xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision d65cd7a57bf0600b722afc770838a5d0c1c3a8e1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/linker_set.h>
36 
37 #include <ctype.h>
38 #include <errno.h>
39 #include <pthread.h>
40 #include <stdio.h>
41 #include <stdlib.h>
42 #include <string.h>
43 #include <strings.h>
44 #include <assert.h>
45 #include <stdbool.h>
46 
47 #include <machine/vmm.h>
48 #include <machine/vmm_snapshot.h>
49 #include <vmmapi.h>
50 
51 #include "acpi.h"
52 #include "bhyverun.h"
53 #include "debug.h"
54 #include "inout.h"
55 #include "ioapic.h"
56 #include "mem.h"
57 #include "pci_emul.h"
58 #include "pci_irq.h"
59 #include "pci_lpc.h"
60 
61 #define CONF1_ADDR_PORT	   0x0cf8
62 #define CONF1_DATA_PORT	   0x0cfc
63 
64 #define CONF1_ENABLE	   0x80000000ul
65 
66 #define	MAXBUSES	(PCI_BUSMAX + 1)
67 #define MAXSLOTS	(PCI_SLOTMAX + 1)
68 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
69 
70 struct funcinfo {
71 	char	*fi_name;
72 	char	*fi_param;
73 	struct pci_devinst *fi_devi;
74 };
75 
76 struct intxinfo {
77 	int	ii_count;
78 	int	ii_pirq_pin;
79 	int	ii_ioapic_irq;
80 };
81 
82 struct slotinfo {
83 	struct intxinfo si_intpins[4];
84 	struct funcinfo si_funcs[MAXFUNCS];
85 };
86 
87 struct businfo {
88 	uint16_t iobase, iolimit;		/* I/O window */
89 	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
90 	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
91 	struct slotinfo slotinfo[MAXSLOTS];
92 };
93 
94 static struct businfo *pci_businfo[MAXBUSES];
95 
96 SET_DECLARE(pci_devemu_set, struct pci_devemu);
97 
98 static uint64_t pci_emul_iobase;
99 static uint64_t pci_emul_membase32;
100 static uint64_t pci_emul_membase64;
101 
102 #define	PCI_EMUL_IOBASE		0x2000
103 #define	PCI_EMUL_IOLIMIT	0x10000
104 
105 #define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
106 #define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
107 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
108 
109 #define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
110 
111 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
112 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
113 
114 static struct pci_devemu *pci_emul_finddev(char *name);
115 static void pci_lintr_route(struct pci_devinst *pi);
116 static void pci_lintr_update(struct pci_devinst *pi);
117 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
118     int func, int coff, int bytes, uint32_t *val);
119 
120 static __inline void
121 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
122 {
123 
124 	if (bytes == 1)
125 		pci_set_cfgdata8(pi, coff, val);
126 	else if (bytes == 2)
127 		pci_set_cfgdata16(pi, coff, val);
128 	else
129 		pci_set_cfgdata32(pi, coff, val);
130 }
131 
132 static __inline uint32_t
133 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
134 {
135 
136 	if (bytes == 1)
137 		return (pci_get_cfgdata8(pi, coff));
138 	else if (bytes == 2)
139 		return (pci_get_cfgdata16(pi, coff));
140 	else
141 		return (pci_get_cfgdata32(pi, coff));
142 }
143 
144 /*
145  * I/O access
146  */
147 
148 /*
149  * Slot options are in the form:
150  *
151  *  <bus>:<slot>:<func>,<emul>[,<config>]
152  *  <slot>[:<func>],<emul>[,<config>]
153  *
154  *  slot is 0..31
155  *  func is 0..7
156  *  emul is a string describing the type of PCI device e.g. virtio-net
157  *  config is an optional string, depending on the device, that can be
158  *  used for configuration.
159  *   Examples are:
160  *     1,virtio-net,tap0
161  *     3:0,dummy
162  */
163 static void
164 pci_parse_slot_usage(char *aopt)
165 {
166 
167 	EPRINTLN("Invalid PCI slot info field \"%s\"", aopt);
168 }
169 
170 int
171 pci_parse_slot(char *opt)
172 {
173 	struct businfo *bi;
174 	struct slotinfo *si;
175 	char *emul, *config, *str, *cp;
176 	int error, bnum, snum, fnum;
177 
178 	error = -1;
179 	str = strdup(opt);
180 
181 	emul = config = NULL;
182 	if ((cp = strchr(str, ',')) != NULL) {
183 		*cp = '\0';
184 		emul = cp + 1;
185 		if ((cp = strchr(emul, ',')) != NULL) {
186 			*cp = '\0';
187 			config = cp + 1;
188 		}
189 	} else {
190 		pci_parse_slot_usage(opt);
191 		goto done;
192 	}
193 
194 	/* <bus>:<slot>:<func> */
195 	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
196 		bnum = 0;
197 		/* <slot>:<func> */
198 		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
199 			fnum = 0;
200 			/* <slot> */
201 			if (sscanf(str, "%d", &snum) != 1) {
202 				snum = -1;
203 			}
204 		}
205 	}
206 
207 	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
208 	    fnum < 0 || fnum >= MAXFUNCS) {
209 		pci_parse_slot_usage(opt);
210 		goto done;
211 	}
212 
213 	if (pci_businfo[bnum] == NULL)
214 		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
215 
216 	bi = pci_businfo[bnum];
217 	si = &bi->slotinfo[snum];
218 
219 	if (si->si_funcs[fnum].fi_name != NULL) {
220 		EPRINTLN("pci slot %d:%d already occupied!",
221 			snum, fnum);
222 		goto done;
223 	}
224 
225 	if (pci_emul_finddev(emul) == NULL) {
226 		EPRINTLN("pci slot %d:%d: unknown device \"%s\"",
227 			snum, fnum, emul);
228 		goto done;
229 	}
230 
231 	error = 0;
232 	si->si_funcs[fnum].fi_name = emul;
233 	si->si_funcs[fnum].fi_param = config;
234 
235 done:
236 	if (error)
237 		free(str);
238 
239 	return (error);
240 }
241 
242 void
243 pci_print_supported_devices()
244 {
245 	struct pci_devemu **pdpp, *pdp;
246 
247 	SET_FOREACH(pdpp, pci_devemu_set) {
248 		pdp = *pdpp;
249 		printf("%s\n", pdp->pe_emu);
250 	}
251 }
252 
253 static int
254 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
255 {
256 
257 	if (offset < pi->pi_msix.pba_offset)
258 		return (0);
259 
260 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
261 		return (0);
262 	}
263 
264 	return (1);
265 }
266 
267 int
268 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
269 		     uint64_t value)
270 {
271 	int msix_entry_offset;
272 	int tab_index;
273 	char *dest;
274 
275 	/* support only 4 or 8 byte writes */
276 	if (size != 4 && size != 8)
277 		return (-1);
278 
279 	/*
280 	 * Return if table index is beyond what device supports
281 	 */
282 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
283 	if (tab_index >= pi->pi_msix.table_count)
284 		return (-1);
285 
286 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
287 
288 	/* support only aligned writes */
289 	if ((msix_entry_offset % size) != 0)
290 		return (-1);
291 
292 	dest = (char *)(pi->pi_msix.table + tab_index);
293 	dest += msix_entry_offset;
294 
295 	if (size == 4)
296 		*((uint32_t *)dest) = value;
297 	else
298 		*((uint64_t *)dest) = value;
299 
300 	return (0);
301 }
302 
303 uint64_t
304 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
305 {
306 	char *dest;
307 	int msix_entry_offset;
308 	int tab_index;
309 	uint64_t retval = ~0;
310 
311 	/*
312 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
313 	 * table but we also allow 1 byte access to accommodate reads from
314 	 * ddb.
315 	 */
316 	if (size != 1 && size != 4 && size != 8)
317 		return (retval);
318 
319 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
320 
321 	/* support only aligned reads */
322 	if ((msix_entry_offset % size) != 0) {
323 		return (retval);
324 	}
325 
326 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
327 
328 	if (tab_index < pi->pi_msix.table_count) {
329 		/* valid MSI-X Table access */
330 		dest = (char *)(pi->pi_msix.table + tab_index);
331 		dest += msix_entry_offset;
332 
333 		if (size == 1)
334 			retval = *((uint8_t *)dest);
335 		else if (size == 4)
336 			retval = *((uint32_t *)dest);
337 		else
338 			retval = *((uint64_t *)dest);
339 	} else if (pci_valid_pba_offset(pi, offset)) {
340 		/* return 0 for PBA access */
341 		retval = 0;
342 	}
343 
344 	return (retval);
345 }
346 
347 int
348 pci_msix_table_bar(struct pci_devinst *pi)
349 {
350 
351 	if (pi->pi_msix.table != NULL)
352 		return (pi->pi_msix.table_bar);
353 	else
354 		return (-1);
355 }
356 
357 int
358 pci_msix_pba_bar(struct pci_devinst *pi)
359 {
360 
361 	if (pi->pi_msix.table != NULL)
362 		return (pi->pi_msix.pba_bar);
363 	else
364 		return (-1);
365 }
366 
367 static int
368 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
369 		    uint32_t *eax, void *arg)
370 {
371 	struct pci_devinst *pdi = arg;
372 	struct pci_devemu *pe = pdi->pi_d;
373 	uint64_t offset;
374 	int i;
375 
376 	for (i = 0; i <= PCI_BARMAX; i++) {
377 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
378 		    port >= pdi->pi_bar[i].addr &&
379 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
380 			offset = port - pdi->pi_bar[i].addr;
381 			if (in)
382 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
383 							 offset, bytes);
384 			else
385 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
386 						   bytes, *eax);
387 			return (0);
388 		}
389 	}
390 	return (-1);
391 }
392 
393 static int
394 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
395 		     int size, uint64_t *val, void *arg1, long arg2)
396 {
397 	struct pci_devinst *pdi = arg1;
398 	struct pci_devemu *pe = pdi->pi_d;
399 	uint64_t offset;
400 	int bidx = (int) arg2;
401 
402 	assert(bidx <= PCI_BARMAX);
403 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
404 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
405 	assert(addr >= pdi->pi_bar[bidx].addr &&
406 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
407 
408 	offset = addr - pdi->pi_bar[bidx].addr;
409 
410 	if (dir == MEM_F_WRITE) {
411 		if (size == 8) {
412 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
413 					   4, *val & 0xffffffff);
414 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
415 					   4, *val >> 32);
416 		} else {
417 			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
418 					   size, *val);
419 		}
420 	} else {
421 		if (size == 8) {
422 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
423 						 offset, 4);
424 			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
425 						  offset + 4, 4) << 32;
426 		} else {
427 			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
428 						 offset, size);
429 		}
430 	}
431 
432 	return (0);
433 }
434 
435 
436 static int
437 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
438 			uint64_t *addr)
439 {
440 	uint64_t base;
441 
442 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
443 
444 	base = roundup2(*baseptr, size);
445 
446 	if (base + size <= limit) {
447 		*addr = base;
448 		*baseptr = base + size;
449 		return (0);
450 	} else
451 		return (-1);
452 }
453 
454 int
455 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
456 		   uint64_t size)
457 {
458 
459 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
460 }
461 
462 /*
463  * Register (or unregister) the MMIO or I/O region associated with the BAR
464  * register 'idx' of an emulated pci device.
465  */
466 static void
467 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
468 {
469 	int error;
470 	struct inout_port iop;
471 	struct mem_range mr;
472 
473 	switch (pi->pi_bar[idx].type) {
474 	case PCIBAR_IO:
475 		bzero(&iop, sizeof(struct inout_port));
476 		iop.name = pi->pi_name;
477 		iop.port = pi->pi_bar[idx].addr;
478 		iop.size = pi->pi_bar[idx].size;
479 		if (registration) {
480 			iop.flags = IOPORT_F_INOUT;
481 			iop.handler = pci_emul_io_handler;
482 			iop.arg = pi;
483 			error = register_inout(&iop);
484 		} else
485 			error = unregister_inout(&iop);
486 		break;
487 	case PCIBAR_MEM32:
488 	case PCIBAR_MEM64:
489 		bzero(&mr, sizeof(struct mem_range));
490 		mr.name = pi->pi_name;
491 		mr.base = pi->pi_bar[idx].addr;
492 		mr.size = pi->pi_bar[idx].size;
493 		if (registration) {
494 			mr.flags = MEM_F_RW;
495 			mr.handler = pci_emul_mem_handler;
496 			mr.arg1 = pi;
497 			mr.arg2 = idx;
498 			error = register_mem(&mr);
499 		} else
500 			error = unregister_mem(&mr);
501 		break;
502 	default:
503 		error = EINVAL;
504 		break;
505 	}
506 	assert(error == 0);
507 }
508 
509 static void
510 unregister_bar(struct pci_devinst *pi, int idx)
511 {
512 
513 	modify_bar_registration(pi, idx, 0);
514 }
515 
516 static void
517 register_bar(struct pci_devinst *pi, int idx)
518 {
519 
520 	modify_bar_registration(pi, idx, 1);
521 }
522 
523 /* Are we decoding i/o port accesses for the emulated pci device? */
524 static int
525 porten(struct pci_devinst *pi)
526 {
527 	uint16_t cmd;
528 
529 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
530 
531 	return (cmd & PCIM_CMD_PORTEN);
532 }
533 
534 /* Are we decoding memory accesses for the emulated pci device? */
535 static int
536 memen(struct pci_devinst *pi)
537 {
538 	uint16_t cmd;
539 
540 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
541 
542 	return (cmd & PCIM_CMD_MEMEN);
543 }
544 
545 /*
546  * Update the MMIO or I/O address that is decoded by the BAR register.
547  *
548  * If the pci device has enabled the address space decoding then intercept
549  * the address range decoded by the BAR register.
550  */
551 static void
552 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
553 {
554 	int decode;
555 
556 	if (pi->pi_bar[idx].type == PCIBAR_IO)
557 		decode = porten(pi);
558 	else
559 		decode = memen(pi);
560 
561 	if (decode)
562 		unregister_bar(pi, idx);
563 
564 	switch (type) {
565 	case PCIBAR_IO:
566 	case PCIBAR_MEM32:
567 		pi->pi_bar[idx].addr = addr;
568 		break;
569 	case PCIBAR_MEM64:
570 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
571 		pi->pi_bar[idx].addr |= addr;
572 		break;
573 	case PCIBAR_MEMHI64:
574 		pi->pi_bar[idx].addr &= 0xffffffff;
575 		pi->pi_bar[idx].addr |= addr;
576 		break;
577 	default:
578 		assert(0);
579 	}
580 
581 	if (decode)
582 		register_bar(pi, idx);
583 }
584 
585 int
586 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
587 		    enum pcibar_type type, uint64_t size)
588 {
589 	int error;
590 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
591 	uint16_t cmd, enbit;
592 
593 	assert(idx >= 0 && idx <= PCI_BARMAX);
594 
595 	if ((size & (size - 1)) != 0)
596 		size = 1UL << flsl(size);	/* round up to a power of 2 */
597 
598 	/* Enforce minimum BAR sizes required by the PCI standard */
599 	if (type == PCIBAR_IO) {
600 		if (size < 4)
601 			size = 4;
602 	} else {
603 		if (size < 16)
604 			size = 16;
605 	}
606 
607 	switch (type) {
608 	case PCIBAR_NONE:
609 		baseptr = NULL;
610 		addr = mask = lobits = enbit = 0;
611 		break;
612 	case PCIBAR_IO:
613 		baseptr = &pci_emul_iobase;
614 		limit = PCI_EMUL_IOLIMIT;
615 		mask = PCIM_BAR_IO_BASE;
616 		lobits = PCIM_BAR_IO_SPACE;
617 		enbit = PCIM_CMD_PORTEN;
618 		break;
619 	case PCIBAR_MEM64:
620 		/*
621 		 * XXX
622 		 * Some drivers do not work well if the 64-bit BAR is allocated
623 		 * above 4GB. Allow for this by allocating small requests under
624 		 * 4GB unless then allocation size is larger than some arbitrary
625 		 * number (32MB currently).
626 		 */
627 		if (size > 32 * 1024 * 1024) {
628 			/*
629 			 * XXX special case for device requiring peer-peer DMA
630 			 */
631 			if (size == 0x100000000UL)
632 				baseptr = &hostbase;
633 			else
634 				baseptr = &pci_emul_membase64;
635 			limit = PCI_EMUL_MEMLIMIT64;
636 			mask = PCIM_BAR_MEM_BASE;
637 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
638 				 PCIM_BAR_MEM_PREFETCH;
639 		} else {
640 			baseptr = &pci_emul_membase32;
641 			limit = PCI_EMUL_MEMLIMIT32;
642 			mask = PCIM_BAR_MEM_BASE;
643 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
644 		}
645 		enbit = PCIM_CMD_MEMEN;
646 		break;
647 	case PCIBAR_MEM32:
648 		baseptr = &pci_emul_membase32;
649 		limit = PCI_EMUL_MEMLIMIT32;
650 		mask = PCIM_BAR_MEM_BASE;
651 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
652 		enbit = PCIM_CMD_MEMEN;
653 		break;
654 	default:
655 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
656 		assert(0);
657 	}
658 
659 	if (baseptr != NULL) {
660 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
661 		if (error != 0)
662 			return (error);
663 	}
664 
665 	pdi->pi_bar[idx].type = type;
666 	pdi->pi_bar[idx].addr = addr;
667 	pdi->pi_bar[idx].size = size;
668 
669 	/* Initialize the BAR register in config space */
670 	bar = (addr & mask) | lobits;
671 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
672 
673 	if (type == PCIBAR_MEM64) {
674 		assert(idx + 1 <= PCI_BARMAX);
675 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
676 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
677 	}
678 
679 	cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
680 	if ((cmd & enbit) != enbit)
681 		pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
682 	register_bar(pdi, idx);
683 
684 	return (0);
685 }
686 
687 #define	CAP_START_OFFSET	0x40
688 static int
689 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
690 {
691 	int i, capoff, reallen;
692 	uint16_t sts;
693 
694 	assert(caplen > 0);
695 
696 	reallen = roundup2(caplen, 4);		/* dword aligned */
697 
698 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
699 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
700 		capoff = CAP_START_OFFSET;
701 	else
702 		capoff = pi->pi_capend + 1;
703 
704 	/* Check if we have enough space */
705 	if (capoff + reallen > PCI_REGMAX + 1)
706 		return (-1);
707 
708 	/* Set the previous capability pointer */
709 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
710 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
711 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
712 	} else
713 		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
714 
715 	/* Copy the capability */
716 	for (i = 0; i < caplen; i++)
717 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
718 
719 	/* Set the next capability pointer */
720 	pci_set_cfgdata8(pi, capoff + 1, 0);
721 
722 	pi->pi_prevcap = capoff;
723 	pi->pi_capend = capoff + reallen - 1;
724 	return (0);
725 }
726 
727 static struct pci_devemu *
728 pci_emul_finddev(char *name)
729 {
730 	struct pci_devemu **pdpp, *pdp;
731 
732 	SET_FOREACH(pdpp, pci_devemu_set) {
733 		pdp = *pdpp;
734 		if (!strcmp(pdp->pe_emu, name)) {
735 			return (pdp);
736 		}
737 	}
738 
739 	return (NULL);
740 }
741 
742 static int
743 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
744     int func, struct funcinfo *fi)
745 {
746 	struct pci_devinst *pdi;
747 	int err;
748 
749 	pdi = calloc(1, sizeof(struct pci_devinst));
750 
751 	pdi->pi_vmctx = ctx;
752 	pdi->pi_bus = bus;
753 	pdi->pi_slot = slot;
754 	pdi->pi_func = func;
755 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
756 	pdi->pi_lintr.pin = 0;
757 	pdi->pi_lintr.state = IDLE;
758 	pdi->pi_lintr.pirq_pin = 0;
759 	pdi->pi_lintr.ioapic_irq = 0;
760 	pdi->pi_d = pde;
761 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
762 
763 	/* Disable legacy interrupts */
764 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
765 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
766 
767 	pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
768 
769 	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
770 	if (err == 0)
771 		fi->fi_devi = pdi;
772 	else
773 		free(pdi);
774 
775 	return (err);
776 }
777 
778 void
779 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
780 {
781 	int mmc;
782 
783 	/* Number of msi messages must be a power of 2 between 1 and 32 */
784 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
785 	mmc = ffs(msgnum) - 1;
786 
787 	bzero(msicap, sizeof(struct msicap));
788 	msicap->capid = PCIY_MSI;
789 	msicap->nextptr = nextptr;
790 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
791 }
792 
793 int
794 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
795 {
796 	struct msicap msicap;
797 
798 	pci_populate_msicap(&msicap, msgnum, 0);
799 
800 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
801 }
802 
803 static void
804 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
805 		     uint32_t msix_tab_size)
806 {
807 
808 	assert(msix_tab_size % 4096 == 0);
809 
810 	bzero(msixcap, sizeof(struct msixcap));
811 	msixcap->capid = PCIY_MSIX;
812 
813 	/*
814 	 * Message Control Register, all fields set to
815 	 * zero except for the Table Size.
816 	 * Note: Table size N is encoded as N-1
817 	 */
818 	msixcap->msgctrl = msgnum - 1;
819 
820 	/*
821 	 * MSI-X BAR setup:
822 	 * - MSI-X table start at offset 0
823 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
824 	 */
825 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
826 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
827 }
828 
829 static void
830 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
831 {
832 	int i, table_size;
833 
834 	assert(table_entries > 0);
835 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
836 
837 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
838 	pi->pi_msix.table = calloc(1, table_size);
839 
840 	/* set mask bit of vector control register */
841 	for (i = 0; i < table_entries; i++)
842 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
843 }
844 
845 int
846 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
847 {
848 	uint32_t tab_size;
849 	struct msixcap msixcap;
850 
851 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
852 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
853 
854 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
855 
856 	/* Align table size to nearest 4K */
857 	tab_size = roundup2(tab_size, 4096);
858 
859 	pi->pi_msix.table_bar = barnum;
860 	pi->pi_msix.pba_bar   = barnum;
861 	pi->pi_msix.table_offset = 0;
862 	pi->pi_msix.table_count = msgnum;
863 	pi->pi_msix.pba_offset = tab_size;
864 	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
865 
866 	pci_msix_table_init(pi, msgnum);
867 
868 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
869 
870 	/* allocate memory for MSI-X Table and PBA */
871 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
872 				tab_size + pi->pi_msix.pba_size);
873 
874 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
875 					sizeof(msixcap)));
876 }
877 
878 void
879 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
880 		 int bytes, uint32_t val)
881 {
882 	uint16_t msgctrl, rwmask;
883 	int off;
884 
885 	off = offset - capoff;
886 	/* Message Control Register */
887 	if (off == 2 && bytes == 2) {
888 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
889 		msgctrl = pci_get_cfgdata16(pi, offset);
890 		msgctrl &= ~rwmask;
891 		msgctrl |= val & rwmask;
892 		val = msgctrl;
893 
894 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
895 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
896 		pci_lintr_update(pi);
897 	}
898 
899 	CFGWRITE(pi, offset, val, bytes);
900 }
901 
902 void
903 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
904 		int bytes, uint32_t val)
905 {
906 	uint16_t msgctrl, rwmask, msgdata, mme;
907 	uint32_t addrlo;
908 
909 	/*
910 	 * If guest is writing to the message control register make sure
911 	 * we do not overwrite read-only fields.
912 	 */
913 	if ((offset - capoff) == 2 && bytes == 2) {
914 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
915 		msgctrl = pci_get_cfgdata16(pi, offset);
916 		msgctrl &= ~rwmask;
917 		msgctrl |= val & rwmask;
918 		val = msgctrl;
919 	}
920 	CFGWRITE(pi, offset, val, bytes);
921 
922 	msgctrl = pci_get_cfgdata16(pi, capoff + 2);
923 	addrlo = pci_get_cfgdata32(pi, capoff + 4);
924 	if (msgctrl & PCIM_MSICTRL_64BIT)
925 		msgdata = pci_get_cfgdata16(pi, capoff + 12);
926 	else
927 		msgdata = pci_get_cfgdata16(pi, capoff + 8);
928 
929 	mme = msgctrl & PCIM_MSICTRL_MME_MASK;
930 	pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
931 	if (pi->pi_msi.enabled) {
932 		pi->pi_msi.addr = addrlo;
933 		pi->pi_msi.msg_data = msgdata;
934 		pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
935 	} else {
936 		pi->pi_msi.maxmsgnum = 0;
937 	}
938 	pci_lintr_update(pi);
939 }
940 
941 void
942 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
943 		 int bytes, uint32_t val)
944 {
945 
946 	/* XXX don't write to the readonly parts */
947 	CFGWRITE(pi, offset, val, bytes);
948 }
949 
950 #define	PCIECAP_VERSION	0x2
951 int
952 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
953 {
954 	int err;
955 	struct pciecap pciecap;
956 
957 	bzero(&pciecap, sizeof(pciecap));
958 
959 	/*
960 	 * Use the integrated endpoint type for endpoints on a root complex bus.
961 	 *
962 	 * NB: bhyve currently only supports a single PCI bus that is the root
963 	 * complex bus, so all endpoints are integrated.
964 	 */
965 	if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
966 		type = PCIEM_TYPE_ROOT_INT_EP;
967 
968 	pciecap.capid = PCIY_EXPRESS;
969 	pciecap.pcie_capabilities = PCIECAP_VERSION | type;
970 	if (type != PCIEM_TYPE_ROOT_INT_EP) {
971 		pciecap.link_capabilities = 0x411;	/* gen1, x1 */
972 		pciecap.link_status = 0x11;		/* gen1, x1 */
973 	}
974 
975 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
976 	return (err);
977 }
978 
979 /*
980  * This function assumes that 'coff' is in the capabilities region of the
981  * config space.
982  */
983 static void
984 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
985 {
986 	int capid;
987 	uint8_t capoff, nextoff;
988 
989 	/* Do not allow un-aligned writes */
990 	if ((offset & (bytes - 1)) != 0)
991 		return;
992 
993 	/* Find the capability that we want to update */
994 	capoff = CAP_START_OFFSET;
995 	while (1) {
996 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
997 		if (nextoff == 0)
998 			break;
999 		if (offset >= capoff && offset < nextoff)
1000 			break;
1001 
1002 		capoff = nextoff;
1003 	}
1004 	assert(offset >= capoff);
1005 
1006 	/*
1007 	 * Capability ID and Next Capability Pointer are readonly.
1008 	 * However, some o/s's do 4-byte writes that include these.
1009 	 * For this case, trim the write back to 2 bytes and adjust
1010 	 * the data.
1011 	 */
1012 	if (offset == capoff || offset == capoff + 1) {
1013 		if (offset == capoff && bytes == 4) {
1014 			bytes = 2;
1015 			offset += 2;
1016 			val >>= 16;
1017 		} else
1018 			return;
1019 	}
1020 
1021 	capid = pci_get_cfgdata8(pi, capoff);
1022 	switch (capid) {
1023 	case PCIY_MSI:
1024 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1025 		break;
1026 	case PCIY_MSIX:
1027 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1028 		break;
1029 	case PCIY_EXPRESS:
1030 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1031 		break;
1032 	default:
1033 		break;
1034 	}
1035 }
1036 
1037 static int
1038 pci_emul_iscap(struct pci_devinst *pi, int offset)
1039 {
1040 	uint16_t sts;
1041 
1042 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1043 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1044 		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1045 			return (1);
1046 	}
1047 	return (0);
1048 }
1049 
1050 static int
1051 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1052 			  int size, uint64_t *val, void *arg1, long arg2)
1053 {
1054 	/*
1055 	 * Ignore writes; return 0xff's for reads. The mem read code
1056 	 * will take care of truncating to the correct size.
1057 	 */
1058 	if (dir == MEM_F_READ) {
1059 		*val = 0xffffffffffffffff;
1060 	}
1061 
1062 	return (0);
1063 }
1064 
1065 static int
1066 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1067     int bytes, uint64_t *val, void *arg1, long arg2)
1068 {
1069 	int bus, slot, func, coff, in;
1070 
1071 	coff = addr & 0xfff;
1072 	func = (addr >> 12) & 0x7;
1073 	slot = (addr >> 15) & 0x1f;
1074 	bus = (addr >> 20) & 0xff;
1075 	in = (dir == MEM_F_READ);
1076 	if (in)
1077 		*val = ~0UL;
1078 	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1079 	return (0);
1080 }
1081 
1082 uint64_t
1083 pci_ecfg_base(void)
1084 {
1085 
1086 	return (PCI_EMUL_ECFG_BASE);
1087 }
1088 
1089 #define	BUSIO_ROUNDUP		32
1090 #define	BUSMEM_ROUNDUP		(1024 * 1024)
1091 
1092 int
1093 init_pci(struct vmctx *ctx)
1094 {
1095 	struct mem_range mr;
1096 	struct pci_devemu *pde;
1097 	struct businfo *bi;
1098 	struct slotinfo *si;
1099 	struct funcinfo *fi;
1100 	size_t lowmem;
1101 	int bus, slot, func;
1102 	int error;
1103 
1104 	pci_emul_iobase = PCI_EMUL_IOBASE;
1105 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1106 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1107 
1108 	for (bus = 0; bus < MAXBUSES; bus++) {
1109 		if ((bi = pci_businfo[bus]) == NULL)
1110 			continue;
1111 		/*
1112 		 * Keep track of the i/o and memory resources allocated to
1113 		 * this bus.
1114 		 */
1115 		bi->iobase = pci_emul_iobase;
1116 		bi->membase32 = pci_emul_membase32;
1117 		bi->membase64 = pci_emul_membase64;
1118 
1119 		for (slot = 0; slot < MAXSLOTS; slot++) {
1120 			si = &bi->slotinfo[slot];
1121 			for (func = 0; func < MAXFUNCS; func++) {
1122 				fi = &si->si_funcs[func];
1123 				if (fi->fi_name == NULL)
1124 					continue;
1125 				pde = pci_emul_finddev(fi->fi_name);
1126 				assert(pde != NULL);
1127 				error = pci_emul_init(ctx, pde, bus, slot,
1128 				    func, fi);
1129 				if (error)
1130 					return (error);
1131 			}
1132 		}
1133 
1134 		/*
1135 		 * Add some slop to the I/O and memory resources decoded by
1136 		 * this bus to give a guest some flexibility if it wants to
1137 		 * reprogram the BARs.
1138 		 */
1139 		pci_emul_iobase += BUSIO_ROUNDUP;
1140 		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1141 		bi->iolimit = pci_emul_iobase;
1142 
1143 		pci_emul_membase32 += BUSMEM_ROUNDUP;
1144 		pci_emul_membase32 = roundup2(pci_emul_membase32,
1145 		    BUSMEM_ROUNDUP);
1146 		bi->memlimit32 = pci_emul_membase32;
1147 
1148 		pci_emul_membase64 += BUSMEM_ROUNDUP;
1149 		pci_emul_membase64 = roundup2(pci_emul_membase64,
1150 		    BUSMEM_ROUNDUP);
1151 		bi->memlimit64 = pci_emul_membase64;
1152 	}
1153 
1154 	/*
1155 	 * PCI backends are initialized before routing INTx interrupts
1156 	 * so that LPC devices are able to reserve ISA IRQs before
1157 	 * routing PIRQ pins.
1158 	 */
1159 	for (bus = 0; bus < MAXBUSES; bus++) {
1160 		if ((bi = pci_businfo[bus]) == NULL)
1161 			continue;
1162 
1163 		for (slot = 0; slot < MAXSLOTS; slot++) {
1164 			si = &bi->slotinfo[slot];
1165 			for (func = 0; func < MAXFUNCS; func++) {
1166 				fi = &si->si_funcs[func];
1167 				if (fi->fi_devi == NULL)
1168 					continue;
1169 				pci_lintr_route(fi->fi_devi);
1170 			}
1171 		}
1172 	}
1173 	lpc_pirq_routed();
1174 
1175 	/*
1176 	 * The guest physical memory map looks like the following:
1177 	 * [0,		    lowmem)		guest system memory
1178 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1179 	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1180 	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1181 	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1182 	 * [4GB,	    4GB + highmem)
1183 	 */
1184 
1185 	/*
1186 	 * Accesses to memory addresses that are not allocated to system
1187 	 * memory or PCI devices return 0xff's.
1188 	 */
1189 	lowmem = vm_get_lowmem_size(ctx);
1190 	bzero(&mr, sizeof(struct mem_range));
1191 	mr.name = "PCI hole";
1192 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1193 	mr.base = lowmem;
1194 	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1195 	mr.handler = pci_emul_fallback_handler;
1196 	error = register_mem_fallback(&mr);
1197 	assert(error == 0);
1198 
1199 	/* PCI extended config space */
1200 	bzero(&mr, sizeof(struct mem_range));
1201 	mr.name = "PCI ECFG";
1202 	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1203 	mr.base = PCI_EMUL_ECFG_BASE;
1204 	mr.size = PCI_EMUL_ECFG_SIZE;
1205 	mr.handler = pci_emul_ecfg_handler;
1206 	error = register_mem(&mr);
1207 	assert(error == 0);
1208 
1209 	return (0);
1210 }
1211 
1212 static void
1213 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1214     void *arg)
1215 {
1216 
1217 	dsdt_line("  Package ()");
1218 	dsdt_line("  {");
1219 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1220 	dsdt_line("    0x%02X,", pin - 1);
1221 	dsdt_line("    Zero,");
1222 	dsdt_line("    0x%X", ioapic_irq);
1223 	dsdt_line("  },");
1224 }
1225 
1226 static void
1227 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1228     void *arg)
1229 {
1230 	char *name;
1231 
1232 	name = lpc_pirq_name(pirq_pin);
1233 	if (name == NULL)
1234 		return;
1235 	dsdt_line("  Package ()");
1236 	dsdt_line("  {");
1237 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1238 	dsdt_line("    0x%02X,", pin - 1);
1239 	dsdt_line("    %s,", name);
1240 	dsdt_line("    0x00");
1241 	dsdt_line("  },");
1242 	free(name);
1243 }
1244 
1245 /*
1246  * A bhyve virtual machine has a flat PCI hierarchy with a root port
1247  * corresponding to each PCI bus.
1248  */
1249 static void
1250 pci_bus_write_dsdt(int bus)
1251 {
1252 	struct businfo *bi;
1253 	struct slotinfo *si;
1254 	struct pci_devinst *pi;
1255 	int count, func, slot;
1256 
1257 	/*
1258 	 * If there are no devices on this 'bus' then just return.
1259 	 */
1260 	if ((bi = pci_businfo[bus]) == NULL) {
1261 		/*
1262 		 * Bus 0 is special because it decodes the I/O ports used
1263 		 * for PCI config space access even if there are no devices
1264 		 * on it.
1265 		 */
1266 		if (bus != 0)
1267 			return;
1268 	}
1269 
1270 	dsdt_line("  Device (PC%02X)", bus);
1271 	dsdt_line("  {");
1272 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1273 
1274 	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1275 	dsdt_line("    {");
1276 	dsdt_line("        Return (0x%08X)", bus);
1277 	dsdt_line("    }");
1278 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1279 	dsdt_line("    {");
1280 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1281 	    "MaxFixed, PosDecode,");
1282 	dsdt_line("        0x0000,             // Granularity");
1283 	dsdt_line("        0x%04X,             // Range Minimum", bus);
1284 	dsdt_line("        0x%04X,             // Range Maximum", bus);
1285 	dsdt_line("        0x0000,             // Translation Offset");
1286 	dsdt_line("        0x0001,             // Length");
1287 	dsdt_line("        ,, )");
1288 
1289 	if (bus == 0) {
1290 		dsdt_indent(3);
1291 		dsdt_fixed_ioport(0xCF8, 8);
1292 		dsdt_unindent(3);
1293 
1294 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1295 		    "PosDecode, EntireRange,");
1296 		dsdt_line("        0x0000,             // Granularity");
1297 		dsdt_line("        0x0000,             // Range Minimum");
1298 		dsdt_line("        0x0CF7,             // Range Maximum");
1299 		dsdt_line("        0x0000,             // Translation Offset");
1300 		dsdt_line("        0x0CF8,             // Length");
1301 		dsdt_line("        ,, , TypeStatic)");
1302 
1303 		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1304 		    "PosDecode, EntireRange,");
1305 		dsdt_line("        0x0000,             // Granularity");
1306 		dsdt_line("        0x0D00,             // Range Minimum");
1307 		dsdt_line("        0x%04X,             // Range Maximum",
1308 		    PCI_EMUL_IOBASE - 1);
1309 		dsdt_line("        0x0000,             // Translation Offset");
1310 		dsdt_line("        0x%04X,             // Length",
1311 		    PCI_EMUL_IOBASE - 0x0D00);
1312 		dsdt_line("        ,, , TypeStatic)");
1313 
1314 		if (bi == NULL) {
1315 			dsdt_line("    })");
1316 			goto done;
1317 		}
1318 	}
1319 	assert(bi != NULL);
1320 
1321 	/* i/o window */
1322 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1323 	    "PosDecode, EntireRange,");
1324 	dsdt_line("        0x0000,             // Granularity");
1325 	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1326 	dsdt_line("        0x%04X,             // Range Maximum",
1327 	    bi->iolimit - 1);
1328 	dsdt_line("        0x0000,             // Translation Offset");
1329 	dsdt_line("        0x%04X,             // Length",
1330 	    bi->iolimit - bi->iobase);
1331 	dsdt_line("        ,, , TypeStatic)");
1332 
1333 	/* mmio window (32-bit) */
1334 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1335 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1336 	dsdt_line("        0x00000000,         // Granularity");
1337 	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1338 	dsdt_line("        0x%08X,         // Range Maximum\n",
1339 	    bi->memlimit32 - 1);
1340 	dsdt_line("        0x00000000,         // Translation Offset");
1341 	dsdt_line("        0x%08X,         // Length\n",
1342 	    bi->memlimit32 - bi->membase32);
1343 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1344 
1345 	/* mmio window (64-bit) */
1346 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1347 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1348 	dsdt_line("        0x0000000000000000, // Granularity");
1349 	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1350 	dsdt_line("        0x%016lX, // Range Maximum\n",
1351 	    bi->memlimit64 - 1);
1352 	dsdt_line("        0x0000000000000000, // Translation Offset");
1353 	dsdt_line("        0x%016lX, // Length\n",
1354 	    bi->memlimit64 - bi->membase64);
1355 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1356 	dsdt_line("    })");
1357 
1358 	count = pci_count_lintr(bus);
1359 	if (count != 0) {
1360 		dsdt_indent(2);
1361 		dsdt_line("Name (PPRT, Package ()");
1362 		dsdt_line("{");
1363 		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1364 		dsdt_line("})");
1365 		dsdt_line("Name (APRT, Package ()");
1366 		dsdt_line("{");
1367 		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1368 		dsdt_line("})");
1369 		dsdt_line("Method (_PRT, 0, NotSerialized)");
1370 		dsdt_line("{");
1371 		dsdt_line("  If (PICM)");
1372 		dsdt_line("  {");
1373 		dsdt_line("    Return (APRT)");
1374 		dsdt_line("  }");
1375 		dsdt_line("  Else");
1376 		dsdt_line("  {");
1377 		dsdt_line("    Return (PPRT)");
1378 		dsdt_line("  }");
1379 		dsdt_line("}");
1380 		dsdt_unindent(2);
1381 	}
1382 
1383 	dsdt_indent(2);
1384 	for (slot = 0; slot < MAXSLOTS; slot++) {
1385 		si = &bi->slotinfo[slot];
1386 		for (func = 0; func < MAXFUNCS; func++) {
1387 			pi = si->si_funcs[func].fi_devi;
1388 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1389 				pi->pi_d->pe_write_dsdt(pi);
1390 		}
1391 	}
1392 	dsdt_unindent(2);
1393 done:
1394 	dsdt_line("  }");
1395 }
1396 
1397 void
1398 pci_write_dsdt(void)
1399 {
1400 	int bus;
1401 
1402 	dsdt_indent(1);
1403 	dsdt_line("Name (PICM, 0x00)");
1404 	dsdt_line("Method (_PIC, 1, NotSerialized)");
1405 	dsdt_line("{");
1406 	dsdt_line("  Store (Arg0, PICM)");
1407 	dsdt_line("}");
1408 	dsdt_line("");
1409 	dsdt_line("Scope (_SB)");
1410 	dsdt_line("{");
1411 	for (bus = 0; bus < MAXBUSES; bus++)
1412 		pci_bus_write_dsdt(bus);
1413 	dsdt_line("}");
1414 	dsdt_unindent(1);
1415 }
1416 
1417 int
1418 pci_bus_configured(int bus)
1419 {
1420 	assert(bus >= 0 && bus < MAXBUSES);
1421 	return (pci_businfo[bus] != NULL);
1422 }
1423 
1424 int
1425 pci_msi_enabled(struct pci_devinst *pi)
1426 {
1427 	return (pi->pi_msi.enabled);
1428 }
1429 
1430 int
1431 pci_msi_maxmsgnum(struct pci_devinst *pi)
1432 {
1433 	if (pi->pi_msi.enabled)
1434 		return (pi->pi_msi.maxmsgnum);
1435 	else
1436 		return (0);
1437 }
1438 
1439 int
1440 pci_msix_enabled(struct pci_devinst *pi)
1441 {
1442 
1443 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1444 }
1445 
1446 void
1447 pci_generate_msix(struct pci_devinst *pi, int index)
1448 {
1449 	struct msix_table_entry *mte;
1450 
1451 	if (!pci_msix_enabled(pi))
1452 		return;
1453 
1454 	if (pi->pi_msix.function_mask)
1455 		return;
1456 
1457 	if (index >= pi->pi_msix.table_count)
1458 		return;
1459 
1460 	mte = &pi->pi_msix.table[index];
1461 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1462 		/* XXX Set PBA bit if interrupt is disabled */
1463 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1464 	}
1465 }
1466 
1467 void
1468 pci_generate_msi(struct pci_devinst *pi, int index)
1469 {
1470 
1471 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1472 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1473 			     pi->pi_msi.msg_data + index);
1474 	}
1475 }
1476 
1477 static bool
1478 pci_lintr_permitted(struct pci_devinst *pi)
1479 {
1480 	uint16_t cmd;
1481 
1482 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1483 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1484 		(cmd & PCIM_CMD_INTxDIS)));
1485 }
1486 
1487 void
1488 pci_lintr_request(struct pci_devinst *pi)
1489 {
1490 	struct businfo *bi;
1491 	struct slotinfo *si;
1492 	int bestpin, bestcount, pin;
1493 
1494 	bi = pci_businfo[pi->pi_bus];
1495 	assert(bi != NULL);
1496 
1497 	/*
1498 	 * Just allocate a pin from our slot.  The pin will be
1499 	 * assigned IRQs later when interrupts are routed.
1500 	 */
1501 	si = &bi->slotinfo[pi->pi_slot];
1502 	bestpin = 0;
1503 	bestcount = si->si_intpins[0].ii_count;
1504 	for (pin = 1; pin < 4; pin++) {
1505 		if (si->si_intpins[pin].ii_count < bestcount) {
1506 			bestpin = pin;
1507 			bestcount = si->si_intpins[pin].ii_count;
1508 		}
1509 	}
1510 
1511 	si->si_intpins[bestpin].ii_count++;
1512 	pi->pi_lintr.pin = bestpin + 1;
1513 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1514 }
1515 
1516 static void
1517 pci_lintr_route(struct pci_devinst *pi)
1518 {
1519 	struct businfo *bi;
1520 	struct intxinfo *ii;
1521 
1522 	if (pi->pi_lintr.pin == 0)
1523 		return;
1524 
1525 	bi = pci_businfo[pi->pi_bus];
1526 	assert(bi != NULL);
1527 	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1528 
1529 	/*
1530 	 * Attempt to allocate an I/O APIC pin for this intpin if one
1531 	 * is not yet assigned.
1532 	 */
1533 	if (ii->ii_ioapic_irq == 0)
1534 		ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1535 	assert(ii->ii_ioapic_irq > 0);
1536 
1537 	/*
1538 	 * Attempt to allocate a PIRQ pin for this intpin if one is
1539 	 * not yet assigned.
1540 	 */
1541 	if (ii->ii_pirq_pin == 0)
1542 		ii->ii_pirq_pin = pirq_alloc_pin(pi);
1543 	assert(ii->ii_pirq_pin > 0);
1544 
1545 	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1546 	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1547 	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1548 }
1549 
1550 void
1551 pci_lintr_assert(struct pci_devinst *pi)
1552 {
1553 
1554 	assert(pi->pi_lintr.pin > 0);
1555 
1556 	pthread_mutex_lock(&pi->pi_lintr.lock);
1557 	if (pi->pi_lintr.state == IDLE) {
1558 		if (pci_lintr_permitted(pi)) {
1559 			pi->pi_lintr.state = ASSERTED;
1560 			pci_irq_assert(pi);
1561 		} else
1562 			pi->pi_lintr.state = PENDING;
1563 	}
1564 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1565 }
1566 
1567 void
1568 pci_lintr_deassert(struct pci_devinst *pi)
1569 {
1570 
1571 	assert(pi->pi_lintr.pin > 0);
1572 
1573 	pthread_mutex_lock(&pi->pi_lintr.lock);
1574 	if (pi->pi_lintr.state == ASSERTED) {
1575 		pi->pi_lintr.state = IDLE;
1576 		pci_irq_deassert(pi);
1577 	} else if (pi->pi_lintr.state == PENDING)
1578 		pi->pi_lintr.state = IDLE;
1579 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1580 }
1581 
1582 static void
1583 pci_lintr_update(struct pci_devinst *pi)
1584 {
1585 
1586 	pthread_mutex_lock(&pi->pi_lintr.lock);
1587 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1588 		pci_irq_deassert(pi);
1589 		pi->pi_lintr.state = PENDING;
1590 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1591 		pi->pi_lintr.state = ASSERTED;
1592 		pci_irq_assert(pi);
1593 	}
1594 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1595 }
1596 
1597 int
1598 pci_count_lintr(int bus)
1599 {
1600 	int count, slot, pin;
1601 	struct slotinfo *slotinfo;
1602 
1603 	count = 0;
1604 	if (pci_businfo[bus] != NULL) {
1605 		for (slot = 0; slot < MAXSLOTS; slot++) {
1606 			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1607 			for (pin = 0; pin < 4; pin++) {
1608 				if (slotinfo->si_intpins[pin].ii_count != 0)
1609 					count++;
1610 			}
1611 		}
1612 	}
1613 	return (count);
1614 }
1615 
1616 void
1617 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1618 {
1619 	struct businfo *bi;
1620 	struct slotinfo *si;
1621 	struct intxinfo *ii;
1622 	int slot, pin;
1623 
1624 	if ((bi = pci_businfo[bus]) == NULL)
1625 		return;
1626 
1627 	for (slot = 0; slot < MAXSLOTS; slot++) {
1628 		si = &bi->slotinfo[slot];
1629 		for (pin = 0; pin < 4; pin++) {
1630 			ii = &si->si_intpins[pin];
1631 			if (ii->ii_count != 0)
1632 				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1633 				    ii->ii_ioapic_irq, arg);
1634 		}
1635 	}
1636 }
1637 
1638 /*
1639  * Return 1 if the emulated device in 'slot' is a multi-function device.
1640  * Return 0 otherwise.
1641  */
1642 static int
1643 pci_emul_is_mfdev(int bus, int slot)
1644 {
1645 	struct businfo *bi;
1646 	struct slotinfo *si;
1647 	int f, numfuncs;
1648 
1649 	numfuncs = 0;
1650 	if ((bi = pci_businfo[bus]) != NULL) {
1651 		si = &bi->slotinfo[slot];
1652 		for (f = 0; f < MAXFUNCS; f++) {
1653 			if (si->si_funcs[f].fi_devi != NULL) {
1654 				numfuncs++;
1655 			}
1656 		}
1657 	}
1658 	return (numfuncs > 1);
1659 }
1660 
1661 /*
1662  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1663  * whether or not is a multi-function being emulated in the pci 'slot'.
1664  */
1665 static void
1666 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1667 {
1668 	int mfdev;
1669 
1670 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1671 		mfdev = pci_emul_is_mfdev(bus, slot);
1672 		switch (bytes) {
1673 		case 1:
1674 		case 2:
1675 			*rv &= ~PCIM_MFDEV;
1676 			if (mfdev) {
1677 				*rv |= PCIM_MFDEV;
1678 			}
1679 			break;
1680 		case 4:
1681 			*rv &= ~(PCIM_MFDEV << 16);
1682 			if (mfdev) {
1683 				*rv |= (PCIM_MFDEV << 16);
1684 			}
1685 			break;
1686 		}
1687 	}
1688 }
1689 
1690 /*
1691  * Update device state in response to changes to the PCI command
1692  * register.
1693  */
1694 void
1695 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
1696 {
1697 	int i;
1698 	uint16_t changed, new;
1699 
1700 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);
1701 	changed = old ^ new;
1702 
1703 	/*
1704 	 * If the MMIO or I/O address space decoding has changed then
1705 	 * register/unregister all BARs that decode that address space.
1706 	 */
1707 	for (i = 0; i <= PCI_BARMAX; i++) {
1708 		switch (pi->pi_bar[i].type) {
1709 			case PCIBAR_NONE:
1710 			case PCIBAR_MEMHI64:
1711 				break;
1712 			case PCIBAR_IO:
1713 				/* I/O address space decoding changed? */
1714 				if (changed & PCIM_CMD_PORTEN) {
1715 					if (new & PCIM_CMD_PORTEN)
1716 						register_bar(pi, i);
1717 					else
1718 						unregister_bar(pi, i);
1719 				}
1720 				break;
1721 			case PCIBAR_MEM32:
1722 			case PCIBAR_MEM64:
1723 				/* MMIO address space decoding changed? */
1724 				if (changed & PCIM_CMD_MEMEN) {
1725 					if (new & PCIM_CMD_MEMEN)
1726 						register_bar(pi, i);
1727 					else
1728 						unregister_bar(pi, i);
1729 				}
1730 				break;
1731 			default:
1732 				assert(0);
1733 		}
1734 	}
1735 
1736 	/*
1737 	 * If INTx has been unmasked and is pending, assert the
1738 	 * interrupt.
1739 	 */
1740 	pci_lintr_update(pi);
1741 }
1742 
1743 static void
1744 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1745 {
1746 	int rshift;
1747 	uint32_t cmd, old, readonly;
1748 
1749 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1750 
1751 	/*
1752 	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1753 	 *
1754 	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1755 	 * 'write 1 to clear'. However these bits are not set to '1' by
1756 	 * any device emulation so it is simpler to treat them as readonly.
1757 	 */
1758 	rshift = (coff & 0x3) * 8;
1759 	readonly = 0xFFFFF880 >> rshift;
1760 
1761 	old = CFGREAD(pi, coff, bytes);
1762 	new &= ~readonly;
1763 	new |= (old & readonly);
1764 	CFGWRITE(pi, coff, new, bytes);			/* update config */
1765 
1766 	pci_emul_cmd_changed(pi, cmd);
1767 }
1768 
1769 static void
1770 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1771     int coff, int bytes, uint32_t *eax)
1772 {
1773 	struct businfo *bi;
1774 	struct slotinfo *si;
1775 	struct pci_devinst *pi;
1776 	struct pci_devemu *pe;
1777 	int idx, needcfg;
1778 	uint64_t addr, bar, mask;
1779 
1780 	if ((bi = pci_businfo[bus]) != NULL) {
1781 		si = &bi->slotinfo[slot];
1782 		pi = si->si_funcs[func].fi_devi;
1783 	} else
1784 		pi = NULL;
1785 
1786 	/*
1787 	 * Just return if there is no device at this slot:func or if the
1788 	 * the guest is doing an un-aligned access.
1789 	 */
1790 	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1791 	    (coff & (bytes - 1)) != 0) {
1792 		if (in)
1793 			*eax = 0xffffffff;
1794 		return;
1795 	}
1796 
1797 	/*
1798 	 * Ignore all writes beyond the standard config space and return all
1799 	 * ones on reads.
1800 	 */
1801 	if (coff >= PCI_REGMAX + 1) {
1802 		if (in) {
1803 			*eax = 0xffffffff;
1804 			/*
1805 			 * Extended capabilities begin at offset 256 in config
1806 			 * space. Absence of extended capabilities is signaled
1807 			 * with all 0s in the extended capability header at
1808 			 * offset 256.
1809 			 */
1810 			if (coff <= PCI_REGMAX + 4)
1811 				*eax = 0x00000000;
1812 		}
1813 		return;
1814 	}
1815 
1816 	pe = pi->pi_d;
1817 
1818 	/*
1819 	 * Config read
1820 	 */
1821 	if (in) {
1822 		/* Let the device emulation override the default handler */
1823 		if (pe->pe_cfgread != NULL) {
1824 			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1825 			    eax);
1826 		} else {
1827 			needcfg = 1;
1828 		}
1829 
1830 		if (needcfg)
1831 			*eax = CFGREAD(pi, coff, bytes);
1832 
1833 		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1834 	} else {
1835 		/* Let the device emulation override the default handler */
1836 		if (pe->pe_cfgwrite != NULL &&
1837 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1838 			return;
1839 
1840 		/*
1841 		 * Special handling for write to BAR registers
1842 		 */
1843 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1844 			/*
1845 			 * Ignore writes to BAR registers that are not
1846 			 * 4-byte aligned.
1847 			 */
1848 			if (bytes != 4 || (coff & 0x3) != 0)
1849 				return;
1850 			idx = (coff - PCIR_BAR(0)) / 4;
1851 			mask = ~(pi->pi_bar[idx].size - 1);
1852 			switch (pi->pi_bar[idx].type) {
1853 			case PCIBAR_NONE:
1854 				pi->pi_bar[idx].addr = bar = 0;
1855 				break;
1856 			case PCIBAR_IO:
1857 				addr = *eax & mask;
1858 				addr &= 0xffff;
1859 				bar = addr | PCIM_BAR_IO_SPACE;
1860 				/*
1861 				 * Register the new BAR value for interception
1862 				 */
1863 				if (addr != pi->pi_bar[idx].addr) {
1864 					update_bar_address(pi, addr, idx,
1865 							   PCIBAR_IO);
1866 				}
1867 				break;
1868 			case PCIBAR_MEM32:
1869 				addr = bar = *eax & mask;
1870 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1871 				if (addr != pi->pi_bar[idx].addr) {
1872 					update_bar_address(pi, addr, idx,
1873 							   PCIBAR_MEM32);
1874 				}
1875 				break;
1876 			case PCIBAR_MEM64:
1877 				addr = bar = *eax & mask;
1878 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1879 				       PCIM_BAR_MEM_PREFETCH;
1880 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1881 					update_bar_address(pi, addr, idx,
1882 							   PCIBAR_MEM64);
1883 				}
1884 				break;
1885 			case PCIBAR_MEMHI64:
1886 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1887 				addr = ((uint64_t)*eax << 32) & mask;
1888 				bar = addr >> 32;
1889 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1890 					update_bar_address(pi, addr, idx - 1,
1891 							   PCIBAR_MEMHI64);
1892 				}
1893 				break;
1894 			default:
1895 				assert(0);
1896 			}
1897 			pci_set_cfgdata32(pi, coff, bar);
1898 
1899 		} else if (pci_emul_iscap(pi, coff)) {
1900 			pci_emul_capwrite(pi, coff, bytes, *eax);
1901 		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1902 			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1903 		} else {
1904 			CFGWRITE(pi, coff, *eax, bytes);
1905 		}
1906 	}
1907 }
1908 
1909 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1910 
1911 static int
1912 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1913 		 uint32_t *eax, void *arg)
1914 {
1915 	uint32_t x;
1916 
1917 	if (bytes != 4) {
1918 		if (in)
1919 			*eax = (bytes == 2) ? 0xffff : 0xff;
1920 		return (0);
1921 	}
1922 
1923 	if (in) {
1924 		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1925 		if (cfgenable)
1926 			x |= CONF1_ENABLE;
1927 		*eax = x;
1928 	} else {
1929 		x = *eax;
1930 		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1931 		cfgoff = x & PCI_REGMAX;
1932 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1933 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1934 		cfgbus = (x >> 16) & PCI_BUSMAX;
1935 	}
1936 
1937 	return (0);
1938 }
1939 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1940 
1941 static int
1942 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1943 		 uint32_t *eax, void *arg)
1944 {
1945 	int coff;
1946 
1947 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1948 
1949 	coff = cfgoff + (port - CONF1_DATA_PORT);
1950 	if (cfgenable) {
1951 		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1952 		    eax);
1953 	} else {
1954 		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1955 		if (in)
1956 			*eax = 0xffffffff;
1957 	}
1958 	return (0);
1959 }
1960 
1961 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1962 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1963 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1964 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1965 
1966 #ifdef BHYVE_SNAPSHOT
1967 /*
1968  * Saves/restores PCI device emulated state. Returns 0 on success.
1969  */
1970 static int
1971 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta)
1972 {
1973 	struct pci_devinst *pi;
1974 	int i;
1975 	int ret;
1976 
1977 	pi = meta->dev_data;
1978 
1979 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done);
1980 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done);
1981 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done);
1982 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done);
1983 
1984 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done);
1985 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done);
1986 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done);
1987 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done);
1988 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done);
1989 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done);
1990 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done);
1991 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done);
1992 	SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_page_offset, meta, ret, done);
1993 
1994 	SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata),
1995 			      meta, ret, done);
1996 
1997 	for (i = 0; i < nitems(pi->pi_bar); i++) {
1998 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done);
1999 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done);
2000 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done);
2001 	}
2002 
2003 	/* Restore MSI-X table. */
2004 	for (i = 0; i < pi->pi_msix.table_count; i++) {
2005 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr,
2006 				      meta, ret, done);
2007 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data,
2008 				      meta, ret, done);
2009 		SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control,
2010 				      meta, ret, done);
2011 	}
2012 
2013 done:
2014 	return (ret);
2015 }
2016 
2017 static int
2018 pci_find_slotted_dev(const char *dev_name, struct pci_devemu **pde,
2019 		     struct pci_devinst **pdi)
2020 {
2021 	struct businfo *bi;
2022 	struct slotinfo *si;
2023 	struct funcinfo *fi;
2024 	int bus, slot, func;
2025 
2026 	assert(dev_name != NULL);
2027 	assert(pde != NULL);
2028 	assert(pdi != NULL);
2029 
2030 	for (bus = 0; bus < MAXBUSES; bus++) {
2031 		if ((bi = pci_businfo[bus]) == NULL)
2032 			continue;
2033 
2034 		for (slot = 0; slot < MAXSLOTS; slot++) {
2035 			si = &bi->slotinfo[slot];
2036 			for (func = 0; func < MAXFUNCS; func++) {
2037 				fi = &si->si_funcs[func];
2038 				if (fi->fi_name == NULL)
2039 					continue;
2040 				if (strcmp(dev_name, fi->fi_name))
2041 					continue;
2042 
2043 				*pde = pci_emul_finddev(fi->fi_name);
2044 				assert(*pde != NULL);
2045 
2046 				*pdi = fi->fi_devi;
2047 				return (0);
2048 			}
2049 		}
2050 	}
2051 
2052 	return (EINVAL);
2053 }
2054 
2055 int
2056 pci_snapshot(struct vm_snapshot_meta *meta)
2057 {
2058 	struct pci_devemu *pde;
2059 	struct pci_devinst *pdi;
2060 	int ret;
2061 
2062 	assert(meta->dev_name != NULL);
2063 
2064 	ret = pci_find_slotted_dev(meta->dev_name, &pde, &pdi);
2065 	if (ret != 0) {
2066 		fprintf(stderr, "%s: no such name: %s\r\n",
2067 			__func__, meta->dev_name);
2068 		memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
2069 		return (0);
2070 	}
2071 
2072 	meta->dev_data = pdi;
2073 
2074 	if (pde->pe_snapshot == NULL) {
2075 		fprintf(stderr, "%s: not implemented yet for: %s\r\n",
2076 			__func__, meta->dev_name);
2077 		return (-1);
2078 	}
2079 
2080 	ret = pci_snapshot_pci_dev(meta);
2081 	if (ret != 0) {
2082 		fprintf(stderr, "%s: failed to snapshot pci dev\r\n",
2083 			__func__);
2084 		return (-1);
2085 	}
2086 
2087 	ret = (*pde->pe_snapshot)(meta);
2088 
2089 	return (ret);
2090 }
2091 
2092 int
2093 pci_pause(struct vmctx *ctx, const char *dev_name)
2094 {
2095 	struct pci_devemu *pde;
2096 	struct pci_devinst *pdi;
2097 	int ret;
2098 
2099 	assert(dev_name != NULL);
2100 
2101 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2102 	if (ret != 0) {
2103 		/*
2104 		 * It is possible to call this function without
2105 		 * checking that the device is inserted first.
2106 		 */
2107 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2108 		return (0);
2109 	}
2110 
2111 	if (pde->pe_pause == NULL) {
2112 		/* The pause/resume functionality is optional. */
2113 		fprintf(stderr, "%s: not implemented for: %s\n",
2114 			__func__, dev_name);
2115 		return (0);
2116 	}
2117 
2118 	return (*pde->pe_pause)(ctx, pdi);
2119 }
2120 
2121 int
2122 pci_resume(struct vmctx *ctx, const char *dev_name)
2123 {
2124 	struct pci_devemu *pde;
2125 	struct pci_devinst *pdi;
2126 	int ret;
2127 
2128 	assert(dev_name != NULL);
2129 
2130 	ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2131 	if (ret != 0) {
2132 		/*
2133 		 * It is possible to call this function without
2134 		 * checking that the device is inserted first.
2135 		 */
2136 		fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2137 		return (0);
2138 	}
2139 
2140 	if (pde->pe_resume == NULL) {
2141 		/* The pause/resume functionality is optional. */
2142 		fprintf(stderr, "%s: not implemented for: %s\n",
2143 			__func__, dev_name);
2144 		return (0);
2145 	}
2146 
2147 	return (*pde->pe_resume)(ctx, pdi);
2148 }
2149 #endif
2150 
2151 #define PCI_EMUL_TEST
2152 #ifdef PCI_EMUL_TEST
2153 /*
2154  * Define a dummy test device
2155  */
2156 #define DIOSZ	8
2157 #define DMEMSZ	4096
2158 struct pci_emul_dsoftc {
2159 	uint8_t   ioregs[DIOSZ];
2160 	uint8_t	  memregs[2][DMEMSZ];
2161 };
2162 
2163 #define	PCI_EMUL_MSI_MSGS	 4
2164 #define	PCI_EMUL_MSIX_MSGS	16
2165 
2166 static int
2167 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2168 {
2169 	int error;
2170 	struct pci_emul_dsoftc *sc;
2171 
2172 	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
2173 
2174 	pi->pi_arg = sc;
2175 
2176 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
2177 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
2178 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
2179 
2180 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
2181 	assert(error == 0);
2182 
2183 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
2184 	assert(error == 0);
2185 
2186 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2187 	assert(error == 0);
2188 
2189 	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2190 	assert(error == 0);
2191 
2192 	return (0);
2193 }
2194 
2195 static void
2196 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2197 	      uint64_t offset, int size, uint64_t value)
2198 {
2199 	int i;
2200 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2201 
2202 	if (baridx == 0) {
2203 		if (offset + size > DIOSZ) {
2204 			printf("diow: iow too large, offset %ld size %d\n",
2205 			       offset, size);
2206 			return;
2207 		}
2208 
2209 		if (size == 1) {
2210 			sc->ioregs[offset] = value & 0xff;
2211 		} else if (size == 2) {
2212 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2213 		} else if (size == 4) {
2214 			*(uint32_t *)&sc->ioregs[offset] = value;
2215 		} else {
2216 			printf("diow: iow unknown size %d\n", size);
2217 		}
2218 
2219 		/*
2220 		 * Special magic value to generate an interrupt
2221 		 */
2222 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2223 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2224 
2225 		if (value == 0xabcdef) {
2226 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2227 				pci_generate_msi(pi, i);
2228 		}
2229 	}
2230 
2231 	if (baridx == 1 || baridx == 2) {
2232 		if (offset + size > DMEMSZ) {
2233 			printf("diow: memw too large, offset %ld size %d\n",
2234 			       offset, size);
2235 			return;
2236 		}
2237 
2238 		i = baridx - 1;		/* 'memregs' index */
2239 
2240 		if (size == 1) {
2241 			sc->memregs[i][offset] = value;
2242 		} else if (size == 2) {
2243 			*(uint16_t *)&sc->memregs[i][offset] = value;
2244 		} else if (size == 4) {
2245 			*(uint32_t *)&sc->memregs[i][offset] = value;
2246 		} else if (size == 8) {
2247 			*(uint64_t *)&sc->memregs[i][offset] = value;
2248 		} else {
2249 			printf("diow: memw unknown size %d\n", size);
2250 		}
2251 
2252 		/*
2253 		 * magic interrupt ??
2254 		 */
2255 	}
2256 
2257 	if (baridx > 2 || baridx < 0) {
2258 		printf("diow: unknown bar idx %d\n", baridx);
2259 	}
2260 }
2261 
2262 static uint64_t
2263 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2264 	      uint64_t offset, int size)
2265 {
2266 	struct pci_emul_dsoftc *sc = pi->pi_arg;
2267 	uint32_t value;
2268 	int i;
2269 
2270 	if (baridx == 0) {
2271 		if (offset + size > DIOSZ) {
2272 			printf("dior: ior too large, offset %ld size %d\n",
2273 			       offset, size);
2274 			return (0);
2275 		}
2276 
2277 		value = 0;
2278 		if (size == 1) {
2279 			value = sc->ioregs[offset];
2280 		} else if (size == 2) {
2281 			value = *(uint16_t *) &sc->ioregs[offset];
2282 		} else if (size == 4) {
2283 			value = *(uint32_t *) &sc->ioregs[offset];
2284 		} else {
2285 			printf("dior: ior unknown size %d\n", size);
2286 		}
2287 	}
2288 
2289 	if (baridx == 1 || baridx == 2) {
2290 		if (offset + size > DMEMSZ) {
2291 			printf("dior: memr too large, offset %ld size %d\n",
2292 			       offset, size);
2293 			return (0);
2294 		}
2295 
2296 		i = baridx - 1;		/* 'memregs' index */
2297 
2298 		if (size == 1) {
2299 			value = sc->memregs[i][offset];
2300 		} else if (size == 2) {
2301 			value = *(uint16_t *) &sc->memregs[i][offset];
2302 		} else if (size == 4) {
2303 			value = *(uint32_t *) &sc->memregs[i][offset];
2304 		} else if (size == 8) {
2305 			value = *(uint64_t *) &sc->memregs[i][offset];
2306 		} else {
2307 			printf("dior: ior unknown size %d\n", size);
2308 		}
2309 	}
2310 
2311 
2312 	if (baridx > 2 || baridx < 0) {
2313 		printf("dior: unknown bar idx %d\n", baridx);
2314 		return (0);
2315 	}
2316 
2317 	return (value);
2318 }
2319 
2320 #ifdef BHYVE_SNAPSHOT
2321 int
2322 pci_emul_snapshot(struct vm_snapshot_meta *meta)
2323 {
2324 
2325 	return (0);
2326 }
2327 #endif
2328 
2329 struct pci_devemu pci_dummy = {
2330 	.pe_emu = "dummy",
2331 	.pe_init = pci_emul_dinit,
2332 	.pe_barwrite = pci_emul_diow,
2333 	.pe_barread = pci_emul_dior,
2334 #ifdef BHYVE_SNAPSHOT
2335 	.pe_snapshot = pci_emul_snapshot,
2336 #endif
2337 };
2338 PCI_EMUL_SET(pci_dummy);
2339 
2340 #endif /* PCI_EMUL_TEST */
2341