1482 				db_printf(", ");
1483 			db_printf("ADDING_THREAD");
1484 			comma = 1;
1485 		}
1486 		if (it != NULL && it->it_need) {
1487 			if (comma)
1488 				db_printf(", ");
1489 			db_printf("NEED");
1490 		}
1491 		db_printf("}");
1492 	}
1493 	db_printf("\n");
1494 
1495 	if (handlers)
1496 		CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
1497 		    db_dump_intrhand(ih);
1498 }
1499 
1500 /*
1501  * Dump data about interrupt handlers
1502  */
1503 DB_SHOW_COMMAND(intr, db_show_intr)
1504 {
1505 	struct intr_event *ie;
1506 	int all, verbose;
1507 
1508 	verbose = strchr(modif, 'v') != NULL;
1509 	all = strchr(modif, 'a') != NULL;
1510 	TAILQ_FOREACH(ie, &event_list, ie_list) {
1511 		if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
1512 			continue;
1513 		db_dump_intr_event(ie, verbose);
1514 		if (db_pager_quit)
1515 			break;
1516 	}
1517 }
1518 #endif /* DDB */
1519 
1520 /*
1521  * Start standard software interrupt threads
1522  */
1523 static void
1524 start_softintr(void *dummy)
1525 {
1526 
1527 	if (swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
1528 		panic("died while creating vm swi ithread");
1529 }
1530 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1531     NULL);
1532 
1533 /*
1534  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1535  * The data for this machine dependent, and the declarations are in machine
1536  * dependent code.  The layout of intrnames and intrcnt however is machine
1537  * independent.
1538  *
1539  * We do not know the length of intrcnt and intrnames at compile time, so
1540  * calculate things at run time.
1541  */
1542 static int
1543 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1544 {
1545 	return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1546 }
1547 
1548 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1549     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1550 
1551 static int
1552 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1553 {
1554 #ifdef SCTL_MASK32
1555 	uint32_t *intrcnt32;
1556 	unsigned i;
1557 	int error;
1558 
1559 	if (req->flags & SCTL_MASK32) {
1560 		if (!req->oldptr)
1561 			return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
1562 		intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
1563 		if (intrcnt32 == NULL)
1564 			return (ENOMEM);
1565 		for (i = 0; i < sintrcnt / sizeof (u_long); i++)
1566 			intrcnt32[i] = intrcnt[i];
1567 		error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
1568 		free(intrcnt32, M_TEMP);
1569 		return (error);
1570 	}
1571 #endif
1572 	return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1573 }
1574 
1575 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1576     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1577 
1578 #ifdef DDB
1579 /*
1580  * DDB command to dump the interrupt statistics.
1581  */
1582 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
1583 {
1584 	u_long *i;
1585 	char *cp;
1586 	u_int j;
1587 
1588 	cp = intrnames;
1589 	j = 0;
1590 	for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1591 	    i++, j++) {
1592 		if (*cp == '\0')
1593 			break;
1594 		if (*i != 0)
1595 			db_printf("%s\t%lu\n", cp, *i);
1596 		cp += strlen(cp) + 1;
1597 	}
1598 }
1599 #endif
1600