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>
37 #include <vm/vm_param.h>
50 #include <machine/vmm.h>
51 #include <machine/vmm_snapshot.h>
52 #include <machine/cpufunc.h>
53 #include <machine/specialreg.h>
66 #define CONF1_ADDR_PORT 0x0cf8
67 #define CONF1_DATA_PORT 0x0cfc
69 #define CONF1_ENABLE 0x80000000ul
71 #define MAXBUSES (PCI_BUSMAX + 1)
72 #define MAXSLOTS (PCI_SLOTMAX + 1)
73 #define MAXFUNCS (PCI_FUNCMAX + 1)
78 struct pci_devinst *fi_devi;
88 struct intxinfo si_intpins[4];
89 struct funcinfo si_funcs[MAXFUNCS];
93 uint16_t iobase, iolimit; /* I/O window */
94 uint32_t membase32, memlimit32; /* mmio window below 4GB */
95 uint64_t membase64, memlimit64; /* mmio window above 4GB */
96 struct slotinfo slotinfo[MAXSLOTS];
99 static struct businfo *pci_businfo[MAXBUSES];
101 SET_DECLARE(pci_devemu_set, struct pci_devemu);
103 static uint64_t pci_emul_iobase;
104 static uint64_t pci_emul_membase32;
105 static uint64_t pci_emul_membase64;
106 static uint64_t pci_emul_memlim64;
108 #define PCI_EMUL_IOBASE 0x2000
109 #define PCI_EMUL_IOLIMIT 0x10000
111 #define PCI_EMUL_ECFG_BASE 0xE0000000 /* 3.5GB */
112 #define PCI_EMUL_ECFG_SIZE (MAXBUSES * 1024 * 1024) /* 1MB per bus */
113 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE);
115 #define PCI_EMUL_MEMLIMIT32 PCI_EMUL_ECFG_BASE
117 static struct pci_devemu *pci_emul_finddev(char *name);
118 static void pci_lintr_route(struct pci_devinst *pi);
119 static void pci_lintr_update(struct pci_devinst *pi);
120 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot,
121 int func, int coff, int bytes, uint32_t *val);
124 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes)
128 pci_set_cfgdata8(pi, coff, val);
130 pci_set_cfgdata16(pi, coff, val);
132 pci_set_cfgdata32(pi, coff, val);
135 static __inline uint32_t
136 CFGREAD(struct pci_devinst *pi, int coff, int bytes)
140 return (pci_get_cfgdata8(pi, coff));
142 return (pci_get_cfgdata16(pi, coff));
144 return (pci_get_cfgdata32(pi, coff));
152 * Slot options are in the form:
154 * <bus>:<slot>:<func>,<emul>[,<config>]
155 * <slot>[:<func>],<emul>[,<config>]
159 * emul is a string describing the type of PCI device e.g. virtio-net
160 * config is an optional string, depending on the device, that can be
161 * used for configuration.
167 pci_parse_slot_usage(char *aopt)
170 EPRINTLN("Invalid PCI slot info field \"%s\"", aopt);
174 pci_parse_slot(char *opt)
178 char *emul, *config, *str, *cp;
179 int error, bnum, snum, fnum;
184 emul = config = NULL;
185 if ((cp = strchr(str, ',')) != NULL) {
188 if ((cp = strchr(emul, ',')) != NULL) {
193 pci_parse_slot_usage(opt);
197 /* <bus>:<slot>:<func> */
198 if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) {
201 if (sscanf(str, "%d:%d", &snum, &fnum) != 2) {
204 if (sscanf(str, "%d", &snum) != 1) {
210 if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS ||
211 fnum < 0 || fnum >= MAXFUNCS) {
212 pci_parse_slot_usage(opt);
216 if (pci_businfo[bnum] == NULL)
217 pci_businfo[bnum] = calloc(1, sizeof(struct businfo));
219 bi = pci_businfo[bnum];
220 si = &bi->slotinfo[snum];
222 if (si->si_funcs[fnum].fi_name != NULL) {
223 EPRINTLN("pci slot %d:%d already occupied!",
228 if (pci_emul_finddev(emul) == NULL) {
229 EPRINTLN("pci slot %d:%d: unknown device \"%s\"",
235 si->si_funcs[fnum].fi_name = emul;
236 si->si_funcs[fnum].fi_param = config;
246 pci_print_supported_devices()
248 struct pci_devemu **pdpp, *pdp;
250 SET_FOREACH(pdpp, pci_devemu_set) {
252 printf("%s\n", pdp->pe_emu);
257 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset)
260 if (offset < pi->pi_msix.pba_offset)
263 if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) {
271 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size,
274 int msix_entry_offset;
278 /* support only 4 or 8 byte writes */
279 if (size != 4 && size != 8)
283 * Return if table index is beyond what device supports
285 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
286 if (tab_index >= pi->pi_msix.table_count)
289 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
291 /* support only aligned writes */
292 if ((msix_entry_offset % size) != 0)
295 dest = (char *)(pi->pi_msix.table + tab_index);
296 dest += msix_entry_offset;
299 *((uint32_t *)dest) = value;
301 *((uint64_t *)dest) = value;
307 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size)
310 int msix_entry_offset;
312 uint64_t retval = ~0;
315 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X
316 * table but we also allow 1 byte access to accommodate reads from
319 if (size != 1 && size != 4 && size != 8)
322 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE;
324 /* support only aligned reads */
325 if ((msix_entry_offset % size) != 0) {
329 tab_index = offset / MSIX_TABLE_ENTRY_SIZE;
331 if (tab_index < pi->pi_msix.table_count) {
332 /* valid MSI-X Table access */
333 dest = (char *)(pi->pi_msix.table + tab_index);
334 dest += msix_entry_offset;
337 retval = *((uint8_t *)dest);
339 retval = *((uint32_t *)dest);
341 retval = *((uint64_t *)dest);
342 } else if (pci_valid_pba_offset(pi, offset)) {
343 /* return 0 for PBA access */
351 pci_msix_table_bar(struct pci_devinst *pi)
354 if (pi->pi_msix.table != NULL)
355 return (pi->pi_msix.table_bar);
361 pci_msix_pba_bar(struct pci_devinst *pi)
364 if (pi->pi_msix.table != NULL)
365 return (pi->pi_msix.pba_bar);
371 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
372 uint32_t *eax, void *arg)
374 struct pci_devinst *pdi = arg;
375 struct pci_devemu *pe = pdi->pi_d;
379 for (i = 0; i <= PCI_BARMAX; i++) {
380 if (pdi->pi_bar[i].type == PCIBAR_IO &&
381 port >= pdi->pi_bar[i].addr &&
382 port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) {
383 offset = port - pdi->pi_bar[i].addr;
385 *eax = (*pe->pe_barread)(ctx, vcpu, pdi, i,
388 (*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset,
397 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
398 int size, uint64_t *val, void *arg1, long arg2)
400 struct pci_devinst *pdi = arg1;
401 struct pci_devemu *pe = pdi->pi_d;
403 int bidx = (int) arg2;
405 assert(bidx <= PCI_BARMAX);
406 assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 ||
407 pdi->pi_bar[bidx].type == PCIBAR_MEM64);
408 assert(addr >= pdi->pi_bar[bidx].addr &&
409 addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size);
411 offset = addr - pdi->pi_bar[bidx].addr;
413 if (dir == MEM_F_WRITE) {
415 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
416 4, *val & 0xffffffff);
417 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4,
420 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset,
425 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
427 *val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
428 offset + 4, 4) << 32;
430 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx,
440 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size,
445 assert((size & (size - 1)) == 0); /* must be a power of 2 */
447 base = roundup2(*baseptr, size);
449 if (base + size <= limit) {
451 *baseptr = base + size;
458 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type,
462 return (pci_emul_alloc_pbar(pdi, idx, 0, type, size));
466 * Register (or unregister) the MMIO or I/O region associated with the BAR
467 * register 'idx' of an emulated pci device.
470 modify_bar_registration(struct pci_devinst *pi, int idx, int registration)
473 struct inout_port iop;
476 switch (pi->pi_bar[idx].type) {
478 bzero(&iop, sizeof(struct inout_port));
479 iop.name = pi->pi_name;
480 iop.port = pi->pi_bar[idx].addr;
481 iop.size = pi->pi_bar[idx].size;
483 iop.flags = IOPORT_F_INOUT;
484 iop.handler = pci_emul_io_handler;
486 error = register_inout(&iop);
488 error = unregister_inout(&iop);
492 bzero(&mr, sizeof(struct mem_range));
493 mr.name = pi->pi_name;
494 mr.base = pi->pi_bar[idx].addr;
495 mr.size = pi->pi_bar[idx].size;
498 mr.handler = pci_emul_mem_handler;
501 error = register_mem(&mr);
503 error = unregister_mem(&mr);
513 unregister_bar(struct pci_devinst *pi, int idx)
516 modify_bar_registration(pi, idx, 0);
520 register_bar(struct pci_devinst *pi, int idx)
523 modify_bar_registration(pi, idx, 1);
526 /* Are we decoding i/o port accesses for the emulated pci device? */
528 porten(struct pci_devinst *pi)
532 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
534 return (cmd & PCIM_CMD_PORTEN);
537 /* Are we decoding memory accesses for the emulated pci device? */
539 memen(struct pci_devinst *pi)
543 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
545 return (cmd & PCIM_CMD_MEMEN);
549 * Update the MMIO or I/O address that is decoded by the BAR register.
551 * If the pci device has enabled the address space decoding then intercept
552 * the address range decoded by the BAR register.
555 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type)
559 if (pi->pi_bar[idx].type == PCIBAR_IO)
565 unregister_bar(pi, idx);
570 pi->pi_bar[idx].addr = addr;
573 pi->pi_bar[idx].addr &= ~0xffffffffUL;
574 pi->pi_bar[idx].addr |= addr;
577 pi->pi_bar[idx].addr &= 0xffffffff;
578 pi->pi_bar[idx].addr |= addr;
585 register_bar(pi, idx);
589 pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase,
590 enum pcibar_type type, uint64_t size)
593 uint64_t *baseptr, limit, addr, mask, lobits, bar;
596 assert(idx >= 0 && idx <= PCI_BARMAX);
598 if ((size & (size - 1)) != 0)
599 size = 1UL << flsl(size); /* round up to a power of 2 */
601 /* Enforce minimum BAR sizes required by the PCI standard */
602 if (type == PCIBAR_IO) {
613 addr = mask = lobits = enbit = 0;
616 baseptr = &pci_emul_iobase;
617 limit = PCI_EMUL_IOLIMIT;
618 mask = PCIM_BAR_IO_BASE;
619 lobits = PCIM_BAR_IO_SPACE;
620 enbit = PCIM_CMD_PORTEN;
625 * Some drivers do not work well if the 64-bit BAR is allocated
626 * above 4GB. Allow for this by allocating small requests under
627 * 4GB unless then allocation size is larger than some arbitrary
628 * number (32MB currently).
630 if (size > 32 * 1024 * 1024) {
632 * XXX special case for device requiring peer-peer DMA
634 if (size == 0x100000000UL)
637 baseptr = &pci_emul_membase64;
638 limit = pci_emul_memlim64;
639 mask = PCIM_BAR_MEM_BASE;
640 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
641 PCIM_BAR_MEM_PREFETCH;
643 baseptr = &pci_emul_membase32;
644 limit = PCI_EMUL_MEMLIMIT32;
645 mask = PCIM_BAR_MEM_BASE;
646 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64;
648 enbit = PCIM_CMD_MEMEN;
651 baseptr = &pci_emul_membase32;
652 limit = PCI_EMUL_MEMLIMIT32;
653 mask = PCIM_BAR_MEM_BASE;
654 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
655 enbit = PCIM_CMD_MEMEN;
658 printf("pci_emul_alloc_base: invalid bar type %d\n", type);
662 if (baseptr != NULL) {
663 error = pci_emul_alloc_resource(baseptr, limit, size, &addr);
668 pdi->pi_bar[idx].type = type;
669 pdi->pi_bar[idx].addr = addr;
670 pdi->pi_bar[idx].size = size;
672 /* Initialize the BAR register in config space */
673 bar = (addr & mask) | lobits;
674 pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar);
676 if (type == PCIBAR_MEM64) {
677 assert(idx + 1 <= PCI_BARMAX);
678 pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64;
679 pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32);
682 cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND);
683 if ((cmd & enbit) != enbit)
684 pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit);
685 register_bar(pdi, idx);
690 #define CAP_START_OFFSET 0x40
692 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen)
694 int i, capoff, reallen;
699 reallen = roundup2(caplen, 4); /* dword aligned */
701 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
702 if ((sts & PCIM_STATUS_CAPPRESENT) == 0)
703 capoff = CAP_START_OFFSET;
705 capoff = pi->pi_capend + 1;
707 /* Check if we have enough space */
708 if (capoff + reallen > PCI_REGMAX + 1)
711 /* Set the previous capability pointer */
712 if ((sts & PCIM_STATUS_CAPPRESENT) == 0) {
713 pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff);
714 pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT);
716 pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff);
718 /* Copy the capability */
719 for (i = 0; i < caplen; i++)
720 pci_set_cfgdata8(pi, capoff + i, capdata[i]);
722 /* Set the next capability pointer */
723 pci_set_cfgdata8(pi, capoff + 1, 0);
725 pi->pi_prevcap = capoff;
726 pi->pi_capend = capoff + reallen - 1;
730 static struct pci_devemu *
731 pci_emul_finddev(char *name)
733 struct pci_devemu **pdpp, *pdp;
735 SET_FOREACH(pdpp, pci_devemu_set) {
737 if (!strcmp(pdp->pe_emu, name)) {
746 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot,
747 int func, struct funcinfo *fi)
749 struct pci_devinst *pdi;
752 pdi = calloc(1, sizeof(struct pci_devinst));
758 pthread_mutex_init(&pdi->pi_lintr.lock, NULL);
759 pdi->pi_lintr.pin = 0;
760 pdi->pi_lintr.state = IDLE;
761 pdi->pi_lintr.pirq_pin = 0;
762 pdi->pi_lintr.ioapic_irq = 0;
764 snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot);
766 /* Disable legacy interrupts */
767 pci_set_cfgdata8(pdi, PCIR_INTLINE, 255);
768 pci_set_cfgdata8(pdi, PCIR_INTPIN, 0);
770 pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN);
772 err = (*pde->pe_init)(ctx, pdi, fi->fi_param);
782 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr)
786 /* Number of msi messages must be a power of 2 between 1 and 32 */
787 assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32);
788 mmc = ffs(msgnum) - 1;
790 bzero(msicap, sizeof(struct msicap));
791 msicap->capid = PCIY_MSI;
792 msicap->nextptr = nextptr;
793 msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1);
797 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum)
799 struct msicap msicap;
801 pci_populate_msicap(&msicap, msgnum, 0);
803 return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap)));
807 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum,
808 uint32_t msix_tab_size)
811 assert(msix_tab_size % 4096 == 0);
813 bzero(msixcap, sizeof(struct msixcap));
814 msixcap->capid = PCIY_MSIX;
817 * Message Control Register, all fields set to
818 * zero except for the Table Size.
819 * Note: Table size N is encoded as N-1
821 msixcap->msgctrl = msgnum - 1;
825 * - MSI-X table start at offset 0
826 * - PBA table starts at a 4K aligned offset after the MSI-X table
828 msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK;
829 msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK);
833 pci_msix_table_init(struct pci_devinst *pi, int table_entries)
837 assert(table_entries > 0);
838 assert(table_entries <= MAX_MSIX_TABLE_ENTRIES);
840 table_size = table_entries * MSIX_TABLE_ENTRY_SIZE;
841 pi->pi_msix.table = calloc(1, table_size);
843 /* set mask bit of vector control register */
844 for (i = 0; i < table_entries; i++)
845 pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK;
849 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum)
852 struct msixcap msixcap;
854 assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES);
855 assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0);
857 tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE;
859 /* Align table size to nearest 4K */
860 tab_size = roundup2(tab_size, 4096);
862 pi->pi_msix.table_bar = barnum;
863 pi->pi_msix.pba_bar = barnum;
864 pi->pi_msix.table_offset = 0;
865 pi->pi_msix.table_count = msgnum;
866 pi->pi_msix.pba_offset = tab_size;
867 pi->pi_msix.pba_size = PBA_SIZE(msgnum);
869 pci_msix_table_init(pi, msgnum);
871 pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size);
873 /* allocate memory for MSI-X Table and PBA */
874 pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32,
875 tab_size + pi->pi_msix.pba_size);
877 return (pci_emul_add_capability(pi, (u_char *)&msixcap,
882 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
883 int bytes, uint32_t val)
885 uint16_t msgctrl, rwmask;
888 off = offset - capoff;
889 /* Message Control Register */
890 if (off == 2 && bytes == 2) {
891 rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK;
892 msgctrl = pci_get_cfgdata16(pi, offset);
894 msgctrl |= val & rwmask;
897 pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE;
898 pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK;
899 pci_lintr_update(pi);
902 CFGWRITE(pi, offset, val, bytes);
906 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
907 int bytes, uint32_t val)
909 uint16_t msgctrl, rwmask, msgdata, mme;
913 * If guest is writing to the message control register make sure
914 * we do not overwrite read-only fields.
916 if ((offset - capoff) == 2 && bytes == 2) {
917 rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE;
918 msgctrl = pci_get_cfgdata16(pi, offset);
920 msgctrl |= val & rwmask;
923 CFGWRITE(pi, offset, val, bytes);
925 msgctrl = pci_get_cfgdata16(pi, capoff + 2);
926 addrlo = pci_get_cfgdata32(pi, capoff + 4);
927 if (msgctrl & PCIM_MSICTRL_64BIT)
928 msgdata = pci_get_cfgdata16(pi, capoff + 12);
930 msgdata = pci_get_cfgdata16(pi, capoff + 8);
932 mme = msgctrl & PCIM_MSICTRL_MME_MASK;
933 pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0;
934 if (pi->pi_msi.enabled) {
935 pi->pi_msi.addr = addrlo;
936 pi->pi_msi.msg_data = msgdata;
937 pi->pi_msi.maxmsgnum = 1 << (mme >> 4);
939 pi->pi_msi.maxmsgnum = 0;
941 pci_lintr_update(pi);
945 pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset,
946 int bytes, uint32_t val)
949 /* XXX don't write to the readonly parts */
950 CFGWRITE(pi, offset, val, bytes);
953 #define PCIECAP_VERSION 0x2
955 pci_emul_add_pciecap(struct pci_devinst *pi, int type)
958 struct pciecap pciecap;
960 bzero(&pciecap, sizeof(pciecap));
963 * Use the integrated endpoint type for endpoints on a root complex bus.
965 * NB: bhyve currently only supports a single PCI bus that is the root
966 * complex bus, so all endpoints are integrated.
968 if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0))
969 type = PCIEM_TYPE_ROOT_INT_EP;
971 pciecap.capid = PCIY_EXPRESS;
972 pciecap.pcie_capabilities = PCIECAP_VERSION | type;
973 if (type != PCIEM_TYPE_ROOT_INT_EP) {
974 pciecap.link_capabilities = 0x411; /* gen1, x1 */
975 pciecap.link_status = 0x11; /* gen1, x1 */
978 err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap));
983 * This function assumes that 'coff' is in the capabilities region of the
984 * config space. A capoff parameter of zero will force a search for the
988 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val,
989 uint8_t capoff, int capid)
993 /* Do not allow un-aligned writes */
994 if ((offset & (bytes - 1)) != 0)
998 /* Find the capability that we want to update */
999 capoff = CAP_START_OFFSET;
1001 nextoff = pci_get_cfgdata8(pi, capoff + 1);
1004 if (offset >= capoff && offset < nextoff)
1009 assert(offset >= capoff);
1010 capid = pci_get_cfgdata8(pi, capoff);
1014 * Capability ID and Next Capability Pointer are readonly.
1015 * However, some o/s's do 4-byte writes that include these.
1016 * For this case, trim the write back to 2 bytes and adjust
1019 if (offset == capoff || offset == capoff + 1) {
1020 if (offset == capoff && bytes == 4) {
1030 msicap_cfgwrite(pi, capoff, offset, bytes, val);
1033 msixcap_cfgwrite(pi, capoff, offset, bytes, val);
1036 pciecap_cfgwrite(pi, capoff, offset, bytes, val);
1044 pci_emul_iscap(struct pci_devinst *pi, int offset)
1048 sts = pci_get_cfgdata16(pi, PCIR_STATUS);
1049 if ((sts & PCIM_STATUS_CAPPRESENT) != 0) {
1050 if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend)
1057 pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1058 int size, uint64_t *val, void *arg1, long arg2)
1061 * Ignore writes; return 0xff's for reads. The mem read code
1062 * will take care of truncating to the correct size.
1064 if (dir == MEM_F_READ) {
1065 *val = 0xffffffffffffffff;
1072 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr,
1073 int bytes, uint64_t *val, void *arg1, long arg2)
1075 int bus, slot, func, coff, in;
1077 coff = addr & 0xfff;
1078 func = (addr >> 12) & 0x7;
1079 slot = (addr >> 15) & 0x1f;
1080 bus = (addr >> 20) & 0xff;
1081 in = (dir == MEM_F_READ);
1084 pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val);
1092 return (PCI_EMUL_ECFG_BASE);
1095 #define BUSIO_ROUNDUP 32
1096 #define BUSMEM_ROUNDUP (1024 * 1024)
1099 init_pci(struct vmctx *ctx)
1101 struct mem_range mr;
1102 struct pci_devemu *pde;
1104 struct slotinfo *si;
1105 struct funcinfo *fi;
1107 uint64_t cpu_maxphysaddr, pci_emul_memresv64;
1109 int bus, slot, func, error;
1111 pci_emul_iobase = PCI_EMUL_IOBASE;
1112 pci_emul_membase32 = vm_get_lowmem_limit(ctx);
1114 do_cpuid(0x80000008, regs);
1115 cpu_maxphysaddr = 1ULL << (regs[0] & 0xff);
1116 if (cpu_maxphysaddr > VM_MAXUSER_ADDRESS_LA48)
1117 cpu_maxphysaddr = VM_MAXUSER_ADDRESS_LA48;
1118 pci_emul_memresv64 = cpu_maxphysaddr / 4;
1120 * Max power of 2 that is less then
1121 * cpu_maxphysaddr - pci_emul_memresv64.
1123 pci_emul_membase64 = 1ULL << (flsl(cpu_maxphysaddr -
1124 pci_emul_memresv64) - 1);
1125 pci_emul_memlim64 = cpu_maxphysaddr;
1127 for (bus = 0; bus < MAXBUSES; bus++) {
1128 if ((bi = pci_businfo[bus]) == NULL)
1131 * Keep track of the i/o and memory resources allocated to
1134 bi->iobase = pci_emul_iobase;
1135 bi->membase32 = pci_emul_membase32;
1136 bi->membase64 = pci_emul_membase64;
1138 for (slot = 0; slot < MAXSLOTS; slot++) {
1139 si = &bi->slotinfo[slot];
1140 for (func = 0; func < MAXFUNCS; func++) {
1141 fi = &si->si_funcs[func];
1142 if (fi->fi_name == NULL)
1144 pde = pci_emul_finddev(fi->fi_name);
1145 assert(pde != NULL);
1146 error = pci_emul_init(ctx, pde, bus, slot,
1154 * Add some slop to the I/O and memory resources decoded by
1155 * this bus to give a guest some flexibility if it wants to
1156 * reprogram the BARs.
1158 pci_emul_iobase += BUSIO_ROUNDUP;
1159 pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP);
1160 bi->iolimit = pci_emul_iobase;
1162 pci_emul_membase32 += BUSMEM_ROUNDUP;
1163 pci_emul_membase32 = roundup2(pci_emul_membase32,
1165 bi->memlimit32 = pci_emul_membase32;
1167 pci_emul_membase64 += BUSMEM_ROUNDUP;
1168 pci_emul_membase64 = roundup2(pci_emul_membase64,
1170 bi->memlimit64 = pci_emul_membase64;
1174 * PCI backends are initialized before routing INTx interrupts
1175 * so that LPC devices are able to reserve ISA IRQs before
1176 * routing PIRQ pins.
1178 for (bus = 0; bus < MAXBUSES; bus++) {
1179 if ((bi = pci_businfo[bus]) == NULL)
1182 for (slot = 0; slot < MAXSLOTS; slot++) {
1183 si = &bi->slotinfo[slot];
1184 for (func = 0; func < MAXFUNCS; func++) {
1185 fi = &si->si_funcs[func];
1186 if (fi->fi_devi == NULL)
1188 pci_lintr_route(fi->fi_devi);
1195 * The guest physical memory map looks like the following:
1196 * [0, lowmem) guest system memory
1197 * [lowmem, lowmem_limit) memory hole (may be absent)
1198 * [lowmem_limit, 0xE0000000) PCI hole (32-bit BAR allocation)
1199 * [0xE0000000, 0xF0000000) PCI extended config window
1200 * [0xF0000000, 4GB) LAPIC, IOAPIC, HPET, firmware
1201 * [4GB, 4GB + highmem)
1205 * Accesses to memory addresses that are not allocated to system
1206 * memory or PCI devices return 0xff's.
1208 lowmem = vm_get_lowmem_size(ctx);
1209 bzero(&mr, sizeof(struct mem_range));
1210 mr.name = "PCI hole";
1211 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1213 mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem;
1214 mr.handler = pci_emul_fallback_handler;
1215 error = register_mem_fallback(&mr);
1218 /* PCI extended config space */
1219 bzero(&mr, sizeof(struct mem_range));
1220 mr.name = "PCI ECFG";
1221 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE;
1222 mr.base = PCI_EMUL_ECFG_BASE;
1223 mr.size = PCI_EMUL_ECFG_SIZE;
1224 mr.handler = pci_emul_ecfg_handler;
1225 error = register_mem(&mr);
1232 pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1236 dsdt_line(" Package ()");
1238 dsdt_line(" 0x%X,", slot << 16 | 0xffff);
1239 dsdt_line(" 0x%02X,", pin - 1);
1240 dsdt_line(" Zero,");
1241 dsdt_line(" 0x%X", ioapic_irq);
1246 pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq,
1251 name = lpc_pirq_name(pirq_pin);
1254 dsdt_line(" Package ()");
1256 dsdt_line(" 0x%X,", slot << 16 | 0xffff);
1257 dsdt_line(" 0x%02X,", pin - 1);
1258 dsdt_line(" %s,", name);
1265 * A bhyve virtual machine has a flat PCI hierarchy with a root port
1266 * corresponding to each PCI bus.
1269 pci_bus_write_dsdt(int bus)
1272 struct slotinfo *si;
1273 struct pci_devinst *pi;
1274 int count, func, slot;
1277 * If there are no devices on this 'bus' then just return.
1279 if ((bi = pci_businfo[bus]) == NULL) {
1281 * Bus 0 is special because it decodes the I/O ports used
1282 * for PCI config space access even if there are no devices
1289 dsdt_line(" Device (PC%02X)", bus);
1291 dsdt_line(" Name (_HID, EisaId (\"PNP0A03\"))");
1293 dsdt_line(" Method (_BBN, 0, NotSerialized)");
1295 dsdt_line(" Return (0x%08X)", bus);
1297 dsdt_line(" Name (_CRS, ResourceTemplate ()");
1299 dsdt_line(" WordBusNumber (ResourceProducer, MinFixed, "
1300 "MaxFixed, PosDecode,");
1301 dsdt_line(" 0x0000, // Granularity");
1302 dsdt_line(" 0x%04X, // Range Minimum", bus);
1303 dsdt_line(" 0x%04X, // Range Maximum", bus);
1304 dsdt_line(" 0x0000, // Translation Offset");
1305 dsdt_line(" 0x0001, // Length");
1310 dsdt_fixed_ioport(0xCF8, 8);
1313 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1314 "PosDecode, EntireRange,");
1315 dsdt_line(" 0x0000, // Granularity");
1316 dsdt_line(" 0x0000, // Range Minimum");
1317 dsdt_line(" 0x0CF7, // Range Maximum");
1318 dsdt_line(" 0x0000, // Translation Offset");
1319 dsdt_line(" 0x0CF8, // Length");
1320 dsdt_line(" ,, , TypeStatic)");
1322 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1323 "PosDecode, EntireRange,");
1324 dsdt_line(" 0x0000, // Granularity");
1325 dsdt_line(" 0x0D00, // Range Minimum");
1326 dsdt_line(" 0x%04X, // Range Maximum",
1327 PCI_EMUL_IOBASE - 1);
1328 dsdt_line(" 0x0000, // Translation Offset");
1329 dsdt_line(" 0x%04X, // Length",
1330 PCI_EMUL_IOBASE - 0x0D00);
1331 dsdt_line(" ,, , TypeStatic)");
1341 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, "
1342 "PosDecode, EntireRange,");
1343 dsdt_line(" 0x0000, // Granularity");
1344 dsdt_line(" 0x%04X, // Range Minimum", bi->iobase);
1345 dsdt_line(" 0x%04X, // Range Maximum",
1347 dsdt_line(" 0x0000, // Translation Offset");
1348 dsdt_line(" 0x%04X, // Length",
1349 bi->iolimit - bi->iobase);
1350 dsdt_line(" ,, , TypeStatic)");
1352 /* mmio window (32-bit) */
1353 dsdt_line(" DWordMemory (ResourceProducer, PosDecode, "
1354 "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1355 dsdt_line(" 0x00000000, // Granularity");
1356 dsdt_line(" 0x%08X, // Range Minimum\n", bi->membase32);
1357 dsdt_line(" 0x%08X, // Range Maximum\n",
1358 bi->memlimit32 - 1);
1359 dsdt_line(" 0x00000000, // Translation Offset");
1360 dsdt_line(" 0x%08X, // Length\n",
1361 bi->memlimit32 - bi->membase32);
1362 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)");
1364 /* mmio window (64-bit) */
1365 dsdt_line(" QWordMemory (ResourceProducer, PosDecode, "
1366 "MinFixed, MaxFixed, NonCacheable, ReadWrite,");
1367 dsdt_line(" 0x0000000000000000, // Granularity");
1368 dsdt_line(" 0x%016lX, // Range Minimum\n", bi->membase64);
1369 dsdt_line(" 0x%016lX, // Range Maximum\n",
1370 bi->memlimit64 - 1);
1371 dsdt_line(" 0x0000000000000000, // Translation Offset");
1372 dsdt_line(" 0x%016lX, // Length\n",
1373 bi->memlimit64 - bi->membase64);
1374 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)");
1377 count = pci_count_lintr(bus);
1380 dsdt_line("Name (PPRT, Package ()");
1382 pci_walk_lintr(bus, pci_pirq_prt_entry, NULL);
1384 dsdt_line("Name (APRT, Package ()");
1386 pci_walk_lintr(bus, pci_apic_prt_entry, NULL);
1388 dsdt_line("Method (_PRT, 0, NotSerialized)");
1390 dsdt_line(" If (PICM)");
1392 dsdt_line(" Return (APRT)");
1396 dsdt_line(" Return (PPRT)");
1403 for (slot = 0; slot < MAXSLOTS; slot++) {
1404 si = &bi->slotinfo[slot];
1405 for (func = 0; func < MAXFUNCS; func++) {
1406 pi = si->si_funcs[func].fi_devi;
1407 if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL)
1408 pi->pi_d->pe_write_dsdt(pi);
1417 pci_write_dsdt(void)
1422 dsdt_line("Name (PICM, 0x00)");
1423 dsdt_line("Method (_PIC, 1, NotSerialized)");
1425 dsdt_line(" Store (Arg0, PICM)");
1428 dsdt_line("Scope (_SB)");
1430 for (bus = 0; bus < MAXBUSES; bus++)
1431 pci_bus_write_dsdt(bus);
1437 pci_bus_configured(int bus)
1439 assert(bus >= 0 && bus < MAXBUSES);
1440 return (pci_businfo[bus] != NULL);
1444 pci_msi_enabled(struct pci_devinst *pi)
1446 return (pi->pi_msi.enabled);
1450 pci_msi_maxmsgnum(struct pci_devinst *pi)
1452 if (pi->pi_msi.enabled)
1453 return (pi->pi_msi.maxmsgnum);
1459 pci_msix_enabled(struct pci_devinst *pi)
1462 return (pi->pi_msix.enabled && !pi->pi_msi.enabled);
1466 pci_generate_msix(struct pci_devinst *pi, int index)
1468 struct msix_table_entry *mte;
1470 if (!pci_msix_enabled(pi))
1473 if (pi->pi_msix.function_mask)
1476 if (index >= pi->pi_msix.table_count)
1479 mte = &pi->pi_msix.table[index];
1480 if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
1481 /* XXX Set PBA bit if interrupt is disabled */
1482 vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data);
1487 pci_generate_msi(struct pci_devinst *pi, int index)
1490 if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) {
1491 vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr,
1492 pi->pi_msi.msg_data + index);
1497 pci_lintr_permitted(struct pci_devinst *pi)
1501 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND);
1502 return (!(pi->pi_msi.enabled || pi->pi_msix.enabled ||
1503 (cmd & PCIM_CMD_INTxDIS)));
1507 pci_lintr_request(struct pci_devinst *pi)
1510 struct slotinfo *si;
1511 int bestpin, bestcount, pin;
1513 bi = pci_businfo[pi->pi_bus];
1517 * Just allocate a pin from our slot. The pin will be
1518 * assigned IRQs later when interrupts are routed.
1520 si = &bi->slotinfo[pi->pi_slot];
1522 bestcount = si->si_intpins[0].ii_count;
1523 for (pin = 1; pin < 4; pin++) {
1524 if (si->si_intpins[pin].ii_count < bestcount) {
1526 bestcount = si->si_intpins[pin].ii_count;
1530 si->si_intpins[bestpin].ii_count++;
1531 pi->pi_lintr.pin = bestpin + 1;
1532 pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1);
1536 pci_lintr_route(struct pci_devinst *pi)
1539 struct intxinfo *ii;
1541 if (pi->pi_lintr.pin == 0)
1544 bi = pci_businfo[pi->pi_bus];
1546 ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1];
1549 * Attempt to allocate an I/O APIC pin for this intpin if one
1550 * is not yet assigned.
1552 if (ii->ii_ioapic_irq == 0)
1553 ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi);
1554 assert(ii->ii_ioapic_irq > 0);
1557 * Attempt to allocate a PIRQ pin for this intpin if one is
1560 if (ii->ii_pirq_pin == 0)
1561 ii->ii_pirq_pin = pirq_alloc_pin(pi);
1562 assert(ii->ii_pirq_pin > 0);
1564 pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq;
1565 pi->pi_lintr.pirq_pin = ii->ii_pirq_pin;
1566 pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin));
1570 pci_lintr_assert(struct pci_devinst *pi)
1573 assert(pi->pi_lintr.pin > 0);
1575 pthread_mutex_lock(&pi->pi_lintr.lock);
1576 if (pi->pi_lintr.state == IDLE) {
1577 if (pci_lintr_permitted(pi)) {
1578 pi->pi_lintr.state = ASSERTED;
1581 pi->pi_lintr.state = PENDING;
1583 pthread_mutex_unlock(&pi->pi_lintr.lock);
1587 pci_lintr_deassert(struct pci_devinst *pi)
1590 assert(pi->pi_lintr.pin > 0);
1592 pthread_mutex_lock(&pi->pi_lintr.lock);
1593 if (pi->pi_lintr.state == ASSERTED) {
1594 pi->pi_lintr.state = IDLE;
1595 pci_irq_deassert(pi);
1596 } else if (pi->pi_lintr.state == PENDING)
1597 pi->pi_lintr.state = IDLE;
1598 pthread_mutex_unlock(&pi->pi_lintr.lock);
1602 pci_lintr_update(struct pci_devinst *pi)
1605 pthread_mutex_lock(&pi->pi_lintr.lock);
1606 if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) {
1607 pci_irq_deassert(pi);
1608 pi->pi_lintr.state = PENDING;
1609 } else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) {
1610 pi->pi_lintr.state = ASSERTED;
1613 pthread_mutex_unlock(&pi->pi_lintr.lock);
1617 pci_count_lintr(int bus)
1619 int count, slot, pin;
1620 struct slotinfo *slotinfo;
1623 if (pci_businfo[bus] != NULL) {
1624 for (slot = 0; slot < MAXSLOTS; slot++) {
1625 slotinfo = &pci_businfo[bus]->slotinfo[slot];
1626 for (pin = 0; pin < 4; pin++) {
1627 if (slotinfo->si_intpins[pin].ii_count != 0)
1636 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg)
1639 struct slotinfo *si;
1640 struct intxinfo *ii;
1643 if ((bi = pci_businfo[bus]) == NULL)
1646 for (slot = 0; slot < MAXSLOTS; slot++) {
1647 si = &bi->slotinfo[slot];
1648 for (pin = 0; pin < 4; pin++) {
1649 ii = &si->si_intpins[pin];
1650 if (ii->ii_count != 0)
1651 cb(bus, slot, pin + 1, ii->ii_pirq_pin,
1652 ii->ii_ioapic_irq, arg);
1658 * Return 1 if the emulated device in 'slot' is a multi-function device.
1659 * Return 0 otherwise.
1662 pci_emul_is_mfdev(int bus, int slot)
1665 struct slotinfo *si;
1669 if ((bi = pci_businfo[bus]) != NULL) {
1670 si = &bi->slotinfo[slot];
1671 for (f = 0; f < MAXFUNCS; f++) {
1672 if (si->si_funcs[f].fi_devi != NULL) {
1677 return (numfuncs > 1);
1681 * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on
1682 * whether or not is a multi-function being emulated in the pci 'slot'.
1685 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv)
1689 if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) {
1690 mfdev = pci_emul_is_mfdev(bus, slot);
1700 *rv &= ~(PCIM_MFDEV << 16);
1702 *rv |= (PCIM_MFDEV << 16);
1710 * Update device state in response to changes to the PCI command
1714 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old)
1717 uint16_t changed, new;
1719 new = pci_get_cfgdata16(pi, PCIR_COMMAND);
1720 changed = old ^ new;
1723 * If the MMIO or I/O address space decoding has changed then
1724 * register/unregister all BARs that decode that address space.
1726 for (i = 0; i <= PCI_BARMAX; i++) {
1727 switch (pi->pi_bar[i].type) {
1729 case PCIBAR_MEMHI64:
1732 /* I/O address space decoding changed? */
1733 if (changed & PCIM_CMD_PORTEN) {
1734 if (new & PCIM_CMD_PORTEN)
1735 register_bar(pi, i);
1737 unregister_bar(pi, i);
1742 /* MMIO address space decoding changed? */
1743 if (changed & PCIM_CMD_MEMEN) {
1744 if (new & PCIM_CMD_MEMEN)
1745 register_bar(pi, i);
1747 unregister_bar(pi, i);
1756 * If INTx has been unmasked and is pending, assert the
1759 pci_lintr_update(pi);
1763 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes)
1766 uint32_t cmd, old, readonly;
1768 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); /* stash old value */
1771 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3.
1773 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are
1774 * 'write 1 to clear'. However these bits are not set to '1' by
1775 * any device emulation so it is simpler to treat them as readonly.
1777 rshift = (coff & 0x3) * 8;
1778 readonly = 0xFFFFF880 >> rshift;
1780 old = CFGREAD(pi, coff, bytes);
1782 new |= (old & readonly);
1783 CFGWRITE(pi, coff, new, bytes); /* update config */
1785 pci_emul_cmd_changed(pi, cmd);
1789 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func,
1790 int coff, int bytes, uint32_t *eax)
1793 struct slotinfo *si;
1794 struct pci_devinst *pi;
1795 struct pci_devemu *pe;
1797 uint64_t addr, bar, mask;
1799 if ((bi = pci_businfo[bus]) != NULL) {
1800 si = &bi->slotinfo[slot];
1801 pi = si->si_funcs[func].fi_devi;
1806 * Just return if there is no device at this slot:func or if the
1807 * the guest is doing an un-aligned access.
1809 if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) ||
1810 (coff & (bytes - 1)) != 0) {
1817 * Ignore all writes beyond the standard config space and return all
1820 if (coff >= PCI_REGMAX + 1) {
1824 * Extended capabilities begin at offset 256 in config
1825 * space. Absence of extended capabilities is signaled
1826 * with all 0s in the extended capability header at
1829 if (coff <= PCI_REGMAX + 4)
1841 /* Let the device emulation override the default handler */
1842 if (pe->pe_cfgread != NULL) {
1843 needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes,
1850 *eax = CFGREAD(pi, coff, bytes);
1852 pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax);
1854 /* Let the device emulation override the default handler */
1855 if (pe->pe_cfgwrite != NULL &&
1856 (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0)
1860 * Special handling for write to BAR registers
1862 if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) {
1864 * Ignore writes to BAR registers that are not
1867 if (bytes != 4 || (coff & 0x3) != 0)
1869 idx = (coff - PCIR_BAR(0)) / 4;
1870 mask = ~(pi->pi_bar[idx].size - 1);
1871 switch (pi->pi_bar[idx].type) {
1873 pi->pi_bar[idx].addr = bar = 0;
1878 bar = addr | PCIM_BAR_IO_SPACE;
1880 * Register the new BAR value for interception
1882 if (addr != pi->pi_bar[idx].addr) {
1883 update_bar_address(pi, addr, idx,
1888 addr = bar = *eax & mask;
1889 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32;
1890 if (addr != pi->pi_bar[idx].addr) {
1891 update_bar_address(pi, addr, idx,
1896 addr = bar = *eax & mask;
1897 bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 |
1898 PCIM_BAR_MEM_PREFETCH;
1899 if (addr != (uint32_t)pi->pi_bar[idx].addr) {
1900 update_bar_address(pi, addr, idx,
1904 case PCIBAR_MEMHI64:
1905 mask = ~(pi->pi_bar[idx - 1].size - 1);
1906 addr = ((uint64_t)*eax << 32) & mask;
1908 if (bar != pi->pi_bar[idx - 1].addr >> 32) {
1909 update_bar_address(pi, addr, idx - 1,
1916 pci_set_cfgdata32(pi, coff, bar);
1918 } else if (pci_emul_iscap(pi, coff)) {
1919 pci_emul_capwrite(pi, coff, bytes, *eax, 0, 0);
1920 } else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) {
1921 pci_emul_cmdsts_write(pi, coff, *eax, bytes);
1923 CFGWRITE(pi, coff, *eax, bytes);
1928 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff;
1931 pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1932 uint32_t *eax, void *arg)
1938 *eax = (bytes == 2) ? 0xffff : 0xff;
1943 x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff;
1949 cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE;
1950 cfgoff = x & PCI_REGMAX;
1951 cfgfunc = (x >> 8) & PCI_FUNCMAX;
1952 cfgslot = (x >> 11) & PCI_SLOTMAX;
1953 cfgbus = (x >> 16) & PCI_BUSMAX;
1958 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr);
1961 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes,
1962 uint32_t *eax, void *arg)
1966 assert(bytes == 1 || bytes == 2 || bytes == 4);
1968 coff = cfgoff + (port - CONF1_DATA_PORT);
1970 pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes,
1973 /* Ignore accesses to cfgdata if not enabled by cfgaddr */
1980 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata);
1981 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata);
1982 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata);
1983 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata);
1985 #ifdef BHYVE_SNAPSHOT
1987 * Saves/restores PCI device emulated state. Returns 0 on success.
1990 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta)
1992 struct pci_devinst *pi;
1996 pi = meta->dev_data;
1998 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done);
1999 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done);
2000 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done);
2001 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done);
2003 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done);
2004 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done);
2005 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done);
2006 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done);
2007 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done);
2008 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done);
2009 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done);
2010 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done);
2011 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_page_offset, meta, ret, done);
2013 SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata),
2016 for (i = 0; i < nitems(pi->pi_bar); i++) {
2017 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done);
2018 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done);
2019 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done);
2022 /* Restore MSI-X table. */
2023 for (i = 0; i < pi->pi_msix.table_count; i++) {
2024 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr,
2026 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data,
2028 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control,
2037 pci_find_slotted_dev(const char *dev_name, struct pci_devemu **pde,
2038 struct pci_devinst **pdi)
2041 struct slotinfo *si;
2042 struct funcinfo *fi;
2043 int bus, slot, func;
2045 assert(dev_name != NULL);
2046 assert(pde != NULL);
2047 assert(pdi != NULL);
2049 for (bus = 0; bus < MAXBUSES; bus++) {
2050 if ((bi = pci_businfo[bus]) == NULL)
2053 for (slot = 0; slot < MAXSLOTS; slot++) {
2054 si = &bi->slotinfo[slot];
2055 for (func = 0; func < MAXFUNCS; func++) {
2056 fi = &si->si_funcs[func];
2057 if (fi->fi_name == NULL)
2059 if (strcmp(dev_name, fi->fi_name))
2062 *pde = pci_emul_finddev(fi->fi_name);
2063 assert(*pde != NULL);
2075 pci_snapshot(struct vm_snapshot_meta *meta)
2077 struct pci_devemu *pde;
2078 struct pci_devinst *pdi;
2081 assert(meta->dev_name != NULL);
2083 ret = pci_find_slotted_dev(meta->dev_name, &pde, &pdi);
2085 fprintf(stderr, "%s: no such name: %s\r\n",
2086 __func__, meta->dev_name);
2087 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
2091 meta->dev_data = pdi;
2093 if (pde->pe_snapshot == NULL) {
2094 fprintf(stderr, "%s: not implemented yet for: %s\r\n",
2095 __func__, meta->dev_name);
2099 ret = pci_snapshot_pci_dev(meta);
2101 fprintf(stderr, "%s: failed to snapshot pci dev\r\n",
2106 ret = (*pde->pe_snapshot)(meta);
2112 pci_pause(struct vmctx *ctx, const char *dev_name)
2114 struct pci_devemu *pde;
2115 struct pci_devinst *pdi;
2118 assert(dev_name != NULL);
2120 ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2123 * It is possible to call this function without
2124 * checking that the device is inserted first.
2126 fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2130 if (pde->pe_pause == NULL) {
2131 /* The pause/resume functionality is optional. */
2132 fprintf(stderr, "%s: not implemented for: %s\n",
2133 __func__, dev_name);
2137 return (*pde->pe_pause)(ctx, pdi);
2141 pci_resume(struct vmctx *ctx, const char *dev_name)
2143 struct pci_devemu *pde;
2144 struct pci_devinst *pdi;
2147 assert(dev_name != NULL);
2149 ret = pci_find_slotted_dev(dev_name, &pde, &pdi);
2152 * It is possible to call this function without
2153 * checking that the device is inserted first.
2155 fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name);
2159 if (pde->pe_resume == NULL) {
2160 /* The pause/resume functionality is optional. */
2161 fprintf(stderr, "%s: not implemented for: %s\n",
2162 __func__, dev_name);
2166 return (*pde->pe_resume)(ctx, pdi);
2170 #define PCI_EMUL_TEST
2171 #ifdef PCI_EMUL_TEST
2173 * Define a dummy test device
2177 struct pci_emul_dsoftc {
2178 uint8_t ioregs[DIOSZ];
2179 uint8_t memregs[2][DMEMSZ];
2182 #define PCI_EMUL_MSI_MSGS 4
2183 #define PCI_EMUL_MSIX_MSGS 16
2186 pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2189 struct pci_emul_dsoftc *sc;
2191 sc = calloc(1, sizeof(struct pci_emul_dsoftc));
2195 pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001);
2196 pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD);
2197 pci_set_cfgdata8(pi, PCIR_CLASS, 0x02);
2199 error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS);
2202 error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ);
2205 error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ);
2208 error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ);
2215 pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2216 uint64_t offset, int size, uint64_t value)
2219 struct pci_emul_dsoftc *sc = pi->pi_arg;
2222 if (offset + size > DIOSZ) {
2223 printf("diow: iow too large, offset %ld size %d\n",
2229 sc->ioregs[offset] = value & 0xff;
2230 } else if (size == 2) {
2231 *(uint16_t *)&sc->ioregs[offset] = value & 0xffff;
2232 } else if (size == 4) {
2233 *(uint32_t *)&sc->ioregs[offset] = value;
2235 printf("diow: iow unknown size %d\n", size);
2239 * Special magic value to generate an interrupt
2241 if (offset == 4 && size == 4 && pci_msi_enabled(pi))
2242 pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi));
2244 if (value == 0xabcdef) {
2245 for (i = 0; i < pci_msi_maxmsgnum(pi); i++)
2246 pci_generate_msi(pi, i);
2250 if (baridx == 1 || baridx == 2) {
2251 if (offset + size > DMEMSZ) {
2252 printf("diow: memw too large, offset %ld size %d\n",
2257 i = baridx - 1; /* 'memregs' index */
2260 sc->memregs[i][offset] = value;
2261 } else if (size == 2) {
2262 *(uint16_t *)&sc->memregs[i][offset] = value;
2263 } else if (size == 4) {
2264 *(uint32_t *)&sc->memregs[i][offset] = value;
2265 } else if (size == 8) {
2266 *(uint64_t *)&sc->memregs[i][offset] = value;
2268 printf("diow: memw unknown size %d\n", size);
2272 * magic interrupt ??
2276 if (baridx > 2 || baridx < 0) {
2277 printf("diow: unknown bar idx %d\n", baridx);
2282 pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2283 uint64_t offset, int size)
2285 struct pci_emul_dsoftc *sc = pi->pi_arg;
2290 if (offset + size > DIOSZ) {
2291 printf("dior: ior too large, offset %ld size %d\n",
2298 value = sc->ioregs[offset];
2299 } else if (size == 2) {
2300 value = *(uint16_t *) &sc->ioregs[offset];
2301 } else if (size == 4) {
2302 value = *(uint32_t *) &sc->ioregs[offset];
2304 printf("dior: ior unknown size %d\n", size);
2308 if (baridx == 1 || baridx == 2) {
2309 if (offset + size > DMEMSZ) {
2310 printf("dior: memr too large, offset %ld size %d\n",
2315 i = baridx - 1; /* 'memregs' index */
2318 value = sc->memregs[i][offset];
2319 } else if (size == 2) {
2320 value = *(uint16_t *) &sc->memregs[i][offset];
2321 } else if (size == 4) {
2322 value = *(uint32_t *) &sc->memregs[i][offset];
2323 } else if (size == 8) {
2324 value = *(uint64_t *) &sc->memregs[i][offset];
2326 printf("dior: ior unknown size %d\n", size);
2331 if (baridx > 2 || baridx < 0) {
2332 printf("dior: unknown bar idx %d\n", baridx);
2339 #ifdef BHYVE_SNAPSHOT
2341 pci_emul_snapshot(struct vm_snapshot_meta *meta)
2348 struct pci_devemu pci_dummy = {
2350 .pe_init = pci_emul_dinit,
2351 .pe_barwrite = pci_emul_diow,
2352 .pe_barread = pci_emul_dior,
2353 #ifdef BHYVE_SNAPSHOT
2354 .pe_snapshot = pci_emul_snapshot,
2357 PCI_EMUL_SET(pci_dummy);
2359 #endif /* PCI_EMUL_TEST */