xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/errno.h>
35 
36 #include <ctype.h>
37 #include <pthread.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <strings.h>
42 #include <assert.h>
43 #include <stdbool.h>
44 
45 #include <machine/vmm.h>
46 #include <vmmapi.h>
47 
48 #include "acpi.h"
49 #include "bhyverun.h"
50 #include "inout.h"
51 #include "ioapic.h"
52 #include "mem.h"
53 #include "pci_emul.h"
54 #include "pci_irq.h"
55 #include "pci_lpc.h"
56 
57 #define CONF1_ADDR_PORT    0x0cf8
58 #define CONF1_DATA_PORT    0x0cfc
59 
60 #define CONF1_ENABLE	   0x80000000ul
61 
62 #define	CFGWRITE(pi,off,val,b)						\
63 do {									\
64 	if ((b) == 1) {							\
65 		pci_set_cfgdata8((pi),(off),(val));			\
66 	} else if ((b) == 2) {						\
67 		pci_set_cfgdata16((pi),(off),(val));			\
68 	} else {							\
69 		pci_set_cfgdata32((pi),(off),(val));			\
70 	}								\
71 } while (0)
72 
73 #define	MAXBUSES	(PCI_BUSMAX + 1)
74 #define MAXSLOTS	(PCI_SLOTMAX + 1)
75 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
76 
77 struct funcinfo {
78 	char	*fi_name;
79 	char	*fi_param;
80 	struct pci_devinst *fi_devi;
81 };
82 
83 struct intxinfo {
84 	int	ii_count;
85 	int	ii_pirq_pin;
86 	int	ii_ioapic_irq;
87 };
88 
89 struct slotinfo {
90 	struct intxinfo si_intpins[4];
91 	struct funcinfo si_funcs[MAXFUNCS];
92 };
93 
94 struct businfo {
95 	uint16_t iobase, iolimit;		/* I/O window */
96 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
97 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
98 	struct slotinfo slotinfo[MAXSLOTS];
99 };
100 
101 static struct businfo *pci_businfo[MAXBUSES];
102 
103 SET_DECLARE(pci_devemu_set, struct pci_devemu);
104 
105 static uint64_t pci_emul_iobase;
106 static uint64_t pci_emul_membase32;
107 static uint64_t pci_emul_membase64;
108 
109 #define	PCI_EMUL_IOBASE		0x2000
110 #define	PCI_EMUL_IOLIMIT	0x10000
111 
112 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
113 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
114 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
115 
116 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
117 
118 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
119 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
120 
121 static struct pci_devemu *pci_emul_finddev(char *name);
122 static void pci_lintr_route(struct pci_devinst *pi);
123 static void pci_lintr_update(struct pci_devinst *pi);
124 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
125     int func, int coff, int bytes, uint32_t *val);
126 
127 /*
128  * I/O access
129  */
130 
131 /*
132  * Slot options are in the form:
133  *
134  *  <bus>:<slot>:<func>,<emul>[,<config>]
135  *  <slot>[:<func>],<emul>[,<config>]
136  *
137  *  slot is 0..31
138  *  func is 0..7
139  *  emul is a string describing the type of PCI device e.g. virtio-net
140  *  config is an optional string, depending on the device, that can be
141  *  used for configuration.
142  *   Examples are:
143  *     1,virtio-net,tap0
144  *     3:0,dummy
145  */
146 static void
147 pci_parse_slot_usage(char *aopt)
148 {
149 
150 	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
151 }
152 
153 int
154 pci_parse_slot(char *opt)
155 {
156 	struct businfo *bi;
157 	struct slotinfo *si;
158 	char *emul, *config, *str, *cp;
159 	int error, bnum, snum, fnum;
160 
161 	error = -1;
162 	str = strdup(opt);
163 
164 	emul = config = NULL;
165 	if ((cp = strchr(str, ',')) != NULL) {
166 		*cp = '\0';
167 		emul = cp + 1;
168 		if ((cp = strchr(emul, ',')) != NULL) {
169 			*cp = '\0';
170 			config = cp + 1;
171 		}
172 	} else {
173 		pci_parse_slot_usage(opt);
174 		goto done;
175 	}
176 
177 	/* <bus>:<slot>:<func> */
178 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
179 		bnum = 0;
180 		/* <slot>:<func> */
181 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
182 			fnum = 0;
183 			/* <slot> */
184 			if (sscanf(str, "%d", &snum) != 1) {
185 				snum = -1;
186 			}
187 		}
188 	}
189 
190 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
191 	    fnum < 0 || fnum >= MAXFUNCS) {
192 		pci_parse_slot_usage(opt);
193 		goto done;
194 	}
195 
196 	if (pci_businfo[bnum] == NULL)
197 		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
198 
199 	bi = pci_businfo[bnum];
200 	si = &bi->slotinfo[snum];
201 
202 	if (si->si_funcs[fnum].fi_name != NULL) {
203 		fprintf(stderr, "pci slot %d:%d already occupied!\n",
204 			snum, fnum);
205 		goto done;
206 	}
207 
208 	if (pci_emul_finddev(emul) == NULL) {
209 		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
210 			snum, fnum, emul);
211 		goto done;
212 	}
213 
214 	error = 0;
215 	si->si_funcs[fnum].fi_name = emul;
216 	si->si_funcs[fnum].fi_param = config;
217 
218 done:
219 	if (error)
220 		free(str);
221 
222 	return (error);
223 }
224 
225 static int
226 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
227 {
228 
229 	if (offset < pi->pi_msix.pba_offset)
230 		return (0);
231 
232 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
233 		return (0);
234 	}
235 
236 	return (1);
237 }
238 
239 int
240 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
241 		     uint64_t value)
242 {
243 	int msix_entry_offset;
244 	int tab_index;
245 	char *dest;
246 
247 	/* support only 4 or 8 byte writes */
248 	if (size != 4 && size != 8)
249 		return (-1);
250 
251 	/*
252 	 * Return if table index is beyond what device supports
253 	 */
254 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
255 	if (tab_index >= pi->pi_msix.table_count)
256 		return (-1);
257 
258 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
259 
260 	/* support only aligned writes */
261 	if ((msix_entry_offset % size) != 0)
262 		return (-1);
263 
264 	dest = (char *)(pi->pi_msix.table + tab_index);
265 	dest += msix_entry_offset;
266 
267 	if (size == 4)
268 		*((uint32_t *)dest) = value;
269 	else
270 		*((uint64_t *)dest) = value;
271 
272 	return (0);
273 }
274 
275 uint64_t
276 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
277 {
278 	char *dest;
279 	int msix_entry_offset;
280 	int tab_index;
281 	uint64_t retval = ~0;
282 
283 	/*
284 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
285 	 * table but we also allow 1 byte access to accomodate reads from
286 	 * ddb.
287 	 */
288 	if (size != 1 && size != 4 && size != 8)
289 		return (retval);
290 
291 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
292 
293 	/* support only aligned reads */
294 	if ((msix_entry_offset % size) != 0) {
295 		return (retval);
296 	}
297 
298 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
299 
300 	if (tab_index < pi->pi_msix.table_count) {
301 		/* valid MSI-X Table access */
302 		dest = (char *)(pi->pi_msix.table + tab_index);
303 		dest += msix_entry_offset;
304 
305 		if (size == 1)
306 			retval = *((uint8_t *)dest);
307 		else if (size == 4)
308 			retval = *((uint32_t *)dest);
309 		else
310 			retval = *((uint64_t *)dest);
311 	} else if (pci_valid_pba_offset(pi, offset)) {
312 		/* return 0 for PBA access */
313 		retval = 0;
314 	}
315 
316 	return (retval);
317 }
318 
319 int
320 pci_msix_table_bar(struct pci_devinst *pi)
321 {
322 
323 	if (pi->pi_msix.table != NULL)
324 		return (pi->pi_msix.table_bar);
325 	else
326 		return (-1);
327 }
328 
329 int
330 pci_msix_pba_bar(struct pci_devinst *pi)
331 {
332 
333 	if (pi->pi_msix.table != NULL)
334 		return (pi->pi_msix.pba_bar);
335 	else
336 		return (-1);
337 }
338 
339 static int
340 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
341 		    uint32_t *eax, void *arg)
342 {
343 	struct pci_devinst *pdi = arg;
344 	struct pci_devemu *pe = pdi->pi_d;
345 	uint64_t offset;
346 	int i;
347 
348 	for (i = 0; i <= PCI_BARMAX; i++) {
349 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
350 		    port >= pdi->pi_bar[i].addr &&
351 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
352 			offset = port - pdi->pi_bar[i].addr;
353 			if (in)
354 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
355 							 offset, bytes);
356 			else
357 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
358 						   bytes, *eax);
359 			return (0);
360 		}
361 	}
362 	return (-1);
363 }
364 
365 static int
366 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
367 		     int size, uint64_t *val, void *arg1, long arg2)
368 {
369 	struct pci_devinst *pdi = arg1;
370 	struct pci_devemu *pe = pdi->pi_d;
371 	uint64_t offset;
372 	int bidx = (int) arg2;
373 
374 	assert(bidx <= PCI_BARMAX);
375 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
376 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
377 	assert(addr >= pdi->pi_bar[bidx].addr &&
378 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
379 
380 	offset = addr - pdi->pi_bar[bidx].addr;
381 
382 	if (dir == MEM_F_WRITE) {
383 		if (size == 8) {
384 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
385 					   4, *val & 0xffffffff);
386 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
387 					   4, *val >> 32);
388 		} else {
389 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
390 					   size, *val);
391 		}
392 	} else {
393 		if (size == 8) {
394 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
395 						 offset, 4);
396 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
397 						  offset + 4, 4) << 32;
398 		} else {
399 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
400 						 offset, size);
401 		}
402 	}
403 
404 	return (0);
405 }
406 
407 
408 static int
409 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
410 			uint64_t *addr)
411 {
412 	uint64_t base;
413 
414 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
415 
416 	base = roundup2(*baseptr, size);
417 
418 	if (base + size <= limit) {
419 		*addr = base;
420 		*baseptr = base + size;
421 		return (0);
422 	} else
423 		return (-1);
424 }
425 
426 int
427 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
428 		   uint64_t size)
429 {
430 
431 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
432 }
433 
434 /*
435  * Register (or unregister) the MMIO or I/O region associated with the BAR
436  * register 'idx' of an emulated pci device.
437  */
438 static void
439 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
440 {
441 	int error;
442 	struct inout_port iop;
443 	struct mem_range mr;
444 
445 	switch (pi->pi_bar[idx].type) {
446 	case PCIBAR_IO:
447 		bzero(&iop, sizeof(struct inout_port));
448 		iop.name = pi->pi_name;
449 		iop.port = pi->pi_bar[idx].addr;
450 		iop.size = pi->pi_bar[idx].size;
451 		if (registration) {
452 			iop.flags = IOPORT_F_INOUT;
453 			iop.handler = pci_emul_io_handler;
454 			iop.arg = pi;
455 			error = register_inout(&iop);
456 		} else
457 			error = unregister_inout(&iop);
458 		break;
459 	case PCIBAR_MEM32:
460 	case PCIBAR_MEM64:
461 		bzero(&mr, sizeof(struct mem_range));
462 		mr.name = pi->pi_name;
463 		mr.base = pi->pi_bar[idx].addr;
464 		mr.size = pi->pi_bar[idx].size;
465 		if (registration) {
466 			mr.flags = MEM_F_RW;
467 			mr.handler = pci_emul_mem_handler;
468 			mr.arg1 = pi;
469 			mr.arg2 = idx;
470 			error = register_mem(&mr);
471 		} else
472 			error = unregister_mem(&mr);
473 		break;
474 	default:
475 		error = EINVAL;
476 		break;
477 	}
478 	assert(error == 0);
479 }
480 
481 static void
482 unregister_bar(struct pci_devinst *pi, int idx)
483 {
484 
485 	modify_bar_registration(pi, idx, 0);
486 }
487 
488 static void
489 register_bar(struct pci_devinst *pi, int idx)
490 {
491 
492 	modify_bar_registration(pi, idx, 1);
493 }
494 
495 /* Are we decoding i/o port accesses for the emulated pci device? */
496 static int
497 porten(struct pci_devinst *pi)
498 {
499 	uint16_t cmd;
500 
501 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
502 
503 	return (cmd & PCIM_CMD_PORTEN);
504 }
505 
506 /* Are we decoding memory accesses for the emulated pci device? */
507 static int
508 memen(struct pci_devinst *pi)
509 {
510 	uint16_t cmd;
511 
512 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
513 
514 	return (cmd & PCIM_CMD_MEMEN);
515 }
516 
517 /*
518  * Update the MMIO or I/O address that is decoded by the BAR register.
519  *
520  * If the pci device has enabled the address space decoding then intercept
521  * the address range decoded by the BAR register.
522  */
523 static void
524 update_bar_address(struct  pci_devinst *pi, uint64_t addr, int idx, int type)
525 {
526 	int decode;
527 
528 	if (pi->pi_bar[idx].type == PCIBAR_IO)
529 		decode = porten(pi);
530 	else
531 		decode = memen(pi);
532 
533 	if (decode)
534 		unregister_bar(pi, idx);
535 
536 	switch (type) {
537 	case PCIBAR_IO:
538 	case PCIBAR_MEM32:
539 		pi->pi_bar[idx].addr = addr;
540 		break;
541 	case PCIBAR_MEM64:
542 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
543 		pi->pi_bar[idx].addr |= addr;
544 		break;
545 	case PCIBAR_MEMHI64:
546 		pi->pi_bar[idx].addr &= 0xffffffff;
547 		pi->pi_bar[idx].addr |= addr;
548 		break;
549 	default:
550 		assert(0);
551 	}
552 
553 	if (decode)
554 		register_bar(pi, idx);
555 }
556 
557 int
558 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
559 		    enum pcibar_type type, uint64_t size)
560 {
561 	int error;
562 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
563 
564 	assert(idx >= 0 && idx <= PCI_BARMAX);
565 
566 	if ((size & (size - 1)) != 0)
567 		size = 1UL << flsl(size);	/* round up to a power of 2 */
568 
569 	/* Enforce minimum BAR sizes required by the PCI standard */
570 	if (type == PCIBAR_IO) {
571 		if (size < 4)
572 			size = 4;
573 	} else {
574 		if (size < 16)
575 			size = 16;
576 	}
577 
578 	switch (type) {
579 	case PCIBAR_NONE:
580 		baseptr = NULL;
581 		addr = mask = lobits = 0;
582 		break;
583 	case PCIBAR_IO:
584 		baseptr = &pci_emul_iobase;
585 		limit = PCI_EMUL_IOLIMIT;
586 		mask = PCIM_BAR_IO_BASE;
587 		lobits = PCIM_BAR_IO_SPACE;
588 		break;
589 	case PCIBAR_MEM64:
590 		/*
591 		 * XXX
592 		 * Some drivers do not work well if the 64-bit BAR is allocated
593 		 * above 4GB. Allow for this by allocating small requests under
594 		 * 4GB unless then allocation size is larger than some arbitrary
595 		 * number (32MB currently).
596 		 */
597 		if (size > 32 * 1024 * 1024) {
598 			/*
599 			 * XXX special case for device requiring peer-peer DMA
600 			 */
601 			if (size == 0x100000000UL)
602 				baseptr = &hostbase;
603 			else
604 				baseptr = &pci_emul_membase64;
605 			limit = PCI_EMUL_MEMLIMIT64;
606 			mask = PCIM_BAR_MEM_BASE;
607 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
608 				 PCIM_BAR_MEM_PREFETCH;
609 			break;
610 		} else {
611 			baseptr = &pci_emul_membase32;
612 			limit = PCI_EMUL_MEMLIMIT32;
613 			mask = PCIM_BAR_MEM_BASE;
614 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
615 		}
616 		break;
617 	case PCIBAR_MEM32:
618 		baseptr = &pci_emul_membase32;
619 		limit = PCI_EMUL_MEMLIMIT32;
620 		mask = PCIM_BAR_MEM_BASE;
621 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
622 		break;
623 	default:
624 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
625 		assert(0);
626 	}
627 
628 	if (baseptr != NULL) {
629 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
630 		if (error != 0)
631 			return (error);
632 	}
633 
634 	pdi->pi_bar[idx].type = type;
635 	pdi->pi_bar[idx].addr = addr;
636 	pdi->pi_bar[idx].size = size;
637 
638 	/* Initialize the BAR register in config space */
639 	bar = (addr & mask) | lobits;
640 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
641 
642 	if (type == PCIBAR_MEM64) {
643 		assert(idx + 1 <= PCI_BARMAX);
644 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
645 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
646 	}
647 
648 	register_bar(pdi, idx);
649 
650 	return (0);
651 }
652 
653 #define	CAP_START_OFFSET	0x40
654 static int
655 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
656 {
657 	int i, capoff, reallen;
658 	uint16_t sts;
659 
660 	assert(caplen > 0);
661 
662 	reallen = roundup2(caplen, 4);		/* dword aligned */
663 
664 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
665 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
666 		capoff = CAP_START_OFFSET;
667 	else
668 		capoff = pi->pi_capend + 1;
669 
670 	/* Check if we have enough space */
671 	if (capoff + reallen > PCI_REGMAX + 1)
672 		return (-1);
673 
674 	/* Set the previous capability pointer */
675 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
676 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
677 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
678 	} else
679 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
680 
681 	/* Copy the capability */
682 	for (i = 0; i < caplen; i++)
683 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
684 
685 	/* Set the next capability pointer */
686 	pci_set_cfgdata8(pi, capoff + 1, 0);
687 
688 	pi->pi_prevcap = capoff;
689 	pi->pi_capend = capoff + reallen - 1;
690 	return (0);
691 }
692 
693 static struct pci_devemu *
694 pci_emul_finddev(char *name)
695 {
696 	struct pci_devemu **pdpp, *pdp;
697 
698 	SET_FOREACH(pdpp, pci_devemu_set) {
699 		pdp = *pdpp;
700 		if (!strcmp(pdp->pe_emu, name)) {
701 			return (pdp);
702 		}
703 	}
704 
705 	return (NULL);
706 }
707 
708 static int
709 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
710     int func, struct funcinfo *fi)
711 {
712 	struct pci_devinst *pdi;
713 	int err;
714 
715 	pdi = calloc(1, sizeof(struct pci_devinst));
716 
717 	pdi->pi_vmctx = ctx;
718 	pdi->pi_bus = bus;
719 	pdi->pi_slot = slot;
720 	pdi->pi_func = func;
721 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
722 	pdi->pi_lintr.pin = 0;
723 	pdi->pi_lintr.state = IDLE;
724 	pdi->pi_lintr.pirq_pin = 0;
725 	pdi->pi_lintr.ioapic_irq = 0;
726 	pdi->pi_d = pde;
727 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
728 
729 	/* Disable legacy interrupts */
730 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
731 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
732 
733 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
734 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
735 
736 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
737 	if (err == 0)
738 		fi->fi_devi = pdi;
739 	else
740 		free(pdi);
741 
742 	return (err);
743 }
744 
745 void
746 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
747 {
748 	int mmc;
749 
750 	CTASSERT(sizeof(struct msicap) == 14);
751 
752 	/* Number of msi messages must be a power of 2 between 1 and 32 */
753 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
754 	mmc = ffs(msgnum) - 1;
755 
756 	bzero(msicap, sizeof(struct msicap));
757 	msicap->capid = PCIY_MSI;
758 	msicap->nextptr = nextptr;
759 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
760 }
761 
762 int
763 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
764 {
765 	struct msicap msicap;
766 
767 	pci_populate_msicap(&msicap, msgnum, 0);
768 
769 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
770 }
771 
772 static void
773 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
774 		     uint32_t msix_tab_size)
775 {
776 	CTASSERT(sizeof(struct msixcap) == 12);
777 
778 	assert(msix_tab_size % 4096 == 0);
779 
780 	bzero(msixcap, sizeof(struct msixcap));
781 	msixcap->capid = PCIY_MSIX;
782 
783 	/*
784 	 * Message Control Register, all fields set to
785 	 * zero except for the Table Size.
786 	 * Note: Table size N is encoded as N-1
787 	 */
788 	msixcap->msgctrl = msgnum - 1;
789 
790 	/*
791 	 * MSI-X BAR setup:
792 	 * - MSI-X table start at offset 0
793 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
794 	 */
795 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
796 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
797 }
798 
799 static void
800 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
801 {
802 	int i, table_size;
803 
804 	assert(table_entries > 0);
805 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
806 
807 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
808 	pi->pi_msix.table = calloc(1, table_size);
809 
810 	/* set mask bit of vector control register */
811 	for (i = 0; i < table_entries; i++)
812 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
813 }
814 
815 int
816 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
817 {
818 	uint32_t tab_size;
819 	struct msixcap msixcap;
820 
821 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
822 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
823 
824 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
825 
826 	/* Align table size to nearest 4K */
827 	tab_size = roundup2(tab_size, 4096);
828 
829 	pi->pi_msix.table_bar = barnum;
830 	pi->pi_msix.pba_bar   = barnum;
831 	pi->pi_msix.table_offset = 0;
832 	pi->pi_msix.table_count = msgnum;
833 	pi->pi_msix.pba_offset = tab_size;
834 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
835 
836 	pci_msix_table_init(pi, msgnum);
837 
838 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
839 
840 	/* allocate memory for MSI-X Table and PBA */
841 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
842 				tab_size + pi->pi_msix.pba_size);
843 
844 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
845 					sizeof(msixcap)));
846 }
847 
848 void
849 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
850 		 int bytes, uint32_t val)
851 {
852 	uint16_t msgctrl, rwmask;
853 	int off, table_bar;
854 
855 	off = offset - capoff;
856 	table_bar = pi->pi_msix.table_bar;
857 	/* Message Control Register */
858 	if (off == 2 && bytes == 2) {
859 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
860 		msgctrl = pci_get_cfgdata16(pi, offset);
861 		msgctrl &= ~rwmask;
862 		msgctrl |= val & rwmask;
863 		val = msgctrl;
864 
865 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
866 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
867 		pci_lintr_update(pi);
868 	}
869 
870 	CFGWRITE(pi, offset, val, bytes);
871 }
872 
873 void
874 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
875 		int bytes, uint32_t val)
876 {
877 	uint16_t msgctrl, rwmask, msgdata, mme;
878 	uint32_t addrlo;
879 
880 	/*
881 	 * If guest is writing to the message control register make sure
882 	 * we do not overwrite read-only fields.
883 	 */
884 	if ((offset - capoff) == 2 && bytes == 2) {
885 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
886 		msgctrl = pci_get_cfgdata16(pi, offset);
887 		msgctrl &= ~rwmask;
888 		msgctrl |= val & rwmask;
889 		val = msgctrl;
890 
891 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
892 		if (msgctrl & PCIM_MSICTRL_64BIT)
893 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
894 		else
895 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
896 
897 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
898 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
899 		if (pi->pi_msi.enabled) {
900 			pi->pi_msi.addr = addrlo;
901 			pi->pi_msi.msg_data = msgdata;
902 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
903 		} else {
904 			pi->pi_msi.maxmsgnum = 0;
905 		}
906 		pci_lintr_update(pi);
907 	}
908 
909 	CFGWRITE(pi, offset, val, bytes);
910 }
911 
912 void
913 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
914 		 int bytes, uint32_t val)
915 {
916 
917 	/* XXX don't write to the readonly parts */
918 	CFGWRITE(pi, offset, val, bytes);
919 }
920 
921 #define	PCIECAP_VERSION	0x2
922 int
923 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
924 {
925 	int err;
926 	struct pciecap pciecap;
927 
928 	CTASSERT(sizeof(struct pciecap) == 60);
929 
930 	if (type != PCIEM_TYPE_ROOT_PORT)
931 		return (-1);
932 
933 	bzero(&pciecap, sizeof(pciecap));
934 
935 	pciecap.capid = PCIY_EXPRESS;
936 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
937 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
938 	pciecap.link_status = 0x11;		/* gen1, x1 */
939 
940 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
941 	return (err);
942 }
943 
944 /*
945  * This function assumes that 'coff' is in the capabilities region of the
946  * config space.
947  */
948 static void
949 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
950 {
951 	int capid;
952 	uint8_t capoff, nextoff;
953 
954 	/* Do not allow un-aligned writes */
955 	if ((offset & (bytes - 1)) != 0)
956 		return;
957 
958 	/* Find the capability that we want to update */
959 	capoff = CAP_START_OFFSET;
960 	while (1) {
961 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
962 		if (nextoff == 0)
963 			break;
964 		if (offset >= capoff && offset < nextoff)
965 			break;
966 
967 		capoff = nextoff;
968 	}
969 	assert(offset >= capoff);
970 
971 	/*
972 	 * Capability ID and Next Capability Pointer are readonly.
973 	 * However, some o/s's do 4-byte writes that include these.
974 	 * For this case, trim the write back to 2 bytes and adjust
975 	 * the data.
976 	 */
977 	if (offset == capoff || offset == capoff + 1) {
978 		if (offset == capoff && bytes == 4) {
979 			bytes = 2;
980 			offset += 2;
981 			val >>= 16;
982 		} else
983 			return;
984 	}
985 
986 	capid = pci_get_cfgdata8(pi, capoff);
987 	switch (capid) {
988 	case PCIY_MSI:
989 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
990 		break;
991 	case PCIY_MSIX:
992 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
993 		break;
994 	case PCIY_EXPRESS:
995 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
996 		break;
997 	default:
998 		break;
999 	}
1000 }
1001 
1002 static int
1003 pci_emul_iscap(struct pci_devinst *pi, int offset)
1004 {
1005 	uint16_t sts;
1006 
1007 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1008 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1009 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1010 			return (1);
1011 	}
1012 	return (0);
1013 }
1014 
1015 static int
1016 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1017 			  int size, uint64_t *val, void *arg1, long arg2)
1018 {
1019 	/*
1020 	 * Ignore writes; return 0xff's for reads. The mem read code
1021 	 * will take care of truncating to the correct size.
1022 	 */
1023 	if (dir == MEM_F_READ) {
1024 		*val = 0xffffffffffffffff;
1025 	}
1026 
1027 	return (0);
1028 }
1029 
1030 static int
1031 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1032     int bytes, uint64_t *val, void *arg1, long arg2)
1033 {
1034 	int bus, slot, func, coff, in;
1035 
1036 	coff = addr & 0xfff;
1037 	func = (addr >> 12) & 0x7;
1038 	slot = (addr >> 15) & 0x1f;
1039 	bus = (addr >> 20) & 0xff;
1040 	in = (dir == MEM_F_READ);
1041 	if (in)
1042 		*val = ~0UL;
1043 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1044 	return (0);
1045 }
1046 
1047 uint64_t
1048 pci_ecfg_base(void)
1049 {
1050 
1051 	return (PCI_EMUL_ECFG_BASE);
1052 }
1053 
1054 #define	BUSIO_ROUNDUP		32
1055 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1056 
1057 int
1058 init_pci(struct vmctx *ctx)
1059 {
1060 	struct mem_range mr;
1061 	struct pci_devemu *pde;
1062 	struct businfo *bi;
1063 	struct slotinfo *si;
1064 	struct funcinfo *fi;
1065 	size_t lowmem;
1066 	int bus, slot, func;
1067 	int error;
1068 
1069 	pci_emul_iobase = PCI_EMUL_IOBASE;
1070 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1071 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1072 
1073 	for (bus = 0; bus < MAXBUSES; bus++) {
1074 		if ((bi = pci_businfo[bus]) == NULL)
1075 			continue;
1076 		/*
1077 		 * Keep track of the i/o and memory resources allocated to
1078 		 * this bus.
1079 		 */
1080 		bi->iobase = pci_emul_iobase;
1081 		bi->membase32 = pci_emul_membase32;
1082 		bi->membase64 = pci_emul_membase64;
1083 
1084 		for (slot = 0; slot < MAXSLOTS; slot++) {
1085 			si = &bi->slotinfo[slot];
1086 			for (func = 0; func < MAXFUNCS; func++) {
1087 				fi = &si->si_funcs[func];
1088 				if (fi->fi_name == NULL)
1089 					continue;
1090 				pde = pci_emul_finddev(fi->fi_name);
1091 				assert(pde != NULL);
1092 				error = pci_emul_init(ctx, pde, bus, slot,
1093 				    func, fi);
1094 				if (error)
1095 					return (error);
1096 			}
1097 		}
1098 
1099 		/*
1100 		 * Add some slop to the I/O and memory resources decoded by
1101 		 * this bus to give a guest some flexibility if it wants to
1102 		 * reprogram the BARs.
1103 		 */
1104 		pci_emul_iobase += BUSIO_ROUNDUP;
1105 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1106 		bi->iolimit = pci_emul_iobase;
1107 
1108 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1109 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1110 		    BUSMEM_ROUNDUP);
1111 		bi->memlimit32 = pci_emul_membase32;
1112 
1113 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1114 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1115 		    BUSMEM_ROUNDUP);
1116 		bi->memlimit64 = pci_emul_membase64;
1117 	}
1118 
1119 	/*
1120 	 * PCI backends are initialized before routing INTx interrupts
1121 	 * so that LPC devices are able to reserve ISA IRQs before
1122 	 * routing PIRQ pins.
1123 	 */
1124 	for (bus = 0; bus < MAXBUSES; bus++) {
1125 		if ((bi = pci_businfo[bus]) == NULL)
1126 			continue;
1127 
1128 		for (slot = 0; slot < MAXSLOTS; slot++) {
1129 			si = &bi->slotinfo[slot];
1130 			for (func = 0; func < MAXFUNCS; func++) {
1131 				fi = &si->si_funcs[func];
1132 				if (fi->fi_devi == NULL)
1133 					continue;
1134 				pci_lintr_route(fi->fi_devi);
1135 			}
1136 		}
1137 	}
1138 	lpc_pirq_routed();
1139 
1140 	/*
1141 	 * The guest physical memory map looks like the following:
1142 	 * [0,		    lowmem)		guest system memory
1143 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1144 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1145 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1146 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1147 	 * [4GB,	    4GB + highmem)
1148 	 */
1149 
1150 	/*
1151 	 * Accesses to memory addresses that are not allocated to system
1152 	 * memory or PCI devices return 0xff's.
1153 	 */
1154 	lowmem = vm_get_lowmem_size(ctx);
1155 	bzero(&mr, sizeof(struct mem_range));
1156 	mr.name = "PCI hole";
1157 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1158 	mr.base = lowmem;
1159 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1160 	mr.handler = pci_emul_fallback_handler;
1161 	error = register_mem_fallback(&mr);
1162 	assert(error == 0);
1163 
1164 	/* PCI extended config space */
1165 	bzero(&mr, sizeof(struct mem_range));
1166 	mr.name = "PCI ECFG";
1167 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1168 	mr.base = PCI_EMUL_ECFG_BASE;
1169 	mr.size = PCI_EMUL_ECFG_SIZE;
1170 	mr.handler = pci_emul_ecfg_handler;
1171 	error = register_mem(&mr);
1172 	assert(error == 0);
1173 
1174 	return (0);
1175 }
1176 
1177 static void
1178 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1179     void *arg)
1180 {
1181 
1182 	dsdt_line("  Package ()");
1183 	dsdt_line("  {");
1184 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1185 	dsdt_line("    0x%02X,", pin - 1);
1186 	dsdt_line("    Zero,");
1187 	dsdt_line("    0x%X", ioapic_irq);
1188 	dsdt_line("  },");
1189 }
1190 
1191 static void
1192 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1193     void *arg)
1194 {
1195 	char *name;
1196 
1197 	name = lpc_pirq_name(pirq_pin);
1198 	if (name == NULL)
1199 		return;
1200 	dsdt_line("  Package ()");
1201 	dsdt_line("  {");
1202 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1203 	dsdt_line("    0x%02X,", pin - 1);
1204 	dsdt_line("    %s,", name);
1205 	dsdt_line("    0x00");
1206 	dsdt_line("  },");
1207 	free(name);
1208 }
1209 
1210 /*
1211  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1212  * corresponding to each PCI bus.
1213  */
1214 static void
1215 pci_bus_write_dsdt(int bus)
1216 {
1217 	struct businfo *bi;
1218 	struct slotinfo *si;
1219 	struct pci_devinst *pi;
1220 	int count, func, slot;
1221 
1222 	/*
1223 	 * If there are no devices on this 'bus' then just return.
1224 	 */
1225 	if ((bi = pci_businfo[bus]) == NULL) {
1226 		/*
1227 		 * Bus 0 is special because it decodes the I/O ports used
1228 		 * for PCI config space access even if there are no devices
1229 		 * on it.
1230 		 */
1231 		if (bus != 0)
1232 			return;
1233 	}
1234 
1235 	dsdt_line("  Device (PC%02X)", bus);
1236 	dsdt_line("  {");
1237 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1238 	dsdt_line("    Name (_ADR, Zero)");
1239 
1240 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1241 	dsdt_line("    {");
1242 	dsdt_line("        Return (0x%08X)", bus);
1243 	dsdt_line("    }");
1244 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1245 	dsdt_line("    {");
1246 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1247 	    "MaxFixed, PosDecode,");
1248 	dsdt_line("        0x0000,             // Granularity");
1249 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1250 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1251 	dsdt_line("        0x0000,             // Translation Offset");
1252 	dsdt_line("        0x0001,             // Length");
1253 	dsdt_line("        ,, )");
1254 
1255 	if (bus == 0) {
1256 		dsdt_indent(3);
1257 		dsdt_fixed_ioport(0xCF8, 8);
1258 		dsdt_unindent(3);
1259 
1260 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1261 		    "PosDecode, EntireRange,");
1262 		dsdt_line("        0x0000,             // Granularity");
1263 		dsdt_line("        0x0000,             // Range Minimum");
1264 		dsdt_line("        0x0CF7,             // Range Maximum");
1265 		dsdt_line("        0x0000,             // Translation Offset");
1266 		dsdt_line("        0x0CF8,             // Length");
1267 		dsdt_line("        ,, , TypeStatic)");
1268 
1269 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1270 		    "PosDecode, EntireRange,");
1271 		dsdt_line("        0x0000,             // Granularity");
1272 		dsdt_line("        0x0D00,             // Range Minimum");
1273 		dsdt_line("        0x%04X,             // Range Maximum",
1274 		    PCI_EMUL_IOBASE - 1);
1275 		dsdt_line("        0x0000,             // Translation Offset");
1276 		dsdt_line("        0x%04X,             // Length",
1277 		    PCI_EMUL_IOBASE - 0x0D00);
1278 		dsdt_line("        ,, , TypeStatic)");
1279 
1280 		if (bi == NULL) {
1281 			dsdt_line("    })");
1282 			goto done;
1283 		}
1284 	}
1285 	assert(bi != NULL);
1286 
1287 	/* i/o window */
1288 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1289 	    "PosDecode, EntireRange,");
1290 	dsdt_line("        0x0000,             // Granularity");
1291 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1292 	dsdt_line("        0x%04X,             // Range Maximum",
1293 	    bi->iolimit - 1);
1294 	dsdt_line("        0x0000,             // Translation Offset");
1295 	dsdt_line("        0x%04X,             // Length",
1296 	    bi->iolimit - bi->iobase);
1297 	dsdt_line("        ,, , TypeStatic)");
1298 
1299 	/* mmio window (32-bit) */
1300 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1301 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1302 	dsdt_line("        0x00000000,         // Granularity");
1303 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1304 	dsdt_line("        0x%08X,         // Range Maximum\n",
1305 	    bi->memlimit32 - 1);
1306 	dsdt_line("        0x00000000,         // Translation Offset");
1307 	dsdt_line("        0x%08X,         // Length\n",
1308 	    bi->memlimit32 - bi->membase32);
1309 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1310 
1311 	/* mmio window (64-bit) */
1312 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1313 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1314 	dsdt_line("        0x0000000000000000, // Granularity");
1315 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1316 	dsdt_line("        0x%016lX, // Range Maximum\n",
1317 	    bi->memlimit64 - 1);
1318 	dsdt_line("        0x0000000000000000, // Translation Offset");
1319 	dsdt_line("        0x%016lX, // Length\n",
1320 	    bi->memlimit64 - bi->membase64);
1321 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1322 	dsdt_line("    })");
1323 
1324 	count = pci_count_lintr(bus);
1325 	if (count != 0) {
1326 		dsdt_indent(2);
1327 		dsdt_line("Name (PPRT, Package ()");
1328 		dsdt_line("{");
1329 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1330  		dsdt_line("})");
1331 		dsdt_line("Name (APRT, Package ()");
1332 		dsdt_line("{");
1333 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1334  		dsdt_line("})");
1335 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1336 		dsdt_line("{");
1337 		dsdt_line("  If (PICM)");
1338 		dsdt_line("  {");
1339 		dsdt_line("    Return (APRT)");
1340 		dsdt_line("  }");
1341 		dsdt_line("  Else");
1342 		dsdt_line("  {");
1343 		dsdt_line("    Return (PPRT)");
1344 		dsdt_line("  }");
1345 		dsdt_line("}");
1346 		dsdt_unindent(2);
1347 	}
1348 
1349 	dsdt_indent(2);
1350 	for (slot = 0; slot < MAXSLOTS; slot++) {
1351 		si = &bi->slotinfo[slot];
1352 		for (func = 0; func < MAXFUNCS; func++) {
1353 			pi = si->si_funcs[func].fi_devi;
1354 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1355 				pi->pi_d->pe_write_dsdt(pi);
1356 		}
1357 	}
1358 	dsdt_unindent(2);
1359 done:
1360 	dsdt_line("  }");
1361 }
1362 
1363 void
1364 pci_write_dsdt(void)
1365 {
1366 	int bus;
1367 
1368 	dsdt_indent(1);
1369 	dsdt_line("Name (PICM, 0x00)");
1370 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1371 	dsdt_line("{");
1372 	dsdt_line("  Store (Arg0, PICM)");
1373 	dsdt_line("}");
1374 	dsdt_line("");
1375 	dsdt_line("Scope (_SB)");
1376 	dsdt_line("{");
1377 	for (bus = 0; bus < MAXBUSES; bus++)
1378 		pci_bus_write_dsdt(bus);
1379 	dsdt_line("}");
1380 	dsdt_unindent(1);
1381 }
1382 
1383 int
1384 pci_bus_configured(int bus)
1385 {
1386 	assert(bus >= 0 && bus < MAXBUSES);
1387 	return (pci_businfo[bus] != NULL);
1388 }
1389 
1390 int
1391 pci_msi_enabled(struct pci_devinst *pi)
1392 {
1393 	return (pi->pi_msi.enabled);
1394 }
1395 
1396 int
1397 pci_msi_maxmsgnum(struct pci_devinst *pi)
1398 {
1399 	if (pi->pi_msi.enabled)
1400 		return (pi->pi_msi.maxmsgnum);
1401 	else
1402 		return (0);
1403 }
1404 
1405 int
1406 pci_msix_enabled(struct pci_devinst *pi)
1407 {
1408 
1409 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1410 }
1411 
1412 void
1413 pci_generate_msix(struct pci_devinst *pi, int index)
1414 {
1415 	struct msix_table_entry *mte;
1416 
1417 	if (!pci_msix_enabled(pi))
1418 		return;
1419 
1420 	if (pi->pi_msix.function_mask)
1421 		return;
1422 
1423 	if (index >= pi->pi_msix.table_count)
1424 		return;
1425 
1426 	mte = &pi->pi_msix.table[index];
1427 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1428 		/* XXX Set PBA bit if interrupt is disabled */
1429 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1430 	}
1431 }
1432 
1433 void
1434 pci_generate_msi(struct pci_devinst *pi, int index)
1435 {
1436 
1437 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1438 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1439 			     pi->pi_msi.msg_data + index);
1440 	}
1441 }
1442 
1443 static bool
1444 pci_lintr_permitted(struct pci_devinst *pi)
1445 {
1446 	uint16_t cmd;
1447 
1448 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1449 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1450 		(cmd & PCIM_CMD_INTxDIS)));
1451 }
1452 
1453 void
1454 pci_lintr_request(struct pci_devinst *pi)
1455 {
1456 	struct businfo *bi;
1457 	struct slotinfo *si;
1458 	int bestpin, bestcount, pin;
1459 
1460 	bi = pci_businfo[pi->pi_bus];
1461 	assert(bi != NULL);
1462 
1463 	/*
1464 	 * Just allocate a pin from our slot.  The pin will be
1465 	 * assigned IRQs later when interrupts are routed.
1466 	 */
1467 	si = &bi->slotinfo[pi->pi_slot];
1468 	bestpin = 0;
1469 	bestcount = si->si_intpins[0].ii_count;
1470 	for (pin = 1; pin < 4; pin++) {
1471 		if (si->si_intpins[pin].ii_count < bestcount) {
1472 			bestpin = pin;
1473 			bestcount = si->si_intpins[pin].ii_count;
1474 		}
1475 	}
1476 
1477 	si->si_intpins[bestpin].ii_count++;
1478 	pi->pi_lintr.pin = bestpin + 1;
1479 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1480 }
1481 
1482 static void
1483 pci_lintr_route(struct pci_devinst *pi)
1484 {
1485 	struct businfo *bi;
1486 	struct intxinfo *ii;
1487 
1488 	if (pi->pi_lintr.pin == 0)
1489 		return;
1490 
1491 	bi = pci_businfo[pi->pi_bus];
1492 	assert(bi != NULL);
1493 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1494 
1495 	/*
1496 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1497 	 * is not yet assigned.
1498 	 */
1499 	if (ii->ii_ioapic_irq == 0)
1500 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq();
1501 	assert(ii->ii_ioapic_irq > 0);
1502 
1503 	/*
1504 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1505 	 * not yet assigned.
1506 	 */
1507 	if (ii->ii_pirq_pin == 0)
1508 		ii->ii_pirq_pin = pirq_alloc_pin(pi->pi_vmctx);
1509 	assert(ii->ii_pirq_pin > 0);
1510 
1511 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1512 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1513 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1514 }
1515 
1516 void
1517 pci_lintr_assert(struct pci_devinst *pi)
1518 {
1519 
1520 	assert(pi->pi_lintr.pin > 0);
1521 
1522 	pthread_mutex_lock(&pi->pi_lintr.lock);
1523 	if (pi->pi_lintr.state == IDLE) {
1524 		if (pci_lintr_permitted(pi)) {
1525 			pi->pi_lintr.state = ASSERTED;
1526 			pci_irq_assert(pi);
1527 		} else
1528 			pi->pi_lintr.state = PENDING;
1529 	}
1530 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1531 }
1532 
1533 void
1534 pci_lintr_deassert(struct pci_devinst *pi)
1535 {
1536 
1537 	assert(pi->pi_lintr.pin > 0);
1538 
1539 	pthread_mutex_lock(&pi->pi_lintr.lock);
1540 	if (pi->pi_lintr.state == ASSERTED) {
1541 		pi->pi_lintr.state = IDLE;
1542 		pci_irq_deassert(pi);
1543 	} else if (pi->pi_lintr.state == PENDING)
1544 		pi->pi_lintr.state = IDLE;
1545 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1546 }
1547 
1548 static void
1549 pci_lintr_update(struct pci_devinst *pi)
1550 {
1551 
1552 	pthread_mutex_lock(&pi->pi_lintr.lock);
1553 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1554 		pci_irq_deassert(pi);
1555 		pi->pi_lintr.state = PENDING;
1556 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1557 		pi->pi_lintr.state = ASSERTED;
1558 		pci_irq_assert(pi);
1559 	}
1560 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1561 }
1562 
1563 int
1564 pci_count_lintr(int bus)
1565 {
1566 	int count, slot, pin;
1567 	struct slotinfo *slotinfo;
1568 
1569 	count = 0;
1570 	if (pci_businfo[bus] != NULL) {
1571 		for (slot = 0; slot < MAXSLOTS; slot++) {
1572 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1573 			for (pin = 0; pin < 4; pin++) {
1574 				if (slotinfo->si_intpins[pin].ii_count != 0)
1575 					count++;
1576 			}
1577 		}
1578 	}
1579 	return (count);
1580 }
1581 
1582 void
1583 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1584 {
1585 	struct businfo *bi;
1586 	struct slotinfo *si;
1587 	struct intxinfo *ii;
1588 	int slot, pin;
1589 
1590 	if ((bi = pci_businfo[bus]) == NULL)
1591 		return;
1592 
1593 	for (slot = 0; slot < MAXSLOTS; slot++) {
1594 		si = &bi->slotinfo[slot];
1595 		for (pin = 0; pin < 4; pin++) {
1596 			ii = &si->si_intpins[pin];
1597 			if (ii->ii_count != 0)
1598 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1599 				    ii->ii_ioapic_irq, arg);
1600 		}
1601 	}
1602 }
1603 
1604 /*
1605  * Return 1 if the emulated device in 'slot' is a multi-function device.
1606  * Return 0 otherwise.
1607  */
1608 static int
1609 pci_emul_is_mfdev(int bus, int slot)
1610 {
1611 	struct businfo *bi;
1612 	struct slotinfo *si;
1613 	int f, numfuncs;
1614 
1615 	numfuncs = 0;
1616 	if ((bi = pci_businfo[bus]) != NULL) {
1617 		si = &bi->slotinfo[slot];
1618 		for (f = 0; f < MAXFUNCS; f++) {
1619 			if (si->si_funcs[f].fi_devi != NULL) {
1620 				numfuncs++;
1621 			}
1622 		}
1623 	}
1624 	return (numfuncs > 1);
1625 }
1626 
1627 /*
1628  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1629  * whether or not is a multi-function being emulated in the pci 'slot'.
1630  */
1631 static void
1632 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1633 {
1634 	int mfdev;
1635 
1636 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1637 		mfdev = pci_emul_is_mfdev(bus, slot);
1638 		switch (bytes) {
1639 		case 1:
1640 		case 2:
1641 			*rv &= ~PCIM_MFDEV;
1642 			if (mfdev) {
1643 				*rv |= PCIM_MFDEV;
1644 			}
1645 			break;
1646 		case 4:
1647 			*rv &= ~(PCIM_MFDEV << 16);
1648 			if (mfdev) {
1649 				*rv |= (PCIM_MFDEV << 16);
1650 			}
1651 			break;
1652 		}
1653 	}
1654 }
1655 
1656 static uint32_t
1657 bits_changed(uint32_t old, uint32_t new, uint32_t mask)
1658 {
1659 
1660 	return ((old ^ new) & mask);
1661 }
1662 
1663 static void
1664 pci_emul_cmdwrite(struct pci_devinst *pi, uint32_t new, int bytes)
1665 {
1666 	int i;
1667 	uint16_t old;
1668 
1669 	/*
1670 	 * The command register is at an offset of 4 bytes and thus the
1671 	 * guest could write 1, 2 or 4 bytes starting at this offset.
1672 	 */
1673 
1674 	old = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1675 	CFGWRITE(pi, PCIR_COMMAND, new, bytes);		/* update config */
1676 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1677 
1678 	/*
1679 	 * If the MMIO or I/O address space decoding has changed then
1680 	 * register/unregister all BARs that decode that address space.
1681 	 */
1682 	for (i = 0; i <= PCI_BARMAX; i++) {
1683 		switch (pi->pi_bar[i].type) {
1684 			case PCIBAR_NONE:
1685 			case PCIBAR_MEMHI64:
1686 				break;
1687 			case PCIBAR_IO:
1688 				/* I/O address space decoding changed? */
1689 				if (bits_changed(old, new, PCIM_CMD_PORTEN)) {
1690 					if (porten(pi))
1691 						register_bar(pi, i);
1692 					else
1693 						unregister_bar(pi, i);
1694 				}
1695 				break;
1696 			case PCIBAR_MEM32:
1697 			case PCIBAR_MEM64:
1698 				/* MMIO address space decoding changed? */
1699 				if (bits_changed(old, new, PCIM_CMD_MEMEN)) {
1700 					if (memen(pi))
1701 						register_bar(pi, i);
1702 					else
1703 						unregister_bar(pi, i);
1704 				}
1705 				break;
1706 			default:
1707 				assert(0);
1708 		}
1709 	}
1710 
1711 	/*
1712 	 * If INTx has been unmasked and is pending, assert the
1713 	 * interrupt.
1714 	 */
1715 	pci_lintr_update(pi);
1716 }
1717 
1718 static void
1719 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1720     int coff, int bytes, uint32_t *eax)
1721 {
1722 	struct businfo *bi;
1723 	struct slotinfo *si;
1724 	struct pci_devinst *pi;
1725 	struct pci_devemu *pe;
1726 	int idx, needcfg;
1727 	uint64_t addr, bar, mask;
1728 
1729 	if ((bi = pci_businfo[bus]) != NULL) {
1730 		si = &bi->slotinfo[slot];
1731 		pi = si->si_funcs[func].fi_devi;
1732 	} else
1733 		pi = NULL;
1734 
1735 	/*
1736 	 * Just return if there is no device at this slot:func or if the
1737 	 * the guest is doing an un-aligned access.
1738 	 */
1739 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1740 	    (coff & (bytes - 1)) != 0) {
1741 		if (in)
1742 			*eax = 0xffffffff;
1743 		return;
1744 	}
1745 
1746 	/*
1747 	 * Ignore all writes beyond the standard config space and return all
1748 	 * ones on reads.
1749 	 */
1750 	if (coff >= PCI_REGMAX + 1) {
1751 		if (in) {
1752 			*eax = 0xffffffff;
1753 			/*
1754 			 * Extended capabilities begin at offset 256 in config
1755 			 * space. Absence of extended capabilities is signaled
1756 			 * with all 0s in the extended capability header at
1757 			 * offset 256.
1758 			 */
1759 			if (coff <= PCI_REGMAX + 4)
1760 				*eax = 0x00000000;
1761 		}
1762 		return;
1763 	}
1764 
1765 	pe = pi->pi_d;
1766 
1767 	/*
1768 	 * Config read
1769 	 */
1770 	if (in) {
1771 		/* Let the device emulation override the default handler */
1772 		if (pe->pe_cfgread != NULL) {
1773 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1774 			    eax);
1775 		} else {
1776 			needcfg = 1;
1777 		}
1778 
1779 		if (needcfg) {
1780 			if (bytes == 1)
1781 				*eax = pci_get_cfgdata8(pi, coff);
1782 			else if (bytes == 2)
1783 				*eax = pci_get_cfgdata16(pi, coff);
1784 			else
1785 				*eax = pci_get_cfgdata32(pi, coff);
1786 		}
1787 
1788 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1789 	} else {
1790 		/* Let the device emulation override the default handler */
1791 		if (pe->pe_cfgwrite != NULL &&
1792 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1793 			return;
1794 
1795 		/*
1796 		 * Special handling for write to BAR registers
1797 		 */
1798 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1799 			/*
1800 			 * Ignore writes to BAR registers that are not
1801 			 * 4-byte aligned.
1802 			 */
1803 			if (bytes != 4 || (coff & 0x3) != 0)
1804 				return;
1805 			idx = (coff - PCIR_BAR(0)) / 4;
1806 			mask = ~(pi->pi_bar[idx].size - 1);
1807 			switch (pi->pi_bar[idx].type) {
1808 			case PCIBAR_NONE:
1809 				pi->pi_bar[idx].addr = bar = 0;
1810 				break;
1811 			case PCIBAR_IO:
1812 				addr = *eax & mask;
1813 				addr &= 0xffff;
1814 				bar = addr | PCIM_BAR_IO_SPACE;
1815 				/*
1816 				 * Register the new BAR value for interception
1817 				 */
1818 				if (addr != pi->pi_bar[idx].addr) {
1819 					update_bar_address(pi, addr, idx,
1820 							   PCIBAR_IO);
1821 				}
1822 				break;
1823 			case PCIBAR_MEM32:
1824 				addr = bar = *eax & mask;
1825 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1826 				if (addr != pi->pi_bar[idx].addr) {
1827 					update_bar_address(pi, addr, idx,
1828 							   PCIBAR_MEM32);
1829 				}
1830 				break;
1831 			case PCIBAR_MEM64:
1832 				addr = bar = *eax & mask;
1833 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1834 				       PCIM_BAR_MEM_PREFETCH;
1835 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1836 					update_bar_address(pi, addr, idx,
1837 							   PCIBAR_MEM64);
1838 				}
1839 				break;
1840 			case PCIBAR_MEMHI64:
1841 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1842 				addr = ((uint64_t)*eax << 32) & mask;
1843 				bar = addr >> 32;
1844 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1845 					update_bar_address(pi, addr, idx - 1,
1846 							   PCIBAR_MEMHI64);
1847 				}
1848 				break;
1849 			default:
1850 				assert(0);
1851 			}
1852 			pci_set_cfgdata32(pi, coff, bar);
1853 
1854 		} else if (pci_emul_iscap(pi, coff)) {
1855 			pci_emul_capwrite(pi, coff, bytes, *eax);
1856 		} else if (coff == PCIR_COMMAND) {
1857 			pci_emul_cmdwrite(pi, *eax, bytes);
1858 		} else {
1859 			CFGWRITE(pi, coff, *eax, bytes);
1860 		}
1861 	}
1862 }
1863 
1864 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1865 
1866 static int
1867 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1868 		 uint32_t *eax, void *arg)
1869 {
1870 	uint32_t x;
1871 
1872 	if (bytes != 4) {
1873 		if (in)
1874 			*eax = (bytes == 2) ? 0xffff : 0xff;
1875 		return (0);
1876 	}
1877 
1878 	if (in) {
1879 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1880 		if (cfgenable)
1881 			x |= CONF1_ENABLE;
1882 		*eax = x;
1883 	} else {
1884 		x = *eax;
1885 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1886 		cfgoff = x & PCI_REGMAX;
1887 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1888 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1889 		cfgbus = (x >> 16) & PCI_BUSMAX;
1890 	}
1891 
1892 	return (0);
1893 }
1894 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1895 
1896 static int
1897 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1898 		 uint32_t *eax, void *arg)
1899 {
1900 	int coff;
1901 
1902 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1903 
1904 	coff = cfgoff + (port - CONF1_DATA_PORT);
1905 	if (cfgenable) {
1906 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1907 		    eax);
1908 	} else {
1909 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1910 		if (in)
1911 			*eax = 0xffffffff;
1912 	}
1913 	return (0);
1914 }
1915 
1916 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1917 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1918 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1919 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1920 
1921 #define PCI_EMUL_TEST
1922 #ifdef PCI_EMUL_TEST
1923 /*
1924  * Define a dummy test device
1925  */
1926 #define DIOSZ	8
1927 #define DMEMSZ	4096
1928 struct pci_emul_dsoftc {
1929 	uint8_t   ioregs[DIOSZ];
1930 	uint8_t	  memregs[DMEMSZ];
1931 };
1932 
1933 #define	PCI_EMUL_MSI_MSGS	 4
1934 #define	PCI_EMUL_MSIX_MSGS	16
1935 
1936 static int
1937 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1938 {
1939 	int error;
1940 	struct pci_emul_dsoftc *sc;
1941 
1942 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1943 
1944 	pi->pi_arg = sc;
1945 
1946 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1947 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1948 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1949 
1950 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1951 	assert(error == 0);
1952 
1953 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1954 	assert(error == 0);
1955 
1956 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1957 	assert(error == 0);
1958 
1959 	return (0);
1960 }
1961 
1962 static void
1963 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1964 	      uint64_t offset, int size, uint64_t value)
1965 {
1966 	int i;
1967 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1968 
1969 	if (baridx == 0) {
1970 		if (offset + size > DIOSZ) {
1971 			printf("diow: iow too large, offset %ld size %d\n",
1972 			       offset, size);
1973 			return;
1974 		}
1975 
1976 		if (size == 1) {
1977 			sc->ioregs[offset] = value & 0xff;
1978 		} else if (size == 2) {
1979 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1980 		} else if (size == 4) {
1981 			*(uint32_t *)&sc->ioregs[offset] = value;
1982 		} else {
1983 			printf("diow: iow unknown size %d\n", size);
1984 		}
1985 
1986 		/*
1987 		 * Special magic value to generate an interrupt
1988 		 */
1989 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
1990 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
1991 
1992 		if (value == 0xabcdef) {
1993 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
1994 				pci_generate_msi(pi, i);
1995 		}
1996 	}
1997 
1998 	if (baridx == 1) {
1999 		if (offset + size > DMEMSZ) {
2000 			printf("diow: memw too large, offset %ld size %d\n",
2001 			       offset, size);
2002 			return;
2003 		}
2004 
2005 		if (size == 1) {
2006 			sc->memregs[offset] = value;
2007 		} else if (size == 2) {
2008 			*(uint16_t *)&sc->memregs[offset] = value;
2009 		} else if (size == 4) {
2010 			*(uint32_t *)&sc->memregs[offset] = value;
2011 		} else if (size == 8) {
2012 			*(uint64_t *)&sc->memregs[offset] = value;
2013 		} else {
2014 			printf("diow: memw unknown size %d\n", size);
2015 		}
2016 
2017 		/*
2018 		 * magic interrupt ??
2019 		 */
2020 	}
2021 
2022 	if (baridx > 1) {
2023 		printf("diow: unknown bar idx %d\n", baridx);
2024 	}
2025 }
2026 
2027 static uint64_t
2028 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2029 	      uint64_t offset, int size)
2030 {
2031 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2032 	uint32_t value;
2033 
2034 	if (baridx == 0) {
2035 		if (offset + size > DIOSZ) {
2036 			printf("dior: ior too large, offset %ld size %d\n",
2037 			       offset, size);
2038 			return (0);
2039 		}
2040 
2041 		if (size == 1) {
2042 			value = sc->ioregs[offset];
2043 		} else if (size == 2) {
2044 			value = *(uint16_t *) &sc->ioregs[offset];
2045 		} else if (size == 4) {
2046 			value = *(uint32_t *) &sc->ioregs[offset];
2047 		} else {
2048 			printf("dior: ior unknown size %d\n", size);
2049 		}
2050 	}
2051 
2052 	if (baridx == 1) {
2053 		if (offset + size > DMEMSZ) {
2054 			printf("dior: memr too large, offset %ld size %d\n",
2055 			       offset, size);
2056 			return (0);
2057 		}
2058 
2059 		if (size == 1) {
2060 			value = sc->memregs[offset];
2061 		} else if (size == 2) {
2062 			value = *(uint16_t *) &sc->memregs[offset];
2063 		} else if (size == 4) {
2064 			value = *(uint32_t *) &sc->memregs[offset];
2065 		} else if (size == 8) {
2066 			value = *(uint64_t *) &sc->memregs[offset];
2067 		} else {
2068 			printf("dior: ior unknown size %d\n", size);
2069 		}
2070 	}
2071 
2072 
2073 	if (baridx > 1) {
2074 		printf("dior: unknown bar idx %d\n", baridx);
2075 		return (0);
2076 	}
2077 
2078 	return (value);
2079 }
2080 
2081 struct pci_devemu pci_dummy = {
2082 	.pe_emu = "dummy",
2083 	.pe_init = pci_emul_dinit,
2084 	.pe_barwrite = pci_emul_diow,
2085 	.pe_barread = pci_emul_dior
2086 };
2087 PCI_EMUL_SET(pci_dummy);
2088 
2089 #endif /* PCI_EMUL_TEST */
2090