xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision 397e83df75e0fcd0d3fcb95ae4d794cb7600fc89)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
29 #include <sys/param.h>
30 #include <sys/linker_set.h>
31 #include <sys/mman.h>
32 
33 #include <ctype.h>
34 #include <err.h>
35 #include <errno.h>
36 #include <pthread.h>
37 #include <stdio.h>
38 #include <stdlib.h>
39 #include <string.h>
40 #include <strings.h>
41 #include <assert.h>
42 #include <stdbool.h>
43 #include <sysexits.h>
44 
45 #include <machine/vmm.h>
46 #include <machine/vmm_snapshot.h>
47 #include <vmmapi.h>
48 
49 #include "acpi.h"
50 #include "bhyverun.h"
51 #include "config.h"
52 #include "debug.h"
53 #ifdef __amd64__
54 #include "amd64/inout.h"
55 #endif
56 #include "mem.h"
57 #include "pci_emul.h"
58 #ifdef __amd64__
59 #include "amd64/pci_lpc.h"
60 #include "pci_passthru.h"
61 #endif
62 #include "qemu_fwcfg.h"
63 
64 #define CONF1_ADDR_PORT	   0x0cf8
65 #define CONF1_DATA_PORT	   0x0cfc
66 
67 #define CONF1_ENABLE	   0x80000000ul
68 
69 #define	MAXBUSES	(PCI_BUSMAX + 1)
70 #define MAXSLOTS	(PCI_SLOTMAX + 1)
71 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
72 
73 #define GB		(1024 * 1024 * 1024UL)
74 
75 struct funcinfo {
76 	nvlist_t *fi_config;
77 	struct pci_devemu *fi_pde;
78 	struct pci_devinst *fi_devi;
79 };
80 
81 struct intxinfo {
82 	int		ii_count;
83 	struct pci_irq	ii_irq;
84 };
85 
86 struct slotinfo {
87 	struct intxinfo si_intpins[4];
88 	struct funcinfo si_funcs[MAXFUNCS];
89 };
90 
91 struct businfo {
92 	uint16_t iobase, iolimit;		/* I/O window */
93 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
94 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
95 	struct slotinfo slotinfo[MAXSLOTS];
96 };
97 
98 static struct businfo *pci_businfo[MAXBUSES];
99 
100 SET_DECLARE(pci_devemu_set, struct pci_devemu);
101 
102 static uint64_t pci_emul_iobase;
103 static uint8_t *pci_emul_rombase;
104 static uint64_t pci_emul_romoffset;
105 static uint8_t *pci_emul_romlim;
106 static uint64_t pci_emul_membase32;
107 static uint64_t pci_emul_membase64;
108 static uint64_t pci_emul_memlim64;
109 
110 struct pci_bar_allocation {
111 	TAILQ_ENTRY(pci_bar_allocation) chain;
112 	struct pci_devinst *pdi;
113 	int idx;
114 	enum pcibar_type type;
115 	uint64_t size;
116 };
117 
118 static TAILQ_HEAD(pci_bar_list, pci_bar_allocation) pci_bars =
119     TAILQ_HEAD_INITIALIZER(pci_bars);
120 
121 struct boot_device {
122 	TAILQ_ENTRY(boot_device) boot_device_chain;
123 	struct pci_devinst *pdi;
124 	int bootindex;
125 };
126 static TAILQ_HEAD(boot_list, boot_device) boot_devices = TAILQ_HEAD_INITIALIZER(
127     boot_devices);
128 
129 #if defined(__amd64__)
130 #define	PCI_EMUL_IOBASE		0x2000
131 #define	PCI_EMUL_IOLIMIT	0x10000
132 #define	PCI_EMUL_IOMASK		0xffff
133 /*
134  * OVMF always uses 0xc0000000 as base address for 32 bit PCI MMIO. Don't
135  * change this address without changing it in OVMF.
136  */
137 #define	PCI_EMUL_MEMBASE32	0xc0000000
138 #elif defined(__aarch64__)
139 #define	PCI_EMUL_IOBASE		0xdf000000UL
140 #define	PCI_EMUL_IOLIMIT	0xe0000000UL
141 #define	PCI_EMUL_MEMBASE32	0xa0000000UL
142 #else
143 #error Unsupported platform
144 #endif
145 
146 #define	PCI_EMUL_ROMSIZE	0x10000000
147 
148 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
149 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
150 #ifdef __amd64__
151 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
152 #endif
153 
154 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
155 #define PCI_EMUL_MEMSIZE64	(32*GB)
156 
157 static void pci_lintr_route(struct pci_devinst *pi);
158 static void pci_lintr_update(struct pci_devinst *pi);
159 
160 static struct pci_devemu *pci_emul_finddev(const char *name);
161 static void pci_cfgrw(int in, int bus, int slot, int func, int coff,
162     int bytes, uint32_t *val);
163 
164 static __inline void
165 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
166 {
167 
168 	if (bytes == 1)
169 		pci_set_cfgdata8(pi, coff, val);
170 	else if (bytes == 2)
171 		pci_set_cfgdata16(pi, coff, val);
172 	else
173 		pci_set_cfgdata32(pi, coff, val);
174 }
175 
176 static __inline uint32_t
177 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
178 {
179 
180 	if (bytes == 1)
181 		return (pci_get_cfgdata8(pi, coff));
182 	else if (bytes == 2)
183 		return (pci_get_cfgdata16(pi, coff));
184 	else
185 		return (pci_get_cfgdata32(pi, coff));
186 }
187 
188 static int
189 is_pcir_bar(int coff)
190 {
191 	return (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1));
192 }
193 
194 static int
195 is_pcir_bios(int coff)
196 {
197 	return (coff >= PCIR_BIOS && coff < PCIR_BIOS + 4);
198 }
199 
200 /*
201  * I/O access
202  */
203 
204 /*
205  * Slot options are in the form:
206  *
207  *  <bus>:<slot>:<func>,<emul>[,<config>]
208  *  <slot>[:<func>],<emul>[,<config>]
209  *
210  *  slot is 0..31
211  *  func is 0..7
212  *  emul is a string describing the type of PCI device e.g. virtio-net
213  *  config is an optional string, depending on the device, that can be
214  *  used for configuration.
215  *   Examples are:
216  *     1,virtio-net,tap0
217  *     3:0,dummy
218  */
219 static void
220 pci_parse_slot_usage(char *aopt)
221 {
222 
223 	EPRINTLN("Invalid PCI slot info field \"%s\"", aopt);
224 }
225 
226 /*
227  * Helper function to parse a list of comma-separated options where
228  * each option is formatted as "name[=value]".  If no value is
229  * provided, the option is treated as a boolean and is given a value
230  * of true.
231  */
232 int
233 pci_parse_legacy_config(nvlist_t *nvl, const char *opt)
234 {
235 	char *config, *name, *tofree, *value;
236 
237 	if (opt == NULL)
238 		return (0);
239 
240 	config = tofree = strdup(opt);
241 	while ((name = strsep(&config, ",")) != NULL) {
242 		value = strchr(name, '=');
243 		if (value != NULL) {
244 			*value = '\0';
245 			value++;
246 			set_config_value_node(nvl, name, value);
247 		} else
248 			set_config_bool_node(nvl, name, true);
249 	}
250 	free(tofree);
251 	return (0);
252 }
253 
254 /*
255  * PCI device configuration is stored in MIBs that encode the device's
256  * location:
257  *
258  * pci.<bus>.<slot>.<func>
259  *
260  * Where "bus", "slot", and "func" are all decimal values without
261  * leading zeroes.  Each valid device must have a "device" node which
262  * identifies the driver model of the device.
263  *
264  * Device backends can provide a parser for the "config" string.  If
265  * a custom parser is not provided, pci_parse_legacy_config() is used
266  * to parse the string.
267  */
268 int
269 pci_parse_slot(char *opt)
270 {
271 	char node_name[sizeof("pci.XXX.XX.X")];
272 	struct pci_devemu *pde;
273 	char *emul, *config, *str, *cp;
274 	int error, bnum, snum, fnum;
275 	nvlist_t *nvl;
276 
277 	error = -1;
278 	str = strdup(opt);
279 
280 	emul = config = NULL;
281 	if ((cp = strchr(str, ',')) != NULL) {
282 		*cp = '\0';
283 		emul = cp + 1;
284 		if ((cp = strchr(emul, ',')) != NULL) {
285 			*cp = '\0';
286 			config = cp + 1;
287 		}
288 	} else {
289 		pci_parse_slot_usage(opt);
290 		goto done;
291 	}
292 
293 	/* <bus>:<slot>:<func> */
294 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
295 		bnum = 0;
296 		/* <slot>:<func> */
297 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
298 			fnum = 0;
299 			/* <slot> */
300 			if (sscanf(str, "%d", &snum) != 1) {
301 				snum = -1;
302 			}
303 		}
304 	}
305 
306 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
307 	    fnum < 0 || fnum >= MAXFUNCS) {
308 		pci_parse_slot_usage(opt);
309 		goto done;
310 	}
311 
312 	pde = pci_emul_finddev(emul);
313 	if (pde == NULL) {
314 		EPRINTLN("pci slot %d:%d:%d: unknown device \"%s\"", bnum, snum,
315 		    fnum, emul);
316 		goto done;
317 	}
318 
319 	snprintf(node_name, sizeof(node_name), "pci.%d.%d.%d", bnum, snum,
320 	    fnum);
321 	nvl = find_config_node(node_name);
322 	if (nvl != NULL) {
323 		EPRINTLN("pci slot %d:%d:%d already occupied!", bnum, snum,
324 		    fnum);
325 		goto done;
326 	}
327 	nvl = create_config_node(node_name);
328 	if (pde->pe_alias != NULL)
329 		set_config_value_node(nvl, "device", pde->pe_alias);
330 	else
331 		set_config_value_node(nvl, "device", pde->pe_emu);
332 
333 	if (pde->pe_legacy_config != NULL)
334 		error = pde->pe_legacy_config(nvl, config);
335 	else
336 		error = pci_parse_legacy_config(nvl, config);
337 done:
338 	free(str);
339 	return (error);
340 }
341 
342 void
343 pci_print_supported_devices(void)
344 {
345 	struct pci_devemu **pdpp, *pdp;
346 
347 	SET_FOREACH(pdpp, pci_devemu_set) {
348 		pdp = *pdpp;
349 		printf("%s\n", pdp->pe_emu);
350 	}
351 }
352 
353 uint32_t
354 pci_config_read_reg(const struct pcisel *const host_sel, nvlist_t *nvl,
355     const uint32_t reg, const uint8_t size, const uint32_t def)
356 {
357 	const char *config;
358 	const nvlist_t *pci_regs;
359 
360 	assert(size == 1 || size == 2 || size == 4);
361 
362 	pci_regs = find_relative_config_node(nvl, "pcireg");
363 	if (pci_regs == NULL) {
364 		return def;
365 	}
366 
367 	switch (reg) {
368 	case PCIR_DEVICE:
369 		config = get_config_value_node(pci_regs, "device");
370 		break;
371 	case PCIR_VENDOR:
372 		config = get_config_value_node(pci_regs, "vendor");
373 		break;
374 	case PCIR_REVID:
375 		config = get_config_value_node(pci_regs, "revid");
376 		break;
377 	case PCIR_SUBVEND_0:
378 		config = get_config_value_node(pci_regs, "subvendor");
379 		break;
380 	case PCIR_SUBDEV_0:
381 		config = get_config_value_node(pci_regs, "subdevice");
382 		break;
383 	default:
384 		return (-1);
385 	}
386 
387 	if (config == NULL) {
388 		return def;
389 	} else if (host_sel != NULL && strcmp(config, "host") == 0) {
390 #ifdef __amd64__
391 		return pci_host_read_config(host_sel, reg, size);
392 #else
393 		errx(1, "cannot fetch host PCI configuration");
394 #endif
395 	} else {
396 		return strtol(config, NULL, 16);
397 	}
398 }
399 
400 static int
401 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
402 {
403 
404 	if (offset < pi->pi_msix.pba_offset)
405 		return (0);
406 
407 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
408 		return (0);
409 	}
410 
411 	return (1);
412 }
413 
414 int
415 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
416 		     uint64_t value)
417 {
418 	int msix_entry_offset;
419 	int tab_index;
420 	char *dest;
421 
422 	/* support only 4 or 8 byte writes */
423 	if (size != 4 && size != 8)
424 		return (-1);
425 
426 	/*
427 	 * Return if table index is beyond what device supports
428 	 */
429 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
430 	if (tab_index >= pi->pi_msix.table_count)
431 		return (-1);
432 
433 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
434 
435 	/* support only aligned writes */
436 	if ((msix_entry_offset % size) != 0)
437 		return (-1);
438 
439 	dest = (char *)(pi->pi_msix.table + tab_index);
440 	dest += msix_entry_offset;
441 
442 	if (size == 4)
443 		*((uint32_t *)dest) = value;
444 	else
445 		*((uint64_t *)dest) = value;
446 
447 	return (0);
448 }
449 
450 uint64_t
451 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
452 {
453 	char *dest;
454 	int msix_entry_offset;
455 	int tab_index;
456 	uint64_t retval = ~0;
457 
458 	/*
459 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
460 	 * table but we also allow 1 byte access to accommodate reads from
461 	 * ddb.
462 	 */
463 	if (size != 1 && size != 4 && size != 8)
464 		return (retval);
465 
466 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
467 
468 	/* support only aligned reads */
469 	if ((msix_entry_offset % size) != 0) {
470 		return (retval);
471 	}
472 
473 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
474 
475 	if (tab_index < pi->pi_msix.table_count) {
476 		/* valid MSI-X Table access */
477 		dest = (char *)(pi->pi_msix.table + tab_index);
478 		dest += msix_entry_offset;
479 
480 		if (size == 1)
481 			retval = *((uint8_t *)dest);
482 		else if (size == 4)
483 			retval = *((uint32_t *)dest);
484 		else
485 			retval = *((uint64_t *)dest);
486 	} else if (pci_valid_pba_offset(pi, offset)) {
487 		/* return 0 for PBA access */
488 		retval = 0;
489 	}
490 
491 	return (retval);
492 }
493 
494 int
495 pci_msix_table_bar(struct pci_devinst *pi)
496 {
497 
498 	if (pi->pi_msix.table != NULL)
499 		return (pi->pi_msix.table_bar);
500 	else
501 		return (-1);
502 }
503 
504 int
505 pci_msix_pba_bar(struct pci_devinst *pi)
506 {
507 
508 	if (pi->pi_msix.table != NULL)
509 		return (pi->pi_msix.pba_bar);
510 	else
511 		return (-1);
512 }
513 
514 #ifdef __amd64__
515 static int
516 pci_emul_io_handler(struct vmctx *ctx __unused, int in, int port,
517     int bytes, uint32_t *eax, void *arg)
518 {
519 	struct pci_devinst *pdi = arg;
520 	struct pci_devemu *pe = pdi->pi_d;
521 	uint64_t offset;
522 	int i;
523 
524 	assert(port >= 0);
525 
526 	for (i = 0; i <= PCI_BARMAX; i++) {
527 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
528 		    (uint64_t)port >= pdi->pi_bar[i].addr &&
529 		    (uint64_t)port + bytes <=
530 		    pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
531 			offset = port - pdi->pi_bar[i].addr;
532 			if (in)
533 				*eax = (*pe->pe_barread)(pdi, i,
534 							 offset, bytes);
535 			else
536 				(*pe->pe_barwrite)(pdi, i, offset,
537 						   bytes, *eax);
538 			return (0);
539 		}
540 	}
541 	return (-1);
542 }
543 #else
544 static int
545 pci_emul_iomem_handler(struct vcpu *vcpu __unused, int dir,
546     uint64_t addr, int size, uint64_t *val, void *arg1, long arg2)
547 {
548 	struct pci_devinst *pdi = arg1;
549 	struct pci_devemu *pe = pdi->pi_d;
550 	uint64_t offset;
551 	int bidx = (int)arg2;
552 
553 	assert(bidx <= PCI_BARMAX);
554 	assert(pdi->pi_bar[bidx].type == PCIBAR_IO);
555 	assert(addr >= pdi->pi_bar[bidx].addr &&
556 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
557 	assert(size == 1 || size == 2 || size == 4);
558 
559 	offset = addr - pdi->pi_bar[bidx].addr;
560 	if (dir == MEM_F_READ)
561 		*val = (*pe->pe_barread)(pdi, bidx, offset, size);
562 	else
563 		(*pe->pe_barwrite)(pdi, bidx, offset, size, *val);
564 
565 	return (0);
566 }
567 #endif /* !__amd64__ */
568 
569 static int
570 pci_emul_mem_handler(struct vcpu *vcpu __unused, int dir,
571     uint64_t addr, int size, uint64_t *val, void *arg1, long arg2)
572 {
573 	struct pci_devinst *pdi = arg1;
574 	struct pci_devemu *pe = pdi->pi_d;
575 	uint64_t offset;
576 	int bidx = (int)arg2;
577 
578 	assert(bidx <= PCI_BARMAX);
579 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
580 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
581 	assert(addr >= pdi->pi_bar[bidx].addr &&
582 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
583 
584 	offset = addr - pdi->pi_bar[bidx].addr;
585 
586 	if (dir == MEM_F_WRITE) {
587 		if (size == 8) {
588 			(*pe->pe_barwrite)(pdi, bidx, offset,
589 					   4, *val & 0xffffffff);
590 			(*pe->pe_barwrite)(pdi, bidx, offset + 4,
591 					   4, *val >> 32);
592 		} else {
593 			(*pe->pe_barwrite)(pdi, bidx, offset,
594 					   size, *val);
595 		}
596 	} else {
597 		if (size == 8) {
598 			*val = (*pe->pe_barread)(pdi, bidx,
599 						 offset, 4);
600 			*val |= (*pe->pe_barread)(pdi, bidx,
601 						  offset + 4, 4) << 32;
602 		} else {
603 			*val = (*pe->pe_barread)(pdi, bidx,
604 						 offset, size);
605 		}
606 	}
607 
608 	return (0);
609 }
610 
611 
612 static int
613 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
614 			uint64_t *addr)
615 {
616 	uint64_t base;
617 
618 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
619 
620 	base = roundup2(*baseptr, size);
621 
622 	if (base + size <= limit) {
623 		*addr = base;
624 		*baseptr = base + size;
625 		return (0);
626 	} else
627 		return (-1);
628 }
629 
630 /*
631  * Register (or unregister) the MMIO or I/O region associated with the BAR
632  * register 'idx' of an emulated pci device.
633  */
634 static void
635 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
636 {
637 	struct pci_devemu *pe;
638 	int error;
639 	enum pcibar_type type;
640 
641 	pe = pi->pi_d;
642 	type = pi->pi_bar[idx].type;
643 	switch (type) {
644 	case PCIBAR_IO:
645 	{
646 #ifdef __amd64__
647 		struct inout_port iop;
648 
649 		bzero(&iop, sizeof(struct inout_port));
650 		iop.name = pi->pi_name;
651 		iop.port = pi->pi_bar[idx].addr;
652 		iop.size = pi->pi_bar[idx].size;
653 		if (registration) {
654 			iop.flags = IOPORT_F_INOUT;
655 			iop.handler = pci_emul_io_handler;
656 			iop.arg = pi;
657 			error = register_inout(&iop);
658 		} else
659 			error = unregister_inout(&iop);
660 #else
661 		struct mem_range mr;
662 
663 		bzero(&mr, sizeof(struct mem_range));
664 		mr.name = pi->pi_name;
665 		mr.base = pi->pi_bar[idx].addr;
666 		mr.size = pi->pi_bar[idx].size;
667 		if (registration) {
668 			mr.flags = MEM_F_RW;
669 			mr.handler = pci_emul_iomem_handler;
670 			mr.arg1 = pi;
671 			mr.arg2 = idx;
672 			error = register_mem(&mr);
673 		} else
674 			error = unregister_mem(&mr);
675 #endif
676 		break;
677 	}
678 	case PCIBAR_MEM32:
679 	case PCIBAR_MEM64:
680 	{
681 		struct mem_range mr;
682 
683 		bzero(&mr, sizeof(struct mem_range));
684 		mr.name = pi->pi_name;
685 		mr.base = pi->pi_bar[idx].addr;
686 		mr.size = pi->pi_bar[idx].size;
687 		if (registration) {
688 			mr.flags = MEM_F_RW;
689 			mr.handler = pci_emul_mem_handler;
690 			mr.arg1 = pi;
691 			mr.arg2 = idx;
692 			error = register_mem(&mr);
693 		} else
694 			error = unregister_mem(&mr);
695 		break;
696 	}
697 	case PCIBAR_ROM:
698 		error = 0;
699 		break;
700 	default:
701 		error = EINVAL;
702 		break;
703 	}
704 	assert(error == 0);
705 
706 	if (pe->pe_baraddr != NULL)
707 		(*pe->pe_baraddr)(pi, idx, registration, pi->pi_bar[idx].addr);
708 }
709 
710 static void
711 unregister_bar(struct pci_devinst *pi, int idx)
712 {
713 
714 	modify_bar_registration(pi, idx, 0);
715 }
716 
717 static void
718 register_bar(struct pci_devinst *pi, int idx)
719 {
720 
721 	modify_bar_registration(pi, idx, 1);
722 }
723 
724 /* Is the ROM enabled for the emulated pci device? */
725 static int
726 romen(struct pci_devinst *pi)
727 {
728 	return (pi->pi_bar[PCI_ROM_IDX].lobits & PCIM_BIOS_ENABLE) ==
729 	    PCIM_BIOS_ENABLE;
730 }
731 
732 /* Are we decoding i/o port accesses for the emulated pci device? */
733 static int
734 porten(struct pci_devinst *pi)
735 {
736 	uint16_t cmd;
737 
738 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
739 
740 	return (cmd & PCIM_CMD_PORTEN);
741 }
742 
743 /* Are we decoding memory accesses for the emulated pci device? */
744 static int
745 memen(struct pci_devinst *pi)
746 {
747 	uint16_t cmd;
748 
749 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
750 
751 	return (cmd & PCIM_CMD_MEMEN);
752 }
753 
754 /*
755  * Update the MMIO or I/O address that is decoded by the BAR register.
756  *
757  * If the pci device has enabled the address space decoding then intercept
758  * the address range decoded by the BAR register.
759  */
760 static void
761 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
762 {
763 	int decode;
764 
765 	if (pi->pi_bar[idx].type == PCIBAR_IO)
766 		decode = porten(pi);
767 	else
768 		decode = memen(pi);
769 
770 	if (decode)
771 		unregister_bar(pi, idx);
772 
773 	switch (type) {
774 	case PCIBAR_IO:
775 	case PCIBAR_MEM32:
776 		pi->pi_bar[idx].addr = addr;
777 		break;
778 	case PCIBAR_MEM64:
779 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
780 		pi->pi_bar[idx].addr |= addr;
781 		break;
782 	case PCIBAR_MEMHI64:
783 		pi->pi_bar[idx].addr &= 0xffffffff;
784 		pi->pi_bar[idx].addr |= addr;
785 		break;
786 	default:
787 		assert(0);
788 	}
789 
790 	if (decode)
791 		register_bar(pi, idx);
792 }
793 
794 int
795 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
796     uint64_t size)
797 {
798 	assert((type == PCIBAR_ROM) || (idx >= 0 && idx <= PCI_BARMAX));
799 	assert((type != PCIBAR_ROM) || (idx == PCI_ROM_IDX));
800 
801 	if ((size & (size - 1)) != 0)
802 		size = 1UL << flsl(size);	/* round up to a power of 2 */
803 
804 	/* Enforce minimum BAR sizes required by the PCI standard */
805 	if (type == PCIBAR_IO) {
806 		if (size < 4)
807 			size = 4;
808 	} else if (type == PCIBAR_ROM) {
809 		if (size < ~PCIM_BIOS_ADDR_MASK + 1)
810 			size = ~PCIM_BIOS_ADDR_MASK + 1;
811 	} else {
812 		if (size < 16)
813 			size = 16;
814 	}
815 
816 	/*
817 	 * To reduce fragmentation of the MMIO space, we allocate the BARs by
818 	 * size. Therefore, don't allocate the BAR yet. We create a list of all
819 	 * BAR allocation which is sorted by BAR size. When all PCI devices are
820 	 * initialized, we will assign an address to the BARs.
821 	 */
822 
823 	/* create a new list entry */
824 	struct pci_bar_allocation *const new_bar = malloc(sizeof(*new_bar));
825 	memset(new_bar, 0, sizeof(*new_bar));
826 	new_bar->pdi = pdi;
827 	new_bar->idx = idx;
828 	new_bar->type = type;
829 	new_bar->size = size;
830 
831 	/*
832 	 * Search for a BAR which size is lower than the size of our newly
833 	 * allocated BAR.
834 	 */
835 	struct pci_bar_allocation *bar = NULL;
836 	TAILQ_FOREACH(bar, &pci_bars, chain) {
837 		if (bar->size < size) {
838 			break;
839 		}
840 	}
841 
842 	if (bar == NULL) {
843 		/*
844 		 * Either the list is empty or new BAR is the smallest BAR of
845 		 * the list. Append it to the end of our list.
846 		 */
847 		TAILQ_INSERT_TAIL(&pci_bars, new_bar, chain);
848 	} else {
849 		/*
850 		 * The found BAR is smaller than our new BAR. For that reason,
851 		 * insert our new BAR before the found BAR.
852 		 */
853 		TAILQ_INSERT_BEFORE(bar, new_bar, chain);
854 	}
855 
856 	/*
857 	 * pci_passthru devices synchronize their physical and virtual command
858 	 * register on init. For that reason, the virtual cmd reg should be
859 	 * updated as early as possible.
860 	 */
861 	uint16_t enbit = 0;
862 	switch (type) {
863 	case PCIBAR_IO:
864 		enbit = PCIM_CMD_PORTEN;
865 		break;
866 	case PCIBAR_MEM64:
867 	case PCIBAR_MEM32:
868 		enbit = PCIM_CMD_MEMEN;
869 		break;
870 	default:
871 		enbit = 0;
872 		break;
873 	}
874 
875 	const uint16_t cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
876 	pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
877 
878 	return (0);
879 }
880 
881 static int
882 pci_emul_assign_bar(struct pci_devinst *const pdi, const int idx,
883     const enum pcibar_type type, const uint64_t size)
884 {
885 	int error;
886 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
887 
888 	switch (type) {
889 	case PCIBAR_NONE:
890 		baseptr = NULL;
891 		addr = mask = lobits = 0;
892 		break;
893 	case PCIBAR_IO:
894 		baseptr = &pci_emul_iobase;
895 		limit = PCI_EMUL_IOLIMIT;
896 		mask = PCIM_BAR_IO_BASE;
897 		lobits = PCIM_BAR_IO_SPACE;
898 		break;
899 	case PCIBAR_MEM64:
900 		/*
901 		 * XXX
902 		 * Some drivers do not work well if the 64-bit BAR is allocated
903 		 * above 4GB. Allow for this by allocating small requests under
904 		 * 4GB unless then allocation size is larger than some arbitrary
905 		 * number (128MB currently).
906 		 */
907 		if (size > 128 * 1024 * 1024) {
908 			baseptr = &pci_emul_membase64;
909 			limit = pci_emul_memlim64;
910 			mask = PCIM_BAR_MEM_BASE;
911 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
912 				 PCIM_BAR_MEM_PREFETCH;
913 		} else {
914 			baseptr = &pci_emul_membase32;
915 			limit = PCI_EMUL_MEMLIMIT32;
916 			mask = PCIM_BAR_MEM_BASE;
917 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
918 		}
919 		break;
920 	case PCIBAR_MEM32:
921 		baseptr = &pci_emul_membase32;
922 		limit = PCI_EMUL_MEMLIMIT32;
923 		mask = PCIM_BAR_MEM_BASE;
924 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
925 		break;
926 	case PCIBAR_ROM:
927 		/* do not claim memory for ROM. OVMF will do it for us. */
928 		baseptr = NULL;
929 		limit = 0;
930 		mask = PCIM_BIOS_ADDR_MASK;
931 		lobits = 0;
932 		break;
933 	default:
934 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
935 		assert(0);
936 	}
937 
938 	if (baseptr != NULL) {
939 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
940 		if (error != 0)
941 			return (error);
942 	} else {
943 		addr = 0;
944 	}
945 
946 	pdi->pi_bar[idx].type = type;
947 	pdi->pi_bar[idx].addr = addr;
948 	pdi->pi_bar[idx].size = size;
949 	/*
950 	 * passthru devices are using same lobits as physical device they set
951 	 * this property
952 	 */
953 	if (pdi->pi_bar[idx].lobits != 0) {
954 		lobits = pdi->pi_bar[idx].lobits;
955 	} else {
956 		pdi->pi_bar[idx].lobits = lobits;
957 	}
958 
959 	/* Initialize the BAR register in config space */
960 	bar = (addr & mask) | lobits;
961 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
962 
963 	if (type == PCIBAR_MEM64) {
964 		assert(idx + 1 <= PCI_BARMAX);
965 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
966 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
967 	}
968 
969 	if (type != PCIBAR_ROM) {
970 		register_bar(pdi, idx);
971 	}
972 
973 	return (0);
974 }
975 
976 int
977 pci_emul_alloc_rom(struct pci_devinst *const pdi, const uint64_t size,
978     void **const addr)
979 {
980 	/* allocate ROM space once on first call */
981 	if (pci_emul_rombase == 0) {
982 		pci_emul_rombase = vm_create_devmem(pdi->pi_vmctx, VM_PCIROM,
983 		    "pcirom", PCI_EMUL_ROMSIZE);
984 		if (pci_emul_rombase == MAP_FAILED) {
985 			warnx("%s: failed to create rom segment", __func__);
986 			return (-1);
987 		}
988 		pci_emul_romlim = pci_emul_rombase + PCI_EMUL_ROMSIZE;
989 		pci_emul_romoffset = 0;
990 	}
991 
992 	/* ROM size should be a power of 2 and greater than 2 KB */
993 	const uint64_t rom_size = MAX(1UL << flsl(size),
994 	    ~PCIM_BIOS_ADDR_MASK + 1);
995 
996 	/* check if ROM fits into ROM space */
997 	if (pci_emul_romoffset + rom_size > PCI_EMUL_ROMSIZE) {
998 		warnx("%s: no space left in rom segment:", __func__);
999 		warnx("%16lu bytes left",
1000 		    PCI_EMUL_ROMSIZE - pci_emul_romoffset);
1001 		warnx("%16lu bytes required by %d/%d/%d", rom_size, pdi->pi_bus,
1002 		    pdi->pi_slot, pdi->pi_func);
1003 		return (-1);
1004 	}
1005 
1006 	/* allocate ROM BAR */
1007 	const int error = pci_emul_alloc_bar(pdi, PCI_ROM_IDX, PCIBAR_ROM,
1008 	    rom_size);
1009 	if (error)
1010 		return error;
1011 
1012 	/* return address */
1013 	*addr = pci_emul_rombase + pci_emul_romoffset;
1014 
1015 	/* save offset into ROM Space */
1016 	pdi->pi_romoffset = pci_emul_romoffset;
1017 
1018 	/* increase offset for next ROM */
1019 	pci_emul_romoffset += rom_size;
1020 
1021 	return (0);
1022 }
1023 
1024 int
1025 pci_emul_add_boot_device(struct pci_devinst *pi, int bootindex)
1026 {
1027 	struct boot_device *new_device, *device;
1028 
1029 	/* don't permit a negative bootindex */
1030 	if (bootindex < 0) {
1031 		errx(4, "Invalid bootindex %d for %s", bootindex, pi->pi_name);
1032 	}
1033 
1034 	/* alloc new boot device */
1035 	new_device = calloc(1, sizeof(struct boot_device));
1036 	if (new_device == NULL) {
1037 		return (ENOMEM);
1038 	}
1039 	new_device->pdi = pi;
1040 	new_device->bootindex = bootindex;
1041 
1042 	/* search for boot device with higher boot index */
1043 	TAILQ_FOREACH(device, &boot_devices, boot_device_chain) {
1044 		if (device->bootindex == bootindex) {
1045 			errx(4,
1046 			    "Could not set bootindex %d for %s. Bootindex already occupied by %s",
1047 			    bootindex, pi->pi_name, device->pdi->pi_name);
1048 		} else if (device->bootindex > bootindex) {
1049 			break;
1050 		}
1051 	}
1052 
1053 	/* add boot device to queue */
1054 	if (device == NULL) {
1055 		TAILQ_INSERT_TAIL(&boot_devices, new_device, boot_device_chain);
1056 	} else {
1057 		TAILQ_INSERT_BEFORE(device, new_device, boot_device_chain);
1058 	}
1059 
1060 	return (0);
1061 }
1062 
1063 #define	CAP_START_OFFSET	0x40
1064 static int
1065 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
1066 {
1067 	int i, capoff, reallen;
1068 	uint16_t sts;
1069 
1070 	assert(caplen > 0);
1071 
1072 	reallen = roundup2(caplen, 4);		/* dword aligned */
1073 
1074 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1075 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
1076 		capoff = CAP_START_OFFSET;
1077 	else
1078 		capoff = pi->pi_capend + 1;
1079 
1080 	/* Check if we have enough space */
1081 	if (capoff + reallen > PCI_REGMAX + 1)
1082 		return (-1);
1083 
1084 	/* Set the previous capability pointer */
1085 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
1086 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
1087 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
1088 	} else
1089 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
1090 
1091 	/* Copy the capability */
1092 	for (i = 0; i < caplen; i++)
1093 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
1094 
1095 	/* Set the next capability pointer */
1096 	pci_set_cfgdata8(pi, capoff + 1, 0);
1097 
1098 	pi->pi_prevcap = capoff;
1099 	pi->pi_capend = capoff + reallen - 1;
1100 	return (0);
1101 }
1102 
1103 static struct pci_devemu *
1104 pci_emul_finddev(const char *name)
1105 {
1106 	struct pci_devemu **pdpp, *pdp;
1107 
1108 	SET_FOREACH(pdpp, pci_devemu_set) {
1109 		pdp = *pdpp;
1110 		if (!strcmp(pdp->pe_emu, name)) {
1111 			return (pdp);
1112 		}
1113 	}
1114 
1115 	return (NULL);
1116 }
1117 
1118 static int
1119 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
1120     int func, struct funcinfo *fi)
1121 {
1122 	struct pci_devinst *pdi;
1123 	int err;
1124 
1125 	pdi = calloc(1, sizeof(struct pci_devinst));
1126 
1127 	pdi->pi_vmctx = ctx;
1128 	pdi->pi_bus = bus;
1129 	pdi->pi_slot = slot;
1130 	pdi->pi_func = func;
1131 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
1132 	pdi->pi_lintr.pin = 0;
1133 	pdi->pi_lintr.state = IDLE;
1134 	pci_irq_init_irq(&pdi->pi_lintr.irq);
1135 	pdi->pi_d = pde;
1136 	snprintf(pdi->pi_name, PI_NAMESZ, "%s@pci.%d.%d.%d", pde->pe_emu, bus,
1137 	    slot, func);
1138 
1139 	/* Disable legacy interrupts */
1140 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
1141 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
1142 
1143 	pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
1144 
1145 	err = (*pde->pe_init)(pdi, fi->fi_config);
1146 	if (err == 0)
1147 		fi->fi_devi = pdi;
1148 	else
1149 		free(pdi);
1150 
1151 	return (err);
1152 }
1153 
1154 void
1155 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
1156 {
1157 	int mmc;
1158 
1159 	/* Number of msi messages must be a power of 2 between 1 and 32 */
1160 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
1161 	mmc = ffs(msgnum) - 1;
1162 
1163 	bzero(msicap, sizeof(struct msicap));
1164 	msicap->capid = PCIY_MSI;
1165 	msicap->nextptr = nextptr;
1166 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
1167 }
1168 
1169 int
1170 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
1171 {
1172 	struct msicap msicap;
1173 
1174 	pci_populate_msicap(&msicap, msgnum, 0);
1175 
1176 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
1177 }
1178 
1179 static void
1180 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
1181 		     uint32_t msix_tab_size)
1182 {
1183 
1184 	assert(msix_tab_size % 4096 == 0);
1185 
1186 	bzero(msixcap, sizeof(struct msixcap));
1187 	msixcap->capid = PCIY_MSIX;
1188 
1189 	/*
1190 	 * Message Control Register, all fields set to
1191 	 * zero except for the Table Size.
1192 	 * Note: Table size N is encoded as N-1
1193 	 */
1194 	msixcap->msgctrl = msgnum - 1;
1195 
1196 	/*
1197 	 * MSI-X BAR setup:
1198 	 * - MSI-X table start at offset 0
1199 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
1200 	 */
1201 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
1202 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
1203 }
1204 
1205 static void
1206 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
1207 {
1208 	int i, table_size;
1209 
1210 	assert(table_entries > 0);
1211 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
1212 
1213 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
1214 	pi->pi_msix.table = calloc(1, table_size);
1215 
1216 	/* set mask bit of vector control register */
1217 	for (i = 0; i < table_entries; i++)
1218 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
1219 }
1220 
1221 int
1222 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
1223 {
1224 	uint32_t tab_size;
1225 	struct msixcap msixcap;
1226 
1227 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
1228 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
1229 
1230 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
1231 
1232 	/* Align table size to nearest 4K */
1233 	tab_size = roundup2(tab_size, 4096);
1234 
1235 	pi->pi_msix.table_bar = barnum;
1236 	pi->pi_msix.pba_bar   = barnum;
1237 	pi->pi_msix.table_offset = 0;
1238 	pi->pi_msix.table_count = msgnum;
1239 	pi->pi_msix.pba_offset = tab_size;
1240 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
1241 
1242 	pci_msix_table_init(pi, msgnum);
1243 
1244 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
1245 
1246 	/* allocate memory for MSI-X Table and PBA */
1247 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
1248 				tab_size + pi->pi_msix.pba_size);
1249 
1250 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
1251 					sizeof(msixcap)));
1252 }
1253 
1254 static void
1255 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1256 		 int bytes, uint32_t val)
1257 {
1258 	uint16_t msgctrl, rwmask;
1259 	int off;
1260 
1261 	off = offset - capoff;
1262 	/* Message Control Register */
1263 	if (off == 2 && bytes == 2) {
1264 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
1265 		msgctrl = pci_get_cfgdata16(pi, offset);
1266 		msgctrl &= ~rwmask;
1267 		msgctrl |= val & rwmask;
1268 		val = msgctrl;
1269 
1270 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
1271 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
1272 		pci_lintr_update(pi);
1273 	}
1274 
1275 	CFGWRITE(pi, offset, val, bytes);
1276 }
1277 
1278 static void
1279 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1280 		int bytes, uint32_t val)
1281 {
1282 	uint16_t msgctrl, rwmask, msgdata, mme;
1283 	uint32_t addrlo;
1284 
1285 	/*
1286 	 * If guest is writing to the message control register make sure
1287 	 * we do not overwrite read-only fields.
1288 	 */
1289 	if ((offset - capoff) == 2 && bytes == 2) {
1290 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
1291 		msgctrl = pci_get_cfgdata16(pi, offset);
1292 		msgctrl &= ~rwmask;
1293 		msgctrl |= val & rwmask;
1294 		val = msgctrl;
1295 	}
1296 	CFGWRITE(pi, offset, val, bytes);
1297 
1298 	msgctrl = pci_get_cfgdata16(pi, capoff + 2);
1299 	addrlo = pci_get_cfgdata32(pi, capoff + 4);
1300 	if (msgctrl & PCIM_MSICTRL_64BIT)
1301 		msgdata = pci_get_cfgdata16(pi, capoff + 12);
1302 	else
1303 		msgdata = pci_get_cfgdata16(pi, capoff + 8);
1304 
1305 	mme = msgctrl & PCIM_MSICTRL_MME_MASK;
1306 	pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
1307 	if (pi->pi_msi.enabled) {
1308 		pi->pi_msi.addr = addrlo;
1309 		pi->pi_msi.msg_data = msgdata;
1310 		pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
1311 	} else {
1312 		pi->pi_msi.maxmsgnum = 0;
1313 	}
1314 	pci_lintr_update(pi);
1315 }
1316 
1317 static void
1318 pciecap_cfgwrite(struct pci_devinst *pi, int capoff __unused, int offset,
1319     int bytes, uint32_t val)
1320 {
1321 
1322 	/* XXX don't write to the readonly parts */
1323 	CFGWRITE(pi, offset, val, bytes);
1324 }
1325 
1326 #define	PCIECAP_VERSION	0x2
1327 int
1328 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
1329 {
1330 	int err;
1331 	struct pciecap pciecap;
1332 
1333 	bzero(&pciecap, sizeof(pciecap));
1334 
1335 	/*
1336 	 * Use the integrated endpoint type for endpoints on a root complex bus.
1337 	 *
1338 	 * NB: bhyve currently only supports a single PCI bus that is the root
1339 	 * complex bus, so all endpoints are integrated.
1340 	 */
1341 	if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
1342 		type = PCIEM_TYPE_ROOT_INT_EP;
1343 
1344 	pciecap.capid = PCIY_EXPRESS;
1345 	pciecap.pcie_capabilities = PCIECAP_VERSION | type;
1346 	if (type != PCIEM_TYPE_ROOT_INT_EP) {
1347 		pciecap.link_capabilities = 0x411;	/* gen1, x1 */
1348 		pciecap.link_status = 0x11;		/* gen1, x1 */
1349 	}
1350 
1351 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
1352 	return (err);
1353 }
1354 
1355 /*
1356  * This function assumes that 'coff' is in the capabilities region of the
1357  * config space. A capoff parameter of zero will force a search for the
1358  * offset and type.
1359  */
1360 void
1361 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val,
1362     uint8_t capoff, int capid)
1363 {
1364 	uint8_t nextoff;
1365 
1366 	/* Do not allow un-aligned writes */
1367 	if ((offset & (bytes - 1)) != 0)
1368 		return;
1369 
1370 	if (capoff == 0) {
1371 		/* Find the capability that we want to update */
1372 		capoff = CAP_START_OFFSET;
1373 		while (1) {
1374 			nextoff = pci_get_cfgdata8(pi, capoff + 1);
1375 			if (nextoff == 0)
1376 				break;
1377 			if (offset >= capoff && offset < nextoff)
1378 				break;
1379 
1380 			capoff = nextoff;
1381 		}
1382 		assert(offset >= capoff);
1383 		capid = pci_get_cfgdata8(pi, capoff);
1384 	}
1385 
1386 	/*
1387 	 * Capability ID and Next Capability Pointer are readonly.
1388 	 * However, some o/s's do 4-byte writes that include these.
1389 	 * For this case, trim the write back to 2 bytes and adjust
1390 	 * the data.
1391 	 */
1392 	if (offset == capoff || offset == capoff + 1) {
1393 		if (offset == capoff && bytes == 4) {
1394 			bytes = 2;
1395 			offset += 2;
1396 			val >>= 16;
1397 		} else
1398 			return;
1399 	}
1400 
1401 	switch (capid) {
1402 	case PCIY_MSI:
1403 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1404 		break;
1405 	case PCIY_MSIX:
1406 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1407 		break;
1408 	case PCIY_EXPRESS:
1409 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1410 		break;
1411 	default:
1412 		break;
1413 	}
1414 }
1415 
1416 static int
1417 pci_emul_iscap(struct pci_devinst *pi, int offset)
1418 {
1419 	uint16_t sts;
1420 
1421 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1422 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1423 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1424 			return (1);
1425 	}
1426 	return (0);
1427 }
1428 
1429 static int
1430 pci_emul_fallback_handler(struct vcpu *vcpu __unused, int dir,
1431     uint64_t addr __unused, int size __unused, uint64_t *val,
1432     void *arg1 __unused, long arg2 __unused)
1433 {
1434 	/*
1435 	 * Ignore writes; return 0xff's for reads. The mem read code
1436 	 * will take care of truncating to the correct size.
1437 	 */
1438 	if (dir == MEM_F_READ) {
1439 		*val = 0xffffffffffffffff;
1440 	}
1441 
1442 	return (0);
1443 }
1444 
1445 static int
1446 pci_emul_ecfg_handler(struct vcpu *vcpu __unused, int dir, uint64_t addr,
1447     int bytes, uint64_t *val, void *arg1 __unused, long arg2 __unused)
1448 {
1449 	int bus, slot, func, coff, in;
1450 
1451 	coff = addr & 0xfff;
1452 	func = (addr >> 12) & 0x7;
1453 	slot = (addr >> 15) & 0x1f;
1454 	bus = (addr >> 20) & 0xff;
1455 	in = (dir == MEM_F_READ);
1456 	if (in)
1457 		*val = ~0UL;
1458 	pci_cfgrw(in, bus, slot, func, coff, bytes, (uint32_t *)val);
1459 	return (0);
1460 }
1461 
1462 uint64_t
1463 pci_ecfg_base(void)
1464 {
1465 
1466 	return (PCI_EMUL_ECFG_BASE);
1467 }
1468 
1469 static int
1470 init_bootorder(void)
1471 {
1472 	struct boot_device *device;
1473 	FILE *fp;
1474 	char *bootorder;
1475 	size_t bootorder_len;
1476 
1477 	if (TAILQ_EMPTY(&boot_devices))
1478 		return (0);
1479 
1480 	fp = open_memstream(&bootorder, &bootorder_len);
1481 	TAILQ_FOREACH(device, &boot_devices, boot_device_chain) {
1482 		fprintf(fp, "/pci@i0cf8/pci@%d,%d\n",
1483 		    device->pdi->pi_slot, device->pdi->pi_func);
1484 	}
1485 	fclose(fp);
1486 
1487 	return (qemu_fwcfg_add_file("bootorder", bootorder_len, bootorder));
1488 }
1489 
1490 #define	BUSIO_ROUNDUP		32
1491 #define	BUSMEM32_ROUNDUP	(1024 * 1024)
1492 #define	BUSMEM64_ROUNDUP	(512 * 1024 * 1024)
1493 
1494 int
1495 init_pci(struct vmctx *ctx)
1496 {
1497 	char node_name[sizeof("pci.XXX.XX.X")];
1498 	struct mem_range mr;
1499 	struct pci_devemu *pde;
1500 	struct businfo *bi;
1501 	struct slotinfo *si;
1502 	struct funcinfo *fi;
1503 	nvlist_t *nvl;
1504 	const char *emul;
1505 	size_t lowmem;
1506 	int bus, slot, func;
1507 	int error;
1508 
1509 	if (vm_get_lowmem_limit(ctx) > PCI_EMUL_MEMBASE32)
1510 		errx(EX_OSERR, "Invalid lowmem limit");
1511 
1512 	pci_emul_iobase = PCI_EMUL_IOBASE;
1513 	pci_emul_membase32 = PCI_EMUL_MEMBASE32;
1514 
1515 	pci_emul_membase64 = vm_get_highmem_base(ctx) +
1516 	    vm_get_highmem_size(ctx);
1517 	pci_emul_membase64 = roundup2(pci_emul_membase64, PCI_EMUL_MEMSIZE64);
1518 	pci_emul_memlim64 = pci_emul_membase64 + PCI_EMUL_MEMSIZE64;
1519 
1520 	TAILQ_INIT(&boot_devices);
1521 
1522 	for (bus = 0; bus < MAXBUSES; bus++) {
1523 		snprintf(node_name, sizeof(node_name), "pci.%d", bus);
1524 		nvl = find_config_node(node_name);
1525 		if (nvl == NULL)
1526 			continue;
1527 		pci_businfo[bus] = calloc(1, sizeof(struct businfo));
1528 		bi = pci_businfo[bus];
1529 
1530 		/*
1531 		 * Keep track of the i/o and memory resources allocated to
1532 		 * this bus.
1533 		 */
1534 		bi->iobase = pci_emul_iobase;
1535 		bi->membase32 = pci_emul_membase32;
1536 		bi->membase64 = pci_emul_membase64;
1537 
1538 		/* first run: init devices */
1539 		for (slot = 0; slot < MAXSLOTS; slot++) {
1540 			si = &bi->slotinfo[slot];
1541 			for (func = 0; func < MAXFUNCS; func++) {
1542 				fi = &si->si_funcs[func];
1543 				snprintf(node_name, sizeof(node_name),
1544 				    "pci.%d.%d.%d", bus, slot, func);
1545 				nvl = find_config_node(node_name);
1546 				if (nvl == NULL)
1547 					continue;
1548 
1549 				fi->fi_config = nvl;
1550 				emul = get_config_value_node(nvl, "device");
1551 				if (emul == NULL) {
1552 					EPRINTLN("pci slot %d:%d:%d: missing "
1553 					    "\"device\" value", bus, slot, func);
1554 					return (EINVAL);
1555 				}
1556 				pde = pci_emul_finddev(emul);
1557 				if (pde == NULL) {
1558 					EPRINTLN("pci slot %d:%d:%d: unknown "
1559 					    "device \"%s\"", bus, slot, func,
1560 					    emul);
1561 					return (EINVAL);
1562 				}
1563 				if (pde->pe_alias != NULL) {
1564 					EPRINTLN("pci slot %d:%d:%d: legacy "
1565 					    "device \"%s\", use \"%s\" instead",
1566 					    bus, slot, func, emul,
1567 					    pde->pe_alias);
1568 					return (EINVAL);
1569 				}
1570 				fi->fi_pde = pde;
1571 				error = pci_emul_init(ctx, pde, bus, slot,
1572 				    func, fi);
1573 				if (error)
1574 					return (error);
1575 			}
1576 		}
1577 
1578 		/* second run: assign BARs and free list */
1579 		struct pci_bar_allocation *bar;
1580 		struct pci_bar_allocation *bar_tmp;
1581 		TAILQ_FOREACH_SAFE(bar, &pci_bars, chain, bar_tmp) {
1582 			pci_emul_assign_bar(bar->pdi, bar->idx, bar->type,
1583 			    bar->size);
1584 			free(bar);
1585 		}
1586 		TAILQ_INIT(&pci_bars);
1587 
1588 		/*
1589 		 * Add some slop to the I/O and memory resources decoded by
1590 		 * this bus to give a guest some flexibility if it wants to
1591 		 * reprogram the BARs.
1592 		 */
1593 		pci_emul_iobase += BUSIO_ROUNDUP;
1594 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1595 		bi->iolimit = pci_emul_iobase;
1596 
1597 		pci_emul_membase32 += BUSMEM32_ROUNDUP;
1598 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1599 		    BUSMEM32_ROUNDUP);
1600 		bi->memlimit32 = pci_emul_membase32;
1601 
1602 		pci_emul_membase64 += BUSMEM64_ROUNDUP;
1603 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1604 		    BUSMEM64_ROUNDUP);
1605 		bi->memlimit64 = pci_emul_membase64;
1606 	}
1607 
1608 	/*
1609 	 * PCI backends are initialized before routing INTx interrupts
1610 	 * so that LPC devices are able to reserve ISA IRQs before
1611 	 * routing PIRQ pins.
1612 	 */
1613 	for (bus = 0; bus < MAXBUSES; bus++) {
1614 		if ((bi = pci_businfo[bus]) == NULL)
1615 			continue;
1616 
1617 		for (slot = 0; slot < MAXSLOTS; slot++) {
1618 			si = &bi->slotinfo[slot];
1619 			for (func = 0; func < MAXFUNCS; func++) {
1620 				fi = &si->si_funcs[func];
1621 				if (fi->fi_devi == NULL)
1622 					continue;
1623 				pci_lintr_route(fi->fi_devi);
1624 			}
1625 		}
1626 	}
1627 #ifdef __amd64__
1628 	lpc_pirq_routed();
1629 #endif
1630 
1631 	if ((error = init_bootorder()) != 0) {
1632 		warnx("%s: Unable to init bootorder", __func__);
1633 		return (error);
1634 	}
1635 
1636 	/*
1637 	 * The guest physical memory map looks like the following on amd64:
1638 	 * [0,		    lowmem)		guest system memory
1639 	 * [lowmem,	    0xC0000000)		memory hole (may be absent)
1640 	 * [0xC0000000,     0xE0000000)		PCI hole (32-bit BAR allocation)
1641 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1642 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1643 	 * [4GB,	    4GB + highmem)	guest system memory
1644 	 * [roundup(4GB + highmem, 32GB), ...)	PCI 64-bit BAR allocation
1645 	 *
1646 	 * On arm64 the guest physical memory map looks like this:
1647 	 * [0x0DF00000,	    0x10000000)		PCI I/O memory
1648 	 * [0xA0000000,	    0xE0000000)		PCI 32-bit BAR allocation
1649 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1650 	 * [4GB,	    4GB + highmem)	guest system memory
1651 	 * [roundup(4GB + highmem, 32GB), ...)	PCI 64-bit BAR allocation
1652 	 *
1653 	 * "lowmem" is guest memory below 0xC0000000.  amd64 guests provisioned
1654 	 * with less than 3GB of RAM will have no memory above the 4GB boundary.
1655 	 * System memory for arm64 guests is all above the 4GB boundary.
1656 	 */
1657 
1658 	/*
1659 	 * Accesses to memory addresses that are not allocated to system
1660 	 * memory or PCI devices return 0xff's.
1661 	 */
1662 	lowmem = vm_get_lowmem_size(ctx);
1663 	bzero(&mr, sizeof(struct mem_range));
1664 	mr.name = "PCI hole";
1665 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1666 	mr.base = lowmem;
1667 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1668 	mr.handler = pci_emul_fallback_handler;
1669 	error = register_mem_fallback(&mr);
1670 	assert(error == 0);
1671 
1672 	/* PCI extended config space */
1673 	bzero(&mr, sizeof(struct mem_range));
1674 	mr.name = "PCI ECFG";
1675 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1676 	mr.base = PCI_EMUL_ECFG_BASE;
1677 	mr.size = PCI_EMUL_ECFG_SIZE;
1678 	mr.handler = pci_emul_ecfg_handler;
1679 	error = register_mem(&mr);
1680 	assert(error == 0);
1681 
1682 	return (0);
1683 }
1684 
1685 #ifdef __amd64__
1686 static void
1687 pci_apic_prt_entry(int bus __unused, int slot, int pin, struct pci_irq *irq,
1688     void *arg __unused)
1689 {
1690 
1691 	dsdt_line("  Package ()");
1692 	dsdt_line("  {");
1693 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1694 	dsdt_line("    0x%02X,", pin - 1);
1695 	dsdt_line("    Zero,");
1696 	dsdt_line("    0x%X", irq->ioapic_irq);
1697 	dsdt_line("  },");
1698 }
1699 
1700 static void
1701 pci_pirq_prt_entry(int bus __unused, int slot, int pin, struct pci_irq *irq,
1702     void *arg __unused)
1703 {
1704 	char *name;
1705 
1706 	name = lpc_pirq_name(irq->pirq_pin);
1707 	if (name == NULL)
1708 		return;
1709 	dsdt_line("  Package ()");
1710 	dsdt_line("  {");
1711 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1712 	dsdt_line("    0x%02X,", pin - 1);
1713 	dsdt_line("    %s,", name);
1714 	dsdt_line("    0x00");
1715 	dsdt_line("  },");
1716 	free(name);
1717 }
1718 #endif
1719 
1720 /*
1721  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1722  * corresponding to each PCI bus.
1723  */
1724 static void
1725 pci_bus_write_dsdt(int bus)
1726 {
1727 	struct businfo *bi;
1728 	struct slotinfo *si;
1729 	struct pci_devinst *pi;
1730 	int func, slot;
1731 
1732 	/*
1733 	 * If there are no devices on this 'bus' then just return.
1734 	 */
1735 	if ((bi = pci_businfo[bus]) == NULL) {
1736 		/*
1737 		 * Bus 0 is special because it decodes the I/O ports used
1738 		 * for PCI config space access even if there are no devices
1739 		 * on it.
1740 		 */
1741 		if (bus != 0)
1742 			return;
1743 	}
1744 
1745 	dsdt_line("  Device (PC%02X)", bus);
1746 	dsdt_line("  {");
1747 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1748 
1749 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1750 	dsdt_line("    {");
1751 	dsdt_line("        Return (0x%08X)", bus);
1752 	dsdt_line("    }");
1753 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1754 	dsdt_line("    {");
1755 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1756 	    "MaxFixed, PosDecode,");
1757 	dsdt_line("        0x0000,             // Granularity");
1758 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1759 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1760 	dsdt_line("        0x0000,             // Translation Offset");
1761 	dsdt_line("        0x0001,             // Length");
1762 	dsdt_line("        ,, )");
1763 
1764 #ifdef __amd64__
1765 	if (bus == 0) {
1766 		dsdt_indent(3);
1767 		dsdt_fixed_ioport(0xCF8, 8);
1768 		dsdt_unindent(3);
1769 
1770 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1771 		    "PosDecode, EntireRange,");
1772 		dsdt_line("        0x0000,             // Granularity");
1773 		dsdt_line("        0x0000,             // Range Minimum");
1774 		dsdt_line("        0x0CF7,             // Range Maximum");
1775 		dsdt_line("        0x0000,             // Translation Offset");
1776 		dsdt_line("        0x0CF8,             // Length");
1777 		dsdt_line("        ,, , TypeStatic)");
1778 
1779 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1780 		    "PosDecode, EntireRange,");
1781 		dsdt_line("        0x0000,             // Granularity");
1782 		dsdt_line("        0x0D00,             // Range Minimum");
1783 		dsdt_line("        0x%04X,             // Range Maximum",
1784 		    PCI_EMUL_IOBASE - 1);
1785 		dsdt_line("        0x0000,             // Translation Offset");
1786 		dsdt_line("        0x%04X,             // Length",
1787 		    PCI_EMUL_IOBASE - 0x0D00);
1788 		dsdt_line("        ,, , TypeStatic)");
1789 
1790 		if (bi == NULL) {
1791 			dsdt_line("    })");
1792 			goto done;
1793 		}
1794 	}
1795 #endif
1796 	assert(bi != NULL);
1797 
1798 	/* i/o window */
1799 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1800 	    "PosDecode, EntireRange,");
1801 	dsdt_line("        0x0000,             // Granularity");
1802 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1803 	dsdt_line("        0x%04X,             // Range Maximum",
1804 	    bi->iolimit - 1);
1805 	dsdt_line("        0x0000,             // Translation Offset");
1806 	dsdt_line("        0x%04X,             // Length",
1807 	    bi->iolimit - bi->iobase);
1808 	dsdt_line("        ,, , TypeStatic)");
1809 
1810 	/* mmio window (32-bit) */
1811 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1812 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1813 	dsdt_line("        0x00000000,         // Granularity");
1814 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1815 	dsdt_line("        0x%08X,         // Range Maximum\n",
1816 	    bi->memlimit32 - 1);
1817 	dsdt_line("        0x00000000,         // Translation Offset");
1818 	dsdt_line("        0x%08X,         // Length\n",
1819 	    bi->memlimit32 - bi->membase32);
1820 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1821 
1822 	/* mmio window (64-bit) */
1823 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1824 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1825 	dsdt_line("        0x0000000000000000, // Granularity");
1826 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1827 	dsdt_line("        0x%016lX, // Range Maximum\n",
1828 	    bi->memlimit64 - 1);
1829 	dsdt_line("        0x0000000000000000, // Translation Offset");
1830 	dsdt_line("        0x%016lX, // Length\n",
1831 	    bi->memlimit64 - bi->membase64);
1832 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1833 	dsdt_line("    })");
1834 
1835 #ifdef __amd64__
1836 	if (pci_count_lintr(bus) != 0) {
1837 		dsdt_indent(2);
1838 		dsdt_line("Name (PPRT, Package ()");
1839 		dsdt_line("{");
1840 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1841 		dsdt_line("})");
1842 		dsdt_line("Name (APRT, Package ()");
1843 		dsdt_line("{");
1844 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1845 		dsdt_line("})");
1846 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1847 		dsdt_line("{");
1848 		dsdt_line("  If (PICM)");
1849 		dsdt_line("  {");
1850 		dsdt_line("    Return (APRT)");
1851 		dsdt_line("  }");
1852 		dsdt_line("  Else");
1853 		dsdt_line("  {");
1854 		dsdt_line("    Return (PPRT)");
1855 		dsdt_line("  }");
1856 		dsdt_line("}");
1857 		dsdt_unindent(2);
1858 	}
1859 #endif
1860 
1861 	dsdt_indent(2);
1862 	for (slot = 0; slot < MAXSLOTS; slot++) {
1863 		si = &bi->slotinfo[slot];
1864 		for (func = 0; func < MAXFUNCS; func++) {
1865 			pi = si->si_funcs[func].fi_devi;
1866 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1867 				pi->pi_d->pe_write_dsdt(pi);
1868 		}
1869 	}
1870 	dsdt_unindent(2);
1871 #ifdef __amd64__
1872 done:
1873 #endif
1874 	dsdt_line("  }");
1875 }
1876 
1877 void
1878 pci_write_dsdt(void)
1879 {
1880 	int bus;
1881 
1882 	dsdt_indent(1);
1883 	dsdt_line("Name (PICM, 0x00)");
1884 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1885 	dsdt_line("{");
1886 	dsdt_line("  Store (Arg0, PICM)");
1887 	dsdt_line("}");
1888 	dsdt_line("");
1889 	dsdt_line("Scope (_SB)");
1890 	dsdt_line("{");
1891 	for (bus = 0; bus < MAXBUSES; bus++)
1892 		pci_bus_write_dsdt(bus);
1893 	dsdt_line("}");
1894 	dsdt_unindent(1);
1895 }
1896 
1897 int
1898 pci_bus_configured(int bus)
1899 {
1900 	assert(bus >= 0 && bus < MAXBUSES);
1901 	return (pci_businfo[bus] != NULL);
1902 }
1903 
1904 int
1905 pci_msi_enabled(struct pci_devinst *pi)
1906 {
1907 	return (pi->pi_msi.enabled);
1908 }
1909 
1910 int
1911 pci_msi_maxmsgnum(struct pci_devinst *pi)
1912 {
1913 	if (pi->pi_msi.enabled)
1914 		return (pi->pi_msi.maxmsgnum);
1915 	else
1916 		return (0);
1917 }
1918 
1919 int
1920 pci_msix_enabled(struct pci_devinst *pi)
1921 {
1922 
1923 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1924 }
1925 
1926 void
1927 pci_generate_msix(struct pci_devinst *pi, int index)
1928 {
1929 	struct msix_table_entry *mte;
1930 
1931 	if (!pci_msix_enabled(pi))
1932 		return;
1933 
1934 	if (pi->pi_msix.function_mask)
1935 		return;
1936 
1937 	if (index >= pi->pi_msix.table_count)
1938 		return;
1939 
1940 	mte = &pi->pi_msix.table[index];
1941 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1942 		/* XXX Set PBA bit if interrupt is disabled */
1943 		vm_raise_msi(pi->pi_vmctx, mte->addr, mte->msg_data,
1944 		    pi->pi_bus, pi->pi_slot, pi->pi_func);
1945 	}
1946 }
1947 
1948 void
1949 pci_generate_msi(struct pci_devinst *pi, int index)
1950 {
1951 
1952 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1953 		vm_raise_msi(pi->pi_vmctx, pi->pi_msi.addr,
1954 		    pi->pi_msi.msg_data + index,
1955 		    pi->pi_bus, pi->pi_slot, pi->pi_func);
1956 	}
1957 }
1958 
1959 static bool
1960 pci_lintr_permitted(struct pci_devinst *pi)
1961 {
1962 	uint16_t cmd;
1963 
1964 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1965 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1966 		(cmd & PCIM_CMD_INTxDIS)));
1967 }
1968 
1969 void
1970 pci_lintr_request(struct pci_devinst *pi)
1971 {
1972 	struct businfo *bi;
1973 	struct slotinfo *si;
1974 	int bestpin, bestcount, pin;
1975 
1976 	bi = pci_businfo[pi->pi_bus];
1977 	assert(bi != NULL);
1978 
1979 	/*
1980 	 * Just allocate a pin from our slot.  The pin will be
1981 	 * assigned IRQs later when interrupts are routed.
1982 	 */
1983 	si = &bi->slotinfo[pi->pi_slot];
1984 	bestpin = 0;
1985 	bestcount = si->si_intpins[0].ii_count;
1986 	for (pin = 1; pin < 4; pin++) {
1987 		if (si->si_intpins[pin].ii_count < bestcount) {
1988 			bestpin = pin;
1989 			bestcount = si->si_intpins[pin].ii_count;
1990 		}
1991 	}
1992 
1993 	si->si_intpins[bestpin].ii_count++;
1994 	pi->pi_lintr.pin = bestpin + 1;
1995 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1996 }
1997 
1998 static void
1999 pci_lintr_route(struct pci_devinst *pi)
2000 {
2001 	struct businfo *bi;
2002 	struct intxinfo *ii;
2003 	struct pci_irq *irq;
2004 
2005 	if (pi->pi_lintr.pin == 0)
2006 		return;
2007 
2008 	bi = pci_businfo[pi->pi_bus];
2009 	assert(bi != NULL);
2010 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
2011 	irq = &ii->ii_irq;
2012 	pci_irq_route(pi, irq);
2013 	pi->pi_lintr.irq = *irq;
2014 	pci_set_cfgdata8(pi, PCIR_INTLINE, pci_irq_intline(irq));
2015 }
2016 
2017 void
2018 pci_lintr_assert(struct pci_devinst *pi)
2019 {
2020 
2021 	assert(pi->pi_lintr.pin > 0);
2022 
2023 	pthread_mutex_lock(&pi->pi_lintr.lock);
2024 	if (pi->pi_lintr.state == IDLE) {
2025 		if (pci_lintr_permitted(pi)) {
2026 			pi->pi_lintr.state = ASSERTED;
2027 			pci_irq_assert(pi);
2028 		} else
2029 			pi->pi_lintr.state = PENDING;
2030 	}
2031 	pthread_mutex_unlock(&pi->pi_lintr.lock);
2032 }
2033 
2034 void
2035 pci_lintr_deassert(struct pci_devinst *pi)
2036 {
2037 
2038 	assert(pi->pi_lintr.pin > 0);
2039 
2040 	pthread_mutex_lock(&pi->pi_lintr.lock);
2041 	if (pi->pi_lintr.state == ASSERTED) {
2042 		pi->pi_lintr.state = IDLE;
2043 		pci_irq_deassert(pi);
2044 	} else if (pi->pi_lintr.state == PENDING)
2045 		pi->pi_lintr.state = IDLE;
2046 	pthread_mutex_unlock(&pi->pi_lintr.lock);
2047 }
2048 
2049 static void
2050 pci_lintr_update(struct pci_devinst *pi)
2051 {
2052 
2053 	pthread_mutex_lock(&pi->pi_lintr.lock);
2054 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
2055 		pci_irq_deassert(pi);
2056 		pi->pi_lintr.state = PENDING;
2057 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
2058 		pi->pi_lintr.state = ASSERTED;
2059 		pci_irq_assert(pi);
2060 	}
2061 	pthread_mutex_unlock(&pi->pi_lintr.lock);
2062 }
2063 
2064 int
2065 pci_count_lintr(int bus)
2066 {
2067 	int count, slot, pin;
2068 	struct slotinfo *slotinfo;
2069 
2070 	count = 0;
2071 	if (pci_businfo[bus] != NULL) {
2072 		for (slot = 0; slot < MAXSLOTS; slot++) {
2073 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
2074 			for (pin = 0; pin < 4; pin++) {
2075 				if (slotinfo->si_intpins[pin].ii_count != 0)
2076 					count++;
2077 			}
2078 		}
2079 	}
2080 	return (count);
2081 }
2082 
2083 void
2084 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
2085 {
2086 	struct businfo *bi;
2087 	struct slotinfo *si;
2088 	struct intxinfo *ii;
2089 	int slot, pin;
2090 
2091 	if ((bi = pci_businfo[bus]) == NULL)
2092 		return;
2093 
2094 	for (slot = 0; slot < MAXSLOTS; slot++) {
2095 		si = &bi->slotinfo[slot];
2096 		for (pin = 0; pin < 4; pin++) {
2097 			ii = &si->si_intpins[pin];
2098 			if (ii->ii_count != 0)
2099 				cb(bus, slot, pin + 1, &ii->ii_irq, arg);
2100 		}
2101 	}
2102 }
2103 
2104 /*
2105  * Return 1 if the emulated device in 'slot' is a multi-function device.
2106  * Return 0 otherwise.
2107  */
2108 static int
2109 pci_emul_is_mfdev(int bus, int slot)
2110 {
2111 	struct businfo *bi;
2112 	struct slotinfo *si;
2113 	int f, numfuncs;
2114 
2115 	numfuncs = 0;
2116 	if ((bi = pci_businfo[bus]) != NULL) {
2117 		si = &bi->slotinfo[slot];
2118 		for (f = 0; f < MAXFUNCS; f++) {
2119 			if (si->si_funcs[f].fi_devi != NULL) {
2120 				numfuncs++;
2121 			}
2122 		}
2123 	}
2124 	return (numfuncs > 1);
2125 }
2126 
2127 /*
2128  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
2129  * whether or not is a multi-function being emulated in the pci 'slot'.
2130  */
2131 static void
2132 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
2133 {
2134 	int mfdev;
2135 
2136 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
2137 		mfdev = pci_emul_is_mfdev(bus, slot);
2138 		switch (bytes) {
2139 		case 1:
2140 		case 2:
2141 			*rv &= ~PCIM_MFDEV;
2142 			if (mfdev) {
2143 				*rv |= PCIM_MFDEV;
2144 			}
2145 			break;
2146 		case 4:
2147 			*rv &= ~(PCIM_MFDEV << 16);
2148 			if (mfdev) {
2149 				*rv |= (PCIM_MFDEV << 16);
2150 			}
2151 			break;
2152 		}
2153 	}
2154 }
2155 
2156 /*
2157  * Update device state in response to changes to the PCI command
2158  * register.
2159  */
2160 void
2161 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
2162 {
2163 	int i;
2164 	uint16_t changed, new;
2165 
2166 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);
2167 	changed = old ^ new;
2168 
2169 	/*
2170 	 * If the MMIO or I/O address space decoding has changed then
2171 	 * register/unregister all BARs that decode that address space.
2172 	 */
2173 	for (i = 0; i <= PCI_BARMAX_WITH_ROM; i++) {
2174 		switch (pi->pi_bar[i].type) {
2175 			case PCIBAR_NONE:
2176 			case PCIBAR_MEMHI64:
2177 				break;
2178 			case PCIBAR_IO:
2179 				/* I/O address space decoding changed? */
2180 				if (changed & PCIM_CMD_PORTEN) {
2181 					if (new & PCIM_CMD_PORTEN)
2182 						register_bar(pi, i);
2183 					else
2184 						unregister_bar(pi, i);
2185 				}
2186 				break;
2187 			case PCIBAR_ROM:
2188 				/* skip (un-)register of ROM if it disabled */
2189 				if (!romen(pi))
2190 					break;
2191 				/* fallthrough */
2192 			case PCIBAR_MEM32:
2193 			case PCIBAR_MEM64:
2194 				/* MMIO address space decoding changed? */
2195 				if (changed & PCIM_CMD_MEMEN) {
2196 					if (new & PCIM_CMD_MEMEN)
2197 						register_bar(pi, i);
2198 					else
2199 						unregister_bar(pi, i);
2200 				}
2201 				break;
2202 			default:
2203 				assert(0);
2204 		}
2205 	}
2206 
2207 	/*
2208 	 * If INTx has been unmasked and is pending, assert the
2209 	 * interrupt.
2210 	 */
2211 	pci_lintr_update(pi);
2212 }
2213 
2214 static void
2215 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
2216 {
2217 	int rshift;
2218 	uint32_t cmd, old, readonly;
2219 
2220 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
2221 
2222 	/*
2223 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
2224 	 *
2225 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
2226 	 * 'write 1 to clear'. However these bits are not set to '1' by
2227 	 * any device emulation so it is simpler to treat them as readonly.
2228 	 */
2229 	rshift = (coff & 0x3) * 8;
2230 	readonly = 0xFFFFF880 >> rshift;
2231 
2232 	old = CFGREAD(pi, coff, bytes);
2233 	new &= ~readonly;
2234 	new |= (old & readonly);
2235 	CFGWRITE(pi, coff, new, bytes);			/* update config */
2236 
2237 	pci_emul_cmd_changed(pi, cmd);
2238 }
2239 
2240 static void
2241 pci_cfgrw(int in, int bus, int slot, int func, int coff, int bytes,
2242     uint32_t *valp)
2243 {
2244 	struct businfo *bi;
2245 	struct slotinfo *si;
2246 	struct pci_devinst *pi;
2247 	struct pci_devemu *pe;
2248 	int idx, needcfg;
2249 	uint64_t addr, bar, mask;
2250 
2251 	if ((bi = pci_businfo[bus]) != NULL) {
2252 		si = &bi->slotinfo[slot];
2253 		pi = si->si_funcs[func].fi_devi;
2254 	} else
2255 		pi = NULL;
2256 
2257 	/*
2258 	 * Just return if there is no device at this slot:func or if the
2259 	 * guest is doing an un-aligned access.
2260 	 */
2261 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
2262 	    (coff & (bytes - 1)) != 0) {
2263 		if (in)
2264 			*valp = 0xffffffff;
2265 		return;
2266 	}
2267 
2268 	/*
2269 	 * Ignore all writes beyond the standard config space and return all
2270 	 * ones on reads.
2271 	 */
2272 	if (coff >= PCI_REGMAX + 1) {
2273 		if (in) {
2274 			*valp = 0xffffffff;
2275 			/*
2276 			 * Extended capabilities begin at offset 256 in config
2277 			 * space. Absence of extended capabilities is signaled
2278 			 * with all 0s in the extended capability header at
2279 			 * offset 256.
2280 			 */
2281 			if (coff <= PCI_REGMAX + 4)
2282 				*valp = 0x00000000;
2283 		}
2284 		return;
2285 	}
2286 
2287 	pe = pi->pi_d;
2288 
2289 	/*
2290 	 * Config read
2291 	 */
2292 	if (in) {
2293 		/* Let the device emulation override the default handler */
2294 		if (pe->pe_cfgread != NULL) {
2295 			needcfg = pe->pe_cfgread(pi, coff, bytes, valp);
2296 		} else {
2297 			needcfg = 1;
2298 		}
2299 
2300 		if (needcfg)
2301 			*valp = CFGREAD(pi, coff, bytes);
2302 
2303 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, valp);
2304 	} else {
2305 		/* Let the device emulation override the default handler */
2306 		if (pe->pe_cfgwrite != NULL &&
2307 		    (*pe->pe_cfgwrite)(pi, coff, bytes, *valp) == 0)
2308 			return;
2309 
2310 		/*
2311 		 * Special handling for write to BAR and ROM registers
2312 		 */
2313 		if (is_pcir_bar(coff) || is_pcir_bios(coff)) {
2314 			/*
2315 			 * Ignore writes to BAR registers that are not
2316 			 * 4-byte aligned.
2317 			 */
2318 			if (bytes != 4 || (coff & 0x3) != 0)
2319 				return;
2320 
2321 			if (is_pcir_bar(coff)) {
2322 				idx = (coff - PCIR_BAR(0)) / 4;
2323 			} else if (is_pcir_bios(coff)) {
2324 				idx = PCI_ROM_IDX;
2325 			} else {
2326 				errx(4, "%s: invalid BAR offset %d", __func__,
2327 				    coff);
2328 			}
2329 
2330 			mask = ~(pi->pi_bar[idx].size - 1);
2331 			switch (pi->pi_bar[idx].type) {
2332 			case PCIBAR_NONE:
2333 				pi->pi_bar[idx].addr = bar = 0;
2334 				break;
2335 			case PCIBAR_IO:
2336 				addr = *valp & mask;
2337 #if defined(PCI_EMUL_IOMASK)
2338 				addr &= PCI_EMUL_IOMASK;
2339 #endif
2340 				bar = addr | pi->pi_bar[idx].lobits;
2341 				/*
2342 				 * Register the new BAR value for interception
2343 				 */
2344 				if (addr != pi->pi_bar[idx].addr) {
2345 					update_bar_address(pi, addr, idx,
2346 							   PCIBAR_IO);
2347 				}
2348 				break;
2349 			case PCIBAR_MEM32:
2350 				addr = bar = *valp & mask;
2351 				bar |= pi->pi_bar[idx].lobits;
2352 				if (addr != pi->pi_bar[idx].addr) {
2353 					update_bar_address(pi, addr, idx,
2354 							   PCIBAR_MEM32);
2355 				}
2356 				break;
2357 			case PCIBAR_MEM64:
2358 				addr = bar = *valp & mask;
2359 				bar |= pi->pi_bar[idx].lobits;
2360 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
2361 					update_bar_address(pi, addr, idx,
2362 							   PCIBAR_MEM64);
2363 				}
2364 				break;
2365 			case PCIBAR_MEMHI64:
2366 				mask = ~(pi->pi_bar[idx - 1].size - 1);
2367 				addr = ((uint64_t)*valp << 32) & mask;
2368 				bar = addr >> 32;
2369 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
2370 					update_bar_address(pi, addr, idx - 1,
2371 							   PCIBAR_MEMHI64);
2372 				}
2373 				break;
2374 			case PCIBAR_ROM:
2375 				addr = bar = *valp & mask;
2376 				if (memen(pi) && romen(pi)) {
2377 					unregister_bar(pi, idx);
2378 				}
2379 				pi->pi_bar[idx].addr = addr;
2380 				pi->pi_bar[idx].lobits = *valp &
2381 				    PCIM_BIOS_ENABLE;
2382 				/* romen could have changed it value */
2383 				if (memen(pi) && romen(pi)) {
2384 					register_bar(pi, idx);
2385 				}
2386 				bar |= pi->pi_bar[idx].lobits;
2387 				break;
2388 			default:
2389 				assert(0);
2390 			}
2391 			pci_set_cfgdata32(pi, coff, bar);
2392 
2393 		} else if (pci_emul_iscap(pi, coff)) {
2394 			pci_emul_capwrite(pi, coff, bytes, *valp, 0, 0);
2395 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
2396 			pci_emul_cmdsts_write(pi, coff, *valp, bytes);
2397 		} else {
2398 			CFGWRITE(pi, coff, *valp, bytes);
2399 		}
2400 	}
2401 }
2402 
2403 #ifdef __amd64__
2404 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
2405 
2406 static int
2407 pci_emul_cfgaddr(struct vmctx *ctx __unused, int in,
2408     int port __unused, int bytes, uint32_t *eax, void *arg __unused)
2409 {
2410 	uint32_t x;
2411 
2412 	if (bytes != 4) {
2413 		if (in)
2414 			*eax = (bytes == 2) ? 0xffff : 0xff;
2415 		return (0);
2416 	}
2417 
2418 	if (in) {
2419 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
2420 		if (cfgenable)
2421 			x |= CONF1_ENABLE;
2422 		*eax = x;
2423 	} else {
2424 		x = *eax;
2425 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
2426 		cfgoff = (x & PCI_REGMAX) & ~0x03;
2427 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
2428 		cfgslot = (x >> 11) & PCI_SLOTMAX;
2429 		cfgbus = (x >> 16) & PCI_BUSMAX;
2430 	}
2431 
2432 	return (0);
2433 }
2434 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
2435 
2436 static int
2437 pci_emul_cfgdata(struct vmctx *ctx __unused, int in, int port,
2438     int bytes, uint32_t *eax, void *arg __unused)
2439 {
2440 	int coff;
2441 
2442 	assert(bytes == 1 || bytes == 2 || bytes == 4);
2443 
2444 	coff = cfgoff + (port - CONF1_DATA_PORT);
2445 	if (cfgenable) {
2446 		pci_cfgrw(in, cfgbus, cfgslot, cfgfunc, coff, bytes, eax);
2447 	} else {
2448 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
2449 		if (in)
2450 			*eax = 0xffffffff;
2451 	}
2452 	return (0);
2453 }
2454 
2455 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
2456 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
2457 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
2458 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
2459 #endif
2460 
2461 #ifdef BHYVE_SNAPSHOT
2462 /*
2463  * Saves/restores PCI device emulated state. Returns 0 on success.
2464  */
2465 static int
2466 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta)
2467 {
2468 	struct pci_devinst *pi;
2469 	int i;
2470 	int ret;
2471 
2472 	pi = meta->dev_data;
2473 
2474 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done);
2475 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done);
2476 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done);
2477 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done);
2478 
2479 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done);
2480 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done);
2481 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done);
2482 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done);
2483 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done);
2484 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done);
2485 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done);
2486 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done);
2487 
2488 	SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata),
2489 			      meta, ret, done);
2490 
2491 	for (i = 0; i < (int)nitems(pi->pi_bar); i++) {
2492 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done);
2493 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done);
2494 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done);
2495 	}
2496 
2497 	/* Restore MSI-X table. */
2498 	for (i = 0; i < pi->pi_msix.table_count; i++) {
2499 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr,
2500 				      meta, ret, done);
2501 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data,
2502 				      meta, ret, done);
2503 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control,
2504 				      meta, ret, done);
2505 	}
2506 
2507 done:
2508 	return (ret);
2509 }
2510 
2511 int
2512 pci_snapshot(struct vm_snapshot_meta *meta)
2513 {
2514 	struct pci_devemu *pde;
2515 	struct pci_devinst *pdi;
2516 	int ret;
2517 
2518 	assert(meta->dev_name != NULL);
2519 
2520 	pdi = meta->dev_data;
2521 	pde = pdi->pi_d;
2522 
2523 	if (pde->pe_snapshot == NULL)
2524 		return (ENOTSUP);
2525 
2526 	ret = pci_snapshot_pci_dev(meta);
2527 	if (ret == 0)
2528 		ret = (*pde->pe_snapshot)(meta);
2529 
2530 	return (ret);
2531 }
2532 
2533 int
2534 pci_pause(struct pci_devinst *pdi)
2535 {
2536 	struct pci_devemu *pde = pdi->pi_d;
2537 
2538 	if (pde->pe_pause == NULL) {
2539 		/* The pause/resume functionality is optional. */
2540 		return (0);
2541 	}
2542 
2543 	return (*pde->pe_pause)(pdi);
2544 }
2545 
2546 int
2547 pci_resume(struct pci_devinst *pdi)
2548 {
2549 	struct pci_devemu *pde = pdi->pi_d;
2550 
2551 	if (pde->pe_resume == NULL) {
2552 		/* The pause/resume functionality is optional. */
2553 		return (0);
2554 	}
2555 
2556 	return (*pde->pe_resume)(pdi);
2557 }
2558 #endif
2559 
2560 #define PCI_EMUL_TEST
2561 #ifdef PCI_EMUL_TEST
2562 /*
2563  * Define a dummy test device
2564  */
2565 #define DIOSZ	8
2566 #define DMEMSZ	4096
2567 struct pci_emul_dsoftc {
2568 	uint8_t   ioregs[DIOSZ];
2569 	uint8_t	  memregs[2][DMEMSZ];
2570 };
2571 
2572 #define	PCI_EMUL_MSI_MSGS	 4
2573 #define	PCI_EMUL_MSIX_MSGS	16
2574 
2575 static int
2576 pci_emul_dinit(struct pci_devinst *pi, nvlist_t *nvl __unused)
2577 {
2578 	int error;
2579 	struct pci_emul_dsoftc *sc;
2580 
2581 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
2582 
2583 	pi->pi_arg = sc;
2584 
2585 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
2586 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
2587 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
2588 
2589 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
2590 	assert(error == 0);
2591 
2592 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
2593 	assert(error == 0);
2594 
2595 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2596 	assert(error == 0);
2597 
2598 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2599 	assert(error == 0);
2600 
2601 	return (0);
2602 }
2603 
2604 static void
2605 pci_emul_diow(struct pci_devinst *pi, int baridx, uint64_t offset, int size,
2606     uint64_t value)
2607 {
2608 	int i;
2609 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2610 
2611 	if (baridx == 0) {
2612 		if (offset + size > DIOSZ) {
2613 			printf("diow: iow too large, offset %ld size %d\n",
2614 			       offset, size);
2615 			return;
2616 		}
2617 
2618 		if (size == 1) {
2619 			sc->ioregs[offset] = value & 0xff;
2620 		} else if (size == 2) {
2621 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2622 		} else if (size == 4) {
2623 			*(uint32_t *)&sc->ioregs[offset] = value;
2624 		} else {
2625 			printf("diow: iow unknown size %d\n", size);
2626 		}
2627 
2628 		/*
2629 		 * Special magic value to generate an interrupt
2630 		 */
2631 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2632 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2633 
2634 		if (value == 0xabcdef) {
2635 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2636 				pci_generate_msi(pi, i);
2637 		}
2638 	}
2639 
2640 	if (baridx == 1 || baridx == 2) {
2641 		if (offset + size > DMEMSZ) {
2642 			printf("diow: memw too large, offset %ld size %d\n",
2643 			       offset, size);
2644 			return;
2645 		}
2646 
2647 		i = baridx - 1;		/* 'memregs' index */
2648 
2649 		if (size == 1) {
2650 			sc->memregs[i][offset] = value;
2651 		} else if (size == 2) {
2652 			*(uint16_t *)&sc->memregs[i][offset] = value;
2653 		} else if (size == 4) {
2654 			*(uint32_t *)&sc->memregs[i][offset] = value;
2655 		} else if (size == 8) {
2656 			*(uint64_t *)&sc->memregs[i][offset] = value;
2657 		} else {
2658 			printf("diow: memw unknown size %d\n", size);
2659 		}
2660 
2661 		/*
2662 		 * magic interrupt ??
2663 		 */
2664 	}
2665 
2666 	if (baridx > 2 || baridx < 0) {
2667 		printf("diow: unknown bar idx %d\n", baridx);
2668 	}
2669 }
2670 
2671 static uint64_t
2672 pci_emul_dior(struct pci_devinst *pi, int baridx, uint64_t offset, int size)
2673 {
2674 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2675 	uint32_t value;
2676 	int i;
2677 
2678 	if (baridx == 0) {
2679 		if (offset + size > DIOSZ) {
2680 			printf("dior: ior too large, offset %ld size %d\n",
2681 			       offset, size);
2682 			return (0);
2683 		}
2684 
2685 		value = 0;
2686 		if (size == 1) {
2687 			value = sc->ioregs[offset];
2688 		} else if (size == 2) {
2689 			value = *(uint16_t *) &sc->ioregs[offset];
2690 		} else if (size == 4) {
2691 			value = *(uint32_t *) &sc->ioregs[offset];
2692 		} else {
2693 			printf("dior: ior unknown size %d\n", size);
2694 		}
2695 	}
2696 
2697 	if (baridx == 1 || baridx == 2) {
2698 		if (offset + size > DMEMSZ) {
2699 			printf("dior: memr too large, offset %ld size %d\n",
2700 			       offset, size);
2701 			return (0);
2702 		}
2703 
2704 		i = baridx - 1;		/* 'memregs' index */
2705 
2706 		if (size == 1) {
2707 			value = sc->memregs[i][offset];
2708 		} else if (size == 2) {
2709 			value = *(uint16_t *) &sc->memregs[i][offset];
2710 		} else if (size == 4) {
2711 			value = *(uint32_t *) &sc->memregs[i][offset];
2712 		} else if (size == 8) {
2713 			value = *(uint64_t *) &sc->memregs[i][offset];
2714 		} else {
2715 			printf("dior: ior unknown size %d\n", size);
2716 		}
2717 	}
2718 
2719 
2720 	if (baridx > 2 || baridx < 0) {
2721 		printf("dior: unknown bar idx %d\n", baridx);
2722 		return (0);
2723 	}
2724 
2725 	return (value);
2726 }
2727 
2728 #ifdef BHYVE_SNAPSHOT
2729 struct pci_devinst *
2730 pci_next(const struct pci_devinst *cursor)
2731 {
2732 	unsigned bus = 0, slot = 0, func = 0;
2733 	struct businfo *bi;
2734 	struct slotinfo *si;
2735 	struct funcinfo *fi;
2736 
2737 	bus = cursor ? cursor->pi_bus : 0;
2738 	slot = cursor ? cursor->pi_slot : 0;
2739 	func = cursor ? (cursor->pi_func + 1) : 0;
2740 
2741 	for (; bus < MAXBUSES; bus++) {
2742 		if ((bi = pci_businfo[bus]) == NULL)
2743 			continue;
2744 
2745 		if (slot >= MAXSLOTS)
2746 			slot = 0;
2747 
2748 		for (; slot < MAXSLOTS; slot++) {
2749 			si = &bi->slotinfo[slot];
2750 			if (func >= MAXFUNCS)
2751 				func = 0;
2752 			for (; func < MAXFUNCS; func++) {
2753 				fi = &si->si_funcs[func];
2754 				if (fi->fi_devi == NULL)
2755 					continue;
2756 
2757 				return (fi->fi_devi);
2758 			}
2759 		}
2760 	}
2761 
2762 	return (NULL);
2763 }
2764 
2765 static int
2766 pci_emul_snapshot(struct vm_snapshot_meta *meta __unused)
2767 {
2768 	return (0);
2769 }
2770 #endif
2771 
2772 static const struct pci_devemu pci_dummy = {
2773 	.pe_emu = "dummy",
2774 	.pe_init = pci_emul_dinit,
2775 	.pe_barwrite = pci_emul_diow,
2776 	.pe_barread = pci_emul_dior,
2777 #ifdef BHYVE_SNAPSHOT
2778 	.pe_snapshot = pci_emul_snapshot,
2779 #endif
2780 };
2781 PCI_EMUL_SET(pci_dummy);
2782 
2783 #endif /* PCI_EMUL_TEST */
2784