xref: /freebsd/usr.sbin/bhyve/pci_passthru.c (revision f1d705d4f431a5553aa5e64310bdc703ba27eab4)
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 #include <dev/vmm/vmm_mem.h>
42 
43 #include <vm/vm.h>
44 
45 #include <machine/iodev.h>
46 #include <machine/vm.h>
47 
48 #ifndef WITHOUT_CAPSICUM
49 #include <capsicum_helpers.h>
50 #endif
51 #include <ctype.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <string.h>
55 #include <err.h>
56 #include <errno.h>
57 #include <fcntl.h>
58 #include <sysexits.h>
59 #include <unistd.h>
60 
61 #include <machine/vmm.h>
62 
63 #include "debug.h"
64 #include "mem.h"
65 #include "pci_passthru.h"
66 
67 #ifndef _PATH_DEVPCI
68 #define	_PATH_DEVPCI	"/dev/pci"
69 #endif
70 
71 #define	LEGACY_SUPPORT	1
72 
73 #define MSIX_TABLE_COUNT(ctrl) (((ctrl) & PCIM_MSIXCTRL_TABLE_SIZE) + 1)
74 #define MSIX_CAPLEN 12
75 
76 #define PASSTHRU_MMIO_MAX 3
77 
78 static int pcifd = -1;
79 
80 SET_DECLARE(passthru_dev_set, struct passthru_dev);
81 
82 struct passthru_softc {
83 	struct pci_devinst *psc_pi;
84 	/* ROM is handled like a BAR */
85 	struct pcibar psc_bar[PCI_BARMAX_WITH_ROM + 1];
86 	struct {
87 		int		capoff;
88 		int		msgctrl;
89 		int		emulated;
90 	} psc_msi;
91 	struct {
92 		int		capoff;
93 	} psc_msix;
94 	struct pcisel psc_sel;
95 
96 	struct passthru_mmio_mapping psc_mmio_map[PASSTHRU_MMIO_MAX];
97 	cfgread_handler psc_pcir_rhandler[PCI_REGMAX + 1];
98 	cfgwrite_handler psc_pcir_whandler[PCI_REGMAX + 1];
99 };
100 
101 static int
msi_caplen(int msgctrl)102 msi_caplen(int msgctrl)
103 {
104 	int len;
105 
106 	len = 10;		/* minimum length of msi capability */
107 
108 	if (msgctrl & PCIM_MSICTRL_64BIT)
109 		len += 4;
110 
111 #if 0
112 	/*
113 	 * Ignore the 'mask' and 'pending' bits in the MSI capability.
114 	 * We'll let the guest manipulate them directly.
115 	 */
116 	if (msgctrl & PCIM_MSICTRL_VECTOR)
117 		len += 10;
118 #endif
119 
120 	return (len);
121 }
122 
123 static int
pcifd_open(void)124 pcifd_open(void)
125 {
126 	int fd;
127 
128 	fd = open(_PATH_DEVPCI, O_RDWR, 0);
129 	if (fd < 0) {
130 		warn("failed to open %s", _PATH_DEVPCI);
131 		return (-1);
132 	}
133 	return (fd);
134 }
135 
136 static int
pcifd_init(void)137 pcifd_init(void)
138 {
139 	pcifd = pcifd_open();
140 	if (pcifd < 0)
141 		return (1);
142 
143 #ifndef WITHOUT_CAPSICUM
144 	cap_rights_t pcifd_rights;
145 	cap_rights_init(&pcifd_rights, CAP_IOCTL, CAP_READ, CAP_WRITE);
146 	if (caph_rights_limit(pcifd, &pcifd_rights) == -1)
147 		errx(EX_OSERR, "Unable to apply rights for sandbox");
148 
149 	const cap_ioctl_t pcifd_ioctls[] = { PCIOCREAD, PCIOCWRITE, PCIOCGETBAR,
150 		PCIOCBARIO, PCIOCBARMMAP, PCIOCGETCONF };
151 	if (caph_ioctls_limit(pcifd, pcifd_ioctls, nitems(pcifd_ioctls)) == -1)
152 		errx(EX_OSERR, "Unable to apply rights for sandbox");
153 #endif
154 
155 	return (0);
156 }
157 
158 static uint32_t
host_read_config(int fd,const struct pcisel * sel,long reg,int width)159 host_read_config(int fd, const struct pcisel *sel, long reg, int width)
160 {
161 	struct pci_io pi;
162 
163 	bzero(&pi, sizeof(pi));
164 	pi.pi_sel = *sel;
165 	pi.pi_reg = reg;
166 	pi.pi_width = width;
167 
168 	if (ioctl(fd, PCIOCREAD, &pi) < 0)
169 		return (0);			/* XXX */
170 	else
171 		return (pi.pi_data);
172 }
173 
174 static uint32_t
passthru_read_config(const struct pcisel * sel,long reg,int width)175 passthru_read_config(const struct pcisel *sel, long reg, int width)
176 {
177 	return (host_read_config(pcifd, sel, reg, width));
178 }
179 
180 uint32_t
pci_host_read_config(const struct pcisel * sel,long reg,int width)181 pci_host_read_config(const struct pcisel *sel, long reg, int width)
182 {
183 	uint32_t ret;
184 	int fd;
185 
186 	fd = pcifd_open();
187 	if (fd < 0)
188 		return (0);
189 	ret = host_read_config(fd, sel, reg, width);
190 	(void)close(fd);
191 	return (ret);
192 }
193 
194 static void
host_write_config(int fd,const struct pcisel * sel,long reg,int width,uint32_t data)195 host_write_config(int fd, const struct pcisel *sel, long reg, int width,
196     uint32_t data)
197 {
198 	struct pci_io pi;
199 
200 	bzero(&pi, sizeof(pi));
201 	pi.pi_sel = *sel;
202 	pi.pi_reg = reg;
203 	pi.pi_width = width;
204 	pi.pi_data = data;
205 
206 	(void)ioctl(fd, PCIOCWRITE, &pi);		/* XXX */
207 }
208 
209 static void
passthru_write_config(const struct pcisel * sel,long reg,int width,uint32_t data)210 passthru_write_config(const struct pcisel *sel, long reg, int width,
211     uint32_t data)
212 {
213 	host_write_config(pcifd, sel, reg, width, data);
214 }
215 
216 void
pci_host_write_config(const struct pcisel * sel,long reg,int width,uint32_t data)217 pci_host_write_config(const struct pcisel *sel, long reg, int width,
218     uint32_t data)
219 {
220 	int fd;
221 
222 	fd = pcifd_open();
223 	if (fd < 0)
224 		return;
225 	host_write_config(fd, sel, reg, width, data);
226 	(void)close(fd);
227 }
228 
229 #ifdef LEGACY_SUPPORT
230 static int
passthru_add_msicap(struct pci_devinst * pi,int msgnum,int nextptr)231 passthru_add_msicap(struct pci_devinst *pi, int msgnum, int nextptr)
232 {
233 	int capoff;
234 	struct msicap msicap;
235 	u_char *capdata;
236 
237 	pci_populate_msicap(&msicap, msgnum, nextptr);
238 
239 	/*
240 	 * XXX
241 	 * Copy the msi capability structure in the last 16 bytes of the
242 	 * config space. This is wrong because it could shadow something
243 	 * useful to the device.
244 	 */
245 	capoff = 256 - roundup(sizeof(msicap), 4);
246 	capdata = (u_char *)&msicap;
247 	for (size_t i = 0; i < sizeof(msicap); i++)
248 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
249 
250 	return (capoff);
251 }
252 #endif	/* LEGACY_SUPPORT */
253 
254 static int
cfginitmsi(struct passthru_softc * sc)255 cfginitmsi(struct passthru_softc *sc)
256 {
257 	int i, ptr, capptr, cap, sts, caplen, table_size;
258 	uint32_t u32;
259 	struct pcisel sel;
260 	struct pci_devinst *pi;
261 	struct msixcap msixcap;
262 	char *msixcap_ptr;
263 
264 	pi = sc->psc_pi;
265 	sel = sc->psc_sel;
266 
267 	/*
268 	 * Parse the capabilities and cache the location of the MSI
269 	 * and MSI-X capabilities.
270 	 */
271 	sts = passthru_read_config(&sel, PCIR_STATUS, 2);
272 	if (sts & PCIM_STATUS_CAPPRESENT) {
273 		ptr = passthru_read_config(&sel, PCIR_CAP_PTR, 1);
274 		while (ptr != 0 && ptr != 0xff) {
275 			cap = passthru_read_config(&sel, ptr + PCICAP_ID, 1);
276 			if (cap == PCIY_MSI) {
277 				/*
278 				 * Copy the MSI capability into the config
279 				 * space of the emulated pci device
280 				 */
281 				sc->psc_msi.capoff = ptr;
282 				sc->psc_msi.msgctrl =
283 				    passthru_read_config(&sel, ptr + 2, 2);
284 				sc->psc_msi.emulated = 0;
285 				caplen = msi_caplen(sc->psc_msi.msgctrl);
286 				capptr = ptr;
287 				while (caplen > 0) {
288 					u32 = passthru_read_config(&sel, capptr,
289 					    4);
290 					pci_set_cfgdata32(pi, capptr, u32);
291 					caplen -= 4;
292 					capptr += 4;
293 				}
294 			} else if (cap == PCIY_MSIX) {
295 				/*
296 				 * Copy the MSI-X capability
297 				 */
298 				sc->psc_msix.capoff = ptr;
299 				caplen = 12;
300 				msixcap_ptr = (char *)&msixcap;
301 				capptr = ptr;
302 				while (caplen > 0) {
303 					u32 = passthru_read_config(&sel, capptr,
304 					    4);
305 					memcpy(msixcap_ptr, &u32, 4);
306 					pci_set_cfgdata32(pi, capptr, u32);
307 					caplen -= 4;
308 					capptr += 4;
309 					msixcap_ptr += 4;
310 				}
311 			}
312 			ptr = passthru_read_config(&sel, ptr + PCICAP_NEXTPTR,
313 			    1);
314 		}
315 	}
316 
317 	if (sc->psc_msix.capoff != 0) {
318 		pi->pi_msix.pba_bar =
319 		    msixcap.pba_info & PCIM_MSIX_BIR_MASK;
320 		pi->pi_msix.pba_offset =
321 		    msixcap.pba_info & ~PCIM_MSIX_BIR_MASK;
322 		pi->pi_msix.table_bar =
323 		    msixcap.table_info & PCIM_MSIX_BIR_MASK;
324 		pi->pi_msix.table_offset =
325 		    msixcap.table_info & ~PCIM_MSIX_BIR_MASK;
326 		pi->pi_msix.table_count = MSIX_TABLE_COUNT(msixcap.msgctrl);
327 		pi->pi_msix.pba_size = PBA_SIZE(pi->pi_msix.table_count);
328 
329 		/* Allocate the emulated MSI-X table array */
330 		table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
331 		pi->pi_msix.table = calloc(1, table_size);
332 
333 		/* Mask all table entries */
334 		for (i = 0; i < pi->pi_msix.table_count; i++) {
335 			pi->pi_msix.table[i].vector_control |=
336 						PCIM_MSIX_VCTRL_MASK;
337 		}
338 	}
339 
340 #ifdef LEGACY_SUPPORT
341 	/*
342 	 * If the passthrough device does not support MSI then craft a
343 	 * MSI capability for it. We link the new MSI capability at the
344 	 * head of the list of capabilities.
345 	 */
346 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0 && sc->psc_msi.capoff == 0) {
347 		int origptr, msiptr;
348 		origptr = passthru_read_config(&sel, PCIR_CAP_PTR, 1);
349 		msiptr = passthru_add_msicap(pi, 1, origptr);
350 		sc->psc_msi.capoff = msiptr;
351 		sc->psc_msi.msgctrl = pci_get_cfgdata16(pi, msiptr + 2);
352 		sc->psc_msi.emulated = 1;
353 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, msiptr);
354 	}
355 #endif
356 
357 	/* Make sure one of the capabilities is present */
358 	if (sc->psc_msi.capoff == 0 && sc->psc_msix.capoff == 0)
359 		return (-1);
360 	else
361 		return (0);
362 }
363 
364 static uint64_t
msix_table_read(struct passthru_softc * sc,uint64_t offset,int size)365 msix_table_read(struct passthru_softc *sc, uint64_t offset, int size)
366 {
367 	struct pci_devinst *pi;
368 	struct msix_table_entry *entry;
369 	uint8_t *src8;
370 	uint16_t *src16;
371 	uint32_t *src32;
372 	uint64_t *src64;
373 	uint64_t data;
374 	size_t entry_offset;
375 	uint32_t table_offset;
376 	int index, table_count;
377 
378 	pi = sc->psc_pi;
379 
380 	table_offset = pi->pi_msix.table_offset;
381 	table_count = pi->pi_msix.table_count;
382 	if (offset < table_offset ||
383 	    offset >= table_offset + table_count * MSIX_TABLE_ENTRY_SIZE) {
384 		switch (size) {
385 		case 1:
386 			src8 = (uint8_t *)(pi->pi_msix.mapped_addr + offset);
387 			data = *src8;
388 			break;
389 		case 2:
390 			src16 = (uint16_t *)(pi->pi_msix.mapped_addr + offset);
391 			data = *src16;
392 			break;
393 		case 4:
394 			src32 = (uint32_t *)(pi->pi_msix.mapped_addr + offset);
395 			data = *src32;
396 			break;
397 		case 8:
398 			src64 = (uint64_t *)(pi->pi_msix.mapped_addr + offset);
399 			data = *src64;
400 			break;
401 		default:
402 			return (-1);
403 		}
404 		return (data);
405 	}
406 
407 	offset -= table_offset;
408 	index = offset / MSIX_TABLE_ENTRY_SIZE;
409 	assert(index < table_count);
410 
411 	entry = &pi->pi_msix.table[index];
412 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
413 
414 	switch (size) {
415 	case 1:
416 		src8 = (uint8_t *)((uint8_t *)entry + entry_offset);
417 		data = *src8;
418 		break;
419 	case 2:
420 		src16 = (uint16_t *)((uint8_t *)entry + entry_offset);
421 		data = *src16;
422 		break;
423 	case 4:
424 		src32 = (uint32_t *)((uint8_t *)entry + entry_offset);
425 		data = *src32;
426 		break;
427 	case 8:
428 		src64 = (uint64_t *)((uint8_t *)entry + entry_offset);
429 		data = *src64;
430 		break;
431 	default:
432 		return (-1);
433 	}
434 
435 	return (data);
436 }
437 
438 static void
msix_table_write(struct passthru_softc * sc,uint64_t offset,int size,uint64_t data)439 msix_table_write(struct passthru_softc *sc, uint64_t offset, int size,
440     uint64_t data)
441 {
442 	struct pci_devinst *pi;
443 	struct msix_table_entry *entry;
444 	uint8_t *dest8;
445 	uint16_t *dest16;
446 	uint32_t *dest32;
447 	uint64_t *dest64;
448 	size_t entry_offset;
449 	uint32_t table_offset, vector_control;
450 	int index, table_count;
451 
452 	pi = sc->psc_pi;
453 
454 	table_offset = pi->pi_msix.table_offset;
455 	table_count = pi->pi_msix.table_count;
456 	if (offset < table_offset ||
457 	    offset >= table_offset + table_count * MSIX_TABLE_ENTRY_SIZE) {
458 		switch (size) {
459 		case 1:
460 			dest8 = (uint8_t *)(pi->pi_msix.mapped_addr + offset);
461 			*dest8 = data;
462 			break;
463 		case 2:
464 			dest16 = (uint16_t *)(pi->pi_msix.mapped_addr + offset);
465 			*dest16 = data;
466 			break;
467 		case 4:
468 			dest32 = (uint32_t *)(pi->pi_msix.mapped_addr + offset);
469 			*dest32 = data;
470 			break;
471 		case 8:
472 			dest64 = (uint64_t *)(pi->pi_msix.mapped_addr + offset);
473 			*dest64 = data;
474 			break;
475 		}
476 		return;
477 	}
478 
479 	offset -= table_offset;
480 	index = offset / MSIX_TABLE_ENTRY_SIZE;
481 	assert(index < table_count);
482 
483 	entry = &pi->pi_msix.table[index];
484 	entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
485 
486 	/* Only 4 byte naturally-aligned writes are supported */
487 	assert(size == 4);
488 	assert(entry_offset % 4 == 0);
489 
490 	vector_control = entry->vector_control;
491 	dest32 = (uint32_t *)((uint8_t *)entry + entry_offset);
492 	*dest32 = data;
493 	/* If MSI-X hasn't been enabled, do nothing */
494 	if (pi->pi_msix.enabled) {
495 		/* If the entry is masked, don't set it up */
496 		if ((entry->vector_control & PCIM_MSIX_VCTRL_MASK) == 0 ||
497 		    (vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
498 			(void)vm_setup_pptdev_msix(sc->psc_pi->pi_vmctx,
499 			    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
500 			    sc->psc_sel.pc_func, index, entry->addr,
501 			    entry->msg_data, entry->vector_control);
502 		}
503 	}
504 }
505 
506 static int
init_msix_table(struct passthru_softc * sc)507 init_msix_table(struct passthru_softc *sc)
508 {
509 	struct pci_devinst *pi = sc->psc_pi;
510 	struct pci_bar_mmap pbm;
511 	int b, s, f;
512 	uint32_t table_size, table_offset;
513 
514 	assert(pci_msix_table_bar(pi) >= 0 && pci_msix_pba_bar(pi) >= 0);
515 
516 	b = sc->psc_sel.pc_bus;
517 	s = sc->psc_sel.pc_dev;
518 	f = sc->psc_sel.pc_func;
519 
520 	/*
521 	 * Map the region of the BAR containing the MSI-X table.  This is
522 	 * necessary for two reasons:
523 	 * 1. The PBA may reside in the first or last page containing the MSI-X
524 	 *    table.
525 	 * 2. While PCI devices are not supposed to use the page(s) containing
526 	 *    the MSI-X table for other purposes, some do in practice.
527 	 */
528 	memset(&pbm, 0, sizeof(pbm));
529 	pbm.pbm_sel = sc->psc_sel;
530 	pbm.pbm_flags = PCIIO_BAR_MMAP_RW;
531 	pbm.pbm_reg = PCIR_BAR(pi->pi_msix.table_bar);
532 	pbm.pbm_memattr = VM_MEMATTR_DEVICE;
533 
534 	if (ioctl(pcifd, PCIOCBARMMAP, &pbm) != 0) {
535 		warn("Failed to map MSI-X table BAR on %d/%d/%d", b, s, f);
536 		return (-1);
537 	}
538 	assert(pbm.pbm_bar_off == 0);
539 	pi->pi_msix.mapped_addr = (uint8_t *)(uintptr_t)pbm.pbm_map_base;
540 	pi->pi_msix.mapped_size = pbm.pbm_map_length;
541 
542 	table_offset = rounddown2(pi->pi_msix.table_offset, 4096);
543 
544 	table_size = pi->pi_msix.table_offset - table_offset;
545 	table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
546 	table_size = roundup2(table_size, 4096);
547 
548 	/*
549 	 * Unmap any pages not containing the table, we do not need to emulate
550 	 * accesses to them.  Avoid releasing address space to help ensure that
551 	 * a buggy out-of-bounds access causes a crash.
552 	 */
553 	if (table_offset != 0)
554 		if (mprotect(pi->pi_msix.mapped_addr, table_offset,
555 		    PROT_NONE) != 0)
556 			warn("Failed to unmap MSI-X table BAR region");
557 	if (table_offset + table_size != pi->pi_msix.mapped_size)
558 		if (mprotect(
559 		    pi->pi_msix.mapped_addr + table_offset + table_size,
560 		    pi->pi_msix.mapped_size - (table_offset + table_size),
561 		    PROT_NONE) != 0)
562 			warn("Failed to unmap MSI-X table BAR region");
563 
564 	return (0);
565 }
566 
567 static int
cfginitbar(struct passthru_softc * sc)568 cfginitbar(struct passthru_softc *sc)
569 {
570 	int i, error;
571 	struct pci_devinst *pi;
572 	struct pci_bar_io bar;
573 	enum pcibar_type bartype;
574 	uint64_t base, size;
575 
576 	pi = sc->psc_pi;
577 
578 	/*
579 	 * Initialize BAR registers
580 	 */
581 	for (i = 0; i <= PCI_BARMAX; i++) {
582 		uint8_t lobits;
583 
584 		bzero(&bar, sizeof(bar));
585 		bar.pbi_sel = sc->psc_sel;
586 		bar.pbi_reg = PCIR_BAR(i);
587 
588 		if (ioctl(pcifd, PCIOCGETBAR, &bar) < 0)
589 			continue;
590 
591 		if (PCI_BAR_IO(bar.pbi_base)) {
592 			bartype = PCIBAR_IO;
593 			base = bar.pbi_base & PCIM_BAR_IO_BASE;
594 		} else {
595 			switch (bar.pbi_base & PCIM_BAR_MEM_TYPE) {
596 			case PCIM_BAR_MEM_64:
597 				bartype = PCIBAR_MEM64;
598 				break;
599 			default:
600 				bartype = PCIBAR_MEM32;
601 				break;
602 			}
603 			base = bar.pbi_base & PCIM_BAR_MEM_BASE;
604 		}
605 		size = bar.pbi_length;
606 
607 		if (bartype != PCIBAR_IO) {
608 			if (((base | size) & PAGE_MASK) != 0) {
609 				warnx("passthru device %d/%d/%d BAR %d: "
610 				    "base %#lx or size %#lx not page aligned\n",
611 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
612 				    sc->psc_sel.pc_func, i, base, size);
613 				return (-1);
614 			}
615 		}
616 
617 		/* Cache information about the "real" BAR */
618 		sc->psc_bar[i].type = bartype;
619 		sc->psc_bar[i].size = size;
620 		sc->psc_bar[i].addr = base;
621 		sc->psc_bar[i].lobits = 0;
622 
623 		/* Allocate the BAR in the guest I/O or MMIO space */
624 		error = pci_emul_alloc_bar(pi, i, bartype, size);
625 		if (error)
626 			return (-1);
627 
628 		/* Use same lobits as physical bar */
629 		lobits = (uint8_t)passthru_read_config(&sc->psc_sel,
630 		    PCIR_BAR(i), 0x01);
631 		if (bartype == PCIBAR_MEM32 || bartype == PCIBAR_MEM64) {
632 			lobits &= ~PCIM_BAR_MEM_BASE;
633 		} else {
634 			lobits &= ~PCIM_BAR_IO_BASE;
635 		}
636 		sc->psc_bar[i].lobits = lobits;
637 		pi->pi_bar[i].lobits = lobits;
638 
639 		/*
640 		 * 64-bit BAR takes up two slots so skip the next one.
641 		 */
642 		if (bartype == PCIBAR_MEM64) {
643 			i++;
644 			assert(i <= PCI_BARMAX);
645 			sc->psc_bar[i].type = PCIBAR_MEMHI64;
646 		}
647 	}
648 	return (0);
649 }
650 
651 static int
cfginit(struct pci_devinst * pi,int bus,int slot,int func)652 cfginit(struct pci_devinst *pi, int bus, int slot, int func)
653 {
654 	int error;
655 	struct passthru_softc *sc;
656 	uint16_t cmd;
657 	uint8_t intline, intpin;
658 
659 	error = 1;
660 	sc = pi->pi_arg;
661 
662 	bzero(&sc->psc_sel, sizeof(struct pcisel));
663 	sc->psc_sel.pc_bus = bus;
664 	sc->psc_sel.pc_dev = slot;
665 	sc->psc_sel.pc_func = func;
666 
667 	/*
668 	 * Copy physical PCI header to virtual config space.  COMMAND,
669 	 * INTLINE, and INTPIN shouldn't be aligned with their
670 	 * physical value and they are already set by pci_emul_init().
671 	 */
672 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
673 	intline = pci_get_cfgdata8(pi, PCIR_INTLINE);
674 	intpin = pci_get_cfgdata8(pi, PCIR_INTPIN);
675 	for (int i = 0; i <= PCIR_MAXLAT; i += 4) {
676 		pci_set_cfgdata32(pi, i,
677 		    passthru_read_config(&sc->psc_sel, i, 4));
678 	}
679 	pci_set_cfgdata16(pi, PCIR_COMMAND, cmd);
680 	pci_set_cfgdata8(pi, PCIR_INTLINE, intline);
681 	pci_set_cfgdata8(pi, PCIR_INTPIN, intpin);
682 
683 	if (cfginitmsi(sc) != 0) {
684 		warnx("failed to initialize MSI for PCI %d/%d/%d",
685 		    bus, slot, func);
686 		goto done;
687 	}
688 
689 	if (cfginitbar(sc) != 0) {
690 		warnx("failed to initialize BARs for PCI %d/%d/%d",
691 		    bus, slot, func);
692 		goto done;
693 	}
694 
695 	if (pci_msix_table_bar(pi) >= 0) {
696 		error = init_msix_table(sc);
697 		if (error != 0) {
698 			warnx(
699 			    "failed to initialize MSI-X table for PCI %d/%d/%d: %d",
700 			    bus, slot, func, error);
701 			goto done;
702 		}
703 	}
704 
705 	error = 0;				/* success */
706 done:
707 	return (error);
708 }
709 
710 struct passthru_mmio_mapping *
passthru_get_mmio(struct passthru_softc * sc,int num)711 passthru_get_mmio(struct passthru_softc *sc, int num)
712 {
713 	assert(sc != NULL);
714 	assert(num < PASSTHRU_MMIO_MAX);
715 
716 	return (&sc->psc_mmio_map[num]);
717 }
718 
719 struct pcisel *
passthru_get_sel(struct passthru_softc * sc)720 passthru_get_sel(struct passthru_softc *sc)
721 {
722 	assert(sc != NULL);
723 
724 	return (&sc->psc_sel);
725 }
726 
727 int
set_pcir_handler(struct passthru_softc * sc,int reg,int len,cfgread_handler rhandler,cfgwrite_handler whandler)728 set_pcir_handler(struct passthru_softc *sc, int reg, int len,
729     cfgread_handler rhandler, cfgwrite_handler whandler)
730 {
731 	if (reg > PCI_REGMAX || reg + len > PCI_REGMAX + 1)
732 		return (-1);
733 
734 	for (int i = reg; i < reg + len; ++i) {
735 		assert(sc->psc_pcir_rhandler[i] == NULL || rhandler == NULL);
736 		assert(sc->psc_pcir_whandler[i] == NULL || whandler == NULL);
737 		sc->psc_pcir_rhandler[i] = rhandler;
738 		sc->psc_pcir_whandler[i] = whandler;
739 	}
740 
741 	return (0);
742 }
743 
744 static int
passthru_legacy_config(nvlist_t * nvl,const char * opts)745 passthru_legacy_config(nvlist_t *nvl, const char *opts)
746 {
747 	const char *cp;
748 	char *tofree;
749 	char value[16];
750 	int bus, slot, func;
751 
752 	if (opts == NULL)
753 		return (0);
754 
755 	cp = strchr(opts, ',');
756 
757 	if (strncmp(opts, "ppt", strlen("ppt")) == 0) {
758 		tofree = strndup(opts, cp - opts);
759 		set_config_value_node(nvl, "pptdev", tofree);
760 		free(tofree);
761 	} else if (sscanf(opts, "pci0:%d:%d:%d", &bus, &slot, &func) == 3 ||
762 	    sscanf(opts, "pci%d:%d:%d", &bus, &slot, &func) == 3 ||
763 	    sscanf(opts, "%d/%d/%d", &bus, &slot, &func) == 3) {
764 		snprintf(value, sizeof(value), "%d", bus);
765 		set_config_value_node(nvl, "bus", value);
766 		snprintf(value, sizeof(value), "%d", slot);
767 		set_config_value_node(nvl, "slot", value);
768 		snprintf(value, sizeof(value), "%d", func);
769 		set_config_value_node(nvl, "func", value);
770 	} else {
771 		EPRINTLN("passthru: invalid options \"%s\"", opts);
772 		return (-1);
773 	}
774 
775 	if (cp == NULL) {
776 		return (0);
777 	}
778 
779 	return (pci_parse_legacy_config(nvl, cp + 1));
780 }
781 
782 static int
passthru_init_rom(struct passthru_softc * const sc,const char * const romfile)783 passthru_init_rom(struct passthru_softc *const sc, const char *const romfile)
784 {
785 	if (romfile == NULL) {
786 		return (0);
787 	}
788 
789 	const int fd = open(romfile, O_RDONLY);
790 	if (fd < 0) {
791 		warnx("%s: can't open romfile \"%s\"", __func__, romfile);
792 		return (-1);
793 	}
794 
795 	struct stat sbuf;
796 	if (fstat(fd, &sbuf) < 0) {
797 		warnx("%s: can't fstat romfile \"%s\"", __func__, romfile);
798 		close(fd);
799 		return (-1);
800 	}
801 	const uint64_t rom_size = sbuf.st_size;
802 
803 	void *const rom_data = mmap(NULL, rom_size, PROT_READ, MAP_SHARED, fd,
804 	    0);
805 	if (rom_data == MAP_FAILED) {
806 		warnx("%s: unable to mmap romfile \"%s\" (%d)", __func__,
807 		    romfile, errno);
808 		close(fd);
809 		return (-1);
810 	}
811 
812 	void *rom_addr;
813 	int error = pci_emul_alloc_rom(sc->psc_pi, rom_size, &rom_addr);
814 	if (error) {
815 		warnx("%s: failed to alloc rom segment", __func__);
816 		munmap(rom_data, rom_size);
817 		close(fd);
818 		return (error);
819 	}
820 	memcpy(rom_addr, rom_data, rom_size);
821 
822 	sc->psc_bar[PCI_ROM_IDX].type = PCIBAR_ROM;
823 	sc->psc_bar[PCI_ROM_IDX].addr = (uint64_t)rom_addr;
824 	sc->psc_bar[PCI_ROM_IDX].size = rom_size;
825 
826 	munmap(rom_data, rom_size);
827 	close(fd);
828 
829 	return (0);
830 }
831 
832 static bool
passthru_lookup_pptdev(const char * name,int * bus,int * slot,int * func)833 passthru_lookup_pptdev(const char *name, int *bus, int *slot, int *func)
834 {
835 	struct pci_conf_io pc;
836 	struct pci_conf conf[1];
837 	struct pci_match_conf patterns[1];
838 	char *cp;
839 
840 	bzero(&pc, sizeof(struct pci_conf_io));
841 	pc.match_buf_len = sizeof(conf);
842 	pc.matches = conf;
843 
844 	bzero(&patterns, sizeof(patterns));
845 
846 	/*
847 	 * The pattern structure requires the unit to be split out from
848 	 * the driver name.  Walk backwards from the end of the name to
849 	 * find the start of the unit.
850 	 */
851 	cp = strchr(name, '\0');
852 	assert(cp != NULL);
853 	while (cp != name && isdigit(cp[-1]))
854 		cp--;
855 	if (cp == name || !isdigit(*cp)) {
856 		EPRINTLN("Invalid passthru device name %s", name);
857 		return (false);
858 	}
859 	if ((size_t)(cp - name) + 1 > sizeof(patterns[0].pd_name)) {
860 		EPRINTLN("Passthru device name %s is too long", name);
861 		return (false);
862 	}
863 	memcpy(patterns[0].pd_name, name, cp - name);
864 	patterns[0].pd_unit = strtol(cp, &cp, 10);
865 	if (*cp != '\0') {
866 		EPRINTLN("Invalid passthru device name %s", name);
867 		return (false);
868 	}
869 	patterns[0].flags = PCI_GETCONF_MATCH_NAME | PCI_GETCONF_MATCH_UNIT;
870 	pc.num_patterns = 1;
871 	pc.pat_buf_len = sizeof(patterns);
872 	pc.patterns = patterns;
873 
874 	if (ioctl(pcifd, PCIOCGETCONF, &pc) == -1) {
875 		EPRINTLN("ioctl(PCIOCGETCONF): %s", strerror(errno));
876 		return (false);
877 	}
878 	if (pc.status != PCI_GETCONF_LAST_DEVICE &&
879 	    pc.status != PCI_GETCONF_MORE_DEVS) {
880 		EPRINTLN("error returned from PCIOCGETCONF ioctl");
881 		return (false);
882 	}
883 	if (pc.num_matches == 0) {
884 		EPRINTLN("Passthru device %s not found", name);
885 		return (false);
886 	}
887 
888 	if (conf[0].pc_sel.pc_domain != 0) {
889 		EPRINTLN("Passthru device %s on unsupported domain", name);
890 		return (false);
891 	}
892 	*bus = conf[0].pc_sel.pc_bus;
893 	*slot = conf[0].pc_sel.pc_dev;
894 	*func = conf[0].pc_sel.pc_func;
895 	return (true);
896 }
897 
898 static int
passthru_init(struct pci_devinst * pi,nvlist_t * nvl)899 passthru_init(struct pci_devinst *pi, nvlist_t *nvl)
900 {
901 	int bus, slot, func, error, memflags;
902 	struct passthru_softc *sc;
903 	struct passthru_dev **devpp;
904 	struct passthru_dev *devp, *dev = NULL;
905 	const char *value;
906 
907 	sc = NULL;
908 	error = 1;
909 
910 	memflags = vm_get_memflags(pi->pi_vmctx);
911 	if (!(memflags & VM_MEM_F_WIRED)) {
912 		warnx("passthru requires guest memory to be wired");
913 		return (error);
914 	}
915 
916 	if (pcifd < 0 && pcifd_init()) {
917 		return (error);
918 	}
919 
920 #define GET_INT_CONFIG(var, name) do {					\
921 	value = get_config_value_node(nvl, name);			\
922 	if (value == NULL) {						\
923 		EPRINTLN("passthru: missing required %s setting", name); \
924 		return (error);						\
925 	}								\
926 	var = atoi(value);						\
927 } while (0)
928 
929 	value = get_config_value_node(nvl, "pptdev");
930 	if (value != NULL) {
931 		if (!passthru_lookup_pptdev(value, &bus, &slot, &func))
932 			return (error);
933 	} else {
934 		GET_INT_CONFIG(bus, "bus");
935 		GET_INT_CONFIG(slot, "slot");
936 		GET_INT_CONFIG(func, "func");
937 	}
938 
939 	if (vm_assign_pptdev(pi->pi_vmctx, bus, slot, func) != 0) {
940 		warnx("PCI device at %d/%d/%d is not using the ppt(4) driver",
941 		    bus, slot, func);
942 		goto done;
943 	}
944 
945 	sc = calloc(1, sizeof(struct passthru_softc));
946 
947 	pi->pi_arg = sc;
948 	sc->psc_pi = pi;
949 
950 	/* initialize config space */
951 	if ((error = cfginit(pi, bus, slot, func)) != 0)
952 		goto done;
953 
954 	/* initialize ROM */
955 	if ((error = passthru_init_rom(sc,
956             get_config_value_node(nvl, "rom"))) != 0)
957 		goto done;
958 
959 	/* Emulate most PCI header register. */
960 	if ((error = set_pcir_handler(sc, 0, PCIR_MAXLAT + 1,
961 	    passthru_cfgread_emulate, passthru_cfgwrite_emulate)) != 0)
962 		goto done;
963 
964 	/* Allow access to the physical status register. */
965 	if ((error = set_pcir_handler(sc, PCIR_COMMAND, 0x04, NULL, NULL)) != 0)
966 		goto done;
967 
968 	SET_FOREACH(devpp, passthru_dev_set) {
969 		devp = *devpp;
970 		assert(devp->probe != NULL);
971 		if (devp->probe(pi) == 0) {
972 			dev = devp;
973 			break;
974 		}
975 	}
976 
977 	if (dev != NULL) {
978 		error = dev->init(pi, nvl);
979 		if (error != 0)
980 			goto done;
981 	}
982 
983 	error = 0;		/* success */
984 done:
985 	if (error) {
986 		if (dev != NULL)
987 			dev->deinit(pi);
988 		free(sc);
989 		vm_unassign_pptdev(pi->pi_vmctx, bus, slot, func);
990 	}
991 	return (error);
992 }
993 
994 static int
msicap_access(struct passthru_softc * sc,int coff)995 msicap_access(struct passthru_softc *sc, int coff)
996 {
997 	int caplen;
998 
999 	if (sc->psc_msi.capoff == 0)
1000 		return (0);
1001 
1002 	caplen = msi_caplen(sc->psc_msi.msgctrl);
1003 
1004 	if (coff >= sc->psc_msi.capoff && coff < sc->psc_msi.capoff + caplen)
1005 		return (1);
1006 	else
1007 		return (0);
1008 }
1009 
1010 static int
msixcap_access(struct passthru_softc * sc,int coff)1011 msixcap_access(struct passthru_softc *sc, int coff)
1012 {
1013 	if (sc->psc_msix.capoff == 0)
1014 		return (0);
1015 
1016 	return (coff >= sc->psc_msix.capoff &&
1017 	        coff < sc->psc_msix.capoff + MSIX_CAPLEN);
1018 }
1019 
1020 static int
passthru_cfgread_default(struct passthru_softc * sc,struct pci_devinst * pi __unused,int coff,int bytes,uint32_t * rv)1021 passthru_cfgread_default(struct passthru_softc *sc,
1022     struct pci_devinst *pi __unused, int coff, int bytes, uint32_t *rv)
1023 {
1024 	/*
1025 	 * MSI capability is emulated.
1026 	 */
1027 	if (msicap_access(sc, coff) || msixcap_access(sc, coff))
1028 		return (-1);
1029 
1030 	/*
1031 	 * Emulate the command register.  If a single read reads both the
1032 	 * command and status registers, read the status register from the
1033 	 * device's config space.
1034 	 */
1035 	if (coff == PCIR_COMMAND) {
1036 		uint32_t st;
1037 
1038 		if (bytes <= 2)
1039 			return (-1);
1040 		st = passthru_read_config(&sc->psc_sel, PCIR_STATUS, 2);
1041 		*rv = (st << 16) | pci_get_cfgdata16(pi, PCIR_COMMAND);
1042 		return (0);
1043 	}
1044 
1045 	/* Everything else just read from the device's config space */
1046 	*rv = passthru_read_config(&sc->psc_sel, coff, bytes);
1047 
1048 	return (0);
1049 }
1050 
1051 int
passthru_cfgread_emulate(struct passthru_softc * sc __unused,struct pci_devinst * pi __unused,int coff __unused,int bytes __unused,uint32_t * rv __unused)1052 passthru_cfgread_emulate(struct passthru_softc *sc __unused,
1053     struct pci_devinst *pi __unused, int coff __unused, int bytes __unused,
1054     uint32_t *rv __unused)
1055 {
1056 	return (-1);
1057 }
1058 
1059 static int
passthru_cfgread(struct pci_devinst * pi,int coff,int bytes,uint32_t * rv)1060 passthru_cfgread(struct pci_devinst *pi, int coff, int bytes, uint32_t *rv)
1061 {
1062 	struct passthru_softc *sc;
1063 
1064 	sc = pi->pi_arg;
1065 
1066 	if (sc->psc_pcir_rhandler[coff] != NULL)
1067 		return (sc->psc_pcir_rhandler[coff](sc, pi, coff, bytes, rv));
1068 
1069 	return (passthru_cfgread_default(sc, pi, coff, bytes, rv));
1070 }
1071 
1072 static int
passthru_cfgwrite_default(struct passthru_softc * sc,struct pci_devinst * pi,int coff,int bytes,uint32_t val)1073 passthru_cfgwrite_default(struct passthru_softc *sc, struct pci_devinst *pi,
1074     int coff, int bytes, uint32_t val)
1075 {
1076 	int error, msix_table_entries, i;
1077 	uint16_t cmd_old;
1078 
1079 	/*
1080 	 * MSI capability is emulated
1081 	 */
1082 	if (msicap_access(sc, coff)) {
1083 		pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msi.capoff,
1084 		    PCIY_MSI);
1085 		error = vm_setup_pptdev_msi(pi->pi_vmctx, sc->psc_sel.pc_bus,
1086 			sc->psc_sel.pc_dev, sc->psc_sel.pc_func,
1087 			pi->pi_msi.addr, pi->pi_msi.msg_data,
1088 			pi->pi_msi.maxmsgnum);
1089 		if (error != 0)
1090 			err(1, "vm_setup_pptdev_msi");
1091 		return (0);
1092 	}
1093 
1094 	if (msixcap_access(sc, coff)) {
1095 		pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msix.capoff,
1096 		    PCIY_MSIX);
1097 		if (pi->pi_msix.enabled) {
1098 			msix_table_entries = pi->pi_msix.table_count;
1099 			for (i = 0; i < msix_table_entries; i++) {
1100 				error = vm_setup_pptdev_msix(pi->pi_vmctx,
1101 				    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
1102 				    sc->psc_sel.pc_func, i,
1103 				    pi->pi_msix.table[i].addr,
1104 				    pi->pi_msix.table[i].msg_data,
1105 				    pi->pi_msix.table[i].vector_control);
1106 
1107 				if (error)
1108 					err(1, "vm_setup_pptdev_msix");
1109 			}
1110 		} else {
1111 			error = vm_disable_pptdev_msix(pi->pi_vmctx,
1112 			    sc->psc_sel.pc_bus, sc->psc_sel.pc_dev,
1113 			    sc->psc_sel.pc_func);
1114 			if (error)
1115 				err(1, "vm_disable_pptdev_msix");
1116 		}
1117 		return (0);
1118 	}
1119 
1120 	/*
1121 	 * The command register is emulated, but the status register
1122 	 * is passed through.
1123 	 */
1124 	if (coff == PCIR_COMMAND) {
1125 		if (bytes <= 2)
1126 			return (-1);
1127 
1128 		/* Update the physical status register. */
1129 		passthru_write_config(&sc->psc_sel, PCIR_STATUS, val >> 16, 2);
1130 
1131 		/* Update the virtual command register. */
1132 		cmd_old = pci_get_cfgdata16(pi, PCIR_COMMAND);
1133 		pci_set_cfgdata16(pi, PCIR_COMMAND, val & 0xffff);
1134 		pci_emul_cmd_changed(pi, cmd_old);
1135 		return (0);
1136 	}
1137 
1138 	passthru_write_config(&sc->psc_sel, coff, bytes, val);
1139 
1140 	return (0);
1141 }
1142 
1143 int
passthru_cfgwrite_emulate(struct passthru_softc * sc __unused,struct pci_devinst * pi __unused,int coff __unused,int bytes __unused,uint32_t val __unused)1144 passthru_cfgwrite_emulate(struct passthru_softc *sc __unused,
1145     struct pci_devinst *pi __unused, int coff __unused, int bytes __unused,
1146     uint32_t val __unused)
1147 {
1148 	return (-1);
1149 }
1150 
1151 static int
passthru_cfgwrite(struct pci_devinst * pi,int coff,int bytes,uint32_t val)1152 passthru_cfgwrite(struct pci_devinst *pi, int coff, int bytes, uint32_t val)
1153 {
1154 	struct passthru_softc *sc;
1155 
1156 	sc = pi->pi_arg;
1157 
1158 	if (sc->psc_pcir_whandler[coff] != NULL)
1159 		return (sc->psc_pcir_whandler[coff](sc, pi, coff, bytes, val));
1160 
1161 	return (passthru_cfgwrite_default(sc, pi, coff, bytes, val));
1162 }
1163 
1164 static void
passthru_write(struct pci_devinst * pi,int baridx,uint64_t offset,int size,uint64_t value)1165 passthru_write(struct pci_devinst *pi, int baridx, uint64_t offset, int size,
1166     uint64_t value)
1167 {
1168 	struct passthru_softc *sc;
1169 	struct pci_bar_ioreq pio;
1170 
1171 	sc = pi->pi_arg;
1172 
1173 	if (baridx == pci_msix_table_bar(pi)) {
1174 		msix_table_write(sc, offset, size, value);
1175 	} else {
1176 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
1177 		assert(size == 1 || size == 2 || size == 4);
1178 		assert(offset <= UINT32_MAX && offset + size <= UINT32_MAX);
1179 
1180 		bzero(&pio, sizeof(pio));
1181 		pio.pbi_sel = sc->psc_sel;
1182 		pio.pbi_op = PCIBARIO_WRITE;
1183 		pio.pbi_bar = baridx;
1184 		pio.pbi_offset = (uint32_t)offset;
1185 		pio.pbi_width = size;
1186 		pio.pbi_value = (uint32_t)value;
1187 
1188 		(void)ioctl(pcifd, PCIOCBARIO, &pio);
1189 	}
1190 }
1191 
1192 static uint64_t
passthru_read(struct pci_devinst * pi,int baridx,uint64_t offset,int size)1193 passthru_read(struct pci_devinst *pi, int baridx, uint64_t offset, int size)
1194 {
1195 	struct passthru_softc *sc;
1196 	struct pci_bar_ioreq pio;
1197 	uint64_t val;
1198 
1199 	sc = pi->pi_arg;
1200 
1201 	if (baridx == pci_msix_table_bar(pi)) {
1202 		val = msix_table_read(sc, offset, size);
1203 	} else {
1204 		assert(pi->pi_bar[baridx].type == PCIBAR_IO);
1205 		assert(size == 1 || size == 2 || size == 4);
1206 		assert(offset <= UINT32_MAX && offset + size <= UINT32_MAX);
1207 
1208 		bzero(&pio, sizeof(pio));
1209 		pio.pbi_sel = sc->psc_sel;
1210 		pio.pbi_op = PCIBARIO_READ;
1211 		pio.pbi_bar = baridx;
1212 		pio.pbi_offset = (uint32_t)offset;
1213 		pio.pbi_width = size;
1214 
1215 		(void)ioctl(pcifd, PCIOCBARIO, &pio);
1216 
1217 		val = pio.pbi_value;
1218 	}
1219 
1220 	return (val);
1221 }
1222 
1223 static void
passthru_msix_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1224 passthru_msix_addr(struct pci_devinst *pi, int baridx, int enabled,
1225     uint64_t address)
1226 {
1227 	struct passthru_softc *sc;
1228 	size_t remaining;
1229 	uint32_t table_size, table_offset;
1230 
1231 	sc = pi->pi_arg;
1232 	table_offset = rounddown2(pi->pi_msix.table_offset, 4096);
1233 	if (table_offset > 0) {
1234 		if (!enabled) {
1235 			if (vm_unmap_pptdev_mmio(pi->pi_vmctx,
1236 						 sc->psc_sel.pc_bus,
1237 						 sc->psc_sel.pc_dev,
1238 						 sc->psc_sel.pc_func, address,
1239 						 table_offset) != 0)
1240 				warnx("pci_passthru: unmap_pptdev_mmio failed");
1241 		} else {
1242 			if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1243 					       sc->psc_sel.pc_dev,
1244 					       sc->psc_sel.pc_func, address,
1245 					       table_offset,
1246 					       sc->psc_bar[baridx].addr) != 0)
1247 				warnx("pci_passthru: map_pptdev_mmio failed");
1248 		}
1249 	}
1250 	table_size = pi->pi_msix.table_offset - table_offset;
1251 	table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE;
1252 	table_size = roundup2(table_size, 4096);
1253 	remaining = pi->pi_bar[baridx].size - table_offset - table_size;
1254 	if (remaining > 0) {
1255 		address += table_offset + table_size;
1256 		if (!enabled) {
1257 			if (vm_unmap_pptdev_mmio(pi->pi_vmctx,
1258 						 sc->psc_sel.pc_bus,
1259 						 sc->psc_sel.pc_dev,
1260 						 sc->psc_sel.pc_func, address,
1261 						 remaining) != 0)
1262 				warnx("pci_passthru: unmap_pptdev_mmio failed");
1263 		} else {
1264 			if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1265 					       sc->psc_sel.pc_dev,
1266 					       sc->psc_sel.pc_func, address,
1267 					       remaining,
1268 					       sc->psc_bar[baridx].addr +
1269 					       table_offset + table_size) != 0)
1270 				warnx("pci_passthru: map_pptdev_mmio failed");
1271 		}
1272 	}
1273 }
1274 
1275 static void
passthru_mmio_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1276 passthru_mmio_addr(struct pci_devinst *pi, int baridx, int enabled,
1277     uint64_t address)
1278 {
1279 	struct passthru_softc *sc;
1280 
1281 	sc = pi->pi_arg;
1282 	if (!enabled) {
1283 		if (vm_unmap_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1284 					 sc->psc_sel.pc_dev,
1285 					 sc->psc_sel.pc_func, address,
1286 					 sc->psc_bar[baridx].size) != 0)
1287 			warnx("pci_passthru: unmap_pptdev_mmio failed");
1288 	} else {
1289 		if (vm_map_pptdev_mmio(pi->pi_vmctx, sc->psc_sel.pc_bus,
1290 				       sc->psc_sel.pc_dev,
1291 				       sc->psc_sel.pc_func, address,
1292 				       sc->psc_bar[baridx].size,
1293 				       sc->psc_bar[baridx].addr) != 0)
1294 			warnx("pci_passthru: map_pptdev_mmio failed");
1295 	}
1296 }
1297 
1298 static void
passthru_addr_rom(struct pci_devinst * const pi,const int idx,const int enabled)1299 passthru_addr_rom(struct pci_devinst *const pi, const int idx,
1300     const int enabled)
1301 {
1302 	const uint64_t addr = pi->pi_bar[idx].addr;
1303 	const uint64_t size = pi->pi_bar[idx].size;
1304 
1305 	if (!enabled) {
1306 		if (vm_munmap_memseg(pi->pi_vmctx, addr, size) != 0) {
1307 			errx(4, "%s: munmap_memseg @ [%016lx - %016lx] failed",
1308 			    __func__, addr, addr + size);
1309 		}
1310 
1311 	} else {
1312 		if (vm_mmap_memseg(pi->pi_vmctx, addr, VM_PCIROM,
1313 			pi->pi_romoffset, size, PROT_READ | PROT_EXEC) != 0) {
1314 			errx(4, "%s: mmap_memseg @ [%016lx - %016lx]  failed",
1315 			    __func__, addr, addr + size);
1316 		}
1317 	}
1318 }
1319 
1320 static void
passthru_addr(struct pci_devinst * pi,int baridx,int enabled,uint64_t address)1321 passthru_addr(struct pci_devinst *pi, int baridx, int enabled, uint64_t address)
1322 {
1323 	switch (pi->pi_bar[baridx].type) {
1324 	case PCIBAR_IO:
1325 		/* IO BARs are emulated */
1326 		break;
1327 	case PCIBAR_ROM:
1328 		passthru_addr_rom(pi, baridx, enabled);
1329 		break;
1330 	case PCIBAR_MEM32:
1331 	case PCIBAR_MEM64:
1332 		if (baridx == pci_msix_table_bar(pi))
1333 			passthru_msix_addr(pi, baridx, enabled, address);
1334 		else
1335 			passthru_mmio_addr(pi, baridx, enabled, address);
1336 		break;
1337 	default:
1338 		errx(4, "%s: invalid BAR type %d", __func__,
1339 		    pi->pi_bar[baridx].type);
1340 	}
1341 }
1342 
1343 static const struct pci_devemu passthru = {
1344 	.pe_emu		= "passthru",
1345 	.pe_init	= passthru_init,
1346 	.pe_legacy_config = passthru_legacy_config,
1347 	.pe_cfgwrite	= passthru_cfgwrite,
1348 	.pe_cfgread	= passthru_cfgread,
1349 	.pe_barwrite 	= passthru_write,
1350 	.pe_barread    	= passthru_read,
1351 	.pe_baraddr	= passthru_addr,
1352 };
1353 PCI_EMUL_SET(passthru);
1354