xref: /freebsd/sys/kern/subr_intr.c (revision b00ab754)
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 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/syslog.h>
46 #include <sys/malloc.h>
47 #include <sys/proc.h>
48 #include <sys/queue.h>
49 #include <sys/bus.h>
50 #include <sys/interrupt.h>
51 #include <sys/conf.h>
52 #include <sys/cpuset.h>
53 #include <sys/rman.h>
54 #include <sys/sched.h>
55 #include <sys/smp.h>
56 #include <sys/vmmeter.h>
57 #ifdef HWPMC_HOOKS
58 #include <sys/pmckern.h>
59 #endif
60 
61 #include <machine/atomic.h>
62 #include <machine/intr.h>
63 #include <machine/cpu.h>
64 #include <machine/smp.h>
65 #include <machine/stdarg.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70 
71 #include "pic_if.h"
72 #include "msi_if.h"
73 
74 #define	INTRNAME_LEN	(2*MAXCOMLEN + 1)
75 
76 #ifdef DEBUG
77 #define debugf(fmt, args...) do { printf("%s(): ", __func__);	\
78     printf(fmt,##args); } while (0)
79 #else
80 #define debugf(fmt, args...)
81 #endif
82 
83 MALLOC_DECLARE(M_INTRNG);
84 MALLOC_DEFINE(M_INTRNG, "intr", "intr interrupt handling");
85 
86 /* Main interrupt handler called from assembler -> 'hidden' for C code. */
87 void intr_irq_handler(struct trapframe *tf);
88 
89 /* Root interrupt controller stuff. */
90 device_t intr_irq_root_dev;
91 static intr_irq_filter_t *irq_root_filter;
92 static void *irq_root_arg;
93 static u_int irq_root_ipicount;
94 
95 struct intr_pic_child {
96 	SLIST_ENTRY(intr_pic_child)	 pc_next;
97 	struct intr_pic			*pc_pic;
98 	intr_child_irq_filter_t		*pc_filter;
99 	void				*pc_filter_arg;
100 	uintptr_t			 pc_start;
101 	uintptr_t			 pc_length;
102 };
103 
104 /* Interrupt controller definition. */
105 struct intr_pic {
106 	SLIST_ENTRY(intr_pic)	pic_next;
107 	intptr_t		pic_xref;	/* hardware identification */
108 	device_t		pic_dev;
109 /* Only one of FLAG_PIC or FLAG_MSI may be set */
110 #define	FLAG_PIC	(1 << 0)
111 #define	FLAG_MSI	(1 << 1)
112 #define	FLAG_TYPE_MASK	(FLAG_PIC | FLAG_MSI)
113 	u_int			pic_flags;
114 	struct mtx		pic_child_lock;
115 	SLIST_HEAD(, intr_pic_child) pic_children;
116 };
117 
118 static struct mtx pic_list_lock;
119 static SLIST_HEAD(, intr_pic) pic_list;
120 
121 static struct intr_pic *pic_lookup(device_t dev, intptr_t xref, int flags);
122 
123 /* Interrupt source definition. */
124 static struct mtx isrc_table_lock;
125 static struct intr_irqsrc *irq_sources[NIRQ];
126 u_int irq_next_free;
127 
128 #ifdef SMP
129 static boolean_t irq_assign_cpu = FALSE;
130 #endif
131 
132 /*
133  * - 2 counters for each I/O interrupt.
134  * - MAXCPU counters for each IPI counters for SMP.
135  */
136 #ifdef SMP
137 #define INTRCNT_COUNT   (NIRQ * 2 + INTR_IPI_COUNT * MAXCPU)
138 #else
139 #define INTRCNT_COUNT   (NIRQ * 2)
140 #endif
141 
142 /* Data for MI statistics reporting. */
143 u_long intrcnt[INTRCNT_COUNT];
144 char intrnames[INTRCNT_COUNT * INTRNAME_LEN];
145 size_t sintrcnt = sizeof(intrcnt);
146 size_t sintrnames = sizeof(intrnames);
147 static u_int intrcnt_index;
148 
149 static struct intr_irqsrc *intr_map_get_isrc(u_int res_id);
150 static void intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc);
151 static struct intr_map_data * intr_map_get_map_data(u_int res_id);
152 static void intr_map_copy_map_data(u_int res_id, device_t *dev, intptr_t *xref,
153     struct intr_map_data **data);
154 
155 /*
156  *  Interrupt framework initialization routine.
157  */
158 static void
159 intr_irq_init(void *dummy __unused)
160 {
161 
162 	SLIST_INIT(&pic_list);
163 	mtx_init(&pic_list_lock, "intr pic list", NULL, MTX_DEF);
164 
165 	mtx_init(&isrc_table_lock, "intr isrc table", NULL, MTX_DEF);
166 }
167 SYSINIT(intr_irq_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_irq_init, NULL);
168 
169 static void
170 intrcnt_setname(const char *name, int index)
171 {
172 
173 	snprintf(intrnames + INTRNAME_LEN * index, INTRNAME_LEN, "%-*s",
174 	    INTRNAME_LEN - 1, name);
175 }
176 
177 /*
178  *  Update name for interrupt source with interrupt event.
179  */
180 static void
181 intrcnt_updatename(struct intr_irqsrc *isrc)
182 {
183 
184 	/* QQQ: What about stray counter name? */
185 	mtx_assert(&isrc_table_lock, MA_OWNED);
186 	intrcnt_setname(isrc->isrc_event->ie_fullname, isrc->isrc_index);
187 }
188 
189 /*
190  *  Virtualization for interrupt source interrupt counter increment.
191  */
192 static inline void
193 isrc_increment_count(struct intr_irqsrc *isrc)
194 {
195 
196 	if (isrc->isrc_flags & INTR_ISRCF_PPI)
197 		atomic_add_long(&isrc->isrc_count[0], 1);
198 	else
199 		isrc->isrc_count[0]++;
200 }
201 
202 /*
203  *  Virtualization for interrupt source interrupt stray counter increment.
204  */
205 static inline void
206 isrc_increment_straycount(struct intr_irqsrc *isrc)
207 {
208 
209 	isrc->isrc_count[1]++;
210 }
211 
212 /*
213  *  Virtualization for interrupt source interrupt name update.
214  */
215 static void
216 isrc_update_name(struct intr_irqsrc *isrc, const char *name)
217 {
218 	char str[INTRNAME_LEN];
219 
220 	mtx_assert(&isrc_table_lock, MA_OWNED);
221 
222 	if (name != NULL) {
223 		snprintf(str, INTRNAME_LEN, "%s: %s", isrc->isrc_name, name);
224 		intrcnt_setname(str, isrc->isrc_index);
225 		snprintf(str, INTRNAME_LEN, "stray %s: %s", isrc->isrc_name,
226 		    name);
227 		intrcnt_setname(str, isrc->isrc_index + 1);
228 	} else {
229 		snprintf(str, INTRNAME_LEN, "%s:", isrc->isrc_name);
230 		intrcnt_setname(str, isrc->isrc_index);
231 		snprintf(str, INTRNAME_LEN, "stray %s:", isrc->isrc_name);
232 		intrcnt_setname(str, isrc->isrc_index + 1);
233 	}
234 }
235 
236 /*
237  *  Virtualization for interrupt source interrupt counters setup.
238  */
239 static void
240 isrc_setup_counters(struct intr_irqsrc *isrc)
241 {
242 	u_int index;
243 
244 	/*
245 	 *  XXX - it does not work well with removable controllers and
246 	 *        interrupt sources !!!
247 	 */
248 	index = atomic_fetchadd_int(&intrcnt_index, 2);
249 	isrc->isrc_index = index;
250 	isrc->isrc_count = &intrcnt[index];
251 	isrc_update_name(isrc, NULL);
252 }
253 
254 /*
255  *  Virtualization for interrupt source interrupt counters release.
256  */
257 static void
258 isrc_release_counters(struct intr_irqsrc *isrc)
259 {
260 
261 	panic("%s: not implemented", __func__);
262 }
263 
264 #ifdef SMP
265 /*
266  *  Virtualization for interrupt source IPI counters setup.
267  */
268 u_long *
269 intr_ipi_setup_counters(const char *name)
270 {
271 	u_int index, i;
272 	char str[INTRNAME_LEN];
273 
274 	index = atomic_fetchadd_int(&intrcnt_index, MAXCPU);
275 	for (i = 0; i < MAXCPU; i++) {
276 		snprintf(str, INTRNAME_LEN, "cpu%d:%s", i, name);
277 		intrcnt_setname(str, index + i);
278 	}
279 	return (&intrcnt[index]);
280 }
281 #endif
282 
283 /*
284  *  Main interrupt dispatch handler. It's called straight
285  *  from the assembler, where CPU interrupt is served.
286  */
287 void
288 intr_irq_handler(struct trapframe *tf)
289 {
290 	struct trapframe * oldframe;
291 	struct thread * td;
292 
293 	KASSERT(irq_root_filter != NULL, ("%s: no filter", __func__));
294 
295 	VM_CNT_INC(v_intr);
296 	critical_enter();
297 	td = curthread;
298 	oldframe = td->td_intr_frame;
299 	td->td_intr_frame = tf;
300 	irq_root_filter(irq_root_arg);
301 	td->td_intr_frame = oldframe;
302 	critical_exit();
303 #ifdef HWPMC_HOOKS
304 	if (pmc_hook && TRAPF_USERMODE(tf) &&
305 	    (PCPU_GET(curthread)->td_pflags & TDP_CALLCHAIN))
306 		pmc_hook(PCPU_GET(curthread), PMC_FN_USER_CALLCHAIN, tf);
307 #endif
308 }
309 
310 int
311 intr_child_irq_handler(struct intr_pic *parent, uintptr_t irq)
312 {
313 	struct intr_pic_child *child;
314 	bool found;
315 
316 	found = false;
317 	mtx_lock_spin(&parent->pic_child_lock);
318 	SLIST_FOREACH(child, &parent->pic_children, pc_next) {
319 		if (child->pc_start <= irq &&
320 		    irq < (child->pc_start + child->pc_length)) {
321 			found = true;
322 			break;
323 		}
324 	}
325 	mtx_unlock_spin(&parent->pic_child_lock);
326 
327 	if (found)
328 		return (child->pc_filter(child->pc_filter_arg, irq));
329 
330 	return (FILTER_STRAY);
331 }
332 
333 /*
334  *  interrupt controller dispatch function for interrupts. It should
335  *  be called straight from the interrupt controller, when associated interrupt
336  *  source is learned.
337  */
338 int
339 intr_isrc_dispatch(struct intr_irqsrc *isrc, struct trapframe *tf)
340 {
341 
342 	KASSERT(isrc != NULL, ("%s: no source", __func__));
343 
344 	isrc_increment_count(isrc);
345 
346 #ifdef INTR_SOLO
347 	if (isrc->isrc_filter != NULL) {
348 		int error;
349 		error = isrc->isrc_filter(isrc->isrc_arg, tf);
350 		PIC_POST_FILTER(isrc->isrc_dev, isrc);
351 		if (error == FILTER_HANDLED)
352 			return (0);
353 	} else
354 #endif
355 	if (isrc->isrc_event != NULL) {
356 		if (intr_event_handle(isrc->isrc_event, tf) == 0)
357 			return (0);
358 	}
359 
360 	isrc_increment_straycount(isrc);
361 	return (EINVAL);
362 }
363 
364 /*
365  *  Alloc unique interrupt number (resource handle) for interrupt source.
366  *
367  *  There could be various strategies how to allocate free interrupt number
368  *  (resource handle) for new interrupt source.
369  *
370  *  1. Handles are always allocated forward, so handles are not recycled
371  *     immediately. However, if only one free handle left which is reused
372  *     constantly...
373  */
374 static inline int
375 isrc_alloc_irq(struct intr_irqsrc *isrc)
376 {
377 	u_int maxirqs, irq;
378 
379 	mtx_assert(&isrc_table_lock, MA_OWNED);
380 
381 	maxirqs = nitems(irq_sources);
382 	if (irq_next_free >= maxirqs)
383 		return (ENOSPC);
384 
385 	for (irq = irq_next_free; irq < maxirqs; irq++) {
386 		if (irq_sources[irq] == NULL)
387 			goto found;
388 	}
389 	for (irq = 0; irq < irq_next_free; irq++) {
390 		if (irq_sources[irq] == NULL)
391 			goto found;
392 	}
393 
394 	irq_next_free = maxirqs;
395 	return (ENOSPC);
396 
397 found:
398 	isrc->isrc_irq = irq;
399 	irq_sources[irq] = isrc;
400 
401 	irq_next_free = irq + 1;
402 	if (irq_next_free >= maxirqs)
403 		irq_next_free = 0;
404 	return (0);
405 }
406 
407 /*
408  *  Free unique interrupt number (resource handle) from interrupt source.
409  */
410 static inline int
411 isrc_free_irq(struct intr_irqsrc *isrc)
412 {
413 
414 	mtx_assert(&isrc_table_lock, MA_OWNED);
415 
416 	if (isrc->isrc_irq >= nitems(irq_sources))
417 		return (EINVAL);
418 	if (irq_sources[isrc->isrc_irq] != isrc)
419 		return (EINVAL);
420 
421 	irq_sources[isrc->isrc_irq] = NULL;
422 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
423 	return (0);
424 }
425 
426 /*
427  *  Initialize interrupt source and register it into global interrupt table.
428  */
429 int
430 intr_isrc_register(struct intr_irqsrc *isrc, device_t dev, u_int flags,
431     const char *fmt, ...)
432 {
433 	int error;
434 	va_list ap;
435 
436 	bzero(isrc, sizeof(struct intr_irqsrc));
437 	isrc->isrc_dev = dev;
438 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
439 	isrc->isrc_flags = flags;
440 
441 	va_start(ap, fmt);
442 	vsnprintf(isrc->isrc_name, INTR_ISRC_NAMELEN, fmt, ap);
443 	va_end(ap);
444 
445 	mtx_lock(&isrc_table_lock);
446 	error = isrc_alloc_irq(isrc);
447 	if (error != 0) {
448 		mtx_unlock(&isrc_table_lock);
449 		return (error);
450 	}
451 	/*
452 	 * Setup interrupt counters, but not for IPI sources. Those are setup
453 	 * later and only for used ones (up to INTR_IPI_COUNT) to not exhaust
454 	 * our counter pool.
455 	 */
456 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
457 		isrc_setup_counters(isrc);
458 	mtx_unlock(&isrc_table_lock);
459 	return (0);
460 }
461 
462 /*
463  *  Deregister interrupt source from global interrupt table.
464  */
465 int
466 intr_isrc_deregister(struct intr_irqsrc *isrc)
467 {
468 	int error;
469 
470 	mtx_lock(&isrc_table_lock);
471 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
472 		isrc_release_counters(isrc);
473 	error = isrc_free_irq(isrc);
474 	mtx_unlock(&isrc_table_lock);
475 	return (error);
476 }
477 
478 #ifdef SMP
479 /*
480  *  A support function for a PIC to decide if provided ISRC should be inited
481  *  on given cpu. The logic of INTR_ISRCF_BOUND flag and isrc_cpu member of
482  *  struct intr_irqsrc is the following:
483  *
484  *     If INTR_ISRCF_BOUND is set, the ISRC should be inited only on cpus
485  *     set in isrc_cpu. If not, the ISRC should be inited on every cpu and
486  *     isrc_cpu is kept consistent with it. Thus isrc_cpu is always correct.
487  */
488 bool
489 intr_isrc_init_on_cpu(struct intr_irqsrc *isrc, u_int cpu)
490 {
491 
492 	if (isrc->isrc_handlers == 0)
493 		return (false);
494 	if ((isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI)) == 0)
495 		return (false);
496 	if (isrc->isrc_flags & INTR_ISRCF_BOUND)
497 		return (CPU_ISSET(cpu, &isrc->isrc_cpu));
498 
499 	CPU_SET(cpu, &isrc->isrc_cpu);
500 	return (true);
501 }
502 #endif
503 
504 #ifdef INTR_SOLO
505 /*
506  *  Setup filter into interrupt source.
507  */
508 static int
509 iscr_setup_filter(struct intr_irqsrc *isrc, const char *name,
510     intr_irq_filter_t *filter, void *arg, void **cookiep)
511 {
512 
513 	if (filter == NULL)
514 		return (EINVAL);
515 
516 	mtx_lock(&isrc_table_lock);
517 	/*
518 	 * Make sure that we do not mix the two ways
519 	 * how we handle interrupt sources.
520 	 */
521 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
522 		mtx_unlock(&isrc_table_lock);
523 		return (EBUSY);
524 	}
525 	isrc->isrc_filter = filter;
526 	isrc->isrc_arg = arg;
527 	isrc_update_name(isrc, name);
528 	mtx_unlock(&isrc_table_lock);
529 
530 	*cookiep = isrc;
531 	return (0);
532 }
533 #endif
534 
535 /*
536  *  Interrupt source pre_ithread method for MI interrupt framework.
537  */
538 static void
539 intr_isrc_pre_ithread(void *arg)
540 {
541 	struct intr_irqsrc *isrc = arg;
542 
543 	PIC_PRE_ITHREAD(isrc->isrc_dev, isrc);
544 }
545 
546 /*
547  *  Interrupt source post_ithread method for MI interrupt framework.
548  */
549 static void
550 intr_isrc_post_ithread(void *arg)
551 {
552 	struct intr_irqsrc *isrc = arg;
553 
554 	PIC_POST_ITHREAD(isrc->isrc_dev, isrc);
555 }
556 
557 /*
558  *  Interrupt source post_filter method for MI interrupt framework.
559  */
560 static void
561 intr_isrc_post_filter(void *arg)
562 {
563 	struct intr_irqsrc *isrc = arg;
564 
565 	PIC_POST_FILTER(isrc->isrc_dev, isrc);
566 }
567 
568 /*
569  *  Interrupt source assign_cpu method for MI interrupt framework.
570  */
571 static int
572 intr_isrc_assign_cpu(void *arg, int cpu)
573 {
574 #ifdef SMP
575 	struct intr_irqsrc *isrc = arg;
576 	int error;
577 
578 	if (isrc->isrc_dev != intr_irq_root_dev)
579 		return (EINVAL);
580 
581 	mtx_lock(&isrc_table_lock);
582 	if (cpu == NOCPU) {
583 		CPU_ZERO(&isrc->isrc_cpu);
584 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
585 	} else {
586 		CPU_SETOF(cpu, &isrc->isrc_cpu);
587 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
588 	}
589 
590 	/*
591 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
592 	 * re-balance it to another CPU or enable it on more CPUs. However,
593 	 * PIC is expected to change isrc_cpu appropriately to keep us well
594 	 * informed if the call is successful.
595 	 */
596 	if (irq_assign_cpu) {
597 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
598 		if (error) {
599 			CPU_ZERO(&isrc->isrc_cpu);
600 			mtx_unlock(&isrc_table_lock);
601 			return (error);
602 		}
603 	}
604 	mtx_unlock(&isrc_table_lock);
605 	return (0);
606 #else
607 	return (EOPNOTSUPP);
608 #endif
609 }
610 
611 /*
612  *  Create interrupt event for interrupt source.
613  */
614 static int
615 isrc_event_create(struct intr_irqsrc *isrc)
616 {
617 	struct intr_event *ie;
618 	int error;
619 
620 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
621 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
622 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
623 	if (error)
624 		return (error);
625 
626 	mtx_lock(&isrc_table_lock);
627 	/*
628 	 * Make sure that we do not mix the two ways
629 	 * how we handle interrupt sources. Let contested event wins.
630 	 */
631 #ifdef INTR_SOLO
632 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
633 #else
634 	if (isrc->isrc_event != NULL) {
635 #endif
636 		mtx_unlock(&isrc_table_lock);
637 		intr_event_destroy(ie);
638 		return (isrc->isrc_event != NULL ? EBUSY : 0);
639 	}
640 	isrc->isrc_event = ie;
641 	mtx_unlock(&isrc_table_lock);
642 
643 	return (0);
644 }
645 #ifdef notyet
646 /*
647  *  Destroy interrupt event for interrupt source.
648  */
649 static void
650 isrc_event_destroy(struct intr_irqsrc *isrc)
651 {
652 	struct intr_event *ie;
653 
654 	mtx_lock(&isrc_table_lock);
655 	ie = isrc->isrc_event;
656 	isrc->isrc_event = NULL;
657 	mtx_unlock(&isrc_table_lock);
658 
659 	if (ie != NULL)
660 		intr_event_destroy(ie);
661 }
662 #endif
663 /*
664  *  Add handler to interrupt source.
665  */
666 static int
667 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
668     driver_filter_t filter, driver_intr_t handler, void *arg,
669     enum intr_type flags, void **cookiep)
670 {
671 	int error;
672 
673 	if (isrc->isrc_event == NULL) {
674 		error = isrc_event_create(isrc);
675 		if (error)
676 			return (error);
677 	}
678 
679 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
680 	    arg, intr_priority(flags), flags, cookiep);
681 	if (error == 0) {
682 		mtx_lock(&isrc_table_lock);
683 		intrcnt_updatename(isrc);
684 		mtx_unlock(&isrc_table_lock);
685 	}
686 
687 	return (error);
688 }
689 
690 /*
691  *  Lookup interrupt controller locked.
692  */
693 static inline struct intr_pic *
694 pic_lookup_locked(device_t dev, intptr_t xref, int flags)
695 {
696 	struct intr_pic *pic;
697 
698 	mtx_assert(&pic_list_lock, MA_OWNED);
699 
700 	if (dev == NULL && xref == 0)
701 		return (NULL);
702 
703 	/* Note that pic->pic_dev is never NULL on registered PIC. */
704 	SLIST_FOREACH(pic, &pic_list, pic_next) {
705 		if ((pic->pic_flags & FLAG_TYPE_MASK) !=
706 		    (flags & FLAG_TYPE_MASK))
707 			continue;
708 
709 		if (dev == NULL) {
710 			if (xref == pic->pic_xref)
711 				return (pic);
712 		} else if (xref == 0 || pic->pic_xref == 0) {
713 			if (dev == pic->pic_dev)
714 				return (pic);
715 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
716 				return (pic);
717 	}
718 	return (NULL);
719 }
720 
721 /*
722  *  Lookup interrupt controller.
723  */
724 static struct intr_pic *
725 pic_lookup(device_t dev, intptr_t xref, int flags)
726 {
727 	struct intr_pic *pic;
728 
729 	mtx_lock(&pic_list_lock);
730 	pic = pic_lookup_locked(dev, xref, flags);
731 	mtx_unlock(&pic_list_lock);
732 	return (pic);
733 }
734 
735 /*
736  *  Create interrupt controller.
737  */
738 static struct intr_pic *
739 pic_create(device_t dev, intptr_t xref, int flags)
740 {
741 	struct intr_pic *pic;
742 
743 	mtx_lock(&pic_list_lock);
744 	pic = pic_lookup_locked(dev, xref, flags);
745 	if (pic != NULL) {
746 		mtx_unlock(&pic_list_lock);
747 		return (pic);
748 	}
749 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
750 	if (pic == NULL) {
751 		mtx_unlock(&pic_list_lock);
752 		return (NULL);
753 	}
754 	pic->pic_xref = xref;
755 	pic->pic_dev = dev;
756 	pic->pic_flags = flags;
757 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
758 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
759 	mtx_unlock(&pic_list_lock);
760 
761 	return (pic);
762 }
763 #ifdef notyet
764 /*
765  *  Destroy interrupt controller.
766  */
767 static void
768 pic_destroy(device_t dev, intptr_t xref, int flags)
769 {
770 	struct intr_pic *pic;
771 
772 	mtx_lock(&pic_list_lock);
773 	pic = pic_lookup_locked(dev, xref, flags);
774 	if (pic == NULL) {
775 		mtx_unlock(&pic_list_lock);
776 		return;
777 	}
778 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
779 	mtx_unlock(&pic_list_lock);
780 
781 	free(pic, M_INTRNG);
782 }
783 #endif
784 /*
785  *  Register interrupt controller.
786  */
787 struct intr_pic *
788 intr_pic_register(device_t dev, intptr_t xref)
789 {
790 	struct intr_pic *pic;
791 
792 	if (dev == NULL)
793 		return (NULL);
794 	pic = pic_create(dev, xref, FLAG_PIC);
795 	if (pic == NULL)
796 		return (NULL);
797 
798 	debugf("PIC %p registered for %s <dev %p, xref %x>\n", pic,
799 	    device_get_nameunit(dev), dev, xref);
800 	return (pic);
801 }
802 
803 /*
804  *  Unregister interrupt controller.
805  */
806 int
807 intr_pic_deregister(device_t dev, intptr_t xref)
808 {
809 
810 	panic("%s: not implemented", __func__);
811 }
812 
813 /*
814  *  Mark interrupt controller (itself) as a root one.
815  *
816  *  Note that only an interrupt controller can really know its position
817  *  in interrupt controller's tree. So root PIC must claim itself as a root.
818  *
819  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
820  *  page 30:
821  *    "The root of the interrupt tree is determined when traversal
822  *     of the interrupt tree reaches an interrupt controller node without
823  *     an interrupts property and thus no explicit interrupt parent."
824  */
825 int
826 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
827     void *arg, u_int ipicount)
828 {
829 	struct intr_pic *pic;
830 
831 	pic = pic_lookup(dev, xref, FLAG_PIC);
832 	if (pic == NULL) {
833 		device_printf(dev, "not registered\n");
834 		return (EINVAL);
835 	}
836 
837 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
838 	    ("%s: Found a non-PIC controller: %s", __func__,
839 	     device_get_name(pic->pic_dev)));
840 
841 	if (filter == NULL) {
842 		device_printf(dev, "filter missing\n");
843 		return (EINVAL);
844 	}
845 
846 	/*
847 	 * Only one interrupt controllers could be on the root for now.
848 	 * Note that we further suppose that there is not threaded interrupt
849 	 * routine (handler) on the root. See intr_irq_handler().
850 	 */
851 	if (intr_irq_root_dev != NULL) {
852 		device_printf(dev, "another root already set\n");
853 		return (EBUSY);
854 	}
855 
856 	intr_irq_root_dev = dev;
857 	irq_root_filter = filter;
858 	irq_root_arg = arg;
859 	irq_root_ipicount = ipicount;
860 
861 	debugf("irq root set to %s\n", device_get_nameunit(dev));
862 	return (0);
863 }
864 
865 /*
866  * Add a handler to manage a sub range of a parents interrupts.
867  */
868 struct intr_pic *
869 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
870     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
871     uintptr_t length)
872 {
873 	struct intr_pic *parent_pic;
874 	struct intr_pic_child *newchild;
875 #ifdef INVARIANTS
876 	struct intr_pic_child *child;
877 #endif
878 
879 	/* Find the parent PIC */
880 	parent_pic = pic_lookup(parent, 0, FLAG_PIC);
881 	if (parent_pic == NULL)
882 		return (NULL);
883 
884 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
885 	newchild->pc_pic = pic;
886 	newchild->pc_filter = filter;
887 	newchild->pc_filter_arg = arg;
888 	newchild->pc_start = start;
889 	newchild->pc_length = length;
890 
891 	mtx_lock_spin(&parent_pic->pic_child_lock);
892 #ifdef INVARIANTS
893 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
894 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
895 		    __func__));
896 	}
897 #endif
898 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
899 	mtx_unlock_spin(&parent_pic->pic_child_lock);
900 
901 	return (pic);
902 }
903 
904 static int
905 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
906     struct intr_irqsrc **isrc)
907 {
908 	struct intr_pic *pic;
909 	struct intr_map_data_msi *msi;
910 
911 	if (data == NULL)
912 		return (EINVAL);
913 
914 	pic = pic_lookup(dev, xref,
915 	    (data->type == INTR_MAP_DATA_MSI) ? FLAG_MSI : FLAG_PIC);
916 	if (pic == NULL)
917 		return (ESRCH);
918 
919 	switch (data->type) {
920 	case INTR_MAP_DATA_MSI:
921 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
922 		    ("%s: Found a non-MSI controller: %s", __func__,
923 		     device_get_name(pic->pic_dev)));
924 		msi = (struct intr_map_data_msi *)data;
925 		*isrc = msi->isrc;
926 		return (0);
927 
928 	default:
929 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
930 		    ("%s: Found a non-PIC controller: %s", __func__,
931 		     device_get_name(pic->pic_dev)));
932 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
933 
934 	}
935 }
936 
937 int
938 intr_activate_irq(device_t dev, struct resource *res)
939 {
940 	device_t map_dev;
941 	intptr_t map_xref;
942 	struct intr_map_data *data;
943 	struct intr_irqsrc *isrc;
944 	u_int res_id;
945 	int error;
946 
947 	KASSERT(rman_get_start(res) == rman_get_end(res),
948 	    ("%s: more interrupts in resource", __func__));
949 
950 	res_id = (u_int)rman_get_start(res);
951 	if (intr_map_get_isrc(res_id) != NULL)
952 		panic("Attempt to double activation of resource id: %u\n",
953 		    res_id);
954 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
955 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
956 	if (error != 0) {
957 		free(data, M_INTRNG);
958 		/* XXX TODO DISCONECTED PICs */
959 		/* if (error == EINVAL) return(0); */
960 		return (error);
961 	}
962 	intr_map_set_isrc(res_id, isrc);
963 	rman_set_virtual(res, data);
964 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
965 }
966 
967 int
968 intr_deactivate_irq(device_t dev, struct resource *res)
969 {
970 	struct intr_map_data *data;
971 	struct intr_irqsrc *isrc;
972 	u_int res_id;
973 	int error;
974 
975 	KASSERT(rman_get_start(res) == rman_get_end(res),
976 	    ("%s: more interrupts in resource", __func__));
977 
978 	res_id = (u_int)rman_get_start(res);
979 	isrc = intr_map_get_isrc(res_id);
980 	if (isrc == NULL)
981 		panic("Attempt to deactivate non-active resource id: %u\n",
982 		    res_id);
983 
984 	data = rman_get_virtual(res);
985 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
986 	intr_map_set_isrc(res_id, NULL);
987 	rman_set_virtual(res, NULL);
988 	free(data, M_INTRNG);
989 	return (error);
990 }
991 
992 int
993 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
994     driver_intr_t hand, void *arg, int flags, void **cookiep)
995 {
996 	int error;
997 	struct intr_map_data *data;
998 	struct intr_irqsrc *isrc;
999 	const char *name;
1000 	u_int res_id;
1001 
1002 	KASSERT(rman_get_start(res) == rman_get_end(res),
1003 	    ("%s: more interrupts in resource", __func__));
1004 
1005 	res_id = (u_int)rman_get_start(res);
1006 	isrc = intr_map_get_isrc(res_id);
1007 	if (isrc == NULL) {
1008 		/* XXX TODO DISCONECTED PICs */
1009 		return (EINVAL);
1010 	}
1011 
1012 	data = rman_get_virtual(res);
1013 	name = device_get_nameunit(dev);
1014 
1015 #ifdef INTR_SOLO
1016 	/*
1017 	 * Standard handling is done through MI interrupt framework. However,
1018 	 * some interrupts could request solely own special handling. This
1019 	 * non standard handling can be used for interrupt controllers without
1020 	 * handler (filter only), so in case that interrupt controllers are
1021 	 * chained, MI interrupt framework is called only in leaf controller.
1022 	 *
1023 	 * Note that root interrupt controller routine is served as well,
1024 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1025 	 */
1026 	if (flags & INTR_SOLO && hand != NULL) {
1027 		debugf("irq %u cannot solo on %s\n", irq, name);
1028 		return (EINVAL);
1029 	}
1030 
1031 	if (flags & INTR_SOLO) {
1032 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1033 		    arg, cookiep);
1034 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1035 		    name);
1036 	} else
1037 #endif
1038 		{
1039 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1040 		    cookiep);
1041 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1042 	}
1043 	if (error != 0)
1044 		return (error);
1045 
1046 	mtx_lock(&isrc_table_lock);
1047 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1048 	if (error == 0) {
1049 		isrc->isrc_handlers++;
1050 		if (isrc->isrc_handlers == 1)
1051 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1052 	}
1053 	mtx_unlock(&isrc_table_lock);
1054 	if (error != 0)
1055 		intr_event_remove_handler(*cookiep);
1056 	return (error);
1057 }
1058 
1059 int
1060 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1061 {
1062 	int error;
1063 	struct intr_map_data *data;
1064 	struct intr_irqsrc *isrc;
1065 	u_int res_id;
1066 
1067 	KASSERT(rman_get_start(res) == rman_get_end(res),
1068 	    ("%s: more interrupts in resource", __func__));
1069 
1070 	res_id = (u_int)rman_get_start(res);
1071 	isrc = intr_map_get_isrc(res_id);
1072 	if (isrc == NULL || isrc->isrc_handlers == 0)
1073 		return (EINVAL);
1074 
1075 	data = rman_get_virtual(res);
1076 
1077 #ifdef INTR_SOLO
1078 	if (isrc->isrc_filter != NULL) {
1079 		if (isrc != cookie)
1080 			return (EINVAL);
1081 
1082 		mtx_lock(&isrc_table_lock);
1083 		isrc->isrc_filter = NULL;
1084 		isrc->isrc_arg = NULL;
1085 		isrc->isrc_handlers = 0;
1086 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1087 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1088 		isrc_update_name(isrc, NULL);
1089 		mtx_unlock(&isrc_table_lock);
1090 		return (0);
1091 	}
1092 #endif
1093 	if (isrc != intr_handler_source(cookie))
1094 		return (EINVAL);
1095 
1096 	error = intr_event_remove_handler(cookie);
1097 	if (error == 0) {
1098 		mtx_lock(&isrc_table_lock);
1099 		isrc->isrc_handlers--;
1100 		if (isrc->isrc_handlers == 0)
1101 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1102 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1103 		intrcnt_updatename(isrc);
1104 		mtx_unlock(&isrc_table_lock);
1105 	}
1106 	return (error);
1107 }
1108 
1109 int
1110 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1111     const char *descr)
1112 {
1113 	int error;
1114 	struct intr_irqsrc *isrc;
1115 	u_int res_id;
1116 
1117 	KASSERT(rman_get_start(res) == rman_get_end(res),
1118 	    ("%s: more interrupts in resource", __func__));
1119 
1120 	res_id = (u_int)rman_get_start(res);
1121 	isrc = intr_map_get_isrc(res_id);
1122 	if (isrc == NULL || isrc->isrc_handlers == 0)
1123 		return (EINVAL);
1124 #ifdef INTR_SOLO
1125 	if (isrc->isrc_filter != NULL) {
1126 		if (isrc != cookie)
1127 			return (EINVAL);
1128 
1129 		mtx_lock(&isrc_table_lock);
1130 		isrc_update_name(isrc, descr);
1131 		mtx_unlock(&isrc_table_lock);
1132 		return (0);
1133 	}
1134 #endif
1135 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1136 	if (error == 0) {
1137 		mtx_lock(&isrc_table_lock);
1138 		intrcnt_updatename(isrc);
1139 		mtx_unlock(&isrc_table_lock);
1140 	}
1141 	return (error);
1142 }
1143 
1144 #ifdef SMP
1145 int
1146 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1147 {
1148 	struct intr_irqsrc *isrc;
1149 	u_int res_id;
1150 
1151 	KASSERT(rman_get_start(res) == rman_get_end(res),
1152 	    ("%s: more interrupts in resource", __func__));
1153 
1154 	res_id = (u_int)rman_get_start(res);
1155 	isrc = intr_map_get_isrc(res_id);
1156 	if (isrc == NULL || isrc->isrc_handlers == 0)
1157 		return (EINVAL);
1158 #ifdef INTR_SOLO
1159 	if (isrc->isrc_filter != NULL)
1160 		return (intr_isrc_assign_cpu(isrc, cpu));
1161 #endif
1162 	return (intr_event_bind(isrc->isrc_event, cpu));
1163 }
1164 
1165 /*
1166  * Return the CPU that the next interrupt source should use.
1167  * For now just returns the next CPU according to round-robin.
1168  */
1169 u_int
1170 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1171 {
1172 	u_int cpu;
1173 
1174 	KASSERT(!CPU_EMPTY(cpumask), ("%s: Empty CPU mask", __func__));
1175 	if (!irq_assign_cpu || mp_ncpus == 1) {
1176 		cpu = PCPU_GET(cpuid);
1177 
1178 		if (CPU_ISSET(cpu, cpumask))
1179 			return (curcpu);
1180 
1181 		return (CPU_FFS(cpumask) - 1);
1182 	}
1183 
1184 	do {
1185 		last_cpu++;
1186 		if (last_cpu > mp_maxid)
1187 			last_cpu = 0;
1188 	} while (!CPU_ISSET(last_cpu, cpumask));
1189 	return (last_cpu);
1190 }
1191 
1192 /*
1193  *  Distribute all the interrupt sources among the available
1194  *  CPUs once the AP's have been launched.
1195  */
1196 static void
1197 intr_irq_shuffle(void *arg __unused)
1198 {
1199 	struct intr_irqsrc *isrc;
1200 	u_int i;
1201 
1202 	if (mp_ncpus == 1)
1203 		return;
1204 
1205 	mtx_lock(&isrc_table_lock);
1206 	irq_assign_cpu = TRUE;
1207 	for (i = 0; i < NIRQ; i++) {
1208 		isrc = irq_sources[i];
1209 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1210 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1211 			continue;
1212 
1213 		if (isrc->isrc_event != NULL &&
1214 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1215 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1216 			panic("%s: CPU inconsistency", __func__);
1217 
1218 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1219 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1220 
1221 		/*
1222 		 * We are in wicked position here if the following call fails
1223 		 * for bound ISRC. The best thing we can do is to clear
1224 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1225 		 */
1226 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1227 			CPU_ZERO(&isrc->isrc_cpu);
1228 	}
1229 	mtx_unlock(&isrc_table_lock);
1230 }
1231 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1232 
1233 #else
1234 u_int
1235 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1236 {
1237 
1238 	return (PCPU_GET(cpuid));
1239 }
1240 #endif
1241 
1242 /*
1243  * Allocate memory for new intr_map_data structure.
1244  * Initialize common fields.
1245  */
1246 struct intr_map_data *
1247 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1248 {
1249 	struct intr_map_data *data;
1250 
1251 	data = malloc(len, M_INTRNG, flags);
1252 	data->type = type;
1253 	data->len = len;
1254 	return (data);
1255 }
1256 
1257 void intr_free_intr_map_data(struct intr_map_data *data)
1258 {
1259 
1260 	free(data, M_INTRNG);
1261 }
1262 
1263 
1264 /*
1265  *  Register a MSI/MSI-X interrupt controller
1266  */
1267 int
1268 intr_msi_register(device_t dev, intptr_t xref)
1269 {
1270 	struct intr_pic *pic;
1271 
1272 	if (dev == NULL)
1273 		return (EINVAL);
1274 	pic = pic_create(dev, xref, FLAG_MSI);
1275 	if (pic == NULL)
1276 		return (ENOMEM);
1277 
1278 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1279 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1280 	return (0);
1281 }
1282 
1283 int
1284 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1285     int maxcount, int *irqs)
1286 {
1287 	struct intr_irqsrc **isrc;
1288 	struct intr_pic *pic;
1289 	device_t pdev;
1290 	struct intr_map_data_msi *msi;
1291 	int err, i;
1292 
1293 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1294 	if (pic == NULL)
1295 		return (ESRCH);
1296 
1297 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1298 	    ("%s: Found a non-MSI controller: %s", __func__,
1299 	     device_get_name(pic->pic_dev)));
1300 
1301 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1302 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1303 	if (err != 0) {
1304 		free(isrc, M_INTRNG);
1305 		return (err);
1306 	}
1307 
1308 	for (i = 0; i < count; i++) {
1309 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1310 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1311 		msi-> isrc = isrc[i];
1312 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1313 		    (struct intr_map_data *)msi);
1314 
1315 	}
1316 	free(isrc, M_INTRNG);
1317 
1318 	return (err);
1319 }
1320 
1321 int
1322 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1323     int *irqs)
1324 {
1325 	struct intr_irqsrc **isrc;
1326 	struct intr_pic *pic;
1327 	struct intr_map_data_msi *msi;
1328 	int i, err;
1329 
1330 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1331 	if (pic == NULL)
1332 		return (ESRCH);
1333 
1334 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1335 	    ("%s: Found a non-MSI controller: %s", __func__,
1336 	     device_get_name(pic->pic_dev)));
1337 
1338 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1339 
1340 	for (i = 0; i < count; i++) {
1341 		msi = (struct intr_map_data_msi *)
1342 		    intr_map_get_map_data(irqs[i]);
1343 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1344 		    ("%s: irq %d map data is not MSI", __func__,
1345 		    irqs[i]));
1346 		isrc[i] = msi->isrc;
1347 	}
1348 
1349 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1350 
1351 	for (i = 0; i < count; i++) {
1352 		if (isrc[i] != NULL)
1353 			intr_unmap_irq(irqs[i]);
1354 	}
1355 
1356 	free(isrc, M_INTRNG);
1357 	return (err);
1358 }
1359 
1360 int
1361 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1362 {
1363 	struct intr_irqsrc *isrc;
1364 	struct intr_pic *pic;
1365 	device_t pdev;
1366 	struct intr_map_data_msi *msi;
1367 	int err;
1368 
1369 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1370 	if (pic == NULL)
1371 		return (ESRCH);
1372 
1373 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1374 	    ("%s: Found a non-MSI controller: %s", __func__,
1375 	     device_get_name(pic->pic_dev)));
1376 
1377 
1378 	err = MSI_ALLOC_MSIX(pic->pic_dev, child, &pdev, &isrc);
1379 	if (err != 0)
1380 		return (err);
1381 
1382 	msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1383 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1384 	msi->isrc = isrc;
1385 	*irq = intr_map_irq(pic->pic_dev, xref, (struct intr_map_data *)msi);
1386 	return (0);
1387 }
1388 
1389 int
1390 intr_release_msix(device_t pci, device_t child, intptr_t xref, int irq)
1391 {
1392 	struct intr_irqsrc *isrc;
1393 	struct intr_pic *pic;
1394 	struct intr_map_data_msi *msi;
1395 	int err;
1396 
1397 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1398 	if (pic == NULL)
1399 		return (ESRCH);
1400 
1401 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1402 	    ("%s: Found a non-MSI controller: %s", __func__,
1403 	     device_get_name(pic->pic_dev)));
1404 
1405 	msi = (struct intr_map_data_msi *)
1406 	    intr_map_get_map_data(irq);
1407 	KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1408 	    ("%s: irq %d map data is not MSI", __func__,
1409 	    irq));
1410 	isrc = msi->isrc;
1411 	if (isrc == NULL) {
1412 		intr_unmap_irq(irq);
1413 		return (EINVAL);
1414 	}
1415 
1416 	err = MSI_RELEASE_MSIX(pic->pic_dev, child, isrc);
1417 	intr_unmap_irq(irq);
1418 
1419 	return (err);
1420 }
1421 
1422 int
1423 intr_map_msi(device_t pci, device_t child, intptr_t xref, int irq,
1424     uint64_t *addr, uint32_t *data)
1425 {
1426 	struct intr_irqsrc *isrc;
1427 	struct intr_pic *pic;
1428 	int err;
1429 
1430 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1431 	if (pic == NULL)
1432 		return (ESRCH);
1433 
1434 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1435 	    ("%s: Found a non-MSI controller: %s", __func__,
1436 	     device_get_name(pic->pic_dev)));
1437 
1438 	isrc = intr_map_get_isrc(irq);
1439 	if (isrc == NULL)
1440 		return (EINVAL);
1441 
1442 	err = MSI_MAP_MSI(pic->pic_dev, child, isrc, addr, data);
1443 	return (err);
1444 }
1445 
1446 
1447 void dosoftints(void);
1448 void
1449 dosoftints(void)
1450 {
1451 }
1452 
1453 #ifdef SMP
1454 /*
1455  *  Init interrupt controller on another CPU.
1456  */
1457 void
1458 intr_pic_init_secondary(void)
1459 {
1460 
1461 	/*
1462 	 * QQQ: Only root PIC is aware of other CPUs ???
1463 	 */
1464 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1465 
1466 	//mtx_lock(&isrc_table_lock);
1467 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1468 	//mtx_unlock(&isrc_table_lock);
1469 }
1470 #endif
1471 
1472 #ifdef DDB
1473 DB_SHOW_COMMAND(irqs, db_show_irqs)
1474 {
1475 	u_int i, irqsum;
1476 	u_long num;
1477 	struct intr_irqsrc *isrc;
1478 
1479 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1480 		isrc = irq_sources[i];
1481 		if (isrc == NULL)
1482 			continue;
1483 
1484 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1485 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1486 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1487 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1488 		irqsum += num;
1489 	}
1490 	db_printf("irq total %u\n", irqsum);
1491 }
1492 #endif
1493 
1494 /*
1495  * Interrupt mapping table functions.
1496  *
1497  * Please, keep this part separately, it can be transformed to
1498  * extension of standard resources.
1499  */
1500 struct intr_map_entry
1501 {
1502 	device_t 		dev;
1503 	intptr_t 		xref;
1504 	struct intr_map_data 	*map_data;
1505 	struct intr_irqsrc 	*isrc;
1506 	/* XXX TODO DISCONECTED PICs */
1507 	/*int			flags */
1508 };
1509 
1510 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1511 static struct intr_map_entry *irq_map[2 * NIRQ];
1512 static int irq_map_count = nitems(irq_map);
1513 static int irq_map_first_free_idx;
1514 static struct mtx irq_map_lock;
1515 
1516 static struct intr_irqsrc *
1517 intr_map_get_isrc(u_int res_id)
1518 {
1519 	struct intr_irqsrc *isrc;
1520 
1521 	mtx_lock(&irq_map_lock);
1522 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1523 		mtx_unlock(&irq_map_lock);
1524 		return (NULL);
1525 	}
1526 	isrc = irq_map[res_id]->isrc;
1527 	mtx_unlock(&irq_map_lock);
1528 	return (isrc);
1529 }
1530 
1531 static void
1532 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1533 {
1534 
1535 	mtx_lock(&irq_map_lock);
1536 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1537 		mtx_unlock(&irq_map_lock);
1538 		return;
1539 	}
1540 	irq_map[res_id]->isrc = isrc;
1541 	mtx_unlock(&irq_map_lock);
1542 }
1543 
1544 /*
1545  * Get a copy of intr_map_entry data
1546  */
1547 static struct intr_map_data *
1548 intr_map_get_map_data(u_int res_id)
1549 {
1550 	struct intr_map_data *data;
1551 
1552 	data = NULL;
1553 	mtx_lock(&irq_map_lock);
1554 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1555 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1556 	data = irq_map[res_id]->map_data;
1557 	mtx_unlock(&irq_map_lock);
1558 
1559 	return (data);
1560 }
1561 
1562 /*
1563  * Get a copy of intr_map_entry data
1564  */
1565 static void
1566 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1567     struct intr_map_data **data)
1568 {
1569 	size_t len;
1570 
1571 	len = 0;
1572 	mtx_lock(&irq_map_lock);
1573 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1574 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1575 	if (irq_map[res_id]->map_data != NULL)
1576 		len = irq_map[res_id]->map_data->len;
1577 	mtx_unlock(&irq_map_lock);
1578 
1579 	if (len == 0)
1580 		*data = NULL;
1581 	else
1582 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1583 	mtx_lock(&irq_map_lock);
1584 	if (irq_map[res_id] == NULL)
1585 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1586 	if (len != 0) {
1587 		if (len != irq_map[res_id]->map_data->len)
1588 			panic("Resource id: %u has changed.\n", res_id);
1589 		memcpy(*data, irq_map[res_id]->map_data, len);
1590 	}
1591 	*map_dev = irq_map[res_id]->dev;
1592 	*map_xref = irq_map[res_id]->xref;
1593 	mtx_unlock(&irq_map_lock);
1594 }
1595 
1596 
1597 /*
1598  * Allocate and fill new entry in irq_map table.
1599  */
1600 u_int
1601 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1602 {
1603 	u_int i;
1604 	struct intr_map_entry *entry;
1605 
1606 	/* Prepare new entry first. */
1607 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1608 
1609 	entry->dev = dev;
1610 	entry->xref = xref;
1611 	entry->map_data = data;
1612 	entry->isrc = NULL;
1613 
1614 	mtx_lock(&irq_map_lock);
1615 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1616 		if (irq_map[i] == NULL) {
1617 			irq_map[i] = entry;
1618 			irq_map_first_free_idx = i + 1;
1619 			mtx_unlock(&irq_map_lock);
1620 			return (i);
1621 		}
1622 	}
1623 	mtx_unlock(&irq_map_lock);
1624 
1625 	/* XXX Expand irq_map table */
1626 	panic("IRQ mapping table is full.");
1627 }
1628 
1629 /*
1630  * Remove and free mapping entry.
1631  */
1632 void
1633 intr_unmap_irq(u_int res_id)
1634 {
1635 	struct intr_map_entry *entry;
1636 
1637 	mtx_lock(&irq_map_lock);
1638 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1639 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1640 	entry = irq_map[res_id];
1641 	irq_map[res_id] = NULL;
1642 	irq_map_first_free_idx = res_id;
1643 	mtx_unlock(&irq_map_lock);
1644 	intr_free_intr_map_data(entry->map_data);
1645 	free(entry, M_INTRNG);
1646 }
1647 
1648 /*
1649  * Clone mapping entry.
1650  */
1651 u_int
1652 intr_map_clone_irq(u_int old_res_id)
1653 {
1654 	device_t map_dev;
1655 	intptr_t map_xref;
1656 	struct intr_map_data *data;
1657 
1658 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1659 	return (intr_map_irq(map_dev, map_xref, data));
1660 }
1661 
1662 static void
1663 intr_map_init(void *dummy __unused)
1664 {
1665 
1666 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1667 }
1668 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1669