xref: /freebsd/usr.sbin/bhyve/pci_passthru.c (revision 649a910e34b0314647d16a94f8af6de0f4cfd4b5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
29 #include <sys/param.h>
30 #ifndef WITHOUT_CAPSICUM
31 #include <sys/capsicum.h>
32 #endif
33 #include <sys/types.h>
34 #include <sys/mman.h>
35 #include <sys/pciio.h>
36 #include <sys/ioctl.h>
37 #include <sys/stat.h>
38 
39 #include <dev/io/iodev.h>
40 #include <dev/pci/pcireg.h>
41 
42 #include <vm/vm.h>
43 
44 #include <machine/iodev.h>
45 #include <machine/vm.h>
46 
47 #ifndef WITHOUT_CAPSICUM
48 #include <capsicum_helpers.h>
49 #endif
50 #include <ctype.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <err.h>
55 #include <errno.h>
56 #include <fcntl.h>
57 #include <sysexits.h>
58 #include <unistd.h>
59 
60 #include <machine/vmm.h>
61 
62 #include "debug.h"
63 #include "mem.h"
64 #include "pci_passthru.h"
65 
66 #ifndef _PATH_DEVPCI
67 #define	_PATH_DEVPCI	"/dev/pci"
68 #endif
69 
70 #define	LEGACY_SUPPORT	1
71 
72 #define MSIX_TABLE_COUNT(ctrl) (((ctrl) & PCIM_MSIXCTRL_TABLE_SIZE) + 1)
73 #define MSIX_CAPLEN 12
74 
75 #define PASSTHRU_MMIO_MAX 2
76 
77 static int pcifd = -1;
78 
79 SET_DECLARE(passthru_dev_set, struct passthru_dev);
80 
81 struct passthru_softc {
82 	struct pci_devinst *psc_pi;
83 	/* ROM is handled like a BAR */
84 	struct pcibar psc_bar[PCI_BARMAX_WITH_ROM + 1];
85 	struct {
86 		int		capoff;
87 		int		msgctrl;
88 		int		emulated;
89 	} psc_msi;
90 	struct {
91 		int		capoff;
92 	} psc_msix;
93 	struct pcisel psc_sel;
94 
95 	struct passthru_mmio_mapping psc_mmio_map[PASSTHRU_MMIO_MAX];
96 	cfgread_handler psc_pcir_rhandler[PCI_REGMAX + 1];
97 	cfgwrite_handler psc_pcir_whandler[PCI_REGMAX + 1];
98 };
99 
100 static int
msi_caplen(int msgctrl)101 msi_caplen(int msgctrl)
102 {
103 	int len;
104 
105 	len = 10;		/* minimum length of msi capability */
106 
107 	if (msgctrl & PCIM_MSICTRL_64BIT)
108 		len += 4;
109 
110 #if 0
111 	/*
112 	 * Ignore the 'mask' and 'pending' bits in the MSI capability.
113 	 * We'll let the guest manipulate them directly.
114 	 */
115 	if (msgctrl & PCIM_MSICTRL_VECTOR)
116 		len += 10;
117 #endif
118 
119 	return (len);
120 }
121 
122 static int
pcifd_open(void)123 pcifd_open(void)
124 {
125 	int fd;
126 
127 	fd = open(_PATH_DEVPCI, O_RDWR, 0);
128 	if (fd < 0) {
129 		warn("failed to open %s", _PATH_DEVPCI);
130 		return (-1);
131 	}
132 	return (fd);
133 }
134 
135 static int
pcifd_init(void)136 pcifd_init(void)
137 {
138 	pcifd = pcifd_open();
139 	if (pcifd < 0)
140 		return (1);
141 
142 #ifndef WITHOUT_CAPSICUM
143 	cap_rights_t pcifd_rights;
144 	cap_rights_init(&pcifd_rights, CAP_IOCTL, CAP_READ, CAP_WRITE);
145 	if (caph_rights_limit(pcifd, &pcifd_rights) == -1)
146 		errx(EX_OSERR, "Unable to apply rights for sandbox");
147 
148 	const cap_ioctl_t pcifd_ioctls[] = { PCIOCREAD, PCIOCWRITE, PCIOCGETBAR,
149 		PCIOCBARIO, PCIOCBARMMAP, PCIOCGETCONF };
150 	if (caph_ioctls_limit(pcifd, pcifd_ioctls, nitems(pcifd_ioctls)) == -1)
151 		errx(EX_OSERR, "Unable to apply rights for sandbox");
152 #endif
153 
154 	return (0);
155 }
156 
157 static uint32_t
host_read_config(int fd,const struct pcisel * sel,long reg,int width)158 host_read_config(int fd, const struct pcisel *sel, long reg, int width)
159 {
160 	struct pci_io pi;
161 
162 	bzero(&pi, sizeof(pi));
163 	pi.pi_sel = *sel;
164 	pi.pi_reg = reg;
165 	pi.pi_width = width;
166 
167 	if (ioctl(fd, PCIOCREAD, &pi) < 0)
168 		return (0);			/* XXX */
169 	else
170 		return (pi.pi_data);
171 }
172 
173 static uint32_t
passthru_read_config(const struct pcisel * sel,long reg,int width)174 passthru_read_config(const struct pcisel *sel, long reg, int width)
175 {
176 	return (host_read_config(pcifd, sel, reg, width));
177 }
178 
179 uint32_t
pci_host_read_config(const struct pcisel * sel,long reg,int width)180 pci_host_read_config(const struct pcisel *sel, long reg, int width)
181 {
182 	uint32_t ret;
183 	int fd;
184 
185 	fd = pcifd_open();
186 	if (fd < 0)
187 		return (0);
188 	ret = host_read_config(fd, sel, reg, width);
189 	(void)close(fd);
190 	return (ret);
191 }
192 
193 static void
host_write_config(int fd,const struct pcisel * sel,long reg,int width,uint32_t data)194 host_write_config(int fd, const struct pcisel *sel, long reg, int width,
195     uint32_t data)
196 {
197 	struct pci_io pi;
198 
199 	bzero(&pi, sizeof(pi));
200 	pi.pi_sel = *sel;
201 	pi.pi_reg = reg;
202 	pi.pi_width = width;
203 	pi.pi_data = data;
204 
205 	(void)ioctl(fd, PCIOCWRITE, &pi);		/* XXX */
206 }
207 
208 static void
passthru_write_config(const struct pcisel * sel,long reg,int width,uint32_t data)209 passthru_write_config(const struct pcisel *sel, long reg, int width,
210     uint32_t data)
211 {
212 	host_write_config(pcifd, sel, reg, width, data);
213 }
214 
215 void
pci_host_write_config(const struct pcisel * sel,long reg,int width,uint32_t data)216 pci_host_write_config(const struct pcisel *sel, long reg, int width,
217     uint32_t data)
218 {
219 	int fd;
220 
221 	fd = pcifd_open();
222 	if (fd < 0)
223 		return;
224 	host_write_config(fd, sel, reg, width, data);
225 	(void)close(fd);
226 }
227 
228 #ifdef LEGACY_SUPPORT
229 static int
passthru_add_msicap(struct pci_devinst * pi,int msgnum,int nextptr)230 passthru_add_msicap(struct pci_devinst *pi, int msgnum, int nextptr)
231 {
232 	int capoff;
233 	struct msicap msicap;
234 	u_char *capdata;
235 
236 	pci_populate_msicap(&msicap, msgnum, nextptr);
237 
238 	/*
239 	 * XXX
240 	 * Copy the msi capability structure in the last 16 bytes of the
241 	 * config space. This is wrong because it could shadow something
242 	 * useful to the device.
243 	 */
244 	capoff = 256 - roundup(sizeof(msicap), 4);
245 	capdata = (u_char *)&msicap;
246 	for (size_t i = 0; i < sizeof(msicap); i++)
247 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
248 
249 	return (capoff);
250 }
251 #endif	/* LEGACY_SUPPORT */
252 
253 static int
cfginitmsi(struct passthru_softc * sc)254 cfginitmsi(struct passthru_softc *sc)
255 {
256 	int i, ptr, capptr, cap, sts, caplen, table_size;
257 	uint32_t u32;
258 	struct pcisel sel;
259 	struct pci_devinst *pi;
260 	struct msixcap msixcap;
261 	char *msixcap_ptr;
262 
263 	pi = sc->psc_pi;
264 	sel = sc->psc_sel;
265 
266 	/*
267 	 * Parse the capabilities and cache the location of the MSI
268 	 * and MSI-X capabilities.
269 	 */
270 	sts = passthru_read_config(&sel, PCIR_STATUS, 2);
271 	if (sts & PCIM_STATUS_CAPPRESENT) {
272 		ptr = passthru_read_config(&sel, PCIR_CAP_PTR, 1);
273 		while (ptr != 0 && ptr != 0xff) {
274 			cap = passthru_read_config(&sel, ptr + PCICAP_ID, 1);
275 			if (cap == PCIY_MSI) {
276 				/*
277 				 * Copy the MSI capability into the config
278 				 * space of the emulated pci device
279 				 */
280 				sc->psc_msi.capoff = ptr;
281 				sc->psc_msi.msgctrl =
282 				    passthru_read_config(&sel, ptr + 2, 2);
283 				sc->psc_msi.emulated = 0;
284 				caplen = msi_caplen(sc->psc_msi.msgctrl);
285 				capptr = ptr;
286 				while (caplen > 0) {
287 					u32 = passthru_read_config(&sel, capptr,
288 					    4);
289 					pci_set_cfgdata32(pi, capptr, u32);
290 					caplen -= 4;
291 					capptr += 4;
292 				}
293 			} else if (cap == PCIY_MSIX) {
294 				/*
295 				 * Copy the MSI-X capability
296 				 */
297 				sc->psc_msix.capoff = ptr;
298 				caplen = 12;
299 				msixcap_ptr = (char *)&msixcap;
300 				capptr = ptr;
301 				while (caplen > 0) {
302 					u32 = passthru_read_config(&sel, capptr,
303 					    4);
304 					memcpy(msixcap_ptr, &u32, 4);
305 					pci_set_cfgdata32(pi, capptr, u32);
306 					caplen -= 4;
307 					capptr += 4;
308 					msixcap_ptr += 4;
309 				}
310 			}
311 			ptr = passthru_read_config(&sel, ptr + PCICAP_NEXTPTR,
312 			    1);
313 		}
314 	}
315 
316 	if (sc->psc_msix.capoff != 0) {
317 		pi->pi_msix.pba_bar =
318 		    msixcap.pba_info & PCIM_MSIX_BIR_MASK;
319 		pi->pi_msix.pba_offset =
320 		    msixcap.pba_info & ~PCIM_MSIX_BIR_MASK;
321 		pi->pi_msix.table_bar =
322 		    msixcap.table_info & PCIM_MSIX_BIR_MASK;
323 		pi->pi_msix.table_offset =
324 		    msixcap.table_info & ~PCIM_MSIX_BIR_MASK;
325 		pi->pi_msix.table_count = MSIX_TABLE_COUNT(msixcap.msgctrl);
326 		pi->pi_msix.pba_size = PBA_SIZE(pi->pi_msix.table_count);
327 
328 		/* Allocate the emulated MSI-X table array */
329 		table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
330 		pi->pi_msix.table = calloc(1, table_size);
331 
332 		/* Mask all table entries */
333 		for (i = 0; i < pi->pi_msix.table_count; i++) {
334 			pi->pi_msix.table[i].vector_control |=
335 						PCIM_MSIX_VCTRL_MASK;
336 		}
337 	}
338 
339 #ifdef LEGACY_SUPPORT
340 	/*
341 	 * If the passthrough device does not support MSI then craft a
342 	 * MSI capability for it. We link the new MSI capability at the
343 	 * head of the list of capabilities.
344 	 */
345 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0 && sc->psc_msi.capoff == 0) {
346 		int origptr, msiptr;
347 		origptr = passthru_read_config(&sel, PCIR_CAP_PTR, 1);
348 		msiptr = passthru_add_msicap(pi, 1, origptr);
349 		sc->psc_msi.capoff = msiptr;
350 		sc->psc_msi.msgctrl = pci_get_cfgdata16(pi, msiptr + 2);
351 		sc->psc_msi.emulated = 1;
352 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, msiptr);
353 	}
354 #endif
355 
356 	/* Make sure one of the capabilities is present */
357 	if (sc->psc_msi.capoff == 0 && sc->psc_msix.capoff == 0)
358 		return (-1);
359 	else
360 		return (0);
361 }
362 
363 static uint64_t
msix_table_read(struct passthru_softc * sc,uint64_t offset,int size)364 msix_table_read(struct passthru_softc *sc, uint64_t offset, int size)
365 {
366 	struct pci_devinst *pi;
367 	struct msix_table_entry *entry;
368 	uint8_t *src8;
369 	uint16_t *src16;
370 	uint32_t *src32;
371 	uint64_t *src64;
372 	uint64_t data;
373 	size_t entry_offset;
374 	uint32_t table_offset;
375 	int index, table_count;
376 
377 	pi = sc->psc_pi;
378 
379 	table_offset = pi->pi_msix.table_offset;
380 	table_count = pi->pi_msix.table_count;
381 	if (offset < table_offset ||
382 	    offset >= table_offset + table_count * MSIX_TABLE_ENTRY_SIZE) {
383 		switch (size) {
384 		case 1:
385 			src8 = (uint8_t *)(pi->pi_msix.mapped_addr + offset);
386 			data = *src8;
387 			break;
388 		case 2:
389 			src16 = (uint16_t *)(pi->pi_msix.mapped_addr + offset);
390 			data = *src16;
391 			break;
392 		case 4:
393 			src32 = (uint32_t *)(pi->pi_msix.mapped_addr + offset);
394 			data = *src32;
395 			break;
396 		case 8:
397 			src64 = (uint64_t *)(pi->pi_msix.mapped_addr + offset);
398 			data = *src64;
399 			break;
400 		default:
401 			return (-1);
402 		}
403 		return (data);
404 	}
405 
406 	offset -= table_offset;
407 	index = offset / MSIX_TABLE_ENTRY_SIZE;
408 	assert(index < table_count);
409 
410 	entry = &pi->pi_msix.table[index];
411 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
412 
413 	switch (size) {
414 	case 1:
415 		src8 = (uint8_t *)((uint8_t *)entry + entry_offset);
416 		data = *src8;
417 		break;
418 	case 2:
419 		src16 = (uint16_t *)((uint8_t *)entry + entry_offset);
420 		data = *src16;
421 		break;
422 	case 4:
423 		src32 = (uint32_t *)((uint8_t *)entry + entry_offset);
424 		data = *src32;
425 		break;
426 	case 8:
427 		src64 = (uint64_t *)((uint8_t *)entry + entry_offset);
428 		data = *src64;
429 		break;
430 	default:
431 		return (-1);
432 	}
433 
434 	return (data);
435 }
436 
437 static void
msix_table_write(struct passthru_softc * sc,uint64_t offset,int size,uint64_t data)438 msix_table_write(struct passthru_softc *sc, uint64_t offset, int size,
439     uint64_t data)
440 {
441 	struct pci_devinst *pi;
442 	struct msix_table_entry *entry;
443 	uint8_t *dest8;
444 	uint16_t *dest16;
445 	uint32_t *dest32;
446 	uint64_t *dest64;
447 	size_t entry_offset;
448 	uint32_t table_offset, vector_control;
449 	int index, table_count;
450 
451 	pi = sc->psc_pi;
452 
453 	table_offset = pi->pi_msix.table_offset;
454 	table_count = pi->pi_msix.table_count;
455 	if (offset < table_offset ||
456 	    offset >= table_offset + table_count * MSIX_TABLE_ENTRY_SIZE) {
457 		switch (size) {
458 		case 1:
459 			dest8 = (uint8_t *)(pi->pi_msix.mapped_addr + offset);
460 			*dest8 = data;
461 			break;
462 		case 2:
463 			dest16 = (uint16_t *)(pi->pi_msix.mapped_addr + offset);
464 			*dest16 = data;
465 			break;
466 		case 4:
467 			dest32 = (uint32_t *)(pi->pi_msix.mapped_addr + offset);
468 			*dest32 = data;
469 			break;
470 		case 8:
471 			dest64 = (uint64_t *)(pi->pi_msix.mapped_addr + offset);
472 			*dest64 = data;
473 			break;
474 		}
475 		return;
476 	}
477 
478 	offset -= table_offset;
479 	index = offset / MSIX_TABLE_ENTRY_SIZE;
480 	assert(index < table_count);
481 
482 	entry = &pi->pi_msix.table[index];
483 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
484 
485 	/* Only 4 byte naturally-aligned writes are supported */
486 	assert(size == 4);
487 	assert(entry_offset % 4 == 0);
488 
489 	vector_control = entry->vector_control;
490 	dest32 = (uint32_t *)((uint8_t *)entry + entry_offset);
491 	*dest32 = data;
492 	/* If MSI-X hasn't been enabled, do nothing */
493 	if (pi->pi_msix.enabled) {
494 		/* If the entry is masked, don't set it up */
495 		if ((entry->vector_control & PCIM_MSIX_VCTRL_MASK) == 0 ||
496 		    (vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
497 			(void)vm_setup_pptdev_msix(sc->psc_pi->pi_vmctx,
498 			    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
499 			    sc->psc_sel.pc_func, index, entry->addr,
500 			    entry->msg_data, entry->vector_control);
501 		}
502 	}
503 }
504 
505 static int
init_msix_table(struct passthru_softc * sc)506 init_msix_table(struct passthru_softc *sc)
507 {
508 	struct pci_devinst *pi = sc->psc_pi;
509 	struct pci_bar_mmap pbm;
510 	int b, s, f;
511 	uint32_t table_size, table_offset;
512 
513 	assert(pci_msix_table_bar(pi) >= 0 && pci_msix_pba_bar(pi) >= 0);
514 
515 	b = sc->psc_sel.pc_bus;
516 	s = sc->psc_sel.pc_dev;
517 	f = sc->psc_sel.pc_func;
518 
519 	/*
520 	 * Map the region of the BAR containing the MSI-X table.  This is
521 	 * necessary for two reasons:
522 	 * 1. The PBA may reside in the first or last page containing the MSI-X
523 	 *    table.
524 	 * 2. While PCI devices are not supposed to use the page(s) containing
525 	 *    the MSI-X table for other purposes, some do in practice.
526 	 */
527 	memset(&pbm, 0, sizeof(pbm));
528 	pbm.pbm_sel = sc->psc_sel;
529 	pbm.pbm_flags = PCIIO_BAR_MMAP_RW;
530 	pbm.pbm_reg = PCIR_BAR(pi->pi_msix.table_bar);
531 	pbm.pbm_memattr = VM_MEMATTR_DEVICE;
532 
533 	if (ioctl(pcifd, PCIOCBARMMAP, &pbm) != 0) {
534 		warn("Failed to map MSI-X table BAR on %d/%d/%d", b, s, f);
535 		return (-1);
536 	}
537 	assert(pbm.pbm_bar_off == 0);
538 	pi->pi_msix.mapped_addr = (uint8_t *)(uintptr_t)pbm.pbm_map_base;
539 	pi->pi_msix.mapped_size = pbm.pbm_map_length;
540 
541 	table_offset = rounddown2(pi->pi_msix.table_offset, 4096);
542 
543 	table_size = pi->pi_msix.table_offset - table_offset;
544 	table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
545 	table_size = roundup2(table_size, 4096);
546 
547 	/*
548 	 * Unmap any pages not containing the table, we do not need to emulate
549 	 * accesses to them.  Avoid releasing address space to help ensure that
550 	 * a buggy out-of-bounds access causes a crash.
551 	 */
552 	if (table_offset != 0)
553 		if (mprotect(pi->pi_msix.mapped_addr, table_offset,
554 		    PROT_NONE) != 0)
555 			warn("Failed to unmap MSI-X table BAR region");
556 	if (table_offset + table_size != pi->pi_msix.mapped_size)
557 		if (mprotect(
558 		    pi->pi_msix.mapped_addr + table_offset + table_size,
559 		    pi->pi_msix.mapped_size - (table_offset + table_size),
560 		    PROT_NONE) != 0)
561 			warn("Failed to unmap MSI-X table BAR region");
562 
563 	return (0);
564 }
565 
566 static int
cfginitbar(struct passthru_softc * sc)567 cfginitbar(struct passthru_softc *sc)
568 {
569 	int i, error;
570 	struct pci_devinst *pi;
571 	struct pci_bar_io bar;
572 	enum pcibar_type bartype;
573 	uint64_t base, size;
574 
575 	pi = sc->psc_pi;
576 
577 	/*
578 	 * Initialize BAR registers
579 	 */
580 	for (i = 0; i <= PCI_BARMAX; i++) {
581 		uint8_t lobits;
582 
583 		bzero(&bar, sizeof(bar));
584 		bar.pbi_sel = sc->psc_sel;
585 		bar.pbi_reg = PCIR_BAR(i);
586 
587 		if (ioctl(pcifd, PCIOCGETBAR, &bar) < 0)
588 			continue;
589 
590 		if (PCI_BAR_IO(bar.pbi_base)) {
591 			bartype = PCIBAR_IO;
592 			base = bar.pbi_base & PCIM_BAR_IO_BASE;
593 		} else {
594 			switch (bar.pbi_base & PCIM_BAR_MEM_TYPE) {
595 			case PCIM_BAR_MEM_64:
596 				bartype = PCIBAR_MEM64;
597 				break;
598 			default:
599 				bartype = PCIBAR_MEM32;
600 				break;
601 			}
602 			base = bar.pbi_base & PCIM_BAR_MEM_BASE;
603 		}
604 		size = bar.pbi_length;
605 
606 		if (bartype != PCIBAR_IO) {
607 			if (((base | size) & PAGE_MASK) != 0) {
608 				warnx("passthru device %d/%d/%d BAR %d: "
609 				    "base %#lx or size %#lx not page aligned\n",
610 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
611 				    sc->psc_sel.pc_func, i, base, size);
612 				return (-1);
613 			}
614 		}
615 
616 		/* Cache information about the "real" BAR */
617 		sc->psc_bar[i].type = bartype;
618 		sc->psc_bar[i].size = size;
619 		sc->psc_bar[i].addr = base;
620 		sc->psc_bar[i].lobits = 0;
621 
622 		/* Allocate the BAR in the guest I/O or MMIO space */
623 		error = pci_emul_alloc_bar(pi, i, bartype, size);
624 		if (error)
625 			return (-1);
626 
627 		/* Use same lobits as physical bar */
628 		lobits = (uint8_t)passthru_read_config(&sc->psc_sel,
629 		    PCIR_BAR(i), 0x01);
630 		if (bartype == PCIBAR_MEM32 || bartype == PCIBAR_MEM64) {
631 			lobits &= ~PCIM_BAR_MEM_BASE;
632 		} else {
633 			lobits &= ~PCIM_BAR_IO_BASE;
634 		}
635 		sc->psc_bar[i].lobits = lobits;
636 		pi->pi_bar[i].lobits = lobits;
637 
638 		/*
639 		 * 64-bit BAR takes up two slots so skip the next one.
640 		 */
641 		if (bartype == PCIBAR_MEM64) {
642 			i++;
643 			assert(i <= PCI_BARMAX);
644 			sc->psc_bar[i].type = PCIBAR_MEMHI64;
645 		}
646 	}
647 	return (0);
648 }
649 
650 static int
cfginit(struct pci_devinst * pi,int bus,int slot,int func)651 cfginit(struct pci_devinst *pi, int bus, int slot, int func)
652 {
653 	int error;
654 	struct passthru_softc *sc;
655 	uint16_t cmd;
656 	uint8_t intline, intpin;
657 
658 	error = 1;
659 	sc = pi->pi_arg;
660 
661 	bzero(&sc->psc_sel, sizeof(struct pcisel));
662 	sc->psc_sel.pc_bus = bus;
663 	sc->psc_sel.pc_dev = slot;
664 	sc->psc_sel.pc_func = func;
665 
666 	/*
667 	 * Copy physical PCI header to virtual config space.  COMMAND,
668 	 * INTLINE, and INTPIN shouldn't be aligned with their
669 	 * physical value and they are already set by pci_emul_init().
670 	 */
671 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
672 	intline = pci_get_cfgdata8(pi, PCIR_INTLINE);
673 	intpin = pci_get_cfgdata8(pi, PCIR_INTPIN);
674 	for (int i = 0; i <= PCIR_MAXLAT; i += 4) {
675 		pci_set_cfgdata32(pi, i,
676 		    passthru_read_config(&sc->psc_sel, i, 4));
677 	}
678 	pci_set_cfgdata16(pi, PCIR_COMMAND, cmd);
679 	pci_set_cfgdata8(pi, PCIR_INTLINE, intline);
680 	pci_set_cfgdata8(pi, PCIR_INTPIN, intpin);
681 
682 	if (cfginitmsi(sc) != 0) {
683 		warnx("failed to initialize MSI for PCI %d/%d/%d",
684 		    bus, slot, func);
685 		goto done;
686 	}
687 
688 	if (cfginitbar(sc) != 0) {
689 		warnx("failed to initialize BARs for PCI %d/%d/%d",
690 		    bus, slot, func);
691 		goto done;
692 	}
693 
694 	if (pci_msix_table_bar(pi) >= 0) {
695 		error = init_msix_table(sc);
696 		if (error != 0) {
697 			warnx(
698 			    "failed to initialize MSI-X table for PCI %d/%d/%d: %d",
699 			    bus, slot, func, error);
700 			goto done;
701 		}
702 	}
703 
704 	error = 0;				/* success */
705 done:
706 	return (error);
707 }
708 
709 struct passthru_mmio_mapping *
passthru_get_mmio(struct passthru_softc * sc,int num)710 passthru_get_mmio(struct passthru_softc *sc, int num)
711 {
712 	assert(sc != NULL);
713 	assert(num < PASSTHRU_MMIO_MAX);
714 
715 	return (&sc->psc_mmio_map[num]);
716 }
717 
718 struct pcisel *
passthru_get_sel(struct passthru_softc * sc)719 passthru_get_sel(struct passthru_softc *sc)
720 {
721 	assert(sc != NULL);
722 
723 	return (&sc->psc_sel);
724 }
725 
726 int
set_pcir_handler(struct passthru_softc * sc,int reg,int len,cfgread_handler rhandler,cfgwrite_handler whandler)727 set_pcir_handler(struct passthru_softc *sc, int reg, int len,
728     cfgread_handler rhandler, cfgwrite_handler whandler)
729 {
730 	if (reg > PCI_REGMAX || reg + len > PCI_REGMAX + 1)
731 		return (-1);
732 
733 	for (int i = reg; i < reg + len; ++i) {
734 		assert(sc->psc_pcir_rhandler[i] == NULL || rhandler == NULL);
735 		assert(sc->psc_pcir_whandler[i] == NULL || whandler == NULL);
736 		sc->psc_pcir_rhandler[i] = rhandler;
737 		sc->psc_pcir_whandler[i] = whandler;
738 	}
739 
740 	return (0);
741 }
742 
743 static int
passthru_legacy_config(nvlist_t * nvl,const char * opts)744 passthru_legacy_config(nvlist_t *nvl, const char *opts)
745 {
746 	const char *cp;
747 	char *tofree;
748 	char value[16];
749 	int bus, slot, func;
750 
751 	if (opts == NULL)
752 		return (0);
753 
754 	cp = strchr(opts, ',');
755 
756 	if (strncmp(opts, "ppt", strlen("ppt")) == 0) {
757 		tofree = strndup(opts, cp - opts);
758 		set_config_value_node(nvl, "pptdev", tofree);
759 		free(tofree);
760 	} else if (sscanf(opts, "pci0:%d:%d:%d", &bus, &slot, &func) == 3 ||
761 	    sscanf(opts, "pci%d:%d:%d", &bus, &slot, &func) == 3 ||
762 	    sscanf(opts, "%d/%d/%d", &bus, &slot, &func) == 3) {
763 		snprintf(value, sizeof(value), "%d", bus);
764 		set_config_value_node(nvl, "bus", value);
765 		snprintf(value, sizeof(value), "%d", slot);
766 		set_config_value_node(nvl, "slot", value);
767 		snprintf(value, sizeof(value), "%d", func);
768 		set_config_value_node(nvl, "func", value);
769 	} else {
770 		EPRINTLN("passthru: invalid options \"%s\"", opts);
771 		return (-1);
772 	}
773 
774 	if (cp == NULL) {
775 		return (0);
776 	}
777 
778 	return (pci_parse_legacy_config(nvl, cp + 1));
779 }
780 
781 static int
passthru_init_rom(struct passthru_softc * const sc,const char * const romfile)782 passthru_init_rom(struct passthru_softc *const sc, const char *const romfile)
783 {
784 	if (romfile == NULL) {
785 		return (0);
786 	}
787 
788 	const int fd = open(romfile, O_RDONLY);
789 	if (fd < 0) {
790 		warnx("%s: can't open romfile \"%s\"", __func__, romfile);
791 		return (-1);
792 	}
793 
794 	struct stat sbuf;
795 	if (fstat(fd, &sbuf) < 0) {
796 		warnx("%s: can't fstat romfile \"%s\"", __func__, romfile);
797 		close(fd);
798 		return (-1);
799 	}
800 	const uint64_t rom_size = sbuf.st_size;
801 
802 	void *const rom_data = mmap(NULL, rom_size, PROT_READ, MAP_SHARED, fd,
803 	    0);
804 	if (rom_data == MAP_FAILED) {
805 		warnx("%s: unable to mmap romfile \"%s\" (%d)", __func__,
806 		    romfile, errno);
807 		close(fd);
808 		return (-1);
809 	}
810 
811 	void *rom_addr;
812 	int error = pci_emul_alloc_rom(sc->psc_pi, rom_size, &rom_addr);
813 	if (error) {
814 		warnx("%s: failed to alloc rom segment", __func__);
815 		munmap(rom_data, rom_size);
816 		close(fd);
817 		return (error);
818 	}
819 	memcpy(rom_addr, rom_data, rom_size);
820 
821 	sc->psc_bar[PCI_ROM_IDX].type = PCIBAR_ROM;
822 	sc->psc_bar[PCI_ROM_IDX].addr = (uint64_t)rom_addr;
823 	sc->psc_bar[PCI_ROM_IDX].size = rom_size;
824 
825 	munmap(rom_data, rom_size);
826 	close(fd);
827 
828 	return (0);
829 }
830 
831 static bool
passthru_lookup_pptdev(const char * name,int * bus,int * slot,int * func)832 passthru_lookup_pptdev(const char *name, int *bus, int *slot, int *func)
833 {
834 	struct pci_conf_io pc;
835 	struct pci_conf conf[1];
836 	struct pci_match_conf patterns[1];
837 	char *cp;
838 
839 	bzero(&pc, sizeof(struct pci_conf_io));
840 	pc.match_buf_len = sizeof(conf);
841 	pc.matches = conf;
842 
843 	bzero(&patterns, sizeof(patterns));
844 
845 	/*
846 	 * The pattern structure requires the unit to be split out from
847 	 * the driver name.  Walk backwards from the end of the name to
848 	 * find the start of the unit.
849 	 */
850 	cp = strchr(name, '\0');
851 	assert(cp != NULL);
852 	while (cp != name && isdigit(cp[-1]))
853 		cp--;
854 	if (cp == name || !isdigit(*cp)) {
855 		EPRINTLN("Invalid passthru device name %s", name);
856 		return (false);
857 	}
858 	if ((size_t)(cp - name) + 1 > sizeof(patterns[0].pd_name)) {
859 		EPRINTLN("Passthru device name %s is too long", name);
860 		return (false);
861 	}
862 	memcpy(patterns[0].pd_name, name, cp - name);
863 	patterns[0].pd_unit = strtol(cp, &cp, 10);
864 	if (*cp != '\0') {
865 		EPRINTLN("Invalid passthru device name %s", name);
866 		return (false);
867 	}
868 	patterns[0].flags = PCI_GETCONF_MATCH_NAME | PCI_GETCONF_MATCH_UNIT;
869 	pc.num_patterns = 1;
870 	pc.pat_buf_len = sizeof(patterns);
871 	pc.patterns = patterns;
872 
873 	if (ioctl(pcifd, PCIOCGETCONF, &pc) == -1) {
874 		EPRINTLN("ioctl(PCIOCGETCONF): %s", strerror(errno));
875 		return (false);
876 	}
877 	if (pc.status != PCI_GETCONF_LAST_DEVICE &&
878 	    pc.status != PCI_GETCONF_MORE_DEVS) {
879 		EPRINTLN("error returned from PCIOCGETCONF ioctl");
880 		return (false);
881 	}
882 	if (pc.num_matches == 0) {
883 		EPRINTLN("Passthru device %s not found", name);
884 		return (false);
885 	}
886 
887 	if (conf[0].pc_sel.pc_domain != 0) {
888 		EPRINTLN("Passthru device %s on unsupported domain", name);
889 		return (false);
890 	}
891 	*bus = conf[0].pc_sel.pc_bus;
892 	*slot = conf[0].pc_sel.pc_dev;
893 	*func = conf[0].pc_sel.pc_func;
894 	return (true);
895 }
896 
897 static int
passthru_init(struct pci_devinst * pi,nvlist_t * nvl)898 passthru_init(struct pci_devinst *pi, nvlist_t *nvl)
899 {
900 	int bus, slot, func, error, memflags;
901 	struct passthru_softc *sc;
902 	struct passthru_dev **devpp;
903 	struct passthru_dev *devp, *dev = NULL;
904 	const char *value;
905 
906 	sc = NULL;
907 	error = 1;
908 
909 	memflags = vm_get_memflags(pi->pi_vmctx);
910 	if (!(memflags & VM_MEM_F_WIRED)) {
911 		warnx("passthru requires guest memory to be wired");
912 		return (error);
913 	}
914 
915 	if (pcifd < 0 && pcifd_init()) {
916 		return (error);
917 	}
918 
919 #define GET_INT_CONFIG(var, name) do {					\
920 	value = get_config_value_node(nvl, name);			\
921 	if (value == NULL) {						\
922 		EPRINTLN("passthru: missing required %s setting", name); \
923 		return (error);						\
924 	}								\
925 	var = atoi(value);						\
926 } while (0)
927 
928 	value = get_config_value_node(nvl, "pptdev");
929 	if (value != NULL) {
930 		if (!passthru_lookup_pptdev(value, &bus, &slot, &func))
931 			return (error);
932 	} else {
933 		GET_INT_CONFIG(bus, "bus");
934 		GET_INT_CONFIG(slot, "slot");
935 		GET_INT_CONFIG(func, "func");
936 	}
937 
938 	if (vm_assign_pptdev(pi->pi_vmctx, bus, slot, func) != 0) {
939 		warnx("PCI device at %d/%d/%d is not using the ppt(4) driver",
940 		    bus, slot, func);
941 		goto done;
942 	}
943 
944 	sc = calloc(1, sizeof(struct passthru_softc));
945 
946 	pi->pi_arg = sc;
947 	sc->psc_pi = pi;
948 
949 	/* initialize config space */
950 	if ((error = cfginit(pi, bus, slot, func)) != 0)
951 		goto done;
952 
953 	/* initialize ROM */
954 	if ((error = passthru_init_rom(sc,
955             get_config_value_node(nvl, "rom"))) != 0)
956 		goto done;
957 
958 	/* Emulate most PCI header register. */
959 	if ((error = set_pcir_handler(sc, 0, PCIR_MAXLAT + 1,
960 	    passthru_cfgread_emulate, passthru_cfgwrite_emulate)) != 0)
961 		goto done;
962 
963 	/* Allow access to the physical status register. */
964 	if ((error = set_pcir_handler(sc, PCIR_COMMAND, 0x04, NULL, NULL)) != 0)
965 		goto done;
966 
967 	SET_FOREACH(devpp, passthru_dev_set) {
968 		devp = *devpp;
969 		assert(devp->probe != NULL);
970 		if (devp->probe(pi) == 0) {
971 			dev = devp;
972 			break;
973 		}
974 	}
975 
976 	if (dev != NULL) {
977 		error = dev->init(pi, nvl);
978 		if (error != 0)
979 			goto done;
980 	}
981 
982 	error = 0;		/* success */
983 done:
984 	if (error) {
985 		if (dev != NULL)
986 			dev->deinit(pi);
987 		free(sc);
988 		vm_unassign_pptdev(pi->pi_vmctx, bus, slot, func);
989 	}
990 	return (error);
991 }
992 
993 static int
msicap_access(struct passthru_softc * sc,int coff)994 msicap_access(struct passthru_softc *sc, int coff)
995 {
996 	int caplen;
997 
998 	if (sc->psc_msi.capoff == 0)
999 		return (0);
1000 
1001 	caplen = msi_caplen(sc->psc_msi.msgctrl);
1002 
1003 	if (coff >= sc->psc_msi.capoff && coff < sc->psc_msi.capoff + caplen)
1004 		return (1);
1005 	else
1006 		return (0);
1007 }
1008 
1009 static int
msixcap_access(struct passthru_softc * sc,int coff)1010 msixcap_access(struct passthru_softc *sc, int coff)
1011 {
1012 	if (sc->psc_msix.capoff == 0)
1013 		return (0);
1014 
1015 	return (coff >= sc->psc_msix.capoff &&
1016 	        coff < sc->psc_msix.capoff + MSIX_CAPLEN);
1017 }
1018 
1019 static int
passthru_cfgread_default(struct passthru_softc * sc,struct pci_devinst * pi __unused,int coff,int bytes,uint32_t * rv)1020 passthru_cfgread_default(struct passthru_softc *sc,
1021     struct pci_devinst *pi __unused, int coff, int bytes, uint32_t *rv)
1022 {
1023 	/*
1024 	 * MSI capability is emulated.
1025 	 */
1026 	if (msicap_access(sc, coff) || msixcap_access(sc, coff))
1027 		return (-1);
1028 
1029 	/*
1030 	 * Emulate the command register.  If a single read reads both the
1031 	 * command and status registers, read the status register from the
1032 	 * device's config space.
1033 	 */
1034 	if (coff == PCIR_COMMAND) {
1035 		uint32_t st;
1036 
1037 		if (bytes <= 2)
1038 			return (-1);
1039 		st = passthru_read_config(&sc->psc_sel, PCIR_STATUS, 2);
1040 		*rv = (st << 16) | pci_get_cfgdata16(pi, PCIR_COMMAND);
1041 		return (0);
1042 	}
1043 
1044 	/* Everything else just read from the device's config space */
1045 	*rv = passthru_read_config(&sc->psc_sel, coff, bytes);
1046 
1047 	return (0);
1048 }
1049 
1050 int
passthru_cfgread_emulate(struct passthru_softc * sc __unused,struct pci_devinst * pi __unused,int coff __unused,int bytes __unused,uint32_t * rv __unused)1051 passthru_cfgread_emulate(struct passthru_softc *sc __unused,
1052     struct pci_devinst *pi __unused, int coff __unused, int bytes __unused,
1053     uint32_t *rv __unused)
1054 {
1055 	return (-1);
1056 }
1057 
1058 static int
passthru_cfgread(struct pci_devinst * pi,int coff,int bytes,uint32_t * rv)1059 passthru_cfgread(struct pci_devinst *pi, int coff, int bytes, uint32_t *rv)
1060 {
1061 	struct passthru_softc *sc;
1062 
1063 	sc = pi->pi_arg;
1064 
1065 	if (sc->psc_pcir_rhandler[coff] != NULL)
1066 		return (sc->psc_pcir_rhandler[coff](sc, pi, coff, bytes, rv));
1067 
1068 	return (passthru_cfgread_default(sc, pi, coff, bytes, rv));
1069 }
1070 
1071 static int
passthru_cfgwrite_default(struct passthru_softc * sc,struct pci_devinst * pi,int coff,int bytes,uint32_t val)1072 passthru_cfgwrite_default(struct passthru_softc *sc, struct pci_devinst *pi,
1073     int coff, int bytes, uint32_t val)
1074 {
1075 	int error, msix_table_entries, i;
1076 	uint16_t cmd_old;
1077 
1078 	/*
1079 	 * MSI capability is emulated
1080 	 */
1081 	if (msicap_access(sc, coff)) {
1082 		pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msi.capoff,
1083 		    PCIY_MSI);
1084 		error = vm_setup_pptdev_msi(pi->pi_vmctx, sc->psc_sel.pc_bus,
1085 			sc->psc_sel.pc_dev, sc->psc_sel.pc_func,
1086 			pi->pi_msi.addr, pi->pi_msi.msg_data,
1087 			pi->pi_msi.maxmsgnum);
1088 		if (error != 0)
1089 			err(1, "vm_setup_pptdev_msi");
1090 		return (0);
1091 	}
1092 
1093 	if (msixcap_access(sc, coff)) {
1094 		pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msix.capoff,
1095 		    PCIY_MSIX);
1096 		if (pi->pi_msix.enabled) {
1097 			msix_table_entries = pi->pi_msix.table_count;
1098 			for (i = 0; i < msix_table_entries; i++) {
1099 				error = vm_setup_pptdev_msix(pi->pi_vmctx,
1100 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
1101 				    sc->psc_sel.pc_func, i,
1102 				    pi->pi_msix.table[i].addr,
1103 				    pi->pi_msix.table[i].msg_data,
1104 				    pi->pi_msix.table[i].vector_control);
1105 
1106 				if (error)
1107 					err(1, "vm_setup_pptdev_msix");
1108 			}
1109 		} else {
1110 			error = vm_disable_pptdev_msix(pi->pi_vmctx,
1111 			    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
1112 			    sc->psc_sel.pc_func);
1113 			if (error)
1114 				err(1, "vm_disable_pptdev_msix");
1115 		}
1116 		return (0);
1117 	}
1118 
1119 	/*
1120 	 * The command register is emulated, but the status register
1121 	 * is passed through.
1122 	 */
1123 	if (coff == PCIR_COMMAND) {
1124 		if (bytes <= 2)
1125 			return (-1);
1126 
1127 		/* Update the physical status register. */
1128 		passthru_write_config(&sc->psc_sel, PCIR_STATUS, val >> 16, 2);
1129 
1130 		/* Update the virtual command register. */
1131 		cmd_old = pci_get_cfgdata16(pi, PCIR_COMMAND);
1132 		pci_set_cfgdata16(pi, PCIR_COMMAND, val & 0xffff);
1133 		pci_emul_cmd_changed(pi, cmd_old);
1134 		return (0);
1135 	}
1136 
1137 	passthru_write_config(&sc->psc_sel, coff, bytes, val);
1138 
1139 	return (0);
1140 }
1141 
1142 int
passthru_cfgwrite_emulate(struct passthru_softc * sc __unused,struct pci_devinst * pi __unused,int coff __unused,int bytes __unused,uint32_t val __unused)1143 passthru_cfgwrite_emulate(struct passthru_softc *sc __unused,
1144     struct pci_devinst *pi __unused, int coff __unused, int bytes __unused,
1145     uint32_t val __unused)
1146 {
1147 	return (-1);
1148 }
1149 
1150 static int
passthru_cfgwrite(struct pci_devinst * pi,int coff,int bytes,uint32_t val)1151 passthru_cfgwrite(struct pci_devinst *pi, int coff, int bytes, uint32_t val)
1152 {
1153 	struct passthru_softc *sc;
1154 
1155 	sc = pi->pi_arg;
1156 
1157 	if (sc->psc_pcir_whandler[coff] != NULL)
1158 		return (sc->psc_pcir_whandler[coff](sc, pi, coff, bytes, val));
1159 
1160 	return (passthru_cfgwrite_default(sc, pi, coff, bytes, val));
1161 }
1162 
1163 static void
passthru_write(struct pci_devinst * pi,int baridx,uint64_t offset,int size,uint64_t value)1164 passthru_write(struct pci_devinst *pi, int baridx, uint64_t offset, int size,
1165     uint64_t value)
1166 {
1167 	struct passthru_softc *sc;
1168 	struct pci_bar_ioreq pio;
1169 
1170 	sc = pi->pi_arg;
1171 
1172 	if (baridx == pci_msix_table_bar(pi)) {
1173 		msix_table_write(sc, offset, size, value);
1174 	} else {
1175 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
1176 		assert(size == 1 || size == 2 || size == 4);
1177 		assert(offset <= UINT32_MAX && offset + size <= UINT32_MAX);
1178 
1179 		bzero(&pio, sizeof(pio));
1180 		pio.pbi_sel = sc->psc_sel;
1181 		pio.pbi_op = PCIBARIO_WRITE;
1182 		pio.pbi_bar = baridx;
1183 		pio.pbi_offset = (uint32_t)offset;
1184 		pio.pbi_width = size;
1185 		pio.pbi_value = (uint32_t)value;
1186 
1187 		(void)ioctl(pcifd, PCIOCBARIO, &pio);
1188 	}
1189 }
1190 
1191 static uint64_t
passthru_read(struct pci_devinst * pi,int baridx,uint64_t offset,int size)1192 passthru_read(struct pci_devinst *pi, int baridx, uint64_t offset, int size)
1193 {
1194 	struct passthru_softc *sc;
1195 	struct pci_bar_ioreq pio;
1196 	uint64_t val;
1197 
1198 	sc = pi->pi_arg;
1199 
1200 	if (baridx == pci_msix_table_bar(pi)) {
1201 		val = msix_table_read(sc, offset, size);
1202 	} else {
1203 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
1204 		assert(size == 1 || size == 2 || size == 4);
1205 		assert(offset <= UINT32_MAX && offset + size <= UINT32_MAX);
1206 
1207 		bzero(&pio, sizeof(pio));
1208 		pio.pbi_sel = sc->psc_sel;
1209 		pio.pbi_op = PCIBARIO_READ;
1210 		pio.pbi_bar = baridx;
1211 		pio.pbi_offset = (uint32_t)offset;
1212 		pio.pbi_width = size;
1213 
1214 		(void)ioctl(pcifd, PCIOCBARIO, &pio);
1215 
1216 		val = pio.pbi_value;
1217 	}
1218 
1219 	return (val);
1220 }
1221 
1222 static void
passthru_msix_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1223 passthru_msix_addr(struct pci_devinst *pi, int baridx, int enabled,
1224     uint64_t address)
1225 {
1226 	struct passthru_softc *sc;
1227 	size_t remaining;
1228 	uint32_t table_size, table_offset;
1229 
1230 	sc = pi->pi_arg;
1231 	table_offset = rounddown2(pi->pi_msix.table_offset, 4096);
1232 	if (table_offset > 0) {
1233 		if (!enabled) {
1234 			if (vm_unmap_pptdev_mmio(pi->pi_vmctx,
1235 						 sc->psc_sel.pc_bus,
1236 						 sc->psc_sel.pc_dev,
1237 						 sc->psc_sel.pc_func, address,
1238 						 table_offset) != 0)
1239 				warnx("pci_passthru: unmap_pptdev_mmio failed");
1240 		} else {
1241 			if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1242 					       sc->psc_sel.pc_dev,
1243 					       sc->psc_sel.pc_func, address,
1244 					       table_offset,
1245 					       sc->psc_bar[baridx].addr) != 0)
1246 				warnx("pci_passthru: map_pptdev_mmio failed");
1247 		}
1248 	}
1249 	table_size = pi->pi_msix.table_offset - table_offset;
1250 	table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
1251 	table_size = roundup2(table_size, 4096);
1252 	remaining = pi->pi_bar[baridx].size - table_offset - table_size;
1253 	if (remaining > 0) {
1254 		address += table_offset + table_size;
1255 		if (!enabled) {
1256 			if (vm_unmap_pptdev_mmio(pi->pi_vmctx,
1257 						 sc->psc_sel.pc_bus,
1258 						 sc->psc_sel.pc_dev,
1259 						 sc->psc_sel.pc_func, address,
1260 						 remaining) != 0)
1261 				warnx("pci_passthru: unmap_pptdev_mmio failed");
1262 		} else {
1263 			if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1264 					       sc->psc_sel.pc_dev,
1265 					       sc->psc_sel.pc_func, address,
1266 					       remaining,
1267 					       sc->psc_bar[baridx].addr +
1268 					       table_offset + table_size) != 0)
1269 				warnx("pci_passthru: map_pptdev_mmio failed");
1270 		}
1271 	}
1272 }
1273 
1274 static void
passthru_mmio_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1275 passthru_mmio_addr(struct pci_devinst *pi, int baridx, int enabled,
1276     uint64_t address)
1277 {
1278 	struct passthru_softc *sc;
1279 
1280 	sc = pi->pi_arg;
1281 	if (!enabled) {
1282 		if (vm_unmap_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1283 					 sc->psc_sel.pc_dev,
1284 					 sc->psc_sel.pc_func, address,
1285 					 sc->psc_bar[baridx].size) != 0)
1286 			warnx("pci_passthru: unmap_pptdev_mmio failed");
1287 	} else {
1288 		if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1289 				       sc->psc_sel.pc_dev,
1290 				       sc->psc_sel.pc_func, address,
1291 				       sc->psc_bar[baridx].size,
1292 				       sc->psc_bar[baridx].addr) != 0)
1293 			warnx("pci_passthru: map_pptdev_mmio failed");
1294 	}
1295 }
1296 
1297 static void
passthru_addr_rom(struct pci_devinst * const pi,const int idx,const int enabled)1298 passthru_addr_rom(struct pci_devinst *const pi, const int idx,
1299     const int enabled)
1300 {
1301 	const uint64_t addr = pi->pi_bar[idx].addr;
1302 	const uint64_t size = pi->pi_bar[idx].size;
1303 
1304 	if (!enabled) {
1305 		if (vm_munmap_memseg(pi->pi_vmctx, addr, size) != 0) {
1306 			errx(4, "%s: munmap_memseg @ [%016lx - %016lx] failed",
1307 			    __func__, addr, addr + size);
1308 		}
1309 
1310 	} else {
1311 		if (vm_mmap_memseg(pi->pi_vmctx, addr, VM_PCIROM,
1312 			pi->pi_romoffset, size, PROT_READ | PROT_EXEC) != 0) {
1313 			errx(4, "%s: mmap_memseg @ [%016lx - %016lx]  failed",
1314 			    __func__, addr, addr + size);
1315 		}
1316 	}
1317 }
1318 
1319 static void
passthru_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1320 passthru_addr(struct pci_devinst *pi, int baridx, int enabled, uint64_t address)
1321 {
1322 	switch (pi->pi_bar[baridx].type) {
1323 	case PCIBAR_IO:
1324 		/* IO BARs are emulated */
1325 		break;
1326 	case PCIBAR_ROM:
1327 		passthru_addr_rom(pi, baridx, enabled);
1328 		break;
1329 	case PCIBAR_MEM32:
1330 	case PCIBAR_MEM64:
1331 		if (baridx == pci_msix_table_bar(pi))
1332 			passthru_msix_addr(pi, baridx, enabled, address);
1333 		else
1334 			passthru_mmio_addr(pi, baridx, enabled, address);
1335 		break;
1336 	default:
1337 		errx(4, "%s: invalid BAR type %d", __func__,
1338 		    pi->pi_bar[baridx].type);
1339 	}
1340 }
1341 
1342 static const struct pci_devemu passthru = {
1343 	.pe_emu		= "passthru",
1344 	.pe_init	= passthru_init,
1345 	.pe_legacy_config = passthru_legacy_config,
1346 	.pe_cfgwrite	= passthru_cfgwrite,
1347 	.pe_cfgread	= passthru_cfgread,
1348 	.pe_barwrite 	= passthru_write,
1349 	.pe_barread    	= passthru_read,
1350 	.pe_baraddr	= passthru_addr,
1351 };
1352 PCI_EMUL_SET(passthru);
1353