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: stable/12/sys/kern/subr_intr.c 368629 2020-12-14 10:14:32Z mmel $");
30 
31 /*
32  *	New-style Interrupt Framework
33  *
34  *  TODO: - add support for disconnected PICs.
35  *        - to support IPI (PPI) enabling on other CPUs if already started.
36  *        - to complete things for removable PICs.
37  */
38 
39 #include "opt_ddb.h"
40 #include "opt_hwpmc_hooks.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/syslog.h>
46 #include <sys/malloc.h>
47 #include <sys/proc.h>
48 #include <sys/queue.h>
49 #include <sys/bus.h>
50 #include <sys/interrupt.h>
51 #include <sys/conf.h>
52 #include <sys/cpuset.h>
53 #include <sys/rman.h>
54 #include <sys/sched.h>
55 #include <sys/smp.h>
56 #include <sys/vmmeter.h>
57 #ifdef HWPMC_HOOKS
58 #include <sys/pmckern.h>
59 #endif
60 
61 #include <machine/atomic.h>
62 #include <machine/intr.h>
63 #include <machine/cpu.h>
64 #include <machine/smp.h>
65 #include <machine/stdarg.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70 
71 #include "pic_if.h"
72 #include "msi_if.h"
73 
74 #define	INTRNAME_LEN	(2*MAXCOMLEN + 1)
75 
76 #ifdef DEBUG
77 #define debugf(fmt, args...) do { printf("%s(): ", __func__);	\
78     printf(fmt,##args); } while (0)
79 #else
80 #define debugf(fmt, args...)
81 #endif
82 
83 MALLOC_DECLARE(M_INTRNG);
84 MALLOC_DEFINE(M_INTRNG, "intr", "intr interrupt handling");
85 
86 /* Main interrupt handler called from assembler -> 'hidden' for C code. */
87 void intr_irq_handler(struct trapframe *tf);
88 
89 /* Root interrupt controller stuff. */
90 device_t intr_irq_root_dev;
91 static intr_irq_filter_t *irq_root_filter;
92 static void *irq_root_arg;
93 static u_int irq_root_ipicount;
94 
95 struct intr_pic_child {
96 	SLIST_ENTRY(intr_pic_child)	 pc_next;
97 	struct intr_pic			*pc_pic;
98 	intr_child_irq_filter_t		*pc_filter;
99 	void				*pc_filter_arg;
100 	uintptr_t			 pc_start;
101 	uintptr_t			 pc_length;
102 };
103 
104 /* Interrupt controller definition. */
105 struct intr_pic {
106 	SLIST_ENTRY(intr_pic)	pic_next;
107 	intptr_t		pic_xref;	/* hardware identification */
108 	device_t		pic_dev;
109 /* Only one of FLAG_PIC or FLAG_MSI may be set */
110 #define	FLAG_PIC	(1 << 0)
111 #define	FLAG_MSI	(1 << 1)
112 #define	FLAG_TYPE_MASK	(FLAG_PIC | FLAG_MSI)
113 	u_int			pic_flags;
114 	struct mtx		pic_child_lock;
115 	SLIST_HEAD(, intr_pic_child) pic_children;
116 };
117 
118 static struct mtx pic_list_lock;
119 static SLIST_HEAD(, intr_pic) pic_list;
120 
121 static struct intr_pic *pic_lookup(device_t dev, intptr_t xref, int flags);
122 
123 /* Interrupt source definition. */
124 static struct mtx isrc_table_lock;
125 static struct intr_irqsrc *irq_sources[NIRQ];
126 u_int irq_next_free;
127 
128 #ifdef SMP
129 static boolean_t irq_assign_cpu = FALSE;
130 #endif
131 
132 /*
133  * - 2 counters for each I/O interrupt.
134  * - MAXCPU counters for each IPI counters for SMP.
135  */
136 #ifdef SMP
137 #define INTRCNT_COUNT   (NIRQ * 2 + INTR_IPI_COUNT * MAXCPU)
138 #else
139 #define INTRCNT_COUNT   (NIRQ * 2)
140 #endif
141 
142 /* Data for MI statistics reporting. */
143 u_long intrcnt[INTRCNT_COUNT];
144 char intrnames[INTRCNT_COUNT * INTRNAME_LEN];
145 size_t sintrcnt = sizeof(intrcnt);
146 size_t sintrnames = sizeof(intrnames);
147 static u_int intrcnt_index;
148 
149 static struct intr_irqsrc *intr_map_get_isrc(u_int res_id);
150 static void intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc);
151 static struct intr_map_data * intr_map_get_map_data(u_int res_id);
152 static void intr_map_copy_map_data(u_int res_id, device_t *dev, intptr_t *xref,
153     struct intr_map_data **data);
154 
155 /*
156  *  Interrupt framework initialization routine.
157  */
158 static void
intr_irq_init(void * dummy __unused)159 intr_irq_init(void *dummy __unused)
160 {
161 
162 	SLIST_INIT(&pic_list);
163 	mtx_init(&pic_list_lock, "intr pic list", NULL, MTX_DEF);
164 
165 	mtx_init(&isrc_table_lock, "intr isrc table", NULL, MTX_DEF);
166 }
167 SYSINIT(intr_irq_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_irq_init, NULL);
168 
169 static void
intrcnt_setname(const char * name,int index)170 intrcnt_setname(const char *name, int index)
171 {
172 
173 	snprintf(intrnames + INTRNAME_LEN * index, INTRNAME_LEN, "%-*s",
174 	    INTRNAME_LEN - 1, name);
175 }
176 
177 /*
178  *  Update name for interrupt source with interrupt event.
179  */
180 static void
intrcnt_updatename(struct intr_irqsrc * isrc)181 intrcnt_updatename(struct intr_irqsrc *isrc)
182 {
183 
184 	/* QQQ: What about stray counter name? */
185 	mtx_assert(&isrc_table_lock, MA_OWNED);
186 	intrcnt_setname(isrc->isrc_event->ie_fullname, isrc->isrc_index);
187 }
188 
189 /*
190  *  Virtualization for interrupt source interrupt counter increment.
191  */
192 static inline void
isrc_increment_count(struct intr_irqsrc * isrc)193 isrc_increment_count(struct intr_irqsrc *isrc)
194 {
195 
196 	if (isrc->isrc_flags & INTR_ISRCF_PPI)
197 		atomic_add_long(&isrc->isrc_count[0], 1);
198 	else
199 		isrc->isrc_count[0]++;
200 }
201 
202 /*
203  *  Virtualization for interrupt source interrupt stray counter increment.
204  */
205 static inline void
isrc_increment_straycount(struct intr_irqsrc * isrc)206 isrc_increment_straycount(struct intr_irqsrc *isrc)
207 {
208 
209 	isrc->isrc_count[1]++;
210 }
211 
212 /*
213  *  Virtualization for interrupt source interrupt name update.
214  */
215 static void
isrc_update_name(struct intr_irqsrc * isrc,const char * name)216 isrc_update_name(struct intr_irqsrc *isrc, const char *name)
217 {
218 	char str[INTRNAME_LEN];
219 
220 	mtx_assert(&isrc_table_lock, MA_OWNED);
221 
222 	if (name != NULL) {
223 		snprintf(str, INTRNAME_LEN, "%s: %s", isrc->isrc_name, name);
224 		intrcnt_setname(str, isrc->isrc_index);
225 		snprintf(str, INTRNAME_LEN, "stray %s: %s", isrc->isrc_name,
226 		    name);
227 		intrcnt_setname(str, isrc->isrc_index + 1);
228 	} else {
229 		snprintf(str, INTRNAME_LEN, "%s:", isrc->isrc_name);
230 		intrcnt_setname(str, isrc->isrc_index);
231 		snprintf(str, INTRNAME_LEN, "stray %s:", isrc->isrc_name);
232 		intrcnt_setname(str, isrc->isrc_index + 1);
233 	}
234 }
235 
236 /*
237  *  Virtualization for interrupt source interrupt counters setup.
238  */
239 static void
isrc_setup_counters(struct intr_irqsrc * isrc)240 isrc_setup_counters(struct intr_irqsrc *isrc)
241 {
242 	u_int index;
243 
244 	/*
245 	 *  XXX - it does not work well with removable controllers and
246 	 *        interrupt sources !!!
247 	 */
248 	index = atomic_fetchadd_int(&intrcnt_index, 2);
249 	isrc->isrc_index = index;
250 	isrc->isrc_count = &intrcnt[index];
251 	isrc_update_name(isrc, NULL);
252 }
253 
254 /*
255  *  Virtualization for interrupt source interrupt counters release.
256  */
257 static void
isrc_release_counters(struct intr_irqsrc * isrc)258 isrc_release_counters(struct intr_irqsrc *isrc)
259 {
260 
261 	panic("%s: not implemented", __func__);
262 }
263 
264 #ifdef SMP
265 /*
266  *  Virtualization for interrupt source IPI counters setup.
267  */
268 u_long *
intr_ipi_setup_counters(const char * name)269 intr_ipi_setup_counters(const char *name)
270 {
271 	u_int index, i;
272 	char str[INTRNAME_LEN];
273 
274 	index = atomic_fetchadd_int(&intrcnt_index, MAXCPU);
275 	for (i = 0; i < MAXCPU; i++) {
276 		snprintf(str, INTRNAME_LEN, "cpu%d:%s", i, name);
277 		intrcnt_setname(str, index + i);
278 	}
279 	return (&intrcnt[index]);
280 }
281 #endif
282 
283 /*
284  *  Main interrupt dispatch handler. It's called straight
285  *  from the assembler, where CPU interrupt is served.
286  */
287 void
intr_irq_handler(struct trapframe * tf)288 intr_irq_handler(struct trapframe *tf)
289 {
290 	struct trapframe * oldframe;
291 	struct thread * td;
292 
293 	KASSERT(irq_root_filter != NULL, ("%s: no filter", __func__));
294 
295 	VM_CNT_INC(v_intr);
296 	critical_enter();
297 	td = curthread;
298 	oldframe = td->td_intr_frame;
299 	td->td_intr_frame = tf;
300 	irq_root_filter(irq_root_arg);
301 	td->td_intr_frame = oldframe;
302 	critical_exit();
303 #ifdef HWPMC_HOOKS
304 	if (pmc_hook && TRAPF_USERMODE(tf) &&
305 	    (PCPU_GET(curthread)->td_pflags & TDP_CALLCHAIN))
306 		pmc_hook(PCPU_GET(curthread), PMC_FN_USER_CALLCHAIN, tf);
307 #endif
308 }
309 
310 int
intr_child_irq_handler(struct intr_pic * parent,uintptr_t irq)311 intr_child_irq_handler(struct intr_pic *parent, uintptr_t irq)
312 {
313 	struct intr_pic_child *child;
314 	bool found;
315 
316 	found = false;
317 	mtx_lock_spin(&parent->pic_child_lock);
318 	SLIST_FOREACH(child, &parent->pic_children, pc_next) {
319 		if (child->pc_start <= irq &&
320 		    irq < (child->pc_start + child->pc_length)) {
321 			found = true;
322 			break;
323 		}
324 	}
325 	mtx_unlock_spin(&parent->pic_child_lock);
326 
327 	if (found)
328 		return (child->pc_filter(child->pc_filter_arg, irq));
329 
330 	return (FILTER_STRAY);
331 }
332 
333 /*
334  *  interrupt controller dispatch function for interrupts. It should
335  *  be called straight from the interrupt controller, when associated interrupt
336  *  source is learned.
337  */
338 int
intr_isrc_dispatch(struct intr_irqsrc * isrc,struct trapframe * tf)339 intr_isrc_dispatch(struct intr_irqsrc *isrc, struct trapframe *tf)
340 {
341 
342 	KASSERT(isrc != NULL, ("%s: no source", __func__));
343 
344 	isrc_increment_count(isrc);
345 
346 #ifdef INTR_SOLO
347 	if (isrc->isrc_filter != NULL) {
348 		int error;
349 		error = isrc->isrc_filter(isrc->isrc_arg, tf);
350 		PIC_POST_FILTER(isrc->isrc_dev, isrc);
351 		if (error == FILTER_HANDLED)
352 			return (0);
353 	} else
354 #endif
355 	if (isrc->isrc_event != NULL) {
356 		if (intr_event_handle(isrc->isrc_event, tf) == 0)
357 			return (0);
358 	}
359 
360 	isrc_increment_straycount(isrc);
361 	return (EINVAL);
362 }
363 
364 /*
365  *  Alloc unique interrupt number (resource handle) for interrupt source.
366  *
367  *  There could be various strategies how to allocate free interrupt number
368  *  (resource handle) for new interrupt source.
369  *
370  *  1. Handles are always allocated forward, so handles are not recycled
371  *     immediately. However, if only one free handle left which is reused
372  *     constantly...
373  */
374 static inline int
isrc_alloc_irq(struct intr_irqsrc * isrc)375 isrc_alloc_irq(struct intr_irqsrc *isrc)
376 {
377 	u_int maxirqs, irq;
378 
379 	mtx_assert(&isrc_table_lock, MA_OWNED);
380 
381 	maxirqs = nitems(irq_sources);
382 	if (irq_next_free >= maxirqs)
383 		return (ENOSPC);
384 
385 	for (irq = irq_next_free; irq < maxirqs; irq++) {
386 		if (irq_sources[irq] == NULL)
387 			goto found;
388 	}
389 	for (irq = 0; irq < irq_next_free; irq++) {
390 		if (irq_sources[irq] == NULL)
391 			goto found;
392 	}
393 
394 	irq_next_free = maxirqs;
395 	return (ENOSPC);
396 
397 found:
398 	isrc->isrc_irq = irq;
399 	irq_sources[irq] = isrc;
400 
401 	irq_next_free = irq + 1;
402 	if (irq_next_free >= maxirqs)
403 		irq_next_free = 0;
404 	return (0);
405 }
406 
407 /*
408  *  Free unique interrupt number (resource handle) from interrupt source.
409  */
410 static inline int
isrc_free_irq(struct intr_irqsrc * isrc)411 isrc_free_irq(struct intr_irqsrc *isrc)
412 {
413 
414 	mtx_assert(&isrc_table_lock, MA_OWNED);
415 
416 	if (isrc->isrc_irq >= nitems(irq_sources))
417 		return (EINVAL);
418 	if (irq_sources[isrc->isrc_irq] != isrc)
419 		return (EINVAL);
420 
421 	irq_sources[isrc->isrc_irq] = NULL;
422 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
423 	return (0);
424 }
425 
426 /*
427  *  Initialize interrupt source and register it into global interrupt table.
428  */
429 int
intr_isrc_register(struct intr_irqsrc * isrc,device_t dev,u_int flags,const char * fmt,...)430 intr_isrc_register(struct intr_irqsrc *isrc, device_t dev, u_int flags,
431     const char *fmt, ...)
432 {
433 	int error;
434 	va_list ap;
435 
436 	bzero(isrc, sizeof(struct intr_irqsrc));
437 	isrc->isrc_dev = dev;
438 	isrc->isrc_irq = INTR_IRQ_INVALID;	/* just to be safe */
439 	isrc->isrc_flags = flags;
440 
441 	va_start(ap, fmt);
442 	vsnprintf(isrc->isrc_name, INTR_ISRC_NAMELEN, fmt, ap);
443 	va_end(ap);
444 
445 	mtx_lock(&isrc_table_lock);
446 	error = isrc_alloc_irq(isrc);
447 	if (error != 0) {
448 		mtx_unlock(&isrc_table_lock);
449 		return (error);
450 	}
451 	/*
452 	 * Setup interrupt counters, but not for IPI sources. Those are setup
453 	 * later and only for used ones (up to INTR_IPI_COUNT) to not exhaust
454 	 * our counter pool.
455 	 */
456 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
457 		isrc_setup_counters(isrc);
458 	mtx_unlock(&isrc_table_lock);
459 	return (0);
460 }
461 
462 /*
463  *  Deregister interrupt source from global interrupt table.
464  */
465 int
intr_isrc_deregister(struct intr_irqsrc * isrc)466 intr_isrc_deregister(struct intr_irqsrc *isrc)
467 {
468 	int error;
469 
470 	mtx_lock(&isrc_table_lock);
471 	if ((isrc->isrc_flags & INTR_ISRCF_IPI) == 0)
472 		isrc_release_counters(isrc);
473 	error = isrc_free_irq(isrc);
474 	mtx_unlock(&isrc_table_lock);
475 	return (error);
476 }
477 
478 #ifdef SMP
479 /*
480  *  A support function for a PIC to decide if provided ISRC should be inited
481  *  on given cpu. The logic of INTR_ISRCF_BOUND flag and isrc_cpu member of
482  *  struct intr_irqsrc is the following:
483  *
484  *     If INTR_ISRCF_BOUND is set, the ISRC should be inited only on cpus
485  *     set in isrc_cpu. If not, the ISRC should be inited on every cpu and
486  *     isrc_cpu is kept consistent with it. Thus isrc_cpu is always correct.
487  */
488 bool
intr_isrc_init_on_cpu(struct intr_irqsrc * isrc,u_int cpu)489 intr_isrc_init_on_cpu(struct intr_irqsrc *isrc, u_int cpu)
490 {
491 
492 	if (isrc->isrc_handlers == 0)
493 		return (false);
494 	if ((isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI)) == 0)
495 		return (false);
496 	if (isrc->isrc_flags & INTR_ISRCF_BOUND)
497 		return (CPU_ISSET(cpu, &isrc->isrc_cpu));
498 
499 	CPU_SET(cpu, &isrc->isrc_cpu);
500 	return (true);
501 }
502 #endif
503 
504 #ifdef INTR_SOLO
505 /*
506  *  Setup filter into interrupt source.
507  */
508 static int
iscr_setup_filter(struct intr_irqsrc * isrc,const char * name,intr_irq_filter_t * filter,void * arg,void ** cookiep)509 iscr_setup_filter(struct intr_irqsrc *isrc, const char *name,
510     intr_irq_filter_t *filter, void *arg, void **cookiep)
511 {
512 
513 	if (filter == NULL)
514 		return (EINVAL);
515 
516 	mtx_lock(&isrc_table_lock);
517 	/*
518 	 * Make sure that we do not mix the two ways
519 	 * how we handle interrupt sources.
520 	 */
521 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
522 		mtx_unlock(&isrc_table_lock);
523 		return (EBUSY);
524 	}
525 	isrc->isrc_filter = filter;
526 	isrc->isrc_arg = arg;
527 	isrc_update_name(isrc, name);
528 	mtx_unlock(&isrc_table_lock);
529 
530 	*cookiep = isrc;
531 	return (0);
532 }
533 #endif
534 
535 /*
536  *  Interrupt source pre_ithread method for MI interrupt framework.
537  */
538 static void
intr_isrc_pre_ithread(void * arg)539 intr_isrc_pre_ithread(void *arg)
540 {
541 	struct intr_irqsrc *isrc = arg;
542 
543 	PIC_PRE_ITHREAD(isrc->isrc_dev, isrc);
544 }
545 
546 /*
547  *  Interrupt source post_ithread method for MI interrupt framework.
548  */
549 static void
intr_isrc_post_ithread(void * arg)550 intr_isrc_post_ithread(void *arg)
551 {
552 	struct intr_irqsrc *isrc = arg;
553 
554 	PIC_POST_ITHREAD(isrc->isrc_dev, isrc);
555 }
556 
557 /*
558  *  Interrupt source post_filter method for MI interrupt framework.
559  */
560 static void
intr_isrc_post_filter(void * arg)561 intr_isrc_post_filter(void *arg)
562 {
563 	struct intr_irqsrc *isrc = arg;
564 
565 	PIC_POST_FILTER(isrc->isrc_dev, isrc);
566 }
567 
568 /*
569  *  Interrupt source assign_cpu method for MI interrupt framework.
570  */
571 static int
intr_isrc_assign_cpu(void * arg,int cpu)572 intr_isrc_assign_cpu(void *arg, int cpu)
573 {
574 #ifdef SMP
575 	struct intr_irqsrc *isrc = arg;
576 	int error;
577 
578 	mtx_lock(&isrc_table_lock);
579 	if (cpu == NOCPU) {
580 		CPU_ZERO(&isrc->isrc_cpu);
581 		isrc->isrc_flags &= ~INTR_ISRCF_BOUND;
582 	} else {
583 		CPU_SETOF(cpu, &isrc->isrc_cpu);
584 		isrc->isrc_flags |= INTR_ISRCF_BOUND;
585 	}
586 
587 	/*
588 	 * In NOCPU case, it's up to PIC to either leave ISRC on same CPU or
589 	 * re-balance it to another CPU or enable it on more CPUs. However,
590 	 * PIC is expected to change isrc_cpu appropriately to keep us well
591 	 * informed if the call is successful.
592 	 */
593 	if (irq_assign_cpu) {
594 		error = PIC_BIND_INTR(isrc->isrc_dev, isrc);
595 		if (error) {
596 			CPU_ZERO(&isrc->isrc_cpu);
597 			mtx_unlock(&isrc_table_lock);
598 			return (error);
599 		}
600 	}
601 	mtx_unlock(&isrc_table_lock);
602 	return (0);
603 #else
604 	return (EOPNOTSUPP);
605 #endif
606 }
607 
608 /*
609  *  Create interrupt event for interrupt source.
610  */
611 static int
isrc_event_create(struct intr_irqsrc * isrc)612 isrc_event_create(struct intr_irqsrc *isrc)
613 {
614 	struct intr_event *ie;
615 	int error;
616 
617 	error = intr_event_create(&ie, isrc, 0, isrc->isrc_irq,
618 	    intr_isrc_pre_ithread, intr_isrc_post_ithread, intr_isrc_post_filter,
619 	    intr_isrc_assign_cpu, "%s:", isrc->isrc_name);
620 	if (error)
621 		return (error);
622 
623 	mtx_lock(&isrc_table_lock);
624 	/*
625 	 * Make sure that we do not mix the two ways
626 	 * how we handle interrupt sources. Let contested event wins.
627 	 */
628 #ifdef INTR_SOLO
629 	if (isrc->isrc_filter != NULL || isrc->isrc_event != NULL) {
630 #else
631 	if (isrc->isrc_event != NULL) {
632 #endif
633 		mtx_unlock(&isrc_table_lock);
634 		intr_event_destroy(ie);
635 		return (isrc->isrc_event != NULL ? EBUSY : 0);
636 	}
637 	isrc->isrc_event = ie;
638 	mtx_unlock(&isrc_table_lock);
639 
640 	return (0);
641 }
642 #ifdef notyet
643 /*
644  *  Destroy interrupt event for interrupt source.
645  */
646 static void
647 isrc_event_destroy(struct intr_irqsrc *isrc)
648 {
649 	struct intr_event *ie;
650 
651 	mtx_lock(&isrc_table_lock);
652 	ie = isrc->isrc_event;
653 	isrc->isrc_event = NULL;
654 	mtx_unlock(&isrc_table_lock);
655 
656 	if (ie != NULL)
657 		intr_event_destroy(ie);
658 }
659 #endif
660 /*
661  *  Add handler to interrupt source.
662  */
663 static int
664 isrc_add_handler(struct intr_irqsrc *isrc, const char *name,
665     driver_filter_t filter, driver_intr_t handler, void *arg,
666     enum intr_type flags, void **cookiep)
667 {
668 	int error;
669 
670 	if (isrc->isrc_event == NULL) {
671 		error = isrc_event_create(isrc);
672 		if (error)
673 			return (error);
674 	}
675 
676 	error = intr_event_add_handler(isrc->isrc_event, name, filter, handler,
677 	    arg, intr_priority(flags), flags, cookiep);
678 	if (error == 0) {
679 		mtx_lock(&isrc_table_lock);
680 		intrcnt_updatename(isrc);
681 		mtx_unlock(&isrc_table_lock);
682 	}
683 
684 	return (error);
685 }
686 
687 /*
688  *  Lookup interrupt controller locked.
689  */
690 static inline struct intr_pic *
691 pic_lookup_locked(device_t dev, intptr_t xref, int flags)
692 {
693 	struct intr_pic *pic;
694 
695 	mtx_assert(&pic_list_lock, MA_OWNED);
696 
697 	if (dev == NULL && xref == 0)
698 		return (NULL);
699 
700 	/* Note that pic->pic_dev is never NULL on registered PIC. */
701 	SLIST_FOREACH(pic, &pic_list, pic_next) {
702 		if ((pic->pic_flags & FLAG_TYPE_MASK) !=
703 		    (flags & FLAG_TYPE_MASK))
704 			continue;
705 
706 		if (dev == NULL) {
707 			if (xref == pic->pic_xref)
708 				return (pic);
709 		} else if (xref == 0 || pic->pic_xref == 0) {
710 			if (dev == pic->pic_dev)
711 				return (pic);
712 		} else if (xref == pic->pic_xref && dev == pic->pic_dev)
713 				return (pic);
714 	}
715 	return (NULL);
716 }
717 
718 /*
719  *  Lookup interrupt controller.
720  */
721 static struct intr_pic *
722 pic_lookup(device_t dev, intptr_t xref, int flags)
723 {
724 	struct intr_pic *pic;
725 
726 	mtx_lock(&pic_list_lock);
727 	pic = pic_lookup_locked(dev, xref, flags);
728 	mtx_unlock(&pic_list_lock);
729 	return (pic);
730 }
731 
732 /*
733  *  Create interrupt controller.
734  */
735 static struct intr_pic *
736 pic_create(device_t dev, intptr_t xref, int flags)
737 {
738 	struct intr_pic *pic;
739 
740 	mtx_lock(&pic_list_lock);
741 	pic = pic_lookup_locked(dev, xref, flags);
742 	if (pic != NULL) {
743 		mtx_unlock(&pic_list_lock);
744 		return (pic);
745 	}
746 	pic = malloc(sizeof(*pic), M_INTRNG, M_NOWAIT | M_ZERO);
747 	if (pic == NULL) {
748 		mtx_unlock(&pic_list_lock);
749 		return (NULL);
750 	}
751 	pic->pic_xref = xref;
752 	pic->pic_dev = dev;
753 	pic->pic_flags = flags;
754 	mtx_init(&pic->pic_child_lock, "pic child lock", NULL, MTX_SPIN);
755 	SLIST_INSERT_HEAD(&pic_list, pic, pic_next);
756 	mtx_unlock(&pic_list_lock);
757 
758 	return (pic);
759 }
760 #ifdef notyet
761 /*
762  *  Destroy interrupt controller.
763  */
764 static void
765 pic_destroy(device_t dev, intptr_t xref, int flags)
766 {
767 	struct intr_pic *pic;
768 
769 	mtx_lock(&pic_list_lock);
770 	pic = pic_lookup_locked(dev, xref, flags);
771 	if (pic == NULL) {
772 		mtx_unlock(&pic_list_lock);
773 		return;
774 	}
775 	SLIST_REMOVE(&pic_list, pic, intr_pic, pic_next);
776 	mtx_unlock(&pic_list_lock);
777 
778 	free(pic, M_INTRNG);
779 }
780 #endif
781 /*
782  *  Register interrupt controller.
783  */
784 struct intr_pic *
785 intr_pic_register(device_t dev, intptr_t xref)
786 {
787 	struct intr_pic *pic;
788 
789 	if (dev == NULL)
790 		return (NULL);
791 	pic = pic_create(dev, xref, FLAG_PIC);
792 	if (pic == NULL)
793 		return (NULL);
794 
795 	debugf("PIC %p registered for %s <dev %p, xref %x>\n", pic,
796 	    device_get_nameunit(dev), dev, xref);
797 	return (pic);
798 }
799 
800 /*
801  *  Unregister interrupt controller.
802  */
803 int
804 intr_pic_deregister(device_t dev, intptr_t xref)
805 {
806 
807 	panic("%s: not implemented", __func__);
808 }
809 
810 /*
811  *  Mark interrupt controller (itself) as a root one.
812  *
813  *  Note that only an interrupt controller can really know its position
814  *  in interrupt controller's tree. So root PIC must claim itself as a root.
815  *
816  *  In FDT case, according to ePAPR approved version 1.1 from 08 April 2011,
817  *  page 30:
818  *    "The root of the interrupt tree is determined when traversal
819  *     of the interrupt tree reaches an interrupt controller node without
820  *     an interrupts property and thus no explicit interrupt parent."
821  */
822 int
823 intr_pic_claim_root(device_t dev, intptr_t xref, intr_irq_filter_t *filter,
824     void *arg, u_int ipicount)
825 {
826 	struct intr_pic *pic;
827 
828 	pic = pic_lookup(dev, xref, FLAG_PIC);
829 	if (pic == NULL) {
830 		device_printf(dev, "not registered\n");
831 		return (EINVAL);
832 	}
833 
834 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
835 	    ("%s: Found a non-PIC controller: %s", __func__,
836 	     device_get_name(pic->pic_dev)));
837 
838 	if (filter == NULL) {
839 		device_printf(dev, "filter missing\n");
840 		return (EINVAL);
841 	}
842 
843 	/*
844 	 * Only one interrupt controllers could be on the root for now.
845 	 * Note that we further suppose that there is not threaded interrupt
846 	 * routine (handler) on the root. See intr_irq_handler().
847 	 */
848 	if (intr_irq_root_dev != NULL) {
849 		device_printf(dev, "another root already set\n");
850 		return (EBUSY);
851 	}
852 
853 	intr_irq_root_dev = dev;
854 	irq_root_filter = filter;
855 	irq_root_arg = arg;
856 	irq_root_ipicount = ipicount;
857 
858 	debugf("irq root set to %s\n", device_get_nameunit(dev));
859 	return (0);
860 }
861 
862 /*
863  * Add a handler to manage a sub range of a parents interrupts.
864  */
865 struct intr_pic *
866 intr_pic_add_handler(device_t parent, struct intr_pic *pic,
867     intr_child_irq_filter_t *filter, void *arg, uintptr_t start,
868     uintptr_t length)
869 {
870 	struct intr_pic *parent_pic;
871 	struct intr_pic_child *newchild;
872 #ifdef INVARIANTS
873 	struct intr_pic_child *child;
874 #endif
875 
876 	/* Find the parent PIC */
877 	parent_pic = pic_lookup(parent, 0, FLAG_PIC);
878 	if (parent_pic == NULL)
879 		return (NULL);
880 
881 	newchild = malloc(sizeof(*newchild), M_INTRNG, M_WAITOK | M_ZERO);
882 	newchild->pc_pic = pic;
883 	newchild->pc_filter = filter;
884 	newchild->pc_filter_arg = arg;
885 	newchild->pc_start = start;
886 	newchild->pc_length = length;
887 
888 	mtx_lock_spin(&parent_pic->pic_child_lock);
889 #ifdef INVARIANTS
890 	SLIST_FOREACH(child, &parent_pic->pic_children, pc_next) {
891 		KASSERT(child->pc_pic != pic, ("%s: Adding a child PIC twice",
892 		    __func__));
893 	}
894 #endif
895 	SLIST_INSERT_HEAD(&parent_pic->pic_children, newchild, pc_next);
896 	mtx_unlock_spin(&parent_pic->pic_child_lock);
897 
898 	return (pic);
899 }
900 
901 static int
902 intr_resolve_irq(device_t dev, intptr_t xref, struct intr_map_data *data,
903     struct intr_irqsrc **isrc)
904 {
905 	struct intr_pic *pic;
906 	struct intr_map_data_msi *msi;
907 
908 	if (data == NULL)
909 		return (EINVAL);
910 
911 	pic = pic_lookup(dev, xref,
912 	    (data->type == INTR_MAP_DATA_MSI) ? FLAG_MSI : FLAG_PIC);
913 	if (pic == NULL)
914 		return (ESRCH);
915 
916 	switch (data->type) {
917 	case INTR_MAP_DATA_MSI:
918 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
919 		    ("%s: Found a non-MSI controller: %s", __func__,
920 		     device_get_name(pic->pic_dev)));
921 		msi = (struct intr_map_data_msi *)data;
922 		*isrc = msi->isrc;
923 		return (0);
924 
925 	default:
926 		KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_PIC,
927 		    ("%s: Found a non-PIC controller: %s", __func__,
928 		     device_get_name(pic->pic_dev)));
929 		return (PIC_MAP_INTR(pic->pic_dev, data, isrc));
930 
931 	}
932 }
933 
934 bool
935 intr_is_per_cpu(struct resource *res)
936 {
937 	u_int res_id;
938 	struct intr_irqsrc *isrc;
939 
940 	res_id = (u_int)rman_get_start(res);
941 	isrc = intr_map_get_isrc(res_id);
942 
943 	if (isrc == NULL)
944 		panic("Attempt to get isrc for non-active resource id: %u\n",
945 		    res_id);
946 	return ((isrc->isrc_flags & INTR_ISRCF_PPI) != 0);
947 }
948 
949 int
950 intr_activate_irq(device_t dev, struct resource *res)
951 {
952 	device_t map_dev;
953 	intptr_t map_xref;
954 	struct intr_map_data *data;
955 	struct intr_irqsrc *isrc;
956 	u_int res_id;
957 	int error;
958 
959 	KASSERT(rman_get_start(res) == rman_get_end(res),
960 	    ("%s: more interrupts in resource", __func__));
961 
962 	res_id = (u_int)rman_get_start(res);
963 	if (intr_map_get_isrc(res_id) != NULL)
964 		panic("Attempt to double activation of resource id: %u\n",
965 		    res_id);
966 	intr_map_copy_map_data(res_id, &map_dev, &map_xref, &data);
967 	error = intr_resolve_irq(map_dev, map_xref, data, &isrc);
968 	if (error != 0) {
969 		free(data, M_INTRNG);
970 		/* XXX TODO DISCONECTED PICs */
971 		/* if (error == EINVAL) return(0); */
972 		return (error);
973 	}
974 	intr_map_set_isrc(res_id, isrc);
975 	rman_set_virtual(res, data);
976 	return (PIC_ACTIVATE_INTR(isrc->isrc_dev, isrc, res, data));
977 }
978 
979 int
980 intr_deactivate_irq(device_t dev, struct resource *res)
981 {
982 	struct intr_map_data *data;
983 	struct intr_irqsrc *isrc;
984 	u_int res_id;
985 	int error;
986 
987 	KASSERT(rman_get_start(res) == rman_get_end(res),
988 	    ("%s: more interrupts in resource", __func__));
989 
990 	res_id = (u_int)rman_get_start(res);
991 	isrc = intr_map_get_isrc(res_id);
992 	if (isrc == NULL)
993 		panic("Attempt to deactivate non-active resource id: %u\n",
994 		    res_id);
995 
996 	data = rman_get_virtual(res);
997 	error = PIC_DEACTIVATE_INTR(isrc->isrc_dev, isrc, res, data);
998 	intr_map_set_isrc(res_id, NULL);
999 	rman_set_virtual(res, NULL);
1000 	free(data, M_INTRNG);
1001 	return (error);
1002 }
1003 
1004 int
1005 intr_setup_irq(device_t dev, struct resource *res, driver_filter_t filt,
1006     driver_intr_t hand, void *arg, int flags, void **cookiep)
1007 {
1008 	int error;
1009 	struct intr_map_data *data;
1010 	struct intr_irqsrc *isrc;
1011 	const char *name;
1012 	u_int res_id;
1013 
1014 	KASSERT(rman_get_start(res) == rman_get_end(res),
1015 	    ("%s: more interrupts in resource", __func__));
1016 
1017 	res_id = (u_int)rman_get_start(res);
1018 	isrc = intr_map_get_isrc(res_id);
1019 	if (isrc == NULL) {
1020 		/* XXX TODO DISCONECTED PICs */
1021 		return (EINVAL);
1022 	}
1023 
1024 	data = rman_get_virtual(res);
1025 	name = device_get_nameunit(dev);
1026 
1027 #ifdef INTR_SOLO
1028 	/*
1029 	 * Standard handling is done through MI interrupt framework. However,
1030 	 * some interrupts could request solely own special handling. This
1031 	 * non standard handling can be used for interrupt controllers without
1032 	 * handler (filter only), so in case that interrupt controllers are
1033 	 * chained, MI interrupt framework is called only in leaf controller.
1034 	 *
1035 	 * Note that root interrupt controller routine is served as well,
1036 	 * however in intr_irq_handler(), i.e. main system dispatch routine.
1037 	 */
1038 	if (flags & INTR_SOLO && hand != NULL) {
1039 		debugf("irq %u cannot solo on %s\n", irq, name);
1040 		return (EINVAL);
1041 	}
1042 
1043 	if (flags & INTR_SOLO) {
1044 		error = iscr_setup_filter(isrc, name, (intr_irq_filter_t *)filt,
1045 		    arg, cookiep);
1046 		debugf("irq %u setup filter error %d on %s\n", isrc->isrc_irq, error,
1047 		    name);
1048 	} else
1049 #endif
1050 		{
1051 		error = isrc_add_handler(isrc, name, filt, hand, arg, flags,
1052 		    cookiep);
1053 		debugf("irq %u add handler error %d on %s\n", isrc->isrc_irq, error, name);
1054 	}
1055 	if (error != 0)
1056 		return (error);
1057 
1058 	mtx_lock(&isrc_table_lock);
1059 	error = PIC_SETUP_INTR(isrc->isrc_dev, isrc, res, data);
1060 	if (error == 0) {
1061 		isrc->isrc_handlers++;
1062 		if (isrc->isrc_handlers == 1)
1063 			PIC_ENABLE_INTR(isrc->isrc_dev, isrc);
1064 	}
1065 	mtx_unlock(&isrc_table_lock);
1066 	if (error != 0)
1067 		intr_event_remove_handler(*cookiep);
1068 	return (error);
1069 }
1070 
1071 int
1072 intr_teardown_irq(device_t dev, struct resource *res, void *cookie)
1073 {
1074 	int error;
1075 	struct intr_map_data *data;
1076 	struct intr_irqsrc *isrc;
1077 	u_int res_id;
1078 
1079 	KASSERT(rman_get_start(res) == rman_get_end(res),
1080 	    ("%s: more interrupts in resource", __func__));
1081 
1082 	res_id = (u_int)rman_get_start(res);
1083 	isrc = intr_map_get_isrc(res_id);
1084 	if (isrc == NULL || isrc->isrc_handlers == 0)
1085 		return (EINVAL);
1086 
1087 	data = rman_get_virtual(res);
1088 
1089 #ifdef INTR_SOLO
1090 	if (isrc->isrc_filter != NULL) {
1091 		if (isrc != cookie)
1092 			return (EINVAL);
1093 
1094 		mtx_lock(&isrc_table_lock);
1095 		isrc->isrc_filter = NULL;
1096 		isrc->isrc_arg = NULL;
1097 		isrc->isrc_handlers = 0;
1098 		PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1099 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1100 		isrc_update_name(isrc, NULL);
1101 		mtx_unlock(&isrc_table_lock);
1102 		return (0);
1103 	}
1104 #endif
1105 	if (isrc != intr_handler_source(cookie))
1106 		return (EINVAL);
1107 
1108 	error = intr_event_remove_handler(cookie);
1109 	if (error == 0) {
1110 		mtx_lock(&isrc_table_lock);
1111 		isrc->isrc_handlers--;
1112 		if (isrc->isrc_handlers == 0)
1113 			PIC_DISABLE_INTR(isrc->isrc_dev, isrc);
1114 		PIC_TEARDOWN_INTR(isrc->isrc_dev, isrc, res, data);
1115 		intrcnt_updatename(isrc);
1116 		mtx_unlock(&isrc_table_lock);
1117 	}
1118 	return (error);
1119 }
1120 
1121 int
1122 intr_describe_irq(device_t dev, struct resource *res, void *cookie,
1123     const char *descr)
1124 {
1125 	int error;
1126 	struct intr_irqsrc *isrc;
1127 	u_int res_id;
1128 
1129 	KASSERT(rman_get_start(res) == rman_get_end(res),
1130 	    ("%s: more interrupts in resource", __func__));
1131 
1132 	res_id = (u_int)rman_get_start(res);
1133 	isrc = intr_map_get_isrc(res_id);
1134 	if (isrc == NULL || isrc->isrc_handlers == 0)
1135 		return (EINVAL);
1136 #ifdef INTR_SOLO
1137 	if (isrc->isrc_filter != NULL) {
1138 		if (isrc != cookie)
1139 			return (EINVAL);
1140 
1141 		mtx_lock(&isrc_table_lock);
1142 		isrc_update_name(isrc, descr);
1143 		mtx_unlock(&isrc_table_lock);
1144 		return (0);
1145 	}
1146 #endif
1147 	error = intr_event_describe_handler(isrc->isrc_event, cookie, descr);
1148 	if (error == 0) {
1149 		mtx_lock(&isrc_table_lock);
1150 		intrcnt_updatename(isrc);
1151 		mtx_unlock(&isrc_table_lock);
1152 	}
1153 	return (error);
1154 }
1155 
1156 #ifdef SMP
1157 int
1158 intr_bind_irq(device_t dev, struct resource *res, int cpu)
1159 {
1160 	struct intr_irqsrc *isrc;
1161 	u_int res_id;
1162 
1163 	KASSERT(rman_get_start(res) == rman_get_end(res),
1164 	    ("%s: more interrupts in resource", __func__));
1165 
1166 	res_id = (u_int)rman_get_start(res);
1167 	isrc = intr_map_get_isrc(res_id);
1168 	if (isrc == NULL || isrc->isrc_handlers == 0)
1169 		return (EINVAL);
1170 #ifdef INTR_SOLO
1171 	if (isrc->isrc_filter != NULL)
1172 		return (intr_isrc_assign_cpu(isrc, cpu));
1173 #endif
1174 	return (intr_event_bind(isrc->isrc_event, cpu));
1175 }
1176 
1177 /*
1178  * Return the CPU that the next interrupt source should use.
1179  * For now just returns the next CPU according to round-robin.
1180  */
1181 u_int
1182 intr_irq_next_cpu(u_int last_cpu, cpuset_t *cpumask)
1183 {
1184 	u_int cpu;
1185 
1186 	KASSERT(!CPU_EMPTY(cpumask), ("%s: Empty CPU mask", __func__));
1187 	if (!irq_assign_cpu || mp_ncpus == 1) {
1188 		cpu = PCPU_GET(cpuid);
1189 
1190 		if (CPU_ISSET(cpu, cpumask))
1191 			return (curcpu);
1192 
1193 		return (CPU_FFS(cpumask) - 1);
1194 	}
1195 
1196 	do {
1197 		last_cpu++;
1198 		if (last_cpu > mp_maxid)
1199 			last_cpu = 0;
1200 	} while (!CPU_ISSET(last_cpu, cpumask));
1201 	return (last_cpu);
1202 }
1203 
1204 /*
1205  *  Distribute all the interrupt sources among the available
1206  *  CPUs once the AP's have been launched.
1207  */
1208 static void
1209 intr_irq_shuffle(void *arg __unused)
1210 {
1211 	struct intr_irqsrc *isrc;
1212 	u_int i;
1213 
1214 	if (mp_ncpus == 1)
1215 		return;
1216 
1217 	mtx_lock(&isrc_table_lock);
1218 	irq_assign_cpu = TRUE;
1219 	for (i = 0; i < NIRQ; i++) {
1220 		isrc = irq_sources[i];
1221 		if (isrc == NULL || isrc->isrc_handlers == 0 ||
1222 		    isrc->isrc_flags & (INTR_ISRCF_PPI | INTR_ISRCF_IPI))
1223 			continue;
1224 
1225 		if (isrc->isrc_event != NULL &&
1226 		    isrc->isrc_flags & INTR_ISRCF_BOUND &&
1227 		    isrc->isrc_event->ie_cpu != CPU_FFS(&isrc->isrc_cpu) - 1)
1228 			panic("%s: CPU inconsistency", __func__);
1229 
1230 		if ((isrc->isrc_flags & INTR_ISRCF_BOUND) == 0)
1231 			CPU_ZERO(&isrc->isrc_cpu); /* start again */
1232 
1233 		/*
1234 		 * We are in wicked position here if the following call fails
1235 		 * for bound ISRC. The best thing we can do is to clear
1236 		 * isrc_cpu so inconsistency with ie_cpu will be detectable.
1237 		 */
1238 		if (PIC_BIND_INTR(isrc->isrc_dev, isrc) != 0)
1239 			CPU_ZERO(&isrc->isrc_cpu);
1240 	}
1241 	mtx_unlock(&isrc_table_lock);
1242 }
1243 SYSINIT(intr_irq_shuffle, SI_SUB_SMP, SI_ORDER_SECOND, intr_irq_shuffle, NULL);
1244 
1245 #else
1246 u_int
1247 intr_irq_next_cpu(u_int current_cpu, cpuset_t *cpumask)
1248 {
1249 
1250 	return (PCPU_GET(cpuid));
1251 }
1252 #endif
1253 
1254 /*
1255  * Allocate memory for new intr_map_data structure.
1256  * Initialize common fields.
1257  */
1258 struct intr_map_data *
1259 intr_alloc_map_data(enum intr_map_data_type type, size_t len, int flags)
1260 {
1261 	struct intr_map_data *data;
1262 
1263 	data = malloc(len, M_INTRNG, flags);
1264 	data->type = type;
1265 	data->len = len;
1266 	return (data);
1267 }
1268 
1269 void intr_free_intr_map_data(struct intr_map_data *data)
1270 {
1271 
1272 	free(data, M_INTRNG);
1273 }
1274 
1275 
1276 /*
1277  *  Register a MSI/MSI-X interrupt controller
1278  */
1279 int
1280 intr_msi_register(device_t dev, intptr_t xref)
1281 {
1282 	struct intr_pic *pic;
1283 
1284 	if (dev == NULL)
1285 		return (EINVAL);
1286 	pic = pic_create(dev, xref, FLAG_MSI);
1287 	if (pic == NULL)
1288 		return (ENOMEM);
1289 
1290 	debugf("PIC %p registered for %s <dev %p, xref %jx>\n", pic,
1291 	    device_get_nameunit(dev), dev, (uintmax_t)xref);
1292 	return (0);
1293 }
1294 
1295 int
1296 intr_alloc_msi(device_t pci, device_t child, intptr_t xref, int count,
1297     int maxcount, int *irqs)
1298 {
1299 	struct intr_irqsrc **isrc;
1300 	struct intr_pic *pic;
1301 	device_t pdev;
1302 	struct intr_map_data_msi *msi;
1303 	int err, i;
1304 
1305 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1306 	if (pic == NULL)
1307 		return (ESRCH);
1308 
1309 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1310 	    ("%s: Found a non-MSI controller: %s", __func__,
1311 	     device_get_name(pic->pic_dev)));
1312 
1313 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1314 	err = MSI_ALLOC_MSI(pic->pic_dev, child, count, maxcount, &pdev, isrc);
1315 	if (err != 0) {
1316 		free(isrc, M_INTRNG);
1317 		return (err);
1318 	}
1319 
1320 	for (i = 0; i < count; i++) {
1321 		msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1322 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1323 		msi-> isrc = isrc[i];
1324 		irqs[i] = intr_map_irq(pic->pic_dev, xref,
1325 		    (struct intr_map_data *)msi);
1326 
1327 	}
1328 	free(isrc, M_INTRNG);
1329 
1330 	return (err);
1331 }
1332 
1333 int
1334 intr_release_msi(device_t pci, device_t child, intptr_t xref, int count,
1335     int *irqs)
1336 {
1337 	struct intr_irqsrc **isrc;
1338 	struct intr_pic *pic;
1339 	struct intr_map_data_msi *msi;
1340 	int i, err;
1341 
1342 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1343 	if (pic == NULL)
1344 		return (ESRCH);
1345 
1346 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1347 	    ("%s: Found a non-MSI controller: %s", __func__,
1348 	     device_get_name(pic->pic_dev)));
1349 
1350 	isrc = malloc(sizeof(*isrc) * count, M_INTRNG, M_WAITOK);
1351 
1352 	for (i = 0; i < count; i++) {
1353 		msi = (struct intr_map_data_msi *)
1354 		    intr_map_get_map_data(irqs[i]);
1355 		KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1356 		    ("%s: irq %d map data is not MSI", __func__,
1357 		    irqs[i]));
1358 		isrc[i] = msi->isrc;
1359 	}
1360 
1361 	err = MSI_RELEASE_MSI(pic->pic_dev, child, count, isrc);
1362 
1363 	for (i = 0; i < count; i++) {
1364 		if (isrc[i] != NULL)
1365 			intr_unmap_irq(irqs[i]);
1366 	}
1367 
1368 	free(isrc, M_INTRNG);
1369 	return (err);
1370 }
1371 
1372 int
1373 intr_alloc_msix(device_t pci, device_t child, intptr_t xref, int *irq)
1374 {
1375 	struct intr_irqsrc *isrc;
1376 	struct intr_pic *pic;
1377 	device_t pdev;
1378 	struct intr_map_data_msi *msi;
1379 	int err;
1380 
1381 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1382 	if (pic == NULL)
1383 		return (ESRCH);
1384 
1385 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1386 	    ("%s: Found a non-MSI controller: %s", __func__,
1387 	     device_get_name(pic->pic_dev)));
1388 
1389 
1390 	err = MSI_ALLOC_MSIX(pic->pic_dev, child, &pdev, &isrc);
1391 	if (err != 0)
1392 		return (err);
1393 
1394 	msi = (struct intr_map_data_msi *)intr_alloc_map_data(
1395 		    INTR_MAP_DATA_MSI, sizeof(*msi), M_WAITOK | M_ZERO);
1396 	msi->isrc = isrc;
1397 	*irq = intr_map_irq(pic->pic_dev, xref, (struct intr_map_data *)msi);
1398 	return (0);
1399 }
1400 
1401 int
1402 intr_release_msix(device_t pci, device_t child, intptr_t xref, int irq)
1403 {
1404 	struct intr_irqsrc *isrc;
1405 	struct intr_pic *pic;
1406 	struct intr_map_data_msi *msi;
1407 	int err;
1408 
1409 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1410 	if (pic == NULL)
1411 		return (ESRCH);
1412 
1413 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1414 	    ("%s: Found a non-MSI controller: %s", __func__,
1415 	     device_get_name(pic->pic_dev)));
1416 
1417 	msi = (struct intr_map_data_msi *)
1418 	    intr_map_get_map_data(irq);
1419 	KASSERT(msi->hdr.type == INTR_MAP_DATA_MSI,
1420 	    ("%s: irq %d map data is not MSI", __func__,
1421 	    irq));
1422 	isrc = msi->isrc;
1423 	if (isrc == NULL) {
1424 		intr_unmap_irq(irq);
1425 		return (EINVAL);
1426 	}
1427 
1428 	err = MSI_RELEASE_MSIX(pic->pic_dev, child, isrc);
1429 	intr_unmap_irq(irq);
1430 
1431 	return (err);
1432 }
1433 
1434 int
1435 intr_map_msi(device_t pci, device_t child, intptr_t xref, int irq,
1436     uint64_t *addr, uint32_t *data)
1437 {
1438 	struct intr_irqsrc *isrc;
1439 	struct intr_pic *pic;
1440 	int err;
1441 
1442 	pic = pic_lookup(NULL, xref, FLAG_MSI);
1443 	if (pic == NULL)
1444 		return (ESRCH);
1445 
1446 	KASSERT((pic->pic_flags & FLAG_TYPE_MASK) == FLAG_MSI,
1447 	    ("%s: Found a non-MSI controller: %s", __func__,
1448 	     device_get_name(pic->pic_dev)));
1449 
1450 	isrc = intr_map_get_isrc(irq);
1451 	if (isrc == NULL)
1452 		return (EINVAL);
1453 
1454 	err = MSI_MAP_MSI(pic->pic_dev, child, isrc, addr, data);
1455 	return (err);
1456 }
1457 
1458 
1459 void dosoftints(void);
1460 void
1461 dosoftints(void)
1462 {
1463 }
1464 
1465 #ifdef SMP
1466 /*
1467  *  Init interrupt controller on another CPU.
1468  */
1469 void
1470 intr_pic_init_secondary(void)
1471 {
1472 
1473 	/*
1474 	 * QQQ: Only root PIC is aware of other CPUs ???
1475 	 */
1476 	KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__));
1477 
1478 	//mtx_lock(&isrc_table_lock);
1479 	PIC_INIT_SECONDARY(intr_irq_root_dev);
1480 	//mtx_unlock(&isrc_table_lock);
1481 }
1482 #endif
1483 
1484 #ifdef DDB
1485 DB_SHOW_COMMAND(irqs, db_show_irqs)
1486 {
1487 	u_int i, irqsum;
1488 	u_long num;
1489 	struct intr_irqsrc *isrc;
1490 
1491 	for (irqsum = 0, i = 0; i < NIRQ; i++) {
1492 		isrc = irq_sources[i];
1493 		if (isrc == NULL)
1494 			continue;
1495 
1496 		num = isrc->isrc_count != NULL ? isrc->isrc_count[0] : 0;
1497 		db_printf("irq%-3u <%s>: cpu %02lx%s cnt %lu\n", i,
1498 		    isrc->isrc_name, isrc->isrc_cpu.__bits[0],
1499 		    isrc->isrc_flags & INTR_ISRCF_BOUND ? " (bound)" : "", num);
1500 		irqsum += num;
1501 	}
1502 	db_printf("irq total %u\n", irqsum);
1503 }
1504 #endif
1505 
1506 /*
1507  * Interrupt mapping table functions.
1508  *
1509  * Please, keep this part separately, it can be transformed to
1510  * extension of standard resources.
1511  */
1512 struct intr_map_entry
1513 {
1514 	device_t 		dev;
1515 	intptr_t 		xref;
1516 	struct intr_map_data 	*map_data;
1517 	struct intr_irqsrc 	*isrc;
1518 	/* XXX TODO DISCONECTED PICs */
1519 	/*int			flags */
1520 };
1521 
1522 /* XXX Convert irq_map[] to dynamicaly expandable one. */
1523 static struct intr_map_entry *irq_map[2 * NIRQ];
1524 static int irq_map_count = nitems(irq_map);
1525 static int irq_map_first_free_idx;
1526 static struct mtx irq_map_lock;
1527 
1528 static struct intr_irqsrc *
1529 intr_map_get_isrc(u_int res_id)
1530 {
1531 	struct intr_irqsrc *isrc;
1532 
1533 	mtx_lock(&irq_map_lock);
1534 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1535 		mtx_unlock(&irq_map_lock);
1536 		return (NULL);
1537 	}
1538 	isrc = irq_map[res_id]->isrc;
1539 	mtx_unlock(&irq_map_lock);
1540 	return (isrc);
1541 }
1542 
1543 static void
1544 intr_map_set_isrc(u_int res_id, struct intr_irqsrc *isrc)
1545 {
1546 
1547 	mtx_lock(&irq_map_lock);
1548 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL)) {
1549 		mtx_unlock(&irq_map_lock);
1550 		return;
1551 	}
1552 	irq_map[res_id]->isrc = isrc;
1553 	mtx_unlock(&irq_map_lock);
1554 }
1555 
1556 /*
1557  * Get a copy of intr_map_entry data
1558  */
1559 static struct intr_map_data *
1560 intr_map_get_map_data(u_int res_id)
1561 {
1562 	struct intr_map_data *data;
1563 
1564 	data = NULL;
1565 	mtx_lock(&irq_map_lock);
1566 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1567 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1568 	data = irq_map[res_id]->map_data;
1569 	mtx_unlock(&irq_map_lock);
1570 
1571 	return (data);
1572 }
1573 
1574 /*
1575  * Get a copy of intr_map_entry data
1576  */
1577 static void
1578 intr_map_copy_map_data(u_int res_id, device_t *map_dev, intptr_t *map_xref,
1579     struct intr_map_data **data)
1580 {
1581 	size_t len;
1582 
1583 	len = 0;
1584 	mtx_lock(&irq_map_lock);
1585 	if (res_id >= irq_map_count || irq_map[res_id] == NULL)
1586 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1587 	if (irq_map[res_id]->map_data != NULL)
1588 		len = irq_map[res_id]->map_data->len;
1589 	mtx_unlock(&irq_map_lock);
1590 
1591 	if (len == 0)
1592 		*data = NULL;
1593 	else
1594 		*data = malloc(len, M_INTRNG, M_WAITOK | M_ZERO);
1595 	mtx_lock(&irq_map_lock);
1596 	if (irq_map[res_id] == NULL)
1597 		panic("Attempt to copy invalid resource id: %u\n", res_id);
1598 	if (len != 0) {
1599 		if (len != irq_map[res_id]->map_data->len)
1600 			panic("Resource id: %u has changed.\n", res_id);
1601 		memcpy(*data, irq_map[res_id]->map_data, len);
1602 	}
1603 	*map_dev = irq_map[res_id]->dev;
1604 	*map_xref = irq_map[res_id]->xref;
1605 	mtx_unlock(&irq_map_lock);
1606 }
1607 
1608 
1609 /*
1610  * Allocate and fill new entry in irq_map table.
1611  */
1612 u_int
1613 intr_map_irq(device_t dev, intptr_t xref, struct intr_map_data *data)
1614 {
1615 	u_int i;
1616 	struct intr_map_entry *entry;
1617 
1618 	/* Prepare new entry first. */
1619 	entry = malloc(sizeof(*entry), M_INTRNG, M_WAITOK | M_ZERO);
1620 
1621 	entry->dev = dev;
1622 	entry->xref = xref;
1623 	entry->map_data = data;
1624 	entry->isrc = NULL;
1625 
1626 	mtx_lock(&irq_map_lock);
1627 	for (i = irq_map_first_free_idx; i < irq_map_count; i++) {
1628 		if (irq_map[i] == NULL) {
1629 			irq_map[i] = entry;
1630 			irq_map_first_free_idx = i + 1;
1631 			mtx_unlock(&irq_map_lock);
1632 			return (i);
1633 		}
1634 	}
1635 	mtx_unlock(&irq_map_lock);
1636 
1637 	/* XXX Expand irq_map table */
1638 	panic("IRQ mapping table is full.");
1639 }
1640 
1641 /*
1642  * Remove and free mapping entry.
1643  */
1644 void
1645 intr_unmap_irq(u_int res_id)
1646 {
1647 	struct intr_map_entry *entry;
1648 
1649 	mtx_lock(&irq_map_lock);
1650 	if ((res_id >= irq_map_count) || (irq_map[res_id] == NULL))
1651 		panic("Attempt to unmap invalid resource id: %u\n", res_id);
1652 	entry = irq_map[res_id];
1653 	irq_map[res_id] = NULL;
1654 	irq_map_first_free_idx = res_id;
1655 	mtx_unlock(&irq_map_lock);
1656 	intr_free_intr_map_data(entry->map_data);
1657 	free(entry, M_INTRNG);
1658 }
1659 
1660 /*
1661  * Clone mapping entry.
1662  */
1663 u_int
1664 intr_map_clone_irq(u_int old_res_id)
1665 {
1666 	device_t map_dev;
1667 	intptr_t map_xref;
1668 	struct intr_map_data *data;
1669 
1670 	intr_map_copy_map_data(old_res_id, &map_dev, &map_xref, &data);
1671 	return (intr_map_irq(map_dev, map_xref, data));
1672 }
1673 
1674 static void
1675 intr_map_init(void *dummy __unused)
1676 {
1677 
1678 	mtx_init(&irq_map_lock, "intr map table", NULL, MTX_DEF);
1679 }
1680 SYSINIT(intr_map_init, SI_SUB_INTR, SI_ORDER_FIRST, intr_map_init, NULL);
1681