xref: /freebsd/sys/kern/subr_intr.c (revision 96fbe51956c9df8bdb8317413b1c487b14e4ee68)
1 /*-
2  * Copyright (c) 2015-2016 Svatopluk Kraus
3  * Copyright (c) 2015-2016 Michal Meloun
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following 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 AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 /*
32  *	New-style Interrupt Framework
33  *
34  *  TODO: - add support for disconnected PICs.
35  *        - to support IPI (PPI) enabling on other CPUs if already started.
36  *        - to complete things for removable PICs.
37  */
38 
39 #include "opt_ddb.h"
40 #include "opt_hwpmc_hooks.h"
41 #include "opt_iommu.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/syslog.h>
49 #include <sys/malloc.h>
50 #include <sys/proc.h>
51 #include <sys/queue.h>
52 #include <sys/bus.h>
53 #include <sys/interrupt.h>
54 #include <sys/taskqueue.h>
55 #include <sys/tree.h>
56 #include <sys/conf.h>
57 #include <sys/cpuset.h>
58 #include <sys/rman.h>
59 #include <sys/sched.h>
60 #include <sys/smp.h>
61 #include <sys/vmmeter.h>
62 #ifdef HWPMC_HOOKS
63 #include <sys/pmckern.h>
64 #endif
65 
66 #include <machine/atomic.h>
67 #include <machine/intr.h>
68 #include <machine/cpu.h>
69 #include <machine/smp.h>
70 #include <machine/stdarg.h>
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #ifdef IOMMU
77 #include <dev/iommu/iommu_msi.h>
78 #endif
79 
80 #include "pic_if.h"
81 #include "msi_if.h"
82 
83 #define	INTRNAME_LEN	(2*MAXCOMLEN + 1)
84 
85 #ifdef DEBUG
86 #define debugf(fmt, args...) do { printf("%s(): ", __func__);	\
87     printf(fmt,##args); } while (0)
88 #else
89 #define debugf(fmt, args...)
90 #endif
91 
92 MALLOC_DECLARE(M_INTRNG);
93 MALLOC_DEFINE(M_INTRNG, "intr", "intr interrupt handling");
94 
95 /* Main interrupt handler called from assembler -> 'hidden' for C code. */
96 void intr_irq_handler(struct trapframe *tf);
97 
98 /* Root interrupt controller stuff. */
99 device_t intr_irq_root_dev;
100 static intr_irq_filter_t *irq_root_filter;
101 static void *irq_root_arg;
102 static u_int irq_root_ipicount;
103 
104 struct intr_pic_child {
105 	SLIST_ENTRY(intr_pic_child)	 pc_next;
106 	struct intr_pic			*pc_pic;
107 	intr_child_irq_filter_t		*pc_filter;
108 	void				*pc_filter_arg;
109 	uintptr_t			 pc_start;
110 	uintptr_t			 pc_length;
111 };
112 
113 /* Interrupt controller definition. */
114 struct intr_pic {
115 	SLIST_ENTRY(intr_pic)	pic_next;
116 	intptr_t		pic_xref;	/* hardware identification */
117 	device_t		pic_dev;
118 /* Only one of FLAG_PIC or FLAG_MSI may be set */
119 #define	FLAG_PIC	(1 << 0)
120 #define	FLAG_MSI	(1 << 1)
121 #define	FLAG_TYPE_MASK	(FLAG_PIC | FLAG_MSI)
122 	u_int			pic_flags;
123 	struct mtx		pic_child_lock;
124 	SLIST_HEAD(, intr_pic_child) pic_children;
125 };
126 
127 static struct mtx pic_list_lock;
128 static SLIST_HEAD(, intr_pic) pic_list;
129 
130 static struct intr_pic *pic_lookup(device_t dev, intptr_t xref, int flags);
131 
132 /* Interrupt source definition. */
133 static struct mtx isrc_table_lock;
134 static struct intr_irqsrc *irq_sources[NIRQ];
135 u_int irq_next_free;
136 
137 #ifdef SMP
138 #ifdef EARLY_AP_STARTUP
139 static bool irq_assign_cpu = true;
140 #else
141 static bool irq_assign_cpu = false;
142 #endif
143 #endif
144 
145 /*
146  * - 2 counters for each I/O interrupt.
147  * - MAXCPU counters for each IPI counters for SMP.
148  */
149 #ifdef SMP
150 #define INTRCNT_COUNT   (NIRQ * 2 + INTR_IPI_COUNT * MAXCPU)
151 #else
152 #define INTRCNT_COUNT   (NIRQ * 2)
153 #endif
154 
155 /* Data for MI statistics reporting. */
156 u_long intrcnt[INTRCNT_COUNT];
157 char intrnames[INTRCNT_COUNT * INTRNAME_LEN];
158 size_t sintrcnt = sizeof(intrcnt);
159 size_t sintrnames = sizeof(intrnames);
160 static u_int intrcnt_index;
161 
162 static struct intr_irqsrc *intr_map_get_isrc(u_int res_id);
163 static void intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc);
164 static struct intr_map_data * intr_map_get_map_data(u_int res_id);
165 static void intr_map_copy_map_data(u_int res_id, device_t *dev, intptr_t *xref,
166     struct intr_map_data **data);
167 
168 /*
169  *  Interrupt framework initialization routine.
170  */
171 static void
172 intr_irq_init(void *dummy __unused)
173 {
174 
175 	SLIST_INIT(&pic_list);
176 	mtx_init(&pic_list_lock, "intr pic list", NULL, MTX_DEF);
177 
178 	mtx_init(&isrc_table_lock, "intr isrc table", NULL, MTX_DEF);
179 }
180 SYSINIT(intr_irq_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_irq_init, NULL);
181 
182 static void
183 intrcnt_setname(const char *name, int index)
184 {
185 
186 	snprintf(intrnames + INTRNAME_LEN * index, INTRNAME_LEN, "%-*s",
187 	    INTRNAME_LEN - 1, name);
188 }
189 
190 /*
191  *  Update name for interrupt source with interrupt event.
192  */
193 static void
194 intrcnt_updatename(struct intr_irqsrc *isrc)
195 {
196 
197 	/* QQQ: What about stray counter name? */
198 	mtx_assert(&isrc_table_lock, MA_OWNED);
199 	intrcnt_setname(isrc->isrc_event->ie_fullname, isrc->isrc_index);
200 }
201 
202 /*
203  *  Virtualization for interrupt source interrupt counter increment.
204  */
205 static inline void
206 isrc_increment_count(struct intr_irqsrc *isrc)
207 {
208 
209 	if (isrc->isrc_flags & INTR_ISRCF_PPI)
210 		atomic_add_long(&isrc->isrc_count[0], 1);
211 	else
212 		isrc->isrc_count[0]++;
213 }
214 
215 /*
216  *  Virtualization for interrupt source interrupt stray counter increment.
217  */
218 static inline void
219 isrc_increment_straycount(struct intr_irqsrc *isrc)
220 {
221 
222 	isrc->isrc_count[1]++;
223 }
224 
225 /*
226  *  Virtualization for interrupt source interrupt name update.
227  */
228 static void
229 isrc_update_name(struct intr_irqsrc *isrc, const char *name)
230 {
231 	char str[INTRNAME_LEN];
232 
233 	mtx_assert(&isrc_table_lock, MA_OWNED);
234 
235 	if (name != NULL) {
236 		snprintf(str, INTRNAME_LEN, "%s: %s", isrc->isrc_name, name);
237 		intrcnt_setname(str, isrc->isrc_index);
238 		snprintf(str, INTRNAME_LEN, "stray %s: %s", isrc->isrc_name,
239 		    name);
240 		intrcnt_setname(str, isrc->isrc_index + 1);
241 	} else {
242 		snprintf(str, INTRNAME_LEN, "%s:", isrc->isrc_name);
243 		intrcnt_setname(str, isrc->isrc_index);
244 		snprintf(str, INTRNAME_LEN, "stray %s:", isrc->isrc_name);
245 		intrcnt_setname(str, isrc->isrc_index + 1);
246 	}
247 }
248 
249 /*
250  *  Virtualization for interrupt source interrupt counters setup.
251  */
252 static void
253 isrc_setup_counters(struct intr_irqsrc *isrc)
254 {
255 	u_int index;
256 
257 	/*
258 	 *  XXX - it does not work well with removable controllers and
259 	 *        interrupt sources !!!
260 	 */
261 	index = atomic_fetchadd_int(&intrcnt_index, 2);
262 	isrc->isrc_index = index;
263 	isrc->isrc_count = &intrcnt[index];
264 	isrc_update_name(isrc, NULL);
265 }
266 
267 /*
268  *  Virtualization for interrupt source interrupt counters release.
269  */
270 static void
271 isrc_release_counters(struct intr_irqsrc *isrc)
272 {
273 
274 	panic("%s: not implemented", __func__);
275 }
276 
277 #ifdef SMP
278 /*
279  *  Virtualization for interrupt source IPI counters setup.
280  */
281 u_long *
282 intr_ipi_setup_counters(const char *name)
283 {
284 	u_int index, i;
285 	char str[INTRNAME_LEN];
286 
287 	index = atomic_fetchadd_int(&intrcnt_index, MAXCPU);
288 	for (i = 0; i < MAXCPU; i++) {
289 		snprintf(str, INTRNAME_LEN, "cpu%d:%s", i, name);
290 		intrcnt_setname(str, index + i);
291 	}
292 	return (&intrcnt[index]);
293 }
294 #endif
295 
296 /*
297  *  Main interrupt dispatch handler. It's called straight
298  *  from the assembler, where CPU interrupt is served.
299  */
300 void
301 intr_irq_handler(struct trapframe *tf)
302 {
303 	struct trapframe * oldframe;
304 	struct thread * td;
305 
306 	KASSERT(irq_root_filter != NULL, ("%s: no filter", __func__));
307 
308 	VM_CNT_INC(v_intr);
309 	critical_enter();
310 	td = curthread;
311 	oldframe = td->td_intr_frame;
312 	td->td_intr_frame = tf;
313 	irq_root_filter(irq_root_arg);
314 	td->td_intr_frame = oldframe;
315 	critical_exit();
316 #ifdef HWPMC_HOOKS
317 	if (pmc_hook && TRAPF_USERMODE(tf) &&
318 	    (PCPU_GET(curthread)->td_pflags & TDP_CALLCHAIN))
319 		pmc_hook(PCPU_GET(curthread), PMC_FN_USER_CALLCHAIN, tf);
320 #endif
321 }
322 
323 int
324 intr_child_irq_handler(struct intr_pic *parent, uintptr_t irq)
325 {
326 	struct intr_pic_child *child;
327 	bool found;
328 
329 	found = false;
330 	mtx_lock_spin(&parent->pic_child_lock);
331 	SLIST_FOREACH(child, &parent->pic_children, pc_next) {
332 		if (child->pc_start <= irq &&
333 		    irq < (child->pc_start + child->pc_length)) {
334 			found = true;
335 			break;
336 		}
337 	}
338 	mtx_unlock_spin(&parent->pic_child_lock);
339 
340 	if (found)
341 		return (child->pc_filter(child->pc_filter_arg, irq));
342 
343 	return (FILTER_STRAY);
344 }
345 
346 /*
347  *  interrupt controller dispatch function for interrupts. It should
348  *  be called straight from the interrupt controller, when associated interrupt
349  *  source is learned.
350  */
351 int
352 intr_isrc_dispatch(struct intr_irqsrc *isrc, struct trapframe *tf)
353 {
354 
355 	KASSERT(isrc != NULL, ("%s: no source", __func__));
356 
357 	isrc_increment_count(isrc);
358 
359 #ifdef INTR_SOLO
360 	if (isrc->isrc_filter != NULL) {
361 		int error;
362 		error = isrc->isrc_filter(isrc->isrc_arg, tf);
363 		PIC_POST_FILTER(isrc->isrc_dev, isrc);
364 		if (error == FILTER_HANDLED)
365 			return (0);
366 	} else
367 #endif
368 	if (isrc->isrc_event != NULL) {
369 		if (intr_event_handle(isrc->isrc_event, tf) == 0)
370 			return (0);
371 	}
372 
373 	isrc_increment_straycount(isrc);
374 	return (EINVAL);
375 }
376 
377 /*
378  *  Alloc unique interrupt number (resource handle) for interrupt source.
379  *
380  *  There could be various strategies how to allocate free interrupt number
381  *  (resource handle) for new interrupt source.
382  *
383  *  1. Handles are always allocated forward, so handles are not recycled
384  *     immediately. However, if only one free handle left which is reused
385  *     constantly...
386  */
387 static inline int
388 isrc_alloc_irq(struct intr_irqsrc *isrc)
389 {
390 	u_int maxirqs, irq;
391 
392 	mtx_assert(&isrc_table_lock, MA_OWNED);
393 
394 	maxirqs = nitems(irq_sources);
395 	if (irq_next_free >= maxirqs)
396 		return (ENOSPC);
397 
398 	for (irq = irq_next_free; irq < maxirqs; irq++) {
399 		if (irq_sources[irq] == NULL)
400 			goto found;
401 	}
402 	for (irq = 0; irq < irq_next_free; irq++) {
403 		if (irq_sources[irq] == NULL)
404 			goto found;
405 	}
406 
407 	irq_next_free = maxirqs;
408 	return (ENOSPC);
409 
410 found:
411 	isrc->isrc_irq = irq;
412 	irq_sources[irq] = isrc;
413 
414 	irq_next_free = irq + 1;
415 	if (irq_next_free >= maxirqs)
416 		irq_next_free = 0;
417 	return (0);
418 }
419 
420 /*
421  *  Free unique interrupt number (resource handle) from interrupt source.
422  */
423 static inline int
424 isrc_free_irq(struct intr_irqsrc *isrc)
425 {
426 
427 	mtx_assert(&isrc_table_lock, MA_OWNED);
428 
429 	if (isrc->isrc_irq >= nitems(irq_sources))
430 		return (EINVAL);
431 	if (irq_sources[isrc->isrc_irq] != isrc)
432 		return (EINVAL);
433 
434 	irq_sources[isrc->isrc_irq] = NULL;
435 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
436 	return (0);
437 }
438 
439 /*
440  *  Initialize interrupt source and register it into global interrupt table.
441  */
442 int
443 intr_isrc_register(struct intr_irqsrc *isrc, device_t dev, u_int flags,
444     const char *fmt, ...)
445 {
446 	int error;
447 	va_list ap;
448 
449 	bzero(isrc, sizeof(struct intr_irqsrc));
450 	isrc->isrc_dev = dev;
451 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
452 	isrc->isrc_flags = flags;
453 
454 	va_start(ap, fmt);
455 	vsnprintf(isrc->isrc_name, INTR_ISRC_NAMELEN, fmt, ap);
456 	va_end(ap);
457 
458 	mtx_lock(&isrc_table_lock);
459 	error = isrc_alloc_irq(isrc);
460 	if (error != 0) {
461 		mtx_unlock(&isrc_table_lock);
462 		return (error);
463 	}
464 	/*
465 	 * Setup interrupt counters, but not for IPI sources. Those are setup
466 	 * later and only for used ones (up to INTR_IPI_COUNT) to not exhaust
467 	 * our counter pool.
468 	 */
469 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
470 		isrc_setup_counters(isrc);
471 	mtx_unlock(&isrc_table_lock);
472 	return (0);
473 }
474 
475 /*
476  *  Deregister interrupt source from global interrupt table.
477  */
478 int
479 intr_isrc_deregister(struct intr_irqsrc *isrc)
480 {
481 	int error;
482 
483 	mtx_lock(&isrc_table_lock);
484 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
485 		isrc_release_counters(isrc);
486 	error = isrc_free_irq(isrc);
487 	mtx_unlock(&isrc_table_lock);
488 	return (error);
489 }
490 
491 #ifdef SMP
492 /*
493  *  A support function for a PIC to decide if provided ISRC should be inited
494  *  on given cpu. The logic of INTR_ISRCF_BOUND flag and isrc_cpu member of
495  *  struct intr_irqsrc is the following:
496  *
497  *     If INTR_ISRCF_BOUND is set, the ISRC should be inited only on cpus
498  *     set in isrc_cpu. If not, the ISRC should be inited on every cpu and
499  *     isrc_cpu is kept consistent with it. Thus isrc_cpu is always correct.
500  */
501 bool
502 intr_isrc_init_on_cpu(struct intr_irqsrc *isrc, u_int cpu)
503 {
504 
505 	if (isrc->isrc_handlers == 0)
506 		return (false);
507 	if ((isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI)) == 0)
508 		return (false);
509 	if (isrc->isrc_flags & INTR_ISRCF_BOUND)
510 		return (CPU_ISSET(cpu, &isrc->isrc_cpu));
511 
512 	CPU_SET(cpu, &isrc->isrc_cpu);
513 	return (true);
514 }
515 #endif
516 
517 #ifdef INTR_SOLO
518 /*
519  *  Setup filter into interrupt source.
520  */
521 static int
522 iscr_setup_filter(struct intr_irqsrc *isrc, const char *name,
523     intr_irq_filter_t *filter, void *arg, void **cookiep)
524 {
525 
526 	if (filter == NULL)
527 		return (EINVAL);
528 
529 	mtx_lock(&isrc_table_lock);
530 	/*
531 	 * Make sure that we do not mix the two ways
532 	 * how we handle interrupt sources.
533 	 */
534 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
535 		mtx_unlock(&isrc_table_lock);
536 		return (EBUSY);
537 	}
538 	isrc->isrc_filter = filter;
539 	isrc->isrc_arg = arg;
540 	isrc_update_name(isrc, name);
541 	mtx_unlock(&isrc_table_lock);
542 
543 	*cookiep = isrc;
544 	return (0);
545 }
546 #endif
547 
548 /*
549  *  Interrupt source pre_ithread method for MI interrupt framework.
550  */
551 static void
552 intr_isrc_pre_ithread(void *arg)
553 {
554 	struct intr_irqsrc *isrc = arg;
555 
556 	PIC_PRE_ITHREAD(isrc->isrc_dev, isrc);
557 }
558 
559 /*
560  *  Interrupt source post_ithread method for MI interrupt framework.
561  */
562 static void
563 intr_isrc_post_ithread(void *arg)
564 {
565 	struct intr_irqsrc *isrc = arg;
566 
567 	PIC_POST_ITHREAD(isrc->isrc_dev, isrc);
568 }
569 
570 /*
571  *  Interrupt source post_filter method for MI interrupt framework.
572  */
573 static void
574 intr_isrc_post_filter(void *arg)
575 {
576 	struct intr_irqsrc *isrc = arg;
577 
578 	PIC_POST_FILTER(isrc->isrc_dev, isrc);
579 }
580 
581 /*
582  *  Interrupt source assign_cpu method for MI interrupt framework.
583  */
584 static int
585 intr_isrc_assign_cpu(void *arg, int cpu)
586 {
587 #ifdef SMP
588 	struct intr_irqsrc *isrc = arg;
589 	int error;
590 
591 	mtx_lock(&isrc_table_lock);
592 	if (cpu == NOCPU) {
593 		CPU_ZERO(&isrc->isrc_cpu);
594 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
595 	} else {
596 		CPU_SETOF(cpu, &isrc->isrc_cpu);
597 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
598 	}
599 
600 	/*
601 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
602 	 * re-balance it to another CPU or enable it on more CPUs. However,
603 	 * PIC is expected to change isrc_cpu appropriately to keep us well
604 	 * informed if the call is successful.
605 	 */
606 	if (irq_assign_cpu) {
607 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
608 		if (error) {
609 			CPU_ZERO(&isrc->isrc_cpu);
610 			mtx_unlock(&isrc_table_lock);
611 			return (error);
612 		}
613 	}
614 	mtx_unlock(&isrc_table_lock);
615 	return (0);
616 #else
617 	return (EOPNOTSUPP);
618 #endif
619 }
620 
621 /*
622  *  Create interrupt event for interrupt source.
623  */
624 static int
625 isrc_event_create(struct intr_irqsrc *isrc)
626 {
627 	struct intr_event *ie;
628 	int error;
629 
630 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
631 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
632 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
633 	if (error)
634 		return (error);
635 
636 	mtx_lock(&isrc_table_lock);
637 	/*
638 	 * Make sure that we do not mix the two ways
639 	 * how we handle interrupt sources. Let contested event wins.
640 	 */
641 #ifdef INTR_SOLO
642 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
643 #else
644 	if (isrc->isrc_event != NULL) {
645 #endif
646 		mtx_unlock(&isrc_table_lock);
647 		intr_event_destroy(ie);
648 		return (isrc->isrc_event != NULL ? EBUSY : 0);
649 	}
650 	isrc->isrc_event = ie;
651 	mtx_unlock(&isrc_table_lock);
652 
653 	return (0);
654 }
655 #ifdef notyet
656 /*
657  *  Destroy interrupt event for interrupt source.
658  */
659 static void
660 isrc_event_destroy(struct intr_irqsrc *isrc)
661 {
662 	struct intr_event *ie;
663 
664 	mtx_lock(&isrc_table_lock);
665 	ie = isrc->isrc_event;
666 	isrc->isrc_event = NULL;
667 	mtx_unlock(&isrc_table_lock);
668 
669 	if (ie != NULL)
670 		intr_event_destroy(ie);
671 }
672 #endif
673 /*
674  *  Add handler to interrupt source.
675  */
676 static int
677 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
678     driver_filter_t filter, driver_intr_t handler, void *arg,
679     enum intr_type flags, void **cookiep)
680 {
681 	int error;
682 
683 	if (isrc->isrc_event == NULL) {
684 		error = isrc_event_create(isrc);
685 		if (error)
686 			return (error);
687 	}
688 
689 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
690 	    arg, intr_priority(flags), flags, cookiep);
691 	if (error == 0) {
692 		mtx_lock(&isrc_table_lock);
693 		intrcnt_updatename(isrc);
694 		mtx_unlock(&isrc_table_lock);
695 	}
696 
697 	return (error);
698 }
699 
700 /*
701  *  Lookup interrupt controller locked.
702  */
703 static inline struct intr_pic *
704 pic_lookup_locked(device_t dev, intptr_t xref, int flags)
705 {
706 	struct intr_pic *pic;
707 
708 	mtx_assert(&pic_list_lock, MA_OWNED);
709 
710 	if (dev == NULL && xref == 0)
711 		return (NULL);
712 
713 	/* Note that pic->pic_dev is never NULL on registered PIC. */
714 	SLIST_FOREACH(pic, &pic_list, pic_next) {
715 		if ((pic->pic_flags & FLAG_TYPE_MASK) !=
716 		    (flags & FLAG_TYPE_MASK))
717 			continue;
718 
719 		if (dev == NULL) {
720 			if (xref == pic->pic_xref)
721 				return (pic);
722 		} else if (xref == 0 || pic->pic_xref == 0) {
723 			if (dev == pic->pic_dev)
724 				return (pic);
725 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
726 				return (pic);
727 	}
728 	return (NULL);
729 }
730 
731 /*
732  *  Lookup interrupt controller.
733  */
734 static struct intr_pic *
735 pic_lookup(device_t dev, intptr_t xref, int flags)
736 {
737 	struct intr_pic *pic;
738 
739 	mtx_lock(&pic_list_lock);
740 	pic = pic_lookup_locked(dev, xref, flags);
741 	mtx_unlock(&pic_list_lock);
742 	return (pic);
743 }
744 
745 /*
746  *  Create interrupt controller.
747  */
748 static struct intr_pic *
749 pic_create(device_t dev, intptr_t xref, int flags)
750 {
751 	struct intr_pic *pic;
752 
753 	mtx_lock(&pic_list_lock);
754 	pic = pic_lookup_locked(dev, xref, flags);
755 	if (pic != NULL) {
756 		mtx_unlock(&pic_list_lock);
757 		return (pic);
758 	}
759 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
760 	if (pic == NULL) {
761 		mtx_unlock(&pic_list_lock);
762 		return (NULL);
763 	}
764 	pic->pic_xref = xref;
765 	pic->pic_dev = dev;
766 	pic->pic_flags = flags;
767 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
768 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
769 	mtx_unlock(&pic_list_lock);
770 
771 	return (pic);
772 }
773 #ifdef notyet
774 /*
775  *  Destroy interrupt controller.
776  */
777 static void
778 pic_destroy(device_t dev, intptr_t xref, int flags)
779 {
780 	struct intr_pic *pic;
781 
782 	mtx_lock(&pic_list_lock);
783 	pic = pic_lookup_locked(dev, xref, flags);
784 	if (pic == NULL) {
785 		mtx_unlock(&pic_list_lock);
786 		return;
787 	}
788 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
789 	mtx_unlock(&pic_list_lock);
790 
791 	free(pic, M_INTRNG);
792 }
793 #endif
794 /*
795  *  Register interrupt controller.
796  */
797 struct intr_pic *
798 intr_pic_register(device_t dev, intptr_t xref)
799 {
800 	struct intr_pic *pic;
801 
802 	if (dev == NULL)
803 		return (NULL);
804 	pic = pic_create(dev, xref, FLAG_PIC);
805 	if (pic == NULL)
806 		return (NULL);
807 
808 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
809 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
810 	return (pic);
811 }
812 
813 /*
814  *  Unregister interrupt controller.
815  */
816 int
817 intr_pic_deregister(device_t dev, intptr_t xref)
818 {
819 
820 	panic("%s: not implemented", __func__);
821 }
822 
823 /*
824  *  Mark interrupt controller (itself) as a root one.
825  *
826  *  Note that only an interrupt controller can really know its position
827  *  in interrupt controller's tree. So root PIC must claim itself as a root.
828  *
829  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
830  *  page 30:
831  *    "The root of the interrupt tree is determined when traversal
832  *     of the interrupt tree reaches an interrupt controller node without
833  *     an interrupts property and thus no explicit interrupt parent."
834  */
835 int
836 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
837     void *arg, u_int ipicount)
838 {
839 	struct intr_pic *pic;
840 
841 	pic = pic_lookup(dev, xref, FLAG_PIC);
842 	if (pic == NULL) {
843 		device_printf(dev, "not registered\n");
844 		return (EINVAL);
845 	}
846 
847 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
848 	    ("%s: Found a non-PIC controller: %s", __func__,
849 	     device_get_name(pic->pic_dev)));
850 
851 	if (filter == NULL) {
852 		device_printf(dev, "filter missing\n");
853 		return (EINVAL);
854 	}
855 
856 	/*
857 	 * Only one interrupt controllers could be on the root for now.
858 	 * Note that we further suppose that there is not threaded interrupt
859 	 * routine (handler) on the root. See intr_irq_handler().
860 	 */
861 	if (intr_irq_root_dev != NULL) {
862 		device_printf(dev, "another root already set\n");
863 		return (EBUSY);
864 	}
865 
866 	intr_irq_root_dev = dev;
867 	irq_root_filter = filter;
868 	irq_root_arg = arg;
869 	irq_root_ipicount = ipicount;
870 
871 	debugf("irq root set to %s\n", device_get_nameunit(dev));
872 	return (0);
873 }
874 
875 /*
876  * Add a handler to manage a sub range of a parents interrupts.
877  */
878 struct intr_pic *
879 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
880     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
881     uintptr_t length)
882 {
883 	struct intr_pic *parent_pic;
884 	struct intr_pic_child *newchild;
885 #ifdef INVARIANTS
886 	struct intr_pic_child *child;
887 #endif
888 
889 	/* Find the parent PIC */
890 	parent_pic = pic_lookup(parent, 0, FLAG_PIC);
891 	if (parent_pic == NULL)
892 		return (NULL);
893 
894 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
895 	newchild->pc_pic = pic;
896 	newchild->pc_filter = filter;
897 	newchild->pc_filter_arg = arg;
898 	newchild->pc_start = start;
899 	newchild->pc_length = length;
900 
901 	mtx_lock_spin(&parent_pic->pic_child_lock);
902 #ifdef INVARIANTS
903 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
904 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
905 		    __func__));
906 	}
907 #endif
908 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
909 	mtx_unlock_spin(&parent_pic->pic_child_lock);
910 
911 	return (pic);
912 }
913 
914 static int
915 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
916     struct intr_irqsrc **isrc)
917 {
918 	struct intr_pic *pic;
919 	struct intr_map_data_msi *msi;
920 
921 	if (data == NULL)
922 		return (EINVAL);
923 
924 	pic = pic_lookup(dev, xref,
925 	    (data->type == INTR_MAP_DATA_MSI) ? FLAG_MSI : FLAG_PIC);
926 	if (pic == NULL)
927 		return (ESRCH);
928 
929 	switch (data->type) {
930 	case INTR_MAP_DATA_MSI:
931 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
932 		    ("%s: Found a non-MSI controller: %s", __func__,
933 		     device_get_name(pic->pic_dev)));
934 		msi = (struct intr_map_data_msi *)data;
935 		*isrc = msi->isrc;
936 		return (0);
937 
938 	default:
939 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
940 		    ("%s: Found a non-PIC controller: %s", __func__,
941 		     device_get_name(pic->pic_dev)));
942 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
943 	}
944 }
945 
946 bool
947 intr_is_per_cpu(struct resource *res)
948 {
949 	u_int res_id;
950 	struct intr_irqsrc *isrc;
951 
952 	res_id = (u_int)rman_get_start(res);
953 	isrc = intr_map_get_isrc(res_id);
954 
955 	if (isrc == NULL)
956 		panic("Attempt to get isrc for non-active resource id: %u\n",
957 		    res_id);
958 	return ((isrc->isrc_flags & INTR_ISRCF_PPI) != 0);
959 }
960 
961 int
962 intr_activate_irq(device_t dev, struct resource *res)
963 {
964 	device_t map_dev;
965 	intptr_t map_xref;
966 	struct intr_map_data *data;
967 	struct intr_irqsrc *isrc;
968 	u_int res_id;
969 	int error;
970 
971 	KASSERT(rman_get_start(res) == rman_get_end(res),
972 	    ("%s: more interrupts in resource", __func__));
973 
974 	res_id = (u_int)rman_get_start(res);
975 	if (intr_map_get_isrc(res_id) != NULL)
976 		panic("Attempt to double activation of resource id: %u\n",
977 		    res_id);
978 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
979 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
980 	if (error != 0) {
981 		free(data, M_INTRNG);
982 		/* XXX TODO DISCONECTED PICs */
983 		/* if (error == EINVAL) return(0); */
984 		return (error);
985 	}
986 	intr_map_set_isrc(res_id, isrc);
987 	rman_set_virtual(res, data);
988 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
989 }
990 
991 int
992 intr_deactivate_irq(device_t dev, struct resource *res)
993 {
994 	struct intr_map_data *data;
995 	struct intr_irqsrc *isrc;
996 	u_int res_id;
997 	int error;
998 
999 	KASSERT(rman_get_start(res) == rman_get_end(res),
1000 	    ("%s: more interrupts in resource", __func__));
1001 
1002 	res_id = (u_int)rman_get_start(res);
1003 	isrc = intr_map_get_isrc(res_id);
1004 	if (isrc == NULL)
1005 		panic("Attempt to deactivate non-active resource id: %u\n",
1006 		    res_id);
1007 
1008 	data = rman_get_virtual(res);
1009 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
1010 	intr_map_set_isrc(res_id, NULL);
1011 	rman_set_virtual(res, NULL);
1012 	free(data, M_INTRNG);
1013 	return (error);
1014 }
1015 
1016 int
1017 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
1018     driver_intr_t hand, void *arg, int flags, void **cookiep)
1019 {
1020 	int error;
1021 	struct intr_map_data *data;
1022 	struct intr_irqsrc *isrc;
1023 	const char *name;
1024 	u_int res_id;
1025 
1026 	KASSERT(rman_get_start(res) == rman_get_end(res),
1027 	    ("%s: more interrupts in resource", __func__));
1028 
1029 	res_id = (u_int)rman_get_start(res);
1030 	isrc = intr_map_get_isrc(res_id);
1031 	if (isrc == NULL) {
1032 		/* XXX TODO DISCONECTED PICs */
1033 		return (EINVAL);
1034 	}
1035 
1036 	data = rman_get_virtual(res);
1037 	name = device_get_nameunit(dev);
1038 
1039 #ifdef INTR_SOLO
1040 	/*
1041 	 * Standard handling is done through MI interrupt framework. However,
1042 	 * some interrupts could request solely own special handling. This
1043 	 * non standard handling can be used for interrupt controllers without
1044 	 * handler (filter only), so in case that interrupt controllers are
1045 	 * chained, MI interrupt framework is called only in leaf controller.
1046 	 *
1047 	 * Note that root interrupt controller routine is served as well,
1048 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1049 	 */
1050 	if (flags & INTR_SOLO && hand != NULL) {
1051 		debugf("irq %u cannot solo on %s\n", irq, name);
1052 		return (EINVAL);
1053 	}
1054 
1055 	if (flags & INTR_SOLO) {
1056 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1057 		    arg, cookiep);
1058 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1059 		    name);
1060 	} else
1061 #endif
1062 		{
1063 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1064 		    cookiep);
1065 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1066 	}
1067 	if (error != 0)
1068 		return (error);
1069 
1070 	mtx_lock(&isrc_table_lock);
1071 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1072 	if (error == 0) {
1073 		isrc->isrc_handlers++;
1074 		if (isrc->isrc_handlers == 1)
1075 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1076 	}
1077 	mtx_unlock(&isrc_table_lock);
1078 	if (error != 0)
1079 		intr_event_remove_handler(*cookiep);
1080 	return (error);
1081 }
1082 
1083 int
1084 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1085 {
1086 	int error;
1087 	struct intr_map_data *data;
1088 	struct intr_irqsrc *isrc;
1089 	u_int res_id;
1090 
1091 	KASSERT(rman_get_start(res) == rman_get_end(res),
1092 	    ("%s: more interrupts in resource", __func__));
1093 
1094 	res_id = (u_int)rman_get_start(res);
1095 	isrc = intr_map_get_isrc(res_id);
1096 	if (isrc == NULL || isrc->isrc_handlers == 0)
1097 		return (EINVAL);
1098 
1099 	data = rman_get_virtual(res);
1100 
1101 #ifdef INTR_SOLO
1102 	if (isrc->isrc_filter != NULL) {
1103 		if (isrc != cookie)
1104 			return (EINVAL);
1105 
1106 		mtx_lock(&isrc_table_lock);
1107 		isrc->isrc_filter = NULL;
1108 		isrc->isrc_arg = NULL;
1109 		isrc->isrc_handlers = 0;
1110 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1111 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1112 		isrc_update_name(isrc, NULL);
1113 		mtx_unlock(&isrc_table_lock);
1114 		return (0);
1115 	}
1116 #endif
1117 	if (isrc != intr_handler_source(cookie))
1118 		return (EINVAL);
1119 
1120 	error = intr_event_remove_handler(cookie);
1121 	if (error == 0) {
1122 		mtx_lock(&isrc_table_lock);
1123 		isrc->isrc_handlers--;
1124 		if (isrc->isrc_handlers == 0)
1125 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1126 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1127 		intrcnt_updatename(isrc);
1128 		mtx_unlock(&isrc_table_lock);
1129 	}
1130 	return (error);
1131 }
1132 
1133 int
1134 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1135     const char *descr)
1136 {
1137 	int error;
1138 	struct intr_irqsrc *isrc;
1139 	u_int res_id;
1140 
1141 	KASSERT(rman_get_start(res) == rman_get_end(res),
1142 	    ("%s: more interrupts in resource", __func__));
1143 
1144 	res_id = (u_int)rman_get_start(res);
1145 	isrc = intr_map_get_isrc(res_id);
1146 	if (isrc == NULL || isrc->isrc_handlers == 0)
1147 		return (EINVAL);
1148 #ifdef INTR_SOLO
1149 	if (isrc->isrc_filter != NULL) {
1150 		if (isrc != cookie)
1151 			return (EINVAL);
1152 
1153 		mtx_lock(&isrc_table_lock);
1154 		isrc_update_name(isrc, descr);
1155 		mtx_unlock(&isrc_table_lock);
1156 		return (0);
1157 	}
1158 #endif
1159 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1160 	if (error == 0) {
1161 		mtx_lock(&isrc_table_lock);
1162 		intrcnt_updatename(isrc);
1163 		mtx_unlock(&isrc_table_lock);
1164 	}
1165 	return (error);
1166 }
1167 
1168 #ifdef SMP
1169 int
1170 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1171 {
1172 	struct intr_irqsrc *isrc;
1173 	u_int res_id;
1174 
1175 	KASSERT(rman_get_start(res) == rman_get_end(res),
1176 	    ("%s: more interrupts in resource", __func__));
1177 
1178 	res_id = (u_int)rman_get_start(res);
1179 	isrc = intr_map_get_isrc(res_id);
1180 	if (isrc == NULL || isrc->isrc_handlers == 0)
1181 		return (EINVAL);
1182 #ifdef INTR_SOLO
1183 	if (isrc->isrc_filter != NULL)
1184 		return (intr_isrc_assign_cpu(isrc, cpu));
1185 #endif
1186 	return (intr_event_bind(isrc->isrc_event, cpu));
1187 }
1188 
1189 /*
1190  * Return the CPU that the next interrupt source should use.
1191  * For now just returns the next CPU according to round-robin.
1192  */
1193 u_int
1194 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1195 {
1196 	u_int cpu;
1197 
1198 	KASSERT(!CPU_EMPTY(cpumask), ("%s: Empty CPU mask", __func__));
1199 	if (!irq_assign_cpu || mp_ncpus == 1) {
1200 		cpu = PCPU_GET(cpuid);
1201 
1202 		if (CPU_ISSET(cpu, cpumask))
1203 			return (curcpu);
1204 
1205 		return (CPU_FFS(cpumask) - 1);
1206 	}
1207 
1208 	do {
1209 		last_cpu++;
1210 		if (last_cpu > mp_maxid)
1211 			last_cpu = 0;
1212 	} while (!CPU_ISSET(last_cpu, cpumask));
1213 	return (last_cpu);
1214 }
1215 
1216 #ifndef EARLY_AP_STARTUP
1217 /*
1218  *  Distribute all the interrupt sources among the available
1219  *  CPUs once the AP's have been launched.
1220  */
1221 static void
1222 intr_irq_shuffle(void *arg __unused)
1223 {
1224 	struct intr_irqsrc *isrc;
1225 	u_int i;
1226 
1227 	if (mp_ncpus == 1)
1228 		return;
1229 
1230 	mtx_lock(&isrc_table_lock);
1231 	irq_assign_cpu = true;
1232 	for (i = 0; i < NIRQ; i++) {
1233 		isrc = irq_sources[i];
1234 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1235 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1236 			continue;
1237 
1238 		if (isrc->isrc_event != NULL &&
1239 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1240 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1241 			panic("%s: CPU inconsistency", __func__);
1242 
1243 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1244 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1245 
1246 		/*
1247 		 * We are in wicked position here if the following call fails
1248 		 * for bound ISRC. The best thing we can do is to clear
1249 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1250 		 */
1251 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1252 			CPU_ZERO(&isrc->isrc_cpu);
1253 	}
1254 	mtx_unlock(&isrc_table_lock);
1255 }
1256 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1257 #endif /* !EARLY_AP_STARTUP */
1258 
1259 #else
1260 u_int
1261 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1262 {
1263 
1264 	return (PCPU_GET(cpuid));
1265 }
1266 #endif /* SMP */
1267 
1268 /*
1269  * Allocate memory for new intr_map_data structure.
1270  * Initialize common fields.
1271  */
1272 struct intr_map_data *
1273 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1274 {
1275 	struct intr_map_data *data;
1276 
1277 	data = malloc(len, M_INTRNG, flags);
1278 	data->type = type;
1279 	data->len = len;
1280 	return (data);
1281 }
1282 
1283 void intr_free_intr_map_data(struct intr_map_data *data)
1284 {
1285 
1286 	free(data, M_INTRNG);
1287 }
1288 
1289 /*
1290  *  Register a MSI/MSI-X interrupt controller
1291  */
1292 int
1293 intr_msi_register(device_t dev, intptr_t xref)
1294 {
1295 	struct intr_pic *pic;
1296 
1297 	if (dev == NULL)
1298 		return (EINVAL);
1299 	pic = pic_create(dev, xref, FLAG_MSI);
1300 	if (pic == NULL)
1301 		return (ENOMEM);
1302 
1303 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1304 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1305 	return (0);
1306 }
1307 
1308 int
1309 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1310     int maxcount, int *irqs)
1311 {
1312 	struct iommu_domain *domain;
1313 	struct intr_irqsrc **isrc;
1314 	struct intr_pic *pic;
1315 	device_t pdev;
1316 	struct intr_map_data_msi *msi;
1317 	int err, i;
1318 
1319 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1320 	if (pic == NULL)
1321 		return (ESRCH);
1322 
1323 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1324 	    ("%s: Found a non-MSI controller: %s", __func__,
1325 	     device_get_name(pic->pic_dev)));
1326 
1327 	/*
1328 	 * If this is the first time we have used this context ask the
1329 	 * interrupt controller to map memory the msi source will need.
1330 	 */
1331 	err = MSI_IOMMU_INIT(pic->pic_dev, child, &domain);
1332 	if (err != 0)
1333 		return (err);
1334 
1335 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1336 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1337 	if (err != 0) {
1338 		free(isrc, M_INTRNG);
1339 		return (err);
1340 	}
1341 
1342 	for (i = 0; i < count; i++) {
1343 		isrc[i]->isrc_iommu = domain;
1344 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1345 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1346 		msi-> isrc = isrc[i];
1347 
1348 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1349 		    (struct intr_map_data *)msi);
1350 	}
1351 	free(isrc, M_INTRNG);
1352 
1353 	return (err);
1354 }
1355 
1356 int
1357 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1358     int *irqs)
1359 {
1360 	struct intr_irqsrc **isrc;
1361 	struct intr_pic *pic;
1362 	struct intr_map_data_msi *msi;
1363 	int i, err;
1364 
1365 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1366 	if (pic == NULL)
1367 		return (ESRCH);
1368 
1369 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1370 	    ("%s: Found a non-MSI controller: %s", __func__,
1371 	     device_get_name(pic->pic_dev)));
1372 
1373 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1374 
1375 	for (i = 0; i < count; i++) {
1376 		msi = (struct intr_map_data_msi *)
1377 		    intr_map_get_map_data(irqs[i]);
1378 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1379 		    ("%s: irq %d map data is not MSI", __func__,
1380 		    irqs[i]));
1381 		isrc[i] = msi->isrc;
1382 	}
1383 
1384 	MSI_IOMMU_DEINIT(pic->pic_dev, child);
1385 
1386 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1387 
1388 	for (i = 0; i < count; i++) {
1389 		if (isrc[i] != NULL)
1390 			intr_unmap_irq(irqs[i]);
1391 	}
1392 
1393 	free(isrc, M_INTRNG);
1394 	return (err);
1395 }
1396 
1397 int
1398 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1399 {
1400 	struct iommu_domain *domain;
1401 	struct intr_irqsrc *isrc;
1402 	struct intr_pic *pic;
1403 	device_t pdev;
1404 	struct intr_map_data_msi *msi;
1405 	int err;
1406 
1407 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1408 	if (pic == NULL)
1409 		return (ESRCH);
1410 
1411 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1412 	    ("%s: Found a non-MSI controller: %s", __func__,
1413 	     device_get_name(pic->pic_dev)));
1414 
1415 	/*
1416 	 * If this is the first time we have used this context ask the
1417 	 * interrupt controller to map memory the msi source will need.
1418 	 */
1419 	err = MSI_IOMMU_INIT(pic->pic_dev, child, &domain);
1420 	if (err != 0)
1421 		return (err);
1422 
1423 	err = MSI_ALLOC_MSIX(pic->pic_dev, child, &pdev, &isrc);
1424 	if (err != 0)
1425 		return (err);
1426 
1427 	isrc->isrc_iommu = domain;
1428 	msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1429 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1430 	msi->isrc = isrc;
1431 	*irq = intr_map_irq(pic->pic_dev, xref, (struct intr_map_data *)msi);
1432 	return (0);
1433 }
1434 
1435 int
1436 intr_release_msix(device_t pci, device_t child, intptr_t xref, int irq)
1437 {
1438 	struct intr_irqsrc *isrc;
1439 	struct intr_pic *pic;
1440 	struct intr_map_data_msi *msi;
1441 	int err;
1442 
1443 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1444 	if (pic == NULL)
1445 		return (ESRCH);
1446 
1447 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1448 	    ("%s: Found a non-MSI controller: %s", __func__,
1449 	     device_get_name(pic->pic_dev)));
1450 
1451 	msi = (struct intr_map_data_msi *)
1452 	    intr_map_get_map_data(irq);
1453 	KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1454 	    ("%s: irq %d map data is not MSI", __func__,
1455 	    irq));
1456 	isrc = msi->isrc;
1457 	if (isrc == NULL) {
1458 		intr_unmap_irq(irq);
1459 		return (EINVAL);
1460 	}
1461 
1462 	MSI_IOMMU_DEINIT(pic->pic_dev, child);
1463 
1464 	err = MSI_RELEASE_MSIX(pic->pic_dev, child, isrc);
1465 	intr_unmap_irq(irq);
1466 
1467 	return (err);
1468 }
1469 
1470 int
1471 intr_map_msi(device_t pci, device_t child, intptr_t xref, int irq,
1472     uint64_t *addr, uint32_t *data)
1473 {
1474 	struct intr_irqsrc *isrc;
1475 	struct intr_pic *pic;
1476 	int err;
1477 
1478 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1479 	if (pic == NULL)
1480 		return (ESRCH);
1481 
1482 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1483 	    ("%s: Found a non-MSI controller: %s", __func__,
1484 	     device_get_name(pic->pic_dev)));
1485 
1486 	isrc = intr_map_get_isrc(irq);
1487 	if (isrc == NULL)
1488 		return (EINVAL);
1489 
1490 	err = MSI_MAP_MSI(pic->pic_dev, child, isrc, addr, data);
1491 
1492 #ifdef IOMMU
1493 	if (isrc->isrc_iommu != NULL)
1494 		iommu_translate_msi(isrc->isrc_iommu, addr);
1495 #endif
1496 
1497 	return (err);
1498 }
1499 
1500 void dosoftints(void);
1501 void
1502 dosoftints(void)
1503 {
1504 }
1505 
1506 #ifdef SMP
1507 /*
1508  *  Init interrupt controller on another CPU.
1509  */
1510 void
1511 intr_pic_init_secondary(void)
1512 {
1513 
1514 	/*
1515 	 * QQQ: Only root PIC is aware of other CPUs ???
1516 	 */
1517 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1518 
1519 	//mtx_lock(&isrc_table_lock);
1520 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1521 	//mtx_unlock(&isrc_table_lock);
1522 }
1523 #endif
1524 
1525 #ifdef DDB
1526 DB_SHOW_COMMAND(irqs, db_show_irqs)
1527 {
1528 	u_int i, irqsum;
1529 	u_long num;
1530 	struct intr_irqsrc *isrc;
1531 
1532 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1533 		isrc = irq_sources[i];
1534 		if (isrc == NULL)
1535 			continue;
1536 
1537 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1538 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1539 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1540 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1541 		irqsum += num;
1542 	}
1543 	db_printf("irq total %u\n", irqsum);
1544 }
1545 #endif
1546 
1547 /*
1548  * Interrupt mapping table functions.
1549  *
1550  * Please, keep this part separately, it can be transformed to
1551  * extension of standard resources.
1552  */
1553 struct intr_map_entry
1554 {
1555 	device_t 		dev;
1556 	intptr_t 		xref;
1557 	struct intr_map_data 	*map_data;
1558 	struct intr_irqsrc 	*isrc;
1559 	/* XXX TODO DISCONECTED PICs */
1560 	/*int			flags */
1561 };
1562 
1563 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1564 static struct intr_map_entry *irq_map[2 * NIRQ];
1565 static int irq_map_count = nitems(irq_map);
1566 static int irq_map_first_free_idx;
1567 static struct mtx irq_map_lock;
1568 
1569 static struct intr_irqsrc *
1570 intr_map_get_isrc(u_int res_id)
1571 {
1572 	struct intr_irqsrc *isrc;
1573 
1574 	isrc = NULL;
1575 	mtx_lock(&irq_map_lock);
1576 	if (res_id < irq_map_count && irq_map[res_id] != NULL)
1577 		isrc = irq_map[res_id]->isrc;
1578 	mtx_unlock(&irq_map_lock);
1579 
1580 	return (isrc);
1581 }
1582 
1583 static void
1584 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1585 {
1586 
1587 	mtx_lock(&irq_map_lock);
1588 	if (res_id < irq_map_count && irq_map[res_id] != NULL)
1589 		irq_map[res_id]->isrc = isrc;
1590 	mtx_unlock(&irq_map_lock);
1591 }
1592 
1593 /*
1594  * Get a copy of intr_map_entry data
1595  */
1596 static struct intr_map_data *
1597 intr_map_get_map_data(u_int res_id)
1598 {
1599 	struct intr_map_data *data;
1600 
1601 	data = NULL;
1602 	mtx_lock(&irq_map_lock);
1603 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1604 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1605 	data = irq_map[res_id]->map_data;
1606 	mtx_unlock(&irq_map_lock);
1607 
1608 	return (data);
1609 }
1610 
1611 /*
1612  * Get a copy of intr_map_entry data
1613  */
1614 static void
1615 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1616     struct intr_map_data **data)
1617 {
1618 	size_t len;
1619 
1620 	len = 0;
1621 	mtx_lock(&irq_map_lock);
1622 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1623 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1624 	if (irq_map[res_id]->map_data != NULL)
1625 		len = irq_map[res_id]->map_data->len;
1626 	mtx_unlock(&irq_map_lock);
1627 
1628 	if (len == 0)
1629 		*data = NULL;
1630 	else
1631 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1632 	mtx_lock(&irq_map_lock);
1633 	if (irq_map[res_id] == NULL)
1634 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1635 	if (len != 0) {
1636 		if (len != irq_map[res_id]->map_data->len)
1637 			panic("Resource id: %u has changed.\n", res_id);
1638 		memcpy(*data, irq_map[res_id]->map_data, len);
1639 	}
1640 	*map_dev = irq_map[res_id]->dev;
1641 	*map_xref = irq_map[res_id]->xref;
1642 	mtx_unlock(&irq_map_lock);
1643 }
1644 
1645 /*
1646  * Allocate and fill new entry in irq_map table.
1647  */
1648 u_int
1649 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1650 {
1651 	u_int i;
1652 	struct intr_map_entry *entry;
1653 
1654 	/* Prepare new entry first. */
1655 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1656 
1657 	entry->dev = dev;
1658 	entry->xref = xref;
1659 	entry->map_data = data;
1660 	entry->isrc = NULL;
1661 
1662 	mtx_lock(&irq_map_lock);
1663 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1664 		if (irq_map[i] == NULL) {
1665 			irq_map[i] = entry;
1666 			irq_map_first_free_idx = i + 1;
1667 			mtx_unlock(&irq_map_lock);
1668 			return (i);
1669 		}
1670 	}
1671 	mtx_unlock(&irq_map_lock);
1672 
1673 	/* XXX Expand irq_map table */
1674 	panic("IRQ mapping table is full.");
1675 }
1676 
1677 /*
1678  * Remove and free mapping entry.
1679  */
1680 void
1681 intr_unmap_irq(u_int res_id)
1682 {
1683 	struct intr_map_entry *entry;
1684 
1685 	mtx_lock(&irq_map_lock);
1686 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1687 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1688 	entry = irq_map[res_id];
1689 	irq_map[res_id] = NULL;
1690 	irq_map_first_free_idx = res_id;
1691 	mtx_unlock(&irq_map_lock);
1692 	intr_free_intr_map_data(entry->map_data);
1693 	free(entry, M_INTRNG);
1694 }
1695 
1696 /*
1697  * Clone mapping entry.
1698  */
1699 u_int
1700 intr_map_clone_irq(u_int old_res_id)
1701 {
1702 	device_t map_dev;
1703 	intptr_t map_xref;
1704 	struct intr_map_data *data;
1705 
1706 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1707 	return (intr_map_irq(map_dev, map_xref, data));
1708 }
1709 
1710 static void
1711 intr_map_init(void *dummy __unused)
1712 {
1713 
1714 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1715 }
1716 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1717