xref: /freebsd/sys/kern/kern_intr.c (revision ee41f1b1cf5e3d4f586cb85b46123b416275862c)
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  * $FreeBSD$
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/ipl.h>
36 #include <sys/interrupt.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/ktr.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/resourcevar.h>
44 #include <sys/unistd.h>
45 #include <sys/vmmeter.h>
46 #include <machine/atomic.h>
47 #include <machine/cpu.h>
48 #include <machine/md_var.h>
49 #include <machine/stdarg.h>
50 
51 #include <net/netisr.h>		/* prototype for legacy_setsoftnet */
52 
53 void	*net_ih;
54 void	*vm_ih;
55 void	*softclock_ih;
56 struct	ithd *clk_ithd;
57 struct	ithd *tty_ithd;
58 
59 static struct	mtx ithread_list_lock;
60 
61 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
62 
63 static void	ithread_update(struct ithd *);
64 static void	ithread_loop(void *);
65 static void	ithread_init(void *);
66 static void	start_softintr(void *);
67 static void	swi_net(void *);
68 
69 u_char
70 ithread_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);
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_CLK:
94 		pri = PI_REALTIME;
95 		break;
96 	case INTR_TYPE_MISC:
97 		pri = PI_DULL;          /* don't care */
98 		break;
99 	default:
100 		/* We didn't specify an interrupt level. */
101 		panic("ithread_priority: no interrupt type in flags");
102 	}
103 
104 	return pri;
105 }
106 
107 /*
108  * Regenerate the name (p_comm) and priority for a threaded interrupt thread.
109  */
110 static void
111 ithread_update(struct ithd *ithd)
112 {
113 	struct intrhand *ih;
114 	struct proc *p;
115 	int entropy;
116 
117 	p = ithd->it_proc;
118 	if (p == NULL)
119 		return;
120 
121 	strncpy(p->p_comm, ithd->it_name, sizeof(ithd->it_name));
122 	ih = TAILQ_FIRST(&ithd->it_handlers);
123 	if (ih == NULL) {
124 		p->p_pri.pri_level = PRI_MAX_ITHD;
125 		ithd->it_flags &= ~IT_ENTROPY;
126 		return;
127 	}
128 
129 	entropy = 0;
130 	p->p_pri.pri_level = ih->ih_pri;
131 	TAILQ_FOREACH(ih, &ithd->it_handlers, ih_next) {
132 		if (strlen(p->p_comm) + strlen(ih->ih_name) + 1 <
133 		    sizeof(p->p_comm)) {
134 			strcat(p->p_comm, " ");
135 			strcat(p->p_comm, ih->ih_name);
136 		} else if (strlen(p->p_comm) + 1 == sizeof(p->p_comm)) {
137 			if (p->p_comm[sizeof(p->p_comm) - 2] == '+')
138 				p->p_comm[sizeof(p->p_comm) - 2] = '*';
139 			else
140 				p->p_comm[sizeof(p->p_comm) - 2] = '+';
141 		} else
142 			strcat(p->p_comm, "+");
143 		if (ih->ih_flags & IH_ENTROPY)
144 			entropy++;
145 	}
146 
147 	if (entropy) {
148 		printf("Warning, ithread (%d, %s) is an entropy source.\n",
149 		    p->p_pid, p->p_comm);
150 		ithd->it_flags |= IT_ENTROPY;
151 	}
152 	else
153 		ithd->it_flags &= ~IT_ENTROPY;
154 }
155 
156 int
157 ithread_create(struct ithd **ithread, int vector, int flags,
158     void (*disable)(int), void (*enable)(int), const char *fmt, ...)
159 {
160 	struct ithd *ithd;
161 	struct proc *p;
162 	int error;
163 	va_list ap;
164 
165 	ithd = malloc(sizeof(struct ithd), M_ITHREAD, M_WAITOK | M_ZERO);
166 	ithd->it_vector = vector;
167 	ithd->it_disable = disable;
168 	ithd->it_enable = enable;
169 	ithd->it_flags = flags;
170 	TAILQ_INIT(&ithd->it_handlers);
171 
172 	va_start(ap, fmt);
173 	vsnprintf(ithd->it_name, sizeof(ithd->it_name), fmt, ap);
174 	va_end(ap);
175 
176 	error = kthread_create(ithread_loop, ithd, &p, RFSTOPPED | RFHIGHPID,
177 	    ithd->it_name);
178 	if (error) {
179 		free(ithd, M_ITHREAD);
180 		return (error);
181 	}
182 	p->p_pri.pri_class = PRI_ITHD;
183 	p->p_pri.pri_level = PRI_MAX_ITHD;
184 	p->p_stat = SWAIT;
185 	ithd->it_proc = p;
186 	p->p_ithd = ithd;
187 	if (ithread != NULL)
188 		*ithread = ithd;
189 
190 	return (0);
191 }
192 
193 int
194 ithread_destroy(struct ithd *ithread)
195 {
196 
197 	if (ithread == NULL || !TAILQ_EMPTY(&ithread->it_handlers))
198 		return (EINVAL);
199 
200 	mtx_lock_spin(&sched_lock);
201 	ithread->it_flags |= IT_DEAD;
202 	if (ithread->it_proc->p_stat == SWAIT) {
203 		ithread->it_proc->p_stat = SRUN;
204 		setrunqueue(ithread->it_proc);
205 	}
206 	mtx_unlock_spin(&sched_lock);
207 	return (0);
208 }
209 
210 int
211 ithread_add_handler(struct ithd* ithread, const char *name,
212     driver_intr_t handler, void *arg, u_char pri, enum intr_type flags,
213     void **cookiep)
214 {
215 	struct intrhand *ih, *temp_ih;
216 
217 	if (ithread == NULL || name == NULL || handler == NULL)
218 		return (EINVAL);
219 	if ((flags & INTR_FAST) !=0)
220 		flags |= INTR_EXCL;
221 
222 	ih = malloc(sizeof(struct intrhand), M_ITHREAD, M_WAITOK | M_ZERO);
223 	ih->ih_handler = handler;
224 	ih->ih_argument = arg;
225 	ih->ih_name = name;
226 	ih->ih_ithread = ithread;
227 	ih->ih_pri = pri;
228 	if (flags & INTR_FAST)
229 		ih->ih_flags = IH_FAST | IH_EXCLUSIVE;
230 	else if (flags & INTR_EXCL)
231 		ih->ih_flags = IH_EXCLUSIVE;
232 	if (flags & INTR_MPSAFE)
233 		ih->ih_flags |= IH_MPSAFE;
234 	if (flags & INTR_ENTROPY)
235 		ih->ih_flags |= IH_ENTROPY;
236 
237 	mtx_lock_spin(&ithread_list_lock);
238 	if ((flags & INTR_EXCL) !=0 && !TAILQ_EMPTY(&ithread->it_handlers))
239 		goto fail;
240 	if (!TAILQ_EMPTY(&ithread->it_handlers) &&
241 	    (TAILQ_FIRST(&ithread->it_handlers)->ih_flags & IH_EXCLUSIVE) != 0)
242 		goto fail;
243 
244 	TAILQ_FOREACH(temp_ih, &ithread->it_handlers, ih_next)
245 	    if (temp_ih->ih_pri > ih->ih_pri)
246 		    break;
247 	if (temp_ih == NULL)
248 		TAILQ_INSERT_TAIL(&ithread->it_handlers, ih, ih_next);
249 	else
250 		TAILQ_INSERT_BEFORE(temp_ih, ih, ih_next);
251 	ithread_update(ithread);
252 	mtx_unlock_spin(&ithread_list_lock);
253 
254 	if (cookiep != NULL)
255 		*cookiep = ih;
256 	return (0);
257 
258 fail:
259 	mtx_unlock_spin(&ithread_list_lock);
260 	free(ih, M_ITHREAD);
261 	return (EINVAL);
262 }
263 
264 int
265 ithread_remove_handler(void *cookie)
266 {
267 	struct intrhand *handler = (struct intrhand *)cookie;
268 	struct ithd *ithread;
269 #ifdef INVARIANTS
270 	struct intrhand *ih;
271 	int found;
272 #endif
273 
274 	if (handler == NULL || (ithread = handler->ih_ithread) == NULL)
275 		return (EINVAL);
276 
277 	mtx_lock_spin(&ithread_list_lock);
278 #ifdef INVARIANTS
279 	found = 0;
280 	TAILQ_FOREACH(ih, &ithread->it_handlers, ih_next)
281 		if (ih == handler) {
282 			found++;
283 			break;
284 		}
285 	if (found == 0) {
286 		mtx_unlock_spin(&ithread_list_lock);
287 		return (EINVAL);
288 	}
289 #endif
290 	TAILQ_REMOVE(&ithread->it_handlers, handler, ih_next);
291 	ithread_update(ithread);
292 	mtx_unlock_spin(&ithread_list_lock);
293 
294 	free(handler, M_ITHREAD);
295 	return (0);
296 }
297 
298 int
299 swi_add(struct ithd **ithdp, const char *name, driver_intr_t handler,
300 	    void *arg, int pri, enum intr_type flags, void **cookiep)
301 {
302 	struct proc *p;
303 	struct ithd *ithd;
304 	int error;
305 
306 	ithd = (ithdp != NULL) ? *ithdp : NULL;
307 
308 	if (ithd == NULL) {
309 		error = ithread_create(&ithd, pri, IT_SOFT, NULL, NULL,
310 		    "swi%d:", pri);
311 		if (error)
312 			return (error);
313 
314 		/* XXX - some hacks are _really_ gross */
315 		p = ithd->it_proc;
316 		PROC_LOCK(p);
317 		if (pri == SWI_CLOCK)
318 			p->p_flag |= P_NOLOAD;
319 		PROC_UNLOCK(p);
320 		if (ithdp != NULL)
321 			*ithdp = ithd;
322 	}
323 	return (ithread_add_handler(ithd, name, handler, arg,
324 		    (pri * RQ_PPQ) + PI_SOFT, flags, cookiep));
325 }
326 
327 
328 /*
329  * Schedule a heavyweight software interrupt process.
330  */
331 void
332 swi_sched(void *cookie, int flags)
333 {
334 	struct intrhand *ih = (struct intrhand *)cookie;
335 	struct ithd *it = ih->ih_ithread;
336 	struct proc *p = it->it_proc;
337 
338 	atomic_add_int(&cnt.v_intr, 1); /* one more global interrupt */
339 
340 	CTR3(KTR_INTR, "swi_sched pid %d(%s) need=%d",
341 		p->p_pid, p->p_comm, it->it_need);
342 
343 	/*
344 	 * Set it_need so that if the thread is already running but close
345 	 * to done, it will do another go-round.  Then get the sched lock
346 	 * and see if the thread is on whichkqs yet.  If not, put it on
347 	 * there.  In any case, kick everyone so that if the new thread
348 	 * is higher priority than their current thread, it gets run now.
349 	 */
350 	atomic_store_rel_int(&ih->ih_need, 1);
351 	if (!(flags & SWI_DELAY)) {
352 		it->it_need = 1;
353 		mtx_lock_spin(&sched_lock);
354 		if (p->p_stat == SWAIT) { /* not on run queue */
355 			CTR1(KTR_INTR, "swi_sched: setrunqueue %d", p->p_pid);
356 			p->p_stat = SRUN;
357 			setrunqueue(p);
358 			if (!cold && flags & SWI_SWITCH) {
359 				if (curproc != PCPU_GET(idleproc))
360 					setrunqueue(curproc);
361 				curproc->p_stats->p_ru.ru_nvcsw++;
362 				mi_switch();
363 			} else
364 				need_resched();
365 		}
366 		else {
367 			CTR3(KTR_INTR, "swi_sched %d: it_need %d, state %d",
368 				p->p_pid, it->it_need, p->p_stat );
369 		}
370 		mtx_unlock_spin(&sched_lock);
371 	}
372 }
373 
374 /*
375  * This is the main code for interrupt threads.
376  */
377 void
378 ithread_loop(void *arg)
379 {
380 	struct ithd *ithd;		/* our thread context */
381 	struct intrhand *ih;		/* and our interrupt handler chain */
382 	struct proc *p;
383 
384 	p = curproc;
385 	ithd = (struct ithd *)arg;	/* point to myself */
386 	KASSERT(ithd->it_proc == p && p->p_ithd == ithd,
387 	    (__func__ ": ithread and proc linkage out of sync"));
388 
389 	/*
390 	 * As long as we have interrupts outstanding, go through the
391 	 * list of handlers, giving each one a go at it.
392 	 */
393 	for (;;) {
394 		/*
395 		 * If we are an orphaned thread, then just die.
396 		 */
397 		if (ithd->it_flags & IT_DEAD) {
398 			CTR2(KTR_INTR, __func__ ": pid %d: (%s) exiting",
399 			    p->p_pid, p->p_comm);
400 			p->p_ithd = NULL;
401 			mtx_lock(&Giant);
402 			free(ithd, M_ITHREAD);
403 			kthread_exit(0);
404 		}
405 
406 		CTR3(KTR_INTR, __func__ ": pid %d: (%s) need=%d",
407 		     p->p_pid, p->p_comm, ithd->it_need);
408 		while (ithd->it_need) {
409 			/*
410 			 * Service interrupts.  If another interrupt
411 			 * arrives while we are running, they will set
412 			 * it_need to denote that we should make
413 			 * another pass.
414 			 */
415 			atomic_store_rel_int(&ithd->it_need, 0);
416 			TAILQ_FOREACH(ih, &ithd->it_handlers, ih_next) {
417 				if (ithd->it_flags & IT_SOFT && !ih->ih_need)
418 					continue;
419 				atomic_store_rel_int(&ih->ih_need, 0);
420 				CTR5(KTR_INTR,
421 				    __func__ ": pid %d ih=%p: %p(%p) flg=%x",
422 				    p->p_pid, (void *)ih,
423 				    (void *)ih->ih_handler, ih->ih_argument,
424 				    ih->ih_flags);
425 
426 				if ((ih->ih_flags & IH_MPSAFE) == 0)
427 					mtx_lock(&Giant);
428 				ih->ih_handler(ih->ih_argument);
429 				if ((ih->ih_flags & IH_MPSAFE) == 0)
430 					mtx_unlock(&Giant);
431 			}
432 		}
433 
434 		/*
435 		 * Processed all our interrupts.  Now get the sched
436 		 * lock.  This may take a while and it_need may get
437 		 * set again, so we have to check it again.
438 		 */
439 		mtx_assert(&Giant, MA_NOTOWNED);
440 		mtx_lock_spin(&sched_lock);
441 		if (!ithd->it_need) {
442 			/*
443 			 * Should we call this earlier in the loop above?
444 			 */
445 			if (ithd->it_enable != NULL)
446 				ithd->it_enable(ithd->it_vector);
447 			p->p_stat = SWAIT; /* we're idle */
448 			CTR1(KTR_INTR, __func__ ": pid %d: done", p->p_pid);
449 			mi_switch();
450 			CTR1(KTR_INTR, __func__ ": pid %d: resumed", p->p_pid);
451 		}
452 		mtx_unlock_spin(&sched_lock);
453 	}
454 }
455 
456 /*
457  * Initialize mutex used to protect ithread handler lists.
458  */
459 static void
460 ithread_init(void *dummy)
461 {
462 
463 	mtx_init(&ithread_list_lock, "ithread list lock", MTX_SPIN);
464 }
465 SYSINIT(ithread_init, SI_SUB_INTR, SI_ORDER_FIRST, ithread_init, NULL);
466 
467 /*
468  * Start standard software interrupt threads
469  */
470 static void
471 start_softintr(void *dummy)
472 {
473 
474 	if (swi_add(NULL, "net", swi_net, NULL, SWI_NET, 0, &net_ih) ||
475 	    swi_add(&clk_ithd, "clock", softclock, NULL, SWI_CLOCK,
476 		INTR_MPSAFE, &softclock_ih) ||
477 	    swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, 0, &vm_ih))
478 		panic("died while creating standard software ithreads");
479 }
480 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr, NULL)
481 
482 void
483 legacy_setsoftnet(void)
484 {
485 	swi_sched(net_ih, SWI_NOSWITCH);
486 }
487 
488 /*
489  * XXX: This should really be in the network code somewhere and installed
490  * via a SI_SUB_SOFINTR, SI_ORDER_MIDDLE sysinit.
491  */
492 void	(*netisrs[32]) __P((void));
493 u_int	netisr;
494 
495 int
496 register_netisr(num, handler)
497 	int num;
498 	netisr_t *handler;
499 {
500 
501 	if (num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs)) ) {
502 		printf("register_netisr: bad isr number: %d\n", num);
503 		return (EINVAL);
504 	}
505 	netisrs[num] = handler;
506 	return (0);
507 }
508 
509 int
510 unregister_netisr(num)
511 	int num;
512 {
513 
514 	if (num < 0 || num >= (sizeof(netisrs)/sizeof(*netisrs)) ) {
515 		printf("unregister_netisr: bad isr number: %d\n", num);
516 		return (EINVAL);
517 	}
518 	netisrs[num] = NULL;
519 	return (0);
520 }
521 
522 static void
523 swi_net(void *dummy)
524 {
525 	u_int bits;
526 	int i;
527 
528 	bits = atomic_readandclear_int(&netisr);
529 	while ((i = ffs(bits)) != 0) {
530 		i--;
531 		if (netisrs[i] != NULL)
532 			netisrs[i]();
533 		else
534 			printf("swi_net: unregistered isr number: %d.\n", i);
535 		bits &= ~(1 << i);
536 	}
537 }
538