xref: /freebsd/sys/kern/kern_intr.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
1 /*-
2  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_ddb.h"
31 
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/conf.h>
35 #include <sys/rtprio.h>
36 #include <sys/systm.h>
37 #include <sys/interrupt.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/ktr.h>
41 #include <sys/limits.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/random.h>
47 #include <sys/resourcevar.h>
48 #include <sys/sched.h>
49 #include <sys/sysctl.h>
50 #include <sys/unistd.h>
51 #include <sys/vmmeter.h>
52 #include <machine/atomic.h>
53 #include <machine/cpu.h>
54 #include <machine/md_var.h>
55 #include <machine/stdarg.h>
56 #ifdef DDB
57 #include <ddb/ddb.h>
58 #include <ddb/db_sym.h>
59 #endif
60 
61 /*
62  * Describe an interrupt thread.  There is one of these per interrupt event.
63  */
64 struct intr_thread {
65 	struct intr_event *it_event;
66 	struct thread *it_thread;	/* Kernel thread. */
67 	int	it_flags;		/* (j) IT_* flags. */
68 	int	it_need;		/* Needs service. */
69 };
70 
71 /* Interrupt thread flags kept in it_flags */
72 #define	IT_DEAD		0x000001	/* Thread is waiting to exit. */
73 
74 struct	intr_entropy {
75 	struct	thread *td;
76 	uintptr_t event;
77 };
78 
79 struct	intr_event *clk_intr_event;
80 struct	intr_event *tty_intr_event;
81 void	*softclock_ih;
82 void	*vm_ih;
83 
84 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
85 
86 static int intr_storm_threshold = 500;
87 TUNABLE_INT("hw.intr_storm_threshold", &intr_storm_threshold);
88 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RW,
89     &intr_storm_threshold, 0,
90     "Number of consecutive interrupts before storm protection is enabled");
91 static TAILQ_HEAD(, intr_event) event_list =
92     TAILQ_HEAD_INITIALIZER(event_list);
93 
94 static void	intr_event_update(struct intr_event *ie);
95 static struct intr_thread *ithread_create(const char *name);
96 static void	ithread_destroy(struct intr_thread *ithread);
97 static void	ithread_execute_handlers(struct proc *p, struct intr_event *ie);
98 static void	ithread_loop(void *);
99 static void	ithread_update(struct intr_thread *ithd);
100 static void	start_softintr(void *);
101 
102 u_char
103 intr_priority(enum intr_type flags)
104 {
105 	u_char pri;
106 
107 	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
108 	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
109 	switch (flags) {
110 	case INTR_TYPE_TTY:
111 		pri = PI_TTYLOW;
112 		break;
113 	case INTR_TYPE_BIO:
114 		/*
115 		 * XXX We need to refine this.  BSD/OS distinguishes
116 		 * between tape and disk priorities.
117 		 */
118 		pri = PI_DISK;
119 		break;
120 	case INTR_TYPE_NET:
121 		pri = PI_NET;
122 		break;
123 	case INTR_TYPE_CAM:
124 		pri = PI_DISK;          /* XXX or PI_CAM? */
125 		break;
126 	case INTR_TYPE_AV:		/* Audio/video */
127 		pri = PI_AV;
128 		break;
129 	case INTR_TYPE_CLK:
130 		pri = PI_REALTIME;
131 		break;
132 	case INTR_TYPE_MISC:
133 		pri = PI_DULL;          /* don't care */
134 		break;
135 	default:
136 		/* We didn't specify an interrupt level. */
137 		panic("intr_priority: no interrupt type in flags");
138 	}
139 
140 	return pri;
141 }
142 
143 /*
144  * Update an ithread based on the associated intr_event.
145  */
146 static void
147 ithread_update(struct intr_thread *ithd)
148 {
149 	struct intr_event *ie;
150 	struct thread *td;
151 	u_char pri;
152 
153 	ie = ithd->it_event;
154 	td = ithd->it_thread;
155 
156 	/* Determine the overall priority of this event. */
157 	if (TAILQ_EMPTY(&ie->ie_handlers))
158 		pri = PRI_MAX_ITHD;
159 	else
160 		pri = TAILQ_FIRST(&ie->ie_handlers)->ih_pri;
161 
162 	/* Update name and priority. */
163 	strlcpy(td->td_proc->p_comm, ie->ie_fullname,
164 	    sizeof(td->td_proc->p_comm));
165 	mtx_lock_spin(&sched_lock);
166 	sched_prio(td, pri);
167 	mtx_unlock_spin(&sched_lock);
168 }
169 
170 /*
171  * Regenerate the full name of an interrupt event and update its priority.
172  */
173 static void
174 intr_event_update(struct intr_event *ie)
175 {
176 	struct intr_handler *ih;
177 	char *last;
178 	int missed, space;
179 
180 	/* Start off with no entropy and just the name of the event. */
181 	mtx_assert(&ie->ie_lock, MA_OWNED);
182 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
183 	ie->ie_flags &= ~IE_ENTROPY;
184 	missed = 0;
185 	space = 1;
186 
187 	/* Run through all the handlers updating values. */
188 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
189 		if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
190 		    sizeof(ie->ie_fullname)) {
191 			strcat(ie->ie_fullname, " ");
192 			strcat(ie->ie_fullname, ih->ih_name);
193 			space = 0;
194 		} else
195 			missed++;
196 		if (ih->ih_flags & IH_ENTROPY)
197 			ie->ie_flags |= IE_ENTROPY;
198 	}
199 
200 	/*
201 	 * If the handler names were too long, add +'s to indicate missing
202 	 * names. If we run out of room and still have +'s to add, change
203 	 * the last character from a + to a *.
204 	 */
205 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
206 	while (missed-- > 0) {
207 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
208 			if (*last == '+') {
209 				*last = '*';
210 				break;
211 			} else
212 				*last = '+';
213 		} else if (space) {
214 			strcat(ie->ie_fullname, " +");
215 			space = 0;
216 		} else
217 			strcat(ie->ie_fullname, "+");
218 	}
219 
220 	/*
221 	 * If this event has an ithread, update it's priority and
222 	 * name.
223 	 */
224 	if (ie->ie_thread != NULL)
225 		ithread_update(ie->ie_thread);
226 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
227 }
228 
229 int
230 intr_event_create(struct intr_event **event, void *source, int flags,
231     void (*enable)(void *), const char *fmt, ...)
232 {
233 	struct intr_event *ie;
234 	va_list ap;
235 
236 	/* The only valid flag during creation is IE_SOFT. */
237 	if ((flags & ~IE_SOFT) != 0)
238 		return (EINVAL);
239 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
240 	ie->ie_source = source;
241 	ie->ie_enable = enable;
242 	ie->ie_flags = flags;
243 	TAILQ_INIT(&ie->ie_handlers);
244 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
245 
246 	va_start(ap, fmt);
247 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
248 	va_end(ap);
249 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
250 	mtx_pool_lock(mtxpool_sleep, &event_list);
251 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
252 	mtx_pool_unlock(mtxpool_sleep, &event_list);
253 	if (event != NULL)
254 		*event = ie;
255 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
256 	return (0);
257 }
258 
259 int
260 intr_event_destroy(struct intr_event *ie)
261 {
262 
263 	mtx_lock(&ie->ie_lock);
264 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
265 		mtx_unlock(&ie->ie_lock);
266 		return (EBUSY);
267 	}
268 	mtx_pool_lock(mtxpool_sleep, &event_list);
269 	TAILQ_REMOVE(&event_list, ie, ie_list);
270 	mtx_pool_unlock(mtxpool_sleep, &event_list);
271 #ifndef notyet
272 	if (ie->ie_thread != NULL) {
273 		ithread_destroy(ie->ie_thread);
274 		ie->ie_thread = NULL;
275 	}
276 #endif
277 	mtx_unlock(&ie->ie_lock);
278 	mtx_destroy(&ie->ie_lock);
279 	free(ie, M_ITHREAD);
280 	return (0);
281 }
282 
283 static struct intr_thread *
284 ithread_create(const char *name)
285 {
286 	struct intr_thread *ithd;
287 	struct thread *td;
288 	struct proc *p;
289 	int error;
290 
291 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
292 
293 	error = kthread_create(ithread_loop, ithd, &p, RFSTOPPED | RFHIGHPID,
294 	    0, "%s", name);
295 	if (error)
296 		panic("kthread_create() failed with %d", error);
297 	td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
298 	mtx_lock_spin(&sched_lock);
299 	td->td_ksegrp->kg_pri_class = PRI_ITHD;
300 	TD_SET_IWAIT(td);
301 	mtx_unlock_spin(&sched_lock);
302 	td->td_pflags |= TDP_ITHREAD;
303 	ithd->it_thread = td;
304 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
305 	return (ithd);
306 }
307 
308 static void
309 ithread_destroy(struct intr_thread *ithread)
310 {
311 	struct thread *td;
312 
313 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
314 	td = ithread->it_thread;
315 	mtx_lock_spin(&sched_lock);
316 	ithread->it_flags |= IT_DEAD;
317 	if (TD_AWAITING_INTR(td)) {
318 		TD_CLR_IWAIT(td);
319 		setrunqueue(td, SRQ_INTR);
320 	}
321 	mtx_unlock_spin(&sched_lock);
322 }
323 
324 int
325 intr_event_add_handler(struct intr_event *ie, const char *name,
326     driver_intr_t handler, void *arg, u_char pri, enum intr_type flags,
327     void **cookiep)
328 {
329 	struct intr_handler *ih, *temp_ih;
330 	struct intr_thread *it;
331 
332 	if (ie == NULL || name == NULL || handler == NULL)
333 		return (EINVAL);
334 
335 	/* Allocate and populate an interrupt handler structure. */
336 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
337 	ih->ih_handler = handler;
338 	ih->ih_argument = arg;
339 	ih->ih_name = name;
340 	ih->ih_event = ie;
341 	ih->ih_pri = pri;
342 	if (flags & INTR_FAST)
343 		ih->ih_flags = IH_FAST;
344 	else if (flags & INTR_EXCL)
345 		ih->ih_flags = IH_EXCLUSIVE;
346 	if (flags & INTR_MPSAFE)
347 		ih->ih_flags |= IH_MPSAFE;
348 	if (flags & INTR_ENTROPY)
349 		ih->ih_flags |= IH_ENTROPY;
350 
351 	/* We can only have one exclusive handler in a event. */
352 	mtx_lock(&ie->ie_lock);
353 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
354 		if ((flags & INTR_EXCL) ||
355 		    (TAILQ_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
356 			mtx_unlock(&ie->ie_lock);
357 			free(ih, M_ITHREAD);
358 			return (EINVAL);
359 		}
360 	}
361 
362 	/* Add the new handler to the event in priority order. */
363 	TAILQ_FOREACH(temp_ih, &ie->ie_handlers, ih_next) {
364 		if (temp_ih->ih_pri > ih->ih_pri)
365 			break;
366 	}
367 	if (temp_ih == NULL)
368 		TAILQ_INSERT_TAIL(&ie->ie_handlers, ih, ih_next);
369 	else
370 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
371 	intr_event_update(ie);
372 
373 	/* Create a thread if we need one. */
374 	while (ie->ie_thread == NULL && !(flags & INTR_FAST)) {
375 		if (ie->ie_flags & IE_ADDING_THREAD)
376 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
377 		else {
378 			ie->ie_flags |= IE_ADDING_THREAD;
379 			mtx_unlock(&ie->ie_lock);
380 			it = ithread_create("intr: newborn");
381 			mtx_lock(&ie->ie_lock);
382 			ie->ie_flags &= ~IE_ADDING_THREAD;
383 			ie->ie_thread = it;
384 			it->it_event = ie;
385 			ithread_update(it);
386 			wakeup(ie);
387 		}
388 	}
389 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
390 	    ie->ie_name);
391 	mtx_unlock(&ie->ie_lock);
392 
393 	if (cookiep != NULL)
394 		*cookiep = ih;
395 	return (0);
396 }
397 
398 int
399 intr_event_remove_handler(void *cookie)
400 {
401 	struct intr_handler *handler = (struct intr_handler *)cookie;
402 	struct intr_event *ie;
403 #ifdef INVARIANTS
404 	struct intr_handler *ih;
405 #endif
406 #ifdef notyet
407 	int dead;
408 #endif
409 
410 	if (handler == NULL)
411 		return (EINVAL);
412 	ie = handler->ih_event;
413 	KASSERT(ie != NULL,
414 	    ("interrupt handler \"%s\" has a NULL interrupt event",
415 		handler->ih_name));
416 	mtx_lock(&ie->ie_lock);
417 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
418 	    ie->ie_name);
419 #ifdef INVARIANTS
420 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
421 		if (ih == handler)
422 			goto ok;
423 	mtx_unlock(&ie->ie_lock);
424 	panic("interrupt handler \"%s\" not found in interrupt event \"%s\"",
425 	    ih->ih_name, ie->ie_name);
426 ok:
427 #endif
428 	/*
429 	 * If there is no ithread, then just remove the handler and return.
430 	 * XXX: Note that an INTR_FAST handler might be running on another
431 	 * CPU!
432 	 */
433 	if (ie->ie_thread == NULL) {
434 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
435 		mtx_unlock(&ie->ie_lock);
436 		free(handler, M_ITHREAD);
437 		return (0);
438 	}
439 
440 	/*
441 	 * If the interrupt thread is already running, then just mark this
442 	 * handler as being dead and let the ithread do the actual removal.
443 	 *
444 	 * During a cold boot while cold is set, msleep() does not sleep,
445 	 * so we have to remove the handler here rather than letting the
446 	 * thread do it.
447 	 */
448 	mtx_lock_spin(&sched_lock);
449 	if (!TD_AWAITING_INTR(ie->ie_thread->it_thread) && !cold) {
450 		handler->ih_flags |= IH_DEAD;
451 
452 		/*
453 		 * Ensure that the thread will process the handler list
454 		 * again and remove this handler if it has already passed
455 		 * it on the list.
456 		 */
457 		ie->ie_thread->it_need = 1;
458 	} else
459 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
460 	mtx_unlock_spin(&sched_lock);
461 	while (handler->ih_flags & IH_DEAD)
462 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
463 	intr_event_update(ie);
464 #ifdef notyet
465 	/*
466 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
467 	 * this could lead to races of stale data when servicing an
468 	 * interrupt.
469 	 */
470 	dead = 1;
471 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
472 		if (!(ih->ih_flags & IH_FAST)) {
473 			dead = 0;
474 			break;
475 		}
476 	}
477 	if (dead) {
478 		ithread_destroy(ie->ie_thread);
479 		ie->ie_thread = NULL;
480 	}
481 #endif
482 	mtx_unlock(&ie->ie_lock);
483 	free(handler, M_ITHREAD);
484 	return (0);
485 }
486 
487 int
488 intr_event_schedule_thread(struct intr_event *ie)
489 {
490 	struct intr_entropy entropy;
491 	struct intr_thread *it;
492 	struct thread *td;
493 	struct thread *ctd;
494 	struct proc *p;
495 
496 	/*
497 	 * If no ithread or no handlers, then we have a stray interrupt.
498 	 */
499 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers) ||
500 	    ie->ie_thread == NULL)
501 		return (EINVAL);
502 
503 	ctd = curthread;
504 	it = ie->ie_thread;
505 	td = it->it_thread;
506 	p = td->td_proc;
507 
508 	/*
509 	 * If any of the handlers for this ithread claim to be good
510 	 * sources of entropy, then gather some.
511 	 */
512 	if (harvest.interrupt && ie->ie_flags & IE_ENTROPY) {
513 		CTR3(KTR_INTR, "%s: pid %d (%s) gathering entropy", __func__,
514 		    p->p_pid, p->p_comm);
515 		entropy.event = (uintptr_t)ie;
516 		entropy.td = ctd;
517 		random_harvest(&entropy, sizeof(entropy), 2, 0,
518 		    RANDOM_INTERRUPT);
519 	}
520 
521 	KASSERT(p != NULL, ("ithread %s has no process", ie->ie_name));
522 
523 	/*
524 	 * Set it_need to tell the thread to keep running if it is already
525 	 * running.  Then, grab sched_lock and see if we actually need to
526 	 * put this thread on the runqueue.
527 	 */
528 	it->it_need = 1;
529 	mtx_lock_spin(&sched_lock);
530 	if (TD_AWAITING_INTR(td)) {
531 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, p->p_pid,
532 		    p->p_comm);
533 		TD_CLR_IWAIT(td);
534 		setrunqueue(td, SRQ_INTR);
535 	} else {
536 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
537 		    __func__, p->p_pid, p->p_comm, it->it_need, td->td_state);
538 	}
539 	mtx_unlock_spin(&sched_lock);
540 
541 	return (0);
542 }
543 
544 /*
545  * Add a software interrupt handler to a specified event.  If a given event
546  * is not specified, then a new event is created.
547  */
548 int
549 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
550 	    void *arg, int pri, enum intr_type flags, void **cookiep)
551 {
552 	struct intr_event *ie;
553 	int error;
554 
555 	if (flags & (INTR_FAST | INTR_ENTROPY))
556 		return (EINVAL);
557 
558 	ie = (eventp != NULL) ? *eventp : NULL;
559 
560 	if (ie != NULL) {
561 		if (!(ie->ie_flags & IE_SOFT))
562 			return (EINVAL);
563 	} else {
564 		error = intr_event_create(&ie, NULL, IE_SOFT, NULL,
565 		    "swi%d:", pri);
566 		if (error)
567 			return (error);
568 		if (eventp != NULL)
569 			*eventp = ie;
570 	}
571 	return (intr_event_add_handler(ie, name, handler, arg,
572 		    (pri * RQ_PPQ) + PI_SOFT, flags, cookiep));
573 		    /* XXKSE.. think of a better way to get separate queues */
574 }
575 
576 /*
577  * Schedule a software interrupt thread.
578  */
579 void
580 swi_sched(void *cookie, int flags)
581 {
582 	struct intr_handler *ih = (struct intr_handler *)cookie;
583 	struct intr_event *ie = ih->ih_event;
584 	int error;
585 
586 	PCPU_LAZY_INC(cnt.v_intr);
587 
588 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
589 	    ih->ih_need);
590 
591 	/*
592 	 * Set ih_need for this handler so that if the ithread is already
593 	 * running it will execute this handler on the next pass.  Otherwise,
594 	 * it will execute it the next time it runs.
595 	 */
596 	atomic_store_rel_int(&ih->ih_need, 1);
597 	if (!(flags & SWI_DELAY)) {
598 		error = intr_event_schedule_thread(ie);
599 		KASSERT(error == 0, ("stray software interrupt"));
600 	}
601 }
602 
603 /*
604  * Remove a software interrupt handler.  Currently this code does not
605  * remove the associated interrupt event if it becomes empty.  Calling code
606  * may do so manually via intr_event_destroy(), but that's not really
607  * an optimal interface.
608  */
609 int
610 swi_remove(void *cookie)
611 {
612 
613 	return (intr_event_remove_handler(cookie));
614 }
615 
616 static void
617 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
618 {
619 	struct intr_handler *ih, *ihn;
620 
621 	/* Interrupt handlers should not sleep. */
622 	if (!(ie->ie_flags & IE_SOFT))
623 		THREAD_NO_SLEEPING();
624 	TAILQ_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
625 
626 		/*
627 		 * If this handler is marked for death, remove it from
628 		 * the list of handlers and wake up the sleeper.
629 		 */
630 		if (ih->ih_flags & IH_DEAD) {
631 			mtx_lock(&ie->ie_lock);
632 			TAILQ_REMOVE(&ie->ie_handlers, ih, ih_next);
633 			ih->ih_flags &= ~IH_DEAD;
634 			wakeup(ih);
635 			mtx_unlock(&ie->ie_lock);
636 			continue;
637 		}
638 
639 		/*
640 		 * For software interrupt threads, we only execute
641 		 * handlers that have their need flag set.  Hardware
642 		 * interrupt threads always invoke all of their handlers.
643 		 */
644 		if (ie->ie_flags & IE_SOFT) {
645 			if (!ih->ih_need)
646 				continue;
647 			else
648 				atomic_store_rel_int(&ih->ih_need, 0);
649 		}
650 
651 		/* Fast handlers are handled in primary interrupt context. */
652 		if (ih->ih_flags & IH_FAST)
653 			continue;
654 
655 		/* Execute this handler. */
656 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
657 		    __func__, p->p_pid, (void *)ih->ih_handler, ih->ih_argument,
658 		    ih->ih_name, ih->ih_flags);
659 
660 		if (!(ih->ih_flags & IH_MPSAFE))
661 			mtx_lock(&Giant);
662 		ih->ih_handler(ih->ih_argument);
663 		if (!(ih->ih_flags & IH_MPSAFE))
664 			mtx_unlock(&Giant);
665 	}
666 	if (!(ie->ie_flags & IE_SOFT))
667 		THREAD_SLEEPING_OK();
668 
669 	/*
670 	 * Interrupt storm handling:
671 	 *
672 	 * If this interrupt source is currently storming, then throttle
673 	 * it to only fire the handler once  per clock tick.
674 	 *
675 	 * If this interrupt source is not currently storming, but the
676 	 * number of back to back interrupts exceeds the storm threshold,
677 	 * then enter storming mode.
678 	 */
679 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold) {
680 		if (ie->ie_warned == 0) {
681 			printf(
682 	"Interrupt storm detected on \"%s\"; throttling interrupt source\n",
683 			    ie->ie_name);
684 			ie->ie_warned = 1;
685 		}
686 		tsleep(&ie->ie_count, 0, "istorm", 1);
687 	} else
688 		ie->ie_count++;
689 
690 	/*
691 	 * Now that all the handlers have had a chance to run, reenable
692 	 * the interrupt source.
693 	 */
694 	if (ie->ie_enable != NULL)
695 		ie->ie_enable(ie->ie_source);
696 }
697 
698 /*
699  * This is the main code for interrupt threads.
700  */
701 static void
702 ithread_loop(void *arg)
703 {
704 	struct intr_thread *ithd;
705 	struct intr_event *ie;
706 	struct thread *td;
707 	struct proc *p;
708 
709 	td = curthread;
710 	p = td->td_proc;
711 	ithd = (struct intr_thread *)arg;
712 	KASSERT(ithd->it_thread == td,
713 	    ("%s: ithread and proc linkage out of sync", __func__));
714 	ie = ithd->it_event;
715 	ie->ie_count = 0;
716 
717 	/*
718 	 * As long as we have interrupts outstanding, go through the
719 	 * list of handlers, giving each one a go at it.
720 	 */
721 	for (;;) {
722 		/*
723 		 * If we are an orphaned thread, then just die.
724 		 */
725 		if (ithd->it_flags & IT_DEAD) {
726 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
727 			    p->p_pid, p->p_comm);
728 			free(ithd, M_ITHREAD);
729 			kthread_exit(0);
730 		}
731 
732 		/*
733 		 * Service interrupts.  If another interrupt arrives while
734 		 * we are running, it will set it_need to note that we
735 		 * should make another pass.
736 		 */
737 		while (ithd->it_need) {
738 			/*
739 			 * This might need a full read and write barrier
740 			 * to make sure that this write posts before any
741 			 * of the memory or device accesses in the
742 			 * handlers.
743 			 */
744 			atomic_store_rel_int(&ithd->it_need, 0);
745 			ithread_execute_handlers(p, ie);
746 		}
747 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
748 		mtx_assert(&Giant, MA_NOTOWNED);
749 
750 		/*
751 		 * Processed all our interrupts.  Now get the sched
752 		 * lock.  This may take a while and it_need may get
753 		 * set again, so we have to check it again.
754 		 */
755 		mtx_lock_spin(&sched_lock);
756 		if (!ithd->it_need && !(ithd->it_flags & IT_DEAD)) {
757 			TD_SET_IWAIT(td);
758 			ie->ie_count = 0;
759 			mi_switch(SW_VOL, NULL);
760 		}
761 		mtx_unlock_spin(&sched_lock);
762 	}
763 }
764 
765 #ifdef DDB
766 /*
767  * Dump details about an interrupt handler
768  */
769 static void
770 db_dump_intrhand(struct intr_handler *ih)
771 {
772 	int comma;
773 
774 	db_printf("\t%-10s ", ih->ih_name);
775 	switch (ih->ih_pri) {
776 	case PI_REALTIME:
777 		db_printf("CLK ");
778 		break;
779 	case PI_AV:
780 		db_printf("AV  ");
781 		break;
782 	case PI_TTYHIGH:
783 	case PI_TTYLOW:
784 		db_printf("TTY ");
785 		break;
786 	case PI_TAPE:
787 		db_printf("TAPE");
788 		break;
789 	case PI_NET:
790 		db_printf("NET ");
791 		break;
792 	case PI_DISK:
793 	case PI_DISKLOW:
794 		db_printf("DISK");
795 		break;
796 	case PI_DULL:
797 		db_printf("DULL");
798 		break;
799 	default:
800 		if (ih->ih_pri >= PI_SOFT)
801 			db_printf("SWI ");
802 		else
803 			db_printf("%4u", ih->ih_pri);
804 		break;
805 	}
806 	db_printf(" ");
807 	db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
808 	db_printf("(%p)", ih->ih_argument);
809 	if (ih->ih_need ||
810 	    (ih->ih_flags & (IH_FAST | IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
811 	    IH_MPSAFE)) != 0) {
812 		db_printf(" {");
813 		comma = 0;
814 		if (ih->ih_flags & IH_FAST) {
815 			db_printf("FAST");
816 			comma = 1;
817 		}
818 		if (ih->ih_flags & IH_EXCLUSIVE) {
819 			if (comma)
820 				db_printf(", ");
821 			db_printf("EXCL");
822 			comma = 1;
823 		}
824 		if (ih->ih_flags & IH_ENTROPY) {
825 			if (comma)
826 				db_printf(", ");
827 			db_printf("ENTROPY");
828 			comma = 1;
829 		}
830 		if (ih->ih_flags & IH_DEAD) {
831 			if (comma)
832 				db_printf(", ");
833 			db_printf("DEAD");
834 			comma = 1;
835 		}
836 		if (ih->ih_flags & IH_MPSAFE) {
837 			if (comma)
838 				db_printf(", ");
839 			db_printf("MPSAFE");
840 			comma = 1;
841 		}
842 		if (ih->ih_need) {
843 			if (comma)
844 				db_printf(", ");
845 			db_printf("NEED");
846 		}
847 		db_printf("}");
848 	}
849 	db_printf("\n");
850 }
851 
852 /*
853  * Dump details about a event.
854  */
855 void
856 db_dump_intr_event(struct intr_event *ie, int handlers)
857 {
858 	struct intr_handler *ih;
859 	struct intr_thread *it;
860 	int comma;
861 
862 	db_printf("%s ", ie->ie_fullname);
863 	it = ie->ie_thread;
864 	if (it != NULL)
865 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
866 	else
867 		db_printf("(no thread)");
868 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
869 	    (it != NULL && it->it_need)) {
870 		db_printf(" {");
871 		comma = 0;
872 		if (ie->ie_flags & IE_SOFT) {
873 			db_printf("SOFT");
874 			comma = 1;
875 		}
876 		if (ie->ie_flags & IE_ENTROPY) {
877 			if (comma)
878 				db_printf(", ");
879 			db_printf("ENTROPY");
880 			comma = 1;
881 		}
882 		if (ie->ie_flags & IE_ADDING_THREAD) {
883 			if (comma)
884 				db_printf(", ");
885 			db_printf("ADDING_THREAD");
886 			comma = 1;
887 		}
888 		if (it != NULL && it->it_need) {
889 			if (comma)
890 				db_printf(", ");
891 			db_printf("NEED");
892 		}
893 		db_printf("}");
894 	}
895 	db_printf("\n");
896 
897 	if (handlers)
898 		TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
899 		    db_dump_intrhand(ih);
900 }
901 
902 /*
903  * Dump data about interrupt handlers
904  */
905 DB_SHOW_COMMAND(intr, db_show_intr)
906 {
907 	struct intr_event *ie;
908 	int all, verbose;
909 
910 	verbose = index(modif, 'v') != NULL;
911 	all = index(modif, 'a') != NULL;
912 	TAILQ_FOREACH(ie, &event_list, ie_list) {
913 		if (!all && TAILQ_EMPTY(&ie->ie_handlers))
914 			continue;
915 		db_dump_intr_event(ie, verbose);
916 		if (db_pager_quit)
917 			break;
918 	}
919 }
920 #endif /* DDB */
921 
922 /*
923  * Start standard software interrupt threads
924  */
925 static void
926 start_softintr(void *dummy)
927 {
928 	struct proc *p;
929 
930 	if (swi_add(&clk_intr_event, "clock", softclock, NULL, SWI_CLOCK,
931 		INTR_MPSAFE, &softclock_ih) ||
932 	    swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
933 		panic("died while creating standard software ithreads");
934 
935 	p = clk_intr_event->ie_thread->it_thread->td_proc;
936 	PROC_LOCK(p);
937 	p->p_flag |= P_NOLOAD;
938 	PROC_UNLOCK(p);
939 }
940 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr, NULL)
941 
942 /*
943  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
944  * The data for this machine dependent, and the declarations are in machine
945  * dependent code.  The layout of intrnames and intrcnt however is machine
946  * independent.
947  *
948  * We do not know the length of intrcnt and intrnames at compile time, so
949  * calculate things at run time.
950  */
951 static int
952 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
953 {
954 	return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
955 	   req));
956 }
957 
958 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
959     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
960 
961 static int
962 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
963 {
964 	return (sysctl_handle_opaque(oidp, intrcnt,
965 	    (char *)eintrcnt - (char *)intrcnt, req));
966 }
967 
968 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
969     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
970 
971 #ifdef DDB
972 /*
973  * DDB command to dump the interrupt statistics.
974  */
975 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
976 {
977 	u_long *i;
978 	char *cp;
979 
980 	cp = intrnames;
981 	for (i = intrcnt; i != eintrcnt && !db_pager_quit; i++) {
982 		if (*cp == '\0')
983 			break;
984 		if (*i != 0)
985 			db_printf("%s\t%lu\n", cp, *i);
986 		cp += strlen(cp) + 1;
987 	}
988 }
989 #endif
990