xref: /freebsd/sys/kern/subr_intr.c (revision e707c8be4e8d5f4d542a1344dc6d96bcc08d0b4f)
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 	if (isrc->isrc_dev != intr_irq_root_dev)
592 		return (EINVAL);
593 
594 	mtx_lock(&isrc_table_lock);
595 	if (cpu == NOCPU) {
596 		CPU_ZERO(&isrc->isrc_cpu);
597 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
598 	} else {
599 		CPU_SETOF(cpu, &isrc->isrc_cpu);
600 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
601 	}
602 
603 	/*
604 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
605 	 * re-balance it to another CPU or enable it on more CPUs. However,
606 	 * PIC is expected to change isrc_cpu appropriately to keep us well
607 	 * informed if the call is successful.
608 	 */
609 	if (irq_assign_cpu) {
610 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
611 		if (error) {
612 			CPU_ZERO(&isrc->isrc_cpu);
613 			mtx_unlock(&isrc_table_lock);
614 			return (error);
615 		}
616 	}
617 	mtx_unlock(&isrc_table_lock);
618 	return (0);
619 #else
620 	return (EOPNOTSUPP);
621 #endif
622 }
623 
624 /*
625  *  Create interrupt event for interrupt source.
626  */
627 static int
628 isrc_event_create(struct intr_irqsrc *isrc)
629 {
630 	struct intr_event *ie;
631 	int error;
632 
633 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
634 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
635 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
636 	if (error)
637 		return (error);
638 
639 	mtx_lock(&isrc_table_lock);
640 	/*
641 	 * Make sure that we do not mix the two ways
642 	 * how we handle interrupt sources. Let contested event wins.
643 	 */
644 #ifdef INTR_SOLO
645 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
646 #else
647 	if (isrc->isrc_event != NULL) {
648 #endif
649 		mtx_unlock(&isrc_table_lock);
650 		intr_event_destroy(ie);
651 		return (isrc->isrc_event != NULL ? EBUSY : 0);
652 	}
653 	isrc->isrc_event = ie;
654 	mtx_unlock(&isrc_table_lock);
655 
656 	return (0);
657 }
658 #ifdef notyet
659 /*
660  *  Destroy interrupt event for interrupt source.
661  */
662 static void
663 isrc_event_destroy(struct intr_irqsrc *isrc)
664 {
665 	struct intr_event *ie;
666 
667 	mtx_lock(&isrc_table_lock);
668 	ie = isrc->isrc_event;
669 	isrc->isrc_event = NULL;
670 	mtx_unlock(&isrc_table_lock);
671 
672 	if (ie != NULL)
673 		intr_event_destroy(ie);
674 }
675 #endif
676 /*
677  *  Add handler to interrupt source.
678  */
679 static int
680 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
681     driver_filter_t filter, driver_intr_t handler, void *arg,
682     enum intr_type flags, void **cookiep)
683 {
684 	int error;
685 
686 	if (isrc->isrc_event == NULL) {
687 		error = isrc_event_create(isrc);
688 		if (error)
689 			return (error);
690 	}
691 
692 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
693 	    arg, intr_priority(flags), flags, cookiep);
694 	if (error == 0) {
695 		mtx_lock(&isrc_table_lock);
696 		intrcnt_updatename(isrc);
697 		mtx_unlock(&isrc_table_lock);
698 	}
699 
700 	return (error);
701 }
702 
703 /*
704  *  Lookup interrupt controller locked.
705  */
706 static inline struct intr_pic *
707 pic_lookup_locked(device_t dev, intptr_t xref, int flags)
708 {
709 	struct intr_pic *pic;
710 
711 	mtx_assert(&pic_list_lock, MA_OWNED);
712 
713 	if (dev == NULL && xref == 0)
714 		return (NULL);
715 
716 	/* Note that pic->pic_dev is never NULL on registered PIC. */
717 	SLIST_FOREACH(pic, &pic_list, pic_next) {
718 		if ((pic->pic_flags & FLAG_TYPE_MASK) !=
719 		    (flags & FLAG_TYPE_MASK))
720 			continue;
721 
722 		if (dev == NULL) {
723 			if (xref == pic->pic_xref)
724 				return (pic);
725 		} else if (xref == 0 || pic->pic_xref == 0) {
726 			if (dev == pic->pic_dev)
727 				return (pic);
728 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
729 				return (pic);
730 	}
731 	return (NULL);
732 }
733 
734 /*
735  *  Lookup interrupt controller.
736  */
737 static struct intr_pic *
738 pic_lookup(device_t dev, intptr_t xref, int flags)
739 {
740 	struct intr_pic *pic;
741 
742 	mtx_lock(&pic_list_lock);
743 	pic = pic_lookup_locked(dev, xref, flags);
744 	mtx_unlock(&pic_list_lock);
745 	return (pic);
746 }
747 
748 /*
749  *  Create interrupt controller.
750  */
751 static struct intr_pic *
752 pic_create(device_t dev, intptr_t xref, int flags)
753 {
754 	struct intr_pic *pic;
755 
756 	mtx_lock(&pic_list_lock);
757 	pic = pic_lookup_locked(dev, xref, flags);
758 	if (pic != NULL) {
759 		mtx_unlock(&pic_list_lock);
760 		return (pic);
761 	}
762 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
763 	if (pic == NULL) {
764 		mtx_unlock(&pic_list_lock);
765 		return (NULL);
766 	}
767 	pic->pic_xref = xref;
768 	pic->pic_dev = dev;
769 	pic->pic_flags = flags;
770 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
771 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
772 	mtx_unlock(&pic_list_lock);
773 
774 	return (pic);
775 }
776 #ifdef notyet
777 /*
778  *  Destroy interrupt controller.
779  */
780 static void
781 pic_destroy(device_t dev, intptr_t xref, int flags)
782 {
783 	struct intr_pic *pic;
784 
785 	mtx_lock(&pic_list_lock);
786 	pic = pic_lookup_locked(dev, xref, flags);
787 	if (pic == NULL) {
788 		mtx_unlock(&pic_list_lock);
789 		return;
790 	}
791 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
792 	mtx_unlock(&pic_list_lock);
793 
794 	free(pic, M_INTRNG);
795 }
796 #endif
797 /*
798  *  Register interrupt controller.
799  */
800 struct intr_pic *
801 intr_pic_register(device_t dev, intptr_t xref)
802 {
803 	struct intr_pic *pic;
804 
805 	if (dev == NULL)
806 		return (NULL);
807 	pic = pic_create(dev, xref, FLAG_PIC);
808 	if (pic == NULL)
809 		return (NULL);
810 
811 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
812 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
813 	return (pic);
814 }
815 
816 /*
817  *  Unregister interrupt controller.
818  */
819 int
820 intr_pic_deregister(device_t dev, intptr_t xref)
821 {
822 
823 	panic("%s: not implemented", __func__);
824 }
825 
826 /*
827  *  Mark interrupt controller (itself) as a root one.
828  *
829  *  Note that only an interrupt controller can really know its position
830  *  in interrupt controller's tree. So root PIC must claim itself as a root.
831  *
832  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
833  *  page 30:
834  *    "The root of the interrupt tree is determined when traversal
835  *     of the interrupt tree reaches an interrupt controller node without
836  *     an interrupts property and thus no explicit interrupt parent."
837  */
838 int
839 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
840     void *arg, u_int ipicount)
841 {
842 	struct intr_pic *pic;
843 
844 	pic = pic_lookup(dev, xref, FLAG_PIC);
845 	if (pic == NULL) {
846 		device_printf(dev, "not registered\n");
847 		return (EINVAL);
848 	}
849 
850 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
851 	    ("%s: Found a non-PIC controller: %s", __func__,
852 	     device_get_name(pic->pic_dev)));
853 
854 	if (filter == NULL) {
855 		device_printf(dev, "filter missing\n");
856 		return (EINVAL);
857 	}
858 
859 	/*
860 	 * Only one interrupt controllers could be on the root for now.
861 	 * Note that we further suppose that there is not threaded interrupt
862 	 * routine (handler) on the root. See intr_irq_handler().
863 	 */
864 	if (intr_irq_root_dev != NULL) {
865 		device_printf(dev, "another root already set\n");
866 		return (EBUSY);
867 	}
868 
869 	intr_irq_root_dev = dev;
870 	irq_root_filter = filter;
871 	irq_root_arg = arg;
872 	irq_root_ipicount = ipicount;
873 
874 	debugf("irq root set to %s\n", device_get_nameunit(dev));
875 	return (0);
876 }
877 
878 /*
879  * Add a handler to manage a sub range of a parents interrupts.
880  */
881 struct intr_pic *
882 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
883     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
884     uintptr_t length)
885 {
886 	struct intr_pic *parent_pic;
887 	struct intr_pic_child *newchild;
888 #ifdef INVARIANTS
889 	struct intr_pic_child *child;
890 #endif
891 
892 	/* Find the parent PIC */
893 	parent_pic = pic_lookup(parent, 0, FLAG_PIC);
894 	if (parent_pic == NULL)
895 		return (NULL);
896 
897 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
898 	newchild->pc_pic = pic;
899 	newchild->pc_filter = filter;
900 	newchild->pc_filter_arg = arg;
901 	newchild->pc_start = start;
902 	newchild->pc_length = length;
903 
904 	mtx_lock_spin(&parent_pic->pic_child_lock);
905 #ifdef INVARIANTS
906 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
907 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
908 		    __func__));
909 	}
910 #endif
911 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
912 	mtx_unlock_spin(&parent_pic->pic_child_lock);
913 
914 	return (pic);
915 }
916 
917 static int
918 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
919     struct intr_irqsrc **isrc)
920 {
921 	struct intr_pic *pic;
922 	struct intr_map_data_msi *msi;
923 
924 	if (data == NULL)
925 		return (EINVAL);
926 
927 	pic = pic_lookup(dev, xref,
928 	    (data->type == INTR_MAP_DATA_MSI) ? FLAG_MSI : FLAG_PIC);
929 	if (pic == NULL)
930 		return (ESRCH);
931 
932 	switch (data->type) {
933 	case INTR_MAP_DATA_MSI:
934 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
935 		    ("%s: Found a non-MSI controller: %s", __func__,
936 		     device_get_name(pic->pic_dev)));
937 		msi = (struct intr_map_data_msi *)data;
938 		*isrc = msi->isrc;
939 		return (0);
940 
941 	default:
942 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
943 		    ("%s: Found a non-PIC controller: %s", __func__,
944 		     device_get_name(pic->pic_dev)));
945 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
946 	}
947 }
948 
949 int
950 intr_activate_irq(device_t dev, struct resource *res)
951 {
952 	device_t map_dev;
953 	intptr_t map_xref;
954 	struct intr_map_data *data;
955 	struct intr_irqsrc *isrc;
956 	u_int res_id;
957 	int error;
958 
959 	KASSERT(rman_get_start(res) == rman_get_end(res),
960 	    ("%s: more interrupts in resource", __func__));
961 
962 	res_id = (u_int)rman_get_start(res);
963 	if (intr_map_get_isrc(res_id) != NULL)
964 		panic("Attempt to double activation of resource id: %u\n",
965 		    res_id);
966 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
967 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
968 	if (error != 0) {
969 		free(data, M_INTRNG);
970 		/* XXX TODO DISCONECTED PICs */
971 		/* if (error == EINVAL) return(0); */
972 		return (error);
973 	}
974 	intr_map_set_isrc(res_id, isrc);
975 	rman_set_virtual(res, data);
976 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
977 }
978 
979 int
980 intr_deactivate_irq(device_t dev, struct resource *res)
981 {
982 	struct intr_map_data *data;
983 	struct intr_irqsrc *isrc;
984 	u_int res_id;
985 	int error;
986 
987 	KASSERT(rman_get_start(res) == rman_get_end(res),
988 	    ("%s: more interrupts in resource", __func__));
989 
990 	res_id = (u_int)rman_get_start(res);
991 	isrc = intr_map_get_isrc(res_id);
992 	if (isrc == NULL)
993 		panic("Attempt to deactivate non-active resource id: %u\n",
994 		    res_id);
995 
996 	data = rman_get_virtual(res);
997 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
998 	intr_map_set_isrc(res_id, NULL);
999 	rman_set_virtual(res, NULL);
1000 	free(data, M_INTRNG);
1001 	return (error);
1002 }
1003 
1004 int
1005 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
1006     driver_intr_t hand, void *arg, int flags, void **cookiep)
1007 {
1008 	int error;
1009 	struct intr_map_data *data;
1010 	struct intr_irqsrc *isrc;
1011 	const char *name;
1012 	u_int res_id;
1013 
1014 	KASSERT(rman_get_start(res) == rman_get_end(res),
1015 	    ("%s: more interrupts in resource", __func__));
1016 
1017 	res_id = (u_int)rman_get_start(res);
1018 	isrc = intr_map_get_isrc(res_id);
1019 	if (isrc == NULL) {
1020 		/* XXX TODO DISCONECTED PICs */
1021 		return (EINVAL);
1022 	}
1023 
1024 	data = rman_get_virtual(res);
1025 	name = device_get_nameunit(dev);
1026 
1027 #ifdef INTR_SOLO
1028 	/*
1029 	 * Standard handling is done through MI interrupt framework. However,
1030 	 * some interrupts could request solely own special handling. This
1031 	 * non standard handling can be used for interrupt controllers without
1032 	 * handler (filter only), so in case that interrupt controllers are
1033 	 * chained, MI interrupt framework is called only in leaf controller.
1034 	 *
1035 	 * Note that root interrupt controller routine is served as well,
1036 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1037 	 */
1038 	if (flags & INTR_SOLO && hand != NULL) {
1039 		debugf("irq %u cannot solo on %s\n", irq, name);
1040 		return (EINVAL);
1041 	}
1042 
1043 	if (flags & INTR_SOLO) {
1044 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1045 		    arg, cookiep);
1046 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1047 		    name);
1048 	} else
1049 #endif
1050 		{
1051 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1052 		    cookiep);
1053 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1054 	}
1055 	if (error != 0)
1056 		return (error);
1057 
1058 	mtx_lock(&isrc_table_lock);
1059 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1060 	if (error == 0) {
1061 		isrc->isrc_handlers++;
1062 		if (isrc->isrc_handlers == 1)
1063 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1064 	}
1065 	mtx_unlock(&isrc_table_lock);
1066 	if (error != 0)
1067 		intr_event_remove_handler(*cookiep);
1068 	return (error);
1069 }
1070 
1071 int
1072 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1073 {
1074 	int error;
1075 	struct intr_map_data *data;
1076 	struct intr_irqsrc *isrc;
1077 	u_int res_id;
1078 
1079 	KASSERT(rman_get_start(res) == rman_get_end(res),
1080 	    ("%s: more interrupts in resource", __func__));
1081 
1082 	res_id = (u_int)rman_get_start(res);
1083 	isrc = intr_map_get_isrc(res_id);
1084 	if (isrc == NULL || isrc->isrc_handlers == 0)
1085 		return (EINVAL);
1086 
1087 	data = rman_get_virtual(res);
1088 
1089 #ifdef INTR_SOLO
1090 	if (isrc->isrc_filter != NULL) {
1091 		if (isrc != cookie)
1092 			return (EINVAL);
1093 
1094 		mtx_lock(&isrc_table_lock);
1095 		isrc->isrc_filter = NULL;
1096 		isrc->isrc_arg = NULL;
1097 		isrc->isrc_handlers = 0;
1098 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1099 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1100 		isrc_update_name(isrc, NULL);
1101 		mtx_unlock(&isrc_table_lock);
1102 		return (0);
1103 	}
1104 #endif
1105 	if (isrc != intr_handler_source(cookie))
1106 		return (EINVAL);
1107 
1108 	error = intr_event_remove_handler(cookie);
1109 	if (error == 0) {
1110 		mtx_lock(&isrc_table_lock);
1111 		isrc->isrc_handlers--;
1112 		if (isrc->isrc_handlers == 0)
1113 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1114 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1115 		intrcnt_updatename(isrc);
1116 		mtx_unlock(&isrc_table_lock);
1117 	}
1118 	return (error);
1119 }
1120 
1121 int
1122 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1123     const char *descr)
1124 {
1125 	int error;
1126 	struct intr_irqsrc *isrc;
1127 	u_int res_id;
1128 
1129 	KASSERT(rman_get_start(res) == rman_get_end(res),
1130 	    ("%s: more interrupts in resource", __func__));
1131 
1132 	res_id = (u_int)rman_get_start(res);
1133 	isrc = intr_map_get_isrc(res_id);
1134 	if (isrc == NULL || isrc->isrc_handlers == 0)
1135 		return (EINVAL);
1136 #ifdef INTR_SOLO
1137 	if (isrc->isrc_filter != NULL) {
1138 		if (isrc != cookie)
1139 			return (EINVAL);
1140 
1141 		mtx_lock(&isrc_table_lock);
1142 		isrc_update_name(isrc, descr);
1143 		mtx_unlock(&isrc_table_lock);
1144 		return (0);
1145 	}
1146 #endif
1147 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1148 	if (error == 0) {
1149 		mtx_lock(&isrc_table_lock);
1150 		intrcnt_updatename(isrc);
1151 		mtx_unlock(&isrc_table_lock);
1152 	}
1153 	return (error);
1154 }
1155 
1156 #ifdef SMP
1157 int
1158 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1159 {
1160 	struct intr_irqsrc *isrc;
1161 	u_int res_id;
1162 
1163 	KASSERT(rman_get_start(res) == rman_get_end(res),
1164 	    ("%s: more interrupts in resource", __func__));
1165 
1166 	res_id = (u_int)rman_get_start(res);
1167 	isrc = intr_map_get_isrc(res_id);
1168 	if (isrc == NULL || isrc->isrc_handlers == 0)
1169 		return (EINVAL);
1170 #ifdef INTR_SOLO
1171 	if (isrc->isrc_filter != NULL)
1172 		return (intr_isrc_assign_cpu(isrc, cpu));
1173 #endif
1174 	return (intr_event_bind(isrc->isrc_event, cpu));
1175 }
1176 
1177 /*
1178  * Return the CPU that the next interrupt source should use.
1179  * For now just returns the next CPU according to round-robin.
1180  */
1181 u_int
1182 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1183 {
1184 	u_int cpu;
1185 
1186 	KASSERT(!CPU_EMPTY(cpumask), ("%s: Empty CPU mask", __func__));
1187 	if (!irq_assign_cpu || mp_ncpus == 1) {
1188 		cpu = PCPU_GET(cpuid);
1189 
1190 		if (CPU_ISSET(cpu, cpumask))
1191 			return (curcpu);
1192 
1193 		return (CPU_FFS(cpumask) - 1);
1194 	}
1195 
1196 	do {
1197 		last_cpu++;
1198 		if (last_cpu > mp_maxid)
1199 			last_cpu = 0;
1200 	} while (!CPU_ISSET(last_cpu, cpumask));
1201 	return (last_cpu);
1202 }
1203 
1204 #ifndef EARLY_AP_STARTUP
1205 /*
1206  *  Distribute all the interrupt sources among the available
1207  *  CPUs once the AP's have been launched.
1208  */
1209 static void
1210 intr_irq_shuffle(void *arg __unused)
1211 {
1212 	struct intr_irqsrc *isrc;
1213 	u_int i;
1214 
1215 	if (mp_ncpus == 1)
1216 		return;
1217 
1218 	mtx_lock(&isrc_table_lock);
1219 	irq_assign_cpu = true;
1220 	for (i = 0; i < NIRQ; i++) {
1221 		isrc = irq_sources[i];
1222 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1223 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1224 			continue;
1225 
1226 		if (isrc->isrc_event != NULL &&
1227 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1228 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1229 			panic("%s: CPU inconsistency", __func__);
1230 
1231 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1232 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1233 
1234 		/*
1235 		 * We are in wicked position here if the following call fails
1236 		 * for bound ISRC. The best thing we can do is to clear
1237 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1238 		 */
1239 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1240 			CPU_ZERO(&isrc->isrc_cpu);
1241 	}
1242 	mtx_unlock(&isrc_table_lock);
1243 }
1244 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1245 #endif /* !EARLY_AP_STARTUP */
1246 
1247 #else
1248 u_int
1249 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1250 {
1251 
1252 	return (PCPU_GET(cpuid));
1253 }
1254 #endif /* SMP */
1255 
1256 /*
1257  * Allocate memory for new intr_map_data structure.
1258  * Initialize common fields.
1259  */
1260 struct intr_map_data *
1261 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1262 {
1263 	struct intr_map_data *data;
1264 
1265 	data = malloc(len, M_INTRNG, flags);
1266 	data->type = type;
1267 	data->len = len;
1268 	return (data);
1269 }
1270 
1271 void intr_free_intr_map_data(struct intr_map_data *data)
1272 {
1273 
1274 	free(data, M_INTRNG);
1275 }
1276 
1277 /*
1278  *  Register a MSI/MSI-X interrupt controller
1279  */
1280 int
1281 intr_msi_register(device_t dev, intptr_t xref)
1282 {
1283 	struct intr_pic *pic;
1284 
1285 	if (dev == NULL)
1286 		return (EINVAL);
1287 	pic = pic_create(dev, xref, FLAG_MSI);
1288 	if (pic == NULL)
1289 		return (ENOMEM);
1290 
1291 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1292 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1293 	return (0);
1294 }
1295 
1296 int
1297 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1298     int maxcount, int *irqs)
1299 {
1300 #ifdef IOMMU
1301 	struct iommu_domain *domain;
1302 #endif
1303 	struct intr_irqsrc **isrc;
1304 	struct intr_pic *pic;
1305 	device_t pdev;
1306 	struct intr_map_data_msi *msi;
1307 	int err, i;
1308 
1309 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1310 	if (pic == NULL)
1311 		return (ESRCH);
1312 
1313 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1314 	    ("%s: Found a non-MSI controller: %s", __func__,
1315 	     device_get_name(pic->pic_dev)));
1316 
1317 #ifdef IOMMU
1318 	/*
1319 	 * If this is the first time we have used this context ask the
1320 	 * interrupt controller to map memory the msi source will need.
1321 	 */
1322 	err = MSI_IOMMU_INIT(pic->pic_dev, child, &domain);
1323 	if (err != 0)
1324 		return (err);
1325 #endif
1326 
1327 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1328 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1329 	if (err != 0) {
1330 		free(isrc, M_INTRNG);
1331 		return (err);
1332 	}
1333 
1334 	for (i = 0; i < count; i++) {
1335 #ifdef IOMMU
1336 		isrc[i]->isrc_iommu = domain;
1337 #endif
1338 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1339 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1340 		msi-> isrc = isrc[i];
1341 
1342 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1343 		    (struct intr_map_data *)msi);
1344 	}
1345 	free(isrc, M_INTRNG);
1346 
1347 	return (err);
1348 }
1349 
1350 int
1351 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1352     int *irqs)
1353 {
1354 	struct intr_irqsrc **isrc;
1355 	struct intr_pic *pic;
1356 	struct intr_map_data_msi *msi;
1357 	int i, err;
1358 
1359 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1360 	if (pic == NULL)
1361 		return (ESRCH);
1362 
1363 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1364 	    ("%s: Found a non-MSI controller: %s", __func__,
1365 	     device_get_name(pic->pic_dev)));
1366 
1367 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1368 
1369 	for (i = 0; i < count; i++) {
1370 		msi = (struct intr_map_data_msi *)
1371 		    intr_map_get_map_data(irqs[i]);
1372 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1373 		    ("%s: irq %d map data is not MSI", __func__,
1374 		    irqs[i]));
1375 		isrc[i] = msi->isrc;
1376 	}
1377 
1378 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1379 
1380 	for (i = 0; i < count; i++) {
1381 		if (isrc[i] != NULL)
1382 			intr_unmap_irq(irqs[i]);
1383 	}
1384 
1385 	free(isrc, M_INTRNG);
1386 	return (err);
1387 }
1388 
1389 int
1390 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1391 {
1392 #ifdef IOMMU
1393 	struct iommu_domain *domain;
1394 #endif
1395 	struct intr_irqsrc *isrc;
1396 	struct intr_pic *pic;
1397 	device_t pdev;
1398 	struct intr_map_data_msi *msi;
1399 	int err;
1400 
1401 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1402 	if (pic == NULL)
1403 		return (ESRCH);
1404 
1405 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1406 	    ("%s: Found a non-MSI controller: %s", __func__,
1407 	     device_get_name(pic->pic_dev)));
1408 
1409 #ifdef IOMMU
1410 	/*
1411 	 * If this is the first time we have used this context ask the
1412 	 * interrupt controller to map memory the msi source will need.
1413 	 */
1414 	err = MSI_IOMMU_INIT(pic->pic_dev, child, &domain);
1415 	if (err != 0)
1416 		return (err);
1417 #endif
1418 
1419 	err = MSI_ALLOC_MSIX(pic->pic_dev, child, &pdev, &isrc);
1420 	if (err != 0)
1421 		return (err);
1422 
1423 #ifdef IOMMU
1424 	isrc->isrc_iommu = domain;
1425 #endif
1426 	msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1427 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1428 	msi->isrc = isrc;
1429 	*irq = intr_map_irq(pic->pic_dev, xref, (struct intr_map_data *)msi);
1430 	return (0);
1431 }
1432 
1433 int
1434 intr_release_msix(device_t pci, device_t child, intptr_t xref, int irq)
1435 {
1436 	struct intr_irqsrc *isrc;
1437 	struct intr_pic *pic;
1438 	struct intr_map_data_msi *msi;
1439 	int err;
1440 
1441 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1442 	if (pic == NULL)
1443 		return (ESRCH);
1444 
1445 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1446 	    ("%s: Found a non-MSI controller: %s", __func__,
1447 	     device_get_name(pic->pic_dev)));
1448 
1449 	msi = (struct intr_map_data_msi *)
1450 	    intr_map_get_map_data(irq);
1451 	KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1452 	    ("%s: irq %d map data is not MSI", __func__,
1453 	    irq));
1454 	isrc = msi->isrc;
1455 	if (isrc == NULL) {
1456 		intr_unmap_irq(irq);
1457 		return (EINVAL);
1458 	}
1459 
1460 	err = MSI_RELEASE_MSIX(pic->pic_dev, child, isrc);
1461 	intr_unmap_irq(irq);
1462 
1463 	return (err);
1464 }
1465 
1466 int
1467 intr_map_msi(device_t pci, device_t child, intptr_t xref, int irq,
1468     uint64_t *addr, uint32_t *data)
1469 {
1470 	struct intr_irqsrc *isrc;
1471 	struct intr_pic *pic;
1472 	int err;
1473 
1474 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1475 	if (pic == NULL)
1476 		return (ESRCH);
1477 
1478 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1479 	    ("%s: Found a non-MSI controller: %s", __func__,
1480 	     device_get_name(pic->pic_dev)));
1481 
1482 	isrc = intr_map_get_isrc(irq);
1483 	if (isrc == NULL)
1484 		return (EINVAL);
1485 
1486 	err = MSI_MAP_MSI(pic->pic_dev, child, isrc, addr, data);
1487 
1488 #ifdef IOMMU
1489 	if (isrc->isrc_iommu != NULL)
1490 		iommu_translate_msi(isrc->isrc_iommu, addr);
1491 #endif
1492 
1493 	return (err);
1494 }
1495 
1496 void dosoftints(void);
1497 void
1498 dosoftints(void)
1499 {
1500 }
1501 
1502 #ifdef SMP
1503 /*
1504  *  Init interrupt controller on another CPU.
1505  */
1506 void
1507 intr_pic_init_secondary(void)
1508 {
1509 
1510 	/*
1511 	 * QQQ: Only root PIC is aware of other CPUs ???
1512 	 */
1513 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1514 
1515 	//mtx_lock(&isrc_table_lock);
1516 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1517 	//mtx_unlock(&isrc_table_lock);
1518 }
1519 #endif
1520 
1521 #ifdef DDB
1522 DB_SHOW_COMMAND(irqs, db_show_irqs)
1523 {
1524 	u_int i, irqsum;
1525 	u_long num;
1526 	struct intr_irqsrc *isrc;
1527 
1528 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1529 		isrc = irq_sources[i];
1530 		if (isrc == NULL)
1531 			continue;
1532 
1533 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1534 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1535 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1536 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1537 		irqsum += num;
1538 	}
1539 	db_printf("irq total %u\n", irqsum);
1540 }
1541 #endif
1542 
1543 /*
1544  * Interrupt mapping table functions.
1545  *
1546  * Please, keep this part separately, it can be transformed to
1547  * extension of standard resources.
1548  */
1549 struct intr_map_entry
1550 {
1551 	device_t 		dev;
1552 	intptr_t 		xref;
1553 	struct intr_map_data 	*map_data;
1554 	struct intr_irqsrc 	*isrc;
1555 	/* XXX TODO DISCONECTED PICs */
1556 	/*int			flags */
1557 };
1558 
1559 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1560 static struct intr_map_entry *irq_map[2 * NIRQ];
1561 static int irq_map_count = nitems(irq_map);
1562 static int irq_map_first_free_idx;
1563 static struct mtx irq_map_lock;
1564 
1565 static struct intr_irqsrc *
1566 intr_map_get_isrc(u_int res_id)
1567 {
1568 	struct intr_irqsrc *isrc;
1569 
1570 	isrc = NULL;
1571 	mtx_lock(&irq_map_lock);
1572 	if (res_id < irq_map_count && irq_map[res_id] != NULL)
1573 		isrc = irq_map[res_id]->isrc;
1574 	mtx_unlock(&irq_map_lock);
1575 
1576 	return (isrc);
1577 }
1578 
1579 static void
1580 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1581 {
1582 
1583 	mtx_lock(&irq_map_lock);
1584 	if (res_id < irq_map_count && irq_map[res_id] != NULL)
1585 		irq_map[res_id]->isrc = isrc;
1586 	mtx_unlock(&irq_map_lock);
1587 }
1588 
1589 /*
1590  * Get a copy of intr_map_entry data
1591  */
1592 static struct intr_map_data *
1593 intr_map_get_map_data(u_int res_id)
1594 {
1595 	struct intr_map_data *data;
1596 
1597 	data = NULL;
1598 	mtx_lock(&irq_map_lock);
1599 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1600 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1601 	data = irq_map[res_id]->map_data;
1602 	mtx_unlock(&irq_map_lock);
1603 
1604 	return (data);
1605 }
1606 
1607 /*
1608  * Get a copy of intr_map_entry data
1609  */
1610 static void
1611 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1612     struct intr_map_data **data)
1613 {
1614 	size_t len;
1615 
1616 	len = 0;
1617 	mtx_lock(&irq_map_lock);
1618 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1619 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1620 	if (irq_map[res_id]->map_data != NULL)
1621 		len = irq_map[res_id]->map_data->len;
1622 	mtx_unlock(&irq_map_lock);
1623 
1624 	if (len == 0)
1625 		*data = NULL;
1626 	else
1627 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1628 	mtx_lock(&irq_map_lock);
1629 	if (irq_map[res_id] == NULL)
1630 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1631 	if (len != 0) {
1632 		if (len != irq_map[res_id]->map_data->len)
1633 			panic("Resource id: %u has changed.\n", res_id);
1634 		memcpy(*data, irq_map[res_id]->map_data, len);
1635 	}
1636 	*map_dev = irq_map[res_id]->dev;
1637 	*map_xref = irq_map[res_id]->xref;
1638 	mtx_unlock(&irq_map_lock);
1639 }
1640 
1641 /*
1642  * Allocate and fill new entry in irq_map table.
1643  */
1644 u_int
1645 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1646 {
1647 	u_int i;
1648 	struct intr_map_entry *entry;
1649 
1650 	/* Prepare new entry first. */
1651 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1652 
1653 	entry->dev = dev;
1654 	entry->xref = xref;
1655 	entry->map_data = data;
1656 	entry->isrc = NULL;
1657 
1658 	mtx_lock(&irq_map_lock);
1659 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1660 		if (irq_map[i] == NULL) {
1661 			irq_map[i] = entry;
1662 			irq_map_first_free_idx = i + 1;
1663 			mtx_unlock(&irq_map_lock);
1664 			return (i);
1665 		}
1666 	}
1667 	mtx_unlock(&irq_map_lock);
1668 
1669 	/* XXX Expand irq_map table */
1670 	panic("IRQ mapping table is full.");
1671 }
1672 
1673 /*
1674  * Remove and free mapping entry.
1675  */
1676 void
1677 intr_unmap_irq(u_int res_id)
1678 {
1679 	struct intr_map_entry *entry;
1680 
1681 	mtx_lock(&irq_map_lock);
1682 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1683 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1684 	entry = irq_map[res_id];
1685 	irq_map[res_id] = NULL;
1686 	irq_map_first_free_idx = res_id;
1687 	mtx_unlock(&irq_map_lock);
1688 	intr_free_intr_map_data(entry->map_data);
1689 	free(entry, M_INTRNG);
1690 }
1691 
1692 /*
1693  * Clone mapping entry.
1694  */
1695 u_int
1696 intr_map_clone_irq(u_int old_res_id)
1697 {
1698 	device_t map_dev;
1699 	intptr_t map_xref;
1700 	struct intr_map_data *data;
1701 
1702 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1703 	return (intr_map_irq(map_dev, map_xref, data));
1704 }
1705 
1706 static void
1707 intr_map_init(void *dummy __unused)
1708 {
1709 
1710 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1711 }
1712 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1713