pci_emul.c revision 293290
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: stable/10/usr.sbin/bhyve/pci_emul.c 293290 2016-01-07 00:40:51Z bdrewery $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_emul.c 293290 2016-01-07 00:40:51Z bdrewery $");
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_irq.h"
55#include "pci_lpc.h"
56
57#define CONF1_ADDR_PORT    0x0cf8
58#define CONF1_DATA_PORT    0x0cfc
59
60#define CONF1_ENABLE	   0x80000000ul
61
62#define	MAXBUSES	(PCI_BUSMAX + 1)
63#define MAXSLOTS	(PCI_SLOTMAX + 1)
64#define	MAXFUNCS	(PCI_FUNCMAX + 1)
65
66struct funcinfo {
67	char	*fi_name;
68	char	*fi_param;
69	struct pci_devinst *fi_devi;
70};
71
72struct intxinfo {
73	int	ii_count;
74	int	ii_pirq_pin;
75	int	ii_ioapic_irq;
76};
77
78struct slotinfo {
79	struct intxinfo si_intpins[4];
80	struct funcinfo si_funcs[MAXFUNCS];
81};
82
83struct businfo {
84	uint16_t iobase, iolimit;		/* I/O window */
85	uint32_t membase32, memlimit32;		/* mmio window below 4GB */
86	uint64_t membase64, memlimit64;		/* mmio window above 4GB */
87	struct slotinfo slotinfo[MAXSLOTS];
88};
89
90static struct businfo *pci_businfo[MAXBUSES];
91
92SET_DECLARE(pci_devemu_set, struct pci_devemu);
93
94static uint64_t pci_emul_iobase;
95static uint64_t pci_emul_membase32;
96static uint64_t pci_emul_membase64;
97
98#define	PCI_EMUL_IOBASE		0x2000
99#define	PCI_EMUL_IOLIMIT	0x10000
100
101#define	PCI_EMUL_ECFG_BASE	0xE0000000		    /* 3.5GB */
102#define	PCI_EMUL_ECFG_SIZE	(MAXBUSES * 1024 * 1024)    /* 1MB per bus */
103SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
104
105#define	PCI_EMUL_MEMLIMIT32	PCI_EMUL_ECFG_BASE
106
107#define	PCI_EMUL_MEMBASE64	0xD000000000UL
108#define	PCI_EMUL_MEMLIMIT64	0xFD00000000UL
109
110static struct pci_devemu *pci_emul_finddev(char *name);
111static void pci_lintr_route(struct pci_devinst *pi);
112static void pci_lintr_update(struct pci_devinst *pi);
113static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
114    int func, int coff, int bytes, uint32_t *val);
115
116static __inline void
117CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
118{
119
120	if (bytes == 1)
121		pci_set_cfgdata8(pi, coff, val);
122	else if (bytes == 2)
123		pci_set_cfgdata16(pi, coff, val);
124	else
125		pci_set_cfgdata32(pi, coff, val);
126}
127
128static __inline uint32_t
129CFGREAD(struct pci_devinst *pi, int coff, int bytes)
130{
131
132	if (bytes == 1)
133		return (pci_get_cfgdata8(pi, coff));
134	else if (bytes == 2)
135		return (pci_get_cfgdata16(pi, coff));
136	else
137		return (pci_get_cfgdata32(pi, coff));
138}
139
140/*
141 * I/O access
142 */
143
144/*
145 * Slot options are in the form:
146 *
147 *  <bus>:<slot>:<func>,<emul>[,<config>]
148 *  <slot>[:<func>],<emul>[,<config>]
149 *
150 *  slot is 0..31
151 *  func is 0..7
152 *  emul is a string describing the type of PCI device e.g. virtio-net
153 *  config is an optional string, depending on the device, that can be
154 *  used for configuration.
155 *   Examples are:
156 *     1,virtio-net,tap0
157 *     3:0,dummy
158 */
159static void
160pci_parse_slot_usage(char *aopt)
161{
162
163	fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
164}
165
166int
167pci_parse_slot(char *opt)
168{
169	struct businfo *bi;
170	struct slotinfo *si;
171	char *emul, *config, *str, *cp;
172	int error, bnum, snum, fnum;
173
174	error = -1;
175	str = strdup(opt);
176
177	emul = config = NULL;
178	if ((cp = strchr(str, ',')) != NULL) {
179		*cp = '\0';
180		emul = cp + 1;
181		if ((cp = strchr(emul, ',')) != NULL) {
182			*cp = '\0';
183			config = cp + 1;
184		}
185	} else {
186		pci_parse_slot_usage(opt);
187		goto done;
188	}
189
190	/* <bus>:<slot>:<func> */
191	if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
192		bnum = 0;
193		/* <slot>:<func> */
194		if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
195			fnum = 0;
196			/* <slot> */
197			if (sscanf(str, "%d", &snum) != 1) {
198				snum = -1;
199			}
200		}
201	}
202
203	if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
204	    fnum < 0 || fnum >= MAXFUNCS) {
205		pci_parse_slot_usage(opt);
206		goto done;
207	}
208
209	if (pci_businfo[bnum] == NULL)
210		pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
211
212	bi = pci_businfo[bnum];
213	si = &bi->slotinfo[snum];
214
215	if (si->si_funcs[fnum].fi_name != NULL) {
216		fprintf(stderr, "pci slot %d:%d already occupied!\n",
217			snum, fnum);
218		goto done;
219	}
220
221	if (pci_emul_finddev(emul) == NULL) {
222		fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
223			snum, fnum, emul);
224		goto done;
225	}
226
227	error = 0;
228	si->si_funcs[fnum].fi_name = emul;
229	si->si_funcs[fnum].fi_param = config;
230
231done:
232	if (error)
233		free(str);
234
235	return (error);
236}
237
238static int
239pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
240{
241
242	if (offset < pi->pi_msix.pba_offset)
243		return (0);
244
245	if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
246		return (0);
247	}
248
249	return (1);
250}
251
252int
253pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
254		     uint64_t value)
255{
256	int msix_entry_offset;
257	int tab_index;
258	char *dest;
259
260	/* support only 4 or 8 byte writes */
261	if (size != 4 && size != 8)
262		return (-1);
263
264	/*
265	 * Return if table index is beyond what device supports
266	 */
267	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
268	if (tab_index >= pi->pi_msix.table_count)
269		return (-1);
270
271	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
272
273	/* support only aligned writes */
274	if ((msix_entry_offset % size) != 0)
275		return (-1);
276
277	dest = (char *)(pi->pi_msix.table + tab_index);
278	dest += msix_entry_offset;
279
280	if (size == 4)
281		*((uint32_t *)dest) = value;
282	else
283		*((uint64_t *)dest) = value;
284
285	return (0);
286}
287
288uint64_t
289pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
290{
291	char *dest;
292	int msix_entry_offset;
293	int tab_index;
294	uint64_t retval = ~0;
295
296	/*
297	 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
298	 * table but we also allow 1 byte access to accommodate reads from
299	 * ddb.
300	 */
301	if (size != 1 && size != 4 && size != 8)
302		return (retval);
303
304	msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
305
306	/* support only aligned reads */
307	if ((msix_entry_offset % size) != 0) {
308		return (retval);
309	}
310
311	tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
312
313	if (tab_index < pi->pi_msix.table_count) {
314		/* valid MSI-X Table access */
315		dest = (char *)(pi->pi_msix.table + tab_index);
316		dest += msix_entry_offset;
317
318		if (size == 1)
319			retval = *((uint8_t *)dest);
320		else if (size == 4)
321			retval = *((uint32_t *)dest);
322		else
323			retval = *((uint64_t *)dest);
324	} else if (pci_valid_pba_offset(pi, offset)) {
325		/* return 0 for PBA access */
326		retval = 0;
327	}
328
329	return (retval);
330}
331
332int
333pci_msix_table_bar(struct pci_devinst *pi)
334{
335
336	if (pi->pi_msix.table != NULL)
337		return (pi->pi_msix.table_bar);
338	else
339		return (-1);
340}
341
342int
343pci_msix_pba_bar(struct pci_devinst *pi)
344{
345
346	if (pi->pi_msix.table != NULL)
347		return (pi->pi_msix.pba_bar);
348	else
349		return (-1);
350}
351
352static int
353pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
354		    uint32_t *eax, void *arg)
355{
356	struct pci_devinst *pdi = arg;
357	struct pci_devemu *pe = pdi->pi_d;
358	uint64_t offset;
359	int i;
360
361	for (i = 0; i <= PCI_BARMAX; i++) {
362		if (pdi->pi_bar[i].type == PCIBAR_IO &&
363		    port >= pdi->pi_bar[i].addr &&
364		    port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
365			offset = port - pdi->pi_bar[i].addr;
366			if (in)
367				*eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
368							 offset, bytes);
369			else
370				(*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
371						   bytes, *eax);
372			return (0);
373		}
374	}
375	return (-1);
376}
377
378static int
379pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
380		     int size, uint64_t *val, void *arg1, long arg2)
381{
382	struct pci_devinst *pdi = arg1;
383	struct pci_devemu *pe = pdi->pi_d;
384	uint64_t offset;
385	int bidx = (int) arg2;
386
387	assert(bidx <= PCI_BARMAX);
388	assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
389	       pdi->pi_bar[bidx].type == PCIBAR_MEM64);
390	assert(addr >= pdi->pi_bar[bidx].addr &&
391	       addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
392
393	offset = addr - pdi->pi_bar[bidx].addr;
394
395	if (dir == MEM_F_WRITE) {
396		if (size == 8) {
397			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
398					   4, *val & 0xffffffff);
399			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
400					   4, *val >> 32);
401		} else {
402			(*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
403					   size, *val);
404		}
405	} else {
406		if (size == 8) {
407			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
408						 offset, 4);
409			*val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
410						  offset + 4, 4) << 32;
411		} else {
412			*val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
413						 offset, size);
414		}
415	}
416
417	return (0);
418}
419
420
421static int
422pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
423			uint64_t *addr)
424{
425	uint64_t base;
426
427	assert((size & (size - 1)) == 0);	/* must be a power of 2 */
428
429	base = roundup2(*baseptr, size);
430
431	if (base + size <= limit) {
432		*addr = base;
433		*baseptr = base + size;
434		return (0);
435	} else
436		return (-1);
437}
438
439int
440pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
441		   uint64_t size)
442{
443
444	return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
445}
446
447/*
448 * Register (or unregister) the MMIO or I/O region associated with the BAR
449 * register 'idx' of an emulated pci device.
450 */
451static void
452modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
453{
454	int error;
455	struct inout_port iop;
456	struct mem_range mr;
457
458	switch (pi->pi_bar[idx].type) {
459	case PCIBAR_IO:
460		bzero(&iop, sizeof(struct inout_port));
461		iop.name = pi->pi_name;
462		iop.port = pi->pi_bar[idx].addr;
463		iop.size = pi->pi_bar[idx].size;
464		if (registration) {
465			iop.flags = IOPORT_F_INOUT;
466			iop.handler = pci_emul_io_handler;
467			iop.arg = pi;
468			error = register_inout(&iop);
469		} else
470			error = unregister_inout(&iop);
471		break;
472	case PCIBAR_MEM32:
473	case PCIBAR_MEM64:
474		bzero(&mr, sizeof(struct mem_range));
475		mr.name = pi->pi_name;
476		mr.base = pi->pi_bar[idx].addr;
477		mr.size = pi->pi_bar[idx].size;
478		if (registration) {
479			mr.flags = MEM_F_RW;
480			mr.handler = pci_emul_mem_handler;
481			mr.arg1 = pi;
482			mr.arg2 = idx;
483			error = register_mem(&mr);
484		} else
485			error = unregister_mem(&mr);
486		break;
487	default:
488		error = EINVAL;
489		break;
490	}
491	assert(error == 0);
492}
493
494static void
495unregister_bar(struct pci_devinst *pi, int idx)
496{
497
498	modify_bar_registration(pi, idx, 0);
499}
500
501static void
502register_bar(struct pci_devinst *pi, int idx)
503{
504
505	modify_bar_registration(pi, idx, 1);
506}
507
508/* Are we decoding i/o port accesses for the emulated pci device? */
509static int
510porten(struct pci_devinst *pi)
511{
512	uint16_t cmd;
513
514	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
515
516	return (cmd & PCIM_CMD_PORTEN);
517}
518
519/* Are we decoding memory accesses for the emulated pci device? */
520static int
521memen(struct pci_devinst *pi)
522{
523	uint16_t cmd;
524
525	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
526
527	return (cmd & PCIM_CMD_MEMEN);
528}
529
530/*
531 * Update the MMIO or I/O address that is decoded by the BAR register.
532 *
533 * If the pci device has enabled the address space decoding then intercept
534 * the address range decoded by the BAR register.
535 */
536static void
537update_bar_address(struct  pci_devinst *pi, uint64_t addr, int idx, int type)
538{
539	int decode;
540
541	if (pi->pi_bar[idx].type == PCIBAR_IO)
542		decode = porten(pi);
543	else
544		decode = memen(pi);
545
546	if (decode)
547		unregister_bar(pi, idx);
548
549	switch (type) {
550	case PCIBAR_IO:
551	case PCIBAR_MEM32:
552		pi->pi_bar[idx].addr = addr;
553		break;
554	case PCIBAR_MEM64:
555		pi->pi_bar[idx].addr &= ~0xffffffffUL;
556		pi->pi_bar[idx].addr |= addr;
557		break;
558	case PCIBAR_MEMHI64:
559		pi->pi_bar[idx].addr &= 0xffffffff;
560		pi->pi_bar[idx].addr |= addr;
561		break;
562	default:
563		assert(0);
564	}
565
566	if (decode)
567		register_bar(pi, idx);
568}
569
570int
571pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
572		    enum pcibar_type type, uint64_t size)
573{
574	int error;
575	uint64_t *baseptr, limit, addr, mask, lobits, bar;
576
577	assert(idx >= 0 && idx <= PCI_BARMAX);
578
579	if ((size & (size - 1)) != 0)
580		size = 1UL << flsl(size);	/* round up to a power of 2 */
581
582	/* Enforce minimum BAR sizes required by the PCI standard */
583	if (type == PCIBAR_IO) {
584		if (size < 4)
585			size = 4;
586	} else {
587		if (size < 16)
588			size = 16;
589	}
590
591	switch (type) {
592	case PCIBAR_NONE:
593		baseptr = NULL;
594		addr = mask = lobits = 0;
595		break;
596	case PCIBAR_IO:
597		baseptr = &pci_emul_iobase;
598		limit = PCI_EMUL_IOLIMIT;
599		mask = PCIM_BAR_IO_BASE;
600		lobits = PCIM_BAR_IO_SPACE;
601		break;
602	case PCIBAR_MEM64:
603		/*
604		 * XXX
605		 * Some drivers do not work well if the 64-bit BAR is allocated
606		 * above 4GB. Allow for this by allocating small requests under
607		 * 4GB unless then allocation size is larger than some arbitrary
608		 * number (32MB currently).
609		 */
610		if (size > 32 * 1024 * 1024) {
611			/*
612			 * XXX special case for device requiring peer-peer DMA
613			 */
614			if (size == 0x100000000UL)
615				baseptr = &hostbase;
616			else
617				baseptr = &pci_emul_membase64;
618			limit = PCI_EMUL_MEMLIMIT64;
619			mask = PCIM_BAR_MEM_BASE;
620			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
621				 PCIM_BAR_MEM_PREFETCH;
622			break;
623		} else {
624			baseptr = &pci_emul_membase32;
625			limit = PCI_EMUL_MEMLIMIT32;
626			mask = PCIM_BAR_MEM_BASE;
627			lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
628		}
629		break;
630	case PCIBAR_MEM32:
631		baseptr = &pci_emul_membase32;
632		limit = PCI_EMUL_MEMLIMIT32;
633		mask = PCIM_BAR_MEM_BASE;
634		lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
635		break;
636	default:
637		printf("pci_emul_alloc_base: invalid bar type %d\n", type);
638		assert(0);
639	}
640
641	if (baseptr != NULL) {
642		error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
643		if (error != 0)
644			return (error);
645	}
646
647	pdi->pi_bar[idx].type = type;
648	pdi->pi_bar[idx].addr = addr;
649	pdi->pi_bar[idx].size = size;
650
651	/* Initialize the BAR register in config space */
652	bar = (addr & mask) | lobits;
653	pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
654
655	if (type == PCIBAR_MEM64) {
656		assert(idx + 1 <= PCI_BARMAX);
657		pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
658		pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
659	}
660
661	register_bar(pdi, idx);
662
663	return (0);
664}
665
666#define	CAP_START_OFFSET	0x40
667static int
668pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
669{
670	int i, capoff, reallen;
671	uint16_t sts;
672
673	assert(caplen > 0);
674
675	reallen = roundup2(caplen, 4);		/* dword aligned */
676
677	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
678	if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
679		capoff = CAP_START_OFFSET;
680	else
681		capoff = pi->pi_capend + 1;
682
683	/* Check if we have enough space */
684	if (capoff + reallen > PCI_REGMAX + 1)
685		return (-1);
686
687	/* Set the previous capability pointer */
688	if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
689		pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
690		pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
691	} else
692		pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
693
694	/* Copy the capability */
695	for (i = 0; i < caplen; i++)
696		pci_set_cfgdata8(pi, capoff + i, capdata[i]);
697
698	/* Set the next capability pointer */
699	pci_set_cfgdata8(pi, capoff + 1, 0);
700
701	pi->pi_prevcap = capoff;
702	pi->pi_capend = capoff + reallen - 1;
703	return (0);
704}
705
706static struct pci_devemu *
707pci_emul_finddev(char *name)
708{
709	struct pci_devemu **pdpp, *pdp;
710
711	SET_FOREACH(pdpp, pci_devemu_set) {
712		pdp = *pdpp;
713		if (!strcmp(pdp->pe_emu, name)) {
714			return (pdp);
715		}
716	}
717
718	return (NULL);
719}
720
721static int
722pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
723    int func, struct funcinfo *fi)
724{
725	struct pci_devinst *pdi;
726	int err;
727
728	pdi = calloc(1, sizeof(struct pci_devinst));
729
730	pdi->pi_vmctx = ctx;
731	pdi->pi_bus = bus;
732	pdi->pi_slot = slot;
733	pdi->pi_func = func;
734	pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
735	pdi->pi_lintr.pin = 0;
736	pdi->pi_lintr.state = IDLE;
737	pdi->pi_lintr.pirq_pin = 0;
738	pdi->pi_lintr.ioapic_irq = 0;
739	pdi->pi_d = pde;
740	snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
741
742	/* Disable legacy interrupts */
743	pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
744	pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
745
746	pci_set_cfgdata8(pdi, PCIR_COMMAND,
747		    PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
748
749	err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
750	if (err == 0)
751		fi->fi_devi = pdi;
752	else
753		free(pdi);
754
755	return (err);
756}
757
758void
759pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
760{
761	int mmc;
762
763	CTASSERT(sizeof(struct msicap) == 14);
764
765	/* Number of msi messages must be a power of 2 between 1 and 32 */
766	assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
767	mmc = ffs(msgnum) - 1;
768
769	bzero(msicap, sizeof(struct msicap));
770	msicap->capid = PCIY_MSI;
771	msicap->nextptr = nextptr;
772	msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
773}
774
775int
776pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
777{
778	struct msicap msicap;
779
780	pci_populate_msicap(&msicap, msgnum, 0);
781
782	return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
783}
784
785static void
786pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
787		     uint32_t msix_tab_size)
788{
789	CTASSERT(sizeof(struct msixcap) == 12);
790
791	assert(msix_tab_size % 4096 == 0);
792
793	bzero(msixcap, sizeof(struct msixcap));
794	msixcap->capid = PCIY_MSIX;
795
796	/*
797	 * Message Control Register, all fields set to
798	 * zero except for the Table Size.
799	 * Note: Table size N is encoded as N-1
800	 */
801	msixcap->msgctrl = msgnum - 1;
802
803	/*
804	 * MSI-X BAR setup:
805	 * - MSI-X table start at offset 0
806	 * - PBA table starts at a 4K aligned offset after the MSI-X table
807	 */
808	msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
809	msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
810}
811
812static void
813pci_msix_table_init(struct pci_devinst *pi, int table_entries)
814{
815	int i, table_size;
816
817	assert(table_entries > 0);
818	assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
819
820	table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
821	pi->pi_msix.table = calloc(1, table_size);
822
823	/* set mask bit of vector control register */
824	for (i = 0; i < table_entries; i++)
825		pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
826}
827
828int
829pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
830{
831	uint32_t tab_size;
832	struct msixcap msixcap;
833
834	assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
835	assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
836
837	tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
838
839	/* Align table size to nearest 4K */
840	tab_size = roundup2(tab_size, 4096);
841
842	pi->pi_msix.table_bar = barnum;
843	pi->pi_msix.pba_bar   = barnum;
844	pi->pi_msix.table_offset = 0;
845	pi->pi_msix.table_count = msgnum;
846	pi->pi_msix.pba_offset = tab_size;
847	pi->pi_msix.pba_size = PBA_SIZE(msgnum);
848
849	pci_msix_table_init(pi, msgnum);
850
851	pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
852
853	/* allocate memory for MSI-X Table and PBA */
854	pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
855				tab_size + pi->pi_msix.pba_size);
856
857	return (pci_emul_add_capability(pi, (u_char *)&msixcap,
858					sizeof(msixcap)));
859}
860
861void
862msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
863		 int bytes, uint32_t val)
864{
865	uint16_t msgctrl, rwmask;
866	int off, table_bar;
867
868	off = offset - capoff;
869	table_bar = pi->pi_msix.table_bar;
870	/* Message Control Register */
871	if (off == 2 && bytes == 2) {
872		rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
873		msgctrl = pci_get_cfgdata16(pi, offset);
874		msgctrl &= ~rwmask;
875		msgctrl |= val & rwmask;
876		val = msgctrl;
877
878		pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
879		pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
880		pci_lintr_update(pi);
881	}
882
883	CFGWRITE(pi, offset, val, bytes);
884}
885
886void
887msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
888		int bytes, uint32_t val)
889{
890	uint16_t msgctrl, rwmask, msgdata, mme;
891	uint32_t addrlo;
892
893	/*
894	 * If guest is writing to the message control register make sure
895	 * we do not overwrite read-only fields.
896	 */
897	if ((offset - capoff) == 2 && bytes == 2) {
898		rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
899		msgctrl = pci_get_cfgdata16(pi, offset);
900		msgctrl &= ~rwmask;
901		msgctrl |= val & rwmask;
902		val = msgctrl;
903
904		addrlo = pci_get_cfgdata32(pi, capoff + 4);
905		if (msgctrl & PCIM_MSICTRL_64BIT)
906			msgdata = pci_get_cfgdata16(pi, capoff + 12);
907		else
908			msgdata = pci_get_cfgdata16(pi, capoff + 8);
909
910		mme = msgctrl & PCIM_MSICTRL_MME_MASK;
911		pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
912		if (pi->pi_msi.enabled) {
913			pi->pi_msi.addr = addrlo;
914			pi->pi_msi.msg_data = msgdata;
915			pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
916		} else {
917			pi->pi_msi.maxmsgnum = 0;
918		}
919		pci_lintr_update(pi);
920	}
921
922	CFGWRITE(pi, offset, val, bytes);
923}
924
925void
926pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
927		 int bytes, uint32_t val)
928{
929
930	/* XXX don't write to the readonly parts */
931	CFGWRITE(pi, offset, val, bytes);
932}
933
934#define	PCIECAP_VERSION	0x2
935int
936pci_emul_add_pciecap(struct pci_devinst *pi, int type)
937{
938	int err;
939	struct pciecap pciecap;
940
941	CTASSERT(sizeof(struct pciecap) == 60);
942
943	if (type != PCIEM_TYPE_ROOT_PORT)
944		return (-1);
945
946	bzero(&pciecap, sizeof(pciecap));
947
948	pciecap.capid = PCIY_EXPRESS;
949	pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
950	pciecap.link_capabilities = 0x411;	/* gen1, x1 */
951	pciecap.link_status = 0x11;		/* gen1, x1 */
952
953	err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
954	return (err);
955}
956
957/*
958 * This function assumes that 'coff' is in the capabilities region of the
959 * config space.
960 */
961static void
962pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
963{
964	int capid;
965	uint8_t capoff, nextoff;
966
967	/* Do not allow un-aligned writes */
968	if ((offset & (bytes - 1)) != 0)
969		return;
970
971	/* Find the capability that we want to update */
972	capoff = CAP_START_OFFSET;
973	while (1) {
974		nextoff = pci_get_cfgdata8(pi, capoff + 1);
975		if (nextoff == 0)
976			break;
977		if (offset >= capoff && offset < nextoff)
978			break;
979
980		capoff = nextoff;
981	}
982	assert(offset >= capoff);
983
984	/*
985	 * Capability ID and Next Capability Pointer are readonly.
986	 * However, some o/s's do 4-byte writes that include these.
987	 * For this case, trim the write back to 2 bytes and adjust
988	 * the data.
989	 */
990	if (offset == capoff || offset == capoff + 1) {
991		if (offset == capoff && bytes == 4) {
992			bytes = 2;
993			offset += 2;
994			val >>= 16;
995		} else
996			return;
997	}
998
999	capid = pci_get_cfgdata8(pi, capoff);
1000	switch (capid) {
1001	case PCIY_MSI:
1002		msicap_cfgwrite(pi, capoff, offset, bytes, val);
1003		break;
1004	case PCIY_MSIX:
1005		msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1006		break;
1007	case PCIY_EXPRESS:
1008		pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1009		break;
1010	default:
1011		break;
1012	}
1013}
1014
1015static int
1016pci_emul_iscap(struct pci_devinst *pi, int offset)
1017{
1018	uint16_t sts;
1019
1020	sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1021	if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1022		if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1023			return (1);
1024	}
1025	return (0);
1026}
1027
1028static int
1029pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1030			  int size, uint64_t *val, void *arg1, long arg2)
1031{
1032	/*
1033	 * Ignore writes; return 0xff's for reads. The mem read code
1034	 * will take care of truncating to the correct size.
1035	 */
1036	if (dir == MEM_F_READ) {
1037		*val = 0xffffffffffffffff;
1038	}
1039
1040	return (0);
1041}
1042
1043static int
1044pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1045    int bytes, uint64_t *val, void *arg1, long arg2)
1046{
1047	int bus, slot, func, coff, in;
1048
1049	coff = addr & 0xfff;
1050	func = (addr >> 12) & 0x7;
1051	slot = (addr >> 15) & 0x1f;
1052	bus = (addr >> 20) & 0xff;
1053	in = (dir == MEM_F_READ);
1054	if (in)
1055		*val = ~0UL;
1056	pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1057	return (0);
1058}
1059
1060uint64_t
1061pci_ecfg_base(void)
1062{
1063
1064	return (PCI_EMUL_ECFG_BASE);
1065}
1066
1067#define	BUSIO_ROUNDUP		32
1068#define	BUSMEM_ROUNDUP		(1024 * 1024)
1069
1070int
1071init_pci(struct vmctx *ctx)
1072{
1073	struct mem_range mr;
1074	struct pci_devemu *pde;
1075	struct businfo *bi;
1076	struct slotinfo *si;
1077	struct funcinfo *fi;
1078	size_t lowmem;
1079	int bus, slot, func;
1080	int error;
1081
1082	pci_emul_iobase = PCI_EMUL_IOBASE;
1083	pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1084	pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1085
1086	for (bus = 0; bus < MAXBUSES; bus++) {
1087		if ((bi = pci_businfo[bus]) == NULL)
1088			continue;
1089		/*
1090		 * Keep track of the i/o and memory resources allocated to
1091		 * this bus.
1092		 */
1093		bi->iobase = pci_emul_iobase;
1094		bi->membase32 = pci_emul_membase32;
1095		bi->membase64 = pci_emul_membase64;
1096
1097		for (slot = 0; slot < MAXSLOTS; slot++) {
1098			si = &bi->slotinfo[slot];
1099			for (func = 0; func < MAXFUNCS; func++) {
1100				fi = &si->si_funcs[func];
1101				if (fi->fi_name == NULL)
1102					continue;
1103				pde = pci_emul_finddev(fi->fi_name);
1104				assert(pde != NULL);
1105				error = pci_emul_init(ctx, pde, bus, slot,
1106				    func, fi);
1107				if (error)
1108					return (error);
1109			}
1110		}
1111
1112		/*
1113		 * Add some slop to the I/O and memory resources decoded by
1114		 * this bus to give a guest some flexibility if it wants to
1115		 * reprogram the BARs.
1116		 */
1117		pci_emul_iobase += BUSIO_ROUNDUP;
1118		pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1119		bi->iolimit = pci_emul_iobase;
1120
1121		pci_emul_membase32 += BUSMEM_ROUNDUP;
1122		pci_emul_membase32 = roundup2(pci_emul_membase32,
1123		    BUSMEM_ROUNDUP);
1124		bi->memlimit32 = pci_emul_membase32;
1125
1126		pci_emul_membase64 += BUSMEM_ROUNDUP;
1127		pci_emul_membase64 = roundup2(pci_emul_membase64,
1128		    BUSMEM_ROUNDUP);
1129		bi->memlimit64 = pci_emul_membase64;
1130	}
1131
1132	/*
1133	 * PCI backends are initialized before routing INTx interrupts
1134	 * so that LPC devices are able to reserve ISA IRQs before
1135	 * routing PIRQ pins.
1136	 */
1137	for (bus = 0; bus < MAXBUSES; bus++) {
1138		if ((bi = pci_businfo[bus]) == NULL)
1139			continue;
1140
1141		for (slot = 0; slot < MAXSLOTS; slot++) {
1142			si = &bi->slotinfo[slot];
1143			for (func = 0; func < MAXFUNCS; func++) {
1144				fi = &si->si_funcs[func];
1145				if (fi->fi_devi == NULL)
1146					continue;
1147				pci_lintr_route(fi->fi_devi);
1148			}
1149		}
1150	}
1151	lpc_pirq_routed();
1152
1153	/*
1154	 * The guest physical memory map looks like the following:
1155	 * [0,		    lowmem)		guest system memory
1156	 * [lowmem,	    lowmem_limit)	memory hole (may be absent)
1157	 * [lowmem_limit,   0xE0000000)		PCI hole (32-bit BAR allocation)
1158	 * [0xE0000000,	    0xF0000000)		PCI extended config window
1159	 * [0xF0000000,	    4GB)		LAPIC, IOAPIC, HPET, firmware
1160	 * [4GB,	    4GB + highmem)
1161	 */
1162
1163	/*
1164	 * Accesses to memory addresses that are not allocated to system
1165	 * memory or PCI devices return 0xff's.
1166	 */
1167	lowmem = vm_get_lowmem_size(ctx);
1168	bzero(&mr, sizeof(struct mem_range));
1169	mr.name = "PCI hole";
1170	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1171	mr.base = lowmem;
1172	mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1173	mr.handler = pci_emul_fallback_handler;
1174	error = register_mem_fallback(&mr);
1175	assert(error == 0);
1176
1177	/* PCI extended config space */
1178	bzero(&mr, sizeof(struct mem_range));
1179	mr.name = "PCI ECFG";
1180	mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1181	mr.base = PCI_EMUL_ECFG_BASE;
1182	mr.size = PCI_EMUL_ECFG_SIZE;
1183	mr.handler = pci_emul_ecfg_handler;
1184	error = register_mem(&mr);
1185	assert(error == 0);
1186
1187	return (0);
1188}
1189
1190static void
1191pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1192    void *arg)
1193{
1194
1195	dsdt_line("  Package ()");
1196	dsdt_line("  {");
1197	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1198	dsdt_line("    0x%02X,", pin - 1);
1199	dsdt_line("    Zero,");
1200	dsdt_line("    0x%X", ioapic_irq);
1201	dsdt_line("  },");
1202}
1203
1204static void
1205pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1206    void *arg)
1207{
1208	char *name;
1209
1210	name = lpc_pirq_name(pirq_pin);
1211	if (name == NULL)
1212		return;
1213	dsdt_line("  Package ()");
1214	dsdt_line("  {");
1215	dsdt_line("    0x%X,", slot << 16 | 0xffff);
1216	dsdt_line("    0x%02X,", pin - 1);
1217	dsdt_line("    %s,", name);
1218	dsdt_line("    0x00");
1219	dsdt_line("  },");
1220	free(name);
1221}
1222
1223/*
1224 * A bhyve virtual machine has a flat PCI hierarchy with a root port
1225 * corresponding to each PCI bus.
1226 */
1227static void
1228pci_bus_write_dsdt(int bus)
1229{
1230	struct businfo *bi;
1231	struct slotinfo *si;
1232	struct pci_devinst *pi;
1233	int count, func, slot;
1234
1235	/*
1236	 * If there are no devices on this 'bus' then just return.
1237	 */
1238	if ((bi = pci_businfo[bus]) == NULL) {
1239		/*
1240		 * Bus 0 is special because it decodes the I/O ports used
1241		 * for PCI config space access even if there are no devices
1242		 * on it.
1243		 */
1244		if (bus != 0)
1245			return;
1246	}
1247
1248	dsdt_line("  Device (PC%02X)", bus);
1249	dsdt_line("  {");
1250	dsdt_line("    Name (_HID, EisaId (\"PNP0A03\"))");
1251	dsdt_line("    Name (_ADR, Zero)");
1252
1253	dsdt_line("    Method (_BBN, 0, NotSerialized)");
1254	dsdt_line("    {");
1255	dsdt_line("        Return (0x%08X)", bus);
1256	dsdt_line("    }");
1257	dsdt_line("    Name (_CRS, ResourceTemplate ()");
1258	dsdt_line("    {");
1259	dsdt_line("      WordBusNumber (ResourceProducer, MinFixed, "
1260	    "MaxFixed, PosDecode,");
1261	dsdt_line("        0x0000,             // Granularity");
1262	dsdt_line("        0x%04X,             // Range Minimum", bus);
1263	dsdt_line("        0x%04X,             // Range Maximum", bus);
1264	dsdt_line("        0x0000,             // Translation Offset");
1265	dsdt_line("        0x0001,             // Length");
1266	dsdt_line("        ,, )");
1267
1268	if (bus == 0) {
1269		dsdt_indent(3);
1270		dsdt_fixed_ioport(0xCF8, 8);
1271		dsdt_unindent(3);
1272
1273		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1274		    "PosDecode, EntireRange,");
1275		dsdt_line("        0x0000,             // Granularity");
1276		dsdt_line("        0x0000,             // Range Minimum");
1277		dsdt_line("        0x0CF7,             // Range Maximum");
1278		dsdt_line("        0x0000,             // Translation Offset");
1279		dsdt_line("        0x0CF8,             // Length");
1280		dsdt_line("        ,, , TypeStatic)");
1281
1282		dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1283		    "PosDecode, EntireRange,");
1284		dsdt_line("        0x0000,             // Granularity");
1285		dsdt_line("        0x0D00,             // Range Minimum");
1286		dsdt_line("        0x%04X,             // Range Maximum",
1287		    PCI_EMUL_IOBASE - 1);
1288		dsdt_line("        0x0000,             // Translation Offset");
1289		dsdt_line("        0x%04X,             // Length",
1290		    PCI_EMUL_IOBASE - 0x0D00);
1291		dsdt_line("        ,, , TypeStatic)");
1292
1293		if (bi == NULL) {
1294			dsdt_line("    })");
1295			goto done;
1296		}
1297	}
1298	assert(bi != NULL);
1299
1300	/* i/o window */
1301	dsdt_line("      WordIO (ResourceProducer, MinFixed, MaxFixed, "
1302	    "PosDecode, EntireRange,");
1303	dsdt_line("        0x0000,             // Granularity");
1304	dsdt_line("        0x%04X,             // Range Minimum", bi->iobase);
1305	dsdt_line("        0x%04X,             // Range Maximum",
1306	    bi->iolimit - 1);
1307	dsdt_line("        0x0000,             // Translation Offset");
1308	dsdt_line("        0x%04X,             // Length",
1309	    bi->iolimit - bi->iobase);
1310	dsdt_line("        ,, , TypeStatic)");
1311
1312	/* mmio window (32-bit) */
1313	dsdt_line("      DWordMemory (ResourceProducer, PosDecode, "
1314	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1315	dsdt_line("        0x00000000,         // Granularity");
1316	dsdt_line("        0x%08X,         // Range Minimum\n", bi->membase32);
1317	dsdt_line("        0x%08X,         // Range Maximum\n",
1318	    bi->memlimit32 - 1);
1319	dsdt_line("        0x00000000,         // Translation Offset");
1320	dsdt_line("        0x%08X,         // Length\n",
1321	    bi->memlimit32 - bi->membase32);
1322	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1323
1324	/* mmio window (64-bit) */
1325	dsdt_line("      QWordMemory (ResourceProducer, PosDecode, "
1326	    "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1327	dsdt_line("        0x0000000000000000, // Granularity");
1328	dsdt_line("        0x%016lX, // Range Minimum\n", bi->membase64);
1329	dsdt_line("        0x%016lX, // Range Maximum\n",
1330	    bi->memlimit64 - 1);
1331	dsdt_line("        0x0000000000000000, // Translation Offset");
1332	dsdt_line("        0x%016lX, // Length\n",
1333	    bi->memlimit64 - bi->membase64);
1334	dsdt_line("        ,, , AddressRangeMemory, TypeStatic)");
1335	dsdt_line("    })");
1336
1337	count = pci_count_lintr(bus);
1338	if (count != 0) {
1339		dsdt_indent(2);
1340		dsdt_line("Name (PPRT, Package ()");
1341		dsdt_line("{");
1342		pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1343 		dsdt_line("})");
1344		dsdt_line("Name (APRT, Package ()");
1345		dsdt_line("{");
1346		pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1347 		dsdt_line("})");
1348		dsdt_line("Method (_PRT, 0, NotSerialized)");
1349		dsdt_line("{");
1350		dsdt_line("  If (PICM)");
1351		dsdt_line("  {");
1352		dsdt_line("    Return (APRT)");
1353		dsdt_line("  }");
1354		dsdt_line("  Else");
1355		dsdt_line("  {");
1356		dsdt_line("    Return (PPRT)");
1357		dsdt_line("  }");
1358		dsdt_line("}");
1359		dsdt_unindent(2);
1360	}
1361
1362	dsdt_indent(2);
1363	for (slot = 0; slot < MAXSLOTS; slot++) {
1364		si = &bi->slotinfo[slot];
1365		for (func = 0; func < MAXFUNCS; func++) {
1366			pi = si->si_funcs[func].fi_devi;
1367			if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1368				pi->pi_d->pe_write_dsdt(pi);
1369		}
1370	}
1371	dsdt_unindent(2);
1372done:
1373	dsdt_line("  }");
1374}
1375
1376void
1377pci_write_dsdt(void)
1378{
1379	int bus;
1380
1381	dsdt_indent(1);
1382	dsdt_line("Name (PICM, 0x00)");
1383	dsdt_line("Method (_PIC, 1, NotSerialized)");
1384	dsdt_line("{");
1385	dsdt_line("  Store (Arg0, PICM)");
1386	dsdt_line("}");
1387	dsdt_line("");
1388	dsdt_line("Scope (_SB)");
1389	dsdt_line("{");
1390	for (bus = 0; bus < MAXBUSES; bus++)
1391		pci_bus_write_dsdt(bus);
1392	dsdt_line("}");
1393	dsdt_unindent(1);
1394}
1395
1396int
1397pci_bus_configured(int bus)
1398{
1399	assert(bus >= 0 && bus < MAXBUSES);
1400	return (pci_businfo[bus] != NULL);
1401}
1402
1403int
1404pci_msi_enabled(struct pci_devinst *pi)
1405{
1406	return (pi->pi_msi.enabled);
1407}
1408
1409int
1410pci_msi_maxmsgnum(struct pci_devinst *pi)
1411{
1412	if (pi->pi_msi.enabled)
1413		return (pi->pi_msi.maxmsgnum);
1414	else
1415		return (0);
1416}
1417
1418int
1419pci_msix_enabled(struct pci_devinst *pi)
1420{
1421
1422	return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1423}
1424
1425void
1426pci_generate_msix(struct pci_devinst *pi, int index)
1427{
1428	struct msix_table_entry *mte;
1429
1430	if (!pci_msix_enabled(pi))
1431		return;
1432
1433	if (pi->pi_msix.function_mask)
1434		return;
1435
1436	if (index >= pi->pi_msix.table_count)
1437		return;
1438
1439	mte = &pi->pi_msix.table[index];
1440	if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1441		/* XXX Set PBA bit if interrupt is disabled */
1442		vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1443	}
1444}
1445
1446void
1447pci_generate_msi(struct pci_devinst *pi, int index)
1448{
1449
1450	if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1451		vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1452			     pi->pi_msi.msg_data + index);
1453	}
1454}
1455
1456static bool
1457pci_lintr_permitted(struct pci_devinst *pi)
1458{
1459	uint16_t cmd;
1460
1461	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1462	return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1463		(cmd & PCIM_CMD_INTxDIS)));
1464}
1465
1466void
1467pci_lintr_request(struct pci_devinst *pi)
1468{
1469	struct businfo *bi;
1470	struct slotinfo *si;
1471	int bestpin, bestcount, pin;
1472
1473	bi = pci_businfo[pi->pi_bus];
1474	assert(bi != NULL);
1475
1476	/*
1477	 * Just allocate a pin from our slot.  The pin will be
1478	 * assigned IRQs later when interrupts are routed.
1479	 */
1480	si = &bi->slotinfo[pi->pi_slot];
1481	bestpin = 0;
1482	bestcount = si->si_intpins[0].ii_count;
1483	for (pin = 1; pin < 4; pin++) {
1484		if (si->si_intpins[pin].ii_count < bestcount) {
1485			bestpin = pin;
1486			bestcount = si->si_intpins[pin].ii_count;
1487		}
1488	}
1489
1490	si->si_intpins[bestpin].ii_count++;
1491	pi->pi_lintr.pin = bestpin + 1;
1492	pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1493}
1494
1495static void
1496pci_lintr_route(struct pci_devinst *pi)
1497{
1498	struct businfo *bi;
1499	struct intxinfo *ii;
1500
1501	if (pi->pi_lintr.pin == 0)
1502		return;
1503
1504	bi = pci_businfo[pi->pi_bus];
1505	assert(bi != NULL);
1506	ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1507
1508	/*
1509	 * Attempt to allocate an I/O APIC pin for this intpin if one
1510	 * is not yet assigned.
1511	 */
1512	if (ii->ii_ioapic_irq == 0)
1513		ii->ii_ioapic_irq = ioapic_pci_alloc_irq();
1514	assert(ii->ii_ioapic_irq > 0);
1515
1516	/*
1517	 * Attempt to allocate a PIRQ pin for this intpin if one is
1518	 * not yet assigned.
1519	 */
1520	if (ii->ii_pirq_pin == 0)
1521		ii->ii_pirq_pin = pirq_alloc_pin(pi->pi_vmctx);
1522	assert(ii->ii_pirq_pin > 0);
1523
1524	pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1525	pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1526	pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1527}
1528
1529void
1530pci_lintr_assert(struct pci_devinst *pi)
1531{
1532
1533	assert(pi->pi_lintr.pin > 0);
1534
1535	pthread_mutex_lock(&pi->pi_lintr.lock);
1536	if (pi->pi_lintr.state == IDLE) {
1537		if (pci_lintr_permitted(pi)) {
1538			pi->pi_lintr.state = ASSERTED;
1539			pci_irq_assert(pi);
1540		} else
1541			pi->pi_lintr.state = PENDING;
1542	}
1543	pthread_mutex_unlock(&pi->pi_lintr.lock);
1544}
1545
1546void
1547pci_lintr_deassert(struct pci_devinst *pi)
1548{
1549
1550	assert(pi->pi_lintr.pin > 0);
1551
1552	pthread_mutex_lock(&pi->pi_lintr.lock);
1553	if (pi->pi_lintr.state == ASSERTED) {
1554		pi->pi_lintr.state = IDLE;
1555		pci_irq_deassert(pi);
1556	} else if (pi->pi_lintr.state == PENDING)
1557		pi->pi_lintr.state = IDLE;
1558	pthread_mutex_unlock(&pi->pi_lintr.lock);
1559}
1560
1561static void
1562pci_lintr_update(struct pci_devinst *pi)
1563{
1564
1565	pthread_mutex_lock(&pi->pi_lintr.lock);
1566	if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1567		pci_irq_deassert(pi);
1568		pi->pi_lintr.state = PENDING;
1569	} else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1570		pi->pi_lintr.state = ASSERTED;
1571		pci_irq_assert(pi);
1572	}
1573	pthread_mutex_unlock(&pi->pi_lintr.lock);
1574}
1575
1576int
1577pci_count_lintr(int bus)
1578{
1579	int count, slot, pin;
1580	struct slotinfo *slotinfo;
1581
1582	count = 0;
1583	if (pci_businfo[bus] != NULL) {
1584		for (slot = 0; slot < MAXSLOTS; slot++) {
1585			slotinfo = &pci_businfo[bus]->slotinfo[slot];
1586			for (pin = 0; pin < 4; pin++) {
1587				if (slotinfo->si_intpins[pin].ii_count != 0)
1588					count++;
1589			}
1590		}
1591	}
1592	return (count);
1593}
1594
1595void
1596pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1597{
1598	struct businfo *bi;
1599	struct slotinfo *si;
1600	struct intxinfo *ii;
1601	int slot, pin;
1602
1603	if ((bi = pci_businfo[bus]) == NULL)
1604		return;
1605
1606	for (slot = 0; slot < MAXSLOTS; slot++) {
1607		si = &bi->slotinfo[slot];
1608		for (pin = 0; pin < 4; pin++) {
1609			ii = &si->si_intpins[pin];
1610			if (ii->ii_count != 0)
1611				cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1612				    ii->ii_ioapic_irq, arg);
1613		}
1614	}
1615}
1616
1617/*
1618 * Return 1 if the emulated device in 'slot' is a multi-function device.
1619 * Return 0 otherwise.
1620 */
1621static int
1622pci_emul_is_mfdev(int bus, int slot)
1623{
1624	struct businfo *bi;
1625	struct slotinfo *si;
1626	int f, numfuncs;
1627
1628	numfuncs = 0;
1629	if ((bi = pci_businfo[bus]) != NULL) {
1630		si = &bi->slotinfo[slot];
1631		for (f = 0; f < MAXFUNCS; f++) {
1632			if (si->si_funcs[f].fi_devi != NULL) {
1633				numfuncs++;
1634			}
1635		}
1636	}
1637	return (numfuncs > 1);
1638}
1639
1640/*
1641 * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1642 * whether or not is a multi-function being emulated in the pci 'slot'.
1643 */
1644static void
1645pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1646{
1647	int mfdev;
1648
1649	if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1650		mfdev = pci_emul_is_mfdev(bus, slot);
1651		switch (bytes) {
1652		case 1:
1653		case 2:
1654			*rv &= ~PCIM_MFDEV;
1655			if (mfdev) {
1656				*rv |= PCIM_MFDEV;
1657			}
1658			break;
1659		case 4:
1660			*rv &= ~(PCIM_MFDEV << 16);
1661			if (mfdev) {
1662				*rv |= (PCIM_MFDEV << 16);
1663			}
1664			break;
1665		}
1666	}
1667}
1668
1669static void
1670pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1671{
1672	int i, rshift;
1673	uint32_t cmd, cmd2, changed, old, readonly;
1674
1675	cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* stash old value */
1676
1677	/*
1678	 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1679	 *
1680	 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1681	 * 'write 1 to clear'. However these bits are not set to '1' by
1682	 * any device emulation so it is simpler to treat them as readonly.
1683	 */
1684	rshift = (coff & 0x3) * 8;
1685	readonly = 0xFFFFF880 >> rshift;
1686
1687	old = CFGREAD(pi, coff, bytes);
1688	new &= ~readonly;
1689	new |= (old & readonly);
1690	CFGWRITE(pi, coff, new, bytes);			/* update config */
1691
1692	cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND);	/* get updated value */
1693	changed = cmd ^ cmd2;
1694
1695	/*
1696	 * If the MMIO or I/O address space decoding has changed then
1697	 * register/unregister all BARs that decode that address space.
1698	 */
1699	for (i = 0; i <= PCI_BARMAX; i++) {
1700		switch (pi->pi_bar[i].type) {
1701			case PCIBAR_NONE:
1702			case PCIBAR_MEMHI64:
1703				break;
1704			case PCIBAR_IO:
1705				/* I/O address space decoding changed? */
1706				if (changed & PCIM_CMD_PORTEN) {
1707					if (porten(pi))
1708						register_bar(pi, i);
1709					else
1710						unregister_bar(pi, i);
1711				}
1712				break;
1713			case PCIBAR_MEM32:
1714			case PCIBAR_MEM64:
1715				/* MMIO address space decoding changed? */
1716				if (changed & PCIM_CMD_MEMEN) {
1717					if (memen(pi))
1718						register_bar(pi, i);
1719					else
1720						unregister_bar(pi, i);
1721				}
1722				break;
1723			default:
1724				assert(0);
1725		}
1726	}
1727
1728	/*
1729	 * If INTx has been unmasked and is pending, assert the
1730	 * interrupt.
1731	 */
1732	pci_lintr_update(pi);
1733}
1734
1735static void
1736pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1737    int coff, int bytes, uint32_t *eax)
1738{
1739	struct businfo *bi;
1740	struct slotinfo *si;
1741	struct pci_devinst *pi;
1742	struct pci_devemu *pe;
1743	int idx, needcfg;
1744	uint64_t addr, bar, mask;
1745
1746	if ((bi = pci_businfo[bus]) != NULL) {
1747		si = &bi->slotinfo[slot];
1748		pi = si->si_funcs[func].fi_devi;
1749	} else
1750		pi = NULL;
1751
1752	/*
1753	 * Just return if there is no device at this slot:func or if the
1754	 * the guest is doing an un-aligned access.
1755	 */
1756	if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1757	    (coff & (bytes - 1)) != 0) {
1758		if (in)
1759			*eax = 0xffffffff;
1760		return;
1761	}
1762
1763	/*
1764	 * Ignore all writes beyond the standard config space and return all
1765	 * ones on reads.
1766	 */
1767	if (coff >= PCI_REGMAX + 1) {
1768		if (in) {
1769			*eax = 0xffffffff;
1770			/*
1771			 * Extended capabilities begin at offset 256 in config
1772			 * space. Absence of extended capabilities is signaled
1773			 * with all 0s in the extended capability header at
1774			 * offset 256.
1775			 */
1776			if (coff <= PCI_REGMAX + 4)
1777				*eax = 0x00000000;
1778		}
1779		return;
1780	}
1781
1782	pe = pi->pi_d;
1783
1784	/*
1785	 * Config read
1786	 */
1787	if (in) {
1788		/* Let the device emulation override the default handler */
1789		if (pe->pe_cfgread != NULL) {
1790			needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1791			    eax);
1792		} else {
1793			needcfg = 1;
1794		}
1795
1796		if (needcfg)
1797			*eax = CFGREAD(pi, coff, bytes);
1798
1799		pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1800	} else {
1801		/* Let the device emulation override the default handler */
1802		if (pe->pe_cfgwrite != NULL &&
1803		    (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1804			return;
1805
1806		/*
1807		 * Special handling for write to BAR registers
1808		 */
1809		if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1810			/*
1811			 * Ignore writes to BAR registers that are not
1812			 * 4-byte aligned.
1813			 */
1814			if (bytes != 4 || (coff & 0x3) != 0)
1815				return;
1816			idx = (coff - PCIR_BAR(0)) / 4;
1817			mask = ~(pi->pi_bar[idx].size - 1);
1818			switch (pi->pi_bar[idx].type) {
1819			case PCIBAR_NONE:
1820				pi->pi_bar[idx].addr = bar = 0;
1821				break;
1822			case PCIBAR_IO:
1823				addr = *eax & mask;
1824				addr &= 0xffff;
1825				bar = addr | PCIM_BAR_IO_SPACE;
1826				/*
1827				 * Register the new BAR value for interception
1828				 */
1829				if (addr != pi->pi_bar[idx].addr) {
1830					update_bar_address(pi, addr, idx,
1831							   PCIBAR_IO);
1832				}
1833				break;
1834			case PCIBAR_MEM32:
1835				addr = bar = *eax & mask;
1836				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1837				if (addr != pi->pi_bar[idx].addr) {
1838					update_bar_address(pi, addr, idx,
1839							   PCIBAR_MEM32);
1840				}
1841				break;
1842			case PCIBAR_MEM64:
1843				addr = bar = *eax & mask;
1844				bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1845				       PCIM_BAR_MEM_PREFETCH;
1846				if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1847					update_bar_address(pi, addr, idx,
1848							   PCIBAR_MEM64);
1849				}
1850				break;
1851			case PCIBAR_MEMHI64:
1852				mask = ~(pi->pi_bar[idx - 1].size - 1);
1853				addr = ((uint64_t)*eax << 32) & mask;
1854				bar = addr >> 32;
1855				if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1856					update_bar_address(pi, addr, idx - 1,
1857							   PCIBAR_MEMHI64);
1858				}
1859				break;
1860			default:
1861				assert(0);
1862			}
1863			pci_set_cfgdata32(pi, coff, bar);
1864
1865		} else if (pci_emul_iscap(pi, coff)) {
1866			pci_emul_capwrite(pi, coff, bytes, *eax);
1867		} else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1868			pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1869		} else {
1870			CFGWRITE(pi, coff, *eax, bytes);
1871		}
1872	}
1873}
1874
1875static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1876
1877static int
1878pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1879		 uint32_t *eax, void *arg)
1880{
1881	uint32_t x;
1882
1883	if (bytes != 4) {
1884		if (in)
1885			*eax = (bytes == 2) ? 0xffff : 0xff;
1886		return (0);
1887	}
1888
1889	if (in) {
1890		x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1891		if (cfgenable)
1892			x |= CONF1_ENABLE;
1893		*eax = x;
1894	} else {
1895		x = *eax;
1896		cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1897		cfgoff = x & PCI_REGMAX;
1898		cfgfunc = (x >> 8) & PCI_FUNCMAX;
1899		cfgslot = (x >> 11) & PCI_SLOTMAX;
1900		cfgbus = (x >> 16) & PCI_BUSMAX;
1901	}
1902
1903	return (0);
1904}
1905INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1906
1907static int
1908pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1909		 uint32_t *eax, void *arg)
1910{
1911	int coff;
1912
1913	assert(bytes == 1 || bytes == 2 || bytes == 4);
1914
1915	coff = cfgoff + (port - CONF1_DATA_PORT);
1916	if (cfgenable) {
1917		pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1918		    eax);
1919	} else {
1920		/* Ignore accesses to cfgdata if not enabled by cfgaddr */
1921		if (in)
1922			*eax = 0xffffffff;
1923	}
1924	return (0);
1925}
1926
1927INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1928INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1929INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1930INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1931
1932#define PCI_EMUL_TEST
1933#ifdef PCI_EMUL_TEST
1934/*
1935 * Define a dummy test device
1936 */
1937#define DIOSZ	8
1938#define DMEMSZ	4096
1939struct pci_emul_dsoftc {
1940	uint8_t   ioregs[DIOSZ];
1941	uint8_t	  memregs[2][DMEMSZ];
1942};
1943
1944#define	PCI_EMUL_MSI_MSGS	 4
1945#define	PCI_EMUL_MSIX_MSGS	16
1946
1947static int
1948pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1949{
1950	int error;
1951	struct pci_emul_dsoftc *sc;
1952
1953	sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1954
1955	pi->pi_arg = sc;
1956
1957	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1958	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1959	pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1960
1961	error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1962	assert(error == 0);
1963
1964	error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1965	assert(error == 0);
1966
1967	error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1968	assert(error == 0);
1969
1970	error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1971	assert(error == 0);
1972
1973	return (0);
1974}
1975
1976static void
1977pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1978	      uint64_t offset, int size, uint64_t value)
1979{
1980	int i;
1981	struct pci_emul_dsoftc *sc = pi->pi_arg;
1982
1983	if (baridx == 0) {
1984		if (offset + size > DIOSZ) {
1985			printf("diow: iow too large, offset %ld size %d\n",
1986			       offset, size);
1987			return;
1988		}
1989
1990		if (size == 1) {
1991			sc->ioregs[offset] = value & 0xff;
1992		} else if (size == 2) {
1993			*(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1994		} else if (size == 4) {
1995			*(uint32_t *)&sc->ioregs[offset] = value;
1996		} else {
1997			printf("diow: iow unknown size %d\n", size);
1998		}
1999
2000		/*
2001		 * Special magic value to generate an interrupt
2002		 */
2003		if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2004			pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2005
2006		if (value == 0xabcdef) {
2007			for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2008				pci_generate_msi(pi, i);
2009		}
2010	}
2011
2012	if (baridx == 1 || baridx == 2) {
2013		if (offset + size > DMEMSZ) {
2014			printf("diow: memw too large, offset %ld size %d\n",
2015			       offset, size);
2016			return;
2017		}
2018
2019		i = baridx - 1;		/* 'memregs' index */
2020
2021		if (size == 1) {
2022			sc->memregs[i][offset] = value;
2023		} else if (size == 2) {
2024			*(uint16_t *)&sc->memregs[i][offset] = value;
2025		} else if (size == 4) {
2026			*(uint32_t *)&sc->memregs[i][offset] = value;
2027		} else if (size == 8) {
2028			*(uint64_t *)&sc->memregs[i][offset] = value;
2029		} else {
2030			printf("diow: memw unknown size %d\n", size);
2031		}
2032
2033		/*
2034		 * magic interrupt ??
2035		 */
2036	}
2037
2038	if (baridx > 2) {
2039		printf("diow: unknown bar idx %d\n", baridx);
2040	}
2041}
2042
2043static uint64_t
2044pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2045	      uint64_t offset, int size)
2046{
2047	struct pci_emul_dsoftc *sc = pi->pi_arg;
2048	uint32_t value;
2049	int i;
2050
2051	if (baridx == 0) {
2052		if (offset + size > DIOSZ) {
2053			printf("dior: ior too large, offset %ld size %d\n",
2054			       offset, size);
2055			return (0);
2056		}
2057
2058		if (size == 1) {
2059			value = sc->ioregs[offset];
2060		} else if (size == 2) {
2061			value = *(uint16_t *) &sc->ioregs[offset];
2062		} else if (size == 4) {
2063			value = *(uint32_t *) &sc->ioregs[offset];
2064		} else {
2065			printf("dior: ior unknown size %d\n", size);
2066		}
2067	}
2068
2069	if (baridx == 1 || baridx == 2) {
2070		if (offset + size > DMEMSZ) {
2071			printf("dior: memr too large, offset %ld size %d\n",
2072			       offset, size);
2073			return (0);
2074		}
2075
2076		i = baridx - 1;		/* 'memregs' index */
2077
2078		if (size == 1) {
2079			value = sc->memregs[i][offset];
2080		} else if (size == 2) {
2081			value = *(uint16_t *) &sc->memregs[i][offset];
2082		} else if (size == 4) {
2083			value = *(uint32_t *) &sc->memregs[i][offset];
2084		} else if (size == 8) {
2085			value = *(uint64_t *) &sc->memregs[i][offset];
2086		} else {
2087			printf("dior: ior unknown size %d\n", size);
2088		}
2089	}
2090
2091
2092	if (baridx > 2) {
2093		printf("dior: unknown bar idx %d\n", baridx);
2094		return (0);
2095	}
2096
2097	return (value);
2098}
2099
2100struct pci_devemu pci_dummy = {
2101	.pe_emu = "dummy",
2102	.pe_init = pci_emul_dinit,
2103	.pe_barwrite = pci_emul_diow,
2104	.pe_barread = pci_emul_dior
2105};
2106PCI_EMUL_SET(pci_dummy);
2107
2108#endif /* PCI_EMUL_TEST */
2109