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