xref: /freebsd/sys/kern/kern_intr.c (revision acd3428b7d3e94cef0e1881c868cb4b131d4ff41)
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 #ifdef KSE
300 	td->td_ksegrp->kg_pri_class = PRI_ITHD;
301 #else
302 	td->td_pri_class = PRI_ITHD;
303 #endif
304 	TD_SET_IWAIT(td);
305 	mtx_unlock_spin(&sched_lock);
306 	td->td_pflags |= TDP_ITHREAD;
307 	ithd->it_thread = td;
308 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
309 	return (ithd);
310 }
311 
312 static void
313 ithread_destroy(struct intr_thread *ithread)
314 {
315 	struct thread *td;
316 
317 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
318 	td = ithread->it_thread;
319 	mtx_lock_spin(&sched_lock);
320 	ithread->it_flags |= IT_DEAD;
321 	if (TD_AWAITING_INTR(td)) {
322 		TD_CLR_IWAIT(td);
323 		setrunqueue(td, SRQ_INTR);
324 	}
325 	mtx_unlock_spin(&sched_lock);
326 }
327 
328 int
329 intr_event_add_handler(struct intr_event *ie, const char *name,
330     driver_intr_t handler, void *arg, u_char pri, enum intr_type flags,
331     void **cookiep)
332 {
333 	struct intr_handler *ih, *temp_ih;
334 	struct intr_thread *it;
335 
336 	if (ie == NULL || name == NULL || handler == NULL)
337 		return (EINVAL);
338 
339 	/* Allocate and populate an interrupt handler structure. */
340 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
341 	ih->ih_handler = handler;
342 	ih->ih_argument = arg;
343 	ih->ih_name = name;
344 	ih->ih_event = ie;
345 	ih->ih_pri = pri;
346 	if (flags & INTR_FAST)
347 		ih->ih_flags = IH_FAST;
348 	else if (flags & INTR_EXCL)
349 		ih->ih_flags = IH_EXCLUSIVE;
350 	if (flags & INTR_MPSAFE)
351 		ih->ih_flags |= IH_MPSAFE;
352 	if (flags & INTR_ENTROPY)
353 		ih->ih_flags |= IH_ENTROPY;
354 
355 	/* We can only have one exclusive handler in a event. */
356 	mtx_lock(&ie->ie_lock);
357 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
358 		if ((flags & INTR_EXCL) ||
359 		    (TAILQ_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
360 			mtx_unlock(&ie->ie_lock);
361 			free(ih, M_ITHREAD);
362 			return (EINVAL);
363 		}
364 	}
365 
366 	/* Add the new handler to the event in priority order. */
367 	TAILQ_FOREACH(temp_ih, &ie->ie_handlers, ih_next) {
368 		if (temp_ih->ih_pri > ih->ih_pri)
369 			break;
370 	}
371 	if (temp_ih == NULL)
372 		TAILQ_INSERT_TAIL(&ie->ie_handlers, ih, ih_next);
373 	else
374 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
375 	intr_event_update(ie);
376 
377 	/* Create a thread if we need one. */
378 	while (ie->ie_thread == NULL && !(flags & INTR_FAST)) {
379 		if (ie->ie_flags & IE_ADDING_THREAD)
380 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
381 		else {
382 			ie->ie_flags |= IE_ADDING_THREAD;
383 			mtx_unlock(&ie->ie_lock);
384 			it = ithread_create("intr: newborn");
385 			mtx_lock(&ie->ie_lock);
386 			ie->ie_flags &= ~IE_ADDING_THREAD;
387 			ie->ie_thread = it;
388 			it->it_event = ie;
389 			ithread_update(it);
390 			wakeup(ie);
391 		}
392 	}
393 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
394 	    ie->ie_name);
395 	mtx_unlock(&ie->ie_lock);
396 
397 	if (cookiep != NULL)
398 		*cookiep = ih;
399 	return (0);
400 }
401 
402 int
403 intr_event_remove_handler(void *cookie)
404 {
405 	struct intr_handler *handler = (struct intr_handler *)cookie;
406 	struct intr_event *ie;
407 #ifdef INVARIANTS
408 	struct intr_handler *ih;
409 #endif
410 #ifdef notyet
411 	int dead;
412 #endif
413 
414 	if (handler == NULL)
415 		return (EINVAL);
416 	ie = handler->ih_event;
417 	KASSERT(ie != NULL,
418 	    ("interrupt handler \"%s\" has a NULL interrupt event",
419 		handler->ih_name));
420 	mtx_lock(&ie->ie_lock);
421 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
422 	    ie->ie_name);
423 #ifdef INVARIANTS
424 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
425 		if (ih == handler)
426 			goto ok;
427 	mtx_unlock(&ie->ie_lock);
428 	panic("interrupt handler \"%s\" not found in interrupt event \"%s\"",
429 	    ih->ih_name, ie->ie_name);
430 ok:
431 #endif
432 	/*
433 	 * If there is no ithread, then just remove the handler and return.
434 	 * XXX: Note that an INTR_FAST handler might be running on another
435 	 * CPU!
436 	 */
437 	if (ie->ie_thread == NULL) {
438 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
439 		mtx_unlock(&ie->ie_lock);
440 		free(handler, M_ITHREAD);
441 		return (0);
442 	}
443 
444 	/*
445 	 * If the interrupt thread is already running, then just mark this
446 	 * handler as being dead and let the ithread do the actual removal.
447 	 *
448 	 * During a cold boot while cold is set, msleep() does not sleep,
449 	 * so we have to remove the handler here rather than letting the
450 	 * thread do it.
451 	 */
452 	mtx_lock_spin(&sched_lock);
453 	if (!TD_AWAITING_INTR(ie->ie_thread->it_thread) && !cold) {
454 		handler->ih_flags |= IH_DEAD;
455 
456 		/*
457 		 * Ensure that the thread will process the handler list
458 		 * again and remove this handler if it has already passed
459 		 * it on the list.
460 		 */
461 		ie->ie_thread->it_need = 1;
462 	} else
463 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
464 	mtx_unlock_spin(&sched_lock);
465 	while (handler->ih_flags & IH_DEAD)
466 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
467 	intr_event_update(ie);
468 #ifdef notyet
469 	/*
470 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
471 	 * this could lead to races of stale data when servicing an
472 	 * interrupt.
473 	 */
474 	dead = 1;
475 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
476 		if (!(ih->ih_flags & IH_FAST)) {
477 			dead = 0;
478 			break;
479 		}
480 	}
481 	if (dead) {
482 		ithread_destroy(ie->ie_thread);
483 		ie->ie_thread = NULL;
484 	}
485 #endif
486 	mtx_unlock(&ie->ie_lock);
487 	free(handler, M_ITHREAD);
488 	return (0);
489 }
490 
491 int
492 intr_event_schedule_thread(struct intr_event *ie)
493 {
494 	struct intr_entropy entropy;
495 	struct intr_thread *it;
496 	struct thread *td;
497 	struct thread *ctd;
498 	struct proc *p;
499 
500 	/*
501 	 * If no ithread or no handlers, then we have a stray interrupt.
502 	 */
503 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers) ||
504 	    ie->ie_thread == NULL)
505 		return (EINVAL);
506 
507 	ctd = curthread;
508 	it = ie->ie_thread;
509 	td = it->it_thread;
510 	p = td->td_proc;
511 
512 	/*
513 	 * If any of the handlers for this ithread claim to be good
514 	 * sources of entropy, then gather some.
515 	 */
516 	if (harvest.interrupt && ie->ie_flags & IE_ENTROPY) {
517 		CTR3(KTR_INTR, "%s: pid %d (%s) gathering entropy", __func__,
518 		    p->p_pid, p->p_comm);
519 		entropy.event = (uintptr_t)ie;
520 		entropy.td = ctd;
521 		random_harvest(&entropy, sizeof(entropy), 2, 0,
522 		    RANDOM_INTERRUPT);
523 	}
524 
525 	KASSERT(p != NULL, ("ithread %s has no process", ie->ie_name));
526 
527 	/*
528 	 * Set it_need to tell the thread to keep running if it is already
529 	 * running.  Then, grab sched_lock and see if we actually need to
530 	 * put this thread on the runqueue.
531 	 */
532 	it->it_need = 1;
533 	mtx_lock_spin(&sched_lock);
534 	if (TD_AWAITING_INTR(td)) {
535 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, p->p_pid,
536 		    p->p_comm);
537 		TD_CLR_IWAIT(td);
538 		setrunqueue(td, SRQ_INTR);
539 	} else {
540 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
541 		    __func__, p->p_pid, p->p_comm, it->it_need, td->td_state);
542 	}
543 	mtx_unlock_spin(&sched_lock);
544 
545 	return (0);
546 }
547 
548 /*
549  * Add a software interrupt handler to a specified event.  If a given event
550  * is not specified, then a new event is created.
551  */
552 int
553 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
554 	    void *arg, int pri, enum intr_type flags, void **cookiep)
555 {
556 	struct intr_event *ie;
557 	int error;
558 
559 	if (flags & (INTR_FAST | INTR_ENTROPY))
560 		return (EINVAL);
561 
562 	ie = (eventp != NULL) ? *eventp : NULL;
563 
564 	if (ie != NULL) {
565 		if (!(ie->ie_flags & IE_SOFT))
566 			return (EINVAL);
567 	} else {
568 		error = intr_event_create(&ie, NULL, IE_SOFT, NULL,
569 		    "swi%d:", pri);
570 		if (error)
571 			return (error);
572 		if (eventp != NULL)
573 			*eventp = ie;
574 	}
575 	return (intr_event_add_handler(ie, name, handler, arg,
576 		    (pri * RQ_PPQ) + PI_SOFT, flags, cookiep));
577 		    /* XXKSE.. think of a better way to get separate queues */
578 }
579 
580 /*
581  * Schedule a software interrupt thread.
582  */
583 void
584 swi_sched(void *cookie, int flags)
585 {
586 	struct intr_handler *ih = (struct intr_handler *)cookie;
587 	struct intr_event *ie = ih->ih_event;
588 	int error;
589 
590 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
591 	    ih->ih_need);
592 
593 	/*
594 	 * Set ih_need for this handler so that if the ithread is already
595 	 * running it will execute this handler on the next pass.  Otherwise,
596 	 * it will execute it the next time it runs.
597 	 */
598 	atomic_store_rel_int(&ih->ih_need, 1);
599 
600 	if (!(flags & SWI_DELAY)) {
601 		PCPU_LAZY_INC(cnt.v_soft);
602 		error = intr_event_schedule_thread(ie);
603 		KASSERT(error == 0, ("stray software interrupt"));
604 	}
605 }
606 
607 /*
608  * Remove a software interrupt handler.  Currently this code does not
609  * remove the associated interrupt event if it becomes empty.  Calling code
610  * may do so manually via intr_event_destroy(), but that's not really
611  * an optimal interface.
612  */
613 int
614 swi_remove(void *cookie)
615 {
616 
617 	return (intr_event_remove_handler(cookie));
618 }
619 
620 static void
621 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
622 {
623 	struct intr_handler *ih, *ihn;
624 
625 	/* Interrupt handlers should not sleep. */
626 	if (!(ie->ie_flags & IE_SOFT))
627 		THREAD_NO_SLEEPING();
628 	TAILQ_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
629 
630 		/*
631 		 * If this handler is marked for death, remove it from
632 		 * the list of handlers and wake up the sleeper.
633 		 */
634 		if (ih->ih_flags & IH_DEAD) {
635 			mtx_lock(&ie->ie_lock);
636 			TAILQ_REMOVE(&ie->ie_handlers, ih, ih_next);
637 			ih->ih_flags &= ~IH_DEAD;
638 			wakeup(ih);
639 			mtx_unlock(&ie->ie_lock);
640 			continue;
641 		}
642 
643 		/*
644 		 * For software interrupt threads, we only execute
645 		 * handlers that have their need flag set.  Hardware
646 		 * interrupt threads always invoke all of their handlers.
647 		 */
648 		if (ie->ie_flags & IE_SOFT) {
649 			if (!ih->ih_need)
650 				continue;
651 			else
652 				atomic_store_rel_int(&ih->ih_need, 0);
653 		}
654 
655 		/* Fast handlers are handled in primary interrupt context. */
656 		if (ih->ih_flags & IH_FAST)
657 			continue;
658 
659 		/* Execute this handler. */
660 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
661 		    __func__, p->p_pid, (void *)ih->ih_handler, ih->ih_argument,
662 		    ih->ih_name, ih->ih_flags);
663 
664 		if (!(ih->ih_flags & IH_MPSAFE))
665 			mtx_lock(&Giant);
666 		ih->ih_handler(ih->ih_argument);
667 		if (!(ih->ih_flags & IH_MPSAFE))
668 			mtx_unlock(&Giant);
669 	}
670 	if (!(ie->ie_flags & IE_SOFT))
671 		THREAD_SLEEPING_OK();
672 
673 	/*
674 	 * Interrupt storm handling:
675 	 *
676 	 * If this interrupt source is currently storming, then throttle
677 	 * it to only fire the handler once  per clock tick.
678 	 *
679 	 * If this interrupt source is not currently storming, but the
680 	 * number of back to back interrupts exceeds the storm threshold,
681 	 * then enter storming mode.
682 	 */
683 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold) {
684 		if (ie->ie_warned == 0) {
685 			printf(
686 	"Interrupt storm detected on \"%s\"; throttling interrupt source\n",
687 			    ie->ie_name);
688 			ie->ie_warned = 1;
689 		}
690 		tsleep(&ie->ie_count, 0, "istorm", 1);
691 	} else
692 		ie->ie_count++;
693 
694 	/*
695 	 * Now that all the handlers have had a chance to run, reenable
696 	 * the interrupt source.
697 	 */
698 	if (ie->ie_enable != NULL)
699 		ie->ie_enable(ie->ie_source);
700 }
701 
702 /*
703  * This is the main code for interrupt threads.
704  */
705 static void
706 ithread_loop(void *arg)
707 {
708 	struct intr_thread *ithd;
709 	struct intr_event *ie;
710 	struct thread *td;
711 	struct proc *p;
712 
713 	td = curthread;
714 	p = td->td_proc;
715 	ithd = (struct intr_thread *)arg;
716 	KASSERT(ithd->it_thread == td,
717 	    ("%s: ithread and proc linkage out of sync", __func__));
718 	ie = ithd->it_event;
719 	ie->ie_count = 0;
720 
721 	/*
722 	 * As long as we have interrupts outstanding, go through the
723 	 * list of handlers, giving each one a go at it.
724 	 */
725 	for (;;) {
726 		/*
727 		 * If we are an orphaned thread, then just die.
728 		 */
729 		if (ithd->it_flags & IT_DEAD) {
730 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
731 			    p->p_pid, p->p_comm);
732 			free(ithd, M_ITHREAD);
733 			kthread_exit(0);
734 		}
735 
736 		/*
737 		 * Service interrupts.  If another interrupt arrives while
738 		 * we are running, it will set it_need to note that we
739 		 * should make another pass.
740 		 */
741 		while (ithd->it_need) {
742 			/*
743 			 * This might need a full read and write barrier
744 			 * to make sure that this write posts before any
745 			 * of the memory or device accesses in the
746 			 * handlers.
747 			 */
748 			atomic_store_rel_int(&ithd->it_need, 0);
749 			ithread_execute_handlers(p, ie);
750 		}
751 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
752 		mtx_assert(&Giant, MA_NOTOWNED);
753 
754 		/*
755 		 * Processed all our interrupts.  Now get the sched
756 		 * lock.  This may take a while and it_need may get
757 		 * set again, so we have to check it again.
758 		 */
759 		mtx_lock_spin(&sched_lock);
760 		if (!ithd->it_need && !(ithd->it_flags & IT_DEAD)) {
761 			TD_SET_IWAIT(td);
762 			ie->ie_count = 0;
763 			mi_switch(SW_VOL, NULL);
764 		}
765 		mtx_unlock_spin(&sched_lock);
766 	}
767 }
768 
769 #ifdef DDB
770 /*
771  * Dump details about an interrupt handler
772  */
773 static void
774 db_dump_intrhand(struct intr_handler *ih)
775 {
776 	int comma;
777 
778 	db_printf("\t%-10s ", ih->ih_name);
779 	switch (ih->ih_pri) {
780 	case PI_REALTIME:
781 		db_printf("CLK ");
782 		break;
783 	case PI_AV:
784 		db_printf("AV  ");
785 		break;
786 	case PI_TTYHIGH:
787 	case PI_TTYLOW:
788 		db_printf("TTY ");
789 		break;
790 	case PI_TAPE:
791 		db_printf("TAPE");
792 		break;
793 	case PI_NET:
794 		db_printf("NET ");
795 		break;
796 	case PI_DISK:
797 	case PI_DISKLOW:
798 		db_printf("DISK");
799 		break;
800 	case PI_DULL:
801 		db_printf("DULL");
802 		break;
803 	default:
804 		if (ih->ih_pri >= PI_SOFT)
805 			db_printf("SWI ");
806 		else
807 			db_printf("%4u", ih->ih_pri);
808 		break;
809 	}
810 	db_printf(" ");
811 	db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
812 	db_printf("(%p)", ih->ih_argument);
813 	if (ih->ih_need ||
814 	    (ih->ih_flags & (IH_FAST | IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
815 	    IH_MPSAFE)) != 0) {
816 		db_printf(" {");
817 		comma = 0;
818 		if (ih->ih_flags & IH_FAST) {
819 			db_printf("FAST");
820 			comma = 1;
821 		}
822 		if (ih->ih_flags & IH_EXCLUSIVE) {
823 			if (comma)
824 				db_printf(", ");
825 			db_printf("EXCL");
826 			comma = 1;
827 		}
828 		if (ih->ih_flags & IH_ENTROPY) {
829 			if (comma)
830 				db_printf(", ");
831 			db_printf("ENTROPY");
832 			comma = 1;
833 		}
834 		if (ih->ih_flags & IH_DEAD) {
835 			if (comma)
836 				db_printf(", ");
837 			db_printf("DEAD");
838 			comma = 1;
839 		}
840 		if (ih->ih_flags & IH_MPSAFE) {
841 			if (comma)
842 				db_printf(", ");
843 			db_printf("MPSAFE");
844 			comma = 1;
845 		}
846 		if (ih->ih_need) {
847 			if (comma)
848 				db_printf(", ");
849 			db_printf("NEED");
850 		}
851 		db_printf("}");
852 	}
853 	db_printf("\n");
854 }
855 
856 /*
857  * Dump details about a event.
858  */
859 void
860 db_dump_intr_event(struct intr_event *ie, int handlers)
861 {
862 	struct intr_handler *ih;
863 	struct intr_thread *it;
864 	int comma;
865 
866 	db_printf("%s ", ie->ie_fullname);
867 	it = ie->ie_thread;
868 	if (it != NULL)
869 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
870 	else
871 		db_printf("(no thread)");
872 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
873 	    (it != NULL && it->it_need)) {
874 		db_printf(" {");
875 		comma = 0;
876 		if (ie->ie_flags & IE_SOFT) {
877 			db_printf("SOFT");
878 			comma = 1;
879 		}
880 		if (ie->ie_flags & IE_ENTROPY) {
881 			if (comma)
882 				db_printf(", ");
883 			db_printf("ENTROPY");
884 			comma = 1;
885 		}
886 		if (ie->ie_flags & IE_ADDING_THREAD) {
887 			if (comma)
888 				db_printf(", ");
889 			db_printf("ADDING_THREAD");
890 			comma = 1;
891 		}
892 		if (it != NULL && it->it_need) {
893 			if (comma)
894 				db_printf(", ");
895 			db_printf("NEED");
896 		}
897 		db_printf("}");
898 	}
899 	db_printf("\n");
900 
901 	if (handlers)
902 		TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
903 		    db_dump_intrhand(ih);
904 }
905 
906 /*
907  * Dump data about interrupt handlers
908  */
909 DB_SHOW_COMMAND(intr, db_show_intr)
910 {
911 	struct intr_event *ie;
912 	int all, verbose;
913 
914 	verbose = index(modif, 'v') != NULL;
915 	all = index(modif, 'a') != NULL;
916 	TAILQ_FOREACH(ie, &event_list, ie_list) {
917 		if (!all && TAILQ_EMPTY(&ie->ie_handlers))
918 			continue;
919 		db_dump_intr_event(ie, verbose);
920 		if (db_pager_quit)
921 			break;
922 	}
923 }
924 #endif /* DDB */
925 
926 /*
927  * Start standard software interrupt threads
928  */
929 static void
930 start_softintr(void *dummy)
931 {
932 	struct proc *p;
933 
934 	if (swi_add(&clk_intr_event, "clock", softclock, NULL, SWI_CLOCK,
935 		INTR_MPSAFE, &softclock_ih) ||
936 	    swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
937 		panic("died while creating standard software ithreads");
938 
939 	p = clk_intr_event->ie_thread->it_thread->td_proc;
940 	PROC_LOCK(p);
941 	p->p_flag |= P_NOLOAD;
942 	PROC_UNLOCK(p);
943 }
944 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr, NULL)
945 
946 /*
947  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
948  * The data for this machine dependent, and the declarations are in machine
949  * dependent code.  The layout of intrnames and intrcnt however is machine
950  * independent.
951  *
952  * We do not know the length of intrcnt and intrnames at compile time, so
953  * calculate things at run time.
954  */
955 static int
956 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
957 {
958 	return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
959 	   req));
960 }
961 
962 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
963     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
964 
965 static int
966 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
967 {
968 	return (sysctl_handle_opaque(oidp, intrcnt,
969 	    (char *)eintrcnt - (char *)intrcnt, req));
970 }
971 
972 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
973     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
974 
975 #ifdef DDB
976 /*
977  * DDB command to dump the interrupt statistics.
978  */
979 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
980 {
981 	u_long *i;
982 	char *cp;
983 
984 	cp = intrnames;
985 	for (i = intrcnt; i != eintrcnt && !db_pager_quit; i++) {
986 		if (*cp == '\0')
987 			break;
988 		if (*i != 0)
989 			db_printf("%s\t%lu\n", cp, *i);
990 		cp += strlen(cp) + 1;
991 	}
992 }
993 #endif
994