xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision fbbd9655e5107c68e4e0146ff22b73d7350475bc)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 
35 #include <ctype.h>
36 #include <errno.h>
37 #include <pthread.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <strings.h>
42 #include <assert.h>
43 #include <stdbool.h>
44 
45 #include <machine/vmm.h>
46 #include <vmmapi.h>
47 
48 #include "acpi.h"
49 #include "bhyverun.h"
50 #include "inout.h"
51 #include "ioapic.h"
52 #include "mem.h"
53 #include "pci_emul.h"
54 #include "pci_irq.h"
55 #include "pci_lpc.h"
56 
57 #define CONF1_ADDR_PORT    0x0cf8
58 #define CONF1_DATA_PORT    0x0cfc
59 
60 #define CONF1_ENABLE	   0x80000000ul
61 
62 #define	MAXBUSES	(PCI_BUSMAX + 1)
63 #define MAXSLOTS	(PCI_SLOTMAX + 1)
64 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
65 
66 struct funcinfo {
67 	char	*fi_name;
68 	char	*fi_param;
69 	struct pci_devinst *fi_devi;
70 };
71 
72 struct intxinfo {
73 	int	ii_count;
74 	int	ii_pirq_pin;
75 	int	ii_ioapic_irq;
76 };
77 
78 struct slotinfo {
79 	struct intxinfo si_intpins[4];
80 	struct funcinfo si_funcs[MAXFUNCS];
81 };
82 
83 struct businfo {
84 	uint16_t iobase, iolimit;		/* I/O window */
85 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
86 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
87 	struct slotinfo slotinfo[MAXSLOTS];
88 };
89 
90 static struct businfo *pci_businfo[MAXBUSES];
91 
92 SET_DECLARE(pci_devemu_set, struct pci_devemu);
93 
94 static uint64_t pci_emul_iobase;
95 static uint64_t pci_emul_membase32;
96 static uint64_t pci_emul_membase64;
97 
98 #define	PCI_EMUL_IOBASE		0x2000
99 #define	PCI_EMUL_IOLIMIT	0x10000
100 
101 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
102 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
103 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
104 
105 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
106 
107 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
108 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
109 
110 static struct pci_devemu *pci_emul_finddev(char *name);
111 static void pci_lintr_route(struct pci_devinst *pi);
112 static void pci_lintr_update(struct pci_devinst *pi);
113 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
114     int func, int coff, int bytes, uint32_t *val);
115 
116 static __inline void
117 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
118 {
119 
120 	if (bytes == 1)
121 		pci_set_cfgdata8(pi, coff, val);
122 	else if (bytes == 2)
123 		pci_set_cfgdata16(pi, coff, val);
124 	else
125 		pci_set_cfgdata32(pi, coff, val);
126 }
127 
128 static __inline uint32_t
129 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
130 {
131 
132 	if (bytes == 1)
133 		return (pci_get_cfgdata8(pi, coff));
134 	else if (bytes == 2)
135 		return (pci_get_cfgdata16(pi, coff));
136 	else
137 		return (pci_get_cfgdata32(pi, coff));
138 }
139 
140 /*
141  * I/O access
142  */
143 
144 /*
145  * Slot options are in the form:
146  *
147  *  <bus>:<slot>:<func>,<emul>[,<config>]
148  *  <slot>[:<func>],<emul>[,<config>]
149  *
150  *  slot is 0..31
151  *  func is 0..7
152  *  emul is a string describing the type of PCI device e.g. virtio-net
153  *  config is an optional string, depending on the device, that can be
154  *  used for configuration.
155  *   Examples are:
156  *     1,virtio-net,tap0
157  *     3:0,dummy
158  */
159 static void
160 pci_parse_slot_usage(char *aopt)
161 {
162 
163 	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
164 }
165 
166 int
167 pci_parse_slot(char *opt)
168 {
169 	struct businfo *bi;
170 	struct slotinfo *si;
171 	char *emul, *config, *str, *cp;
172 	int error, bnum, snum, fnum;
173 
174 	error = -1;
175 	str = strdup(opt);
176 
177 	emul = config = NULL;
178 	if ((cp = strchr(str, ',')) != NULL) {
179 		*cp = '\0';
180 		emul = cp + 1;
181 		if ((cp = strchr(emul, ',')) != NULL) {
182 			*cp = '\0';
183 			config = cp + 1;
184 		}
185 	} else {
186 		pci_parse_slot_usage(opt);
187 		goto done;
188 	}
189 
190 	/* <bus>:<slot>:<func> */
191 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
192 		bnum = 0;
193 		/* <slot>:<func> */
194 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
195 			fnum = 0;
196 			/* <slot> */
197 			if (sscanf(str, "%d", &snum) != 1) {
198 				snum = -1;
199 			}
200 		}
201 	}
202 
203 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
204 	    fnum < 0 || fnum >= MAXFUNCS) {
205 		pci_parse_slot_usage(opt);
206 		goto done;
207 	}
208 
209 	if (pci_businfo[bnum] == NULL)
210 		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
211 
212 	bi = pci_businfo[bnum];
213 	si = &bi->slotinfo[snum];
214 
215 	if (si->si_funcs[fnum].fi_name != NULL) {
216 		fprintf(stderr, "pci slot %d:%d already occupied!\n",
217 			snum, fnum);
218 		goto done;
219 	}
220 
221 	if (pci_emul_finddev(emul) == NULL) {
222 		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
223 			snum, fnum, emul);
224 		goto done;
225 	}
226 
227 	error = 0;
228 	si->si_funcs[fnum].fi_name = emul;
229 	si->si_funcs[fnum].fi_param = config;
230 
231 done:
232 	if (error)
233 		free(str);
234 
235 	return (error);
236 }
237 
238 static int
239 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
240 {
241 
242 	if (offset < pi->pi_msix.pba_offset)
243 		return (0);
244 
245 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
246 		return (0);
247 	}
248 
249 	return (1);
250 }
251 
252 int
253 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
254 		     uint64_t value)
255 {
256 	int msix_entry_offset;
257 	int tab_index;
258 	char *dest;
259 
260 	/* support only 4 or 8 byte writes */
261 	if (size != 4 && size != 8)
262 		return (-1);
263 
264 	/*
265 	 * Return if table index is beyond what device supports
266 	 */
267 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
268 	if (tab_index >= pi->pi_msix.table_count)
269 		return (-1);
270 
271 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
272 
273 	/* support only aligned writes */
274 	if ((msix_entry_offset % size) != 0)
275 		return (-1);
276 
277 	dest = (char *)(pi->pi_msix.table + tab_index);
278 	dest += msix_entry_offset;
279 
280 	if (size == 4)
281 		*((uint32_t *)dest) = value;
282 	else
283 		*((uint64_t *)dest) = value;
284 
285 	return (0);
286 }
287 
288 uint64_t
289 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
290 {
291 	char *dest;
292 	int msix_entry_offset;
293 	int tab_index;
294 	uint64_t retval = ~0;
295 
296 	/*
297 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
298 	 * table but we also allow 1 byte access to accommodate reads from
299 	 * ddb.
300 	 */
301 	if (size != 1 && size != 4 && size != 8)
302 		return (retval);
303 
304 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
305 
306 	/* support only aligned reads */
307 	if ((msix_entry_offset % size) != 0) {
308 		return (retval);
309 	}
310 
311 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
312 
313 	if (tab_index < pi->pi_msix.table_count) {
314 		/* valid MSI-X Table access */
315 		dest = (char *)(pi->pi_msix.table + tab_index);
316 		dest += msix_entry_offset;
317 
318 		if (size == 1)
319 			retval = *((uint8_t *)dest);
320 		else if (size == 4)
321 			retval = *((uint32_t *)dest);
322 		else
323 			retval = *((uint64_t *)dest);
324 	} else if (pci_valid_pba_offset(pi, offset)) {
325 		/* return 0 for PBA access */
326 		retval = 0;
327 	}
328 
329 	return (retval);
330 }
331 
332 int
333 pci_msix_table_bar(struct pci_devinst *pi)
334 {
335 
336 	if (pi->pi_msix.table != NULL)
337 		return (pi->pi_msix.table_bar);
338 	else
339 		return (-1);
340 }
341 
342 int
343 pci_msix_pba_bar(struct pci_devinst *pi)
344 {
345 
346 	if (pi->pi_msix.table != NULL)
347 		return (pi->pi_msix.pba_bar);
348 	else
349 		return (-1);
350 }
351 
352 static int
353 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
354 		    uint32_t *eax, void *arg)
355 {
356 	struct pci_devinst *pdi = arg;
357 	struct pci_devemu *pe = pdi->pi_d;
358 	uint64_t offset;
359 	int i;
360 
361 	for (i = 0; i <= PCI_BARMAX; i++) {
362 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
363 		    port >= pdi->pi_bar[i].addr &&
364 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
365 			offset = port - pdi->pi_bar[i].addr;
366 			if (in)
367 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
368 							 offset, bytes);
369 			else
370 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
371 						   bytes, *eax);
372 			return (0);
373 		}
374 	}
375 	return (-1);
376 }
377 
378 static int
379 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
380 		     int size, uint64_t *val, void *arg1, long arg2)
381 {
382 	struct pci_devinst *pdi = arg1;
383 	struct pci_devemu *pe = pdi->pi_d;
384 	uint64_t offset;
385 	int bidx = (int) arg2;
386 
387 	assert(bidx <= PCI_BARMAX);
388 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
389 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
390 	assert(addr >= pdi->pi_bar[bidx].addr &&
391 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
392 
393 	offset = addr - pdi->pi_bar[bidx].addr;
394 
395 	if (dir == MEM_F_WRITE) {
396 		if (size == 8) {
397 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
398 					   4, *val & 0xffffffff);
399 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
400 					   4, *val >> 32);
401 		} else {
402 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
403 					   size, *val);
404 		}
405 	} else {
406 		if (size == 8) {
407 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
408 						 offset, 4);
409 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
410 						  offset + 4, 4) << 32;
411 		} else {
412 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
413 						 offset, size);
414 		}
415 	}
416 
417 	return (0);
418 }
419 
420 
421 static int
422 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
423 			uint64_t *addr)
424 {
425 	uint64_t base;
426 
427 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
428 
429 	base = roundup2(*baseptr, size);
430 
431 	if (base + size <= limit) {
432 		*addr = base;
433 		*baseptr = base + size;
434 		return (0);
435 	} else
436 		return (-1);
437 }
438 
439 int
440 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
441 		   uint64_t size)
442 {
443 
444 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
445 }
446 
447 /*
448  * Register (or unregister) the MMIO or I/O region associated with the BAR
449  * register 'idx' of an emulated pci device.
450  */
451 static void
452 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
453 {
454 	int error;
455 	struct inout_port iop;
456 	struct mem_range mr;
457 
458 	switch (pi->pi_bar[idx].type) {
459 	case PCIBAR_IO:
460 		bzero(&iop, sizeof(struct inout_port));
461 		iop.name = pi->pi_name;
462 		iop.port = pi->pi_bar[idx].addr;
463 		iop.size = pi->pi_bar[idx].size;
464 		if (registration) {
465 			iop.flags = IOPORT_F_INOUT;
466 			iop.handler = pci_emul_io_handler;
467 			iop.arg = pi;
468 			error = register_inout(&iop);
469 		} else
470 			error = unregister_inout(&iop);
471 		break;
472 	case PCIBAR_MEM32:
473 	case PCIBAR_MEM64:
474 		bzero(&mr, sizeof(struct mem_range));
475 		mr.name = pi->pi_name;
476 		mr.base = pi->pi_bar[idx].addr;
477 		mr.size = pi->pi_bar[idx].size;
478 		if (registration) {
479 			mr.flags = MEM_F_RW;
480 			mr.handler = pci_emul_mem_handler;
481 			mr.arg1 = pi;
482 			mr.arg2 = idx;
483 			error = register_mem(&mr);
484 		} else
485 			error = unregister_mem(&mr);
486 		break;
487 	default:
488 		error = EINVAL;
489 		break;
490 	}
491 	assert(error == 0);
492 }
493 
494 static void
495 unregister_bar(struct pci_devinst *pi, int idx)
496 {
497 
498 	modify_bar_registration(pi, idx, 0);
499 }
500 
501 static void
502 register_bar(struct pci_devinst *pi, int idx)
503 {
504 
505 	modify_bar_registration(pi, idx, 1);
506 }
507 
508 /* Are we decoding i/o port accesses for the emulated pci device? */
509 static int
510 porten(struct pci_devinst *pi)
511 {
512 	uint16_t cmd;
513 
514 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
515 
516 	return (cmd & PCIM_CMD_PORTEN);
517 }
518 
519 /* Are we decoding memory accesses for the emulated pci device? */
520 static int
521 memen(struct pci_devinst *pi)
522 {
523 	uint16_t cmd;
524 
525 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
526 
527 	return (cmd & PCIM_CMD_MEMEN);
528 }
529 
530 /*
531  * Update the MMIO or I/O address that is decoded by the BAR register.
532  *
533  * If the pci device has enabled the address space decoding then intercept
534  * the address range decoded by the BAR register.
535  */
536 static void
537 update_bar_address(struct  pci_devinst *pi, uint64_t addr, int idx, int type)
538 {
539 	int decode;
540 
541 	if (pi->pi_bar[idx].type == PCIBAR_IO)
542 		decode = porten(pi);
543 	else
544 		decode = memen(pi);
545 
546 	if (decode)
547 		unregister_bar(pi, idx);
548 
549 	switch (type) {
550 	case PCIBAR_IO:
551 	case PCIBAR_MEM32:
552 		pi->pi_bar[idx].addr = addr;
553 		break;
554 	case PCIBAR_MEM64:
555 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
556 		pi->pi_bar[idx].addr |= addr;
557 		break;
558 	case PCIBAR_MEMHI64:
559 		pi->pi_bar[idx].addr &= 0xffffffff;
560 		pi->pi_bar[idx].addr |= addr;
561 		break;
562 	default:
563 		assert(0);
564 	}
565 
566 	if (decode)
567 		register_bar(pi, idx);
568 }
569 
570 int
571 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
572 		    enum pcibar_type type, uint64_t size)
573 {
574 	int error;
575 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
576 
577 	assert(idx >= 0 && idx <= PCI_BARMAX);
578 
579 	if ((size & (size - 1)) != 0)
580 		size = 1UL << flsl(size);	/* round up to a power of 2 */
581 
582 	/* Enforce minimum BAR sizes required by the PCI standard */
583 	if (type == PCIBAR_IO) {
584 		if (size < 4)
585 			size = 4;
586 	} else {
587 		if (size < 16)
588 			size = 16;
589 	}
590 
591 	switch (type) {
592 	case PCIBAR_NONE:
593 		baseptr = NULL;
594 		addr = mask = lobits = 0;
595 		break;
596 	case PCIBAR_IO:
597 		baseptr = &pci_emul_iobase;
598 		limit = PCI_EMUL_IOLIMIT;
599 		mask = PCIM_BAR_IO_BASE;
600 		lobits = PCIM_BAR_IO_SPACE;
601 		break;
602 	case PCIBAR_MEM64:
603 		/*
604 		 * XXX
605 		 * Some drivers do not work well if the 64-bit BAR is allocated
606 		 * above 4GB. Allow for this by allocating small requests under
607 		 * 4GB unless then allocation size is larger than some arbitrary
608 		 * number (32MB currently).
609 		 */
610 		if (size > 32 * 1024 * 1024) {
611 			/*
612 			 * XXX special case for device requiring peer-peer DMA
613 			 */
614 			if (size == 0x100000000UL)
615 				baseptr = &hostbase;
616 			else
617 				baseptr = &pci_emul_membase64;
618 			limit = PCI_EMUL_MEMLIMIT64;
619 			mask = PCIM_BAR_MEM_BASE;
620 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
621 				 PCIM_BAR_MEM_PREFETCH;
622 			break;
623 		} else {
624 			baseptr = &pci_emul_membase32;
625 			limit = PCI_EMUL_MEMLIMIT32;
626 			mask = PCIM_BAR_MEM_BASE;
627 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
628 		}
629 		break;
630 	case PCIBAR_MEM32:
631 		baseptr = &pci_emul_membase32;
632 		limit = PCI_EMUL_MEMLIMIT32;
633 		mask = PCIM_BAR_MEM_BASE;
634 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
635 		break;
636 	default:
637 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
638 		assert(0);
639 	}
640 
641 	if (baseptr != NULL) {
642 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
643 		if (error != 0)
644 			return (error);
645 	}
646 
647 	pdi->pi_bar[idx].type = type;
648 	pdi->pi_bar[idx].addr = addr;
649 	pdi->pi_bar[idx].size = size;
650 
651 	/* Initialize the BAR register in config space */
652 	bar = (addr & mask) | lobits;
653 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
654 
655 	if (type == PCIBAR_MEM64) {
656 		assert(idx + 1 <= PCI_BARMAX);
657 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
658 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
659 	}
660 
661 	register_bar(pdi, idx);
662 
663 	return (0);
664 }
665 
666 #define	CAP_START_OFFSET	0x40
667 static int
668 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
669 {
670 	int i, capoff, reallen;
671 	uint16_t sts;
672 
673 	assert(caplen > 0);
674 
675 	reallen = roundup2(caplen, 4);		/* dword aligned */
676 
677 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
678 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
679 		capoff = CAP_START_OFFSET;
680 	else
681 		capoff = pi->pi_capend + 1;
682 
683 	/* Check if we have enough space */
684 	if (capoff + reallen > PCI_REGMAX + 1)
685 		return (-1);
686 
687 	/* Set the previous capability pointer */
688 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
689 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
690 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
691 	} else
692 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
693 
694 	/* Copy the capability */
695 	for (i = 0; i < caplen; i++)
696 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
697 
698 	/* Set the next capability pointer */
699 	pci_set_cfgdata8(pi, capoff + 1, 0);
700 
701 	pi->pi_prevcap = capoff;
702 	pi->pi_capend = capoff + reallen - 1;
703 	return (0);
704 }
705 
706 static struct pci_devemu *
707 pci_emul_finddev(char *name)
708 {
709 	struct pci_devemu **pdpp, *pdp;
710 
711 	SET_FOREACH(pdpp, pci_devemu_set) {
712 		pdp = *pdpp;
713 		if (!strcmp(pdp->pe_emu, name)) {
714 			return (pdp);
715 		}
716 	}
717 
718 	return (NULL);
719 }
720 
721 static int
722 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
723     int func, struct funcinfo *fi)
724 {
725 	struct pci_devinst *pdi;
726 	int err;
727 
728 	pdi = calloc(1, sizeof(struct pci_devinst));
729 
730 	pdi->pi_vmctx = ctx;
731 	pdi->pi_bus = bus;
732 	pdi->pi_slot = slot;
733 	pdi->pi_func = func;
734 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
735 	pdi->pi_lintr.pin = 0;
736 	pdi->pi_lintr.state = IDLE;
737 	pdi->pi_lintr.pirq_pin = 0;
738 	pdi->pi_lintr.ioapic_irq = 0;
739 	pdi->pi_d = pde;
740 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
741 
742 	/* Disable legacy interrupts */
743 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
744 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
745 
746 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
747 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
748 
749 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
750 	if (err == 0)
751 		fi->fi_devi = pdi;
752 	else
753 		free(pdi);
754 
755 	return (err);
756 }
757 
758 void
759 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
760 {
761 	int mmc;
762 
763 	/* Number of msi messages must be a power of 2 between 1 and 32 */
764 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
765 	mmc = ffs(msgnum) - 1;
766 
767 	bzero(msicap, sizeof(struct msicap));
768 	msicap->capid = PCIY_MSI;
769 	msicap->nextptr = nextptr;
770 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
771 }
772 
773 int
774 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
775 {
776 	struct msicap msicap;
777 
778 	pci_populate_msicap(&msicap, msgnum, 0);
779 
780 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
781 }
782 
783 static void
784 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
785 		     uint32_t msix_tab_size)
786 {
787 
788 	assert(msix_tab_size % 4096 == 0);
789 
790 	bzero(msixcap, sizeof(struct msixcap));
791 	msixcap->capid = PCIY_MSIX;
792 
793 	/*
794 	 * Message Control Register, all fields set to
795 	 * zero except for the Table Size.
796 	 * Note: Table size N is encoded as N-1
797 	 */
798 	msixcap->msgctrl = msgnum - 1;
799 
800 	/*
801 	 * MSI-X BAR setup:
802 	 * - MSI-X table start at offset 0
803 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
804 	 */
805 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
806 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
807 }
808 
809 static void
810 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
811 {
812 	int i, table_size;
813 
814 	assert(table_entries > 0);
815 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
816 
817 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
818 	pi->pi_msix.table = calloc(1, table_size);
819 
820 	/* set mask bit of vector control register */
821 	for (i = 0; i < table_entries; i++)
822 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
823 }
824 
825 int
826 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
827 {
828 	uint32_t tab_size;
829 	struct msixcap msixcap;
830 
831 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
832 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
833 
834 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
835 
836 	/* Align table size to nearest 4K */
837 	tab_size = roundup2(tab_size, 4096);
838 
839 	pi->pi_msix.table_bar = barnum;
840 	pi->pi_msix.pba_bar   = barnum;
841 	pi->pi_msix.table_offset = 0;
842 	pi->pi_msix.table_count = msgnum;
843 	pi->pi_msix.pba_offset = tab_size;
844 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
845 
846 	pci_msix_table_init(pi, msgnum);
847 
848 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
849 
850 	/* allocate memory for MSI-X Table and PBA */
851 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
852 				tab_size + pi->pi_msix.pba_size);
853 
854 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
855 					sizeof(msixcap)));
856 }
857 
858 void
859 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
860 		 int bytes, uint32_t val)
861 {
862 	uint16_t msgctrl, rwmask;
863 	int off;
864 
865 	off = offset - capoff;
866 	/* Message Control Register */
867 	if (off == 2 && bytes == 2) {
868 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
869 		msgctrl = pci_get_cfgdata16(pi, offset);
870 		msgctrl &= ~rwmask;
871 		msgctrl |= val & rwmask;
872 		val = msgctrl;
873 
874 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
875 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
876 		pci_lintr_update(pi);
877 	}
878 
879 	CFGWRITE(pi, offset, val, bytes);
880 }
881 
882 void
883 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
884 		int bytes, uint32_t val)
885 {
886 	uint16_t msgctrl, rwmask, msgdata, mme;
887 	uint32_t addrlo;
888 
889 	/*
890 	 * If guest is writing to the message control register make sure
891 	 * we do not overwrite read-only fields.
892 	 */
893 	if ((offset - capoff) == 2 && bytes == 2) {
894 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
895 		msgctrl = pci_get_cfgdata16(pi, offset);
896 		msgctrl &= ~rwmask;
897 		msgctrl |= val & rwmask;
898 		val = msgctrl;
899 
900 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
901 		if (msgctrl & PCIM_MSICTRL_64BIT)
902 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
903 		else
904 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
905 
906 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
907 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
908 		if (pi->pi_msi.enabled) {
909 			pi->pi_msi.addr = addrlo;
910 			pi->pi_msi.msg_data = msgdata;
911 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
912 		} else {
913 			pi->pi_msi.maxmsgnum = 0;
914 		}
915 		pci_lintr_update(pi);
916 	}
917 
918 	CFGWRITE(pi, offset, val, bytes);
919 }
920 
921 void
922 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
923 		 int bytes, uint32_t val)
924 {
925 
926 	/* XXX don't write to the readonly parts */
927 	CFGWRITE(pi, offset, val, bytes);
928 }
929 
930 #define	PCIECAP_VERSION	0x2
931 int
932 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
933 {
934 	int err;
935 	struct pciecap pciecap;
936 
937 	if (type != PCIEM_TYPE_ROOT_PORT)
938 		return (-1);
939 
940 	bzero(&pciecap, sizeof(pciecap));
941 
942 	pciecap.capid = PCIY_EXPRESS;
943 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
944 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
945 	pciecap.link_status = 0x11;		/* gen1, x1 */
946 
947 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
948 	return (err);
949 }
950 
951 /*
952  * This function assumes that 'coff' is in the capabilities region of the
953  * config space.
954  */
955 static void
956 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
957 {
958 	int capid;
959 	uint8_t capoff, nextoff;
960 
961 	/* Do not allow un-aligned writes */
962 	if ((offset & (bytes - 1)) != 0)
963 		return;
964 
965 	/* Find the capability that we want to update */
966 	capoff = CAP_START_OFFSET;
967 	while (1) {
968 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
969 		if (nextoff == 0)
970 			break;
971 		if (offset >= capoff && offset < nextoff)
972 			break;
973 
974 		capoff = nextoff;
975 	}
976 	assert(offset >= capoff);
977 
978 	/*
979 	 * Capability ID and Next Capability Pointer are readonly.
980 	 * However, some o/s's do 4-byte writes that include these.
981 	 * For this case, trim the write back to 2 bytes and adjust
982 	 * the data.
983 	 */
984 	if (offset == capoff || offset == capoff + 1) {
985 		if (offset == capoff && bytes == 4) {
986 			bytes = 2;
987 			offset += 2;
988 			val >>= 16;
989 		} else
990 			return;
991 	}
992 
993 	capid = pci_get_cfgdata8(pi, capoff);
994 	switch (capid) {
995 	case PCIY_MSI:
996 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
997 		break;
998 	case PCIY_MSIX:
999 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1000 		break;
1001 	case PCIY_EXPRESS:
1002 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1003 		break;
1004 	default:
1005 		break;
1006 	}
1007 }
1008 
1009 static int
1010 pci_emul_iscap(struct pci_devinst *pi, int offset)
1011 {
1012 	uint16_t sts;
1013 
1014 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1015 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1016 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1017 			return (1);
1018 	}
1019 	return (0);
1020 }
1021 
1022 static int
1023 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1024 			  int size, uint64_t *val, void *arg1, long arg2)
1025 {
1026 	/*
1027 	 * Ignore writes; return 0xff's for reads. The mem read code
1028 	 * will take care of truncating to the correct size.
1029 	 */
1030 	if (dir == MEM_F_READ) {
1031 		*val = 0xffffffffffffffff;
1032 	}
1033 
1034 	return (0);
1035 }
1036 
1037 static int
1038 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1039     int bytes, uint64_t *val, void *arg1, long arg2)
1040 {
1041 	int bus, slot, func, coff, in;
1042 
1043 	coff = addr & 0xfff;
1044 	func = (addr >> 12) & 0x7;
1045 	slot = (addr >> 15) & 0x1f;
1046 	bus = (addr >> 20) & 0xff;
1047 	in = (dir == MEM_F_READ);
1048 	if (in)
1049 		*val = ~0UL;
1050 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1051 	return (0);
1052 }
1053 
1054 uint64_t
1055 pci_ecfg_base(void)
1056 {
1057 
1058 	return (PCI_EMUL_ECFG_BASE);
1059 }
1060 
1061 #define	BUSIO_ROUNDUP		32
1062 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1063 
1064 int
1065 init_pci(struct vmctx *ctx)
1066 {
1067 	struct mem_range mr;
1068 	struct pci_devemu *pde;
1069 	struct businfo *bi;
1070 	struct slotinfo *si;
1071 	struct funcinfo *fi;
1072 	size_t lowmem;
1073 	int bus, slot, func;
1074 	int error;
1075 
1076 	pci_emul_iobase = PCI_EMUL_IOBASE;
1077 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1078 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1079 
1080 	for (bus = 0; bus < MAXBUSES; bus++) {
1081 		if ((bi = pci_businfo[bus]) == NULL)
1082 			continue;
1083 		/*
1084 		 * Keep track of the i/o and memory resources allocated to
1085 		 * this bus.
1086 		 */
1087 		bi->iobase = pci_emul_iobase;
1088 		bi->membase32 = pci_emul_membase32;
1089 		bi->membase64 = pci_emul_membase64;
1090 
1091 		for (slot = 0; slot < MAXSLOTS; slot++) {
1092 			si = &bi->slotinfo[slot];
1093 			for (func = 0; func < MAXFUNCS; func++) {
1094 				fi = &si->si_funcs[func];
1095 				if (fi->fi_name == NULL)
1096 					continue;
1097 				pde = pci_emul_finddev(fi->fi_name);
1098 				assert(pde != NULL);
1099 				error = pci_emul_init(ctx, pde, bus, slot,
1100 				    func, fi);
1101 				if (error)
1102 					return (error);
1103 			}
1104 		}
1105 
1106 		/*
1107 		 * Add some slop to the I/O and memory resources decoded by
1108 		 * this bus to give a guest some flexibility if it wants to
1109 		 * reprogram the BARs.
1110 		 */
1111 		pci_emul_iobase += BUSIO_ROUNDUP;
1112 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1113 		bi->iolimit = pci_emul_iobase;
1114 
1115 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1116 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1117 		    BUSMEM_ROUNDUP);
1118 		bi->memlimit32 = pci_emul_membase32;
1119 
1120 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1121 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1122 		    BUSMEM_ROUNDUP);
1123 		bi->memlimit64 = pci_emul_membase64;
1124 	}
1125 
1126 	/*
1127 	 * PCI backends are initialized before routing INTx interrupts
1128 	 * so that LPC devices are able to reserve ISA IRQs before
1129 	 * routing PIRQ pins.
1130 	 */
1131 	for (bus = 0; bus < MAXBUSES; bus++) {
1132 		if ((bi = pci_businfo[bus]) == NULL)
1133 			continue;
1134 
1135 		for (slot = 0; slot < MAXSLOTS; slot++) {
1136 			si = &bi->slotinfo[slot];
1137 			for (func = 0; func < MAXFUNCS; func++) {
1138 				fi = &si->si_funcs[func];
1139 				if (fi->fi_devi == NULL)
1140 					continue;
1141 				pci_lintr_route(fi->fi_devi);
1142 			}
1143 		}
1144 	}
1145 	lpc_pirq_routed();
1146 
1147 	/*
1148 	 * The guest physical memory map looks like the following:
1149 	 * [0,		    lowmem)		guest system memory
1150 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1151 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1152 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1153 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1154 	 * [4GB,	    4GB + highmem)
1155 	 */
1156 
1157 	/*
1158 	 * Accesses to memory addresses that are not allocated to system
1159 	 * memory or PCI devices return 0xff's.
1160 	 */
1161 	lowmem = vm_get_lowmem_size(ctx);
1162 	bzero(&mr, sizeof(struct mem_range));
1163 	mr.name = "PCI hole";
1164 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1165 	mr.base = lowmem;
1166 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1167 	mr.handler = pci_emul_fallback_handler;
1168 	error = register_mem_fallback(&mr);
1169 	assert(error == 0);
1170 
1171 	/* PCI extended config space */
1172 	bzero(&mr, sizeof(struct mem_range));
1173 	mr.name = "PCI ECFG";
1174 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1175 	mr.base = PCI_EMUL_ECFG_BASE;
1176 	mr.size = PCI_EMUL_ECFG_SIZE;
1177 	mr.handler = pci_emul_ecfg_handler;
1178 	error = register_mem(&mr);
1179 	assert(error == 0);
1180 
1181 	return (0);
1182 }
1183 
1184 static void
1185 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1186     void *arg)
1187 {
1188 
1189 	dsdt_line("  Package ()");
1190 	dsdt_line("  {");
1191 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1192 	dsdt_line("    0x%02X,", pin - 1);
1193 	dsdt_line("    Zero,");
1194 	dsdt_line("    0x%X", ioapic_irq);
1195 	dsdt_line("  },");
1196 }
1197 
1198 static void
1199 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1200     void *arg)
1201 {
1202 	char *name;
1203 
1204 	name = lpc_pirq_name(pirq_pin);
1205 	if (name == NULL)
1206 		return;
1207 	dsdt_line("  Package ()");
1208 	dsdt_line("  {");
1209 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1210 	dsdt_line("    0x%02X,", pin - 1);
1211 	dsdt_line("    %s,", name);
1212 	dsdt_line("    0x00");
1213 	dsdt_line("  },");
1214 	free(name);
1215 }
1216 
1217 /*
1218  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1219  * corresponding to each PCI bus.
1220  */
1221 static void
1222 pci_bus_write_dsdt(int bus)
1223 {
1224 	struct businfo *bi;
1225 	struct slotinfo *si;
1226 	struct pci_devinst *pi;
1227 	int count, func, slot;
1228 
1229 	/*
1230 	 * If there are no devices on this 'bus' then just return.
1231 	 */
1232 	if ((bi = pci_businfo[bus]) == NULL) {
1233 		/*
1234 		 * Bus 0 is special because it decodes the I/O ports used
1235 		 * for PCI config space access even if there are no devices
1236 		 * on it.
1237 		 */
1238 		if (bus != 0)
1239 			return;
1240 	}
1241 
1242 	dsdt_line("  Device (PC%02X)", bus);
1243 	dsdt_line("  {");
1244 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1245 	dsdt_line("    Name (_ADR, Zero)");
1246 
1247 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1248 	dsdt_line("    {");
1249 	dsdt_line("        Return (0x%08X)", bus);
1250 	dsdt_line("    }");
1251 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1252 	dsdt_line("    {");
1253 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1254 	    "MaxFixed, PosDecode,");
1255 	dsdt_line("        0x0000,             // Granularity");
1256 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1257 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1258 	dsdt_line("        0x0000,             // Translation Offset");
1259 	dsdt_line("        0x0001,             // Length");
1260 	dsdt_line("        ,, )");
1261 
1262 	if (bus == 0) {
1263 		dsdt_indent(3);
1264 		dsdt_fixed_ioport(0xCF8, 8);
1265 		dsdt_unindent(3);
1266 
1267 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1268 		    "PosDecode, EntireRange,");
1269 		dsdt_line("        0x0000,             // Granularity");
1270 		dsdt_line("        0x0000,             // Range Minimum");
1271 		dsdt_line("        0x0CF7,             // Range Maximum");
1272 		dsdt_line("        0x0000,             // Translation Offset");
1273 		dsdt_line("        0x0CF8,             // Length");
1274 		dsdt_line("        ,, , TypeStatic)");
1275 
1276 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1277 		    "PosDecode, EntireRange,");
1278 		dsdt_line("        0x0000,             // Granularity");
1279 		dsdt_line("        0x0D00,             // Range Minimum");
1280 		dsdt_line("        0x%04X,             // Range Maximum",
1281 		    PCI_EMUL_IOBASE - 1);
1282 		dsdt_line("        0x0000,             // Translation Offset");
1283 		dsdt_line("        0x%04X,             // Length",
1284 		    PCI_EMUL_IOBASE - 0x0D00);
1285 		dsdt_line("        ,, , TypeStatic)");
1286 
1287 		if (bi == NULL) {
1288 			dsdt_line("    })");
1289 			goto done;
1290 		}
1291 	}
1292 	assert(bi != NULL);
1293 
1294 	/* i/o window */
1295 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1296 	    "PosDecode, EntireRange,");
1297 	dsdt_line("        0x0000,             // Granularity");
1298 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1299 	dsdt_line("        0x%04X,             // Range Maximum",
1300 	    bi->iolimit - 1);
1301 	dsdt_line("        0x0000,             // Translation Offset");
1302 	dsdt_line("        0x%04X,             // Length",
1303 	    bi->iolimit - bi->iobase);
1304 	dsdt_line("        ,, , TypeStatic)");
1305 
1306 	/* mmio window (32-bit) */
1307 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1308 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1309 	dsdt_line("        0x00000000,         // Granularity");
1310 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1311 	dsdt_line("        0x%08X,         // Range Maximum\n",
1312 	    bi->memlimit32 - 1);
1313 	dsdt_line("        0x00000000,         // Translation Offset");
1314 	dsdt_line("        0x%08X,         // Length\n",
1315 	    bi->memlimit32 - bi->membase32);
1316 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1317 
1318 	/* mmio window (64-bit) */
1319 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1320 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1321 	dsdt_line("        0x0000000000000000, // Granularity");
1322 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1323 	dsdt_line("        0x%016lX, // Range Maximum\n",
1324 	    bi->memlimit64 - 1);
1325 	dsdt_line("        0x0000000000000000, // Translation Offset");
1326 	dsdt_line("        0x%016lX, // Length\n",
1327 	    bi->memlimit64 - bi->membase64);
1328 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1329 	dsdt_line("    })");
1330 
1331 	count = pci_count_lintr(bus);
1332 	if (count != 0) {
1333 		dsdt_indent(2);
1334 		dsdt_line("Name (PPRT, Package ()");
1335 		dsdt_line("{");
1336 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1337  		dsdt_line("})");
1338 		dsdt_line("Name (APRT, Package ()");
1339 		dsdt_line("{");
1340 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1341  		dsdt_line("})");
1342 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1343 		dsdt_line("{");
1344 		dsdt_line("  If (PICM)");
1345 		dsdt_line("  {");
1346 		dsdt_line("    Return (APRT)");
1347 		dsdt_line("  }");
1348 		dsdt_line("  Else");
1349 		dsdt_line("  {");
1350 		dsdt_line("    Return (PPRT)");
1351 		dsdt_line("  }");
1352 		dsdt_line("}");
1353 		dsdt_unindent(2);
1354 	}
1355 
1356 	dsdt_indent(2);
1357 	for (slot = 0; slot < MAXSLOTS; slot++) {
1358 		si = &bi->slotinfo[slot];
1359 		for (func = 0; func < MAXFUNCS; func++) {
1360 			pi = si->si_funcs[func].fi_devi;
1361 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1362 				pi->pi_d->pe_write_dsdt(pi);
1363 		}
1364 	}
1365 	dsdt_unindent(2);
1366 done:
1367 	dsdt_line("  }");
1368 }
1369 
1370 void
1371 pci_write_dsdt(void)
1372 {
1373 	int bus;
1374 
1375 	dsdt_indent(1);
1376 	dsdt_line("Name (PICM, 0x00)");
1377 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1378 	dsdt_line("{");
1379 	dsdt_line("  Store (Arg0, PICM)");
1380 	dsdt_line("}");
1381 	dsdt_line("");
1382 	dsdt_line("Scope (_SB)");
1383 	dsdt_line("{");
1384 	for (bus = 0; bus < MAXBUSES; bus++)
1385 		pci_bus_write_dsdt(bus);
1386 	dsdt_line("}");
1387 	dsdt_unindent(1);
1388 }
1389 
1390 int
1391 pci_bus_configured(int bus)
1392 {
1393 	assert(bus >= 0 && bus < MAXBUSES);
1394 	return (pci_businfo[bus] != NULL);
1395 }
1396 
1397 int
1398 pci_msi_enabled(struct pci_devinst *pi)
1399 {
1400 	return (pi->pi_msi.enabled);
1401 }
1402 
1403 int
1404 pci_msi_maxmsgnum(struct pci_devinst *pi)
1405 {
1406 	if (pi->pi_msi.enabled)
1407 		return (pi->pi_msi.maxmsgnum);
1408 	else
1409 		return (0);
1410 }
1411 
1412 int
1413 pci_msix_enabled(struct pci_devinst *pi)
1414 {
1415 
1416 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1417 }
1418 
1419 void
1420 pci_generate_msix(struct pci_devinst *pi, int index)
1421 {
1422 	struct msix_table_entry *mte;
1423 
1424 	if (!pci_msix_enabled(pi))
1425 		return;
1426 
1427 	if (pi->pi_msix.function_mask)
1428 		return;
1429 
1430 	if (index >= pi->pi_msix.table_count)
1431 		return;
1432 
1433 	mte = &pi->pi_msix.table[index];
1434 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1435 		/* XXX Set PBA bit if interrupt is disabled */
1436 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1437 	}
1438 }
1439 
1440 void
1441 pci_generate_msi(struct pci_devinst *pi, int index)
1442 {
1443 
1444 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1445 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1446 			     pi->pi_msi.msg_data + index);
1447 	}
1448 }
1449 
1450 static bool
1451 pci_lintr_permitted(struct pci_devinst *pi)
1452 {
1453 	uint16_t cmd;
1454 
1455 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1456 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1457 		(cmd & PCIM_CMD_INTxDIS)));
1458 }
1459 
1460 void
1461 pci_lintr_request(struct pci_devinst *pi)
1462 {
1463 	struct businfo *bi;
1464 	struct slotinfo *si;
1465 	int bestpin, bestcount, pin;
1466 
1467 	bi = pci_businfo[pi->pi_bus];
1468 	assert(bi != NULL);
1469 
1470 	/*
1471 	 * Just allocate a pin from our slot.  The pin will be
1472 	 * assigned IRQs later when interrupts are routed.
1473 	 */
1474 	si = &bi->slotinfo[pi->pi_slot];
1475 	bestpin = 0;
1476 	bestcount = si->si_intpins[0].ii_count;
1477 	for (pin = 1; pin < 4; pin++) {
1478 		if (si->si_intpins[pin].ii_count < bestcount) {
1479 			bestpin = pin;
1480 			bestcount = si->si_intpins[pin].ii_count;
1481 		}
1482 	}
1483 
1484 	si->si_intpins[bestpin].ii_count++;
1485 	pi->pi_lintr.pin = bestpin + 1;
1486 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1487 }
1488 
1489 static void
1490 pci_lintr_route(struct pci_devinst *pi)
1491 {
1492 	struct businfo *bi;
1493 	struct intxinfo *ii;
1494 
1495 	if (pi->pi_lintr.pin == 0)
1496 		return;
1497 
1498 	bi = pci_businfo[pi->pi_bus];
1499 	assert(bi != NULL);
1500 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1501 
1502 	/*
1503 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1504 	 * is not yet assigned.
1505 	 */
1506 	if (ii->ii_ioapic_irq == 0)
1507 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1508 	assert(ii->ii_ioapic_irq > 0);
1509 
1510 	/*
1511 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1512 	 * not yet assigned.
1513 	 */
1514 	if (ii->ii_pirq_pin == 0)
1515 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1516 	assert(ii->ii_pirq_pin > 0);
1517 
1518 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1519 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1520 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1521 }
1522 
1523 void
1524 pci_lintr_assert(struct pci_devinst *pi)
1525 {
1526 
1527 	assert(pi->pi_lintr.pin > 0);
1528 
1529 	pthread_mutex_lock(&pi->pi_lintr.lock);
1530 	if (pi->pi_lintr.state == IDLE) {
1531 		if (pci_lintr_permitted(pi)) {
1532 			pi->pi_lintr.state = ASSERTED;
1533 			pci_irq_assert(pi);
1534 		} else
1535 			pi->pi_lintr.state = PENDING;
1536 	}
1537 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1538 }
1539 
1540 void
1541 pci_lintr_deassert(struct pci_devinst *pi)
1542 {
1543 
1544 	assert(pi->pi_lintr.pin > 0);
1545 
1546 	pthread_mutex_lock(&pi->pi_lintr.lock);
1547 	if (pi->pi_lintr.state == ASSERTED) {
1548 		pi->pi_lintr.state = IDLE;
1549 		pci_irq_deassert(pi);
1550 	} else if (pi->pi_lintr.state == PENDING)
1551 		pi->pi_lintr.state = IDLE;
1552 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1553 }
1554 
1555 static void
1556 pci_lintr_update(struct pci_devinst *pi)
1557 {
1558 
1559 	pthread_mutex_lock(&pi->pi_lintr.lock);
1560 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1561 		pci_irq_deassert(pi);
1562 		pi->pi_lintr.state = PENDING;
1563 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1564 		pi->pi_lintr.state = ASSERTED;
1565 		pci_irq_assert(pi);
1566 	}
1567 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1568 }
1569 
1570 int
1571 pci_count_lintr(int bus)
1572 {
1573 	int count, slot, pin;
1574 	struct slotinfo *slotinfo;
1575 
1576 	count = 0;
1577 	if (pci_businfo[bus] != NULL) {
1578 		for (slot = 0; slot < MAXSLOTS; slot++) {
1579 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1580 			for (pin = 0; pin < 4; pin++) {
1581 				if (slotinfo->si_intpins[pin].ii_count != 0)
1582 					count++;
1583 			}
1584 		}
1585 	}
1586 	return (count);
1587 }
1588 
1589 void
1590 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1591 {
1592 	struct businfo *bi;
1593 	struct slotinfo *si;
1594 	struct intxinfo *ii;
1595 	int slot, pin;
1596 
1597 	if ((bi = pci_businfo[bus]) == NULL)
1598 		return;
1599 
1600 	for (slot = 0; slot < MAXSLOTS; slot++) {
1601 		si = &bi->slotinfo[slot];
1602 		for (pin = 0; pin < 4; pin++) {
1603 			ii = &si->si_intpins[pin];
1604 			if (ii->ii_count != 0)
1605 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1606 				    ii->ii_ioapic_irq, arg);
1607 		}
1608 	}
1609 }
1610 
1611 /*
1612  * Return 1 if the emulated device in 'slot' is a multi-function device.
1613  * Return 0 otherwise.
1614  */
1615 static int
1616 pci_emul_is_mfdev(int bus, int slot)
1617 {
1618 	struct businfo *bi;
1619 	struct slotinfo *si;
1620 	int f, numfuncs;
1621 
1622 	numfuncs = 0;
1623 	if ((bi = pci_businfo[bus]) != NULL) {
1624 		si = &bi->slotinfo[slot];
1625 		for (f = 0; f < MAXFUNCS; f++) {
1626 			if (si->si_funcs[f].fi_devi != NULL) {
1627 				numfuncs++;
1628 			}
1629 		}
1630 	}
1631 	return (numfuncs > 1);
1632 }
1633 
1634 /*
1635  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1636  * whether or not is a multi-function being emulated in the pci 'slot'.
1637  */
1638 static void
1639 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1640 {
1641 	int mfdev;
1642 
1643 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1644 		mfdev = pci_emul_is_mfdev(bus, slot);
1645 		switch (bytes) {
1646 		case 1:
1647 		case 2:
1648 			*rv &= ~PCIM_MFDEV;
1649 			if (mfdev) {
1650 				*rv |= PCIM_MFDEV;
1651 			}
1652 			break;
1653 		case 4:
1654 			*rv &= ~(PCIM_MFDEV << 16);
1655 			if (mfdev) {
1656 				*rv |= (PCIM_MFDEV << 16);
1657 			}
1658 			break;
1659 		}
1660 	}
1661 }
1662 
1663 static void
1664 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1665 {
1666 	int i, rshift;
1667 	uint32_t cmd, cmd2, changed, old, readonly;
1668 
1669 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1670 
1671 	/*
1672 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1673 	 *
1674 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1675 	 * 'write 1 to clear'. However these bits are not set to '1' by
1676 	 * any device emulation so it is simpler to treat them as readonly.
1677 	 */
1678 	rshift = (coff & 0x3) * 8;
1679 	readonly = 0xFFFFF880 >> rshift;
1680 
1681 	old = CFGREAD(pi, coff, bytes);
1682 	new &= ~readonly;
1683 	new |= (old & readonly);
1684 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1685 
1686 	cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1687 	changed = cmd ^ cmd2;
1688 
1689 	/*
1690 	 * If the MMIO or I/O address space decoding has changed then
1691 	 * register/unregister all BARs that decode that address space.
1692 	 */
1693 	for (i = 0; i <= PCI_BARMAX; i++) {
1694 		switch (pi->pi_bar[i].type) {
1695 			case PCIBAR_NONE:
1696 			case PCIBAR_MEMHI64:
1697 				break;
1698 			case PCIBAR_IO:
1699 				/* I/O address space decoding changed? */
1700 				if (changed & PCIM_CMD_PORTEN) {
1701 					if (porten(pi))
1702 						register_bar(pi, i);
1703 					else
1704 						unregister_bar(pi, i);
1705 				}
1706 				break;
1707 			case PCIBAR_MEM32:
1708 			case PCIBAR_MEM64:
1709 				/* MMIO address space decoding changed? */
1710 				if (changed & PCIM_CMD_MEMEN) {
1711 					if (memen(pi))
1712 						register_bar(pi, i);
1713 					else
1714 						unregister_bar(pi, i);
1715 				}
1716 				break;
1717 			default:
1718 				assert(0);
1719 		}
1720 	}
1721 
1722 	/*
1723 	 * If INTx has been unmasked and is pending, assert the
1724 	 * interrupt.
1725 	 */
1726 	pci_lintr_update(pi);
1727 }
1728 
1729 static void
1730 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1731     int coff, int bytes, uint32_t *eax)
1732 {
1733 	struct businfo *bi;
1734 	struct slotinfo *si;
1735 	struct pci_devinst *pi;
1736 	struct pci_devemu *pe;
1737 	int idx, needcfg;
1738 	uint64_t addr, bar, mask;
1739 
1740 	if ((bi = pci_businfo[bus]) != NULL) {
1741 		si = &bi->slotinfo[slot];
1742 		pi = si->si_funcs[func].fi_devi;
1743 	} else
1744 		pi = NULL;
1745 
1746 	/*
1747 	 * Just return if there is no device at this slot:func or if the
1748 	 * the guest is doing an un-aligned access.
1749 	 */
1750 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1751 	    (coff & (bytes - 1)) != 0) {
1752 		if (in)
1753 			*eax = 0xffffffff;
1754 		return;
1755 	}
1756 
1757 	/*
1758 	 * Ignore all writes beyond the standard config space and return all
1759 	 * ones on reads.
1760 	 */
1761 	if (coff >= PCI_REGMAX + 1) {
1762 		if (in) {
1763 			*eax = 0xffffffff;
1764 			/*
1765 			 * Extended capabilities begin at offset 256 in config
1766 			 * space. Absence of extended capabilities is signaled
1767 			 * with all 0s in the extended capability header at
1768 			 * offset 256.
1769 			 */
1770 			if (coff <= PCI_REGMAX + 4)
1771 				*eax = 0x00000000;
1772 		}
1773 		return;
1774 	}
1775 
1776 	pe = pi->pi_d;
1777 
1778 	/*
1779 	 * Config read
1780 	 */
1781 	if (in) {
1782 		/* Let the device emulation override the default handler */
1783 		if (pe->pe_cfgread != NULL) {
1784 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1785 			    eax);
1786 		} else {
1787 			needcfg = 1;
1788 		}
1789 
1790 		if (needcfg)
1791 			*eax = CFGREAD(pi, coff, bytes);
1792 
1793 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1794 	} else {
1795 		/* Let the device emulation override the default handler */
1796 		if (pe->pe_cfgwrite != NULL &&
1797 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1798 			return;
1799 
1800 		/*
1801 		 * Special handling for write to BAR registers
1802 		 */
1803 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1804 			/*
1805 			 * Ignore writes to BAR registers that are not
1806 			 * 4-byte aligned.
1807 			 */
1808 			if (bytes != 4 || (coff & 0x3) != 0)
1809 				return;
1810 			idx = (coff - PCIR_BAR(0)) / 4;
1811 			mask = ~(pi->pi_bar[idx].size - 1);
1812 			switch (pi->pi_bar[idx].type) {
1813 			case PCIBAR_NONE:
1814 				pi->pi_bar[idx].addr = bar = 0;
1815 				break;
1816 			case PCIBAR_IO:
1817 				addr = *eax & mask;
1818 				addr &= 0xffff;
1819 				bar = addr | PCIM_BAR_IO_SPACE;
1820 				/*
1821 				 * Register the new BAR value for interception
1822 				 */
1823 				if (addr != pi->pi_bar[idx].addr) {
1824 					update_bar_address(pi, addr, idx,
1825 							   PCIBAR_IO);
1826 				}
1827 				break;
1828 			case PCIBAR_MEM32:
1829 				addr = bar = *eax & mask;
1830 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1831 				if (addr != pi->pi_bar[idx].addr) {
1832 					update_bar_address(pi, addr, idx,
1833 							   PCIBAR_MEM32);
1834 				}
1835 				break;
1836 			case PCIBAR_MEM64:
1837 				addr = bar = *eax & mask;
1838 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1839 				       PCIM_BAR_MEM_PREFETCH;
1840 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1841 					update_bar_address(pi, addr, idx,
1842 							   PCIBAR_MEM64);
1843 				}
1844 				break;
1845 			case PCIBAR_MEMHI64:
1846 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1847 				addr = ((uint64_t)*eax << 32) & mask;
1848 				bar = addr >> 32;
1849 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1850 					update_bar_address(pi, addr, idx - 1,
1851 							   PCIBAR_MEMHI64);
1852 				}
1853 				break;
1854 			default:
1855 				assert(0);
1856 			}
1857 			pci_set_cfgdata32(pi, coff, bar);
1858 
1859 		} else if (pci_emul_iscap(pi, coff)) {
1860 			pci_emul_capwrite(pi, coff, bytes, *eax);
1861 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1862 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1863 		} else {
1864 			CFGWRITE(pi, coff, *eax, bytes);
1865 		}
1866 	}
1867 }
1868 
1869 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1870 
1871 static int
1872 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1873 		 uint32_t *eax, void *arg)
1874 {
1875 	uint32_t x;
1876 
1877 	if (bytes != 4) {
1878 		if (in)
1879 			*eax = (bytes == 2) ? 0xffff : 0xff;
1880 		return (0);
1881 	}
1882 
1883 	if (in) {
1884 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1885 		if (cfgenable)
1886 			x |= CONF1_ENABLE;
1887 		*eax = x;
1888 	} else {
1889 		x = *eax;
1890 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1891 		cfgoff = x & PCI_REGMAX;
1892 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1893 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1894 		cfgbus = (x >> 16) & PCI_BUSMAX;
1895 	}
1896 
1897 	return (0);
1898 }
1899 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1900 
1901 static int
1902 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1903 		 uint32_t *eax, void *arg)
1904 {
1905 	int coff;
1906 
1907 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1908 
1909 	coff = cfgoff + (port - CONF1_DATA_PORT);
1910 	if (cfgenable) {
1911 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1912 		    eax);
1913 	} else {
1914 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1915 		if (in)
1916 			*eax = 0xffffffff;
1917 	}
1918 	return (0);
1919 }
1920 
1921 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1922 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1923 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1924 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1925 
1926 #define PCI_EMUL_TEST
1927 #ifdef PCI_EMUL_TEST
1928 /*
1929  * Define a dummy test device
1930  */
1931 #define DIOSZ	8
1932 #define DMEMSZ	4096
1933 struct pci_emul_dsoftc {
1934 	uint8_t   ioregs[DIOSZ];
1935 	uint8_t	  memregs[2][DMEMSZ];
1936 };
1937 
1938 #define	PCI_EMUL_MSI_MSGS	 4
1939 #define	PCI_EMUL_MSIX_MSGS	16
1940 
1941 static int
1942 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1943 {
1944 	int error;
1945 	struct pci_emul_dsoftc *sc;
1946 
1947 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1948 
1949 	pi->pi_arg = sc;
1950 
1951 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1952 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1953 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1954 
1955 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1956 	assert(error == 0);
1957 
1958 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1959 	assert(error == 0);
1960 
1961 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1962 	assert(error == 0);
1963 
1964 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1965 	assert(error == 0);
1966 
1967 	return (0);
1968 }
1969 
1970 static void
1971 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1972 	      uint64_t offset, int size, uint64_t value)
1973 {
1974 	int i;
1975 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1976 
1977 	if (baridx == 0) {
1978 		if (offset + size > DIOSZ) {
1979 			printf("diow: iow too large, offset %ld size %d\n",
1980 			       offset, size);
1981 			return;
1982 		}
1983 
1984 		if (size == 1) {
1985 			sc->ioregs[offset] = value & 0xff;
1986 		} else if (size == 2) {
1987 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1988 		} else if (size == 4) {
1989 			*(uint32_t *)&sc->ioregs[offset] = value;
1990 		} else {
1991 			printf("diow: iow unknown size %d\n", size);
1992 		}
1993 
1994 		/*
1995 		 * Special magic value to generate an interrupt
1996 		 */
1997 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
1998 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
1999 
2000 		if (value == 0xabcdef) {
2001 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2002 				pci_generate_msi(pi, i);
2003 		}
2004 	}
2005 
2006 	if (baridx == 1 || baridx == 2) {
2007 		if (offset + size > DMEMSZ) {
2008 			printf("diow: memw too large, offset %ld size %d\n",
2009 			       offset, size);
2010 			return;
2011 		}
2012 
2013 		i = baridx - 1;		/* 'memregs' index */
2014 
2015 		if (size == 1) {
2016 			sc->memregs[i][offset] = value;
2017 		} else if (size == 2) {
2018 			*(uint16_t *)&sc->memregs[i][offset] = value;
2019 		} else if (size == 4) {
2020 			*(uint32_t *)&sc->memregs[i][offset] = value;
2021 		} else if (size == 8) {
2022 			*(uint64_t *)&sc->memregs[i][offset] = value;
2023 		} else {
2024 			printf("diow: memw unknown size %d\n", size);
2025 		}
2026 
2027 		/*
2028 		 * magic interrupt ??
2029 		 */
2030 	}
2031 
2032 	if (baridx > 2 || baridx < 0) {
2033 		printf("diow: unknown bar idx %d\n", baridx);
2034 	}
2035 }
2036 
2037 static uint64_t
2038 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2039 	      uint64_t offset, int size)
2040 {
2041 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2042 	uint32_t value;
2043 	int i;
2044 
2045 	if (baridx == 0) {
2046 		if (offset + size > DIOSZ) {
2047 			printf("dior: ior too large, offset %ld size %d\n",
2048 			       offset, size);
2049 			return (0);
2050 		}
2051 
2052 		value = 0;
2053 		if (size == 1) {
2054 			value = sc->ioregs[offset];
2055 		} else if (size == 2) {
2056 			value = *(uint16_t *) &sc->ioregs[offset];
2057 		} else if (size == 4) {
2058 			value = *(uint32_t *) &sc->ioregs[offset];
2059 		} else {
2060 			printf("dior: ior unknown size %d\n", size);
2061 		}
2062 	}
2063 
2064 	if (baridx == 1 || baridx == 2) {
2065 		if (offset + size > DMEMSZ) {
2066 			printf("dior: memr too large, offset %ld size %d\n",
2067 			       offset, size);
2068 			return (0);
2069 		}
2070 
2071 		i = baridx - 1;		/* 'memregs' index */
2072 
2073 		if (size == 1) {
2074 			value = sc->memregs[i][offset];
2075 		} else if (size == 2) {
2076 			value = *(uint16_t *) &sc->memregs[i][offset];
2077 		} else if (size == 4) {
2078 			value = *(uint32_t *) &sc->memregs[i][offset];
2079 		} else if (size == 8) {
2080 			value = *(uint64_t *) &sc->memregs[i][offset];
2081 		} else {
2082 			printf("dior: ior unknown size %d\n", size);
2083 		}
2084 	}
2085 
2086 
2087 	if (baridx > 2 || baridx < 0) {
2088 		printf("dior: unknown bar idx %d\n", baridx);
2089 		return (0);
2090 	}
2091 
2092 	return (value);
2093 }
2094 
2095 struct pci_devemu pci_dummy = {
2096 	.pe_emu = "dummy",
2097 	.pe_init = pci_emul_dinit,
2098 	.pe_barwrite = pci_emul_diow,
2099 	.pe_barread = pci_emul_dior
2100 };
2101 PCI_EMUL_SET(pci_dummy);
2102 
2103 #endif /* PCI_EMUL_TEST */
2104