1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
12 * disclaimer.
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.
16 *
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.
27 */
28
29 #include <sys/cdefs.h>
30 #include "opt_ddb.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kstack_usage_prof.h"
33
34 #include <sys/param.h>
35 #include <sys/bus.h>
36 #include <sys/conf.h>
37 #include <sys/cpuset.h>
38 #include <sys/rtprio.h>
39 #include <sys/systm.h>
40 #include <sys/interrupt.h>
41 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/ktr.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mutex.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/epoch.h>
51 #include <sys/random.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
54 #include <sys/smp.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/smp.h>
63 #include <machine/stdarg.h>
64 #ifdef DDB
65 #include <ddb/ddb.h>
66 #include <ddb/db_sym.h>
67 #endif
68
69 /*
70 * Describe an interrupt thread. There is one of these per interrupt event.
71 */
72 struct intr_thread {
73 struct intr_event *it_event;
74 struct thread *it_thread; /* Kernel thread. */
75 int it_flags; /* (j) IT_* flags. */
76 int it_need; /* Needs service. */
77 int it_waiting; /* Waiting in the runq. */
78 };
79
80 /* Interrupt thread flags kept in it_flags */
81 #define IT_DEAD 0x000001 /* Thread is waiting to exit. */
82 #define IT_WAIT 0x000002 /* Thread is waiting for completion. */
83
84 struct intr_entropy {
85 struct thread *td;
86 uintptr_t event;
87 };
88
89 struct intr_event *clk_intr_event;
90 struct proc *intrproc;
91
92 static MALLOC_DEFINE(M_ITHREAD, "ithread", "Interrupt Threads");
93
94 static int intr_storm_threshold = 0;
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 int intr_epoch_batch = 1000;
99 SYSCTL_INT(_hw, OID_AUTO, intr_epoch_batch, CTLFLAG_RWTUN, &intr_epoch_batch,
100 0, "Maximum interrupt handler executions without re-entering epoch(9)");
101 #ifdef HWPMC_HOOKS
102 static int intr_hwpmc_waiting_report_threshold = 1;
103 SYSCTL_INT(_hw, OID_AUTO, intr_hwpmc_waiting_report_threshold, CTLFLAG_RWTUN,
104 &intr_hwpmc_waiting_report_threshold, 1,
105 "Threshold for reporting number of events in a workq");
106 #define PMC_HOOK_INSTALLED_ANY() __predict_false(pmc_hook != NULL)
107 #endif
108 static TAILQ_HEAD(, intr_event) event_list =
109 TAILQ_HEAD_INITIALIZER(event_list);
110 static struct mtx event_lock;
111 MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
112
113 static void intr_event_update(struct intr_event *ie);
114 static int intr_event_schedule_thread(struct intr_event *ie, struct trapframe *frame);
115 static struct intr_thread *ithread_create(const char *name);
116 static void ithread_destroy(struct intr_thread *ithread);
117 static void ithread_execute_handlers(struct proc *p,
118 struct intr_event *ie);
119 static void ithread_loop(void *);
120 static void ithread_update(struct intr_thread *ithd);
121 static void start_softintr(void *);
122
123 #ifdef HWPMC_HOOKS
124 #include <sys/pmckern.h>
125 PMC_SOFT_DEFINE( , , intr, all);
126 PMC_SOFT_DEFINE( , , intr, ithread);
127 PMC_SOFT_DEFINE( , , intr, filter);
128 PMC_SOFT_DEFINE( , , intr, stray);
129 PMC_SOFT_DEFINE( , , intr, schedule);
130 PMC_SOFT_DEFINE( , , intr, waiting);
131
132 #define PMC_SOFT_CALL_INTR_HLPR(event, frame) \
133 do { \
134 if (frame != NULL) \
135 PMC_SOFT_CALL_TF( , , intr, event, frame); \
136 else \
137 PMC_SOFT_CALL( , , intr, event); \
138 } while (0)
139 #endif
140
141 /* Map an interrupt type to an ithread priority. */
142 u_char
intr_priority(enum intr_type flags)143 intr_priority(enum intr_type flags)
144 {
145 u_char pri;
146
147 flags &= (INTR_TYPE_TTY | INTR_TYPE_BIO | INTR_TYPE_NET |
148 INTR_TYPE_CAM | INTR_TYPE_MISC | INTR_TYPE_CLK | INTR_TYPE_AV);
149 switch (flags) {
150 case INTR_TYPE_TTY:
151 pri = PI_TTY;
152 break;
153 case INTR_TYPE_BIO:
154 pri = PI_DISK;
155 break;
156 case INTR_TYPE_NET:
157 pri = PI_NET;
158 break;
159 case INTR_TYPE_CAM:
160 pri = PI_DISK;
161 break;
162 case INTR_TYPE_AV:
163 pri = PI_AV;
164 break;
165 case INTR_TYPE_CLK:
166 pri = PI_REALTIME;
167 break;
168 case INTR_TYPE_MISC:
169 pri = PI_DULL; /* don't care */
170 break;
171 default:
172 /* We didn't specify an interrupt level. */
173 panic("intr_priority: no interrupt type in flags");
174 }
175
176 return pri;
177 }
178
179 /*
180 * Update an ithread based on the associated intr_event.
181 */
182 static void
ithread_update(struct intr_thread * ithd)183 ithread_update(struct intr_thread *ithd)
184 {
185 struct intr_event *ie;
186 struct thread *td;
187 u_char pri;
188
189 ie = ithd->it_event;
190 td = ithd->it_thread;
191 mtx_assert(&ie->ie_lock, MA_OWNED);
192
193 /* Determine the overall priority of this event. */
194 if (CK_SLIST_EMPTY(&ie->ie_handlers))
195 pri = PRI_MAX_ITHD;
196 else
197 pri = CK_SLIST_FIRST(&ie->ie_handlers)->ih_pri;
198
199 /* Update name and priority. */
200 strlcpy(td->td_name, ie->ie_fullname, sizeof(td->td_name));
201 #ifdef KTR
202 sched_clear_tdname(td);
203 #endif
204 thread_lock(td);
205 sched_ithread_prio(td, pri);
206 thread_unlock(td);
207 }
208
209 /*
210 * Regenerate the full name of an interrupt event and update its priority.
211 */
212 static void
intr_event_update(struct intr_event * ie)213 intr_event_update(struct intr_event *ie)
214 {
215 struct intr_handler *ih;
216 char *last;
217 int missed, space, flags;
218
219 /* Start off with no entropy and just the name of the event. */
220 mtx_assert(&ie->ie_lock, MA_OWNED);
221 strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
222 flags = 0;
223 missed = 0;
224 space = 1;
225
226 /* Run through all the handlers updating values. */
227 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
228 if (strlen(ie->ie_fullname) + strlen(ih->ih_name) + 1 <
229 sizeof(ie->ie_fullname)) {
230 strcat(ie->ie_fullname, " ");
231 strcat(ie->ie_fullname, ih->ih_name);
232 space = 0;
233 } else
234 missed++;
235 flags |= ih->ih_flags;
236 }
237 ie->ie_hflags = flags;
238
239 /*
240 * If there is only one handler and its name is too long, just copy in
241 * as much of the end of the name (includes the unit number) as will
242 * fit. Otherwise, we have multiple handlers and not all of the names
243 * will fit. Add +'s to indicate missing names. If we run out of room
244 * and still have +'s to add, change the last character from a + to a *.
245 */
246 if (missed == 1 && space == 1) {
247 ih = CK_SLIST_FIRST(&ie->ie_handlers);
248 missed = strlen(ie->ie_fullname) + strlen(ih->ih_name) + 2 -
249 sizeof(ie->ie_fullname);
250 strcat(ie->ie_fullname, (missed == 0) ? " " : "-");
251 strcat(ie->ie_fullname, &ih->ih_name[missed]);
252 missed = 0;
253 }
254 last = &ie->ie_fullname[sizeof(ie->ie_fullname) - 2];
255 while (missed-- > 0) {
256 if (strlen(ie->ie_fullname) + 1 == sizeof(ie->ie_fullname)) {
257 if (*last == '+') {
258 *last = '*';
259 break;
260 } else
261 *last = '+';
262 } else if (space) {
263 strcat(ie->ie_fullname, " +");
264 space = 0;
265 } else
266 strcat(ie->ie_fullname, "+");
267 }
268
269 /*
270 * If this event has an ithread, update it's priority and
271 * name.
272 */
273 if (ie->ie_thread != NULL)
274 ithread_update(ie->ie_thread);
275 CTR2(KTR_INTR, "%s: updated %s", __func__, ie->ie_fullname);
276 }
277
278 int
intr_event_create(struct intr_event ** event,void * source,int flags,u_int irq,void (* pre_ithread)(void *),void (* post_ithread)(void *),void (* post_filter)(void *),int (* assign_cpu)(void *,int),const char * fmt,...)279 intr_event_create(struct intr_event **event, void *source, int flags, u_int irq,
280 void (*pre_ithread)(void *), void (*post_ithread)(void *),
281 void (*post_filter)(void *), int (*assign_cpu)(void *, int),
282 const char *fmt, ...)
283 {
284 struct intr_event *ie;
285 va_list ap;
286
287 /* The only valid flag during creation is IE_SOFT. */
288 if ((flags & ~IE_SOFT) != 0)
289 return (EINVAL);
290 ie = malloc(sizeof(struct intr_event), M_ITHREAD, M_WAITOK | M_ZERO);
291 ie->ie_source = source;
292 ie->ie_pre_ithread = pre_ithread;
293 ie->ie_post_ithread = post_ithread;
294 ie->ie_post_filter = post_filter;
295 ie->ie_assign_cpu = assign_cpu;
296 ie->ie_flags = flags;
297 ie->ie_irq = irq;
298 ie->ie_cpu = NOCPU;
299 CK_SLIST_INIT(&ie->ie_handlers);
300 mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
301
302 va_start(ap, fmt);
303 vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
304 va_end(ap);
305 strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
306 mtx_lock(&event_lock);
307 TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
308 mtx_unlock(&event_lock);
309 if (event != NULL)
310 *event = ie;
311 CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
312 return (0);
313 }
314
315 /*
316 * Bind an interrupt event to the specified CPU. Note that not all
317 * platforms support binding an interrupt to a CPU. For those
318 * platforms this request will fail. Using a cpu id of NOCPU unbinds
319 * the interrupt event.
320 */
321 static int
_intr_event_bind(struct intr_event * ie,int cpu,bool bindirq,bool bindithread)322 _intr_event_bind(struct intr_event *ie, int cpu, bool bindirq, bool bindithread)
323 {
324 lwpid_t id;
325 int error;
326
327 /* Need a CPU to bind to. */
328 if (cpu != NOCPU && CPU_ABSENT(cpu))
329 return (EINVAL);
330
331 if (ie->ie_assign_cpu == NULL)
332 return (EOPNOTSUPP);
333
334 error = priv_check(curthread, PRIV_SCHED_CPUSET_INTR);
335 if (error)
336 return (error);
337
338 /*
339 * If we have any ithreads try to set their mask first to verify
340 * permissions, etc.
341 */
342 if (bindithread) {
343 mtx_lock(&ie->ie_lock);
344 if (ie->ie_thread != NULL) {
345 id = ie->ie_thread->it_thread->td_tid;
346 mtx_unlock(&ie->ie_lock);
347 error = cpuset_setithread(id, cpu);
348 if (error)
349 return (error);
350 } else
351 mtx_unlock(&ie->ie_lock);
352 }
353 if (bindirq)
354 error = ie->ie_assign_cpu(ie->ie_source, cpu);
355 if (error) {
356 if (bindithread) {
357 mtx_lock(&ie->ie_lock);
358 if (ie->ie_thread != NULL) {
359 cpu = ie->ie_cpu;
360 id = ie->ie_thread->it_thread->td_tid;
361 mtx_unlock(&ie->ie_lock);
362 (void)cpuset_setithread(id, cpu);
363 } else
364 mtx_unlock(&ie->ie_lock);
365 }
366 return (error);
367 }
368
369 if (bindirq) {
370 mtx_lock(&ie->ie_lock);
371 ie->ie_cpu = cpu;
372 mtx_unlock(&ie->ie_lock);
373 }
374
375 return (error);
376 }
377
378 /*
379 * Bind an interrupt event to the specified CPU. For supported platforms, any
380 * associated ithreads as well as the primary interrupt context will be bound
381 * to the specificed CPU.
382 */
383 int
intr_event_bind(struct intr_event * ie,int cpu)384 intr_event_bind(struct intr_event *ie, int cpu)
385 {
386
387 return (_intr_event_bind(ie, cpu, true, true));
388 }
389
390 /*
391 * Bind an interrupt event to the specified CPU, but do not bind associated
392 * ithreads.
393 */
394 int
intr_event_bind_irqonly(struct intr_event * ie,int cpu)395 intr_event_bind_irqonly(struct intr_event *ie, int cpu)
396 {
397
398 return (_intr_event_bind(ie, cpu, true, false));
399 }
400
401 /*
402 * Bind an interrupt event's ithread to the specified CPU.
403 */
404 int
intr_event_bind_ithread(struct intr_event * ie,int cpu)405 intr_event_bind_ithread(struct intr_event *ie, int cpu)
406 {
407
408 return (_intr_event_bind(ie, cpu, false, true));
409 }
410
411 /*
412 * Bind an interrupt event's ithread to the specified cpuset.
413 */
414 int
intr_event_bind_ithread_cpuset(struct intr_event * ie,cpuset_t * cs)415 intr_event_bind_ithread_cpuset(struct intr_event *ie, cpuset_t *cs)
416 {
417 lwpid_t id;
418
419 mtx_lock(&ie->ie_lock);
420 if (ie->ie_thread != NULL) {
421 id = ie->ie_thread->it_thread->td_tid;
422 mtx_unlock(&ie->ie_lock);
423 return (cpuset_setthread(id, cs));
424 } else {
425 mtx_unlock(&ie->ie_lock);
426 }
427 return (ENODEV);
428 }
429
430 static struct intr_event *
intr_lookup(int irq)431 intr_lookup(int irq)
432 {
433 struct intr_event *ie;
434
435 mtx_lock(&event_lock);
436 TAILQ_FOREACH(ie, &event_list, ie_list)
437 if (ie->ie_irq == irq &&
438 (ie->ie_flags & IE_SOFT) == 0 &&
439 CK_SLIST_FIRST(&ie->ie_handlers) != NULL)
440 break;
441 mtx_unlock(&event_lock);
442 return (ie);
443 }
444
445 int
intr_setaffinity(int irq,int mode,const void * m)446 intr_setaffinity(int irq, int mode, const void *m)
447 {
448 struct intr_event *ie;
449 const cpuset_t *mask;
450 int cpu, n;
451
452 mask = m;
453 cpu = NOCPU;
454 /*
455 * If we're setting all cpus we can unbind. Otherwise make sure
456 * only one cpu is in the set.
457 */
458 if (CPU_CMP(cpuset_root, mask)) {
459 for (n = 0; n < CPU_SETSIZE; n++) {
460 if (!CPU_ISSET(n, mask))
461 continue;
462 if (cpu != NOCPU)
463 return (EINVAL);
464 cpu = n;
465 }
466 }
467 ie = intr_lookup(irq);
468 if (ie == NULL)
469 return (ESRCH);
470 switch (mode) {
471 case CPU_WHICH_IRQ:
472 return (intr_event_bind(ie, cpu));
473 case CPU_WHICH_INTRHANDLER:
474 return (intr_event_bind_irqonly(ie, cpu));
475 case CPU_WHICH_ITHREAD:
476 return (intr_event_bind_ithread(ie, cpu));
477 default:
478 return (EINVAL);
479 }
480 }
481
482 int
intr_getaffinity(int irq,int mode,void * m)483 intr_getaffinity(int irq, int mode, void *m)
484 {
485 struct intr_event *ie;
486 struct thread *td;
487 struct proc *p;
488 cpuset_t *mask;
489 lwpid_t id;
490 int error;
491
492 mask = m;
493 ie = intr_lookup(irq);
494 if (ie == NULL)
495 return (ESRCH);
496
497 error = 0;
498 CPU_ZERO(mask);
499 switch (mode) {
500 case CPU_WHICH_IRQ:
501 case CPU_WHICH_INTRHANDLER:
502 mtx_lock(&ie->ie_lock);
503 if (ie->ie_cpu == NOCPU)
504 CPU_COPY(cpuset_root, mask);
505 else
506 CPU_SET(ie->ie_cpu, mask);
507 mtx_unlock(&ie->ie_lock);
508 break;
509 case CPU_WHICH_ITHREAD:
510 mtx_lock(&ie->ie_lock);
511 if (ie->ie_thread == NULL) {
512 mtx_unlock(&ie->ie_lock);
513 CPU_COPY(cpuset_root, mask);
514 } else {
515 id = ie->ie_thread->it_thread->td_tid;
516 mtx_unlock(&ie->ie_lock);
517 error = cpuset_which(CPU_WHICH_TID, id, &p, &td, NULL);
518 if (error != 0)
519 return (error);
520 CPU_COPY(&td->td_cpuset->cs_mask, mask);
521 PROC_UNLOCK(p);
522 }
523 default:
524 return (EINVAL);
525 }
526 return (0);
527 }
528
529 int
intr_event_destroy(struct intr_event * ie)530 intr_event_destroy(struct intr_event *ie)
531 {
532
533 if (ie == NULL)
534 return (EINVAL);
535
536 mtx_lock(&event_lock);
537 mtx_lock(&ie->ie_lock);
538 if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
539 mtx_unlock(&ie->ie_lock);
540 mtx_unlock(&event_lock);
541 return (EBUSY);
542 }
543 TAILQ_REMOVE(&event_list, ie, ie_list);
544 mtx_unlock(&event_lock);
545 if (ie->ie_thread != NULL)
546 ithread_destroy(ie->ie_thread);
547 mtx_unlock(&ie->ie_lock);
548 mtx_destroy(&ie->ie_lock);
549 free(ie, M_ITHREAD);
550 return (0);
551 }
552
553 static struct intr_thread *
ithread_create(const char * name)554 ithread_create(const char *name)
555 {
556 struct intr_thread *ithd;
557 struct thread *td;
558 int error;
559
560 ithd = malloc(sizeof(struct intr_thread), M_ITHREAD, M_WAITOK | M_ZERO);
561
562 error = kproc_kthread_add(ithread_loop, ithd, &intrproc,
563 &td, RFSTOPPED | RFHIGHPID,
564 0, "intr", "%s", name);
565 if (error)
566 panic("kproc_create() failed with %d", error);
567 thread_lock(td);
568 sched_class(td, PRI_ITHD);
569 TD_SET_IWAIT(td);
570 thread_unlock(td);
571 td->td_pflags |= TDP_ITHREAD;
572 ithd->it_thread = td;
573 CTR2(KTR_INTR, "%s: created %s", __func__, name);
574 return (ithd);
575 }
576
577 static void
ithread_destroy(struct intr_thread * ithread)578 ithread_destroy(struct intr_thread *ithread)
579 {
580 struct intr_event *ie;
581 struct thread *td;
582
583 td = ithread->it_thread;
584 ie = ithread->it_event;
585
586 mtx_assert(&ie->ie_lock, MA_OWNED);
587
588 CTR2(KTR_INTR, "%s: killing %s", __func__, ie->ie_name);
589
590 thread_lock(td);
591 ithread->it_flags |= IT_DEAD;
592 if (TD_AWAITING_INTR(td)) {
593 TD_CLR_IWAIT(td);
594 sched_wakeup(td, SRQ_INTR);
595 } else
596 thread_unlock(td);
597 while (ie->ie_thread != NULL)
598 msleep(ithread, &ie->ie_lock, 0, "ithd_dth", 0);
599 }
600
601 int
intr_event_add_handler(struct intr_event * ie,const char * name,driver_filter_t filter,driver_intr_t handler,void * arg,u_char pri,enum intr_type flags,void ** cookiep)602 intr_event_add_handler(struct intr_event *ie, const char *name,
603 driver_filter_t filter, driver_intr_t handler, void *arg, u_char pri,
604 enum intr_type flags, void **cookiep)
605 {
606 struct intr_handler *ih, *temp_ih;
607 struct intr_handler **prevptr;
608 struct intr_thread *it;
609
610 if (ie == NULL || name == NULL || (handler == NULL && filter == NULL))
611 return (EINVAL);
612
613 if ((flags & INTR_SLEEPABLE) != 0 && (flags & INTR_EXCL) == 0) {
614 printf("%s: INTR_SLEEPABLE requires INTR_EXCL to be set\n",
615 __func__);
616 return (EINVAL);
617 }
618
619 /* Allocate and populate an interrupt handler structure. */
620 ih = malloc(sizeof(struct intr_handler), M_ITHREAD, M_WAITOK | M_ZERO);
621 ih->ih_filter = filter;
622 ih->ih_handler = handler;
623 ih->ih_argument = arg;
624 strlcpy(ih->ih_name, name, sizeof(ih->ih_name));
625 ih->ih_event = ie;
626 ih->ih_pri = pri;
627 if (flags & INTR_EXCL)
628 ih->ih_flags = IH_EXCLUSIVE;
629 if (flags & INTR_MPSAFE)
630 ih->ih_flags |= IH_MPSAFE;
631 if (flags & INTR_ENTROPY)
632 ih->ih_flags |= IH_ENTROPY;
633 if (flags & INTR_TYPE_NET)
634 ih->ih_flags |= IH_NET;
635
636 /* We can only have one exclusive or sleepable handler in a event. */
637 mtx_lock(&ie->ie_lock);
638 if (!CK_SLIST_EMPTY(&ie->ie_handlers)) {
639 if ((flags & (INTR_EXCL | INTR_SLEEPABLE)) ||
640 (CK_SLIST_FIRST(&ie->ie_handlers)->ih_flags & IH_EXCLUSIVE)) {
641 mtx_unlock(&ie->ie_lock);
642 free(ih, M_ITHREAD);
643 return (EINVAL);
644 }
645 }
646 if (flags & INTR_SLEEPABLE)
647 ie->ie_flags |= IE_SLEEPABLE;
648
649 /* Create a thread if we need one. */
650 while (ie->ie_thread == NULL && handler != NULL) {
651 if (ie->ie_flags & IE_ADDING_THREAD)
652 msleep(ie, &ie->ie_lock, 0, "ithread", 0);
653 else {
654 ie->ie_flags |= IE_ADDING_THREAD;
655 mtx_unlock(&ie->ie_lock);
656 it = ithread_create("intr: newborn");
657 mtx_lock(&ie->ie_lock);
658 ie->ie_flags &= ~IE_ADDING_THREAD;
659 ie->ie_thread = it;
660 it->it_event = ie;
661 ithread_update(it);
662 wakeup(ie);
663 }
664 }
665
666 /* Add the new handler to the event in priority order. */
667 CK_SLIST_FOREACH_PREVPTR(temp_ih, prevptr, &ie->ie_handlers, ih_next) {
668 if (temp_ih->ih_pri > ih->ih_pri)
669 break;
670 }
671 CK_SLIST_INSERT_PREVPTR(prevptr, temp_ih, ih, ih_next);
672
673 intr_event_update(ie);
674
675 CTR3(KTR_INTR, "%s: added %s to %s", __func__, ih->ih_name,
676 ie->ie_name);
677 mtx_unlock(&ie->ie_lock);
678
679 if (cookiep != NULL)
680 *cookiep = ih;
681 return (0);
682 }
683
684 /*
685 * Append a description preceded by a ':' to the name of the specified
686 * interrupt handler.
687 */
688 int
intr_event_describe_handler(struct intr_event * ie,void * cookie,const char * descr)689 intr_event_describe_handler(struct intr_event *ie, void *cookie,
690 const char *descr)
691 {
692 struct intr_handler *ih;
693 size_t space;
694 char *start;
695
696 mtx_lock(&ie->ie_lock);
697 #ifdef INVARIANTS
698 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
699 if (ih == cookie)
700 break;
701 }
702 if (ih == NULL) {
703 mtx_unlock(&ie->ie_lock);
704 panic("handler %p not found in interrupt event %p", cookie, ie);
705 }
706 #endif
707 ih = cookie;
708
709 /*
710 * Look for an existing description by checking for an
711 * existing ":". This assumes device names do not include
712 * colons. If one is found, prepare to insert the new
713 * description at that point. If one is not found, find the
714 * end of the name to use as the insertion point.
715 */
716 start = strchr(ih->ih_name, ':');
717 if (start == NULL)
718 start = strchr(ih->ih_name, 0);
719
720 /*
721 * See if there is enough remaining room in the string for the
722 * description + ":". The "- 1" leaves room for the trailing
723 * '\0'. The "+ 1" accounts for the colon.
724 */
725 space = sizeof(ih->ih_name) - (start - ih->ih_name) - 1;
726 if (strlen(descr) + 1 > space) {
727 mtx_unlock(&ie->ie_lock);
728 return (ENOSPC);
729 }
730
731 /* Append a colon followed by the description. */
732 *start = ':';
733 strcpy(start + 1, descr);
734 intr_event_update(ie);
735 mtx_unlock(&ie->ie_lock);
736 return (0);
737 }
738
739 /*
740 * Return the ie_source field from the intr_event an intr_handler is
741 * associated with.
742 */
743 void *
intr_handler_source(void * cookie)744 intr_handler_source(void *cookie)
745 {
746 struct intr_handler *ih;
747 struct intr_event *ie;
748
749 ih = (struct intr_handler *)cookie;
750 if (ih == NULL)
751 return (NULL);
752 ie = ih->ih_event;
753 KASSERT(ie != NULL,
754 ("interrupt handler \"%s\" has a NULL interrupt event",
755 ih->ih_name));
756 return (ie->ie_source);
757 }
758
759 /*
760 * If intr_event_handle() is running in the ISR context at the time of the call,
761 * then wait for it to complete.
762 */
763 static void
intr_event_barrier(struct intr_event * ie)764 intr_event_barrier(struct intr_event *ie)
765 {
766 int phase;
767
768 mtx_assert(&ie->ie_lock, MA_OWNED);
769 phase = ie->ie_phase;
770
771 /*
772 * Switch phase to direct future interrupts to the other active counter.
773 * Make sure that any preceding stores are visible before the switch.
774 */
775 KASSERT(ie->ie_active[!phase] == 0, ("idle phase has activity"));
776 atomic_store_rel_int(&ie->ie_phase, !phase);
777
778 /*
779 * This code cooperates with wait-free iteration of ie_handlers
780 * in intr_event_handle.
781 * Make sure that the removal and the phase update are not reordered
782 * with the active count check.
783 * Note that no combination of acquire and release fences can provide
784 * that guarantee as Store->Load sequences can always be reordered.
785 */
786 atomic_thread_fence_seq_cst();
787
788 /*
789 * Now wait on the inactive phase.
790 * The acquire fence is needed so that all post-barrier accesses
791 * are after the check.
792 */
793 while (ie->ie_active[phase] > 0)
794 cpu_spinwait();
795 atomic_thread_fence_acq();
796 }
797
798 static void
intr_handler_barrier(struct intr_handler * handler)799 intr_handler_barrier(struct intr_handler *handler)
800 {
801 struct intr_event *ie;
802
803 ie = handler->ih_event;
804 mtx_assert(&ie->ie_lock, MA_OWNED);
805 KASSERT((handler->ih_flags & IH_DEAD) == 0,
806 ("update for a removed handler"));
807
808 if (ie->ie_thread == NULL) {
809 intr_event_barrier(ie);
810 return;
811 }
812 if ((handler->ih_flags & IH_CHANGED) == 0) {
813 handler->ih_flags |= IH_CHANGED;
814 intr_event_schedule_thread(ie, NULL);
815 }
816 while ((handler->ih_flags & IH_CHANGED) != 0)
817 msleep(handler, &ie->ie_lock, 0, "ih_barr", 0);
818 }
819
820 /*
821 * Sleep until an ithread finishes executing an interrupt handler.
822 *
823 * XXX Doesn't currently handle interrupt filters or fast interrupt
824 * handlers. This is intended for LinuxKPI drivers only.
825 * Do not use in BSD code.
826 */
827 void
_intr_drain(int irq)828 _intr_drain(int irq)
829 {
830 struct intr_event *ie;
831 struct intr_thread *ithd;
832 struct thread *td;
833
834 ie = intr_lookup(irq);
835 if (ie == NULL)
836 return;
837 if (ie->ie_thread == NULL)
838 return;
839 ithd = ie->ie_thread;
840 td = ithd->it_thread;
841 /*
842 * We set the flag and wait for it to be cleared to avoid
843 * long delays with potentially busy interrupt handlers
844 * were we to only sample TD_AWAITING_INTR() every tick.
845 */
846 thread_lock(td);
847 if (!TD_AWAITING_INTR(td)) {
848 ithd->it_flags |= IT_WAIT;
849 while (ithd->it_flags & IT_WAIT) {
850 thread_unlock(td);
851 pause("idrain", 1);
852 thread_lock(td);
853 }
854 }
855 thread_unlock(td);
856 return;
857 }
858
859 int
intr_event_remove_handler(void * cookie)860 intr_event_remove_handler(void *cookie)
861 {
862 struct intr_handler *handler = (struct intr_handler *)cookie;
863 struct intr_event *ie;
864 struct intr_handler *ih;
865 struct intr_handler **prevptr;
866
867 if (handler == NULL)
868 return (EINVAL);
869 ie = handler->ih_event;
870 KASSERT(ie != NULL,
871 ("interrupt handler \"%s\" has a NULL interrupt event",
872 handler->ih_name));
873
874 mtx_lock(&ie->ie_lock);
875 CTR3(KTR_INTR, "%s: removing %s from %s", __func__, handler->ih_name,
876 ie->ie_name);
877 CK_SLIST_FOREACH_PREVPTR(ih, prevptr, &ie->ie_handlers, ih_next) {
878 if (ih == handler)
879 break;
880 }
881 if (ih == NULL) {
882 panic("interrupt handler \"%s\" not found in "
883 "interrupt event \"%s\"", handler->ih_name, ie->ie_name);
884 }
885
886 if (ie->ie_thread == NULL) {
887 /*
888 * If there is no ithread, then directly remove the handler.
889 * Note that intr_event_handle() iterates ie_handlers in a
890 * lock-less fashion, so care needs to be taken to keep
891 * ie_handlers consistent and to free the removed handler only
892 * when ie_handlers is quiescent.
893 */
894 CK_SLIST_REMOVE_PREVPTR(prevptr, ih, ih_next);
895 intr_event_barrier(ie);
896 } else {
897 /*
898 * Let the interrupt thread do the job. The interrupt source is
899 * disabled when the interrupt thread is running, so it does not
900 * have to worry about interaction with intr_event_handle().
901 */
902 KASSERT((handler->ih_flags & IH_DEAD) == 0,
903 ("duplicate handle remove"));
904 handler->ih_flags |= IH_DEAD;
905 intr_event_schedule_thread(ie, NULL);
906 while (handler->ih_flags & IH_DEAD)
907 msleep(handler, &ie->ie_lock, 0, "iev_rmh", 0);
908 }
909 intr_event_update(ie);
910 mtx_unlock(&ie->ie_lock);
911 free(handler, M_ITHREAD);
912 return (0);
913 }
914
915 int
intr_event_suspend_handler(void * cookie)916 intr_event_suspend_handler(void *cookie)
917 {
918 struct intr_handler *handler = (struct intr_handler *)cookie;
919 struct intr_event *ie;
920
921 if (handler == NULL)
922 return (EINVAL);
923 ie = handler->ih_event;
924 KASSERT(ie != NULL,
925 ("interrupt handler \"%s\" has a NULL interrupt event",
926 handler->ih_name));
927 mtx_lock(&ie->ie_lock);
928 handler->ih_flags |= IH_SUSP;
929 intr_handler_barrier(handler);
930 mtx_unlock(&ie->ie_lock);
931 return (0);
932 }
933
934 int
intr_event_resume_handler(void * cookie)935 intr_event_resume_handler(void *cookie)
936 {
937 struct intr_handler *handler = (struct intr_handler *)cookie;
938 struct intr_event *ie;
939
940 if (handler == NULL)
941 return (EINVAL);
942 ie = handler->ih_event;
943 KASSERT(ie != NULL,
944 ("interrupt handler \"%s\" has a NULL interrupt event",
945 handler->ih_name));
946
947 /*
948 * intr_handler_barrier() acts not only as a barrier,
949 * it also allows to check for any pending interrupts.
950 */
951 mtx_lock(&ie->ie_lock);
952 handler->ih_flags &= ~IH_SUSP;
953 intr_handler_barrier(handler);
954 mtx_unlock(&ie->ie_lock);
955 return (0);
956 }
957
958 static int
intr_event_schedule_thread(struct intr_event * ie,struct trapframe * frame)959 intr_event_schedule_thread(struct intr_event *ie, struct trapframe *frame)
960 {
961 struct intr_entropy entropy;
962 struct intr_thread *it;
963 struct thread *td;
964 struct thread *ctd;
965
966 /*
967 * If no ithread or no handlers, then we have a stray interrupt.
968 */
969 if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers) ||
970 ie->ie_thread == NULL)
971 return (EINVAL);
972
973 ctd = curthread;
974 it = ie->ie_thread;
975 td = it->it_thread;
976
977 /*
978 * If any of the handlers for this ithread claim to be good
979 * sources of entropy, then gather some.
980 */
981 if (ie->ie_hflags & IH_ENTROPY) {
982 entropy.event = (uintptr_t)ie;
983 entropy.td = ctd;
984 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_INTERRUPT);
985 }
986
987 KASSERT(td->td_proc != NULL, ("ithread %s has no process", ie->ie_name));
988
989 /*
990 * Set it_need to tell the thread to keep running if it is already
991 * running. Then, lock the thread and see if we actually need to
992 * put it on the runqueue.
993 *
994 * Use store_rel to arrange that the store to ih_need in
995 * swi_sched() is before the store to it_need and prepare for
996 * transfer of this order to loads in the ithread.
997 */
998 atomic_store_rel_int(&it->it_need, 1);
999 thread_lock(td);
1000 if (TD_AWAITING_INTR(td)) {
1001 #ifdef HWPMC_HOOKS
1002 it->it_waiting = 0;
1003 if (PMC_HOOK_INSTALLED_ANY())
1004 PMC_SOFT_CALL_INTR_HLPR(schedule, frame);
1005 #endif
1006 CTR3(KTR_INTR, "%s: schedule pid %d (%s)", __func__, td->td_proc->p_pid,
1007 td->td_name);
1008 TD_CLR_IWAIT(td);
1009 sched_wakeup(td, SRQ_INTR);
1010 } else {
1011 #ifdef HWPMC_HOOKS
1012 it->it_waiting++;
1013 if (PMC_HOOK_INSTALLED_ANY() &&
1014 (it->it_waiting >= intr_hwpmc_waiting_report_threshold))
1015 PMC_SOFT_CALL_INTR_HLPR(waiting, frame);
1016 #endif
1017 CTR5(KTR_INTR, "%s: pid %d (%s): it_need %d, state %d",
1018 __func__, td->td_proc->p_pid, td->td_name, it->it_need, TD_GET_STATE(td));
1019 thread_unlock(td);
1020 }
1021
1022 return (0);
1023 }
1024
1025 /*
1026 * Allow interrupt event binding for software interrupt handlers -- a no-op,
1027 * since interrupts are generated in software rather than being directed by
1028 * a PIC.
1029 */
1030 static int
swi_assign_cpu(void * arg,int cpu)1031 swi_assign_cpu(void *arg, int cpu)
1032 {
1033
1034 return (0);
1035 }
1036
1037 /*
1038 * Add a software interrupt handler to a specified event. If a given event
1039 * is not specified, then a new event is created.
1040 */
1041 int
swi_add(struct intr_event ** eventp,const char * name,driver_intr_t handler,void * arg,int pri,enum intr_type flags,void ** cookiep)1042 swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
1043 void *arg, int pri, enum intr_type flags, void **cookiep)
1044 {
1045 struct intr_event *ie;
1046 int error = 0;
1047
1048 if (flags & INTR_ENTROPY)
1049 return (EINVAL);
1050
1051 ie = (eventp != NULL) ? *eventp : NULL;
1052
1053 if (ie != NULL) {
1054 if (!(ie->ie_flags & IE_SOFT))
1055 return (EINVAL);
1056 } else {
1057 error = intr_event_create(&ie, NULL, IE_SOFT, 0,
1058 NULL, NULL, NULL, swi_assign_cpu, "swi%d:", pri);
1059 if (error)
1060 return (error);
1061 if (eventp != NULL)
1062 *eventp = ie;
1063 }
1064 if (handler != NULL) {
1065 error = intr_event_add_handler(ie, name, NULL, handler, arg,
1066 PI_SWI(pri), flags, cookiep);
1067 }
1068 return (error);
1069 }
1070
1071 /*
1072 * Schedule a software interrupt thread.
1073 */
1074 void
swi_sched(void * cookie,int flags)1075 swi_sched(void *cookie, int flags)
1076 {
1077 struct intr_handler *ih = (struct intr_handler *)cookie;
1078 struct intr_event *ie = ih->ih_event;
1079 struct intr_entropy entropy;
1080 int error __unused;
1081
1082 CTR3(KTR_INTR, "swi_sched: %s %s need=%d", ie->ie_name, ih->ih_name,
1083 ih->ih_need);
1084
1085 if ((flags & SWI_FROMNMI) == 0) {
1086 entropy.event = (uintptr_t)ih;
1087 entropy.td = curthread;
1088 random_harvest_queue(&entropy, sizeof(entropy), RANDOM_SWI);
1089 }
1090
1091 /*
1092 * Set ih_need for this handler so that if the ithread is already
1093 * running it will execute this handler on the next pass. Otherwise,
1094 * it will execute it the next time it runs.
1095 */
1096 ih->ih_need = 1;
1097
1098 if (flags & SWI_DELAY)
1099 return;
1100
1101 if (flags & SWI_FROMNMI) {
1102 #if defined(SMP) && (defined(__i386__) || defined(__amd64__))
1103 KASSERT(ie == clk_intr_event,
1104 ("SWI_FROMNMI used not with clk_intr_event"));
1105 ipi_self_from_nmi(IPI_SWI);
1106 #endif
1107 } else {
1108 VM_CNT_INC(v_soft);
1109 error = intr_event_schedule_thread(ie, NULL);
1110 KASSERT(error == 0, ("stray software interrupt"));
1111 }
1112 }
1113
1114 /*
1115 * Remove a software interrupt handler. Currently this code does not
1116 * remove the associated interrupt event if it becomes empty. Calling code
1117 * may do so manually via intr_event_destroy(), but that's not really
1118 * an optimal interface.
1119 */
1120 int
swi_remove(void * cookie)1121 swi_remove(void *cookie)
1122 {
1123
1124 return (intr_event_remove_handler(cookie));
1125 }
1126
1127 static void
intr_event_execute_handlers(struct proc * p,struct intr_event * ie)1128 intr_event_execute_handlers(struct proc *p, struct intr_event *ie)
1129 {
1130 struct intr_handler *ih, *ihn, *ihp;
1131
1132 ihp = NULL;
1133 CK_SLIST_FOREACH_SAFE(ih, &ie->ie_handlers, ih_next, ihn) {
1134 /*
1135 * If this handler is marked for death, remove it from
1136 * the list of handlers and wake up the sleeper.
1137 */
1138 if (ih->ih_flags & IH_DEAD) {
1139 mtx_lock(&ie->ie_lock);
1140 if (ihp == NULL)
1141 CK_SLIST_REMOVE_HEAD(&ie->ie_handlers, ih_next);
1142 else
1143 CK_SLIST_REMOVE_AFTER(ihp, ih_next);
1144 ih->ih_flags &= ~IH_DEAD;
1145 wakeup(ih);
1146 mtx_unlock(&ie->ie_lock);
1147 continue;
1148 }
1149
1150 /*
1151 * Now that we know that the current element won't be removed
1152 * update the previous element.
1153 */
1154 ihp = ih;
1155
1156 if ((ih->ih_flags & IH_CHANGED) != 0) {
1157 mtx_lock(&ie->ie_lock);
1158 ih->ih_flags &= ~IH_CHANGED;
1159 wakeup(ih);
1160 mtx_unlock(&ie->ie_lock);
1161 }
1162
1163 /* Skip filter only handlers */
1164 if (ih->ih_handler == NULL)
1165 continue;
1166
1167 /* Skip suspended handlers */
1168 if ((ih->ih_flags & IH_SUSP) != 0)
1169 continue;
1170
1171 /*
1172 * For software interrupt threads, we only execute
1173 * handlers that have their need flag set. Hardware
1174 * interrupt threads always invoke all of their handlers.
1175 *
1176 * ih_need can only be 0 or 1. Failed cmpset below
1177 * means that there is no request to execute handlers,
1178 * so a retry of the cmpset is not needed.
1179 */
1180 if ((ie->ie_flags & IE_SOFT) != 0 &&
1181 atomic_cmpset_int(&ih->ih_need, 1, 0) == 0)
1182 continue;
1183
1184 /* Execute this handler. */
1185 CTR6(KTR_INTR, "%s: pid %d exec %p(%p) for %s flg=%x",
1186 __func__, p->p_pid, (void *)ih->ih_handler,
1187 ih->ih_argument, ih->ih_name, ih->ih_flags);
1188
1189 if (!(ih->ih_flags & IH_MPSAFE))
1190 mtx_lock(&Giant);
1191 ih->ih_handler(ih->ih_argument);
1192 if (!(ih->ih_flags & IH_MPSAFE))
1193 mtx_unlock(&Giant);
1194 }
1195 }
1196
1197 static void
ithread_execute_handlers(struct proc * p,struct intr_event * ie)1198 ithread_execute_handlers(struct proc *p, struct intr_event *ie)
1199 {
1200
1201 /* Only specifically marked sleepable interrupt handlers can sleep. */
1202 if (!(ie->ie_flags & (IE_SOFT | IE_SLEEPABLE)))
1203 THREAD_NO_SLEEPING();
1204 intr_event_execute_handlers(p, ie);
1205 if (!(ie->ie_flags & (IE_SOFT | IE_SLEEPABLE)))
1206 THREAD_SLEEPING_OK();
1207
1208 /*
1209 * Interrupt storm handling:
1210 *
1211 * If this interrupt source is currently storming, then throttle
1212 * it to only fire the handler once per clock tick.
1213 *
1214 * If this interrupt source is not currently storming, but the
1215 * number of back to back interrupts exceeds the storm threshold,
1216 * then enter storming mode.
1217 */
1218 if (__predict_false(intr_storm_threshold != 0 &&
1219 ie->ie_count >= intr_storm_threshold &&
1220 (ie->ie_flags & IE_SOFT) == 0)) {
1221 /* Report the message only once every second. */
1222 if (ppsratecheck(&ie->ie_warntm, &ie->ie_warncnt, 1)) {
1223 printf(
1224 "interrupt storm detected on \"%s\"; throttling interrupt source\n",
1225 ie->ie_name);
1226 }
1227 pause("istorm", 1);
1228 } else
1229 ie->ie_count++;
1230
1231 /*
1232 * Now that all the handlers have had a chance to run, reenable
1233 * the interrupt source.
1234 */
1235 if (ie->ie_post_ithread != NULL)
1236 ie->ie_post_ithread(ie->ie_source);
1237 }
1238
1239 /*
1240 * This is the main code for interrupt threads.
1241 */
1242 static void
ithread_loop(void * arg)1243 ithread_loop(void *arg)
1244 {
1245 struct epoch_tracker et;
1246 struct intr_thread *ithd;
1247 struct intr_event *ie;
1248 struct thread *td;
1249 struct proc *p;
1250 int epoch_count;
1251 bool needs_epoch;
1252
1253 td = curthread;
1254 p = td->td_proc;
1255 ithd = (struct intr_thread *)arg;
1256 KASSERT(ithd->it_thread == td,
1257 ("%s: ithread and proc linkage out of sync", __func__));
1258 ie = ithd->it_event;
1259 ie->ie_count = 0;
1260
1261 /*
1262 * As long as we have interrupts outstanding, go through the
1263 * list of handlers, giving each one a go at it.
1264 */
1265 for (;;) {
1266 /*
1267 * If we are an orphaned thread, then just die.
1268 */
1269 if (__predict_false((ithd->it_flags & IT_DEAD) != 0)) {
1270 CTR3(KTR_INTR, "%s: pid %d (%s) exiting", __func__,
1271 p->p_pid, td->td_name);
1272 mtx_lock(&ie->ie_lock);
1273 ie->ie_thread = NULL;
1274 wakeup(ithd);
1275 mtx_unlock(&ie->ie_lock);
1276
1277 free(ithd, M_ITHREAD);
1278 kthread_exit();
1279 }
1280
1281 /*
1282 * Service interrupts. If another interrupt arrives while
1283 * we are running, it will set it_need to note that we
1284 * should make another pass.
1285 *
1286 * The load_acq part of the following cmpset ensures
1287 * that the load of ih_need in ithread_execute_handlers()
1288 * is ordered after the load of it_need here.
1289 */
1290 needs_epoch =
1291 (atomic_load_int(&ie->ie_hflags) & IH_NET) != 0;
1292 if (needs_epoch) {
1293 epoch_count = 0;
1294 NET_EPOCH_ENTER(et);
1295 }
1296 while (atomic_cmpset_acq_int(&ithd->it_need, 1, 0) != 0) {
1297 ithread_execute_handlers(p, ie);
1298 if (needs_epoch &&
1299 ++epoch_count >= intr_epoch_batch) {
1300 NET_EPOCH_EXIT(et);
1301 epoch_count = 0;
1302 NET_EPOCH_ENTER(et);
1303 }
1304 }
1305 if (needs_epoch)
1306 NET_EPOCH_EXIT(et);
1307 WITNESS_WARN(WARN_PANIC, NULL, "suspending ithread");
1308 mtx_assert(&Giant, MA_NOTOWNED);
1309
1310 /*
1311 * Processed all our interrupts. Now get the sched
1312 * lock. This may take a while and it_need may get
1313 * set again, so we have to check it again.
1314 */
1315 thread_lock(td);
1316 if (atomic_load_acq_int(&ithd->it_need) == 0 &&
1317 (ithd->it_flags & (IT_DEAD | IT_WAIT)) == 0) {
1318 TD_SET_IWAIT(td);
1319 ie->ie_count = 0;
1320 mi_switch(SW_VOL | SWT_IWAIT);
1321 } else if ((ithd->it_flags & IT_WAIT) != 0) {
1322 ithd->it_flags &= ~IT_WAIT;
1323 thread_unlock(td);
1324 wakeup(ithd);
1325 } else
1326 thread_unlock(td);
1327 }
1328 }
1329
1330 /*
1331 * Main interrupt handling body.
1332 *
1333 * Input:
1334 * o ie: the event connected to this interrupt.
1335 --------------------------------------------------------------------------------
1336 * o frame: the current trap frame. If the client interrupt
1337 * handler needs this frame, they should get it
1338 * via curthread->td_intr_frame.
1339 *
1340 * Return value:
1341 * o 0: everything ok.
1342 * o EINVAL: stray interrupt.
1343 */
1344 int
intr_event_handle(struct intr_event * ie,struct trapframe * frame)1345 intr_event_handle(struct intr_event *ie, struct trapframe *frame)
1346 {
1347 struct intr_handler *ih;
1348 struct trapframe *oldframe;
1349 struct thread *td;
1350 int phase;
1351 int ret;
1352 bool filter, thread;
1353
1354 td = curthread;
1355
1356 #ifdef KSTACK_USAGE_PROF
1357 intr_prof_stack_use(td, frame);
1358 #endif
1359
1360 /* An interrupt with no event or handlers is a stray interrupt. */
1361 if (ie == NULL || CK_SLIST_EMPTY(&ie->ie_handlers))
1362 return (EINVAL);
1363
1364 /*
1365 * Execute fast interrupt handlers directly.
1366 */
1367 td->td_intr_nesting_level++;
1368 filter = false;
1369 thread = false;
1370 ret = 0;
1371 critical_enter();
1372 oldframe = td->td_intr_frame;
1373 td->td_intr_frame = frame;
1374
1375 phase = ie->ie_phase;
1376 atomic_add_int(&ie->ie_active[phase], 1);
1377
1378 /*
1379 * This fence is required to ensure that no later loads are
1380 * re-ordered before the ie_active store.
1381 */
1382 atomic_thread_fence_seq_cst();
1383
1384 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next) {
1385 if ((ih->ih_flags & IH_SUSP) != 0)
1386 continue;
1387 if ((ie->ie_flags & IE_SOFT) != 0 && ih->ih_need == 0)
1388 continue;
1389 if (ih->ih_filter == NULL) {
1390 thread = true;
1391 continue;
1392 }
1393 CTR4(KTR_INTR, "%s: exec %p(%p) for %s", __func__,
1394 ih->ih_filter, ih->ih_argument, ih->ih_name);
1395 ret = ih->ih_filter(ih->ih_argument);
1396 #ifdef HWPMC_HOOKS
1397 PMC_SOFT_CALL_TF( , , intr, all, frame);
1398 #endif
1399 KASSERT(ret == FILTER_STRAY ||
1400 ((ret & (FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) != 0 &&
1401 (ret & ~(FILTER_SCHEDULE_THREAD | FILTER_HANDLED)) == 0),
1402 ("%s: incorrect return value %#x from %s", __func__, ret,
1403 ih->ih_name));
1404 filter = filter || ret == FILTER_HANDLED;
1405 #ifdef HWPMC_HOOKS
1406 if (ret & FILTER_SCHEDULE_THREAD)
1407 PMC_SOFT_CALL_TF( , , intr, ithread, frame);
1408 else if (ret & FILTER_HANDLED)
1409 PMC_SOFT_CALL_TF( , , intr, filter, frame);
1410 else if (ret == FILTER_STRAY)
1411 PMC_SOFT_CALL_TF( , , intr, stray, frame);
1412 #endif
1413
1414 /*
1415 * Wrapper handler special handling:
1416 *
1417 * in some particular cases (like pccard and pccbb),
1418 * the _real_ device handler is wrapped in a couple of
1419 * functions - a filter wrapper and an ithread wrapper.
1420 * In this case (and just in this case), the filter wrapper
1421 * could ask the system to schedule the ithread and mask
1422 * the interrupt source if the wrapped handler is composed
1423 * of just an ithread handler.
1424 *
1425 * TODO: write a generic wrapper to avoid people rolling
1426 * their own.
1427 */
1428 if (!thread) {
1429 if (ret == FILTER_SCHEDULE_THREAD)
1430 thread = true;
1431 }
1432 }
1433 atomic_add_rel_int(&ie->ie_active[phase], -1);
1434
1435 td->td_intr_frame = oldframe;
1436
1437 if (thread) {
1438 if (ie->ie_pre_ithread != NULL)
1439 ie->ie_pre_ithread(ie->ie_source);
1440 } else {
1441 if (ie->ie_post_filter != NULL)
1442 ie->ie_post_filter(ie->ie_source);
1443 }
1444
1445 /* Schedule the ithread if needed. */
1446 if (thread) {
1447 int error __unused;
1448
1449 error = intr_event_schedule_thread(ie, frame);
1450 KASSERT(error == 0, ("bad stray interrupt"));
1451 }
1452 critical_exit();
1453 td->td_intr_nesting_level--;
1454 #ifdef notyet
1455 /* The interrupt is not aknowledged by any filter and has no ithread. */
1456 if (!thread && !filter)
1457 return (EINVAL);
1458 #endif
1459 return (0);
1460 }
1461
1462 #ifdef DDB
1463 /*
1464 * Dump details about an interrupt handler
1465 */
1466 static void
db_dump_intrhand(struct intr_handler * ih)1467 db_dump_intrhand(struct intr_handler *ih)
1468 {
1469 int comma;
1470
1471 db_printf("\t%-10s ", ih->ih_name);
1472 switch (ih->ih_pri) {
1473 case PI_REALTIME:
1474 db_printf("CLK ");
1475 break;
1476 case PI_INTR:
1477 db_printf("INTR");
1478 break;
1479 default:
1480 if (ih->ih_pri >= PI_SOFT)
1481 db_printf("SWI ");
1482 else
1483 db_printf("%4u", ih->ih_pri);
1484 break;
1485 }
1486 db_printf(" ");
1487 if (ih->ih_filter != NULL) {
1488 db_printf("[F]");
1489 db_printsym((uintptr_t)ih->ih_filter, DB_STGY_PROC);
1490 }
1491 if (ih->ih_handler != NULL) {
1492 if (ih->ih_filter != NULL)
1493 db_printf(",");
1494 db_printf("[H]");
1495 db_printsym((uintptr_t)ih->ih_handler, DB_STGY_PROC);
1496 }
1497 db_printf("(%p)", ih->ih_argument);
1498 if (ih->ih_need ||
1499 (ih->ih_flags & (IH_EXCLUSIVE | IH_ENTROPY | IH_DEAD |
1500 IH_MPSAFE)) != 0) {
1501 db_printf(" {");
1502 comma = 0;
1503 if (ih->ih_flags & IH_EXCLUSIVE) {
1504 if (comma)
1505 db_printf(", ");
1506 db_printf("EXCL");
1507 comma = 1;
1508 }
1509 if (ih->ih_flags & IH_ENTROPY) {
1510 if (comma)
1511 db_printf(", ");
1512 db_printf("ENTROPY");
1513 comma = 1;
1514 }
1515 if (ih->ih_flags & IH_DEAD) {
1516 if (comma)
1517 db_printf(", ");
1518 db_printf("DEAD");
1519 comma = 1;
1520 }
1521 if (ih->ih_flags & IH_MPSAFE) {
1522 if (comma)
1523 db_printf(", ");
1524 db_printf("MPSAFE");
1525 comma = 1;
1526 }
1527 if (ih->ih_need) {
1528 if (comma)
1529 db_printf(", ");
1530 db_printf("NEED");
1531 }
1532 db_printf("}");
1533 }
1534 db_printf("\n");
1535 }
1536
1537 /*
1538 * Dump details about a event.
1539 */
1540 void
db_dump_intr_event(struct intr_event * ie,int handlers)1541 db_dump_intr_event(struct intr_event *ie, int handlers)
1542 {
1543 struct intr_handler *ih;
1544 struct intr_thread *it;
1545 int comma;
1546
1547 db_printf("%s ", ie->ie_fullname);
1548 it = ie->ie_thread;
1549 if (it != NULL)
1550 db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
1551 else
1552 db_printf("(no thread)");
1553 if ((ie->ie_flags & (IE_SOFT | IE_ADDING_THREAD)) != 0 ||
1554 (it != NULL && it->it_need)) {
1555 db_printf(" {");
1556 comma = 0;
1557 if (ie->ie_flags & IE_SOFT) {
1558 db_printf("SOFT");
1559 comma = 1;
1560 }
1561 if (ie->ie_flags & IE_ADDING_THREAD) {
1562 if (comma)
1563 db_printf(", ");
1564 db_printf("ADDING_THREAD");
1565 comma = 1;
1566 }
1567 if (it != NULL && it->it_need) {
1568 if (comma)
1569 db_printf(", ");
1570 db_printf("NEED");
1571 }
1572 db_printf("}");
1573 }
1574 db_printf("\n");
1575
1576 if (handlers)
1577 CK_SLIST_FOREACH(ih, &ie->ie_handlers, ih_next)
1578 db_dump_intrhand(ih);
1579 }
1580
1581 /*
1582 * Dump data about interrupt handlers
1583 */
DB_SHOW_COMMAND_FLAGS(intr,db_show_intr,DB_CMD_MEMSAFE)1584 DB_SHOW_COMMAND_FLAGS(intr, db_show_intr, DB_CMD_MEMSAFE)
1585 {
1586 struct intr_event *ie;
1587 int all, verbose;
1588
1589 verbose = strchr(modif, 'v') != NULL;
1590 all = strchr(modif, 'a') != NULL;
1591 TAILQ_FOREACH(ie, &event_list, ie_list) {
1592 if (!all && CK_SLIST_EMPTY(&ie->ie_handlers))
1593 continue;
1594 db_dump_intr_event(ie, verbose);
1595 if (db_pager_quit)
1596 break;
1597 }
1598 }
1599 #endif /* DDB */
1600
1601 /*
1602 * Start standard software interrupt threads
1603 */
1604 static void
start_softintr(void * dummy)1605 start_softintr(void *dummy)
1606 {
1607
1608 if (swi_add(&clk_intr_event, "clk", NULL, NULL, SWI_CLOCK,
1609 INTR_MPSAFE, NULL))
1610 panic("died while creating clk swi ithread");
1611 }
1612 SYSINIT(start_softintr, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softintr,
1613 NULL);
1614
1615 /*
1616 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt.
1617 * The data for this machine dependent, and the declarations are in machine
1618 * dependent code. The layout of intrnames and intrcnt however is machine
1619 * independent.
1620 *
1621 * We do not know the length of intrcnt and intrnames at compile time, so
1622 * calculate things at run time.
1623 */
1624 static int
sysctl_intrnames(SYSCTL_HANDLER_ARGS)1625 sysctl_intrnames(SYSCTL_HANDLER_ARGS)
1626 {
1627 return (sysctl_handle_opaque(oidp, intrnames, sintrnames, req));
1628 }
1629
1630 SYSCTL_PROC(_hw, OID_AUTO, intrnames,
1631 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1632 sysctl_intrnames, "",
1633 "Interrupt Names");
1634
1635 static int
sysctl_intrcnt(SYSCTL_HANDLER_ARGS)1636 sysctl_intrcnt(SYSCTL_HANDLER_ARGS)
1637 {
1638 #ifdef SCTL_MASK32
1639 uint32_t *intrcnt32;
1640 unsigned i;
1641 int error;
1642
1643 if (req->flags & SCTL_MASK32) {
1644 if (!req->oldptr)
1645 return (sysctl_handle_opaque(oidp, NULL, sintrcnt / 2, req));
1646 intrcnt32 = malloc(sintrcnt / 2, M_TEMP, M_NOWAIT);
1647 if (intrcnt32 == NULL)
1648 return (ENOMEM);
1649 for (i = 0; i < sintrcnt / sizeof (u_long); i++)
1650 intrcnt32[i] = intrcnt[i];
1651 error = sysctl_handle_opaque(oidp, intrcnt32, sintrcnt / 2, req);
1652 free(intrcnt32, M_TEMP);
1653 return (error);
1654 }
1655 #endif
1656 return (sysctl_handle_opaque(oidp, intrcnt, sintrcnt, req));
1657 }
1658
1659 SYSCTL_PROC(_hw, OID_AUTO, intrcnt,
1660 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1661 sysctl_intrcnt, "",
1662 "Interrupt Counts");
1663
1664 #ifdef DDB
1665 /*
1666 * DDB command to dump the interrupt statistics.
1667 */
DB_SHOW_COMMAND_FLAGS(intrcnt,db_show_intrcnt,DB_CMD_MEMSAFE)1668 DB_SHOW_COMMAND_FLAGS(intrcnt, db_show_intrcnt, DB_CMD_MEMSAFE)
1669 {
1670 u_long *i;
1671 char *cp;
1672 u_int j;
1673
1674 cp = intrnames;
1675 j = 0;
1676 for (i = intrcnt; j < (sintrcnt / sizeof(u_long)) && !db_pager_quit;
1677 i++, j++) {
1678 if (*cp == '\0')
1679 break;
1680 if (*i != 0)
1681 db_printf("%s\t%lu\n", cp, *i);
1682 cp += strlen(cp) + 1;
1683 }
1684 }
1685 #endif
1686