xref: /freebsd/usr.sbin/bhyve/pci_emul.c (revision 1a61beb0549e05b33df31380e427d90f6e46ff7e)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/errno.h>
35 
36 #include <ctype.h>
37 #include <pthread.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <strings.h>
42 #include <assert.h>
43 #include <stdbool.h>
44 
45 #include <machine/vmm.h>
46 #include <vmmapi.h>
47 
48 #include "acpi.h"
49 #include "bhyverun.h"
50 #include "inout.h"
51 #include "ioapic.h"
52 #include "mem.h"
53 #include "pci_emul.h"
54 #include "pci_lpc.h"
55 
56 #define CONF1_ADDR_PORT    0x0cf8
57 #define CONF1_DATA_PORT    0x0cfc
58 
59 #define CONF1_ENABLE	   0x80000000ul
60 
61 #define	CFGWRITE(pi,off,val,b)						\
62 do {									\
63 	if ((b) == 1) {							\
64 		pci_set_cfgdata8((pi),(off),(val));			\
65 	} else if ((b) == 2) {						\
66 		pci_set_cfgdata16((pi),(off),(val));			\
67 	} else {							\
68 		pci_set_cfgdata32((pi),(off),(val));			\
69 	}								\
70 } while (0)
71 
72 #define MAXSLOTS	(PCI_SLOTMAX + 1)
73 #define	MAXFUNCS	(PCI_FUNCMAX + 1)
74 
75 struct funcinfo {
76 	char	*fi_name;
77 	char	*fi_param;
78 	struct pci_devinst *fi_devi;
79 };
80 
81 struct intxinfo {
82 	int	ii_count;
83 	int	ii_ioapic_irq;
84 };
85 
86 struct slotinfo {
87 	struct intxinfo si_intpins[4];
88 	struct funcinfo si_funcs[MAXFUNCS];
89 } pci_slotinfo[MAXSLOTS];
90 
91 SET_DECLARE(pci_devemu_set, struct pci_devemu);
92 
93 static uint64_t pci_emul_iobase;
94 static uint64_t pci_emul_membase32;
95 static uint64_t pci_emul_membase64;
96 
97 #define	PCI_EMUL_IOBASE		0x2000
98 #define	PCI_EMUL_IOLIMIT	0x10000
99 
100 #define	PCI_EMUL_MEMLIMIT32	0xE0000000		/* 3.5GB */
101 
102 #define	PCI_EMUL_MEMBASE64	0xD000000000UL
103 #define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
104 
105 static struct pci_devemu *pci_emul_finddev(char *name);
106 static void	pci_lintr_update(struct pci_devinst *pi);
107 
108 static int pci_emul_devices;
109 static struct mem_range pci_mem_hole;
110 
111 /*
112  * I/O access
113  */
114 
115 /*
116  * Slot options are in the form:
117  *
118  *  <slot>[:<func>],<emul>[,<config>]
119  *
120  *  slot is 0..31
121  *  func is 0..7
122  *  emul is a string describing the type of PCI device e.g. virtio-net
123  *  config is an optional string, depending on the device, that can be
124  *  used for configuration.
125  *   Examples are:
126  *     1,virtio-net,tap0
127  *     3:0,dummy
128  */
129 static void
130 pci_parse_slot_usage(char *aopt)
131 {
132 
133 	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
134 }
135 
136 int
137 pci_parse_slot(char *opt)
138 {
139 	char *slot, *func, *emul, *config;
140 	char *str, *cpy;
141 	int error, snum, fnum;
142 
143 	error = -1;
144 	str = cpy = strdup(opt);
145 
146         slot = strsep(&str, ",");
147         func = NULL;
148         if (strchr(slot, ':') != NULL) {
149 		func = cpy;
150 		(void) strsep(&func, ":");
151         }
152 
153 	emul = strsep(&str, ",");
154 	config = str;
155 
156 	if (emul == NULL) {
157 		pci_parse_slot_usage(opt);
158 		goto done;
159 	}
160 
161 	snum = atoi(slot);
162 	fnum = func ? atoi(func) : 0;
163 
164 	if (snum < 0 || snum >= MAXSLOTS || fnum < 0 || fnum >= MAXFUNCS) {
165 		pci_parse_slot_usage(opt);
166 		goto done;
167 	}
168 
169 	if (pci_slotinfo[snum].si_funcs[fnum].fi_name != NULL) {
170 		fprintf(stderr, "pci slot %d:%d already occupied!\n",
171 			snum, fnum);
172 		goto done;
173 	}
174 
175 	if (pci_emul_finddev(emul) == NULL) {
176 		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
177 			snum, fnum, emul);
178 		goto done;
179 	}
180 
181 	error = 0;
182 	pci_slotinfo[snum].si_funcs[fnum].fi_name = emul;
183 	pci_slotinfo[snum].si_funcs[fnum].fi_param = config;
184 
185 done:
186 	if (error)
187 		free(cpy);
188 
189 	return (error);
190 }
191 
192 static int
193 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
194 {
195 
196 	if (offset < pi->pi_msix.pba_offset)
197 		return (0);
198 
199 	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
200 		return (0);
201 	}
202 
203 	return (1);
204 }
205 
206 int
207 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
208 		     uint64_t value)
209 {
210 	int msix_entry_offset;
211 	int tab_index;
212 	char *dest;
213 
214 	/* support only 4 or 8 byte writes */
215 	if (size != 4 && size != 8)
216 		return (-1);
217 
218 	/*
219 	 * Return if table index is beyond what device supports
220 	 */
221 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
222 	if (tab_index >= pi->pi_msix.table_count)
223 		return (-1);
224 
225 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
226 
227 	/* support only aligned writes */
228 	if ((msix_entry_offset % size) != 0)
229 		return (-1);
230 
231 	dest = (char *)(pi->pi_msix.table + tab_index);
232 	dest += msix_entry_offset;
233 
234 	if (size == 4)
235 		*((uint32_t *)dest) = value;
236 	else
237 		*((uint64_t *)dest) = value;
238 
239 	return (0);
240 }
241 
242 uint64_t
243 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
244 {
245 	char *dest;
246 	int msix_entry_offset;
247 	int tab_index;
248 	uint64_t retval = ~0;
249 
250 	/*
251 	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
252 	 * table but we also allow 1 byte access to accomodate reads from
253 	 * ddb.
254 	 */
255 	if (size != 1 && size != 4 && size != 8)
256 		return (retval);
257 
258 	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
259 
260 	/* support only aligned reads */
261 	if ((msix_entry_offset % size) != 0) {
262 		return (retval);
263 	}
264 
265 	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
266 
267 	if (tab_index < pi->pi_msix.table_count) {
268 		/* valid MSI-X Table access */
269 		dest = (char *)(pi->pi_msix.table + tab_index);
270 		dest += msix_entry_offset;
271 
272 		if (size == 1)
273 			retval = *((uint8_t *)dest);
274 		else if (size == 4)
275 			retval = *((uint32_t *)dest);
276 		else
277 			retval = *((uint64_t *)dest);
278 	} else if (pci_valid_pba_offset(pi, offset)) {
279 		/* return 0 for PBA access */
280 		retval = 0;
281 	}
282 
283 	return (retval);
284 }
285 
286 int
287 pci_msix_table_bar(struct pci_devinst *pi)
288 {
289 
290 	if (pi->pi_msix.table != NULL)
291 		return (pi->pi_msix.table_bar);
292 	else
293 		return (-1);
294 }
295 
296 int
297 pci_msix_pba_bar(struct pci_devinst *pi)
298 {
299 
300 	if (pi->pi_msix.table != NULL)
301 		return (pi->pi_msix.pba_bar);
302 	else
303 		return (-1);
304 }
305 
306 static int
307 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
308 		    uint32_t *eax, void *arg)
309 {
310 	struct pci_devinst *pdi = arg;
311 	struct pci_devemu *pe = pdi->pi_d;
312 	uint64_t offset;
313 	int i;
314 
315 	for (i = 0; i <= PCI_BARMAX; i++) {
316 		if (pdi->pi_bar[i].type == PCIBAR_IO &&
317 		    port >= pdi->pi_bar[i].addr &&
318 		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
319 			offset = port - pdi->pi_bar[i].addr;
320 			if (in)
321 				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
322 							 offset, bytes);
323 			else
324 				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
325 						   bytes, *eax);
326 			return (0);
327 		}
328 	}
329 	return (-1);
330 }
331 
332 static int
333 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
334 		     int size, uint64_t *val, void *arg1, long arg2)
335 {
336 	struct pci_devinst *pdi = arg1;
337 	struct pci_devemu *pe = pdi->pi_d;
338 	uint64_t offset;
339 	int bidx = (int) arg2;
340 
341 	assert(bidx <= PCI_BARMAX);
342 	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
343 	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
344 	assert(addr >= pdi->pi_bar[bidx].addr &&
345 	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
346 
347 	offset = addr - pdi->pi_bar[bidx].addr;
348 
349 	if (dir == MEM_F_WRITE)
350 		(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, size, *val);
351 	else
352 		*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, offset, size);
353 
354 	return (0);
355 }
356 
357 
358 static int
359 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
360 			uint64_t *addr)
361 {
362 	uint64_t base;
363 
364 	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
365 
366 	base = roundup2(*baseptr, size);
367 
368 	if (base + size <= limit) {
369 		*addr = base;
370 		*baseptr = base + size;
371 		return (0);
372 	} else
373 		return (-1);
374 }
375 
376 int
377 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
378 		   uint64_t size)
379 {
380 
381 	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
382 }
383 
384 /*
385  * Register (or unregister) the MMIO or I/O region associated with the BAR
386  * register 'idx' of an emulated pci device.
387  */
388 static void
389 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
390 {
391 	int error;
392 	struct inout_port iop;
393 	struct mem_range mr;
394 
395 	switch (pi->pi_bar[idx].type) {
396 	case PCIBAR_IO:
397 		bzero(&iop, sizeof(struct inout_port));
398 		iop.name = pi->pi_name;
399 		iop.port = pi->pi_bar[idx].addr;
400 		iop.size = pi->pi_bar[idx].size;
401 		if (registration) {
402 			iop.flags = IOPORT_F_INOUT;
403 			iop.handler = pci_emul_io_handler;
404 			iop.arg = pi;
405 			error = register_inout(&iop);
406 		} else
407 			error = unregister_inout(&iop);
408 		break;
409 	case PCIBAR_MEM32:
410 	case PCIBAR_MEM64:
411 		bzero(&mr, sizeof(struct mem_range));
412 		mr.name = pi->pi_name;
413 		mr.base = pi->pi_bar[idx].addr;
414 		mr.size = pi->pi_bar[idx].size;
415 		if (registration) {
416 			mr.flags = MEM_F_RW;
417 			mr.handler = pci_emul_mem_handler;
418 			mr.arg1 = pi;
419 			mr.arg2 = idx;
420 			error = register_mem(&mr);
421 		} else
422 			error = unregister_mem(&mr);
423 		break;
424 	default:
425 		error = EINVAL;
426 		break;
427 	}
428 	assert(error == 0);
429 }
430 
431 static void
432 unregister_bar(struct pci_devinst *pi, int idx)
433 {
434 
435 	modify_bar_registration(pi, idx, 0);
436 }
437 
438 static void
439 register_bar(struct pci_devinst *pi, int idx)
440 {
441 
442 	modify_bar_registration(pi, idx, 1);
443 }
444 
445 /* Are we decoding i/o port accesses for the emulated pci device? */
446 static int
447 porten(struct pci_devinst *pi)
448 {
449 	uint16_t cmd;
450 
451 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
452 
453 	return (cmd & PCIM_CMD_PORTEN);
454 }
455 
456 /* Are we decoding memory accesses for the emulated pci device? */
457 static int
458 memen(struct pci_devinst *pi)
459 {
460 	uint16_t cmd;
461 
462 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
463 
464 	return (cmd & PCIM_CMD_MEMEN);
465 }
466 
467 /*
468  * Update the MMIO or I/O address that is decoded by the BAR register.
469  *
470  * If the pci device has enabled the address space decoding then intercept
471  * the address range decoded by the BAR register.
472  */
473 static void
474 update_bar_address(struct  pci_devinst *pi, uint64_t addr, int idx, int type)
475 {
476 	int decode;
477 
478 	if (pi->pi_bar[idx].type == PCIBAR_IO)
479 		decode = porten(pi);
480 	else
481 		decode = memen(pi);
482 
483 	if (decode)
484 		unregister_bar(pi, idx);
485 
486 	switch (type) {
487 	case PCIBAR_IO:
488 	case PCIBAR_MEM32:
489 		pi->pi_bar[idx].addr = addr;
490 		break;
491 	case PCIBAR_MEM64:
492 		pi->pi_bar[idx].addr &= ~0xffffffffUL;
493 		pi->pi_bar[idx].addr |= addr;
494 		break;
495 	case PCIBAR_MEMHI64:
496 		pi->pi_bar[idx].addr &= 0xffffffff;
497 		pi->pi_bar[idx].addr |= addr;
498 		break;
499 	default:
500 		assert(0);
501 	}
502 
503 	if (decode)
504 		register_bar(pi, idx);
505 }
506 
507 int
508 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
509 		    enum pcibar_type type, uint64_t size)
510 {
511 	int error;
512 	uint64_t *baseptr, limit, addr, mask, lobits, bar;
513 
514 	assert(idx >= 0 && idx <= PCI_BARMAX);
515 
516 	if ((size & (size - 1)) != 0)
517 		size = 1UL << flsl(size);	/* round up to a power of 2 */
518 
519 	/* Enforce minimum BAR sizes required by the PCI standard */
520 	if (type == PCIBAR_IO) {
521 		if (size < 4)
522 			size = 4;
523 	} else {
524 		if (size < 16)
525 			size = 16;
526 	}
527 
528 	switch (type) {
529 	case PCIBAR_NONE:
530 		baseptr = NULL;
531 		addr = mask = lobits = 0;
532 		break;
533 	case PCIBAR_IO:
534 		baseptr = &pci_emul_iobase;
535 		limit = PCI_EMUL_IOLIMIT;
536 		mask = PCIM_BAR_IO_BASE;
537 		lobits = PCIM_BAR_IO_SPACE;
538 		break;
539 	case PCIBAR_MEM64:
540 		/*
541 		 * XXX
542 		 * Some drivers do not work well if the 64-bit BAR is allocated
543 		 * above 4GB. Allow for this by allocating small requests under
544 		 * 4GB unless then allocation size is larger than some arbitrary
545 		 * number (32MB currently).
546 		 */
547 		if (size > 32 * 1024 * 1024) {
548 			/*
549 			 * XXX special case for device requiring peer-peer DMA
550 			 */
551 			if (size == 0x100000000UL)
552 				baseptr = &hostbase;
553 			else
554 				baseptr = &pci_emul_membase64;
555 			limit = PCI_EMUL_MEMLIMIT64;
556 			mask = PCIM_BAR_MEM_BASE;
557 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
558 				 PCIM_BAR_MEM_PREFETCH;
559 			break;
560 		} else {
561 			baseptr = &pci_emul_membase32;
562 			limit = PCI_EMUL_MEMLIMIT32;
563 			mask = PCIM_BAR_MEM_BASE;
564 			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
565 		}
566 		break;
567 	case PCIBAR_MEM32:
568 		baseptr = &pci_emul_membase32;
569 		limit = PCI_EMUL_MEMLIMIT32;
570 		mask = PCIM_BAR_MEM_BASE;
571 		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
572 		break;
573 	default:
574 		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
575 		assert(0);
576 	}
577 
578 	if (baseptr != NULL) {
579 		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
580 		if (error != 0)
581 			return (error);
582 	}
583 
584 	pdi->pi_bar[idx].type = type;
585 	pdi->pi_bar[idx].addr = addr;
586 	pdi->pi_bar[idx].size = size;
587 
588 	/* Initialize the BAR register in config space */
589 	bar = (addr & mask) | lobits;
590 	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
591 
592 	if (type == PCIBAR_MEM64) {
593 		assert(idx + 1 <= PCI_BARMAX);
594 		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
595 		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
596 	}
597 
598 	register_bar(pdi, idx);
599 
600 	return (0);
601 }
602 
603 #define	CAP_START_OFFSET	0x40
604 static int
605 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
606 {
607 	int i, capoff, capid, reallen;
608 	uint16_t sts;
609 
610 	static u_char endofcap[4] = {
611 		PCIY_RESERVED, 0, 0, 0
612 	};
613 
614 	assert(caplen > 0 && capdata[0] != PCIY_RESERVED);
615 
616 	reallen = roundup2(caplen, 4);		/* dword aligned */
617 
618 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
619 	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
620 		capoff = CAP_START_OFFSET;
621 		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
622 		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
623 	} else {
624 		capoff = pci_get_cfgdata8(pi, PCIR_CAP_PTR);
625 		while (1) {
626 			assert((capoff & 0x3) == 0);
627 			capid = pci_get_cfgdata8(pi, capoff);
628 			if (capid == PCIY_RESERVED)
629 				break;
630 			capoff = pci_get_cfgdata8(pi, capoff + 1);
631 		}
632 	}
633 
634 	/* Check if we have enough space */
635 	if (capoff + reallen + sizeof(endofcap) > PCI_REGMAX + 1)
636 		return (-1);
637 
638 	/* Copy the capability */
639 	for (i = 0; i < caplen; i++)
640 		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
641 
642 	/* Set the next capability pointer */
643 	pci_set_cfgdata8(pi, capoff + 1, capoff + reallen);
644 
645 	/* Copy of the reserved capability which serves as the end marker */
646 	for (i = 0; i < sizeof(endofcap); i++)
647 		pci_set_cfgdata8(pi, capoff + reallen + i, endofcap[i]);
648 
649 	return (0);
650 }
651 
652 static struct pci_devemu *
653 pci_emul_finddev(char *name)
654 {
655 	struct pci_devemu **pdpp, *pdp;
656 
657 	SET_FOREACH(pdpp, pci_devemu_set) {
658 		pdp = *pdpp;
659 		if (!strcmp(pdp->pe_emu, name)) {
660 			return (pdp);
661 		}
662 	}
663 
664 	return (NULL);
665 }
666 
667 static int
668 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int slot, int func,
669 	      char *params)
670 {
671 	struct pci_devinst *pdi;
672 	int err;
673 
674 	pdi = malloc(sizeof(struct pci_devinst));
675 	bzero(pdi, sizeof(*pdi));
676 
677 	pdi->pi_vmctx = ctx;
678 	pdi->pi_bus = 0;
679 	pdi->pi_slot = slot;
680 	pdi->pi_func = func;
681 	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
682 	pdi->pi_lintr.pin = 0;
683 	pdi->pi_lintr.state = IDLE;
684 	pdi->pi_lintr.ioapic_irq = 0;
685 	pdi->pi_d = pde;
686 	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
687 
688 	/* Disable legacy interrupts */
689 	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
690 	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
691 
692 	pci_set_cfgdata8(pdi, PCIR_COMMAND,
693 		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
694 
695 	err = (*pde->pe_init)(ctx, pdi, params);
696 	if (err != 0) {
697 		free(pdi);
698 	} else {
699 		pci_emul_devices++;
700 		pci_slotinfo[slot].si_funcs[func].fi_devi = pdi;
701 	}
702 
703 	return (err);
704 }
705 
706 void
707 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
708 {
709 	int mmc;
710 
711 	CTASSERT(sizeof(struct msicap) == 14);
712 
713 	/* Number of msi messages must be a power of 2 between 1 and 32 */
714 	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
715 	mmc = ffs(msgnum) - 1;
716 
717 	bzero(msicap, sizeof(struct msicap));
718 	msicap->capid = PCIY_MSI;
719 	msicap->nextptr = nextptr;
720 	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
721 }
722 
723 int
724 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
725 {
726 	struct msicap msicap;
727 
728 	pci_populate_msicap(&msicap, msgnum, 0);
729 
730 	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
731 }
732 
733 static void
734 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
735 		     uint32_t msix_tab_size, int nextptr)
736 {
737 	CTASSERT(sizeof(struct msixcap) == 12);
738 
739 	assert(msix_tab_size % 4096 == 0);
740 
741 	bzero(msixcap, sizeof(struct msixcap));
742 	msixcap->capid = PCIY_MSIX;
743 	msixcap->nextptr = nextptr;
744 
745 	/*
746 	 * Message Control Register, all fields set to
747 	 * zero except for the Table Size.
748 	 * Note: Table size N is encoded as N-1
749 	 */
750 	msixcap->msgctrl = msgnum - 1;
751 
752 	/*
753 	 * MSI-X BAR setup:
754 	 * - MSI-X table start at offset 0
755 	 * - PBA table starts at a 4K aligned offset after the MSI-X table
756 	 */
757 	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
758 	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
759 }
760 
761 static void
762 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
763 {
764 	int i, table_size;
765 
766 	assert(table_entries > 0);
767 	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
768 
769 	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
770 	pi->pi_msix.table = malloc(table_size);
771 	bzero(pi->pi_msix.table, table_size);
772 
773 	/* set mask bit of vector control register */
774 	for (i = 0; i < table_entries; i++)
775 		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
776 }
777 
778 int
779 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
780 {
781 	uint16_t pba_index;
782 	uint32_t tab_size;
783 	struct msixcap msixcap;
784 
785 	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
786 	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
787 
788 	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
789 
790 	/* Align table size to nearest 4K */
791 	tab_size = roundup2(tab_size, 4096);
792 
793 	pi->pi_msix.table_bar = barnum;
794 	pi->pi_msix.pba_bar   = barnum;
795 	pi->pi_msix.table_offset = 0;
796 	pi->pi_msix.table_count = msgnum;
797 	pi->pi_msix.pba_offset = tab_size;
798 
799 	/* calculate the MMIO size required for MSI-X PBA */
800 	pba_index = (msgnum - 1) / (PBA_TABLE_ENTRY_SIZE * 8);
801 	pi->pi_msix.pba_size = (pba_index + 1) * PBA_TABLE_ENTRY_SIZE;
802 
803 	pci_msix_table_init(pi, msgnum);
804 
805 	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size, 0);
806 
807 	/* allocate memory for MSI-X Table and PBA */
808 	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
809 				tab_size + pi->pi_msix.pba_size);
810 
811 	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
812 					sizeof(msixcap)));
813 }
814 
815 void
816 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
817 		 int bytes, uint32_t val)
818 {
819 	uint16_t msgctrl, rwmask;
820 	int off, table_bar;
821 
822 	off = offset - capoff;
823 	table_bar = pi->pi_msix.table_bar;
824 	/* Message Control Register */
825 	if (off == 2 && bytes == 2) {
826 		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
827 		msgctrl = pci_get_cfgdata16(pi, offset);
828 		msgctrl &= ~rwmask;
829 		msgctrl |= val & rwmask;
830 		val = msgctrl;
831 
832 		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
833 		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
834 		pci_lintr_update(pi);
835 	}
836 
837 	CFGWRITE(pi, offset, val, bytes);
838 }
839 
840 void
841 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
842 		int bytes, uint32_t val)
843 {
844 	uint16_t msgctrl, rwmask, msgdata, mme;
845 	uint32_t addrlo;
846 
847 	/*
848 	 * If guest is writing to the message control register make sure
849 	 * we do not overwrite read-only fields.
850 	 */
851 	if ((offset - capoff) == 2 && bytes == 2) {
852 		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
853 		msgctrl = pci_get_cfgdata16(pi, offset);
854 		msgctrl &= ~rwmask;
855 		msgctrl |= val & rwmask;
856 		val = msgctrl;
857 
858 		addrlo = pci_get_cfgdata32(pi, capoff + 4);
859 		if (msgctrl & PCIM_MSICTRL_64BIT)
860 			msgdata = pci_get_cfgdata16(pi, capoff + 12);
861 		else
862 			msgdata = pci_get_cfgdata16(pi, capoff + 8);
863 
864 		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
865 		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
866 		if (pi->pi_msi.enabled) {
867 			pi->pi_msi.addr = addrlo;
868 			pi->pi_msi.msg_data = msgdata;
869 			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
870 		} else {
871 			pi->pi_msi.maxmsgnum = 0;
872 		}
873 		pci_lintr_update(pi);
874 	}
875 
876 	CFGWRITE(pi, offset, val, bytes);
877 }
878 
879 void
880 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
881 		 int bytes, uint32_t val)
882 {
883 
884 	/* XXX don't write to the readonly parts */
885 	CFGWRITE(pi, offset, val, bytes);
886 }
887 
888 #define	PCIECAP_VERSION	0x2
889 int
890 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
891 {
892 	int err;
893 	struct pciecap pciecap;
894 
895 	CTASSERT(sizeof(struct pciecap) == 60);
896 
897 	if (type != PCIEM_TYPE_ROOT_PORT)
898 		return (-1);
899 
900 	bzero(&pciecap, sizeof(pciecap));
901 
902 	pciecap.capid = PCIY_EXPRESS;
903 	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
904 	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
905 	pciecap.link_status = 0x11;		/* gen1, x1 */
906 
907 	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
908 	return (err);
909 }
910 
911 /*
912  * This function assumes that 'coff' is in the capabilities region of the
913  * config space.
914  */
915 static void
916 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
917 {
918 	int capid;
919 	uint8_t capoff, nextoff;
920 
921 	/* Do not allow un-aligned writes */
922 	if ((offset & (bytes - 1)) != 0)
923 		return;
924 
925 	/* Find the capability that we want to update */
926 	capoff = CAP_START_OFFSET;
927 	while (1) {
928 		capid = pci_get_cfgdata8(pi, capoff);
929 		if (capid == PCIY_RESERVED)
930 			break;
931 
932 		nextoff = pci_get_cfgdata8(pi, capoff + 1);
933 		if (offset >= capoff && offset < nextoff)
934 			break;
935 
936 		capoff = nextoff;
937 	}
938 	assert(offset >= capoff);
939 
940 	/*
941 	 * Capability ID and Next Capability Pointer are readonly.
942 	 * However, some o/s's do 4-byte writes that include these.
943 	 * For this case, trim the write back to 2 bytes and adjust
944 	 * the data.
945 	 */
946 	if (offset == capoff || offset == capoff + 1) {
947 		if (offset == capoff && bytes == 4) {
948 			bytes = 2;
949 			offset += 2;
950 			val >>= 16;
951 		} else
952 			return;
953 	}
954 
955 	switch (capid) {
956 	case PCIY_MSI:
957 		msicap_cfgwrite(pi, capoff, offset, bytes, val);
958 		break;
959 	case PCIY_MSIX:
960 		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
961 		break;
962 	case PCIY_EXPRESS:
963 		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
964 		break;
965 	default:
966 		break;
967 	}
968 }
969 
970 static int
971 pci_emul_iscap(struct pci_devinst *pi, int offset)
972 {
973 	int found;
974 	uint16_t sts;
975 	uint8_t capid, lastoff;
976 
977 	found = 0;
978 	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
979 	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
980 		lastoff = pci_get_cfgdata8(pi, PCIR_CAP_PTR);
981 		while (1) {
982 			assert((lastoff & 0x3) == 0);
983 			capid = pci_get_cfgdata8(pi, lastoff);
984 			if (capid == PCIY_RESERVED)
985 				break;
986 			lastoff = pci_get_cfgdata8(pi, lastoff + 1);
987 		}
988 		if (offset >= CAP_START_OFFSET && offset <= lastoff)
989 			found = 1;
990 	}
991 	return (found);
992 }
993 
994 static int
995 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
996 			  int size, uint64_t *val, void *arg1, long arg2)
997 {
998 	/*
999 	 * Ignore writes; return 0xff's for reads. The mem read code
1000 	 * will take care of truncating to the correct size.
1001 	 */
1002 	if (dir == MEM_F_READ) {
1003 		*val = 0xffffffffffffffff;
1004 	}
1005 
1006 	return (0);
1007 }
1008 
1009 int
1010 init_pci(struct vmctx *ctx)
1011 {
1012 	struct pci_devemu *pde;
1013 	struct funcinfo *fi;
1014 	size_t lowmem;
1015 	int slot, func;
1016 	int error;
1017 
1018 	pci_emul_iobase = PCI_EMUL_IOBASE;
1019 	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1020 	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1021 
1022 	for (slot = 0; slot < MAXSLOTS; slot++) {
1023 		for (func = 0; func < MAXFUNCS; func++) {
1024 			fi = &pci_slotinfo[slot].si_funcs[func];
1025 			if (fi->fi_name != NULL) {
1026 				pde = pci_emul_finddev(fi->fi_name);
1027 				assert(pde != NULL);
1028 				error = pci_emul_init(ctx, pde, slot, func,
1029 					    fi->fi_param);
1030 				if (error)
1031 					return (error);
1032 			}
1033 		}
1034 	}
1035 
1036 	/*
1037 	 * The guest physical memory map looks like the following:
1038 	 * [0,		    lowmem)		guest system memory
1039 	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1040 	 * [lowmem_limit,   4GB)		PCI hole (32-bit BAR allocation)
1041 	 * [4GB,	    4GB + highmem)
1042 	 *
1043 	 * Accesses to memory addresses that are not allocated to system
1044 	 * memory or PCI devices return 0xff's.
1045 	 */
1046 	error = vm_get_memory_seg(ctx, 0, &lowmem, NULL);
1047 	assert(error == 0);
1048 
1049 	memset(&pci_mem_hole, 0, sizeof(struct mem_range));
1050 	pci_mem_hole.name = "PCI hole";
1051 	pci_mem_hole.flags = MEM_F_RW;
1052 	pci_mem_hole.base = lowmem;
1053 	pci_mem_hole.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1054 	pci_mem_hole.handler = pci_emul_fallback_handler;
1055 
1056 	error = register_mem_fallback(&pci_mem_hole);
1057 	assert(error == 0);
1058 
1059 	return (0);
1060 }
1061 
1062 static void
1063 pci_prt_entry(int slot, int pin, int ioapic_irq, void *arg)
1064 {
1065 	int *count;
1066 
1067 	count = arg;
1068 	dsdt_line("  Package (0x04)");
1069 	dsdt_line("  {");
1070 	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1071 	dsdt_line("    0x%02X,", pin - 1);
1072 	dsdt_line("    Zero,");
1073 	dsdt_line("    0x%X", ioapic_irq);
1074 	dsdt_line("  }%s", *count == 1 ? "" : ",");
1075 	(*count)--;
1076 }
1077 
1078 void
1079 pci_write_dsdt(void)
1080 {
1081 	struct pci_devinst *pi;
1082 	int count, slot, func;
1083 
1084 	dsdt_indent(1);
1085 	dsdt_line("Scope (_SB)");
1086 	dsdt_line("{");
1087 	dsdt_line("  Device (PCI0)");
1088 	dsdt_line("  {");
1089 	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1090 	dsdt_line("    Name (_ADR, Zero)");
1091 	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1092 	dsdt_line("    {");
1093 	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1094 	    "MaxFixed, PosDecode,");
1095 	dsdt_line("        0x0000,             // Granularity");
1096 	dsdt_line("        0x0000,             // Range Minimum");
1097 	dsdt_line("        0x00FF,             // Range Maximum");
1098 	dsdt_line("        0x0000,             // Translation Offset");
1099 	dsdt_line("        0x0100,             // Length");
1100 	dsdt_line("        ,, )");
1101 	dsdt_indent(3);
1102 	dsdt_fixed_ioport(0xCF8, 8);
1103 	dsdt_unindent(3);
1104 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1105 	    "PosDecode, EntireRange,");
1106 	dsdt_line("        0x0000,             // Granularity");
1107 	dsdt_line("        0x0000,             // Range Minimum");
1108 	dsdt_line("        0x0CF7,             // Range Maximum");
1109 	dsdt_line("        0x0000,             // Translation Offset");
1110 	dsdt_line("        0x0CF8,             // Length");
1111 	dsdt_line("        ,, , TypeStatic)");
1112 	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1113 	    "PosDecode, EntireRange,");
1114 	dsdt_line("        0x0000,             // Granularity");
1115 	dsdt_line("        0x0D00,             // Range Minimum");
1116 	dsdt_line("        0xFFFF,             // Range Maximum");
1117 	dsdt_line("        0x0000,             // Translation Offset");
1118 	dsdt_line("        0xF300,             // Length");
1119 	dsdt_line("        ,, , TypeStatic)");
1120 	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1121 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1122 	dsdt_line("        0x00000000,         // Granularity");
1123 	dsdt_line("        0x%08lX,         // Range Minimum\n",
1124 	    pci_mem_hole.base);
1125 	dsdt_line("        0x%08X,         // Range Maximum\n",
1126 	    PCI_EMUL_MEMLIMIT32 - 1);
1127 	dsdt_line("        0x00000000,         // Translation Offset");
1128 	dsdt_line("        0x%08lX,         // Length\n",
1129 	    PCI_EMUL_MEMLIMIT32 - pci_mem_hole.base);
1130 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1131 	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1132 	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1133 	dsdt_line("        0x0000000000000000, // Granularity");
1134 	dsdt_line("        0x%016lX, // Range Minimum\n",
1135 	    PCI_EMUL_MEMBASE64);
1136 	dsdt_line("        0x%016lX, // Range Maximum\n",
1137 	    PCI_EMUL_MEMLIMIT64 - 1);
1138 	dsdt_line("        0x0000000000000000, // Translation Offset");
1139 	dsdt_line("        0x%016lX, // Length\n",
1140 	    PCI_EMUL_MEMLIMIT64 - PCI_EMUL_MEMBASE64);
1141 	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1142 	dsdt_line("    })");
1143 	count = pci_count_lintr();
1144 	if (count != 0) {
1145 		dsdt_indent(2);
1146 		dsdt_line("Name (_PRT, Package (0x%02X)", count);
1147 		dsdt_line("{");
1148 		pci_walk_lintr(pci_prt_entry, &count);
1149 		dsdt_line("})");
1150 		dsdt_unindent(2);
1151 	}
1152 
1153 	dsdt_indent(2);
1154 	for (slot = 0; slot < MAXSLOTS; slot++) {
1155 		for (func = 0; func < MAXFUNCS; func++) {
1156 			pi = pci_slotinfo[slot].si_funcs[func].fi_devi;
1157 			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1158 				pi->pi_d->pe_write_dsdt(pi);
1159 		}
1160 	}
1161 	dsdt_unindent(2);
1162 
1163 	dsdt_line("  }");
1164 	dsdt_line("}");
1165 	dsdt_unindent(1);
1166 }
1167 
1168 int
1169 pci_msi_enabled(struct pci_devinst *pi)
1170 {
1171 	return (pi->pi_msi.enabled);
1172 }
1173 
1174 int
1175 pci_msi_maxmsgnum(struct pci_devinst *pi)
1176 {
1177 	if (pi->pi_msi.enabled)
1178 		return (pi->pi_msi.maxmsgnum);
1179 	else
1180 		return (0);
1181 }
1182 
1183 int
1184 pci_msix_enabled(struct pci_devinst *pi)
1185 {
1186 
1187 	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1188 }
1189 
1190 void
1191 pci_generate_msix(struct pci_devinst *pi, int index)
1192 {
1193 	struct msix_table_entry *mte;
1194 
1195 	if (!pci_msix_enabled(pi))
1196 		return;
1197 
1198 	if (pi->pi_msix.function_mask)
1199 		return;
1200 
1201 	if (index >= pi->pi_msix.table_count)
1202 		return;
1203 
1204 	mte = &pi->pi_msix.table[index];
1205 	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1206 		/* XXX Set PBA bit if interrupt is disabled */
1207 		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1208 	}
1209 }
1210 
1211 void
1212 pci_generate_msi(struct pci_devinst *pi, int index)
1213 {
1214 
1215 	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1216 		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1217 			     pi->pi_msi.msg_data + index);
1218 	}
1219 }
1220 
1221 static bool
1222 pci_lintr_permitted(struct pci_devinst *pi)
1223 {
1224 	uint16_t cmd;
1225 
1226 	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1227 	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1228 		(cmd & PCIM_CMD_INTxDIS)));
1229 }
1230 
1231 int
1232 pci_lintr_request(struct pci_devinst *pi)
1233 {
1234 	struct slotinfo *si;
1235 	int bestpin, bestcount, irq, pin;
1236 
1237 	/*
1238 	 * First, allocate a pin from our slot.
1239 	 */
1240 	si = &pci_slotinfo[pi->pi_slot];
1241 	bestpin = 0;
1242 	bestcount = si->si_intpins[0].ii_count;
1243 	for (pin = 1; pin < 4; pin++) {
1244 		if (si->si_intpins[pin].ii_count < bestcount) {
1245 			bestpin = pin;
1246 			bestcount = si->si_intpins[pin].ii_count;
1247 		}
1248 	}
1249 
1250 	/*
1251 	 * Attempt to allocate an I/O APIC pin for this intpin.  If
1252 	 * 8259A support is added we will need a separate field to
1253 	 * assign the intpin to an input pin on the PCI interrupt
1254 	 * router.
1255 	 */
1256 	if (si->si_intpins[bestpin].ii_count == 0) {
1257 		irq = ioapic_pci_alloc_irq();
1258 		if (irq < 0)
1259 			return (-1);
1260 		si->si_intpins[bestpin].ii_ioapic_irq = irq;
1261 	} else
1262 		irq = si->si_intpins[bestpin].ii_ioapic_irq;
1263 	si->si_intpins[bestpin].ii_count++;
1264 
1265 	pi->pi_lintr.pin = bestpin + 1;
1266 	pi->pi_lintr.ioapic_irq = irq;
1267 	pci_set_cfgdata8(pi, PCIR_INTLINE, irq);
1268 	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1269 	return (0);
1270 }
1271 
1272 void
1273 pci_lintr_assert(struct pci_devinst *pi)
1274 {
1275 
1276 	assert(pi->pi_lintr.pin > 0);
1277 
1278 	pthread_mutex_lock(&pi->pi_lintr.lock);
1279 	if (pi->pi_lintr.state == IDLE) {
1280 		if (pci_lintr_permitted(pi)) {
1281 			pi->pi_lintr.state = ASSERTED;
1282 			vm_ioapic_assert_irq(pi->pi_vmctx,
1283 			    pi->pi_lintr.ioapic_irq);
1284 		} else
1285 			pi->pi_lintr.state = PENDING;
1286 	}
1287 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1288 }
1289 
1290 void
1291 pci_lintr_deassert(struct pci_devinst *pi)
1292 {
1293 
1294 	assert(pi->pi_lintr.pin > 0);
1295 
1296 	pthread_mutex_lock(&pi->pi_lintr.lock);
1297 	if (pi->pi_lintr.state == ASSERTED) {
1298 		pi->pi_lintr.state = IDLE;
1299 		vm_ioapic_deassert_irq(pi->pi_vmctx, pi->pi_lintr.ioapic_irq);
1300 	} else if (pi->pi_lintr.state == PENDING)
1301 		pi->pi_lintr.state = IDLE;
1302 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1303 }
1304 
1305 static void
1306 pci_lintr_update(struct pci_devinst *pi)
1307 {
1308 
1309 	pthread_mutex_lock(&pi->pi_lintr.lock);
1310 	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1311 		vm_ioapic_deassert_irq(pi->pi_vmctx, pi->pi_lintr.ioapic_irq);
1312 		pi->pi_lintr.state = PENDING;
1313 	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1314 		pi->pi_lintr.state = ASSERTED;
1315 		vm_ioapic_assert_irq(pi->pi_vmctx, pi->pi_lintr.ioapic_irq);
1316 	}
1317 	pthread_mutex_unlock(&pi->pi_lintr.lock);
1318 }
1319 
1320 int
1321 pci_count_lintr(void)
1322 {
1323 	int count, slot, pin;
1324 
1325 	count = 0;
1326 	for (slot = 0; slot < MAXSLOTS; slot++) {
1327 		for (pin = 0; pin < 4; pin++) {
1328 			if (pci_slotinfo[slot].si_intpins[pin].ii_count != 0)
1329 				count++;
1330 		}
1331 	}
1332 	return (count);
1333 }
1334 
1335 void
1336 pci_walk_lintr(pci_lintr_cb cb, void *arg)
1337 {
1338 	struct intxinfo *ii;
1339 	int slot, pin;
1340 
1341 	for (slot = 0; slot < MAXSLOTS; slot++) {
1342 		for (pin = 0; pin < 4; pin++) {
1343 			ii = &pci_slotinfo[slot].si_intpins[pin];
1344 			if (ii->ii_count != 0)
1345 				cb(slot, pin + 1, ii->ii_ioapic_irq, arg);
1346 		}
1347 	}
1348 }
1349 
1350 /*
1351  * Return 1 if the emulated device in 'slot' is a multi-function device.
1352  * Return 0 otherwise.
1353  */
1354 static int
1355 pci_emul_is_mfdev(int slot)
1356 {
1357 	int f, numfuncs;
1358 
1359 	numfuncs = 0;
1360 	for (f = 0; f < MAXFUNCS; f++) {
1361 		if (pci_slotinfo[slot].si_funcs[f].fi_devi != NULL) {
1362 			numfuncs++;
1363 		}
1364 	}
1365 	return (numfuncs > 1);
1366 }
1367 
1368 /*
1369  * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1370  * whether or not is a multi-function being emulated in the pci 'slot'.
1371  */
1372 static void
1373 pci_emul_hdrtype_fixup(int slot, int off, int bytes, uint32_t *rv)
1374 {
1375 	int mfdev;
1376 
1377 	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1378 		mfdev = pci_emul_is_mfdev(slot);
1379 		switch (bytes) {
1380 		case 1:
1381 		case 2:
1382 			*rv &= ~PCIM_MFDEV;
1383 			if (mfdev) {
1384 				*rv |= PCIM_MFDEV;
1385 			}
1386 			break;
1387 		case 4:
1388 			*rv &= ~(PCIM_MFDEV << 16);
1389 			if (mfdev) {
1390 				*rv |= (PCIM_MFDEV << 16);
1391 			}
1392 			break;
1393 		}
1394 	}
1395 }
1396 
1397 static int cfgbus, cfgslot, cfgfunc, cfgoff;
1398 
1399 static int
1400 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1401 		 uint32_t *eax, void *arg)
1402 {
1403 	uint32_t x;
1404 
1405 	if (bytes != 4) {
1406 		if (in)
1407 			*eax = (bytes == 2) ? 0xffff : 0xff;
1408 		return (0);
1409 	}
1410 
1411 	if (in) {
1412 		x = (cfgbus << 16) |
1413 		    (cfgslot << 11) |
1414 		    (cfgfunc << 8) |
1415 		    cfgoff;
1416 		*eax = x | CONF1_ENABLE;
1417 	} else {
1418 		x = *eax;
1419 		cfgoff = x & PCI_REGMAX;
1420 		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1421 		cfgslot = (x >> 11) & PCI_SLOTMAX;
1422 		cfgbus = (x >> 16) & PCI_BUSMAX;
1423 	}
1424 
1425 	return (0);
1426 }
1427 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1428 
1429 static uint32_t
1430 bits_changed(uint32_t old, uint32_t new, uint32_t mask)
1431 {
1432 
1433 	return ((old ^ new) & mask);
1434 }
1435 
1436 static void
1437 pci_emul_cmdwrite(struct pci_devinst *pi, uint32_t new, int bytes)
1438 {
1439 	int i;
1440 	uint16_t old;
1441 
1442 	/*
1443 	 * The command register is at an offset of 4 bytes and thus the
1444 	 * guest could write 1, 2 or 4 bytes starting at this offset.
1445 	 */
1446 
1447 	old = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1448 	CFGWRITE(pi, PCIR_COMMAND, new, bytes);		/* update config */
1449 	new = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1450 
1451 	/*
1452 	 * If the MMIO or I/O address space decoding has changed then
1453 	 * register/unregister all BARs that decode that address space.
1454 	 */
1455 	for (i = 0; i <= PCI_BARMAX; i++) {
1456 		switch (pi->pi_bar[i].type) {
1457 			case PCIBAR_NONE:
1458 			case PCIBAR_MEMHI64:
1459 				break;
1460 			case PCIBAR_IO:
1461 				/* I/O address space decoding changed? */
1462 				if (bits_changed(old, new, PCIM_CMD_PORTEN)) {
1463 					if (porten(pi))
1464 						register_bar(pi, i);
1465 					else
1466 						unregister_bar(pi, i);
1467 				}
1468 				break;
1469 			case PCIBAR_MEM32:
1470 			case PCIBAR_MEM64:
1471 				/* MMIO address space decoding changed? */
1472 				if (bits_changed(old, new, PCIM_CMD_MEMEN)) {
1473 					if (memen(pi))
1474 						register_bar(pi, i);
1475 					else
1476 						unregister_bar(pi, i);
1477 				}
1478 				break;
1479 			default:
1480 				assert(0);
1481 		}
1482 	}
1483 
1484 	/*
1485 	 * If INTx has been unmasked and is pending, assert the
1486 	 * interrupt.
1487 	 */
1488 	pci_lintr_update(pi);
1489 }
1490 
1491 static int
1492 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1493 		 uint32_t *eax, void *arg)
1494 {
1495 	struct pci_devinst *pi;
1496 	struct pci_devemu *pe;
1497 	int coff, idx, needcfg;
1498 	uint64_t addr, bar, mask;
1499 
1500 	assert(bytes == 1 || bytes == 2 || bytes == 4);
1501 
1502 	if (cfgbus == 0)
1503 		pi = pci_slotinfo[cfgslot].si_funcs[cfgfunc].fi_devi;
1504 	else
1505 		pi = NULL;
1506 
1507 	coff = cfgoff + (port - CONF1_DATA_PORT);
1508 
1509 #if 0
1510 	printf("pcicfg-%s from 0x%0x of %d bytes (%d/%d/%d)\n\r",
1511 		in ? "read" : "write", coff, bytes, cfgbus, cfgslot, cfgfunc);
1512 #endif
1513 
1514 	/*
1515 	 * Just return if there is no device at this cfgslot:cfgfunc or
1516 	 * if the guest is doing an un-aligned access
1517 	 */
1518 	if (pi == NULL || (coff & (bytes - 1)) != 0) {
1519 		if (in)
1520 			*eax = 0xffffffff;
1521 		return (0);
1522 	}
1523 
1524 	pe = pi->pi_d;
1525 
1526 	/*
1527 	 * Config read
1528 	 */
1529 	if (in) {
1530 		/* Let the device emulation override the default handler */
1531 		if (pe->pe_cfgread != NULL) {
1532 			needcfg = pe->pe_cfgread(ctx, vcpu, pi,
1533 						    coff, bytes, eax);
1534 		} else {
1535 			needcfg = 1;
1536 		}
1537 
1538 		if (needcfg) {
1539 			if (bytes == 1)
1540 				*eax = pci_get_cfgdata8(pi, coff);
1541 			else if (bytes == 2)
1542 				*eax = pci_get_cfgdata16(pi, coff);
1543 			else
1544 				*eax = pci_get_cfgdata32(pi, coff);
1545 		}
1546 
1547 		pci_emul_hdrtype_fixup(cfgslot, coff, bytes, eax);
1548 	} else {
1549 		/* Let the device emulation override the default handler */
1550 		if (pe->pe_cfgwrite != NULL &&
1551 		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1552 			return (0);
1553 
1554 		/*
1555 		 * Special handling for write to BAR registers
1556 		 */
1557 		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1558 			/*
1559 			 * Ignore writes to BAR registers that are not
1560 			 * 4-byte aligned.
1561 			 */
1562 			if (bytes != 4 || (coff & 0x3) != 0)
1563 				return (0);
1564 			idx = (coff - PCIR_BAR(0)) / 4;
1565 			mask = ~(pi->pi_bar[idx].size - 1);
1566 			switch (pi->pi_bar[idx].type) {
1567 			case PCIBAR_NONE:
1568 				pi->pi_bar[idx].addr = bar = 0;
1569 				break;
1570 			case PCIBAR_IO:
1571 				addr = *eax & mask;
1572 				addr &= 0xffff;
1573 				bar = addr | PCIM_BAR_IO_SPACE;
1574 				/*
1575 				 * Register the new BAR value for interception
1576 				 */
1577 				if (addr != pi->pi_bar[idx].addr) {
1578 					update_bar_address(pi, addr, idx,
1579 							   PCIBAR_IO);
1580 				}
1581 				break;
1582 			case PCIBAR_MEM32:
1583 				addr = bar = *eax & mask;
1584 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1585 				if (addr != pi->pi_bar[idx].addr) {
1586 					update_bar_address(pi, addr, idx,
1587 							   PCIBAR_MEM32);
1588 				}
1589 				break;
1590 			case PCIBAR_MEM64:
1591 				addr = bar = *eax & mask;
1592 				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1593 				       PCIM_BAR_MEM_PREFETCH;
1594 				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1595 					update_bar_address(pi, addr, idx,
1596 							   PCIBAR_MEM64);
1597 				}
1598 				break;
1599 			case PCIBAR_MEMHI64:
1600 				mask = ~(pi->pi_bar[idx - 1].size - 1);
1601 				addr = ((uint64_t)*eax << 32) & mask;
1602 				bar = addr >> 32;
1603 				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1604 					update_bar_address(pi, addr, idx - 1,
1605 							   PCIBAR_MEMHI64);
1606 				}
1607 				break;
1608 			default:
1609 				assert(0);
1610 			}
1611 			pci_set_cfgdata32(pi, coff, bar);
1612 
1613 		} else if (pci_emul_iscap(pi, coff)) {
1614 			pci_emul_capwrite(pi, coff, bytes, *eax);
1615 		} else if (coff == PCIR_COMMAND) {
1616 			pci_emul_cmdwrite(pi, *eax, bytes);
1617 		} else {
1618 			CFGWRITE(pi, coff, *eax, bytes);
1619 		}
1620 	}
1621 
1622 	return (0);
1623 }
1624 
1625 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1626 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1627 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1628 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1629 
1630 /*
1631  * I/O ports to configure PCI IRQ routing. We ignore all writes to it.
1632  */
1633 static int
1634 pci_irq_port_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1635 		     uint32_t *eax, void *arg)
1636 {
1637 	assert(in == 0);
1638 	return (0);
1639 }
1640 INOUT_PORT(pci_irq, 0xC00, IOPORT_F_OUT, pci_irq_port_handler);
1641 INOUT_PORT(pci_irq, 0xC01, IOPORT_F_OUT, pci_irq_port_handler);
1642 SYSRES_IO(0xC00, 2);
1643 
1644 #define PCI_EMUL_TEST
1645 #ifdef PCI_EMUL_TEST
1646 /*
1647  * Define a dummy test device
1648  */
1649 #define DIOSZ	20
1650 #define DMEMSZ	4096
1651 struct pci_emul_dsoftc {
1652 	uint8_t   ioregs[DIOSZ];
1653 	uint8_t	  memregs[DMEMSZ];
1654 };
1655 
1656 #define	PCI_EMUL_MSI_MSGS	 4
1657 #define	PCI_EMUL_MSIX_MSGS	16
1658 
1659 static int
1660 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1661 {
1662 	int error;
1663 	struct pci_emul_dsoftc *sc;
1664 
1665 	sc = malloc(sizeof(struct pci_emul_dsoftc));
1666 	memset(sc, 0, sizeof(struct pci_emul_dsoftc));
1667 
1668 	pi->pi_arg = sc;
1669 
1670 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1671 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1672 	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1673 
1674 	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1675 	assert(error == 0);
1676 
1677 	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1678 	assert(error == 0);
1679 
1680 	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1681 	assert(error == 0);
1682 
1683 	return (0);
1684 }
1685 
1686 static void
1687 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1688 	      uint64_t offset, int size, uint64_t value)
1689 {
1690 	int i;
1691 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1692 
1693 	if (baridx == 0) {
1694 		if (offset + size > DIOSZ) {
1695 			printf("diow: iow too large, offset %ld size %d\n",
1696 			       offset, size);
1697 			return;
1698 		}
1699 
1700 		if (size == 1) {
1701 			sc->ioregs[offset] = value & 0xff;
1702 		} else if (size == 2) {
1703 			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1704 		} else if (size == 4) {
1705 			*(uint32_t *)&sc->ioregs[offset] = value;
1706 		} else {
1707 			printf("diow: iow unknown size %d\n", size);
1708 		}
1709 
1710 		/*
1711 		 * Special magic value to generate an interrupt
1712 		 */
1713 		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
1714 			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
1715 
1716 		if (value == 0xabcdef) {
1717 			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
1718 				pci_generate_msi(pi, i);
1719 		}
1720 	}
1721 
1722 	if (baridx == 1) {
1723 		if (offset + size > DMEMSZ) {
1724 			printf("diow: memw too large, offset %ld size %d\n",
1725 			       offset, size);
1726 			return;
1727 		}
1728 
1729 		if (size == 1) {
1730 			sc->memregs[offset] = value;
1731 		} else if (size == 2) {
1732 			*(uint16_t *)&sc->memregs[offset] = value;
1733 		} else if (size == 4) {
1734 			*(uint32_t *)&sc->memregs[offset] = value;
1735 		} else if (size == 8) {
1736 			*(uint64_t *)&sc->memregs[offset] = value;
1737 		} else {
1738 			printf("diow: memw unknown size %d\n", size);
1739 		}
1740 
1741 		/*
1742 		 * magic interrupt ??
1743 		 */
1744 	}
1745 
1746 	if (baridx > 1) {
1747 		printf("diow: unknown bar idx %d\n", baridx);
1748 	}
1749 }
1750 
1751 static uint64_t
1752 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1753 	      uint64_t offset, int size)
1754 {
1755 	struct pci_emul_dsoftc *sc = pi->pi_arg;
1756 	uint32_t value;
1757 
1758 	if (baridx == 0) {
1759 		if (offset + size > DIOSZ) {
1760 			printf("dior: ior too large, offset %ld size %d\n",
1761 			       offset, size);
1762 			return (0);
1763 		}
1764 
1765 		if (size == 1) {
1766 			value = sc->ioregs[offset];
1767 		} else if (size == 2) {
1768 			value = *(uint16_t *) &sc->ioregs[offset];
1769 		} else if (size == 4) {
1770 			value = *(uint32_t *) &sc->ioregs[offset];
1771 		} else {
1772 			printf("dior: ior unknown size %d\n", size);
1773 		}
1774 	}
1775 
1776 	if (baridx == 1) {
1777 		if (offset + size > DMEMSZ) {
1778 			printf("dior: memr too large, offset %ld size %d\n",
1779 			       offset, size);
1780 			return (0);
1781 		}
1782 
1783 		if (size == 1) {
1784 			value = sc->memregs[offset];
1785 		} else if (size == 2) {
1786 			value = *(uint16_t *) &sc->memregs[offset];
1787 		} else if (size == 4) {
1788 			value = *(uint32_t *) &sc->memregs[offset];
1789 		} else if (size == 8) {
1790 			value = *(uint64_t *) &sc->memregs[offset];
1791 		} else {
1792 			printf("dior: ior unknown size %d\n", size);
1793 		}
1794 	}
1795 
1796 
1797 	if (baridx > 1) {
1798 		printf("dior: unknown bar idx %d\n", baridx);
1799 		return (0);
1800 	}
1801 
1802 	return (value);
1803 }
1804 
1805 struct pci_devemu pci_dummy = {
1806 	.pe_emu = "dummy",
1807 	.pe_init = pci_emul_dinit,
1808 	.pe_barwrite = pci_emul_diow,
1809 	.pe_barread = pci_emul_dior
1810 };
1811 PCI_EMUL_SET(pci_dummy);
1812 
1813 #endif /* PCI_EMUL_TEST */
1814