1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 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. 15 * 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 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/linker_set.h> 36 #include <sys/mman.h> 37 38 #include <ctype.h> 39 #include <err.h> 40 #include <errno.h> 41 #include <pthread.h> 42 #include <stdio.h> 43 #include <stdlib.h> 44 #include <string.h> 45 #include <strings.h> 46 #include <assert.h> 47 #include <stdbool.h> 48 #include <sysexits.h> 49 50 #include <machine/vmm.h> 51 #include <machine/vmm_snapshot.h> 52 #include <vmmapi.h> 53 54 #include "acpi.h" 55 #include "bhyverun.h" 56 #include "config.h" 57 #include "debug.h" 58 #include "inout.h" 59 #include "ioapic.h" 60 #include "mem.h" 61 #include "pci_emul.h" 62 #include "pci_irq.h" 63 #include "pci_lpc.h" 64 65 #define CONF1_ADDR_PORT 0x0cf8 66 #define CONF1_DATA_PORT 0x0cfc 67 68 #define CONF1_ENABLE 0x80000000ul 69 70 #define MAXBUSES (PCI_BUSMAX + 1) 71 #define MAXSLOTS (PCI_SLOTMAX + 1) 72 #define MAXFUNCS (PCI_FUNCMAX + 1) 73 74 #define GB (1024 * 1024 * 1024UL) 75 76 struct funcinfo { 77 nvlist_t *fi_config; 78 struct pci_devemu *fi_pde; 79 struct pci_devinst *fi_devi; 80 }; 81 82 struct intxinfo { 83 int ii_count; 84 int ii_pirq_pin; 85 int ii_ioapic_irq; 86 }; 87 88 struct slotinfo { 89 struct intxinfo si_intpins[4]; 90 struct funcinfo si_funcs[MAXFUNCS]; 91 }; 92 93 struct businfo { 94 uint16_t iobase, iolimit; /* I/O window */ 95 uint32_t membase32, memlimit32; /* mmio window below 4GB */ 96 uint64_t membase64, memlimit64; /* mmio window above 4GB */ 97 struct slotinfo slotinfo[MAXSLOTS]; 98 }; 99 100 static struct businfo *pci_businfo[MAXBUSES]; 101 102 SET_DECLARE(pci_devemu_set, struct pci_devemu); 103 104 static uint64_t pci_emul_iobase; 105 static uint8_t *pci_emul_rombase; 106 static uint64_t pci_emul_romoffset; 107 static uint8_t *pci_emul_romlim; 108 static uint64_t pci_emul_membase32; 109 static uint64_t pci_emul_membase64; 110 static uint64_t pci_emul_memlim64; 111 112 struct pci_bar_allocation { 113 TAILQ_ENTRY(pci_bar_allocation) chain; 114 struct pci_devinst *pdi; 115 int idx; 116 enum pcibar_type type; 117 uint64_t size; 118 }; 119 120 static TAILQ_HEAD(pci_bar_list, pci_bar_allocation) pci_bars = 121 TAILQ_HEAD_INITIALIZER(pci_bars); 122 123 #define PCI_EMUL_IOBASE 0x2000 124 #define PCI_EMUL_IOLIMIT 0x10000 125 126 #define PCI_EMUL_ROMSIZE 0x10000000 127 128 #define PCI_EMUL_ECFG_BASE 0xE0000000 /* 3.5GB */ 129 #define PCI_EMUL_ECFG_SIZE (MAXBUSES * 1024 * 1024) /* 1MB per bus */ 130 SYSRES_MEM(PCI_EMUL_ECFG_BASE, PCI_EMUL_ECFG_SIZE); 131 132 /* 133 * OVMF always uses 0xC0000000 as base address for 32 bit PCI MMIO. Don't 134 * change this address without changing it in OVMF. 135 */ 136 #define PCI_EMUL_MEMBASE32 0xC0000000 137 #define PCI_EMUL_MEMLIMIT32 PCI_EMUL_ECFG_BASE 138 #define PCI_EMUL_MEMSIZE64 (32*GB) 139 140 static struct pci_devemu *pci_emul_finddev(const char *name); 141 static void pci_lintr_route(struct pci_devinst *pi); 142 static void pci_lintr_update(struct pci_devinst *pi); 143 static void pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, 144 int func, int coff, int bytes, uint32_t *val); 145 146 static __inline void 147 CFGWRITE(struct pci_devinst *pi, int coff, uint32_t val, int bytes) 148 { 149 150 if (bytes == 1) 151 pci_set_cfgdata8(pi, coff, val); 152 else if (bytes == 2) 153 pci_set_cfgdata16(pi, coff, val); 154 else 155 pci_set_cfgdata32(pi, coff, val); 156 } 157 158 static __inline uint32_t 159 CFGREAD(struct pci_devinst *pi, int coff, int bytes) 160 { 161 162 if (bytes == 1) 163 return (pci_get_cfgdata8(pi, coff)); 164 else if (bytes == 2) 165 return (pci_get_cfgdata16(pi, coff)); 166 else 167 return (pci_get_cfgdata32(pi, coff)); 168 } 169 170 static int 171 is_pcir_bar(int coff) 172 { 173 return (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)); 174 } 175 176 static int 177 is_pcir_bios(int coff) 178 { 179 return (coff >= PCIR_BIOS && coff < PCIR_BIOS + 4); 180 } 181 182 /* 183 * I/O access 184 */ 185 186 /* 187 * Slot options are in the form: 188 * 189 * <bus>:<slot>:<func>,<emul>[,<config>] 190 * <slot>[:<func>],<emul>[,<config>] 191 * 192 * slot is 0..31 193 * func is 0..7 194 * emul is a string describing the type of PCI device e.g. virtio-net 195 * config is an optional string, depending on the device, that can be 196 * used for configuration. 197 * Examples are: 198 * 1,virtio-net,tap0 199 * 3:0,dummy 200 */ 201 static void 202 pci_parse_slot_usage(char *aopt) 203 { 204 205 EPRINTLN("Invalid PCI slot info field \"%s\"", aopt); 206 } 207 208 /* 209 * Helper function to parse a list of comma-separated options where 210 * each option is formatted as "name[=value]". If no value is 211 * provided, the option is treated as a boolean and is given a value 212 * of true. 213 */ 214 int 215 pci_parse_legacy_config(nvlist_t *nvl, const char *opt) 216 { 217 char *config, *name, *tofree, *value; 218 219 if (opt == NULL) 220 return (0); 221 222 config = tofree = strdup(opt); 223 while ((name = strsep(&config, ",")) != NULL) { 224 value = strchr(name, '='); 225 if (value != NULL) { 226 *value = '\0'; 227 value++; 228 set_config_value_node(nvl, name, value); 229 } else 230 set_config_bool_node(nvl, name, true); 231 } 232 free(tofree); 233 return (0); 234 } 235 236 /* 237 * PCI device configuration is stored in MIBs that encode the device's 238 * location: 239 * 240 * pci.<bus>.<slot>.<func> 241 * 242 * Where "bus", "slot", and "func" are all decimal values without 243 * leading zeroes. Each valid device must have a "device" node which 244 * identifies the driver model of the device. 245 * 246 * Device backends can provide a parser for the "config" string. If 247 * a custom parser is not provided, pci_parse_legacy_config() is used 248 * to parse the string. 249 */ 250 int 251 pci_parse_slot(char *opt) 252 { 253 char node_name[sizeof("pci.XXX.XX.X")]; 254 struct pci_devemu *pde; 255 char *emul, *config, *str, *cp; 256 int error, bnum, snum, fnum; 257 nvlist_t *nvl; 258 259 error = -1; 260 str = strdup(opt); 261 262 emul = config = NULL; 263 if ((cp = strchr(str, ',')) != NULL) { 264 *cp = '\0'; 265 emul = cp + 1; 266 if ((cp = strchr(emul, ',')) != NULL) { 267 *cp = '\0'; 268 config = cp + 1; 269 } 270 } else { 271 pci_parse_slot_usage(opt); 272 goto done; 273 } 274 275 /* <bus>:<slot>:<func> */ 276 if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) { 277 bnum = 0; 278 /* <slot>:<func> */ 279 if (sscanf(str, "%d:%d", &snum, &fnum) != 2) { 280 fnum = 0; 281 /* <slot> */ 282 if (sscanf(str, "%d", &snum) != 1) { 283 snum = -1; 284 } 285 } 286 } 287 288 if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS || 289 fnum < 0 || fnum >= MAXFUNCS) { 290 pci_parse_slot_usage(opt); 291 goto done; 292 } 293 294 pde = pci_emul_finddev(emul); 295 if (pde == NULL) { 296 EPRINTLN("pci slot %d:%d:%d: unknown device \"%s\"", bnum, snum, 297 fnum, emul); 298 goto done; 299 } 300 301 snprintf(node_name, sizeof(node_name), "pci.%d.%d.%d", bnum, snum, 302 fnum); 303 nvl = find_config_node(node_name); 304 if (nvl != NULL) { 305 EPRINTLN("pci slot %d:%d:%d already occupied!", bnum, snum, 306 fnum); 307 goto done; 308 } 309 nvl = create_config_node(node_name); 310 if (pde->pe_alias != NULL) 311 set_config_value_node(nvl, "device", pde->pe_alias); 312 else 313 set_config_value_node(nvl, "device", pde->pe_emu); 314 315 if (pde->pe_legacy_config != NULL) 316 error = pde->pe_legacy_config(nvl, config); 317 else 318 error = pci_parse_legacy_config(nvl, config); 319 done: 320 free(str); 321 return (error); 322 } 323 324 void 325 pci_print_supported_devices(void) 326 { 327 struct pci_devemu **pdpp, *pdp; 328 329 SET_FOREACH(pdpp, pci_devemu_set) { 330 pdp = *pdpp; 331 printf("%s\n", pdp->pe_emu); 332 } 333 } 334 335 static int 336 pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset) 337 { 338 339 if (offset < pi->pi_msix.pba_offset) 340 return (0); 341 342 if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) { 343 return (0); 344 } 345 346 return (1); 347 } 348 349 int 350 pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size, 351 uint64_t value) 352 { 353 int msix_entry_offset; 354 int tab_index; 355 char *dest; 356 357 /* support only 4 or 8 byte writes */ 358 if (size != 4 && size != 8) 359 return (-1); 360 361 /* 362 * Return if table index is beyond what device supports 363 */ 364 tab_index = offset / MSIX_TABLE_ENTRY_SIZE; 365 if (tab_index >= pi->pi_msix.table_count) 366 return (-1); 367 368 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; 369 370 /* support only aligned writes */ 371 if ((msix_entry_offset % size) != 0) 372 return (-1); 373 374 dest = (char *)(pi->pi_msix.table + tab_index); 375 dest += msix_entry_offset; 376 377 if (size == 4) 378 *((uint32_t *)dest) = value; 379 else 380 *((uint64_t *)dest) = value; 381 382 return (0); 383 } 384 385 uint64_t 386 pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size) 387 { 388 char *dest; 389 int msix_entry_offset; 390 int tab_index; 391 uint64_t retval = ~0; 392 393 /* 394 * The PCI standard only allows 4 and 8 byte accesses to the MSI-X 395 * table but we also allow 1 byte access to accommodate reads from 396 * ddb. 397 */ 398 if (size != 1 && size != 4 && size != 8) 399 return (retval); 400 401 msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; 402 403 /* support only aligned reads */ 404 if ((msix_entry_offset % size) != 0) { 405 return (retval); 406 } 407 408 tab_index = offset / MSIX_TABLE_ENTRY_SIZE; 409 410 if (tab_index < pi->pi_msix.table_count) { 411 /* valid MSI-X Table access */ 412 dest = (char *)(pi->pi_msix.table + tab_index); 413 dest += msix_entry_offset; 414 415 if (size == 1) 416 retval = *((uint8_t *)dest); 417 else if (size == 4) 418 retval = *((uint32_t *)dest); 419 else 420 retval = *((uint64_t *)dest); 421 } else if (pci_valid_pba_offset(pi, offset)) { 422 /* return 0 for PBA access */ 423 retval = 0; 424 } 425 426 return (retval); 427 } 428 429 int 430 pci_msix_table_bar(struct pci_devinst *pi) 431 { 432 433 if (pi->pi_msix.table != NULL) 434 return (pi->pi_msix.table_bar); 435 else 436 return (-1); 437 } 438 439 int 440 pci_msix_pba_bar(struct pci_devinst *pi) 441 { 442 443 if (pi->pi_msix.table != NULL) 444 return (pi->pi_msix.pba_bar); 445 else 446 return (-1); 447 } 448 449 static int 450 pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes, 451 uint32_t *eax, void *arg) 452 { 453 struct pci_devinst *pdi = arg; 454 struct pci_devemu *pe = pdi->pi_d; 455 uint64_t offset; 456 int i; 457 458 assert(port >= 0); 459 460 for (i = 0; i <= PCI_BARMAX; i++) { 461 if (pdi->pi_bar[i].type == PCIBAR_IO && 462 (uint64_t)port >= pdi->pi_bar[i].addr && 463 (uint64_t)port + bytes <= 464 pdi->pi_bar[i].addr + pdi->pi_bar[i].size) { 465 offset = port - pdi->pi_bar[i].addr; 466 if (in) 467 *eax = (*pe->pe_barread)(ctx, vcpu, pdi, i, 468 offset, bytes); 469 else 470 (*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset, 471 bytes, *eax); 472 return (0); 473 } 474 } 475 return (-1); 476 } 477 478 static int 479 pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr, 480 int size, uint64_t *val, void *arg1, long arg2) 481 { 482 struct pci_devinst *pdi = arg1; 483 struct pci_devemu *pe = pdi->pi_d; 484 uint64_t offset; 485 int bidx = (int) arg2; 486 487 assert(bidx <= PCI_BARMAX); 488 assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 || 489 pdi->pi_bar[bidx].type == PCIBAR_MEM64); 490 assert(addr >= pdi->pi_bar[bidx].addr && 491 addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size); 492 493 offset = addr - pdi->pi_bar[bidx].addr; 494 495 if (dir == MEM_F_WRITE) { 496 if (size == 8) { 497 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, 498 4, *val & 0xffffffff); 499 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4, 500 4, *val >> 32); 501 } else { 502 (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, 503 size, *val); 504 } 505 } else { 506 if (size == 8) { 507 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, 508 offset, 4); 509 *val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx, 510 offset + 4, 4) << 32; 511 } else { 512 *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, 513 offset, size); 514 } 515 } 516 517 return (0); 518 } 519 520 521 static int 522 pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size, 523 uint64_t *addr) 524 { 525 uint64_t base; 526 527 assert((size & (size - 1)) == 0); /* must be a power of 2 */ 528 529 base = roundup2(*baseptr, size); 530 531 if (base + size <= limit) { 532 *addr = base; 533 *baseptr = base + size; 534 return (0); 535 } else 536 return (-1); 537 } 538 539 /* 540 * Register (or unregister) the MMIO or I/O region associated with the BAR 541 * register 'idx' of an emulated pci device. 542 */ 543 static void 544 modify_bar_registration(struct pci_devinst *pi, int idx, int registration) 545 { 546 struct pci_devemu *pe; 547 int error; 548 struct inout_port iop; 549 struct mem_range mr; 550 551 pe = pi->pi_d; 552 switch (pi->pi_bar[idx].type) { 553 case PCIBAR_IO: 554 bzero(&iop, sizeof(struct inout_port)); 555 iop.name = pi->pi_name; 556 iop.port = pi->pi_bar[idx].addr; 557 iop.size = pi->pi_bar[idx].size; 558 if (registration) { 559 iop.flags = IOPORT_F_INOUT; 560 iop.handler = pci_emul_io_handler; 561 iop.arg = pi; 562 error = register_inout(&iop); 563 } else 564 error = unregister_inout(&iop); 565 if (pe->pe_baraddr != NULL) 566 (*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration, 567 pi->pi_bar[idx].addr); 568 break; 569 case PCIBAR_MEM32: 570 case PCIBAR_MEM64: 571 bzero(&mr, sizeof(struct mem_range)); 572 mr.name = pi->pi_name; 573 mr.base = pi->pi_bar[idx].addr; 574 mr.size = pi->pi_bar[idx].size; 575 if (registration) { 576 mr.flags = MEM_F_RW; 577 mr.handler = pci_emul_mem_handler; 578 mr.arg1 = pi; 579 mr.arg2 = idx; 580 error = register_mem(&mr); 581 } else 582 error = unregister_mem(&mr); 583 if (pe->pe_baraddr != NULL) 584 (*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration, 585 pi->pi_bar[idx].addr); 586 break; 587 case PCIBAR_ROM: 588 error = 0; 589 if (pe->pe_baraddr != NULL) 590 (*pe->pe_baraddr)(pi->pi_vmctx, pi, idx, registration, 591 pi->pi_bar[idx].addr); 592 break; 593 default: 594 error = EINVAL; 595 break; 596 } 597 assert(error == 0); 598 } 599 600 static void 601 unregister_bar(struct pci_devinst *pi, int idx) 602 { 603 604 modify_bar_registration(pi, idx, 0); 605 } 606 607 static void 608 register_bar(struct pci_devinst *pi, int idx) 609 { 610 611 modify_bar_registration(pi, idx, 1); 612 } 613 614 /* Is the ROM enabled for the emulated pci device? */ 615 static int 616 romen(struct pci_devinst *pi) 617 { 618 return (pi->pi_bar[PCI_ROM_IDX].lobits & PCIM_BIOS_ENABLE) == 619 PCIM_BIOS_ENABLE; 620 } 621 622 /* Are we decoding i/o port accesses for the emulated pci device? */ 623 static int 624 porten(struct pci_devinst *pi) 625 { 626 uint16_t cmd; 627 628 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); 629 630 return (cmd & PCIM_CMD_PORTEN); 631 } 632 633 /* Are we decoding memory accesses for the emulated pci device? */ 634 static int 635 memen(struct pci_devinst *pi) 636 { 637 uint16_t cmd; 638 639 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); 640 641 return (cmd & PCIM_CMD_MEMEN); 642 } 643 644 /* 645 * Update the MMIO or I/O address that is decoded by the BAR register. 646 * 647 * If the pci device has enabled the address space decoding then intercept 648 * the address range decoded by the BAR register. 649 */ 650 static void 651 update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type) 652 { 653 int decode; 654 655 if (pi->pi_bar[idx].type == PCIBAR_IO) 656 decode = porten(pi); 657 else 658 decode = memen(pi); 659 660 if (decode) 661 unregister_bar(pi, idx); 662 663 switch (type) { 664 case PCIBAR_IO: 665 case PCIBAR_MEM32: 666 pi->pi_bar[idx].addr = addr; 667 break; 668 case PCIBAR_MEM64: 669 pi->pi_bar[idx].addr &= ~0xffffffffUL; 670 pi->pi_bar[idx].addr |= addr; 671 break; 672 case PCIBAR_MEMHI64: 673 pi->pi_bar[idx].addr &= 0xffffffff; 674 pi->pi_bar[idx].addr |= addr; 675 break; 676 default: 677 assert(0); 678 } 679 680 if (decode) 681 register_bar(pi, idx); 682 } 683 684 int 685 pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type, 686 uint64_t size) 687 { 688 assert((type == PCIBAR_ROM) || (idx >= 0 && idx <= PCI_BARMAX)); 689 assert((type != PCIBAR_ROM) || (idx == PCI_ROM_IDX)); 690 691 if ((size & (size - 1)) != 0) 692 size = 1UL << flsl(size); /* round up to a power of 2 */ 693 694 /* Enforce minimum BAR sizes required by the PCI standard */ 695 if (type == PCIBAR_IO) { 696 if (size < 4) 697 size = 4; 698 } else if (type == PCIBAR_ROM) { 699 if (size < ~PCIM_BIOS_ADDR_MASK + 1) 700 size = ~PCIM_BIOS_ADDR_MASK + 1; 701 } else { 702 if (size < 16) 703 size = 16; 704 } 705 706 /* 707 * To reduce fragmentation of the MMIO space, we allocate the BARs by 708 * size. Therefore, don't allocate the BAR yet. We create a list of all 709 * BAR allocation which is sorted by BAR size. When all PCI devices are 710 * initialized, we will assign an address to the BARs. 711 */ 712 713 /* create a new list entry */ 714 struct pci_bar_allocation *const new_bar = malloc(sizeof(*new_bar)); 715 memset(new_bar, 0, sizeof(*new_bar)); 716 new_bar->pdi = pdi; 717 new_bar->idx = idx; 718 new_bar->type = type; 719 new_bar->size = size; 720 721 /* 722 * Search for a BAR which size is lower than the size of our newly 723 * allocated BAR. 724 */ 725 struct pci_bar_allocation *bar = NULL; 726 TAILQ_FOREACH(bar, &pci_bars, chain) { 727 if (bar->size < size) { 728 break; 729 } 730 } 731 732 if (bar == NULL) { 733 /* 734 * Either the list is empty or new BAR is the smallest BAR of 735 * the list. Append it to the end of our list. 736 */ 737 TAILQ_INSERT_TAIL(&pci_bars, new_bar, chain); 738 } else { 739 /* 740 * The found BAR is smaller than our new BAR. For that reason, 741 * insert our new BAR before the found BAR. 742 */ 743 TAILQ_INSERT_BEFORE(bar, new_bar, chain); 744 } 745 746 /* 747 * pci_passthru devices synchronize their physical and virtual command 748 * register on init. For that reason, the virtual cmd reg should be 749 * updated as early as possible. 750 */ 751 uint16_t enbit = 0; 752 switch (type) { 753 case PCIBAR_IO: 754 enbit = PCIM_CMD_PORTEN; 755 break; 756 case PCIBAR_MEM64: 757 case PCIBAR_MEM32: 758 enbit = PCIM_CMD_MEMEN; 759 break; 760 default: 761 enbit = 0; 762 break; 763 } 764 765 const uint16_t cmd = pci_get_cfgdata16(pdi, PCIR_COMMAND); 766 pci_set_cfgdata16(pdi, PCIR_COMMAND, cmd | enbit); 767 768 return (0); 769 } 770 771 static int 772 pci_emul_assign_bar(struct pci_devinst *const pdi, const int idx, 773 const enum pcibar_type type, const uint64_t size) 774 { 775 int error; 776 uint64_t *baseptr, limit, addr, mask, lobits, bar; 777 778 switch (type) { 779 case PCIBAR_NONE: 780 baseptr = NULL; 781 addr = mask = lobits = 0; 782 break; 783 case PCIBAR_IO: 784 baseptr = &pci_emul_iobase; 785 limit = PCI_EMUL_IOLIMIT; 786 mask = PCIM_BAR_IO_BASE; 787 lobits = PCIM_BAR_IO_SPACE; 788 break; 789 case PCIBAR_MEM64: 790 /* 791 * XXX 792 * Some drivers do not work well if the 64-bit BAR is allocated 793 * above 4GB. Allow for this by allocating small requests under 794 * 4GB unless then allocation size is larger than some arbitrary 795 * number (128MB currently). 796 */ 797 if (size > 128 * 1024 * 1024) { 798 baseptr = &pci_emul_membase64; 799 limit = pci_emul_memlim64; 800 mask = PCIM_BAR_MEM_BASE; 801 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 | 802 PCIM_BAR_MEM_PREFETCH; 803 } else { 804 baseptr = &pci_emul_membase32; 805 limit = PCI_EMUL_MEMLIMIT32; 806 mask = PCIM_BAR_MEM_BASE; 807 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64; 808 } 809 break; 810 case PCIBAR_MEM32: 811 baseptr = &pci_emul_membase32; 812 limit = PCI_EMUL_MEMLIMIT32; 813 mask = PCIM_BAR_MEM_BASE; 814 lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32; 815 break; 816 case PCIBAR_ROM: 817 /* do not claim memory for ROM. OVMF will do it for us. */ 818 baseptr = NULL; 819 limit = 0; 820 mask = PCIM_BIOS_ADDR_MASK; 821 lobits = 0; 822 break; 823 default: 824 printf("pci_emul_alloc_base: invalid bar type %d\n", type); 825 assert(0); 826 } 827 828 if (baseptr != NULL) { 829 error = pci_emul_alloc_resource(baseptr, limit, size, &addr); 830 if (error != 0) 831 return (error); 832 } else { 833 addr = 0; 834 } 835 836 pdi->pi_bar[idx].type = type; 837 pdi->pi_bar[idx].addr = addr; 838 pdi->pi_bar[idx].size = size; 839 /* 840 * passthru devices are using same lobits as physical device they set 841 * this property 842 */ 843 if (pdi->pi_bar[idx].lobits != 0) { 844 lobits = pdi->pi_bar[idx].lobits; 845 } else { 846 pdi->pi_bar[idx].lobits = lobits; 847 } 848 849 /* Initialize the BAR register in config space */ 850 bar = (addr & mask) | lobits; 851 pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar); 852 853 if (type == PCIBAR_MEM64) { 854 assert(idx + 1 <= PCI_BARMAX); 855 pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64; 856 pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32); 857 } 858 859 if (type != PCIBAR_ROM) { 860 register_bar(pdi, idx); 861 } 862 863 return (0); 864 } 865 866 int 867 pci_emul_alloc_rom(struct pci_devinst *const pdi, const uint64_t size, 868 void **const addr) 869 { 870 /* allocate ROM space once on first call */ 871 if (pci_emul_rombase == 0) { 872 pci_emul_rombase = vm_create_devmem(pdi->pi_vmctx, VM_PCIROM, 873 "pcirom", PCI_EMUL_ROMSIZE); 874 if (pci_emul_rombase == MAP_FAILED) { 875 warnx("%s: failed to create rom segment", __func__); 876 return (-1); 877 } 878 pci_emul_romlim = pci_emul_rombase + PCI_EMUL_ROMSIZE; 879 pci_emul_romoffset = 0; 880 } 881 882 /* ROM size should be a power of 2 and greater than 2 KB */ 883 const uint64_t rom_size = MAX(1UL << flsl(size), 884 ~PCIM_BIOS_ADDR_MASK + 1); 885 886 /* check if ROM fits into ROM space */ 887 if (pci_emul_romoffset + rom_size > PCI_EMUL_ROMSIZE) { 888 warnx("%s: no space left in rom segment:", __func__); 889 warnx("%16lu bytes left", 890 PCI_EMUL_ROMSIZE - pci_emul_romoffset); 891 warnx("%16lu bytes required by %d/%d/%d", rom_size, pdi->pi_bus, 892 pdi->pi_slot, pdi->pi_func); 893 return (-1); 894 } 895 896 /* allocate ROM BAR */ 897 const int error = pci_emul_alloc_bar(pdi, PCI_ROM_IDX, PCIBAR_ROM, 898 rom_size); 899 if (error) 900 return error; 901 902 /* return address */ 903 *addr = pci_emul_rombase + pci_emul_romoffset; 904 905 /* save offset into ROM Space */ 906 pdi->pi_romoffset = pci_emul_romoffset; 907 908 /* increase offset for next ROM */ 909 pci_emul_romoffset += rom_size; 910 911 return (0); 912 } 913 914 #define CAP_START_OFFSET 0x40 915 static int 916 pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen) 917 { 918 int i, capoff, reallen; 919 uint16_t sts; 920 921 assert(caplen > 0); 922 923 reallen = roundup2(caplen, 4); /* dword aligned */ 924 925 sts = pci_get_cfgdata16(pi, PCIR_STATUS); 926 if ((sts & PCIM_STATUS_CAPPRESENT) == 0) 927 capoff = CAP_START_OFFSET; 928 else 929 capoff = pi->pi_capend + 1; 930 931 /* Check if we have enough space */ 932 if (capoff + reallen > PCI_REGMAX + 1) 933 return (-1); 934 935 /* Set the previous capability pointer */ 936 if ((sts & PCIM_STATUS_CAPPRESENT) == 0) { 937 pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff); 938 pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT); 939 } else 940 pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff); 941 942 /* Copy the capability */ 943 for (i = 0; i < caplen; i++) 944 pci_set_cfgdata8(pi, capoff + i, capdata[i]); 945 946 /* Set the next capability pointer */ 947 pci_set_cfgdata8(pi, capoff + 1, 0); 948 949 pi->pi_prevcap = capoff; 950 pi->pi_capend = capoff + reallen - 1; 951 return (0); 952 } 953 954 static struct pci_devemu * 955 pci_emul_finddev(const char *name) 956 { 957 struct pci_devemu **pdpp, *pdp; 958 959 SET_FOREACH(pdpp, pci_devemu_set) { 960 pdp = *pdpp; 961 if (!strcmp(pdp->pe_emu, name)) { 962 return (pdp); 963 } 964 } 965 966 return (NULL); 967 } 968 969 static int 970 pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot, 971 int func, struct funcinfo *fi) 972 { 973 struct pci_devinst *pdi; 974 int err; 975 976 pdi = calloc(1, sizeof(struct pci_devinst)); 977 978 pdi->pi_vmctx = ctx; 979 pdi->pi_bus = bus; 980 pdi->pi_slot = slot; 981 pdi->pi_func = func; 982 pthread_mutex_init(&pdi->pi_lintr.lock, NULL); 983 pdi->pi_lintr.pin = 0; 984 pdi->pi_lintr.state = IDLE; 985 pdi->pi_lintr.pirq_pin = 0; 986 pdi->pi_lintr.ioapic_irq = 0; 987 pdi->pi_d = pde; 988 snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot); 989 990 /* Disable legacy interrupts */ 991 pci_set_cfgdata8(pdi, PCIR_INTLINE, 255); 992 pci_set_cfgdata8(pdi, PCIR_INTPIN, 0); 993 994 pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_BUSMASTEREN); 995 996 err = (*pde->pe_init)(ctx, pdi, fi->fi_config); 997 if (err == 0) 998 fi->fi_devi = pdi; 999 else 1000 free(pdi); 1001 1002 return (err); 1003 } 1004 1005 void 1006 pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr) 1007 { 1008 int mmc; 1009 1010 /* Number of msi messages must be a power of 2 between 1 and 32 */ 1011 assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32); 1012 mmc = ffs(msgnum) - 1; 1013 1014 bzero(msicap, sizeof(struct msicap)); 1015 msicap->capid = PCIY_MSI; 1016 msicap->nextptr = nextptr; 1017 msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1); 1018 } 1019 1020 int 1021 pci_emul_add_msicap(struct pci_devinst *pi, int msgnum) 1022 { 1023 struct msicap msicap; 1024 1025 pci_populate_msicap(&msicap, msgnum, 0); 1026 1027 return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap))); 1028 } 1029 1030 static void 1031 pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum, 1032 uint32_t msix_tab_size) 1033 { 1034 1035 assert(msix_tab_size % 4096 == 0); 1036 1037 bzero(msixcap, sizeof(struct msixcap)); 1038 msixcap->capid = PCIY_MSIX; 1039 1040 /* 1041 * Message Control Register, all fields set to 1042 * zero except for the Table Size. 1043 * Note: Table size N is encoded as N-1 1044 */ 1045 msixcap->msgctrl = msgnum - 1; 1046 1047 /* 1048 * MSI-X BAR setup: 1049 * - MSI-X table start at offset 0 1050 * - PBA table starts at a 4K aligned offset after the MSI-X table 1051 */ 1052 msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK; 1053 msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK); 1054 } 1055 1056 static void 1057 pci_msix_table_init(struct pci_devinst *pi, int table_entries) 1058 { 1059 int i, table_size; 1060 1061 assert(table_entries > 0); 1062 assert(table_entries <= MAX_MSIX_TABLE_ENTRIES); 1063 1064 table_size = table_entries * MSIX_TABLE_ENTRY_SIZE; 1065 pi->pi_msix.table = calloc(1, table_size); 1066 1067 /* set mask bit of vector control register */ 1068 for (i = 0; i < table_entries; i++) 1069 pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK; 1070 } 1071 1072 int 1073 pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum) 1074 { 1075 uint32_t tab_size; 1076 struct msixcap msixcap; 1077 1078 assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES); 1079 assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0); 1080 1081 tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE; 1082 1083 /* Align table size to nearest 4K */ 1084 tab_size = roundup2(tab_size, 4096); 1085 1086 pi->pi_msix.table_bar = barnum; 1087 pi->pi_msix.pba_bar = barnum; 1088 pi->pi_msix.table_offset = 0; 1089 pi->pi_msix.table_count = msgnum; 1090 pi->pi_msix.pba_offset = tab_size; 1091 pi->pi_msix.pba_size = PBA_SIZE(msgnum); 1092 1093 pci_msix_table_init(pi, msgnum); 1094 1095 pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size); 1096 1097 /* allocate memory for MSI-X Table and PBA */ 1098 pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32, 1099 tab_size + pi->pi_msix.pba_size); 1100 1101 return (pci_emul_add_capability(pi, (u_char *)&msixcap, 1102 sizeof(msixcap))); 1103 } 1104 1105 static void 1106 msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset, 1107 int bytes, uint32_t val) 1108 { 1109 uint16_t msgctrl, rwmask; 1110 int off; 1111 1112 off = offset - capoff; 1113 /* Message Control Register */ 1114 if (off == 2 && bytes == 2) { 1115 rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK; 1116 msgctrl = pci_get_cfgdata16(pi, offset); 1117 msgctrl &= ~rwmask; 1118 msgctrl |= val & rwmask; 1119 val = msgctrl; 1120 1121 pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE; 1122 pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK; 1123 pci_lintr_update(pi); 1124 } 1125 1126 CFGWRITE(pi, offset, val, bytes); 1127 } 1128 1129 static void 1130 msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset, 1131 int bytes, uint32_t val) 1132 { 1133 uint16_t msgctrl, rwmask, msgdata, mme; 1134 uint32_t addrlo; 1135 1136 /* 1137 * If guest is writing to the message control register make sure 1138 * we do not overwrite read-only fields. 1139 */ 1140 if ((offset - capoff) == 2 && bytes == 2) { 1141 rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE; 1142 msgctrl = pci_get_cfgdata16(pi, offset); 1143 msgctrl &= ~rwmask; 1144 msgctrl |= val & rwmask; 1145 val = msgctrl; 1146 } 1147 CFGWRITE(pi, offset, val, bytes); 1148 1149 msgctrl = pci_get_cfgdata16(pi, capoff + 2); 1150 addrlo = pci_get_cfgdata32(pi, capoff + 4); 1151 if (msgctrl & PCIM_MSICTRL_64BIT) 1152 msgdata = pci_get_cfgdata16(pi, capoff + 12); 1153 else 1154 msgdata = pci_get_cfgdata16(pi, capoff + 8); 1155 1156 mme = msgctrl & PCIM_MSICTRL_MME_MASK; 1157 pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0; 1158 if (pi->pi_msi.enabled) { 1159 pi->pi_msi.addr = addrlo; 1160 pi->pi_msi.msg_data = msgdata; 1161 pi->pi_msi.maxmsgnum = 1 << (mme >> 4); 1162 } else { 1163 pi->pi_msi.maxmsgnum = 0; 1164 } 1165 pci_lintr_update(pi); 1166 } 1167 1168 static void 1169 pciecap_cfgwrite(struct pci_devinst *pi, int capoff __unused, int offset, 1170 int bytes, uint32_t val) 1171 { 1172 1173 /* XXX don't write to the readonly parts */ 1174 CFGWRITE(pi, offset, val, bytes); 1175 } 1176 1177 #define PCIECAP_VERSION 0x2 1178 int 1179 pci_emul_add_pciecap(struct pci_devinst *pi, int type) 1180 { 1181 int err; 1182 struct pciecap pciecap; 1183 1184 bzero(&pciecap, sizeof(pciecap)); 1185 1186 /* 1187 * Use the integrated endpoint type for endpoints on a root complex bus. 1188 * 1189 * NB: bhyve currently only supports a single PCI bus that is the root 1190 * complex bus, so all endpoints are integrated. 1191 */ 1192 if ((type == PCIEM_TYPE_ENDPOINT) && (pi->pi_bus == 0)) 1193 type = PCIEM_TYPE_ROOT_INT_EP; 1194 1195 pciecap.capid = PCIY_EXPRESS; 1196 pciecap.pcie_capabilities = PCIECAP_VERSION | type; 1197 if (type != PCIEM_TYPE_ROOT_INT_EP) { 1198 pciecap.link_capabilities = 0x411; /* gen1, x1 */ 1199 pciecap.link_status = 0x11; /* gen1, x1 */ 1200 } 1201 1202 err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap)); 1203 return (err); 1204 } 1205 1206 /* 1207 * This function assumes that 'coff' is in the capabilities region of the 1208 * config space. A capoff parameter of zero will force a search for the 1209 * offset and type. 1210 */ 1211 void 1212 pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val, 1213 uint8_t capoff, int capid) 1214 { 1215 uint8_t nextoff; 1216 1217 /* Do not allow un-aligned writes */ 1218 if ((offset & (bytes - 1)) != 0) 1219 return; 1220 1221 if (capoff == 0) { 1222 /* Find the capability that we want to update */ 1223 capoff = CAP_START_OFFSET; 1224 while (1) { 1225 nextoff = pci_get_cfgdata8(pi, capoff + 1); 1226 if (nextoff == 0) 1227 break; 1228 if (offset >= capoff && offset < nextoff) 1229 break; 1230 1231 capoff = nextoff; 1232 } 1233 assert(offset >= capoff); 1234 capid = pci_get_cfgdata8(pi, capoff); 1235 } 1236 1237 /* 1238 * Capability ID and Next Capability Pointer are readonly. 1239 * However, some o/s's do 4-byte writes that include these. 1240 * For this case, trim the write back to 2 bytes and adjust 1241 * the data. 1242 */ 1243 if (offset == capoff || offset == capoff + 1) { 1244 if (offset == capoff && bytes == 4) { 1245 bytes = 2; 1246 offset += 2; 1247 val >>= 16; 1248 } else 1249 return; 1250 } 1251 1252 switch (capid) { 1253 case PCIY_MSI: 1254 msicap_cfgwrite(pi, capoff, offset, bytes, val); 1255 break; 1256 case PCIY_MSIX: 1257 msixcap_cfgwrite(pi, capoff, offset, bytes, val); 1258 break; 1259 case PCIY_EXPRESS: 1260 pciecap_cfgwrite(pi, capoff, offset, bytes, val); 1261 break; 1262 default: 1263 break; 1264 } 1265 } 1266 1267 static int 1268 pci_emul_iscap(struct pci_devinst *pi, int offset) 1269 { 1270 uint16_t sts; 1271 1272 sts = pci_get_cfgdata16(pi, PCIR_STATUS); 1273 if ((sts & PCIM_STATUS_CAPPRESENT) != 0) { 1274 if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend) 1275 return (1); 1276 } 1277 return (0); 1278 } 1279 1280 static int 1281 pci_emul_fallback_handler(struct vmctx *ctx __unused, int vcpu __unused, 1282 int dir, uint64_t addr __unused, int size __unused, uint64_t *val, 1283 void *arg1 __unused, long arg2 __unused) 1284 { 1285 /* 1286 * Ignore writes; return 0xff's for reads. The mem read code 1287 * will take care of truncating to the correct size. 1288 */ 1289 if (dir == MEM_F_READ) { 1290 *val = 0xffffffffffffffff; 1291 } 1292 1293 return (0); 1294 } 1295 1296 static int 1297 pci_emul_ecfg_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr, 1298 int bytes, uint64_t *val, void *arg1 __unused, long arg2 __unused) 1299 { 1300 int bus, slot, func, coff, in; 1301 1302 coff = addr & 0xfff; 1303 func = (addr >> 12) & 0x7; 1304 slot = (addr >> 15) & 0x1f; 1305 bus = (addr >> 20) & 0xff; 1306 in = (dir == MEM_F_READ); 1307 if (in) 1308 *val = ~0UL; 1309 pci_cfgrw(ctx, vcpu, in, bus, slot, func, coff, bytes, (uint32_t *)val); 1310 return (0); 1311 } 1312 1313 uint64_t 1314 pci_ecfg_base(void) 1315 { 1316 1317 return (PCI_EMUL_ECFG_BASE); 1318 } 1319 1320 #define BUSIO_ROUNDUP 32 1321 #define BUSMEM32_ROUNDUP (1024 * 1024) 1322 #define BUSMEM64_ROUNDUP (512 * 1024 * 1024) 1323 1324 int 1325 init_pci(struct vmctx *ctx) 1326 { 1327 char node_name[sizeof("pci.XXX.XX.X")]; 1328 struct mem_range mr; 1329 struct pci_devemu *pde; 1330 struct businfo *bi; 1331 struct slotinfo *si; 1332 struct funcinfo *fi; 1333 nvlist_t *nvl; 1334 const char *emul; 1335 size_t lowmem; 1336 int bus, slot, func; 1337 int error; 1338 1339 if (vm_get_lowmem_limit(ctx) > PCI_EMUL_MEMBASE32) 1340 errx(EX_OSERR, "Invalid lowmem limit"); 1341 1342 pci_emul_iobase = PCI_EMUL_IOBASE; 1343 pci_emul_membase32 = PCI_EMUL_MEMBASE32; 1344 1345 pci_emul_membase64 = 4*GB + vm_get_highmem_size(ctx); 1346 pci_emul_membase64 = roundup2(pci_emul_membase64, PCI_EMUL_MEMSIZE64); 1347 pci_emul_memlim64 = pci_emul_membase64 + PCI_EMUL_MEMSIZE64; 1348 1349 for (bus = 0; bus < MAXBUSES; bus++) { 1350 snprintf(node_name, sizeof(node_name), "pci.%d", bus); 1351 nvl = find_config_node(node_name); 1352 if (nvl == NULL) 1353 continue; 1354 pci_businfo[bus] = calloc(1, sizeof(struct businfo)); 1355 bi = pci_businfo[bus]; 1356 1357 /* 1358 * Keep track of the i/o and memory resources allocated to 1359 * this bus. 1360 */ 1361 bi->iobase = pci_emul_iobase; 1362 bi->membase32 = pci_emul_membase32; 1363 bi->membase64 = pci_emul_membase64; 1364 1365 /* first run: init devices */ 1366 for (slot = 0; slot < MAXSLOTS; slot++) { 1367 si = &bi->slotinfo[slot]; 1368 for (func = 0; func < MAXFUNCS; func++) { 1369 fi = &si->si_funcs[func]; 1370 snprintf(node_name, sizeof(node_name), 1371 "pci.%d.%d.%d", bus, slot, func); 1372 nvl = find_config_node(node_name); 1373 if (nvl == NULL) 1374 continue; 1375 1376 fi->fi_config = nvl; 1377 emul = get_config_value_node(nvl, "device"); 1378 if (emul == NULL) { 1379 EPRINTLN("pci slot %d:%d:%d: missing " 1380 "\"device\" value", bus, slot, func); 1381 return (EINVAL); 1382 } 1383 pde = pci_emul_finddev(emul); 1384 if (pde == NULL) { 1385 EPRINTLN("pci slot %d:%d:%d: unknown " 1386 "device \"%s\"", bus, slot, func, 1387 emul); 1388 return (EINVAL); 1389 } 1390 if (pde->pe_alias != NULL) { 1391 EPRINTLN("pci slot %d:%d:%d: legacy " 1392 "device \"%s\", use \"%s\" instead", 1393 bus, slot, func, emul, 1394 pde->pe_alias); 1395 return (EINVAL); 1396 } 1397 fi->fi_pde = pde; 1398 error = pci_emul_init(ctx, pde, bus, slot, 1399 func, fi); 1400 if (error) 1401 return (error); 1402 } 1403 } 1404 1405 /* second run: assign BARs and free list */ 1406 struct pci_bar_allocation *bar; 1407 struct pci_bar_allocation *bar_tmp; 1408 TAILQ_FOREACH_SAFE(bar, &pci_bars, chain, bar_tmp) { 1409 pci_emul_assign_bar(bar->pdi, bar->idx, bar->type, 1410 bar->size); 1411 free(bar); 1412 } 1413 TAILQ_INIT(&pci_bars); 1414 1415 /* 1416 * Add some slop to the I/O and memory resources decoded by 1417 * this bus to give a guest some flexibility if it wants to 1418 * reprogram the BARs. 1419 */ 1420 pci_emul_iobase += BUSIO_ROUNDUP; 1421 pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP); 1422 bi->iolimit = pci_emul_iobase; 1423 1424 pci_emul_membase32 += BUSMEM32_ROUNDUP; 1425 pci_emul_membase32 = roundup2(pci_emul_membase32, 1426 BUSMEM32_ROUNDUP); 1427 bi->memlimit32 = pci_emul_membase32; 1428 1429 pci_emul_membase64 += BUSMEM64_ROUNDUP; 1430 pci_emul_membase64 = roundup2(pci_emul_membase64, 1431 BUSMEM64_ROUNDUP); 1432 bi->memlimit64 = pci_emul_membase64; 1433 } 1434 1435 /* 1436 * PCI backends are initialized before routing INTx interrupts 1437 * so that LPC devices are able to reserve ISA IRQs before 1438 * routing PIRQ pins. 1439 */ 1440 for (bus = 0; bus < MAXBUSES; bus++) { 1441 if ((bi = pci_businfo[bus]) == NULL) 1442 continue; 1443 1444 for (slot = 0; slot < MAXSLOTS; slot++) { 1445 si = &bi->slotinfo[slot]; 1446 for (func = 0; func < MAXFUNCS; func++) { 1447 fi = &si->si_funcs[func]; 1448 if (fi->fi_devi == NULL) 1449 continue; 1450 pci_lintr_route(fi->fi_devi); 1451 } 1452 } 1453 } 1454 lpc_pirq_routed(); 1455 1456 /* 1457 * The guest physical memory map looks like the following: 1458 * [0, lowmem) guest system memory 1459 * [lowmem, 0xC0000000) memory hole (may be absent) 1460 * [0xC0000000, 0xE0000000) PCI hole (32-bit BAR allocation) 1461 * [0xE0000000, 0xF0000000) PCI extended config window 1462 * [0xF0000000, 4GB) LAPIC, IOAPIC, HPET, firmware 1463 * [4GB, 4GB + highmem) 1464 */ 1465 1466 /* 1467 * Accesses to memory addresses that are not allocated to system 1468 * memory or PCI devices return 0xff's. 1469 */ 1470 lowmem = vm_get_lowmem_size(ctx); 1471 bzero(&mr, sizeof(struct mem_range)); 1472 mr.name = "PCI hole"; 1473 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE; 1474 mr.base = lowmem; 1475 mr.size = (4ULL * 1024 * 1024 * 1024) - lowmem; 1476 mr.handler = pci_emul_fallback_handler; 1477 error = register_mem_fallback(&mr); 1478 assert(error == 0); 1479 1480 /* PCI extended config space */ 1481 bzero(&mr, sizeof(struct mem_range)); 1482 mr.name = "PCI ECFG"; 1483 mr.flags = MEM_F_RW | MEM_F_IMMUTABLE; 1484 mr.base = PCI_EMUL_ECFG_BASE; 1485 mr.size = PCI_EMUL_ECFG_SIZE; 1486 mr.handler = pci_emul_ecfg_handler; 1487 error = register_mem(&mr); 1488 assert(error == 0); 1489 1490 return (0); 1491 } 1492 1493 static void 1494 pci_apic_prt_entry(int bus __unused, int slot, int pin, int pirq_pin __unused, 1495 int ioapic_irq, void *arg __unused) 1496 { 1497 1498 dsdt_line(" Package ()"); 1499 dsdt_line(" {"); 1500 dsdt_line(" 0x%X,", slot << 16 | 0xffff); 1501 dsdt_line(" 0x%02X,", pin - 1); 1502 dsdt_line(" Zero,"); 1503 dsdt_line(" 0x%X", ioapic_irq); 1504 dsdt_line(" },"); 1505 } 1506 1507 static void 1508 pci_pirq_prt_entry(int bus __unused, int slot, int pin, int pirq_pin, 1509 int ioapic_irq __unused, void *arg __unused) 1510 { 1511 char *name; 1512 1513 name = lpc_pirq_name(pirq_pin); 1514 if (name == NULL) 1515 return; 1516 dsdt_line(" Package ()"); 1517 dsdt_line(" {"); 1518 dsdt_line(" 0x%X,", slot << 16 | 0xffff); 1519 dsdt_line(" 0x%02X,", pin - 1); 1520 dsdt_line(" %s,", name); 1521 dsdt_line(" 0x00"); 1522 dsdt_line(" },"); 1523 free(name); 1524 } 1525 1526 /* 1527 * A bhyve virtual machine has a flat PCI hierarchy with a root port 1528 * corresponding to each PCI bus. 1529 */ 1530 static void 1531 pci_bus_write_dsdt(int bus) 1532 { 1533 struct businfo *bi; 1534 struct slotinfo *si; 1535 struct pci_devinst *pi; 1536 int count, func, slot; 1537 1538 /* 1539 * If there are no devices on this 'bus' then just return. 1540 */ 1541 if ((bi = pci_businfo[bus]) == NULL) { 1542 /* 1543 * Bus 0 is special because it decodes the I/O ports used 1544 * for PCI config space access even if there are no devices 1545 * on it. 1546 */ 1547 if (bus != 0) 1548 return; 1549 } 1550 1551 dsdt_line(" Device (PC%02X)", bus); 1552 dsdt_line(" {"); 1553 dsdt_line(" Name (_HID, EisaId (\"PNP0A03\"))"); 1554 1555 dsdt_line(" Method (_BBN, 0, NotSerialized)"); 1556 dsdt_line(" {"); 1557 dsdt_line(" Return (0x%08X)", bus); 1558 dsdt_line(" }"); 1559 dsdt_line(" Name (_CRS, ResourceTemplate ()"); 1560 dsdt_line(" {"); 1561 dsdt_line(" WordBusNumber (ResourceProducer, MinFixed, " 1562 "MaxFixed, PosDecode,"); 1563 dsdt_line(" 0x0000, // Granularity"); 1564 dsdt_line(" 0x%04X, // Range Minimum", bus); 1565 dsdt_line(" 0x%04X, // Range Maximum", bus); 1566 dsdt_line(" 0x0000, // Translation Offset"); 1567 dsdt_line(" 0x0001, // Length"); 1568 dsdt_line(" ,, )"); 1569 1570 if (bus == 0) { 1571 dsdt_indent(3); 1572 dsdt_fixed_ioport(0xCF8, 8); 1573 dsdt_unindent(3); 1574 1575 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " 1576 "PosDecode, EntireRange,"); 1577 dsdt_line(" 0x0000, // Granularity"); 1578 dsdt_line(" 0x0000, // Range Minimum"); 1579 dsdt_line(" 0x0CF7, // Range Maximum"); 1580 dsdt_line(" 0x0000, // Translation Offset"); 1581 dsdt_line(" 0x0CF8, // Length"); 1582 dsdt_line(" ,, , TypeStatic)"); 1583 1584 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " 1585 "PosDecode, EntireRange,"); 1586 dsdt_line(" 0x0000, // Granularity"); 1587 dsdt_line(" 0x0D00, // Range Minimum"); 1588 dsdt_line(" 0x%04X, // Range Maximum", 1589 PCI_EMUL_IOBASE - 1); 1590 dsdt_line(" 0x0000, // Translation Offset"); 1591 dsdt_line(" 0x%04X, // Length", 1592 PCI_EMUL_IOBASE - 0x0D00); 1593 dsdt_line(" ,, , TypeStatic)"); 1594 1595 if (bi == NULL) { 1596 dsdt_line(" })"); 1597 goto done; 1598 } 1599 } 1600 assert(bi != NULL); 1601 1602 /* i/o window */ 1603 dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " 1604 "PosDecode, EntireRange,"); 1605 dsdt_line(" 0x0000, // Granularity"); 1606 dsdt_line(" 0x%04X, // Range Minimum", bi->iobase); 1607 dsdt_line(" 0x%04X, // Range Maximum", 1608 bi->iolimit - 1); 1609 dsdt_line(" 0x0000, // Translation Offset"); 1610 dsdt_line(" 0x%04X, // Length", 1611 bi->iolimit - bi->iobase); 1612 dsdt_line(" ,, , TypeStatic)"); 1613 1614 /* mmio window (32-bit) */ 1615 dsdt_line(" DWordMemory (ResourceProducer, PosDecode, " 1616 "MinFixed, MaxFixed, NonCacheable, ReadWrite,"); 1617 dsdt_line(" 0x00000000, // Granularity"); 1618 dsdt_line(" 0x%08X, // Range Minimum\n", bi->membase32); 1619 dsdt_line(" 0x%08X, // Range Maximum\n", 1620 bi->memlimit32 - 1); 1621 dsdt_line(" 0x00000000, // Translation Offset"); 1622 dsdt_line(" 0x%08X, // Length\n", 1623 bi->memlimit32 - bi->membase32); 1624 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)"); 1625 1626 /* mmio window (64-bit) */ 1627 dsdt_line(" QWordMemory (ResourceProducer, PosDecode, " 1628 "MinFixed, MaxFixed, NonCacheable, ReadWrite,"); 1629 dsdt_line(" 0x0000000000000000, // Granularity"); 1630 dsdt_line(" 0x%016lX, // Range Minimum\n", bi->membase64); 1631 dsdt_line(" 0x%016lX, // Range Maximum\n", 1632 bi->memlimit64 - 1); 1633 dsdt_line(" 0x0000000000000000, // Translation Offset"); 1634 dsdt_line(" 0x%016lX, // Length\n", 1635 bi->memlimit64 - bi->membase64); 1636 dsdt_line(" ,, , AddressRangeMemory, TypeStatic)"); 1637 dsdt_line(" })"); 1638 1639 count = pci_count_lintr(bus); 1640 if (count != 0) { 1641 dsdt_indent(2); 1642 dsdt_line("Name (PPRT, Package ()"); 1643 dsdt_line("{"); 1644 pci_walk_lintr(bus, pci_pirq_prt_entry, NULL); 1645 dsdt_line("})"); 1646 dsdt_line("Name (APRT, Package ()"); 1647 dsdt_line("{"); 1648 pci_walk_lintr(bus, pci_apic_prt_entry, NULL); 1649 dsdt_line("})"); 1650 dsdt_line("Method (_PRT, 0, NotSerialized)"); 1651 dsdt_line("{"); 1652 dsdt_line(" If (PICM)"); 1653 dsdt_line(" {"); 1654 dsdt_line(" Return (APRT)"); 1655 dsdt_line(" }"); 1656 dsdt_line(" Else"); 1657 dsdt_line(" {"); 1658 dsdt_line(" Return (PPRT)"); 1659 dsdt_line(" }"); 1660 dsdt_line("}"); 1661 dsdt_unindent(2); 1662 } 1663 1664 dsdt_indent(2); 1665 for (slot = 0; slot < MAXSLOTS; slot++) { 1666 si = &bi->slotinfo[slot]; 1667 for (func = 0; func < MAXFUNCS; func++) { 1668 pi = si->si_funcs[func].fi_devi; 1669 if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL) 1670 pi->pi_d->pe_write_dsdt(pi); 1671 } 1672 } 1673 dsdt_unindent(2); 1674 done: 1675 dsdt_line(" }"); 1676 } 1677 1678 void 1679 pci_write_dsdt(void) 1680 { 1681 int bus; 1682 1683 dsdt_indent(1); 1684 dsdt_line("Name (PICM, 0x00)"); 1685 dsdt_line("Method (_PIC, 1, NotSerialized)"); 1686 dsdt_line("{"); 1687 dsdt_line(" Store (Arg0, PICM)"); 1688 dsdt_line("}"); 1689 dsdt_line(""); 1690 dsdt_line("Scope (_SB)"); 1691 dsdt_line("{"); 1692 for (bus = 0; bus < MAXBUSES; bus++) 1693 pci_bus_write_dsdt(bus); 1694 dsdt_line("}"); 1695 dsdt_unindent(1); 1696 } 1697 1698 int 1699 pci_bus_configured(int bus) 1700 { 1701 assert(bus >= 0 && bus < MAXBUSES); 1702 return (pci_businfo[bus] != NULL); 1703 } 1704 1705 int 1706 pci_msi_enabled(struct pci_devinst *pi) 1707 { 1708 return (pi->pi_msi.enabled); 1709 } 1710 1711 int 1712 pci_msi_maxmsgnum(struct pci_devinst *pi) 1713 { 1714 if (pi->pi_msi.enabled) 1715 return (pi->pi_msi.maxmsgnum); 1716 else 1717 return (0); 1718 } 1719 1720 int 1721 pci_msix_enabled(struct pci_devinst *pi) 1722 { 1723 1724 return (pi->pi_msix.enabled && !pi->pi_msi.enabled); 1725 } 1726 1727 void 1728 pci_generate_msix(struct pci_devinst *pi, int index) 1729 { 1730 struct msix_table_entry *mte; 1731 1732 if (!pci_msix_enabled(pi)) 1733 return; 1734 1735 if (pi->pi_msix.function_mask) 1736 return; 1737 1738 if (index >= pi->pi_msix.table_count) 1739 return; 1740 1741 mte = &pi->pi_msix.table[index]; 1742 if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { 1743 /* XXX Set PBA bit if interrupt is disabled */ 1744 vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data); 1745 } 1746 } 1747 1748 void 1749 pci_generate_msi(struct pci_devinst *pi, int index) 1750 { 1751 1752 if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) { 1753 vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr, 1754 pi->pi_msi.msg_data + index); 1755 } 1756 } 1757 1758 static bool 1759 pci_lintr_permitted(struct pci_devinst *pi) 1760 { 1761 uint16_t cmd; 1762 1763 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); 1764 return (!(pi->pi_msi.enabled || pi->pi_msix.enabled || 1765 (cmd & PCIM_CMD_INTxDIS))); 1766 } 1767 1768 void 1769 pci_lintr_request(struct pci_devinst *pi) 1770 { 1771 struct businfo *bi; 1772 struct slotinfo *si; 1773 int bestpin, bestcount, pin; 1774 1775 bi = pci_businfo[pi->pi_bus]; 1776 assert(bi != NULL); 1777 1778 /* 1779 * Just allocate a pin from our slot. The pin will be 1780 * assigned IRQs later when interrupts are routed. 1781 */ 1782 si = &bi->slotinfo[pi->pi_slot]; 1783 bestpin = 0; 1784 bestcount = si->si_intpins[0].ii_count; 1785 for (pin = 1; pin < 4; pin++) { 1786 if (si->si_intpins[pin].ii_count < bestcount) { 1787 bestpin = pin; 1788 bestcount = si->si_intpins[pin].ii_count; 1789 } 1790 } 1791 1792 si->si_intpins[bestpin].ii_count++; 1793 pi->pi_lintr.pin = bestpin + 1; 1794 pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1); 1795 } 1796 1797 static void 1798 pci_lintr_route(struct pci_devinst *pi) 1799 { 1800 struct businfo *bi; 1801 struct intxinfo *ii; 1802 1803 if (pi->pi_lintr.pin == 0) 1804 return; 1805 1806 bi = pci_businfo[pi->pi_bus]; 1807 assert(bi != NULL); 1808 ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1]; 1809 1810 /* 1811 * Attempt to allocate an I/O APIC pin for this intpin if one 1812 * is not yet assigned. 1813 */ 1814 if (ii->ii_ioapic_irq == 0) 1815 ii->ii_ioapic_irq = ioapic_pci_alloc_irq(pi); 1816 assert(ii->ii_ioapic_irq > 0); 1817 1818 /* 1819 * Attempt to allocate a PIRQ pin for this intpin if one is 1820 * not yet assigned. 1821 */ 1822 if (ii->ii_pirq_pin == 0) 1823 ii->ii_pirq_pin = pirq_alloc_pin(pi); 1824 assert(ii->ii_pirq_pin > 0); 1825 1826 pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq; 1827 pi->pi_lintr.pirq_pin = ii->ii_pirq_pin; 1828 pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin)); 1829 } 1830 1831 void 1832 pci_lintr_assert(struct pci_devinst *pi) 1833 { 1834 1835 assert(pi->pi_lintr.pin > 0); 1836 1837 pthread_mutex_lock(&pi->pi_lintr.lock); 1838 if (pi->pi_lintr.state == IDLE) { 1839 if (pci_lintr_permitted(pi)) { 1840 pi->pi_lintr.state = ASSERTED; 1841 pci_irq_assert(pi); 1842 } else 1843 pi->pi_lintr.state = PENDING; 1844 } 1845 pthread_mutex_unlock(&pi->pi_lintr.lock); 1846 } 1847 1848 void 1849 pci_lintr_deassert(struct pci_devinst *pi) 1850 { 1851 1852 assert(pi->pi_lintr.pin > 0); 1853 1854 pthread_mutex_lock(&pi->pi_lintr.lock); 1855 if (pi->pi_lintr.state == ASSERTED) { 1856 pi->pi_lintr.state = IDLE; 1857 pci_irq_deassert(pi); 1858 } else if (pi->pi_lintr.state == PENDING) 1859 pi->pi_lintr.state = IDLE; 1860 pthread_mutex_unlock(&pi->pi_lintr.lock); 1861 } 1862 1863 static void 1864 pci_lintr_update(struct pci_devinst *pi) 1865 { 1866 1867 pthread_mutex_lock(&pi->pi_lintr.lock); 1868 if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) { 1869 pci_irq_deassert(pi); 1870 pi->pi_lintr.state = PENDING; 1871 } else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) { 1872 pi->pi_lintr.state = ASSERTED; 1873 pci_irq_assert(pi); 1874 } 1875 pthread_mutex_unlock(&pi->pi_lintr.lock); 1876 } 1877 1878 int 1879 pci_count_lintr(int bus) 1880 { 1881 int count, slot, pin; 1882 struct slotinfo *slotinfo; 1883 1884 count = 0; 1885 if (pci_businfo[bus] != NULL) { 1886 for (slot = 0; slot < MAXSLOTS; slot++) { 1887 slotinfo = &pci_businfo[bus]->slotinfo[slot]; 1888 for (pin = 0; pin < 4; pin++) { 1889 if (slotinfo->si_intpins[pin].ii_count != 0) 1890 count++; 1891 } 1892 } 1893 } 1894 return (count); 1895 } 1896 1897 void 1898 pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg) 1899 { 1900 struct businfo *bi; 1901 struct slotinfo *si; 1902 struct intxinfo *ii; 1903 int slot, pin; 1904 1905 if ((bi = pci_businfo[bus]) == NULL) 1906 return; 1907 1908 for (slot = 0; slot < MAXSLOTS; slot++) { 1909 si = &bi->slotinfo[slot]; 1910 for (pin = 0; pin < 4; pin++) { 1911 ii = &si->si_intpins[pin]; 1912 if (ii->ii_count != 0) 1913 cb(bus, slot, pin + 1, ii->ii_pirq_pin, 1914 ii->ii_ioapic_irq, arg); 1915 } 1916 } 1917 } 1918 1919 /* 1920 * Return 1 if the emulated device in 'slot' is a multi-function device. 1921 * Return 0 otherwise. 1922 */ 1923 static int 1924 pci_emul_is_mfdev(int bus, int slot) 1925 { 1926 struct businfo *bi; 1927 struct slotinfo *si; 1928 int f, numfuncs; 1929 1930 numfuncs = 0; 1931 if ((bi = pci_businfo[bus]) != NULL) { 1932 si = &bi->slotinfo[slot]; 1933 for (f = 0; f < MAXFUNCS; f++) { 1934 if (si->si_funcs[f].fi_devi != NULL) { 1935 numfuncs++; 1936 } 1937 } 1938 } 1939 return (numfuncs > 1); 1940 } 1941 1942 /* 1943 * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on 1944 * whether or not is a multi-function being emulated in the pci 'slot'. 1945 */ 1946 static void 1947 pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv) 1948 { 1949 int mfdev; 1950 1951 if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) { 1952 mfdev = pci_emul_is_mfdev(bus, slot); 1953 switch (bytes) { 1954 case 1: 1955 case 2: 1956 *rv &= ~PCIM_MFDEV; 1957 if (mfdev) { 1958 *rv |= PCIM_MFDEV; 1959 } 1960 break; 1961 case 4: 1962 *rv &= ~(PCIM_MFDEV << 16); 1963 if (mfdev) { 1964 *rv |= (PCIM_MFDEV << 16); 1965 } 1966 break; 1967 } 1968 } 1969 } 1970 1971 /* 1972 * Update device state in response to changes to the PCI command 1973 * register. 1974 */ 1975 void 1976 pci_emul_cmd_changed(struct pci_devinst *pi, uint16_t old) 1977 { 1978 int i; 1979 uint16_t changed, new; 1980 1981 new = pci_get_cfgdata16(pi, PCIR_COMMAND); 1982 changed = old ^ new; 1983 1984 /* 1985 * If the MMIO or I/O address space decoding has changed then 1986 * register/unregister all BARs that decode that address space. 1987 */ 1988 for (i = 0; i <= PCI_BARMAX_WITH_ROM; i++) { 1989 switch (pi->pi_bar[i].type) { 1990 case PCIBAR_NONE: 1991 case PCIBAR_MEMHI64: 1992 break; 1993 case PCIBAR_IO: 1994 /* I/O address space decoding changed? */ 1995 if (changed & PCIM_CMD_PORTEN) { 1996 if (new & PCIM_CMD_PORTEN) 1997 register_bar(pi, i); 1998 else 1999 unregister_bar(pi, i); 2000 } 2001 break; 2002 case PCIBAR_ROM: 2003 /* skip (un-)register of ROM if it disabled */ 2004 if (!romen(pi)) 2005 break; 2006 /* fallthrough */ 2007 case PCIBAR_MEM32: 2008 case PCIBAR_MEM64: 2009 /* MMIO address space decoding changed? */ 2010 if (changed & PCIM_CMD_MEMEN) { 2011 if (new & PCIM_CMD_MEMEN) 2012 register_bar(pi, i); 2013 else 2014 unregister_bar(pi, i); 2015 } 2016 break; 2017 default: 2018 assert(0); 2019 } 2020 } 2021 2022 /* 2023 * If INTx has been unmasked and is pending, assert the 2024 * interrupt. 2025 */ 2026 pci_lintr_update(pi); 2027 } 2028 2029 static void 2030 pci_emul_cmdsts_write(struct pci_devinst *pi, int coff, uint32_t new, int bytes) 2031 { 2032 int rshift; 2033 uint32_t cmd, old, readonly; 2034 2035 cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); /* stash old value */ 2036 2037 /* 2038 * From PCI Local Bus Specification 3.0 sections 6.2.2 and 6.2.3. 2039 * 2040 * XXX Bits 8, 11, 12, 13, 14 and 15 in the status register are 2041 * 'write 1 to clear'. However these bits are not set to '1' by 2042 * any device emulation so it is simpler to treat them as readonly. 2043 */ 2044 rshift = (coff & 0x3) * 8; 2045 readonly = 0xFFFFF880 >> rshift; 2046 2047 old = CFGREAD(pi, coff, bytes); 2048 new &= ~readonly; 2049 new |= (old & readonly); 2050 CFGWRITE(pi, coff, new, bytes); /* update config */ 2051 2052 pci_emul_cmd_changed(pi, cmd); 2053 } 2054 2055 static void 2056 pci_cfgrw(struct vmctx *ctx, int vcpu, int in, int bus, int slot, int func, 2057 int coff, int bytes, uint32_t *eax) 2058 { 2059 struct businfo *bi; 2060 struct slotinfo *si; 2061 struct pci_devinst *pi; 2062 struct pci_devemu *pe; 2063 int idx, needcfg; 2064 uint64_t addr, bar, mask; 2065 2066 if ((bi = pci_businfo[bus]) != NULL) { 2067 si = &bi->slotinfo[slot]; 2068 pi = si->si_funcs[func].fi_devi; 2069 } else 2070 pi = NULL; 2071 2072 /* 2073 * Just return if there is no device at this slot:func or if the 2074 * the guest is doing an un-aligned access. 2075 */ 2076 if (pi == NULL || (bytes != 1 && bytes != 2 && bytes != 4) || 2077 (coff & (bytes - 1)) != 0) { 2078 if (in) 2079 *eax = 0xffffffff; 2080 return; 2081 } 2082 2083 /* 2084 * Ignore all writes beyond the standard config space and return all 2085 * ones on reads. 2086 */ 2087 if (coff >= PCI_REGMAX + 1) { 2088 if (in) { 2089 *eax = 0xffffffff; 2090 /* 2091 * Extended capabilities begin at offset 256 in config 2092 * space. Absence of extended capabilities is signaled 2093 * with all 0s in the extended capability header at 2094 * offset 256. 2095 */ 2096 if (coff <= PCI_REGMAX + 4) 2097 *eax = 0x00000000; 2098 } 2099 return; 2100 } 2101 2102 pe = pi->pi_d; 2103 2104 /* 2105 * Config read 2106 */ 2107 if (in) { 2108 /* Let the device emulation override the default handler */ 2109 if (pe->pe_cfgread != NULL) { 2110 needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes, 2111 eax); 2112 } else { 2113 needcfg = 1; 2114 } 2115 2116 if (needcfg) 2117 *eax = CFGREAD(pi, coff, bytes); 2118 2119 pci_emul_hdrtype_fixup(bus, slot, coff, bytes, eax); 2120 } else { 2121 /* Let the device emulation override the default handler */ 2122 if (pe->pe_cfgwrite != NULL && 2123 (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0) 2124 return; 2125 2126 /* 2127 * Special handling for write to BAR and ROM registers 2128 */ 2129 if (is_pcir_bar(coff) || is_pcir_bios(coff)) { 2130 /* 2131 * Ignore writes to BAR registers that are not 2132 * 4-byte aligned. 2133 */ 2134 if (bytes != 4 || (coff & 0x3) != 0) 2135 return; 2136 2137 if (is_pcir_bar(coff)) { 2138 idx = (coff - PCIR_BAR(0)) / 4; 2139 } else if (is_pcir_bios(coff)) { 2140 idx = PCI_ROM_IDX; 2141 } else { 2142 errx(4, "%s: invalid BAR offset %d", __func__, 2143 coff); 2144 } 2145 2146 mask = ~(pi->pi_bar[idx].size - 1); 2147 switch (pi->pi_bar[idx].type) { 2148 case PCIBAR_NONE: 2149 pi->pi_bar[idx].addr = bar = 0; 2150 break; 2151 case PCIBAR_IO: 2152 addr = *eax & mask; 2153 addr &= 0xffff; 2154 bar = addr | pi->pi_bar[idx].lobits; 2155 /* 2156 * Register the new BAR value for interception 2157 */ 2158 if (addr != pi->pi_bar[idx].addr) { 2159 update_bar_address(pi, addr, idx, 2160 PCIBAR_IO); 2161 } 2162 break; 2163 case PCIBAR_MEM32: 2164 addr = bar = *eax & mask; 2165 bar |= pi->pi_bar[idx].lobits; 2166 if (addr != pi->pi_bar[idx].addr) { 2167 update_bar_address(pi, addr, idx, 2168 PCIBAR_MEM32); 2169 } 2170 break; 2171 case PCIBAR_MEM64: 2172 addr = bar = *eax & mask; 2173 bar |= pi->pi_bar[idx].lobits; 2174 if (addr != (uint32_t)pi->pi_bar[idx].addr) { 2175 update_bar_address(pi, addr, idx, 2176 PCIBAR_MEM64); 2177 } 2178 break; 2179 case PCIBAR_MEMHI64: 2180 mask = ~(pi->pi_bar[idx - 1].size - 1); 2181 addr = ((uint64_t)*eax << 32) & mask; 2182 bar = addr >> 32; 2183 if (bar != pi->pi_bar[idx - 1].addr >> 32) { 2184 update_bar_address(pi, addr, idx - 1, 2185 PCIBAR_MEMHI64); 2186 } 2187 break; 2188 case PCIBAR_ROM: 2189 addr = bar = *eax & mask; 2190 if (memen(pi) && romen(pi)) { 2191 unregister_bar(pi, idx); 2192 } 2193 pi->pi_bar[idx].addr = addr; 2194 pi->pi_bar[idx].lobits = *eax & 2195 PCIM_BIOS_ENABLE; 2196 /* romen could have changed it value */ 2197 if (memen(pi) && romen(pi)) { 2198 register_bar(pi, idx); 2199 } 2200 bar |= pi->pi_bar[idx].lobits; 2201 break; 2202 default: 2203 assert(0); 2204 } 2205 pci_set_cfgdata32(pi, coff, bar); 2206 2207 } else if (pci_emul_iscap(pi, coff)) { 2208 pci_emul_capwrite(pi, coff, bytes, *eax, 0, 0); 2209 } else if (coff >= PCIR_COMMAND && coff < PCIR_REVID) { 2210 pci_emul_cmdsts_write(pi, coff, *eax, bytes); 2211 } else { 2212 CFGWRITE(pi, coff, *eax, bytes); 2213 } 2214 } 2215 } 2216 2217 static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff; 2218 2219 static int 2220 pci_emul_cfgaddr(struct vmctx *ctx __unused, int vcpu __unused, int in, 2221 int port __unused, int bytes, uint32_t *eax, void *arg __unused) 2222 { 2223 uint32_t x; 2224 2225 if (bytes != 4) { 2226 if (in) 2227 *eax = (bytes == 2) ? 0xffff : 0xff; 2228 return (0); 2229 } 2230 2231 if (in) { 2232 x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff; 2233 if (cfgenable) 2234 x |= CONF1_ENABLE; 2235 *eax = x; 2236 } else { 2237 x = *eax; 2238 cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE; 2239 cfgoff = (x & PCI_REGMAX) & ~0x03; 2240 cfgfunc = (x >> 8) & PCI_FUNCMAX; 2241 cfgslot = (x >> 11) & PCI_SLOTMAX; 2242 cfgbus = (x >> 16) & PCI_BUSMAX; 2243 } 2244 2245 return (0); 2246 } 2247 INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr); 2248 2249 static int 2250 pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes, 2251 uint32_t *eax, void *arg __unused) 2252 { 2253 int coff; 2254 2255 assert(bytes == 1 || bytes == 2 || bytes == 4); 2256 2257 coff = cfgoff + (port - CONF1_DATA_PORT); 2258 if (cfgenable) { 2259 pci_cfgrw(ctx, vcpu, in, cfgbus, cfgslot, cfgfunc, coff, bytes, 2260 eax); 2261 } else { 2262 /* Ignore accesses to cfgdata if not enabled by cfgaddr */ 2263 if (in) 2264 *eax = 0xffffffff; 2265 } 2266 return (0); 2267 } 2268 2269 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata); 2270 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata); 2271 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata); 2272 INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata); 2273 2274 #ifdef BHYVE_SNAPSHOT 2275 /* 2276 * Saves/restores PCI device emulated state. Returns 0 on success. 2277 */ 2278 static int 2279 pci_snapshot_pci_dev(struct vm_snapshot_meta *meta) 2280 { 2281 struct pci_devinst *pi; 2282 int i; 2283 int ret; 2284 2285 pi = meta->dev_data; 2286 2287 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.enabled, meta, ret, done); 2288 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.addr, meta, ret, done); 2289 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.msg_data, meta, ret, done); 2290 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msi.maxmsgnum, meta, ret, done); 2291 2292 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.enabled, meta, ret, done); 2293 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_bar, meta, ret, done); 2294 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_bar, meta, ret, done); 2295 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_offset, meta, ret, done); 2296 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table_count, meta, ret, done); 2297 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_offset, meta, ret, done); 2298 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.pba_size, meta, ret, done); 2299 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.function_mask, meta, ret, done); 2300 2301 SNAPSHOT_BUF_OR_LEAVE(pi->pi_cfgdata, sizeof(pi->pi_cfgdata), 2302 meta, ret, done); 2303 2304 for (i = 0; i < (int)nitems(pi->pi_bar); i++) { 2305 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].type, meta, ret, done); 2306 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].size, meta, ret, done); 2307 SNAPSHOT_VAR_OR_LEAVE(pi->pi_bar[i].addr, meta, ret, done); 2308 } 2309 2310 /* Restore MSI-X table. */ 2311 for (i = 0; i < pi->pi_msix.table_count; i++) { 2312 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].addr, 2313 meta, ret, done); 2314 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].msg_data, 2315 meta, ret, done); 2316 SNAPSHOT_VAR_OR_LEAVE(pi->pi_msix.table[i].vector_control, 2317 meta, ret, done); 2318 } 2319 2320 done: 2321 return (ret); 2322 } 2323 2324 static int 2325 pci_find_slotted_dev(const char *dev_name, struct pci_devemu **pde, 2326 struct pci_devinst **pdi) 2327 { 2328 struct businfo *bi; 2329 struct slotinfo *si; 2330 struct funcinfo *fi; 2331 int bus, slot, func; 2332 2333 assert(dev_name != NULL); 2334 assert(pde != NULL); 2335 assert(pdi != NULL); 2336 2337 for (bus = 0; bus < MAXBUSES; bus++) { 2338 if ((bi = pci_businfo[bus]) == NULL) 2339 continue; 2340 2341 for (slot = 0; slot < MAXSLOTS; slot++) { 2342 si = &bi->slotinfo[slot]; 2343 for (func = 0; func < MAXFUNCS; func++) { 2344 fi = &si->si_funcs[func]; 2345 if (fi->fi_pde == NULL) 2346 continue; 2347 if (strcmp(dev_name, fi->fi_pde->pe_emu) != 0) 2348 continue; 2349 2350 *pde = fi->fi_pde; 2351 *pdi = fi->fi_devi; 2352 return (0); 2353 } 2354 } 2355 } 2356 2357 return (EINVAL); 2358 } 2359 2360 int 2361 pci_snapshot(struct vm_snapshot_meta *meta) 2362 { 2363 struct pci_devemu *pde; 2364 struct pci_devinst *pdi; 2365 int ret; 2366 2367 assert(meta->dev_name != NULL); 2368 2369 ret = pci_find_slotted_dev(meta->dev_name, &pde, &pdi); 2370 if (ret != 0) { 2371 fprintf(stderr, "%s: no such name: %s\r\n", 2372 __func__, meta->dev_name); 2373 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size); 2374 return (0); 2375 } 2376 2377 meta->dev_data = pdi; 2378 2379 if (pde->pe_snapshot == NULL) { 2380 fprintf(stderr, "%s: not implemented yet for: %s\r\n", 2381 __func__, meta->dev_name); 2382 return (-1); 2383 } 2384 2385 ret = pci_snapshot_pci_dev(meta); 2386 if (ret != 0) { 2387 fprintf(stderr, "%s: failed to snapshot pci dev\r\n", 2388 __func__); 2389 return (-1); 2390 } 2391 2392 ret = (*pde->pe_snapshot)(meta); 2393 2394 return (ret); 2395 } 2396 2397 int 2398 pci_pause(struct vmctx *ctx, const char *dev_name) 2399 { 2400 struct pci_devemu *pde; 2401 struct pci_devinst *pdi; 2402 int ret; 2403 2404 assert(dev_name != NULL); 2405 2406 ret = pci_find_slotted_dev(dev_name, &pde, &pdi); 2407 if (ret != 0) { 2408 /* 2409 * It is possible to call this function without 2410 * checking that the device is inserted first. 2411 */ 2412 fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name); 2413 return (0); 2414 } 2415 2416 if (pde->pe_pause == NULL) { 2417 /* The pause/resume functionality is optional. */ 2418 fprintf(stderr, "%s: not implemented for: %s\n", 2419 __func__, dev_name); 2420 return (0); 2421 } 2422 2423 return (*pde->pe_pause)(ctx, pdi); 2424 } 2425 2426 int 2427 pci_resume(struct vmctx *ctx, const char *dev_name) 2428 { 2429 struct pci_devemu *pde; 2430 struct pci_devinst *pdi; 2431 int ret; 2432 2433 assert(dev_name != NULL); 2434 2435 ret = pci_find_slotted_dev(dev_name, &pde, &pdi); 2436 if (ret != 0) { 2437 /* 2438 * It is possible to call this function without 2439 * checking that the device is inserted first. 2440 */ 2441 fprintf(stderr, "%s: no such name: %s\n", __func__, dev_name); 2442 return (0); 2443 } 2444 2445 if (pde->pe_resume == NULL) { 2446 /* The pause/resume functionality is optional. */ 2447 fprintf(stderr, "%s: not implemented for: %s\n", 2448 __func__, dev_name); 2449 return (0); 2450 } 2451 2452 return (*pde->pe_resume)(ctx, pdi); 2453 } 2454 #endif 2455 2456 #define PCI_EMUL_TEST 2457 #ifdef PCI_EMUL_TEST 2458 /* 2459 * Define a dummy test device 2460 */ 2461 #define DIOSZ 8 2462 #define DMEMSZ 4096 2463 struct pci_emul_dsoftc { 2464 uint8_t ioregs[DIOSZ]; 2465 uint8_t memregs[2][DMEMSZ]; 2466 }; 2467 2468 #define PCI_EMUL_MSI_MSGS 4 2469 #define PCI_EMUL_MSIX_MSGS 16 2470 2471 static int 2472 pci_emul_dinit(struct vmctx *ctx __unused, struct pci_devinst *pi, 2473 nvlist_t *nvl __unused) 2474 { 2475 int error; 2476 struct pci_emul_dsoftc *sc; 2477 2478 sc = calloc(1, sizeof(struct pci_emul_dsoftc)); 2479 2480 pi->pi_arg = sc; 2481 2482 pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001); 2483 pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD); 2484 pci_set_cfgdata8(pi, PCIR_CLASS, 0x02); 2485 2486 error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS); 2487 assert(error == 0); 2488 2489 error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ); 2490 assert(error == 0); 2491 2492 error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ); 2493 assert(error == 0); 2494 2495 error = pci_emul_alloc_bar(pi, 2, PCIBAR_MEM32, DMEMSZ); 2496 assert(error == 0); 2497 2498 return (0); 2499 } 2500 2501 static void 2502 pci_emul_diow(struct vmctx *ctx __unused, int vcpu __unused, 2503 struct pci_devinst *pi, int baridx, uint64_t offset, int size, 2504 uint64_t value) 2505 { 2506 int i; 2507 struct pci_emul_dsoftc *sc = pi->pi_arg; 2508 2509 if (baridx == 0) { 2510 if (offset + size > DIOSZ) { 2511 printf("diow: iow too large, offset %ld size %d\n", 2512 offset, size); 2513 return; 2514 } 2515 2516 if (size == 1) { 2517 sc->ioregs[offset] = value & 0xff; 2518 } else if (size == 2) { 2519 *(uint16_t *)&sc->ioregs[offset] = value & 0xffff; 2520 } else if (size == 4) { 2521 *(uint32_t *)&sc->ioregs[offset] = value; 2522 } else { 2523 printf("diow: iow unknown size %d\n", size); 2524 } 2525 2526 /* 2527 * Special magic value to generate an interrupt 2528 */ 2529 if (offset == 4 && size == 4 && pci_msi_enabled(pi)) 2530 pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi)); 2531 2532 if (value == 0xabcdef) { 2533 for (i = 0; i < pci_msi_maxmsgnum(pi); i++) 2534 pci_generate_msi(pi, i); 2535 } 2536 } 2537 2538 if (baridx == 1 || baridx == 2) { 2539 if (offset + size > DMEMSZ) { 2540 printf("diow: memw too large, offset %ld size %d\n", 2541 offset, size); 2542 return; 2543 } 2544 2545 i = baridx - 1; /* 'memregs' index */ 2546 2547 if (size == 1) { 2548 sc->memregs[i][offset] = value; 2549 } else if (size == 2) { 2550 *(uint16_t *)&sc->memregs[i][offset] = value; 2551 } else if (size == 4) { 2552 *(uint32_t *)&sc->memregs[i][offset] = value; 2553 } else if (size == 8) { 2554 *(uint64_t *)&sc->memregs[i][offset] = value; 2555 } else { 2556 printf("diow: memw unknown size %d\n", size); 2557 } 2558 2559 /* 2560 * magic interrupt ?? 2561 */ 2562 } 2563 2564 if (baridx > 2 || baridx < 0) { 2565 printf("diow: unknown bar idx %d\n", baridx); 2566 } 2567 } 2568 2569 static uint64_t 2570 pci_emul_dior(struct vmctx *ctx __unused, int vcpu __unused, 2571 struct pci_devinst *pi, int baridx, uint64_t offset, int size) 2572 { 2573 struct pci_emul_dsoftc *sc = pi->pi_arg; 2574 uint32_t value; 2575 int i; 2576 2577 if (baridx == 0) { 2578 if (offset + size > DIOSZ) { 2579 printf("dior: ior too large, offset %ld size %d\n", 2580 offset, size); 2581 return (0); 2582 } 2583 2584 value = 0; 2585 if (size == 1) { 2586 value = sc->ioregs[offset]; 2587 } else if (size == 2) { 2588 value = *(uint16_t *) &sc->ioregs[offset]; 2589 } else if (size == 4) { 2590 value = *(uint32_t *) &sc->ioregs[offset]; 2591 } else { 2592 printf("dior: ior unknown size %d\n", size); 2593 } 2594 } 2595 2596 if (baridx == 1 || baridx == 2) { 2597 if (offset + size > DMEMSZ) { 2598 printf("dior: memr too large, offset %ld size %d\n", 2599 offset, size); 2600 return (0); 2601 } 2602 2603 i = baridx - 1; /* 'memregs' index */ 2604 2605 if (size == 1) { 2606 value = sc->memregs[i][offset]; 2607 } else if (size == 2) { 2608 value = *(uint16_t *) &sc->memregs[i][offset]; 2609 } else if (size == 4) { 2610 value = *(uint32_t *) &sc->memregs[i][offset]; 2611 } else if (size == 8) { 2612 value = *(uint64_t *) &sc->memregs[i][offset]; 2613 } else { 2614 printf("dior: ior unknown size %d\n", size); 2615 } 2616 } 2617 2618 2619 if (baridx > 2 || baridx < 0) { 2620 printf("dior: unknown bar idx %d\n", baridx); 2621 return (0); 2622 } 2623 2624 return (value); 2625 } 2626 2627 #ifdef BHYVE_SNAPSHOT 2628 static int 2629 pci_emul_snapshot(struct vm_snapshot_meta *meta __unused) 2630 { 2631 return (0); 2632 } 2633 #endif 2634 2635 static const struct pci_devemu pci_dummy = { 2636 .pe_emu = "dummy", 2637 .pe_init = pci_emul_dinit, 2638 .pe_barwrite = pci_emul_diow, 2639 .pe_barread = pci_emul_dior, 2640 #ifdef BHYVE_SNAPSHOT 2641 .pe_snapshot = pci_emul_snapshot, 2642 #endif 2643 }; 2644 PCI_EMUL_SET(pci_dummy); 2645 2646 #endif /* PCI_EMUL_TEST */ 2647