xref: /illumos-gate/usr/src/cmd/bhyve/pci_emul.c (revision 4c87aefe8930bd07275b8dd2e96ea5f24d93a52e)
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  * This file and its contents are supplied under the terms of the
32  * Common Development and Distribution License ("CDDL"), version 1.0.
33  * You may only use this file in accordance with the terms of version
34  * 1.0 of the CDDL.
35  *
36  * A full copy of the text of the CDDL should have accompanied this
37  * source.  A copy of the CDDL is also available via the Internet at
38  * http://www.illumos.org/license/CDDL.
39  *
40  * Copyright 2014 Pluribus Networks Inc.
41  * Copyright 2018 Joyent, Inc.
42  */
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include <sys/param.h>
48 #include <sys/linker_set.h>
49 
50 #include <ctype.h>
51 #include <errno.h>
52 #include <pthread.h>
53 #include <stdio.h>
54 #include <stdlib.h>
55 #include <string.h>
56 #include <strings.h>
57 #include <assert.h>
58 #include <stdbool.h>
59 
60 #include <machine/vmm.h>
61 #include <vmmapi.h>
62 
63 #include "acpi.h"
64 #include "bhyverun.h"
65 #include "inout.h"
66 #include "ioapic.h"
67 #include "mem.h"
68 #include "pci_emul.h"
69 #include "pci_irq.h"
70 #include "pci_lpc.h"
71 
72 #define CONF1_ADDR_PORT	   0x0cf8
73 #define CONF1_DATA_PORT	   0x0cfc
74 
75 #define CONF1_ENABLE	   0x80000000ul
76 
77 #define	MAXBUSES	(PCI_BUSMAX + 1)
78 #define MAXSLOTS	(PCI_SLOTMAX + 1)
79 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
80 
81 struct funcinfo {
82 	char	*fi_name;
83 	char	*fi_param;
84 	struct pci_devinst *fi_devi;
85 };
86 
87 struct intxinfo {
88 	int	ii_count;
89 	int	ii_pirq_pin;
90 	int	ii_ioapic_irq;
91 };
92 
93 struct slotinfo {
94 	struct intxinfo si_intpins[4];
95 	struct funcinfo si_funcs[MAXFUNCS];
96 };
97 
98 struct businfo {
99 	uint16_t iobase, iolimit;		/* I/O window */
100 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
101 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
102 	struct slotinfo slotinfo[MAXSLOTS];
103 };
104 
105 static struct businfo *pci_businfo[MAXBUSES];
106 
107 SET_DECLARE(pci_devemu_set, struct pci_devemu);
108 
109 static uint64_t pci_emul_iobase;
110 static uint64_t pci_emul_membase32;
111 static uint64_t pci_emul_membase64;
112 
113 #define	PCI_EMUL_IOBASE		0x2000
114 #define	PCI_EMUL_IOLIMIT	0x10000
115 
116 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
117 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
118 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
119 
120 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
121 
122 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
123 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
124 
125 static struct pci_devemu *pci_emul_finddev(char *name);
126 static void pci_lintr_route(struct pci_devinst *pi);
127 static void pci_lintr_update(struct pci_devinst *pi);
128 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
129     int func, int coff, int bytes, uint32_t *val);
130 
131 static __inline void
132 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
133 {
134 
135 	if (bytes == 1)
136 		pci_set_cfgdata8(pi, coff, val);
137 	else if (bytes == 2)
138 		pci_set_cfgdata16(pi, coff, val);
139 	else
140 		pci_set_cfgdata32(pi, coff, val);
141 }
142 
143 static __inline uint32_t
144 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
145 {
146 
147 	if (bytes == 1)
148 		return (pci_get_cfgdata8(pi, coff));
149 	else if (bytes == 2)
150 		return (pci_get_cfgdata16(pi, coff));
151 	else
152 		return (pci_get_cfgdata32(pi, coff));
153 }
154 
155 /*
156  * I/O access
157  */
158 
159 /*
160  * Slot options are in the form:
161  *
162  *  <bus>:<slot>:<func>,<emul>[,<config>]
163  *  <slot>[:<func>],<emul>[,<config>]
164  *
165  *  slot is 0..31
166  *  func is 0..7
167  *  emul is a string describing the type of PCI device e.g. virtio-net
168  *  config is an optional string, depending on the device, that can be
169  *  used for configuration.
170  *   Examples are:
171  *     1,virtio-net,tap0
172  *     3:0,dummy
173  */
174 static void
175 pci_parse_slot_usage(char *aopt)
176 {
177 
178 	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
179 }
180 
181 int
182 pci_parse_slot(char *opt)
183 {
184 	struct businfo *bi;
185 	struct slotinfo *si;
186 	char *emul, *config, *str, *cp;
187 	int error, bnum, snum, fnum;
188 
189 	error = -1;
190 	str = strdup(opt);
191 
192 	emul = config = NULL;
193 	if ((cp = strchr(str, ',')) != NULL) {
194 		*cp = '\0';
195 		emul = cp + 1;
196 		if ((cp = strchr(emul, ',')) != NULL) {
197 			*cp = '\0';
198 			config = cp + 1;
199 		}
200 	} else {
201 		pci_parse_slot_usage(opt);
202 		goto done;
203 	}
204 
205 	/* <bus>:<slot>:<func> */
206 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
207 		bnum = 0;
208 		/* <slot>:<func> */
209 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
210 			fnum = 0;
211 			/* <slot> */
212 			if (sscanf(str, "%d", &snum) != 1) {
213 				snum = -1;
214 			}
215 		}
216 	}
217 
218 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
219 	    fnum < 0 || fnum >= MAXFUNCS) {
220 		pci_parse_slot_usage(opt);
221 		goto done;
222 	}
223 
224 	if (pci_businfo[bnum] == NULL)
225 		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
226 
227 	bi = pci_businfo[bnum];
228 	si = &bi->slotinfo[snum];
229 
230 	if (si->si_funcs[fnum].fi_name != NULL) {
231 		fprintf(stderr, "pci slot %d:%d already occupied!\n",
232 			snum, fnum);
233 		goto done;
234 	}
235 
236 	if (pci_emul_finddev(emul) == NULL) {
237 		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
238 			snum, fnum, emul);
239 		goto done;
240 	}
241 
242 	error = 0;
243 	si->si_funcs[fnum].fi_name = emul;
244 	si->si_funcs[fnum].fi_param = config;
245 
246 done:
247 	if (error)
248 		free(str);
249 
250 	return (error);
251 }
252 
253 void
254 pci_print_supported_devices()
255 {
256 	struct pci_devemu **pdpp, *pdp;
257 
258 	SET_FOREACH(pdpp, pci_devemu_set) {
259 		pdp = *pdpp;
260 		printf("%s\n", pdp->pe_emu);
261 	}
262 }
263 
264 static int
265 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
266 {
267 
268 	if (offset < pi->pi_msix.pba_offset)
269 		return (0);
270 
271 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
272 		return (0);
273 	}
274 
275 	return (1);
276 }
277 
278 int
279 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
280 		     uint64_t value)
281 {
282 	int msix_entry_offset;
283 	int tab_index;
284 	char *dest;
285 
286 	/* support only 4 or 8 byte writes */
287 	if (size != 4 && size != 8)
288 		return (-1);
289 
290 	/*
291 	 * Return if table index is beyond what device supports
292 	 */
293 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
294 	if (tab_index >= pi->pi_msix.table_count)
295 		return (-1);
296 
297 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
298 
299 	/* support only aligned writes */
300 	if ((msix_entry_offset % size) != 0)
301 		return (-1);
302 
303 	dest = (char *)(pi->pi_msix.table + tab_index);
304 	dest += msix_entry_offset;
305 
306 	if (size == 4)
307 		*((uint32_t *)dest) = value;
308 	else
309 		*((uint64_t *)dest) = value;
310 
311 	return (0);
312 }
313 
314 uint64_t
315 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
316 {
317 	char *dest;
318 	int msix_entry_offset;
319 	int tab_index;
320 	uint64_t retval = ~0;
321 
322 	/*
323 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
324 	 * table but we also allow 1 byte access to accommodate reads from
325 	 * ddb.
326 	 */
327 	if (size != 1 && size != 4 && size != 8)
328 		return (retval);
329 
330 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
331 
332 	/* support only aligned reads */
333 	if ((msix_entry_offset % size) != 0) {
334 		return (retval);
335 	}
336 
337 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
338 
339 	if (tab_index < pi->pi_msix.table_count) {
340 		/* valid MSI-X Table access */
341 		dest = (char *)(pi->pi_msix.table + tab_index);
342 		dest += msix_entry_offset;
343 
344 		if (size == 1)
345 			retval = *((uint8_t *)dest);
346 		else if (size == 4)
347 			retval = *((uint32_t *)dest);
348 		else
349 			retval = *((uint64_t *)dest);
350 	} else if (pci_valid_pba_offset(pi, offset)) {
351 		/* return 0 for PBA access */
352 		retval = 0;
353 	}
354 
355 	return (retval);
356 }
357 
358 int
359 pci_msix_table_bar(struct pci_devinst *pi)
360 {
361 
362 	if (pi->pi_msix.table != NULL)
363 		return (pi->pi_msix.table_bar);
364 	else
365 		return (-1);
366 }
367 
368 int
369 pci_msix_pba_bar(struct pci_devinst *pi)
370 {
371 
372 	if (pi->pi_msix.table != NULL)
373 		return (pi->pi_msix.pba_bar);
374 	else
375 		return (-1);
376 }
377 
378 static int
379 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
380 		    uint32_t *eax, void *arg)
381 {
382 	struct pci_devinst *pdi = arg;
383 	struct pci_devemu *pe = pdi->pi_d;
384 	uint64_t offset;
385 	int i;
386 
387 	for (i = 0; i <= PCI_BARMAX; i++) {
388 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
389 		    port >= pdi->pi_bar[i].addr &&
390 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
391 			offset = port - pdi->pi_bar[i].addr;
392 			if (in)
393 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
394 							 offset, bytes);
395 			else
396 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
397 						   bytes, *eax);
398 			return (0);
399 		}
400 	}
401 	return (-1);
402 }
403 
404 static int
405 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
406 		     int size, uint64_t *val, void *arg1, long arg2)
407 {
408 	struct pci_devinst *pdi = arg1;
409 	struct pci_devemu *pe = pdi->pi_d;
410 	uint64_t offset;
411 	int bidx = (int) arg2;
412 
413 	assert(bidx <= PCI_BARMAX);
414 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
415 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
416 	assert(addr >= pdi->pi_bar[bidx].addr &&
417 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
418 
419 	offset = addr - pdi->pi_bar[bidx].addr;
420 
421 	if (dir == MEM_F_WRITE) {
422 		if (size == 8) {
423 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
424 					   4, *val & 0xffffffff);
425 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
426 					   4, *val >> 32);
427 		} else {
428 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
429 					   size, *val);
430 		}
431 	} else {
432 		if (size == 8) {
433 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
434 						 offset, 4);
435 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
436 						  offset + 4, 4) << 32;
437 		} else {
438 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
439 						 offset, size);
440 		}
441 	}
442 
443 	return (0);
444 }
445 
446 
447 static int
448 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
449 			uint64_t *addr)
450 {
451 	uint64_t base;
452 
453 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
454 
455 	base = roundup2(*baseptr, size);
456 
457 	if (base + size <= limit) {
458 		*addr = base;
459 		*baseptr = base + size;
460 		return (0);
461 	} else
462 		return (-1);
463 }
464 
465 int
466 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
467 		   uint64_t size)
468 {
469 
470 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
471 }
472 
473 /*
474  * Register (or unregister) the MMIO or I/O region associated with the BAR
475  * register 'idx' of an emulated pci device.
476  */
477 static void
478 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
479 {
480 	int error;
481 	struct inout_port iop;
482 	struct mem_range mr;
483 
484 	switch (pi->pi_bar[idx].type) {
485 	case PCIBAR_IO:
486 		bzero(&iop, sizeof(struct inout_port));
487 		iop.name = pi->pi_name;
488 		iop.port = pi->pi_bar[idx].addr;
489 		iop.size = pi->pi_bar[idx].size;
490 		if (registration) {
491 			iop.flags = IOPORT_F_INOUT;
492 			iop.handler = pci_emul_io_handler;
493 			iop.arg = pi;
494 			error = register_inout(&iop);
495 		} else
496 			error = unregister_inout(&iop);
497 		break;
498 	case PCIBAR_MEM32:
499 	case PCIBAR_MEM64:
500 		bzero(&mr, sizeof(struct mem_range));
501 		mr.name = pi->pi_name;
502 		mr.base = pi->pi_bar[idx].addr;
503 		mr.size = pi->pi_bar[idx].size;
504 		if (registration) {
505 			mr.flags = MEM_F_RW;
506 			mr.handler = pci_emul_mem_handler;
507 			mr.arg1 = pi;
508 			mr.arg2 = idx;
509 			error = register_mem(&mr);
510 		} else
511 			error = unregister_mem(&mr);
512 		break;
513 	default:
514 		error = EINVAL;
515 		break;
516 	}
517 	assert(error == 0);
518 }
519 
520 static void
521 unregister_bar(struct pci_devinst *pi, int idx)
522 {
523 
524 	modify_bar_registration(pi, idx, 0);
525 }
526 
527 static void
528 register_bar(struct pci_devinst *pi, int idx)
529 {
530 
531 	modify_bar_registration(pi, idx, 1);
532 }
533 
534 /* Are we decoding i/o port accesses for the emulated pci device? */
535 static int
536 porten(struct pci_devinst *pi)
537 {
538 	uint16_t cmd;
539 
540 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
541 
542 	return (cmd & PCIM_CMD_PORTEN);
543 }
544 
545 /* Are we decoding memory accesses for the emulated pci device? */
546 static int
547 memen(struct pci_devinst *pi)
548 {
549 	uint16_t cmd;
550 
551 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
552 
553 	return (cmd & PCIM_CMD_MEMEN);
554 }
555 
556 /*
557  * Update the MMIO or I/O address that is decoded by the BAR register.
558  *
559  * If the pci device has enabled the address space decoding then intercept
560  * the address range decoded by the BAR register.
561  */
562 static void
563 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
564 {
565 	int decode;
566 
567 	if (pi->pi_bar[idx].type == PCIBAR_IO)
568 		decode = porten(pi);
569 	else
570 		decode = memen(pi);
571 
572 	if (decode)
573 		unregister_bar(pi, idx);
574 
575 	switch (type) {
576 	case PCIBAR_IO:
577 	case PCIBAR_MEM32:
578 		pi->pi_bar[idx].addr = addr;
579 		break;
580 	case PCIBAR_MEM64:
581 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
582 		pi->pi_bar[idx].addr |= addr;
583 		break;
584 	case PCIBAR_MEMHI64:
585 		pi->pi_bar[idx].addr &= 0xffffffff;
586 		pi->pi_bar[idx].addr |= addr;
587 		break;
588 	default:
589 		assert(0);
590 	}
591 
592 	if (decode)
593 		register_bar(pi, idx);
594 }
595 
596 int
597 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
598 		    enum pcibar_type type, uint64_t size)
599 {
600 	uint64_t *baseptr = NULL;
601 	uint64_t limit = 0, lobits = 0;
602 	uint64_t addr, mask, bar;
603 	int error;
604 
605 	assert(idx >= 0 && idx <= PCI_BARMAX);
606 
607 	if ((size & (size - 1)) != 0)
608 		size = 1UL << flsl(size);	/* round up to a power of 2 */
609 
610 	/* Enforce minimum BAR sizes required by the PCI standard */
611 	if (type == PCIBAR_IO) {
612 		if (size < 4)
613 			size = 4;
614 	} else {
615 		if (size < 16)
616 			size = 16;
617 	}
618 
619 	switch (type) {
620 	case PCIBAR_NONE:
621 		baseptr = NULL;
622 		addr = mask = lobits = 0;
623 		break;
624 	case PCIBAR_IO:
625 		baseptr = &pci_emul_iobase;
626 		limit = PCI_EMUL_IOLIMIT;
627 		mask = PCIM_BAR_IO_BASE;
628 		lobits = PCIM_BAR_IO_SPACE;
629 		break;
630 	case PCIBAR_MEM64:
631 		/*
632 		 * XXX
633 		 * Some drivers do not work well if the 64-bit BAR is allocated
634 		 * above 4GB. Allow for this by allocating small requests under
635 		 * 4GB unless then allocation size is larger than some arbitrary
636 		 * number (32MB currently).
637 		 */
638 		if (size > 32 * 1024 * 1024) {
639 			/*
640 			 * XXX special case for device requiring peer-peer DMA
641 			 */
642 			if (size == 0x100000000UL)
643 				baseptr = &hostbase;
644 			else
645 				baseptr = &pci_emul_membase64;
646 			limit = PCI_EMUL_MEMLIMIT64;
647 			mask = PCIM_BAR_MEM_BASE;
648 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
649 				 PCIM_BAR_MEM_PREFETCH;
650 			break;
651 		} else {
652 			baseptr = &pci_emul_membase32;
653 			limit = PCI_EMUL_MEMLIMIT32;
654 			mask = PCIM_BAR_MEM_BASE;
655 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
656 		}
657 		break;
658 	case PCIBAR_MEM32:
659 		baseptr = &pci_emul_membase32;
660 		limit = PCI_EMUL_MEMLIMIT32;
661 		mask = PCIM_BAR_MEM_BASE;
662 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
663 		break;
664 	default:
665 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
666 #ifdef FreeBSD
667 		assert(0);
668 #else
669 		abort();
670 #endif
671 	}
672 
673 	if (baseptr != NULL) {
674 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
675 		if (error != 0)
676 			return (error);
677 	}
678 
679 	pdi->pi_bar[idx].type = type;
680 	pdi->pi_bar[idx].addr = addr;
681 	pdi->pi_bar[idx].size = size;
682 
683 	/* Initialize the BAR register in config space */
684 	bar = (addr & mask) | lobits;
685 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
686 
687 	if (type == PCIBAR_MEM64) {
688 		assert(idx + 1 <= PCI_BARMAX);
689 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
690 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
691 	}
692 
693 	register_bar(pdi, idx);
694 
695 	return (0);
696 }
697 
698 #define	CAP_START_OFFSET	0x40
699 static int
700 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
701 {
702 	int i, capoff, reallen;
703 	uint16_t sts;
704 
705 	assert(caplen > 0);
706 
707 	reallen = roundup2(caplen, 4);		/* dword aligned */
708 
709 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
710 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
711 		capoff = CAP_START_OFFSET;
712 	else
713 		capoff = pi->pi_capend + 1;
714 
715 	/* Check if we have enough space */
716 	if (capoff + reallen > PCI_REGMAX + 1)
717 		return (-1);
718 
719 	/* Set the previous capability pointer */
720 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
721 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
722 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
723 	} else
724 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
725 
726 	/* Copy the capability */
727 	for (i = 0; i < caplen; i++)
728 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
729 
730 	/* Set the next capability pointer */
731 	pci_set_cfgdata8(pi, capoff + 1, 0);
732 
733 	pi->pi_prevcap = capoff;
734 	pi->pi_capend = capoff + reallen - 1;
735 	return (0);
736 }
737 
738 static struct pci_devemu *
739 pci_emul_finddev(char *name)
740 {
741 	struct pci_devemu **pdpp, *pdp;
742 
743 	SET_FOREACH(pdpp, pci_devemu_set) {
744 		pdp = *pdpp;
745 		if (!strcmp(pdp->pe_emu, name)) {
746 			return (pdp);
747 		}
748 	}
749 
750 	return (NULL);
751 }
752 
753 static int
754 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
755     int func, struct funcinfo *fi)
756 {
757 	struct pci_devinst *pdi;
758 	int err;
759 
760 	pdi = calloc(1, sizeof(struct pci_devinst));
761 
762 	pdi->pi_vmctx = ctx;
763 	pdi->pi_bus = bus;
764 	pdi->pi_slot = slot;
765 	pdi->pi_func = func;
766 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
767 	pdi->pi_lintr.pin = 0;
768 	pdi->pi_lintr.state = IDLE;
769 	pdi->pi_lintr.pirq_pin = 0;
770 	pdi->pi_lintr.ioapic_irq = 0;
771 	pdi->pi_d = pde;
772 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
773 
774 	/* Disable legacy interrupts */
775 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
776 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
777 
778 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
779 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
780 
781 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
782 	if (err == 0)
783 		fi->fi_devi = pdi;
784 	else
785 		free(pdi);
786 
787 	return (err);
788 }
789 
790 void
791 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
792 {
793 	int mmc;
794 
795 	/* Number of msi messages must be a power of 2 between 1 and 32 */
796 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
797 	mmc = ffs(msgnum) - 1;
798 
799 	bzero(msicap, sizeof(struct msicap));
800 	msicap->capid = PCIY_MSI;
801 	msicap->nextptr = nextptr;
802 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
803 }
804 
805 int
806 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
807 {
808 	struct msicap msicap;
809 
810 	pci_populate_msicap(&msicap, msgnum, 0);
811 
812 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
813 }
814 
815 static void
816 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
817 		     uint32_t msix_tab_size)
818 {
819 
820 	assert(msix_tab_size % 4096 == 0);
821 
822 	bzero(msixcap, sizeof(struct msixcap));
823 	msixcap->capid = PCIY_MSIX;
824 
825 	/*
826 	 * Message Control Register, all fields set to
827 	 * zero except for the Table Size.
828 	 * Note: Table size N is encoded as N-1
829 	 */
830 	msixcap->msgctrl = msgnum - 1;
831 
832 	/*
833 	 * MSI-X BAR setup:
834 	 * - MSI-X table start at offset 0
835 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
836 	 */
837 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
838 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
839 }
840 
841 static void
842 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
843 {
844 	int i, table_size;
845 
846 	assert(table_entries > 0);
847 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
848 
849 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
850 	pi->pi_msix.table = calloc(1, table_size);
851 
852 	/* set mask bit of vector control register */
853 	for (i = 0; i < table_entries; i++)
854 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
855 }
856 
857 int
858 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
859 {
860 	uint32_t tab_size;
861 	struct msixcap msixcap;
862 
863 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
864 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
865 
866 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
867 
868 	/* Align table size to nearest 4K */
869 	tab_size = roundup2(tab_size, 4096);
870 
871 	pi->pi_msix.table_bar = barnum;
872 	pi->pi_msix.pba_bar   = barnum;
873 	pi->pi_msix.table_offset = 0;
874 	pi->pi_msix.table_count = msgnum;
875 	pi->pi_msix.pba_offset = tab_size;
876 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
877 
878 	pci_msix_table_init(pi, msgnum);
879 
880 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
881 
882 	/* allocate memory for MSI-X Table and PBA */
883 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
884 				tab_size + pi->pi_msix.pba_size);
885 
886 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
887 					sizeof(msixcap)));
888 }
889 
890 void
891 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
892 		 int bytes, uint32_t val)
893 {
894 	uint16_t msgctrl, rwmask;
895 	int off;
896 
897 	off = offset - capoff;
898 	/* Message Control Register */
899 	if (off == 2 && bytes == 2) {
900 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
901 		msgctrl = pci_get_cfgdata16(pi, offset);
902 		msgctrl &= ~rwmask;
903 		msgctrl |= val & rwmask;
904 		val = msgctrl;
905 
906 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
907 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
908 		pci_lintr_update(pi);
909 	}
910 
911 	CFGWRITE(pi, offset, val, bytes);
912 }
913 
914 void
915 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
916 		int bytes, uint32_t val)
917 {
918 	uint16_t msgctrl, rwmask, msgdata, mme;
919 	uint32_t addrlo;
920 
921 	/*
922 	 * If guest is writing to the message control register make sure
923 	 * we do not overwrite read-only fields.
924 	 */
925 	if ((offset - capoff) == 2 && bytes == 2) {
926 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
927 		msgctrl = pci_get_cfgdata16(pi, offset);
928 		msgctrl &= ~rwmask;
929 		msgctrl |= val & rwmask;
930 		val = msgctrl;
931 
932 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
933 		if (msgctrl & PCIM_MSICTRL_64BIT)
934 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
935 		else
936 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
937 
938 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
939 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
940 		if (pi->pi_msi.enabled) {
941 			pi->pi_msi.addr = addrlo;
942 			pi->pi_msi.msg_data = msgdata;
943 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
944 		} else {
945 			pi->pi_msi.maxmsgnum = 0;
946 		}
947 		pci_lintr_update(pi);
948 	}
949 
950 	CFGWRITE(pi, offset, val, bytes);
951 }
952 
953 void
954 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
955 		 int bytes, uint32_t val)
956 {
957 
958 	/* XXX don't write to the readonly parts */
959 	CFGWRITE(pi, offset, val, bytes);
960 }
961 
962 #define	PCIECAP_VERSION	0x2
963 int
964 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
965 {
966 	int err;
967 	struct pciecap pciecap;
968 
969 	if (type != PCIEM_TYPE_ROOT_PORT)
970 		return (-1);
971 
972 	bzero(&pciecap, sizeof(pciecap));
973 
974 	pciecap.capid = PCIY_EXPRESS;
975 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
976 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
977 	pciecap.link_status = 0x11;		/* gen1, x1 */
978 
979 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
980 	return (err);
981 }
982 
983 /*
984  * This function assumes that 'coff' is in the capabilities region of the
985  * config space.
986  */
987 static void
988 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
989 {
990 	int capid;
991 	uint8_t capoff, nextoff;
992 
993 	/* Do not allow un-aligned writes */
994 	if ((offset & (bytes - 1)) != 0)
995 		return;
996 
997 	/* Find the capability that we want to update */
998 	capoff = CAP_START_OFFSET;
999 	while (1) {
1000 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
1001 		if (nextoff == 0)
1002 			break;
1003 		if (offset >= capoff && offset < nextoff)
1004 			break;
1005 
1006 		capoff = nextoff;
1007 	}
1008 	assert(offset >= capoff);
1009 
1010 	/*
1011 	 * Capability ID and Next Capability Pointer are readonly.
1012 	 * However, some o/s's do 4-byte writes that include these.
1013 	 * For this case, trim the write back to 2 bytes and adjust
1014 	 * the data.
1015 	 */
1016 	if (offset == capoff || offset == capoff + 1) {
1017 		if (offset == capoff && bytes == 4) {
1018 			bytes = 2;
1019 			offset += 2;
1020 			val >>= 16;
1021 		} else
1022 			return;
1023 	}
1024 
1025 	capid = pci_get_cfgdata8(pi, capoff);
1026 	switch (capid) {
1027 	case PCIY_MSI:
1028 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1029 		break;
1030 	case PCIY_MSIX:
1031 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1032 		break;
1033 	case PCIY_EXPRESS:
1034 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1035 		break;
1036 	default:
1037 		break;
1038 	}
1039 }
1040 
1041 static int
1042 pci_emul_iscap(struct pci_devinst *pi, int offset)
1043 {
1044 	uint16_t sts;
1045 
1046 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1047 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1048 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1049 			return (1);
1050 	}
1051 	return (0);
1052 }
1053 
1054 static int
1055 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1056 			  int size, uint64_t *val, void *arg1, long arg2)
1057 {
1058 	/*
1059 	 * Ignore writes; return 0xff's for reads. The mem read code
1060 	 * will take care of truncating to the correct size.
1061 	 */
1062 	if (dir == MEM_F_READ) {
1063 		*val = 0xffffffffffffffff;
1064 	}
1065 
1066 	return (0);
1067 }
1068 
1069 static int
1070 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1071     int bytes, uint64_t *val, void *arg1, long arg2)
1072 {
1073 	int bus, slot, func, coff, in;
1074 
1075 	coff = addr & 0xfff;
1076 	func = (addr >> 12) & 0x7;
1077 	slot = (addr >> 15) & 0x1f;
1078 	bus = (addr >> 20) & 0xff;
1079 	in = (dir == MEM_F_READ);
1080 	if (in)
1081 		*val = ~0UL;
1082 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1083 	return (0);
1084 }
1085 
1086 uint64_t
1087 pci_ecfg_base(void)
1088 {
1089 
1090 	return (PCI_EMUL_ECFG_BASE);
1091 }
1092 
1093 #define	BUSIO_ROUNDUP		32
1094 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1095 
1096 int
1097 init_pci(struct vmctx *ctx)
1098 {
1099 	struct mem_range mr;
1100 	struct pci_devemu *pde;
1101 	struct businfo *bi;
1102 	struct slotinfo *si;
1103 	struct funcinfo *fi;
1104 	size_t lowmem;
1105 	int bus, slot, func;
1106 	int error;
1107 
1108 	pci_emul_iobase = PCI_EMUL_IOBASE;
1109 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1110 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1111 
1112 	for (bus = 0; bus < MAXBUSES; bus++) {
1113 		if ((bi = pci_businfo[bus]) == NULL)
1114 			continue;
1115 		/*
1116 		 * Keep track of the i/o and memory resources allocated to
1117 		 * this bus.
1118 		 */
1119 		bi->iobase = pci_emul_iobase;
1120 		bi->membase32 = pci_emul_membase32;
1121 		bi->membase64 = pci_emul_membase64;
1122 
1123 		for (slot = 0; slot < MAXSLOTS; slot++) {
1124 			si = &bi->slotinfo[slot];
1125 			for (func = 0; func < MAXFUNCS; func++) {
1126 				fi = &si->si_funcs[func];
1127 				if (fi->fi_name == NULL)
1128 					continue;
1129 				pde = pci_emul_finddev(fi->fi_name);
1130 				assert(pde != NULL);
1131 				error = pci_emul_init(ctx, pde, bus, slot,
1132 				    func, fi);
1133 				if (error)
1134 					return (error);
1135 			}
1136 		}
1137 
1138 		/*
1139 		 * Add some slop to the I/O and memory resources decoded by
1140 		 * this bus to give a guest some flexibility if it wants to
1141 		 * reprogram the BARs.
1142 		 */
1143 		pci_emul_iobase += BUSIO_ROUNDUP;
1144 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1145 		bi->iolimit = pci_emul_iobase;
1146 
1147 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1148 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1149 		    BUSMEM_ROUNDUP);
1150 		bi->memlimit32 = pci_emul_membase32;
1151 
1152 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1153 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1154 		    BUSMEM_ROUNDUP);
1155 		bi->memlimit64 = pci_emul_membase64;
1156 	}
1157 
1158 	/*
1159 	 * PCI backends are initialized before routing INTx interrupts
1160 	 * so that LPC devices are able to reserve ISA IRQs before
1161 	 * routing PIRQ pins.
1162 	 */
1163 	for (bus = 0; bus < MAXBUSES; bus++) {
1164 		if ((bi = pci_businfo[bus]) == NULL)
1165 			continue;
1166 
1167 		for (slot = 0; slot < MAXSLOTS; slot++) {
1168 			si = &bi->slotinfo[slot];
1169 			for (func = 0; func < MAXFUNCS; func++) {
1170 				fi = &si->si_funcs[func];
1171 				if (fi->fi_devi == NULL)
1172 					continue;
1173 				pci_lintr_route(fi->fi_devi);
1174 			}
1175 		}
1176 	}
1177 	lpc_pirq_routed();
1178 
1179 	/*
1180 	 * The guest physical memory map looks like the following:
1181 	 * [0,		    lowmem)		guest system memory
1182 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1183 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1184 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1185 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1186 	 * [4GB,	    4GB + highmem)
1187 	 */
1188 
1189 	/*
1190 	 * Accesses to memory addresses that are not allocated to system
1191 	 * memory or PCI devices return 0xff's.
1192 	 */
1193 	lowmem = vm_get_lowmem_size(ctx);
1194 	bzero(&mr, sizeof(struct mem_range));
1195 	mr.name = "PCI hole";
1196 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1197 	mr.base = lowmem;
1198 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1199 	mr.handler = pci_emul_fallback_handler;
1200 	error = register_mem_fallback(&mr);
1201 	assert(error == 0);
1202 
1203 	/* PCI extended config space */
1204 	bzero(&mr, sizeof(struct mem_range));
1205 	mr.name = "PCI ECFG";
1206 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1207 	mr.base = PCI_EMUL_ECFG_BASE;
1208 	mr.size = PCI_EMUL_ECFG_SIZE;
1209 	mr.handler = pci_emul_ecfg_handler;
1210 	error = register_mem(&mr);
1211 	assert(error == 0);
1212 
1213 	return (0);
1214 }
1215 
1216 static void
1217 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1218     void *arg)
1219 {
1220 
1221 	dsdt_line("  Package ()");
1222 	dsdt_line("  {");
1223 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1224 	dsdt_line("    0x%02X,", pin - 1);
1225 	dsdt_line("    Zero,");
1226 	dsdt_line("    0x%X", ioapic_irq);
1227 	dsdt_line("  },");
1228 }
1229 
1230 static void
1231 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1232     void *arg)
1233 {
1234 	char *name;
1235 
1236 	name = lpc_pirq_name(pirq_pin);
1237 	if (name == NULL)
1238 		return;
1239 	dsdt_line("  Package ()");
1240 	dsdt_line("  {");
1241 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1242 	dsdt_line("    0x%02X,", pin - 1);
1243 	dsdt_line("    %s,", name);
1244 	dsdt_line("    0x00");
1245 	dsdt_line("  },");
1246 	free(name);
1247 }
1248 
1249 /*
1250  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1251  * corresponding to each PCI bus.
1252  */
1253 static void
1254 pci_bus_write_dsdt(int bus)
1255 {
1256 	struct businfo *bi;
1257 	struct slotinfo *si;
1258 	struct pci_devinst *pi;
1259 	int count, func, slot;
1260 
1261 	/*
1262 	 * If there are no devices on this 'bus' then just return.
1263 	 */
1264 	if ((bi = pci_businfo[bus]) == NULL) {
1265 		/*
1266 		 * Bus 0 is special because it decodes the I/O ports used
1267 		 * for PCI config space access even if there are no devices
1268 		 * on it.
1269 		 */
1270 		if (bus != 0)
1271 			return;
1272 	}
1273 
1274 	dsdt_line("  Device (PC%02X)", bus);
1275 	dsdt_line("  {");
1276 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1277 	dsdt_line("    Name (_ADR, Zero)");
1278 
1279 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1280 	dsdt_line("    {");
1281 	dsdt_line("        Return (0x%08X)", bus);
1282 	dsdt_line("    }");
1283 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1284 	dsdt_line("    {");
1285 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1286 	    "MaxFixed, PosDecode,");
1287 	dsdt_line("        0x0000,             // Granularity");
1288 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1289 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1290 	dsdt_line("        0x0000,             // Translation Offset");
1291 	dsdt_line("        0x0001,             // Length");
1292 	dsdt_line("        ,, )");
1293 
1294 	if (bus == 0) {
1295 		dsdt_indent(3);
1296 		dsdt_fixed_ioport(0xCF8, 8);
1297 		dsdt_unindent(3);
1298 
1299 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1300 		    "PosDecode, EntireRange,");
1301 		dsdt_line("        0x0000,             // Granularity");
1302 		dsdt_line("        0x0000,             // Range Minimum");
1303 		dsdt_line("        0x0CF7,             // Range Maximum");
1304 		dsdt_line("        0x0000,             // Translation Offset");
1305 		dsdt_line("        0x0CF8,             // Length");
1306 		dsdt_line("        ,, , TypeStatic)");
1307 
1308 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1309 		    "PosDecode, EntireRange,");
1310 		dsdt_line("        0x0000,             // Granularity");
1311 		dsdt_line("        0x0D00,             // Range Minimum");
1312 		dsdt_line("        0x%04X,             // Range Maximum",
1313 		    PCI_EMUL_IOBASE - 1);
1314 		dsdt_line("        0x0000,             // Translation Offset");
1315 		dsdt_line("        0x%04X,             // Length",
1316 		    PCI_EMUL_IOBASE - 0x0D00);
1317 		dsdt_line("        ,, , TypeStatic)");
1318 
1319 		if (bi == NULL) {
1320 			dsdt_line("    })");
1321 			goto done;
1322 		}
1323 	}
1324 	assert(bi != NULL);
1325 
1326 	/* i/o window */
1327 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1328 	    "PosDecode, EntireRange,");
1329 	dsdt_line("        0x0000,             // Granularity");
1330 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1331 	dsdt_line("        0x%04X,             // Range Maximum",
1332 	    bi->iolimit - 1);
1333 	dsdt_line("        0x0000,             // Translation Offset");
1334 	dsdt_line("        0x%04X,             // Length",
1335 	    bi->iolimit - bi->iobase);
1336 	dsdt_line("        ,, , TypeStatic)");
1337 
1338 	/* mmio window (32-bit) */
1339 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1340 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1341 	dsdt_line("        0x00000000,         // Granularity");
1342 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1343 	dsdt_line("        0x%08X,         // Range Maximum\n",
1344 	    bi->memlimit32 - 1);
1345 	dsdt_line("        0x00000000,         // Translation Offset");
1346 	dsdt_line("        0x%08X,         // Length\n",
1347 	    bi->memlimit32 - bi->membase32);
1348 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1349 
1350 	/* mmio window (64-bit) */
1351 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1352 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1353 	dsdt_line("        0x0000000000000000, // Granularity");
1354 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1355 	dsdt_line("        0x%016lX, // Range Maximum\n",
1356 	    bi->memlimit64 - 1);
1357 	dsdt_line("        0x0000000000000000, // Translation Offset");
1358 	dsdt_line("        0x%016lX, // Length\n",
1359 	    bi->memlimit64 - bi->membase64);
1360 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1361 	dsdt_line("    })");
1362 
1363 	count = pci_count_lintr(bus);
1364 	if (count != 0) {
1365 		dsdt_indent(2);
1366 		dsdt_line("Name (PPRT, Package ()");
1367 		dsdt_line("{");
1368 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1369 		dsdt_line("})");
1370 		dsdt_line("Name (APRT, Package ()");
1371 		dsdt_line("{");
1372 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1373 		dsdt_line("})");
1374 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1375 		dsdt_line("{");
1376 		dsdt_line("  If (PICM)");
1377 		dsdt_line("  {");
1378 		dsdt_line("    Return (APRT)");
1379 		dsdt_line("  }");
1380 		dsdt_line("  Else");
1381 		dsdt_line("  {");
1382 		dsdt_line("    Return (PPRT)");
1383 		dsdt_line("  }");
1384 		dsdt_line("}");
1385 		dsdt_unindent(2);
1386 	}
1387 
1388 	dsdt_indent(2);
1389 	for (slot = 0; slot < MAXSLOTS; slot++) {
1390 		si = &bi->slotinfo[slot];
1391 		for (func = 0; func < MAXFUNCS; func++) {
1392 			pi = si->si_funcs[func].fi_devi;
1393 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1394 				pi->pi_d->pe_write_dsdt(pi);
1395 		}
1396 	}
1397 	dsdt_unindent(2);
1398 done:
1399 	dsdt_line("  }");
1400 }
1401 
1402 void
1403 pci_write_dsdt(void)
1404 {
1405 	int bus;
1406 
1407 	dsdt_indent(1);
1408 	dsdt_line("Name (PICM, 0x00)");
1409 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1410 	dsdt_line("{");
1411 	dsdt_line("  Store (Arg0, PICM)");
1412 	dsdt_line("}");
1413 	dsdt_line("");
1414 	dsdt_line("Scope (_SB)");
1415 	dsdt_line("{");
1416 	for (bus = 0; bus < MAXBUSES; bus++)
1417 		pci_bus_write_dsdt(bus);
1418 	dsdt_line("}");
1419 	dsdt_unindent(1);
1420 }
1421 
1422 int
1423 pci_bus_configured(int bus)
1424 {
1425 	assert(bus >= 0 && bus < MAXBUSES);
1426 	return (pci_businfo[bus] != NULL);
1427 }
1428 
1429 int
1430 pci_msi_enabled(struct pci_devinst *pi)
1431 {
1432 	return (pi->pi_msi.enabled);
1433 }
1434 
1435 int
1436 pci_msi_maxmsgnum(struct pci_devinst *pi)
1437 {
1438 	if (pi->pi_msi.enabled)
1439 		return (pi->pi_msi.maxmsgnum);
1440 	else
1441 		return (0);
1442 }
1443 
1444 int
1445 pci_msix_enabled(struct pci_devinst *pi)
1446 {
1447 
1448 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1449 }
1450 
1451 void
1452 pci_generate_msix(struct pci_devinst *pi, int index)
1453 {
1454 	struct msix_table_entry *mte;
1455 
1456 	if (!pci_msix_enabled(pi))
1457 		return;
1458 
1459 	if (pi->pi_msix.function_mask)
1460 		return;
1461 
1462 	if (index >= pi->pi_msix.table_count)
1463 		return;
1464 
1465 	mte = &pi->pi_msix.table[index];
1466 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1467 		/* XXX Set PBA bit if interrupt is disabled */
1468 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1469 	}
1470 }
1471 
1472 void
1473 pci_generate_msi(struct pci_devinst *pi, int index)
1474 {
1475 
1476 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1477 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1478 			     pi->pi_msi.msg_data + index);
1479 	}
1480 }
1481 
1482 static bool
1483 pci_lintr_permitted(struct pci_devinst *pi)
1484 {
1485 	uint16_t cmd;
1486 
1487 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1488 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1489 		(cmd & PCIM_CMD_INTxDIS)));
1490 }
1491 
1492 void
1493 pci_lintr_request(struct pci_devinst *pi)
1494 {
1495 	struct businfo *bi;
1496 	struct slotinfo *si;
1497 	int bestpin, bestcount, pin;
1498 
1499 	bi = pci_businfo[pi->pi_bus];
1500 	assert(bi != NULL);
1501 
1502 	/*
1503 	 * Just allocate a pin from our slot.  The pin will be
1504 	 * assigned IRQs later when interrupts are routed.
1505 	 */
1506 	si = &bi->slotinfo[pi->pi_slot];
1507 	bestpin = 0;
1508 	bestcount = si->si_intpins[0].ii_count;
1509 	for (pin = 1; pin < 4; pin++) {
1510 		if (si->si_intpins[pin].ii_count < bestcount) {
1511 			bestpin = pin;
1512 			bestcount = si->si_intpins[pin].ii_count;
1513 		}
1514 	}
1515 
1516 	si->si_intpins[bestpin].ii_count++;
1517 	pi->pi_lintr.pin = bestpin + 1;
1518 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1519 }
1520 
1521 static void
1522 pci_lintr_route(struct pci_devinst *pi)
1523 {
1524 	struct businfo *bi;
1525 	struct intxinfo *ii;
1526 
1527 	if (pi->pi_lintr.pin == 0)
1528 		return;
1529 
1530 	bi = pci_businfo[pi->pi_bus];
1531 	assert(bi != NULL);
1532 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1533 
1534 	/*
1535 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1536 	 * is not yet assigned.
1537 	 */
1538 	if (ii->ii_ioapic_irq == 0)
1539 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1540 	assert(ii->ii_ioapic_irq > 0);
1541 
1542 	/*
1543 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1544 	 * not yet assigned.
1545 	 */
1546 	if (ii->ii_pirq_pin == 0)
1547 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1548 	assert(ii->ii_pirq_pin > 0);
1549 
1550 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1551 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1552 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1553 }
1554 
1555 void
1556 pci_lintr_assert(struct pci_devinst *pi)
1557 {
1558 
1559 	assert(pi->pi_lintr.pin > 0);
1560 
1561 	pthread_mutex_lock(&pi->pi_lintr.lock);
1562 	if (pi->pi_lintr.state == IDLE) {
1563 		if (pci_lintr_permitted(pi)) {
1564 			pi->pi_lintr.state = ASSERTED;
1565 			pci_irq_assert(pi);
1566 		} else
1567 			pi->pi_lintr.state = PENDING;
1568 	}
1569 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1570 }
1571 
1572 void
1573 pci_lintr_deassert(struct pci_devinst *pi)
1574 {
1575 
1576 	assert(pi->pi_lintr.pin > 0);
1577 
1578 	pthread_mutex_lock(&pi->pi_lintr.lock);
1579 	if (pi->pi_lintr.state == ASSERTED) {
1580 		pi->pi_lintr.state = IDLE;
1581 		pci_irq_deassert(pi);
1582 	} else if (pi->pi_lintr.state == PENDING)
1583 		pi->pi_lintr.state = IDLE;
1584 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1585 }
1586 
1587 static void
1588 pci_lintr_update(struct pci_devinst *pi)
1589 {
1590 
1591 	pthread_mutex_lock(&pi->pi_lintr.lock);
1592 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1593 		pci_irq_deassert(pi);
1594 		pi->pi_lintr.state = PENDING;
1595 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1596 		pi->pi_lintr.state = ASSERTED;
1597 		pci_irq_assert(pi);
1598 	}
1599 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1600 }
1601 
1602 int
1603 pci_count_lintr(int bus)
1604 {
1605 	int count, slot, pin;
1606 	struct slotinfo *slotinfo;
1607 
1608 	count = 0;
1609 	if (pci_businfo[bus] != NULL) {
1610 		for (slot = 0; slot < MAXSLOTS; slot++) {
1611 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1612 			for (pin = 0; pin < 4; pin++) {
1613 				if (slotinfo->si_intpins[pin].ii_count != 0)
1614 					count++;
1615 			}
1616 		}
1617 	}
1618 	return (count);
1619 }
1620 
1621 void
1622 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1623 {
1624 	struct businfo *bi;
1625 	struct slotinfo *si;
1626 	struct intxinfo *ii;
1627 	int slot, pin;
1628 
1629 	if ((bi = pci_businfo[bus]) == NULL)
1630 		return;
1631 
1632 	for (slot = 0; slot < MAXSLOTS; slot++) {
1633 		si = &bi->slotinfo[slot];
1634 		for (pin = 0; pin < 4; pin++) {
1635 			ii = &si->si_intpins[pin];
1636 			if (ii->ii_count != 0)
1637 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1638 				    ii->ii_ioapic_irq, arg);
1639 		}
1640 	}
1641 }
1642 
1643 /*
1644  * Return 1 if the emulated device in 'slot' is a multi-function device.
1645  * Return 0 otherwise.
1646  */
1647 static int
1648 pci_emul_is_mfdev(int bus, int slot)
1649 {
1650 	struct businfo *bi;
1651 	struct slotinfo *si;
1652 	int f, numfuncs;
1653 
1654 	numfuncs = 0;
1655 	if ((bi = pci_businfo[bus]) != NULL) {
1656 		si = &bi->slotinfo[slot];
1657 		for (f = 0; f < MAXFUNCS; f++) {
1658 			if (si->si_funcs[f].fi_devi != NULL) {
1659 				numfuncs++;
1660 			}
1661 		}
1662 	}
1663 	return (numfuncs > 1);
1664 }
1665 
1666 /*
1667  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1668  * whether or not is a multi-function being emulated in the pci 'slot'.
1669  */
1670 static void
1671 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1672 {
1673 	int mfdev;
1674 
1675 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1676 		mfdev = pci_emul_is_mfdev(bus, slot);
1677 		switch (bytes) {
1678 		case 1:
1679 		case 2:
1680 			*rv &= ~PCIM_MFDEV;
1681 			if (mfdev) {
1682 				*rv |= PCIM_MFDEV;
1683 			}
1684 			break;
1685 		case 4:
1686 			*rv &= ~(PCIM_MFDEV << 16);
1687 			if (mfdev) {
1688 				*rv |= (PCIM_MFDEV << 16);
1689 			}
1690 			break;
1691 		}
1692 	}
1693 }
1694 
1695 static void
1696 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1697 {
1698 	int i, rshift;
1699 	uint32_t cmd, cmd2, changed, old, readonly;
1700 
1701 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1702 
1703 	/*
1704 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1705 	 *
1706 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1707 	 * 'write 1 to clear'. However these bits are not set to '1' by
1708 	 * any device emulation so it is simpler to treat them as readonly.
1709 	 */
1710 	rshift = (coff & 0x3) * 8;
1711 	readonly = 0xFFFFF880 >> rshift;
1712 
1713 	old = CFGREAD(pi, coff, bytes);
1714 	new &= ~readonly;
1715 	new |= (old & readonly);
1716 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1717 
1718 	cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1719 	changed = cmd ^ cmd2;
1720 
1721 	/*
1722 	 * If the MMIO or I/O address space decoding has changed then
1723 	 * register/unregister all BARs that decode that address space.
1724 	 */
1725 	for (i = 0; i <= PCI_BARMAX; i++) {
1726 		switch (pi->pi_bar[i].type) {
1727 			case PCIBAR_NONE:
1728 			case PCIBAR_MEMHI64:
1729 				break;
1730 			case PCIBAR_IO:
1731 				/* I/O address space decoding changed? */
1732 				if (changed & PCIM_CMD_PORTEN) {
1733 					if (porten(pi))
1734 						register_bar(pi, i);
1735 					else
1736 						unregister_bar(pi, i);
1737 				}
1738 				break;
1739 			case PCIBAR_MEM32:
1740 			case PCIBAR_MEM64:
1741 				/* MMIO address space decoding changed? */
1742 				if (changed & PCIM_CMD_MEMEN) {
1743 					if (memen(pi))
1744 						register_bar(pi, i);
1745 					else
1746 						unregister_bar(pi, i);
1747 				}
1748 				break;
1749 			default:
1750 				assert(0);
1751 		}
1752 	}
1753 
1754 	/*
1755 	 * If INTx has been unmasked and is pending, assert the
1756 	 * interrupt.
1757 	 */
1758 	pci_lintr_update(pi);
1759 }
1760 
1761 static void
1762 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1763     int coff, int bytes, uint32_t *eax)
1764 {
1765 	struct businfo *bi;
1766 	struct slotinfo *si;
1767 	struct pci_devinst *pi;
1768 	struct pci_devemu *pe;
1769 	int idx, needcfg;
1770 	uint64_t addr, mask;
1771 	uint64_t bar = 0;
1772 
1773 	if ((bi = pci_businfo[bus]) != NULL) {
1774 		si = &bi->slotinfo[slot];
1775 		pi = si->si_funcs[func].fi_devi;
1776 	} else
1777 		pi = NULL;
1778 
1779 	/*
1780 	 * Just return if there is no device at this slot:func or if the
1781 	 * the guest is doing an un-aligned access.
1782 	 */
1783 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1784 	    (coff & (bytes - 1)) != 0) {
1785 		if (in)
1786 			*eax = 0xffffffff;
1787 		return;
1788 	}
1789 
1790 	/*
1791 	 * Ignore all writes beyond the standard config space and return all
1792 	 * ones on reads.
1793 	 */
1794 	if (coff >= PCI_REGMAX + 1) {
1795 		if (in) {
1796 			*eax = 0xffffffff;
1797 			/*
1798 			 * Extended capabilities begin at offset 256 in config
1799 			 * space. Absence of extended capabilities is signaled
1800 			 * with all 0s in the extended capability header at
1801 			 * offset 256.
1802 			 */
1803 			if (coff <= PCI_REGMAX + 4)
1804 				*eax = 0x00000000;
1805 		}
1806 		return;
1807 	}
1808 
1809 	pe = pi->pi_d;
1810 
1811 	/*
1812 	 * Config read
1813 	 */
1814 	if (in) {
1815 		/* Let the device emulation override the default handler */
1816 		if (pe->pe_cfgread != NULL) {
1817 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1818 			    eax);
1819 		} else {
1820 			needcfg = 1;
1821 		}
1822 
1823 		if (needcfg)
1824 			*eax = CFGREAD(pi, coff, bytes);
1825 
1826 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1827 	} else {
1828 		/* Let the device emulation override the default handler */
1829 		if (pe->pe_cfgwrite != NULL &&
1830 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1831 			return;
1832 
1833 		/*
1834 		 * Special handling for write to BAR registers
1835 		 */
1836 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1837 			/*
1838 			 * Ignore writes to BAR registers that are not
1839 			 * 4-byte aligned.
1840 			 */
1841 			if (bytes != 4 || (coff & 0x3) != 0)
1842 				return;
1843 			idx = (coff - PCIR_BAR(0)) / 4;
1844 			mask = ~(pi->pi_bar[idx].size - 1);
1845 			switch (pi->pi_bar[idx].type) {
1846 			case PCIBAR_NONE:
1847 				pi->pi_bar[idx].addr = bar = 0;
1848 				break;
1849 			case PCIBAR_IO:
1850 				addr = *eax & mask;
1851 				addr &= 0xffff;
1852 				bar = addr | PCIM_BAR_IO_SPACE;
1853 				/*
1854 				 * Register the new BAR value for interception
1855 				 */
1856 				if (addr != pi->pi_bar[idx].addr) {
1857 					update_bar_address(pi, addr, idx,
1858 							   PCIBAR_IO);
1859 				}
1860 				break;
1861 			case PCIBAR_MEM32:
1862 				addr = bar = *eax & mask;
1863 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1864 				if (addr != pi->pi_bar[idx].addr) {
1865 					update_bar_address(pi, addr, idx,
1866 							   PCIBAR_MEM32);
1867 				}
1868 				break;
1869 			case PCIBAR_MEM64:
1870 				addr = bar = *eax & mask;
1871 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1872 				       PCIM_BAR_MEM_PREFETCH;
1873 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1874 					update_bar_address(pi, addr, idx,
1875 							   PCIBAR_MEM64);
1876 				}
1877 				break;
1878 			case PCIBAR_MEMHI64:
1879 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1880 				addr = ((uint64_t)*eax << 32) & mask;
1881 				bar = addr >> 32;
1882 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1883 					update_bar_address(pi, addr, idx - 1,
1884 							   PCIBAR_MEMHI64);
1885 				}
1886 				break;
1887 			default:
1888 				assert(0);
1889 			}
1890 			pci_set_cfgdata32(pi, coff, bar);
1891 
1892 		} else if (pci_emul_iscap(pi, coff)) {
1893 			pci_emul_capwrite(pi, coff, bytes, *eax);
1894 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1895 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1896 		} else {
1897 			CFGWRITE(pi, coff, *eax, bytes);
1898 		}
1899 	}
1900 }
1901 
1902 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1903 
1904 static int
1905 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1906 		 uint32_t *eax, void *arg)
1907 {
1908 	uint32_t x;
1909 
1910 	if (bytes != 4) {
1911 		if (in)
1912 			*eax = (bytes == 2) ? 0xffff : 0xff;
1913 		return (0);
1914 	}
1915 
1916 	if (in) {
1917 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1918 		if (cfgenable)
1919 			x |= CONF1_ENABLE;
1920 		*eax = x;
1921 	} else {
1922 		x = *eax;
1923 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1924 		cfgoff = x & PCI_REGMAX;
1925 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1926 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1927 		cfgbus = (x >> 16) & PCI_BUSMAX;
1928 	}
1929 
1930 	return (0);
1931 }
1932 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1933 
1934 static int
1935 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1936 		 uint32_t *eax, void *arg)
1937 {
1938 	int coff;
1939 
1940 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1941 
1942 	coff = cfgoff + (port - CONF1_DATA_PORT);
1943 	if (cfgenable) {
1944 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1945 		    eax);
1946 	} else {
1947 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1948 		if (in)
1949 			*eax = 0xffffffff;
1950 	}
1951 	return (0);
1952 }
1953 
1954 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1955 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1956 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1957 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1958 
1959 #define PCI_EMUL_TEST
1960 #ifdef PCI_EMUL_TEST
1961 /*
1962  * Define a dummy test device
1963  */
1964 #define DIOSZ	8
1965 #define DMEMSZ	4096
1966 struct pci_emul_dsoftc {
1967 	uint8_t	  ioregs[DIOSZ];
1968 	uint8_t	  memregs[2][DMEMSZ];
1969 };
1970 
1971 #define	PCI_EMUL_MSI_MSGS	 4
1972 #define	PCI_EMUL_MSIX_MSGS	16
1973 
1974 static int
1975 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1976 {
1977 	int error;
1978 	struct pci_emul_dsoftc *sc;
1979 
1980 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1981 
1982 	pi->pi_arg = sc;
1983 
1984 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1985 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1986 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1987 
1988 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1989 	assert(error == 0);
1990 
1991 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1992 	assert(error == 0);
1993 
1994 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1995 	assert(error == 0);
1996 
1997 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1998 	assert(error == 0);
1999 
2000 	return (0);
2001 }
2002 
2003 static void
2004 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2005 	      uint64_t offset, int size, uint64_t value)
2006 {
2007 	int i;
2008 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2009 
2010 	if (baridx == 0) {
2011 		if (offset + size > DIOSZ) {
2012 			printf("diow: iow too large, offset %ld size %d\n",
2013 			       offset, size);
2014 			return;
2015 		}
2016 
2017 		if (size == 1) {
2018 			sc->ioregs[offset] = value & 0xff;
2019 		} else if (size == 2) {
2020 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2021 		} else if (size == 4) {
2022 			*(uint32_t *)&sc->ioregs[offset] = value;
2023 		} else {
2024 			printf("diow: iow unknown size %d\n", size);
2025 		}
2026 
2027 		/*
2028 		 * Special magic value to generate an interrupt
2029 		 */
2030 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2031 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2032 
2033 		if (value == 0xabcdef) {
2034 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2035 				pci_generate_msi(pi, i);
2036 		}
2037 	}
2038 
2039 	if (baridx == 1 || baridx == 2) {
2040 		if (offset + size > DMEMSZ) {
2041 			printf("diow: memw too large, offset %ld size %d\n",
2042 			       offset, size);
2043 			return;
2044 		}
2045 
2046 		i = baridx - 1;		/* 'memregs' index */
2047 
2048 		if (size == 1) {
2049 			sc->memregs[i][offset] = value;
2050 		} else if (size == 2) {
2051 			*(uint16_t *)&sc->memregs[i][offset] = value;
2052 		} else if (size == 4) {
2053 			*(uint32_t *)&sc->memregs[i][offset] = value;
2054 		} else if (size == 8) {
2055 			*(uint64_t *)&sc->memregs[i][offset] = value;
2056 		} else {
2057 			printf("diow: memw unknown size %d\n", size);
2058 		}
2059 
2060 		/*
2061 		 * magic interrupt ??
2062 		 */
2063 	}
2064 
2065 	if (baridx > 2 || baridx < 0) {
2066 		printf("diow: unknown bar idx %d\n", baridx);
2067 	}
2068 }
2069 
2070 static uint64_t
2071 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2072 	      uint64_t offset, int size)
2073 {
2074 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2075 	uint32_t value;
2076 	int i;
2077 
2078 	value = 0;
2079 	if (baridx == 0) {
2080 		if (offset + size > DIOSZ) {
2081 			printf("dior: ior too large, offset %ld size %d\n",
2082 			       offset, size);
2083 			return (0);
2084 		}
2085 
2086 		value = 0;
2087 		if (size == 1) {
2088 			value = sc->ioregs[offset];
2089 		} else if (size == 2) {
2090 			value = *(uint16_t *) &sc->ioregs[offset];
2091 		} else if (size == 4) {
2092 			value = *(uint32_t *) &sc->ioregs[offset];
2093 		} else {
2094 			printf("dior: ior unknown size %d\n", size);
2095 		}
2096 	}
2097 
2098 	if (baridx == 1 || baridx == 2) {
2099 		if (offset + size > DMEMSZ) {
2100 			printf("dior: memr too large, offset %ld size %d\n",
2101 			       offset, size);
2102 			return (0);
2103 		}
2104 
2105 		i = baridx - 1;		/* 'memregs' index */
2106 
2107 		if (size == 1) {
2108 			value = sc->memregs[i][offset];
2109 		} else if (size == 2) {
2110 			value = *(uint16_t *) &sc->memregs[i][offset];
2111 		} else if (size == 4) {
2112 			value = *(uint32_t *) &sc->memregs[i][offset];
2113 		} else if (size == 8) {
2114 			value = *(uint64_t *) &sc->memregs[i][offset];
2115 		} else {
2116 			printf("dior: ior unknown size %d\n", size);
2117 		}
2118 	}
2119 
2120 
2121 	if (baridx > 2 || baridx < 0) {
2122 		printf("dior: unknown bar idx %d\n", baridx);
2123 		return (0);
2124 	}
2125 
2126 	return (value);
2127 }
2128 
2129 struct pci_devemu pci_dummy = {
2130 	.pe_emu = "dummy",
2131 	.pe_init = pci_emul_dinit,
2132 	.pe_barwrite = pci_emul_diow,
2133 	.pe_barread = pci_emul_dior
2134 };
2135 PCI_EMUL_SET(pci_dummy);
2136 
2137 #endif /* PCI_EMUL_TEST */
2138