xref: /dflybsd-src/sys/kern/kern_intr.c (revision c9fbf0d3b1d54097180190816f8fa4f5d415b174)
1 /*
2  * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved.
3  * Copyright (c) 1997, Stefan Esser <se@freebsd.org> 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  * $FreeBSD: src/sys/kern/kern_intr.c,v 1.24.2.1 2001/10/14 20:05:50 luigi Exp $
27  * $DragonFly: src/sys/kern/kern_intr.c,v 1.33 2005/11/04 08:17:19 dillon Exp $
28  *
29  */
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/malloc.h>
34 #include <sys/kernel.h>
35 #include <sys/sysctl.h>
36 #include <sys/thread.h>
37 #include <sys/proc.h>
38 #include <sys/thread2.h>
39 #include <sys/random.h>
40 #include <sys/serialize.h>
41 #include <sys/bus.h>
42 #include <sys/machintr.h>
43 
44 #include <machine/ipl.h>
45 #include <machine/frame.h>
46 
47 #include <sys/interrupt.h>
48 
49 typedef struct intrec {
50     struct intrec *next;
51     inthand2_t	*handler;
52     void	*argument;
53     char	*name;
54     int		intr;
55     int		intr_flags;
56     struct lwkt_serialize *serializer;
57 } *intrec_t;
58 
59 struct intr_info {
60 	intrec_t	i_reclist;
61 	struct thread	i_thread;
62 	struct random_softc i_random;
63 	int		i_running;
64 	long		i_count;
65 	int		i_fast;
66 	int		i_slow;
67 	int		i_state;
68 } intr_info_ary[MAX_INTS];
69 
70 int max_installed_hard_intr;
71 int max_installed_soft_intr;
72 
73 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000
74 
75 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS);
76 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS);
77 static void emergency_intr_timer_callback(systimer_t, struct intrframe *);
78 static void ithread_handler(void *arg);
79 static void ithread_emergency(void *arg);
80 
81 int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]);
82 
83 static struct systimer emergency_intr_timer;
84 static struct thread emergency_intr_thread;
85 
86 #define ISTATE_NOTHREAD		0
87 #define ISTATE_NORMAL		1
88 #define ISTATE_LIVELOCKED	2
89 
90 static int livelock_limit = 50000;
91 static int livelock_lowater = 20000;
92 SYSCTL_INT(_kern, OID_AUTO, livelock_limit,
93         CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit");
94 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater,
95         CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore");
96 
97 static int emergency_intr_enable = 0;	/* emergency interrupt polling */
98 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable);
99 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW,
100         0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable");
101 
102 static int emergency_intr_freq = 10;	/* emergency polling frequency */
103 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq);
104 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW,
105         0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency");
106 
107 /*
108  * Sysctl support routines
109  */
110 static int
111 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS)
112 {
113 	int error, enabled;
114 
115 	enabled = emergency_intr_enable;
116 	error = sysctl_handle_int(oidp, &enabled, 0, req);
117 	if (error || req->newptr == NULL)
118 		return error;
119 	emergency_intr_enable = enabled;
120 	if (emergency_intr_enable) {
121 		emergency_intr_timer.periodic =
122 			sys_cputimer->fromhz(emergency_intr_freq);
123 	} else {
124 		emergency_intr_timer.periodic = sys_cputimer->fromhz(1);
125 	}
126 	return 0;
127 }
128 
129 static int
130 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS)
131 {
132         int error, phz;
133 
134         phz = emergency_intr_freq;
135         error = sysctl_handle_int(oidp, &phz, 0, req);
136         if (error || req->newptr == NULL)
137                 return error;
138         if (phz <= 0)
139                 return EINVAL;
140         else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX)
141                 phz = EMERGENCY_INTR_POLLING_FREQ_MAX;
142 
143         emergency_intr_freq = phz;
144 	if (emergency_intr_enable) {
145 		emergency_intr_timer.periodic =
146 			sys_cputimer->fromhz(emergency_intr_freq);
147 	} else {
148 		emergency_intr_timer.periodic = sys_cputimer->fromhz(1);
149 	}
150         return 0;
151 }
152 
153 /*
154  * Register an SWI or INTerrupt handler.
155  */
156 void *
157 register_swi(int intr, inthand2_t *handler, void *arg, const char *name,
158 		struct lwkt_serialize *serializer)
159 {
160     if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
161 	panic("register_swi: bad intr %d", intr);
162     return(register_int(intr, handler, arg, name, serializer, 0));
163 }
164 
165 void *
166 register_int(int intr, inthand2_t *handler, void *arg, const char *name,
167 		struct lwkt_serialize *serializer, int intr_flags)
168 {
169     struct intr_info *info;
170     struct intrec **list;
171     intrec_t rec;
172 
173     if (intr < 0 || intr >= MAX_INTS)
174 	panic("register_int: bad intr %d", intr);
175     if (name == NULL)
176 	name = "???";
177     info = &intr_info_ary[intr];
178 
179     rec = malloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT);
180     rec->name = malloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT);
181     strcpy(rec->name, name);
182 
183     rec->handler = handler;
184     rec->argument = arg;
185     rec->intr = intr;
186     rec->intr_flags = intr_flags;
187     rec->next = NULL;
188     rec->serializer = serializer;
189 
190     list = &info->i_reclist;
191 
192     /*
193      * Keep track of how many fast and slow interrupts we have.
194      */
195     if (intr_flags & INTR_FAST)
196 	++info->i_fast;
197     else
198 	++info->i_slow;
199 
200     /*
201      * Create an emergency polling thread and set up a systimer to wake
202      * it up.
203      */
204     if (emergency_intr_thread.td_kstack == NULL) {
205 	lwkt_create(ithread_emergency, NULL, NULL,
206 		    &emergency_intr_thread, TDF_STOPREQ|TDF_INTTHREAD, -1,
207 		    "ithread emerg");
208 	systimer_init_periodic_nq(&emergency_intr_timer,
209 		    emergency_intr_timer_callback, &emergency_intr_thread,
210 		    (emergency_intr_enable ? emergency_intr_freq : 1));
211     }
212 
213     /*
214      * Create an interrupt thread if necessary, leave it in an unscheduled
215      * state.
216      */
217     if (info->i_state == ISTATE_NOTHREAD) {
218 	info->i_state = ISTATE_NORMAL;
219 	lwkt_create((void *)ithread_handler, (void *)intr, NULL,
220 	    &info->i_thread, TDF_STOPREQ|TDF_INTTHREAD, -1,
221 	    "ithread %d", intr);
222 	if (intr >= FIRST_SOFTINT)
223 	    lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM);
224 	else
225 	    lwkt_setpri(&info->i_thread, TDPRI_INT_MED);
226 	info->i_thread.td_preemptable = lwkt_preempt;
227     }
228 
229     /*
230      * Add the record to the interrupt list
231      */
232     crit_enter();	/* token */
233     while (*list != NULL)
234 	list = &(*list)->next;
235     *list = rec;
236     crit_exit();
237 
238     /*
239      * Update max_installed_hard_intr to make the emergency intr poll
240      * a bit more efficient.
241      */
242     if (intr < FIRST_SOFTINT) {
243 	if (max_installed_hard_intr <= intr)
244 	    max_installed_hard_intr = intr + 1;
245     } else {
246 	if (max_installed_soft_intr <= intr)
247 	    max_installed_soft_intr = intr + 1;
248     }
249     return(rec);
250 }
251 
252 int
253 unregister_swi(void *id)
254 {
255     return(unregister_int(id));
256 }
257 
258 int
259 unregister_int(void *id)
260 {
261     struct intr_info *info;
262     struct intrec **list;
263     intrec_t rec;
264     int intr;
265 
266     intr = ((intrec_t)id)->intr;
267 
268     if (intr < 0 || intr >= MAX_INTS)
269 	panic("register_int: bad intr %d", intr);
270 
271     info = &intr_info_ary[intr];
272 
273     /*
274      * Remove the interrupt descriptor
275      */
276     crit_enter();
277     list = &info->i_reclist;
278     while ((rec = *list) != NULL) {
279 	if (rec == id) {
280 	    *list = rec->next;
281 	    break;
282 	}
283 	list = &rec->next;
284     }
285     crit_exit();
286 
287     /*
288      * Free it, adjust interrupt type counts
289      */
290     if (rec != NULL) {
291 	if (rec->intr_flags & INTR_FAST)
292 	    --info->i_fast;
293 	else
294 	    --info->i_slow;
295 	free(rec->name, M_DEVBUF);
296 	free(rec, M_DEVBUF);
297     } else {
298 	printf("warning: unregister_int: int %d handler for %s not found\n",
299 		intr, ((intrec_t)id)->name);
300     }
301 
302     /*
303      * Return the number of interrupt vectors still registered on this intr
304      */
305     return(info->i_fast + info->i_slow);
306 }
307 
308 int
309 get_registered_intr(void *id)
310 {
311     return(((intrec_t)id)->intr);
312 }
313 
314 const char *
315 get_registered_name(int intr)
316 {
317     intrec_t rec;
318 
319     if (intr < 0 || intr >= MAX_INTS)
320 	panic("register_int: bad intr %d", intr);
321 
322     if ((rec = intr_info_ary[intr].i_reclist) == NULL)
323 	return(NULL);
324     else if (rec->next)
325 	return("mux");
326     else
327 	return(rec->name);
328 }
329 
330 int
331 count_registered_ints(int intr)
332 {
333     struct intr_info *info;
334 
335     if (intr < 0 || intr >= MAX_INTS)
336 	panic("register_int: bad intr %d", intr);
337     info = &intr_info_ary[intr];
338     return(info->i_fast + info->i_slow);
339 }
340 
341 long
342 get_interrupt_counter(int intr)
343 {
344     struct intr_info *info;
345 
346     if (intr < 0 || intr >= MAX_INTS)
347 	panic("register_int: bad intr %d", intr);
348     info = &intr_info_ary[intr];
349     return(info->i_count);
350 }
351 
352 
353 void
354 swi_setpriority(int intr, int pri)
355 {
356     struct intr_info *info;
357 
358     if (intr < FIRST_SOFTINT || intr >= MAX_INTS)
359 	panic("register_swi: bad intr %d", intr);
360     info = &intr_info_ary[intr];
361     if (info->i_state != ISTATE_NOTHREAD)
362 	lwkt_setpri(&info->i_thread, pri);
363 }
364 
365 void
366 register_randintr(int intr)
367 {
368     struct intr_info *info;
369 
370     if (intr < 0 || intr >= MAX_INTS)
371 	panic("register_randintr: bad intr %d", intr);
372     info = &intr_info_ary[intr];
373     info->i_random.sc_intr = intr;
374     info->i_random.sc_enabled = 1;
375 }
376 
377 void
378 unregister_randintr(int intr)
379 {
380     struct intr_info *info;
381 
382     if (intr < 0 || intr >= MAX_INTS)
383 	panic("register_swi: bad intr %d", intr);
384     info = &intr_info_ary[intr];
385     info->i_random.sc_enabled = 0;
386 }
387 
388 int
389 next_registered_randintr(int intr)
390 {
391     struct intr_info *info;
392 
393     if (intr < 0 || intr >= MAX_INTS)
394 	panic("register_swi: bad intr %d", intr);
395     while (intr < MAX_INTS) {
396 	info = &intr_info_ary[intr];
397 	if (info->i_random.sc_enabled)
398 	    break;
399 	++intr;
400     }
401     return(intr);
402 }
403 
404 /*
405  * Dispatch an interrupt.  If there's nothing to do we have a stray
406  * interrupt and can just return, leaving the interrupt masked.
407  *
408  * We need to schedule the interrupt and set its i_running bit.  If
409  * we are not on the interrupt thread's cpu we have to send a message
410  * to the correct cpu that will issue the desired action (interlocking
411  * with the interrupt thread's critical section).  We do NOT attempt to
412  * reschedule interrupts whos i_running bit is already set because
413  * this would prematurely wakeup a livelock-limited interrupt thread.
414  *
415  * i_running is only tested/set on the same cpu as the interrupt thread.
416  *
417  * We are NOT in a critical section, which will allow the scheduled
418  * interrupt to preempt us.  The MP lock might *NOT* be held here.
419  */
420 #ifdef SMP
421 
422 static void
423 sched_ithd_remote(void *arg)
424 {
425     sched_ithd((int)arg);
426 }
427 
428 #endif
429 
430 void
431 sched_ithd(int intr)
432 {
433     struct intr_info *info;
434 
435     info = &intr_info_ary[intr];
436 
437     ++info->i_count;
438     if (info->i_state != ISTATE_NOTHREAD) {
439 	if (info->i_reclist == NULL) {
440 	    printf("sched_ithd: stray interrupt %d\n", intr);
441 	} else {
442 #ifdef SMP
443 	    if (info->i_thread.td_gd == mycpu) {
444 		if (info->i_running == 0) {
445 		    info->i_running = 1;
446 		    if (info->i_state != ISTATE_LIVELOCKED)
447 			lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
448 		}
449 	    } else {
450 		lwkt_send_ipiq(info->i_thread.td_gd,
451 				sched_ithd_remote, (void *)intr);
452 	    }
453 #else
454 	    if (info->i_running == 0) {
455 		info->i_running = 1;
456 		if (info->i_state != ISTATE_LIVELOCKED)
457 		    lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */
458 	    }
459 #endif
460 	}
461     } else {
462 	printf("sched_ithd: stray interrupt %d\n", intr);
463     }
464 }
465 
466 /*
467  * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL
468  * might not be held).
469  */
470 static void
471 ithread_livelock_wakeup(systimer_t st)
472 {
473     struct intr_info *info;
474 
475     info = &intr_info_ary[(int)st->data];
476     if (info->i_state != ISTATE_NOTHREAD)
477 	lwkt_schedule(&info->i_thread);
478 }
479 
480 /*
481  * This function is called drectly from the ICU or APIC vector code assembly
482  * to process an interrupt.  The critical section and interrupt deferral
483  * checks have already been done but the function is entered WITHOUT
484  * a critical section held.  The BGL may or may not be held.
485  *
486  * Must return non-zero if we do not want the vector code to re-enable
487  * the interrupt (which we don't if we have to schedule the interrupt)
488  */
489 int ithread_fast_handler(struct intrframe frame);
490 
491 int
492 ithread_fast_handler(struct intrframe frame)
493 {
494     int intr;
495     struct intr_info *info;
496     struct intrec **list;
497     int must_schedule;
498 #ifdef SMP
499     int got_mplock;
500 #endif
501     intrec_t rec, next_rec;
502     globaldata_t gd;
503 
504     intr = frame.if_vec;
505     gd = mycpu;
506 
507     info = &intr_info_ary[intr];
508 
509     /*
510      * If we are not processing any FAST interrupts, just schedule the thing.
511      * (since we aren't in a critical section, this can result in a
512      * preemption)
513      */
514     if (info->i_fast == 0) {
515 	sched_ithd(intr);
516 	return(1);
517     }
518 
519     /*
520      * This should not normally occur since interrupts ought to be
521      * masked if the ithread has been scheduled or is running.
522      */
523     if (info->i_running)
524 	return(1);
525 
526     /*
527      * Bump the interrupt nesting level to process any FAST interrupts.
528      * Obtain the MP lock as necessary.  If the MP lock cannot be obtained,
529      * schedule the interrupt thread to deal with the issue instead.
530      *
531      * To reduce overhead, just leave the MP lock held once it has been
532      * obtained.
533      */
534     crit_enter_gd(gd);
535     ++gd->gd_intr_nesting_level;
536     ++gd->gd_cnt.v_intr;
537     must_schedule = info->i_slow;
538 #ifdef SMP
539     got_mplock = 0;
540 #endif
541 
542     list = &info->i_reclist;
543     for (rec = *list; rec; rec = next_rec) {
544 	next_rec = rec->next;	/* rec may be invalid after call */
545 
546 	if (rec->intr_flags & INTR_FAST) {
547 #ifdef SMP
548 	    if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) {
549 		if (try_mplock() == 0) {
550 		    int owner;
551 
552 		    /*
553 		     * If we couldn't get the MP lock try to forward it
554 		     * to the cpu holding the MP lock, setting must_schedule
555 		     * to -1 so we do not schedule and also do not unmask
556 		     * the interrupt.  Otherwise just schedule it.
557 		     */
558 		    owner = owner_mplock();
559 		    if (owner >= 0 && owner != gd->gd_cpuid) {
560 			lwkt_send_ipiq_bycpu(owner, forward_fastint_remote,
561 						(void *)intr);
562 			must_schedule = -1;
563 			++gd->gd_cnt.v_forwarded_ints;
564 		    } else {
565 			must_schedule = 1;
566 		    }
567 		    break;
568 		}
569 		got_mplock = 1;
570 	    }
571 #endif
572 	    if (rec->serializer) {
573 		must_schedule += lwkt_serialize_handler_try(
574 					rec->serializer, rec->handler,
575 					rec->argument, &frame);
576 	    } else {
577 		rec->handler(rec->argument, &frame);
578 	    }
579 	}
580     }
581 
582     /*
583      * Cleanup
584      */
585     --gd->gd_intr_nesting_level;
586 #ifdef SMP
587     if (got_mplock)
588 	rel_mplock();
589 #endif
590     crit_exit_gd(gd);
591 
592     /*
593      * If we had a problem, schedule the thread to catch the missed
594      * records (it will just re-run all of them).  A return value of 0
595      * indicates that all handlers have been run and the interrupt can
596      * be re-enabled, and a non-zero return indicates that the interrupt
597      * thread controls re-enablement.
598      */
599     if (must_schedule > 0)
600 	sched_ithd(intr);
601     else if (must_schedule == 0)
602 	++info->i_count;
603     return(must_schedule);
604 }
605 
606 #if 0
607 
608 6: ;                                                                    \
609         /* could not get the MP lock, forward the interrupt */          \
610         movl    mp_lock, %eax ;          /* check race */               \
611         cmpl    $MP_FREE_LOCK,%eax ;                                    \
612         je      2b ;                                                    \
613         incl    PCPU(cnt)+V_FORWARDED_INTS ;                            \
614         subl    $12,%esp ;                                              \
615         movl    $irq_num,8(%esp) ;                                      \
616         movl    $forward_fastint_remote,4(%esp) ;                       \
617         movl    %eax,(%esp) ;                                           \
618         call    lwkt_send_ipiq_bycpu ;                                  \
619         addl    $12,%esp ;                                              \
620         jmp     5f ;
621 
622 #endif
623 
624 
625 /*
626  * Interrupt threads run this as their main loop.
627  *
628  * The handler begins execution outside a critical section and with the BGL
629  * held.
630  *
631  * The i_running state starts at 0.  When an interrupt occurs, the hardware
632  * interrupt is disabled and sched_ithd() The HW interrupt remains disabled
633  * until all routines have run.  We then call ithread_done() to reenable
634  * the HW interrupt and deschedule us until the next interrupt.
635  *
636  * We are responsible for atomically checking i_running and ithread_done()
637  * is responsible for atomically checking for platform-specific delayed
638  * interrupts.  i_running for our irq is only set in the context of our cpu,
639  * so a critical section is a sufficient interlock.
640  */
641 #define LIVELOCK_TIMEFRAME(freq)	((freq) >> 2)	/* 1/4 second */
642 
643 static void
644 ithread_handler(void *arg)
645 {
646     struct intr_info *info;
647     int use_limit;
648     int lticks;
649     int lcount;
650     int intr;
651     struct intrec **list;
652     intrec_t rec, nrec;
653     globaldata_t gd;
654     struct systimer ill_timer;	/* enforced freq. timer */
655     u_int ill_count;		/* interrupt livelock counter */
656 
657     ill_count = 0;
658     lticks = ticks;
659     lcount = 0;
660     intr = (int)arg;
661     info = &intr_info_ary[intr];
662     list = &info->i_reclist;
663     gd = mycpu;
664 
665     /*
666      * The loop must be entered with one critical section held.
667      */
668     crit_enter_gd(gd);
669 
670     for (;;) {
671 	/*
672 	 * If an interrupt is pending, clear i_running and execute the
673 	 * handlers.  Note that certain types of interrupts can re-trigger
674 	 * and set i_running again.
675 	 *
676 	 * Each handler is run in a critical section.  Note that we run both
677 	 * FAST and SLOW designated service routines.
678 	 */
679 	if (info->i_running) {
680 	    ++ill_count;
681 	    info->i_running = 0;
682 	    for (rec = *list; rec; rec = nrec) {
683 		nrec = rec->next;
684 		if (rec->serializer) {
685 		    lwkt_serialize_handler_call(rec->serializer, rec->handler,
686 						rec->argument, NULL);
687 		} else {
688 		    rec->handler(rec->argument, NULL);
689 		}
690 	    }
691 	}
692 
693 	/*
694 	 * This is our interrupt hook to add rate randomness to the random
695 	 * number generator.
696 	 */
697 	if (info->i_random.sc_enabled)
698 	    add_interrupt_randomness(intr);
699 
700 	/*
701 	 * Unmask the interrupt to allow it to trigger again.  This only
702 	 * applies to certain types of interrupts (typ level interrupts).
703 	 * This can result in the interrupt retriggering, but the retrigger
704 	 * will not be processed until we cycle our critical section.
705 	 *
706 	 * Only unmask interrupts while handlers are installed.  It is
707 	 * possible to hit a situation where no handlers are installed
708 	 * due to a device driver livelocking and then tearing down its
709 	 * interrupt on close (the parallel bus being a good example).
710 	 */
711 	if (*list)
712 	    machintr_intren(intr);
713 
714 	/*
715 	 * Do a quick exit/enter to catch any higher-priority interrupt
716 	 * sources, such as the statclock, so thread time accounting
717 	 * will still work.  This may also cause an interrupt to re-trigger.
718 	 */
719 	crit_exit_gd(gd);
720 	crit_enter_gd(gd);
721 
722 	/*
723 	 * LIVELOCK STATE MACHINE
724 	 */
725 	switch(info->i_state) {
726 	case ISTATE_NORMAL:
727 	    /*
728 	     * Calculate a running average every tick.
729 	     */
730 	    if (lticks != ticks) {
731 		lticks = ticks;
732 		ill_count -= ill_count / hz;
733 	    }
734 
735 	    /*
736 	     * If we did not exceed the frequency limit, we are done.
737 	     * If the interrupt has not retriggered we deschedule ourselves.
738 	     */
739 	    if (ill_count <= livelock_limit) {
740 		if (info->i_running == 0) {
741 		    lwkt_deschedule_self(gd->gd_curthread);
742 		    lwkt_switch();
743 		}
744 		break;
745 	    }
746 
747 	    /*
748 	     * Otherwise we are livelocked.  Set up a periodic systimer
749 	     * to wake the thread up at the limit frequency.
750 	     */
751 	    printf("intr %d at %d > %d hz, livelocked limit engaged!\n",
752 		   intr, livelock_limit, ill_count);
753 	    info->i_state = ISTATE_LIVELOCKED;
754 	    if ((use_limit = livelock_limit) < 100)
755 		use_limit = 100;
756 	    else if (use_limit > 500000)
757 		use_limit = 500000;
758 	    systimer_init_periodic(&ill_timer, ithread_livelock_wakeup,
759 				   (void *)intr, use_limit);
760 	    lcount = 0;
761 	    /* fall through */
762 	case ISTATE_LIVELOCKED:
763 	    /*
764 	     * Wait for our periodic timer to go off.  Since the interrupt
765 	     * has re-armed it can still set i_running, but it will not
766 	     * reschedule us while we are in a livelocked state.
767 	     */
768 	    lwkt_deschedule_self(gd->gd_curthread);
769 	    lwkt_switch();
770 
771 	    /*
772 	     * Check to see if the livelock condition no longer applies.
773 	     * The interrupt must be able to operate normally for one
774 	     * full second before we restore normal operation.
775 	     */
776 	    if (lticks != ticks) {
777 		lticks = ticks;
778 		if (ill_count < livelock_lowater) {
779 		    if (++lcount >= hz) {
780 			info->i_state = ISTATE_NORMAL;
781 			systimer_del(&ill_timer);
782 			printf("intr %d at %d < %d hz, livelock removed\n",
783 			       intr, ill_count, livelock_lowater);
784 		    }
785 		} else {
786 		    lcount = 0;
787 		}
788 		ill_count -= ill_count / hz;
789 	    }
790 	    break;
791 	}
792     }
793     /* not reached */
794 }
795 
796 /*
797  * Emergency interrupt polling thread.  The thread begins execution
798  * outside a critical section with the BGL held.
799  *
800  * If emergency interrupt polling is enabled, this thread will
801  * execute all system interrupts not marked INTR_NOPOLL at the
802  * specified polling frequency.
803  *
804  * WARNING!  This thread runs *ALL* interrupt service routines that
805  * are not marked INTR_NOPOLL, which basically means everything except
806  * the 8254 clock interrupt and the ATA interrupt.  It has very high
807  * overhead and should only be used in situations where the machine
808  * cannot otherwise be made to work.  Due to the severe performance
809  * degredation, it should not be enabled on production machines.
810  */
811 static void
812 ithread_emergency(void *arg __unused)
813 {
814     struct intr_info *info;
815     intrec_t rec, nrec;
816     int intr;
817 
818     for (;;) {
819 	for (intr = 0; intr < max_installed_hard_intr; ++intr) {
820 	    info = &intr_info_ary[intr];
821 	    for (rec = info->i_reclist; rec; rec = nrec) {
822 		if ((rec->intr_flags & INTR_NOPOLL) == 0) {
823 		    if (rec->serializer) {
824 			lwkt_serialize_handler_call(rec->serializer,
825 						rec->handler, rec->argument, NULL);
826 		    } else {
827 			rec->handler(rec->argument, NULL);
828 		    }
829 		}
830 		nrec = rec->next;
831 	    }
832 	}
833 	lwkt_deschedule_self(curthread);
834 	lwkt_switch();
835     }
836 }
837 
838 /*
839  * Systimer callback - schedule the emergency interrupt poll thread
840  * 		       if emergency polling is enabled.
841  */
842 static
843 void
844 emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused)
845 {
846     if (emergency_intr_enable)
847 	lwkt_schedule(info->data);
848 }
849 
850 /*
851  * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
852  * The data for this machine dependent, and the declarations are in machine
853  * dependent code.  The layout of intrnames and intrcnt however is machine
854  * independent.
855  *
856  * We do not know the length of intrcnt and intrnames at compile time, so
857  * calculate things at run time.
858  */
859 
860 static int
861 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
862 {
863     struct intr_info *info;
864     intrec_t rec;
865     int error = 0;
866     int len;
867     int intr;
868     char buf[64];
869 
870     for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) {
871 	info = &intr_info_ary[intr];
872 
873 	len = 0;
874 	buf[0] = 0;
875 	for (rec = info->i_reclist; rec; rec = rec->next) {
876 	    snprintf(buf + len, sizeof(buf) - len, "%s%s",
877 		(len ? "/" : ""), rec->name);
878 	    len += strlen(buf + len);
879 	}
880 	if (len == 0) {
881 	    snprintf(buf, sizeof(buf), "irq%d", intr);
882 	    len = strlen(buf);
883 	}
884 	error = SYSCTL_OUT(req, buf, len + 1);
885     }
886     return (error);
887 }
888 
889 
890 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
891 	NULL, 0, sysctl_intrnames, "", "Interrupt Names");
892 
893 static int
894 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
895 {
896     struct intr_info *info;
897     int error = 0;
898     int intr;
899 
900     for (intr = 0; intr < max_installed_hard_intr; ++intr) {
901 	info = &intr_info_ary[intr];
902 
903 	error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
904 	if (error)
905 		goto failed;
906     }
907     for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) {
908 	info = &intr_info_ary[intr];
909 
910 	error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count));
911 	if (error)
912 		goto failed;
913     }
914 failed:
915     return(error);
916 }
917 
918 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
919 	NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
920 
921