xref: /freebsd/sys/kern/kern_intr.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
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 /* Map an interrupt type to an ithread priority. */
103 u_char
104 intr_priority(enum intr_type flags)
105 {
106 	u_char pri;
107 
108 	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
109 	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
110 	switch (flags) {
111 	case INTR_TYPE_TTY:
112 		pri = PI_TTYLOW;
113 		break;
114 	case INTR_TYPE_BIO:
115 		/*
116 		 * XXX We need to refine this.  BSD/OS distinguishes
117 		 * between tape and disk priorities.
118 		 */
119 		pri = PI_DISK;
120 		break;
121 	case INTR_TYPE_NET:
122 		pri = PI_NET;
123 		break;
124 	case INTR_TYPE_CAM:
125 		pri = PI_DISK;          /* XXX or PI_CAM? */
126 		break;
127 	case INTR_TYPE_AV:		/* Audio/video */
128 		pri = PI_AV;
129 		break;
130 	case INTR_TYPE_CLK:
131 		pri = PI_REALTIME;
132 		break;
133 	case INTR_TYPE_MISC:
134 		pri = PI_DULL;          /* don't care */
135 		break;
136 	default:
137 		/* We didn't specify an interrupt level. */
138 		panic("intr_priority: no interrupt type in flags");
139 	}
140 
141 	return pri;
142 }
143 
144 /*
145  * Update an ithread based on the associated intr_event.
146  */
147 static void
148 ithread_update(struct intr_thread *ithd)
149 {
150 	struct intr_event *ie;
151 	struct thread *td;
152 	u_char pri;
153 
154 	ie = ithd->it_event;
155 	td = ithd->it_thread;
156 
157 	/* Determine the overall priority of this event. */
158 	if (TAILQ_EMPTY(&ie->ie_handlers))
159 		pri = PRI_MAX_ITHD;
160 	else
161 		pri = TAILQ_FIRST(&ie->ie_handlers)->ih_pri;
162 
163 	/* Update name and priority. */
164 	strlcpy(td->td_proc->p_comm, ie->ie_fullname,
165 	    sizeof(td->td_proc->p_comm));
166 	mtx_lock_spin(&sched_lock);
167 	sched_prio(td, pri);
168 	mtx_unlock_spin(&sched_lock);
169 }
170 
171 /*
172  * Regenerate the full name of an interrupt event and update its priority.
173  */
174 static void
175 intr_event_update(struct intr_event *ie)
176 {
177 	struct intr_handler *ih;
178 	char *last;
179 	int missed, space;
180 
181 	/* Start off with no entropy and just the name of the event. */
182 	mtx_assert(&ie->ie_lock, MA_OWNED);
183 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
184 	ie->ie_flags &= ~IE_ENTROPY;
185 	missed = 0;
186 	space = 1;
187 
188 	/* Run through all the handlers updating values. */
189 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
190 		if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
191 		    sizeof(ie->ie_fullname)) {
192 			strcat(ie->ie_fullname, " ");
193 			strcat(ie->ie_fullname, ih->ih_name);
194 			space = 0;
195 		} else
196 			missed++;
197 		if (ih->ih_flags & IH_ENTROPY)
198 			ie->ie_flags |= IE_ENTROPY;
199 	}
200 
201 	/*
202 	 * If the handler names were too long, add +'s to indicate missing
203 	 * names. If we run out of room and still have +'s to add, change
204 	 * the last character from a + to a *.
205 	 */
206 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
207 	while (missed-- > 0) {
208 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
209 			if (*last == '+') {
210 				*last = '*';
211 				break;
212 			} else
213 				*last = '+';
214 		} else if (space) {
215 			strcat(ie->ie_fullname, " +");
216 			space = 0;
217 		} else
218 			strcat(ie->ie_fullname, "+");
219 	}
220 
221 	/*
222 	 * If this event has an ithread, update it's priority and
223 	 * name.
224 	 */
225 	if (ie->ie_thread != NULL)
226 		ithread_update(ie->ie_thread);
227 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
228 }
229 
230 int
231 intr_event_create(struct intr_event **event, void *source, int flags,
232     void (*enable)(void *), const char *fmt, ...)
233 {
234 	struct intr_event *ie;
235 	va_list ap;
236 
237 	/* The only valid flag during creation is IE_SOFT. */
238 	if ((flags & ~IE_SOFT) != 0)
239 		return (EINVAL);
240 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
241 	ie->ie_source = source;
242 	ie->ie_enable = enable;
243 	ie->ie_flags = flags;
244 	TAILQ_INIT(&ie->ie_handlers);
245 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
246 
247 	va_start(ap, fmt);
248 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
249 	va_end(ap);
250 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
251 	mtx_pool_lock(mtxpool_sleep, &event_list);
252 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
253 	mtx_pool_unlock(mtxpool_sleep, &event_list);
254 	if (event != NULL)
255 		*event = ie;
256 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
257 	return (0);
258 }
259 
260 int
261 intr_event_destroy(struct intr_event *ie)
262 {
263 
264 	mtx_lock(&ie->ie_lock);
265 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
266 		mtx_unlock(&ie->ie_lock);
267 		return (EBUSY);
268 	}
269 	mtx_pool_lock(mtxpool_sleep, &event_list);
270 	TAILQ_REMOVE(&event_list, ie, ie_list);
271 	mtx_pool_unlock(mtxpool_sleep, &event_list);
272 #ifndef notyet
273 	if (ie->ie_thread != NULL) {
274 		ithread_destroy(ie->ie_thread);
275 		ie->ie_thread = NULL;
276 	}
277 #endif
278 	mtx_unlock(&ie->ie_lock);
279 	mtx_destroy(&ie->ie_lock);
280 	free(ie, M_ITHREAD);
281 	return (0);
282 }
283 
284 static struct intr_thread *
285 ithread_create(const char *name)
286 {
287 	struct intr_thread *ithd;
288 	struct thread *td;
289 	struct proc *p;
290 	int error;
291 
292 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
293 
294 	error = kthread_create(ithread_loop, ithd, &p, RFSTOPPED | RFHIGHPID,
295 	    0, "%s", name);
296 	if (error)
297 		panic("kthread_create() failed with %d", error);
298 	td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
299 	mtx_lock_spin(&sched_lock);
300 	sched_class(td, PRI_ITHD);
301 	TD_SET_IWAIT(td);
302 	mtx_unlock_spin(&sched_lock);
303 	td->td_pflags |= TDP_ITHREAD;
304 	ithd->it_thread = td;
305 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
306 	return (ithd);
307 }
308 
309 static void
310 ithread_destroy(struct intr_thread *ithread)
311 {
312 	struct thread *td;
313 
314 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
315 	td = ithread->it_thread;
316 	mtx_lock_spin(&sched_lock);
317 	ithread->it_flags |= IT_DEAD;
318 	if (TD_AWAITING_INTR(td)) {
319 		TD_CLR_IWAIT(td);
320 		setrunqueue(td, SRQ_INTR);
321 	}
322 	mtx_unlock_spin(&sched_lock);
323 }
324 
325 int
326 intr_event_add_handler(struct intr_event *ie, const char *name,
327     driver_intr_t handler, void *arg, u_char pri, enum intr_type flags,
328     void **cookiep)
329 {
330 	struct intr_handler *ih, *temp_ih;
331 	struct intr_thread *it;
332 
333 	if (ie == NULL || name == NULL || handler == NULL)
334 		return (EINVAL);
335 
336 	/* Allocate and populate an interrupt handler structure. */
337 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
338 	ih->ih_handler = handler;
339 	ih->ih_argument = arg;
340 	ih->ih_name = name;
341 	ih->ih_event = ie;
342 	ih->ih_pri = pri;
343 	if (flags & INTR_FAST)
344 		ih->ih_flags = IH_FAST;
345 	else if (flags & INTR_EXCL)
346 		ih->ih_flags = IH_EXCLUSIVE;
347 	if (flags & INTR_MPSAFE)
348 		ih->ih_flags |= IH_MPSAFE;
349 	if (flags & INTR_ENTROPY)
350 		ih->ih_flags |= IH_ENTROPY;
351 
352 	/* We can only have one exclusive handler in a event. */
353 	mtx_lock(&ie->ie_lock);
354 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
355 		if ((flags & INTR_EXCL) ||
356 		    (TAILQ_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
357 			mtx_unlock(&ie->ie_lock);
358 			free(ih, M_ITHREAD);
359 			return (EINVAL);
360 		}
361 	}
362 
363 	/* Add the new handler to the event in priority order. */
364 	TAILQ_FOREACH(temp_ih, &ie->ie_handlers, ih_next) {
365 		if (temp_ih->ih_pri > ih->ih_pri)
366 			break;
367 	}
368 	if (temp_ih == NULL)
369 		TAILQ_INSERT_TAIL(&ie->ie_handlers, ih, ih_next);
370 	else
371 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
372 	intr_event_update(ie);
373 
374 	/* Create a thread if we need one. */
375 	while (ie->ie_thread == NULL && !(flags & INTR_FAST)) {
376 		if (ie->ie_flags & IE_ADDING_THREAD)
377 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
378 		else {
379 			ie->ie_flags |= IE_ADDING_THREAD;
380 			mtx_unlock(&ie->ie_lock);
381 			it = ithread_create("intr: newborn");
382 			mtx_lock(&ie->ie_lock);
383 			ie->ie_flags &= ~IE_ADDING_THREAD;
384 			ie->ie_thread = it;
385 			it->it_event = ie;
386 			ithread_update(it);
387 			wakeup(ie);
388 		}
389 	}
390 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
391 	    ie->ie_name);
392 	mtx_unlock(&ie->ie_lock);
393 
394 	if (cookiep != NULL)
395 		*cookiep = ih;
396 	return (0);
397 }
398 
399 /*
400  * Return the ie_source field from the intr_event an intr_handler is
401  * associated with.
402  */
403 void *
404 intr_handler_source(void *cookie)
405 {
406 	struct intr_handler *ih;
407 	struct intr_event *ie;
408 
409 	ih = (struct intr_handler *)cookie;
410 	if (ih == NULL)
411 		return (NULL);
412 	ie = ih->ih_event;
413 	KASSERT(ie != NULL,
414 	    ("interrupt handler \"%s\" has a NULL interrupt event",
415 	    ih->ih_name));
416 	return (ie->ie_source);
417 }
418 
419 int
420 intr_event_remove_handler(void *cookie)
421 {
422 	struct intr_handler *handler = (struct intr_handler *)cookie;
423 	struct intr_event *ie;
424 #ifdef INVARIANTS
425 	struct intr_handler *ih;
426 #endif
427 #ifdef notyet
428 	int dead;
429 #endif
430 
431 	if (handler == NULL)
432 		return (EINVAL);
433 	ie = handler->ih_event;
434 	KASSERT(ie != NULL,
435 	    ("interrupt handler \"%s\" has a NULL interrupt event",
436 	    handler->ih_name));
437 	mtx_lock(&ie->ie_lock);
438 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
439 	    ie->ie_name);
440 #ifdef INVARIANTS
441 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
442 		if (ih == handler)
443 			goto ok;
444 	mtx_unlock(&ie->ie_lock);
445 	panic("interrupt handler \"%s\" not found in interrupt event \"%s\"",
446 	    ih->ih_name, ie->ie_name);
447 ok:
448 #endif
449 	/*
450 	 * If there is no ithread, then just remove the handler and return.
451 	 * XXX: Note that an INTR_FAST handler might be running on another
452 	 * CPU!
453 	 */
454 	if (ie->ie_thread == NULL) {
455 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
456 		mtx_unlock(&ie->ie_lock);
457 		free(handler, M_ITHREAD);
458 		return (0);
459 	}
460 
461 	/*
462 	 * If the interrupt thread is already running, then just mark this
463 	 * handler as being dead and let the ithread do the actual removal.
464 	 *
465 	 * During a cold boot while cold is set, msleep() does not sleep,
466 	 * so we have to remove the handler here rather than letting the
467 	 * thread do it.
468 	 */
469 	mtx_lock_spin(&sched_lock);
470 	if (!TD_AWAITING_INTR(ie->ie_thread->it_thread) && !cold) {
471 		handler->ih_flags |= IH_DEAD;
472 
473 		/*
474 		 * Ensure that the thread will process the handler list
475 		 * again and remove this handler if it has already passed
476 		 * it on the list.
477 		 */
478 		ie->ie_thread->it_need = 1;
479 	} else
480 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
481 	mtx_unlock_spin(&sched_lock);
482 	while (handler->ih_flags & IH_DEAD)
483 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
484 	intr_event_update(ie);
485 #ifdef notyet
486 	/*
487 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
488 	 * this could lead to races of stale data when servicing an
489 	 * interrupt.
490 	 */
491 	dead = 1;
492 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
493 		if (!(ih->ih_flags & IH_FAST)) {
494 			dead = 0;
495 			break;
496 		}
497 	}
498 	if (dead) {
499 		ithread_destroy(ie->ie_thread);
500 		ie->ie_thread = NULL;
501 	}
502 #endif
503 	mtx_unlock(&ie->ie_lock);
504 	free(handler, M_ITHREAD);
505 	return (0);
506 }
507 
508 int
509 intr_event_schedule_thread(struct intr_event *ie)
510 {
511 	struct intr_entropy entropy;
512 	struct intr_thread *it;
513 	struct thread *td;
514 	struct thread *ctd;
515 	struct proc *p;
516 
517 	/*
518 	 * If no ithread or no handlers, then we have a stray interrupt.
519 	 */
520 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers) ||
521 	    ie->ie_thread == NULL)
522 		return (EINVAL);
523 
524 	ctd = curthread;
525 	it = ie->ie_thread;
526 	td = it->it_thread;
527 	p = td->td_proc;
528 
529 	/*
530 	 * If any of the handlers for this ithread claim to be good
531 	 * sources of entropy, then gather some.
532 	 */
533 	if (harvest.interrupt && ie->ie_flags & IE_ENTROPY) {
534 		CTR3(KTR_INTR, "%s: pid %d (%s) gathering entropy", __func__,
535 		    p->p_pid, p->p_comm);
536 		entropy.event = (uintptr_t)ie;
537 		entropy.td = ctd;
538 		random_harvest(&entropy, sizeof(entropy), 2, 0,
539 		    RANDOM_INTERRUPT);
540 	}
541 
542 	KASSERT(p != NULL, ("ithread %s has no process", ie->ie_name));
543 
544 	/*
545 	 * Set it_need to tell the thread to keep running if it is already
546 	 * running.  Then, grab sched_lock and see if we actually need to
547 	 * put this thread on the runqueue.
548 	 */
549 	it->it_need = 1;
550 	mtx_lock_spin(&sched_lock);
551 	if (TD_AWAITING_INTR(td)) {
552 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, p->p_pid,
553 		    p->p_comm);
554 		TD_CLR_IWAIT(td);
555 		setrunqueue(td, SRQ_INTR);
556 	} else {
557 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
558 		    __func__, p->p_pid, p->p_comm, it->it_need, td->td_state);
559 	}
560 	mtx_unlock_spin(&sched_lock);
561 
562 	return (0);
563 }
564 
565 /*
566  * Add a software interrupt handler to a specified event.  If a given event
567  * is not specified, then a new event is created.
568  */
569 int
570 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
571 	    void *arg, int pri, enum intr_type flags, void **cookiep)
572 {
573 	struct intr_event *ie;
574 	int error;
575 
576 	if (flags & (INTR_FAST | INTR_ENTROPY))
577 		return (EINVAL);
578 
579 	ie = (eventp != NULL) ? *eventp : NULL;
580 
581 	if (ie != NULL) {
582 		if (!(ie->ie_flags & IE_SOFT))
583 			return (EINVAL);
584 	} else {
585 		error = intr_event_create(&ie, NULL, IE_SOFT, NULL,
586 		    "swi%d:", pri);
587 		if (error)
588 			return (error);
589 		if (eventp != NULL)
590 			*eventp = ie;
591 	}
592 	return (intr_event_add_handler(ie, name, handler, arg,
593 		    (pri * RQ_PPQ) + PI_SOFT, flags, cookiep));
594 		    /* XXKSE.. think of a better way to get separate queues */
595 }
596 
597 /*
598  * Schedule a software interrupt thread.
599  */
600 void
601 swi_sched(void *cookie, int flags)
602 {
603 	struct intr_handler *ih = (struct intr_handler *)cookie;
604 	struct intr_event *ie = ih->ih_event;
605 	int error;
606 
607 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
608 	    ih->ih_need);
609 
610 	/*
611 	 * Set ih_need for this handler so that if the ithread is already
612 	 * running it will execute this handler on the next pass.  Otherwise,
613 	 * it will execute it the next time it runs.
614 	 */
615 	atomic_store_rel_int(&ih->ih_need, 1);
616 
617 	if (!(flags & SWI_DELAY)) {
618 		PCPU_LAZY_INC(cnt.v_soft);
619 		error = intr_event_schedule_thread(ie);
620 		KASSERT(error == 0, ("stray software interrupt"));
621 	}
622 }
623 
624 /*
625  * Remove a software interrupt handler.  Currently this code does not
626  * remove the associated interrupt event if it becomes empty.  Calling code
627  * may do so manually via intr_event_destroy(), but that's not really
628  * an optimal interface.
629  */
630 int
631 swi_remove(void *cookie)
632 {
633 
634 	return (intr_event_remove_handler(cookie));
635 }
636 
637 static void
638 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
639 {
640 	struct intr_handler *ih, *ihn;
641 
642 	/* Interrupt handlers should not sleep. */
643 	if (!(ie->ie_flags & IE_SOFT))
644 		THREAD_NO_SLEEPING();
645 	TAILQ_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
646 
647 		/*
648 		 * If this handler is marked for death, remove it from
649 		 * the list of handlers and wake up the sleeper.
650 		 */
651 		if (ih->ih_flags & IH_DEAD) {
652 			mtx_lock(&ie->ie_lock);
653 			TAILQ_REMOVE(&ie->ie_handlers, ih, ih_next);
654 			ih->ih_flags &= ~IH_DEAD;
655 			wakeup(ih);
656 			mtx_unlock(&ie->ie_lock);
657 			continue;
658 		}
659 
660 		/*
661 		 * For software interrupt threads, we only execute
662 		 * handlers that have their need flag set.  Hardware
663 		 * interrupt threads always invoke all of their handlers.
664 		 */
665 		if (ie->ie_flags & IE_SOFT) {
666 			if (!ih->ih_need)
667 				continue;
668 			else
669 				atomic_store_rel_int(&ih->ih_need, 0);
670 		}
671 
672 		/* Fast handlers are handled in primary interrupt context. */
673 		if (ih->ih_flags & IH_FAST)
674 			continue;
675 
676 		/* Execute this handler. */
677 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
678 		    __func__, p->p_pid, (void *)ih->ih_handler, ih->ih_argument,
679 		    ih->ih_name, ih->ih_flags);
680 
681 		if (!(ih->ih_flags & IH_MPSAFE))
682 			mtx_lock(&Giant);
683 		ih->ih_handler(ih->ih_argument);
684 		if (!(ih->ih_flags & IH_MPSAFE))
685 			mtx_unlock(&Giant);
686 	}
687 	if (!(ie->ie_flags & IE_SOFT))
688 		THREAD_SLEEPING_OK();
689 
690 	/*
691 	 * Interrupt storm handling:
692 	 *
693 	 * If this interrupt source is currently storming, then throttle
694 	 * it to only fire the handler once  per clock tick.
695 	 *
696 	 * If this interrupt source is not currently storming, but the
697 	 * number of back to back interrupts exceeds the storm threshold,
698 	 * then enter storming mode.
699 	 */
700 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold) {
701 		if (ie->ie_warned == 0) {
702 			printf(
703 	"Interrupt storm detected on \"%s\"; throttling interrupt source\n",
704 			    ie->ie_name);
705 			ie->ie_warned = 1;
706 		}
707 		tsleep(&ie->ie_count, 0, "istorm", 1);
708 	} else
709 		ie->ie_count++;
710 
711 	/*
712 	 * Now that all the handlers have had a chance to run, reenable
713 	 * the interrupt source.
714 	 */
715 	if (ie->ie_enable != NULL)
716 		ie->ie_enable(ie->ie_source);
717 }
718 
719 /*
720  * This is the main code for interrupt threads.
721  */
722 static void
723 ithread_loop(void *arg)
724 {
725 	struct intr_thread *ithd;
726 	struct intr_event *ie;
727 	struct thread *td;
728 	struct proc *p;
729 
730 	td = curthread;
731 	p = td->td_proc;
732 	ithd = (struct intr_thread *)arg;
733 	KASSERT(ithd->it_thread == td,
734 	    ("%s: ithread and proc linkage out of sync", __func__));
735 	ie = ithd->it_event;
736 	ie->ie_count = 0;
737 
738 	/*
739 	 * As long as we have interrupts outstanding, go through the
740 	 * list of handlers, giving each one a go at it.
741 	 */
742 	for (;;) {
743 		/*
744 		 * If we are an orphaned thread, then just die.
745 		 */
746 		if (ithd->it_flags & IT_DEAD) {
747 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
748 			    p->p_pid, p->p_comm);
749 			free(ithd, M_ITHREAD);
750 			kthread_exit(0);
751 		}
752 
753 		/*
754 		 * Service interrupts.  If another interrupt arrives while
755 		 * we are running, it will set it_need to note that we
756 		 * should make another pass.
757 		 */
758 		while (ithd->it_need) {
759 			/*
760 			 * This might need a full read and write barrier
761 			 * to make sure that this write posts before any
762 			 * of the memory or device accesses in the
763 			 * handlers.
764 			 */
765 			atomic_store_rel_int(&ithd->it_need, 0);
766 			ithread_execute_handlers(p, ie);
767 		}
768 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
769 		mtx_assert(&Giant, MA_NOTOWNED);
770 
771 		/*
772 		 * Processed all our interrupts.  Now get the sched
773 		 * lock.  This may take a while and it_need may get
774 		 * set again, so we have to check it again.
775 		 */
776 		mtx_lock_spin(&sched_lock);
777 		if (!ithd->it_need && !(ithd->it_flags & IT_DEAD)) {
778 			TD_SET_IWAIT(td);
779 			ie->ie_count = 0;
780 			mi_switch(SW_VOL, NULL);
781 		}
782 		mtx_unlock_spin(&sched_lock);
783 	}
784 }
785 
786 #ifdef DDB
787 /*
788  * Dump details about an interrupt handler
789  */
790 static void
791 db_dump_intrhand(struct intr_handler *ih)
792 {
793 	int comma;
794 
795 	db_printf("\t%-10s ", ih->ih_name);
796 	switch (ih->ih_pri) {
797 	case PI_REALTIME:
798 		db_printf("CLK ");
799 		break;
800 	case PI_AV:
801 		db_printf("AV  ");
802 		break;
803 	case PI_TTYHIGH:
804 	case PI_TTYLOW:
805 		db_printf("TTY ");
806 		break;
807 	case PI_TAPE:
808 		db_printf("TAPE");
809 		break;
810 	case PI_NET:
811 		db_printf("NET ");
812 		break;
813 	case PI_DISK:
814 	case PI_DISKLOW:
815 		db_printf("DISK");
816 		break;
817 	case PI_DULL:
818 		db_printf("DULL");
819 		break;
820 	default:
821 		if (ih->ih_pri >= PI_SOFT)
822 			db_printf("SWI ");
823 		else
824 			db_printf("%4u", ih->ih_pri);
825 		break;
826 	}
827 	db_printf(" ");
828 	db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
829 	db_printf("(%p)", ih->ih_argument);
830 	if (ih->ih_need ||
831 	    (ih->ih_flags & (IH_FAST | IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
832 	    IH_MPSAFE)) != 0) {
833 		db_printf(" {");
834 		comma = 0;
835 		if (ih->ih_flags & IH_FAST) {
836 			db_printf("FAST");
837 			comma = 1;
838 		}
839 		if (ih->ih_flags & IH_EXCLUSIVE) {
840 			if (comma)
841 				db_printf(", ");
842 			db_printf("EXCL");
843 			comma = 1;
844 		}
845 		if (ih->ih_flags & IH_ENTROPY) {
846 			if (comma)
847 				db_printf(", ");
848 			db_printf("ENTROPY");
849 			comma = 1;
850 		}
851 		if (ih->ih_flags & IH_DEAD) {
852 			if (comma)
853 				db_printf(", ");
854 			db_printf("DEAD");
855 			comma = 1;
856 		}
857 		if (ih->ih_flags & IH_MPSAFE) {
858 			if (comma)
859 				db_printf(", ");
860 			db_printf("MPSAFE");
861 			comma = 1;
862 		}
863 		if (ih->ih_need) {
864 			if (comma)
865 				db_printf(", ");
866 			db_printf("NEED");
867 		}
868 		db_printf("}");
869 	}
870 	db_printf("\n");
871 }
872 
873 /*
874  * Dump details about a event.
875  */
876 void
877 db_dump_intr_event(struct intr_event *ie, int handlers)
878 {
879 	struct intr_handler *ih;
880 	struct intr_thread *it;
881 	int comma;
882 
883 	db_printf("%s ", ie->ie_fullname);
884 	it = ie->ie_thread;
885 	if (it != NULL)
886 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
887 	else
888 		db_printf("(no thread)");
889 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
890 	    (it != NULL && it->it_need)) {
891 		db_printf(" {");
892 		comma = 0;
893 		if (ie->ie_flags & IE_SOFT) {
894 			db_printf("SOFT");
895 			comma = 1;
896 		}
897 		if (ie->ie_flags & IE_ENTROPY) {
898 			if (comma)
899 				db_printf(", ");
900 			db_printf("ENTROPY");
901 			comma = 1;
902 		}
903 		if (ie->ie_flags & IE_ADDING_THREAD) {
904 			if (comma)
905 				db_printf(", ");
906 			db_printf("ADDING_THREAD");
907 			comma = 1;
908 		}
909 		if (it != NULL && it->it_need) {
910 			if (comma)
911 				db_printf(", ");
912 			db_printf("NEED");
913 		}
914 		db_printf("}");
915 	}
916 	db_printf("\n");
917 
918 	if (handlers)
919 		TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
920 		    db_dump_intrhand(ih);
921 }
922 
923 /*
924  * Dump data about interrupt handlers
925  */
926 DB_SHOW_COMMAND(intr, db_show_intr)
927 {
928 	struct intr_event *ie;
929 	int all, verbose;
930 
931 	verbose = index(modif, 'v') != NULL;
932 	all = index(modif, 'a') != NULL;
933 	TAILQ_FOREACH(ie, &event_list, ie_list) {
934 		if (!all && TAILQ_EMPTY(&ie->ie_handlers))
935 			continue;
936 		db_dump_intr_event(ie, verbose);
937 		if (db_pager_quit)
938 			break;
939 	}
940 }
941 #endif /* DDB */
942 
943 /*
944  * Start standard software interrupt threads
945  */
946 static void
947 start_softintr(void *dummy)
948 {
949 	struct proc *p;
950 
951 	if (swi_add(&clk_intr_event, "clock", softclock, NULL, SWI_CLOCK,
952 		INTR_MPSAFE, &softclock_ih) ||
953 	    swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
954 		panic("died while creating standard software ithreads");
955 
956 	p = clk_intr_event->ie_thread->it_thread->td_proc;
957 	PROC_LOCK(p);
958 	p->p_flag |= P_NOLOAD;
959 	PROC_UNLOCK(p);
960 }
961 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr, NULL)
962 
963 /*
964  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
965  * The data for this machine dependent, and the declarations are in machine
966  * dependent code.  The layout of intrnames and intrcnt however is machine
967  * independent.
968  *
969  * We do not know the length of intrcnt and intrnames at compile time, so
970  * calculate things at run time.
971  */
972 static int
973 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
974 {
975 	return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
976 	   req));
977 }
978 
979 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
980     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
981 
982 static int
983 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
984 {
985 	return (sysctl_handle_opaque(oidp, intrcnt,
986 	    (char *)eintrcnt - (char *)intrcnt, req));
987 }
988 
989 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
990     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
991 
992 #ifdef DDB
993 /*
994  * DDB command to dump the interrupt statistics.
995  */
996 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
997 {
998 	u_long *i;
999 	char *cp;
1000 
1001 	cp = intrnames;
1002 	for (i = intrcnt; i != eintrcnt && !db_pager_quit; i++) {
1003 		if (*cp == '\0')
1004 			break;
1005 		if (*i != 0)
1006 			db_printf("%s\t%lu\n", cp, *i);
1007 		cp += strlen(cp) + 1;
1008 	}
1009 }
1010 #endif
1011