2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
33 #include "opt_kstack_usage_prof.h"
35 #include <sys/param.h>
38 #include <sys/cpuset.h>
39 #include <sys/rtprio.h>
40 #include <sys/systm.h>
41 #include <sys/interrupt.h>
42 #include <sys/kernel.h>
43 #include <sys/kthread.h>
45 #include <sys/limits.h>
47 #include <sys/malloc.h>
48 #include <sys/mutex.h>
51 #include <sys/random.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
55 #include <sys/sysctl.h>
56 #include <sys/syslog.h>
57 #include <sys/unistd.h>
58 #include <sys/vmmeter.h>
59 #include <machine/atomic.h>
60 #include <machine/cpu.h>
61 #include <machine/md_var.h>
62 #include <machine/stdarg.h>
65 #include <ddb/db_sym.h>
69 * Describe an interrupt thread. There is one of these per interrupt event.
72 struct intr_event *it_event;
73 struct thread *it_thread; /* Kernel thread. */
74 int it_flags; /* (j) IT_* flags. */
75 int it_need; /* Needs service. */
78 /* Interrupt thread flags kept in it_flags */
79 #define IT_DEAD 0x000001 /* Thread is waiting to exit. */
80 #define IT_WAIT 0x000002 /* Thread is waiting for completion. */
87 struct intr_event *clk_intr_event;
88 struct intr_event *tty_intr_event;
90 struct proc *intrproc;
92 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
94 static int intr_storm_threshold = 1000;
95 SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RWTUN,
96 &intr_storm_threshold, 0,
97 "Number of consecutive interrupts before storm protection is enabled");
98 static TAILQ_HEAD(, intr_event) event_list =
99 TAILQ_HEAD_INITIALIZER(event_list);
100 static struct mtx event_lock;
101 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
103 static void intr_event_update(struct intr_event *ie);
104 static int intr_event_schedule_thread(struct intr_event *ie);
105 static struct intr_thread *ithread_create(const char *name);
106 static void ithread_destroy(struct intr_thread *ithread);
107 static void ithread_execute_handlers(struct proc *p,
108 struct intr_event *ie);
109 static void ithread_loop(void *);
110 static void ithread_update(struct intr_thread *ithd);
111 static void start_softintr(void *);
113 /* Map an interrupt type to an ithread priority. */
115 intr_priority(enum intr_type flags)
119 flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
120 INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
141 pri = PI_DULL; /* don't care */
144 /* We didn't specify an interrupt level. */
145 panic("intr_priority: no interrupt type in flags");
152 * Update an ithread based on the associated intr_event.
155 ithread_update(struct intr_thread *ithd)
157 struct intr_event *ie;
162 td = ithd->it_thread;
163 mtx_assert(&ie->ie_lock, MA_OWNED);
165 /* Determine the overall priority of this event. */
166 if (CK_SLIST_EMPTY(&ie->ie_handlers))
169 pri = CK_SLIST_FIRST(&ie->ie_handlers)->ih_pri;
171 /* Update name and priority. */
172 strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
174 sched_clear_tdname(td);
182 * Regenerate the full name of an interrupt event and update its priority.
185 intr_event_update(struct intr_event *ie)
187 struct intr_handler *ih;
191 /* Start off with no entropy and just the name of the event. */
192 mtx_assert(&ie->ie_lock, MA_OWNED);
193 strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
194 ie->ie_flags &= ~IE_ENTROPY;
198 /* Run through all the handlers updating values. */
199 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
200 if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
201 sizeof(ie->ie_fullname)) {
202 strcat(ie->ie_fullname, " ");
203 strcat(ie->ie_fullname, ih->ih_name);
207 if (ih->ih_flags & IH_ENTROPY)
208 ie->ie_flags |= IE_ENTROPY;
212 * If there is only one handler and its name is too long, just copy in
213 * as much of the end of the name (includes the unit number) as will
214 * fit. Otherwise, we have multiple handlers and not all of the names
215 * will fit. Add +'s to indicate missing names. If we run out of room
216 * and still have +'s to add, change the last character from a + to a *.
218 if (missed == 1 && space == 1) {
219 ih = CK_SLIST_FIRST(&ie->ie_handlers);
220 missed = strlen(ie->ie_fullname) + strlen(ih->ih_name) + 2 -
221 sizeof(ie->ie_fullname);
222 strcat(ie->ie_fullname, (missed == 0) ? " " : "-");
223 strcat(ie->ie_fullname, &ih->ih_name[missed]);
226 last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
227 while (missed-- > 0) {
228 if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
235 strcat(ie->ie_fullname, " +");
238 strcat(ie->ie_fullname, "+");
242 * If this event has an ithread, update it's priority and
245 if (ie->ie_thread != NULL)
246 ithread_update(ie->ie_thread);
247 CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
251 intr_event_create(struct intr_event **event, void *source, int flags, int irq,
252 void (*pre_ithread)(void *), void (*post_ithread)(void *),
253 void (*post_filter)(void *), int (*assign_cpu)(void *, int),
254 const char *fmt, ...)
256 struct intr_event *ie;
259 /* The only valid flag during creation is IE_SOFT. */
260 if ((flags & ~IE_SOFT) != 0)
262 ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
263 ie->ie_source = source;
264 ie->ie_pre_ithread = pre_ithread;
265 ie->ie_post_ithread = post_ithread;
266 ie->ie_post_filter = post_filter;
267 ie->ie_assign_cpu = assign_cpu;
268 ie->ie_flags = flags;
271 CK_SLIST_INIT(&ie->ie_handlers);
272 mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
275 vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
277 strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
278 mtx_lock(&event_lock);
279 TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
280 mtx_unlock(&event_lock);
283 CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
288 * Bind an interrupt event to the specified CPU. Note that not all
289 * platforms support binding an interrupt to a CPU. For those
290 * platforms this request will fail. Using a cpu id of NOCPU unbinds
291 * the interrupt event.
294 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
299 /* Need a CPU to bind to. */
300 if (cpu != NOCPU && CPU_ABSENT(cpu))
303 if (ie->ie_assign_cpu == NULL)
306 error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
311 * If we have any ithreads try to set their mask first to verify
315 mtx_lock(&ie->ie_lock);
316 if (ie->ie_thread != NULL) {
317 id = ie->ie_thread->it_thread->td_tid;
318 mtx_unlock(&ie->ie_lock);
319 error = cpuset_setithread(id, cpu);
323 mtx_unlock(&ie->ie_lock);
326 error = ie->ie_assign_cpu(ie->ie_source, cpu);
329 mtx_lock(&ie->ie_lock);
330 if (ie->ie_thread != NULL) {
332 id = ie->ie_thread->it_thread->td_tid;
333 mtx_unlock(&ie->ie_lock);
334 (void)cpuset_setithread(id, cpu);
336 mtx_unlock(&ie->ie_lock);
342 mtx_lock(&ie->ie_lock);
344 mtx_unlock(&ie->ie_lock);
351 * Bind an interrupt event to the specified CPU. For supported platforms, any
352 * associated ithreads as well as the primary interrupt context will be bound
353 * to the specificed CPU.
356 intr_event_bind(struct intr_event *ie, int cpu)
359 return (_intr_event_bind(ie, cpu, true, true));
363 * Bind an interrupt event to the specified CPU, but do not bind associated
367 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
370 return (_intr_event_bind(ie, cpu, true, false));
374 * Bind an interrupt event's ithread to the specified CPU.
377 intr_event_bind_ithread(struct intr_event *ie, int cpu)
380 return (_intr_event_bind(ie, cpu, false, true));
383 static struct intr_event *
386 struct intr_event *ie;
388 mtx_lock(&event_lock);
389 TAILQ_FOREACH(ie, &event_list, ie_list)
390 if (ie->ie_irq == irq &&
391 (ie->ie_flags & IE_SOFT) == 0 &&
392 CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
394 mtx_unlock(&event_lock);
399 intr_setaffinity(int irq, int mode, void *m)
401 struct intr_event *ie;
408 * If we're setting all cpus we can unbind. Otherwise make sure
409 * only one cpu is in the set.
411 if (CPU_CMP(cpuset_root, mask)) {
412 for (n = 0; n < CPU_SETSIZE; n++) {
413 if (!CPU_ISSET(n, mask))
420 ie = intr_lookup(irq);
425 return (intr_event_bind(ie, cpu));
426 case CPU_WHICH_INTRHANDLER:
427 return (intr_event_bind_irqonly(ie, cpu));
428 case CPU_WHICH_ITHREAD:
429 return (intr_event_bind_ithread(ie, cpu));
436 intr_getaffinity(int irq, int mode, void *m)
438 struct intr_event *ie;
446 ie = intr_lookup(irq);
454 case CPU_WHICH_INTRHANDLER:
455 mtx_lock(&ie->ie_lock);
456 if (ie->ie_cpu == NOCPU)
457 CPU_COPY(cpuset_root, mask);
459 CPU_SET(ie->ie_cpu, mask);
460 mtx_unlock(&ie->ie_lock);
462 case CPU_WHICH_ITHREAD:
463 mtx_lock(&ie->ie_lock);
464 if (ie->ie_thread == NULL) {
465 mtx_unlock(&ie->ie_lock);
466 CPU_COPY(cpuset_root, mask);
468 id = ie->ie_thread->it_thread->td_tid;
469 mtx_unlock(&ie->ie_lock);
470 error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
473 CPU_COPY(&td->td_cpuset->cs_mask, mask);
483 intr_event_destroy(struct intr_event *ie)
486 mtx_lock(&event_lock);
487 mtx_lock(&ie->ie_lock);
488 if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
489 mtx_unlock(&ie->ie_lock);
490 mtx_unlock(&event_lock);
493 TAILQ_REMOVE(&event_list, ie, ie_list);
495 if (ie->ie_thread != NULL) {
496 ithread_destroy(ie->ie_thread);
497 ie->ie_thread = NULL;
500 mtx_unlock(&ie->ie_lock);
501 mtx_unlock(&event_lock);
502 mtx_destroy(&ie->ie_lock);
507 static struct intr_thread *
508 ithread_create(const char *name)
510 struct intr_thread *ithd;
514 ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
516 error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
517 &td, RFSTOPPED | RFHIGHPID,
518 0, "intr", "%s", name);
520 panic("kproc_create() failed with %d", error);
522 sched_class(td, PRI_ITHD);
525 td->td_pflags |= TDP_ITHREAD;
526 ithd->it_thread = td;
527 CTR2(KTR_INTR, "%s: created %s", __func__, name);
532 ithread_destroy(struct intr_thread *ithread)
536 CTR2(KTR_INTR, "%s: killing %s", __func__, ithread->it_event->ie_name);
537 td = ithread->it_thread;
539 ithread->it_flags |= IT_DEAD;
540 if (TD_AWAITING_INTR(td)) {
542 sched_add(td, SRQ_INTR);
548 intr_event_add_handler(struct intr_event *ie, const char *name,
549 driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
550 enum intr_type flags, void **cookiep)
552 struct intr_handler *ih, *temp_ih;
553 struct intr_handler **prevptr;
554 struct intr_thread *it;
556 if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
559 /* Allocate and populate an interrupt handler structure. */
560 ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
561 ih->ih_filter = filter;
562 ih->ih_handler = handler;
563 ih->ih_argument = arg;
564 strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
567 if (flags & INTR_EXCL)
568 ih->ih_flags = IH_EXCLUSIVE;
569 if (flags & INTR_MPSAFE)
570 ih->ih_flags |= IH_MPSAFE;
571 if (flags & INTR_ENTROPY)
572 ih->ih_flags |= IH_ENTROPY;
574 /* We can only have one exclusive handler in a event. */
575 mtx_lock(&ie->ie_lock);
576 if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
577 if ((flags & INTR_EXCL) ||
578 (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
579 mtx_unlock(&ie->ie_lock);
585 /* Create a thread if we need one. */
586 while (ie->ie_thread == NULL && handler != NULL) {
587 if (ie->ie_flags & IE_ADDING_THREAD)
588 msleep(ie, &ie->ie_lock, 0, "ithread", 0);
590 ie->ie_flags |= IE_ADDING_THREAD;
591 mtx_unlock(&ie->ie_lock);
592 it = ithread_create("intr: newborn");
593 mtx_lock(&ie->ie_lock);
594 ie->ie_flags &= ~IE_ADDING_THREAD;
602 /* Add the new handler to the event in priority order. */
603 CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
604 if (temp_ih->ih_pri > ih->ih_pri)
607 CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
609 intr_event_update(ie);
611 CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
613 mtx_unlock(&ie->ie_lock);
621 * Append a description preceded by a ':' to the name of the specified
625 intr_event_describe_handler(struct intr_event *ie, void *cookie,
628 struct intr_handler *ih;
632 mtx_lock(&ie->ie_lock);
634 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
639 mtx_unlock(&ie->ie_lock);
640 panic("handler %p not found in interrupt event %p", cookie, ie);
646 * Look for an existing description by checking for an
647 * existing ":". This assumes device names do not include
648 * colons. If one is found, prepare to insert the new
649 * description at that point. If one is not found, find the
650 * end of the name to use as the insertion point.
652 start = strchr(ih->ih_name, ':');
654 start = strchr(ih->ih_name, 0);
657 * See if there is enough remaining room in the string for the
658 * description + ":". The "- 1" leaves room for the trailing
659 * '\0'. The "+ 1" accounts for the colon.
661 space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
662 if (strlen(descr) + 1 > space) {
663 mtx_unlock(&ie->ie_lock);
667 /* Append a colon followed by the description. */
669 strcpy(start + 1, descr);
670 intr_event_update(ie);
671 mtx_unlock(&ie->ie_lock);
676 * Return the ie_source field from the intr_event an intr_handler is
680 intr_handler_source(void *cookie)
682 struct intr_handler *ih;
683 struct intr_event *ie;
685 ih = (struct intr_handler *)cookie;
690 ("interrupt handler \"%s\" has a NULL interrupt event",
692 return (ie->ie_source);
696 * If intr_event_handle() is running in the ISR context at the time of the call,
697 * then wait for it to complete.
700 intr_event_barrier(struct intr_event *ie)
704 mtx_assert(&ie->ie_lock, MA_OWNED);
705 phase = ie->ie_phase;
708 * Switch phase to direct future interrupts to the other active counter.
709 * Make sure that any preceding stores are visible before the switch.
711 KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
712 atomic_store_rel_int(&ie->ie_phase, !phase);
715 * This code cooperates with wait-free iteration of ie_handlers
716 * in intr_event_handle.
717 * Make sure that the removal and the phase update are not reordered
718 * with the active count check.
719 * Note that no combination of acquire and release fences can provide
720 * that guarantee as Store->Load sequences can always be reordered.
722 atomic_thread_fence_seq_cst();
725 * Now wait on the inactive phase.
726 * The acquire fence is needed so that that all post-barrier accesses
727 * are after the check.
729 while (ie->ie_active[phase] > 0)
731 atomic_thread_fence_acq();
735 intr_handler_barrier(struct intr_handler *handler)
737 struct intr_event *ie;
739 ie = handler->ih_event;
740 mtx_assert(&ie->ie_lock, MA_OWNED);
741 KASSERT((handler->ih_flags & IH_DEAD) == 0,
742 ("update for a removed handler"));
744 if (ie->ie_thread == NULL) {
745 intr_event_barrier(ie);
748 if ((handler->ih_flags & IH_CHANGED) == 0) {
749 handler->ih_flags |= IH_CHANGED;
750 intr_event_schedule_thread(ie);
752 while ((handler->ih_flags & IH_CHANGED) != 0)
753 msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
757 * Sleep until an ithread finishes executing an interrupt handler.
759 * XXX Doesn't currently handle interrupt filters or fast interrupt
760 * handlers. This is intended for compatibility with linux drivers
761 * only. Do not use in BSD code.
766 struct intr_event *ie;
767 struct intr_thread *ithd;
770 ie = intr_lookup(irq);
773 if (ie->ie_thread == NULL)
775 ithd = ie->ie_thread;
776 td = ithd->it_thread;
778 * We set the flag and wait for it to be cleared to avoid
779 * long delays with potentially busy interrupt handlers
780 * were we to only sample TD_AWAITING_INTR() every tick.
783 if (!TD_AWAITING_INTR(td)) {
784 ithd->it_flags |= IT_WAIT;
785 while (ithd->it_flags & IT_WAIT) {
796 intr_event_remove_handler(void *cookie)
798 struct intr_handler *handler = (struct intr_handler *)cookie;
799 struct intr_event *ie;
800 struct intr_handler *ih;
801 struct intr_handler **prevptr;
808 ie = handler->ih_event;
810 ("interrupt handler \"%s\" has a NULL interrupt event",
813 mtx_lock(&ie->ie_lock);
814 CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
816 CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
821 panic("interrupt handler \"%s\" not found in "
822 "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
826 * If there is no ithread, then directly remove the handler. Note that
827 * intr_event_handle() iterates ie_handlers in a lock-less fashion, so
828 * care needs to be taken to keep ie_handlers consistent and to free
829 * the removed handler only when ie_handlers is quiescent.
831 if (ie->ie_thread == NULL) {
832 CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
833 intr_event_barrier(ie);
834 intr_event_update(ie);
835 mtx_unlock(&ie->ie_lock);
836 free(handler, M_ITHREAD);
841 * Let the interrupt thread do the job.
842 * The interrupt source is disabled when the interrupt thread is
843 * running, so it does not have to worry about interaction with
844 * intr_event_handle().
846 KASSERT((handler->ih_flags & IH_DEAD) == 0,
847 ("duplicate handle remove"));
848 handler->ih_flags |= IH_DEAD;
849 intr_event_schedule_thread(ie);
850 while (handler->ih_flags & IH_DEAD)
851 msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
852 intr_event_update(ie);
856 * XXX: This could be bad in the case of ppbus(8). Also, I think
857 * this could lead to races of stale data when servicing an
861 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
862 if (ih->ih_handler != NULL) {
868 ithread_destroy(ie->ie_thread);
869 ie->ie_thread = NULL;
872 mtx_unlock(&ie->ie_lock);
873 free(handler, M_ITHREAD);
878 intr_event_suspend_handler(void *cookie)
880 struct intr_handler *handler = (struct intr_handler *)cookie;
881 struct intr_event *ie;
885 ie = handler->ih_event;
887 ("interrupt handler \"%s\" has a NULL interrupt event",
889 mtx_lock(&ie->ie_lock);
890 handler->ih_flags |= IH_SUSP;
891 intr_handler_barrier(handler);
892 mtx_unlock(&ie->ie_lock);
897 intr_event_resume_handler(void *cookie)
899 struct intr_handler *handler = (struct intr_handler *)cookie;
900 struct intr_event *ie;
904 ie = handler->ih_event;
906 ("interrupt handler \"%s\" has a NULL interrupt event",
910 * intr_handler_barrier() acts not only as a barrier,
911 * it also allows to check for any pending interrupts.
913 mtx_lock(&ie->ie_lock);
914 handler->ih_flags &= ~IH_SUSP;
915 intr_handler_barrier(handler);
916 mtx_unlock(&ie->ie_lock);
921 intr_event_schedule_thread(struct intr_event *ie)
923 struct intr_entropy entropy;
924 struct intr_thread *it;
929 * If no ithread or no handlers, then we have a stray interrupt.
931 if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
932 ie->ie_thread == NULL)
940 * If any of the handlers for this ithread claim to be good
941 * sources of entropy, then gather some.
943 if (ie->ie_flags & IE_ENTROPY) {
944 entropy.event = (uintptr_t)ie;
946 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
949 KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
952 * Set it_need to tell the thread to keep running if it is already
953 * running. Then, lock the thread and see if we actually need to
954 * put it on the runqueue.
956 * Use store_rel to arrange that the store to ih_need in
957 * swi_sched() is before the store to it_need and prepare for
958 * transfer of this order to loads in the ithread.
960 atomic_store_rel_int(&it->it_need, 1);
962 if (TD_AWAITING_INTR(td)) {
963 CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
966 sched_add(td, SRQ_INTR);
968 CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
969 __func__, td->td_proc->p_pid, td->td_name, it->it_need, td->td_state);
977 * Allow interrupt event binding for software interrupt handlers -- a no-op,
978 * since interrupts are generated in software rather than being directed by
982 swi_assign_cpu(void *arg, int cpu)
989 * Add a software interrupt handler to a specified event. If a given event
990 * is not specified, then a new event is created.
993 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
994 void *arg, int pri, enum intr_type flags, void **cookiep)
996 struct intr_event *ie;
999 if (flags & INTR_ENTROPY)
1002 ie = (eventp != NULL) ? *eventp : NULL;
1005 if (!(ie->ie_flags & IE_SOFT))
1008 error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1009 NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1015 error = intr_event_add_handler(ie, name, NULL, handler, arg,
1016 PI_SWI(pri), flags, cookiep);
1021 * Schedule a software interrupt thread.
1024 swi_sched(void *cookie, int flags)
1026 struct intr_handler *ih = (struct intr_handler *)cookie;
1027 struct intr_event *ie = ih->ih_event;
1028 struct intr_entropy entropy;
1031 CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1034 entropy.event = (uintptr_t)ih;
1035 entropy.td = curthread;
1036 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1039 * Set ih_need for this handler so that if the ithread is already
1040 * running it will execute this handler on the next pass. Otherwise,
1041 * it will execute it the next time it runs.
1045 if (!(flags & SWI_DELAY)) {
1047 error = intr_event_schedule_thread(ie);
1048 KASSERT(error == 0, ("stray software interrupt"));
1053 * Remove a software interrupt handler. Currently this code does not
1054 * remove the associated interrupt event if it becomes empty. Calling code
1055 * may do so manually via intr_event_destroy(), but that's not really
1056 * an optimal interface.
1059 swi_remove(void *cookie)
1062 return (intr_event_remove_handler(cookie));
1066 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1068 struct intr_handler *ih, *ihn, *ihp;
1071 CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1073 * If this handler is marked for death, remove it from
1074 * the list of handlers and wake up the sleeper.
1076 if (ih->ih_flags & IH_DEAD) {
1077 mtx_lock(&ie->ie_lock);
1079 CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1081 CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1082 ih->ih_flags &= ~IH_DEAD;
1084 mtx_unlock(&ie->ie_lock);
1089 * Now that we know that the current element won't be removed
1090 * update the previous element.
1094 if ((ih->ih_flags & IH_CHANGED) != 0) {
1095 mtx_lock(&ie->ie_lock);
1096 ih->ih_flags &= ~IH_CHANGED;
1098 mtx_unlock(&ie->ie_lock);
1101 /* Skip filter only handlers */
1102 if (ih->ih_handler == NULL)
1105 /* Skip suspended handlers */
1106 if ((ih->ih_flags & IH_SUSP) != 0)
1110 * For software interrupt threads, we only execute
1111 * handlers that have their need flag set. Hardware
1112 * interrupt threads always invoke all of their handlers.
1114 * ih_need can only be 0 or 1. Failed cmpset below
1115 * means that there is no request to execute handlers,
1116 * so a retry of the cmpset is not needed.
1118 if ((ie->ie_flags & IE_SOFT) != 0 &&
1119 atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1122 /* Execute this handler. */
1123 CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1124 __func__, p->p_pid, (void *)ih->ih_handler,
1125 ih->ih_argument, ih->ih_name, ih->ih_flags);
1127 if (!(ih->ih_flags & IH_MPSAFE))
1129 ih->ih_handler(ih->ih_argument);
1130 if (!(ih->ih_flags & IH_MPSAFE))
1136 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1139 /* Interrupt handlers should not sleep. */
1140 if (!(ie->ie_flags & IE_SOFT))
1141 THREAD_NO_SLEEPING();
1142 intr_event_execute_handlers(p, ie);
1143 if (!(ie->ie_flags & IE_SOFT))
1144 THREAD_SLEEPING_OK();
1147 * Interrupt storm handling:
1149 * If this interrupt source is currently storming, then throttle
1150 * it to only fire the handler once per clock tick.
1152 * If this interrupt source is not currently storming, but the
1153 * number of back to back interrupts exceeds the storm threshold,
1154 * then enter storming mode.
1156 if (intr_storm_threshold != 0 && ie->ie_count >= intr_storm_threshold &&
1157 !(ie->ie_flags & IE_SOFT)) {
1158 /* Report the message only once every second. */
1159 if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1161 "interrupt storm detected on \"%s\"; throttling interrupt source\n",
1169 * Now that all the handlers have had a chance to run, reenable
1170 * the interrupt source.
1172 if (ie->ie_post_ithread != NULL)
1173 ie->ie_post_ithread(ie->ie_source);
1177 * This is the main code for interrupt threads.
1180 ithread_loop(void *arg)
1182 struct intr_thread *ithd;
1183 struct intr_event *ie;
1190 ithd = (struct intr_thread *)arg;
1191 KASSERT(ithd->it_thread == td,
1192 ("%s: ithread and proc linkage out of sync", __func__));
1193 ie = ithd->it_event;
1198 * As long as we have interrupts outstanding, go through the
1199 * list of handlers, giving each one a go at it.
1203 * If we are an orphaned thread, then just die.
1205 if (ithd->it_flags & IT_DEAD) {
1206 CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1207 p->p_pid, td->td_name);
1208 free(ithd, M_ITHREAD);
1213 * Service interrupts. If another interrupt arrives while
1214 * we are running, it will set it_need to note that we
1215 * should make another pass.
1217 * The load_acq part of the following cmpset ensures
1218 * that the load of ih_need in ithread_execute_handlers()
1219 * is ordered after the load of it_need here.
1221 while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0)
1222 ithread_execute_handlers(p, ie);
1223 WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1224 mtx_assert(&Giant, MA_NOTOWNED);
1227 * Processed all our interrupts. Now get the sched
1228 * lock. This may take a while and it_need may get
1229 * set again, so we have to check it again.
1232 if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1233 (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1236 mi_switch(SW_VOL | SWT_IWAIT, NULL);
1238 if (ithd->it_flags & IT_WAIT) {
1240 ithd->it_flags &= ~IT_WAIT;
1251 * Main interrupt handling body.
1254 * o ie: the event connected to this interrupt.
1255 * o frame: some archs (i.e. i386) pass a frame to some.
1256 * handlers as their main argument.
1258 * o 0: everything ok.
1259 * o EINVAL: stray interrupt.
1262 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1264 struct intr_handler *ih;
1265 struct trapframe *oldframe;
1269 bool filter, thread;
1273 #ifdef KSTACK_USAGE_PROF
1274 intr_prof_stack_use(td, frame);
1277 /* An interrupt with no event or handlers is a stray interrupt. */
1278 if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
1282 * Execute fast interrupt handlers directly.
1283 * To support clock handlers, if a handler registers
1284 * with a NULL argument, then we pass it a pointer to
1285 * a trapframe as its argument.
1287 td->td_intr_nesting_level++;
1292 oldframe = td->td_intr_frame;
1293 td->td_intr_frame = frame;
1295 phase = ie->ie_phase;
1296 atomic_add_int(&ie->ie_active[phase], 1);
1299 * This fence is required to ensure that no later loads are
1300 * re-ordered before the ie_active store.
1302 atomic_thread_fence_seq_cst();
1304 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
1305 if ((ih->ih_flags & IH_SUSP) != 0)
1307 if (ih->ih_filter == NULL) {
1311 CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1312 ih->ih_filter, ih->ih_argument == NULL ? frame :
1313 ih->ih_argument, ih->ih_name);
1314 if (ih->ih_argument == NULL)
1315 ret = ih->ih_filter(frame);
1317 ret = ih->ih_filter(ih->ih_argument);
1318 KASSERT(ret == FILTER_STRAY ||
1319 ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1320 (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1321 ("%s: incorrect return value %#x from %s", __func__, ret,
1323 filter = filter || ret == FILTER_HANDLED;
1326 * Wrapper handler special handling:
1328 * in some particular cases (like pccard and pccbb),
1329 * the _real_ device handler is wrapped in a couple of
1330 * functions - a filter wrapper and an ithread wrapper.
1331 * In this case (and just in this case), the filter wrapper
1332 * could ask the system to schedule the ithread and mask
1333 * the interrupt source if the wrapped handler is composed
1334 * of just an ithread handler.
1336 * TODO: write a generic wrapper to avoid people rolling
1340 if (ret == FILTER_SCHEDULE_THREAD)
1344 atomic_add_rel_int(&ie->ie_active[phase], -1);
1346 td->td_intr_frame = oldframe;
1349 if (ie->ie_pre_ithread != NULL)
1350 ie->ie_pre_ithread(ie->ie_source);
1352 if (ie->ie_post_filter != NULL)
1353 ie->ie_post_filter(ie->ie_source);
1356 /* Schedule the ithread if needed. */
1360 error = intr_event_schedule_thread(ie);
1361 KASSERT(error == 0, ("bad stray interrupt"));
1364 td->td_intr_nesting_level--;
1366 /* The interrupt is not aknowledged by any filter and has no ithread. */
1367 if (!thread && !filter)
1375 * Dump details about an interrupt handler
1378 db_dump_intrhand(struct intr_handler *ih)
1382 db_printf("\t%-10s ", ih->ih_name);
1383 switch (ih->ih_pri) {
1403 if (ih->ih_pri >= PI_SOFT)
1406 db_printf("%4u", ih->ih_pri);
1410 if (ih->ih_filter != NULL) {
1412 db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
1414 if (ih->ih_handler != NULL) {
1415 if (ih->ih_filter != NULL)
1418 db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1420 db_printf("(%p)", ih->ih_argument);
1422 (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1426 if (ih->ih_flags & IH_EXCLUSIVE) {
1432 if (ih->ih_flags & IH_ENTROPY) {
1435 db_printf("ENTROPY");
1438 if (ih->ih_flags & IH_DEAD) {
1444 if (ih->ih_flags & IH_MPSAFE) {
1447 db_printf("MPSAFE");
1461 * Dump details about a event.
1464 db_dump_intr_event(struct intr_event *ie, int handlers)
1466 struct intr_handler *ih;
1467 struct intr_thread *it;
1470 db_printf("%s ", ie->ie_fullname);
1473 db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1475 db_printf("(no thread)");
1476 if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
1477 (it != NULL && it->it_need)) {
1480 if (ie->ie_flags & IE_SOFT) {
1484 if (ie->ie_flags & IE_ENTROPY) {
1487 db_printf("ENTROPY");
1490 if (ie->ie_flags & IE_ADDING_THREAD) {
1493 db_printf("ADDING_THREAD");
1496 if (it != NULL && it->it_need) {
1506 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
1507 db_dump_intrhand(ih);
1511 * Dump data about interrupt handlers
1513 DB_SHOW_COMMAND(intr, db_show_intr)
1515 struct intr_event *ie;
1518 verbose = strchr(modif, 'v') != NULL;
1519 all = strchr(modif, 'a') != NULL;
1520 TAILQ_FOREACH(ie, &event_list, ie_list) {
1521 if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
1523 db_dump_intr_event(ie, verbose);
1531 * Start standard software interrupt threads
1534 start_softintr(void *dummy)
1537 if (swi_add(NULL, "vm", swi_vm, NULL, SWI_VM, INTR_MPSAFE, &vm_ih))
1538 panic("died while creating vm swi ithread");
1540 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1544 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1545 * The data for this machine dependent, and the declarations are in machine
1546 * dependent code. The layout of intrnames and intrcnt however is machine
1549 * We do not know the length of intrcnt and intrnames at compile time, so
1550 * calculate things at run time.
1553 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1555 return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1558 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD,
1559 NULL, 0, sysctl_intrnames, "", "Interrupt Names");
1562 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1565 uint32_t *intrcnt32;
1569 if (req->flags & SCTL_MASK32) {
1571 return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
1572 intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
1573 if (intrcnt32 == NULL)
1575 for (i = 0; i < sintrcnt / sizeof (u_long); i++)
1576 intrcnt32[i] = intrcnt[i];
1577 error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
1578 free(intrcnt32, M_TEMP);
1582 return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1585 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD,
1586 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts");
1590 * DDB command to dump the interrupt statistics.
1592 DB_SHOW_COMMAND(intrcnt, db_show_intrcnt)
1600 for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1605 db_printf("%s\t%lu\n", cp, *i);
1606 cp += strlen(cp) + 1;