xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision ffaed739)
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 #include "pci_passthru.h"
65 
66 #define CONF1_ADDR_PORT	   0x0cf8
67 #define CONF1_DATA_PORT	   0x0cfc
68 
69 #define CONF1_ENABLE	   0x80000000ul
70 
71 #define	MAXBUSES	(PCI_BUSMAX + 1)
72 #define MAXSLOTS	(PCI_SLOTMAX + 1)
73 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
74 
75 #define GB		(1024 * 1024 * 1024UL)
76 
77 struct funcinfo {
78 	nvlist_t *fi_config;
79 	struct pci_devemu *fi_pde;
80 	struct pci_devinst *fi_devi;
81 };
82 
83 struct intxinfo {
84 	int	ii_count;
85 	int	ii_pirq_pin;
86 	int	ii_ioapic_irq;
87 };
88 
89 struct slotinfo {
90 	struct intxinfo si_intpins[4];
91 	struct funcinfo si_funcs[MAXFUNCS];
92 };
93 
94 struct businfo {
95 	uint16_t iobase, iolimit;		/* I/O window */
96 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
97 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
98 	struct slotinfo slotinfo[MAXSLOTS];
99 };
100 
101 static struct businfo *pci_businfo[MAXBUSES];
102 
103 SET_DECLARE(pci_devemu_set, struct pci_devemu);
104 
105 static uint64_t pci_emul_iobase;
106 static uint8_t *pci_emul_rombase;
107 static uint64_t pci_emul_romoffset;
108 static uint8_t *pci_emul_romlim;
109 static uint64_t pci_emul_membase32;
110 static uint64_t pci_emul_membase64;
111 static uint64_t pci_emul_memlim64;
112 
113 struct pci_bar_allocation {
114 	TAILQ_ENTRY(pci_bar_allocation) chain;
115 	struct pci_devinst *pdi;
116 	int idx;
117 	enum pcibar_type type;
118 	uint64_t size;
119 };
120 
121 static TAILQ_HEAD(pci_bar_list, pci_bar_allocation) pci_bars =
122     TAILQ_HEAD_INITIALIZER(pci_bars);
123 
124 #define	PCI_EMUL_IOBASE		0x2000
125 #define	PCI_EMUL_IOLIMIT	0x10000
126 
127 #define PCI_EMUL_ROMSIZE 0x10000000
128 
129 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
130 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
131 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
132 
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 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
139 #define PCI_EMUL_MEMSIZE64	(32*GB)
140 
141 static struct pci_devemu *pci_emul_finddev(const char *name);
142 static void pci_lintr_route(struct pci_devinst *pi);
143 static void pci_lintr_update(struct pci_devinst *pi);
144 static void pci_cfgrw(int in, int bus, int slot, int func, int coff,
145     int bytes, uint32_t *val);
146 
147 static __inline void
148 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
149 {
150 
151 	if (bytes == 1)
152 		pci_set_cfgdata8(pi, coff, val);
153 	else if (bytes == 2)
154 		pci_set_cfgdata16(pi, coff, val);
155 	else
156 		pci_set_cfgdata32(pi, coff, val);
157 }
158 
159 static __inline uint32_t
160 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
161 {
162 
163 	if (bytes == 1)
164 		return (pci_get_cfgdata8(pi, coff));
165 	else if (bytes == 2)
166 		return (pci_get_cfgdata16(pi, coff));
167 	else
168 		return (pci_get_cfgdata32(pi, coff));
169 }
170 
171 static int
172 is_pcir_bar(int coff)
173 {
174 	return (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1));
175 }
176 
177 static int
178 is_pcir_bios(int coff)
179 {
180 	return (coff >= PCIR_BIOS && coff < PCIR_BIOS + 4);
181 }
182 
183 /*
184  * I/O access
185  */
186 
187 /*
188  * Slot options are in the form:
189  *
190  *  <bus>:<slot>:<func>,<emul>[,<config>]
191  *  <slot>[:<func>],<emul>[,<config>]
192  *
193  *  slot is 0..31
194  *  func is 0..7
195  *  emul is a string describing the type of PCI device e.g. virtio-net
196  *  config is an optional string, depending on the device, that can be
197  *  used for configuration.
198  *   Examples are:
199  *     1,virtio-net,tap0
200  *     3:0,dummy
201  */
202 static void
203 pci_parse_slot_usage(char *aopt)
204 {
205 
206 	EPRINTLN("Invalid PCI slot info field \"%s\"", aopt);
207 }
208 
209 /*
210  * Helper function to parse a list of comma-separated options where
211  * each option is formatted as "name[=value]".  If no value is
212  * provided, the option is treated as a boolean and is given a value
213  * of true.
214  */
215 int
216 pci_parse_legacy_config(nvlist_t *nvl, const char *opt)
217 {
218 	char *config, *name, *tofree, *value;
219 
220 	if (opt == NULL)
221 		return (0);
222 
223 	config = tofree = strdup(opt);
224 	while ((name = strsep(&config, ",")) != NULL) {
225 		value = strchr(name, '=');
226 		if (value != NULL) {
227 			*value = '\0';
228 			value++;
229 			set_config_value_node(nvl, name, value);
230 		} else
231 			set_config_bool_node(nvl, name, true);
232 	}
233 	free(tofree);
234 	return (0);
235 }
236 
237 /*
238  * PCI device configuration is stored in MIBs that encode the device's
239  * location:
240  *
241  * pci.<bus>.<slot>.<func>
242  *
243  * Where "bus", "slot", and "func" are all decimal values without
244  * leading zeroes.  Each valid device must have a "device" node which
245  * identifies the driver model of the device.
246  *
247  * Device backends can provide a parser for the "config" string.  If
248  * a custom parser is not provided, pci_parse_legacy_config() is used
249  * to parse the string.
250  */
251 int
252 pci_parse_slot(char *opt)
253 {
254 	char node_name[sizeof("pci.XXX.XX.X")];
255 	struct pci_devemu *pde;
256 	char *emul, *config, *str, *cp;
257 	int error, bnum, snum, fnum;
258 	nvlist_t *nvl;
259 
260 	error = -1;
261 	str = strdup(opt);
262 
263 	emul = config = NULL;
264 	if ((cp = strchr(str, ',')) != NULL) {
265 		*cp = '\0';
266 		emul = cp + 1;
267 		if ((cp = strchr(emul, ',')) != NULL) {
268 			*cp = '\0';
269 			config = cp + 1;
270 		}
271 	} else {
272 		pci_parse_slot_usage(opt);
273 		goto done;
274 	}
275 
276 	/* <bus>:<slot>:<func> */
277 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
278 		bnum = 0;
279 		/* <slot>:<func> */
280 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
281 			fnum = 0;
282 			/* <slot> */
283 			if (sscanf(str, "%d", &snum) != 1) {
284 				snum = -1;
285 			}
286 		}
287 	}
288 
289 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
290 	    fnum < 0 || fnum >= MAXFUNCS) {
291 		pci_parse_slot_usage(opt);
292 		goto done;
293 	}
294 
295 	pde = pci_emul_finddev(emul);
296 	if (pde == NULL) {
297 		EPRINTLN("pci slot %d:%d:%d: unknown device \"%s\"", bnum, snum,
298 		    fnum, emul);
299 		goto done;
300 	}
301 
302 	snprintf(node_name, sizeof(node_name), "pci.%d.%d.%d", bnum, snum,
303 	    fnum);
304 	nvl = find_config_node(node_name);
305 	if (nvl != NULL) {
306 		EPRINTLN("pci slot %d:%d:%d already occupied!", bnum, snum,
307 		    fnum);
308 		goto done;
309 	}
310 	nvl = create_config_node(node_name);
311 	if (pde->pe_alias != NULL)
312 		set_config_value_node(nvl, "device", pde->pe_alias);
313 	else
314 		set_config_value_node(nvl, "device", pde->pe_emu);
315 
316 	if (pde->pe_legacy_config != NULL)
317 		error = pde->pe_legacy_config(nvl, config);
318 	else
319 		error = pci_parse_legacy_config(nvl, config);
320 done:
321 	free(str);
322 	return (error);
323 }
324 
325 void
326 pci_print_supported_devices(void)
327 {
328 	struct pci_devemu **pdpp, *pdp;
329 
330 	SET_FOREACH(pdpp, pci_devemu_set) {
331 		pdp = *pdpp;
332 		printf("%s\n", pdp->pe_emu);
333 	}
334 }
335 
336 uint32_t
337 pci_config_read_reg(const struct pcisel *const host_sel, nvlist_t *nvl,
338     const uint32_t reg, const uint8_t size, const uint32_t def)
339 {
340 	const char *config;
341 	const nvlist_t *pci_regs;
342 
343 	assert(size == 1 || size == 2 || size == 4);
344 
345 	pci_regs = find_relative_config_node(nvl, "pcireg");
346 	if (pci_regs == NULL) {
347 		return def;
348 	}
349 
350 	switch (reg) {
351 	case PCIR_DEVICE:
352 		config = get_config_value_node(pci_regs, "device");
353 		break;
354 	case PCIR_VENDOR:
355 		config = get_config_value_node(pci_regs, "vendor");
356 		break;
357 	case PCIR_REVID:
358 		config = get_config_value_node(pci_regs, "revid");
359 		break;
360 	case PCIR_SUBVEND_0:
361 		config = get_config_value_node(pci_regs, "subvendor");
362 		break;
363 	case PCIR_SUBDEV_0:
364 		config = get_config_value_node(pci_regs, "subdevice");
365 		break;
366 	default:
367 		return (-1);
368 	}
369 
370 	if (config == NULL) {
371 		return def;
372 	} else if (host_sel != NULL && strcmp(config, "host") == 0) {
373 		return read_config(host_sel, reg, size);
374 	} else {
375 		return strtol(config, NULL, 16);
376 	}
377 }
378 
379 static int
380 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
381 {
382 
383 	if (offset < pi->pi_msix.pba_offset)
384 		return (0);
385 
386 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
387 		return (0);
388 	}
389 
390 	return (1);
391 }
392 
393 int
394 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
395 		     uint64_t value)
396 {
397 	int msix_entry_offset;
398 	int tab_index;
399 	char *dest;
400 
401 	/* support only 4 or 8 byte writes */
402 	if (size != 4 && size != 8)
403 		return (-1);
404 
405 	/*
406 	 * Return if table index is beyond what device supports
407 	 */
408 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
409 	if (tab_index >= pi->pi_msix.table_count)
410 		return (-1);
411 
412 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
413 
414 	/* support only aligned writes */
415 	if ((msix_entry_offset % size) != 0)
416 		return (-1);
417 
418 	dest = (char *)(pi->pi_msix.table + tab_index);
419 	dest += msix_entry_offset;
420 
421 	if (size == 4)
422 		*((uint32_t *)dest) = value;
423 	else
424 		*((uint64_t *)dest) = value;
425 
426 	return (0);
427 }
428 
429 uint64_t
430 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
431 {
432 	char *dest;
433 	int msix_entry_offset;
434 	int tab_index;
435 	uint64_t retval = ~0;
436 
437 	/*
438 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
439 	 * table but we also allow 1 byte access to accommodate reads from
440 	 * ddb.
441 	 */
442 	if (size != 1 && size != 4 && size != 8)
443 		return (retval);
444 
445 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
446 
447 	/* support only aligned reads */
448 	if ((msix_entry_offset % size) != 0) {
449 		return (retval);
450 	}
451 
452 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
453 
454 	if (tab_index < pi->pi_msix.table_count) {
455 		/* valid MSI-X Table access */
456 		dest = (char *)(pi->pi_msix.table + tab_index);
457 		dest += msix_entry_offset;
458 
459 		if (size == 1)
460 			retval = *((uint8_t *)dest);
461 		else if (size == 4)
462 			retval = *((uint32_t *)dest);
463 		else
464 			retval = *((uint64_t *)dest);
465 	} else if (pci_valid_pba_offset(pi, offset)) {
466 		/* return 0 for PBA access */
467 		retval = 0;
468 	}
469 
470 	return (retval);
471 }
472 
473 int
474 pci_msix_table_bar(struct pci_devinst *pi)
475 {
476 
477 	if (pi->pi_msix.table != NULL)
478 		return (pi->pi_msix.table_bar);
479 	else
480 		return (-1);
481 }
482 
483 int
484 pci_msix_pba_bar(struct pci_devinst *pi)
485 {
486 
487 	if (pi->pi_msix.table != NULL)
488 		return (pi->pi_msix.pba_bar);
489 	else
490 		return (-1);
491 }
492 
493 static int
494 pci_emul_io_handler(struct vmctx *ctx __unused, int in, int port,
495     int bytes, uint32_t *eax, void *arg)
496 {
497 	struct pci_devinst *pdi = arg;
498 	struct pci_devemu *pe = pdi->pi_d;
499 	uint64_t offset;
500 	int i;
501 
502 	assert(port >= 0);
503 
504 	for (i = 0; i <= PCI_BARMAX; i++) {
505 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
506 		    (uint64_t)port >= pdi->pi_bar[i].addr &&
507 		    (uint64_t)port + bytes <=
508 		    pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
509 			offset = port - pdi->pi_bar[i].addr;
510 			if (in)
511 				*eax = (*pe->pe_barread)(pdi, i,
512 							 offset, bytes);
513 			else
514 				(*pe->pe_barwrite)(pdi, i, offset,
515 						   bytes, *eax);
516 			return (0);
517 		}
518 	}
519 	return (-1);
520 }
521 
522 static int
523 pci_emul_mem_handler(struct vcpu *vcpu __unused, int dir,
524     uint64_t addr, int size, uint64_t *val, void *arg1, long arg2)
525 {
526 	struct pci_devinst *pdi = arg1;
527 	struct pci_devemu *pe = pdi->pi_d;
528 	uint64_t offset;
529 	int bidx = (int) arg2;
530 
531 	assert(bidx <= PCI_BARMAX);
532 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
533 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
534 	assert(addr >= pdi->pi_bar[bidx].addr &&
535 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
536 
537 	offset = addr - pdi->pi_bar[bidx].addr;
538 
539 	if (dir == MEM_F_WRITE) {
540 		if (size == 8) {
541 			(*pe->pe_barwrite)(pdi, bidx, offset,
542 					   4, *val & 0xffffffff);
543 			(*pe->pe_barwrite)(pdi, bidx, offset + 4,
544 					   4, *val >> 32);
545 		} else {
546 			(*pe->pe_barwrite)(pdi, bidx, offset,
547 					   size, *val);
548 		}
549 	} else {
550 		if (size == 8) {
551 			*val = (*pe->pe_barread)(pdi, bidx,
552 						 offset, 4);
553 			*val |= (*pe->pe_barread)(pdi, bidx,
554 						  offset + 4, 4) << 32;
555 		} else {
556 			*val = (*pe->pe_barread)(pdi, bidx,
557 						 offset, size);
558 		}
559 	}
560 
561 	return (0);
562 }
563 
564 
565 static int
566 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
567 			uint64_t *addr)
568 {
569 	uint64_t base;
570 
571 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
572 
573 	base = roundup2(*baseptr, size);
574 
575 	if (base + size <= limit) {
576 		*addr = base;
577 		*baseptr = base + size;
578 		return (0);
579 	} else
580 		return (-1);
581 }
582 
583 /*
584  * Register (or unregister) the MMIO or I/O region associated with the BAR
585  * register 'idx' of an emulated pci device.
586  */
587 static void
588 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
589 {
590 	struct pci_devemu *pe;
591 	int error;
592 	struct inout_port iop;
593 	struct mem_range mr;
594 
595 	pe = pi->pi_d;
596 	switch (pi->pi_bar[idx].type) {
597 	case PCIBAR_IO:
598 		bzero(&iop, sizeof(struct inout_port));
599 		iop.name = pi->pi_name;
600 		iop.port = pi->pi_bar[idx].addr;
601 		iop.size = pi->pi_bar[idx].size;
602 		if (registration) {
603 			iop.flags = IOPORT_F_INOUT;
604 			iop.handler = pci_emul_io_handler;
605 			iop.arg = pi;
606 			error = register_inout(&iop);
607 		} else
608 			error = unregister_inout(&iop);
609 		if (pe->pe_baraddr != NULL)
610 			(*pe->pe_baraddr)(pi, idx, registration,
611 					  pi->pi_bar[idx].addr);
612 		break;
613 	case PCIBAR_MEM32:
614 	case PCIBAR_MEM64:
615 		bzero(&mr, sizeof(struct mem_range));
616 		mr.name = pi->pi_name;
617 		mr.base = pi->pi_bar[idx].addr;
618 		mr.size = pi->pi_bar[idx].size;
619 		if (registration) {
620 			mr.flags = MEM_F_RW;
621 			mr.handler = pci_emul_mem_handler;
622 			mr.arg1 = pi;
623 			mr.arg2 = idx;
624 			error = register_mem(&mr);
625 		} else
626 			error = unregister_mem(&mr);
627 		if (pe->pe_baraddr != NULL)
628 			(*pe->pe_baraddr)(pi, idx, registration,
629 					  pi->pi_bar[idx].addr);
630 		break;
631 	case PCIBAR_ROM:
632 		error = 0;
633 		if (pe->pe_baraddr != NULL)
634 			(*pe->pe_baraddr)(pi, idx, registration,
635 			    pi->pi_bar[idx].addr);
636 		break;
637 	default:
638 		error = EINVAL;
639 		break;
640 	}
641 	assert(error == 0);
642 }
643 
644 static void
645 unregister_bar(struct pci_devinst *pi, int idx)
646 {
647 
648 	modify_bar_registration(pi, idx, 0);
649 }
650 
651 static void
652 register_bar(struct pci_devinst *pi, int idx)
653 {
654 
655 	modify_bar_registration(pi, idx, 1);
656 }
657 
658 /* Is the ROM enabled for the emulated pci device? */
659 static int
660 romen(struct pci_devinst *pi)
661 {
662 	return (pi->pi_bar[PCI_ROM_IDX].lobits & PCIM_BIOS_ENABLE) ==
663 	    PCIM_BIOS_ENABLE;
664 }
665 
666 /* Are we decoding i/o port accesses for the emulated pci device? */
667 static int
668 porten(struct pci_devinst *pi)
669 {
670 	uint16_t cmd;
671 
672 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
673 
674 	return (cmd & PCIM_CMD_PORTEN);
675 }
676 
677 /* Are we decoding memory accesses for the emulated pci device? */
678 static int
679 memen(struct pci_devinst *pi)
680 {
681 	uint16_t cmd;
682 
683 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
684 
685 	return (cmd & PCIM_CMD_MEMEN);
686 }
687 
688 /*
689  * Update the MMIO or I/O address that is decoded by the BAR register.
690  *
691  * If the pci device has enabled the address space decoding then intercept
692  * the address range decoded by the BAR register.
693  */
694 static void
695 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
696 {
697 	int decode;
698 
699 	if (pi->pi_bar[idx].type == PCIBAR_IO)
700 		decode = porten(pi);
701 	else
702 		decode = memen(pi);
703 
704 	if (decode)
705 		unregister_bar(pi, idx);
706 
707 	switch (type) {
708 	case PCIBAR_IO:
709 	case PCIBAR_MEM32:
710 		pi->pi_bar[idx].addr = addr;
711 		break;
712 	case PCIBAR_MEM64:
713 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
714 		pi->pi_bar[idx].addr |= addr;
715 		break;
716 	case PCIBAR_MEMHI64:
717 		pi->pi_bar[idx].addr &= 0xffffffff;
718 		pi->pi_bar[idx].addr |= addr;
719 		break;
720 	default:
721 		assert(0);
722 	}
723 
724 	if (decode)
725 		register_bar(pi, idx);
726 }
727 
728 int
729 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
730     uint64_t size)
731 {
732 	assert((type == PCIBAR_ROM) || (idx >= 0 && idx <= PCI_BARMAX));
733 	assert((type != PCIBAR_ROM) || (idx == PCI_ROM_IDX));
734 
735 	if ((size & (size - 1)) != 0)
736 		size = 1UL << flsl(size);	/* round up to a power of 2 */
737 
738 	/* Enforce minimum BAR sizes required by the PCI standard */
739 	if (type == PCIBAR_IO) {
740 		if (size < 4)
741 			size = 4;
742 	} else if (type == PCIBAR_ROM) {
743 		if (size < ~PCIM_BIOS_ADDR_MASK + 1)
744 			size = ~PCIM_BIOS_ADDR_MASK + 1;
745 	} else {
746 		if (size < 16)
747 			size = 16;
748 	}
749 
750 	/*
751 	 * To reduce fragmentation of the MMIO space, we allocate the BARs by
752 	 * size. Therefore, don't allocate the BAR yet. We create a list of all
753 	 * BAR allocation which is sorted by BAR size. When all PCI devices are
754 	 * initialized, we will assign an address to the BARs.
755 	 */
756 
757 	/* create a new list entry */
758 	struct pci_bar_allocation *const new_bar = malloc(sizeof(*new_bar));
759 	memset(new_bar, 0, sizeof(*new_bar));
760 	new_bar->pdi = pdi;
761 	new_bar->idx = idx;
762 	new_bar->type = type;
763 	new_bar->size = size;
764 
765 	/*
766 	 * Search for a BAR which size is lower than the size of our newly
767 	 * allocated BAR.
768 	 */
769 	struct pci_bar_allocation *bar = NULL;
770 	TAILQ_FOREACH(bar, &pci_bars, chain) {
771 		if (bar->size < size) {
772 			break;
773 		}
774 	}
775 
776 	if (bar == NULL) {
777 		/*
778 		 * Either the list is empty or new BAR is the smallest BAR of
779 		 * the list. Append it to the end of our list.
780 		 */
781 		TAILQ_INSERT_TAIL(&pci_bars, new_bar, chain);
782 	} else {
783 		/*
784 		 * The found BAR is smaller than our new BAR. For that reason,
785 		 * insert our new BAR before the found BAR.
786 		 */
787 		TAILQ_INSERT_BEFORE(bar, new_bar, chain);
788 	}
789 
790 	/*
791 	 * pci_passthru devices synchronize their physical and virtual command
792 	 * register on init. For that reason, the virtual cmd reg should be
793 	 * updated as early as possible.
794 	 */
795 	uint16_t enbit = 0;
796 	switch (type) {
797 	case PCIBAR_IO:
798 		enbit = PCIM_CMD_PORTEN;
799 		break;
800 	case PCIBAR_MEM64:
801 	case PCIBAR_MEM32:
802 		enbit = PCIM_CMD_MEMEN;
803 		break;
804 	default:
805 		enbit = 0;
806 		break;
807 	}
808 
809 	const uint16_t cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
810 	pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
811 
812 	return (0);
813 }
814 
815 static int
816 pci_emul_assign_bar(struct pci_devinst *const pdi, const int idx,
817     const enum pcibar_type type, const uint64_t size)
818 {
819 	int error;
820 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
821 
822 	switch (type) {
823 	case PCIBAR_NONE:
824 		baseptr = NULL;
825 		addr = mask = lobits = 0;
826 		break;
827 	case PCIBAR_IO:
828 		baseptr = &pci_emul_iobase;
829 		limit = PCI_EMUL_IOLIMIT;
830 		mask = PCIM_BAR_IO_BASE;
831 		lobits = PCIM_BAR_IO_SPACE;
832 		break;
833 	case PCIBAR_MEM64:
834 		/*
835 		 * XXX
836 		 * Some drivers do not work well if the 64-bit BAR is allocated
837 		 * above 4GB. Allow for this by allocating small requests under
838 		 * 4GB unless then allocation size is larger than some arbitrary
839 		 * number (128MB currently).
840 		 */
841 		if (size > 128 * 1024 * 1024) {
842 			baseptr = &pci_emul_membase64;
843 			limit = pci_emul_memlim64;
844 			mask = PCIM_BAR_MEM_BASE;
845 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
846 				 PCIM_BAR_MEM_PREFETCH;
847 		} else {
848 			baseptr = &pci_emul_membase32;
849 			limit = PCI_EMUL_MEMLIMIT32;
850 			mask = PCIM_BAR_MEM_BASE;
851 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
852 		}
853 		break;
854 	case PCIBAR_MEM32:
855 		baseptr = &pci_emul_membase32;
856 		limit = PCI_EMUL_MEMLIMIT32;
857 		mask = PCIM_BAR_MEM_BASE;
858 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
859 		break;
860 	case PCIBAR_ROM:
861 		/* do not claim memory for ROM. OVMF will do it for us. */
862 		baseptr = NULL;
863 		limit = 0;
864 		mask = PCIM_BIOS_ADDR_MASK;
865 		lobits = 0;
866 		break;
867 	default:
868 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
869 		assert(0);
870 	}
871 
872 	if (baseptr != NULL) {
873 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
874 		if (error != 0)
875 			return (error);
876 	} else {
877 		addr = 0;
878 	}
879 
880 	pdi->pi_bar[idx].type = type;
881 	pdi->pi_bar[idx].addr = addr;
882 	pdi->pi_bar[idx].size = size;
883 	/*
884 	 * passthru devices are using same lobits as physical device they set
885 	 * this property
886 	 */
887 	if (pdi->pi_bar[idx].lobits != 0) {
888 		lobits = pdi->pi_bar[idx].lobits;
889 	} else {
890 		pdi->pi_bar[idx].lobits = lobits;
891 	}
892 
893 	/* Initialize the BAR register in config space */
894 	bar = (addr & mask) | lobits;
895 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
896 
897 	if (type == PCIBAR_MEM64) {
898 		assert(idx + 1 <= PCI_BARMAX);
899 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
900 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
901 	}
902 
903 	if (type != PCIBAR_ROM) {
904 		register_bar(pdi, idx);
905 	}
906 
907 	return (0);
908 }
909 
910 int
911 pci_emul_alloc_rom(struct pci_devinst *const pdi, const uint64_t size,
912     void **const addr)
913 {
914 	/* allocate ROM space once on first call */
915 	if (pci_emul_rombase == 0) {
916 		pci_emul_rombase = vm_create_devmem(pdi->pi_vmctx, VM_PCIROM,
917 		    "pcirom", PCI_EMUL_ROMSIZE);
918 		if (pci_emul_rombase == MAP_FAILED) {
919 			warnx("%s: failed to create rom segment", __func__);
920 			return (-1);
921 		}
922 		pci_emul_romlim = pci_emul_rombase + PCI_EMUL_ROMSIZE;
923 		pci_emul_romoffset = 0;
924 	}
925 
926 	/* ROM size should be a power of 2 and greater than 2 KB */
927 	const uint64_t rom_size = MAX(1UL << flsl(size),
928 	    ~PCIM_BIOS_ADDR_MASK + 1);
929 
930 	/* check if ROM fits into ROM space */
931 	if (pci_emul_romoffset + rom_size > PCI_EMUL_ROMSIZE) {
932 		warnx("%s: no space left in rom segment:", __func__);
933 		warnx("%16lu bytes left",
934 		    PCI_EMUL_ROMSIZE - pci_emul_romoffset);
935 		warnx("%16lu bytes required by %d/%d/%d", rom_size, pdi->pi_bus,
936 		    pdi->pi_slot, pdi->pi_func);
937 		return (-1);
938 	}
939 
940 	/* allocate ROM BAR */
941 	const int error = pci_emul_alloc_bar(pdi, PCI_ROM_IDX, PCIBAR_ROM,
942 	    rom_size);
943 	if (error)
944 		return error;
945 
946 	/* return address */
947 	*addr = pci_emul_rombase + pci_emul_romoffset;
948 
949 	/* save offset into ROM Space */
950 	pdi->pi_romoffset = pci_emul_romoffset;
951 
952 	/* increase offset for next ROM */
953 	pci_emul_romoffset += rom_size;
954 
955 	return (0);
956 }
957 
958 #define	CAP_START_OFFSET	0x40
959 static int
960 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
961 {
962 	int i, capoff, reallen;
963 	uint16_t sts;
964 
965 	assert(caplen > 0);
966 
967 	reallen = roundup2(caplen, 4);		/* dword aligned */
968 
969 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
970 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
971 		capoff = CAP_START_OFFSET;
972 	else
973 		capoff = pi->pi_capend + 1;
974 
975 	/* Check if we have enough space */
976 	if (capoff + reallen > PCI_REGMAX + 1)
977 		return (-1);
978 
979 	/* Set the previous capability pointer */
980 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
981 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
982 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
983 	} else
984 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
985 
986 	/* Copy the capability */
987 	for (i = 0; i < caplen; i++)
988 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
989 
990 	/* Set the next capability pointer */
991 	pci_set_cfgdata8(pi, capoff + 1, 0);
992 
993 	pi->pi_prevcap = capoff;
994 	pi->pi_capend = capoff + reallen - 1;
995 	return (0);
996 }
997 
998 static struct pci_devemu *
999 pci_emul_finddev(const char *name)
1000 {
1001 	struct pci_devemu **pdpp, *pdp;
1002 
1003 	SET_FOREACH(pdpp, pci_devemu_set) {
1004 		pdp = *pdpp;
1005 		if (!strcmp(pdp->pe_emu, name)) {
1006 			return (pdp);
1007 		}
1008 	}
1009 
1010 	return (NULL);
1011 }
1012 
1013 static int
1014 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
1015     int func, struct funcinfo *fi)
1016 {
1017 	struct pci_devinst *pdi;
1018 	int err;
1019 
1020 	pdi = calloc(1, sizeof(struct pci_devinst));
1021 
1022 	pdi->pi_vmctx = ctx;
1023 	pdi->pi_bus = bus;
1024 	pdi->pi_slot = slot;
1025 	pdi->pi_func = func;
1026 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
1027 	pdi->pi_lintr.pin = 0;
1028 	pdi->pi_lintr.state = IDLE;
1029 	pdi->pi_lintr.pirq_pin = 0;
1030 	pdi->pi_lintr.ioapic_irq = 0;
1031 	pdi->pi_d = pde;
1032 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
1033 
1034 	/* Disable legacy interrupts */
1035 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
1036 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
1037 
1038 	pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
1039 
1040 	err = (*pde->pe_init)(pdi, fi->fi_config);
1041 	if (err == 0)
1042 		fi->fi_devi = pdi;
1043 	else
1044 		free(pdi);
1045 
1046 	return (err);
1047 }
1048 
1049 void
1050 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
1051 {
1052 	int mmc;
1053 
1054 	/* Number of msi messages must be a power of 2 between 1 and 32 */
1055 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
1056 	mmc = ffs(msgnum) - 1;
1057 
1058 	bzero(msicap, sizeof(struct msicap));
1059 	msicap->capid = PCIY_MSI;
1060 	msicap->nextptr = nextptr;
1061 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
1062 }
1063 
1064 int
1065 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
1066 {
1067 	struct msicap msicap;
1068 
1069 	pci_populate_msicap(&msicap, msgnum, 0);
1070 
1071 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
1072 }
1073 
1074 static void
1075 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
1076 		     uint32_t msix_tab_size)
1077 {
1078 
1079 	assert(msix_tab_size % 4096 == 0);
1080 
1081 	bzero(msixcap, sizeof(struct msixcap));
1082 	msixcap->capid = PCIY_MSIX;
1083 
1084 	/*
1085 	 * Message Control Register, all fields set to
1086 	 * zero except for the Table Size.
1087 	 * Note: Table size N is encoded as N-1
1088 	 */
1089 	msixcap->msgctrl = msgnum - 1;
1090 
1091 	/*
1092 	 * MSI-X BAR setup:
1093 	 * - MSI-X table start at offset 0
1094 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
1095 	 */
1096 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
1097 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
1098 }
1099 
1100 static void
1101 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
1102 {
1103 	int i, table_size;
1104 
1105 	assert(table_entries > 0);
1106 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
1107 
1108 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
1109 	pi->pi_msix.table = calloc(1, table_size);
1110 
1111 	/* set mask bit of vector control register */
1112 	for (i = 0; i < table_entries; i++)
1113 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
1114 }
1115 
1116 int
1117 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
1118 {
1119 	uint32_t tab_size;
1120 	struct msixcap msixcap;
1121 
1122 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
1123 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
1124 
1125 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
1126 
1127 	/* Align table size to nearest 4K */
1128 	tab_size = roundup2(tab_size, 4096);
1129 
1130 	pi->pi_msix.table_bar = barnum;
1131 	pi->pi_msix.pba_bar   = barnum;
1132 	pi->pi_msix.table_offset = 0;
1133 	pi->pi_msix.table_count = msgnum;
1134 	pi->pi_msix.pba_offset = tab_size;
1135 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
1136 
1137 	pci_msix_table_init(pi, msgnum);
1138 
1139 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
1140 
1141 	/* allocate memory for MSI-X Table and PBA */
1142 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
1143 				tab_size + pi->pi_msix.pba_size);
1144 
1145 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
1146 					sizeof(msixcap)));
1147 }
1148 
1149 static void
1150 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1151 		 int bytes, uint32_t val)
1152 {
1153 	uint16_t msgctrl, rwmask;
1154 	int off;
1155 
1156 	off = offset - capoff;
1157 	/* Message Control Register */
1158 	if (off == 2 && bytes == 2) {
1159 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
1160 		msgctrl = pci_get_cfgdata16(pi, offset);
1161 		msgctrl &= ~rwmask;
1162 		msgctrl |= val & rwmask;
1163 		val = msgctrl;
1164 
1165 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
1166 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
1167 		pci_lintr_update(pi);
1168 	}
1169 
1170 	CFGWRITE(pi, offset, val, bytes);
1171 }
1172 
1173 static void
1174 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
1175 		int bytes, uint32_t val)
1176 {
1177 	uint16_t msgctrl, rwmask, msgdata, mme;
1178 	uint32_t addrlo;
1179 
1180 	/*
1181 	 * If guest is writing to the message control register make sure
1182 	 * we do not overwrite read-only fields.
1183 	 */
1184 	if ((offset - capoff) == 2 && bytes == 2) {
1185 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
1186 		msgctrl = pci_get_cfgdata16(pi, offset);
1187 		msgctrl &= ~rwmask;
1188 		msgctrl |= val & rwmask;
1189 		val = msgctrl;
1190 	}
1191 	CFGWRITE(pi, offset, val, bytes);
1192 
1193 	msgctrl = pci_get_cfgdata16(pi, capoff + 2);
1194 	addrlo = pci_get_cfgdata32(pi, capoff + 4);
1195 	if (msgctrl & PCIM_MSICTRL_64BIT)
1196 		msgdata = pci_get_cfgdata16(pi, capoff + 12);
1197 	else
1198 		msgdata = pci_get_cfgdata16(pi, capoff + 8);
1199 
1200 	mme = msgctrl & PCIM_MSICTRL_MME_MASK;
1201 	pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
1202 	if (pi->pi_msi.enabled) {
1203 		pi->pi_msi.addr = addrlo;
1204 		pi->pi_msi.msg_data = msgdata;
1205 		pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
1206 	} else {
1207 		pi->pi_msi.maxmsgnum = 0;
1208 	}
1209 	pci_lintr_update(pi);
1210 }
1211 
1212 static void
1213 pciecap_cfgwrite(struct pci_devinst *pi, int capoff __unused, int offset,
1214     int bytes, uint32_t val)
1215 {
1216 
1217 	/* XXX don't write to the readonly parts */
1218 	CFGWRITE(pi, offset, val, bytes);
1219 }
1220 
1221 #define	PCIECAP_VERSION	0x2
1222 int
1223 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
1224 {
1225 	int err;
1226 	struct pciecap pciecap;
1227 
1228 	bzero(&pciecap, sizeof(pciecap));
1229 
1230 	/*
1231 	 * Use the integrated endpoint type for endpoints on a root complex bus.
1232 	 *
1233 	 * NB: bhyve currently only supports a single PCI bus that is the root
1234 	 * complex bus, so all endpoints are integrated.
1235 	 */
1236 	if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
1237 		type = PCIEM_TYPE_ROOT_INT_EP;
1238 
1239 	pciecap.capid = PCIY_EXPRESS;
1240 	pciecap.pcie_capabilities = PCIECAP_VERSION | type;
1241 	if (type != PCIEM_TYPE_ROOT_INT_EP) {
1242 		pciecap.link_capabilities = 0x411;	/* gen1, x1 */
1243 		pciecap.link_status = 0x11;		/* gen1, x1 */
1244 	}
1245 
1246 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
1247 	return (err);
1248 }
1249 
1250 /*
1251  * This function assumes that 'coff' is in the capabilities region of the
1252  * config space. A capoff parameter of zero will force a search for the
1253  * offset and type.
1254  */
1255 void
1256 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val,
1257     uint8_t capoff, int capid)
1258 {
1259 	uint8_t nextoff;
1260 
1261 	/* Do not allow un-aligned writes */
1262 	if ((offset & (bytes - 1)) != 0)
1263 		return;
1264 
1265 	if (capoff == 0) {
1266 		/* Find the capability that we want to update */
1267 		capoff = CAP_START_OFFSET;
1268 		while (1) {
1269 			nextoff = pci_get_cfgdata8(pi, capoff + 1);
1270 			if (nextoff == 0)
1271 				break;
1272 			if (offset >= capoff && offset < nextoff)
1273 				break;
1274 
1275 			capoff = nextoff;
1276 		}
1277 		assert(offset >= capoff);
1278 		capid = pci_get_cfgdata8(pi, capoff);
1279 	}
1280 
1281 	/*
1282 	 * Capability ID and Next Capability Pointer are readonly.
1283 	 * However, some o/s's do 4-byte writes that include these.
1284 	 * For this case, trim the write back to 2 bytes and adjust
1285 	 * the data.
1286 	 */
1287 	if (offset == capoff || offset == capoff + 1) {
1288 		if (offset == capoff && bytes == 4) {
1289 			bytes = 2;
1290 			offset += 2;
1291 			val >>= 16;
1292 		} else
1293 			return;
1294 	}
1295 
1296 	switch (capid) {
1297 	case PCIY_MSI:
1298 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1299 		break;
1300 	case PCIY_MSIX:
1301 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1302 		break;
1303 	case PCIY_EXPRESS:
1304 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1305 		break;
1306 	default:
1307 		break;
1308 	}
1309 }
1310 
1311 static int
1312 pci_emul_iscap(struct pci_devinst *pi, int offset)
1313 {
1314 	uint16_t sts;
1315 
1316 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1317 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1318 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1319 			return (1);
1320 	}
1321 	return (0);
1322 }
1323 
1324 static int
1325 pci_emul_fallback_handler(struct vcpu *vcpu __unused, int dir,
1326     uint64_t addr __unused, int size __unused, uint64_t *val,
1327     void *arg1 __unused, long arg2 __unused)
1328 {
1329 	/*
1330 	 * Ignore writes; return 0xff's for reads. The mem read code
1331 	 * will take care of truncating to the correct size.
1332 	 */
1333 	if (dir == MEM_F_READ) {
1334 		*val = 0xffffffffffffffff;
1335 	}
1336 
1337 	return (0);
1338 }
1339 
1340 static int
1341 pci_emul_ecfg_handler(struct vcpu *vcpu __unused, int dir, uint64_t addr,
1342     int bytes, uint64_t *val, void *arg1 __unused, long arg2 __unused)
1343 {
1344 	int bus, slot, func, coff, in;
1345 
1346 	coff = addr & 0xfff;
1347 	func = (addr >> 12) & 0x7;
1348 	slot = (addr >> 15) & 0x1f;
1349 	bus = (addr >> 20) & 0xff;
1350 	in = (dir == MEM_F_READ);
1351 	if (in)
1352 		*val = ~0UL;
1353 	pci_cfgrw(in, bus, slot, func, coff, bytes, (uint32_t *)val);
1354 	return (0);
1355 }
1356 
1357 uint64_t
1358 pci_ecfg_base(void)
1359 {
1360 
1361 	return (PCI_EMUL_ECFG_BASE);
1362 }
1363 
1364 #define	BUSIO_ROUNDUP		32
1365 #define	BUSMEM32_ROUNDUP	(1024 * 1024)
1366 #define	BUSMEM64_ROUNDUP	(512 * 1024 * 1024)
1367 
1368 int
1369 init_pci(struct vmctx *ctx)
1370 {
1371 	char node_name[sizeof("pci.XXX.XX.X")];
1372 	struct mem_range mr;
1373 	struct pci_devemu *pde;
1374 	struct businfo *bi;
1375 	struct slotinfo *si;
1376 	struct funcinfo *fi;
1377 	nvlist_t *nvl;
1378 	const char *emul;
1379 	size_t lowmem;
1380 	int bus, slot, func;
1381 	int error;
1382 
1383 	if (vm_get_lowmem_limit(ctx) > PCI_EMUL_MEMBASE32)
1384 		errx(EX_OSERR, "Invalid lowmem limit");
1385 
1386 	pci_emul_iobase = PCI_EMUL_IOBASE;
1387 	pci_emul_membase32 = PCI_EMUL_MEMBASE32;
1388 
1389 	pci_emul_membase64 = 4*GB + vm_get_highmem_size(ctx);
1390 	pci_emul_membase64 = roundup2(pci_emul_membase64, PCI_EMUL_MEMSIZE64);
1391 	pci_emul_memlim64 = pci_emul_membase64 + PCI_EMUL_MEMSIZE64;
1392 
1393 	for (bus = 0; bus < MAXBUSES; bus++) {
1394 		snprintf(node_name, sizeof(node_name), "pci.%d", bus);
1395 		nvl = find_config_node(node_name);
1396 		if (nvl == NULL)
1397 			continue;
1398 		pci_businfo[bus] = calloc(1, sizeof(struct businfo));
1399 		bi = pci_businfo[bus];
1400 
1401 		/*
1402 		 * Keep track of the i/o and memory resources allocated to
1403 		 * this bus.
1404 		 */
1405 		bi->iobase = pci_emul_iobase;
1406 		bi->membase32 = pci_emul_membase32;
1407 		bi->membase64 = pci_emul_membase64;
1408 
1409 		/* first run: init devices */
1410 		for (slot = 0; slot < MAXSLOTS; slot++) {
1411 			si = &bi->slotinfo[slot];
1412 			for (func = 0; func < MAXFUNCS; func++) {
1413 				fi = &si->si_funcs[func];
1414 				snprintf(node_name, sizeof(node_name),
1415 				    "pci.%d.%d.%d", bus, slot, func);
1416 				nvl = find_config_node(node_name);
1417 				if (nvl == NULL)
1418 					continue;
1419 
1420 				fi->fi_config = nvl;
1421 				emul = get_config_value_node(nvl, "device");
1422 				if (emul == NULL) {
1423 					EPRINTLN("pci slot %d:%d:%d: missing "
1424 					    "\"device\" value", bus, slot, func);
1425 					return (EINVAL);
1426 				}
1427 				pde = pci_emul_finddev(emul);
1428 				if (pde == NULL) {
1429 					EPRINTLN("pci slot %d:%d:%d: unknown "
1430 					    "device \"%s\"", bus, slot, func,
1431 					    emul);
1432 					return (EINVAL);
1433 				}
1434 				if (pde->pe_alias != NULL) {
1435 					EPRINTLN("pci slot %d:%d:%d: legacy "
1436 					    "device \"%s\", use \"%s\" instead",
1437 					    bus, slot, func, emul,
1438 					    pde->pe_alias);
1439 					return (EINVAL);
1440 				}
1441 				fi->fi_pde = pde;
1442 				error = pci_emul_init(ctx, pde, bus, slot,
1443 				    func, fi);
1444 				if (error)
1445 					return (error);
1446 			}
1447 		}
1448 
1449 		/* second run: assign BARs and free list */
1450 		struct pci_bar_allocation *bar;
1451 		struct pci_bar_allocation *bar_tmp;
1452 		TAILQ_FOREACH_SAFE(bar, &pci_bars, chain, bar_tmp) {
1453 			pci_emul_assign_bar(bar->pdi, bar->idx, bar->type,
1454 			    bar->size);
1455 			free(bar);
1456 		}
1457 		TAILQ_INIT(&pci_bars);
1458 
1459 		/*
1460 		 * Add some slop to the I/O and memory resources decoded by
1461 		 * this bus to give a guest some flexibility if it wants to
1462 		 * reprogram the BARs.
1463 		 */
1464 		pci_emul_iobase += BUSIO_ROUNDUP;
1465 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1466 		bi->iolimit = pci_emul_iobase;
1467 
1468 		pci_emul_membase32 += BUSMEM32_ROUNDUP;
1469 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1470 		    BUSMEM32_ROUNDUP);
1471 		bi->memlimit32 = pci_emul_membase32;
1472 
1473 		pci_emul_membase64 += BUSMEM64_ROUNDUP;
1474 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1475 		    BUSMEM64_ROUNDUP);
1476 		bi->memlimit64 = pci_emul_membase64;
1477 	}
1478 
1479 	/*
1480 	 * PCI backends are initialized before routing INTx interrupts
1481 	 * so that LPC devices are able to reserve ISA IRQs before
1482 	 * routing PIRQ pins.
1483 	 */
1484 	for (bus = 0; bus < MAXBUSES; bus++) {
1485 		if ((bi = pci_businfo[bus]) == NULL)
1486 			continue;
1487 
1488 		for (slot = 0; slot < MAXSLOTS; slot++) {
1489 			si = &bi->slotinfo[slot];
1490 			for (func = 0; func < MAXFUNCS; func++) {
1491 				fi = &si->si_funcs[func];
1492 				if (fi->fi_devi == NULL)
1493 					continue;
1494 				pci_lintr_route(fi->fi_devi);
1495 			}
1496 		}
1497 	}
1498 	lpc_pirq_routed();
1499 
1500 	/*
1501 	 * The guest physical memory map looks like the following:
1502 	 * [0,		    lowmem)		guest system memory
1503 	 * [lowmem,	    0xC0000000)		memory hole (may be absent)
1504 	 * [0xC0000000,     0xE0000000)		PCI hole (32-bit BAR allocation)
1505 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1506 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1507 	 * [4GB,	    4GB + highmem)
1508 	 */
1509 
1510 	/*
1511 	 * Accesses to memory addresses that are not allocated to system
1512 	 * memory or PCI devices return 0xff's.
1513 	 */
1514 	lowmem = vm_get_lowmem_size(ctx);
1515 	bzero(&mr, sizeof(struct mem_range));
1516 	mr.name = "PCI hole";
1517 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1518 	mr.base = lowmem;
1519 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1520 	mr.handler = pci_emul_fallback_handler;
1521 	error = register_mem_fallback(&mr);
1522 	assert(error == 0);
1523 
1524 	/* PCI extended config space */
1525 	bzero(&mr, sizeof(struct mem_range));
1526 	mr.name = "PCI ECFG";
1527 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1528 	mr.base = PCI_EMUL_ECFG_BASE;
1529 	mr.size = PCI_EMUL_ECFG_SIZE;
1530 	mr.handler = pci_emul_ecfg_handler;
1531 	error = register_mem(&mr);
1532 	assert(error == 0);
1533 
1534 	return (0);
1535 }
1536 
1537 static void
1538 pci_apic_prt_entry(int bus __unused, int slot, int pin, int pirq_pin __unused,
1539     int ioapic_irq, void *arg __unused)
1540 {
1541 
1542 	dsdt_line("  Package ()");
1543 	dsdt_line("  {");
1544 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1545 	dsdt_line("    0x%02X,", pin - 1);
1546 	dsdt_line("    Zero,");
1547 	dsdt_line("    0x%X", ioapic_irq);
1548 	dsdt_line("  },");
1549 }
1550 
1551 static void
1552 pci_pirq_prt_entry(int bus __unused, int slot, int pin, int pirq_pin,
1553     int ioapic_irq __unused, void *arg __unused)
1554 {
1555 	char *name;
1556 
1557 	name = lpc_pirq_name(pirq_pin);
1558 	if (name == NULL)
1559 		return;
1560 	dsdt_line("  Package ()");
1561 	dsdt_line("  {");
1562 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1563 	dsdt_line("    0x%02X,", pin - 1);
1564 	dsdt_line("    %s,", name);
1565 	dsdt_line("    0x00");
1566 	dsdt_line("  },");
1567 	free(name);
1568 }
1569 
1570 /*
1571  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1572  * corresponding to each PCI bus.
1573  */
1574 static void
1575 pci_bus_write_dsdt(int bus)
1576 {
1577 	struct businfo *bi;
1578 	struct slotinfo *si;
1579 	struct pci_devinst *pi;
1580 	int count, func, slot;
1581 
1582 	/*
1583 	 * If there are no devices on this 'bus' then just return.
1584 	 */
1585 	if ((bi = pci_businfo[bus]) == NULL) {
1586 		/*
1587 		 * Bus 0 is special because it decodes the I/O ports used
1588 		 * for PCI config space access even if there are no devices
1589 		 * on it.
1590 		 */
1591 		if (bus != 0)
1592 			return;
1593 	}
1594 
1595 	dsdt_line("  Device (PC%02X)", bus);
1596 	dsdt_line("  {");
1597 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1598 
1599 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1600 	dsdt_line("    {");
1601 	dsdt_line("        Return (0x%08X)", bus);
1602 	dsdt_line("    }");
1603 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1604 	dsdt_line("    {");
1605 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1606 	    "MaxFixed, PosDecode,");
1607 	dsdt_line("        0x0000,             // Granularity");
1608 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1609 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1610 	dsdt_line("        0x0000,             // Translation Offset");
1611 	dsdt_line("        0x0001,             // Length");
1612 	dsdt_line("        ,, )");
1613 
1614 	if (bus == 0) {
1615 		dsdt_indent(3);
1616 		dsdt_fixed_ioport(0xCF8, 8);
1617 		dsdt_unindent(3);
1618 
1619 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1620 		    "PosDecode, EntireRange,");
1621 		dsdt_line("        0x0000,             // Granularity");
1622 		dsdt_line("        0x0000,             // Range Minimum");
1623 		dsdt_line("        0x0CF7,             // Range Maximum");
1624 		dsdt_line("        0x0000,             // Translation Offset");
1625 		dsdt_line("        0x0CF8,             // Length");
1626 		dsdt_line("        ,, , TypeStatic)");
1627 
1628 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1629 		    "PosDecode, EntireRange,");
1630 		dsdt_line("        0x0000,             // Granularity");
1631 		dsdt_line("        0x0D00,             // Range Minimum");
1632 		dsdt_line("        0x%04X,             // Range Maximum",
1633 		    PCI_EMUL_IOBASE - 1);
1634 		dsdt_line("        0x0000,             // Translation Offset");
1635 		dsdt_line("        0x%04X,             // Length",
1636 		    PCI_EMUL_IOBASE - 0x0D00);
1637 		dsdt_line("        ,, , TypeStatic)");
1638 
1639 		if (bi == NULL) {
1640 			dsdt_line("    })");
1641 			goto done;
1642 		}
1643 	}
1644 	assert(bi != NULL);
1645 
1646 	/* i/o window */
1647 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1648 	    "PosDecode, EntireRange,");
1649 	dsdt_line("        0x0000,             // Granularity");
1650 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1651 	dsdt_line("        0x%04X,             // Range Maximum",
1652 	    bi->iolimit - 1);
1653 	dsdt_line("        0x0000,             // Translation Offset");
1654 	dsdt_line("        0x%04X,             // Length",
1655 	    bi->iolimit - bi->iobase);
1656 	dsdt_line("        ,, , TypeStatic)");
1657 
1658 	/* mmio window (32-bit) */
1659 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1660 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1661 	dsdt_line("        0x00000000,         // Granularity");
1662 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1663 	dsdt_line("        0x%08X,         // Range Maximum\n",
1664 	    bi->memlimit32 - 1);
1665 	dsdt_line("        0x00000000,         // Translation Offset");
1666 	dsdt_line("        0x%08X,         // Length\n",
1667 	    bi->memlimit32 - bi->membase32);
1668 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1669 
1670 	/* mmio window (64-bit) */
1671 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1672 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1673 	dsdt_line("        0x0000000000000000, // Granularity");
1674 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1675 	dsdt_line("        0x%016lX, // Range Maximum\n",
1676 	    bi->memlimit64 - 1);
1677 	dsdt_line("        0x0000000000000000, // Translation Offset");
1678 	dsdt_line("        0x%016lX, // Length\n",
1679 	    bi->memlimit64 - bi->membase64);
1680 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1681 	dsdt_line("    })");
1682 
1683 	count = pci_count_lintr(bus);
1684 	if (count != 0) {
1685 		dsdt_indent(2);
1686 		dsdt_line("Name (PPRT, Package ()");
1687 		dsdt_line("{");
1688 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1689 		dsdt_line("})");
1690 		dsdt_line("Name (APRT, Package ()");
1691 		dsdt_line("{");
1692 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1693 		dsdt_line("})");
1694 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1695 		dsdt_line("{");
1696 		dsdt_line("  If (PICM)");
1697 		dsdt_line("  {");
1698 		dsdt_line("    Return (APRT)");
1699 		dsdt_line("  }");
1700 		dsdt_line("  Else");
1701 		dsdt_line("  {");
1702 		dsdt_line("    Return (PPRT)");
1703 		dsdt_line("  }");
1704 		dsdt_line("}");
1705 		dsdt_unindent(2);
1706 	}
1707 
1708 	dsdt_indent(2);
1709 	for (slot = 0; slot < MAXSLOTS; slot++) {
1710 		si = &bi->slotinfo[slot];
1711 		for (func = 0; func < MAXFUNCS; func++) {
1712 			pi = si->si_funcs[func].fi_devi;
1713 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1714 				pi->pi_d->pe_write_dsdt(pi);
1715 		}
1716 	}
1717 	dsdt_unindent(2);
1718 done:
1719 	dsdt_line("  }");
1720 }
1721 
1722 void
1723 pci_write_dsdt(void)
1724 {
1725 	int bus;
1726 
1727 	dsdt_indent(1);
1728 	dsdt_line("Name (PICM, 0x00)");
1729 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1730 	dsdt_line("{");
1731 	dsdt_line("  Store (Arg0, PICM)");
1732 	dsdt_line("}");
1733 	dsdt_line("");
1734 	dsdt_line("Scope (_SB)");
1735 	dsdt_line("{");
1736 	for (bus = 0; bus < MAXBUSES; bus++)
1737 		pci_bus_write_dsdt(bus);
1738 	dsdt_line("}");
1739 	dsdt_unindent(1);
1740 }
1741 
1742 int
1743 pci_bus_configured(int bus)
1744 {
1745 	assert(bus >= 0 && bus < MAXBUSES);
1746 	return (pci_businfo[bus] != NULL);
1747 }
1748 
1749 int
1750 pci_msi_enabled(struct pci_devinst *pi)
1751 {
1752 	return (pi->pi_msi.enabled);
1753 }
1754 
1755 int
1756 pci_msi_maxmsgnum(struct pci_devinst *pi)
1757 {
1758 	if (pi->pi_msi.enabled)
1759 		return (pi->pi_msi.maxmsgnum);
1760 	else
1761 		return (0);
1762 }
1763 
1764 int
1765 pci_msix_enabled(struct pci_devinst *pi)
1766 {
1767 
1768 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1769 }
1770 
1771 void
1772 pci_generate_msix(struct pci_devinst *pi, int index)
1773 {
1774 	struct msix_table_entry *mte;
1775 
1776 	if (!pci_msix_enabled(pi))
1777 		return;
1778 
1779 	if (pi->pi_msix.function_mask)
1780 		return;
1781 
1782 	if (index >= pi->pi_msix.table_count)
1783 		return;
1784 
1785 	mte = &pi->pi_msix.table[index];
1786 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1787 		/* XXX Set PBA bit if interrupt is disabled */
1788 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1789 	}
1790 }
1791 
1792 void
1793 pci_generate_msi(struct pci_devinst *pi, int index)
1794 {
1795 
1796 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1797 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1798 			     pi->pi_msi.msg_data + index);
1799 	}
1800 }
1801 
1802 static bool
1803 pci_lintr_permitted(struct pci_devinst *pi)
1804 {
1805 	uint16_t cmd;
1806 
1807 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1808 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1809 		(cmd & PCIM_CMD_INTxDIS)));
1810 }
1811 
1812 void
1813 pci_lintr_request(struct pci_devinst *pi)
1814 {
1815 	struct businfo *bi;
1816 	struct slotinfo *si;
1817 	int bestpin, bestcount, pin;
1818 
1819 	bi = pci_businfo[pi->pi_bus];
1820 	assert(bi != NULL);
1821 
1822 	/*
1823 	 * Just allocate a pin from our slot.  The pin will be
1824 	 * assigned IRQs later when interrupts are routed.
1825 	 */
1826 	si = &bi->slotinfo[pi->pi_slot];
1827 	bestpin = 0;
1828 	bestcount = si->si_intpins[0].ii_count;
1829 	for (pin = 1; pin < 4; pin++) {
1830 		if (si->si_intpins[pin].ii_count < bestcount) {
1831 			bestpin = pin;
1832 			bestcount = si->si_intpins[pin].ii_count;
1833 		}
1834 	}
1835 
1836 	si->si_intpins[bestpin].ii_count++;
1837 	pi->pi_lintr.pin = bestpin + 1;
1838 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1839 }
1840 
1841 static void
1842 pci_lintr_route(struct pci_devinst *pi)
1843 {
1844 	struct businfo *bi;
1845 	struct intxinfo *ii;
1846 
1847 	if (pi->pi_lintr.pin == 0)
1848 		return;
1849 
1850 	bi = pci_businfo[pi->pi_bus];
1851 	assert(bi != NULL);
1852 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1853 
1854 	/*
1855 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1856 	 * is not yet assigned.
1857 	 */
1858 	if (ii->ii_ioapic_irq == 0)
1859 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1860 	assert(ii->ii_ioapic_irq > 0);
1861 
1862 	/*
1863 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1864 	 * not yet assigned.
1865 	 */
1866 	if (ii->ii_pirq_pin == 0)
1867 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1868 	assert(ii->ii_pirq_pin > 0);
1869 
1870 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1871 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1872 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1873 }
1874 
1875 void
1876 pci_lintr_assert(struct pci_devinst *pi)
1877 {
1878 
1879 	assert(pi->pi_lintr.pin > 0);
1880 
1881 	pthread_mutex_lock(&pi->pi_lintr.lock);
1882 	if (pi->pi_lintr.state == IDLE) {
1883 		if (pci_lintr_permitted(pi)) {
1884 			pi->pi_lintr.state = ASSERTED;
1885 			pci_irq_assert(pi);
1886 		} else
1887 			pi->pi_lintr.state = PENDING;
1888 	}
1889 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1890 }
1891 
1892 void
1893 pci_lintr_deassert(struct pci_devinst *pi)
1894 {
1895 
1896 	assert(pi->pi_lintr.pin > 0);
1897 
1898 	pthread_mutex_lock(&pi->pi_lintr.lock);
1899 	if (pi->pi_lintr.state == ASSERTED) {
1900 		pi->pi_lintr.state = IDLE;
1901 		pci_irq_deassert(pi);
1902 	} else if (pi->pi_lintr.state == PENDING)
1903 		pi->pi_lintr.state = IDLE;
1904 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1905 }
1906 
1907 static void
1908 pci_lintr_update(struct pci_devinst *pi)
1909 {
1910 
1911 	pthread_mutex_lock(&pi->pi_lintr.lock);
1912 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1913 		pci_irq_deassert(pi);
1914 		pi->pi_lintr.state = PENDING;
1915 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1916 		pi->pi_lintr.state = ASSERTED;
1917 		pci_irq_assert(pi);
1918 	}
1919 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1920 }
1921 
1922 int
1923 pci_count_lintr(int bus)
1924 {
1925 	int count, slot, pin;
1926 	struct slotinfo *slotinfo;
1927 
1928 	count = 0;
1929 	if (pci_businfo[bus] != NULL) {
1930 		for (slot = 0; slot < MAXSLOTS; slot++) {
1931 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1932 			for (pin = 0; pin < 4; pin++) {
1933 				if (slotinfo->si_intpins[pin].ii_count != 0)
1934 					count++;
1935 			}
1936 		}
1937 	}
1938 	return (count);
1939 }
1940 
1941 void
1942 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1943 {
1944 	struct businfo *bi;
1945 	struct slotinfo *si;
1946 	struct intxinfo *ii;
1947 	int slot, pin;
1948 
1949 	if ((bi = pci_businfo[bus]) == NULL)
1950 		return;
1951 
1952 	for (slot = 0; slot < MAXSLOTS; slot++) {
1953 		si = &bi->slotinfo[slot];
1954 		for (pin = 0; pin < 4; pin++) {
1955 			ii = &si->si_intpins[pin];
1956 			if (ii->ii_count != 0)
1957 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1958 				    ii->ii_ioapic_irq, arg);
1959 		}
1960 	}
1961 }
1962 
1963 /*
1964  * Return 1 if the emulated device in 'slot' is a multi-function device.
1965  * Return 0 otherwise.
1966  */
1967 static int
1968 pci_emul_is_mfdev(int bus, int slot)
1969 {
1970 	struct businfo *bi;
1971 	struct slotinfo *si;
1972 	int f, numfuncs;
1973 
1974 	numfuncs = 0;
1975 	if ((bi = pci_businfo[bus]) != NULL) {
1976 		si = &bi->slotinfo[slot];
1977 		for (f = 0; f < MAXFUNCS; f++) {
1978 			if (si->si_funcs[f].fi_devi != NULL) {
1979 				numfuncs++;
1980 			}
1981 		}
1982 	}
1983 	return (numfuncs > 1);
1984 }
1985 
1986 /*
1987  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1988  * whether or not is a multi-function being emulated in the pci 'slot'.
1989  */
1990 static void
1991 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1992 {
1993 	int mfdev;
1994 
1995 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1996 		mfdev = pci_emul_is_mfdev(bus, slot);
1997 		switch (bytes) {
1998 		case 1:
1999 		case 2:
2000 			*rv &= ~PCIM_MFDEV;
2001 			if (mfdev) {
2002 				*rv |= PCIM_MFDEV;
2003 			}
2004 			break;
2005 		case 4:
2006 			*rv &= ~(PCIM_MFDEV << 16);
2007 			if (mfdev) {
2008 				*rv |= (PCIM_MFDEV << 16);
2009 			}
2010 			break;
2011 		}
2012 	}
2013 }
2014 
2015 /*
2016  * Update device state in response to changes to the PCI command
2017  * register.
2018  */
2019 void
2020 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
2021 {
2022 	int i;
2023 	uint16_t changed, new;
2024 
2025 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);
2026 	changed = old ^ new;
2027 
2028 	/*
2029 	 * If the MMIO or I/O address space decoding has changed then
2030 	 * register/unregister all BARs that decode that address space.
2031 	 */
2032 	for (i = 0; i <= PCI_BARMAX_WITH_ROM; i++) {
2033 		switch (pi->pi_bar[i].type) {
2034 			case PCIBAR_NONE:
2035 			case PCIBAR_MEMHI64:
2036 				break;
2037 			case PCIBAR_IO:
2038 				/* I/O address space decoding changed? */
2039 				if (changed & PCIM_CMD_PORTEN) {
2040 					if (new & PCIM_CMD_PORTEN)
2041 						register_bar(pi, i);
2042 					else
2043 						unregister_bar(pi, i);
2044 				}
2045 				break;
2046 			case PCIBAR_ROM:
2047 				/* skip (un-)register of ROM if it disabled */
2048 				if (!romen(pi))
2049 					break;
2050 				/* fallthrough */
2051 			case PCIBAR_MEM32:
2052 			case PCIBAR_MEM64:
2053 				/* MMIO address space decoding changed? */
2054 				if (changed & PCIM_CMD_MEMEN) {
2055 					if (new & PCIM_CMD_MEMEN)
2056 						register_bar(pi, i);
2057 					else
2058 						unregister_bar(pi, i);
2059 				}
2060 				break;
2061 			default:
2062 				assert(0);
2063 		}
2064 	}
2065 
2066 	/*
2067 	 * If INTx has been unmasked and is pending, assert the
2068 	 * interrupt.
2069 	 */
2070 	pci_lintr_update(pi);
2071 }
2072 
2073 static void
2074 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
2075 {
2076 	int rshift;
2077 	uint32_t cmd, old, readonly;
2078 
2079 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
2080 
2081 	/*
2082 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
2083 	 *
2084 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
2085 	 * 'write 1 to clear'. However these bits are not set to '1' by
2086 	 * any device emulation so it is simpler to treat them as readonly.
2087 	 */
2088 	rshift = (coff & 0x3) * 8;
2089 	readonly = 0xFFFFF880 >> rshift;
2090 
2091 	old = CFGREAD(pi, coff, bytes);
2092 	new &= ~readonly;
2093 	new |= (old & readonly);
2094 	CFGWRITE(pi, coff, new, bytes);			/* update config */
2095 
2096 	pci_emul_cmd_changed(pi, cmd);
2097 }
2098 
2099 static void
2100 pci_cfgrw(int in, int bus, int slot, int func, int coff, int bytes,
2101     uint32_t *eax)
2102 {
2103 	struct businfo *bi;
2104 	struct slotinfo *si;
2105 	struct pci_devinst *pi;
2106 	struct pci_devemu *pe;
2107 	int idx, needcfg;
2108 	uint64_t addr, bar, mask;
2109 
2110 	if ((bi = pci_businfo[bus]) != NULL) {
2111 		si = &bi->slotinfo[slot];
2112 		pi = si->si_funcs[func].fi_devi;
2113 	} else
2114 		pi = NULL;
2115 
2116 	/*
2117 	 * Just return if there is no device at this slot:func or if the
2118 	 * the guest is doing an un-aligned access.
2119 	 */
2120 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
2121 	    (coff & (bytes - 1)) != 0) {
2122 		if (in)
2123 			*eax = 0xffffffff;
2124 		return;
2125 	}
2126 
2127 	/*
2128 	 * Ignore all writes beyond the standard config space and return all
2129 	 * ones on reads.
2130 	 */
2131 	if (coff >= PCI_REGMAX + 1) {
2132 		if (in) {
2133 			*eax = 0xffffffff;
2134 			/*
2135 			 * Extended capabilities begin at offset 256 in config
2136 			 * space. Absence of extended capabilities is signaled
2137 			 * with all 0s in the extended capability header at
2138 			 * offset 256.
2139 			 */
2140 			if (coff <= PCI_REGMAX + 4)
2141 				*eax = 0x00000000;
2142 		}
2143 		return;
2144 	}
2145 
2146 	pe = pi->pi_d;
2147 
2148 	/*
2149 	 * Config read
2150 	 */
2151 	if (in) {
2152 		/* Let the device emulation override the default handler */
2153 		if (pe->pe_cfgread != NULL) {
2154 			needcfg = pe->pe_cfgread(pi, coff, bytes, eax);
2155 		} else {
2156 			needcfg = 1;
2157 		}
2158 
2159 		if (needcfg)
2160 			*eax = CFGREAD(pi, coff, bytes);
2161 
2162 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
2163 	} else {
2164 		/* Let the device emulation override the default handler */
2165 		if (pe->pe_cfgwrite != NULL &&
2166 		    (*pe->pe_cfgwrite)(pi, coff, bytes, *eax) == 0)
2167 			return;
2168 
2169 		/*
2170 		 * Special handling for write to BAR and ROM registers
2171 		 */
2172 		if (is_pcir_bar(coff) || is_pcir_bios(coff)) {
2173 			/*
2174 			 * Ignore writes to BAR registers that are not
2175 			 * 4-byte aligned.
2176 			 */
2177 			if (bytes != 4 || (coff & 0x3) != 0)
2178 				return;
2179 
2180 			if (is_pcir_bar(coff)) {
2181 				idx = (coff - PCIR_BAR(0)) / 4;
2182 			} else if (is_pcir_bios(coff)) {
2183 				idx = PCI_ROM_IDX;
2184 			} else {
2185 				errx(4, "%s: invalid BAR offset %d", __func__,
2186 				    coff);
2187 			}
2188 
2189 			mask = ~(pi->pi_bar[idx].size - 1);
2190 			switch (pi->pi_bar[idx].type) {
2191 			case PCIBAR_NONE:
2192 				pi->pi_bar[idx].addr = bar = 0;
2193 				break;
2194 			case PCIBAR_IO:
2195 				addr = *eax & mask;
2196 				addr &= 0xffff;
2197 				bar = addr | pi->pi_bar[idx].lobits;
2198 				/*
2199 				 * Register the new BAR value for interception
2200 				 */
2201 				if (addr != pi->pi_bar[idx].addr) {
2202 					update_bar_address(pi, addr, idx,
2203 							   PCIBAR_IO);
2204 				}
2205 				break;
2206 			case PCIBAR_MEM32:
2207 				addr = bar = *eax & mask;
2208 				bar |= pi->pi_bar[idx].lobits;
2209 				if (addr != pi->pi_bar[idx].addr) {
2210 					update_bar_address(pi, addr, idx,
2211 							   PCIBAR_MEM32);
2212 				}
2213 				break;
2214 			case PCIBAR_MEM64:
2215 				addr = bar = *eax & mask;
2216 				bar |= pi->pi_bar[idx].lobits;
2217 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
2218 					update_bar_address(pi, addr, idx,
2219 							   PCIBAR_MEM64);
2220 				}
2221 				break;
2222 			case PCIBAR_MEMHI64:
2223 				mask = ~(pi->pi_bar[idx - 1].size - 1);
2224 				addr = ((uint64_t)*eax << 32) & mask;
2225 				bar = addr >> 32;
2226 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
2227 					update_bar_address(pi, addr, idx - 1,
2228 							   PCIBAR_MEMHI64);
2229 				}
2230 				break;
2231 			case PCIBAR_ROM:
2232 				addr = bar = *eax & mask;
2233 				if (memen(pi) && romen(pi)) {
2234 					unregister_bar(pi, idx);
2235 				}
2236 				pi->pi_bar[idx].addr = addr;
2237 				pi->pi_bar[idx].lobits = *eax &
2238 				    PCIM_BIOS_ENABLE;
2239 				/* romen could have changed it value */
2240 				if (memen(pi) && romen(pi)) {
2241 					register_bar(pi, idx);
2242 				}
2243 				bar |= pi->pi_bar[idx].lobits;
2244 				break;
2245 			default:
2246 				assert(0);
2247 			}
2248 			pci_set_cfgdata32(pi, coff, bar);
2249 
2250 		} else if (pci_emul_iscap(pi, coff)) {
2251 			pci_emul_capwrite(pi, coff, bytes, *eax, 0, 0);
2252 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
2253 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
2254 		} else {
2255 			CFGWRITE(pi, coff, *eax, bytes);
2256 		}
2257 	}
2258 }
2259 
2260 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
2261 
2262 static int
2263 pci_emul_cfgaddr(struct vmctx *ctx __unused, int in,
2264     int port __unused, int bytes, uint32_t *eax, void *arg __unused)
2265 {
2266 	uint32_t x;
2267 
2268 	if (bytes != 4) {
2269 		if (in)
2270 			*eax = (bytes == 2) ? 0xffff : 0xff;
2271 		return (0);
2272 	}
2273 
2274 	if (in) {
2275 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
2276 		if (cfgenable)
2277 			x |= CONF1_ENABLE;
2278 		*eax = x;
2279 	} else {
2280 		x = *eax;
2281 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
2282 		cfgoff = (x & PCI_REGMAX) & ~0x03;
2283 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
2284 		cfgslot = (x >> 11) & PCI_SLOTMAX;
2285 		cfgbus = (x >> 16) & PCI_BUSMAX;
2286 	}
2287 
2288 	return (0);
2289 }
2290 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
2291 
2292 static int
2293 pci_emul_cfgdata(struct vmctx *ctx __unused, int in, int port,
2294     int bytes, uint32_t *eax, void *arg __unused)
2295 {
2296 	int coff;
2297 
2298 	assert(bytes == 1 || bytes == 2 || bytes == 4);
2299 
2300 	coff = cfgoff + (port - CONF1_DATA_PORT);
2301 	if (cfgenable) {
2302 		pci_cfgrw(in, cfgbus, cfgslot, cfgfunc, coff, bytes, eax);
2303 	} else {
2304 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
2305 		if (in)
2306 			*eax = 0xffffffff;
2307 	}
2308 	return (0);
2309 }
2310 
2311 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
2312 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
2313 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
2314 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
2315 
2316 #ifdef BHYVE_SNAPSHOT
2317 /*
2318  * Saves/restores PCI device emulated state. Returns 0 on success.
2319  */
2320 static int
2321 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta)
2322 {
2323 	struct pci_devinst *pi;
2324 	int i;
2325 	int ret;
2326 
2327 	pi = meta->dev_data;
2328 
2329 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done);
2330 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done);
2331 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done);
2332 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done);
2333 
2334 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done);
2335 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done);
2336 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done);
2337 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done);
2338 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done);
2339 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done);
2340 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done);
2341 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done);
2342 
2343 	SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata),
2344 			      meta, ret, done);
2345 
2346 	for (i = 0; i < (int)nitems(pi->pi_bar); i++) {
2347 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done);
2348 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done);
2349 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done);
2350 	}
2351 
2352 	/* Restore MSI-X table. */
2353 	for (i = 0; i < pi->pi_msix.table_count; i++) {
2354 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr,
2355 				      meta, ret, done);
2356 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data,
2357 				      meta, ret, done);
2358 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control,
2359 				      meta, ret, done);
2360 	}
2361 
2362 done:
2363 	return (ret);
2364 }
2365 
2366 static int
2367 pci_find_slotted_dev(const char *dev_name, struct pci_devemu **pde,
2368 		     struct pci_devinst **pdi)
2369 {
2370 	struct businfo *bi;
2371 	struct slotinfo *si;
2372 	struct funcinfo *fi;
2373 	int bus, slot, func;
2374 
2375 	assert(dev_name != NULL);
2376 	assert(pde != NULL);
2377 	assert(pdi != NULL);
2378 
2379 	for (bus = 0; bus < MAXBUSES; bus++) {
2380 		if ((bi = pci_businfo[bus]) == NULL)
2381 			continue;
2382 
2383 		for (slot = 0; slot < MAXSLOTS; slot++) {
2384 			si = &bi->slotinfo[slot];
2385 			for (func = 0; func < MAXFUNCS; func++) {
2386 				fi = &si->si_funcs[func];
2387 				if (fi->fi_pde == NULL)
2388 					continue;
2389 				if (strcmp(dev_name, fi->fi_pde->pe_emu) != 0)
2390 					continue;
2391 
2392 				*pde = fi->fi_pde;
2393 				*pdi = fi->fi_devi;
2394 				return (0);
2395 			}
2396 		}
2397 	}
2398 
2399 	return (EINVAL);
2400 }
2401 
2402 int
2403 pci_snapshot(struct vm_snapshot_meta *meta)
2404 {
2405 	struct pci_devemu *pde;
2406 	struct pci_devinst *pdi;
2407 	int ret;
2408 
2409 	assert(meta->dev_name != NULL);
2410 
2411 	ret = pci_find_slotted_dev(meta->dev_name, &pde, &pdi);
2412 	if (ret != 0) {
2413 		fprintf(stderr, "%s: no such name: %s\r\n",
2414 			__func__, meta->dev_name);
2415 		memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
2416 		return (0);
2417 	}
2418 
2419 	meta->dev_data = pdi;
2420 
2421 	if (pde->pe_snapshot == NULL) {
2422 		fprintf(stderr, "%s: not implemented yet for: %s\r\n",
2423 			__func__, meta->dev_name);
2424 		return (-1);
2425 	}
2426 
2427 	ret = pci_snapshot_pci_dev(meta);
2428 	if (ret != 0) {
2429 		fprintf(stderr, "%s: failed to snapshot pci dev\r\n",
2430 			__func__);
2431 		return (-1);
2432 	}
2433 
2434 	ret = (*pde->pe_snapshot)(meta);
2435 
2436 	return (ret);
2437 }
2438 
2439 int
2440 pci_pause(const char *dev_name)
2441 {
2442 	struct pci_devemu *pde;
2443 	struct pci_devinst *pdi;
2444 	int ret;
2445 
2446 	assert(dev_name != NULL);
2447 
2448 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2449 	if (ret != 0) {
2450 		/*
2451 		 * It is possible to call this function without
2452 		 * checking that the device is inserted first.
2453 		 */
2454 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2455 		return (0);
2456 	}
2457 
2458 	if (pde->pe_pause == NULL) {
2459 		/* The pause/resume functionality is optional. */
2460 		fprintf(stderr, "%s: not implemented for: %s\n",
2461 			__func__, dev_name);
2462 		return (0);
2463 	}
2464 
2465 	return (*pde->pe_pause)(pdi);
2466 }
2467 
2468 int
2469 pci_resume(const char *dev_name)
2470 {
2471 	struct pci_devemu *pde;
2472 	struct pci_devinst *pdi;
2473 	int ret;
2474 
2475 	assert(dev_name != NULL);
2476 
2477 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2478 	if (ret != 0) {
2479 		/*
2480 		 * It is possible to call this function without
2481 		 * checking that the device is inserted first.
2482 		 */
2483 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2484 		return (0);
2485 	}
2486 
2487 	if (pde->pe_resume == NULL) {
2488 		/* The pause/resume functionality is optional. */
2489 		fprintf(stderr, "%s: not implemented for: %s\n",
2490 			__func__, dev_name);
2491 		return (0);
2492 	}
2493 
2494 	return (*pde->pe_resume)(pdi);
2495 }
2496 #endif
2497 
2498 #define PCI_EMUL_TEST
2499 #ifdef PCI_EMUL_TEST
2500 /*
2501  * Define a dummy test device
2502  */
2503 #define DIOSZ	8
2504 #define DMEMSZ	4096
2505 struct pci_emul_dsoftc {
2506 	uint8_t   ioregs[DIOSZ];
2507 	uint8_t	  memregs[2][DMEMSZ];
2508 };
2509 
2510 #define	PCI_EMUL_MSI_MSGS	 4
2511 #define	PCI_EMUL_MSIX_MSGS	16
2512 
2513 static int
2514 pci_emul_dinit(struct pci_devinst *pi, nvlist_t *nvl __unused)
2515 {
2516 	int error;
2517 	struct pci_emul_dsoftc *sc;
2518 
2519 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
2520 
2521 	pi->pi_arg = sc;
2522 
2523 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
2524 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
2525 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
2526 
2527 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
2528 	assert(error == 0);
2529 
2530 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
2531 	assert(error == 0);
2532 
2533 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2534 	assert(error == 0);
2535 
2536 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2537 	assert(error == 0);
2538 
2539 	return (0);
2540 }
2541 
2542 static void
2543 pci_emul_diow(struct pci_devinst *pi, int baridx, uint64_t offset, int size,
2544     uint64_t value)
2545 {
2546 	int i;
2547 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2548 
2549 	if (baridx == 0) {
2550 		if (offset + size > DIOSZ) {
2551 			printf("diow: iow too large, offset %ld size %d\n",
2552 			       offset, size);
2553 			return;
2554 		}
2555 
2556 		if (size == 1) {
2557 			sc->ioregs[offset] = value & 0xff;
2558 		} else if (size == 2) {
2559 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2560 		} else if (size == 4) {
2561 			*(uint32_t *)&sc->ioregs[offset] = value;
2562 		} else {
2563 			printf("diow: iow unknown size %d\n", size);
2564 		}
2565 
2566 		/*
2567 		 * Special magic value to generate an interrupt
2568 		 */
2569 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2570 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2571 
2572 		if (value == 0xabcdef) {
2573 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2574 				pci_generate_msi(pi, i);
2575 		}
2576 	}
2577 
2578 	if (baridx == 1 || baridx == 2) {
2579 		if (offset + size > DMEMSZ) {
2580 			printf("diow: memw too large, offset %ld size %d\n",
2581 			       offset, size);
2582 			return;
2583 		}
2584 
2585 		i = baridx - 1;		/* 'memregs' index */
2586 
2587 		if (size == 1) {
2588 			sc->memregs[i][offset] = value;
2589 		} else if (size == 2) {
2590 			*(uint16_t *)&sc->memregs[i][offset] = value;
2591 		} else if (size == 4) {
2592 			*(uint32_t *)&sc->memregs[i][offset] = value;
2593 		} else if (size == 8) {
2594 			*(uint64_t *)&sc->memregs[i][offset] = value;
2595 		} else {
2596 			printf("diow: memw unknown size %d\n", size);
2597 		}
2598 
2599 		/*
2600 		 * magic interrupt ??
2601 		 */
2602 	}
2603 
2604 	if (baridx > 2 || baridx < 0) {
2605 		printf("diow: unknown bar idx %d\n", baridx);
2606 	}
2607 }
2608 
2609 static uint64_t
2610 pci_emul_dior(struct pci_devinst *pi, int baridx, uint64_t offset, int size)
2611 {
2612 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2613 	uint32_t value;
2614 	int i;
2615 
2616 	if (baridx == 0) {
2617 		if (offset + size > DIOSZ) {
2618 			printf("dior: ior too large, offset %ld size %d\n",
2619 			       offset, size);
2620 			return (0);
2621 		}
2622 
2623 		value = 0;
2624 		if (size == 1) {
2625 			value = sc->ioregs[offset];
2626 		} else if (size == 2) {
2627 			value = *(uint16_t *) &sc->ioregs[offset];
2628 		} else if (size == 4) {
2629 			value = *(uint32_t *) &sc->ioregs[offset];
2630 		} else {
2631 			printf("dior: ior unknown size %d\n", size);
2632 		}
2633 	}
2634 
2635 	if (baridx == 1 || baridx == 2) {
2636 		if (offset + size > DMEMSZ) {
2637 			printf("dior: memr too large, offset %ld size %d\n",
2638 			       offset, size);
2639 			return (0);
2640 		}
2641 
2642 		i = baridx - 1;		/* 'memregs' index */
2643 
2644 		if (size == 1) {
2645 			value = sc->memregs[i][offset];
2646 		} else if (size == 2) {
2647 			value = *(uint16_t *) &sc->memregs[i][offset];
2648 		} else if (size == 4) {
2649 			value = *(uint32_t *) &sc->memregs[i][offset];
2650 		} else if (size == 8) {
2651 			value = *(uint64_t *) &sc->memregs[i][offset];
2652 		} else {
2653 			printf("dior: ior unknown size %d\n", size);
2654 		}
2655 	}
2656 
2657 
2658 	if (baridx > 2 || baridx < 0) {
2659 		printf("dior: unknown bar idx %d\n", baridx);
2660 		return (0);
2661 	}
2662 
2663 	return (value);
2664 }
2665 
2666 #ifdef BHYVE_SNAPSHOT
2667 static int
2668 pci_emul_snapshot(struct vm_snapshot_meta *meta __unused)
2669 {
2670 	return (0);
2671 }
2672 #endif
2673 
2674 static const struct pci_devemu pci_dummy = {
2675 	.pe_emu = "dummy",
2676 	.pe_init = pci_emul_dinit,
2677 	.pe_barwrite = pci_emul_diow,
2678 	.pe_barread = pci_emul_dior,
2679 #ifdef BHYVE_SNAPSHOT
2680 	.pe_snapshot = pci_emul_snapshot,
2681 #endif
2682 };
2683 PCI_EMUL_SET(pci_dummy);
2684 
2685 #endif /* PCI_EMUL_TEST */
2686