xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision 6683132d54bd6d589889e43dabdc53d35e38a028)
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 void
241 pci_print_supported_devices()
242 {
243 	struct pci_devemu **pdpp, *pdp;
244 
245 	SET_FOREACH(pdpp, pci_devemu_set) {
246 		pdp = *pdpp;
247 		printf("%s\n", pdp->pe_emu);
248 	}
249 }
250 
251 static int
252 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
253 {
254 
255 	if (offset < pi->pi_msix.pba_offset)
256 		return (0);
257 
258 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
259 		return (0);
260 	}
261 
262 	return (1);
263 }
264 
265 int
266 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
267 		     uint64_t value)
268 {
269 	int msix_entry_offset;
270 	int tab_index;
271 	char *dest;
272 
273 	/* support only 4 or 8 byte writes */
274 	if (size != 4 && size != 8)
275 		return (-1);
276 
277 	/*
278 	 * Return if table index is beyond what device supports
279 	 */
280 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
281 	if (tab_index >= pi->pi_msix.table_count)
282 		return (-1);
283 
284 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
285 
286 	/* support only aligned writes */
287 	if ((msix_entry_offset % size) != 0)
288 		return (-1);
289 
290 	dest = (char *)(pi->pi_msix.table + tab_index);
291 	dest += msix_entry_offset;
292 
293 	if (size == 4)
294 		*((uint32_t *)dest) = value;
295 	else
296 		*((uint64_t *)dest) = value;
297 
298 	return (0);
299 }
300 
301 uint64_t
302 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
303 {
304 	char *dest;
305 	int msix_entry_offset;
306 	int tab_index;
307 	uint64_t retval = ~0;
308 
309 	/*
310 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
311 	 * table but we also allow 1 byte access to accommodate reads from
312 	 * ddb.
313 	 */
314 	if (size != 1 && size != 4 && size != 8)
315 		return (retval);
316 
317 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
318 
319 	/* support only aligned reads */
320 	if ((msix_entry_offset % size) != 0) {
321 		return (retval);
322 	}
323 
324 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
325 
326 	if (tab_index < pi->pi_msix.table_count) {
327 		/* valid MSI-X Table access */
328 		dest = (char *)(pi->pi_msix.table + tab_index);
329 		dest += msix_entry_offset;
330 
331 		if (size == 1)
332 			retval = *((uint8_t *)dest);
333 		else if (size == 4)
334 			retval = *((uint32_t *)dest);
335 		else
336 			retval = *((uint64_t *)dest);
337 	} else if (pci_valid_pba_offset(pi, offset)) {
338 		/* return 0 for PBA access */
339 		retval = 0;
340 	}
341 
342 	return (retval);
343 }
344 
345 int
346 pci_msix_table_bar(struct pci_devinst *pi)
347 {
348 
349 	if (pi->pi_msix.table != NULL)
350 		return (pi->pi_msix.table_bar);
351 	else
352 		return (-1);
353 }
354 
355 int
356 pci_msix_pba_bar(struct pci_devinst *pi)
357 {
358 
359 	if (pi->pi_msix.table != NULL)
360 		return (pi->pi_msix.pba_bar);
361 	else
362 		return (-1);
363 }
364 
365 static int
366 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
367 		    uint32_t *eax, void *arg)
368 {
369 	struct pci_devinst *pdi = arg;
370 	struct pci_devemu *pe = pdi->pi_d;
371 	uint64_t offset;
372 	int i;
373 
374 	for (i = 0; i <= PCI_BARMAX; i++) {
375 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
376 		    port >= pdi->pi_bar[i].addr &&
377 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
378 			offset = port - pdi->pi_bar[i].addr;
379 			if (in)
380 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
381 							 offset, bytes);
382 			else
383 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
384 						   bytes, *eax);
385 			return (0);
386 		}
387 	}
388 	return (-1);
389 }
390 
391 static int
392 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
393 		     int size, uint64_t *val, void *arg1, long arg2)
394 {
395 	struct pci_devinst *pdi = arg1;
396 	struct pci_devemu *pe = pdi->pi_d;
397 	uint64_t offset;
398 	int bidx = (int) arg2;
399 
400 	assert(bidx <= PCI_BARMAX);
401 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
402 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
403 	assert(addr >= pdi->pi_bar[bidx].addr &&
404 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
405 
406 	offset = addr - pdi->pi_bar[bidx].addr;
407 
408 	if (dir == MEM_F_WRITE) {
409 		if (size == 8) {
410 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
411 					   4, *val & 0xffffffff);
412 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
413 					   4, *val >> 32);
414 		} else {
415 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
416 					   size, *val);
417 		}
418 	} else {
419 		if (size == 8) {
420 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
421 						 offset, 4);
422 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
423 						  offset + 4, 4) << 32;
424 		} else {
425 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
426 						 offset, size);
427 		}
428 	}
429 
430 	return (0);
431 }
432 
433 
434 static int
435 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
436 			uint64_t *addr)
437 {
438 	uint64_t base;
439 
440 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
441 
442 	base = roundup2(*baseptr, size);
443 
444 	if (base + size <= limit) {
445 		*addr = base;
446 		*baseptr = base + size;
447 		return (0);
448 	} else
449 		return (-1);
450 }
451 
452 int
453 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
454 		   uint64_t size)
455 {
456 
457 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
458 }
459 
460 /*
461  * Register (or unregister) the MMIO or I/O region associated with the BAR
462  * register 'idx' of an emulated pci device.
463  */
464 static void
465 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
466 {
467 	int error;
468 	struct inout_port iop;
469 	struct mem_range mr;
470 
471 	switch (pi->pi_bar[idx].type) {
472 	case PCIBAR_IO:
473 		bzero(&iop, sizeof(struct inout_port));
474 		iop.name = pi->pi_name;
475 		iop.port = pi->pi_bar[idx].addr;
476 		iop.size = pi->pi_bar[idx].size;
477 		if (registration) {
478 			iop.flags = IOPORT_F_INOUT;
479 			iop.handler = pci_emul_io_handler;
480 			iop.arg = pi;
481 			error = register_inout(&iop);
482 		} else
483 			error = unregister_inout(&iop);
484 		break;
485 	case PCIBAR_MEM32:
486 	case PCIBAR_MEM64:
487 		bzero(&mr, sizeof(struct mem_range));
488 		mr.name = pi->pi_name;
489 		mr.base = pi->pi_bar[idx].addr;
490 		mr.size = pi->pi_bar[idx].size;
491 		if (registration) {
492 			mr.flags = MEM_F_RW;
493 			mr.handler = pci_emul_mem_handler;
494 			mr.arg1 = pi;
495 			mr.arg2 = idx;
496 			error = register_mem(&mr);
497 		} else
498 			error = unregister_mem(&mr);
499 		break;
500 	default:
501 		error = EINVAL;
502 		break;
503 	}
504 	assert(error == 0);
505 }
506 
507 static void
508 unregister_bar(struct pci_devinst *pi, int idx)
509 {
510 
511 	modify_bar_registration(pi, idx, 0);
512 }
513 
514 static void
515 register_bar(struct pci_devinst *pi, int idx)
516 {
517 
518 	modify_bar_registration(pi, idx, 1);
519 }
520 
521 /* Are we decoding i/o port accesses for the emulated pci device? */
522 static int
523 porten(struct pci_devinst *pi)
524 {
525 	uint16_t cmd;
526 
527 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
528 
529 	return (cmd & PCIM_CMD_PORTEN);
530 }
531 
532 /* Are we decoding memory accesses for the emulated pci device? */
533 static int
534 memen(struct pci_devinst *pi)
535 {
536 	uint16_t cmd;
537 
538 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
539 
540 	return (cmd & PCIM_CMD_MEMEN);
541 }
542 
543 /*
544  * Update the MMIO or I/O address that is decoded by the BAR register.
545  *
546  * If the pci device has enabled the address space decoding then intercept
547  * the address range decoded by the BAR register.
548  */
549 static void
550 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
551 {
552 	int decode;
553 
554 	if (pi->pi_bar[idx].type == PCIBAR_IO)
555 		decode = porten(pi);
556 	else
557 		decode = memen(pi);
558 
559 	if (decode)
560 		unregister_bar(pi, idx);
561 
562 	switch (type) {
563 	case PCIBAR_IO:
564 	case PCIBAR_MEM32:
565 		pi->pi_bar[idx].addr = addr;
566 		break;
567 	case PCIBAR_MEM64:
568 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
569 		pi->pi_bar[idx].addr |= addr;
570 		break;
571 	case PCIBAR_MEMHI64:
572 		pi->pi_bar[idx].addr &= 0xffffffff;
573 		pi->pi_bar[idx].addr |= addr;
574 		break;
575 	default:
576 		assert(0);
577 	}
578 
579 	if (decode)
580 		register_bar(pi, idx);
581 }
582 
583 int
584 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
585 		    enum pcibar_type type, uint64_t size)
586 {
587 	int error;
588 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
589 	uint16_t cmd, enbit;
590 
591 	assert(idx >= 0 && idx <= PCI_BARMAX);
592 
593 	if ((size & (size - 1)) != 0)
594 		size = 1UL << flsl(size);	/* round up to a power of 2 */
595 
596 	/* Enforce minimum BAR sizes required by the PCI standard */
597 	if (type == PCIBAR_IO) {
598 		if (size < 4)
599 			size = 4;
600 	} else {
601 		if (size < 16)
602 			size = 16;
603 	}
604 
605 	switch (type) {
606 	case PCIBAR_NONE:
607 		baseptr = NULL;
608 		addr = mask = lobits = enbit = 0;
609 		break;
610 	case PCIBAR_IO:
611 		baseptr = &pci_emul_iobase;
612 		limit = PCI_EMUL_IOLIMIT;
613 		mask = PCIM_BAR_IO_BASE;
614 		lobits = PCIM_BAR_IO_SPACE;
615 		enbit = PCIM_CMD_PORTEN;
616 		break;
617 	case PCIBAR_MEM64:
618 		/*
619 		 * XXX
620 		 * Some drivers do not work well if the 64-bit BAR is allocated
621 		 * above 4GB. Allow for this by allocating small requests under
622 		 * 4GB unless then allocation size is larger than some arbitrary
623 		 * number (32MB currently).
624 		 */
625 		if (size > 32 * 1024 * 1024) {
626 			/*
627 			 * XXX special case for device requiring peer-peer DMA
628 			 */
629 			if (size == 0x100000000UL)
630 				baseptr = &hostbase;
631 			else
632 				baseptr = &pci_emul_membase64;
633 			limit = PCI_EMUL_MEMLIMIT64;
634 			mask = PCIM_BAR_MEM_BASE;
635 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
636 				 PCIM_BAR_MEM_PREFETCH;
637 		} else {
638 			baseptr = &pci_emul_membase32;
639 			limit = PCI_EMUL_MEMLIMIT32;
640 			mask = PCIM_BAR_MEM_BASE;
641 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
642 		}
643 		enbit = PCIM_CMD_MEMEN;
644 		break;
645 	case PCIBAR_MEM32:
646 		baseptr = &pci_emul_membase32;
647 		limit = PCI_EMUL_MEMLIMIT32;
648 		mask = PCIM_BAR_MEM_BASE;
649 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
650 		enbit = PCIM_CMD_MEMEN;
651 		break;
652 	default:
653 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
654 		assert(0);
655 	}
656 
657 	if (baseptr != NULL) {
658 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
659 		if (error != 0)
660 			return (error);
661 	}
662 
663 	pdi->pi_bar[idx].type = type;
664 	pdi->pi_bar[idx].addr = addr;
665 	pdi->pi_bar[idx].size = size;
666 
667 	/* Initialize the BAR register in config space */
668 	bar = (addr & mask) | lobits;
669 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
670 
671 	if (type == PCIBAR_MEM64) {
672 		assert(idx + 1 <= PCI_BARMAX);
673 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
674 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
675 	}
676 
677 	cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
678 	if ((cmd & enbit) != enbit)
679 		pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
680 	register_bar(pdi, idx);
681 
682 	return (0);
683 }
684 
685 #define	CAP_START_OFFSET	0x40
686 static int
687 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
688 {
689 	int i, capoff, reallen;
690 	uint16_t sts;
691 
692 	assert(caplen > 0);
693 
694 	reallen = roundup2(caplen, 4);		/* dword aligned */
695 
696 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
697 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
698 		capoff = CAP_START_OFFSET;
699 	else
700 		capoff = pi->pi_capend + 1;
701 
702 	/* Check if we have enough space */
703 	if (capoff + reallen > PCI_REGMAX + 1)
704 		return (-1);
705 
706 	/* Set the previous capability pointer */
707 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
708 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
709 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
710 	} else
711 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
712 
713 	/* Copy the capability */
714 	for (i = 0; i < caplen; i++)
715 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
716 
717 	/* Set the next capability pointer */
718 	pci_set_cfgdata8(pi, capoff + 1, 0);
719 
720 	pi->pi_prevcap = capoff;
721 	pi->pi_capend = capoff + reallen - 1;
722 	return (0);
723 }
724 
725 static struct pci_devemu *
726 pci_emul_finddev(char *name)
727 {
728 	struct pci_devemu **pdpp, *pdp;
729 
730 	SET_FOREACH(pdpp, pci_devemu_set) {
731 		pdp = *pdpp;
732 		if (!strcmp(pdp->pe_emu, name)) {
733 			return (pdp);
734 		}
735 	}
736 
737 	return (NULL);
738 }
739 
740 static int
741 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
742     int func, struct funcinfo *fi)
743 {
744 	struct pci_devinst *pdi;
745 	int err;
746 
747 	pdi = calloc(1, sizeof(struct pci_devinst));
748 
749 	pdi->pi_vmctx = ctx;
750 	pdi->pi_bus = bus;
751 	pdi->pi_slot = slot;
752 	pdi->pi_func = func;
753 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
754 	pdi->pi_lintr.pin = 0;
755 	pdi->pi_lintr.state = IDLE;
756 	pdi->pi_lintr.pirq_pin = 0;
757 	pdi->pi_lintr.ioapic_irq = 0;
758 	pdi->pi_d = pde;
759 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
760 
761 	/* Disable legacy interrupts */
762 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
763 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
764 
765 	pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
766 
767 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
768 	if (err == 0)
769 		fi->fi_devi = pdi;
770 	else
771 		free(pdi);
772 
773 	return (err);
774 }
775 
776 void
777 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
778 {
779 	int mmc;
780 
781 	/* Number of msi messages must be a power of 2 between 1 and 32 */
782 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
783 	mmc = ffs(msgnum) - 1;
784 
785 	bzero(msicap, sizeof(struct msicap));
786 	msicap->capid = PCIY_MSI;
787 	msicap->nextptr = nextptr;
788 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
789 }
790 
791 int
792 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
793 {
794 	struct msicap msicap;
795 
796 	pci_populate_msicap(&msicap, msgnum, 0);
797 
798 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
799 }
800 
801 static void
802 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
803 		     uint32_t msix_tab_size)
804 {
805 
806 	assert(msix_tab_size % 4096 == 0);
807 
808 	bzero(msixcap, sizeof(struct msixcap));
809 	msixcap->capid = PCIY_MSIX;
810 
811 	/*
812 	 * Message Control Register, all fields set to
813 	 * zero except for the Table Size.
814 	 * Note: Table size N is encoded as N-1
815 	 */
816 	msixcap->msgctrl = msgnum - 1;
817 
818 	/*
819 	 * MSI-X BAR setup:
820 	 * - MSI-X table start at offset 0
821 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
822 	 */
823 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
824 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
825 }
826 
827 static void
828 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
829 {
830 	int i, table_size;
831 
832 	assert(table_entries > 0);
833 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
834 
835 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
836 	pi->pi_msix.table = calloc(1, table_size);
837 
838 	/* set mask bit of vector control register */
839 	for (i = 0; i < table_entries; i++)
840 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
841 }
842 
843 int
844 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
845 {
846 	uint32_t tab_size;
847 	struct msixcap msixcap;
848 
849 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
850 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
851 
852 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
853 
854 	/* Align table size to nearest 4K */
855 	tab_size = roundup2(tab_size, 4096);
856 
857 	pi->pi_msix.table_bar = barnum;
858 	pi->pi_msix.pba_bar   = barnum;
859 	pi->pi_msix.table_offset = 0;
860 	pi->pi_msix.table_count = msgnum;
861 	pi->pi_msix.pba_offset = tab_size;
862 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
863 
864 	pci_msix_table_init(pi, msgnum);
865 
866 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
867 
868 	/* allocate memory for MSI-X Table and PBA */
869 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
870 				tab_size + pi->pi_msix.pba_size);
871 
872 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
873 					sizeof(msixcap)));
874 }
875 
876 void
877 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
878 		 int bytes, uint32_t val)
879 {
880 	uint16_t msgctrl, rwmask;
881 	int off;
882 
883 	off = offset - capoff;
884 	/* Message Control Register */
885 	if (off == 2 && bytes == 2) {
886 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
887 		msgctrl = pci_get_cfgdata16(pi, offset);
888 		msgctrl &= ~rwmask;
889 		msgctrl |= val & rwmask;
890 		val = msgctrl;
891 
892 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
893 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
894 		pci_lintr_update(pi);
895 	}
896 
897 	CFGWRITE(pi, offset, val, bytes);
898 }
899 
900 void
901 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
902 		int bytes, uint32_t val)
903 {
904 	uint16_t msgctrl, rwmask, msgdata, mme;
905 	uint32_t addrlo;
906 
907 	/*
908 	 * If guest is writing to the message control register make sure
909 	 * we do not overwrite read-only fields.
910 	 */
911 	if ((offset - capoff) == 2 && bytes == 2) {
912 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
913 		msgctrl = pci_get_cfgdata16(pi, offset);
914 		msgctrl &= ~rwmask;
915 		msgctrl |= val & rwmask;
916 		val = msgctrl;
917 
918 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
919 		if (msgctrl & PCIM_MSICTRL_64BIT)
920 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
921 		else
922 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
923 
924 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
925 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
926 		if (pi->pi_msi.enabled) {
927 			pi->pi_msi.addr = addrlo;
928 			pi->pi_msi.msg_data = msgdata;
929 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
930 		} else {
931 			pi->pi_msi.maxmsgnum = 0;
932 		}
933 		pci_lintr_update(pi);
934 	}
935 
936 	CFGWRITE(pi, offset, val, bytes);
937 }
938 
939 void
940 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
941 		 int bytes, uint32_t val)
942 {
943 
944 	/* XXX don't write to the readonly parts */
945 	CFGWRITE(pi, offset, val, bytes);
946 }
947 
948 #define	PCIECAP_VERSION	0x2
949 int
950 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
951 {
952 	int err;
953 	struct pciecap pciecap;
954 
955 	bzero(&pciecap, sizeof(pciecap));
956 
957 	/*
958 	 * Use the integrated endpoint type for endpoints on a root complex bus.
959 	 *
960 	 * NB: bhyve currently only supports a single PCI bus that is the root
961 	 * complex bus, so all endpoints are integrated.
962 	 */
963 	if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
964 		type = PCIEM_TYPE_ROOT_INT_EP;
965 
966 	pciecap.capid = PCIY_EXPRESS;
967 	pciecap.pcie_capabilities = PCIECAP_VERSION | type;
968 	if (type != PCIEM_TYPE_ROOT_INT_EP) {
969 		pciecap.link_capabilities = 0x411;	/* gen1, x1 */
970 		pciecap.link_status = 0x11;		/* gen1, x1 */
971 	}
972 
973 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
974 	return (err);
975 }
976 
977 /*
978  * This function assumes that 'coff' is in the capabilities region of the
979  * config space.
980  */
981 static void
982 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
983 {
984 	int capid;
985 	uint8_t capoff, nextoff;
986 
987 	/* Do not allow un-aligned writes */
988 	if ((offset & (bytes - 1)) != 0)
989 		return;
990 
991 	/* Find the capability that we want to update */
992 	capoff = CAP_START_OFFSET;
993 	while (1) {
994 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
995 		if (nextoff == 0)
996 			break;
997 		if (offset >= capoff && offset < nextoff)
998 			break;
999 
1000 		capoff = nextoff;
1001 	}
1002 	assert(offset >= capoff);
1003 
1004 	/*
1005 	 * Capability ID and Next Capability Pointer are readonly.
1006 	 * However, some o/s's do 4-byte writes that include these.
1007 	 * For this case, trim the write back to 2 bytes and adjust
1008 	 * the data.
1009 	 */
1010 	if (offset == capoff || offset == capoff + 1) {
1011 		if (offset == capoff && bytes == 4) {
1012 			bytes = 2;
1013 			offset += 2;
1014 			val >>= 16;
1015 		} else
1016 			return;
1017 	}
1018 
1019 	capid = pci_get_cfgdata8(pi, capoff);
1020 	switch (capid) {
1021 	case PCIY_MSI:
1022 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1023 		break;
1024 	case PCIY_MSIX:
1025 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1026 		break;
1027 	case PCIY_EXPRESS:
1028 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1029 		break;
1030 	default:
1031 		break;
1032 	}
1033 }
1034 
1035 static int
1036 pci_emul_iscap(struct pci_devinst *pi, int offset)
1037 {
1038 	uint16_t sts;
1039 
1040 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1041 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1042 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1043 			return (1);
1044 	}
1045 	return (0);
1046 }
1047 
1048 static int
1049 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1050 			  int size, uint64_t *val, void *arg1, long arg2)
1051 {
1052 	/*
1053 	 * Ignore writes; return 0xff's for reads. The mem read code
1054 	 * will take care of truncating to the correct size.
1055 	 */
1056 	if (dir == MEM_F_READ) {
1057 		*val = 0xffffffffffffffff;
1058 	}
1059 
1060 	return (0);
1061 }
1062 
1063 static int
1064 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1065     int bytes, uint64_t *val, void *arg1, long arg2)
1066 {
1067 	int bus, slot, func, coff, in;
1068 
1069 	coff = addr & 0xfff;
1070 	func = (addr >> 12) & 0x7;
1071 	slot = (addr >> 15) & 0x1f;
1072 	bus = (addr >> 20) & 0xff;
1073 	in = (dir == MEM_F_READ);
1074 	if (in)
1075 		*val = ~0UL;
1076 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1077 	return (0);
1078 }
1079 
1080 uint64_t
1081 pci_ecfg_base(void)
1082 {
1083 
1084 	return (PCI_EMUL_ECFG_BASE);
1085 }
1086 
1087 #define	BUSIO_ROUNDUP		32
1088 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1089 
1090 int
1091 init_pci(struct vmctx *ctx)
1092 {
1093 	struct mem_range mr;
1094 	struct pci_devemu *pde;
1095 	struct businfo *bi;
1096 	struct slotinfo *si;
1097 	struct funcinfo *fi;
1098 	size_t lowmem;
1099 	int bus, slot, func;
1100 	int error;
1101 
1102 	pci_emul_iobase = PCI_EMUL_IOBASE;
1103 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1104 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1105 
1106 	for (bus = 0; bus < MAXBUSES; bus++) {
1107 		if ((bi = pci_businfo[bus]) == NULL)
1108 			continue;
1109 		/*
1110 		 * Keep track of the i/o and memory resources allocated to
1111 		 * this bus.
1112 		 */
1113 		bi->iobase = pci_emul_iobase;
1114 		bi->membase32 = pci_emul_membase32;
1115 		bi->membase64 = pci_emul_membase64;
1116 
1117 		for (slot = 0; slot < MAXSLOTS; slot++) {
1118 			si = &bi->slotinfo[slot];
1119 			for (func = 0; func < MAXFUNCS; func++) {
1120 				fi = &si->si_funcs[func];
1121 				if (fi->fi_name == NULL)
1122 					continue;
1123 				pde = pci_emul_finddev(fi->fi_name);
1124 				assert(pde != NULL);
1125 				error = pci_emul_init(ctx, pde, bus, slot,
1126 				    func, fi);
1127 				if (error)
1128 					return (error);
1129 			}
1130 		}
1131 
1132 		/*
1133 		 * Add some slop to the I/O and memory resources decoded by
1134 		 * this bus to give a guest some flexibility if it wants to
1135 		 * reprogram the BARs.
1136 		 */
1137 		pci_emul_iobase += BUSIO_ROUNDUP;
1138 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1139 		bi->iolimit = pci_emul_iobase;
1140 
1141 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1142 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1143 		    BUSMEM_ROUNDUP);
1144 		bi->memlimit32 = pci_emul_membase32;
1145 
1146 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1147 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1148 		    BUSMEM_ROUNDUP);
1149 		bi->memlimit64 = pci_emul_membase64;
1150 	}
1151 
1152 	/*
1153 	 * PCI backends are initialized before routing INTx interrupts
1154 	 * so that LPC devices are able to reserve ISA IRQs before
1155 	 * routing PIRQ pins.
1156 	 */
1157 	for (bus = 0; bus < MAXBUSES; bus++) {
1158 		if ((bi = pci_businfo[bus]) == NULL)
1159 			continue;
1160 
1161 		for (slot = 0; slot < MAXSLOTS; slot++) {
1162 			si = &bi->slotinfo[slot];
1163 			for (func = 0; func < MAXFUNCS; func++) {
1164 				fi = &si->si_funcs[func];
1165 				if (fi->fi_devi == NULL)
1166 					continue;
1167 				pci_lintr_route(fi->fi_devi);
1168 			}
1169 		}
1170 	}
1171 	lpc_pirq_routed();
1172 
1173 	/*
1174 	 * The guest physical memory map looks like the following:
1175 	 * [0,		    lowmem)		guest system memory
1176 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1177 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1178 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1179 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1180 	 * [4GB,	    4GB + highmem)
1181 	 */
1182 
1183 	/*
1184 	 * Accesses to memory addresses that are not allocated to system
1185 	 * memory or PCI devices return 0xff's.
1186 	 */
1187 	lowmem = vm_get_lowmem_size(ctx);
1188 	bzero(&mr, sizeof(struct mem_range));
1189 	mr.name = "PCI hole";
1190 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1191 	mr.base = lowmem;
1192 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1193 	mr.handler = pci_emul_fallback_handler;
1194 	error = register_mem_fallback(&mr);
1195 	assert(error == 0);
1196 
1197 	/* PCI extended config space */
1198 	bzero(&mr, sizeof(struct mem_range));
1199 	mr.name = "PCI ECFG";
1200 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1201 	mr.base = PCI_EMUL_ECFG_BASE;
1202 	mr.size = PCI_EMUL_ECFG_SIZE;
1203 	mr.handler = pci_emul_ecfg_handler;
1204 	error = register_mem(&mr);
1205 	assert(error == 0);
1206 
1207 	return (0);
1208 }
1209 
1210 static void
1211 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1212     void *arg)
1213 {
1214 
1215 	dsdt_line("  Package ()");
1216 	dsdt_line("  {");
1217 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1218 	dsdt_line("    0x%02X,", pin - 1);
1219 	dsdt_line("    Zero,");
1220 	dsdt_line("    0x%X", ioapic_irq);
1221 	dsdt_line("  },");
1222 }
1223 
1224 static void
1225 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1226     void *arg)
1227 {
1228 	char *name;
1229 
1230 	name = lpc_pirq_name(pirq_pin);
1231 	if (name == NULL)
1232 		return;
1233 	dsdt_line("  Package ()");
1234 	dsdt_line("  {");
1235 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1236 	dsdt_line("    0x%02X,", pin - 1);
1237 	dsdt_line("    %s,", name);
1238 	dsdt_line("    0x00");
1239 	dsdt_line("  },");
1240 	free(name);
1241 }
1242 
1243 /*
1244  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1245  * corresponding to each PCI bus.
1246  */
1247 static void
1248 pci_bus_write_dsdt(int bus)
1249 {
1250 	struct businfo *bi;
1251 	struct slotinfo *si;
1252 	struct pci_devinst *pi;
1253 	int count, func, slot;
1254 
1255 	/*
1256 	 * If there are no devices on this 'bus' then just return.
1257 	 */
1258 	if ((bi = pci_businfo[bus]) == NULL) {
1259 		/*
1260 		 * Bus 0 is special because it decodes the I/O ports used
1261 		 * for PCI config space access even if there are no devices
1262 		 * on it.
1263 		 */
1264 		if (bus != 0)
1265 			return;
1266 	}
1267 
1268 	dsdt_line("  Device (PC%02X)", bus);
1269 	dsdt_line("  {");
1270 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1271 	dsdt_line("    Name (_ADR, Zero)");
1272 
1273 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1274 	dsdt_line("    {");
1275 	dsdt_line("        Return (0x%08X)", bus);
1276 	dsdt_line("    }");
1277 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1278 	dsdt_line("    {");
1279 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1280 	    "MaxFixed, PosDecode,");
1281 	dsdt_line("        0x0000,             // Granularity");
1282 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1283 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1284 	dsdt_line("        0x0000,             // Translation Offset");
1285 	dsdt_line("        0x0001,             // Length");
1286 	dsdt_line("        ,, )");
1287 
1288 	if (bus == 0) {
1289 		dsdt_indent(3);
1290 		dsdt_fixed_ioport(0xCF8, 8);
1291 		dsdt_unindent(3);
1292 
1293 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1294 		    "PosDecode, EntireRange,");
1295 		dsdt_line("        0x0000,             // Granularity");
1296 		dsdt_line("        0x0000,             // Range Minimum");
1297 		dsdt_line("        0x0CF7,             // Range Maximum");
1298 		dsdt_line("        0x0000,             // Translation Offset");
1299 		dsdt_line("        0x0CF8,             // Length");
1300 		dsdt_line("        ,, , TypeStatic)");
1301 
1302 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1303 		    "PosDecode, EntireRange,");
1304 		dsdt_line("        0x0000,             // Granularity");
1305 		dsdt_line("        0x0D00,             // Range Minimum");
1306 		dsdt_line("        0x%04X,             // Range Maximum",
1307 		    PCI_EMUL_IOBASE - 1);
1308 		dsdt_line("        0x0000,             // Translation Offset");
1309 		dsdt_line("        0x%04X,             // Length",
1310 		    PCI_EMUL_IOBASE - 0x0D00);
1311 		dsdt_line("        ,, , TypeStatic)");
1312 
1313 		if (bi == NULL) {
1314 			dsdt_line("    })");
1315 			goto done;
1316 		}
1317 	}
1318 	assert(bi != NULL);
1319 
1320 	/* i/o window */
1321 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1322 	    "PosDecode, EntireRange,");
1323 	dsdt_line("        0x0000,             // Granularity");
1324 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1325 	dsdt_line("        0x%04X,             // Range Maximum",
1326 	    bi->iolimit - 1);
1327 	dsdt_line("        0x0000,             // Translation Offset");
1328 	dsdt_line("        0x%04X,             // Length",
1329 	    bi->iolimit - bi->iobase);
1330 	dsdt_line("        ,, , TypeStatic)");
1331 
1332 	/* mmio window (32-bit) */
1333 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1334 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1335 	dsdt_line("        0x00000000,         // Granularity");
1336 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1337 	dsdt_line("        0x%08X,         // Range Maximum\n",
1338 	    bi->memlimit32 - 1);
1339 	dsdt_line("        0x00000000,         // Translation Offset");
1340 	dsdt_line("        0x%08X,         // Length\n",
1341 	    bi->memlimit32 - bi->membase32);
1342 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1343 
1344 	/* mmio window (64-bit) */
1345 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1346 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1347 	dsdt_line("        0x0000000000000000, // Granularity");
1348 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1349 	dsdt_line("        0x%016lX, // Range Maximum\n",
1350 	    bi->memlimit64 - 1);
1351 	dsdt_line("        0x0000000000000000, // Translation Offset");
1352 	dsdt_line("        0x%016lX, // Length\n",
1353 	    bi->memlimit64 - bi->membase64);
1354 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1355 	dsdt_line("    })");
1356 
1357 	count = pci_count_lintr(bus);
1358 	if (count != 0) {
1359 		dsdt_indent(2);
1360 		dsdt_line("Name (PPRT, Package ()");
1361 		dsdt_line("{");
1362 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1363 		dsdt_line("})");
1364 		dsdt_line("Name (APRT, Package ()");
1365 		dsdt_line("{");
1366 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1367 		dsdt_line("})");
1368 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1369 		dsdt_line("{");
1370 		dsdt_line("  If (PICM)");
1371 		dsdt_line("  {");
1372 		dsdt_line("    Return (APRT)");
1373 		dsdt_line("  }");
1374 		dsdt_line("  Else");
1375 		dsdt_line("  {");
1376 		dsdt_line("    Return (PPRT)");
1377 		dsdt_line("  }");
1378 		dsdt_line("}");
1379 		dsdt_unindent(2);
1380 	}
1381 
1382 	dsdt_indent(2);
1383 	for (slot = 0; slot < MAXSLOTS; slot++) {
1384 		si = &bi->slotinfo[slot];
1385 		for (func = 0; func < MAXFUNCS; func++) {
1386 			pi = si->si_funcs[func].fi_devi;
1387 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1388 				pi->pi_d->pe_write_dsdt(pi);
1389 		}
1390 	}
1391 	dsdt_unindent(2);
1392 done:
1393 	dsdt_line("  }");
1394 }
1395 
1396 void
1397 pci_write_dsdt(void)
1398 {
1399 	int bus;
1400 
1401 	dsdt_indent(1);
1402 	dsdt_line("Name (PICM, 0x00)");
1403 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1404 	dsdt_line("{");
1405 	dsdt_line("  Store (Arg0, PICM)");
1406 	dsdt_line("}");
1407 	dsdt_line("");
1408 	dsdt_line("Scope (_SB)");
1409 	dsdt_line("{");
1410 	for (bus = 0; bus < MAXBUSES; bus++)
1411 		pci_bus_write_dsdt(bus);
1412 	dsdt_line("}");
1413 	dsdt_unindent(1);
1414 }
1415 
1416 int
1417 pci_bus_configured(int bus)
1418 {
1419 	assert(bus >= 0 && bus < MAXBUSES);
1420 	return (pci_businfo[bus] != NULL);
1421 }
1422 
1423 int
1424 pci_msi_enabled(struct pci_devinst *pi)
1425 {
1426 	return (pi->pi_msi.enabled);
1427 }
1428 
1429 int
1430 pci_msi_maxmsgnum(struct pci_devinst *pi)
1431 {
1432 	if (pi->pi_msi.enabled)
1433 		return (pi->pi_msi.maxmsgnum);
1434 	else
1435 		return (0);
1436 }
1437 
1438 int
1439 pci_msix_enabled(struct pci_devinst *pi)
1440 {
1441 
1442 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1443 }
1444 
1445 void
1446 pci_generate_msix(struct pci_devinst *pi, int index)
1447 {
1448 	struct msix_table_entry *mte;
1449 
1450 	if (!pci_msix_enabled(pi))
1451 		return;
1452 
1453 	if (pi->pi_msix.function_mask)
1454 		return;
1455 
1456 	if (index >= pi->pi_msix.table_count)
1457 		return;
1458 
1459 	mte = &pi->pi_msix.table[index];
1460 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1461 		/* XXX Set PBA bit if interrupt is disabled */
1462 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1463 	}
1464 }
1465 
1466 void
1467 pci_generate_msi(struct pci_devinst *pi, int index)
1468 {
1469 
1470 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1471 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1472 			     pi->pi_msi.msg_data + index);
1473 	}
1474 }
1475 
1476 static bool
1477 pci_lintr_permitted(struct pci_devinst *pi)
1478 {
1479 	uint16_t cmd;
1480 
1481 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1482 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1483 		(cmd & PCIM_CMD_INTxDIS)));
1484 }
1485 
1486 void
1487 pci_lintr_request(struct pci_devinst *pi)
1488 {
1489 	struct businfo *bi;
1490 	struct slotinfo *si;
1491 	int bestpin, bestcount, pin;
1492 
1493 	bi = pci_businfo[pi->pi_bus];
1494 	assert(bi != NULL);
1495 
1496 	/*
1497 	 * Just allocate a pin from our slot.  The pin will be
1498 	 * assigned IRQs later when interrupts are routed.
1499 	 */
1500 	si = &bi->slotinfo[pi->pi_slot];
1501 	bestpin = 0;
1502 	bestcount = si->si_intpins[0].ii_count;
1503 	for (pin = 1; pin < 4; pin++) {
1504 		if (si->si_intpins[pin].ii_count < bestcount) {
1505 			bestpin = pin;
1506 			bestcount = si->si_intpins[pin].ii_count;
1507 		}
1508 	}
1509 
1510 	si->si_intpins[bestpin].ii_count++;
1511 	pi->pi_lintr.pin = bestpin + 1;
1512 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1513 }
1514 
1515 static void
1516 pci_lintr_route(struct pci_devinst *pi)
1517 {
1518 	struct businfo *bi;
1519 	struct intxinfo *ii;
1520 
1521 	if (pi->pi_lintr.pin == 0)
1522 		return;
1523 
1524 	bi = pci_businfo[pi->pi_bus];
1525 	assert(bi != NULL);
1526 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1527 
1528 	/*
1529 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1530 	 * is not yet assigned.
1531 	 */
1532 	if (ii->ii_ioapic_irq == 0)
1533 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1534 	assert(ii->ii_ioapic_irq > 0);
1535 
1536 	/*
1537 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1538 	 * not yet assigned.
1539 	 */
1540 	if (ii->ii_pirq_pin == 0)
1541 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1542 	assert(ii->ii_pirq_pin > 0);
1543 
1544 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1545 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1546 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1547 }
1548 
1549 void
1550 pci_lintr_assert(struct pci_devinst *pi)
1551 {
1552 
1553 	assert(pi->pi_lintr.pin > 0);
1554 
1555 	pthread_mutex_lock(&pi->pi_lintr.lock);
1556 	if (pi->pi_lintr.state == IDLE) {
1557 		if (pci_lintr_permitted(pi)) {
1558 			pi->pi_lintr.state = ASSERTED;
1559 			pci_irq_assert(pi);
1560 		} else
1561 			pi->pi_lintr.state = PENDING;
1562 	}
1563 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1564 }
1565 
1566 void
1567 pci_lintr_deassert(struct pci_devinst *pi)
1568 {
1569 
1570 	assert(pi->pi_lintr.pin > 0);
1571 
1572 	pthread_mutex_lock(&pi->pi_lintr.lock);
1573 	if (pi->pi_lintr.state == ASSERTED) {
1574 		pi->pi_lintr.state = IDLE;
1575 		pci_irq_deassert(pi);
1576 	} else if (pi->pi_lintr.state == PENDING)
1577 		pi->pi_lintr.state = IDLE;
1578 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1579 }
1580 
1581 static void
1582 pci_lintr_update(struct pci_devinst *pi)
1583 {
1584 
1585 	pthread_mutex_lock(&pi->pi_lintr.lock);
1586 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1587 		pci_irq_deassert(pi);
1588 		pi->pi_lintr.state = PENDING;
1589 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1590 		pi->pi_lintr.state = ASSERTED;
1591 		pci_irq_assert(pi);
1592 	}
1593 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1594 }
1595 
1596 int
1597 pci_count_lintr(int bus)
1598 {
1599 	int count, slot, pin;
1600 	struct slotinfo *slotinfo;
1601 
1602 	count = 0;
1603 	if (pci_businfo[bus] != NULL) {
1604 		for (slot = 0; slot < MAXSLOTS; slot++) {
1605 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1606 			for (pin = 0; pin < 4; pin++) {
1607 				if (slotinfo->si_intpins[pin].ii_count != 0)
1608 					count++;
1609 			}
1610 		}
1611 	}
1612 	return (count);
1613 }
1614 
1615 void
1616 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1617 {
1618 	struct businfo *bi;
1619 	struct slotinfo *si;
1620 	struct intxinfo *ii;
1621 	int slot, pin;
1622 
1623 	if ((bi = pci_businfo[bus]) == NULL)
1624 		return;
1625 
1626 	for (slot = 0; slot < MAXSLOTS; slot++) {
1627 		si = &bi->slotinfo[slot];
1628 		for (pin = 0; pin < 4; pin++) {
1629 			ii = &si->si_intpins[pin];
1630 			if (ii->ii_count != 0)
1631 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1632 				    ii->ii_ioapic_irq, arg);
1633 		}
1634 	}
1635 }
1636 
1637 /*
1638  * Return 1 if the emulated device in 'slot' is a multi-function device.
1639  * Return 0 otherwise.
1640  */
1641 static int
1642 pci_emul_is_mfdev(int bus, int slot)
1643 {
1644 	struct businfo *bi;
1645 	struct slotinfo *si;
1646 	int f, numfuncs;
1647 
1648 	numfuncs = 0;
1649 	if ((bi = pci_businfo[bus]) != NULL) {
1650 		si = &bi->slotinfo[slot];
1651 		for (f = 0; f < MAXFUNCS; f++) {
1652 			if (si->si_funcs[f].fi_devi != NULL) {
1653 				numfuncs++;
1654 			}
1655 		}
1656 	}
1657 	return (numfuncs > 1);
1658 }
1659 
1660 /*
1661  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1662  * whether or not is a multi-function being emulated in the pci 'slot'.
1663  */
1664 static void
1665 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1666 {
1667 	int mfdev;
1668 
1669 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1670 		mfdev = pci_emul_is_mfdev(bus, slot);
1671 		switch (bytes) {
1672 		case 1:
1673 		case 2:
1674 			*rv &= ~PCIM_MFDEV;
1675 			if (mfdev) {
1676 				*rv |= PCIM_MFDEV;
1677 			}
1678 			break;
1679 		case 4:
1680 			*rv &= ~(PCIM_MFDEV << 16);
1681 			if (mfdev) {
1682 				*rv |= (PCIM_MFDEV << 16);
1683 			}
1684 			break;
1685 		}
1686 	}
1687 }
1688 
1689 /*
1690  * Update device state in response to changes to the PCI command
1691  * register.
1692  */
1693 void
1694 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
1695 {
1696 	int i;
1697 	uint16_t changed, new;
1698 
1699 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);
1700 	changed = old ^ new;
1701 
1702 	/*
1703 	 * If the MMIO or I/O address space decoding has changed then
1704 	 * register/unregister all BARs that decode that address space.
1705 	 */
1706 	for (i = 0; i <= PCI_BARMAX; i++) {
1707 		switch (pi->pi_bar[i].type) {
1708 			case PCIBAR_NONE:
1709 			case PCIBAR_MEMHI64:
1710 				break;
1711 			case PCIBAR_IO:
1712 				/* I/O address space decoding changed? */
1713 				if (changed & PCIM_CMD_PORTEN) {
1714 					if (new & PCIM_CMD_PORTEN)
1715 						register_bar(pi, i);
1716 					else
1717 						unregister_bar(pi, i);
1718 				}
1719 				break;
1720 			case PCIBAR_MEM32:
1721 			case PCIBAR_MEM64:
1722 				/* MMIO address space decoding changed? */
1723 				if (changed & PCIM_CMD_MEMEN) {
1724 					if (new & PCIM_CMD_MEMEN)
1725 						register_bar(pi, i);
1726 					else
1727 						unregister_bar(pi, i);
1728 				}
1729 				break;
1730 			default:
1731 				assert(0);
1732 		}
1733 	}
1734 
1735 	/*
1736 	 * If INTx has been unmasked and is pending, assert the
1737 	 * interrupt.
1738 	 */
1739 	pci_lintr_update(pi);
1740 }
1741 
1742 static void
1743 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1744 {
1745 	int rshift;
1746 	uint32_t cmd, old, readonly;
1747 
1748 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1749 
1750 	/*
1751 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1752 	 *
1753 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1754 	 * 'write 1 to clear'. However these bits are not set to '1' by
1755 	 * any device emulation so it is simpler to treat them as readonly.
1756 	 */
1757 	rshift = (coff & 0x3) * 8;
1758 	readonly = 0xFFFFF880 >> rshift;
1759 
1760 	old = CFGREAD(pi, coff, bytes);
1761 	new &= ~readonly;
1762 	new |= (old & readonly);
1763 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1764 
1765 	pci_emul_cmd_changed(pi, cmd);
1766 }
1767 
1768 static void
1769 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1770     int coff, int bytes, uint32_t *eax)
1771 {
1772 	struct businfo *bi;
1773 	struct slotinfo *si;
1774 	struct pci_devinst *pi;
1775 	struct pci_devemu *pe;
1776 	int idx, needcfg;
1777 	uint64_t addr, bar, mask;
1778 
1779 	if ((bi = pci_businfo[bus]) != NULL) {
1780 		si = &bi->slotinfo[slot];
1781 		pi = si->si_funcs[func].fi_devi;
1782 	} else
1783 		pi = NULL;
1784 
1785 	/*
1786 	 * Just return if there is no device at this slot:func or if the
1787 	 * the guest is doing an un-aligned access.
1788 	 */
1789 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1790 	    (coff & (bytes - 1)) != 0) {
1791 		if (in)
1792 			*eax = 0xffffffff;
1793 		return;
1794 	}
1795 
1796 	/*
1797 	 * Ignore all writes beyond the standard config space and return all
1798 	 * ones on reads.
1799 	 */
1800 	if (coff >= PCI_REGMAX + 1) {
1801 		if (in) {
1802 			*eax = 0xffffffff;
1803 			/*
1804 			 * Extended capabilities begin at offset 256 in config
1805 			 * space. Absence of extended capabilities is signaled
1806 			 * with all 0s in the extended capability header at
1807 			 * offset 256.
1808 			 */
1809 			if (coff <= PCI_REGMAX + 4)
1810 				*eax = 0x00000000;
1811 		}
1812 		return;
1813 	}
1814 
1815 	pe = pi->pi_d;
1816 
1817 	/*
1818 	 * Config read
1819 	 */
1820 	if (in) {
1821 		/* Let the device emulation override the default handler */
1822 		if (pe->pe_cfgread != NULL) {
1823 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1824 			    eax);
1825 		} else {
1826 			needcfg = 1;
1827 		}
1828 
1829 		if (needcfg)
1830 			*eax = CFGREAD(pi, coff, bytes);
1831 
1832 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1833 	} else {
1834 		/* Let the device emulation override the default handler */
1835 		if (pe->pe_cfgwrite != NULL &&
1836 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1837 			return;
1838 
1839 		/*
1840 		 * Special handling for write to BAR registers
1841 		 */
1842 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1843 			/*
1844 			 * Ignore writes to BAR registers that are not
1845 			 * 4-byte aligned.
1846 			 */
1847 			if (bytes != 4 || (coff & 0x3) != 0)
1848 				return;
1849 			idx = (coff - PCIR_BAR(0)) / 4;
1850 			mask = ~(pi->pi_bar[idx].size - 1);
1851 			switch (pi->pi_bar[idx].type) {
1852 			case PCIBAR_NONE:
1853 				pi->pi_bar[idx].addr = bar = 0;
1854 				break;
1855 			case PCIBAR_IO:
1856 				addr = *eax & mask;
1857 				addr &= 0xffff;
1858 				bar = addr | PCIM_BAR_IO_SPACE;
1859 				/*
1860 				 * Register the new BAR value for interception
1861 				 */
1862 				if (addr != pi->pi_bar[idx].addr) {
1863 					update_bar_address(pi, addr, idx,
1864 							   PCIBAR_IO);
1865 				}
1866 				break;
1867 			case PCIBAR_MEM32:
1868 				addr = bar = *eax & mask;
1869 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1870 				if (addr != pi->pi_bar[idx].addr) {
1871 					update_bar_address(pi, addr, idx,
1872 							   PCIBAR_MEM32);
1873 				}
1874 				break;
1875 			case PCIBAR_MEM64:
1876 				addr = bar = *eax & mask;
1877 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1878 				       PCIM_BAR_MEM_PREFETCH;
1879 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1880 					update_bar_address(pi, addr, idx,
1881 							   PCIBAR_MEM64);
1882 				}
1883 				break;
1884 			case PCIBAR_MEMHI64:
1885 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1886 				addr = ((uint64_t)*eax << 32) & mask;
1887 				bar = addr >> 32;
1888 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1889 					update_bar_address(pi, addr, idx - 1,
1890 							   PCIBAR_MEMHI64);
1891 				}
1892 				break;
1893 			default:
1894 				assert(0);
1895 			}
1896 			pci_set_cfgdata32(pi, coff, bar);
1897 
1898 		} else if (pci_emul_iscap(pi, coff)) {
1899 			pci_emul_capwrite(pi, coff, bytes, *eax);
1900 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1901 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1902 		} else {
1903 			CFGWRITE(pi, coff, *eax, bytes);
1904 		}
1905 	}
1906 }
1907 
1908 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1909 
1910 static int
1911 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1912 		 uint32_t *eax, void *arg)
1913 {
1914 	uint32_t x;
1915 
1916 	if (bytes != 4) {
1917 		if (in)
1918 			*eax = (bytes == 2) ? 0xffff : 0xff;
1919 		return (0);
1920 	}
1921 
1922 	if (in) {
1923 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1924 		if (cfgenable)
1925 			x |= CONF1_ENABLE;
1926 		*eax = x;
1927 	} else {
1928 		x = *eax;
1929 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1930 		cfgoff = x & PCI_REGMAX;
1931 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1932 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1933 		cfgbus = (x >> 16) & PCI_BUSMAX;
1934 	}
1935 
1936 	return (0);
1937 }
1938 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1939 
1940 static int
1941 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1942 		 uint32_t *eax, void *arg)
1943 {
1944 	int coff;
1945 
1946 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1947 
1948 	coff = cfgoff + (port - CONF1_DATA_PORT);
1949 	if (cfgenable) {
1950 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1951 		    eax);
1952 	} else {
1953 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1954 		if (in)
1955 			*eax = 0xffffffff;
1956 	}
1957 	return (0);
1958 }
1959 
1960 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1961 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1962 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1963 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1964 
1965 #define PCI_EMUL_TEST
1966 #ifdef PCI_EMUL_TEST
1967 /*
1968  * Define a dummy test device
1969  */
1970 #define DIOSZ	8
1971 #define DMEMSZ	4096
1972 struct pci_emul_dsoftc {
1973 	uint8_t	  ioregs[DIOSZ];
1974 	uint8_t	  memregs[2][DMEMSZ];
1975 };
1976 
1977 #define	PCI_EMUL_MSI_MSGS	 4
1978 #define	PCI_EMUL_MSIX_MSGS	16
1979 
1980 static int
1981 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1982 {
1983 	int error;
1984 	struct pci_emul_dsoftc *sc;
1985 
1986 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1987 
1988 	pi->pi_arg = sc;
1989 
1990 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1991 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1992 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1993 
1994 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1995 	assert(error == 0);
1996 
1997 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1998 	assert(error == 0);
1999 
2000 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2001 	assert(error == 0);
2002 
2003 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2004 	assert(error == 0);
2005 
2006 	return (0);
2007 }
2008 
2009 static void
2010 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2011 	      uint64_t offset, int size, uint64_t value)
2012 {
2013 	int i;
2014 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2015 
2016 	if (baridx == 0) {
2017 		if (offset + size > DIOSZ) {
2018 			printf("diow: iow too large, offset %ld size %d\n",
2019 			       offset, size);
2020 			return;
2021 		}
2022 
2023 		if (size == 1) {
2024 			sc->ioregs[offset] = value & 0xff;
2025 		} else if (size == 2) {
2026 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2027 		} else if (size == 4) {
2028 			*(uint32_t *)&sc->ioregs[offset] = value;
2029 		} else {
2030 			printf("diow: iow unknown size %d\n", size);
2031 		}
2032 
2033 		/*
2034 		 * Special magic value to generate an interrupt
2035 		 */
2036 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2037 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2038 
2039 		if (value == 0xabcdef) {
2040 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2041 				pci_generate_msi(pi, i);
2042 		}
2043 	}
2044 
2045 	if (baridx == 1 || baridx == 2) {
2046 		if (offset + size > DMEMSZ) {
2047 			printf("diow: memw too large, offset %ld size %d\n",
2048 			       offset, size);
2049 			return;
2050 		}
2051 
2052 		i = baridx - 1;		/* 'memregs' index */
2053 
2054 		if (size == 1) {
2055 			sc->memregs[i][offset] = value;
2056 		} else if (size == 2) {
2057 			*(uint16_t *)&sc->memregs[i][offset] = value;
2058 		} else if (size == 4) {
2059 			*(uint32_t *)&sc->memregs[i][offset] = value;
2060 		} else if (size == 8) {
2061 			*(uint64_t *)&sc->memregs[i][offset] = value;
2062 		} else {
2063 			printf("diow: memw unknown size %d\n", size);
2064 		}
2065 
2066 		/*
2067 		 * magic interrupt ??
2068 		 */
2069 	}
2070 
2071 	if (baridx > 2 || baridx < 0) {
2072 		printf("diow: unknown bar idx %d\n", baridx);
2073 	}
2074 }
2075 
2076 static uint64_t
2077 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2078 	      uint64_t offset, int size)
2079 {
2080 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2081 	uint32_t value;
2082 	int i;
2083 
2084 	if (baridx == 0) {
2085 		if (offset + size > DIOSZ) {
2086 			printf("dior: ior too large, offset %ld size %d\n",
2087 			       offset, size);
2088 			return (0);
2089 		}
2090 
2091 		value = 0;
2092 		if (size == 1) {
2093 			value = sc->ioregs[offset];
2094 		} else if (size == 2) {
2095 			value = *(uint16_t *) &sc->ioregs[offset];
2096 		} else if (size == 4) {
2097 			value = *(uint32_t *) &sc->ioregs[offset];
2098 		} else {
2099 			printf("dior: ior unknown size %d\n", size);
2100 		}
2101 	}
2102 
2103 	if (baridx == 1 || baridx == 2) {
2104 		if (offset + size > DMEMSZ) {
2105 			printf("dior: memr too large, offset %ld size %d\n",
2106 			       offset, size);
2107 			return (0);
2108 		}
2109 
2110 		i = baridx - 1;		/* 'memregs' index */
2111 
2112 		if (size == 1) {
2113 			value = sc->memregs[i][offset];
2114 		} else if (size == 2) {
2115 			value = *(uint16_t *) &sc->memregs[i][offset];
2116 		} else if (size == 4) {
2117 			value = *(uint32_t *) &sc->memregs[i][offset];
2118 		} else if (size == 8) {
2119 			value = *(uint64_t *) &sc->memregs[i][offset];
2120 		} else {
2121 			printf("dior: ior unknown size %d\n", size);
2122 		}
2123 	}
2124 
2125 
2126 	if (baridx > 2 || baridx < 0) {
2127 		printf("dior: unknown bar idx %d\n", baridx);
2128 		return (0);
2129 	}
2130 
2131 	return (value);
2132 }
2133 
2134 struct pci_devemu pci_dummy = {
2135 	.pe_emu = "dummy",
2136 	.pe_init = pci_emul_dinit,
2137 	.pe_barwrite = pci_emul_diow,
2138 	.pe_barread = pci_emul_dior
2139 };
2140 PCI_EMUL_SET(pci_dummy);
2141 
2142 #endif /* PCI_EMUL_TEST */
2143