xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision 190cef3d)
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 
590 	assert(idx >= 0 && idx <= PCI_BARMAX);
591 
592 	if ((size & (size - 1)) != 0)
593 		size = 1UL << flsl(size);	/* round up to a power of 2 */
594 
595 	/* Enforce minimum BAR sizes required by the PCI standard */
596 	if (type == PCIBAR_IO) {
597 		if (size < 4)
598 			size = 4;
599 	} else {
600 		if (size < 16)
601 			size = 16;
602 	}
603 
604 	switch (type) {
605 	case PCIBAR_NONE:
606 		baseptr = NULL;
607 		addr = mask = lobits = 0;
608 		break;
609 	case PCIBAR_IO:
610 		baseptr = &pci_emul_iobase;
611 		limit = PCI_EMUL_IOLIMIT;
612 		mask = PCIM_BAR_IO_BASE;
613 		lobits = PCIM_BAR_IO_SPACE;
614 		break;
615 	case PCIBAR_MEM64:
616 		/*
617 		 * XXX
618 		 * Some drivers do not work well if the 64-bit BAR is allocated
619 		 * above 4GB. Allow for this by allocating small requests under
620 		 * 4GB unless then allocation size is larger than some arbitrary
621 		 * number (32MB currently).
622 		 */
623 		if (size > 32 * 1024 * 1024) {
624 			/*
625 			 * XXX special case for device requiring peer-peer DMA
626 			 */
627 			if (size == 0x100000000UL)
628 				baseptr = &hostbase;
629 			else
630 				baseptr = &pci_emul_membase64;
631 			limit = PCI_EMUL_MEMLIMIT64;
632 			mask = PCIM_BAR_MEM_BASE;
633 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
634 				 PCIM_BAR_MEM_PREFETCH;
635 			break;
636 		} else {
637 			baseptr = &pci_emul_membase32;
638 			limit = PCI_EMUL_MEMLIMIT32;
639 			mask = PCIM_BAR_MEM_BASE;
640 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
641 		}
642 		break;
643 	case PCIBAR_MEM32:
644 		baseptr = &pci_emul_membase32;
645 		limit = PCI_EMUL_MEMLIMIT32;
646 		mask = PCIM_BAR_MEM_BASE;
647 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
648 		break;
649 	default:
650 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
651 		assert(0);
652 	}
653 
654 	if (baseptr != NULL) {
655 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
656 		if (error != 0)
657 			return (error);
658 	}
659 
660 	pdi->pi_bar[idx].type = type;
661 	pdi->pi_bar[idx].addr = addr;
662 	pdi->pi_bar[idx].size = size;
663 
664 	/* Initialize the BAR register in config space */
665 	bar = (addr & mask) | lobits;
666 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
667 
668 	if (type == PCIBAR_MEM64) {
669 		assert(idx + 1 <= PCI_BARMAX);
670 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
671 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
672 	}
673 
674 	register_bar(pdi, idx);
675 
676 	return (0);
677 }
678 
679 #define	CAP_START_OFFSET	0x40
680 static int
681 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
682 {
683 	int i, capoff, reallen;
684 	uint16_t sts;
685 
686 	assert(caplen > 0);
687 
688 	reallen = roundup2(caplen, 4);		/* dword aligned */
689 
690 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
691 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
692 		capoff = CAP_START_OFFSET;
693 	else
694 		capoff = pi->pi_capend + 1;
695 
696 	/* Check if we have enough space */
697 	if (capoff + reallen > PCI_REGMAX + 1)
698 		return (-1);
699 
700 	/* Set the previous capability pointer */
701 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
702 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
703 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
704 	} else
705 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
706 
707 	/* Copy the capability */
708 	for (i = 0; i < caplen; i++)
709 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
710 
711 	/* Set the next capability pointer */
712 	pci_set_cfgdata8(pi, capoff + 1, 0);
713 
714 	pi->pi_prevcap = capoff;
715 	pi->pi_capend = capoff + reallen - 1;
716 	return (0);
717 }
718 
719 static struct pci_devemu *
720 pci_emul_finddev(char *name)
721 {
722 	struct pci_devemu **pdpp, *pdp;
723 
724 	SET_FOREACH(pdpp, pci_devemu_set) {
725 		pdp = *pdpp;
726 		if (!strcmp(pdp->pe_emu, name)) {
727 			return (pdp);
728 		}
729 	}
730 
731 	return (NULL);
732 }
733 
734 static int
735 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
736     int func, struct funcinfo *fi)
737 {
738 	struct pci_devinst *pdi;
739 	int err;
740 
741 	pdi = calloc(1, sizeof(struct pci_devinst));
742 
743 	pdi->pi_vmctx = ctx;
744 	pdi->pi_bus = bus;
745 	pdi->pi_slot = slot;
746 	pdi->pi_func = func;
747 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
748 	pdi->pi_lintr.pin = 0;
749 	pdi->pi_lintr.state = IDLE;
750 	pdi->pi_lintr.pirq_pin = 0;
751 	pdi->pi_lintr.ioapic_irq = 0;
752 	pdi->pi_d = pde;
753 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
754 
755 	/* Disable legacy interrupts */
756 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
757 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
758 
759 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
760 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
761 
762 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
763 	if (err == 0)
764 		fi->fi_devi = pdi;
765 	else
766 		free(pdi);
767 
768 	return (err);
769 }
770 
771 void
772 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
773 {
774 	int mmc;
775 
776 	/* Number of msi messages must be a power of 2 between 1 and 32 */
777 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
778 	mmc = ffs(msgnum) - 1;
779 
780 	bzero(msicap, sizeof(struct msicap));
781 	msicap->capid = PCIY_MSI;
782 	msicap->nextptr = nextptr;
783 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
784 }
785 
786 int
787 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
788 {
789 	struct msicap msicap;
790 
791 	pci_populate_msicap(&msicap, msgnum, 0);
792 
793 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
794 }
795 
796 static void
797 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
798 		     uint32_t msix_tab_size)
799 {
800 
801 	assert(msix_tab_size % 4096 == 0);
802 
803 	bzero(msixcap, sizeof(struct msixcap));
804 	msixcap->capid = PCIY_MSIX;
805 
806 	/*
807 	 * Message Control Register, all fields set to
808 	 * zero except for the Table Size.
809 	 * Note: Table size N is encoded as N-1
810 	 */
811 	msixcap->msgctrl = msgnum - 1;
812 
813 	/*
814 	 * MSI-X BAR setup:
815 	 * - MSI-X table start at offset 0
816 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
817 	 */
818 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
819 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
820 }
821 
822 static void
823 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
824 {
825 	int i, table_size;
826 
827 	assert(table_entries > 0);
828 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
829 
830 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
831 	pi->pi_msix.table = calloc(1, table_size);
832 
833 	/* set mask bit of vector control register */
834 	for (i = 0; i < table_entries; i++)
835 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
836 }
837 
838 int
839 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
840 {
841 	uint32_t tab_size;
842 	struct msixcap msixcap;
843 
844 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
845 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
846 
847 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
848 
849 	/* Align table size to nearest 4K */
850 	tab_size = roundup2(tab_size, 4096);
851 
852 	pi->pi_msix.table_bar = barnum;
853 	pi->pi_msix.pba_bar   = barnum;
854 	pi->pi_msix.table_offset = 0;
855 	pi->pi_msix.table_count = msgnum;
856 	pi->pi_msix.pba_offset = tab_size;
857 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
858 
859 	pci_msix_table_init(pi, msgnum);
860 
861 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
862 
863 	/* allocate memory for MSI-X Table and PBA */
864 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
865 				tab_size + pi->pi_msix.pba_size);
866 
867 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
868 					sizeof(msixcap)));
869 }
870 
871 void
872 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
873 		 int bytes, uint32_t val)
874 {
875 	uint16_t msgctrl, rwmask;
876 	int off;
877 
878 	off = offset - capoff;
879 	/* Message Control Register */
880 	if (off == 2 && bytes == 2) {
881 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
882 		msgctrl = pci_get_cfgdata16(pi, offset);
883 		msgctrl &= ~rwmask;
884 		msgctrl |= val & rwmask;
885 		val = msgctrl;
886 
887 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
888 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
889 		pci_lintr_update(pi);
890 	}
891 
892 	CFGWRITE(pi, offset, val, bytes);
893 }
894 
895 void
896 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
897 		int bytes, uint32_t val)
898 {
899 	uint16_t msgctrl, rwmask, msgdata, mme;
900 	uint32_t addrlo;
901 
902 	/*
903 	 * If guest is writing to the message control register make sure
904 	 * we do not overwrite read-only fields.
905 	 */
906 	if ((offset - capoff) == 2 && bytes == 2) {
907 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
908 		msgctrl = pci_get_cfgdata16(pi, offset);
909 		msgctrl &= ~rwmask;
910 		msgctrl |= val & rwmask;
911 		val = msgctrl;
912 
913 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
914 		if (msgctrl & PCIM_MSICTRL_64BIT)
915 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
916 		else
917 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
918 
919 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
920 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
921 		if (pi->pi_msi.enabled) {
922 			pi->pi_msi.addr = addrlo;
923 			pi->pi_msi.msg_data = msgdata;
924 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
925 		} else {
926 			pi->pi_msi.maxmsgnum = 0;
927 		}
928 		pci_lintr_update(pi);
929 	}
930 
931 	CFGWRITE(pi, offset, val, bytes);
932 }
933 
934 void
935 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
936 		 int bytes, uint32_t val)
937 {
938 
939 	/* XXX don't write to the readonly parts */
940 	CFGWRITE(pi, offset, val, bytes);
941 }
942 
943 #define	PCIECAP_VERSION	0x2
944 int
945 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
946 {
947 	int err;
948 	struct pciecap pciecap;
949 
950 	if (type != PCIEM_TYPE_ROOT_PORT)
951 		return (-1);
952 
953 	bzero(&pciecap, sizeof(pciecap));
954 
955 	pciecap.capid = PCIY_EXPRESS;
956 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
957 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
958 	pciecap.link_status = 0x11;		/* gen1, x1 */
959 
960 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
961 	return (err);
962 }
963 
964 /*
965  * This function assumes that 'coff' is in the capabilities region of the
966  * config space.
967  */
968 static void
969 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
970 {
971 	int capid;
972 	uint8_t capoff, nextoff;
973 
974 	/* Do not allow un-aligned writes */
975 	if ((offset & (bytes - 1)) != 0)
976 		return;
977 
978 	/* Find the capability that we want to update */
979 	capoff = CAP_START_OFFSET;
980 	while (1) {
981 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
982 		if (nextoff == 0)
983 			break;
984 		if (offset >= capoff && offset < nextoff)
985 			break;
986 
987 		capoff = nextoff;
988 	}
989 	assert(offset >= capoff);
990 
991 	/*
992 	 * Capability ID and Next Capability Pointer are readonly.
993 	 * However, some o/s's do 4-byte writes that include these.
994 	 * For this case, trim the write back to 2 bytes and adjust
995 	 * the data.
996 	 */
997 	if (offset == capoff || offset == capoff + 1) {
998 		if (offset == capoff && bytes == 4) {
999 			bytes = 2;
1000 			offset += 2;
1001 			val >>= 16;
1002 		} else
1003 			return;
1004 	}
1005 
1006 	capid = pci_get_cfgdata8(pi, capoff);
1007 	switch (capid) {
1008 	case PCIY_MSI:
1009 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1010 		break;
1011 	case PCIY_MSIX:
1012 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1013 		break;
1014 	case PCIY_EXPRESS:
1015 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1016 		break;
1017 	default:
1018 		break;
1019 	}
1020 }
1021 
1022 static int
1023 pci_emul_iscap(struct pci_devinst *pi, int offset)
1024 {
1025 	uint16_t sts;
1026 
1027 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1028 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1029 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1030 			return (1);
1031 	}
1032 	return (0);
1033 }
1034 
1035 static int
1036 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1037 			  int size, uint64_t *val, void *arg1, long arg2)
1038 {
1039 	/*
1040 	 * Ignore writes; return 0xff's for reads. The mem read code
1041 	 * will take care of truncating to the correct size.
1042 	 */
1043 	if (dir == MEM_F_READ) {
1044 		*val = 0xffffffffffffffff;
1045 	}
1046 
1047 	return (0);
1048 }
1049 
1050 static int
1051 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1052     int bytes, uint64_t *val, void *arg1, long arg2)
1053 {
1054 	int bus, slot, func, coff, in;
1055 
1056 	coff = addr & 0xfff;
1057 	func = (addr >> 12) & 0x7;
1058 	slot = (addr >> 15) & 0x1f;
1059 	bus = (addr >> 20) & 0xff;
1060 	in = (dir == MEM_F_READ);
1061 	if (in)
1062 		*val = ~0UL;
1063 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1064 	return (0);
1065 }
1066 
1067 uint64_t
1068 pci_ecfg_base(void)
1069 {
1070 
1071 	return (PCI_EMUL_ECFG_BASE);
1072 }
1073 
1074 #define	BUSIO_ROUNDUP		32
1075 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1076 
1077 int
1078 init_pci(struct vmctx *ctx)
1079 {
1080 	struct mem_range mr;
1081 	struct pci_devemu *pde;
1082 	struct businfo *bi;
1083 	struct slotinfo *si;
1084 	struct funcinfo *fi;
1085 	size_t lowmem;
1086 	int bus, slot, func;
1087 	int error;
1088 
1089 	pci_emul_iobase = PCI_EMUL_IOBASE;
1090 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1091 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1092 
1093 	for (bus = 0; bus < MAXBUSES; bus++) {
1094 		if ((bi = pci_businfo[bus]) == NULL)
1095 			continue;
1096 		/*
1097 		 * Keep track of the i/o and memory resources allocated to
1098 		 * this bus.
1099 		 */
1100 		bi->iobase = pci_emul_iobase;
1101 		bi->membase32 = pci_emul_membase32;
1102 		bi->membase64 = pci_emul_membase64;
1103 
1104 		for (slot = 0; slot < MAXSLOTS; slot++) {
1105 			si = &bi->slotinfo[slot];
1106 			for (func = 0; func < MAXFUNCS; func++) {
1107 				fi = &si->si_funcs[func];
1108 				if (fi->fi_name == NULL)
1109 					continue;
1110 				pde = pci_emul_finddev(fi->fi_name);
1111 				assert(pde != NULL);
1112 				error = pci_emul_init(ctx, pde, bus, slot,
1113 				    func, fi);
1114 				if (error)
1115 					return (error);
1116 			}
1117 		}
1118 
1119 		/*
1120 		 * Add some slop to the I/O and memory resources decoded by
1121 		 * this bus to give a guest some flexibility if it wants to
1122 		 * reprogram the BARs.
1123 		 */
1124 		pci_emul_iobase += BUSIO_ROUNDUP;
1125 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1126 		bi->iolimit = pci_emul_iobase;
1127 
1128 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1129 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1130 		    BUSMEM_ROUNDUP);
1131 		bi->memlimit32 = pci_emul_membase32;
1132 
1133 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1134 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1135 		    BUSMEM_ROUNDUP);
1136 		bi->memlimit64 = pci_emul_membase64;
1137 	}
1138 
1139 	/*
1140 	 * PCI backends are initialized before routing INTx interrupts
1141 	 * so that LPC devices are able to reserve ISA IRQs before
1142 	 * routing PIRQ pins.
1143 	 */
1144 	for (bus = 0; bus < MAXBUSES; bus++) {
1145 		if ((bi = pci_businfo[bus]) == NULL)
1146 			continue;
1147 
1148 		for (slot = 0; slot < MAXSLOTS; slot++) {
1149 			si = &bi->slotinfo[slot];
1150 			for (func = 0; func < MAXFUNCS; func++) {
1151 				fi = &si->si_funcs[func];
1152 				if (fi->fi_devi == NULL)
1153 					continue;
1154 				pci_lintr_route(fi->fi_devi);
1155 			}
1156 		}
1157 	}
1158 	lpc_pirq_routed();
1159 
1160 	/*
1161 	 * The guest physical memory map looks like the following:
1162 	 * [0,		    lowmem)		guest system memory
1163 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1164 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1165 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1166 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1167 	 * [4GB,	    4GB + highmem)
1168 	 */
1169 
1170 	/*
1171 	 * Accesses to memory addresses that are not allocated to system
1172 	 * memory or PCI devices return 0xff's.
1173 	 */
1174 	lowmem = vm_get_lowmem_size(ctx);
1175 	bzero(&mr, sizeof(struct mem_range));
1176 	mr.name = "PCI hole";
1177 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1178 	mr.base = lowmem;
1179 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1180 	mr.handler = pci_emul_fallback_handler;
1181 	error = register_mem_fallback(&mr);
1182 	assert(error == 0);
1183 
1184 	/* PCI extended config space */
1185 	bzero(&mr, sizeof(struct mem_range));
1186 	mr.name = "PCI ECFG";
1187 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1188 	mr.base = PCI_EMUL_ECFG_BASE;
1189 	mr.size = PCI_EMUL_ECFG_SIZE;
1190 	mr.handler = pci_emul_ecfg_handler;
1191 	error = register_mem(&mr);
1192 	assert(error == 0);
1193 
1194 	return (0);
1195 }
1196 
1197 static void
1198 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1199     void *arg)
1200 {
1201 
1202 	dsdt_line("  Package ()");
1203 	dsdt_line("  {");
1204 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1205 	dsdt_line("    0x%02X,", pin - 1);
1206 	dsdt_line("    Zero,");
1207 	dsdt_line("    0x%X", ioapic_irq);
1208 	dsdt_line("  },");
1209 }
1210 
1211 static void
1212 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1213     void *arg)
1214 {
1215 	char *name;
1216 
1217 	name = lpc_pirq_name(pirq_pin);
1218 	if (name == NULL)
1219 		return;
1220 	dsdt_line("  Package ()");
1221 	dsdt_line("  {");
1222 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1223 	dsdt_line("    0x%02X,", pin - 1);
1224 	dsdt_line("    %s,", name);
1225 	dsdt_line("    0x00");
1226 	dsdt_line("  },");
1227 	free(name);
1228 }
1229 
1230 /*
1231  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1232  * corresponding to each PCI bus.
1233  */
1234 static void
1235 pci_bus_write_dsdt(int bus)
1236 {
1237 	struct businfo *bi;
1238 	struct slotinfo *si;
1239 	struct pci_devinst *pi;
1240 	int count, func, slot;
1241 
1242 	/*
1243 	 * If there are no devices on this 'bus' then just return.
1244 	 */
1245 	if ((bi = pci_businfo[bus]) == NULL) {
1246 		/*
1247 		 * Bus 0 is special because it decodes the I/O ports used
1248 		 * for PCI config space access even if there are no devices
1249 		 * on it.
1250 		 */
1251 		if (bus != 0)
1252 			return;
1253 	}
1254 
1255 	dsdt_line("  Device (PC%02X)", bus);
1256 	dsdt_line("  {");
1257 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1258 	dsdt_line("    Name (_ADR, Zero)");
1259 
1260 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1261 	dsdt_line("    {");
1262 	dsdt_line("        Return (0x%08X)", bus);
1263 	dsdt_line("    }");
1264 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1265 	dsdt_line("    {");
1266 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1267 	    "MaxFixed, PosDecode,");
1268 	dsdt_line("        0x0000,             // Granularity");
1269 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1270 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1271 	dsdt_line("        0x0000,             // Translation Offset");
1272 	dsdt_line("        0x0001,             // Length");
1273 	dsdt_line("        ,, )");
1274 
1275 	if (bus == 0) {
1276 		dsdt_indent(3);
1277 		dsdt_fixed_ioport(0xCF8, 8);
1278 		dsdt_unindent(3);
1279 
1280 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1281 		    "PosDecode, EntireRange,");
1282 		dsdt_line("        0x0000,             // Granularity");
1283 		dsdt_line("        0x0000,             // Range Minimum");
1284 		dsdt_line("        0x0CF7,             // Range Maximum");
1285 		dsdt_line("        0x0000,             // Translation Offset");
1286 		dsdt_line("        0x0CF8,             // Length");
1287 		dsdt_line("        ,, , TypeStatic)");
1288 
1289 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1290 		    "PosDecode, EntireRange,");
1291 		dsdt_line("        0x0000,             // Granularity");
1292 		dsdt_line("        0x0D00,             // Range Minimum");
1293 		dsdt_line("        0x%04X,             // Range Maximum",
1294 		    PCI_EMUL_IOBASE - 1);
1295 		dsdt_line("        0x0000,             // Translation Offset");
1296 		dsdt_line("        0x%04X,             // Length",
1297 		    PCI_EMUL_IOBASE - 0x0D00);
1298 		dsdt_line("        ,, , TypeStatic)");
1299 
1300 		if (bi == NULL) {
1301 			dsdt_line("    })");
1302 			goto done;
1303 		}
1304 	}
1305 	assert(bi != NULL);
1306 
1307 	/* i/o window */
1308 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1309 	    "PosDecode, EntireRange,");
1310 	dsdt_line("        0x0000,             // Granularity");
1311 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1312 	dsdt_line("        0x%04X,             // Range Maximum",
1313 	    bi->iolimit - 1);
1314 	dsdt_line("        0x0000,             // Translation Offset");
1315 	dsdt_line("        0x%04X,             // Length",
1316 	    bi->iolimit - bi->iobase);
1317 	dsdt_line("        ,, , TypeStatic)");
1318 
1319 	/* mmio window (32-bit) */
1320 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1321 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1322 	dsdt_line("        0x00000000,         // Granularity");
1323 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1324 	dsdt_line("        0x%08X,         // Range Maximum\n",
1325 	    bi->memlimit32 - 1);
1326 	dsdt_line("        0x00000000,         // Translation Offset");
1327 	dsdt_line("        0x%08X,         // Length\n",
1328 	    bi->memlimit32 - bi->membase32);
1329 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1330 
1331 	/* mmio window (64-bit) */
1332 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1333 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1334 	dsdt_line("        0x0000000000000000, // Granularity");
1335 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1336 	dsdt_line("        0x%016lX, // Range Maximum\n",
1337 	    bi->memlimit64 - 1);
1338 	dsdt_line("        0x0000000000000000, // Translation Offset");
1339 	dsdt_line("        0x%016lX, // Length\n",
1340 	    bi->memlimit64 - bi->membase64);
1341 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1342 	dsdt_line("    })");
1343 
1344 	count = pci_count_lintr(bus);
1345 	if (count != 0) {
1346 		dsdt_indent(2);
1347 		dsdt_line("Name (PPRT, Package ()");
1348 		dsdt_line("{");
1349 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1350 		dsdt_line("})");
1351 		dsdt_line("Name (APRT, Package ()");
1352 		dsdt_line("{");
1353 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1354 		dsdt_line("})");
1355 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1356 		dsdt_line("{");
1357 		dsdt_line("  If (PICM)");
1358 		dsdt_line("  {");
1359 		dsdt_line("    Return (APRT)");
1360 		dsdt_line("  }");
1361 		dsdt_line("  Else");
1362 		dsdt_line("  {");
1363 		dsdt_line("    Return (PPRT)");
1364 		dsdt_line("  }");
1365 		dsdt_line("}");
1366 		dsdt_unindent(2);
1367 	}
1368 
1369 	dsdt_indent(2);
1370 	for (slot = 0; slot < MAXSLOTS; slot++) {
1371 		si = &bi->slotinfo[slot];
1372 		for (func = 0; func < MAXFUNCS; func++) {
1373 			pi = si->si_funcs[func].fi_devi;
1374 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1375 				pi->pi_d->pe_write_dsdt(pi);
1376 		}
1377 	}
1378 	dsdt_unindent(2);
1379 done:
1380 	dsdt_line("  }");
1381 }
1382 
1383 void
1384 pci_write_dsdt(void)
1385 {
1386 	int bus;
1387 
1388 	dsdt_indent(1);
1389 	dsdt_line("Name (PICM, 0x00)");
1390 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1391 	dsdt_line("{");
1392 	dsdt_line("  Store (Arg0, PICM)");
1393 	dsdt_line("}");
1394 	dsdt_line("");
1395 	dsdt_line("Scope (_SB)");
1396 	dsdt_line("{");
1397 	for (bus = 0; bus < MAXBUSES; bus++)
1398 		pci_bus_write_dsdt(bus);
1399 	dsdt_line("}");
1400 	dsdt_unindent(1);
1401 }
1402 
1403 int
1404 pci_bus_configured(int bus)
1405 {
1406 	assert(bus >= 0 && bus < MAXBUSES);
1407 	return (pci_businfo[bus] != NULL);
1408 }
1409 
1410 int
1411 pci_msi_enabled(struct pci_devinst *pi)
1412 {
1413 	return (pi->pi_msi.enabled);
1414 }
1415 
1416 int
1417 pci_msi_maxmsgnum(struct pci_devinst *pi)
1418 {
1419 	if (pi->pi_msi.enabled)
1420 		return (pi->pi_msi.maxmsgnum);
1421 	else
1422 		return (0);
1423 }
1424 
1425 int
1426 pci_msix_enabled(struct pci_devinst *pi)
1427 {
1428 
1429 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1430 }
1431 
1432 void
1433 pci_generate_msix(struct pci_devinst *pi, int index)
1434 {
1435 	struct msix_table_entry *mte;
1436 
1437 	if (!pci_msix_enabled(pi))
1438 		return;
1439 
1440 	if (pi->pi_msix.function_mask)
1441 		return;
1442 
1443 	if (index >= pi->pi_msix.table_count)
1444 		return;
1445 
1446 	mte = &pi->pi_msix.table[index];
1447 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1448 		/* XXX Set PBA bit if interrupt is disabled */
1449 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1450 	}
1451 }
1452 
1453 void
1454 pci_generate_msi(struct pci_devinst *pi, int index)
1455 {
1456 
1457 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1458 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1459 			     pi->pi_msi.msg_data + index);
1460 	}
1461 }
1462 
1463 static bool
1464 pci_lintr_permitted(struct pci_devinst *pi)
1465 {
1466 	uint16_t cmd;
1467 
1468 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1469 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1470 		(cmd & PCIM_CMD_INTxDIS)));
1471 }
1472 
1473 void
1474 pci_lintr_request(struct pci_devinst *pi)
1475 {
1476 	struct businfo *bi;
1477 	struct slotinfo *si;
1478 	int bestpin, bestcount, pin;
1479 
1480 	bi = pci_businfo[pi->pi_bus];
1481 	assert(bi != NULL);
1482 
1483 	/*
1484 	 * Just allocate a pin from our slot.  The pin will be
1485 	 * assigned IRQs later when interrupts are routed.
1486 	 */
1487 	si = &bi->slotinfo[pi->pi_slot];
1488 	bestpin = 0;
1489 	bestcount = si->si_intpins[0].ii_count;
1490 	for (pin = 1; pin < 4; pin++) {
1491 		if (si->si_intpins[pin].ii_count < bestcount) {
1492 			bestpin = pin;
1493 			bestcount = si->si_intpins[pin].ii_count;
1494 		}
1495 	}
1496 
1497 	si->si_intpins[bestpin].ii_count++;
1498 	pi->pi_lintr.pin = bestpin + 1;
1499 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1500 }
1501 
1502 static void
1503 pci_lintr_route(struct pci_devinst *pi)
1504 {
1505 	struct businfo *bi;
1506 	struct intxinfo *ii;
1507 
1508 	if (pi->pi_lintr.pin == 0)
1509 		return;
1510 
1511 	bi = pci_businfo[pi->pi_bus];
1512 	assert(bi != NULL);
1513 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1514 
1515 	/*
1516 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1517 	 * is not yet assigned.
1518 	 */
1519 	if (ii->ii_ioapic_irq == 0)
1520 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1521 	assert(ii->ii_ioapic_irq > 0);
1522 
1523 	/*
1524 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1525 	 * not yet assigned.
1526 	 */
1527 	if (ii->ii_pirq_pin == 0)
1528 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1529 	assert(ii->ii_pirq_pin > 0);
1530 
1531 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1532 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1533 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1534 }
1535 
1536 void
1537 pci_lintr_assert(struct pci_devinst *pi)
1538 {
1539 
1540 	assert(pi->pi_lintr.pin > 0);
1541 
1542 	pthread_mutex_lock(&pi->pi_lintr.lock);
1543 	if (pi->pi_lintr.state == IDLE) {
1544 		if (pci_lintr_permitted(pi)) {
1545 			pi->pi_lintr.state = ASSERTED;
1546 			pci_irq_assert(pi);
1547 		} else
1548 			pi->pi_lintr.state = PENDING;
1549 	}
1550 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1551 }
1552 
1553 void
1554 pci_lintr_deassert(struct pci_devinst *pi)
1555 {
1556 
1557 	assert(pi->pi_lintr.pin > 0);
1558 
1559 	pthread_mutex_lock(&pi->pi_lintr.lock);
1560 	if (pi->pi_lintr.state == ASSERTED) {
1561 		pi->pi_lintr.state = IDLE;
1562 		pci_irq_deassert(pi);
1563 	} else if (pi->pi_lintr.state == PENDING)
1564 		pi->pi_lintr.state = IDLE;
1565 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1566 }
1567 
1568 static void
1569 pci_lintr_update(struct pci_devinst *pi)
1570 {
1571 
1572 	pthread_mutex_lock(&pi->pi_lintr.lock);
1573 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1574 		pci_irq_deassert(pi);
1575 		pi->pi_lintr.state = PENDING;
1576 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1577 		pi->pi_lintr.state = ASSERTED;
1578 		pci_irq_assert(pi);
1579 	}
1580 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1581 }
1582 
1583 int
1584 pci_count_lintr(int bus)
1585 {
1586 	int count, slot, pin;
1587 	struct slotinfo *slotinfo;
1588 
1589 	count = 0;
1590 	if (pci_businfo[bus] != NULL) {
1591 		for (slot = 0; slot < MAXSLOTS; slot++) {
1592 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1593 			for (pin = 0; pin < 4; pin++) {
1594 				if (slotinfo->si_intpins[pin].ii_count != 0)
1595 					count++;
1596 			}
1597 		}
1598 	}
1599 	return (count);
1600 }
1601 
1602 void
1603 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1604 {
1605 	struct businfo *bi;
1606 	struct slotinfo *si;
1607 	struct intxinfo *ii;
1608 	int slot, pin;
1609 
1610 	if ((bi = pci_businfo[bus]) == NULL)
1611 		return;
1612 
1613 	for (slot = 0; slot < MAXSLOTS; slot++) {
1614 		si = &bi->slotinfo[slot];
1615 		for (pin = 0; pin < 4; pin++) {
1616 			ii = &si->si_intpins[pin];
1617 			if (ii->ii_count != 0)
1618 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1619 				    ii->ii_ioapic_irq, arg);
1620 		}
1621 	}
1622 }
1623 
1624 /*
1625  * Return 1 if the emulated device in 'slot' is a multi-function device.
1626  * Return 0 otherwise.
1627  */
1628 static int
1629 pci_emul_is_mfdev(int bus, int slot)
1630 {
1631 	struct businfo *bi;
1632 	struct slotinfo *si;
1633 	int f, numfuncs;
1634 
1635 	numfuncs = 0;
1636 	if ((bi = pci_businfo[bus]) != NULL) {
1637 		si = &bi->slotinfo[slot];
1638 		for (f = 0; f < MAXFUNCS; f++) {
1639 			if (si->si_funcs[f].fi_devi != NULL) {
1640 				numfuncs++;
1641 			}
1642 		}
1643 	}
1644 	return (numfuncs > 1);
1645 }
1646 
1647 /*
1648  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1649  * whether or not is a multi-function being emulated in the pci 'slot'.
1650  */
1651 static void
1652 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1653 {
1654 	int mfdev;
1655 
1656 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1657 		mfdev = pci_emul_is_mfdev(bus, slot);
1658 		switch (bytes) {
1659 		case 1:
1660 		case 2:
1661 			*rv &= ~PCIM_MFDEV;
1662 			if (mfdev) {
1663 				*rv |= PCIM_MFDEV;
1664 			}
1665 			break;
1666 		case 4:
1667 			*rv &= ~(PCIM_MFDEV << 16);
1668 			if (mfdev) {
1669 				*rv |= (PCIM_MFDEV << 16);
1670 			}
1671 			break;
1672 		}
1673 	}
1674 }
1675 
1676 static void
1677 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1678 {
1679 	int i, rshift;
1680 	uint32_t cmd, cmd2, changed, old, readonly;
1681 
1682 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1683 
1684 	/*
1685 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1686 	 *
1687 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1688 	 * 'write 1 to clear'. However these bits are not set to '1' by
1689 	 * any device emulation so it is simpler to treat them as readonly.
1690 	 */
1691 	rshift = (coff & 0x3) * 8;
1692 	readonly = 0xFFFFF880 >> rshift;
1693 
1694 	old = CFGREAD(pi, coff, bytes);
1695 	new &= ~readonly;
1696 	new |= (old & readonly);
1697 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1698 
1699 	cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1700 	changed = cmd ^ cmd2;
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 (porten(pi))
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 (memen(pi))
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_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1744     int coff, int bytes, uint32_t *eax)
1745 {
1746 	struct businfo *bi;
1747 	struct slotinfo *si;
1748 	struct pci_devinst *pi;
1749 	struct pci_devemu *pe;
1750 	int idx, needcfg;
1751 	uint64_t addr, bar, mask;
1752 
1753 	if ((bi = pci_businfo[bus]) != NULL) {
1754 		si = &bi->slotinfo[slot];
1755 		pi = si->si_funcs[func].fi_devi;
1756 	} else
1757 		pi = NULL;
1758 
1759 	/*
1760 	 * Just return if there is no device at this slot:func or if the
1761 	 * the guest is doing an un-aligned access.
1762 	 */
1763 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1764 	    (coff & (bytes - 1)) != 0) {
1765 		if (in)
1766 			*eax = 0xffffffff;
1767 		return;
1768 	}
1769 
1770 	/*
1771 	 * Ignore all writes beyond the standard config space and return all
1772 	 * ones on reads.
1773 	 */
1774 	if (coff >= PCI_REGMAX + 1) {
1775 		if (in) {
1776 			*eax = 0xffffffff;
1777 			/*
1778 			 * Extended capabilities begin at offset 256 in config
1779 			 * space. Absence of extended capabilities is signaled
1780 			 * with all 0s in the extended capability header at
1781 			 * offset 256.
1782 			 */
1783 			if (coff <= PCI_REGMAX + 4)
1784 				*eax = 0x00000000;
1785 		}
1786 		return;
1787 	}
1788 
1789 	pe = pi->pi_d;
1790 
1791 	/*
1792 	 * Config read
1793 	 */
1794 	if (in) {
1795 		/* Let the device emulation override the default handler */
1796 		if (pe->pe_cfgread != NULL) {
1797 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1798 			    eax);
1799 		} else {
1800 			needcfg = 1;
1801 		}
1802 
1803 		if (needcfg)
1804 			*eax = CFGREAD(pi, coff, bytes);
1805 
1806 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1807 	} else {
1808 		/* Let the device emulation override the default handler */
1809 		if (pe->pe_cfgwrite != NULL &&
1810 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1811 			return;
1812 
1813 		/*
1814 		 * Special handling for write to BAR registers
1815 		 */
1816 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1817 			/*
1818 			 * Ignore writes to BAR registers that are not
1819 			 * 4-byte aligned.
1820 			 */
1821 			if (bytes != 4 || (coff & 0x3) != 0)
1822 				return;
1823 			idx = (coff - PCIR_BAR(0)) / 4;
1824 			mask = ~(pi->pi_bar[idx].size - 1);
1825 			switch (pi->pi_bar[idx].type) {
1826 			case PCIBAR_NONE:
1827 				pi->pi_bar[idx].addr = bar = 0;
1828 				break;
1829 			case PCIBAR_IO:
1830 				addr = *eax & mask;
1831 				addr &= 0xffff;
1832 				bar = addr | PCIM_BAR_IO_SPACE;
1833 				/*
1834 				 * Register the new BAR value for interception
1835 				 */
1836 				if (addr != pi->pi_bar[idx].addr) {
1837 					update_bar_address(pi, addr, idx,
1838 							   PCIBAR_IO);
1839 				}
1840 				break;
1841 			case PCIBAR_MEM32:
1842 				addr = bar = *eax & mask;
1843 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1844 				if (addr != pi->pi_bar[idx].addr) {
1845 					update_bar_address(pi, addr, idx,
1846 							   PCIBAR_MEM32);
1847 				}
1848 				break;
1849 			case PCIBAR_MEM64:
1850 				addr = bar = *eax & mask;
1851 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1852 				       PCIM_BAR_MEM_PREFETCH;
1853 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1854 					update_bar_address(pi, addr, idx,
1855 							   PCIBAR_MEM64);
1856 				}
1857 				break;
1858 			case PCIBAR_MEMHI64:
1859 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1860 				addr = ((uint64_t)*eax << 32) & mask;
1861 				bar = addr >> 32;
1862 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1863 					update_bar_address(pi, addr, idx - 1,
1864 							   PCIBAR_MEMHI64);
1865 				}
1866 				break;
1867 			default:
1868 				assert(0);
1869 			}
1870 			pci_set_cfgdata32(pi, coff, bar);
1871 
1872 		} else if (pci_emul_iscap(pi, coff)) {
1873 			pci_emul_capwrite(pi, coff, bytes, *eax);
1874 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1875 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1876 		} else {
1877 			CFGWRITE(pi, coff, *eax, bytes);
1878 		}
1879 	}
1880 }
1881 
1882 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1883 
1884 static int
1885 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1886 		 uint32_t *eax, void *arg)
1887 {
1888 	uint32_t x;
1889 
1890 	if (bytes != 4) {
1891 		if (in)
1892 			*eax = (bytes == 2) ? 0xffff : 0xff;
1893 		return (0);
1894 	}
1895 
1896 	if (in) {
1897 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1898 		if (cfgenable)
1899 			x |= CONF1_ENABLE;
1900 		*eax = x;
1901 	} else {
1902 		x = *eax;
1903 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1904 		cfgoff = x & PCI_REGMAX;
1905 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1906 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1907 		cfgbus = (x >> 16) & PCI_BUSMAX;
1908 	}
1909 
1910 	return (0);
1911 }
1912 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1913 
1914 static int
1915 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1916 		 uint32_t *eax, void *arg)
1917 {
1918 	int coff;
1919 
1920 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1921 
1922 	coff = cfgoff + (port - CONF1_DATA_PORT);
1923 	if (cfgenable) {
1924 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1925 		    eax);
1926 	} else {
1927 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1928 		if (in)
1929 			*eax = 0xffffffff;
1930 	}
1931 	return (0);
1932 }
1933 
1934 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1935 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1936 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1937 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1938 
1939 #define PCI_EMUL_TEST
1940 #ifdef PCI_EMUL_TEST
1941 /*
1942  * Define a dummy test device
1943  */
1944 #define DIOSZ	8
1945 #define DMEMSZ	4096
1946 struct pci_emul_dsoftc {
1947 	uint8_t   ioregs[DIOSZ];
1948 	uint8_t	  memregs[2][DMEMSZ];
1949 };
1950 
1951 #define	PCI_EMUL_MSI_MSGS	 4
1952 #define	PCI_EMUL_MSIX_MSGS	16
1953 
1954 static int
1955 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1956 {
1957 	int error;
1958 	struct pci_emul_dsoftc *sc;
1959 
1960 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1961 
1962 	pi->pi_arg = sc;
1963 
1964 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1965 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1966 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1967 
1968 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1969 	assert(error == 0);
1970 
1971 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1972 	assert(error == 0);
1973 
1974 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1975 	assert(error == 0);
1976 
1977 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1978 	assert(error == 0);
1979 
1980 	return (0);
1981 }
1982 
1983 static void
1984 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1985 	      uint64_t offset, int size, uint64_t value)
1986 {
1987 	int i;
1988 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1989 
1990 	if (baridx == 0) {
1991 		if (offset + size > DIOSZ) {
1992 			printf("diow: iow too large, offset %ld size %d\n",
1993 			       offset, size);
1994 			return;
1995 		}
1996 
1997 		if (size == 1) {
1998 			sc->ioregs[offset] = value & 0xff;
1999 		} else if (size == 2) {
2000 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2001 		} else if (size == 4) {
2002 			*(uint32_t *)&sc->ioregs[offset] = value;
2003 		} else {
2004 			printf("diow: iow unknown size %d\n", size);
2005 		}
2006 
2007 		/*
2008 		 * Special magic value to generate an interrupt
2009 		 */
2010 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2011 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2012 
2013 		if (value == 0xabcdef) {
2014 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2015 				pci_generate_msi(pi, i);
2016 		}
2017 	}
2018 
2019 	if (baridx == 1 || baridx == 2) {
2020 		if (offset + size > DMEMSZ) {
2021 			printf("diow: memw too large, offset %ld size %d\n",
2022 			       offset, size);
2023 			return;
2024 		}
2025 
2026 		i = baridx - 1;		/* 'memregs' index */
2027 
2028 		if (size == 1) {
2029 			sc->memregs[i][offset] = value;
2030 		} else if (size == 2) {
2031 			*(uint16_t *)&sc->memregs[i][offset] = value;
2032 		} else if (size == 4) {
2033 			*(uint32_t *)&sc->memregs[i][offset] = value;
2034 		} else if (size == 8) {
2035 			*(uint64_t *)&sc->memregs[i][offset] = value;
2036 		} else {
2037 			printf("diow: memw unknown size %d\n", size);
2038 		}
2039 
2040 		/*
2041 		 * magic interrupt ??
2042 		 */
2043 	}
2044 
2045 	if (baridx > 2 || baridx < 0) {
2046 		printf("diow: unknown bar idx %d\n", baridx);
2047 	}
2048 }
2049 
2050 static uint64_t
2051 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2052 	      uint64_t offset, int size)
2053 {
2054 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2055 	uint32_t value;
2056 	int i;
2057 
2058 	if (baridx == 0) {
2059 		if (offset + size > DIOSZ) {
2060 			printf("dior: ior too large, offset %ld size %d\n",
2061 			       offset, size);
2062 			return (0);
2063 		}
2064 
2065 		value = 0;
2066 		if (size == 1) {
2067 			value = sc->ioregs[offset];
2068 		} else if (size == 2) {
2069 			value = *(uint16_t *) &sc->ioregs[offset];
2070 		} else if (size == 4) {
2071 			value = *(uint32_t *) &sc->ioregs[offset];
2072 		} else {
2073 			printf("dior: ior unknown size %d\n", size);
2074 		}
2075 	}
2076 
2077 	if (baridx == 1 || baridx == 2) {
2078 		if (offset + size > DMEMSZ) {
2079 			printf("dior: memr too large, offset %ld size %d\n",
2080 			       offset, size);
2081 			return (0);
2082 		}
2083 
2084 		i = baridx - 1;		/* 'memregs' index */
2085 
2086 		if (size == 1) {
2087 			value = sc->memregs[i][offset];
2088 		} else if (size == 2) {
2089 			value = *(uint16_t *) &sc->memregs[i][offset];
2090 		} else if (size == 4) {
2091 			value = *(uint32_t *) &sc->memregs[i][offset];
2092 		} else if (size == 8) {
2093 			value = *(uint64_t *) &sc->memregs[i][offset];
2094 		} else {
2095 			printf("dior: ior unknown size %d\n", size);
2096 		}
2097 	}
2098 
2099 
2100 	if (baridx > 2 || baridx < 0) {
2101 		printf("dior: unknown bar idx %d\n", baridx);
2102 		return (0);
2103 	}
2104 
2105 	return (value);
2106 }
2107 
2108 struct pci_devemu pci_dummy = {
2109 	.pe_emu = "dummy",
2110 	.pe_init = pci_emul_dinit,
2111 	.pe_barwrite = pci_emul_diow,
2112 	.pe_barread = pci_emul_dior
2113 };
2114 PCI_EMUL_SET(pci_dummy);
2115 
2116 #endif /* PCI_EMUL_TEST */
2117