xref: /freebsd/sys/kern/kern_intr.c (revision 70e0bbedef95258a4dadc996d641a9bebd3f107d)
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/cpuset.h>
36 #include <sys/rtprio.h>
37 #include <sys/systm.h>
38 #include <sys/interrupt.h>
39 #include <sys/kernel.h>
40 #include <sys/kthread.h>
41 #include <sys/ktr.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mutex.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/random.h>
49 #include <sys/resourcevar.h>
50 #include <sys/sched.h>
51 #include <sys/smp.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/unistd.h>
55 #include <sys/vmmeter.h>
56 #include <machine/atomic.h>
57 #include <machine/cpu.h>
58 #include <machine/md_var.h>
59 #include <machine/stdarg.h>
60 #ifdef DDB
61 #include <ddb/ddb.h>
62 #include <ddb/db_sym.h>
63 #endif
64 
65 /*
66  * Describe an interrupt thread.  There is one of these per interrupt event.
67  */
68 struct intr_thread {
69 	struct intr_event *it_event;
70 	struct thread *it_thread;	/* Kernel thread. */
71 	int	it_flags;		/* (j) IT_* flags. */
72 	int	it_need;		/* Needs service. */
73 };
74 
75 /* Interrupt thread flags kept in it_flags */
76 #define	IT_DEAD		0x000001	/* Thread is waiting to exit. */
77 #define	IT_WAIT		0x000002	/* Thread is waiting for completion. */
78 
79 struct	intr_entropy {
80 	struct	thread *td;
81 	uintptr_t event;
82 };
83 
84 struct	intr_event *clk_intr_event;
85 struct	intr_event *tty_intr_event;
86 void	*vm_ih;
87 struct proc *intrproc;
88 
89 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
90 
91 static int intr_storm_threshold = 1000;
92 TUNABLE_INT("hw.intr_storm_threshold", &intr_storm_threshold);
93 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RW,
94     &intr_storm_threshold, 0,
95     "Number of consecutive interrupts before storm protection is enabled");
96 static TAILQ_HEAD(, intr_event) event_list =
97     TAILQ_HEAD_INITIALIZER(event_list);
98 static struct mtx event_lock;
99 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
100 
101 static void	intr_event_update(struct intr_event *ie);
102 #ifdef INTR_FILTER
103 static int	intr_event_schedule_thread(struct intr_event *ie,
104 		    struct intr_thread *ithd);
105 static int	intr_filter_loop(struct intr_event *ie,
106 		    struct trapframe *frame, struct intr_thread **ithd);
107 static struct intr_thread *ithread_create(const char *name,
108 			      struct intr_handler *ih);
109 #else
110 static int	intr_event_schedule_thread(struct intr_event *ie);
111 static struct intr_thread *ithread_create(const char *name);
112 #endif
113 static void	ithread_destroy(struct intr_thread *ithread);
114 static void	ithread_execute_handlers(struct proc *p,
115 		    struct intr_event *ie);
116 #ifdef INTR_FILTER
117 static void	priv_ithread_execute_handler(struct proc *p,
118 		    struct intr_handler *ih);
119 #endif
120 static void	ithread_loop(void *);
121 static void	ithread_update(struct intr_thread *ithd);
122 static void	start_softintr(void *);
123 
124 /* Map an interrupt type to an ithread priority. */
125 u_char
126 intr_priority(enum intr_type flags)
127 {
128 	u_char pri;
129 
130 	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
131 	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
132 	switch (flags) {
133 	case INTR_TYPE_TTY:
134 		pri = PI_TTY;
135 		break;
136 	case INTR_TYPE_BIO:
137 		pri = PI_DISK;
138 		break;
139 	case INTR_TYPE_NET:
140 		pri = PI_NET;
141 		break;
142 	case INTR_TYPE_CAM:
143 		pri = PI_DISK;
144 		break;
145 	case INTR_TYPE_AV:
146 		pri = PI_AV;
147 		break;
148 	case INTR_TYPE_CLK:
149 		pri = PI_REALTIME;
150 		break;
151 	case INTR_TYPE_MISC:
152 		pri = PI_DULL;          /* don't care */
153 		break;
154 	default:
155 		/* We didn't specify an interrupt level. */
156 		panic("intr_priority: no interrupt type in flags");
157 	}
158 
159 	return pri;
160 }
161 
162 /*
163  * Update an ithread based on the associated intr_event.
164  */
165 static void
166 ithread_update(struct intr_thread *ithd)
167 {
168 	struct intr_event *ie;
169 	struct thread *td;
170 	u_char pri;
171 
172 	ie = ithd->it_event;
173 	td = ithd->it_thread;
174 
175 	/* Determine the overall priority of this event. */
176 	if (TAILQ_EMPTY(&ie->ie_handlers))
177 		pri = PRI_MAX_ITHD;
178 	else
179 		pri = TAILQ_FIRST(&ie->ie_handlers)->ih_pri;
180 
181 	/* Update name and priority. */
182 	strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
183 	thread_lock(td);
184 	sched_prio(td, pri);
185 	thread_unlock(td);
186 }
187 
188 /*
189  * Regenerate the full name of an interrupt event and update its priority.
190  */
191 static void
192 intr_event_update(struct intr_event *ie)
193 {
194 	struct intr_handler *ih;
195 	char *last;
196 	int missed, space;
197 
198 	/* Start off with no entropy and just the name of the event. */
199 	mtx_assert(&ie->ie_lock, MA_OWNED);
200 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
201 	ie->ie_flags &= ~IE_ENTROPY;
202 	missed = 0;
203 	space = 1;
204 
205 	/* Run through all the handlers updating values. */
206 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
207 		if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
208 		    sizeof(ie->ie_fullname)) {
209 			strcat(ie->ie_fullname, " ");
210 			strcat(ie->ie_fullname, ih->ih_name);
211 			space = 0;
212 		} else
213 			missed++;
214 		if (ih->ih_flags & IH_ENTROPY)
215 			ie->ie_flags |= IE_ENTROPY;
216 	}
217 
218 	/*
219 	 * If the handler names were too long, add +'s to indicate missing
220 	 * names. If we run out of room and still have +'s to add, change
221 	 * the last character from a + to a *.
222 	 */
223 	last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
224 	while (missed-- > 0) {
225 		if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
226 			if (*last == '+') {
227 				*last = '*';
228 				break;
229 			} else
230 				*last = '+';
231 		} else if (space) {
232 			strcat(ie->ie_fullname, " +");
233 			space = 0;
234 		} else
235 			strcat(ie->ie_fullname, "+");
236 	}
237 
238 	/*
239 	 * If this event has an ithread, update it's priority and
240 	 * name.
241 	 */
242 	if (ie->ie_thread != NULL)
243 		ithread_update(ie->ie_thread);
244 	CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
245 }
246 
247 int
248 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
249     void (*pre_ithread)(void *), void (*post_ithread)(void *),
250     void (*post_filter)(void *), int (*assign_cpu)(void *, u_char),
251     const char *fmt, ...)
252 {
253 	struct intr_event *ie;
254 	va_list ap;
255 
256 	/* The only valid flag during creation is IE_SOFT. */
257 	if ((flags & ~IE_SOFT) != 0)
258 		return (EINVAL);
259 	ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
260 	ie->ie_source = source;
261 	ie->ie_pre_ithread = pre_ithread;
262 	ie->ie_post_ithread = post_ithread;
263 	ie->ie_post_filter = post_filter;
264 	ie->ie_assign_cpu = assign_cpu;
265 	ie->ie_flags = flags;
266 	ie->ie_irq = irq;
267 	ie->ie_cpu = NOCPU;
268 	TAILQ_INIT(&ie->ie_handlers);
269 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
270 
271 	va_start(ap, fmt);
272 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
273 	va_end(ap);
274 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
275 	mtx_lock(&event_lock);
276 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
277 	mtx_unlock(&event_lock);
278 	if (event != NULL)
279 		*event = ie;
280 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
281 	return (0);
282 }
283 
284 /*
285  * Bind an interrupt event to the specified CPU.  Note that not all
286  * platforms support binding an interrupt to a CPU.  For those
287  * platforms this request will fail.  For supported platforms, any
288  * associated ithreads as well as the primary interrupt context will
289  * be bound to the specificed CPU.  Using a cpu id of NOCPU unbinds
290  * the interrupt event.
291  */
292 int
293 intr_event_bind(struct intr_event *ie, u_char cpu)
294 {
295 	cpuset_t mask;
296 	lwpid_t id;
297 	int error;
298 
299 	/* Need a CPU to bind to. */
300 	if (cpu != NOCPU && CPU_ABSENT(cpu))
301 		return (EINVAL);
302 
303 	if (ie->ie_assign_cpu == NULL)
304 		return (EOPNOTSUPP);
305 
306 	error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
307 	if (error)
308 		return (error);
309 
310 	/*
311 	 * If we have any ithreads try to set their mask first to verify
312 	 * permissions, etc.
313 	 */
314 	mtx_lock(&ie->ie_lock);
315 	if (ie->ie_thread != NULL) {
316 		CPU_ZERO(&mask);
317 		if (cpu == NOCPU)
318 			CPU_COPY(cpuset_root, &mask);
319 		else
320 			CPU_SET(cpu, &mask);
321 		id = ie->ie_thread->it_thread->td_tid;
322 		mtx_unlock(&ie->ie_lock);
323 		error = cpuset_setthread(id, &mask);
324 		if (error)
325 			return (error);
326 	} else
327 		mtx_unlock(&ie->ie_lock);
328 	error = ie->ie_assign_cpu(ie->ie_source, cpu);
329 	if (error) {
330 		mtx_lock(&ie->ie_lock);
331 		if (ie->ie_thread != NULL) {
332 			CPU_ZERO(&mask);
333 			if (ie->ie_cpu == NOCPU)
334 				CPU_COPY(cpuset_root, &mask);
335 			else
336 				CPU_SET(cpu, &mask);
337 			id = ie->ie_thread->it_thread->td_tid;
338 			mtx_unlock(&ie->ie_lock);
339 			(void)cpuset_setthread(id, &mask);
340 		} else
341 			mtx_unlock(&ie->ie_lock);
342 		return (error);
343 	}
344 
345 	mtx_lock(&ie->ie_lock);
346 	ie->ie_cpu = cpu;
347 	mtx_unlock(&ie->ie_lock);
348 
349 	return (error);
350 }
351 
352 static struct intr_event *
353 intr_lookup(int irq)
354 {
355 	struct intr_event *ie;
356 
357 	mtx_lock(&event_lock);
358 	TAILQ_FOREACH(ie, &event_list, ie_list)
359 		if (ie->ie_irq == irq &&
360 		    (ie->ie_flags & IE_SOFT) == 0 &&
361 		    TAILQ_FIRST(&ie->ie_handlers) != NULL)
362 			break;
363 	mtx_unlock(&event_lock);
364 	return (ie);
365 }
366 
367 int
368 intr_setaffinity(int irq, void *m)
369 {
370 	struct intr_event *ie;
371 	cpuset_t *mask;
372 	u_char cpu;
373 	int n;
374 
375 	mask = m;
376 	cpu = NOCPU;
377 	/*
378 	 * If we're setting all cpus we can unbind.  Otherwise make sure
379 	 * only one cpu is in the set.
380 	 */
381 	if (CPU_CMP(cpuset_root, mask)) {
382 		for (n = 0; n < CPU_SETSIZE; n++) {
383 			if (!CPU_ISSET(n, mask))
384 				continue;
385 			if (cpu != NOCPU)
386 				return (EINVAL);
387 			cpu = (u_char)n;
388 		}
389 	}
390 	ie = intr_lookup(irq);
391 	if (ie == NULL)
392 		return (ESRCH);
393 	return (intr_event_bind(ie, cpu));
394 }
395 
396 int
397 intr_getaffinity(int irq, void *m)
398 {
399 	struct intr_event *ie;
400 	cpuset_t *mask;
401 
402 	mask = m;
403 	ie = intr_lookup(irq);
404 	if (ie == NULL)
405 		return (ESRCH);
406 	CPU_ZERO(mask);
407 	mtx_lock(&ie->ie_lock);
408 	if (ie->ie_cpu == NOCPU)
409 		CPU_COPY(cpuset_root, mask);
410 	else
411 		CPU_SET(ie->ie_cpu, mask);
412 	mtx_unlock(&ie->ie_lock);
413 	return (0);
414 }
415 
416 int
417 intr_event_destroy(struct intr_event *ie)
418 {
419 
420 	mtx_lock(&event_lock);
421 	mtx_lock(&ie->ie_lock);
422 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
423 		mtx_unlock(&ie->ie_lock);
424 		mtx_unlock(&event_lock);
425 		return (EBUSY);
426 	}
427 	TAILQ_REMOVE(&event_list, ie, ie_list);
428 #ifndef notyet
429 	if (ie->ie_thread != NULL) {
430 		ithread_destroy(ie->ie_thread);
431 		ie->ie_thread = NULL;
432 	}
433 #endif
434 	mtx_unlock(&ie->ie_lock);
435 	mtx_unlock(&event_lock);
436 	mtx_destroy(&ie->ie_lock);
437 	free(ie, M_ITHREAD);
438 	return (0);
439 }
440 
441 #ifndef INTR_FILTER
442 static struct intr_thread *
443 ithread_create(const char *name)
444 {
445 	struct intr_thread *ithd;
446 	struct thread *td;
447 	int error;
448 
449 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
450 
451 	error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
452 		    &td, RFSTOPPED | RFHIGHPID,
453 	    	    0, "intr", "%s", name);
454 	if (error)
455 		panic("kproc_create() failed with %d", error);
456 	thread_lock(td);
457 	sched_class(td, PRI_ITHD);
458 	TD_SET_IWAIT(td);
459 	thread_unlock(td);
460 	td->td_pflags |= TDP_ITHREAD;
461 	ithd->it_thread = td;
462 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
463 	return (ithd);
464 }
465 #else
466 static struct intr_thread *
467 ithread_create(const char *name, struct intr_handler *ih)
468 {
469 	struct intr_thread *ithd;
470 	struct thread *td;
471 	int error;
472 
473 	ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
474 
475 	error = kproc_kthread_add(ithread_loop, ih, &intrproc,
476 		    &td, RFSTOPPED | RFHIGHPID,
477 	    	    0, "intr", "%s", name);
478 	if (error)
479 		panic("kproc_create() failed with %d", error);
480 	thread_lock(td);
481 	sched_class(td, PRI_ITHD);
482 	TD_SET_IWAIT(td);
483 	thread_unlock(td);
484 	td->td_pflags |= TDP_ITHREAD;
485 	ithd->it_thread = td;
486 	CTR2(KTR_INTR, "%s: created %s", __func__, name);
487 	return (ithd);
488 }
489 #endif
490 
491 static void
492 ithread_destroy(struct intr_thread *ithread)
493 {
494 	struct thread *td;
495 
496 	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
497 	td = ithread->it_thread;
498 	thread_lock(td);
499 	ithread->it_flags |= IT_DEAD;
500 	if (TD_AWAITING_INTR(td)) {
501 		TD_CLR_IWAIT(td);
502 		sched_add(td, SRQ_INTR);
503 	}
504 	thread_unlock(td);
505 }
506 
507 #ifndef INTR_FILTER
508 int
509 intr_event_add_handler(struct intr_event *ie, const char *name,
510     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
511     enum intr_type flags, void **cookiep)
512 {
513 	struct intr_handler *ih, *temp_ih;
514 	struct intr_thread *it;
515 
516 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
517 		return (EINVAL);
518 
519 	/* Allocate and populate an interrupt handler structure. */
520 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
521 	ih->ih_filter = filter;
522 	ih->ih_handler = handler;
523 	ih->ih_argument = arg;
524 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
525 	ih->ih_event = ie;
526 	ih->ih_pri = pri;
527 	if (flags & INTR_EXCL)
528 		ih->ih_flags = IH_EXCLUSIVE;
529 	if (flags & INTR_MPSAFE)
530 		ih->ih_flags |= IH_MPSAFE;
531 	if (flags & INTR_ENTROPY)
532 		ih->ih_flags |= IH_ENTROPY;
533 
534 	/* We can only have one exclusive handler in a event. */
535 	mtx_lock(&ie->ie_lock);
536 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
537 		if ((flags & INTR_EXCL) ||
538 		    (TAILQ_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
539 			mtx_unlock(&ie->ie_lock);
540 			free(ih, M_ITHREAD);
541 			return (EINVAL);
542 		}
543 	}
544 
545 	/* Add the new handler to the event in priority order. */
546 	TAILQ_FOREACH(temp_ih, &ie->ie_handlers, ih_next) {
547 		if (temp_ih->ih_pri > ih->ih_pri)
548 			break;
549 	}
550 	if (temp_ih == NULL)
551 		TAILQ_INSERT_TAIL(&ie->ie_handlers, ih, ih_next);
552 	else
553 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
554 	intr_event_update(ie);
555 
556 	/* Create a thread if we need one. */
557 	while (ie->ie_thread == NULL && handler != NULL) {
558 		if (ie->ie_flags & IE_ADDING_THREAD)
559 			msleep(ie, &ie->ie_lock, 0, "ithread", 0);
560 		else {
561 			ie->ie_flags |= IE_ADDING_THREAD;
562 			mtx_unlock(&ie->ie_lock);
563 			it = ithread_create("intr: newborn");
564 			mtx_lock(&ie->ie_lock);
565 			ie->ie_flags &= ~IE_ADDING_THREAD;
566 			ie->ie_thread = it;
567 			it->it_event = ie;
568 			ithread_update(it);
569 			wakeup(ie);
570 		}
571 	}
572 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
573 	    ie->ie_name);
574 	mtx_unlock(&ie->ie_lock);
575 
576 	if (cookiep != NULL)
577 		*cookiep = ih;
578 	return (0);
579 }
580 #else
581 int
582 intr_event_add_handler(struct intr_event *ie, const char *name,
583     driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
584     enum intr_type flags, void **cookiep)
585 {
586 	struct intr_handler *ih, *temp_ih;
587 	struct intr_thread *it;
588 
589 	if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
590 		return (EINVAL);
591 
592 	/* Allocate and populate an interrupt handler structure. */
593 	ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
594 	ih->ih_filter = filter;
595 	ih->ih_handler = handler;
596 	ih->ih_argument = arg;
597 	strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
598 	ih->ih_event = ie;
599 	ih->ih_pri = pri;
600 	if (flags & INTR_EXCL)
601 		ih->ih_flags = IH_EXCLUSIVE;
602 	if (flags & INTR_MPSAFE)
603 		ih->ih_flags |= IH_MPSAFE;
604 	if (flags & INTR_ENTROPY)
605 		ih->ih_flags |= IH_ENTROPY;
606 
607 	/* We can only have one exclusive handler in a event. */
608 	mtx_lock(&ie->ie_lock);
609 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
610 		if ((flags & INTR_EXCL) ||
611 		    (TAILQ_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
612 			mtx_unlock(&ie->ie_lock);
613 			free(ih, M_ITHREAD);
614 			return (EINVAL);
615 		}
616 	}
617 
618 	/* Add the new handler to the event in priority order. */
619 	TAILQ_FOREACH(temp_ih, &ie->ie_handlers, ih_next) {
620 		if (temp_ih->ih_pri > ih->ih_pri)
621 			break;
622 	}
623 	if (temp_ih == NULL)
624 		TAILQ_INSERT_TAIL(&ie->ie_handlers, ih, ih_next);
625 	else
626 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
627 	intr_event_update(ie);
628 
629 	/* For filtered handlers, create a private ithread to run on. */
630 	if (filter != NULL && handler != NULL) {
631 		mtx_unlock(&ie->ie_lock);
632 		it = ithread_create("intr: newborn", ih);
633 		mtx_lock(&ie->ie_lock);
634 		it->it_event = ie;
635 		ih->ih_thread = it;
636 		ithread_update(it); // XXX - do we really need this?!?!?
637 	} else { /* Create the global per-event thread if we need one. */
638 		while (ie->ie_thread == NULL && handler != NULL) {
639 			if (ie->ie_flags & IE_ADDING_THREAD)
640 				msleep(ie, &ie->ie_lock, 0, "ithread", 0);
641 			else {
642 				ie->ie_flags |= IE_ADDING_THREAD;
643 				mtx_unlock(&ie->ie_lock);
644 				it = ithread_create("intr: newborn", ih);
645 				mtx_lock(&ie->ie_lock);
646 				ie->ie_flags &= ~IE_ADDING_THREAD;
647 				ie->ie_thread = it;
648 				it->it_event = ie;
649 				ithread_update(it);
650 				wakeup(ie);
651 			}
652 		}
653 	}
654 	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
655 	    ie->ie_name);
656 	mtx_unlock(&ie->ie_lock);
657 
658 	if (cookiep != NULL)
659 		*cookiep = ih;
660 	return (0);
661 }
662 #endif
663 
664 /*
665  * Append a description preceded by a ':' to the name of the specified
666  * interrupt handler.
667  */
668 int
669 intr_event_describe_handler(struct intr_event *ie, void *cookie,
670     const char *descr)
671 {
672 	struct intr_handler *ih;
673 	size_t space;
674 	char *start;
675 
676 	mtx_lock(&ie->ie_lock);
677 #ifdef INVARIANTS
678 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
679 		if (ih == cookie)
680 			break;
681 	}
682 	if (ih == NULL) {
683 		mtx_unlock(&ie->ie_lock);
684 		panic("handler %p not found in interrupt event %p", cookie, ie);
685 	}
686 #endif
687 	ih = cookie;
688 
689 	/*
690 	 * Look for an existing description by checking for an
691 	 * existing ":".  This assumes device names do not include
692 	 * colons.  If one is found, prepare to insert the new
693 	 * description at that point.  If one is not found, find the
694 	 * end of the name to use as the insertion point.
695 	 */
696 	start = strchr(ih->ih_name, ':');
697 	if (start == NULL)
698 		start = strchr(ih->ih_name, 0);
699 
700 	/*
701 	 * See if there is enough remaining room in the string for the
702 	 * description + ":".  The "- 1" leaves room for the trailing
703 	 * '\0'.  The "+ 1" accounts for the colon.
704 	 */
705 	space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
706 	if (strlen(descr) + 1 > space) {
707 		mtx_unlock(&ie->ie_lock);
708 		return (ENOSPC);
709 	}
710 
711 	/* Append a colon followed by the description. */
712 	*start = ':';
713 	strcpy(start + 1, descr);
714 	intr_event_update(ie);
715 	mtx_unlock(&ie->ie_lock);
716 	return (0);
717 }
718 
719 /*
720  * Return the ie_source field from the intr_event an intr_handler is
721  * associated with.
722  */
723 void *
724 intr_handler_source(void *cookie)
725 {
726 	struct intr_handler *ih;
727 	struct intr_event *ie;
728 
729 	ih = (struct intr_handler *)cookie;
730 	if (ih == NULL)
731 		return (NULL);
732 	ie = ih->ih_event;
733 	KASSERT(ie != NULL,
734 	    ("interrupt handler \"%s\" has a NULL interrupt event",
735 	    ih->ih_name));
736 	return (ie->ie_source);
737 }
738 
739 /*
740  * Sleep until an ithread finishes executing an interrupt handler.
741  *
742  * XXX Doesn't currently handle interrupt filters or fast interrupt
743  * handlers.  This is intended for compatibility with linux drivers
744  * only.  Do not use in BSD code.
745  */
746 void
747 _intr_drain(int irq)
748 {
749 	struct intr_event *ie;
750 	struct intr_thread *ithd;
751 	struct thread *td;
752 
753 	ie = intr_lookup(irq);
754 	if (ie == NULL)
755 		return;
756 	if (ie->ie_thread == NULL)
757 		return;
758 	ithd = ie->ie_thread;
759 	td = ithd->it_thread;
760 	/*
761 	 * We set the flag and wait for it to be cleared to avoid
762 	 * long delays with potentially busy interrupt handlers
763 	 * were we to only sample TD_AWAITING_INTR() every tick.
764 	 */
765 	thread_lock(td);
766 	if (!TD_AWAITING_INTR(td)) {
767 		ithd->it_flags |= IT_WAIT;
768 		while (ithd->it_flags & IT_WAIT) {
769 			thread_unlock(td);
770 			pause("idrain", 1);
771 			thread_lock(td);
772 		}
773 	}
774 	thread_unlock(td);
775 	return;
776 }
777 
778 
779 #ifndef INTR_FILTER
780 int
781 intr_event_remove_handler(void *cookie)
782 {
783 	struct intr_handler *handler = (struct intr_handler *)cookie;
784 	struct intr_event *ie;
785 #ifdef INVARIANTS
786 	struct intr_handler *ih;
787 #endif
788 #ifdef notyet
789 	int dead;
790 #endif
791 
792 	if (handler == NULL)
793 		return (EINVAL);
794 	ie = handler->ih_event;
795 	KASSERT(ie != NULL,
796 	    ("interrupt handler \"%s\" has a NULL interrupt event",
797 	    handler->ih_name));
798 	mtx_lock(&ie->ie_lock);
799 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
800 	    ie->ie_name);
801 #ifdef INVARIANTS
802 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
803 		if (ih == handler)
804 			goto ok;
805 	mtx_unlock(&ie->ie_lock);
806 	panic("interrupt handler \"%s\" not found in interrupt event \"%s\"",
807 	    ih->ih_name, ie->ie_name);
808 ok:
809 #endif
810 	/*
811 	 * If there is no ithread, then just remove the handler and return.
812 	 * XXX: Note that an INTR_FAST handler might be running on another
813 	 * CPU!
814 	 */
815 	if (ie->ie_thread == NULL) {
816 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
817 		mtx_unlock(&ie->ie_lock);
818 		free(handler, M_ITHREAD);
819 		return (0);
820 	}
821 
822 	/*
823 	 * If the interrupt thread is already running, then just mark this
824 	 * handler as being dead and let the ithread do the actual removal.
825 	 *
826 	 * During a cold boot while cold is set, msleep() does not sleep,
827 	 * so we have to remove the handler here rather than letting the
828 	 * thread do it.
829 	 */
830 	thread_lock(ie->ie_thread->it_thread);
831 	if (!TD_AWAITING_INTR(ie->ie_thread->it_thread) && !cold) {
832 		handler->ih_flags |= IH_DEAD;
833 
834 		/*
835 		 * Ensure that the thread will process the handler list
836 		 * again and remove this handler if it has already passed
837 		 * it on the list.
838 		 */
839 		ie->ie_thread->it_need = 1;
840 	} else
841 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
842 	thread_unlock(ie->ie_thread->it_thread);
843 	while (handler->ih_flags & IH_DEAD)
844 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
845 	intr_event_update(ie);
846 #ifdef notyet
847 	/*
848 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
849 	 * this could lead to races of stale data when servicing an
850 	 * interrupt.
851 	 */
852 	dead = 1;
853 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
854 		if (!(ih->ih_flags & IH_FAST)) {
855 			dead = 0;
856 			break;
857 		}
858 	}
859 	if (dead) {
860 		ithread_destroy(ie->ie_thread);
861 		ie->ie_thread = NULL;
862 	}
863 #endif
864 	mtx_unlock(&ie->ie_lock);
865 	free(handler, M_ITHREAD);
866 	return (0);
867 }
868 
869 static int
870 intr_event_schedule_thread(struct intr_event *ie)
871 {
872 	struct intr_entropy entropy;
873 	struct intr_thread *it;
874 	struct thread *td;
875 	struct thread *ctd;
876 	struct proc *p;
877 
878 	/*
879 	 * If no ithread or no handlers, then we have a stray interrupt.
880 	 */
881 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers) ||
882 	    ie->ie_thread == NULL)
883 		return (EINVAL);
884 
885 	ctd = curthread;
886 	it = ie->ie_thread;
887 	td = it->it_thread;
888 	p = td->td_proc;
889 
890 	/*
891 	 * If any of the handlers for this ithread claim to be good
892 	 * sources of entropy, then gather some.
893 	 */
894 	if (harvest.interrupt && ie->ie_flags & IE_ENTROPY) {
895 		CTR3(KTR_INTR, "%s: pid %d (%s) gathering entropy", __func__,
896 		    p->p_pid, td->td_name);
897 		entropy.event = (uintptr_t)ie;
898 		entropy.td = ctd;
899 		random_harvest(&entropy, sizeof(entropy), 2, 0,
900 		    RANDOM_INTERRUPT);
901 	}
902 
903 	KASSERT(p != NULL, ("ithread %s has no process", ie->ie_name));
904 
905 	/*
906 	 * Set it_need to tell the thread to keep running if it is already
907 	 * running.  Then, lock the thread and see if we actually need to
908 	 * put it on the runqueue.
909 	 */
910 	it->it_need = 1;
911 	thread_lock(td);
912 	if (TD_AWAITING_INTR(td)) {
913 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, p->p_pid,
914 		    td->td_name);
915 		TD_CLR_IWAIT(td);
916 		sched_add(td, SRQ_INTR);
917 	} else {
918 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
919 		    __func__, p->p_pid, td->td_name, it->it_need, td->td_state);
920 	}
921 	thread_unlock(td);
922 
923 	return (0);
924 }
925 #else
926 int
927 intr_event_remove_handler(void *cookie)
928 {
929 	struct intr_handler *handler = (struct intr_handler *)cookie;
930 	struct intr_event *ie;
931 	struct intr_thread *it;
932 #ifdef INVARIANTS
933 	struct intr_handler *ih;
934 #endif
935 #ifdef notyet
936 	int dead;
937 #endif
938 
939 	if (handler == NULL)
940 		return (EINVAL);
941 	ie = handler->ih_event;
942 	KASSERT(ie != NULL,
943 	    ("interrupt handler \"%s\" has a NULL interrupt event",
944 	    handler->ih_name));
945 	mtx_lock(&ie->ie_lock);
946 	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
947 	    ie->ie_name);
948 #ifdef INVARIANTS
949 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
950 		if (ih == handler)
951 			goto ok;
952 	mtx_unlock(&ie->ie_lock);
953 	panic("interrupt handler \"%s\" not found in interrupt event \"%s\"",
954 	    ih->ih_name, ie->ie_name);
955 ok:
956 #endif
957 	/*
958 	 * If there are no ithreads (per event and per handler), then
959 	 * just remove the handler and return.
960 	 * XXX: Note that an INTR_FAST handler might be running on another CPU!
961 	 */
962 	if (ie->ie_thread == NULL && handler->ih_thread == NULL) {
963 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
964 		mtx_unlock(&ie->ie_lock);
965 		free(handler, M_ITHREAD);
966 		return (0);
967 	}
968 
969 	/* Private or global ithread? */
970 	it = (handler->ih_thread) ? handler->ih_thread : ie->ie_thread;
971 	/*
972 	 * If the interrupt thread is already running, then just mark this
973 	 * handler as being dead and let the ithread do the actual removal.
974 	 *
975 	 * During a cold boot while cold is set, msleep() does not sleep,
976 	 * so we have to remove the handler here rather than letting the
977 	 * thread do it.
978 	 */
979 	thread_lock(it->it_thread);
980 	if (!TD_AWAITING_INTR(it->it_thread) && !cold) {
981 		handler->ih_flags |= IH_DEAD;
982 
983 		/*
984 		 * Ensure that the thread will process the handler list
985 		 * again and remove this handler if it has already passed
986 		 * it on the list.
987 		 */
988 		it->it_need = 1;
989 	} else
990 		TAILQ_REMOVE(&ie->ie_handlers, handler, ih_next);
991 	thread_unlock(it->it_thread);
992 	while (handler->ih_flags & IH_DEAD)
993 		msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
994 	/*
995 	 * At this point, the handler has been disconnected from the event,
996 	 * so we can kill the private ithread if any.
997 	 */
998 	if (handler->ih_thread) {
999 		ithread_destroy(handler->ih_thread);
1000 		handler->ih_thread = NULL;
1001 	}
1002 	intr_event_update(ie);
1003 #ifdef notyet
1004 	/*
1005 	 * XXX: This could be bad in the case of ppbus(8).  Also, I think
1006 	 * this could lead to races of stale data when servicing an
1007 	 * interrupt.
1008 	 */
1009 	dead = 1;
1010 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
1011 		if (handler != NULL) {
1012 			dead = 0;
1013 			break;
1014 		}
1015 	}
1016 	if (dead) {
1017 		ithread_destroy(ie->ie_thread);
1018 		ie->ie_thread = NULL;
1019 	}
1020 #endif
1021 	mtx_unlock(&ie->ie_lock);
1022 	free(handler, M_ITHREAD);
1023 	return (0);
1024 }
1025 
1026 static int
1027 intr_event_schedule_thread(struct intr_event *ie, struct intr_thread *it)
1028 {
1029 	struct intr_entropy entropy;
1030 	struct thread *td;
1031 	struct thread *ctd;
1032 	struct proc *p;
1033 
1034 	/*
1035 	 * If no ithread or no handlers, then we have a stray interrupt.
1036 	 */
1037 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers) || it == NULL)
1038 		return (EINVAL);
1039 
1040 	ctd = curthread;
1041 	td = it->it_thread;
1042 	p = td->td_proc;
1043 
1044 	/*
1045 	 * If any of the handlers for this ithread claim to be good
1046 	 * sources of entropy, then gather some.
1047 	 */
1048 	if (harvest.interrupt && ie->ie_flags & IE_ENTROPY) {
1049 		CTR3(KTR_INTR, "%s: pid %d (%s) gathering entropy", __func__,
1050 		    p->p_pid, td->td_name);
1051 		entropy.event = (uintptr_t)ie;
1052 		entropy.td = ctd;
1053 		random_harvest(&entropy, sizeof(entropy), 2, 0,
1054 		    RANDOM_INTERRUPT);
1055 	}
1056 
1057 	KASSERT(p != NULL, ("ithread %s has no process", ie->ie_name));
1058 
1059 	/*
1060 	 * Set it_need to tell the thread to keep running if it is already
1061 	 * running.  Then, lock the thread and see if we actually need to
1062 	 * put it on the runqueue.
1063 	 */
1064 	it->it_need = 1;
1065 	thread_lock(td);
1066 	if (TD_AWAITING_INTR(td)) {
1067 		CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, p->p_pid,
1068 		    td->td_name);
1069 		TD_CLR_IWAIT(td);
1070 		sched_add(td, SRQ_INTR);
1071 	} else {
1072 		CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
1073 		    __func__, p->p_pid, td->td_name, it->it_need, td->td_state);
1074 	}
1075 	thread_unlock(td);
1076 
1077 	return (0);
1078 }
1079 #endif
1080 
1081 /*
1082  * Allow interrupt event binding for software interrupt handlers -- a no-op,
1083  * since interrupts are generated in software rather than being directed by
1084  * a PIC.
1085  */
1086 static int
1087 swi_assign_cpu(void *arg, u_char cpu)
1088 {
1089 
1090 	return (0);
1091 }
1092 
1093 /*
1094  * Add a software interrupt handler to a specified event.  If a given event
1095  * is not specified, then a new event is created.
1096  */
1097 int
1098 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
1099 	    void *arg, int pri, enum intr_type flags, void **cookiep)
1100 {
1101 	struct thread *td;
1102 	struct intr_event *ie;
1103 	int error;
1104 
1105 	if (flags & INTR_ENTROPY)
1106 		return (EINVAL);
1107 
1108 	ie = (eventp != NULL) ? *eventp : NULL;
1109 
1110 	if (ie != NULL) {
1111 		if (!(ie->ie_flags & IE_SOFT))
1112 			return (EINVAL);
1113 	} else {
1114 		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1115 		    NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1116 		if (error)
1117 			return (error);
1118 		if (eventp != NULL)
1119 			*eventp = ie;
1120 	}
1121 	error = intr_event_add_handler(ie, name, NULL, handler, arg,
1122 	    PI_SWI(pri), flags, cookiep);
1123 	if (error)
1124 		return (error);
1125 	if (pri == SWI_CLOCK) {
1126 		td = ie->ie_thread->it_thread;
1127 		thread_lock(td);
1128 		td->td_flags |= TDF_NOLOAD;
1129 		thread_unlock(td);
1130 	}
1131 	return (0);
1132 }
1133 
1134 /*
1135  * Schedule a software interrupt thread.
1136  */
1137 void
1138 swi_sched(void *cookie, int flags)
1139 {
1140 	struct intr_handler *ih = (struct intr_handler *)cookie;
1141 	struct intr_event *ie = ih->ih_event;
1142 	int error;
1143 
1144 	CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1145 	    ih->ih_need);
1146 
1147 	/*
1148 	 * Set ih_need for this handler so that if the ithread is already
1149 	 * running it will execute this handler on the next pass.  Otherwise,
1150 	 * it will execute it the next time it runs.
1151 	 */
1152 	atomic_store_rel_int(&ih->ih_need, 1);
1153 
1154 	if (!(flags & SWI_DELAY)) {
1155 		PCPU_INC(cnt.v_soft);
1156 #ifdef INTR_FILTER
1157 		error = intr_event_schedule_thread(ie, ie->ie_thread);
1158 #else
1159 		error = intr_event_schedule_thread(ie);
1160 #endif
1161 		KASSERT(error == 0, ("stray software interrupt"));
1162 	}
1163 }
1164 
1165 /*
1166  * Remove a software interrupt handler.  Currently this code does not
1167  * remove the associated interrupt event if it becomes empty.  Calling code
1168  * may do so manually via intr_event_destroy(), but that's not really
1169  * an optimal interface.
1170  */
1171 int
1172 swi_remove(void *cookie)
1173 {
1174 
1175 	return (intr_event_remove_handler(cookie));
1176 }
1177 
1178 #ifdef INTR_FILTER
1179 static void
1180 priv_ithread_execute_handler(struct proc *p, struct intr_handler *ih)
1181 {
1182 	struct intr_event *ie;
1183 
1184 	ie = ih->ih_event;
1185 	/*
1186 	 * If this handler is marked for death, remove it from
1187 	 * the list of handlers and wake up the sleeper.
1188 	 */
1189 	if (ih->ih_flags & IH_DEAD) {
1190 		mtx_lock(&ie->ie_lock);
1191 		TAILQ_REMOVE(&ie->ie_handlers, ih, ih_next);
1192 		ih->ih_flags &= ~IH_DEAD;
1193 		wakeup(ih);
1194 		mtx_unlock(&ie->ie_lock);
1195 		return;
1196 	}
1197 
1198 	/* Execute this handler. */
1199 	CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1200 	     __func__, p->p_pid, (void *)ih->ih_handler, ih->ih_argument,
1201 	     ih->ih_name, ih->ih_flags);
1202 
1203 	if (!(ih->ih_flags & IH_MPSAFE))
1204 		mtx_lock(&Giant);
1205 	ih->ih_handler(ih->ih_argument);
1206 	if (!(ih->ih_flags & IH_MPSAFE))
1207 		mtx_unlock(&Giant);
1208 }
1209 #endif
1210 
1211 /*
1212  * This is a public function for use by drivers that mux interrupt
1213  * handlers for child devices from their interrupt handler.
1214  */
1215 void
1216 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1217 {
1218 	struct intr_handler *ih, *ihn;
1219 
1220 	TAILQ_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1221 		/*
1222 		 * If this handler is marked for death, remove it from
1223 		 * the list of handlers and wake up the sleeper.
1224 		 */
1225 		if (ih->ih_flags & IH_DEAD) {
1226 			mtx_lock(&ie->ie_lock);
1227 			TAILQ_REMOVE(&ie->ie_handlers, ih, ih_next);
1228 			ih->ih_flags &= ~IH_DEAD;
1229 			wakeup(ih);
1230 			mtx_unlock(&ie->ie_lock);
1231 			continue;
1232 		}
1233 
1234 		/* Skip filter only handlers */
1235 		if (ih->ih_handler == NULL)
1236 			continue;
1237 
1238 		/*
1239 		 * For software interrupt threads, we only execute
1240 		 * handlers that have their need flag set.  Hardware
1241 		 * interrupt threads always invoke all of their handlers.
1242 		 */
1243 		if (ie->ie_flags & IE_SOFT) {
1244 			if (!ih->ih_need)
1245 				continue;
1246 			else
1247 				atomic_store_rel_int(&ih->ih_need, 0);
1248 		}
1249 
1250 		/* Execute this handler. */
1251 		CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1252 		    __func__, p->p_pid, (void *)ih->ih_handler,
1253 		    ih->ih_argument, ih->ih_name, ih->ih_flags);
1254 
1255 		if (!(ih->ih_flags & IH_MPSAFE))
1256 			mtx_lock(&Giant);
1257 		ih->ih_handler(ih->ih_argument);
1258 		if (!(ih->ih_flags & IH_MPSAFE))
1259 			mtx_unlock(&Giant);
1260 	}
1261 }
1262 
1263 static void
1264 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1265 {
1266 
1267 	/* Interrupt handlers should not sleep. */
1268 	if (!(ie->ie_flags & IE_SOFT))
1269 		THREAD_NO_SLEEPING();
1270 	intr_event_execute_handlers(p, ie);
1271 	if (!(ie->ie_flags & IE_SOFT))
1272 		THREAD_SLEEPING_OK();
1273 
1274 	/*
1275 	 * Interrupt storm handling:
1276 	 *
1277 	 * If this interrupt source is currently storming, then throttle
1278 	 * it to only fire the handler once  per clock tick.
1279 	 *
1280 	 * If this interrupt source is not currently storming, but the
1281 	 * number of back to back interrupts exceeds the storm threshold,
1282 	 * then enter storming mode.
1283 	 */
1284 	if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1285 	    !(ie->ie_flags & IE_SOFT)) {
1286 		/* Report the message only once every second. */
1287 		if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1288 			printf(
1289 	"interrupt storm detected on \"%s\"; throttling interrupt source\n",
1290 			    ie->ie_name);
1291 		}
1292 		pause("istorm", 1);
1293 	} else
1294 		ie->ie_count++;
1295 
1296 	/*
1297 	 * Now that all the handlers have had a chance to run, reenable
1298 	 * the interrupt source.
1299 	 */
1300 	if (ie->ie_post_ithread != NULL)
1301 		ie->ie_post_ithread(ie->ie_source);
1302 }
1303 
1304 #ifndef INTR_FILTER
1305 /*
1306  * This is the main code for interrupt threads.
1307  */
1308 static void
1309 ithread_loop(void *arg)
1310 {
1311 	struct intr_thread *ithd;
1312 	struct intr_event *ie;
1313 	struct thread *td;
1314 	struct proc *p;
1315 	int wake;
1316 
1317 	td = curthread;
1318 	p = td->td_proc;
1319 	ithd = (struct intr_thread *)arg;
1320 	KASSERT(ithd->it_thread == td,
1321 	    ("%s: ithread and proc linkage out of sync", __func__));
1322 	ie = ithd->it_event;
1323 	ie->ie_count = 0;
1324 	wake = 0;
1325 
1326 	/*
1327 	 * As long as we have interrupts outstanding, go through the
1328 	 * list of handlers, giving each one a go at it.
1329 	 */
1330 	for (;;) {
1331 		/*
1332 		 * If we are an orphaned thread, then just die.
1333 		 */
1334 		if (ithd->it_flags & IT_DEAD) {
1335 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1336 			    p->p_pid, td->td_name);
1337 			free(ithd, M_ITHREAD);
1338 			kthread_exit();
1339 		}
1340 
1341 		/*
1342 		 * Service interrupts.  If another interrupt arrives while
1343 		 * we are running, it will set it_need to note that we
1344 		 * should make another pass.
1345 		 */
1346 		while (ithd->it_need) {
1347 			/*
1348 			 * This might need a full read and write barrier
1349 			 * to make sure that this write posts before any
1350 			 * of the memory or device accesses in the
1351 			 * handlers.
1352 			 */
1353 			atomic_store_rel_int(&ithd->it_need, 0);
1354 			ithread_execute_handlers(p, ie);
1355 		}
1356 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1357 		mtx_assert(&Giant, MA_NOTOWNED);
1358 
1359 		/*
1360 		 * Processed all our interrupts.  Now get the sched
1361 		 * lock.  This may take a while and it_need may get
1362 		 * set again, so we have to check it again.
1363 		 */
1364 		thread_lock(td);
1365 		if (!ithd->it_need && !(ithd->it_flags & (IT_DEAD | IT_WAIT))) {
1366 			TD_SET_IWAIT(td);
1367 			ie->ie_count = 0;
1368 			mi_switch(SW_VOL | SWT_IWAIT, NULL);
1369 		}
1370 		if (ithd->it_flags & IT_WAIT) {
1371 			wake = 1;
1372 			ithd->it_flags &= ~IT_WAIT;
1373 		}
1374 		thread_unlock(td);
1375 		if (wake) {
1376 			wakeup(ithd);
1377 			wake = 0;
1378 		}
1379 	}
1380 }
1381 
1382 /*
1383  * Main interrupt handling body.
1384  *
1385  * Input:
1386  * o ie:                        the event connected to this interrupt.
1387  * o frame:                     some archs (i.e. i386) pass a frame to some.
1388  *                              handlers as their main argument.
1389  * Return value:
1390  * o 0:                         everything ok.
1391  * o EINVAL:                    stray interrupt.
1392  */
1393 int
1394 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1395 {
1396 	struct intr_handler *ih;
1397 	struct trapframe *oldframe;
1398 	struct thread *td;
1399 	int error, ret, thread;
1400 
1401 	td = curthread;
1402 
1403 	/* An interrupt with no event or handlers is a stray interrupt. */
1404 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers))
1405 		return (EINVAL);
1406 
1407 	/*
1408 	 * Execute fast interrupt handlers directly.
1409 	 * To support clock handlers, if a handler registers
1410 	 * with a NULL argument, then we pass it a pointer to
1411 	 * a trapframe as its argument.
1412 	 */
1413 	td->td_intr_nesting_level++;
1414 	thread = 0;
1415 	ret = 0;
1416 	critical_enter();
1417 	oldframe = td->td_intr_frame;
1418 	td->td_intr_frame = frame;
1419 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
1420 		if (ih->ih_filter == NULL) {
1421 			thread = 1;
1422 			continue;
1423 		}
1424 		CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1425 		    ih->ih_filter, ih->ih_argument == NULL ? frame :
1426 		    ih->ih_argument, ih->ih_name);
1427 		if (ih->ih_argument == NULL)
1428 			ret = ih->ih_filter(frame);
1429 		else
1430 			ret = ih->ih_filter(ih->ih_argument);
1431 		KASSERT(ret == FILTER_STRAY ||
1432 		    ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1433 		    (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1434 		    ("%s: incorrect return value %#x from %s", __func__, ret,
1435 		    ih->ih_name));
1436 
1437 		/*
1438 		 * Wrapper handler special handling:
1439 		 *
1440 		 * in some particular cases (like pccard and pccbb),
1441 		 * the _real_ device handler is wrapped in a couple of
1442 		 * functions - a filter wrapper and an ithread wrapper.
1443 		 * In this case (and just in this case), the filter wrapper
1444 		 * could ask the system to schedule the ithread and mask
1445 		 * the interrupt source if the wrapped handler is composed
1446 		 * of just an ithread handler.
1447 		 *
1448 		 * TODO: write a generic wrapper to avoid people rolling
1449 		 * their own
1450 		 */
1451 		if (!thread) {
1452 			if (ret == FILTER_SCHEDULE_THREAD)
1453 				thread = 1;
1454 		}
1455 	}
1456 	td->td_intr_frame = oldframe;
1457 
1458 	if (thread) {
1459 		if (ie->ie_pre_ithread != NULL)
1460 			ie->ie_pre_ithread(ie->ie_source);
1461 	} else {
1462 		if (ie->ie_post_filter != NULL)
1463 			ie->ie_post_filter(ie->ie_source);
1464 	}
1465 
1466 	/* Schedule the ithread if needed. */
1467 	if (thread) {
1468 		error = intr_event_schedule_thread(ie);
1469 #ifndef XEN
1470 		KASSERT(error == 0, ("bad stray interrupt"));
1471 #else
1472 		if (error != 0)
1473 			log(LOG_WARNING, "bad stray interrupt");
1474 #endif
1475 	}
1476 	critical_exit();
1477 	td->td_intr_nesting_level--;
1478 	return (0);
1479 }
1480 #else
1481 /*
1482  * This is the main code for interrupt threads.
1483  */
1484 static void
1485 ithread_loop(void *arg)
1486 {
1487 	struct intr_thread *ithd;
1488 	struct intr_handler *ih;
1489 	struct intr_event *ie;
1490 	struct thread *td;
1491 	struct proc *p;
1492 	int priv;
1493 	int wake;
1494 
1495 	td = curthread;
1496 	p = td->td_proc;
1497 	ih = (struct intr_handler *)arg;
1498 	priv = (ih->ih_thread != NULL) ? 1 : 0;
1499 	ithd = (priv) ? ih->ih_thread : ih->ih_event->ie_thread;
1500 	KASSERT(ithd->it_thread == td,
1501 	    ("%s: ithread and proc linkage out of sync", __func__));
1502 	ie = ithd->it_event;
1503 	ie->ie_count = 0;
1504 	wake = 0;
1505 
1506 	/*
1507 	 * As long as we have interrupts outstanding, go through the
1508 	 * list of handlers, giving each one a go at it.
1509 	 */
1510 	for (;;) {
1511 		/*
1512 		 * If we are an orphaned thread, then just die.
1513 		 */
1514 		if (ithd->it_flags & IT_DEAD) {
1515 			CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1516 			    p->p_pid, td->td_name);
1517 			free(ithd, M_ITHREAD);
1518 			kthread_exit();
1519 		}
1520 
1521 		/*
1522 		 * Service interrupts.  If another interrupt arrives while
1523 		 * we are running, it will set it_need to note that we
1524 		 * should make another pass.
1525 		 */
1526 		while (ithd->it_need) {
1527 			/*
1528 			 * This might need a full read and write barrier
1529 			 * to make sure that this write posts before any
1530 			 * of the memory or device accesses in the
1531 			 * handlers.
1532 			 */
1533 			atomic_store_rel_int(&ithd->it_need, 0);
1534 			if (priv)
1535 				priv_ithread_execute_handler(p, ih);
1536 			else
1537 				ithread_execute_handlers(p, ie);
1538 		}
1539 		WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1540 		mtx_assert(&Giant, MA_NOTOWNED);
1541 
1542 		/*
1543 		 * Processed all our interrupts.  Now get the sched
1544 		 * lock.  This may take a while and it_need may get
1545 		 * set again, so we have to check it again.
1546 		 */
1547 		thread_lock(td);
1548 		if (!ithd->it_need && !(ithd->it_flags & (IT_DEAD | IT_WAIT))) {
1549 			TD_SET_IWAIT(td);
1550 			ie->ie_count = 0;
1551 			mi_switch(SW_VOL | SWT_IWAIT, NULL);
1552 		}
1553 		if (ithd->it_flags & IT_WAIT) {
1554 			wake = 1;
1555 			ithd->it_flags &= ~IT_WAIT;
1556 		}
1557 		thread_unlock(td);
1558 		if (wake) {
1559 			wakeup(ithd);
1560 			wake = 0;
1561 		}
1562 	}
1563 }
1564 
1565 /*
1566  * Main loop for interrupt filter.
1567  *
1568  * Some architectures (i386, amd64 and arm) require the optional frame
1569  * parameter, and use it as the main argument for fast handler execution
1570  * when ih_argument == NULL.
1571  *
1572  * Return value:
1573  * o FILTER_STRAY:              No filter recognized the event, and no
1574  *                              filter-less handler is registered on this
1575  *                              line.
1576  * o FILTER_HANDLED:            A filter claimed the event and served it.
1577  * o FILTER_SCHEDULE_THREAD:    No filter claimed the event, but there's at
1578  *                              least one filter-less handler on this line.
1579  * o FILTER_HANDLED |
1580  *   FILTER_SCHEDULE_THREAD:    A filter claimed the event, and asked for
1581  *                              scheduling the per-handler ithread.
1582  *
1583  * In case an ithread has to be scheduled, in *ithd there will be a
1584  * pointer to a struct intr_thread containing the thread to be
1585  * scheduled.
1586  */
1587 
1588 static int
1589 intr_filter_loop(struct intr_event *ie, struct trapframe *frame,
1590 		 struct intr_thread **ithd)
1591 {
1592 	struct intr_handler *ih;
1593 	void *arg;
1594 	int ret, thread_only;
1595 
1596 	ret = 0;
1597 	thread_only = 0;
1598 	TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next) {
1599 		/*
1600 		 * Execute fast interrupt handlers directly.
1601 		 * To support clock handlers, if a handler registers
1602 		 * with a NULL argument, then we pass it a pointer to
1603 		 * a trapframe as its argument.
1604 		 */
1605 		arg = ((ih->ih_argument == NULL) ? frame : ih->ih_argument);
1606 
1607 		CTR5(KTR_INTR, "%s: exec %p/%p(%p) for %s", __func__,
1608 		     ih->ih_filter, ih->ih_handler, arg, ih->ih_name);
1609 
1610 		if (ih->ih_filter != NULL)
1611 			ret = ih->ih_filter(arg);
1612 		else {
1613 			thread_only = 1;
1614 			continue;
1615 		}
1616 		KASSERT(ret == FILTER_STRAY ||
1617 		    ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1618 		    (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1619 		    ("%s: incorrect return value %#x from %s", __func__, ret,
1620 		    ih->ih_name));
1621 		if (ret & FILTER_STRAY)
1622 			continue;
1623 		else {
1624 			*ithd = ih->ih_thread;
1625 			return (ret);
1626 		}
1627 	}
1628 
1629 	/*
1630 	 * No filters handled the interrupt and we have at least
1631 	 * one handler without a filter.  In this case, we schedule
1632 	 * all of the filter-less handlers to run in the ithread.
1633 	 */
1634 	if (thread_only) {
1635 		*ithd = ie->ie_thread;
1636 		return (FILTER_SCHEDULE_THREAD);
1637 	}
1638 	return (FILTER_STRAY);
1639 }
1640 
1641 /*
1642  * Main interrupt handling body.
1643  *
1644  * Input:
1645  * o ie:                        the event connected to this interrupt.
1646  * o frame:                     some archs (i.e. i386) pass a frame to some.
1647  *                              handlers as their main argument.
1648  * Return value:
1649  * o 0:                         everything ok.
1650  * o EINVAL:                    stray interrupt.
1651  */
1652 int
1653 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1654 {
1655 	struct intr_thread *ithd;
1656 	struct trapframe *oldframe;
1657 	struct thread *td;
1658 	int thread;
1659 
1660 	ithd = NULL;
1661 	td = curthread;
1662 
1663 	if (ie == NULL || TAILQ_EMPTY(&ie->ie_handlers))
1664 		return (EINVAL);
1665 
1666 	td->td_intr_nesting_level++;
1667 	thread = 0;
1668 	critical_enter();
1669 	oldframe = td->td_intr_frame;
1670 	td->td_intr_frame = frame;
1671 	thread = intr_filter_loop(ie, frame, &ithd);
1672 	if (thread & FILTER_HANDLED) {
1673 		if (ie->ie_post_filter != NULL)
1674 			ie->ie_post_filter(ie->ie_source);
1675 	} else {
1676 		if (ie->ie_pre_ithread != NULL)
1677 			ie->ie_pre_ithread(ie->ie_source);
1678 	}
1679 	td->td_intr_frame = oldframe;
1680 	critical_exit();
1681 
1682 	/* Interrupt storm logic */
1683 	if (thread & FILTER_STRAY) {
1684 		ie->ie_count++;
1685 		if (ie->ie_count < intr_storm_threshold)
1686 			printf("Interrupt stray detection not present\n");
1687 	}
1688 
1689 	/* Schedule an ithread if needed. */
1690 	if (thread & FILTER_SCHEDULE_THREAD) {
1691 		if (intr_event_schedule_thread(ie, ithd) != 0)
1692 			panic("%s: impossible stray interrupt", __func__);
1693 	}
1694 	td->td_intr_nesting_level--;
1695 	return (0);
1696 }
1697 #endif
1698 
1699 #ifdef DDB
1700 /*
1701  * Dump details about an interrupt handler
1702  */
1703 static void
1704 db_dump_intrhand(struct intr_handler *ih)
1705 {
1706 	int comma;
1707 
1708 	db_printf("\t%-10s ", ih->ih_name);
1709 	switch (ih->ih_pri) {
1710 	case PI_REALTIME:
1711 		db_printf("CLK ");
1712 		break;
1713 	case PI_AV:
1714 		db_printf("AV  ");
1715 		break;
1716 	case PI_TTY:
1717 		db_printf("TTY ");
1718 		break;
1719 	case PI_NET:
1720 		db_printf("NET ");
1721 		break;
1722 	case PI_DISK:
1723 		db_printf("DISK");
1724 		break;
1725 	case PI_DULL:
1726 		db_printf("DULL");
1727 		break;
1728 	default:
1729 		if (ih->ih_pri >= PI_SOFT)
1730 			db_printf("SWI ");
1731 		else
1732 			db_printf("%4u", ih->ih_pri);
1733 		break;
1734 	}
1735 	db_printf(" ");
1736 	db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1737 	db_printf("(%p)", ih->ih_argument);
1738 	if (ih->ih_need ||
1739 	    (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1740 	    IH_MPSAFE)) != 0) {
1741 		db_printf(" {");
1742 		comma = 0;
1743 		if (ih->ih_flags & IH_EXCLUSIVE) {
1744 			if (comma)
1745 				db_printf(", ");
1746 			db_printf("EXCL");
1747 			comma = 1;
1748 		}
1749 		if (ih->ih_flags & IH_ENTROPY) {
1750 			if (comma)
1751 				db_printf(", ");
1752 			db_printf("ENTROPY");
1753 			comma = 1;
1754 		}
1755 		if (ih->ih_flags & IH_DEAD) {
1756 			if (comma)
1757 				db_printf(", ");
1758 			db_printf("DEAD");
1759 			comma = 1;
1760 		}
1761 		if (ih->ih_flags & IH_MPSAFE) {
1762 			if (comma)
1763 				db_printf(", ");
1764 			db_printf("MPSAFE");
1765 			comma = 1;
1766 		}
1767 		if (ih->ih_need) {
1768 			if (comma)
1769 				db_printf(", ");
1770 			db_printf("NEED");
1771 		}
1772 		db_printf("}");
1773 	}
1774 	db_printf("\n");
1775 }
1776 
1777 /*
1778  * Dump details about a event.
1779  */
1780 void
1781 db_dump_intr_event(struct intr_event *ie, int handlers)
1782 {
1783 	struct intr_handler *ih;
1784 	struct intr_thread *it;
1785 	int comma;
1786 
1787 	db_printf("%s ", ie->ie_fullname);
1788 	it = ie->ie_thread;
1789 	if (it != NULL)
1790 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1791 	else
1792 		db_printf("(no thread)");
1793 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
1794 	    (it != NULL && it->it_need)) {
1795 		db_printf(" {");
1796 		comma = 0;
1797 		if (ie->ie_flags & IE_SOFT) {
1798 			db_printf("SOFT");
1799 			comma = 1;
1800 		}
1801 		if (ie->ie_flags & IE_ENTROPY) {
1802 			if (comma)
1803 				db_printf(", ");
1804 			db_printf("ENTROPY");
1805 			comma = 1;
1806 		}
1807 		if (ie->ie_flags & IE_ADDING_THREAD) {
1808 			if (comma)
1809 				db_printf(", ");
1810 			db_printf("ADDING_THREAD");
1811 			comma = 1;
1812 		}
1813 		if (it != NULL && it->it_need) {
1814 			if (comma)
1815 				db_printf(", ");
1816 			db_printf("NEED");
1817 		}
1818 		db_printf("}");
1819 	}
1820 	db_printf("\n");
1821 
1822 	if (handlers)
1823 		TAILQ_FOREACH(ih, &ie->ie_handlers, ih_next)
1824 		    db_dump_intrhand(ih);
1825 }
1826 
1827 /*
1828  * Dump data about interrupt handlers
1829  */
1830 DB_SHOW_COMMAND(intr, db_show_intr)
1831 {
1832 	struct intr_event *ie;
1833 	int all, verbose;
1834 
1835 	verbose = strchr(modif, 'v') != NULL;
1836 	all = strchr(modif, 'a') != NULL;
1837 	TAILQ_FOREACH(ie, &event_list, ie_list) {
1838 		if (!all && TAILQ_EMPTY(&ie->ie_handlers))
1839 			continue;
1840 		db_dump_intr_event(ie, verbose);
1841 		if (db_pager_quit)
1842 			break;
1843 	}
1844 }
1845 #endif /* DDB */
1846 
1847 /*
1848  * Start standard software interrupt threads
1849  */
1850 static void
1851 start_softintr(void *dummy)
1852 {
1853 
1854 	if (swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
1855 		panic("died while creating vm swi ithread");
1856 }
1857 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1858     NULL);
1859 
1860 /*
1861  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1862  * The data for this machine dependent, and the declarations are in machine
1863  * dependent code.  The layout of intrnames and intrcnt however is machine
1864  * independent.
1865  *
1866  * We do not know the length of intrcnt and intrnames at compile time, so
1867  * calculate things at run time.
1868  */
1869 static int
1870 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1871 {
1872 	return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1873 }
1874 
1875 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1876     NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1877 
1878 static int
1879 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1880 {
1881 	return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1882 }
1883 
1884 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1885     NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1886 
1887 #ifdef DDB
1888 /*
1889  * DDB command to dump the interrupt statistics.
1890  */
1891 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
1892 {
1893 	u_long *i;
1894 	char *cp;
1895 	u_int j;
1896 
1897 	cp = intrnames;
1898 	j = 0;
1899 	for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1900 	    i++, j++) {
1901 		if (*cp == '\0')
1902 			break;
1903 		if (*i != 0)
1904 			db_printf("%s\t%lu\n", cp, *i);
1905 		cp += strlen(cp) + 1;
1906 	}
1907 }
1908 #endif
1909