xref: /freebsd-11-stable/sys/kern/subr_intr.c (revision ec4092e367a2a123f534b57e4b8b4e9753054a02)
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 #define	FLAG_PIC	(1 << 0)
110 #define	FLAG_MSI	(1 << 1)
111 	u_int			pic_flags;
112 	struct mtx		pic_child_lock;
113 	SLIST_HEAD(, intr_pic_child) pic_children;
114 };
115 
116 static struct mtx pic_list_lock;
117 static SLIST_HEAD(, intr_pic) pic_list;
118 
119 static struct intr_pic *pic_lookup(device_t dev, intptr_t xref);
120 
121 /* Interrupt source definition. */
122 static struct mtx isrc_table_lock;
123 static struct intr_irqsrc *irq_sources[NIRQ];
124 u_int irq_next_free;
125 
126 #ifdef SMP
127 static boolean_t irq_assign_cpu = FALSE;
128 #endif
129 
130 /*
131  * - 2 counters for each I/O interrupt.
132  * - MAXCPU counters for each IPI counters for SMP.
133  */
134 #ifdef SMP
135 #define INTRCNT_COUNT   (NIRQ * 2 + INTR_IPI_COUNT * MAXCPU)
136 #else
137 #define INTRCNT_COUNT   (NIRQ * 2)
138 #endif
139 
140 /* Data for MI statistics reporting. */
141 u_long intrcnt[INTRCNT_COUNT];
142 char intrnames[INTRCNT_COUNT * INTRNAME_LEN];
143 size_t sintrcnt = sizeof(intrcnt);
144 size_t sintrnames = sizeof(intrnames);
145 static u_int intrcnt_index;
146 
147 static struct intr_irqsrc *intr_map_get_isrc(u_int res_id);
148 static void intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc);
149 static struct intr_map_data * intr_map_get_map_data(u_int res_id);
150 static void intr_map_copy_map_data(u_int res_id, device_t *dev, intptr_t *xref,
151     struct intr_map_data **data);
152 
153 /*
154  *  Interrupt framework initialization routine.
155  */
156 static void
intr_irq_init(void * dummy __unused)157 intr_irq_init(void *dummy __unused)
158 {
159 
160 	SLIST_INIT(&pic_list);
161 	mtx_init(&pic_list_lock, "intr pic list", NULL, MTX_DEF);
162 
163 	mtx_init(&isrc_table_lock, "intr isrc table", NULL, MTX_DEF);
164 }
165 SYSINIT(intr_irq_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_irq_init, NULL);
166 
167 static void
intrcnt_setname(const char * name,int index)168 intrcnt_setname(const char *name, int index)
169 {
170 
171 	snprintf(intrnames + INTRNAME_LEN * index, INTRNAME_LEN, "%-*s",
172 	    INTRNAME_LEN - 1, name);
173 }
174 
175 /*
176  *  Update name for interrupt source with interrupt event.
177  */
178 static void
intrcnt_updatename(struct intr_irqsrc * isrc)179 intrcnt_updatename(struct intr_irqsrc *isrc)
180 {
181 
182 	/* QQQ: What about stray counter name? */
183 	mtx_assert(&isrc_table_lock, MA_OWNED);
184 	intrcnt_setname(isrc->isrc_event->ie_fullname, isrc->isrc_index);
185 }
186 
187 /*
188  *  Virtualization for interrupt source interrupt counter increment.
189  */
190 static inline void
isrc_increment_count(struct intr_irqsrc * isrc)191 isrc_increment_count(struct intr_irqsrc *isrc)
192 {
193 
194 	if (isrc->isrc_flags & INTR_ISRCF_PPI)
195 		atomic_add_long(&isrc->isrc_count[0], 1);
196 	else
197 		isrc->isrc_count[0]++;
198 }
199 
200 /*
201  *  Virtualization for interrupt source interrupt stray counter increment.
202  */
203 static inline void
isrc_increment_straycount(struct intr_irqsrc * isrc)204 isrc_increment_straycount(struct intr_irqsrc *isrc)
205 {
206 
207 	isrc->isrc_count[1]++;
208 }
209 
210 /*
211  *  Virtualization for interrupt source interrupt name update.
212  */
213 static void
isrc_update_name(struct intr_irqsrc * isrc,const char * name)214 isrc_update_name(struct intr_irqsrc *isrc, const char *name)
215 {
216 	char str[INTRNAME_LEN];
217 
218 	mtx_assert(&isrc_table_lock, MA_OWNED);
219 
220 	if (name != NULL) {
221 		snprintf(str, INTRNAME_LEN, "%s: %s", isrc->isrc_name, name);
222 		intrcnt_setname(str, isrc->isrc_index);
223 		snprintf(str, INTRNAME_LEN, "stray %s: %s", isrc->isrc_name,
224 		    name);
225 		intrcnt_setname(str, isrc->isrc_index + 1);
226 	} else {
227 		snprintf(str, INTRNAME_LEN, "%s:", isrc->isrc_name);
228 		intrcnt_setname(str, isrc->isrc_index);
229 		snprintf(str, INTRNAME_LEN, "stray %s:", isrc->isrc_name);
230 		intrcnt_setname(str, isrc->isrc_index + 1);
231 	}
232 }
233 
234 /*
235  *  Virtualization for interrupt source interrupt counters setup.
236  */
237 static void
isrc_setup_counters(struct intr_irqsrc * isrc)238 isrc_setup_counters(struct intr_irqsrc *isrc)
239 {
240 	u_int index;
241 
242 	/*
243 	 *  XXX - it does not work well with removable controllers and
244 	 *        interrupt sources !!!
245 	 */
246 	index = atomic_fetchadd_int(&intrcnt_index, 2);
247 	isrc->isrc_index = index;
248 	isrc->isrc_count = &intrcnt[index];
249 	isrc_update_name(isrc, NULL);
250 }
251 
252 /*
253  *  Virtualization for interrupt source interrupt counters release.
254  */
255 static void
isrc_release_counters(struct intr_irqsrc * isrc)256 isrc_release_counters(struct intr_irqsrc *isrc)
257 {
258 
259 	panic("%s: not implemented", __func__);
260 }
261 
262 #ifdef SMP
263 /*
264  *  Virtualization for interrupt source IPI counters setup.
265  */
266 u_long *
intr_ipi_setup_counters(const char * name)267 intr_ipi_setup_counters(const char *name)
268 {
269 	u_int index, i;
270 	char str[INTRNAME_LEN];
271 
272 	index = atomic_fetchadd_int(&intrcnt_index, MAXCPU);
273 	for (i = 0; i < MAXCPU; i++) {
274 		snprintf(str, INTRNAME_LEN, "cpu%d:%s", i, name);
275 		intrcnt_setname(str, index + i);
276 	}
277 	return (&intrcnt[index]);
278 }
279 #endif
280 
281 /*
282  *  Main interrupt dispatch handler. It's called straight
283  *  from the assembler, where CPU interrupt is served.
284  */
285 void
intr_irq_handler(struct trapframe * tf)286 intr_irq_handler(struct trapframe *tf)
287 {
288 	struct trapframe * oldframe;
289 	struct thread * td;
290 
291 	KASSERT(irq_root_filter != NULL, ("%s: no filter", __func__));
292 
293 	PCPU_INC(cnt.v_intr);
294 	critical_enter();
295 	td = curthread;
296 	oldframe = td->td_intr_frame;
297 	td->td_intr_frame = tf;
298 	irq_root_filter(irq_root_arg);
299 	td->td_intr_frame = oldframe;
300 	critical_exit();
301 #ifdef HWPMC_HOOKS
302 	if (pmc_hook && TRAPF_USERMODE(tf) &&
303 	    (PCPU_GET(curthread)->td_pflags & TDP_CALLCHAIN))
304 		pmc_hook(PCPU_GET(curthread), PMC_FN_USER_CALLCHAIN, tf);
305 #endif
306 }
307 
308 int
intr_child_irq_handler(struct intr_pic * parent,uintptr_t irq)309 intr_child_irq_handler(struct intr_pic *parent, uintptr_t irq)
310 {
311 	struct intr_pic_child *child;
312 	bool found;
313 
314 	found = false;
315 	mtx_lock_spin(&parent->pic_child_lock);
316 	SLIST_FOREACH(child, &parent->pic_children, pc_next) {
317 		if (child->pc_start <= irq &&
318 		    irq < (child->pc_start + child->pc_length)) {
319 			found = true;
320 			break;
321 		}
322 	}
323 	mtx_unlock_spin(&parent->pic_child_lock);
324 
325 	if (found)
326 		return (child->pc_filter(child->pc_filter_arg, irq));
327 
328 	return (FILTER_STRAY);
329 }
330 
331 /*
332  *  interrupt controller dispatch function for interrupts. It should
333  *  be called straight from the interrupt controller, when associated interrupt
334  *  source is learned.
335  */
336 int
intr_isrc_dispatch(struct intr_irqsrc * isrc,struct trapframe * tf)337 intr_isrc_dispatch(struct intr_irqsrc *isrc, struct trapframe *tf)
338 {
339 
340 	KASSERT(isrc != NULL, ("%s: no source", __func__));
341 
342 	isrc_increment_count(isrc);
343 
344 #ifdef INTR_SOLO
345 	if (isrc->isrc_filter != NULL) {
346 		int error;
347 		error = isrc->isrc_filter(isrc->isrc_arg, tf);
348 		PIC_POST_FILTER(isrc->isrc_dev, isrc);
349 		if (error == FILTER_HANDLED)
350 			return (0);
351 	} else
352 #endif
353 	if (isrc->isrc_event != NULL) {
354 		if (intr_event_handle(isrc->isrc_event, tf) == 0)
355 			return (0);
356 	}
357 
358 	isrc_increment_straycount(isrc);
359 	return (EINVAL);
360 }
361 
362 /*
363  *  Alloc unique interrupt number (resource handle) for interrupt source.
364  *
365  *  There could be various strategies how to allocate free interrupt number
366  *  (resource handle) for new interrupt source.
367  *
368  *  1. Handles are always allocated forward, so handles are not recycled
369  *     immediately. However, if only one free handle left which is reused
370  *     constantly...
371  */
372 static inline int
isrc_alloc_irq(struct intr_irqsrc * isrc)373 isrc_alloc_irq(struct intr_irqsrc *isrc)
374 {
375 	u_int maxirqs, irq;
376 
377 	mtx_assert(&isrc_table_lock, MA_OWNED);
378 
379 	maxirqs = nitems(irq_sources);
380 	if (irq_next_free >= maxirqs)
381 		return (ENOSPC);
382 
383 	for (irq = irq_next_free; irq < maxirqs; irq++) {
384 		if (irq_sources[irq] == NULL)
385 			goto found;
386 	}
387 	for (irq = 0; irq < irq_next_free; irq++) {
388 		if (irq_sources[irq] == NULL)
389 			goto found;
390 	}
391 
392 	irq_next_free = maxirqs;
393 	return (ENOSPC);
394 
395 found:
396 	isrc->isrc_irq = irq;
397 	irq_sources[irq] = isrc;
398 
399 	irq_next_free = irq + 1;
400 	if (irq_next_free >= maxirqs)
401 		irq_next_free = 0;
402 	return (0);
403 }
404 
405 /*
406  *  Free unique interrupt number (resource handle) from interrupt source.
407  */
408 static inline int
isrc_free_irq(struct intr_irqsrc * isrc)409 isrc_free_irq(struct intr_irqsrc *isrc)
410 {
411 
412 	mtx_assert(&isrc_table_lock, MA_OWNED);
413 
414 	if (isrc->isrc_irq >= nitems(irq_sources))
415 		return (EINVAL);
416 	if (irq_sources[isrc->isrc_irq] != isrc)
417 		return (EINVAL);
418 
419 	irq_sources[isrc->isrc_irq] = NULL;
420 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
421 	return (0);
422 }
423 
424 /*
425  *  Initialize interrupt source and register it into global interrupt table.
426  */
427 int
intr_isrc_register(struct intr_irqsrc * isrc,device_t dev,u_int flags,const char * fmt,...)428 intr_isrc_register(struct intr_irqsrc *isrc, device_t dev, u_int flags,
429     const char *fmt, ...)
430 {
431 	int error;
432 	va_list ap;
433 
434 	bzero(isrc, sizeof(struct intr_irqsrc));
435 	isrc->isrc_dev = dev;
436 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
437 	isrc->isrc_flags = flags;
438 
439 	va_start(ap, fmt);
440 	vsnprintf(isrc->isrc_name, INTR_ISRC_NAMELEN, fmt, ap);
441 	va_end(ap);
442 
443 	mtx_lock(&isrc_table_lock);
444 	error = isrc_alloc_irq(isrc);
445 	if (error != 0) {
446 		mtx_unlock(&isrc_table_lock);
447 		return (error);
448 	}
449 	/*
450 	 * Setup interrupt counters, but not for IPI sources. Those are setup
451 	 * later and only for used ones (up to INTR_IPI_COUNT) to not exhaust
452 	 * our counter pool.
453 	 */
454 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
455 		isrc_setup_counters(isrc);
456 	mtx_unlock(&isrc_table_lock);
457 	return (0);
458 }
459 
460 /*
461  *  Deregister interrupt source from global interrupt table.
462  */
463 int
intr_isrc_deregister(struct intr_irqsrc * isrc)464 intr_isrc_deregister(struct intr_irqsrc *isrc)
465 {
466 	int error;
467 
468 	mtx_lock(&isrc_table_lock);
469 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
470 		isrc_release_counters(isrc);
471 	error = isrc_free_irq(isrc);
472 	mtx_unlock(&isrc_table_lock);
473 	return (error);
474 }
475 
476 #ifdef SMP
477 /*
478  *  A support function for a PIC to decide if provided ISRC should be inited
479  *  on given cpu. The logic of INTR_ISRCF_BOUND flag and isrc_cpu member of
480  *  struct intr_irqsrc is the following:
481  *
482  *     If INTR_ISRCF_BOUND is set, the ISRC should be inited only on cpus
483  *     set in isrc_cpu. If not, the ISRC should be inited on every cpu and
484  *     isrc_cpu is kept consistent with it. Thus isrc_cpu is always correct.
485  */
486 bool
intr_isrc_init_on_cpu(struct intr_irqsrc * isrc,u_int cpu)487 intr_isrc_init_on_cpu(struct intr_irqsrc *isrc, u_int cpu)
488 {
489 
490 	if (isrc->isrc_handlers == 0)
491 		return (false);
492 	if ((isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI)) == 0)
493 		return (false);
494 	if (isrc->isrc_flags & INTR_ISRCF_BOUND)
495 		return (CPU_ISSET(cpu, &isrc->isrc_cpu));
496 
497 	CPU_SET(cpu, &isrc->isrc_cpu);
498 	return (true);
499 }
500 #endif
501 
502 #ifdef INTR_SOLO
503 /*
504  *  Setup filter into interrupt source.
505  */
506 static int
iscr_setup_filter(struct intr_irqsrc * isrc,const char * name,intr_irq_filter_t * filter,void * arg,void ** cookiep)507 iscr_setup_filter(struct intr_irqsrc *isrc, const char *name,
508     intr_irq_filter_t *filter, void *arg, void **cookiep)
509 {
510 
511 	if (filter == NULL)
512 		return (EINVAL);
513 
514 	mtx_lock(&isrc_table_lock);
515 	/*
516 	 * Make sure that we do not mix the two ways
517 	 * how we handle interrupt sources.
518 	 */
519 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
520 		mtx_unlock(&isrc_table_lock);
521 		return (EBUSY);
522 	}
523 	isrc->isrc_filter = filter;
524 	isrc->isrc_arg = arg;
525 	isrc_update_name(isrc, name);
526 	mtx_unlock(&isrc_table_lock);
527 
528 	*cookiep = isrc;
529 	return (0);
530 }
531 #endif
532 
533 /*
534  *  Interrupt source pre_ithread method for MI interrupt framework.
535  */
536 static void
intr_isrc_pre_ithread(void * arg)537 intr_isrc_pre_ithread(void *arg)
538 {
539 	struct intr_irqsrc *isrc = arg;
540 
541 	PIC_PRE_ITHREAD(isrc->isrc_dev, isrc);
542 }
543 
544 /*
545  *  Interrupt source post_ithread method for MI interrupt framework.
546  */
547 static void
intr_isrc_post_ithread(void * arg)548 intr_isrc_post_ithread(void *arg)
549 {
550 	struct intr_irqsrc *isrc = arg;
551 
552 	PIC_POST_ITHREAD(isrc->isrc_dev, isrc);
553 }
554 
555 /*
556  *  Interrupt source post_filter method for MI interrupt framework.
557  */
558 static void
intr_isrc_post_filter(void * arg)559 intr_isrc_post_filter(void *arg)
560 {
561 	struct intr_irqsrc *isrc = arg;
562 
563 	PIC_POST_FILTER(isrc->isrc_dev, isrc);
564 }
565 
566 /*
567  *  Interrupt source assign_cpu method for MI interrupt framework.
568  */
569 static int
intr_isrc_assign_cpu(void * arg,int cpu)570 intr_isrc_assign_cpu(void *arg, int cpu)
571 {
572 #ifdef SMP
573 	struct intr_irqsrc *isrc = arg;
574 	int error;
575 
576 	if (isrc->isrc_dev != intr_irq_root_dev)
577 		return (EINVAL);
578 
579 	mtx_lock(&isrc_table_lock);
580 	if (cpu == NOCPU) {
581 		CPU_ZERO(&isrc->isrc_cpu);
582 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
583 	} else {
584 		CPU_SETOF(cpu, &isrc->isrc_cpu);
585 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
586 	}
587 
588 	/*
589 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
590 	 * re-balance it to another CPU or enable it on more CPUs. However,
591 	 * PIC is expected to change isrc_cpu appropriately to keep us well
592 	 * informed if the call is successful.
593 	 */
594 	if (irq_assign_cpu) {
595 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
596 		if (error) {
597 			CPU_ZERO(&isrc->isrc_cpu);
598 			mtx_unlock(&isrc_table_lock);
599 			return (error);
600 		}
601 	}
602 	mtx_unlock(&isrc_table_lock);
603 	return (0);
604 #else
605 	return (EOPNOTSUPP);
606 #endif
607 }
608 
609 /*
610  *  Create interrupt event for interrupt source.
611  */
612 static int
isrc_event_create(struct intr_irqsrc * isrc)613 isrc_event_create(struct intr_irqsrc *isrc)
614 {
615 	struct intr_event *ie;
616 	int error;
617 
618 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
619 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
620 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
621 	if (error)
622 		return (error);
623 
624 	mtx_lock(&isrc_table_lock);
625 	/*
626 	 * Make sure that we do not mix the two ways
627 	 * how we handle interrupt sources. Let contested event wins.
628 	 */
629 #ifdef INTR_SOLO
630 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
631 #else
632 	if (isrc->isrc_event != NULL) {
633 #endif
634 		mtx_unlock(&isrc_table_lock);
635 		intr_event_destroy(ie);
636 		return (isrc->isrc_event != NULL ? EBUSY : 0);
637 	}
638 	isrc->isrc_event = ie;
639 	mtx_unlock(&isrc_table_lock);
640 
641 	return (0);
642 }
643 #ifdef notyet
644 /*
645  *  Destroy interrupt event for interrupt source.
646  */
647 static void
648 isrc_event_destroy(struct intr_irqsrc *isrc)
649 {
650 	struct intr_event *ie;
651 
652 	mtx_lock(&isrc_table_lock);
653 	ie = isrc->isrc_event;
654 	isrc->isrc_event = NULL;
655 	mtx_unlock(&isrc_table_lock);
656 
657 	if (ie != NULL)
658 		intr_event_destroy(ie);
659 }
660 #endif
661 /*
662  *  Add handler to interrupt source.
663  */
664 static int
665 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
666     driver_filter_t filter, driver_intr_t handler, void *arg,
667     enum intr_type flags, void **cookiep)
668 {
669 	int error;
670 
671 	if (isrc->isrc_event == NULL) {
672 		error = isrc_event_create(isrc);
673 		if (error)
674 			return (error);
675 	}
676 
677 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
678 	    arg, intr_priority(flags), flags, cookiep);
679 	if (error == 0) {
680 		mtx_lock(&isrc_table_lock);
681 		intrcnt_updatename(isrc);
682 		mtx_unlock(&isrc_table_lock);
683 	}
684 
685 	return (error);
686 }
687 
688 /*
689  *  Lookup interrupt controller locked.
690  */
691 static inline struct intr_pic *
692 pic_lookup_locked(device_t dev, intptr_t xref)
693 {
694 	struct intr_pic *pic;
695 
696 	mtx_assert(&pic_list_lock, MA_OWNED);
697 
698 	if (dev == NULL && xref == 0)
699 		return (NULL);
700 
701 	/* Note that pic->pic_dev is never NULL on registered PIC. */
702 	SLIST_FOREACH(pic, &pic_list, pic_next) {
703 		if (dev == NULL) {
704 			if (xref == pic->pic_xref)
705 				return (pic);
706 		} else if (xref == 0 || pic->pic_xref == 0) {
707 			if (dev == pic->pic_dev)
708 				return (pic);
709 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
710 				return (pic);
711 	}
712 	return (NULL);
713 }
714 
715 /*
716  *  Lookup interrupt controller.
717  */
718 static struct intr_pic *
719 pic_lookup(device_t dev, intptr_t xref)
720 {
721 	struct intr_pic *pic;
722 
723 	mtx_lock(&pic_list_lock);
724 	pic = pic_lookup_locked(dev, xref);
725 	mtx_unlock(&pic_list_lock);
726 	return (pic);
727 }
728 
729 /*
730  *  Create interrupt controller.
731  */
732 static struct intr_pic *
733 pic_create(device_t dev, intptr_t xref)
734 {
735 	struct intr_pic *pic;
736 
737 	mtx_lock(&pic_list_lock);
738 	pic = pic_lookup_locked(dev, xref);
739 	if (pic != NULL) {
740 		mtx_unlock(&pic_list_lock);
741 		return (pic);
742 	}
743 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
744 	if (pic == NULL) {
745 		mtx_unlock(&pic_list_lock);
746 		return (NULL);
747 	}
748 	pic->pic_xref = xref;
749 	pic->pic_dev = dev;
750 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
751 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
752 	mtx_unlock(&pic_list_lock);
753 
754 	return (pic);
755 }
756 #ifdef notyet
757 /*
758  *  Destroy interrupt controller.
759  */
760 static void
761 pic_destroy(device_t dev, intptr_t xref)
762 {
763 	struct intr_pic *pic;
764 
765 	mtx_lock(&pic_list_lock);
766 	pic = pic_lookup_locked(dev, xref);
767 	if (pic == NULL) {
768 		mtx_unlock(&pic_list_lock);
769 		return;
770 	}
771 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
772 	mtx_unlock(&pic_list_lock);
773 
774 	free(pic, M_INTRNG);
775 }
776 #endif
777 /*
778  *  Register interrupt controller.
779  */
780 struct intr_pic *
781 intr_pic_register(device_t dev, intptr_t xref)
782 {
783 	struct intr_pic *pic;
784 
785 	if (dev == NULL)
786 		return (NULL);
787 	pic = pic_create(dev, xref);
788 	if (pic == NULL)
789 		return (NULL);
790 
791 	pic->pic_flags |= FLAG_PIC;
792 
793 	debugf("PIC %p registered for %s <dev %p, xref %x>\n", pic,
794 	    device_get_nameunit(dev), dev, xref);
795 	return (pic);
796 }
797 
798 /*
799  *  Unregister interrupt controller.
800  */
801 int
802 intr_pic_deregister(device_t dev, intptr_t xref)
803 {
804 
805 	panic("%s: not implemented", __func__);
806 }
807 
808 /*
809  *  Mark interrupt controller (itself) as a root one.
810  *
811  *  Note that only an interrupt controller can really know its position
812  *  in interrupt controller's tree. So root PIC must claim itself as a root.
813  *
814  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
815  *  page 30:
816  *    "The root of the interrupt tree is determined when traversal
817  *     of the interrupt tree reaches an interrupt controller node without
818  *     an interrupts property and thus no explicit interrupt parent."
819  */
820 int
821 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
822     void *arg, u_int ipicount)
823 {
824 	struct intr_pic *pic;
825 
826 	pic = pic_lookup(dev, xref);
827 	if (pic == NULL) {
828 		device_printf(dev, "not registered\n");
829 		return (EINVAL);
830 	}
831 
832 	KASSERT((pic->pic_flags & FLAG_PIC) != 0,
833 	    ("%s: Found a non-PIC controller: %s", __func__,
834 	     device_get_name(pic->pic_dev)));
835 
836 	if (filter == NULL) {
837 		device_printf(dev, "filter missing\n");
838 		return (EINVAL);
839 	}
840 
841 	/*
842 	 * Only one interrupt controllers could be on the root for now.
843 	 * Note that we further suppose that there is not threaded interrupt
844 	 * routine (handler) on the root. See intr_irq_handler().
845 	 */
846 	if (intr_irq_root_dev != NULL) {
847 		device_printf(dev, "another root already set\n");
848 		return (EBUSY);
849 	}
850 
851 	intr_irq_root_dev = dev;
852 	irq_root_filter = filter;
853 	irq_root_arg = arg;
854 	irq_root_ipicount = ipicount;
855 
856 	debugf("irq root set to %s\n", device_get_nameunit(dev));
857 	return (0);
858 }
859 
860 /*
861  * Add a handler to manage a sub range of a parents interrupts.
862  */
863 struct intr_pic *
864 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
865     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
866     uintptr_t length)
867 {
868 	struct intr_pic *parent_pic;
869 	struct intr_pic_child *newchild;
870 #ifdef INVARIANTS
871 	struct intr_pic_child *child;
872 #endif
873 
874 	parent_pic = pic_lookup(parent, 0);
875 	if (parent_pic == NULL)
876 		return (NULL);
877 
878 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
879 	newchild->pc_pic = pic;
880 	newchild->pc_filter = filter;
881 	newchild->pc_filter_arg = arg;
882 	newchild->pc_start = start;
883 	newchild->pc_length = length;
884 
885 	mtx_lock_spin(&parent_pic->pic_child_lock);
886 #ifdef INVARIANTS
887 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
888 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
889 		    __func__));
890 	}
891 #endif
892 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
893 	mtx_unlock_spin(&parent_pic->pic_child_lock);
894 
895 	return (pic);
896 }
897 
898 static int
899 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
900     struct intr_irqsrc **isrc)
901 {
902 	struct intr_pic *pic;
903 	struct intr_map_data_msi *msi;
904 
905 	if (data == NULL)
906 		return (EINVAL);
907 
908 	pic = pic_lookup(dev, xref);
909 	if (pic == NULL)
910 		return (ESRCH);
911 
912 	switch (data->type) {
913 	case INTR_MAP_DATA_MSI:
914 		KASSERT((pic->pic_flags & FLAG_MSI) != 0,
915 		    ("%s: Found a non-MSI controller: %s", __func__,
916 		     device_get_name(pic->pic_dev)));
917 		msi = (struct intr_map_data_msi *)data;
918 		*isrc = msi->isrc;
919 		return (0);
920 
921 	default:
922 		KASSERT((pic->pic_flags & FLAG_PIC) != 0,
923 		    ("%s: Found a non-PIC controller: %s", __func__,
924 		     device_get_name(pic->pic_dev)));
925 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
926 
927 	}
928 }
929 
930 int
931 intr_activate_irq(device_t dev, struct resource *res)
932 {
933 	device_t map_dev;
934 	intptr_t map_xref;
935 	struct intr_map_data *data;
936 	struct intr_irqsrc *isrc;
937 	u_int res_id;
938 	int error;
939 
940 	KASSERT(rman_get_start(res) == rman_get_end(res),
941 	    ("%s: more interrupts in resource", __func__));
942 
943 	res_id = (u_int)rman_get_start(res);
944 	if (intr_map_get_isrc(res_id) != NULL)
945 		panic("Attempt to double activation of resource id: %u\n",
946 		    res_id);
947 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
948 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
949 	if (error != 0) {
950 		free(data, M_INTRNG);
951 		/* XXX TODO DISCONECTED PICs */
952 		/* if (error == EINVAL) return(0); */
953 		return (error);
954 	}
955 	intr_map_set_isrc(res_id, isrc);
956 	rman_set_virtual(res, data);
957 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
958 }
959 
960 int
961 intr_deactivate_irq(device_t dev, struct resource *res)
962 {
963 	struct intr_map_data *data;
964 	struct intr_irqsrc *isrc;
965 	u_int res_id;
966 	int error;
967 
968 	KASSERT(rman_get_start(res) == rman_get_end(res),
969 	    ("%s: more interrupts in resource", __func__));
970 
971 	res_id = (u_int)rman_get_start(res);
972 	isrc = intr_map_get_isrc(res_id);
973 	if (isrc == NULL)
974 		panic("Attempt to deactivate non-active resource id: %u\n",
975 		    res_id);
976 
977 	data = rman_get_virtual(res);
978 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
979 	intr_map_set_isrc(res_id, NULL);
980 	rman_set_virtual(res, NULL);
981 	free(data, M_INTRNG);
982 	return (error);
983 }
984 
985 int
986 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
987     driver_intr_t hand, void *arg, int flags, void **cookiep)
988 {
989 	int error;
990 	struct intr_map_data *data;
991 	struct intr_irqsrc *isrc;
992 	const char *name;
993 	u_int res_id;
994 
995 	KASSERT(rman_get_start(res) == rman_get_end(res),
996 	    ("%s: more interrupts in resource", __func__));
997 
998 	res_id = (u_int)rman_get_start(res);
999 	isrc = intr_map_get_isrc(res_id);
1000 	if (isrc == NULL) {
1001 		/* XXX TODO DISCONECTED PICs */
1002 		return (EINVAL);
1003 	}
1004 
1005 	data = rman_get_virtual(res);
1006 	name = device_get_nameunit(dev);
1007 
1008 #ifdef INTR_SOLO
1009 	/*
1010 	 * Standard handling is done through MI interrupt framework. However,
1011 	 * some interrupts could request solely own special handling. This
1012 	 * non standard handling can be used for interrupt controllers without
1013 	 * handler (filter only), so in case that interrupt controllers are
1014 	 * chained, MI interrupt framework is called only in leaf controller.
1015 	 *
1016 	 * Note that root interrupt controller routine is served as well,
1017 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1018 	 */
1019 	if (flags & INTR_SOLO && hand != NULL) {
1020 		debugf("irq %u cannot solo on %s\n", irq, name);
1021 		return (EINVAL);
1022 	}
1023 
1024 	if (flags & INTR_SOLO) {
1025 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1026 		    arg, cookiep);
1027 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1028 		    name);
1029 	} else
1030 #endif
1031 		{
1032 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1033 		    cookiep);
1034 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1035 	}
1036 	if (error != 0)
1037 		return (error);
1038 
1039 	mtx_lock(&isrc_table_lock);
1040 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1041 	if (error == 0) {
1042 		isrc->isrc_handlers++;
1043 		if (isrc->isrc_handlers == 1)
1044 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1045 	}
1046 	mtx_unlock(&isrc_table_lock);
1047 	if (error != 0)
1048 		intr_event_remove_handler(*cookiep);
1049 	return (error);
1050 }
1051 
1052 int
1053 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1054 {
1055 	int error;
1056 	struct intr_map_data *data;
1057 	struct intr_irqsrc *isrc;
1058 	u_int res_id;
1059 
1060 	KASSERT(rman_get_start(res) == rman_get_end(res),
1061 	    ("%s: more interrupts in resource", __func__));
1062 
1063 	res_id = (u_int)rman_get_start(res);
1064 	isrc = intr_map_get_isrc(res_id);
1065 	if (isrc == NULL || isrc->isrc_handlers == 0)
1066 		return (EINVAL);
1067 
1068 	data = rman_get_virtual(res);
1069 
1070 #ifdef INTR_SOLO
1071 	if (isrc->isrc_filter != NULL) {
1072 		if (isrc != cookie)
1073 			return (EINVAL);
1074 
1075 		mtx_lock(&isrc_table_lock);
1076 		isrc->isrc_filter = NULL;
1077 		isrc->isrc_arg = NULL;
1078 		isrc->isrc_handlers = 0;
1079 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1080 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1081 		isrc_update_name(isrc, NULL);
1082 		mtx_unlock(&isrc_table_lock);
1083 		return (0);
1084 	}
1085 #endif
1086 	if (isrc != intr_handler_source(cookie))
1087 		return (EINVAL);
1088 
1089 	error = intr_event_remove_handler(cookie);
1090 	if (error == 0) {
1091 		mtx_lock(&isrc_table_lock);
1092 		isrc->isrc_handlers--;
1093 		if (isrc->isrc_handlers == 0)
1094 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1095 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1096 		intrcnt_updatename(isrc);
1097 		mtx_unlock(&isrc_table_lock);
1098 	}
1099 	return (error);
1100 }
1101 
1102 int
1103 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1104     const char *descr)
1105 {
1106 	int error;
1107 	struct intr_irqsrc *isrc;
1108 	u_int res_id;
1109 
1110 	KASSERT(rman_get_start(res) == rman_get_end(res),
1111 	    ("%s: more interrupts in resource", __func__));
1112 
1113 	res_id = (u_int)rman_get_start(res);
1114 	isrc = intr_map_get_isrc(res_id);
1115 	if (isrc == NULL || isrc->isrc_handlers == 0)
1116 		return (EINVAL);
1117 #ifdef INTR_SOLO
1118 	if (isrc->isrc_filter != NULL) {
1119 		if (isrc != cookie)
1120 			return (EINVAL);
1121 
1122 		mtx_lock(&isrc_table_lock);
1123 		isrc_update_name(isrc, descr);
1124 		mtx_unlock(&isrc_table_lock);
1125 		return (0);
1126 	}
1127 #endif
1128 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1129 	if (error == 0) {
1130 		mtx_lock(&isrc_table_lock);
1131 		intrcnt_updatename(isrc);
1132 		mtx_unlock(&isrc_table_lock);
1133 	}
1134 	return (error);
1135 }
1136 
1137 #ifdef SMP
1138 int
1139 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1140 {
1141 	struct intr_irqsrc *isrc;
1142 	u_int res_id;
1143 
1144 	KASSERT(rman_get_start(res) == rman_get_end(res),
1145 	    ("%s: more interrupts in resource", __func__));
1146 
1147 	res_id = (u_int)rman_get_start(res);
1148 	isrc = intr_map_get_isrc(res_id);
1149 	if (isrc == NULL || isrc->isrc_handlers == 0)
1150 		return (EINVAL);
1151 #ifdef INTR_SOLO
1152 	if (isrc->isrc_filter != NULL)
1153 		return (intr_isrc_assign_cpu(isrc, cpu));
1154 #endif
1155 	return (intr_event_bind(isrc->isrc_event, cpu));
1156 }
1157 
1158 /*
1159  * Return the CPU that the next interrupt source should use.
1160  * For now just returns the next CPU according to round-robin.
1161  */
1162 u_int
1163 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1164 {
1165 
1166 	if (!irq_assign_cpu || mp_ncpus == 1)
1167 		return (PCPU_GET(cpuid));
1168 
1169 	do {
1170 		last_cpu++;
1171 		if (last_cpu > mp_maxid)
1172 			last_cpu = 0;
1173 	} while (!CPU_ISSET(last_cpu, cpumask));
1174 	return (last_cpu);
1175 }
1176 
1177 /*
1178  *  Distribute all the interrupt sources among the available
1179  *  CPUs once the AP's have been launched.
1180  */
1181 static void
1182 intr_irq_shuffle(void *arg __unused)
1183 {
1184 	struct intr_irqsrc *isrc;
1185 	u_int i;
1186 
1187 	if (mp_ncpus == 1)
1188 		return;
1189 
1190 	mtx_lock(&isrc_table_lock);
1191 	irq_assign_cpu = TRUE;
1192 	for (i = 0; i < NIRQ; i++) {
1193 		isrc = irq_sources[i];
1194 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1195 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1196 			continue;
1197 
1198 		if (isrc->isrc_event != NULL &&
1199 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1200 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1201 			panic("%s: CPU inconsistency", __func__);
1202 
1203 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1204 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1205 
1206 		/*
1207 		 * We are in wicked position here if the following call fails
1208 		 * for bound ISRC. The best thing we can do is to clear
1209 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1210 		 */
1211 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1212 			CPU_ZERO(&isrc->isrc_cpu);
1213 	}
1214 	mtx_unlock(&isrc_table_lock);
1215 }
1216 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1217 
1218 #else
1219 u_int
1220 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1221 {
1222 
1223 	return (PCPU_GET(cpuid));
1224 }
1225 #endif
1226 
1227 /*
1228  * Allocate memory for new intr_map_data structure.
1229  * Initialize common fields.
1230  */
1231 struct intr_map_data *
1232 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1233 {
1234 	struct intr_map_data *data;
1235 
1236 	data = malloc(len, M_INTRNG, flags);
1237 	data->type = type;
1238 	data->len = len;
1239 	return (data);
1240 }
1241 
1242 void intr_free_intr_map_data(struct intr_map_data *data)
1243 {
1244 
1245 	free(data, M_INTRNG);
1246 }
1247 
1248 
1249 /*
1250  *  Register a MSI/MSI-X interrupt controller
1251  */
1252 int
1253 intr_msi_register(device_t dev, intptr_t xref)
1254 {
1255 	struct intr_pic *pic;
1256 
1257 	if (dev == NULL)
1258 		return (EINVAL);
1259 	pic = pic_create(dev, xref);
1260 	if (pic == NULL)
1261 		return (ENOMEM);
1262 
1263 	pic->pic_flags |= FLAG_MSI;
1264 
1265 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1266 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1267 	return (0);
1268 }
1269 
1270 int
1271 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1272     int maxcount, int *irqs)
1273 {
1274 	struct intr_irqsrc **isrc;
1275 	struct intr_pic *pic;
1276 	device_t pdev;
1277 	struct intr_map_data_msi *msi;
1278 	int err, i;
1279 
1280 	pic = pic_lookup(NULL, xref);
1281 	if (pic == NULL)
1282 		return (ESRCH);
1283 
1284 	KASSERT((pic->pic_flags & FLAG_MSI) != 0,
1285 	    ("%s: Found a non-MSI controller: %s", __func__,
1286 	     device_get_name(pic->pic_dev)));
1287 
1288 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1289 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1290 	if (err != 0) {
1291 		free(isrc, M_INTRNG);
1292 		return (err);
1293 	}
1294 
1295 	for (i = 0; i < count; i++) {
1296 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1297 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1298 		msi-> isrc = isrc[i];
1299 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1300 		    (struct intr_map_data *)msi);
1301 
1302 	}
1303 	free(isrc, M_INTRNG);
1304 
1305 	return (err);
1306 }
1307 
1308 int
1309 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1310     int *irqs)
1311 {
1312 	struct intr_irqsrc **isrc;
1313 	struct intr_pic *pic;
1314 	struct intr_map_data_msi *msi;
1315 	int i, err;
1316 
1317 	pic = pic_lookup(NULL, xref);
1318 	if (pic == NULL)
1319 		return (ESRCH);
1320 
1321 	KASSERT((pic->pic_flags & FLAG_MSI) != 0,
1322 	    ("%s: Found a non-MSI controller: %s", __func__,
1323 	     device_get_name(pic->pic_dev)));
1324 
1325 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1326 
1327 	for (i = 0; i < count; i++) {
1328 		msi = (struct intr_map_data_msi *)
1329 		    intr_map_get_map_data(irqs[i]);
1330 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1331 		    ("%s: irq %d map data is not MSI", __func__,
1332 		    irqs[i]));
1333 		isrc[i] = msi->isrc;
1334 	}
1335 
1336 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1337 
1338 	for (i = 0; i < count; i++) {
1339 		if (isrc[i] != NULL)
1340 			intr_unmap_irq(irqs[i]);
1341 	}
1342 
1343 	free(isrc, M_INTRNG);
1344 	return (err);
1345 }
1346 
1347 int
1348 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1349 {
1350 	struct intr_irqsrc *isrc;
1351 	struct intr_pic *pic;
1352 	device_t pdev;
1353 	struct intr_map_data_msi *msi;
1354 	int err;
1355 
1356 	pic = pic_lookup(NULL, xref);
1357 	if (pic == NULL)
1358 		return (ESRCH);
1359 
1360 	KASSERT((pic->pic_flags & FLAG_MSI) != 0,
1361 	    ("%s: Found a non-MSI controller: %s", __func__,
1362 	     device_get_name(pic->pic_dev)));
1363 
1364 
1365 	err = MSI_ALLOC_MSIX(pic->pic_dev, child, &pdev, &isrc);
1366 	if (err != 0)
1367 		return (err);
1368 
1369 	msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1370 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1371 	msi->isrc = isrc;
1372 	*irq = intr_map_irq(pic->pic_dev, xref, (struct intr_map_data *)msi);
1373 	return (0);
1374 }
1375 
1376 int
1377 intr_release_msix(device_t pci, device_t child, intptr_t xref, int irq)
1378 {
1379 	struct intr_irqsrc *isrc;
1380 	struct intr_pic *pic;
1381 	struct intr_map_data_msi *msi;
1382 	int err;
1383 
1384 	pic = pic_lookup(NULL, xref);
1385 	if (pic == NULL)
1386 		return (ESRCH);
1387 
1388 	KASSERT((pic->pic_flags & FLAG_MSI) != 0,
1389 	    ("%s: Found a non-MSI controller: %s", __func__,
1390 	     device_get_name(pic->pic_dev)));
1391 
1392 	msi = (struct intr_map_data_msi *)
1393 	    intr_map_get_map_data(irq);
1394 	KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1395 	    ("%s: irq %d map data is not MSI", __func__,
1396 	    irq));
1397 	isrc = msi->isrc;
1398 	if (isrc == NULL) {
1399 		intr_unmap_irq(irq);
1400 		return (EINVAL);
1401 	}
1402 
1403 	err = MSI_RELEASE_MSIX(pic->pic_dev, child, isrc);
1404 	intr_unmap_irq(irq);
1405 
1406 	return (err);
1407 }
1408 
1409 int
1410 intr_map_msi(device_t pci, device_t child, intptr_t xref, int irq,
1411     uint64_t *addr, uint32_t *data)
1412 {
1413 	struct intr_irqsrc *isrc;
1414 	struct intr_pic *pic;
1415 	int err;
1416 
1417 	pic = pic_lookup(NULL, xref);
1418 	if (pic == NULL)
1419 		return (ESRCH);
1420 
1421 	KASSERT((pic->pic_flags & FLAG_MSI) != 0,
1422 	    ("%s: Found a non-MSI controller: %s", __func__,
1423 	     device_get_name(pic->pic_dev)));
1424 
1425 	isrc = intr_map_get_isrc(irq);
1426 	if (isrc == NULL)
1427 		return (EINVAL);
1428 
1429 	err = MSI_MAP_MSI(pic->pic_dev, child, isrc, addr, data);
1430 	return (err);
1431 }
1432 
1433 
1434 void dosoftints(void);
1435 void
1436 dosoftints(void)
1437 {
1438 }
1439 
1440 #ifdef SMP
1441 /*
1442  *  Init interrupt controller on another CPU.
1443  */
1444 void
1445 intr_pic_init_secondary(void)
1446 {
1447 
1448 	/*
1449 	 * QQQ: Only root PIC is aware of other CPUs ???
1450 	 */
1451 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1452 
1453 	//mtx_lock(&isrc_table_lock);
1454 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1455 	//mtx_unlock(&isrc_table_lock);
1456 }
1457 #endif
1458 
1459 #ifdef DDB
1460 DB_SHOW_COMMAND(irqs, db_show_irqs)
1461 {
1462 	u_int i, irqsum;
1463 	u_long num;
1464 	struct intr_irqsrc *isrc;
1465 
1466 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1467 		isrc = irq_sources[i];
1468 		if (isrc == NULL)
1469 			continue;
1470 
1471 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1472 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1473 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1474 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1475 		irqsum += num;
1476 	}
1477 	db_printf("irq total %u\n", irqsum);
1478 }
1479 #endif
1480 
1481 /*
1482  * Interrupt mapping table functions.
1483  *
1484  * Please, keep this part separately, it can be transformed to
1485  * extension of standard resources.
1486  */
1487 struct intr_map_entry
1488 {
1489 	device_t 		dev;
1490 	intptr_t 		xref;
1491 	struct intr_map_data 	*map_data;
1492 	struct intr_irqsrc 	*isrc;
1493 	/* XXX TODO DISCONECTED PICs */
1494 	/*int			flags */
1495 };
1496 
1497 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1498 static struct intr_map_entry *irq_map[2 * NIRQ];
1499 static int irq_map_count = nitems(irq_map);
1500 static int irq_map_first_free_idx;
1501 static struct mtx irq_map_lock;
1502 
1503 static struct intr_irqsrc *
1504 intr_map_get_isrc(u_int res_id)
1505 {
1506 	struct intr_irqsrc *isrc;
1507 
1508 	mtx_lock(&irq_map_lock);
1509 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1510 		mtx_unlock(&irq_map_lock);
1511 		return (NULL);
1512 	}
1513 	isrc = irq_map[res_id]->isrc;
1514 	mtx_unlock(&irq_map_lock);
1515 	return (isrc);
1516 }
1517 
1518 static void
1519 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1520 {
1521 
1522 	mtx_lock(&irq_map_lock);
1523 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1524 		mtx_unlock(&irq_map_lock);
1525 		return;
1526 	}
1527 	irq_map[res_id]->isrc = isrc;
1528 	mtx_unlock(&irq_map_lock);
1529 }
1530 
1531 /*
1532  * Get a copy of intr_map_entry data
1533  */
1534 static struct intr_map_data *
1535 intr_map_get_map_data(u_int res_id)
1536 {
1537 	struct intr_map_data *data;
1538 
1539 	data = NULL;
1540 	mtx_lock(&irq_map_lock);
1541 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1542 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1543 	data = irq_map[res_id]->map_data;
1544 	mtx_unlock(&irq_map_lock);
1545 
1546 	return (data);
1547 }
1548 
1549 /*
1550  * Get a copy of intr_map_entry data
1551  */
1552 static void
1553 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1554     struct intr_map_data **data)
1555 {
1556 	size_t len;
1557 
1558 	len = 0;
1559 	mtx_lock(&irq_map_lock);
1560 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1561 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1562 	if (irq_map[res_id]->map_data != NULL)
1563 		len = irq_map[res_id]->map_data->len;
1564 	mtx_unlock(&irq_map_lock);
1565 
1566 	if (len == 0)
1567 		*data = NULL;
1568 	else
1569 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1570 	mtx_lock(&irq_map_lock);
1571 	if (irq_map[res_id] == NULL)
1572 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1573 	if (len != 0) {
1574 		if (len != irq_map[res_id]->map_data->len)
1575 			panic("Resource id: %u has changed.\n", res_id);
1576 		memcpy(*data, irq_map[res_id]->map_data, len);
1577 	}
1578 	*map_dev = irq_map[res_id]->dev;
1579 	*map_xref = irq_map[res_id]->xref;
1580 	mtx_unlock(&irq_map_lock);
1581 }
1582 
1583 
1584 /*
1585  * Allocate and fill new entry in irq_map table.
1586  */
1587 u_int
1588 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1589 {
1590 	u_int i;
1591 	struct intr_map_entry *entry;
1592 
1593 	/* Prepare new entry first. */
1594 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1595 
1596 	entry->dev = dev;
1597 	entry->xref = xref;
1598 	entry->map_data = data;
1599 	entry->isrc = NULL;
1600 
1601 	mtx_lock(&irq_map_lock);
1602 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1603 		if (irq_map[i] == NULL) {
1604 			irq_map[i] = entry;
1605 			irq_map_first_free_idx = i + 1;
1606 			mtx_unlock(&irq_map_lock);
1607 			return (i);
1608 		}
1609 	}
1610 	mtx_unlock(&irq_map_lock);
1611 
1612 	/* XXX Expand irq_map table */
1613 	panic("IRQ mapping table is full.");
1614 }
1615 
1616 /*
1617  * Remove and free mapping entry.
1618  */
1619 void
1620 intr_unmap_irq(u_int res_id)
1621 {
1622 	struct intr_map_entry *entry;
1623 
1624 	mtx_lock(&irq_map_lock);
1625 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1626 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1627 	entry = irq_map[res_id];
1628 	irq_map[res_id] = NULL;
1629 	irq_map_first_free_idx = res_id;
1630 	mtx_unlock(&irq_map_lock);
1631 	intr_free_intr_map_data(entry->map_data);
1632 	free(entry, M_INTRNG);
1633 }
1634 
1635 /*
1636  * Clone mapping entry.
1637  */
1638 u_int
1639 intr_map_clone_irq(u_int old_res_id)
1640 {
1641 	device_t map_dev;
1642 	intptr_t map_xref;
1643 	struct intr_map_data *data;
1644 
1645 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1646 	return (intr_map_irq(map_dev, map_xref, data));
1647 }
1648 
1649 static void
1650 intr_map_init(void *dummy __unused)
1651 {
1652 
1653 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1654 }
1655 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1656