2 * Copyright (c) 2011 NetApp, Inc.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/errno.h>
44 #include <machine/vmm.h>
53 #define CONF1_ADDR_PORT 0x0cf8
54 #define CONF1_DATA_PORT 0x0cfc
56 #define CONF1_ENABLE 0x80000000ul
58 #define CFGWRITE(pi,off,val,b) \
61 pci_set_cfgdata8((pi),(off),(val)); \
62 } else if ((b) == 2) { \
63 pci_set_cfgdata16((pi),(off),(val)); \
65 pci_set_cfgdata32((pi),(off),(val)); \
69 #define MAXSLOTS (PCI_SLOTMAX + 1)
70 #define MAXFUNCS (PCI_FUNCMAX + 1)
72 static struct slotinfo {
75 struct pci_devinst *si_devi;
77 } pci_slotinfo[MAXSLOTS][MAXFUNCS];
80 * Used to keep track of legacy interrupt owners/requestors
84 static struct lirqinfo {
87 struct pci_devinst *li_owner; /* XXX should be a list */
90 SET_DECLARE(pci_devemu_set, struct pci_devemu);
92 static uint64_t pci_emul_iobase;
93 static uint64_t pci_emul_membase32;
94 static uint64_t pci_emul_membase64;
96 #define PCI_EMUL_IOBASE 0x2000
97 #define PCI_EMUL_IOLIMIT 0x10000
99 #define PCI_EMUL_MEMLIMIT32 0xE0000000 /* 3.5GB */
101 #define PCI_EMUL_MEMBASE64 0xD000000000UL
102 #define PCI_EMUL_MEMLIMIT64 0xFD00000000UL
104 static struct pci_devemu *pci_emul_finddev(char *name);
106 static int pci_emul_devices;
113 * Slot options are in the form:
115 * <slot>[:<func>],<emul>[,<config>]
119 * emul is a string describing the type of PCI device e.g. virtio-net
120 * config is an optional string, depending on the device, that can be
121 * used for configuration.
127 pci_parse_slot_usage(char *aopt)
130 fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt);
134 pci_parse_slot(char *opt, int legacy)
136 char *slot, *func, *emul, *config;
138 int error, snum, fnum;
141 str = cpy = strdup(opt);
143 slot = strsep(&str, ",");
145 if (strchr(slot, ':') != NULL) {
147 (void) strsep(&func, ":");
150 emul = strsep(&str, ",");
154 pci_parse_slot_usage(opt);
159 fnum = func ? atoi(func) : 0;
161 if (snum < 0 || snum >= MAXSLOTS || fnum < 0 || fnum >= MAXFUNCS) {
162 pci_parse_slot_usage(opt);
166 if (pci_slotinfo[snum][fnum].si_name != NULL) {
167 fprintf(stderr, "pci slot %d:%d already occupied!\n",
172 if (pci_emul_finddev(emul) == NULL) {
173 fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n",
179 pci_slotinfo[snum][fnum].si_name = emul;
180 pci_slotinfo[snum][fnum].si_param = config;
181 pci_slotinfo[snum][fnum].si_legacy = legacy;
191 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
194 if (offset < pi->pi_msix.pba_offset)
197 if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
205 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
208 int msix_entry_offset;
212 /* support only 4 or 8 byte writes */
213 if (size != 4 && size != 8)
217 * Return if table index is beyond what device supports
219 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
220 if (tab_index >= pi->pi_msix.table_count)
223 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
225 /* support only aligned writes */
226 if ((msix_entry_offset % size) != 0)
229 dest = (char *)(pi->pi_msix.table + tab_index);
230 dest += msix_entry_offset;
233 *((uint32_t *)dest) = value;
235 *((uint64_t *)dest) = value;
241 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
244 int msix_entry_offset;
246 uint64_t retval = ~0;
249 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
250 * table but we also allow 1 byte access to accomodate reads from
253 if (size != 1 && size != 4 && size != 8)
256 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
258 /* support only aligned reads */
259 if ((msix_entry_offset % size) != 0) {
263 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
265 if (tab_index < pi->pi_msix.table_count) {
266 /* valid MSI-X Table access */
267 dest = (char *)(pi->pi_msix.table + tab_index);
268 dest += msix_entry_offset;
271 retval = *((uint8_t *)dest);
273 retval = *((uint32_t *)dest);
275 retval = *((uint64_t *)dest);
276 } else if (pci_valid_pba_offset(pi, offset)) {
277 /* return 0 for PBA access */
285 pci_msix_table_bar(struct pci_devinst *pi)
288 if (pi->pi_msix.table != NULL)
289 return (pi->pi_msix.table_bar);
295 pci_msix_pba_bar(struct pci_devinst *pi)
298 if (pi->pi_msix.table != NULL)
299 return (pi->pi_msix.pba_bar);
305 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
306 uint32_t *eax, void *arg)
308 struct pci_devinst *pdi = arg;
309 struct pci_devemu *pe = pdi->pi_d;
313 for (i = 0; i <= PCI_BARMAX; i++) {
314 if (pdi->pi_bar[i].type == PCIBAR_IO &&
315 port >= pdi->pi_bar[i].addr &&
316 port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
317 offset = port - pdi->pi_bar[i].addr;
319 *eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
322 (*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
331 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
332 int size, uint64_t *val, void *arg1, long arg2)
334 struct pci_devinst *pdi = arg1;
335 struct pci_devemu *pe = pdi->pi_d;
337 int bidx = (int) arg2;
339 assert(bidx <= PCI_BARMAX);
340 assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
341 pdi->pi_bar[bidx].type == PCIBAR_MEM64);
342 assert(addr >= pdi->pi_bar[bidx].addr &&
343 addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
345 offset = addr - pdi->pi_bar[bidx].addr;
347 if (dir == MEM_F_WRITE)
348 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, size, *val);
350 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, offset, size);
357 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
362 assert((size & (size - 1)) == 0); /* must be a power of 2 */
364 base = roundup2(*baseptr, size);
366 if (base + size <= limit) {
368 *baseptr = base + size;
375 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
379 return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
383 * Register (or unregister) the MMIO or I/O region associated with the BAR
384 * register 'idx' of an emulated pci device.
387 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
390 struct inout_port iop;
393 switch (pi->pi_bar[idx].type) {
395 bzero(&iop, sizeof(struct inout_port));
396 iop.name = pi->pi_name;
397 iop.port = pi->pi_bar[idx].addr;
398 iop.size = pi->pi_bar[idx].size;
400 iop.flags = IOPORT_F_INOUT;
401 iop.handler = pci_emul_io_handler;
403 error = register_inout(&iop);
405 error = unregister_inout(&iop);
409 bzero(&mr, sizeof(struct mem_range));
410 mr.name = pi->pi_name;
411 mr.base = pi->pi_bar[idx].addr;
412 mr.size = pi->pi_bar[idx].size;
415 mr.handler = pci_emul_mem_handler;
418 error = register_mem(&mr);
420 error = unregister_mem(&mr);
430 unregister_bar(struct pci_devinst *pi, int idx)
433 modify_bar_registration(pi, idx, 0);
437 register_bar(struct pci_devinst *pi, int idx)
440 modify_bar_registration(pi, idx, 1);
443 /* Are we decoding i/o port accesses for the emulated pci device? */
445 porten(struct pci_devinst *pi)
449 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
451 return (cmd & PCIM_CMD_PORTEN);
454 /* Are we decoding memory accesses for the emulated pci device? */
456 memen(struct pci_devinst *pi)
460 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
462 return (cmd & PCIM_CMD_MEMEN);
466 * Update the MMIO or I/O address that is decoded by the BAR register.
468 * If the pci device has enabled the address space decoding then intercept
469 * the address range decoded by the BAR register.
472 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
476 if (pi->pi_bar[idx].type == PCIBAR_IO)
482 unregister_bar(pi, idx);
487 pi->pi_bar[idx].addr = addr;
490 pi->pi_bar[idx].addr &= ~0xffffffffUL;
491 pi->pi_bar[idx].addr |= addr;
494 pi->pi_bar[idx].addr &= 0xffffffff;
495 pi->pi_bar[idx].addr |= addr;
502 register_bar(pi, idx);
506 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
507 enum pcibar_type type, uint64_t size)
510 uint64_t *baseptr, limit, addr, mask, lobits, bar;
512 assert(idx >= 0 && idx <= PCI_BARMAX);
514 if ((size & (size - 1)) != 0)
515 size = 1UL << flsl(size); /* round up to a power of 2 */
517 /* Enforce minimum BAR sizes required by the PCI standard */
518 if (type == PCIBAR_IO) {
529 addr = mask = lobits = 0;
533 pci_slotinfo[pdi->pi_slot][pdi->pi_func].si_legacy) {
534 assert(hostbase < PCI_EMUL_IOBASE);
537 baseptr = &pci_emul_iobase;
539 limit = PCI_EMUL_IOLIMIT;
540 mask = PCIM_BAR_IO_BASE;
541 lobits = PCIM_BAR_IO_SPACE;
546 * Some drivers do not work well if the 64-bit BAR is allocated
547 * above 4GB. Allow for this by allocating small requests under
548 * 4GB unless then allocation size is larger than some arbitrary
549 * number (32MB currently).
551 if (size > 32 * 1024 * 1024) {
553 * XXX special case for device requiring peer-peer DMA
555 if (size == 0x100000000UL)
558 baseptr = &pci_emul_membase64;
559 limit = PCI_EMUL_MEMLIMIT64;
560 mask = PCIM_BAR_MEM_BASE;
561 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
562 PCIM_BAR_MEM_PREFETCH;
565 baseptr = &pci_emul_membase32;
566 limit = PCI_EMUL_MEMLIMIT32;
567 mask = PCIM_BAR_MEM_BASE;
568 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
572 baseptr = &pci_emul_membase32;
573 limit = PCI_EMUL_MEMLIMIT32;
574 mask = PCIM_BAR_MEM_BASE;
575 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
578 printf("pci_emul_alloc_base: invalid bar type %d\n", type);
582 if (baseptr != NULL) {
583 error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
588 pdi->pi_bar[idx].type = type;
589 pdi->pi_bar[idx].addr = addr;
590 pdi->pi_bar[idx].size = size;
592 /* Initialize the BAR register in config space */
593 bar = (addr & mask) | lobits;
594 pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
596 if (type == PCIBAR_MEM64) {
597 assert(idx + 1 <= PCI_BARMAX);
598 pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
599 pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
602 register_bar(pdi, idx);
607 #define CAP_START_OFFSET 0x40
609 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
611 int i, capoff, capid, reallen;
614 static u_char endofcap[4] = {
615 PCIY_RESERVED, 0, 0, 0
618 assert(caplen > 0 && capdata[0] != PCIY_RESERVED);
620 reallen = roundup2(caplen, 4); /* dword aligned */
622 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
623 if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
624 capoff = CAP_START_OFFSET;
625 pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
626 pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
628 capoff = pci_get_cfgdata8(pi, PCIR_CAP_PTR);
630 assert((capoff & 0x3) == 0);
631 capid = pci_get_cfgdata8(pi, capoff);
632 if (capid == PCIY_RESERVED)
634 capoff = pci_get_cfgdata8(pi, capoff + 1);
638 /* Check if we have enough space */
639 if (capoff + reallen + sizeof(endofcap) > PCI_REGMAX + 1)
642 /* Copy the capability */
643 for (i = 0; i < caplen; i++)
644 pci_set_cfgdata8(pi, capoff + i, capdata[i]);
646 /* Set the next capability pointer */
647 pci_set_cfgdata8(pi, capoff + 1, capoff + reallen);
649 /* Copy of the reserved capability which serves as the end marker */
650 for (i = 0; i < sizeof(endofcap); i++)
651 pci_set_cfgdata8(pi, capoff + reallen + i, endofcap[i]);
656 static struct pci_devemu *
657 pci_emul_finddev(char *name)
659 struct pci_devemu **pdpp, *pdp;
661 SET_FOREACH(pdpp, pci_devemu_set) {
663 if (!strcmp(pdp->pe_emu, name)) {
672 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int slot, int func,
675 struct pci_devinst *pdi;
678 pdi = malloc(sizeof(struct pci_devinst));
679 bzero(pdi, sizeof(*pdi));
686 snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
688 /* Disable legacy interrupts */
689 pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
690 pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
692 pci_set_cfgdata8(pdi, PCIR_COMMAND,
693 PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
695 err = (*pde->pe_init)(ctx, pdi, params);
700 pci_slotinfo[slot][func].si_devi = pdi;
707 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
711 CTASSERT(sizeof(struct msicap) == 14);
713 /* Number of msi messages must be a power of 2 between 1 and 32 */
714 assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
715 mmc = ffs(msgnum) - 1;
717 bzero(msicap, sizeof(struct msicap));
718 msicap->capid = PCIY_MSI;
719 msicap->nextptr = nextptr;
720 msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
724 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
726 struct msicap msicap;
728 pci_populate_msicap(&msicap, msgnum, 0);
730 return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
734 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
735 uint32_t msix_tab_size, int nextptr)
737 CTASSERT(sizeof(struct msixcap) == 12);
739 assert(msix_tab_size % 4096 == 0);
741 bzero(msixcap, sizeof(struct msixcap));
742 msixcap->capid = PCIY_MSIX;
743 msixcap->nextptr = nextptr;
746 * Message Control Register, all fields set to
747 * zero except for the Table Size.
748 * Note: Table size N is encoded as N-1
750 msixcap->msgctrl = msgnum - 1;
754 * - MSI-X table start at offset 0
755 * - PBA table starts at a 4K aligned offset after the MSI-X table
757 msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
758 msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
762 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
766 assert(table_entries > 0);
767 assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
769 table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
770 pi->pi_msix.table = malloc(table_size);
771 bzero(pi->pi_msix.table, table_size);
773 /* set mask bit of vector control register */
774 for (i = 0; i < table_entries; i++)
775 pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
779 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
783 struct msixcap msixcap;
785 assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
786 assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
788 tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
790 /* Align table size to nearest 4K */
791 tab_size = roundup2(tab_size, 4096);
793 pi->pi_msix.table_bar = barnum;
794 pi->pi_msix.pba_bar = barnum;
795 pi->pi_msix.table_offset = 0;
796 pi->pi_msix.table_count = msgnum;
797 pi->pi_msix.pba_offset = tab_size;
799 /* calculate the MMIO size required for MSI-X PBA */
800 pba_index = (msgnum - 1) / (PBA_TABLE_ENTRY_SIZE * 8);
801 pi->pi_msix.pba_size = (pba_index + 1) * PBA_TABLE_ENTRY_SIZE;
803 pci_msix_table_init(pi, msgnum);
805 pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size, 0);
807 /* allocate memory for MSI-X Table and PBA */
808 pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
809 tab_size + pi->pi_msix.pba_size);
811 return (pci_emul_add_capability(pi, (u_char *)&msixcap,
816 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
817 int bytes, uint32_t val)
819 uint16_t msgctrl, rwmask;
822 off = offset - capoff;
823 table_bar = pi->pi_msix.table_bar;
824 /* Message Control Register */
825 if (off == 2 && bytes == 2) {
826 rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
827 msgctrl = pci_get_cfgdata16(pi, offset);
829 msgctrl |= val & rwmask;
832 pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
833 pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
836 CFGWRITE(pi, offset, val, bytes);
840 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
841 int bytes, uint32_t val)
843 uint16_t msgctrl, rwmask, msgdata, mme;
847 * If guest is writing to the message control register make sure
848 * we do not overwrite read-only fields.
850 if ((offset - capoff) == 2 && bytes == 2) {
851 rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
852 msgctrl = pci_get_cfgdata16(pi, offset);
854 msgctrl |= val & rwmask;
857 addrlo = pci_get_cfgdata32(pi, capoff + 4);
858 if (msgctrl & PCIM_MSICTRL_64BIT)
859 msgdata = pci_get_cfgdata16(pi, capoff + 12);
861 msgdata = pci_get_cfgdata16(pi, capoff + 8);
864 * XXX check delivery mode, destination mode etc
866 mme = msgctrl & PCIM_MSICTRL_MME_MASK;
867 pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
868 if (pi->pi_msi.enabled) {
869 pi->pi_msi.cpu = (addrlo >> 12) & 0xff;
870 pi->pi_msi.vector = msgdata & 0xff;
871 pi->pi_msi.msgnum = 1 << (mme >> 4);
874 pi->pi_msi.vector = 0;
875 pi->pi_msi.msgnum = 0;
879 CFGWRITE(pi, offset, val, bytes);
883 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
884 int bytes, uint32_t val)
887 /* XXX don't write to the readonly parts */
888 CFGWRITE(pi, offset, val, bytes);
891 #define PCIECAP_VERSION 0x2
893 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
896 struct pciecap pciecap;
898 CTASSERT(sizeof(struct pciecap) == 60);
900 if (type != PCIEM_TYPE_ROOT_PORT)
903 bzero(&pciecap, sizeof(pciecap));
905 pciecap.capid = PCIY_EXPRESS;
906 pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT;
907 pciecap.link_capabilities = 0x411; /* gen1, x1 */
908 pciecap.link_status = 0x11; /* gen1, x1 */
910 err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
915 * This function assumes that 'coff' is in the capabilities region of the
919 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val)
922 uint8_t capoff, nextoff;
924 /* Do not allow un-aligned writes */
925 if ((offset & (bytes - 1)) != 0)
928 /* Find the capability that we want to update */
929 capoff = CAP_START_OFFSET;
931 capid = pci_get_cfgdata8(pi, capoff);
932 if (capid == PCIY_RESERVED)
935 nextoff = pci_get_cfgdata8(pi, capoff + 1);
936 if (offset >= capoff && offset < nextoff)
941 assert(offset >= capoff);
944 * Capability ID and Next Capability Pointer are readonly.
945 * However, some o/s's do 4-byte writes that include these.
946 * For this case, trim the write back to 2 bytes and adjust
949 if (offset == capoff || offset == capoff + 1) {
950 if (offset == capoff && bytes == 4) {
960 msicap_cfgwrite(pi, capoff, offset, bytes, val);
963 msixcap_cfgwrite(pi, capoff, offset, bytes, val);
966 pciecap_cfgwrite(pi, capoff, offset, bytes, val);
974 pci_emul_iscap(struct pci_devinst *pi, int offset)
978 uint8_t capid, lastoff;
981 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
982 if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
983 lastoff = pci_get_cfgdata8(pi, PCIR_CAP_PTR);
985 assert((lastoff & 0x3) == 0);
986 capid = pci_get_cfgdata8(pi, lastoff);
987 if (capid == PCIY_RESERVED)
989 lastoff = pci_get_cfgdata8(pi, lastoff + 1);
991 if (offset >= CAP_START_OFFSET && offset <= lastoff)
998 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
999 int size, uint64_t *val, void *arg1, long arg2)
1002 * Ignore writes; return 0xff's for reads. The mem read code
1003 * will take care of truncating to the correct size.
1005 if (dir == MEM_F_READ) {
1006 *val = 0xffffffffffffffff;
1013 init_pci(struct vmctx *ctx)
1015 struct mem_range memp;
1016 struct pci_devemu *pde;
1017 struct slotinfo *si;
1022 pci_emul_iobase = PCI_EMUL_IOBASE;
1023 pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1024 pci_emul_membase64 = PCI_EMUL_MEMBASE64;
1027 * Allow ISA IRQs 5,10,11,12, and 15 to be available for
1030 lirq[5].li_generic = 1;
1031 lirq[10].li_generic = 1;
1032 lirq[11].li_generic = 1;
1033 lirq[12].li_generic = 1;
1034 lirq[15].li_generic = 1;
1036 for (slot = 0; slot < MAXSLOTS; slot++) {
1037 for (func = 0; func < MAXFUNCS; func++) {
1038 si = &pci_slotinfo[slot][func];
1039 if (si->si_name != NULL) {
1040 pde = pci_emul_finddev(si->si_name);
1041 assert(pde != NULL);
1042 error = pci_emul_init(ctx, pde, slot, func,
1051 * The guest physical memory map looks like the following:
1052 * [0, lowmem) guest system memory
1053 * [lowmem, lowmem_limit) memory hole (may be absent)
1054 * [lowmem_limit, 4GB) PCI hole (32-bit BAR allocation)
1055 * [4GB, 4GB + highmem)
1057 * Accesses to memory addresses that are not allocated to system
1058 * memory or PCI devices return 0xff's.
1060 error = vm_get_memory_seg(ctx, 0, &lowmem, NULL);
1063 memset(&memp, 0, sizeof(struct mem_range));
1064 memp.name = "PCI hole";
1065 memp.flags = MEM_F_RW;
1067 memp.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1068 memp.handler = pci_emul_fallback_handler;
1070 error = register_mem_fallback(&memp);
1077 pci_msi_enabled(struct pci_devinst *pi)
1079 return (pi->pi_msi.enabled);
1083 pci_msi_msgnum(struct pci_devinst *pi)
1085 if (pi->pi_msi.enabled)
1086 return (pi->pi_msi.msgnum);
1092 pci_msix_enabled(struct pci_devinst *pi)
1095 return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1099 pci_generate_msix(struct pci_devinst *pi, int index)
1101 struct msix_table_entry *mte;
1103 if (!pci_msix_enabled(pi))
1106 if (pi->pi_msix.function_mask)
1109 if (index >= pi->pi_msix.table_count)
1112 mte = &pi->pi_msix.table[index];
1113 if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1114 /* XXX Set PBA bit if interrupt is disabled */
1115 vm_lapic_irq(pi->pi_vmctx,
1116 (mte->addr >> 12) & 0xff, mte->msg_data & 0xff);
1121 pci_generate_msi(struct pci_devinst *pi, int msg)
1124 if (pci_msi_enabled(pi) && msg < pci_msi_msgnum(pi)) {
1125 vm_lapic_irq(pi->pi_vmctx,
1127 pi->pi_msi.vector + msg);
1132 pci_is_legacy(struct pci_devinst *pi)
1135 return (pci_slotinfo[pi->pi_slot][pi->pi_func].si_legacy);
1139 pci_lintr_alloc(struct pci_devinst *pi, int vec)
1143 assert(vec < NLIRQ);
1146 for (i = 0; i < NLIRQ; i++) {
1147 if (lirq[i].li_generic &&
1148 lirq[i].li_owner == NULL) {
1154 if (lirq[vec].li_owner != NULL) {
1160 lirq[vec].li_owner = pi;
1161 pi->pi_lintr_pin = vec;
1167 pci_lintr_request(struct pci_devinst *pi, int vec)
1170 vec = pci_lintr_alloc(pi, vec);
1171 pci_set_cfgdata8(pi, PCIR_INTLINE, vec);
1172 pci_set_cfgdata8(pi, PCIR_INTPIN, 1);
1177 pci_lintr_assert(struct pci_devinst *pi)
1180 assert(pi->pi_lintr_pin);
1181 ioapic_assert_pin(pi->pi_vmctx, pi->pi_lintr_pin);
1185 pci_lintr_deassert(struct pci_devinst *pi)
1188 assert(pi->pi_lintr_pin);
1189 ioapic_deassert_pin(pi->pi_vmctx, pi->pi_lintr_pin);
1193 * Return 1 if the emulated device in 'slot' is a multi-function device.
1194 * Return 0 otherwise.
1197 pci_emul_is_mfdev(int slot)
1202 for (f = 0; f < MAXFUNCS; f++) {
1203 if (pci_slotinfo[slot][f].si_devi != NULL) {
1207 return (numfuncs > 1);
1211 * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1212 * whether or not is a multi-function being emulated in the pci 'slot'.
1215 pci_emul_hdrtype_fixup(int slot, int off, int bytes, uint32_t *rv)
1219 if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1220 mfdev = pci_emul_is_mfdev(slot);
1230 *rv &= ~(PCIM_MFDEV << 16);
1232 *rv |= (PCIM_MFDEV << 16);
1239 static int cfgbus, cfgslot, cfgfunc, cfgoff;
1242 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1243 uint32_t *eax, void *arg)
1249 *eax = (bytes == 2) ? 0xffff : 0xff;
1254 x = (cfgbus << 16) |
1258 *eax = x | CONF1_ENABLE;
1261 cfgoff = x & PCI_REGMAX;
1262 cfgfunc = (x >> 8) & PCI_FUNCMAX;
1263 cfgslot = (x >> 11) & PCI_SLOTMAX;
1264 cfgbus = (x >> 16) & PCI_BUSMAX;
1269 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1272 bits_changed(uint32_t old, uint32_t new, uint32_t mask)
1275 return ((old ^ new) & mask);
1279 pci_emul_cmdwrite(struct pci_devinst *pi, uint32_t new, int bytes)
1285 * The command register is at an offset of 4 bytes and thus the
1286 * guest could write 1, 2 or 4 bytes starting at this offset.
1289 old = pci_get_cfgdata16(pi, PCIR_COMMAND); /* stash old value */
1290 CFGWRITE(pi, PCIR_COMMAND, new, bytes); /* update config */
1291 new = pci_get_cfgdata16(pi, PCIR_COMMAND); /* get updated value */
1294 * If the MMIO or I/O address space decoding has changed then
1295 * register/unregister all BARs that decode that address space.
1297 for (i = 0; i < PCI_BARMAX; i++) {
1298 switch (pi->pi_bar[i].type) {
1300 case PCIBAR_MEMHI64:
1303 /* I/O address space decoding changed? */
1304 if (bits_changed(old, new, PCIM_CMD_PORTEN)) {
1306 register_bar(pi, i);
1308 unregister_bar(pi, i);
1313 /* MMIO address space decoding changed? */
1314 if (bits_changed(old, new, PCIM_CMD_MEMEN)) {
1316 register_bar(pi, i);
1318 unregister_bar(pi, i);
1328 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1329 uint32_t *eax, void *arg)
1331 struct pci_devinst *pi;
1332 struct pci_devemu *pe;
1333 int coff, idx, needcfg;
1334 uint64_t addr, bar, mask;
1336 assert(bytes == 1 || bytes == 2 || bytes == 4);
1339 pi = pci_slotinfo[cfgslot][cfgfunc].si_devi;
1343 coff = cfgoff + (port - CONF1_DATA_PORT);
1346 printf("pcicfg-%s from 0x%0x of %d bytes (%d/%d/%d)\n\r",
1347 in ? "read" : "write", coff, bytes, cfgbus, cfgslot, cfgfunc);
1351 * Just return if there is no device at this cfgslot:cfgfunc or
1352 * if the guest is doing an un-aligned access
1354 if (pi == NULL || (coff & (bytes - 1)) != 0) {
1366 /* Let the device emulation override the default handler */
1367 if (pe->pe_cfgread != NULL) {
1368 needcfg = pe->pe_cfgread(ctx, vcpu, pi,
1376 *eax = pci_get_cfgdata8(pi, coff);
1377 else if (bytes == 2)
1378 *eax = pci_get_cfgdata16(pi, coff);
1380 *eax = pci_get_cfgdata32(pi, coff);
1383 pci_emul_hdrtype_fixup(cfgslot, coff, bytes, eax);
1385 /* Let the device emulation override the default handler */
1386 if (pe->pe_cfgwrite != NULL &&
1387 (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1391 * Special handling for write to BAR registers
1393 if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1395 * Ignore writes to BAR registers that are not
1398 if (bytes != 4 || (coff & 0x3) != 0)
1400 idx = (coff - PCIR_BAR(0)) / 4;
1401 mask = ~(pi->pi_bar[idx].size - 1);
1402 switch (pi->pi_bar[idx].type) {
1404 pi->pi_bar[idx].addr = bar = 0;
1409 bar = addr | PCIM_BAR_IO_SPACE;
1411 * Register the new BAR value for interception
1413 if (addr != pi->pi_bar[idx].addr) {
1414 update_bar_address(pi, addr, idx,
1419 addr = bar = *eax & mask;
1420 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1421 if (addr != pi->pi_bar[idx].addr) {
1422 update_bar_address(pi, addr, idx,
1427 addr = bar = *eax & mask;
1428 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1429 PCIM_BAR_MEM_PREFETCH;
1430 if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1431 update_bar_address(pi, addr, idx,
1435 case PCIBAR_MEMHI64:
1436 mask = ~(pi->pi_bar[idx - 1].size - 1);
1437 addr = ((uint64_t)*eax << 32) & mask;
1439 if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1440 update_bar_address(pi, addr, idx - 1,
1447 pci_set_cfgdata32(pi, coff, bar);
1449 } else if (pci_emul_iscap(pi, coff)) {
1450 pci_emul_capwrite(pi, coff, bytes, *eax);
1451 } else if (coff == PCIR_COMMAND) {
1452 pci_emul_cmdwrite(pi, *eax, bytes);
1454 CFGWRITE(pi, coff, *eax, bytes);
1461 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1462 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1463 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1464 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1467 * I/O ports to configure PCI IRQ routing. We ignore all writes to it.
1470 pci_irq_port_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1471 uint32_t *eax, void *arg)
1476 INOUT_PORT(pci_irq, 0xC00, IOPORT_F_OUT, pci_irq_port_handler);
1477 INOUT_PORT(pci_irq, 0xC01, IOPORT_F_OUT, pci_irq_port_handler);
1479 #define PCI_EMUL_TEST
1480 #ifdef PCI_EMUL_TEST
1482 * Define a dummy test device
1486 struct pci_emul_dsoftc {
1487 uint8_t ioregs[DIOSZ];
1488 uint8_t memregs[DMEMSZ];
1491 #define PCI_EMUL_MSI_MSGS 4
1492 #define PCI_EMUL_MSIX_MSGS 16
1495 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1498 struct pci_emul_dsoftc *sc;
1500 sc = malloc(sizeof(struct pci_emul_dsoftc));
1501 memset(sc, 0, sizeof(struct pci_emul_dsoftc));
1505 pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
1506 pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
1507 pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
1509 error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
1512 error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
1515 error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
1522 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1523 uint64_t offset, int size, uint64_t value)
1526 struct pci_emul_dsoftc *sc = pi->pi_arg;
1529 if (offset + size > DIOSZ) {
1530 printf("diow: iow too large, offset %ld size %d\n",
1536 sc->ioregs[offset] = value & 0xff;
1537 } else if (size == 2) {
1538 *(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
1539 } else if (size == 4) {
1540 *(uint32_t *)&sc->ioregs[offset] = value;
1542 printf("diow: iow unknown size %d\n", size);
1546 * Special magic value to generate an interrupt
1548 if (offset == 4 && size == 4 && pci_msi_enabled(pi))
1549 pci_generate_msi(pi, value % pci_msi_msgnum(pi));
1551 if (value == 0xabcdef) {
1552 for (i = 0; i < pci_msi_msgnum(pi); i++)
1553 pci_generate_msi(pi, i);
1558 if (offset + size > DMEMSZ) {
1559 printf("diow: memw too large, offset %ld size %d\n",
1565 sc->memregs[offset] = value;
1566 } else if (size == 2) {
1567 *(uint16_t *)&sc->memregs[offset] = value;
1568 } else if (size == 4) {
1569 *(uint32_t *)&sc->memregs[offset] = value;
1570 } else if (size == 8) {
1571 *(uint64_t *)&sc->memregs[offset] = value;
1573 printf("diow: memw unknown size %d\n", size);
1577 * magic interrupt ??
1582 printf("diow: unknown bar idx %d\n", baridx);
1587 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1588 uint64_t offset, int size)
1590 struct pci_emul_dsoftc *sc = pi->pi_arg;
1594 if (offset + size > DIOSZ) {
1595 printf("dior: ior too large, offset %ld size %d\n",
1601 value = sc->ioregs[offset];
1602 } else if (size == 2) {
1603 value = *(uint16_t *) &sc->ioregs[offset];
1604 } else if (size == 4) {
1605 value = *(uint32_t *) &sc->ioregs[offset];
1607 printf("dior: ior unknown size %d\n", size);
1612 if (offset + size > DMEMSZ) {
1613 printf("dior: memr too large, offset %ld size %d\n",
1619 value = sc->memregs[offset];
1620 } else if (size == 2) {
1621 value = *(uint16_t *) &sc->memregs[offset];
1622 } else if (size == 4) {
1623 value = *(uint32_t *) &sc->memregs[offset];
1624 } else if (size == 8) {
1625 value = *(uint64_t *) &sc->memregs[offset];
1627 printf("dior: ior unknown size %d\n", size);
1633 printf("dior: unknown bar idx %d\n", baridx);
1640 struct pci_devemu pci_dummy = {
1642 .pe_init = pci_emul_dinit,
1643 .pe_barwrite = pci_emul_diow,
1644 .pe_barread = pci_emul_dior
1646 PCI_EMUL_SET(pci_dummy);
1648 #endif /* PCI_EMUL_TEST */