2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 NetApp, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/linker_set.h>
47 #include <machine/vmm.h>
59 #define CONF1_ADDR_PORT 0x0cf8
60 #define CONF1_DATA_PORT 0x0cfc
62 #define CONF1_ENABLE 0x80000000ul
64 #define MAXBUSES (PCI_BUSMAX + 1)
65 #define MAXSLOTS (PCI_SLOTMAX + 1)
66 #define MAXFUNCS (PCI_FUNCMAX + 1)
71 struct pci_devinst *fi_devi;
81 struct intxinfo si_intpins[4];
82 struct funcinfo si_funcs[MAXFUNCS];
86 uint16_t iobase, iolimit; /* I/O window */
87 uint32_t membase32, memlimit32; /* mmio window below 4GB */
88 uint64_t membase64, memlimit64; /* mmio window above 4GB */
89 struct slotinfo slotinfo[MAXSLOTS];
92 static struct businfo *pci_businfo[MAXBUSES];
94 SET_DECLARE(pci_devemu_set, struct pci_devemu);
96 static uint64_t pci_emul_iobase;
97 static uint64_t pci_emul_membase32;
98 static uint64_t pci_emul_membase64;
100 #define PCI_EMUL_IOBASE 0x2000
101 #define PCI_EMUL_IOLIMIT 0x10000
103 #define PCI_EMUL_ECFG_BASE 0xE0000000 /* 3.5GB */
104 #define PCI_EMUL_ECFG_SIZE (MAXBUSES * 1024 * 1024) /* 1MB per bus */
105 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
107 #define PCI_EMUL_MEMLIMIT32 PCI_EMUL_ECFG_BASE
109 #define PCI_EMUL_MEMBASE64 0xD000000000UL
110 #define PCI_EMUL_MEMLIMIT64 0xFD00000000UL
112 static struct pci_devemu *pci_emul_finddev(char *name);
113 static void pci_lintr_route(struct pci_devinst *pi);
114 static void pci_lintr_update(struct pci_devinst *pi);
115 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
116 int func, int coff, int bytes, uint32_t *val);
119 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
123 pci_set_cfgdata8(pi, coff, val);
125 pci_set_cfgdata16(pi, coff, val);
127 pci_set_cfgdata32(pi, coff, val);
130 static __inline uint32_t
131 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
135 return (pci_get_cfgdata8(pi, coff));
137 return (pci_get_cfgdata16(pi, coff));
139 return (pci_get_cfgdata32(pi, coff));
147 * Slot options are in the form:
149 * <bus>:<slot>:<func>,<emul>[,<config>]
150 * <slot>[:<func>],<emul>[,<config>]
154 * emul is a string describing the type of PCI device e.g. virtio-net
155 * config is an optional string, depending on the device, that can be
156 * used for configuration.
162 pci_parse_slot_usage(char *aopt)
165 fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
169 pci_parse_slot(char *opt)
173 char *emul, *config, *str, *cp;
174 int error, bnum, snum, fnum;
179 emul = config = NULL;
180 if ((cp = strchr(str, ',')) != NULL) {
183 if ((cp = strchr(emul, ',')) != NULL) {
188 pci_parse_slot_usage(opt);
192 /* <bus>:<slot>:<func> */
193 if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
196 if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
199 if (sscanf(str, "%d", &snum) != 1) {
205 if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
206 fnum < 0 || fnum >= MAXFUNCS) {
207 pci_parse_slot_usage(opt);
211 if (pci_businfo[bnum] == NULL)
212 pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
214 bi = pci_businfo[bnum];
215 si = &bi->slotinfo[snum];
217 if (si->si_funcs[fnum].fi_name != NULL) {
218 fprintf(stderr, "pci slot %d:%d already occupied!\n",
223 if (pci_emul_finddev(emul) == NULL) {
224 fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
230 si->si_funcs[fnum].fi_name = emul;
231 si->si_funcs[fnum].fi_param = config;
240 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
243 if (offset < pi->pi_msix.pba_offset)
246 if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
254 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
257 int msix_entry_offset;
261 /* support only 4 or 8 byte writes */
262 if (size != 4 && size != 8)
266 * Return if table index is beyond what device supports
268 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
269 if (tab_index >= pi->pi_msix.table_count)
272 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
274 /* support only aligned writes */
275 if ((msix_entry_offset % size) != 0)
278 dest = (char *)(pi->pi_msix.table + tab_index);
279 dest += msix_entry_offset;
282 *((uint32_t *)dest) = value;
284 *((uint64_t *)dest) = value;
290 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
293 int msix_entry_offset;
295 uint64_t retval = ~0;
298 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
299 * table but we also allow 1 byte access to accommodate reads from
302 if (size != 1 && size != 4 && size != 8)
305 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
307 /* support only aligned reads */
308 if ((msix_entry_offset % size) != 0) {
312 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
314 if (tab_index < pi->pi_msix.table_count) {
315 /* valid MSI-X Table access */
316 dest = (char *)(pi->pi_msix.table + tab_index);
317 dest += msix_entry_offset;
320 retval = *((uint8_t *)dest);
322 retval = *((uint32_t *)dest);
324 retval = *((uint64_t *)dest);
325 } else if (pci_valid_pba_offset(pi, offset)) {
326 /* return 0 for PBA access */
334 pci_msix_table_bar(struct pci_devinst *pi)
337 if (pi->pi_msix.table != NULL)
338 return (pi->pi_msix.table_bar);
344 pci_msix_pba_bar(struct pci_devinst *pi)
347 if (pi->pi_msix.table != NULL)
348 return (pi->pi_msix.pba_bar);
354 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
355 uint32_t *eax, void *arg)
357 struct pci_devinst *pdi = arg;
358 struct pci_devemu *pe = pdi->pi_d;
362 for (i = 0; i <= PCI_BARMAX; i++) {
363 if (pdi->pi_bar[i].type == PCIBAR_IO &&
364 port >= pdi->pi_bar[i].addr &&
365 port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
366 offset = port - pdi->pi_bar[i].addr;
368 *eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
371 (*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
380 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
381 int size, uint64_t *val, void *arg1, long arg2)
383 struct pci_devinst *pdi = arg1;
384 struct pci_devemu *pe = pdi->pi_d;
386 int bidx = (int) arg2;
388 assert(bidx <= PCI_BARMAX);
389 assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
390 pdi->pi_bar[bidx].type == PCIBAR_MEM64);
391 assert(addr >= pdi->pi_bar[bidx].addr &&
392 addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
394 offset = addr - pdi->pi_bar[bidx].addr;
396 if (dir == MEM_F_WRITE) {
398 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
399 4, *val & 0xffffffff);
400 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
403 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
408 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
410 *val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
411 offset + 4, 4) << 32;
413 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
423 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
428 assert((size & (size - 1)) == 0); /* must be a power of 2 */
430 base = roundup2(*baseptr, size);
432 if (base + size <= limit) {
434 *baseptr = base + size;
441 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
445 return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
449 * Register (or unregister) the MMIO or I/O region associated with the BAR
450 * register 'idx' of an emulated pci device.
453 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
456 struct inout_port iop;
459 switch (pi->pi_bar[idx].type) {
461 bzero(&iop, sizeof(struct inout_port));
462 iop.name = pi->pi_name;
463 iop.port = pi->pi_bar[idx].addr;
464 iop.size = pi->pi_bar[idx].size;
466 iop.flags = IOPORT_F_INOUT;
467 iop.handler = pci_emul_io_handler;
469 error = register_inout(&iop);
471 error = unregister_inout(&iop);
475 bzero(&mr, sizeof(struct mem_range));
476 mr.name = pi->pi_name;
477 mr.base = pi->pi_bar[idx].addr;
478 mr.size = pi->pi_bar[idx].size;
481 mr.handler = pci_emul_mem_handler;
484 error = register_mem(&mr);
486 error = unregister_mem(&mr);
496 unregister_bar(struct pci_devinst *pi, int idx)
499 modify_bar_registration(pi, idx, 0);
503 register_bar(struct pci_devinst *pi, int idx)
506 modify_bar_registration(pi, idx, 1);
509 /* Are we decoding i/o port accesses for the emulated pci device? */
511 porten(struct pci_devinst *pi)
515 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
517 return (cmd & PCIM_CMD_PORTEN);
520 /* Are we decoding memory accesses for the emulated pci device? */
522 memen(struct pci_devinst *pi)
526 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
528 return (cmd & PCIM_CMD_MEMEN);
532 * Update the MMIO or I/O address that is decoded by the BAR register.
534 * If the pci device has enabled the address space decoding then intercept
535 * the address range decoded by the BAR register.
538 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
542 if (pi->pi_bar[idx].type == PCIBAR_IO)
548 unregister_bar(pi, idx);
553 pi->pi_bar[idx].addr = addr;
556 pi->pi_bar[idx].addr &= ~0xffffffffUL;
557 pi->pi_bar[idx].addr |= addr;
560 pi->pi_bar[idx].addr &= 0xffffffff;
561 pi->pi_bar[idx].addr |= addr;
568 register_bar(pi, idx);
572 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
573 enum pcibar_type type, uint64_t size)
576 uint64_t *baseptr, limit, addr, mask, lobits, bar;
578 assert(idx >= 0 && idx <= PCI_BARMAX);
580 if ((size & (size - 1)) != 0)
581 size = 1UL << flsl(size); /* round up to a power of 2 */
583 /* Enforce minimum BAR sizes required by the PCI standard */
584 if (type == PCIBAR_IO) {
595 addr = mask = lobits = 0;
598 baseptr = &pci_emul_iobase;
599 limit = PCI_EMUL_IOLIMIT;
600 mask = PCIM_BAR_IO_BASE;
601 lobits = PCIM_BAR_IO_SPACE;
606 * Some drivers do not work well if the 64-bit BAR is allocated
607 * above 4GB. Allow for this by allocating small requests under
608 * 4GB unless then allocation size is larger than some arbitrary
609 * number (32MB currently).
611 if (size > 32 * 1024 * 1024) {
613 * XXX special case for device requiring peer-peer DMA
615 if (size == 0x100000000UL)
618 baseptr = &pci_emul_membase64;
619 limit = PCI_EMUL_MEMLIMIT64;
620 mask = PCIM_BAR_MEM_BASE;
621 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
622 PCIM_BAR_MEM_PREFETCH;
625 baseptr = &pci_emul_membase32;
626 limit = PCI_EMUL_MEMLIMIT32;
627 mask = PCIM_BAR_MEM_BASE;
628 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
632 baseptr = &pci_emul_membase32;
633 limit = PCI_EMUL_MEMLIMIT32;
634 mask = PCIM_BAR_MEM_BASE;
635 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
638 printf("pci_emul_alloc_base: invalid bar type %d\n", type);
642 if (baseptr != NULL) {
643 error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
648 pdi->pi_bar[idx].type = type;
649 pdi->pi_bar[idx].addr = addr;
650 pdi->pi_bar[idx].size = size;
652 /* Initialize the BAR register in config space */
653 bar = (addr & mask) | lobits;
654 pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
656 if (type == PCIBAR_MEM64) {
657 assert(idx + 1 <= PCI_BARMAX);
658 pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
659 pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
662 register_bar(pdi, idx);
667 #define CAP_START_OFFSET 0x40
669 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
671 int i, capoff, reallen;
676 reallen = roundup2(caplen, 4); /* dword aligned */
678 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
679 if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
680 capoff = CAP_START_OFFSET;
682 capoff = pi->pi_capend + 1;
684 /* Check if we have enough space */
685 if (capoff + reallen > PCI_REGMAX + 1)
688 /* Set the previous capability pointer */
689 if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
690 pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
691 pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
693 pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
695 /* Copy the capability */
696 for (i = 0; i < caplen; i++)
697 pci_set_cfgdata8(pi, capoff + i, capdata[i]);
699 /* Set the next capability pointer */
700 pci_set_cfgdata8(pi, capoff + 1, 0);
702 pi->pi_prevcap = capoff;
703 pi->pi_capend = capoff + reallen - 1;
707 static struct pci_devemu *
708 pci_emul_finddev(char *name)
710 struct pci_devemu **pdpp, *pdp;
712 SET_FOREACH(pdpp, pci_devemu_set) {
714 if (!strcmp(pdp->pe_emu, name)) {
723 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
724 int func, struct funcinfo *fi)
726 struct pci_devinst *pdi;
729 pdi = calloc(1, sizeof(struct pci_devinst));
735 pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
736 pdi->pi_lintr.pin = 0;
737 pdi->pi_lintr.state = IDLE;
738 pdi->pi_lintr.pirq_pin = 0;
739 pdi->pi_lintr.ioapic_irq = 0;
741 snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
743 /* Disable legacy interrupts */
744 pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
745 pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
747 pci_set_cfgdata8(pdi, PCIR_COMMAND,
748 PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
750 err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
760 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
764 /* Number of msi messages must be a power of 2 between 1 and 32 */
765 assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
766 mmc = ffs(msgnum) - 1;
768 bzero(msicap, sizeof(struct msicap));
769 msicap->capid = PCIY_MSI;
770 msicap->nextptr = nextptr;
771 msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
775 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
777 struct msicap msicap;
779 pci_populate_msicap(&msicap, msgnum, 0);
781 return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
785 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
786 uint32_t msix_tab_size)
789 assert(msix_tab_size % 4096 == 0);
791 bzero(msixcap, sizeof(struct msixcap));
792 msixcap->capid = PCIY_MSIX;
795 * Message Control Register, all fields set to
796 * zero except for the Table Size.
797 * Note: Table size N is encoded as N-1
799 msixcap->msgctrl = msgnum - 1;
803 * - MSI-X table start at offset 0
804 * - PBA table starts at a 4K aligned offset after the MSI-X table
806 msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
807 msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
811 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
815 assert(table_entries > 0);
816 assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
818 table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
819 pi->pi_msix.table = calloc(1, table_size);
821 /* set mask bit of vector control register */
822 for (i = 0; i < table_entries; i++)
823 pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
827 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
830 struct msixcap msixcap;
832 assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
833 assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
835 tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
837 /* Align table size to nearest 4K */
838 tab_size = roundup2(tab_size, 4096);
840 pi->pi_msix.table_bar = barnum;
841 pi->pi_msix.pba_bar = barnum;
842 pi->pi_msix.table_offset = 0;
843 pi->pi_msix.table_count = msgnum;
844 pi->pi_msix.pba_offset = tab_size;
845 pi->pi_msix.pba_size = PBA_SIZE(msgnum);
847 pci_msix_table_init(pi, msgnum);
849 pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
851 /* allocate memory for MSI-X Table and PBA */
852 pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
853 tab_size + pi->pi_msix.pba_size);
855 return (pci_emul_add_capability(pi, (u_char *)&msixcap,
860 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
861 int bytes, uint32_t val)
863 uint16_t msgctrl, rwmask;
866 off = offset - capoff;
867 /* Message Control Register */
868 if (off == 2 && bytes == 2) {
869 rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
870 msgctrl = pci_get_cfgdata16(pi, offset);
872 msgctrl |= val & rwmask;
875 pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
876 pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
877 pci_lintr_update(pi);
880 CFGWRITE(pi, offset, val, bytes);
884 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
885 int bytes, uint32_t val)
887 uint16_t msgctrl, rwmask, msgdata, mme;
891 * If guest is writing to the message control register make sure
892 * we do not overwrite read-only fields.
894 if ((offset - capoff) == 2 && bytes == 2) {
895 rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
896 msgctrl = pci_get_cfgdata16(pi, offset);
898 msgctrl |= val & rwmask;
901 addrlo = pci_get_cfgdata32(pi, capoff + 4);
902 if (msgctrl & PCIM_MSICTRL_64BIT)
903 msgdata = pci_get_cfgdata16(pi, capoff + 12);
905 msgdata = pci_get_cfgdata16(pi, capoff + 8);
907 mme = msgctrl & PCIM_MSICTRL_MME_MASK;
908 pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
909 if (pi->pi_msi.enabled) {
910 pi->pi_msi.addr = addrlo;
911 pi->pi_msi.msg_data = msgdata;
912 pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
914 pi->pi_msi.maxmsgnum = 0;
916 pci_lintr_update(pi);
919 CFGWRITE(pi, offset, val, bytes);
923 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
924 int bytes, uint32_t val)
927 /* XXX don't write to the readonly parts */
928 CFGWRITE(pi, offset, val, bytes);
931 #define PCIECAP_VERSION 0x2
933 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
936 struct pciecap pciecap;
938 if (type != PCIEM_TYPE_ROOT_PORT)
941 bzero(&pciecap, sizeof(pciecap));
943 pciecap.capid = PCIY_EXPRESS;
944 pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
945 pciecap.link_capabilities = 0x411; /* gen1, x1 */
946 pciecap.link_status = 0x11; /* gen1, x1 */
948 err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
953 * This function assumes that 'coff' is in the capabilities region of the
957 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
960 uint8_t capoff, nextoff;
962 /* Do not allow un-aligned writes */
963 if ((offset & (bytes - 1)) != 0)
966 /* Find the capability that we want to update */
967 capoff = CAP_START_OFFSET;
969 nextoff = pci_get_cfgdata8(pi, capoff + 1);
972 if (offset >= capoff && offset < nextoff)
977 assert(offset >= capoff);
980 * Capability ID and Next Capability Pointer are readonly.
981 * However, some o/s's do 4-byte writes that include these.
982 * For this case, trim the write back to 2 bytes and adjust
985 if (offset == capoff || offset == capoff + 1) {
986 if (offset == capoff && bytes == 4) {
994 capid = pci_get_cfgdata8(pi, capoff);
997 msicap_cfgwrite(pi, capoff, offset, bytes, val);
1000 msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1003 pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1011 pci_emul_iscap(struct pci_devinst *pi, int offset)
1015 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1016 if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1017 if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1024 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1025 int size, uint64_t *val, void *arg1, long arg2)
1028 * Ignore writes; return 0xff's for reads. The mem read code
1029 * will take care of truncating to the correct size.
1031 if (dir == MEM_F_READ) {
1032 *val = 0xffffffffffffffff;
1039 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1040 int bytes, uint64_t *val, void *arg1, long arg2)
1042 int bus, slot, func, coff, in;
1044 coff = addr & 0xfff;
1045 func = (addr >> 12) & 0x7;
1046 slot = (addr >> 15) & 0x1f;
1047 bus = (addr >> 20) & 0xff;
1048 in = (dir == MEM_F_READ);
1051 pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1059 return (PCI_EMUL_ECFG_BASE);
1062 #define BUSIO_ROUNDUP 32
1063 #define BUSMEM_ROUNDUP (1024 * 1024)
1066 init_pci(struct vmctx *ctx)
1068 struct mem_range mr;
1069 struct pci_devemu *pde;
1071 struct slotinfo *si;
1072 struct funcinfo *fi;
1074 int bus, slot, func;
1077 pci_emul_iobase = PCI_EMUL_IOBASE;
1078 pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1079 pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1081 for (bus = 0; bus < MAXBUSES; bus++) {
1082 if ((bi = pci_businfo[bus]) == NULL)
1085 * Keep track of the i/o and memory resources allocated to
1088 bi->iobase = pci_emul_iobase;
1089 bi->membase32 = pci_emul_membase32;
1090 bi->membase64 = pci_emul_membase64;
1092 for (slot = 0; slot < MAXSLOTS; slot++) {
1093 si = &bi->slotinfo[slot];
1094 for (func = 0; func < MAXFUNCS; func++) {
1095 fi = &si->si_funcs[func];
1096 if (fi->fi_name == NULL)
1098 pde = pci_emul_finddev(fi->fi_name);
1099 assert(pde != NULL);
1100 error = pci_emul_init(ctx, pde, bus, slot,
1108 * Add some slop to the I/O and memory resources decoded by
1109 * this bus to give a guest some flexibility if it wants to
1110 * reprogram the BARs.
1112 pci_emul_iobase += BUSIO_ROUNDUP;
1113 pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1114 bi->iolimit = pci_emul_iobase;
1116 pci_emul_membase32 += BUSMEM_ROUNDUP;
1117 pci_emul_membase32 = roundup2(pci_emul_membase32,
1119 bi->memlimit32 = pci_emul_membase32;
1121 pci_emul_membase64 += BUSMEM_ROUNDUP;
1122 pci_emul_membase64 = roundup2(pci_emul_membase64,
1124 bi->memlimit64 = pci_emul_membase64;
1128 * PCI backends are initialized before routing INTx interrupts
1129 * so that LPC devices are able to reserve ISA IRQs before
1130 * routing PIRQ pins.
1132 for (bus = 0; bus < MAXBUSES; bus++) {
1133 if ((bi = pci_businfo[bus]) == NULL)
1136 for (slot = 0; slot < MAXSLOTS; slot++) {
1137 si = &bi->slotinfo[slot];
1138 for (func = 0; func < MAXFUNCS; func++) {
1139 fi = &si->si_funcs[func];
1140 if (fi->fi_devi == NULL)
1142 pci_lintr_route(fi->fi_devi);
1149 * The guest physical memory map looks like the following:
1150 * [0, lowmem) guest system memory
1151 * [lowmem, lowmem_limit) memory hole (may be absent)
1152 * [lowmem_limit, 0xE0000000) PCI hole (32-bit BAR allocation)
1153 * [0xE0000000, 0xF0000000) PCI extended config window
1154 * [0xF0000000, 4GB) LAPIC, IOAPIC, HPET, firmware
1155 * [4GB, 4GB + highmem)
1159 * Accesses to memory addresses that are not allocated to system
1160 * memory or PCI devices return 0xff's.
1162 lowmem = vm_get_lowmem_size(ctx);
1163 bzero(&mr, sizeof(struct mem_range));
1164 mr.name = "PCI hole";
1165 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1167 mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1168 mr.handler = pci_emul_fallback_handler;
1169 error = register_mem_fallback(&mr);
1172 /* PCI extended config space */
1173 bzero(&mr, sizeof(struct mem_range));
1174 mr.name = "PCI ECFG";
1175 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1176 mr.base = PCI_EMUL_ECFG_BASE;
1177 mr.size = PCI_EMUL_ECFG_SIZE;
1178 mr.handler = pci_emul_ecfg_handler;
1179 error = register_mem(&mr);
1186 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1190 dsdt_line(" Package ()");
1192 dsdt_line(" 0x%X,", slot << 16 | 0xffff);
1193 dsdt_line(" 0x%02X,", pin - 1);
1194 dsdt_line(" Zero,");
1195 dsdt_line(" 0x%X", ioapic_irq);
1200 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1205 name = lpc_pirq_name(pirq_pin);
1208 dsdt_line(" Package ()");
1210 dsdt_line(" 0x%X,", slot << 16 | 0xffff);
1211 dsdt_line(" 0x%02X,", pin - 1);
1212 dsdt_line(" %s,", name);
1219 * A bhyve virtual machine has a flat PCI hierarchy with a root port
1220 * corresponding to each PCI bus.
1223 pci_bus_write_dsdt(int bus)
1226 struct slotinfo *si;
1227 struct pci_devinst *pi;
1228 int count, func, slot;
1231 * If there are no devices on this 'bus' then just return.
1233 if ((bi = pci_businfo[bus]) == NULL) {
1235 * Bus 0 is special because it decodes the I/O ports used
1236 * for PCI config space access even if there are no devices
1243 dsdt_line(" Device (PC%02X)", bus);
1245 dsdt_line(" Name (_HID, EisaId (\"PNP0A03\"))");
1246 dsdt_line(" Name (_ADR, Zero)");
1248 dsdt_line(" Method (_BBN, 0, NotSerialized)");
1250 dsdt_line(" Return (0x%08X)", bus);
1252 dsdt_line(" Name (_CRS, ResourceTemplate ()");
1254 dsdt_line(" WordBusNumber (ResourceProducer, MinFixed, "
1255 "MaxFixed, PosDecode,");
1256 dsdt_line(" 0x0000, // Granularity");
1257 dsdt_line(" 0x%04X, // Range Minimum", bus);
1258 dsdt_line(" 0x%04X, // Range Maximum", bus);
1259 dsdt_line(" 0x0000, // Translation Offset");
1260 dsdt_line(" 0x0001, // Length");
1265 dsdt_fixed_ioport(0xCF8, 8);
1268 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1269 "PosDecode, EntireRange,");
1270 dsdt_line(" 0x0000, // Granularity");
1271 dsdt_line(" 0x0000, // Range Minimum");
1272 dsdt_line(" 0x0CF7, // Range Maximum");
1273 dsdt_line(" 0x0000, // Translation Offset");
1274 dsdt_line(" 0x0CF8, // Length");
1275 dsdt_line(" ,, , TypeStatic)");
1277 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1278 "PosDecode, EntireRange,");
1279 dsdt_line(" 0x0000, // Granularity");
1280 dsdt_line(" 0x0D00, // Range Minimum");
1281 dsdt_line(" 0x%04X, // Range Maximum",
1282 PCI_EMUL_IOBASE - 1);
1283 dsdt_line(" 0x0000, // Translation Offset");
1284 dsdt_line(" 0x%04X, // Length",
1285 PCI_EMUL_IOBASE - 0x0D00);
1286 dsdt_line(" ,, , TypeStatic)");
1296 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1297 "PosDecode, EntireRange,");
1298 dsdt_line(" 0x0000, // Granularity");
1299 dsdt_line(" 0x%04X, // Range Minimum", bi->iobase);
1300 dsdt_line(" 0x%04X, // Range Maximum",
1302 dsdt_line(" 0x0000, // Translation Offset");
1303 dsdt_line(" 0x%04X, // Length",
1304 bi->iolimit - bi->iobase);
1305 dsdt_line(" ,, , TypeStatic)");
1307 /* mmio window (32-bit) */
1308 dsdt_line(" DWordMemory (ResourceProducer, PosDecode, "
1309 "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1310 dsdt_line(" 0x00000000, // Granularity");
1311 dsdt_line(" 0x%08X, // Range Minimum\n", bi->membase32);
1312 dsdt_line(" 0x%08X, // Range Maximum\n",
1313 bi->memlimit32 - 1);
1314 dsdt_line(" 0x00000000, // Translation Offset");
1315 dsdt_line(" 0x%08X, // Length\n",
1316 bi->memlimit32 - bi->membase32);
1317 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)");
1319 /* mmio window (64-bit) */
1320 dsdt_line(" QWordMemory (ResourceProducer, PosDecode, "
1321 "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1322 dsdt_line(" 0x0000000000000000, // Granularity");
1323 dsdt_line(" 0x%016lX, // Range Minimum\n", bi->membase64);
1324 dsdt_line(" 0x%016lX, // Range Maximum\n",
1325 bi->memlimit64 - 1);
1326 dsdt_line(" 0x0000000000000000, // Translation Offset");
1327 dsdt_line(" 0x%016lX, // Length\n",
1328 bi->memlimit64 - bi->membase64);
1329 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)");
1332 count = pci_count_lintr(bus);
1335 dsdt_line("Name (PPRT, Package ()");
1337 pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1339 dsdt_line("Name (APRT, Package ()");
1341 pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1343 dsdt_line("Method (_PRT, 0, NotSerialized)");
1345 dsdt_line(" If (PICM)");
1347 dsdt_line(" Return (APRT)");
1351 dsdt_line(" Return (PPRT)");
1358 for (slot = 0; slot < MAXSLOTS; slot++) {
1359 si = &bi->slotinfo[slot];
1360 for (func = 0; func < MAXFUNCS; func++) {
1361 pi = si->si_funcs[func].fi_devi;
1362 if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1363 pi->pi_d->pe_write_dsdt(pi);
1372 pci_write_dsdt(void)
1377 dsdt_line("Name (PICM, 0x00)");
1378 dsdt_line("Method (_PIC, 1, NotSerialized)");
1380 dsdt_line(" Store (Arg0, PICM)");
1383 dsdt_line("Scope (_SB)");
1385 for (bus = 0; bus < MAXBUSES; bus++)
1386 pci_bus_write_dsdt(bus);
1392 pci_bus_configured(int bus)
1394 assert(bus >= 0 && bus < MAXBUSES);
1395 return (pci_businfo[bus] != NULL);
1399 pci_msi_enabled(struct pci_devinst *pi)
1401 return (pi->pi_msi.enabled);
1405 pci_msi_maxmsgnum(struct pci_devinst *pi)
1407 if (pi->pi_msi.enabled)
1408 return (pi->pi_msi.maxmsgnum);
1414 pci_msix_enabled(struct pci_devinst *pi)
1417 return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1421 pci_generate_msix(struct pci_devinst *pi, int index)
1423 struct msix_table_entry *mte;
1425 if (!pci_msix_enabled(pi))
1428 if (pi->pi_msix.function_mask)
1431 if (index >= pi->pi_msix.table_count)
1434 mte = &pi->pi_msix.table[index];
1435 if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1436 /* XXX Set PBA bit if interrupt is disabled */
1437 vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1442 pci_generate_msi(struct pci_devinst *pi, int index)
1445 if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1446 vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1447 pi->pi_msi.msg_data + index);
1452 pci_lintr_permitted(struct pci_devinst *pi)
1456 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1457 return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1458 (cmd & PCIM_CMD_INTxDIS)));
1462 pci_lintr_request(struct pci_devinst *pi)
1465 struct slotinfo *si;
1466 int bestpin, bestcount, pin;
1468 bi = pci_businfo[pi->pi_bus];
1472 * Just allocate a pin from our slot. The pin will be
1473 * assigned IRQs later when interrupts are routed.
1475 si = &bi->slotinfo[pi->pi_slot];
1477 bestcount = si->si_intpins[0].ii_count;
1478 for (pin = 1; pin < 4; pin++) {
1479 if (si->si_intpins[pin].ii_count < bestcount) {
1481 bestcount = si->si_intpins[pin].ii_count;
1485 si->si_intpins[bestpin].ii_count++;
1486 pi->pi_lintr.pin = bestpin + 1;
1487 pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1491 pci_lintr_route(struct pci_devinst *pi)
1494 struct intxinfo *ii;
1496 if (pi->pi_lintr.pin == 0)
1499 bi = pci_businfo[pi->pi_bus];
1501 ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1504 * Attempt to allocate an I/O APIC pin for this intpin if one
1505 * is not yet assigned.
1507 if (ii->ii_ioapic_irq == 0)
1508 ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1509 assert(ii->ii_ioapic_irq > 0);
1512 * Attempt to allocate a PIRQ pin for this intpin if one is
1515 if (ii->ii_pirq_pin == 0)
1516 ii->ii_pirq_pin = pirq_alloc_pin(pi);
1517 assert(ii->ii_pirq_pin > 0);
1519 pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1520 pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1521 pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1525 pci_lintr_assert(struct pci_devinst *pi)
1528 assert(pi->pi_lintr.pin > 0);
1530 pthread_mutex_lock(&pi->pi_lintr.lock);
1531 if (pi->pi_lintr.state == IDLE) {
1532 if (pci_lintr_permitted(pi)) {
1533 pi->pi_lintr.state = ASSERTED;
1536 pi->pi_lintr.state = PENDING;
1538 pthread_mutex_unlock(&pi->pi_lintr.lock);
1542 pci_lintr_deassert(struct pci_devinst *pi)
1545 assert(pi->pi_lintr.pin > 0);
1547 pthread_mutex_lock(&pi->pi_lintr.lock);
1548 if (pi->pi_lintr.state == ASSERTED) {
1549 pi->pi_lintr.state = IDLE;
1550 pci_irq_deassert(pi);
1551 } else if (pi->pi_lintr.state == PENDING)
1552 pi->pi_lintr.state = IDLE;
1553 pthread_mutex_unlock(&pi->pi_lintr.lock);
1557 pci_lintr_update(struct pci_devinst *pi)
1560 pthread_mutex_lock(&pi->pi_lintr.lock);
1561 if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1562 pci_irq_deassert(pi);
1563 pi->pi_lintr.state = PENDING;
1564 } else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1565 pi->pi_lintr.state = ASSERTED;
1568 pthread_mutex_unlock(&pi->pi_lintr.lock);
1572 pci_count_lintr(int bus)
1574 int count, slot, pin;
1575 struct slotinfo *slotinfo;
1578 if (pci_businfo[bus] != NULL) {
1579 for (slot = 0; slot < MAXSLOTS; slot++) {
1580 slotinfo = &pci_businfo[bus]->slotinfo[slot];
1581 for (pin = 0; pin < 4; pin++) {
1582 if (slotinfo->si_intpins[pin].ii_count != 0)
1591 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1594 struct slotinfo *si;
1595 struct intxinfo *ii;
1598 if ((bi = pci_businfo[bus]) == NULL)
1601 for (slot = 0; slot < MAXSLOTS; slot++) {
1602 si = &bi->slotinfo[slot];
1603 for (pin = 0; pin < 4; pin++) {
1604 ii = &si->si_intpins[pin];
1605 if (ii->ii_count != 0)
1606 cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1607 ii->ii_ioapic_irq, arg);
1613 * Return 1 if the emulated device in 'slot' is a multi-function device.
1614 * Return 0 otherwise.
1617 pci_emul_is_mfdev(int bus, int slot)
1620 struct slotinfo *si;
1624 if ((bi = pci_businfo[bus]) != NULL) {
1625 si = &bi->slotinfo[slot];
1626 for (f = 0; f < MAXFUNCS; f++) {
1627 if (si->si_funcs[f].fi_devi != NULL) {
1632 return (numfuncs > 1);
1636 * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1637 * whether or not is a multi-function being emulated in the pci 'slot'.
1640 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1644 if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1645 mfdev = pci_emul_is_mfdev(bus, slot);
1655 *rv &= ~(PCIM_MFDEV << 16);
1657 *rv |= (PCIM_MFDEV << 16);
1665 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1668 uint32_t cmd, cmd2, changed, old, readonly;
1670 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); /* stash old value */
1673 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1675 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1676 * 'write 1 to clear'. However these bits are not set to '1' by
1677 * any device emulation so it is simpler to treat them as readonly.
1679 rshift = (coff & 0x3) * 8;
1680 readonly = 0xFFFFF880 >> rshift;
1682 old = CFGREAD(pi, coff, bytes);
1684 new |= (old & readonly);
1685 CFGWRITE(pi, coff, new, bytes); /* update config */
1687 cmd2 = pci_get_cfgdata16(pi, PCIR_COMMAND); /* get updated value */
1688 changed = cmd ^ cmd2;
1691 * If the MMIO or I/O address space decoding has changed then
1692 * register/unregister all BARs that decode that address space.
1694 for (i = 0; i <= PCI_BARMAX; i++) {
1695 switch (pi->pi_bar[i].type) {
1697 case PCIBAR_MEMHI64:
1700 /* I/O address space decoding changed? */
1701 if (changed & PCIM_CMD_PORTEN) {
1703 register_bar(pi, i);
1705 unregister_bar(pi, i);
1710 /* MMIO address space decoding changed? */
1711 if (changed & PCIM_CMD_MEMEN) {
1713 register_bar(pi, i);
1715 unregister_bar(pi, i);
1724 * If INTx has been unmasked and is pending, assert the
1727 pci_lintr_update(pi);
1731 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1732 int coff, int bytes, uint32_t *eax)
1735 struct slotinfo *si;
1736 struct pci_devinst *pi;
1737 struct pci_devemu *pe;
1739 uint64_t addr, bar, mask;
1741 if ((bi = pci_businfo[bus]) != NULL) {
1742 si = &bi->slotinfo[slot];
1743 pi = si->si_funcs[func].fi_devi;
1748 * Just return if there is no device at this slot:func or if the
1749 * the guest is doing an un-aligned access.
1751 if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1752 (coff & (bytes - 1)) != 0) {
1759 * Ignore all writes beyond the standard config space and return all
1762 if (coff >= PCI_REGMAX + 1) {
1766 * Extended capabilities begin at offset 256 in config
1767 * space. Absence of extended capabilities is signaled
1768 * with all 0s in the extended capability header at
1771 if (coff <= PCI_REGMAX + 4)
1783 /* Let the device emulation override the default handler */
1784 if (pe->pe_cfgread != NULL) {
1785 needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1792 *eax = CFGREAD(pi, coff, bytes);
1794 pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1796 /* Let the device emulation override the default handler */
1797 if (pe->pe_cfgwrite != NULL &&
1798 (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1802 * Special handling for write to BAR registers
1804 if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1806 * Ignore writes to BAR registers that are not
1809 if (bytes != 4 || (coff & 0x3) != 0)
1811 idx = (coff - PCIR_BAR(0)) / 4;
1812 mask = ~(pi->pi_bar[idx].size - 1);
1813 switch (pi->pi_bar[idx].type) {
1815 pi->pi_bar[idx].addr = bar = 0;
1820 bar = addr | PCIM_BAR_IO_SPACE;
1822 * Register the new BAR value for interception
1824 if (addr != pi->pi_bar[idx].addr) {
1825 update_bar_address(pi, addr, idx,
1830 addr = bar = *eax & mask;
1831 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1832 if (addr != pi->pi_bar[idx].addr) {
1833 update_bar_address(pi, addr, idx,
1838 addr = bar = *eax & mask;
1839 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1840 PCIM_BAR_MEM_PREFETCH;
1841 if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1842 update_bar_address(pi, addr, idx,
1846 case PCIBAR_MEMHI64:
1847 mask = ~(pi->pi_bar[idx - 1].size - 1);
1848 addr = ((uint64_t)*eax << 32) & mask;
1850 if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1851 update_bar_address(pi, addr, idx - 1,
1858 pci_set_cfgdata32(pi, coff, bar);
1860 } else if (pci_emul_iscap(pi, coff)) {
1861 pci_emul_capwrite(pi, coff, bytes, *eax);
1862 } else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1863 pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1865 CFGWRITE(pi, coff, *eax, bytes);
1870 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1873 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1874 uint32_t *eax, void *arg)
1880 *eax = (bytes == 2) ? 0xffff : 0xff;
1885 x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1891 cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1892 cfgoff = x & PCI_REGMAX;
1893 cfgfunc = (x >> 8) & PCI_FUNCMAX;
1894 cfgslot = (x >> 11) & PCI_SLOTMAX;
1895 cfgbus = (x >> 16) & PCI_BUSMAX;
1900 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1903 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1904 uint32_t *eax, void *arg)
1908 assert(bytes == 1 || bytes == 2 || bytes == 4);
1910 coff = cfgoff + (port - CONF1_DATA_PORT);
1912 pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1915 /* Ignore accesses to cfgdata if not enabled by cfgaddr */
1922 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1923 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1924 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1925 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1927 #define PCI_EMUL_TEST
1928 #ifdef PCI_EMUL_TEST
1930 * Define a dummy test device
1934 struct pci_emul_dsoftc {
1935 uint8_t ioregs[DIOSZ];
1936 uint8_t memregs[2][DMEMSZ];
1939 #define PCI_EMUL_MSI_MSGS 4
1940 #define PCI_EMUL_MSIX_MSGS 16
1943 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1946 struct pci_emul_dsoftc *sc;
1948 sc = calloc(1, sizeof(struct pci_emul_dsoftc));
1952 pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1953 pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1954 pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1956 error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1959 error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1962 error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1965 error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
1972 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1973 uint64_t offset, int size, uint64_t value)
1976 struct pci_emul_dsoftc *sc = pi->pi_arg;
1979 if (offset + size > DIOSZ) {
1980 printf("diow: iow too large, offset %ld size %d\n",
1986 sc->ioregs[offset] = value & 0xff;
1987 } else if (size == 2) {
1988 *(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1989 } else if (size == 4) {
1990 *(uint32_t *)&sc->ioregs[offset] = value;
1992 printf("diow: iow unknown size %d\n", size);
1996 * Special magic value to generate an interrupt
1998 if (offset == 4 && size == 4 && pci_msi_enabled(pi))
1999 pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2001 if (value == 0xabcdef) {
2002 for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2003 pci_generate_msi(pi, i);
2007 if (baridx == 1 || baridx == 2) {
2008 if (offset + size > DMEMSZ) {
2009 printf("diow: memw too large, offset %ld size %d\n",
2014 i = baridx - 1; /* 'memregs' index */
2017 sc->memregs[i][offset] = value;
2018 } else if (size == 2) {
2019 *(uint16_t *)&sc->memregs[i][offset] = value;
2020 } else if (size == 4) {
2021 *(uint32_t *)&sc->memregs[i][offset] = value;
2022 } else if (size == 8) {
2023 *(uint64_t *)&sc->memregs[i][offset] = value;
2025 printf("diow: memw unknown size %d\n", size);
2029 * magic interrupt ??
2033 if (baridx > 2 || baridx < 0) {
2034 printf("diow: unknown bar idx %d\n", baridx);
2039 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2040 uint64_t offset, int size)
2042 struct pci_emul_dsoftc *sc = pi->pi_arg;
2047 if (offset + size > DIOSZ) {
2048 printf("dior: ior too large, offset %ld size %d\n",
2055 value = sc->ioregs[offset];
2056 } else if (size == 2) {
2057 value = *(uint16_t *) &sc->ioregs[offset];
2058 } else if (size == 4) {
2059 value = *(uint32_t *) &sc->ioregs[offset];
2061 printf("dior: ior unknown size %d\n", size);
2065 if (baridx == 1 || baridx == 2) {
2066 if (offset + size > DMEMSZ) {
2067 printf("dior: memr too large, offset %ld size %d\n",
2072 i = baridx - 1; /* 'memregs' index */
2075 value = sc->memregs[i][offset];
2076 } else if (size == 2) {
2077 value = *(uint16_t *) &sc->memregs[i][offset];
2078 } else if (size == 4) {
2079 value = *(uint32_t *) &sc->memregs[i][offset];
2080 } else if (size == 8) {
2081 value = *(uint64_t *) &sc->memregs[i][offset];
2083 printf("dior: ior unknown size %d\n", size);
2088 if (baridx > 2 || baridx < 0) {
2089 printf("dior: unknown bar idx %d\n", baridx);
2096 struct pci_devemu pci_dummy = {
2098 .pe_init = pci_emul_dinit,
2099 .pe_barwrite = pci_emul_diow,
2100 .pe_barread = pci_emul_dior
2102 PCI_EMUL_SET(pci_dummy);
2104 #endif /* PCI_EMUL_TEST */