kern_intr.c revision 109623
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 * $FreeBSD: head/sys/kern/kern_intr.c 109623 2003-01-21 08:56:16Z alfred $
27 *
28 */
29
30
31#include <sys/param.h>
32#include <sys/bus.h>
33#include <sys/rtprio.h>
34#include <sys/systm.h>
35#include <sys/interrupt.h>
36#include <sys/kernel.h>
37#include <sys/kthread.h>
38#include <sys/ktr.h>
39#include <sys/lock.h>
40#include <sys/malloc.h>
41#include <sys/mutex.h>
42#include <sys/proc.h>
43#include <sys/random.h>
44#include <sys/resourcevar.h>
45#include <sys/sysctl.h>
46#include <sys/unistd.h>
47#include <sys/vmmeter.h>
48#include <machine/atomic.h>
49#include <machine/cpu.h>
50#include <machine/md_var.h>
51#include <machine/stdarg.h>
52
53struct	int_entropy {
54	struct	proc *proc;
55	int	vector;
56};
57
58void	*vm_ih;
59void	*softclock_ih;
60struct	ithd *clk_ithd;
61struct	ithd *tty_ithd;
62
63static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
64
65static void	ithread_update(struct ithd *);
66static void	ithread_loop(void *);
67static void	start_softintr(void *);
68
69u_char
70ithread_priority(enum intr_type flags)
71{
72	u_char pri;
73
74	flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
75	    INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
76	switch (flags) {
77	case INTR_TYPE_TTY:
78		pri = PI_TTYLOW;
79		break;
80	case INTR_TYPE_BIO:
81		/*
82		 * XXX We need to refine this.  BSD/OS distinguishes
83		 * between tape and disk priorities.
84		 */
85		pri = PI_DISK;
86		break;
87	case INTR_TYPE_NET:
88		pri = PI_NET;
89		break;
90	case INTR_TYPE_CAM:
91		pri = PI_DISK;          /* XXX or PI_CAM? */
92		break;
93	case INTR_TYPE_AV:		/* Audio/video */
94		pri = PI_AV;
95		break;
96	case INTR_TYPE_CLK:
97		pri = PI_REALTIME;
98		break;
99	case INTR_TYPE_MISC:
100		pri = PI_DULL;          /* don't care */
101		break;
102	default:
103		/* We didn't specify an interrupt level. */
104		panic("ithread_priority: no interrupt type in flags");
105	}
106
107	return pri;
108}
109
110/*
111 * Regenerate the name (p_comm) and priority for a threaded interrupt thread.
112 */
113static void
114ithread_update(struct ithd *ithd)
115{
116	struct intrhand *ih;
117	struct thread *td;
118	struct proc *p;
119	int entropy;
120
121	mtx_assert(&ithd->it_lock, MA_OWNED);
122	td = ithd->it_td;
123	if (td == NULL)
124		return;
125	p = td->td_proc;
126
127	strlcpy(p->p_comm, ithd->it_name, sizeof(p->p_comm));
128
129	ih = TAILQ_FIRST(&ithd->it_handlers);
130	if (ih == NULL) {
131		mtx_lock_spin(&sched_lock);
132		td->td_priority = PRI_MAX_ITHD;
133		td->td_base_pri = PRI_MAX_ITHD;
134		mtx_unlock_spin(&sched_lock);
135		ithd->it_flags &= ~IT_ENTROPY;
136		return;
137	}
138	entropy = 0;
139	mtx_lock_spin(&sched_lock);
140	td->td_priority = ih->ih_pri;
141	td->td_base_pri = ih->ih_pri;
142	mtx_unlock_spin(&sched_lock);
143	TAILQ_FOREACH(ih, &ithd->it_handlers, ih_next) {
144		if (strlen(p->p_comm) + strlen(ih->ih_name) + 1 <
145		    sizeof(p->p_comm)) {
146			strcat(p->p_comm, " ");
147			strcat(p->p_comm, ih->ih_name);
148		} else if (strlen(p->p_comm) + 1 == sizeof(p->p_comm)) {
149			if (p->p_comm[sizeof(p->p_comm) - 2] == '+')
150				p->p_comm[sizeof(p->p_comm) - 2] = '*';
151			else
152				p->p_comm[sizeof(p->p_comm) - 2] = '+';
153		} else
154			strcat(p->p_comm, "+");
155		if (ih->ih_flags & IH_ENTROPY)
156			entropy++;
157	}
158	if (entropy)
159		ithd->it_flags |= IT_ENTROPY;
160	else
161		ithd->it_flags &= ~IT_ENTROPY;
162	CTR2(KTR_INTR, "%s: updated %s", __func__, p->p_comm);
163}
164
165int
166ithread_create(struct ithd **ithread, int vector, int flags,
167    void (*disable)(int), void (*enable)(int), const char *fmt, ...)
168{
169	struct ithd *ithd;
170	struct thread *td;
171	struct proc *p;
172	int error;
173	va_list ap;
174
175	/* The only valid flag during creation is IT_SOFT. */
176	if ((flags & ~IT_SOFT) != 0)
177		return (EINVAL);
178
179	ithd = malloc(sizeof(struct ithd), M_ITHREAD, M_ZERO);
180	ithd->it_vector = vector;
181	ithd->it_disable = disable;
182	ithd->it_enable = enable;
183	ithd->it_flags = flags;
184	TAILQ_INIT(&ithd->it_handlers);
185	mtx_init(&ithd->it_lock, "ithread", NULL, MTX_DEF);
186
187	va_start(ap, fmt);
188	vsnprintf(ithd->it_name, sizeof(ithd->it_name), fmt, ap);
189	va_end(ap);
190
191	error = kthread_create(ithread_loop, ithd, &p, RFSTOPPED | RFHIGHPID,
192	    0, "%s", ithd->it_name);
193	if (error) {
194		mtx_destroy(&ithd->it_lock);
195		free(ithd, M_ITHREAD);
196		return (error);
197	}
198	td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
199	td->td_ksegrp->kg_pri_class = PRI_ITHD;
200	td->td_priority = PRI_MAX_ITHD;
201	TD_SET_IWAIT(td);
202	ithd->it_td = td;
203	td->td_ithd = ithd;
204	if (ithread != NULL)
205		*ithread = ithd;
206
207	CTR2(KTR_INTR, "%s: created %s", __func__, ithd->it_name);
208	return (0);
209}
210
211int
212ithread_destroy(struct ithd *ithread)
213{
214
215	struct thread *td;
216	struct proc *p;
217	if (ithread == NULL)
218		return (EINVAL);
219
220	td = ithread->it_td;
221	p = td->td_proc;
222	mtx_lock(&ithread->it_lock);
223	if (!TAILQ_EMPTY(&ithread->it_handlers)) {
224		mtx_unlock(&ithread->it_lock);
225		return (EINVAL);
226	}
227	ithread->it_flags |= IT_DEAD;
228	mtx_lock_spin(&sched_lock);
229	if (TD_AWAITING_INTR(td)) {
230		TD_CLR_IWAIT(td);
231		setrunqueue(td);
232	}
233	mtx_unlock_spin(&sched_lock);
234	mtx_unlock(&ithread->it_lock);
235	CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_name);
236	return (0);
237}
238
239int
240ithread_add_handler(struct ithd* ithread, const char *name,
241    driver_intr_t handler, void *arg, u_char pri, enum intr_type flags,
242    void **cookiep)
243{
244	struct intrhand *ih, *temp_ih;
245
246	if (ithread == NULL || name == NULL || handler == NULL)
247		return (EINVAL);
248	if ((flags & INTR_FAST) !=0)
249		flags |= INTR_EXCL;
250
251	ih = malloc(sizeof(struct intrhand), M_ITHREAD, M_ZERO);
252	ih->ih_handler = handler;
253	ih->ih_argument = arg;
254	ih->ih_name = name;
255	ih->ih_ithread = ithread;
256	ih->ih_pri = pri;
257	if (flags & INTR_FAST)
258		ih->ih_flags = IH_FAST | IH_EXCLUSIVE;
259	else if (flags & INTR_EXCL)
260		ih->ih_flags = IH_EXCLUSIVE;
261	if (flags & INTR_MPSAFE)
262		ih->ih_flags |= IH_MPSAFE;
263	if (flags & INTR_ENTROPY)
264		ih->ih_flags |= IH_ENTROPY;
265
266	mtx_lock(&ithread->it_lock);
267	if ((flags & INTR_EXCL) !=0 && !TAILQ_EMPTY(&ithread->it_handlers))
268		goto fail;
269	if (!TAILQ_EMPTY(&ithread->it_handlers) &&
270	    (TAILQ_FIRST(&ithread->it_handlers)->ih_flags & IH_EXCLUSIVE) != 0)
271		goto fail;
272
273	TAILQ_FOREACH(temp_ih, &ithread->it_handlers, ih_next)
274	    if (temp_ih->ih_pri > ih->ih_pri)
275		    break;
276	if (temp_ih == NULL)
277		TAILQ_INSERT_TAIL(&ithread->it_handlers, ih, ih_next);
278	else
279		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
280	ithread_update(ithread);
281	mtx_unlock(&ithread->it_lock);
282
283	if (cookiep != NULL)
284		*cookiep = ih;
285	CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
286	    ithread->it_name);
287	return (0);
288
289fail:
290	mtx_unlock(&ithread->it_lock);
291	free(ih, M_ITHREAD);
292	return (EINVAL);
293}
294
295int
296ithread_remove_handler(void *cookie)
297{
298	struct intrhand *handler = (struct intrhand *)cookie;
299	struct ithd *ithread;
300#ifdef INVARIANTS
301	struct intrhand *ih;
302#endif
303
304	if (handler == NULL)
305		return (EINVAL);
306	ithread = handler->ih_ithread;
307	KASSERT(ithread != NULL,
308	    ("interrupt handler \"%s\" has a NULL interrupt thread",
309		handler->ih_name));
310	CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
311	    ithread->it_name);
312	mtx_lock(&ithread->it_lock);
313#ifdef INVARIANTS
314	TAILQ_FOREACH(ih, &ithread->it_handlers, ih_next)
315		if (ih == handler)
316			goto ok;
317	mtx_unlock(&ithread->it_lock);
318	panic("interrupt handler \"%s\" not found in interrupt thread \"%s\"",
319	    ih->ih_name, ithread->it_name);
320ok:
321#endif
322	/*
323	 * If the interrupt thread is already running, then just mark this
324	 * handler as being dead and let the ithread do the actual removal.
325	 */
326	mtx_lock_spin(&sched_lock);
327	if (!TD_AWAITING_INTR(ithread->it_td)) {
328		handler->ih_flags |= IH_DEAD;
329
330		/*
331		 * Ensure that the thread will process the handler list
332		 * again and remove this handler if it has already passed
333		 * it on the list.
334		 */
335		ithread->it_need = 1;
336	} else
337		TAILQ_REMOVE(&ithread->it_handlers, handler, ih_next);
338	mtx_unlock_spin(&sched_lock);
339	if ((handler->ih_flags & IH_DEAD) != 0)
340		msleep(handler, &ithread->it_lock, PUSER, "itrmh", 0);
341	ithread_update(ithread);
342	mtx_unlock(&ithread->it_lock);
343	free(handler, M_ITHREAD);
344	return (0);
345}
346
347int
348ithread_schedule(struct ithd *ithread, int do_switch)
349{
350	struct int_entropy entropy;
351	struct thread *td;
352	struct thread *ctd;
353	struct proc *p;
354
355	/*
356	 * If no ithread or no handlers, then we have a stray interrupt.
357	 */
358	if ((ithread == NULL) || TAILQ_EMPTY(&ithread->it_handlers))
359		return (EINVAL);
360
361	ctd = curthread;
362	/*
363	 * If any of the handlers for this ithread claim to be good
364	 * sources of entropy, then gather some.
365	 */
366	if (harvest.interrupt && ithread->it_flags & IT_ENTROPY) {
367		entropy.vector = ithread->it_vector;
368		entropy.proc = ctd->td_proc;
369		random_harvest(&entropy, sizeof(entropy), 2, 0,
370		    RANDOM_INTERRUPT);
371	}
372
373	td = ithread->it_td;
374	p = td->td_proc;
375	KASSERT(p != NULL, ("ithread %s has no process", ithread->it_name));
376	CTR4(KTR_INTR, "%s: pid %d: (%s) need = %d",
377	    __func__, p->p_pid, p->p_comm, ithread->it_need);
378
379	/*
380	 * Set it_need to tell the thread to keep running if it is already
381	 * running.  Then, grab sched_lock and see if we actually need to
382	 * put this thread on the runqueue.  If so and the do_switch flag is
383	 * true and it is safe to switch, then switch to the ithread
384	 * immediately.  Otherwise, set the needresched flag to guarantee
385	 * that this ithread will run before any userland processes.
386	 */
387	ithread->it_need = 1;
388	mtx_lock_spin(&sched_lock);
389	if (TD_AWAITING_INTR(td)) {
390		CTR2(KTR_INTR, "%s: setrunqueue %d", __func__, p->p_pid);
391		TD_CLR_IWAIT(td);
392		setrunqueue(td);
393		if (do_switch &&
394		    (ctd->td_critnest == 1) ) {
395			KASSERT((TD_IS_RUNNING(ctd)),
396			    ("ithread_schedule: Bad state for curthread."));
397			ctd->td_proc->p_stats->p_ru.ru_nivcsw++;
398			if (ctd->td_kse->ke_flags & KEF_IDLEKSE)
399				ctd->td_state = TDS_CAN_RUN; /* XXXKSE */
400			mi_switch();
401		} else {
402			curthread->td_kse->ke_flags |= KEF_NEEDRESCHED;
403		}
404	} else {
405		CTR4(KTR_INTR, "%s: pid %d: it_need %d, state %d",
406		    __func__, p->p_pid, ithread->it_need, p->p_state);
407	}
408	mtx_unlock_spin(&sched_lock);
409
410	return (0);
411}
412
413int
414swi_add(struct ithd **ithdp, const char *name, driver_intr_t handler,
415	    void *arg, int pri, enum intr_type flags, void **cookiep)
416{
417	struct ithd *ithd;
418	int error;
419
420	if (flags & (INTR_FAST | INTR_ENTROPY))
421		return (EINVAL);
422
423	ithd = (ithdp != NULL) ? *ithdp : NULL;
424
425	if (ithd != NULL) {
426		if ((ithd->it_flags & IT_SOFT) == 0)
427			return(EINVAL);
428	} else {
429		error = ithread_create(&ithd, pri, IT_SOFT, NULL, NULL,
430		    "swi%d:", pri);
431		if (error)
432			return (error);
433
434		if (ithdp != NULL)
435			*ithdp = ithd;
436	}
437	return (ithread_add_handler(ithd, name, handler, arg,
438		    (pri * RQ_PPQ) + PI_SOFT, flags, cookiep));
439}
440
441
442/*
443 * Schedule a heavyweight software interrupt process.
444 */
445void
446swi_sched(void *cookie, int flags)
447{
448	struct intrhand *ih = (struct intrhand *)cookie;
449	struct ithd *it = ih->ih_ithread;
450	int error;
451
452	atomic_add_int(&cnt.v_intr, 1); /* one more global interrupt */
453
454	CTR3(KTR_INTR, "swi_sched pid %d(%s) need=%d",
455		it->it_td->td_proc->p_pid, it->it_td->td_proc->p_comm, it->it_need);
456
457	/*
458	 * Set ih_need for this handler so that if the ithread is already
459	 * running it will execute this handler on the next pass.  Otherwise,
460	 * it will execute it the next time it runs.
461	 */
462	atomic_store_rel_int(&ih->ih_need, 1);
463	if (!(flags & SWI_DELAY)) {
464		error = ithread_schedule(it, !cold);
465		KASSERT(error == 0, ("stray software interrupt"));
466	}
467}
468
469/*
470 * This is the main code for interrupt threads.
471 */
472static void
473ithread_loop(void *arg)
474{
475	struct ithd *ithd;		/* our thread context */
476	struct intrhand *ih;		/* and our interrupt handler chain */
477	struct thread *td;
478	struct proc *p;
479
480	td = curthread;
481	p = td->td_proc;
482	ithd = (struct ithd *)arg;	/* point to myself */
483	KASSERT(ithd->it_td == td && td->td_ithd == ithd,
484	    ("%s: ithread and proc linkage out of sync", __func__));
485
486	/*
487	 * As long as we have interrupts outstanding, go through the
488	 * list of handlers, giving each one a go at it.
489	 */
490	for (;;) {
491		/*
492		 * If we are an orphaned thread, then just die.
493		 */
494		if (ithd->it_flags & IT_DEAD) {
495			CTR3(KTR_INTR, "%s: pid %d: (%s) exiting", __func__,
496			    p->p_pid, p->p_comm);
497			td->td_ithd = NULL;
498			mtx_destroy(&ithd->it_lock);
499			mtx_lock(&Giant);
500			free(ithd, M_ITHREAD);
501			kthread_exit(0);
502		}
503
504		CTR4(KTR_INTR, "%s: pid %d: (%s) need=%d", __func__,
505		     p->p_pid, p->p_comm, ithd->it_need);
506		while (ithd->it_need) {
507			/*
508			 * Service interrupts.  If another interrupt
509			 * arrives while we are running, they will set
510			 * it_need to denote that we should make
511			 * another pass.
512			 */
513			atomic_store_rel_int(&ithd->it_need, 0);
514restart:
515			TAILQ_FOREACH(ih, &ithd->it_handlers, ih_next) {
516				if (ithd->it_flags & IT_SOFT && !ih->ih_need)
517					continue;
518				atomic_store_rel_int(&ih->ih_need, 0);
519				CTR6(KTR_INTR,
520				    "%s: pid %d ih=%p: %p(%p) flg=%x", __func__,
521				    p->p_pid, (void *)ih,
522				    (void *)ih->ih_handler, ih->ih_argument,
523				    ih->ih_flags);
524
525				if ((ih->ih_flags & IH_DEAD) != 0) {
526					mtx_lock(&ithd->it_lock);
527					TAILQ_REMOVE(&ithd->it_handlers, ih,
528					    ih_next);
529					wakeup(ih);
530					mtx_unlock(&ithd->it_lock);
531					goto restart;
532				}
533				if ((ih->ih_flags & IH_MPSAFE) == 0)
534					mtx_lock(&Giant);
535				ih->ih_handler(ih->ih_argument);
536				if ((ih->ih_flags & IH_MPSAFE) == 0)
537					mtx_unlock(&Giant);
538			}
539		}
540
541		/*
542		 * Processed all our interrupts.  Now get the sched
543		 * lock.  This may take a while and it_need may get
544		 * set again, so we have to check it again.
545		 */
546		mtx_assert(&Giant, MA_NOTOWNED);
547		mtx_lock_spin(&sched_lock);
548		if (!ithd->it_need) {
549			/*
550			 * Should we call this earlier in the loop above?
551			 */
552			if (ithd->it_enable != NULL)
553				ithd->it_enable(ithd->it_vector);
554			TD_SET_IWAIT(td); /* we're idle */
555			p->p_stats->p_ru.ru_nvcsw++;
556			CTR2(KTR_INTR, "%s: pid %d: done", __func__, p->p_pid);
557			mi_switch();
558			CTR2(KTR_INTR, "%s: pid %d: resumed", __func__, p->p_pid);
559		}
560		mtx_unlock_spin(&sched_lock);
561	}
562}
563
564/*
565 * Start standard software interrupt threads
566 */
567static void
568start_softintr(void *dummy)
569{
570
571	if (swi_add(&clk_ithd, "clock", softclock, NULL, SWI_CLOCK,
572		INTR_MPSAFE, &softclock_ih) ||
573	    swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, 0, &vm_ih))
574		panic("died while creating standard software ithreads");
575
576	PROC_LOCK(clk_ithd->it_td->td_proc);
577	clk_ithd->it_td->td_proc->p_flag |= P_NOLOAD;
578	PROC_UNLOCK(clk_ithd->it_td->td_proc);
579}
580SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr, NULL)
581
582/*
583 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
584 * The data for this machine dependent, and the declarations are in machine
585 * dependent code.  The layout of intrnames and intrcnt however is machine
586 * independent.
587 *
588 * We do not know the length of intrcnt and intrnames at compile time, so
589 * calculate things at run time.
590 */
591static int
592sysctl_intrnames(SYSCTL_HANDLER_ARGS)
593{
594	return (sysctl_handle_opaque(oidp, intrnames, eintrnames - intrnames,
595	   req));
596}
597
598SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
599    NULL, 0, sysctl_intrnames, "", "Interrupt Names");
600
601static int
602sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
603{
604	return (sysctl_handle_opaque(oidp, intrcnt,
605	    (char *)eintrcnt - (char *)intrcnt, req));
606}
607
608SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
609    NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
610