xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision b1c5f60ce87cc2f179dfb81de507d9b7bf59564c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/linker_set.h>
36 #include <sys/mman.h>
37 
38 #include <ctype.h>
39 #include <err.h>
40 #include <errno.h>
41 #include <pthread.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <string.h>
45 #include <strings.h>
46 #include <assert.h>
47 #include <stdbool.h>
48 #include <sysexits.h>
49 
50 #include <machine/vmm.h>
51 #include <machine/vmm_snapshot.h>
52 #include <vmmapi.h>
53 
54 #include "acpi.h"
55 #include "bhyverun.h"
56 #include "config.h"
57 #include "debug.h"
58 #include "inout.h"
59 #include "ioapic.h"
60 #include "mem.h"
61 #include "pci_emul.h"
62 #include "pci_irq.h"
63 #include "pci_lpc.h"
64 
65 #define CONF1_ADDR_PORT	   0x0cf8
66 #define CONF1_DATA_PORT	   0x0cfc
67 
68 #define CONF1_ENABLE	   0x80000000ul
69 
70 #define	MAXBUSES	(PCI_BUSMAX + 1)
71 #define MAXSLOTS	(PCI_SLOTMAX + 1)
72 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
73 
74 #define GB		(1024 * 1024 * 1024UL)
75 
76 struct funcinfo {
77 	nvlist_t *fi_config;
78 	struct pci_devemu *fi_pde;
79 	struct pci_devinst *fi_devi;
80 };
81 
82 struct intxinfo {
83 	int	ii_count;
84 	int	ii_pirq_pin;
85 	int	ii_ioapic_irq;
86 };
87 
88 struct slotinfo {
89 	struct intxinfo si_intpins[4];
90 	struct funcinfo si_funcs[MAXFUNCS];
91 };
92 
93 struct businfo {
94 	uint16_t iobase, iolimit;		/* I/O window */
95 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
96 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
97 	struct slotinfo slotinfo[MAXSLOTS];
98 };
99 
100 static struct businfo *pci_businfo[MAXBUSES];
101 
102 SET_DECLARE(pci_devemu_set, struct pci_devemu);
103 
104 static uint64_t pci_emul_iobase;
105 static uint8_t *pci_emul_rombase;
106 static uint64_t pci_emul_romoffset;
107 static uint8_t *pci_emul_romlim;
108 static uint64_t pci_emul_membase32;
109 static uint64_t pci_emul_membase64;
110 static uint64_t pci_emul_memlim64;
111 
112 struct pci_bar_allocation {
113 	TAILQ_ENTRY(pci_bar_allocation) chain;
114 	struct pci_devinst *pdi;
115 	int idx;
116 	enum pcibar_type type;
117 	uint64_t size;
118 };
119 TAILQ_HEAD(pci_bar_list, pci_bar_allocation) pci_bars = TAILQ_HEAD_INITIALIZER(
120     pci_bars);
121 
122 #define	PCI_EMUL_IOBASE		0x2000
123 #define	PCI_EMUL_IOLIMIT	0x10000
124 
125 #define PCI_EMUL_ROMSIZE 0x10000000
126 
127 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
128 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
129 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
130 
131 /*
132  * OVMF always uses 0xC0000000 as base address for 32 bit PCI MMIO. Don't
133  * change this address without changing it in OVMF.
134  */
135 #define PCI_EMUL_MEMBASE32 0xC0000000
136 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
137 #define PCI_EMUL_MEMSIZE64	(32*GB)
138 
139 static struct pci_devemu *pci_emul_finddev(const char *name);
140 static void pci_lintr_route(struct pci_devinst *pi);
141 static void pci_lintr_update(struct pci_devinst *pi);
142 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
143     int func, int coff, int bytes, uint32_t *val);
144 
145 static __inline void
146 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
147 {
148 
149 	if (bytes == 1)
150 		pci_set_cfgdata8(pi, coff, val);
151 	else if (bytes == 2)
152 		pci_set_cfgdata16(pi, coff, val);
153 	else
154 		pci_set_cfgdata32(pi, coff, val);
155 }
156 
157 static __inline uint32_t
158 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
159 {
160 
161 	if (bytes == 1)
162 		return (pci_get_cfgdata8(pi, coff));
163 	else if (bytes == 2)
164 		return (pci_get_cfgdata16(pi, coff));
165 	else
166 		return (pci_get_cfgdata32(pi, coff));
167 }
168 
169 /*
170  * I/O access
171  */
172 
173 /*
174  * Slot options are in the form:
175  *
176  *  <bus>:<slot>:<func>,<emul>[,<config>]
177  *  <slot>[:<func>],<emul>[,<config>]
178  *
179  *  slot is 0..31
180  *  func is 0..7
181  *  emul is a string describing the type of PCI device e.g. virtio-net
182  *  config is an optional string, depending on the device, that can be
183  *  used for configuration.
184  *   Examples are:
185  *     1,virtio-net,tap0
186  *     3:0,dummy
187  */
188 static void
189 pci_parse_slot_usage(char *aopt)
190 {
191 
192 	EPRINTLN("Invalid PCI slot info field \"%s\"", aopt);
193 }
194 
195 /*
196  * Helper function to parse a list of comma-separated options where
197  * each option is formatted as "name[=value]".  If no value is
198  * provided, the option is treated as a boolean and is given a value
199  * of true.
200  */
201 int
202 pci_parse_legacy_config(nvlist_t *nvl, const char *opt)
203 {
204 	char *config, *name, *tofree, *value;
205 
206 	if (opt == NULL)
207 		return (0);
208 
209 	config = tofree = strdup(opt);
210 	while ((name = strsep(&config, ",")) != NULL) {
211 		value = strchr(name, '=');
212 		if (value != NULL) {
213 			*value = '\0';
214 			value++;
215 			set_config_value_node(nvl, name, value);
216 		} else
217 			set_config_bool_node(nvl, name, true);
218 	}
219 	free(tofree);
220 	return (0);
221 }
222 
223 /*
224  * PCI device configuration is stored in MIBs that encode the device's
225  * location:
226  *
227  * pci.<bus>.<slot>.<func>
228  *
229  * Where "bus", "slot", and "func" are all decimal values without
230  * leading zeroes.  Each valid device must have a "device" node which
231  * identifies the driver model of the device.
232  *
233  * Device backends can provide a parser for the "config" string.  If
234  * a custom parser is not provided, pci_parse_legacy_config() is used
235  * to parse the string.
236  */
237 int
238 pci_parse_slot(char *opt)
239 {
240 	char node_name[sizeof("pci.XXX.XX.X")];
241 	struct pci_devemu *pde;
242 	char *emul, *config, *str, *cp;
243 	int error, bnum, snum, fnum;
244 	nvlist_t *nvl;
245 
246 	error = -1;
247 	str = strdup(opt);
248 
249 	emul = config = NULL;
250 	if ((cp = strchr(str, ',')) != NULL) {
251 		*cp = '\0';
252 		emul = cp + 1;
253 		if ((cp = strchr(emul, ',')) != NULL) {
254 			*cp = '\0';
255 			config = cp + 1;
256 		}
257 	} else {
258 		pci_parse_slot_usage(opt);
259 		goto done;
260 	}
261 
262 	/* <bus>:<slot>:<func> */
263 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
264 		bnum = 0;
265 		/* <slot>:<func> */
266 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
267 			fnum = 0;
268 			/* <slot> */
269 			if (sscanf(str, "%d", &snum) != 1) {
270 				snum = -1;
271 			}
272 		}
273 	}
274 
275 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
276 	    fnum < 0 || fnum >= MAXFUNCS) {
277 		pci_parse_slot_usage(opt);
278 		goto done;
279 	}
280 
281 	pde = pci_emul_finddev(emul);
282 	if (pde == NULL) {
283 		EPRINTLN("pci slot %d:%d:%d: unknown device \"%s\"", bnum, snum,
284 		    fnum, emul);
285 		goto done;
286 	}
287 
288 	snprintf(node_name, sizeof(node_name), "pci.%d.%d.%d", bnum, snum,
289 	    fnum);
290 	nvl = find_config_node(node_name);
291 	if (nvl != NULL) {
292 		EPRINTLN("pci slot %d:%d:%d already occupied!", bnum, snum,
293 		    fnum);
294 		goto done;
295 	}
296 	nvl = create_config_node(node_name);
297 	if (pde->pe_alias != NULL)
298 		set_config_value_node(nvl, "device", pde->pe_alias);
299 	else
300 		set_config_value_node(nvl, "device", pde->pe_emu);
301 
302 	if (pde->pe_legacy_config != NULL)
303 		error = pde->pe_legacy_config(nvl, config);
304 	else
305 		error = pci_parse_legacy_config(nvl, config);
306 done:
307 	free(str);
308 	return (error);
309 }
310 
311 void
312 pci_print_supported_devices()
313 {
314 	struct pci_devemu **pdpp, *pdp;
315 
316 	SET_FOREACH(pdpp, pci_devemu_set) {
317 		pdp = *pdpp;
318 		printf("%s\n", pdp->pe_emu);
319 	}
320 }
321 
322 static int
323 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
324 {
325 
326 	if (offset < pi->pi_msix.pba_offset)
327 		return (0);
328 
329 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
330 		return (0);
331 	}
332 
333 	return (1);
334 }
335 
336 int
337 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
338 		     uint64_t value)
339 {
340 	int msix_entry_offset;
341 	int tab_index;
342 	char *dest;
343 
344 	/* support only 4 or 8 byte writes */
345 	if (size != 4 && size != 8)
346 		return (-1);
347 
348 	/*
349 	 * Return if table index is beyond what device supports
350 	 */
351 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
352 	if (tab_index >= pi->pi_msix.table_count)
353 		return (-1);
354 
355 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
356 
357 	/* support only aligned writes */
358 	if ((msix_entry_offset % size) != 0)
359 		return (-1);
360 
361 	dest = (char *)(pi->pi_msix.table + tab_index);
362 	dest += msix_entry_offset;
363 
364 	if (size == 4)
365 		*((uint32_t *)dest) = value;
366 	else
367 		*((uint64_t *)dest) = value;
368 
369 	return (0);
370 }
371 
372 uint64_t
373 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
374 {
375 	char *dest;
376 	int msix_entry_offset;
377 	int tab_index;
378 	uint64_t retval = ~0;
379 
380 	/*
381 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
382 	 * table but we also allow 1 byte access to accommodate reads from
383 	 * ddb.
384 	 */
385 	if (size != 1 && size != 4 && size != 8)
386 		return (retval);
387 
388 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
389 
390 	/* support only aligned reads */
391 	if ((msix_entry_offset % size) != 0) {
392 		return (retval);
393 	}
394 
395 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
396 
397 	if (tab_index < pi->pi_msix.table_count) {
398 		/* valid MSI-X Table access */
399 		dest = (char *)(pi->pi_msix.table + tab_index);
400 		dest += msix_entry_offset;
401 
402 		if (size == 1)
403 			retval = *((uint8_t *)dest);
404 		else if (size == 4)
405 			retval = *((uint32_t *)dest);
406 		else
407 			retval = *((uint64_t *)dest);
408 	} else if (pci_valid_pba_offset(pi, offset)) {
409 		/* return 0 for PBA access */
410 		retval = 0;
411 	}
412 
413 	return (retval);
414 }
415 
416 int
417 pci_msix_table_bar(struct pci_devinst *pi)
418 {
419 
420 	if (pi->pi_msix.table != NULL)
421 		return (pi->pi_msix.table_bar);
422 	else
423 		return (-1);
424 }
425 
426 int
427 pci_msix_pba_bar(struct pci_devinst *pi)
428 {
429 
430 	if (pi->pi_msix.table != NULL)
431 		return (pi->pi_msix.pba_bar);
432 	else
433 		return (-1);
434 }
435 
436 static int
437 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
438 		    uint32_t *eax, void *arg)
439 {
440 	struct pci_devinst *pdi = arg;
441 	struct pci_devemu *pe = pdi->pi_d;
442 	uint64_t offset;
443 	int i;
444 
445 	for (i = 0; i <= PCI_BARMAX; i++) {
446 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
447 		    port >= pdi->pi_bar[i].addr &&
448 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
449 			offset = port - pdi->pi_bar[i].addr;
450 			if (in)
451 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
452 							 offset, bytes);
453 			else
454 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
455 						   bytes, *eax);
456 			return (0);
457 		}
458 	}
459 	return (-1);
460 }
461 
462 static int
463 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
464 		     int size, uint64_t *val, void *arg1, long arg2)
465 {
466 	struct pci_devinst *pdi = arg1;
467 	struct pci_devemu *pe = pdi->pi_d;
468 	uint64_t offset;
469 	int bidx = (int) arg2;
470 
471 	assert(bidx <= PCI_BARMAX);
472 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
473 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
474 	assert(addr >= pdi->pi_bar[bidx].addr &&
475 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
476 
477 	offset = addr - pdi->pi_bar[bidx].addr;
478 
479 	if (dir == MEM_F_WRITE) {
480 		if (size == 8) {
481 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
482 					   4, *val & 0xffffffff);
483 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
484 					   4, *val >> 32);
485 		} else {
486 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
487 					   size, *val);
488 		}
489 	} else {
490 		if (size == 8) {
491 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
492 						 offset, 4);
493 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
494 						  offset + 4, 4) << 32;
495 		} else {
496 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
497 						 offset, size);
498 		}
499 	}
500 
501 	return (0);
502 }
503 
504 
505 static int
506 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
507 			uint64_t *addr)
508 {
509 	uint64_t base;
510 
511 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
512 
513 	base = roundup2(*baseptr, size);
514 
515 	if (base + size <= limit) {
516 		*addr = base;
517 		*baseptr = base + size;
518 		return (0);
519 	} else
520 		return (-1);
521 }
522 
523 /*
524  * Register (or unregister) the MMIO or I/O region associated with the BAR
525  * register 'idx' of an emulated pci device.
526  */
527 static void
528 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
529 {
530 	struct pci_devemu *pe;
531 	int error;
532 	struct inout_port iop;
533 	struct mem_range mr;
534 
535 	pe = pi->pi_d;
536 	switch (pi->pi_bar[idx].type) {
537 	case PCIBAR_IO:
538 		bzero(&iop, sizeof(struct inout_port));
539 		iop.name = pi->pi_name;
540 		iop.port = pi->pi_bar[idx].addr;
541 		iop.size = pi->pi_bar[idx].size;
542 		if (registration) {
543 			iop.flags = IOPORT_F_INOUT;
544 			iop.handler = pci_emul_io_handler;
545 			iop.arg = pi;
546 			error = register_inout(&iop);
547 		} else
548 			error = unregister_inout(&iop);
549 		if (pe->pe_baraddr != NULL)
550 			(*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration,
551 					  pi->pi_bar[idx].addr);
552 		break;
553 	case PCIBAR_MEM32:
554 	case PCIBAR_MEM64:
555 		bzero(&mr, sizeof(struct mem_range));
556 		mr.name = pi->pi_name;
557 		mr.base = pi->pi_bar[idx].addr;
558 		mr.size = pi->pi_bar[idx].size;
559 		if (registration) {
560 			mr.flags = MEM_F_RW;
561 			mr.handler = pci_emul_mem_handler;
562 			mr.arg1 = pi;
563 			mr.arg2 = idx;
564 			error = register_mem(&mr);
565 		} else
566 			error = unregister_mem(&mr);
567 		if (pe->pe_baraddr != NULL)
568 			(*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration,
569 					  pi->pi_bar[idx].addr);
570 		break;
571 	case PCIBAR_ROM:
572 		error = 0;
573 		if (pe->pe_baraddr != NULL)
574 			(*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration,
575 			    pi->pi_bar[idx].addr);
576 		break;
577 	default:
578 		error = EINVAL;
579 		break;
580 	}
581 	assert(error == 0);
582 }
583 
584 static void
585 unregister_bar(struct pci_devinst *pi, int idx)
586 {
587 
588 	modify_bar_registration(pi, idx, 0);
589 }
590 
591 static void
592 register_bar(struct pci_devinst *pi, int idx)
593 {
594 
595 	modify_bar_registration(pi, idx, 1);
596 }
597 
598 /* Is the ROM enabled for the emulated pci device? */
599 static int
600 romen(struct pci_devinst *pi)
601 {
602 	return (pi->pi_bar[PCI_ROM_IDX].lobits & PCIM_BIOS_ENABLE) ==
603 	    PCIM_BIOS_ENABLE;
604 }
605 
606 /* Are we decoding i/o port accesses for the emulated pci device? */
607 static int
608 porten(struct pci_devinst *pi)
609 {
610 	uint16_t cmd;
611 
612 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
613 
614 	return (cmd & PCIM_CMD_PORTEN);
615 }
616 
617 /* Are we decoding memory accesses for the emulated pci device? */
618 static int
619 memen(struct pci_devinst *pi)
620 {
621 	uint16_t cmd;
622 
623 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
624 
625 	return (cmd & PCIM_CMD_MEMEN);
626 }
627 
628 /*
629  * Update the MMIO or I/O address that is decoded by the BAR register.
630  *
631  * If the pci device has enabled the address space decoding then intercept
632  * the address range decoded by the BAR register.
633  */
634 static void
635 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
636 {
637 	int decode;
638 
639 	if (pi->pi_bar[idx].type == PCIBAR_IO)
640 		decode = porten(pi);
641 	else
642 		decode = memen(pi);
643 
644 	if (decode)
645 		unregister_bar(pi, idx);
646 
647 	switch (type) {
648 	case PCIBAR_IO:
649 	case PCIBAR_MEM32:
650 		pi->pi_bar[idx].addr = addr;
651 		break;
652 	case PCIBAR_MEM64:
653 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
654 		pi->pi_bar[idx].addr |= addr;
655 		break;
656 	case PCIBAR_MEMHI64:
657 		pi->pi_bar[idx].addr &= 0xffffffff;
658 		pi->pi_bar[idx].addr |= addr;
659 		break;
660 	default:
661 		assert(0);
662 	}
663 
664 	if (decode)
665 		register_bar(pi, idx);
666 }
667 
668 int
669 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
670     uint64_t size)
671 {
672 	assert((type == PCIBAR_ROM) || (idx >= 0 && idx <= PCI_BARMAX));
673 	assert((type != PCIBAR_ROM) || (idx == PCI_ROM_IDX));
674 
675 	if ((size & (size - 1)) != 0)
676 		size = 1UL << flsl(size);	/* round up to a power of 2 */
677 
678 	/* Enforce minimum BAR sizes required by the PCI standard */
679 	if (type == PCIBAR_IO) {
680 		if (size < 4)
681 			size = 4;
682 	} else if (type == PCIBAR_ROM) {
683 		if (size < ~PCIM_BIOS_ADDR_MASK + 1)
684 			size = ~PCIM_BIOS_ADDR_MASK + 1;
685 	} else {
686 		if (size < 16)
687 			size = 16;
688 	}
689 
690 	/*
691 	 * To reduce fragmentation of the MMIO space, we allocate the BARs by
692 	 * size. Therefore, don't allocate the BAR yet. We create a list of all
693 	 * BAR allocation which is sorted by BAR size. When all PCI devices are
694 	 * initialized, we will assign an address to the BARs.
695 	 */
696 
697 	/* create a new list entry */
698 	struct pci_bar_allocation *const new_bar = malloc(sizeof(*new_bar));
699 	memset(new_bar, 0, sizeof(*new_bar));
700 	new_bar->pdi = pdi;
701 	new_bar->idx = idx;
702 	new_bar->type = type;
703 	new_bar->size = size;
704 
705 	/*
706 	 * Search for a BAR which size is lower than the size of our newly
707 	 * allocated BAR.
708 	 */
709 	struct pci_bar_allocation *bar = NULL;
710 	TAILQ_FOREACH(bar, &pci_bars, chain) {
711 		if (bar->size < size) {
712 			break;
713 		}
714 	}
715 
716 	if (bar == NULL) {
717 		/*
718 		 * Either the list is empty or new BAR is the smallest BAR of
719 		 * the list. Append it to the end of our list.
720 		 */
721 		TAILQ_INSERT_TAIL(&pci_bars, new_bar, chain);
722 	} else {
723 		/*
724 		 * The found BAR is smaller than our new BAR. For that reason,
725 		 * insert our new BAR before the found BAR.
726 		 */
727 		TAILQ_INSERT_BEFORE(bar, new_bar, chain);
728 	}
729 
730 	/*
731 	 * pci_passthru devices synchronize their physical and virtual command
732 	 * register on init. For that reason, the virtual cmd reg should be
733 	 * updated as early as possible.
734 	 */
735 	uint16_t enbit = 0;
736 	switch (type) {
737 	case PCIBAR_IO:
738 		enbit = PCIM_CMD_PORTEN;
739 		break;
740 	case PCIBAR_MEM64:
741 	case PCIBAR_MEM32:
742 		enbit = PCIM_CMD_MEMEN;
743 		break;
744 	default:
745 		enbit = 0;
746 		break;
747 	}
748 
749 	const uint16_t cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
750 	pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
751 
752 	return (0);
753 }
754 
755 static int
756 pci_emul_assign_bar(struct pci_devinst *const pdi, const int idx,
757     const enum pcibar_type type, const uint64_t size)
758 {
759 	int error;
760 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
761 
762 	switch (type) {
763 	case PCIBAR_NONE:
764 		baseptr = NULL;
765 		addr = mask = lobits = 0;
766 		break;
767 	case PCIBAR_IO:
768 		baseptr = &pci_emul_iobase;
769 		limit = PCI_EMUL_IOLIMIT;
770 		mask = PCIM_BAR_IO_BASE;
771 		lobits = PCIM_BAR_IO_SPACE;
772 		break;
773 	case PCIBAR_MEM64:
774 		/*
775 		 * XXX
776 		 * Some drivers do not work well if the 64-bit BAR is allocated
777 		 * above 4GB. Allow for this by allocating small requests under
778 		 * 4GB unless then allocation size is larger than some arbitrary
779 		 * number (128MB currently).
780 		 */
781 		if (size > 128 * 1024 * 1024) {
782 			baseptr = &pci_emul_membase64;
783 			limit = pci_emul_memlim64;
784 			mask = PCIM_BAR_MEM_BASE;
785 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
786 				 PCIM_BAR_MEM_PREFETCH;
787 		} else {
788 			baseptr = &pci_emul_membase32;
789 			limit = PCI_EMUL_MEMLIMIT32;
790 			mask = PCIM_BAR_MEM_BASE;
791 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
792 		}
793 		break;
794 	case PCIBAR_MEM32:
795 		baseptr = &pci_emul_membase32;
796 		limit = PCI_EMUL_MEMLIMIT32;
797 		mask = PCIM_BAR_MEM_BASE;
798 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
799 		break;
800 	case PCIBAR_ROM:
801 		/* do not claim memory for ROM. OVMF will do it for us. */
802 		baseptr = NULL;
803 		limit = 0;
804 		mask = PCIM_BIOS_ADDR_MASK;
805 		lobits = 0;
806 		break;
807 	default:
808 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
809 		assert(0);
810 	}
811 
812 	if (baseptr != NULL) {
813 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
814 		if (error != 0)
815 			return (error);
816 	}
817 
818 	pdi->pi_bar[idx].type = type;
819 	pdi->pi_bar[idx].addr = addr;
820 	pdi->pi_bar[idx].size = size;
821 	/*
822 	 * passthru devices are using same lobits as physical device they set
823 	 * this property
824 	 */
825 	if (pdi->pi_bar[idx].lobits != 0) {
826 		lobits = pdi->pi_bar[idx].lobits;
827 	} else {
828 		pdi->pi_bar[idx].lobits = lobits;
829 	}
830 
831 	/* Initialize the BAR register in config space */
832 	bar = (addr & mask) | lobits;
833 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
834 
835 	if (type == PCIBAR_MEM64) {
836 		assert(idx + 1 <= PCI_BARMAX);
837 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
838 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
839 	}
840 
841 	if (type != PCIBAR_ROM) {
842 		register_bar(pdi, idx);
843 	}
844 
845 	return (0);
846 }
847 
848 int
849 pci_emul_alloc_rom(struct pci_devinst *const pdi, const uint64_t size,
850     void **const addr)
851 {
852 	/* allocate ROM space once on first call */
853 	if (pci_emul_rombase == 0) {
854 		pci_emul_rombase = vm_create_devmem(pdi->pi_vmctx, VM_PCIROM,
855 		    "pcirom", PCI_EMUL_ROMSIZE);
856 		if (pci_emul_rombase == MAP_FAILED) {
857 			warnx("%s: failed to create rom segment", __func__);
858 			return (-1);
859 		}
860 		pci_emul_romlim = pci_emul_rombase + PCI_EMUL_ROMSIZE;
861 		pci_emul_romoffset = 0;
862 	}
863 
864 	/* ROM size should be a power of 2 and greater than 2 KB */
865 	const uint64_t rom_size = MAX(1UL << flsl(size),
866 	    ~PCIM_BIOS_ADDR_MASK + 1);
867 
868 	/* check if ROM fits into ROM space */
869 	if (pci_emul_romoffset + rom_size > PCI_EMUL_ROMSIZE) {
870 		warnx("%s: no space left in rom segment:", __func__);
871 		warnx("%16lu bytes left",
872 		    PCI_EMUL_ROMSIZE - pci_emul_romoffset);
873 		warnx("%16lu bytes required by %d/%d/%d", rom_size, pdi->pi_bus,
874 		    pdi->pi_slot, pdi->pi_func);
875 		return (-1);
876 	}
877 
878 	/* allocate ROM BAR */
879 	const int error = pci_emul_alloc_bar(pdi, PCI_ROM_IDX, PCIBAR_ROM,
880 	    rom_size);
881 	if (error)
882 		return error;
883 
884 	/* return address */
885 	*addr = pci_emul_rombase + pci_emul_romoffset;
886 
887 	/* save offset into ROM Space */
888 	pdi->pi_romoffset = pci_emul_romoffset;
889 
890 	/* increase offset for next ROM */
891 	pci_emul_romoffset += rom_size;
892 
893 	return (0);
894 }
895 
896 #define	CAP_START_OFFSET	0x40
897 static int
898 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
899 {
900 	int i, capoff, reallen;
901 	uint16_t sts;
902 
903 	assert(caplen > 0);
904 
905 	reallen = roundup2(caplen, 4);		/* dword aligned */
906 
907 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
908 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
909 		capoff = CAP_START_OFFSET;
910 	else
911 		capoff = pi->pi_capend + 1;
912 
913 	/* Check if we have enough space */
914 	if (capoff + reallen > PCI_REGMAX + 1)
915 		return (-1);
916 
917 	/* Set the previous capability pointer */
918 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
919 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
920 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
921 	} else
922 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
923 
924 	/* Copy the capability */
925 	for (i = 0; i < caplen; i++)
926 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
927 
928 	/* Set the next capability pointer */
929 	pci_set_cfgdata8(pi, capoff + 1, 0);
930 
931 	pi->pi_prevcap = capoff;
932 	pi->pi_capend = capoff + reallen - 1;
933 	return (0);
934 }
935 
936 static struct pci_devemu *
937 pci_emul_finddev(const char *name)
938 {
939 	struct pci_devemu **pdpp, *pdp;
940 
941 	SET_FOREACH(pdpp, pci_devemu_set) {
942 		pdp = *pdpp;
943 		if (!strcmp(pdp->pe_emu, name)) {
944 			return (pdp);
945 		}
946 	}
947 
948 	return (NULL);
949 }
950 
951 static int
952 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
953     int func, struct funcinfo *fi)
954 {
955 	struct pci_devinst *pdi;
956 	int err;
957 
958 	pdi = calloc(1, sizeof(struct pci_devinst));
959 
960 	pdi->pi_vmctx = ctx;
961 	pdi->pi_bus = bus;
962 	pdi->pi_slot = slot;
963 	pdi->pi_func = func;
964 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
965 	pdi->pi_lintr.pin = 0;
966 	pdi->pi_lintr.state = IDLE;
967 	pdi->pi_lintr.pirq_pin = 0;
968 	pdi->pi_lintr.ioapic_irq = 0;
969 	pdi->pi_d = pde;
970 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
971 
972 	/* Disable legacy interrupts */
973 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
974 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
975 
976 	pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
977 
978 	err = (*pde->pe_init)(ctx, pdi, fi->fi_config);
979 	if (err == 0)
980 		fi->fi_devi = pdi;
981 	else
982 		free(pdi);
983 
984 	return (err);
985 }
986 
987 void
988 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
989 {
990 	int mmc;
991 
992 	/* Number of msi messages must be a power of 2 between 1 and 32 */
993 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
994 	mmc = ffs(msgnum) - 1;
995 
996 	bzero(msicap, sizeof(struct msicap));
997 	msicap->capid = PCIY_MSI;
998 	msicap->nextptr = nextptr;
999 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
1000 }
1001 
1002 int
1003 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
1004 {
1005 	struct msicap msicap;
1006 
1007 	pci_populate_msicap(&msicap, msgnum, 0);
1008 
1009 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
1010 }
1011 
1012 static void
1013 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
1014 		     uint32_t msix_tab_size)
1015 {
1016 
1017 	assert(msix_tab_size % 4096 == 0);
1018 
1019 	bzero(msixcap, sizeof(struct msixcap));
1020 	msixcap->capid = PCIY_MSIX;
1021 
1022 	/*
1023 	 * Message Control Register, all fields set to
1024 	 * zero except for the Table Size.
1025 	 * Note: Table size N is encoded as N-1
1026 	 */
1027 	msixcap->msgctrl = msgnum - 1;
1028 
1029 	/*
1030 	 * MSI-X BAR setup:
1031 	 * - MSI-X table start at offset 0
1032 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
1033 	 */
1034 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
1035 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
1036 }
1037 
1038 static void
1039 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
1040 {
1041 	int i, table_size;
1042 
1043 	assert(table_entries > 0);
1044 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
1045 
1046 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
1047 	pi->pi_msix.table = calloc(1, table_size);
1048 
1049 	/* set mask bit of vector control register */
1050 	for (i = 0; i < table_entries; i++)
1051 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
1052 }
1053 
1054 int
1055 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
1056 {
1057 	uint32_t tab_size;
1058 	struct msixcap msixcap;
1059 
1060 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
1061 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
1062 
1063 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
1064 
1065 	/* Align table size to nearest 4K */
1066 	tab_size = roundup2(tab_size, 4096);
1067 
1068 	pi->pi_msix.table_bar = barnum;
1069 	pi->pi_msix.pba_bar   = barnum;
1070 	pi->pi_msix.table_offset = 0;
1071 	pi->pi_msix.table_count = msgnum;
1072 	pi->pi_msix.pba_offset = tab_size;
1073 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
1074 
1075 	pci_msix_table_init(pi, msgnum);
1076 
1077 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
1078 
1079 	/* allocate memory for MSI-X Table and PBA */
1080 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
1081 				tab_size + pi->pi_msix.pba_size);
1082 
1083 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
1084 					sizeof(msixcap)));
1085 }
1086 
1087 static void
1088 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1089 		 int bytes, uint32_t val)
1090 {
1091 	uint16_t msgctrl, rwmask;
1092 	int off;
1093 
1094 	off = offset - capoff;
1095 	/* Message Control Register */
1096 	if (off == 2 && bytes == 2) {
1097 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
1098 		msgctrl = pci_get_cfgdata16(pi, offset);
1099 		msgctrl &= ~rwmask;
1100 		msgctrl |= val & rwmask;
1101 		val = msgctrl;
1102 
1103 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
1104 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
1105 		pci_lintr_update(pi);
1106 	}
1107 
1108 	CFGWRITE(pi, offset, val, bytes);
1109 }
1110 
1111 static void
1112 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1113 		int bytes, uint32_t val)
1114 {
1115 	uint16_t msgctrl, rwmask, msgdata, mme;
1116 	uint32_t addrlo;
1117 
1118 	/*
1119 	 * If guest is writing to the message control register make sure
1120 	 * we do not overwrite read-only fields.
1121 	 */
1122 	if ((offset - capoff) == 2 && bytes == 2) {
1123 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
1124 		msgctrl = pci_get_cfgdata16(pi, offset);
1125 		msgctrl &= ~rwmask;
1126 		msgctrl |= val & rwmask;
1127 		val = msgctrl;
1128 	}
1129 	CFGWRITE(pi, offset, val, bytes);
1130 
1131 	msgctrl = pci_get_cfgdata16(pi, capoff + 2);
1132 	addrlo = pci_get_cfgdata32(pi, capoff + 4);
1133 	if (msgctrl & PCIM_MSICTRL_64BIT)
1134 		msgdata = pci_get_cfgdata16(pi, capoff + 12);
1135 	else
1136 		msgdata = pci_get_cfgdata16(pi, capoff + 8);
1137 
1138 	mme = msgctrl & PCIM_MSICTRL_MME_MASK;
1139 	pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
1140 	if (pi->pi_msi.enabled) {
1141 		pi->pi_msi.addr = addrlo;
1142 		pi->pi_msi.msg_data = msgdata;
1143 		pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
1144 	} else {
1145 		pi->pi_msi.maxmsgnum = 0;
1146 	}
1147 	pci_lintr_update(pi);
1148 }
1149 
1150 void
1151 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1152 		 int bytes, uint32_t val)
1153 {
1154 
1155 	/* XXX don't write to the readonly parts */
1156 	CFGWRITE(pi, offset, val, bytes);
1157 }
1158 
1159 #define	PCIECAP_VERSION	0x2
1160 int
1161 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
1162 {
1163 	int err;
1164 	struct pciecap pciecap;
1165 
1166 	bzero(&pciecap, sizeof(pciecap));
1167 
1168 	/*
1169 	 * Use the integrated endpoint type for endpoints on a root complex bus.
1170 	 *
1171 	 * NB: bhyve currently only supports a single PCI bus that is the root
1172 	 * complex bus, so all endpoints are integrated.
1173 	 */
1174 	if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
1175 		type = PCIEM_TYPE_ROOT_INT_EP;
1176 
1177 	pciecap.capid = PCIY_EXPRESS;
1178 	pciecap.pcie_capabilities = PCIECAP_VERSION | type;
1179 	if (type != PCIEM_TYPE_ROOT_INT_EP) {
1180 		pciecap.link_capabilities = 0x411;	/* gen1, x1 */
1181 		pciecap.link_status = 0x11;		/* gen1, x1 */
1182 	}
1183 
1184 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
1185 	return (err);
1186 }
1187 
1188 /*
1189  * This function assumes that 'coff' is in the capabilities region of the
1190  * config space. A capoff parameter of zero will force a search for the
1191  * offset and type.
1192  */
1193 void
1194 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val,
1195     uint8_t capoff, int capid)
1196 {
1197 	uint8_t nextoff;
1198 
1199 	/* Do not allow un-aligned writes */
1200 	if ((offset & (bytes - 1)) != 0)
1201 		return;
1202 
1203 	if (capoff == 0) {
1204 		/* Find the capability that we want to update */
1205 		capoff = CAP_START_OFFSET;
1206 		while (1) {
1207 			nextoff = pci_get_cfgdata8(pi, capoff + 1);
1208 			if (nextoff == 0)
1209 				break;
1210 			if (offset >= capoff && offset < nextoff)
1211 				break;
1212 
1213 			capoff = nextoff;
1214 		}
1215 		assert(offset >= capoff);
1216 		capid = pci_get_cfgdata8(pi, capoff);
1217 	}
1218 
1219 	/*
1220 	 * Capability ID and Next Capability Pointer are readonly.
1221 	 * However, some o/s's do 4-byte writes that include these.
1222 	 * For this case, trim the write back to 2 bytes and adjust
1223 	 * the data.
1224 	 */
1225 	if (offset == capoff || offset == capoff + 1) {
1226 		if (offset == capoff && bytes == 4) {
1227 			bytes = 2;
1228 			offset += 2;
1229 			val >>= 16;
1230 		} else
1231 			return;
1232 	}
1233 
1234 	switch (capid) {
1235 	case PCIY_MSI:
1236 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1237 		break;
1238 	case PCIY_MSIX:
1239 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1240 		break;
1241 	case PCIY_EXPRESS:
1242 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1243 		break;
1244 	default:
1245 		break;
1246 	}
1247 }
1248 
1249 static int
1250 pci_emul_iscap(struct pci_devinst *pi, int offset)
1251 {
1252 	uint16_t sts;
1253 
1254 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1255 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1256 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1257 			return (1);
1258 	}
1259 	return (0);
1260 }
1261 
1262 static int
1263 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1264 			  int size, uint64_t *val, void *arg1, long arg2)
1265 {
1266 	/*
1267 	 * Ignore writes; return 0xff's for reads. The mem read code
1268 	 * will take care of truncating to the correct size.
1269 	 */
1270 	if (dir == MEM_F_READ) {
1271 		*val = 0xffffffffffffffff;
1272 	}
1273 
1274 	return (0);
1275 }
1276 
1277 static int
1278 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1279     int bytes, uint64_t *val, void *arg1, long arg2)
1280 {
1281 	int bus, slot, func, coff, in;
1282 
1283 	coff = addr & 0xfff;
1284 	func = (addr >> 12) & 0x7;
1285 	slot = (addr >> 15) & 0x1f;
1286 	bus = (addr >> 20) & 0xff;
1287 	in = (dir == MEM_F_READ);
1288 	if (in)
1289 		*val = ~0UL;
1290 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1291 	return (0);
1292 }
1293 
1294 uint64_t
1295 pci_ecfg_base(void)
1296 {
1297 
1298 	return (PCI_EMUL_ECFG_BASE);
1299 }
1300 
1301 #define	BUSIO_ROUNDUP		32
1302 #define	BUSMEM32_ROUNDUP	(1024 * 1024)
1303 #define	BUSMEM64_ROUNDUP	(512 * 1024 * 1024)
1304 
1305 int
1306 init_pci(struct vmctx *ctx)
1307 {
1308 	char node_name[sizeof("pci.XXX.XX.X")];
1309 	struct mem_range mr;
1310 	struct pci_devemu *pde;
1311 	struct businfo *bi;
1312 	struct slotinfo *si;
1313 	struct funcinfo *fi;
1314 	nvlist_t *nvl;
1315 	const char *emul;
1316 	size_t lowmem;
1317 	int bus, slot, func;
1318 	int error;
1319 
1320 	if (vm_get_lowmem_limit(ctx) > PCI_EMUL_MEMBASE32)
1321 		errx(EX_OSERR, "Invalid lowmem limit");
1322 
1323 	pci_emul_iobase = PCI_EMUL_IOBASE;
1324 	pci_emul_membase32 = PCI_EMUL_MEMBASE32;
1325 
1326 	pci_emul_membase64 = 4*GB + vm_get_highmem_size(ctx);
1327 	pci_emul_membase64 = roundup2(pci_emul_membase64, PCI_EMUL_MEMSIZE64);
1328 	pci_emul_memlim64 = pci_emul_membase64 + PCI_EMUL_MEMSIZE64;
1329 
1330 	for (bus = 0; bus < MAXBUSES; bus++) {
1331 		snprintf(node_name, sizeof(node_name), "pci.%d", bus);
1332 		nvl = find_config_node(node_name);
1333 		if (nvl == NULL)
1334 			continue;
1335 		pci_businfo[bus] = calloc(1, sizeof(struct businfo));
1336 		bi = pci_businfo[bus];
1337 
1338 		/*
1339 		 * Keep track of the i/o and memory resources allocated to
1340 		 * this bus.
1341 		 */
1342 		bi->iobase = pci_emul_iobase;
1343 		bi->membase32 = pci_emul_membase32;
1344 		bi->membase64 = pci_emul_membase64;
1345 
1346 		/* first run: init devices */
1347 		for (slot = 0; slot < MAXSLOTS; slot++) {
1348 			si = &bi->slotinfo[slot];
1349 			for (func = 0; func < MAXFUNCS; func++) {
1350 				fi = &si->si_funcs[func];
1351 				snprintf(node_name, sizeof(node_name),
1352 				    "pci.%d.%d.%d", bus, slot, func);
1353 				nvl = find_config_node(node_name);
1354 				if (nvl == NULL)
1355 					continue;
1356 
1357 				fi->fi_config = nvl;
1358 				emul = get_config_value_node(nvl, "device");
1359 				if (emul == NULL) {
1360 					EPRINTLN("pci slot %d:%d:%d: missing "
1361 					    "\"device\" value", bus, slot, func);
1362 					return (EINVAL);
1363 				}
1364 				pde = pci_emul_finddev(emul);
1365 				if (pde == NULL) {
1366 					EPRINTLN("pci slot %d:%d:%d: unknown "
1367 					    "device \"%s\"", bus, slot, func,
1368 					    emul);
1369 					return (EINVAL);
1370 				}
1371 				if (pde->pe_alias != NULL) {
1372 					EPRINTLN("pci slot %d:%d:%d: legacy "
1373 					    "device \"%s\", use \"%s\" instead",
1374 					    bus, slot, func, emul,
1375 					    pde->pe_alias);
1376 					return (EINVAL);
1377 				}
1378 				fi->fi_pde = pde;
1379 				error = pci_emul_init(ctx, pde, bus, slot,
1380 				    func, fi);
1381 				if (error)
1382 					return (error);
1383 			}
1384 		}
1385 
1386 		/* second run: assign BARs and free list */
1387 		struct pci_bar_allocation *bar;
1388 		struct pci_bar_allocation *bar_tmp;
1389 		TAILQ_FOREACH_SAFE(bar, &pci_bars, chain, bar_tmp) {
1390 			pci_emul_assign_bar(bar->pdi, bar->idx, bar->type,
1391 			    bar->size);
1392 			free(bar);
1393 		}
1394 		TAILQ_INIT(&pci_bars);
1395 
1396 		/*
1397 		 * Add some slop to the I/O and memory resources decoded by
1398 		 * this bus to give a guest some flexibility if it wants to
1399 		 * reprogram the BARs.
1400 		 */
1401 		pci_emul_iobase += BUSIO_ROUNDUP;
1402 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1403 		bi->iolimit = pci_emul_iobase;
1404 
1405 		pci_emul_membase32 += BUSMEM32_ROUNDUP;
1406 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1407 		    BUSMEM32_ROUNDUP);
1408 		bi->memlimit32 = pci_emul_membase32;
1409 
1410 		pci_emul_membase64 += BUSMEM64_ROUNDUP;
1411 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1412 		    BUSMEM64_ROUNDUP);
1413 		bi->memlimit64 = pci_emul_membase64;
1414 	}
1415 
1416 	/*
1417 	 * PCI backends are initialized before routing INTx interrupts
1418 	 * so that LPC devices are able to reserve ISA IRQs before
1419 	 * routing PIRQ pins.
1420 	 */
1421 	for (bus = 0; bus < MAXBUSES; bus++) {
1422 		if ((bi = pci_businfo[bus]) == NULL)
1423 			continue;
1424 
1425 		for (slot = 0; slot < MAXSLOTS; slot++) {
1426 			si = &bi->slotinfo[slot];
1427 			for (func = 0; func < MAXFUNCS; func++) {
1428 				fi = &si->si_funcs[func];
1429 				if (fi->fi_devi == NULL)
1430 					continue;
1431 				pci_lintr_route(fi->fi_devi);
1432 			}
1433 		}
1434 	}
1435 	lpc_pirq_routed();
1436 
1437 	/*
1438 	 * The guest physical memory map looks like the following:
1439 	 * [0,		    lowmem)		guest system memory
1440 	 * [lowmem,	    0xC0000000)		memory hole (may be absent)
1441 	 * [0xC0000000,     0xE0000000)		PCI hole (32-bit BAR allocation)
1442 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1443 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1444 	 * [4GB,	    4GB + highmem)
1445 	 */
1446 
1447 	/*
1448 	 * Accesses to memory addresses that are not allocated to system
1449 	 * memory or PCI devices return 0xff's.
1450 	 */
1451 	lowmem = vm_get_lowmem_size(ctx);
1452 	bzero(&mr, sizeof(struct mem_range));
1453 	mr.name = "PCI hole";
1454 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1455 	mr.base = lowmem;
1456 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1457 	mr.handler = pci_emul_fallback_handler;
1458 	error = register_mem_fallback(&mr);
1459 	assert(error == 0);
1460 
1461 	/* PCI extended config space */
1462 	bzero(&mr, sizeof(struct mem_range));
1463 	mr.name = "PCI ECFG";
1464 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1465 	mr.base = PCI_EMUL_ECFG_BASE;
1466 	mr.size = PCI_EMUL_ECFG_SIZE;
1467 	mr.handler = pci_emul_ecfg_handler;
1468 	error = register_mem(&mr);
1469 	assert(error == 0);
1470 
1471 	return (0);
1472 }
1473 
1474 static void
1475 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1476     void *arg)
1477 {
1478 
1479 	dsdt_line("  Package ()");
1480 	dsdt_line("  {");
1481 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1482 	dsdt_line("    0x%02X,", pin - 1);
1483 	dsdt_line("    Zero,");
1484 	dsdt_line("    0x%X", ioapic_irq);
1485 	dsdt_line("  },");
1486 }
1487 
1488 static void
1489 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1490     void *arg)
1491 {
1492 	char *name;
1493 
1494 	name = lpc_pirq_name(pirq_pin);
1495 	if (name == NULL)
1496 		return;
1497 	dsdt_line("  Package ()");
1498 	dsdt_line("  {");
1499 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1500 	dsdt_line("    0x%02X,", pin - 1);
1501 	dsdt_line("    %s,", name);
1502 	dsdt_line("    0x00");
1503 	dsdt_line("  },");
1504 	free(name);
1505 }
1506 
1507 /*
1508  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1509  * corresponding to each PCI bus.
1510  */
1511 static void
1512 pci_bus_write_dsdt(int bus)
1513 {
1514 	struct businfo *bi;
1515 	struct slotinfo *si;
1516 	struct pci_devinst *pi;
1517 	int count, func, slot;
1518 
1519 	/*
1520 	 * If there are no devices on this 'bus' then just return.
1521 	 */
1522 	if ((bi = pci_businfo[bus]) == NULL) {
1523 		/*
1524 		 * Bus 0 is special because it decodes the I/O ports used
1525 		 * for PCI config space access even if there are no devices
1526 		 * on it.
1527 		 */
1528 		if (bus != 0)
1529 			return;
1530 	}
1531 
1532 	dsdt_line("  Device (PC%02X)", bus);
1533 	dsdt_line("  {");
1534 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1535 
1536 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1537 	dsdt_line("    {");
1538 	dsdt_line("        Return (0x%08X)", bus);
1539 	dsdt_line("    }");
1540 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1541 	dsdt_line("    {");
1542 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1543 	    "MaxFixed, PosDecode,");
1544 	dsdt_line("        0x0000,             // Granularity");
1545 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1546 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1547 	dsdt_line("        0x0000,             // Translation Offset");
1548 	dsdt_line("        0x0001,             // Length");
1549 	dsdt_line("        ,, )");
1550 
1551 	if (bus == 0) {
1552 		dsdt_indent(3);
1553 		dsdt_fixed_ioport(0xCF8, 8);
1554 		dsdt_unindent(3);
1555 
1556 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1557 		    "PosDecode, EntireRange,");
1558 		dsdt_line("        0x0000,             // Granularity");
1559 		dsdt_line("        0x0000,             // Range Minimum");
1560 		dsdt_line("        0x0CF7,             // Range Maximum");
1561 		dsdt_line("        0x0000,             // Translation Offset");
1562 		dsdt_line("        0x0CF8,             // Length");
1563 		dsdt_line("        ,, , TypeStatic)");
1564 
1565 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1566 		    "PosDecode, EntireRange,");
1567 		dsdt_line("        0x0000,             // Granularity");
1568 		dsdt_line("        0x0D00,             // Range Minimum");
1569 		dsdt_line("        0x%04X,             // Range Maximum",
1570 		    PCI_EMUL_IOBASE - 1);
1571 		dsdt_line("        0x0000,             // Translation Offset");
1572 		dsdt_line("        0x%04X,             // Length",
1573 		    PCI_EMUL_IOBASE - 0x0D00);
1574 		dsdt_line("        ,, , TypeStatic)");
1575 
1576 		if (bi == NULL) {
1577 			dsdt_line("    })");
1578 			goto done;
1579 		}
1580 	}
1581 	assert(bi != NULL);
1582 
1583 	/* i/o window */
1584 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1585 	    "PosDecode, EntireRange,");
1586 	dsdt_line("        0x0000,             // Granularity");
1587 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1588 	dsdt_line("        0x%04X,             // Range Maximum",
1589 	    bi->iolimit - 1);
1590 	dsdt_line("        0x0000,             // Translation Offset");
1591 	dsdt_line("        0x%04X,             // Length",
1592 	    bi->iolimit - bi->iobase);
1593 	dsdt_line("        ,, , TypeStatic)");
1594 
1595 	/* mmio window (32-bit) */
1596 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1597 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1598 	dsdt_line("        0x00000000,         // Granularity");
1599 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1600 	dsdt_line("        0x%08X,         // Range Maximum\n",
1601 	    bi->memlimit32 - 1);
1602 	dsdt_line("        0x00000000,         // Translation Offset");
1603 	dsdt_line("        0x%08X,         // Length\n",
1604 	    bi->memlimit32 - bi->membase32);
1605 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1606 
1607 	/* mmio window (64-bit) */
1608 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1609 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1610 	dsdt_line("        0x0000000000000000, // Granularity");
1611 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1612 	dsdt_line("        0x%016lX, // Range Maximum\n",
1613 	    bi->memlimit64 - 1);
1614 	dsdt_line("        0x0000000000000000, // Translation Offset");
1615 	dsdt_line("        0x%016lX, // Length\n",
1616 	    bi->memlimit64 - bi->membase64);
1617 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1618 	dsdt_line("    })");
1619 
1620 	count = pci_count_lintr(bus);
1621 	if (count != 0) {
1622 		dsdt_indent(2);
1623 		dsdt_line("Name (PPRT, Package ()");
1624 		dsdt_line("{");
1625 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1626 		dsdt_line("})");
1627 		dsdt_line("Name (APRT, Package ()");
1628 		dsdt_line("{");
1629 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1630 		dsdt_line("})");
1631 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1632 		dsdt_line("{");
1633 		dsdt_line("  If (PICM)");
1634 		dsdt_line("  {");
1635 		dsdt_line("    Return (APRT)");
1636 		dsdt_line("  }");
1637 		dsdt_line("  Else");
1638 		dsdt_line("  {");
1639 		dsdt_line("    Return (PPRT)");
1640 		dsdt_line("  }");
1641 		dsdt_line("}");
1642 		dsdt_unindent(2);
1643 	}
1644 
1645 	dsdt_indent(2);
1646 	for (slot = 0; slot < MAXSLOTS; slot++) {
1647 		si = &bi->slotinfo[slot];
1648 		for (func = 0; func < MAXFUNCS; func++) {
1649 			pi = si->si_funcs[func].fi_devi;
1650 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1651 				pi->pi_d->pe_write_dsdt(pi);
1652 		}
1653 	}
1654 	dsdt_unindent(2);
1655 done:
1656 	dsdt_line("  }");
1657 }
1658 
1659 void
1660 pci_write_dsdt(void)
1661 {
1662 	int bus;
1663 
1664 	dsdt_indent(1);
1665 	dsdt_line("Name (PICM, 0x00)");
1666 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1667 	dsdt_line("{");
1668 	dsdt_line("  Store (Arg0, PICM)");
1669 	dsdt_line("}");
1670 	dsdt_line("");
1671 	dsdt_line("Scope (_SB)");
1672 	dsdt_line("{");
1673 	for (bus = 0; bus < MAXBUSES; bus++)
1674 		pci_bus_write_dsdt(bus);
1675 	dsdt_line("}");
1676 	dsdt_unindent(1);
1677 }
1678 
1679 int
1680 pci_bus_configured(int bus)
1681 {
1682 	assert(bus >= 0 && bus < MAXBUSES);
1683 	return (pci_businfo[bus] != NULL);
1684 }
1685 
1686 int
1687 pci_msi_enabled(struct pci_devinst *pi)
1688 {
1689 	return (pi->pi_msi.enabled);
1690 }
1691 
1692 int
1693 pci_msi_maxmsgnum(struct pci_devinst *pi)
1694 {
1695 	if (pi->pi_msi.enabled)
1696 		return (pi->pi_msi.maxmsgnum);
1697 	else
1698 		return (0);
1699 }
1700 
1701 int
1702 pci_msix_enabled(struct pci_devinst *pi)
1703 {
1704 
1705 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1706 }
1707 
1708 void
1709 pci_generate_msix(struct pci_devinst *pi, int index)
1710 {
1711 	struct msix_table_entry *mte;
1712 
1713 	if (!pci_msix_enabled(pi))
1714 		return;
1715 
1716 	if (pi->pi_msix.function_mask)
1717 		return;
1718 
1719 	if (index >= pi->pi_msix.table_count)
1720 		return;
1721 
1722 	mte = &pi->pi_msix.table[index];
1723 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1724 		/* XXX Set PBA bit if interrupt is disabled */
1725 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1726 	}
1727 }
1728 
1729 void
1730 pci_generate_msi(struct pci_devinst *pi, int index)
1731 {
1732 
1733 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1734 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1735 			     pi->pi_msi.msg_data + index);
1736 	}
1737 }
1738 
1739 static bool
1740 pci_lintr_permitted(struct pci_devinst *pi)
1741 {
1742 	uint16_t cmd;
1743 
1744 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1745 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1746 		(cmd & PCIM_CMD_INTxDIS)));
1747 }
1748 
1749 void
1750 pci_lintr_request(struct pci_devinst *pi)
1751 {
1752 	struct businfo *bi;
1753 	struct slotinfo *si;
1754 	int bestpin, bestcount, pin;
1755 
1756 	bi = pci_businfo[pi->pi_bus];
1757 	assert(bi != NULL);
1758 
1759 	/*
1760 	 * Just allocate a pin from our slot.  The pin will be
1761 	 * assigned IRQs later when interrupts are routed.
1762 	 */
1763 	si = &bi->slotinfo[pi->pi_slot];
1764 	bestpin = 0;
1765 	bestcount = si->si_intpins[0].ii_count;
1766 	for (pin = 1; pin < 4; pin++) {
1767 		if (si->si_intpins[pin].ii_count < bestcount) {
1768 			bestpin = pin;
1769 			bestcount = si->si_intpins[pin].ii_count;
1770 		}
1771 	}
1772 
1773 	si->si_intpins[bestpin].ii_count++;
1774 	pi->pi_lintr.pin = bestpin + 1;
1775 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1776 }
1777 
1778 static void
1779 pci_lintr_route(struct pci_devinst *pi)
1780 {
1781 	struct businfo *bi;
1782 	struct intxinfo *ii;
1783 
1784 	if (pi->pi_lintr.pin == 0)
1785 		return;
1786 
1787 	bi = pci_businfo[pi->pi_bus];
1788 	assert(bi != NULL);
1789 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1790 
1791 	/*
1792 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1793 	 * is not yet assigned.
1794 	 */
1795 	if (ii->ii_ioapic_irq == 0)
1796 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1797 	assert(ii->ii_ioapic_irq > 0);
1798 
1799 	/*
1800 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1801 	 * not yet assigned.
1802 	 */
1803 	if (ii->ii_pirq_pin == 0)
1804 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1805 	assert(ii->ii_pirq_pin > 0);
1806 
1807 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1808 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1809 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1810 }
1811 
1812 void
1813 pci_lintr_assert(struct pci_devinst *pi)
1814 {
1815 
1816 	assert(pi->pi_lintr.pin > 0);
1817 
1818 	pthread_mutex_lock(&pi->pi_lintr.lock);
1819 	if (pi->pi_lintr.state == IDLE) {
1820 		if (pci_lintr_permitted(pi)) {
1821 			pi->pi_lintr.state = ASSERTED;
1822 			pci_irq_assert(pi);
1823 		} else
1824 			pi->pi_lintr.state = PENDING;
1825 	}
1826 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1827 }
1828 
1829 void
1830 pci_lintr_deassert(struct pci_devinst *pi)
1831 {
1832 
1833 	assert(pi->pi_lintr.pin > 0);
1834 
1835 	pthread_mutex_lock(&pi->pi_lintr.lock);
1836 	if (pi->pi_lintr.state == ASSERTED) {
1837 		pi->pi_lintr.state = IDLE;
1838 		pci_irq_deassert(pi);
1839 	} else if (pi->pi_lintr.state == PENDING)
1840 		pi->pi_lintr.state = IDLE;
1841 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1842 }
1843 
1844 static void
1845 pci_lintr_update(struct pci_devinst *pi)
1846 {
1847 
1848 	pthread_mutex_lock(&pi->pi_lintr.lock);
1849 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1850 		pci_irq_deassert(pi);
1851 		pi->pi_lintr.state = PENDING;
1852 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1853 		pi->pi_lintr.state = ASSERTED;
1854 		pci_irq_assert(pi);
1855 	}
1856 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1857 }
1858 
1859 int
1860 pci_count_lintr(int bus)
1861 {
1862 	int count, slot, pin;
1863 	struct slotinfo *slotinfo;
1864 
1865 	count = 0;
1866 	if (pci_businfo[bus] != NULL) {
1867 		for (slot = 0; slot < MAXSLOTS; slot++) {
1868 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1869 			for (pin = 0; pin < 4; pin++) {
1870 				if (slotinfo->si_intpins[pin].ii_count != 0)
1871 					count++;
1872 			}
1873 		}
1874 	}
1875 	return (count);
1876 }
1877 
1878 void
1879 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1880 {
1881 	struct businfo *bi;
1882 	struct slotinfo *si;
1883 	struct intxinfo *ii;
1884 	int slot, pin;
1885 
1886 	if ((bi = pci_businfo[bus]) == NULL)
1887 		return;
1888 
1889 	for (slot = 0; slot < MAXSLOTS; slot++) {
1890 		si = &bi->slotinfo[slot];
1891 		for (pin = 0; pin < 4; pin++) {
1892 			ii = &si->si_intpins[pin];
1893 			if (ii->ii_count != 0)
1894 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1895 				    ii->ii_ioapic_irq, arg);
1896 		}
1897 	}
1898 }
1899 
1900 /*
1901  * Return 1 if the emulated device in 'slot' is a multi-function device.
1902  * Return 0 otherwise.
1903  */
1904 static int
1905 pci_emul_is_mfdev(int bus, int slot)
1906 {
1907 	struct businfo *bi;
1908 	struct slotinfo *si;
1909 	int f, numfuncs;
1910 
1911 	numfuncs = 0;
1912 	if ((bi = pci_businfo[bus]) != NULL) {
1913 		si = &bi->slotinfo[slot];
1914 		for (f = 0; f < MAXFUNCS; f++) {
1915 			if (si->si_funcs[f].fi_devi != NULL) {
1916 				numfuncs++;
1917 			}
1918 		}
1919 	}
1920 	return (numfuncs > 1);
1921 }
1922 
1923 /*
1924  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1925  * whether or not is a multi-function being emulated in the pci 'slot'.
1926  */
1927 static void
1928 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1929 {
1930 	int mfdev;
1931 
1932 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1933 		mfdev = pci_emul_is_mfdev(bus, slot);
1934 		switch (bytes) {
1935 		case 1:
1936 		case 2:
1937 			*rv &= ~PCIM_MFDEV;
1938 			if (mfdev) {
1939 				*rv |= PCIM_MFDEV;
1940 			}
1941 			break;
1942 		case 4:
1943 			*rv &= ~(PCIM_MFDEV << 16);
1944 			if (mfdev) {
1945 				*rv |= (PCIM_MFDEV << 16);
1946 			}
1947 			break;
1948 		}
1949 	}
1950 }
1951 
1952 /*
1953  * Update device state in response to changes to the PCI command
1954  * register.
1955  */
1956 void
1957 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
1958 {
1959 	int i;
1960 	uint16_t changed, new;
1961 
1962 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);
1963 	changed = old ^ new;
1964 
1965 	/*
1966 	 * If the MMIO or I/O address space decoding has changed then
1967 	 * register/unregister all BARs that decode that address space.
1968 	 */
1969 	for (i = 0; i <= PCI_BARMAX_WITH_ROM; i++) {
1970 		switch (pi->pi_bar[i].type) {
1971 			case PCIBAR_NONE:
1972 			case PCIBAR_MEMHI64:
1973 				break;
1974 			case PCIBAR_IO:
1975 				/* I/O address space decoding changed? */
1976 				if (changed & PCIM_CMD_PORTEN) {
1977 					if (new & PCIM_CMD_PORTEN)
1978 						register_bar(pi, i);
1979 					else
1980 						unregister_bar(pi, i);
1981 				}
1982 				break;
1983 			case PCIBAR_ROM:
1984 				/* skip (un-)register of ROM if it disabled */
1985 				if (!romen(pi))
1986 					break;
1987 				/* fallthrough */
1988 			case PCIBAR_MEM32:
1989 			case PCIBAR_MEM64:
1990 				/* MMIO address space decoding changed? */
1991 				if (changed & PCIM_CMD_MEMEN) {
1992 					if (new & PCIM_CMD_MEMEN)
1993 						register_bar(pi, i);
1994 					else
1995 						unregister_bar(pi, i);
1996 				}
1997 				break;
1998 			default:
1999 				assert(0);
2000 		}
2001 	}
2002 
2003 	/*
2004 	 * If INTx has been unmasked and is pending, assert the
2005 	 * interrupt.
2006 	 */
2007 	pci_lintr_update(pi);
2008 }
2009 
2010 static void
2011 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
2012 {
2013 	int rshift;
2014 	uint32_t cmd, old, readonly;
2015 
2016 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
2017 
2018 	/*
2019 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
2020 	 *
2021 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
2022 	 * 'write 1 to clear'. However these bits are not set to '1' by
2023 	 * any device emulation so it is simpler to treat them as readonly.
2024 	 */
2025 	rshift = (coff & 0x3) * 8;
2026 	readonly = 0xFFFFF880 >> rshift;
2027 
2028 	old = CFGREAD(pi, coff, bytes);
2029 	new &= ~readonly;
2030 	new |= (old & readonly);
2031 	CFGWRITE(pi, coff, new, bytes);			/* update config */
2032 
2033 	pci_emul_cmd_changed(pi, cmd);
2034 }
2035 
2036 static void
2037 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
2038     int coff, int bytes, uint32_t *eax)
2039 {
2040 	struct businfo *bi;
2041 	struct slotinfo *si;
2042 	struct pci_devinst *pi;
2043 	struct pci_devemu *pe;
2044 	int idx, needcfg;
2045 	uint64_t addr, bar, mask;
2046 
2047 	if ((bi = pci_businfo[bus]) != NULL) {
2048 		si = &bi->slotinfo[slot];
2049 		pi = si->si_funcs[func].fi_devi;
2050 	} else
2051 		pi = NULL;
2052 
2053 	/*
2054 	 * Just return if there is no device at this slot:func or if the
2055 	 * the guest is doing an un-aligned access.
2056 	 */
2057 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
2058 	    (coff & (bytes - 1)) != 0) {
2059 		if (in)
2060 			*eax = 0xffffffff;
2061 		return;
2062 	}
2063 
2064 	/*
2065 	 * Ignore all writes beyond the standard config space and return all
2066 	 * ones on reads.
2067 	 */
2068 	if (coff >= PCI_REGMAX + 1) {
2069 		if (in) {
2070 			*eax = 0xffffffff;
2071 			/*
2072 			 * Extended capabilities begin at offset 256 in config
2073 			 * space. Absence of extended capabilities is signaled
2074 			 * with all 0s in the extended capability header at
2075 			 * offset 256.
2076 			 */
2077 			if (coff <= PCI_REGMAX + 4)
2078 				*eax = 0x00000000;
2079 		}
2080 		return;
2081 	}
2082 
2083 	pe = pi->pi_d;
2084 
2085 	/*
2086 	 * Config read
2087 	 */
2088 	if (in) {
2089 		/* Let the device emulation override the default handler */
2090 		if (pe->pe_cfgread != NULL) {
2091 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
2092 			    eax);
2093 		} else {
2094 			needcfg = 1;
2095 		}
2096 
2097 		if (needcfg)
2098 			*eax = CFGREAD(pi, coff, bytes);
2099 
2100 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
2101 	} else {
2102 		/* Let the device emulation override the default handler */
2103 		if (pe->pe_cfgwrite != NULL &&
2104 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
2105 			return;
2106 
2107 		/*
2108 		 * Special handling for write to BAR and ROM registers
2109 		 */
2110 		if ((coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) ||
2111 		    (coff >= PCIR_BIOS && coff < PCIR_BIOS + 4)) {
2112 			/*
2113 			 * Ignore writes to BAR registers that are not
2114 			 * 4-byte aligned.
2115 			 */
2116 			if (bytes != 4 || (coff & 0x3) != 0)
2117 				return;
2118 			if (coff != PCIR_BIOS) {
2119 				idx = (coff - PCIR_BAR(0)) / 4;
2120 			} else {
2121 				idx = PCI_ROM_IDX;
2122 			}
2123 			mask = ~(pi->pi_bar[idx].size - 1);
2124 			switch (pi->pi_bar[idx].type) {
2125 			case PCIBAR_NONE:
2126 				pi->pi_bar[idx].addr = bar = 0;
2127 				break;
2128 			case PCIBAR_IO:
2129 				addr = *eax & mask;
2130 				addr &= 0xffff;
2131 				bar = addr | pi->pi_bar[idx].lobits;
2132 				/*
2133 				 * Register the new BAR value for interception
2134 				 */
2135 				if (addr != pi->pi_bar[idx].addr) {
2136 					update_bar_address(pi, addr, idx,
2137 							   PCIBAR_IO);
2138 				}
2139 				break;
2140 			case PCIBAR_MEM32:
2141 				addr = bar = *eax & mask;
2142 				bar |= pi->pi_bar[idx].lobits;
2143 				if (addr != pi->pi_bar[idx].addr) {
2144 					update_bar_address(pi, addr, idx,
2145 							   PCIBAR_MEM32);
2146 				}
2147 				break;
2148 			case PCIBAR_MEM64:
2149 				addr = bar = *eax & mask;
2150 				bar |= pi->pi_bar[idx].lobits;
2151 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
2152 					update_bar_address(pi, addr, idx,
2153 							   PCIBAR_MEM64);
2154 				}
2155 				break;
2156 			case PCIBAR_MEMHI64:
2157 				mask = ~(pi->pi_bar[idx - 1].size - 1);
2158 				addr = ((uint64_t)*eax << 32) & mask;
2159 				bar = addr >> 32;
2160 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
2161 					update_bar_address(pi, addr, idx - 1,
2162 							   PCIBAR_MEMHI64);
2163 				}
2164 				break;
2165 			case PCIBAR_ROM:
2166 				addr = bar = *eax & mask;
2167 				if (memen(pi) && romen(pi)) {
2168 					unregister_bar(pi, idx);
2169 				}
2170 				pi->pi_bar[idx].addr = addr;
2171 				pi->pi_bar[idx].lobits = *eax &
2172 				    PCIM_BIOS_ENABLE;
2173 				/* romen could have changed it value */
2174 				if (memen(pi) && romen(pi)) {
2175 					register_bar(pi, idx);
2176 				}
2177 				bar |= pi->pi_bar[idx].lobits;
2178 				break;
2179 			default:
2180 				assert(0);
2181 			}
2182 			pci_set_cfgdata32(pi, coff, bar);
2183 
2184 		} else if (pci_emul_iscap(pi, coff)) {
2185 			pci_emul_capwrite(pi, coff, bytes, *eax, 0, 0);
2186 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
2187 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
2188 		} else {
2189 			CFGWRITE(pi, coff, *eax, bytes);
2190 		}
2191 	}
2192 }
2193 
2194 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
2195 
2196 static int
2197 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
2198 		 uint32_t *eax, void *arg)
2199 {
2200 	uint32_t x;
2201 
2202 	if (bytes != 4) {
2203 		if (in)
2204 			*eax = (bytes == 2) ? 0xffff : 0xff;
2205 		return (0);
2206 	}
2207 
2208 	if (in) {
2209 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
2210 		if (cfgenable)
2211 			x |= CONF1_ENABLE;
2212 		*eax = x;
2213 	} else {
2214 		x = *eax;
2215 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
2216 		cfgoff = (x & PCI_REGMAX) & ~0x03;
2217 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
2218 		cfgslot = (x >> 11) & PCI_SLOTMAX;
2219 		cfgbus = (x >> 16) & PCI_BUSMAX;
2220 	}
2221 
2222 	return (0);
2223 }
2224 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
2225 
2226 static int
2227 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
2228 		 uint32_t *eax, void *arg)
2229 {
2230 	int coff;
2231 
2232 	assert(bytes == 1 || bytes == 2 || bytes == 4);
2233 
2234 	coff = cfgoff + (port - CONF1_DATA_PORT);
2235 	if (cfgenable) {
2236 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
2237 		    eax);
2238 	} else {
2239 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
2240 		if (in)
2241 			*eax = 0xffffffff;
2242 	}
2243 	return (0);
2244 }
2245 
2246 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
2247 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
2248 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
2249 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
2250 
2251 #ifdef BHYVE_SNAPSHOT
2252 /*
2253  * Saves/restores PCI device emulated state. Returns 0 on success.
2254  */
2255 static int
2256 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta)
2257 {
2258 	struct pci_devinst *pi;
2259 	int i;
2260 	int ret;
2261 
2262 	pi = meta->dev_data;
2263 
2264 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done);
2265 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done);
2266 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done);
2267 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done);
2268 
2269 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done);
2270 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done);
2271 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done);
2272 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done);
2273 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done);
2274 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done);
2275 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done);
2276 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done);
2277 
2278 	SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata),
2279 			      meta, ret, done);
2280 
2281 	for (i = 0; i < nitems(pi->pi_bar); i++) {
2282 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done);
2283 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done);
2284 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done);
2285 	}
2286 
2287 	/* Restore MSI-X table. */
2288 	for (i = 0; i < pi->pi_msix.table_count; i++) {
2289 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr,
2290 				      meta, ret, done);
2291 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data,
2292 				      meta, ret, done);
2293 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control,
2294 				      meta, ret, done);
2295 	}
2296 
2297 done:
2298 	return (ret);
2299 }
2300 
2301 static int
2302 pci_find_slotted_dev(const char *dev_name, struct pci_devemu **pde,
2303 		     struct pci_devinst **pdi)
2304 {
2305 	struct businfo *bi;
2306 	struct slotinfo *si;
2307 	struct funcinfo *fi;
2308 	int bus, slot, func;
2309 
2310 	assert(dev_name != NULL);
2311 	assert(pde != NULL);
2312 	assert(pdi != NULL);
2313 
2314 	for (bus = 0; bus < MAXBUSES; bus++) {
2315 		if ((bi = pci_businfo[bus]) == NULL)
2316 			continue;
2317 
2318 		for (slot = 0; slot < MAXSLOTS; slot++) {
2319 			si = &bi->slotinfo[slot];
2320 			for (func = 0; func < MAXFUNCS; func++) {
2321 				fi = &si->si_funcs[func];
2322 				if (fi->fi_pde == NULL)
2323 					continue;
2324 				if (strcmp(dev_name, fi->fi_pde->pe_emu) != 0)
2325 					continue;
2326 
2327 				*pde = fi->fi_pde;
2328 				*pdi = fi->fi_devi;
2329 				return (0);
2330 			}
2331 		}
2332 	}
2333 
2334 	return (EINVAL);
2335 }
2336 
2337 int
2338 pci_snapshot(struct vm_snapshot_meta *meta)
2339 {
2340 	struct pci_devemu *pde;
2341 	struct pci_devinst *pdi;
2342 	int ret;
2343 
2344 	assert(meta->dev_name != NULL);
2345 
2346 	ret = pci_find_slotted_dev(meta->dev_name, &pde, &pdi);
2347 	if (ret != 0) {
2348 		fprintf(stderr, "%s: no such name: %s\r\n",
2349 			__func__, meta->dev_name);
2350 		memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
2351 		return (0);
2352 	}
2353 
2354 	meta->dev_data = pdi;
2355 
2356 	if (pde->pe_snapshot == NULL) {
2357 		fprintf(stderr, "%s: not implemented yet for: %s\r\n",
2358 			__func__, meta->dev_name);
2359 		return (-1);
2360 	}
2361 
2362 	ret = pci_snapshot_pci_dev(meta);
2363 	if (ret != 0) {
2364 		fprintf(stderr, "%s: failed to snapshot pci dev\r\n",
2365 			__func__);
2366 		return (-1);
2367 	}
2368 
2369 	ret = (*pde->pe_snapshot)(meta);
2370 
2371 	return (ret);
2372 }
2373 
2374 int
2375 pci_pause(struct vmctx *ctx, const char *dev_name)
2376 {
2377 	struct pci_devemu *pde;
2378 	struct pci_devinst *pdi;
2379 	int ret;
2380 
2381 	assert(dev_name != NULL);
2382 
2383 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2384 	if (ret != 0) {
2385 		/*
2386 		 * It is possible to call this function without
2387 		 * checking that the device is inserted first.
2388 		 */
2389 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2390 		return (0);
2391 	}
2392 
2393 	if (pde->pe_pause == NULL) {
2394 		/* The pause/resume functionality is optional. */
2395 		fprintf(stderr, "%s: not implemented for: %s\n",
2396 			__func__, dev_name);
2397 		return (0);
2398 	}
2399 
2400 	return (*pde->pe_pause)(ctx, pdi);
2401 }
2402 
2403 int
2404 pci_resume(struct vmctx *ctx, const char *dev_name)
2405 {
2406 	struct pci_devemu *pde;
2407 	struct pci_devinst *pdi;
2408 	int ret;
2409 
2410 	assert(dev_name != NULL);
2411 
2412 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2413 	if (ret != 0) {
2414 		/*
2415 		 * It is possible to call this function without
2416 		 * checking that the device is inserted first.
2417 		 */
2418 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2419 		return (0);
2420 	}
2421 
2422 	if (pde->pe_resume == NULL) {
2423 		/* The pause/resume functionality is optional. */
2424 		fprintf(stderr, "%s: not implemented for: %s\n",
2425 			__func__, dev_name);
2426 		return (0);
2427 	}
2428 
2429 	return (*pde->pe_resume)(ctx, pdi);
2430 }
2431 #endif
2432 
2433 #define PCI_EMUL_TEST
2434 #ifdef PCI_EMUL_TEST
2435 /*
2436  * Define a dummy test device
2437  */
2438 #define DIOSZ	8
2439 #define DMEMSZ	4096
2440 struct pci_emul_dsoftc {
2441 	uint8_t   ioregs[DIOSZ];
2442 	uint8_t	  memregs[2][DMEMSZ];
2443 };
2444 
2445 #define	PCI_EMUL_MSI_MSGS	 4
2446 #define	PCI_EMUL_MSIX_MSGS	16
2447 
2448 static int
2449 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, nvlist_t *nvl)
2450 {
2451 	int error;
2452 	struct pci_emul_dsoftc *sc;
2453 
2454 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
2455 
2456 	pi->pi_arg = sc;
2457 
2458 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
2459 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
2460 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
2461 
2462 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
2463 	assert(error == 0);
2464 
2465 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
2466 	assert(error == 0);
2467 
2468 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2469 	assert(error == 0);
2470 
2471 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2472 	assert(error == 0);
2473 
2474 	return (0);
2475 }
2476 
2477 static void
2478 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2479 	      uint64_t offset, int size, uint64_t value)
2480 {
2481 	int i;
2482 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2483 
2484 	if (baridx == 0) {
2485 		if (offset + size > DIOSZ) {
2486 			printf("diow: iow too large, offset %ld size %d\n",
2487 			       offset, size);
2488 			return;
2489 		}
2490 
2491 		if (size == 1) {
2492 			sc->ioregs[offset] = value & 0xff;
2493 		} else if (size == 2) {
2494 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2495 		} else if (size == 4) {
2496 			*(uint32_t *)&sc->ioregs[offset] = value;
2497 		} else {
2498 			printf("diow: iow unknown size %d\n", size);
2499 		}
2500 
2501 		/*
2502 		 * Special magic value to generate an interrupt
2503 		 */
2504 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2505 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2506 
2507 		if (value == 0xabcdef) {
2508 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2509 				pci_generate_msi(pi, i);
2510 		}
2511 	}
2512 
2513 	if (baridx == 1 || baridx == 2) {
2514 		if (offset + size > DMEMSZ) {
2515 			printf("diow: memw too large, offset %ld size %d\n",
2516 			       offset, size);
2517 			return;
2518 		}
2519 
2520 		i = baridx - 1;		/* 'memregs' index */
2521 
2522 		if (size == 1) {
2523 			sc->memregs[i][offset] = value;
2524 		} else if (size == 2) {
2525 			*(uint16_t *)&sc->memregs[i][offset] = value;
2526 		} else if (size == 4) {
2527 			*(uint32_t *)&sc->memregs[i][offset] = value;
2528 		} else if (size == 8) {
2529 			*(uint64_t *)&sc->memregs[i][offset] = value;
2530 		} else {
2531 			printf("diow: memw unknown size %d\n", size);
2532 		}
2533 
2534 		/*
2535 		 * magic interrupt ??
2536 		 */
2537 	}
2538 
2539 	if (baridx > 2 || baridx < 0) {
2540 		printf("diow: unknown bar idx %d\n", baridx);
2541 	}
2542 }
2543 
2544 static uint64_t
2545 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2546 	      uint64_t offset, int size)
2547 {
2548 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2549 	uint32_t value;
2550 	int i;
2551 
2552 	if (baridx == 0) {
2553 		if (offset + size > DIOSZ) {
2554 			printf("dior: ior too large, offset %ld size %d\n",
2555 			       offset, size);
2556 			return (0);
2557 		}
2558 
2559 		value = 0;
2560 		if (size == 1) {
2561 			value = sc->ioregs[offset];
2562 		} else if (size == 2) {
2563 			value = *(uint16_t *) &sc->ioregs[offset];
2564 		} else if (size == 4) {
2565 			value = *(uint32_t *) &sc->ioregs[offset];
2566 		} else {
2567 			printf("dior: ior unknown size %d\n", size);
2568 		}
2569 	}
2570 
2571 	if (baridx == 1 || baridx == 2) {
2572 		if (offset + size > DMEMSZ) {
2573 			printf("dior: memr too large, offset %ld size %d\n",
2574 			       offset, size);
2575 			return (0);
2576 		}
2577 
2578 		i = baridx - 1;		/* 'memregs' index */
2579 
2580 		if (size == 1) {
2581 			value = sc->memregs[i][offset];
2582 		} else if (size == 2) {
2583 			value = *(uint16_t *) &sc->memregs[i][offset];
2584 		} else if (size == 4) {
2585 			value = *(uint32_t *) &sc->memregs[i][offset];
2586 		} else if (size == 8) {
2587 			value = *(uint64_t *) &sc->memregs[i][offset];
2588 		} else {
2589 			printf("dior: ior unknown size %d\n", size);
2590 		}
2591 	}
2592 
2593 
2594 	if (baridx > 2 || baridx < 0) {
2595 		printf("dior: unknown bar idx %d\n", baridx);
2596 		return (0);
2597 	}
2598 
2599 	return (value);
2600 }
2601 
2602 #ifdef BHYVE_SNAPSHOT
2603 int
2604 pci_emul_snapshot(struct vm_snapshot_meta *meta)
2605 {
2606 
2607 	return (0);
2608 }
2609 #endif
2610 
2611 struct pci_devemu pci_dummy = {
2612 	.pe_emu = "dummy",
2613 	.pe_init = pci_emul_dinit,
2614 	.pe_barwrite = pci_emul_diow,
2615 	.pe_barread = pci_emul_dior,
2616 #ifdef BHYVE_SNAPSHOT
2617 	.pe_snapshot = pci_emul_snapshot,
2618 #endif
2619 };
2620 PCI_EMUL_SET(pci_dummy);
2621 
2622 #endif /* PCI_EMUL_TEST */
2623