xref: /freebsd/usr.sbin/bhyve/pci_passthru.c (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
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 		pi->pi_msix.pba_size = PBA_SIZE(pi->pi_msix.table_count);
232 
233 		/* Allocate the emulated MSI-X table array */
234 		table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
235 		pi->pi_msix.table = malloc(table_size);
236 		bzero(pi->pi_msix.table, table_size);
237 
238 		/* Mask all table entries */
239 		for (i = 0; i < pi->pi_msix.table_count; i++) {
240 			pi->pi_msix.table[i].vector_control |=
241 						PCIM_MSIX_VCTRL_MASK;
242 		}
243 	}
244 
245 #ifdef LEGACY_SUPPORT
246 	/*
247 	 * If the passthrough device does not support MSI then craft a
248 	 * MSI capability for it. We link the new MSI capability at the
249 	 * head of the list of capabilities.
250 	 */
251 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0 && sc->psc_msi.capoff == 0) {
252 		int origptr, msiptr;
253 		origptr = read_config(&sel, PCIR_CAP_PTR, 1);
254 		msiptr = passthru_add_msicap(pi, 1, origptr);
255 		sc->psc_msi.capoff = msiptr;
256 		sc->psc_msi.msgctrl = pci_get_cfgdata16(pi, msiptr + 2);
257 		sc->psc_msi.emulated = 1;
258 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, msiptr);
259 	}
260 #endif
261 
262 	/* Make sure one of the capabilities is present */
263 	if (sc->psc_msi.capoff == 0 && sc->psc_msix.capoff == 0)
264 		return (-1);
265 	else
266 		return (0);
267 }
268 
269 static uint64_t
270 msix_table_read(struct passthru_softc *sc, uint64_t offset, int size)
271 {
272 	struct pci_devinst *pi;
273 	struct msix_table_entry *entry;
274 	uint8_t *src8;
275 	uint16_t *src16;
276 	uint32_t *src32;
277 	uint64_t *src64;
278 	uint64_t data;
279 	size_t entry_offset;
280 	int index;
281 
282 	pi = sc->psc_pi;
283 	if (offset < pi->pi_msix.table_offset)
284 		return (-1);
285 
286 	offset -= pi->pi_msix.table_offset;
287 	index = offset / MSIX_TABLE_ENTRY_SIZE;
288 	if (index >= pi->pi_msix.table_count)
289 		return (-1);
290 
291 	entry = &pi->pi_msix.table[index];
292 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
293 
294 	switch(size) {
295 	case 1:
296 		src8 = (uint8_t *)((void *)entry + entry_offset);
297 		data = *src8;
298 		break;
299 	case 2:
300 		src16 = (uint16_t *)((void *)entry + entry_offset);
301 		data = *src16;
302 		break;
303 	case 4:
304 		src32 = (uint32_t *)((void *)entry + entry_offset);
305 		data = *src32;
306 		break;
307 	case 8:
308 		src64 = (uint64_t *)((void *)entry + entry_offset);
309 		data = *src64;
310 		break;
311 	default:
312 		return (-1);
313 	}
314 
315 	return (data);
316 }
317 
318 static void
319 msix_table_write(struct vmctx *ctx, int vcpu, struct passthru_softc *sc,
320 		 uint64_t offset, int size, uint64_t data)
321 {
322 	struct pci_devinst *pi;
323 	struct msix_table_entry *entry;
324 	uint32_t *dest;
325 	size_t entry_offset;
326 	uint32_t vector_control;
327 	int error, index;
328 
329 	pi = sc->psc_pi;
330 	if (offset < pi->pi_msix.table_offset)
331 		return;
332 
333 	offset -= pi->pi_msix.table_offset;
334 	index = offset / MSIX_TABLE_ENTRY_SIZE;
335 	if (index >= pi->pi_msix.table_count)
336 		return;
337 
338 	entry = &pi->pi_msix.table[index];
339 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
340 
341 	/* Only 4 byte naturally-aligned writes are supported */
342 	assert(size == 4);
343 	assert(entry_offset % 4 == 0);
344 
345 	vector_control = entry->vector_control;
346 	dest = (uint32_t *)((void *)entry + entry_offset);
347 	*dest = data;
348 	/* If MSI-X hasn't been enabled, do nothing */
349 	if (pi->pi_msix.enabled) {
350 		/* If the entry is masked, don't set it up */
351 		if ((entry->vector_control & PCIM_MSIX_VCTRL_MASK) == 0 ||
352 		    (vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
353 			error = vm_setup_pptdev_msix(ctx, vcpu,
354 			    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
355 			    sc->psc_sel.pc_func, index, entry->addr,
356 			    entry->msg_data, entry->vector_control);
357 		}
358 	}
359 }
360 
361 static int
362 init_msix_table(struct vmctx *ctx, struct passthru_softc *sc, uint64_t base)
363 {
364 	int b, s, f;
365 	int error, idx;
366 	size_t len, remaining;
367 	uint32_t table_size, table_offset;
368 	uint32_t pba_size, pba_offset;
369 	vm_paddr_t start;
370 	struct pci_devinst *pi = sc->psc_pi;
371 
372 	assert(pci_msix_table_bar(pi) >= 0 && pci_msix_pba_bar(pi) >= 0);
373 
374 	b = sc->psc_sel.pc_bus;
375 	s = sc->psc_sel.pc_dev;
376 	f = sc->psc_sel.pc_func;
377 
378 	/*
379 	 * If the MSI-X table BAR maps memory intended for
380 	 * other uses, it is at least assured that the table
381 	 * either resides in its own page within the region,
382 	 * or it resides in a page shared with only the PBA.
383 	 */
384 	table_offset = rounddown2(pi->pi_msix.table_offset, 4096);
385 
386 	table_size = pi->pi_msix.table_offset - table_offset;
387 	table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
388 	table_size = roundup2(table_size, 4096);
389 
390 	if (pi->pi_msix.pba_bar == pi->pi_msix.table_bar) {
391 		pba_offset = pi->pi_msix.pba_offset;
392 		pba_size = pi->pi_msix.pba_size;
393 		if (pba_offset >= table_offset + table_size ||
394 		    table_offset >= pba_offset + pba_size) {
395 			/*
396 			 * The PBA can reside in the same BAR as the MSI-x
397 			 * tables as long as it does not overlap with any
398 			 * naturally aligned page occupied by the tables.
399 			 */
400 		} else {
401 			/* Need to also emulate the PBA, not supported yet */
402 			printf("Unsupported MSI-X configuration: %d/%d/%d\n",
403 		            b, s, f);
404 			return (-1);
405 		}
406 	}
407 
408 	idx = pi->pi_msix.table_bar;
409 	start = pi->pi_bar[idx].addr;
410 	remaining = pi->pi_bar[idx].size;
411 
412 	/* Map everything before the MSI-X table */
413 	if (table_offset > 0) {
414 		len = table_offset;
415 		error = vm_map_pptdev_mmio(ctx, b, s, f, start, len, base);
416 		if (error)
417 			return (error);
418 
419 		base += len;
420 		start += len;
421 		remaining -= len;
422 	}
423 
424 	/* Skip the MSI-X table */
425 	base += table_size;
426 	start += table_size;
427 	remaining -= table_size;
428 
429 	/* Map everything beyond the end of the MSI-X table */
430 	if (remaining > 0) {
431 		len = remaining;
432 		error = vm_map_pptdev_mmio(ctx, b, s, f, start, len, base);
433 		if (error)
434 			return (error);
435 	}
436 
437 	return (0);
438 }
439 
440 static int
441 cfginitbar(struct vmctx *ctx, struct passthru_softc *sc)
442 {
443 	int i, error;
444 	struct pci_devinst *pi;
445 	struct pci_bar_io bar;
446 	enum pcibar_type bartype;
447 	uint64_t base, size;
448 
449 	pi = sc->psc_pi;
450 
451 	/*
452 	 * Initialize BAR registers
453 	 */
454 	for (i = 0; i <= PCI_BARMAX; i++) {
455 		bzero(&bar, sizeof(bar));
456 		bar.pbi_sel = sc->psc_sel;
457 		bar.pbi_reg = PCIR_BAR(i);
458 
459 		if (ioctl(pcifd, PCIOCGETBAR, &bar) < 0)
460 			continue;
461 
462 		if (PCI_BAR_IO(bar.pbi_base)) {
463 			bartype = PCIBAR_IO;
464 			base = bar.pbi_base & PCIM_BAR_IO_BASE;
465 		} else {
466 			switch (bar.pbi_base & PCIM_BAR_MEM_TYPE) {
467 			case PCIM_BAR_MEM_64:
468 				bartype = PCIBAR_MEM64;
469 				break;
470 			default:
471 				bartype = PCIBAR_MEM32;
472 				break;
473 			}
474 			base = bar.pbi_base & PCIM_BAR_MEM_BASE;
475 		}
476 		size = bar.pbi_length;
477 
478 		if (bartype != PCIBAR_IO) {
479 			if (((base | size) & PAGE_MASK) != 0) {
480 				printf("passthru device %d/%d/%d BAR %d: "
481 				    "base %#lx or size %#lx not page aligned\n",
482 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
483 				    sc->psc_sel.pc_func, i, base, size);
484 				return (-1);
485 			}
486 		}
487 
488 		/* Cache information about the "real" BAR */
489 		sc->psc_bar[i].type = bartype;
490 		sc->psc_bar[i].size = size;
491 		sc->psc_bar[i].addr = base;
492 
493 		/* Allocate the BAR in the guest I/O or MMIO space */
494 		error = pci_emul_alloc_pbar(pi, i, base, bartype, size);
495 		if (error)
496 			return (-1);
497 
498 		/* The MSI-X table needs special handling */
499 		if (i == pci_msix_table_bar(pi)) {
500 			error = init_msix_table(ctx, sc, base);
501 			if (error)
502 				return (-1);
503 		} else if (bartype != PCIBAR_IO) {
504 			/* Map the physical BAR in the guest MMIO space */
505 			error = vm_map_pptdev_mmio(ctx, sc->psc_sel.pc_bus,
506 				sc->psc_sel.pc_dev, sc->psc_sel.pc_func,
507 				pi->pi_bar[i].addr, pi->pi_bar[i].size, base);
508 			if (error)
509 				return (-1);
510 		}
511 
512 		/*
513 		 * 64-bit BAR takes up two slots so skip the next one.
514 		 */
515 		if (bartype == PCIBAR_MEM64) {
516 			i++;
517 			assert(i <= PCI_BARMAX);
518 			sc->psc_bar[i].type = PCIBAR_MEMHI64;
519 		}
520 	}
521 	return (0);
522 }
523 
524 static int
525 cfginit(struct vmctx *ctx, struct pci_devinst *pi, int bus, int slot, int func)
526 {
527 	int error;
528 	struct passthru_softc *sc;
529 
530 	error = 1;
531 	sc = pi->pi_arg;
532 
533 	bzero(&sc->psc_sel, sizeof(struct pcisel));
534 	sc->psc_sel.pc_bus = bus;
535 	sc->psc_sel.pc_dev = slot;
536 	sc->psc_sel.pc_func = func;
537 
538 	if (cfginitmsi(sc) != 0)
539 		goto done;
540 
541 	if (cfginitbar(ctx, sc) != 0)
542 		goto done;
543 
544 	error = 0;				/* success */
545 done:
546 	return (error);
547 }
548 
549 static int
550 passthru_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
551 {
552 	int bus, slot, func, error;
553 	struct passthru_softc *sc;
554 
555 	sc = NULL;
556 	error = 1;
557 
558 	if (pcifd < 0) {
559 		pcifd = open(_PATH_DEVPCI, O_RDWR, 0);
560 		if (pcifd < 0)
561 			goto done;
562 	}
563 
564 	if (iofd < 0) {
565 		iofd = open(_PATH_DEVIO, O_RDWR, 0);
566 		if (iofd < 0)
567 			goto done;
568 	}
569 
570 	if (opts == NULL ||
571 	    sscanf(opts, "%d/%d/%d", &bus, &slot, &func) != 3)
572 		goto done;
573 
574 	if (vm_assign_pptdev(ctx, bus, slot, func) != 0)
575 		goto done;
576 
577 	sc = malloc(sizeof(struct passthru_softc));
578 	memset(sc, 0, sizeof(struct passthru_softc));
579 
580 	pi->pi_arg = sc;
581 	sc->psc_pi = pi;
582 
583 	/* initialize config space */
584 	if ((error = cfginit(ctx, pi, bus, slot, func)) != 0)
585 		goto done;
586 
587 	error = 0;		/* success */
588 done:
589 	if (error) {
590 		free(sc);
591 		vm_unassign_pptdev(ctx, bus, slot, func);
592 	}
593 	return (error);
594 }
595 
596 static int
597 bar_access(int coff)
598 {
599 	if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1))
600 		return (1);
601 	else
602 		return (0);
603 }
604 
605 static int
606 msicap_access(struct passthru_softc *sc, int coff)
607 {
608 	int caplen;
609 
610 	if (sc->psc_msi.capoff == 0)
611 		return (0);
612 
613 	caplen = msi_caplen(sc->psc_msi.msgctrl);
614 
615 	if (coff >= sc->psc_msi.capoff && coff < sc->psc_msi.capoff + caplen)
616 		return (1);
617 	else
618 		return (0);
619 }
620 
621 static int
622 msixcap_access(struct passthru_softc *sc, int coff)
623 {
624 	if (sc->psc_msix.capoff == 0)
625 		return (0);
626 
627 	return (coff >= sc->psc_msix.capoff &&
628 	        coff < sc->psc_msix.capoff + MSIX_CAPLEN);
629 }
630 
631 static int
632 passthru_cfgread(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
633 		 int coff, int bytes, uint32_t *rv)
634 {
635 	struct passthru_softc *sc;
636 
637 	sc = pi->pi_arg;
638 
639 	/*
640 	 * PCI BARs and MSI capability is emulated.
641 	 */
642 	if (bar_access(coff) || msicap_access(sc, coff))
643 		return (-1);
644 
645 #ifdef LEGACY_SUPPORT
646 	/*
647 	 * Emulate PCIR_CAP_PTR if this device does not support MSI capability
648 	 * natively.
649 	 */
650 	if (sc->psc_msi.emulated) {
651 		if (coff >= PCIR_CAP_PTR && coff < PCIR_CAP_PTR + 4)
652 			return (-1);
653 	}
654 #endif
655 
656 	/* Everything else just read from the device's config space */
657 	*rv = read_config(&sc->psc_sel, coff, bytes);
658 
659 	return (0);
660 }
661 
662 static int
663 passthru_cfgwrite(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
664 		  int coff, int bytes, uint32_t val)
665 {
666 	int error, msix_table_entries, i;
667 	struct passthru_softc *sc;
668 
669 	sc = pi->pi_arg;
670 
671 	/*
672 	 * PCI BARs are emulated
673 	 */
674 	if (bar_access(coff))
675 		return (-1);
676 
677 	/*
678 	 * MSI capability is emulated
679 	 */
680 	if (msicap_access(sc, coff)) {
681 		msicap_cfgwrite(pi, sc->psc_msi.capoff, coff, bytes, val);
682 
683 		error = vm_setup_pptdev_msi(ctx, vcpu, sc->psc_sel.pc_bus,
684 			sc->psc_sel.pc_dev, sc->psc_sel.pc_func,
685 			pi->pi_msi.addr, pi->pi_msi.msg_data,
686 			pi->pi_msi.maxmsgnum);
687 		if (error != 0) {
688 			printf("vm_setup_pptdev_msi error %d\r\n", errno);
689 			exit(1);
690 		}
691 		return (0);
692 	}
693 
694 	if (msixcap_access(sc, coff)) {
695 		msixcap_cfgwrite(pi, sc->psc_msix.capoff, coff, bytes, val);
696 		if (pi->pi_msix.enabled) {
697 			msix_table_entries = pi->pi_msix.table_count;
698 			for (i = 0; i < msix_table_entries; i++) {
699 				error = vm_setup_pptdev_msix(ctx, vcpu,
700 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
701 				    sc->psc_sel.pc_func, i,
702 				    pi->pi_msix.table[i].addr,
703 				    pi->pi_msix.table[i].msg_data,
704 				    pi->pi_msix.table[i].vector_control);
705 
706 				if (error) {
707 					printf("vm_setup_pptdev_msix error "
708 					    "%d\r\n", errno);
709 					exit(1);
710 				}
711 			}
712 		}
713 		return (0);
714 	}
715 
716 #ifdef LEGACY_SUPPORT
717 	/*
718 	 * If this device does not support MSI natively then we cannot let
719 	 * the guest disable legacy interrupts from the device. It is the
720 	 * legacy interrupt that is triggering the virtual MSI to the guest.
721 	 */
722 	if (sc->psc_msi.emulated && pci_msi_enabled(pi)) {
723 		if (coff == PCIR_COMMAND && bytes == 2)
724 			val &= ~PCIM_CMD_INTxDIS;
725 	}
726 #endif
727 
728 	write_config(&sc->psc_sel, coff, bytes, val);
729 
730 	return (0);
731 }
732 
733 static void
734 passthru_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
735 	       uint64_t offset, int size, uint64_t value)
736 {
737 	struct passthru_softc *sc;
738 	struct iodev_pio_req pio;
739 
740 	sc = pi->pi_arg;
741 
742 	if (baridx == pci_msix_table_bar(pi)) {
743 		msix_table_write(ctx, vcpu, sc, offset, size, value);
744 	} else {
745 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
746 		bzero(&pio, sizeof(struct iodev_pio_req));
747 		pio.access = IODEV_PIO_WRITE;
748 		pio.port = sc->psc_bar[baridx].addr + offset;
749 		pio.width = size;
750 		pio.val = value;
751 
752 		(void)ioctl(iofd, IODEV_PIO, &pio);
753 	}
754 }
755 
756 static uint64_t
757 passthru_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
758 	      uint64_t offset, int size)
759 {
760 	struct passthru_softc *sc;
761 	struct iodev_pio_req pio;
762 	uint64_t val;
763 
764 	sc = pi->pi_arg;
765 
766 	if (baridx == pci_msix_table_bar(pi)) {
767 		val = msix_table_read(sc, offset, size);
768 	} else {
769 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
770 		bzero(&pio, sizeof(struct iodev_pio_req));
771 		pio.access = IODEV_PIO_READ;
772 		pio.port = sc->psc_bar[baridx].addr + offset;
773 		pio.width = size;
774 		pio.val = 0;
775 
776 		(void)ioctl(iofd, IODEV_PIO, &pio);
777 
778 		val = pio.val;
779 	}
780 
781 	return (val);
782 }
783 
784 struct pci_devemu passthru = {
785 	.pe_emu		= "passthru",
786 	.pe_init	= passthru_init,
787 	.pe_cfgwrite	= passthru_cfgwrite,
788 	.pe_cfgread	= passthru_cfgread,
789 	.pe_barwrite 	= passthru_write,
790 	.pe_barread    	= passthru_read,
791 };
792 PCI_EMUL_SET(passthru);
793