xref: /freebsd/usr.sbin/bhyve/pci_passthru.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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/types.h>
34 #include <sys/pciio.h>
35 #include <sys/ioctl.h>
36 
37 #include <dev/io/iodev.h>
38 #include <dev/pci/pcireg.h>
39 
40 #include <machine/iodev.h>
41 
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <string.h>
45 #include <errno.h>
46 #include <fcntl.h>
47 #include <unistd.h>
48 
49 #include <machine/vmm.h>
50 #include <vmmapi.h>
51 #include "pci_emul.h"
52 #include "mem.h"
53 
54 #ifndef _PATH_DEVPCI
55 #define	_PATH_DEVPCI	"/dev/pci"
56 #endif
57 
58 #ifndef	_PATH_DEVIO
59 #define	_PATH_DEVIO	"/dev/io"
60 #endif
61 
62 #define	LEGACY_SUPPORT	1
63 
64 #define MSIX_TABLE_COUNT(ctrl) (((ctrl) & PCIM_MSIXCTRL_TABLE_SIZE) + 1)
65 #define MSIX_CAPLEN 12
66 
67 static int pcifd = -1;
68 static int iofd = -1;
69 
70 struct passthru_softc {
71 	struct pci_devinst *psc_pi;
72 	struct pcibar psc_bar[PCI_BARMAX + 1];
73 	struct {
74 		int		capoff;
75 		int		msgctrl;
76 		int		emulated;
77 	} psc_msi;
78 	struct {
79 		int		capoff;
80 	} psc_msix;
81 	struct pcisel psc_sel;
82 };
83 
84 static int
85 msi_caplen(int msgctrl)
86 {
87 	int len;
88 
89 	len = 10;		/* minimum length of msi capability */
90 
91 	if (msgctrl & PCIM_MSICTRL_64BIT)
92 		len += 4;
93 
94 #if 0
95 	/*
96 	 * Ignore the 'mask' and 'pending' bits in the MSI capability.
97 	 * We'll let the guest manipulate them directly.
98 	 */
99 	if (msgctrl & PCIM_MSICTRL_VECTOR)
100 		len += 10;
101 #endif
102 
103 	return (len);
104 }
105 
106 static uint32_t
107 read_config(const struct pcisel *sel, long reg, int width)
108 {
109 	struct pci_io pi;
110 
111 	bzero(&pi, sizeof(pi));
112 	pi.pi_sel = *sel;
113 	pi.pi_reg = reg;
114 	pi.pi_width = width;
115 
116 	if (ioctl(pcifd, PCIOCREAD, &pi) < 0)
117 		return (0);				/* XXX */
118 	else
119 		return (pi.pi_data);
120 }
121 
122 static void
123 write_config(const struct pcisel *sel, long reg, int width, uint32_t data)
124 {
125 	struct pci_io pi;
126 
127 	bzero(&pi, sizeof(pi));
128 	pi.pi_sel = *sel;
129 	pi.pi_reg = reg;
130 	pi.pi_width = width;
131 	pi.pi_data = data;
132 
133 	(void)ioctl(pcifd, PCIOCWRITE, &pi);		/* XXX */
134 }
135 
136 #ifdef LEGACY_SUPPORT
137 static int
138 passthru_add_msicap(struct pci_devinst *pi, int msgnum, int nextptr)
139 {
140 	int capoff, i;
141 	struct msicap msicap;
142 	u_char *capdata;
143 
144 	pci_populate_msicap(&msicap, msgnum, nextptr);
145 
146 	/*
147 	 * XXX
148 	 * Copy the msi capability structure in the last 16 bytes of the
149 	 * config space. This is wrong because it could shadow something
150 	 * useful to the device.
151 	 */
152 	capoff = 256 - roundup(sizeof(msicap), 4);
153 	capdata = (u_char *)&msicap;
154 	for (i = 0; i < sizeof(msicap); i++)
155 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
156 
157 	return (capoff);
158 }
159 #endif	/* LEGACY_SUPPORT */
160 
161 static int
162 cfginitmsi(struct passthru_softc *sc)
163 {
164 	int i, ptr, capptr, cap, sts, caplen, table_size;
165 	uint32_t u32;
166 	struct pcisel sel;
167 	struct pci_devinst *pi;
168 	struct msixcap msixcap;
169 	uint32_t *msixcap_ptr;
170 
171 	pi = sc->psc_pi;
172 	sel = sc->psc_sel;
173 
174 	/*
175 	 * Parse the capabilities and cache the location of the MSI
176 	 * and MSI-X capabilities.
177 	 */
178 	sts = read_config(&sel, PCIR_STATUS, 2);
179 	if (sts & PCIM_STATUS_CAPPRESENT) {
180 		ptr = read_config(&sel, PCIR_CAP_PTR, 1);
181 		while (ptr != 0 && ptr != 0xff) {
182 			cap = read_config(&sel, ptr + PCICAP_ID, 1);
183 			if (cap == PCIY_MSI) {
184 				/*
185 				 * Copy the MSI capability into the config
186 				 * space of the emulated pci device
187 				 */
188 				sc->psc_msi.capoff = ptr;
189 				sc->psc_msi.msgctrl = read_config(&sel,
190 								  ptr + 2, 2);
191 				sc->psc_msi.emulated = 0;
192 				caplen = msi_caplen(sc->psc_msi.msgctrl);
193 				capptr = ptr;
194 				while (caplen > 0) {
195 					u32 = read_config(&sel, capptr, 4);
196 					pci_set_cfgdata32(pi, capptr, u32);
197 					caplen -= 4;
198 					capptr += 4;
199 				}
200 			} else if (cap == PCIY_MSIX) {
201 				/*
202 				 * Copy the MSI-X capability
203 				 */
204 				sc->psc_msix.capoff = ptr;
205 				caplen = 12;
206 				msixcap_ptr = (uint32_t*) &msixcap;
207 				capptr = ptr;
208 				while (caplen > 0) {
209 					u32 = read_config(&sel, capptr, 4);
210 					*msixcap_ptr = u32;
211 					pci_set_cfgdata32(pi, capptr, u32);
212 					caplen -= 4;
213 					capptr += 4;
214 					msixcap_ptr++;
215 				}
216 			}
217 			ptr = read_config(&sel, ptr + PCICAP_NEXTPTR, 1);
218 		}
219 	}
220 
221 	if (sc->psc_msix.capoff != 0) {
222 		pi->pi_msix.pba_bar =
223 		    msixcap.pba_info & PCIM_MSIX_BIR_MASK;
224 		pi->pi_msix.pba_offset =
225 		    msixcap.pba_info & ~PCIM_MSIX_BIR_MASK;
226 		pi->pi_msix.table_bar =
227 		    msixcap.table_info & PCIM_MSIX_BIR_MASK;
228 		pi->pi_msix.table_offset =
229 		    msixcap.table_info & ~PCIM_MSIX_BIR_MASK;
230 		pi->pi_msix.table_count = MSIX_TABLE_COUNT(msixcap.msgctrl);
231 
232 		/* Allocate the emulated MSI-X table array */
233 		table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
234 		pi->pi_msix.table = malloc(table_size);
235 		bzero(pi->pi_msix.table, table_size);
236 
237 		/* Mask all table entries */
238 		for (i = 0; i < pi->pi_msix.table_count; i++) {
239 			pi->pi_msix.table[i].vector_control |=
240 						PCIM_MSIX_VCTRL_MASK;
241 		}
242 	}
243 
244 #ifdef LEGACY_SUPPORT
245 	/*
246 	 * If the passthrough device does not support MSI then craft a
247 	 * MSI capability for it. We link the new MSI capability at the
248 	 * head of the list of capabilities.
249 	 */
250 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0 && sc->psc_msi.capoff == 0) {
251 		int origptr, msiptr;
252 		origptr = read_config(&sel, PCIR_CAP_PTR, 1);
253 		msiptr = passthru_add_msicap(pi, 1, origptr);
254 		sc->psc_msi.capoff = msiptr;
255 		sc->psc_msi.msgctrl = pci_get_cfgdata16(pi, msiptr + 2);
256 		sc->psc_msi.emulated = 1;
257 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, msiptr);
258 	}
259 #endif
260 
261 	/* Make sure one of the capabilities is present */
262 	if (sc->psc_msi.capoff == 0 && sc->psc_msix.capoff == 0)
263 		return (-1);
264 	else
265 		return (0);
266 }
267 
268 static uint64_t
269 msix_table_read(struct passthru_softc *sc, uint64_t offset, int size)
270 {
271 	struct pci_devinst *pi;
272 	struct msix_table_entry *entry;
273 	uint8_t *src8;
274 	uint16_t *src16;
275 	uint32_t *src32;
276 	uint64_t *src64;
277 	uint64_t data;
278 	size_t entry_offset;
279 	int index;
280 
281 	pi = sc->psc_pi;
282 	offset -= pi->pi_msix.table_offset;
283 
284 	index = offset / MSIX_TABLE_ENTRY_SIZE;
285 	if (index >= pi->pi_msix.table_count)
286 		return (-1);
287 
288 	entry = &pi->pi_msix.table[index];
289 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
290 
291 	switch(size) {
292 	case 1:
293 		src8 = (uint8_t *)((void *)entry + entry_offset);
294 		data = *src8;
295 		break;
296 	case 2:
297 		src16 = (uint16_t *)((void *)entry + entry_offset);
298 		data = *src16;
299 		break;
300 	case 4:
301 		src32 = (uint32_t *)((void *)entry + entry_offset);
302 		data = *src32;
303 		break;
304 	case 8:
305 		src64 = (uint64_t *)((void *)entry + entry_offset);
306 		data = *src64;
307 		break;
308 	default:
309 		return (-1);
310 	}
311 
312 	return (data);
313 }
314 
315 static void
316 msix_table_write(struct vmctx *ctx, int vcpu, struct passthru_softc *sc,
317 		 uint64_t offset, int size, uint64_t data)
318 {
319 	struct pci_devinst *pi;
320 	struct msix_table_entry *entry;
321 	uint32_t *dest;
322 	size_t entry_offset;
323 	uint32_t vector_control;
324 	int error, index;
325 
326 	pi = sc->psc_pi;
327 	offset -= pi->pi_msix.table_offset;
328 
329 	index = offset / MSIX_TABLE_ENTRY_SIZE;
330 	if (index >= pi->pi_msix.table_count)
331 		return;
332 
333 	entry = &pi->pi_msix.table[index];
334 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
335 
336 	/* Only 4 byte naturally-aligned writes are supported */
337 	assert(size == 4);
338 	assert(entry_offset % 4 == 0);
339 
340 	vector_control = entry->vector_control;
341 	dest = (uint32_t *)((void *)entry + entry_offset);
342 	*dest = data;
343 	/* If MSI-X hasn't been enabled, do nothing */
344 	if (pi->pi_msix.enabled) {
345 		/* If the entry is masked, don't set it up */
346 		if ((entry->vector_control & PCIM_MSIX_VCTRL_MASK) == 0 ||
347 		    (vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
348 			error = vm_setup_msix(ctx, vcpu, sc->psc_sel.pc_bus,
349 					      sc->psc_sel.pc_dev,
350 					      sc->psc_sel.pc_func,
351 					      index, entry->msg_data,
352 					      entry->vector_control,
353 					      entry->addr);
354 		}
355 	}
356 }
357 
358 static int
359 init_msix_table(struct vmctx *ctx, struct passthru_softc *sc, uint64_t base)
360 {
361 	int b, s, f;
362 	int error, idx;
363 	size_t len, remaining, table_size;
364 	vm_paddr_t start;
365 	struct pci_devinst *pi = sc->psc_pi;
366 
367 	assert(pci_msix_table_bar(pi) >= 0 && pci_msix_pba_bar(pi) >= 0);
368 
369 	b = sc->psc_sel.pc_bus;
370 	s = sc->psc_sel.pc_dev;
371 	f = sc->psc_sel.pc_func;
372 
373 	/*
374 	 * If the MSI-X table BAR maps memory intended for
375 	 * other uses, it is at least assured that the table
376 	 * either resides in its own page within the region,
377 	 * or it resides in a page shared with only the PBA.
378 	 */
379 	if (pi->pi_msix.pba_bar == pi->pi_msix.table_bar &&
380 	    ((pi->pi_msix.pba_offset - pi->pi_msix.table_offset) < 4096)) {
381 		/* Need to also emulate the PBA, not supported yet */
382 		printf("Unsupported MSI-X configuration: %d/%d/%d\n", b, s, f);
383 		return (-1);
384 	}
385 
386 	/* Compute the MSI-X table size */
387 	table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
388 	table_size = roundup2(table_size, 4096);
389 
390 	idx = pi->pi_msix.table_bar;
391 	start = pi->pi_bar[idx].addr;
392 	remaining = pi->pi_bar[idx].size;
393 
394 	/* Map everything before the MSI-X table */
395 	if (pi->pi_msix.table_offset > 0) {
396 		len = pi->pi_msix.table_offset;
397 		error = vm_map_pptdev_mmio(ctx, b, s, f, start, len, base);
398 		if (error)
399 			return (error);
400 
401 		base += len;
402 		start += len;
403 		remaining -= len;
404 	}
405 
406 	/* Skip the MSI-X table */
407 	base += table_size;
408 	start += table_size;
409 	remaining -= table_size;
410 
411 	/* Map everything beyond the end of the MSI-X table */
412 	if (remaining > 0) {
413 		len = remaining;
414 		error = vm_map_pptdev_mmio(ctx, b, s, f, start, len, base);
415 		if (error)
416 			return (error);
417 	}
418 
419 	return (0);
420 }
421 
422 static int
423 cfginitbar(struct vmctx *ctx, struct passthru_softc *sc)
424 {
425 	int i, error;
426 	struct pci_devinst *pi;
427 	struct pci_bar_io bar;
428 	enum pcibar_type bartype;
429 	uint64_t base;
430 
431 	pi = sc->psc_pi;
432 
433 	/*
434 	 * Initialize BAR registers
435 	 */
436 	for (i = 0; i <= PCI_BARMAX; i++) {
437 		bzero(&bar, sizeof(bar));
438 		bar.pbi_sel = sc->psc_sel;
439 		bar.pbi_reg = PCIR_BAR(i);
440 
441 		if (ioctl(pcifd, PCIOCGETBAR, &bar) < 0)
442 			continue;
443 
444 		if (PCI_BAR_IO(bar.pbi_base)) {
445 			bartype = PCIBAR_IO;
446 			base = bar.pbi_base & PCIM_BAR_IO_BASE;
447 		} else {
448 			switch (bar.pbi_base & PCIM_BAR_MEM_TYPE) {
449 			case PCIM_BAR_MEM_64:
450 				bartype = PCIBAR_MEM64;
451 				break;
452 			default:
453 				bartype = PCIBAR_MEM32;
454 				break;
455 			}
456 			base = bar.pbi_base & PCIM_BAR_MEM_BASE;
457 		}
458 
459 		/* Cache information about the "real" BAR */
460 		sc->psc_bar[i].type = bartype;
461 		sc->psc_bar[i].size = bar.pbi_length;
462 		sc->psc_bar[i].addr = base;
463 
464 		/* Allocate the BAR in the guest I/O or MMIO space */
465 		error = pci_emul_alloc_pbar(pi, i, base, bartype,
466 					    bar.pbi_length);
467 		if (error)
468 			return (-1);
469 
470 		/* The MSI-X table needs special handling */
471 		if (i == pci_msix_table_bar(pi)) {
472 			error = init_msix_table(ctx, sc, base);
473 			if (error)
474 				return (-1);
475 		} else if (bartype != PCIBAR_IO) {
476 			/* Map the physical MMIO space in the guest MMIO space */
477 			error = vm_map_pptdev_mmio(ctx, sc->psc_sel.pc_bus,
478 				sc->psc_sel.pc_dev, sc->psc_sel.pc_func,
479 				pi->pi_bar[i].addr, pi->pi_bar[i].size, base);
480 			if (error)
481 				return (-1);
482 		}
483 
484 		/*
485 		 * 64-bit BAR takes up two slots so skip the next one.
486 		 */
487 		if (bartype == PCIBAR_MEM64) {
488 			i++;
489 			assert(i <= PCI_BARMAX);
490 			sc->psc_bar[i].type = PCIBAR_MEMHI64;
491 		}
492 	}
493 	return (0);
494 }
495 
496 static int
497 cfginit(struct vmctx *ctx, struct pci_devinst *pi, int bus, int slot, int func)
498 {
499 	int error;
500 	struct passthru_softc *sc;
501 
502 	error = 1;
503 	sc = pi->pi_arg;
504 
505 	bzero(&sc->psc_sel, sizeof(struct pcisel));
506 	sc->psc_sel.pc_bus = bus;
507 	sc->psc_sel.pc_dev = slot;
508 	sc->psc_sel.pc_func = func;
509 
510 	if (cfginitmsi(sc) != 0)
511 		goto done;
512 
513 	if (cfginitbar(ctx, sc) != 0)
514 		goto done;
515 
516 	error = 0;				/* success */
517 done:
518 	return (error);
519 }
520 
521 static int
522 passthru_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
523 {
524 	int bus, slot, func, error;
525 	struct passthru_softc *sc;
526 
527 	sc = NULL;
528 	error = 1;
529 
530 	if (pcifd < 0) {
531 		pcifd = open(_PATH_DEVPCI, O_RDWR, 0);
532 		if (pcifd < 0)
533 			goto done;
534 	}
535 
536 	if (iofd < 0) {
537 		iofd = open(_PATH_DEVIO, O_RDWR, 0);
538 		if (iofd < 0)
539 			goto done;
540 	}
541 
542 	if (opts == NULL ||
543 	    sscanf(opts, "%d/%d/%d", &bus, &slot, &func) != 3)
544 		goto done;
545 
546 	if (vm_assign_pptdev(ctx, bus, slot, func) != 0)
547 		goto done;
548 
549 	sc = malloc(sizeof(struct passthru_softc));
550 	memset(sc, 0, sizeof(struct passthru_softc));
551 
552 	pi->pi_arg = sc;
553 	sc->psc_pi = pi;
554 
555 	/* initialize config space */
556 	if ((error = cfginit(ctx, pi, bus, slot, func)) != 0)
557 		goto done;
558 
559 	error = 0;		/* success */
560 done:
561 	if (error) {
562 		free(sc);
563 		vm_unassign_pptdev(ctx, bus, slot, func);
564 	}
565 	return (error);
566 }
567 
568 static int
569 bar_access(int coff)
570 {
571 	if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1))
572 		return (1);
573 	else
574 		return (0);
575 }
576 
577 static int
578 msicap_access(struct passthru_softc *sc, int coff)
579 {
580 	int caplen;
581 
582 	if (sc->psc_msi.capoff == 0)
583 		return (0);
584 
585 	caplen = msi_caplen(sc->psc_msi.msgctrl);
586 
587 	if (coff >= sc->psc_msi.capoff && coff < sc->psc_msi.capoff + caplen)
588 		return (1);
589 	else
590 		return (0);
591 }
592 
593 static int
594 msixcap_access(struct passthru_softc *sc, int coff)
595 {
596 	if (sc->psc_msix.capoff == 0)
597 		return (0);
598 
599 	return (coff >= sc->psc_msix.capoff &&
600 	        coff < sc->psc_msix.capoff + MSIX_CAPLEN);
601 }
602 
603 static int
604 passthru_cfgread(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
605 		 int coff, int bytes, uint32_t *rv)
606 {
607 	struct passthru_softc *sc;
608 
609 	sc = pi->pi_arg;
610 
611 	/*
612 	 * PCI BARs and MSI capability is emulated.
613 	 */
614 	if (bar_access(coff) || msicap_access(sc, coff))
615 		return (-1);
616 
617 #ifdef LEGACY_SUPPORT
618 	/*
619 	 * Emulate PCIR_CAP_PTR if this device does not support MSI capability
620 	 * natively.
621 	 */
622 	if (sc->psc_msi.emulated) {
623 		if (coff >= PCIR_CAP_PTR && coff < PCIR_CAP_PTR + 4)
624 			return (-1);
625 	}
626 #endif
627 
628 	/* Everything else just read from the device's config space */
629 	*rv = read_config(&sc->psc_sel, coff, bytes);
630 
631 	return (0);
632 }
633 
634 static int
635 passthru_cfgwrite(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
636 		  int coff, int bytes, uint32_t val)
637 {
638 	int error, msix_table_entries, i;
639 	struct passthru_softc *sc;
640 
641 	sc = pi->pi_arg;
642 
643 	/*
644 	 * PCI BARs are emulated
645 	 */
646 	if (bar_access(coff))
647 		return (-1);
648 
649 	/*
650 	 * MSI capability is emulated
651 	 */
652 	if (msicap_access(sc, coff)) {
653 		msicap_cfgwrite(pi, sc->psc_msi.capoff, coff, bytes, val);
654 
655 		error = vm_setup_msi(ctx, vcpu, sc->psc_sel.pc_bus,
656 			sc->psc_sel.pc_dev, sc->psc_sel.pc_func, pi->pi_msi.cpu,
657 			pi->pi_msi.vector, pi->pi_msi.msgnum);
658 		if (error != 0) {
659 			printf("vm_setup_msi returned error %d\r\n", errno);
660 			exit(1);
661 		}
662 		return (0);
663 	}
664 
665 	if (msixcap_access(sc, coff)) {
666 		msixcap_cfgwrite(pi, sc->psc_msix.capoff, coff, bytes, val);
667 		if (pi->pi_msix.enabled) {
668 			msix_table_entries = pi->pi_msix.table_count;
669 			for (i = 0; i < msix_table_entries; i++) {
670 				error = vm_setup_msix(ctx, vcpu, sc->psc_sel.pc_bus,
671 						      sc->psc_sel.pc_dev,
672 						      sc->psc_sel.pc_func, i,
673 						      pi->pi_msix.table[i].msg_data,
674 						      pi->pi_msix.table[i].vector_control,
675 						      pi->pi_msix.table[i].addr);
676 
677 				if (error) {
678 					printf("vm_setup_msix returned error %d\r\n", errno);
679 					exit(1);
680 				}
681 			}
682 		}
683 		return (0);
684 	}
685 
686 #ifdef LEGACY_SUPPORT
687 	/*
688 	 * If this device does not support MSI natively then we cannot let
689 	 * the guest disable legacy interrupts from the device. It is the
690 	 * legacy interrupt that is triggering the virtual MSI to the guest.
691 	 */
692 	if (sc->psc_msi.emulated && pci_msi_enabled(pi)) {
693 		if (coff == PCIR_COMMAND && bytes == 2)
694 			val &= ~PCIM_CMD_INTxDIS;
695 	}
696 #endif
697 
698 	write_config(&sc->psc_sel, coff, bytes, val);
699 
700 	return (0);
701 }
702 
703 static void
704 passthru_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
705 	       uint64_t offset, int size, uint64_t value)
706 {
707 	struct passthru_softc *sc;
708 	struct iodev_pio_req pio;
709 
710 	sc = pi->pi_arg;
711 
712 	if (baridx == pci_msix_table_bar(pi)) {
713 		msix_table_write(ctx, vcpu, sc, offset, size, value);
714 	} else {
715 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
716 		bzero(&pio, sizeof(struct iodev_pio_req));
717 		pio.access = IODEV_PIO_WRITE;
718 		pio.port = sc->psc_bar[baridx].addr + offset;
719 		pio.width = size;
720 		pio.val = value;
721 
722 		(void)ioctl(iofd, IODEV_PIO, &pio);
723 	}
724 }
725 
726 static uint64_t
727 passthru_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
728 	      uint64_t offset, int size)
729 {
730 	struct passthru_softc *sc;
731 	struct iodev_pio_req pio;
732 	uint64_t val;
733 
734 	sc = pi->pi_arg;
735 
736 	if (baridx == pci_msix_table_bar(pi)) {
737 		val = msix_table_read(sc, offset, size);
738 	} else {
739 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
740 		bzero(&pio, sizeof(struct iodev_pio_req));
741 		pio.access = IODEV_PIO_READ;
742 		pio.port = sc->psc_bar[baridx].addr + offset;
743 		pio.width = size;
744 		pio.val = 0;
745 
746 		(void)ioctl(iofd, IODEV_PIO, &pio);
747 
748 		val = pio.val;
749 	}
750 
751 	return (val);
752 }
753 
754 struct pci_devemu passthru = {
755 	.pe_emu		= "passthru",
756 	.pe_init	= passthru_init,
757 	.pe_cfgwrite	= passthru_cfgwrite,
758 	.pe_cfgread	= passthru_cfgread,
759 	.pe_barwrite 	= passthru_write,
760 	.pe_barread    	= passthru_read,
761 };
762 PCI_EMUL_SET(passthru);
763