xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision f5147e312f43a9050468de539aeafa072caa1a60)
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 
37 #include <ctype.h>
38 #include <errno.h>
39 #include <pthread.h>
40 #include <stdio.h>
41 #include <stdlib.h>
42 #include <string.h>
43 #include <strings.h>
44 #include <assert.h>
45 #include <stdbool.h>
46 
47 #include <machine/vmm.h>
48 #include <vmmapi.h>
49 
50 #include "acpi.h"
51 #include "bhyverun.h"
52 #include "inout.h"
53 #include "ioapic.h"
54 #include "mem.h"
55 #include "pci_emul.h"
56 #include "pci_irq.h"
57 #include "pci_lpc.h"
58 
59 #define CONF1_ADDR_PORT    0x0cf8
60 #define CONF1_DATA_PORT    0x0cfc
61 
62 #define CONF1_ENABLE	   0x80000000ul
63 
64 #define	MAXBUSES	(PCI_BUSMAX + 1)
65 #define MAXSLOTS	(PCI_SLOTMAX + 1)
66 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
67 
68 struct funcinfo {
69 	char	*fi_name;
70 	char	*fi_param;
71 	struct pci_devinst *fi_devi;
72 };
73 
74 struct intxinfo {
75 	int	ii_count;
76 	int	ii_pirq_pin;
77 	int	ii_ioapic_irq;
78 };
79 
80 struct slotinfo {
81 	struct intxinfo si_intpins[4];
82 	struct funcinfo si_funcs[MAXFUNCS];
83 };
84 
85 struct businfo {
86 	uint16_t iobase, iolimit;		/* I/O window */
87 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
88 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
89 	struct slotinfo slotinfo[MAXSLOTS];
90 };
91 
92 static struct businfo *pci_businfo[MAXBUSES];
93 
94 SET_DECLARE(pci_devemu_set, struct pci_devemu);
95 
96 static uint64_t pci_emul_iobase;
97 static uint64_t pci_emul_membase32;
98 static uint64_t pci_emul_membase64;
99 
100 #define	PCI_EMUL_IOBASE		0x2000
101 #define	PCI_EMUL_IOLIMIT	0x10000
102 
103 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
104 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
105 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
106 
107 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
108 
109 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
110 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
111 
112 static struct pci_devemu *pci_emul_finddev(char *name);
113 static void pci_lintr_route(struct pci_devinst *pi);
114 static void pci_lintr_update(struct pci_devinst *pi);
115 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
116     int func, int coff, int bytes, uint32_t *val);
117 
118 static __inline void
119 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
120 {
121 
122 	if (bytes == 1)
123 		pci_set_cfgdata8(pi, coff, val);
124 	else if (bytes == 2)
125 		pci_set_cfgdata16(pi, coff, val);
126 	else
127 		pci_set_cfgdata32(pi, coff, val);
128 }
129 
130 static __inline uint32_t
131 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
132 {
133 
134 	if (bytes == 1)
135 		return (pci_get_cfgdata8(pi, coff));
136 	else if (bytes == 2)
137 		return (pci_get_cfgdata16(pi, coff));
138 	else
139 		return (pci_get_cfgdata32(pi, coff));
140 }
141 
142 /*
143  * I/O access
144  */
145 
146 /*
147  * Slot options are in the form:
148  *
149  *  <bus>:<slot>:<func>,<emul>[,<config>]
150  *  <slot>[:<func>],<emul>[,<config>]
151  *
152  *  slot is 0..31
153  *  func is 0..7
154  *  emul is a string describing the type of PCI device e.g. virtio-net
155  *  config is an optional string, depending on the device, that can be
156  *  used for configuration.
157  *   Examples are:
158  *     1,virtio-net,tap0
159  *     3:0,dummy
160  */
161 static void
162 pci_parse_slot_usage(char *aopt)
163 {
164 
165 	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
166 }
167 
168 int
169 pci_parse_slot(char *opt)
170 {
171 	struct businfo *bi;
172 	struct slotinfo *si;
173 	char *emul, *config, *str, *cp;
174 	int error, bnum, snum, fnum;
175 
176 	error = -1;
177 	str = strdup(opt);
178 
179 	emul = config = NULL;
180 	if ((cp = strchr(str, ',')) != NULL) {
181 		*cp = '\0';
182 		emul = cp + 1;
183 		if ((cp = strchr(emul, ',')) != NULL) {
184 			*cp = '\0';
185 			config = cp + 1;
186 		}
187 	} else {
188 		pci_parse_slot_usage(opt);
189 		goto done;
190 	}
191 
192 	/* <bus>:<slot>:<func> */
193 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
194 		bnum = 0;
195 		/* <slot>:<func> */
196 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
197 			fnum = 0;
198 			/* <slot> */
199 			if (sscanf(str, "%d", &snum) != 1) {
200 				snum = -1;
201 			}
202 		}
203 	}
204 
205 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
206 	    fnum < 0 || fnum >= MAXFUNCS) {
207 		pci_parse_slot_usage(opt);
208 		goto done;
209 	}
210 
211 	if (pci_businfo[bnum] == NULL)
212 		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
213 
214 	bi = pci_businfo[bnum];
215 	si = &bi->slotinfo[snum];
216 
217 	if (si->si_funcs[fnum].fi_name != NULL) {
218 		fprintf(stderr, "pci slot %d:%d already occupied!\n",
219 			snum, fnum);
220 		goto done;
221 	}
222 
223 	if (pci_emul_finddev(emul) == NULL) {
224 		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
225 			snum, fnum, emul);
226 		goto done;
227 	}
228 
229 	error = 0;
230 	si->si_funcs[fnum].fi_name = emul;
231 	si->si_funcs[fnum].fi_param = config;
232 
233 done:
234 	if (error)
235 		free(str);
236 
237 	return (error);
238 }
239 
240 static int
241 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
242 {
243 
244 	if (offset < pi->pi_msix.pba_offset)
245 		return (0);
246 
247 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
248 		return (0);
249 	}
250 
251 	return (1);
252 }
253 
254 int
255 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
256 		     uint64_t value)
257 {
258 	int msix_entry_offset;
259 	int tab_index;
260 	char *dest;
261 
262 	/* support only 4 or 8 byte writes */
263 	if (size != 4 && size != 8)
264 		return (-1);
265 
266 	/*
267 	 * Return if table index is beyond what device supports
268 	 */
269 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
270 	if (tab_index >= pi->pi_msix.table_count)
271 		return (-1);
272 
273 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
274 
275 	/* support only aligned writes */
276 	if ((msix_entry_offset % size) != 0)
277 		return (-1);
278 
279 	dest = (char *)(pi->pi_msix.table + tab_index);
280 	dest += msix_entry_offset;
281 
282 	if (size == 4)
283 		*((uint32_t *)dest) = value;
284 	else
285 		*((uint64_t *)dest) = value;
286 
287 	return (0);
288 }
289 
290 uint64_t
291 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
292 {
293 	char *dest;
294 	int msix_entry_offset;
295 	int tab_index;
296 	uint64_t retval = ~0;
297 
298 	/*
299 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
300 	 * table but we also allow 1 byte access to accommodate reads from
301 	 * ddb.
302 	 */
303 	if (size != 1 && size != 4 && size != 8)
304 		return (retval);
305 
306 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
307 
308 	/* support only aligned reads */
309 	if ((msix_entry_offset % size) != 0) {
310 		return (retval);
311 	}
312 
313 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
314 
315 	if (tab_index < pi->pi_msix.table_count) {
316 		/* valid MSI-X Table access */
317 		dest = (char *)(pi->pi_msix.table + tab_index);
318 		dest += msix_entry_offset;
319 
320 		if (size == 1)
321 			retval = *((uint8_t *)dest);
322 		else if (size == 4)
323 			retval = *((uint32_t *)dest);
324 		else
325 			retval = *((uint64_t *)dest);
326 	} else if (pci_valid_pba_offset(pi, offset)) {
327 		/* return 0 for PBA access */
328 		retval = 0;
329 	}
330 
331 	return (retval);
332 }
333 
334 int
335 pci_msix_table_bar(struct pci_devinst *pi)
336 {
337 
338 	if (pi->pi_msix.table != NULL)
339 		return (pi->pi_msix.table_bar);
340 	else
341 		return (-1);
342 }
343 
344 int
345 pci_msix_pba_bar(struct pci_devinst *pi)
346 {
347 
348 	if (pi->pi_msix.table != NULL)
349 		return (pi->pi_msix.pba_bar);
350 	else
351 		return (-1);
352 }
353 
354 static int
355 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
356 		    uint32_t *eax, void *arg)
357 {
358 	struct pci_devinst *pdi = arg;
359 	struct pci_devemu *pe = pdi->pi_d;
360 	uint64_t offset;
361 	int i;
362 
363 	for (i = 0; i <= PCI_BARMAX; i++) {
364 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
365 		    port >= pdi->pi_bar[i].addr &&
366 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
367 			offset = port - pdi->pi_bar[i].addr;
368 			if (in)
369 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
370 							 offset, bytes);
371 			else
372 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
373 						   bytes, *eax);
374 			return (0);
375 		}
376 	}
377 	return (-1);
378 }
379 
380 static int
381 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
382 		     int size, uint64_t *val, void *arg1, long arg2)
383 {
384 	struct pci_devinst *pdi = arg1;
385 	struct pci_devemu *pe = pdi->pi_d;
386 	uint64_t offset;
387 	int bidx = (int) arg2;
388 
389 	assert(bidx <= PCI_BARMAX);
390 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
391 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
392 	assert(addr >= pdi->pi_bar[bidx].addr &&
393 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
394 
395 	offset = addr - pdi->pi_bar[bidx].addr;
396 
397 	if (dir == MEM_F_WRITE) {
398 		if (size == 8) {
399 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
400 					   4, *val & 0xffffffff);
401 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
402 					   4, *val >> 32);
403 		} else {
404 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
405 					   size, *val);
406 		}
407 	} else {
408 		if (size == 8) {
409 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
410 						 offset, 4);
411 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
412 						  offset + 4, 4) << 32;
413 		} else {
414 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
415 						 offset, size);
416 		}
417 	}
418 
419 	return (0);
420 }
421 
422 
423 static int
424 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
425 			uint64_t *addr)
426 {
427 	uint64_t base;
428 
429 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
430 
431 	base = roundup2(*baseptr, size);
432 
433 	if (base + size <= limit) {
434 		*addr = base;
435 		*baseptr = base + size;
436 		return (0);
437 	} else
438 		return (-1);
439 }
440 
441 int
442 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
443 		   uint64_t size)
444 {
445 
446 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
447 }
448 
449 /*
450  * Register (or unregister) the MMIO or I/O region associated with the BAR
451  * register 'idx' of an emulated pci device.
452  */
453 static void
454 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
455 {
456 	int error;
457 	struct inout_port iop;
458 	struct mem_range mr;
459 
460 	switch (pi->pi_bar[idx].type) {
461 	case PCIBAR_IO:
462 		bzero(&iop, sizeof(struct inout_port));
463 		iop.name = pi->pi_name;
464 		iop.port = pi->pi_bar[idx].addr;
465 		iop.size = pi->pi_bar[idx].size;
466 		if (registration) {
467 			iop.flags = IOPORT_F_INOUT;
468 			iop.handler = pci_emul_io_handler;
469 			iop.arg = pi;
470 			error = register_inout(&iop);
471 		} else
472 			error = unregister_inout(&iop);
473 		break;
474 	case PCIBAR_MEM32:
475 	case PCIBAR_MEM64:
476 		bzero(&mr, sizeof(struct mem_range));
477 		mr.name = pi->pi_name;
478 		mr.base = pi->pi_bar[idx].addr;
479 		mr.size = pi->pi_bar[idx].size;
480 		if (registration) {
481 			mr.flags = MEM_F_RW;
482 			mr.handler = pci_emul_mem_handler;
483 			mr.arg1 = pi;
484 			mr.arg2 = idx;
485 			error = register_mem(&mr);
486 		} else
487 			error = unregister_mem(&mr);
488 		break;
489 	default:
490 		error = EINVAL;
491 		break;
492 	}
493 	assert(error == 0);
494 }
495 
496 static void
497 unregister_bar(struct pci_devinst *pi, int idx)
498 {
499 
500 	modify_bar_registration(pi, idx, 0);
501 }
502 
503 static void
504 register_bar(struct pci_devinst *pi, int idx)
505 {
506 
507 	modify_bar_registration(pi, idx, 1);
508 }
509 
510 /* Are we decoding i/o port accesses for the emulated pci device? */
511 static int
512 porten(struct pci_devinst *pi)
513 {
514 	uint16_t cmd;
515 
516 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
517 
518 	return (cmd & PCIM_CMD_PORTEN);
519 }
520 
521 /* Are we decoding memory accesses for the emulated pci device? */
522 static int
523 memen(struct pci_devinst *pi)
524 {
525 	uint16_t cmd;
526 
527 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
528 
529 	return (cmd & PCIM_CMD_MEMEN);
530 }
531 
532 /*
533  * Update the MMIO or I/O address that is decoded by the BAR register.
534  *
535  * If the pci device has enabled the address space decoding then intercept
536  * the address range decoded by the BAR register.
537  */
538 static void
539 update_bar_address(struct  pci_devinst *pi, uint64_t addr, int idx, int type)
540 {
541 	int decode;
542 
543 	if (pi->pi_bar[idx].type == PCIBAR_IO)
544 		decode = porten(pi);
545 	else
546 		decode = memen(pi);
547 
548 	if (decode)
549 		unregister_bar(pi, idx);
550 
551 	switch (type) {
552 	case PCIBAR_IO:
553 	case PCIBAR_MEM32:
554 		pi->pi_bar[idx].addr = addr;
555 		break;
556 	case PCIBAR_MEM64:
557 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
558 		pi->pi_bar[idx].addr |= addr;
559 		break;
560 	case PCIBAR_MEMHI64:
561 		pi->pi_bar[idx].addr &= 0xffffffff;
562 		pi->pi_bar[idx].addr |= addr;
563 		break;
564 	default:
565 		assert(0);
566 	}
567 
568 	if (decode)
569 		register_bar(pi, idx);
570 }
571 
572 int
573 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
574 		    enum pcibar_type type, uint64_t size)
575 {
576 	int error;
577 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
578 
579 	assert(idx >= 0 && idx <= PCI_BARMAX);
580 
581 	if ((size & (size - 1)) != 0)
582 		size = 1UL << flsl(size);	/* round up to a power of 2 */
583 
584 	/* Enforce minimum BAR sizes required by the PCI standard */
585 	if (type == PCIBAR_IO) {
586 		if (size < 4)
587 			size = 4;
588 	} else {
589 		if (size < 16)
590 			size = 16;
591 	}
592 
593 	switch (type) {
594 	case PCIBAR_NONE:
595 		baseptr = NULL;
596 		addr = mask = lobits = 0;
597 		break;
598 	case PCIBAR_IO:
599 		baseptr = &pci_emul_iobase;
600 		limit = PCI_EMUL_IOLIMIT;
601 		mask = PCIM_BAR_IO_BASE;
602 		lobits = PCIM_BAR_IO_SPACE;
603 		break;
604 	case PCIBAR_MEM64:
605 		/*
606 		 * XXX
607 		 * Some drivers do not work well if the 64-bit BAR is allocated
608 		 * above 4GB. Allow for this by allocating small requests under
609 		 * 4GB unless then allocation size is larger than some arbitrary
610 		 * number (32MB currently).
611 		 */
612 		if (size > 32 * 1024 * 1024) {
613 			/*
614 			 * XXX special case for device requiring peer-peer DMA
615 			 */
616 			if (size == 0x100000000UL)
617 				baseptr = &hostbase;
618 			else
619 				baseptr = &pci_emul_membase64;
620 			limit = PCI_EMUL_MEMLIMIT64;
621 			mask = PCIM_BAR_MEM_BASE;
622 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
623 				 PCIM_BAR_MEM_PREFETCH;
624 			break;
625 		} else {
626 			baseptr = &pci_emul_membase32;
627 			limit = PCI_EMUL_MEMLIMIT32;
628 			mask = PCIM_BAR_MEM_BASE;
629 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
630 		}
631 		break;
632 	case PCIBAR_MEM32:
633 		baseptr = &pci_emul_membase32;
634 		limit = PCI_EMUL_MEMLIMIT32;
635 		mask = PCIM_BAR_MEM_BASE;
636 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
637 		break;
638 	default:
639 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
640 		assert(0);
641 	}
642 
643 	if (baseptr != NULL) {
644 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
645 		if (error != 0)
646 			return (error);
647 	}
648 
649 	pdi->pi_bar[idx].type = type;
650 	pdi->pi_bar[idx].addr = addr;
651 	pdi->pi_bar[idx].size = size;
652 
653 	/* Initialize the BAR register in config space */
654 	bar = (addr & mask) | lobits;
655 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
656 
657 	if (type == PCIBAR_MEM64) {
658 		assert(idx + 1 <= PCI_BARMAX);
659 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
660 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
661 	}
662 
663 	register_bar(pdi, idx);
664 
665 	return (0);
666 }
667 
668 #define	CAP_START_OFFSET	0x40
669 static int
670 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
671 {
672 	int i, capoff, reallen;
673 	uint16_t sts;
674 
675 	assert(caplen > 0);
676 
677 	reallen = roundup2(caplen, 4);		/* dword aligned */
678 
679 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
680 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
681 		capoff = CAP_START_OFFSET;
682 	else
683 		capoff = pi->pi_capend + 1;
684 
685 	/* Check if we have enough space */
686 	if (capoff + reallen > PCI_REGMAX + 1)
687 		return (-1);
688 
689 	/* Set the previous capability pointer */
690 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
691 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
692 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
693 	} else
694 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
695 
696 	/* Copy the capability */
697 	for (i = 0; i < caplen; i++)
698 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
699 
700 	/* Set the next capability pointer */
701 	pci_set_cfgdata8(pi, capoff + 1, 0);
702 
703 	pi->pi_prevcap = capoff;
704 	pi->pi_capend = capoff + reallen - 1;
705 	return (0);
706 }
707 
708 static struct pci_devemu *
709 pci_emul_finddev(char *name)
710 {
711 	struct pci_devemu **pdpp, *pdp;
712 
713 	SET_FOREACH(pdpp, pci_devemu_set) {
714 		pdp = *pdpp;
715 		if (!strcmp(pdp->pe_emu, name)) {
716 			return (pdp);
717 		}
718 	}
719 
720 	return (NULL);
721 }
722 
723 static int
724 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
725     int func, struct funcinfo *fi)
726 {
727 	struct pci_devinst *pdi;
728 	int err;
729 
730 	pdi = calloc(1, sizeof(struct pci_devinst));
731 
732 	pdi->pi_vmctx = ctx;
733 	pdi->pi_bus = bus;
734 	pdi->pi_slot = slot;
735 	pdi->pi_func = func;
736 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
737 	pdi->pi_lintr.pin = 0;
738 	pdi->pi_lintr.state = IDLE;
739 	pdi->pi_lintr.pirq_pin = 0;
740 	pdi->pi_lintr.ioapic_irq = 0;
741 	pdi->pi_d = pde;
742 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
743 
744 	/* Disable legacy interrupts */
745 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
746 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
747 
748 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
749 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
750 
751 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
752 	if (err == 0)
753 		fi->fi_devi = pdi;
754 	else
755 		free(pdi);
756 
757 	return (err);
758 }
759 
760 void
761 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
762 {
763 	int mmc;
764 
765 	/* Number of msi messages must be a power of 2 between 1 and 32 */
766 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
767 	mmc = ffs(msgnum) - 1;
768 
769 	bzero(msicap, sizeof(struct msicap));
770 	msicap->capid = PCIY_MSI;
771 	msicap->nextptr = nextptr;
772 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
773 }
774 
775 int
776 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
777 {
778 	struct msicap msicap;
779 
780 	pci_populate_msicap(&msicap, msgnum, 0);
781 
782 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
783 }
784 
785 static void
786 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
787 		     uint32_t msix_tab_size)
788 {
789 
790 	assert(msix_tab_size % 4096 == 0);
791 
792 	bzero(msixcap, sizeof(struct msixcap));
793 	msixcap->capid = PCIY_MSIX;
794 
795 	/*
796 	 * Message Control Register, all fields set to
797 	 * zero except for the Table Size.
798 	 * Note: Table size N is encoded as N-1
799 	 */
800 	msixcap->msgctrl = msgnum - 1;
801 
802 	/*
803 	 * MSI-X BAR setup:
804 	 * - MSI-X table start at offset 0
805 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
806 	 */
807 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
808 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
809 }
810 
811 static void
812 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
813 {
814 	int i, table_size;
815 
816 	assert(table_entries > 0);
817 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
818 
819 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
820 	pi->pi_msix.table = calloc(1, table_size);
821 
822 	/* set mask bit of vector control register */
823 	for (i = 0; i < table_entries; i++)
824 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
825 }
826 
827 int
828 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
829 {
830 	uint32_t tab_size;
831 	struct msixcap msixcap;
832 
833 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
834 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
835 
836 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
837 
838 	/* Align table size to nearest 4K */
839 	tab_size = roundup2(tab_size, 4096);
840 
841 	pi->pi_msix.table_bar = barnum;
842 	pi->pi_msix.pba_bar   = barnum;
843 	pi->pi_msix.table_offset = 0;
844 	pi->pi_msix.table_count = msgnum;
845 	pi->pi_msix.pba_offset = tab_size;
846 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
847 
848 	pci_msix_table_init(pi, msgnum);
849 
850 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
851 
852 	/* allocate memory for MSI-X Table and PBA */
853 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
854 				tab_size + pi->pi_msix.pba_size);
855 
856 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
857 					sizeof(msixcap)));
858 }
859 
860 void
861 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
862 		 int bytes, uint32_t val)
863 {
864 	uint16_t msgctrl, rwmask;
865 	int off;
866 
867 	off = offset - capoff;
868 	/* Message Control Register */
869 	if (off == 2 && bytes == 2) {
870 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
871 		msgctrl = pci_get_cfgdata16(pi, offset);
872 		msgctrl &= ~rwmask;
873 		msgctrl |= val & rwmask;
874 		val = msgctrl;
875 
876 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
877 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
878 		pci_lintr_update(pi);
879 	}
880 
881 	CFGWRITE(pi, offset, val, bytes);
882 }
883 
884 void
885 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
886 		int bytes, uint32_t val)
887 {
888 	uint16_t msgctrl, rwmask, msgdata, mme;
889 	uint32_t addrlo;
890 
891 	/*
892 	 * If guest is writing to the message control register make sure
893 	 * we do not overwrite read-only fields.
894 	 */
895 	if ((offset - capoff) == 2 && bytes == 2) {
896 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
897 		msgctrl = pci_get_cfgdata16(pi, offset);
898 		msgctrl &= ~rwmask;
899 		msgctrl |= val & rwmask;
900 		val = msgctrl;
901 
902 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
903 		if (msgctrl & PCIM_MSICTRL_64BIT)
904 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
905 		else
906 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
907 
908 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
909 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
910 		if (pi->pi_msi.enabled) {
911 			pi->pi_msi.addr = addrlo;
912 			pi->pi_msi.msg_data = msgdata;
913 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
914 		} else {
915 			pi->pi_msi.maxmsgnum = 0;
916 		}
917 		pci_lintr_update(pi);
918 	}
919 
920 	CFGWRITE(pi, offset, val, bytes);
921 }
922 
923 void
924 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
925 		 int bytes, uint32_t val)
926 {
927 
928 	/* XXX don't write to the readonly parts */
929 	CFGWRITE(pi, offset, val, bytes);
930 }
931 
932 #define	PCIECAP_VERSION	0x2
933 int
934 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
935 {
936 	int err;
937 	struct pciecap pciecap;
938 
939 	if (type != PCIEM_TYPE_ROOT_PORT)
940 		return (-1);
941 
942 	bzero(&pciecap, sizeof(pciecap));
943 
944 	pciecap.capid = PCIY_EXPRESS;
945 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
946 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
947 	pciecap.link_status = 0x11;		/* gen1, x1 */
948 
949 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
950 	return (err);
951 }
952 
953 /*
954  * This function assumes that 'coff' is in the capabilities region of the
955  * config space.
956  */
957 static void
958 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
959 {
960 	int capid;
961 	uint8_t capoff, nextoff;
962 
963 	/* Do not allow un-aligned writes */
964 	if ((offset & (bytes - 1)) != 0)
965 		return;
966 
967 	/* Find the capability that we want to update */
968 	capoff = CAP_START_OFFSET;
969 	while (1) {
970 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
971 		if (nextoff == 0)
972 			break;
973 		if (offset >= capoff && offset < nextoff)
974 			break;
975 
976 		capoff = nextoff;
977 	}
978 	assert(offset >= capoff);
979 
980 	/*
981 	 * Capability ID and Next Capability Pointer are readonly.
982 	 * However, some o/s's do 4-byte writes that include these.
983 	 * For this case, trim the write back to 2 bytes and adjust
984 	 * the data.
985 	 */
986 	if (offset == capoff || offset == capoff + 1) {
987 		if (offset == capoff && bytes == 4) {
988 			bytes = 2;
989 			offset += 2;
990 			val >>= 16;
991 		} else
992 			return;
993 	}
994 
995 	capid = pci_get_cfgdata8(pi, capoff);
996 	switch (capid) {
997 	case PCIY_MSI:
998 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
999 		break;
1000 	case PCIY_MSIX:
1001 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1002 		break;
1003 	case PCIY_EXPRESS:
1004 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1005 		break;
1006 	default:
1007 		break;
1008 	}
1009 }
1010 
1011 static int
1012 pci_emul_iscap(struct pci_devinst *pi, int offset)
1013 {
1014 	uint16_t sts;
1015 
1016 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1017 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1018 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1019 			return (1);
1020 	}
1021 	return (0);
1022 }
1023 
1024 static int
1025 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1026 			  int size, uint64_t *val, void *arg1, long arg2)
1027 {
1028 	/*
1029 	 * Ignore writes; return 0xff's for reads. The mem read code
1030 	 * will take care of truncating to the correct size.
1031 	 */
1032 	if (dir == MEM_F_READ) {
1033 		*val = 0xffffffffffffffff;
1034 	}
1035 
1036 	return (0);
1037 }
1038 
1039 static int
1040 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1041     int bytes, uint64_t *val, void *arg1, long arg2)
1042 {
1043 	int bus, slot, func, coff, in;
1044 
1045 	coff = addr & 0xfff;
1046 	func = (addr >> 12) & 0x7;
1047 	slot = (addr >> 15) & 0x1f;
1048 	bus = (addr >> 20) & 0xff;
1049 	in = (dir == MEM_F_READ);
1050 	if (in)
1051 		*val = ~0UL;
1052 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1053 	return (0);
1054 }
1055 
1056 uint64_t
1057 pci_ecfg_base(void)
1058 {
1059 
1060 	return (PCI_EMUL_ECFG_BASE);
1061 }
1062 
1063 #define	BUSIO_ROUNDUP		32
1064 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1065 
1066 int
1067 init_pci(struct vmctx *ctx)
1068 {
1069 	struct mem_range mr;
1070 	struct pci_devemu *pde;
1071 	struct businfo *bi;
1072 	struct slotinfo *si;
1073 	struct funcinfo *fi;
1074 	size_t lowmem;
1075 	int bus, slot, func;
1076 	int error;
1077 
1078 	pci_emul_iobase = PCI_EMUL_IOBASE;
1079 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1080 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1081 
1082 	for (bus = 0; bus < MAXBUSES; bus++) {
1083 		if ((bi = pci_businfo[bus]) == NULL)
1084 			continue;
1085 		/*
1086 		 * Keep track of the i/o and memory resources allocated to
1087 		 * this bus.
1088 		 */
1089 		bi->iobase = pci_emul_iobase;
1090 		bi->membase32 = pci_emul_membase32;
1091 		bi->membase64 = pci_emul_membase64;
1092 
1093 		for (slot = 0; slot < MAXSLOTS; slot++) {
1094 			si = &bi->slotinfo[slot];
1095 			for (func = 0; func < MAXFUNCS; func++) {
1096 				fi = &si->si_funcs[func];
1097 				if (fi->fi_name == NULL)
1098 					continue;
1099 				pde = pci_emul_finddev(fi->fi_name);
1100 				assert(pde != NULL);
1101 				error = pci_emul_init(ctx, pde, bus, slot,
1102 				    func, fi);
1103 				if (error)
1104 					return (error);
1105 			}
1106 		}
1107 
1108 		/*
1109 		 * Add some slop to the I/O and memory resources decoded by
1110 		 * this bus to give a guest some flexibility if it wants to
1111 		 * reprogram the BARs.
1112 		 */
1113 		pci_emul_iobase += BUSIO_ROUNDUP;
1114 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1115 		bi->iolimit = pci_emul_iobase;
1116 
1117 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1118 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1119 		    BUSMEM_ROUNDUP);
1120 		bi->memlimit32 = pci_emul_membase32;
1121 
1122 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1123 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1124 		    BUSMEM_ROUNDUP);
1125 		bi->memlimit64 = pci_emul_membase64;
1126 	}
1127 
1128 	/*
1129 	 * PCI backends are initialized before routing INTx interrupts
1130 	 * so that LPC devices are able to reserve ISA IRQs before
1131 	 * routing PIRQ pins.
1132 	 */
1133 	for (bus = 0; bus < MAXBUSES; bus++) {
1134 		if ((bi = pci_businfo[bus]) == NULL)
1135 			continue;
1136 
1137 		for (slot = 0; slot < MAXSLOTS; slot++) {
1138 			si = &bi->slotinfo[slot];
1139 			for (func = 0; func < MAXFUNCS; func++) {
1140 				fi = &si->si_funcs[func];
1141 				if (fi->fi_devi == NULL)
1142 					continue;
1143 				pci_lintr_route(fi->fi_devi);
1144 			}
1145 		}
1146 	}
1147 	lpc_pirq_routed();
1148 
1149 	/*
1150 	 * The guest physical memory map looks like the following:
1151 	 * [0,		    lowmem)		guest system memory
1152 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1153 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1154 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1155 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1156 	 * [4GB,	    4GB + highmem)
1157 	 */
1158 
1159 	/*
1160 	 * Accesses to memory addresses that are not allocated to system
1161 	 * memory or PCI devices return 0xff's.
1162 	 */
1163 	lowmem = vm_get_lowmem_size(ctx);
1164 	bzero(&mr, sizeof(struct mem_range));
1165 	mr.name = "PCI hole";
1166 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1167 	mr.base = lowmem;
1168 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1169 	mr.handler = pci_emul_fallback_handler;
1170 	error = register_mem_fallback(&mr);
1171 	assert(error == 0);
1172 
1173 	/* PCI extended config space */
1174 	bzero(&mr, sizeof(struct mem_range));
1175 	mr.name = "PCI ECFG";
1176 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1177 	mr.base = PCI_EMUL_ECFG_BASE;
1178 	mr.size = PCI_EMUL_ECFG_SIZE;
1179 	mr.handler = pci_emul_ecfg_handler;
1180 	error = register_mem(&mr);
1181 	assert(error == 0);
1182 
1183 	return (0);
1184 }
1185 
1186 static void
1187 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1188     void *arg)
1189 {
1190 
1191 	dsdt_line("  Package ()");
1192 	dsdt_line("  {");
1193 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1194 	dsdt_line("    0x%02X,", pin - 1);
1195 	dsdt_line("    Zero,");
1196 	dsdt_line("    0x%X", ioapic_irq);
1197 	dsdt_line("  },");
1198 }
1199 
1200 static void
1201 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1202     void *arg)
1203 {
1204 	char *name;
1205 
1206 	name = lpc_pirq_name(pirq_pin);
1207 	if (name == NULL)
1208 		return;
1209 	dsdt_line("  Package ()");
1210 	dsdt_line("  {");
1211 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1212 	dsdt_line("    0x%02X,", pin - 1);
1213 	dsdt_line("    %s,", name);
1214 	dsdt_line("    0x00");
1215 	dsdt_line("  },");
1216 	free(name);
1217 }
1218 
1219 /*
1220  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1221  * corresponding to each PCI bus.
1222  */
1223 static void
1224 pci_bus_write_dsdt(int bus)
1225 {
1226 	struct businfo *bi;
1227 	struct slotinfo *si;
1228 	struct pci_devinst *pi;
1229 	int count, func, slot;
1230 
1231 	/*
1232 	 * If there are no devices on this 'bus' then just return.
1233 	 */
1234 	if ((bi = pci_businfo[bus]) == NULL) {
1235 		/*
1236 		 * Bus 0 is special because it decodes the I/O ports used
1237 		 * for PCI config space access even if there are no devices
1238 		 * on it.
1239 		 */
1240 		if (bus != 0)
1241 			return;
1242 	}
1243 
1244 	dsdt_line("  Device (PC%02X)", bus);
1245 	dsdt_line("  {");
1246 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1247 	dsdt_line("    Name (_ADR, Zero)");
1248 
1249 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1250 	dsdt_line("    {");
1251 	dsdt_line("        Return (0x%08X)", bus);
1252 	dsdt_line("    }");
1253 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1254 	dsdt_line("    {");
1255 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1256 	    "MaxFixed, PosDecode,");
1257 	dsdt_line("        0x0000,             // Granularity");
1258 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1259 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1260 	dsdt_line("        0x0000,             // Translation Offset");
1261 	dsdt_line("        0x0001,             // Length");
1262 	dsdt_line("        ,, )");
1263 
1264 	if (bus == 0) {
1265 		dsdt_indent(3);
1266 		dsdt_fixed_ioport(0xCF8, 8);
1267 		dsdt_unindent(3);
1268 
1269 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1270 		    "PosDecode, EntireRange,");
1271 		dsdt_line("        0x0000,             // Granularity");
1272 		dsdt_line("        0x0000,             // Range Minimum");
1273 		dsdt_line("        0x0CF7,             // Range Maximum");
1274 		dsdt_line("        0x0000,             // Translation Offset");
1275 		dsdt_line("        0x0CF8,             // Length");
1276 		dsdt_line("        ,, , TypeStatic)");
1277 
1278 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1279 		    "PosDecode, EntireRange,");
1280 		dsdt_line("        0x0000,             // Granularity");
1281 		dsdt_line("        0x0D00,             // Range Minimum");
1282 		dsdt_line("        0x%04X,             // Range Maximum",
1283 		    PCI_EMUL_IOBASE - 1);
1284 		dsdt_line("        0x0000,             // Translation Offset");
1285 		dsdt_line("        0x%04X,             // Length",
1286 		    PCI_EMUL_IOBASE - 0x0D00);
1287 		dsdt_line("        ,, , TypeStatic)");
1288 
1289 		if (bi == NULL) {
1290 			dsdt_line("    })");
1291 			goto done;
1292 		}
1293 	}
1294 	assert(bi != NULL);
1295 
1296 	/* i/o window */
1297 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1298 	    "PosDecode, EntireRange,");
1299 	dsdt_line("        0x0000,             // Granularity");
1300 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1301 	dsdt_line("        0x%04X,             // Range Maximum",
1302 	    bi->iolimit - 1);
1303 	dsdt_line("        0x0000,             // Translation Offset");
1304 	dsdt_line("        0x%04X,             // Length",
1305 	    bi->iolimit - bi->iobase);
1306 	dsdt_line("        ,, , TypeStatic)");
1307 
1308 	/* mmio window (32-bit) */
1309 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1310 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1311 	dsdt_line("        0x00000000,         // Granularity");
1312 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1313 	dsdt_line("        0x%08X,         // Range Maximum\n",
1314 	    bi->memlimit32 - 1);
1315 	dsdt_line("        0x00000000,         // Translation Offset");
1316 	dsdt_line("        0x%08X,         // Length\n",
1317 	    bi->memlimit32 - bi->membase32);
1318 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1319 
1320 	/* mmio window (64-bit) */
1321 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1322 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1323 	dsdt_line("        0x0000000000000000, // Granularity");
1324 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1325 	dsdt_line("        0x%016lX, // Range Maximum\n",
1326 	    bi->memlimit64 - 1);
1327 	dsdt_line("        0x0000000000000000, // Translation Offset");
1328 	dsdt_line("        0x%016lX, // Length\n",
1329 	    bi->memlimit64 - bi->membase64);
1330 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1331 	dsdt_line("    })");
1332 
1333 	count = pci_count_lintr(bus);
1334 	if (count != 0) {
1335 		dsdt_indent(2);
1336 		dsdt_line("Name (PPRT, Package ()");
1337 		dsdt_line("{");
1338 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1339  		dsdt_line("})");
1340 		dsdt_line("Name (APRT, Package ()");
1341 		dsdt_line("{");
1342 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1343  		dsdt_line("})");
1344 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1345 		dsdt_line("{");
1346 		dsdt_line("  If (PICM)");
1347 		dsdt_line("  {");
1348 		dsdt_line("    Return (APRT)");
1349 		dsdt_line("  }");
1350 		dsdt_line("  Else");
1351 		dsdt_line("  {");
1352 		dsdt_line("    Return (PPRT)");
1353 		dsdt_line("  }");
1354 		dsdt_line("}");
1355 		dsdt_unindent(2);
1356 	}
1357 
1358 	dsdt_indent(2);
1359 	for (slot = 0; slot < MAXSLOTS; slot++) {
1360 		si = &bi->slotinfo[slot];
1361 		for (func = 0; func < MAXFUNCS; func++) {
1362 			pi = si->si_funcs[func].fi_devi;
1363 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1364 				pi->pi_d->pe_write_dsdt(pi);
1365 		}
1366 	}
1367 	dsdt_unindent(2);
1368 done:
1369 	dsdt_line("  }");
1370 }
1371 
1372 void
1373 pci_write_dsdt(void)
1374 {
1375 	int bus;
1376 
1377 	dsdt_indent(1);
1378 	dsdt_line("Name (PICM, 0x00)");
1379 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1380 	dsdt_line("{");
1381 	dsdt_line("  Store (Arg0, PICM)");
1382 	dsdt_line("}");
1383 	dsdt_line("");
1384 	dsdt_line("Scope (_SB)");
1385 	dsdt_line("{");
1386 	for (bus = 0; bus < MAXBUSES; bus++)
1387 		pci_bus_write_dsdt(bus);
1388 	dsdt_line("}");
1389 	dsdt_unindent(1);
1390 }
1391 
1392 int
1393 pci_bus_configured(int bus)
1394 {
1395 	assert(bus >= 0 && bus < MAXBUSES);
1396 	return (pci_businfo[bus] != NULL);
1397 }
1398 
1399 int
1400 pci_msi_enabled(struct pci_devinst *pi)
1401 {
1402 	return (pi->pi_msi.enabled);
1403 }
1404 
1405 int
1406 pci_msi_maxmsgnum(struct pci_devinst *pi)
1407 {
1408 	if (pi->pi_msi.enabled)
1409 		return (pi->pi_msi.maxmsgnum);
1410 	else
1411 		return (0);
1412 }
1413 
1414 int
1415 pci_msix_enabled(struct pci_devinst *pi)
1416 {
1417 
1418 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1419 }
1420 
1421 void
1422 pci_generate_msix(struct pci_devinst *pi, int index)
1423 {
1424 	struct msix_table_entry *mte;
1425 
1426 	if (!pci_msix_enabled(pi))
1427 		return;
1428 
1429 	if (pi->pi_msix.function_mask)
1430 		return;
1431 
1432 	if (index >= pi->pi_msix.table_count)
1433 		return;
1434 
1435 	mte = &pi->pi_msix.table[index];
1436 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1437 		/* XXX Set PBA bit if interrupt is disabled */
1438 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1439 	}
1440 }
1441 
1442 void
1443 pci_generate_msi(struct pci_devinst *pi, int index)
1444 {
1445 
1446 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1447 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1448 			     pi->pi_msi.msg_data + index);
1449 	}
1450 }
1451 
1452 static bool
1453 pci_lintr_permitted(struct pci_devinst *pi)
1454 {
1455 	uint16_t cmd;
1456 
1457 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1458 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1459 		(cmd & PCIM_CMD_INTxDIS)));
1460 }
1461 
1462 void
1463 pci_lintr_request(struct pci_devinst *pi)
1464 {
1465 	struct businfo *bi;
1466 	struct slotinfo *si;
1467 	int bestpin, bestcount, pin;
1468 
1469 	bi = pci_businfo[pi->pi_bus];
1470 	assert(bi != NULL);
1471 
1472 	/*
1473 	 * Just allocate a pin from our slot.  The pin will be
1474 	 * assigned IRQs later when interrupts are routed.
1475 	 */
1476 	si = &bi->slotinfo[pi->pi_slot];
1477 	bestpin = 0;
1478 	bestcount = si->si_intpins[0].ii_count;
1479 	for (pin = 1; pin < 4; pin++) {
1480 		if (si->si_intpins[pin].ii_count < bestcount) {
1481 			bestpin = pin;
1482 			bestcount = si->si_intpins[pin].ii_count;
1483 		}
1484 	}
1485 
1486 	si->si_intpins[bestpin].ii_count++;
1487 	pi->pi_lintr.pin = bestpin + 1;
1488 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1489 }
1490 
1491 static void
1492 pci_lintr_route(struct pci_devinst *pi)
1493 {
1494 	struct businfo *bi;
1495 	struct intxinfo *ii;
1496 
1497 	if (pi->pi_lintr.pin == 0)
1498 		return;
1499 
1500 	bi = pci_businfo[pi->pi_bus];
1501 	assert(bi != NULL);
1502 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1503 
1504 	/*
1505 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1506 	 * is not yet assigned.
1507 	 */
1508 	if (ii->ii_ioapic_irq == 0)
1509 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1510 	assert(ii->ii_ioapic_irq > 0);
1511 
1512 	/*
1513 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1514 	 * not yet assigned.
1515 	 */
1516 	if (ii->ii_pirq_pin == 0)
1517 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1518 	assert(ii->ii_pirq_pin > 0);
1519 
1520 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1521 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1522 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1523 }
1524 
1525 void
1526 pci_lintr_assert(struct pci_devinst *pi)
1527 {
1528 
1529 	assert(pi->pi_lintr.pin > 0);
1530 
1531 	pthread_mutex_lock(&pi->pi_lintr.lock);
1532 	if (pi->pi_lintr.state == IDLE) {
1533 		if (pci_lintr_permitted(pi)) {
1534 			pi->pi_lintr.state = ASSERTED;
1535 			pci_irq_assert(pi);
1536 		} else
1537 			pi->pi_lintr.state = PENDING;
1538 	}
1539 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1540 }
1541 
1542 void
1543 pci_lintr_deassert(struct pci_devinst *pi)
1544 {
1545 
1546 	assert(pi->pi_lintr.pin > 0);
1547 
1548 	pthread_mutex_lock(&pi->pi_lintr.lock);
1549 	if (pi->pi_lintr.state == ASSERTED) {
1550 		pi->pi_lintr.state = IDLE;
1551 		pci_irq_deassert(pi);
1552 	} else if (pi->pi_lintr.state == PENDING)
1553 		pi->pi_lintr.state = IDLE;
1554 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1555 }
1556 
1557 static void
1558 pci_lintr_update(struct pci_devinst *pi)
1559 {
1560 
1561 	pthread_mutex_lock(&pi->pi_lintr.lock);
1562 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1563 		pci_irq_deassert(pi);
1564 		pi->pi_lintr.state = PENDING;
1565 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1566 		pi->pi_lintr.state = ASSERTED;
1567 		pci_irq_assert(pi);
1568 	}
1569 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1570 }
1571 
1572 int
1573 pci_count_lintr(int bus)
1574 {
1575 	int count, slot, pin;
1576 	struct slotinfo *slotinfo;
1577 
1578 	count = 0;
1579 	if (pci_businfo[bus] != NULL) {
1580 		for (slot = 0; slot < MAXSLOTS; slot++) {
1581 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1582 			for (pin = 0; pin < 4; pin++) {
1583 				if (slotinfo->si_intpins[pin].ii_count != 0)
1584 					count++;
1585 			}
1586 		}
1587 	}
1588 	return (count);
1589 }
1590 
1591 void
1592 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1593 {
1594 	struct businfo *bi;
1595 	struct slotinfo *si;
1596 	struct intxinfo *ii;
1597 	int slot, pin;
1598 
1599 	if ((bi = pci_businfo[bus]) == NULL)
1600 		return;
1601 
1602 	for (slot = 0; slot < MAXSLOTS; slot++) {
1603 		si = &bi->slotinfo[slot];
1604 		for (pin = 0; pin < 4; pin++) {
1605 			ii = &si->si_intpins[pin];
1606 			if (ii->ii_count != 0)
1607 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1608 				    ii->ii_ioapic_irq, arg);
1609 		}
1610 	}
1611 }
1612 
1613 /*
1614  * Return 1 if the emulated device in 'slot' is a multi-function device.
1615  * Return 0 otherwise.
1616  */
1617 static int
1618 pci_emul_is_mfdev(int bus, int slot)
1619 {
1620 	struct businfo *bi;
1621 	struct slotinfo *si;
1622 	int f, numfuncs;
1623 
1624 	numfuncs = 0;
1625 	if ((bi = pci_businfo[bus]) != NULL) {
1626 		si = &bi->slotinfo[slot];
1627 		for (f = 0; f < MAXFUNCS; f++) {
1628 			if (si->si_funcs[f].fi_devi != NULL) {
1629 				numfuncs++;
1630 			}
1631 		}
1632 	}
1633 	return (numfuncs > 1);
1634 }
1635 
1636 /*
1637  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1638  * whether or not is a multi-function being emulated in the pci 'slot'.
1639  */
1640 static void
1641 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1642 {
1643 	int mfdev;
1644 
1645 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1646 		mfdev = pci_emul_is_mfdev(bus, slot);
1647 		switch (bytes) {
1648 		case 1:
1649 		case 2:
1650 			*rv &= ~PCIM_MFDEV;
1651 			if (mfdev) {
1652 				*rv |= PCIM_MFDEV;
1653 			}
1654 			break;
1655 		case 4:
1656 			*rv &= ~(PCIM_MFDEV << 16);
1657 			if (mfdev) {
1658 				*rv |= (PCIM_MFDEV << 16);
1659 			}
1660 			break;
1661 		}
1662 	}
1663 }
1664 
1665 static void
1666 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1667 {
1668 	int i, rshift;
1669 	uint32_t cmd, cmd2, changed, old, readonly;
1670 
1671 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1672 
1673 	/*
1674 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1675 	 *
1676 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1677 	 * 'write 1 to clear'. However these bits are not set to '1' by
1678 	 * any device emulation so it is simpler to treat them as readonly.
1679 	 */
1680 	rshift = (coff & 0x3) * 8;
1681 	readonly = 0xFFFFF880 >> rshift;
1682 
1683 	old = CFGREAD(pi, coff, bytes);
1684 	new &= ~readonly;
1685 	new |= (old & readonly);
1686 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1687 
1688 	cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1689 	changed = cmd ^ cmd2;
1690 
1691 	/*
1692 	 * If the MMIO or I/O address space decoding has changed then
1693 	 * register/unregister all BARs that decode that address space.
1694 	 */
1695 	for (i = 0; i <= PCI_BARMAX; i++) {
1696 		switch (pi->pi_bar[i].type) {
1697 			case PCIBAR_NONE:
1698 			case PCIBAR_MEMHI64:
1699 				break;
1700 			case PCIBAR_IO:
1701 				/* I/O address space decoding changed? */
1702 				if (changed & PCIM_CMD_PORTEN) {
1703 					if (porten(pi))
1704 						register_bar(pi, i);
1705 					else
1706 						unregister_bar(pi, i);
1707 				}
1708 				break;
1709 			case PCIBAR_MEM32:
1710 			case PCIBAR_MEM64:
1711 				/* MMIO address space decoding changed? */
1712 				if (changed & PCIM_CMD_MEMEN) {
1713 					if (memen(pi))
1714 						register_bar(pi, i);
1715 					else
1716 						unregister_bar(pi, i);
1717 				}
1718 				break;
1719 			default:
1720 				assert(0);
1721 		}
1722 	}
1723 
1724 	/*
1725 	 * If INTx has been unmasked and is pending, assert the
1726 	 * interrupt.
1727 	 */
1728 	pci_lintr_update(pi);
1729 }
1730 
1731 static void
1732 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1733     int coff, int bytes, uint32_t *eax)
1734 {
1735 	struct businfo *bi;
1736 	struct slotinfo *si;
1737 	struct pci_devinst *pi;
1738 	struct pci_devemu *pe;
1739 	int idx, needcfg;
1740 	uint64_t addr, bar, mask;
1741 
1742 	if ((bi = pci_businfo[bus]) != NULL) {
1743 		si = &bi->slotinfo[slot];
1744 		pi = si->si_funcs[func].fi_devi;
1745 	} else
1746 		pi = NULL;
1747 
1748 	/*
1749 	 * Just return if there is no device at this slot:func or if the
1750 	 * the guest is doing an un-aligned access.
1751 	 */
1752 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1753 	    (coff & (bytes - 1)) != 0) {
1754 		if (in)
1755 			*eax = 0xffffffff;
1756 		return;
1757 	}
1758 
1759 	/*
1760 	 * Ignore all writes beyond the standard config space and return all
1761 	 * ones on reads.
1762 	 */
1763 	if (coff >= PCI_REGMAX + 1) {
1764 		if (in) {
1765 			*eax = 0xffffffff;
1766 			/*
1767 			 * Extended capabilities begin at offset 256 in config
1768 			 * space. Absence of extended capabilities is signaled
1769 			 * with all 0s in the extended capability header at
1770 			 * offset 256.
1771 			 */
1772 			if (coff <= PCI_REGMAX + 4)
1773 				*eax = 0x00000000;
1774 		}
1775 		return;
1776 	}
1777 
1778 	pe = pi->pi_d;
1779 
1780 	/*
1781 	 * Config read
1782 	 */
1783 	if (in) {
1784 		/* Let the device emulation override the default handler */
1785 		if (pe->pe_cfgread != NULL) {
1786 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1787 			    eax);
1788 		} else {
1789 			needcfg = 1;
1790 		}
1791 
1792 		if (needcfg)
1793 			*eax = CFGREAD(pi, coff, bytes);
1794 
1795 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1796 	} else {
1797 		/* Let the device emulation override the default handler */
1798 		if (pe->pe_cfgwrite != NULL &&
1799 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1800 			return;
1801 
1802 		/*
1803 		 * Special handling for write to BAR registers
1804 		 */
1805 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1806 			/*
1807 			 * Ignore writes to BAR registers that are not
1808 			 * 4-byte aligned.
1809 			 */
1810 			if (bytes != 4 || (coff & 0x3) != 0)
1811 				return;
1812 			idx = (coff - PCIR_BAR(0)) / 4;
1813 			mask = ~(pi->pi_bar[idx].size - 1);
1814 			switch (pi->pi_bar[idx].type) {
1815 			case PCIBAR_NONE:
1816 				pi->pi_bar[idx].addr = bar = 0;
1817 				break;
1818 			case PCIBAR_IO:
1819 				addr = *eax & mask;
1820 				addr &= 0xffff;
1821 				bar = addr | PCIM_BAR_IO_SPACE;
1822 				/*
1823 				 * Register the new BAR value for interception
1824 				 */
1825 				if (addr != pi->pi_bar[idx].addr) {
1826 					update_bar_address(pi, addr, idx,
1827 							   PCIBAR_IO);
1828 				}
1829 				break;
1830 			case PCIBAR_MEM32:
1831 				addr = bar = *eax & mask;
1832 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1833 				if (addr != pi->pi_bar[idx].addr) {
1834 					update_bar_address(pi, addr, idx,
1835 							   PCIBAR_MEM32);
1836 				}
1837 				break;
1838 			case PCIBAR_MEM64:
1839 				addr = bar = *eax & mask;
1840 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1841 				       PCIM_BAR_MEM_PREFETCH;
1842 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1843 					update_bar_address(pi, addr, idx,
1844 							   PCIBAR_MEM64);
1845 				}
1846 				break;
1847 			case PCIBAR_MEMHI64:
1848 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1849 				addr = ((uint64_t)*eax << 32) & mask;
1850 				bar = addr >> 32;
1851 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1852 					update_bar_address(pi, addr, idx - 1,
1853 							   PCIBAR_MEMHI64);
1854 				}
1855 				break;
1856 			default:
1857 				assert(0);
1858 			}
1859 			pci_set_cfgdata32(pi, coff, bar);
1860 
1861 		} else if (pci_emul_iscap(pi, coff)) {
1862 			pci_emul_capwrite(pi, coff, bytes, *eax);
1863 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1864 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1865 		} else {
1866 			CFGWRITE(pi, coff, *eax, bytes);
1867 		}
1868 	}
1869 }
1870 
1871 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1872 
1873 static int
1874 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1875 		 uint32_t *eax, void *arg)
1876 {
1877 	uint32_t x;
1878 
1879 	if (bytes != 4) {
1880 		if (in)
1881 			*eax = (bytes == 2) ? 0xffff : 0xff;
1882 		return (0);
1883 	}
1884 
1885 	if (in) {
1886 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1887 		if (cfgenable)
1888 			x |= CONF1_ENABLE;
1889 		*eax = x;
1890 	} else {
1891 		x = *eax;
1892 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1893 		cfgoff = x & PCI_REGMAX;
1894 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1895 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1896 		cfgbus = (x >> 16) & PCI_BUSMAX;
1897 	}
1898 
1899 	return (0);
1900 }
1901 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1902 
1903 static int
1904 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1905 		 uint32_t *eax, void *arg)
1906 {
1907 	int coff;
1908 
1909 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1910 
1911 	coff = cfgoff + (port - CONF1_DATA_PORT);
1912 	if (cfgenable) {
1913 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1914 		    eax);
1915 	} else {
1916 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1917 		if (in)
1918 			*eax = 0xffffffff;
1919 	}
1920 	return (0);
1921 }
1922 
1923 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1924 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1925 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1926 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1927 
1928 #define PCI_EMUL_TEST
1929 #ifdef PCI_EMUL_TEST
1930 /*
1931  * Define a dummy test device
1932  */
1933 #define DIOSZ	8
1934 #define DMEMSZ	4096
1935 struct pci_emul_dsoftc {
1936 	uint8_t   ioregs[DIOSZ];
1937 	uint8_t	  memregs[2][DMEMSZ];
1938 };
1939 
1940 #define	PCI_EMUL_MSI_MSGS	 4
1941 #define	PCI_EMUL_MSIX_MSGS	16
1942 
1943 static int
1944 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1945 {
1946 	int error;
1947 	struct pci_emul_dsoftc *sc;
1948 
1949 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1950 
1951 	pi->pi_arg = sc;
1952 
1953 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1954 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1955 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1956 
1957 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1958 	assert(error == 0);
1959 
1960 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1961 	assert(error == 0);
1962 
1963 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1964 	assert(error == 0);
1965 
1966 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1967 	assert(error == 0);
1968 
1969 	return (0);
1970 }
1971 
1972 static void
1973 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1974 	      uint64_t offset, int size, uint64_t value)
1975 {
1976 	int i;
1977 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1978 
1979 	if (baridx == 0) {
1980 		if (offset + size > DIOSZ) {
1981 			printf("diow: iow too large, offset %ld size %d\n",
1982 			       offset, size);
1983 			return;
1984 		}
1985 
1986 		if (size == 1) {
1987 			sc->ioregs[offset] = value & 0xff;
1988 		} else if (size == 2) {
1989 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1990 		} else if (size == 4) {
1991 			*(uint32_t *)&sc->ioregs[offset] = value;
1992 		} else {
1993 			printf("diow: iow unknown size %d\n", size);
1994 		}
1995 
1996 		/*
1997 		 * Special magic value to generate an interrupt
1998 		 */
1999 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2000 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2001 
2002 		if (value == 0xabcdef) {
2003 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2004 				pci_generate_msi(pi, i);
2005 		}
2006 	}
2007 
2008 	if (baridx == 1 || baridx == 2) {
2009 		if (offset + size > DMEMSZ) {
2010 			printf("diow: memw too large, offset %ld size %d\n",
2011 			       offset, size);
2012 			return;
2013 		}
2014 
2015 		i = baridx - 1;		/* 'memregs' index */
2016 
2017 		if (size == 1) {
2018 			sc->memregs[i][offset] = value;
2019 		} else if (size == 2) {
2020 			*(uint16_t *)&sc->memregs[i][offset] = value;
2021 		} else if (size == 4) {
2022 			*(uint32_t *)&sc->memregs[i][offset] = value;
2023 		} else if (size == 8) {
2024 			*(uint64_t *)&sc->memregs[i][offset] = value;
2025 		} else {
2026 			printf("diow: memw unknown size %d\n", size);
2027 		}
2028 
2029 		/*
2030 		 * magic interrupt ??
2031 		 */
2032 	}
2033 
2034 	if (baridx > 2 || baridx < 0) {
2035 		printf("diow: unknown bar idx %d\n", baridx);
2036 	}
2037 }
2038 
2039 static uint64_t
2040 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2041 	      uint64_t offset, int size)
2042 {
2043 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2044 	uint32_t value;
2045 	int i;
2046 
2047 	if (baridx == 0) {
2048 		if (offset + size > DIOSZ) {
2049 			printf("dior: ior too large, offset %ld size %d\n",
2050 			       offset, size);
2051 			return (0);
2052 		}
2053 
2054 		value = 0;
2055 		if (size == 1) {
2056 			value = sc->ioregs[offset];
2057 		} else if (size == 2) {
2058 			value = *(uint16_t *) &sc->ioregs[offset];
2059 		} else if (size == 4) {
2060 			value = *(uint32_t *) &sc->ioregs[offset];
2061 		} else {
2062 			printf("dior: ior unknown size %d\n", size);
2063 		}
2064 	}
2065 
2066 	if (baridx == 1 || baridx == 2) {
2067 		if (offset + size > DMEMSZ) {
2068 			printf("dior: memr too large, offset %ld size %d\n",
2069 			       offset, size);
2070 			return (0);
2071 		}
2072 
2073 		i = baridx - 1;		/* 'memregs' index */
2074 
2075 		if (size == 1) {
2076 			value = sc->memregs[i][offset];
2077 		} else if (size == 2) {
2078 			value = *(uint16_t *) &sc->memregs[i][offset];
2079 		} else if (size == 4) {
2080 			value = *(uint32_t *) &sc->memregs[i][offset];
2081 		} else if (size == 8) {
2082 			value = *(uint64_t *) &sc->memregs[i][offset];
2083 		} else {
2084 			printf("dior: ior unknown size %d\n", size);
2085 		}
2086 	}
2087 
2088 
2089 	if (baridx > 2 || baridx < 0) {
2090 		printf("dior: unknown bar idx %d\n", baridx);
2091 		return (0);
2092 	}
2093 
2094 	return (value);
2095 }
2096 
2097 struct pci_devemu pci_dummy = {
2098 	.pe_emu = "dummy",
2099 	.pe_init = pci_emul_dinit,
2100 	.pe_barwrite = pci_emul_diow,
2101 	.pe_barread = pci_emul_dior
2102 };
2103 PCI_EMUL_SET(pci_dummy);
2104 
2105 #endif /* PCI_EMUL_TEST */
2106