xref: /freebsd/sys/kern/subr_intr.c (revision 6419bb52)
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/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/syslog.h>
48 #include <sys/malloc.h>
49 #include <sys/proc.h>
50 #include <sys/queue.h>
51 #include <sys/bus.h>
52 #include <sys/interrupt.h>
53 #include <sys/conf.h>
54 #include <sys/cpuset.h>
55 #include <sys/rman.h>
56 #include <sys/sched.h>
57 #include <sys/smp.h>
58 #include <sys/vmmeter.h>
59 #ifdef HWPMC_HOOKS
60 #include <sys/pmckern.h>
61 #endif
62 
63 #include <machine/atomic.h>
64 #include <machine/intr.h>
65 #include <machine/cpu.h>
66 #include <machine/smp.h>
67 #include <machine/stdarg.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif
72 
73 #include "pic_if.h"
74 #include "msi_if.h"
75 
76 #define	INTRNAME_LEN	(2*MAXCOMLEN + 1)
77 
78 #ifdef DEBUG
79 #define debugf(fmt, args...) do { printf("%s(): ", __func__);	\
80     printf(fmt,##args); } while (0)
81 #else
82 #define debugf(fmt, args...)
83 #endif
84 
85 MALLOC_DECLARE(M_INTRNG);
86 MALLOC_DEFINE(M_INTRNG, "intr", "intr interrupt handling");
87 
88 /* Main interrupt handler called from assembler -> 'hidden' for C code. */
89 void intr_irq_handler(struct trapframe *tf);
90 
91 /* Root interrupt controller stuff. */
92 device_t intr_irq_root_dev;
93 static intr_irq_filter_t *irq_root_filter;
94 static void *irq_root_arg;
95 static u_int irq_root_ipicount;
96 
97 struct intr_pic_child {
98 	SLIST_ENTRY(intr_pic_child)	 pc_next;
99 	struct intr_pic			*pc_pic;
100 	intr_child_irq_filter_t		*pc_filter;
101 	void				*pc_filter_arg;
102 	uintptr_t			 pc_start;
103 	uintptr_t			 pc_length;
104 };
105 
106 /* Interrupt controller definition. */
107 struct intr_pic {
108 	SLIST_ENTRY(intr_pic)	pic_next;
109 	intptr_t		pic_xref;	/* hardware identification */
110 	device_t		pic_dev;
111 /* Only one of FLAG_PIC or FLAG_MSI may be set */
112 #define	FLAG_PIC	(1 << 0)
113 #define	FLAG_MSI	(1 << 1)
114 #define	FLAG_TYPE_MASK	(FLAG_PIC | FLAG_MSI)
115 	u_int			pic_flags;
116 	struct mtx		pic_child_lock;
117 	SLIST_HEAD(, intr_pic_child) pic_children;
118 };
119 
120 static struct mtx pic_list_lock;
121 static SLIST_HEAD(, intr_pic) pic_list;
122 
123 static struct intr_pic *pic_lookup(device_t dev, intptr_t xref, int flags);
124 
125 /* Interrupt source definition. */
126 static struct mtx isrc_table_lock;
127 static struct intr_irqsrc *irq_sources[NIRQ];
128 u_int irq_next_free;
129 
130 #ifdef SMP
131 static boolean_t irq_assign_cpu = FALSE;
132 #endif
133 
134 /*
135  * - 2 counters for each I/O interrupt.
136  * - MAXCPU counters for each IPI counters for SMP.
137  */
138 #ifdef SMP
139 #define INTRCNT_COUNT   (NIRQ * 2 + INTR_IPI_COUNT * MAXCPU)
140 #else
141 #define INTRCNT_COUNT   (NIRQ * 2)
142 #endif
143 
144 /* Data for MI statistics reporting. */
145 u_long intrcnt[INTRCNT_COUNT];
146 char intrnames[INTRCNT_COUNT * INTRNAME_LEN];
147 size_t sintrcnt = sizeof(intrcnt);
148 size_t sintrnames = sizeof(intrnames);
149 static u_int intrcnt_index;
150 
151 static struct intr_irqsrc *intr_map_get_isrc(u_int res_id);
152 static void intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc);
153 static struct intr_map_data * intr_map_get_map_data(u_int res_id);
154 static void intr_map_copy_map_data(u_int res_id, device_t *dev, intptr_t *xref,
155     struct intr_map_data **data);
156 
157 /*
158  *  Interrupt framework initialization routine.
159  */
160 static void
161 intr_irq_init(void *dummy __unused)
162 {
163 
164 	SLIST_INIT(&pic_list);
165 	mtx_init(&pic_list_lock, "intr pic list", NULL, MTX_DEF);
166 
167 	mtx_init(&isrc_table_lock, "intr isrc table", NULL, MTX_DEF);
168 }
169 SYSINIT(intr_irq_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_irq_init, NULL);
170 
171 static void
172 intrcnt_setname(const char *name, int index)
173 {
174 
175 	snprintf(intrnames + INTRNAME_LEN * index, INTRNAME_LEN, "%-*s",
176 	    INTRNAME_LEN - 1, name);
177 }
178 
179 /*
180  *  Update name for interrupt source with interrupt event.
181  */
182 static void
183 intrcnt_updatename(struct intr_irqsrc *isrc)
184 {
185 
186 	/* QQQ: What about stray counter name? */
187 	mtx_assert(&isrc_table_lock, MA_OWNED);
188 	intrcnt_setname(isrc->isrc_event->ie_fullname, isrc->isrc_index);
189 }
190 
191 /*
192  *  Virtualization for interrupt source interrupt counter increment.
193  */
194 static inline void
195 isrc_increment_count(struct intr_irqsrc *isrc)
196 {
197 
198 	if (isrc->isrc_flags & INTR_ISRCF_PPI)
199 		atomic_add_long(&isrc->isrc_count[0], 1);
200 	else
201 		isrc->isrc_count[0]++;
202 }
203 
204 /*
205  *  Virtualization for interrupt source interrupt stray counter increment.
206  */
207 static inline void
208 isrc_increment_straycount(struct intr_irqsrc *isrc)
209 {
210 
211 	isrc->isrc_count[1]++;
212 }
213 
214 /*
215  *  Virtualization for interrupt source interrupt name update.
216  */
217 static void
218 isrc_update_name(struct intr_irqsrc *isrc, const char *name)
219 {
220 	char str[INTRNAME_LEN];
221 
222 	mtx_assert(&isrc_table_lock, MA_OWNED);
223 
224 	if (name != NULL) {
225 		snprintf(str, INTRNAME_LEN, "%s: %s", isrc->isrc_name, name);
226 		intrcnt_setname(str, isrc->isrc_index);
227 		snprintf(str, INTRNAME_LEN, "stray %s: %s", isrc->isrc_name,
228 		    name);
229 		intrcnt_setname(str, isrc->isrc_index + 1);
230 	} else {
231 		snprintf(str, INTRNAME_LEN, "%s:", isrc->isrc_name);
232 		intrcnt_setname(str, isrc->isrc_index);
233 		snprintf(str, INTRNAME_LEN, "stray %s:", isrc->isrc_name);
234 		intrcnt_setname(str, isrc->isrc_index + 1);
235 	}
236 }
237 
238 /*
239  *  Virtualization for interrupt source interrupt counters setup.
240  */
241 static void
242 isrc_setup_counters(struct intr_irqsrc *isrc)
243 {
244 	u_int index;
245 
246 	/*
247 	 *  XXX - it does not work well with removable controllers and
248 	 *        interrupt sources !!!
249 	 */
250 	index = atomic_fetchadd_int(&intrcnt_index, 2);
251 	isrc->isrc_index = index;
252 	isrc->isrc_count = &intrcnt[index];
253 	isrc_update_name(isrc, NULL);
254 }
255 
256 /*
257  *  Virtualization for interrupt source interrupt counters release.
258  */
259 static void
260 isrc_release_counters(struct intr_irqsrc *isrc)
261 {
262 
263 	panic("%s: not implemented", __func__);
264 }
265 
266 #ifdef SMP
267 /*
268  *  Virtualization for interrupt source IPI counters setup.
269  */
270 u_long *
271 intr_ipi_setup_counters(const char *name)
272 {
273 	u_int index, i;
274 	char str[INTRNAME_LEN];
275 
276 	index = atomic_fetchadd_int(&intrcnt_index, MAXCPU);
277 	for (i = 0; i < MAXCPU; i++) {
278 		snprintf(str, INTRNAME_LEN, "cpu%d:%s", i, name);
279 		intrcnt_setname(str, index + i);
280 	}
281 	return (&intrcnt[index]);
282 }
283 #endif
284 
285 /*
286  *  Main interrupt dispatch handler. It's called straight
287  *  from the assembler, where CPU interrupt is served.
288  */
289 void
290 intr_irq_handler(struct trapframe *tf)
291 {
292 	struct trapframe * oldframe;
293 	struct thread * td;
294 
295 	KASSERT(irq_root_filter != NULL, ("%s: no filter", __func__));
296 
297 	VM_CNT_INC(v_intr);
298 	critical_enter();
299 	td = curthread;
300 	oldframe = td->td_intr_frame;
301 	td->td_intr_frame = tf;
302 	irq_root_filter(irq_root_arg);
303 	td->td_intr_frame = oldframe;
304 	critical_exit();
305 #ifdef HWPMC_HOOKS
306 	if (pmc_hook && TRAPF_USERMODE(tf) &&
307 	    (PCPU_GET(curthread)->td_pflags & TDP_CALLCHAIN))
308 		pmc_hook(PCPU_GET(curthread), PMC_FN_USER_CALLCHAIN, tf);
309 #endif
310 }
311 
312 int
313 intr_child_irq_handler(struct intr_pic *parent, uintptr_t irq)
314 {
315 	struct intr_pic_child *child;
316 	bool found;
317 
318 	found = false;
319 	mtx_lock_spin(&parent->pic_child_lock);
320 	SLIST_FOREACH(child, &parent->pic_children, pc_next) {
321 		if (child->pc_start <= irq &&
322 		    irq < (child->pc_start + child->pc_length)) {
323 			found = true;
324 			break;
325 		}
326 	}
327 	mtx_unlock_spin(&parent->pic_child_lock);
328 
329 	if (found)
330 		return (child->pc_filter(child->pc_filter_arg, irq));
331 
332 	return (FILTER_STRAY);
333 }
334 
335 /*
336  *  interrupt controller dispatch function for interrupts. It should
337  *  be called straight from the interrupt controller, when associated interrupt
338  *  source is learned.
339  */
340 int
341 intr_isrc_dispatch(struct intr_irqsrc *isrc, struct trapframe *tf)
342 {
343 
344 	KASSERT(isrc != NULL, ("%s: no source", __func__));
345 
346 	isrc_increment_count(isrc);
347 
348 #ifdef INTR_SOLO
349 	if (isrc->isrc_filter != NULL) {
350 		int error;
351 		error = isrc->isrc_filter(isrc->isrc_arg, tf);
352 		PIC_POST_FILTER(isrc->isrc_dev, isrc);
353 		if (error == FILTER_HANDLED)
354 			return (0);
355 	} else
356 #endif
357 	if (isrc->isrc_event != NULL) {
358 		if (intr_event_handle(isrc->isrc_event, tf) == 0)
359 			return (0);
360 	}
361 
362 	isrc_increment_straycount(isrc);
363 	return (EINVAL);
364 }
365 
366 /*
367  *  Alloc unique interrupt number (resource handle) for interrupt source.
368  *
369  *  There could be various strategies how to allocate free interrupt number
370  *  (resource handle) for new interrupt source.
371  *
372  *  1. Handles are always allocated forward, so handles are not recycled
373  *     immediately. However, if only one free handle left which is reused
374  *     constantly...
375  */
376 static inline int
377 isrc_alloc_irq(struct intr_irqsrc *isrc)
378 {
379 	u_int maxirqs, irq;
380 
381 	mtx_assert(&isrc_table_lock, MA_OWNED);
382 
383 	maxirqs = nitems(irq_sources);
384 	if (irq_next_free >= maxirqs)
385 		return (ENOSPC);
386 
387 	for (irq = irq_next_free; irq < maxirqs; irq++) {
388 		if (irq_sources[irq] == NULL)
389 			goto found;
390 	}
391 	for (irq = 0; irq < irq_next_free; irq++) {
392 		if (irq_sources[irq] == NULL)
393 			goto found;
394 	}
395 
396 	irq_next_free = maxirqs;
397 	return (ENOSPC);
398 
399 found:
400 	isrc->isrc_irq = irq;
401 	irq_sources[irq] = isrc;
402 
403 	irq_next_free = irq + 1;
404 	if (irq_next_free >= maxirqs)
405 		irq_next_free = 0;
406 	return (0);
407 }
408 
409 /*
410  *  Free unique interrupt number (resource handle) from interrupt source.
411  */
412 static inline int
413 isrc_free_irq(struct intr_irqsrc *isrc)
414 {
415 
416 	mtx_assert(&isrc_table_lock, MA_OWNED);
417 
418 	if (isrc->isrc_irq >= nitems(irq_sources))
419 		return (EINVAL);
420 	if (irq_sources[isrc->isrc_irq] != isrc)
421 		return (EINVAL);
422 
423 	irq_sources[isrc->isrc_irq] = NULL;
424 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
425 	return (0);
426 }
427 
428 /*
429  *  Initialize interrupt source and register it into global interrupt table.
430  */
431 int
432 intr_isrc_register(struct intr_irqsrc *isrc, device_t dev, u_int flags,
433     const char *fmt, ...)
434 {
435 	int error;
436 	va_list ap;
437 
438 	bzero(isrc, sizeof(struct intr_irqsrc));
439 	isrc->isrc_dev = dev;
440 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
441 	isrc->isrc_flags = flags;
442 
443 	va_start(ap, fmt);
444 	vsnprintf(isrc->isrc_name, INTR_ISRC_NAMELEN, fmt, ap);
445 	va_end(ap);
446 
447 	mtx_lock(&isrc_table_lock);
448 	error = isrc_alloc_irq(isrc);
449 	if (error != 0) {
450 		mtx_unlock(&isrc_table_lock);
451 		return (error);
452 	}
453 	/*
454 	 * Setup interrupt counters, but not for IPI sources. Those are setup
455 	 * later and only for used ones (up to INTR_IPI_COUNT) to not exhaust
456 	 * our counter pool.
457 	 */
458 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
459 		isrc_setup_counters(isrc);
460 	mtx_unlock(&isrc_table_lock);
461 	return (0);
462 }
463 
464 /*
465  *  Deregister interrupt source from global interrupt table.
466  */
467 int
468 intr_isrc_deregister(struct intr_irqsrc *isrc)
469 {
470 	int error;
471 
472 	mtx_lock(&isrc_table_lock);
473 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
474 		isrc_release_counters(isrc);
475 	error = isrc_free_irq(isrc);
476 	mtx_unlock(&isrc_table_lock);
477 	return (error);
478 }
479 
480 #ifdef SMP
481 /*
482  *  A support function for a PIC to decide if provided ISRC should be inited
483  *  on given cpu. The logic of INTR_ISRCF_BOUND flag and isrc_cpu member of
484  *  struct intr_irqsrc is the following:
485  *
486  *     If INTR_ISRCF_BOUND is set, the ISRC should be inited only on cpus
487  *     set in isrc_cpu. If not, the ISRC should be inited on every cpu and
488  *     isrc_cpu is kept consistent with it. Thus isrc_cpu is always correct.
489  */
490 bool
491 intr_isrc_init_on_cpu(struct intr_irqsrc *isrc, u_int cpu)
492 {
493 
494 	if (isrc->isrc_handlers == 0)
495 		return (false);
496 	if ((isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI)) == 0)
497 		return (false);
498 	if (isrc->isrc_flags & INTR_ISRCF_BOUND)
499 		return (CPU_ISSET(cpu, &isrc->isrc_cpu));
500 
501 	CPU_SET(cpu, &isrc->isrc_cpu);
502 	return (true);
503 }
504 #endif
505 
506 #ifdef INTR_SOLO
507 /*
508  *  Setup filter into interrupt source.
509  */
510 static int
511 iscr_setup_filter(struct intr_irqsrc *isrc, const char *name,
512     intr_irq_filter_t *filter, void *arg, void **cookiep)
513 {
514 
515 	if (filter == NULL)
516 		return (EINVAL);
517 
518 	mtx_lock(&isrc_table_lock);
519 	/*
520 	 * Make sure that we do not mix the two ways
521 	 * how we handle interrupt sources.
522 	 */
523 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
524 		mtx_unlock(&isrc_table_lock);
525 		return (EBUSY);
526 	}
527 	isrc->isrc_filter = filter;
528 	isrc->isrc_arg = arg;
529 	isrc_update_name(isrc, name);
530 	mtx_unlock(&isrc_table_lock);
531 
532 	*cookiep = isrc;
533 	return (0);
534 }
535 #endif
536 
537 /*
538  *  Interrupt source pre_ithread method for MI interrupt framework.
539  */
540 static void
541 intr_isrc_pre_ithread(void *arg)
542 {
543 	struct intr_irqsrc *isrc = arg;
544 
545 	PIC_PRE_ITHREAD(isrc->isrc_dev, isrc);
546 }
547 
548 /*
549  *  Interrupt source post_ithread method for MI interrupt framework.
550  */
551 static void
552 intr_isrc_post_ithread(void *arg)
553 {
554 	struct intr_irqsrc *isrc = arg;
555 
556 	PIC_POST_ITHREAD(isrc->isrc_dev, isrc);
557 }
558 
559 /*
560  *  Interrupt source post_filter method for MI interrupt framework.
561  */
562 static void
563 intr_isrc_post_filter(void *arg)
564 {
565 	struct intr_irqsrc *isrc = arg;
566 
567 	PIC_POST_FILTER(isrc->isrc_dev, isrc);
568 }
569 
570 /*
571  *  Interrupt source assign_cpu method for MI interrupt framework.
572  */
573 static int
574 intr_isrc_assign_cpu(void *arg, int cpu)
575 {
576 #ifdef SMP
577 	struct intr_irqsrc *isrc = arg;
578 	int error;
579 
580 	if (isrc->isrc_dev != intr_irq_root_dev)
581 		return (EINVAL);
582 
583 	mtx_lock(&isrc_table_lock);
584 	if (cpu == NOCPU) {
585 		CPU_ZERO(&isrc->isrc_cpu);
586 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
587 	} else {
588 		CPU_SETOF(cpu, &isrc->isrc_cpu);
589 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
590 	}
591 
592 	/*
593 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
594 	 * re-balance it to another CPU or enable it on more CPUs. However,
595 	 * PIC is expected to change isrc_cpu appropriately to keep us well
596 	 * informed if the call is successful.
597 	 */
598 	if (irq_assign_cpu) {
599 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
600 		if (error) {
601 			CPU_ZERO(&isrc->isrc_cpu);
602 			mtx_unlock(&isrc_table_lock);
603 			return (error);
604 		}
605 	}
606 	mtx_unlock(&isrc_table_lock);
607 	return (0);
608 #else
609 	return (EOPNOTSUPP);
610 #endif
611 }
612 
613 /*
614  *  Create interrupt event for interrupt source.
615  */
616 static int
617 isrc_event_create(struct intr_irqsrc *isrc)
618 {
619 	struct intr_event *ie;
620 	int error;
621 
622 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
623 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
624 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
625 	if (error)
626 		return (error);
627 
628 	mtx_lock(&isrc_table_lock);
629 	/*
630 	 * Make sure that we do not mix the two ways
631 	 * how we handle interrupt sources. Let contested event wins.
632 	 */
633 #ifdef INTR_SOLO
634 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
635 #else
636 	if (isrc->isrc_event != NULL) {
637 #endif
638 		mtx_unlock(&isrc_table_lock);
639 		intr_event_destroy(ie);
640 		return (isrc->isrc_event != NULL ? EBUSY : 0);
641 	}
642 	isrc->isrc_event = ie;
643 	mtx_unlock(&isrc_table_lock);
644 
645 	return (0);
646 }
647 #ifdef notyet
648 /*
649  *  Destroy interrupt event for interrupt source.
650  */
651 static void
652 isrc_event_destroy(struct intr_irqsrc *isrc)
653 {
654 	struct intr_event *ie;
655 
656 	mtx_lock(&isrc_table_lock);
657 	ie = isrc->isrc_event;
658 	isrc->isrc_event = NULL;
659 	mtx_unlock(&isrc_table_lock);
660 
661 	if (ie != NULL)
662 		intr_event_destroy(ie);
663 }
664 #endif
665 /*
666  *  Add handler to interrupt source.
667  */
668 static int
669 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
670     driver_filter_t filter, driver_intr_t handler, void *arg,
671     enum intr_type flags, void **cookiep)
672 {
673 	int error;
674 
675 	if (isrc->isrc_event == NULL) {
676 		error = isrc_event_create(isrc);
677 		if (error)
678 			return (error);
679 	}
680 
681 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
682 	    arg, intr_priority(flags), flags, cookiep);
683 	if (error == 0) {
684 		mtx_lock(&isrc_table_lock);
685 		intrcnt_updatename(isrc);
686 		mtx_unlock(&isrc_table_lock);
687 	}
688 
689 	return (error);
690 }
691 
692 /*
693  *  Lookup interrupt controller locked.
694  */
695 static inline struct intr_pic *
696 pic_lookup_locked(device_t dev, intptr_t xref, int flags)
697 {
698 	struct intr_pic *pic;
699 
700 	mtx_assert(&pic_list_lock, MA_OWNED);
701 
702 	if (dev == NULL && xref == 0)
703 		return (NULL);
704 
705 	/* Note that pic->pic_dev is never NULL on registered PIC. */
706 	SLIST_FOREACH(pic, &pic_list, pic_next) {
707 		if ((pic->pic_flags & FLAG_TYPE_MASK) !=
708 		    (flags & FLAG_TYPE_MASK))
709 			continue;
710 
711 		if (dev == NULL) {
712 			if (xref == pic->pic_xref)
713 				return (pic);
714 		} else if (xref == 0 || pic->pic_xref == 0) {
715 			if (dev == pic->pic_dev)
716 				return (pic);
717 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
718 				return (pic);
719 	}
720 	return (NULL);
721 }
722 
723 /*
724  *  Lookup interrupt controller.
725  */
726 static struct intr_pic *
727 pic_lookup(device_t dev, intptr_t xref, int flags)
728 {
729 	struct intr_pic *pic;
730 
731 	mtx_lock(&pic_list_lock);
732 	pic = pic_lookup_locked(dev, xref, flags);
733 	mtx_unlock(&pic_list_lock);
734 	return (pic);
735 }
736 
737 /*
738  *  Create interrupt controller.
739  */
740 static struct intr_pic *
741 pic_create(device_t dev, intptr_t xref, int flags)
742 {
743 	struct intr_pic *pic;
744 
745 	mtx_lock(&pic_list_lock);
746 	pic = pic_lookup_locked(dev, xref, flags);
747 	if (pic != NULL) {
748 		mtx_unlock(&pic_list_lock);
749 		return (pic);
750 	}
751 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
752 	if (pic == NULL) {
753 		mtx_unlock(&pic_list_lock);
754 		return (NULL);
755 	}
756 	pic->pic_xref = xref;
757 	pic->pic_dev = dev;
758 	pic->pic_flags = flags;
759 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
760 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
761 	mtx_unlock(&pic_list_lock);
762 
763 	return (pic);
764 }
765 #ifdef notyet
766 /*
767  *  Destroy interrupt controller.
768  */
769 static void
770 pic_destroy(device_t dev, intptr_t xref, int flags)
771 {
772 	struct intr_pic *pic;
773 
774 	mtx_lock(&pic_list_lock);
775 	pic = pic_lookup_locked(dev, xref, flags);
776 	if (pic == NULL) {
777 		mtx_unlock(&pic_list_lock);
778 		return;
779 	}
780 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
781 	mtx_unlock(&pic_list_lock);
782 
783 	free(pic, M_INTRNG);
784 }
785 #endif
786 /*
787  *  Register interrupt controller.
788  */
789 struct intr_pic *
790 intr_pic_register(device_t dev, intptr_t xref)
791 {
792 	struct intr_pic *pic;
793 
794 	if (dev == NULL)
795 		return (NULL);
796 	pic = pic_create(dev, xref, FLAG_PIC);
797 	if (pic == NULL)
798 		return (NULL);
799 
800 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
801 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
802 	return (pic);
803 }
804 
805 /*
806  *  Unregister interrupt controller.
807  */
808 int
809 intr_pic_deregister(device_t dev, intptr_t xref)
810 {
811 
812 	panic("%s: not implemented", __func__);
813 }
814 
815 /*
816  *  Mark interrupt controller (itself) as a root one.
817  *
818  *  Note that only an interrupt controller can really know its position
819  *  in interrupt controller's tree. So root PIC must claim itself as a root.
820  *
821  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
822  *  page 30:
823  *    "The root of the interrupt tree is determined when traversal
824  *     of the interrupt tree reaches an interrupt controller node without
825  *     an interrupts property and thus no explicit interrupt parent."
826  */
827 int
828 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
829     void *arg, u_int ipicount)
830 {
831 	struct intr_pic *pic;
832 
833 	pic = pic_lookup(dev, xref, FLAG_PIC);
834 	if (pic == NULL) {
835 		device_printf(dev, "not registered\n");
836 		return (EINVAL);
837 	}
838 
839 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
840 	    ("%s: Found a non-PIC controller: %s", __func__,
841 	     device_get_name(pic->pic_dev)));
842 
843 	if (filter == NULL) {
844 		device_printf(dev, "filter missing\n");
845 		return (EINVAL);
846 	}
847 
848 	/*
849 	 * Only one interrupt controllers could be on the root for now.
850 	 * Note that we further suppose that there is not threaded interrupt
851 	 * routine (handler) on the root. See intr_irq_handler().
852 	 */
853 	if (intr_irq_root_dev != NULL) {
854 		device_printf(dev, "another root already set\n");
855 		return (EBUSY);
856 	}
857 
858 	intr_irq_root_dev = dev;
859 	irq_root_filter = filter;
860 	irq_root_arg = arg;
861 	irq_root_ipicount = ipicount;
862 
863 	debugf("irq root set to %s\n", device_get_nameunit(dev));
864 	return (0);
865 }
866 
867 /*
868  * Add a handler to manage a sub range of a parents interrupts.
869  */
870 struct intr_pic *
871 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
872     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
873     uintptr_t length)
874 {
875 	struct intr_pic *parent_pic;
876 	struct intr_pic_child *newchild;
877 #ifdef INVARIANTS
878 	struct intr_pic_child *child;
879 #endif
880 
881 	/* Find the parent PIC */
882 	parent_pic = pic_lookup(parent, 0, FLAG_PIC);
883 	if (parent_pic == NULL)
884 		return (NULL);
885 
886 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
887 	newchild->pc_pic = pic;
888 	newchild->pc_filter = filter;
889 	newchild->pc_filter_arg = arg;
890 	newchild->pc_start = start;
891 	newchild->pc_length = length;
892 
893 	mtx_lock_spin(&parent_pic->pic_child_lock);
894 #ifdef INVARIANTS
895 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
896 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
897 		    __func__));
898 	}
899 #endif
900 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
901 	mtx_unlock_spin(&parent_pic->pic_child_lock);
902 
903 	return (pic);
904 }
905 
906 static int
907 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
908     struct intr_irqsrc **isrc)
909 {
910 	struct intr_pic *pic;
911 	struct intr_map_data_msi *msi;
912 
913 	if (data == NULL)
914 		return (EINVAL);
915 
916 	pic = pic_lookup(dev, xref,
917 	    (data->type == INTR_MAP_DATA_MSI) ? FLAG_MSI : FLAG_PIC);
918 	if (pic == NULL)
919 		return (ESRCH);
920 
921 	switch (data->type) {
922 	case INTR_MAP_DATA_MSI:
923 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
924 		    ("%s: Found a non-MSI controller: %s", __func__,
925 		     device_get_name(pic->pic_dev)));
926 		msi = (struct intr_map_data_msi *)data;
927 		*isrc = msi->isrc;
928 		return (0);
929 
930 	default:
931 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
932 		    ("%s: Found a non-PIC controller: %s", __func__,
933 		     device_get_name(pic->pic_dev)));
934 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
935 
936 	}
937 }
938 
939 int
940 intr_activate_irq(device_t dev, struct resource *res)
941 {
942 	device_t map_dev;
943 	intptr_t map_xref;
944 	struct intr_map_data *data;
945 	struct intr_irqsrc *isrc;
946 	u_int res_id;
947 	int error;
948 
949 	KASSERT(rman_get_start(res) == rman_get_end(res),
950 	    ("%s: more interrupts in resource", __func__));
951 
952 	res_id = (u_int)rman_get_start(res);
953 	if (intr_map_get_isrc(res_id) != NULL)
954 		panic("Attempt to double activation of resource id: %u\n",
955 		    res_id);
956 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
957 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
958 	if (error != 0) {
959 		free(data, M_INTRNG);
960 		/* XXX TODO DISCONECTED PICs */
961 		/* if (error == EINVAL) return(0); */
962 		return (error);
963 	}
964 	intr_map_set_isrc(res_id, isrc);
965 	rman_set_virtual(res, data);
966 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
967 }
968 
969 int
970 intr_deactivate_irq(device_t dev, struct resource *res)
971 {
972 	struct intr_map_data *data;
973 	struct intr_irqsrc *isrc;
974 	u_int res_id;
975 	int error;
976 
977 	KASSERT(rman_get_start(res) == rman_get_end(res),
978 	    ("%s: more interrupts in resource", __func__));
979 
980 	res_id = (u_int)rman_get_start(res);
981 	isrc = intr_map_get_isrc(res_id);
982 	if (isrc == NULL)
983 		panic("Attempt to deactivate non-active resource id: %u\n",
984 		    res_id);
985 
986 	data = rman_get_virtual(res);
987 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
988 	intr_map_set_isrc(res_id, NULL);
989 	rman_set_virtual(res, NULL);
990 	free(data, M_INTRNG);
991 	return (error);
992 }
993 
994 int
995 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
996     driver_intr_t hand, void *arg, int flags, void **cookiep)
997 {
998 	int error;
999 	struct intr_map_data *data;
1000 	struct intr_irqsrc *isrc;
1001 	const char *name;
1002 	u_int res_id;
1003 
1004 	KASSERT(rman_get_start(res) == rman_get_end(res),
1005 	    ("%s: more interrupts in resource", __func__));
1006 
1007 	res_id = (u_int)rman_get_start(res);
1008 	isrc = intr_map_get_isrc(res_id);
1009 	if (isrc == NULL) {
1010 		/* XXX TODO DISCONECTED PICs */
1011 		return (EINVAL);
1012 	}
1013 
1014 	data = rman_get_virtual(res);
1015 	name = device_get_nameunit(dev);
1016 
1017 #ifdef INTR_SOLO
1018 	/*
1019 	 * Standard handling is done through MI interrupt framework. However,
1020 	 * some interrupts could request solely own special handling. This
1021 	 * non standard handling can be used for interrupt controllers without
1022 	 * handler (filter only), so in case that interrupt controllers are
1023 	 * chained, MI interrupt framework is called only in leaf controller.
1024 	 *
1025 	 * Note that root interrupt controller routine is served as well,
1026 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1027 	 */
1028 	if (flags & INTR_SOLO && hand != NULL) {
1029 		debugf("irq %u cannot solo on %s\n", irq, name);
1030 		return (EINVAL);
1031 	}
1032 
1033 	if (flags & INTR_SOLO) {
1034 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1035 		    arg, cookiep);
1036 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1037 		    name);
1038 	} else
1039 #endif
1040 		{
1041 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1042 		    cookiep);
1043 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1044 	}
1045 	if (error != 0)
1046 		return (error);
1047 
1048 	mtx_lock(&isrc_table_lock);
1049 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1050 	if (error == 0) {
1051 		isrc->isrc_handlers++;
1052 		if (isrc->isrc_handlers == 1)
1053 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1054 	}
1055 	mtx_unlock(&isrc_table_lock);
1056 	if (error != 0)
1057 		intr_event_remove_handler(*cookiep);
1058 	return (error);
1059 }
1060 
1061 int
1062 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1063 {
1064 	int error;
1065 	struct intr_map_data *data;
1066 	struct intr_irqsrc *isrc;
1067 	u_int res_id;
1068 
1069 	KASSERT(rman_get_start(res) == rman_get_end(res),
1070 	    ("%s: more interrupts in resource", __func__));
1071 
1072 	res_id = (u_int)rman_get_start(res);
1073 	isrc = intr_map_get_isrc(res_id);
1074 	if (isrc == NULL || isrc->isrc_handlers == 0)
1075 		return (EINVAL);
1076 
1077 	data = rman_get_virtual(res);
1078 
1079 #ifdef INTR_SOLO
1080 	if (isrc->isrc_filter != NULL) {
1081 		if (isrc != cookie)
1082 			return (EINVAL);
1083 
1084 		mtx_lock(&isrc_table_lock);
1085 		isrc->isrc_filter = NULL;
1086 		isrc->isrc_arg = NULL;
1087 		isrc->isrc_handlers = 0;
1088 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1089 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1090 		isrc_update_name(isrc, NULL);
1091 		mtx_unlock(&isrc_table_lock);
1092 		return (0);
1093 	}
1094 #endif
1095 	if (isrc != intr_handler_source(cookie))
1096 		return (EINVAL);
1097 
1098 	error = intr_event_remove_handler(cookie);
1099 	if (error == 0) {
1100 		mtx_lock(&isrc_table_lock);
1101 		isrc->isrc_handlers--;
1102 		if (isrc->isrc_handlers == 0)
1103 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1104 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1105 		intrcnt_updatename(isrc);
1106 		mtx_unlock(&isrc_table_lock);
1107 	}
1108 	return (error);
1109 }
1110 
1111 int
1112 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1113     const char *descr)
1114 {
1115 	int error;
1116 	struct intr_irqsrc *isrc;
1117 	u_int res_id;
1118 
1119 	KASSERT(rman_get_start(res) == rman_get_end(res),
1120 	    ("%s: more interrupts in resource", __func__));
1121 
1122 	res_id = (u_int)rman_get_start(res);
1123 	isrc = intr_map_get_isrc(res_id);
1124 	if (isrc == NULL || isrc->isrc_handlers == 0)
1125 		return (EINVAL);
1126 #ifdef INTR_SOLO
1127 	if (isrc->isrc_filter != NULL) {
1128 		if (isrc != cookie)
1129 			return (EINVAL);
1130 
1131 		mtx_lock(&isrc_table_lock);
1132 		isrc_update_name(isrc, descr);
1133 		mtx_unlock(&isrc_table_lock);
1134 		return (0);
1135 	}
1136 #endif
1137 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1138 	if (error == 0) {
1139 		mtx_lock(&isrc_table_lock);
1140 		intrcnt_updatename(isrc);
1141 		mtx_unlock(&isrc_table_lock);
1142 	}
1143 	return (error);
1144 }
1145 
1146 #ifdef SMP
1147 int
1148 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1149 {
1150 	struct intr_irqsrc *isrc;
1151 	u_int res_id;
1152 
1153 	KASSERT(rman_get_start(res) == rman_get_end(res),
1154 	    ("%s: more interrupts in resource", __func__));
1155 
1156 	res_id = (u_int)rman_get_start(res);
1157 	isrc = intr_map_get_isrc(res_id);
1158 	if (isrc == NULL || isrc->isrc_handlers == 0)
1159 		return (EINVAL);
1160 #ifdef INTR_SOLO
1161 	if (isrc->isrc_filter != NULL)
1162 		return (intr_isrc_assign_cpu(isrc, cpu));
1163 #endif
1164 	return (intr_event_bind(isrc->isrc_event, cpu));
1165 }
1166 
1167 /*
1168  * Return the CPU that the next interrupt source should use.
1169  * For now just returns the next CPU according to round-robin.
1170  */
1171 u_int
1172 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1173 {
1174 	u_int cpu;
1175 
1176 	KASSERT(!CPU_EMPTY(cpumask), ("%s: Empty CPU mask", __func__));
1177 	if (!irq_assign_cpu || mp_ncpus == 1) {
1178 		cpu = PCPU_GET(cpuid);
1179 
1180 		if (CPU_ISSET(cpu, cpumask))
1181 			return (curcpu);
1182 
1183 		return (CPU_FFS(cpumask) - 1);
1184 	}
1185 
1186 	do {
1187 		last_cpu++;
1188 		if (last_cpu > mp_maxid)
1189 			last_cpu = 0;
1190 	} while (!CPU_ISSET(last_cpu, cpumask));
1191 	return (last_cpu);
1192 }
1193 
1194 /*
1195  *  Distribute all the interrupt sources among the available
1196  *  CPUs once the AP's have been launched.
1197  */
1198 static void
1199 intr_irq_shuffle(void *arg __unused)
1200 {
1201 	struct intr_irqsrc *isrc;
1202 	u_int i;
1203 
1204 	if (mp_ncpus == 1)
1205 		return;
1206 
1207 	mtx_lock(&isrc_table_lock);
1208 	irq_assign_cpu = TRUE;
1209 	for (i = 0; i < NIRQ; i++) {
1210 		isrc = irq_sources[i];
1211 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1212 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1213 			continue;
1214 
1215 		if (isrc->isrc_event != NULL &&
1216 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1217 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1218 			panic("%s: CPU inconsistency", __func__);
1219 
1220 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1221 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1222 
1223 		/*
1224 		 * We are in wicked position here if the following call fails
1225 		 * for bound ISRC. The best thing we can do is to clear
1226 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1227 		 */
1228 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1229 			CPU_ZERO(&isrc->isrc_cpu);
1230 	}
1231 	mtx_unlock(&isrc_table_lock);
1232 }
1233 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1234 
1235 #else
1236 u_int
1237 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1238 {
1239 
1240 	return (PCPU_GET(cpuid));
1241 }
1242 #endif
1243 
1244 /*
1245  * Allocate memory for new intr_map_data structure.
1246  * Initialize common fields.
1247  */
1248 struct intr_map_data *
1249 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1250 {
1251 	struct intr_map_data *data;
1252 
1253 	data = malloc(len, M_INTRNG, flags);
1254 	data->type = type;
1255 	data->len = len;
1256 	return (data);
1257 }
1258 
1259 void intr_free_intr_map_data(struct intr_map_data *data)
1260 {
1261 
1262 	free(data, M_INTRNG);
1263 }
1264 
1265 /*
1266  *  Register a MSI/MSI-X interrupt controller
1267  */
1268 int
1269 intr_msi_register(device_t dev, intptr_t xref)
1270 {
1271 	struct intr_pic *pic;
1272 
1273 	if (dev == NULL)
1274 		return (EINVAL);
1275 	pic = pic_create(dev, xref, FLAG_MSI);
1276 	if (pic == NULL)
1277 		return (ENOMEM);
1278 
1279 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1280 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1281 	return (0);
1282 }
1283 
1284 int
1285 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1286     int maxcount, int *irqs)
1287 {
1288 	struct intr_irqsrc **isrc;
1289 	struct intr_pic *pic;
1290 	device_t pdev;
1291 	struct intr_map_data_msi *msi;
1292 	int err, i;
1293 
1294 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1295 	if (pic == NULL)
1296 		return (ESRCH);
1297 
1298 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1299 	    ("%s: Found a non-MSI controller: %s", __func__,
1300 	     device_get_name(pic->pic_dev)));
1301 
1302 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1303 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1304 	if (err != 0) {
1305 		free(isrc, M_INTRNG);
1306 		return (err);
1307 	}
1308 
1309 	for (i = 0; i < count; i++) {
1310 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1311 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1312 		msi-> isrc = isrc[i];
1313 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1314 		    (struct intr_map_data *)msi);
1315 
1316 	}
1317 	free(isrc, M_INTRNG);
1318 
1319 	return (err);
1320 }
1321 
1322 int
1323 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1324     int *irqs)
1325 {
1326 	struct intr_irqsrc **isrc;
1327 	struct intr_pic *pic;
1328 	struct intr_map_data_msi *msi;
1329 	int i, err;
1330 
1331 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1332 	if (pic == NULL)
1333 		return (ESRCH);
1334 
1335 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1336 	    ("%s: Found a non-MSI controller: %s", __func__,
1337 	     device_get_name(pic->pic_dev)));
1338 
1339 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1340 
1341 	for (i = 0; i < count; i++) {
1342 		msi = (struct intr_map_data_msi *)
1343 		    intr_map_get_map_data(irqs[i]);
1344 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1345 		    ("%s: irq %d map data is not MSI", __func__,
1346 		    irqs[i]));
1347 		isrc[i] = msi->isrc;
1348 	}
1349 
1350 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1351 
1352 	for (i = 0; i < count; i++) {
1353 		if (isrc[i] != NULL)
1354 			intr_unmap_irq(irqs[i]);
1355 	}
1356 
1357 	free(isrc, M_INTRNG);
1358 	return (err);
1359 }
1360 
1361 int
1362 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1363 {
1364 	struct intr_irqsrc *isrc;
1365 	struct intr_pic *pic;
1366 	device_t pdev;
1367 	struct intr_map_data_msi *msi;
1368 	int err;
1369 
1370 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1371 	if (pic == NULL)
1372 		return (ESRCH);
1373 
1374 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1375 	    ("%s: Found a non-MSI controller: %s", __func__,
1376 	     device_get_name(pic->pic_dev)));
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 void dosoftints(void);
1447 void
1448 dosoftints(void)
1449 {
1450 }
1451 
1452 #ifdef SMP
1453 /*
1454  *  Init interrupt controller on another CPU.
1455  */
1456 void
1457 intr_pic_init_secondary(void)
1458 {
1459 
1460 	/*
1461 	 * QQQ: Only root PIC is aware of other CPUs ???
1462 	 */
1463 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1464 
1465 	//mtx_lock(&isrc_table_lock);
1466 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1467 	//mtx_unlock(&isrc_table_lock);
1468 }
1469 #endif
1470 
1471 #ifdef DDB
1472 DB_SHOW_COMMAND(irqs, db_show_irqs)
1473 {
1474 	u_int i, irqsum;
1475 	u_long num;
1476 	struct intr_irqsrc *isrc;
1477 
1478 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1479 		isrc = irq_sources[i];
1480 		if (isrc == NULL)
1481 			continue;
1482 
1483 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1484 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1485 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1486 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1487 		irqsum += num;
1488 	}
1489 	db_printf("irq total %u\n", irqsum);
1490 }
1491 #endif
1492 
1493 /*
1494  * Interrupt mapping table functions.
1495  *
1496  * Please, keep this part separately, it can be transformed to
1497  * extension of standard resources.
1498  */
1499 struct intr_map_entry
1500 {
1501 	device_t 		dev;
1502 	intptr_t 		xref;
1503 	struct intr_map_data 	*map_data;
1504 	struct intr_irqsrc 	*isrc;
1505 	/* XXX TODO DISCONECTED PICs */
1506 	/*int			flags */
1507 };
1508 
1509 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1510 static struct intr_map_entry *irq_map[2 * NIRQ];
1511 static int irq_map_count = nitems(irq_map);
1512 static int irq_map_first_free_idx;
1513 static struct mtx irq_map_lock;
1514 
1515 static struct intr_irqsrc *
1516 intr_map_get_isrc(u_int res_id)
1517 {
1518 	struct intr_irqsrc *isrc;
1519 
1520 	mtx_lock(&irq_map_lock);
1521 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1522 		mtx_unlock(&irq_map_lock);
1523 		return (NULL);
1524 	}
1525 	isrc = irq_map[res_id]->isrc;
1526 	mtx_unlock(&irq_map_lock);
1527 	return (isrc);
1528 }
1529 
1530 static void
1531 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1532 {
1533 
1534 	mtx_lock(&irq_map_lock);
1535 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1536 		mtx_unlock(&irq_map_lock);
1537 		return;
1538 	}
1539 	irq_map[res_id]->isrc = isrc;
1540 	mtx_unlock(&irq_map_lock);
1541 }
1542 
1543 /*
1544  * Get a copy of intr_map_entry data
1545  */
1546 static struct intr_map_data *
1547 intr_map_get_map_data(u_int res_id)
1548 {
1549 	struct intr_map_data *data;
1550 
1551 	data = NULL;
1552 	mtx_lock(&irq_map_lock);
1553 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1554 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1555 	data = irq_map[res_id]->map_data;
1556 	mtx_unlock(&irq_map_lock);
1557 
1558 	return (data);
1559 }
1560 
1561 /*
1562  * Get a copy of intr_map_entry data
1563  */
1564 static void
1565 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1566     struct intr_map_data **data)
1567 {
1568 	size_t len;
1569 
1570 	len = 0;
1571 	mtx_lock(&irq_map_lock);
1572 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1573 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1574 	if (irq_map[res_id]->map_data != NULL)
1575 		len = irq_map[res_id]->map_data->len;
1576 	mtx_unlock(&irq_map_lock);
1577 
1578 	if (len == 0)
1579 		*data = NULL;
1580 	else
1581 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1582 	mtx_lock(&irq_map_lock);
1583 	if (irq_map[res_id] == NULL)
1584 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1585 	if (len != 0) {
1586 		if (len != irq_map[res_id]->map_data->len)
1587 			panic("Resource id: %u has changed.\n", res_id);
1588 		memcpy(*data, irq_map[res_id]->map_data, len);
1589 	}
1590 	*map_dev = irq_map[res_id]->dev;
1591 	*map_xref = irq_map[res_id]->xref;
1592 	mtx_unlock(&irq_map_lock);
1593 }
1594 
1595 /*
1596  * Allocate and fill new entry in irq_map table.
1597  */
1598 u_int
1599 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1600 {
1601 	u_int i;
1602 	struct intr_map_entry *entry;
1603 
1604 	/* Prepare new entry first. */
1605 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1606 
1607 	entry->dev = dev;
1608 	entry->xref = xref;
1609 	entry->map_data = data;
1610 	entry->isrc = NULL;
1611 
1612 	mtx_lock(&irq_map_lock);
1613 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1614 		if (irq_map[i] == NULL) {
1615 			irq_map[i] = entry;
1616 			irq_map_first_free_idx = i + 1;
1617 			mtx_unlock(&irq_map_lock);
1618 			return (i);
1619 		}
1620 	}
1621 	mtx_unlock(&irq_map_lock);
1622 
1623 	/* XXX Expand irq_map table */
1624 	panic("IRQ mapping table is full.");
1625 }
1626 
1627 /*
1628  * Remove and free mapping entry.
1629  */
1630 void
1631 intr_unmap_irq(u_int res_id)
1632 {
1633 	struct intr_map_entry *entry;
1634 
1635 	mtx_lock(&irq_map_lock);
1636 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1637 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1638 	entry = irq_map[res_id];
1639 	irq_map[res_id] = NULL;
1640 	irq_map_first_free_idx = res_id;
1641 	mtx_unlock(&irq_map_lock);
1642 	intr_free_intr_map_data(entry->map_data);
1643 	free(entry, M_INTRNG);
1644 }
1645 
1646 /*
1647  * Clone mapping entry.
1648  */
1649 u_int
1650 intr_map_clone_irq(u_int old_res_id)
1651 {
1652 	device_t map_dev;
1653 	intptr_t map_xref;
1654 	struct intr_map_data *data;
1655 
1656 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1657 	return (intr_map_irq(map_dev, map_xref, data));
1658 }
1659 
1660 static void
1661 intr_map_init(void *dummy __unused)
1662 {
1663 
1664 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1665 }
1666 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1667