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 #ifndef WITHOUT_CAPSICUM 36 #include <sys/capsicum.h> 37 #endif 38 #include <sys/types.h> 39 #include <sys/mman.h> 40 #include <sys/pciio.h> 41 #include <sys/ioctl.h> 42 43 #include <sys/pci.h> 44 45 #include <dev/io/iodev.h> 46 #include <dev/pci/pcireg.h> 47 48 #include <machine/iodev.h> 49 50 #ifndef WITHOUT_CAPSICUM 51 #include <capsicum_helpers.h> 52 #endif 53 #include <stdio.h> 54 #include <stdlib.h> 55 #include <string.h> 56 #include <err.h> 57 #include <errno.h> 58 #include <fcntl.h> 59 #include <sysexits.h> 60 #include <unistd.h> 61 62 #include <machine/vmm.h> 63 #include <vmmapi.h> 64 #include <sys/ppt_dev.h> 65 66 #include "config.h" 67 #include "debug.h" 68 #include "pci_emul.h" 69 #include "mem.h" 70 71 #define LEGACY_SUPPORT 1 72 73 #define MSIX_TABLE_COUNT(ctrl) (((ctrl) & PCIM_MSIXCTRL_TABLE_SIZE) + 1) 74 #define MSIX_CAPLEN 12 75 76 struct passthru_softc { 77 struct pci_devinst *psc_pi; 78 struct pcibar psc_bar[PCI_BARMAX + 1]; 79 struct { 80 int capoff; 81 int msgctrl; 82 int emulated; 83 } psc_msi; 84 struct { 85 int capoff; 86 } psc_msix; 87 int pptfd; 88 int msi_limit; 89 int msix_limit; 90 }; 91 92 static int 93 msi_caplen(int msgctrl) 94 { 95 int len; 96 97 len = 10; /* minimum length of msi capability */ 98 99 if (msgctrl & PCIM_MSICTRL_64BIT) 100 len += 4; 101 102 #if 0 103 /* 104 * Ignore the 'mask' and 'pending' bits in the MSI capability. 105 * We'll let the guest manipulate them directly. 106 */ 107 if (msgctrl & PCIM_MSICTRL_VECTOR) 108 len += 10; 109 #endif 110 111 return (len); 112 } 113 114 static uint32_t 115 read_config(const struct passthru_softc *sc, long reg, int width) 116 { 117 struct ppt_cfg_io pi; 118 119 pi.pci_off = reg; 120 pi.pci_width = width; 121 122 if (ioctl(sc->pptfd, PPT_CFG_READ, &pi) != 0) { 123 return (0); 124 } 125 return (pi.pci_data); 126 } 127 128 static void 129 write_config(const struct passthru_softc *sc, long reg, int width, 130 uint32_t data) 131 { 132 struct ppt_cfg_io pi; 133 134 pi.pci_off = reg; 135 pi.pci_width = width; 136 pi.pci_data = data; 137 138 (void) ioctl(sc->pptfd, PPT_CFG_WRITE, &pi); 139 } 140 141 static int 142 passthru_get_bar(struct passthru_softc *sc, int bar, enum pcibar_type *type, 143 uint64_t *base, uint64_t *size) 144 { 145 struct ppt_bar_query pb; 146 147 pb.pbq_baridx = bar; 148 149 if (ioctl(sc->pptfd, PPT_BAR_QUERY, &pb) != 0) { 150 return (-1); 151 } 152 153 switch (pb.pbq_type) { 154 case PCI_ADDR_IO: 155 *type = PCIBAR_IO; 156 break; 157 case PCI_ADDR_MEM32: 158 *type = PCIBAR_MEM32; 159 break; 160 case PCI_ADDR_MEM64: 161 *type = PCIBAR_MEM64; 162 break; 163 default: 164 err(1, "unrecognized BAR type: %u\n", pb.pbq_type); 165 break; 166 } 167 168 *base = pb.pbq_base; 169 *size = pb.pbq_size; 170 return (0); 171 } 172 173 static int 174 passthru_dev_open(const char *path, int *pptfdp) 175 { 176 int pptfd; 177 178 if ((pptfd = open(path, O_RDWR)) < 0) { 179 return (errno); 180 } 181 182 /* XXX: verify fd with ioctl? */ 183 *pptfdp = pptfd; 184 return (0); 185 } 186 187 #ifdef LEGACY_SUPPORT 188 static int 189 passthru_add_msicap(struct pci_devinst *pi, int msgnum, int nextptr) 190 { 191 int capoff, i; 192 struct msicap msicap; 193 u_char *capdata; 194 195 pci_populate_msicap(&msicap, msgnum, nextptr); 196 197 /* 198 * XXX 199 * Copy the msi capability structure in the last 16 bytes of the 200 * config space. This is wrong because it could shadow something 201 * useful to the device. 202 */ 203 capoff = 256 - roundup(sizeof(msicap), 4); 204 capdata = (u_char *)&msicap; 205 for (i = 0; i < sizeof(msicap); i++) 206 pci_set_cfgdata8(pi, capoff + i, capdata[i]); 207 208 return (capoff); 209 } 210 #endif /* LEGACY_SUPPORT */ 211 212 static void 213 passthru_intr_limit(struct passthru_softc *sc, struct msixcap *msixcap) 214 { 215 struct pci_devinst *pi = sc->psc_pi; 216 int off; 217 218 /* Reduce the number of MSI vectors if higher than OS limit */ 219 if ((off = sc->psc_msi.capoff) != 0 && sc->msi_limit != -1) { 220 int msi_limit, mmc; 221 222 msi_limit = 223 sc->msi_limit > 16 ? PCIM_MSICTRL_MMC_32 : 224 sc->msi_limit > 8 ? PCIM_MSICTRL_MMC_16 : 225 sc->msi_limit > 4 ? PCIM_MSICTRL_MMC_8 : 226 sc->msi_limit > 2 ? PCIM_MSICTRL_MMC_4 : 227 sc->msi_limit > 1 ? PCIM_MSICTRL_MMC_2 : 228 PCIM_MSICTRL_MMC_1; 229 mmc = sc->psc_msi.msgctrl & PCIM_MSICTRL_MMC_MASK; 230 231 if (mmc > msi_limit) { 232 sc->psc_msi.msgctrl &= ~PCIM_MSICTRL_MMC_MASK; 233 sc->psc_msi.msgctrl |= msi_limit; 234 pci_set_cfgdata16(pi, off + 2, sc->psc_msi.msgctrl); 235 } 236 } 237 238 /* Reduce the number of MSI-X vectors if higher than OS limit */ 239 if ((off = sc->psc_msix.capoff) != 0 && sc->msix_limit != -1) { 240 if (MSIX_TABLE_COUNT(msixcap->msgctrl) > sc->msix_limit) { 241 msixcap->msgctrl &= ~PCIM_MSIXCTRL_TABLE_SIZE; 242 msixcap->msgctrl |= sc->msix_limit - 1; 243 pci_set_cfgdata16(pi, off + 2, msixcap->msgctrl); 244 } 245 } 246 } 247 248 static int 249 cfginitmsi(struct passthru_softc *sc) 250 { 251 int i, ptr, capptr, cap, sts, caplen, table_size; 252 uint32_t u32; 253 struct pci_devinst *pi = sc->psc_pi; 254 struct msixcap msixcap; 255 uint32_t *msixcap_ptr; 256 257 /* 258 * Parse the capabilities and cache the location of the MSI 259 * and MSI-X capabilities. 260 */ 261 sts = read_config(sc, PCIR_STATUS, 2); 262 if (sts & PCIM_STATUS_CAPPRESENT) { 263 ptr = read_config(sc, PCIR_CAP_PTR, 1); 264 while (ptr != 0 && ptr != 0xff) { 265 cap = read_config(sc, ptr + PCICAP_ID, 1); 266 if (cap == PCIY_MSI) { 267 /* 268 * Copy the MSI capability into the config 269 * space of the emulated pci device 270 */ 271 sc->psc_msi.capoff = ptr; 272 sc->psc_msi.msgctrl = read_config(sc, 273 ptr + 2, 2); 274 sc->psc_msi.emulated = 0; 275 caplen = msi_caplen(sc->psc_msi.msgctrl); 276 capptr = ptr; 277 while (caplen > 0) { 278 u32 = read_config(sc, capptr, 4); 279 pci_set_cfgdata32(pi, capptr, u32); 280 caplen -= 4; 281 capptr += 4; 282 } 283 } else if (cap == PCIY_MSIX) { 284 /* 285 * Copy the MSI-X capability 286 */ 287 sc->psc_msix.capoff = ptr; 288 caplen = 12; 289 msixcap_ptr = (uint32_t*) &msixcap; 290 capptr = ptr; 291 while (caplen > 0) { 292 u32 = read_config(sc, capptr, 4); 293 *msixcap_ptr = u32; 294 pci_set_cfgdata32(pi, capptr, u32); 295 caplen -= 4; 296 capptr += 4; 297 msixcap_ptr++; 298 } 299 } 300 ptr = read_config(sc, ptr + PCICAP_NEXTPTR, 1); 301 } 302 } 303 304 passthru_intr_limit(sc, &msixcap); 305 306 if (sc->psc_msix.capoff != 0) { 307 pi->pi_msix.pba_bar = 308 msixcap.pba_info & PCIM_MSIX_BIR_MASK; 309 pi->pi_msix.pba_offset = 310 msixcap.pba_info & ~PCIM_MSIX_BIR_MASK; 311 pi->pi_msix.table_bar = 312 msixcap.table_info & PCIM_MSIX_BIR_MASK; 313 pi->pi_msix.table_offset = 314 msixcap.table_info & ~PCIM_MSIX_BIR_MASK; 315 pi->pi_msix.table_count = MSIX_TABLE_COUNT(msixcap.msgctrl); 316 pi->pi_msix.pba_size = PBA_SIZE(pi->pi_msix.table_count); 317 318 /* Allocate the emulated MSI-X table array */ 319 table_size = pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE; 320 pi->pi_msix.table = calloc(1, table_size); 321 322 /* Mask all table entries */ 323 for (i = 0; i < pi->pi_msix.table_count; i++) { 324 pi->pi_msix.table[i].vector_control |= 325 PCIM_MSIX_VCTRL_MASK; 326 } 327 } 328 329 #ifdef LEGACY_SUPPORT 330 /* 331 * If the passthrough device does not support MSI then craft a 332 * MSI capability for it. We link the new MSI capability at the 333 * head of the list of capabilities. 334 */ 335 if ((sts & PCIM_STATUS_CAPPRESENT) != 0 && sc->psc_msi.capoff == 0) { 336 int origptr, msiptr; 337 origptr = read_config(sc, PCIR_CAP_PTR, 1); 338 msiptr = passthru_add_msicap(pi, 1, origptr); 339 sc->psc_msi.capoff = msiptr; 340 sc->psc_msi.msgctrl = pci_get_cfgdata16(pi, msiptr + 2); 341 sc->psc_msi.emulated = 1; 342 pci_set_cfgdata8(pi, PCIR_CAP_PTR, msiptr); 343 } 344 #endif 345 346 /* Make sure one of the capabilities is present */ 347 if (sc->psc_msi.capoff == 0 && sc->psc_msix.capoff == 0) { 348 return (-1); 349 } else { 350 return (0); 351 } 352 } 353 354 static uint64_t 355 passthru_msix_table_read(struct passthru_softc *sc, uint64_t offset, int size) 356 { 357 struct pci_devinst *pi; 358 struct msix_table_entry *entry; 359 uint8_t *src8; 360 uint16_t *src16; 361 uint32_t *src32; 362 uint64_t *src64; 363 uint64_t data; 364 size_t entry_offset; 365 int index; 366 367 pi = sc->psc_pi; 368 if (offset >= pi->pi_msix.pba_offset && 369 offset < pi->pi_msix.pba_offset + pi->pi_msix.pba_size) { 370 switch(size) { 371 case 1: 372 src8 = (uint8_t *)(pi->pi_msix.pba_page + offset - 373 pi->pi_msix.pba_page_offset); 374 data = *src8; 375 break; 376 case 2: 377 src16 = (uint16_t *)(pi->pi_msix.pba_page + offset - 378 pi->pi_msix.pba_page_offset); 379 data = *src16; 380 break; 381 case 4: 382 src32 = (uint32_t *)(pi->pi_msix.pba_page + offset - 383 pi->pi_msix.pba_page_offset); 384 data = *src32; 385 break; 386 case 8: 387 src64 = (uint64_t *)(pi->pi_msix.pba_page + offset - 388 pi->pi_msix.pba_page_offset); 389 data = *src64; 390 break; 391 default: 392 return (-1); 393 } 394 return (data); 395 } 396 397 if (offset < pi->pi_msix.table_offset) 398 return (-1); 399 400 offset -= pi->pi_msix.table_offset; 401 index = offset / MSIX_TABLE_ENTRY_SIZE; 402 if (index >= pi->pi_msix.table_count) 403 return (-1); 404 405 entry = &pi->pi_msix.table[index]; 406 entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; 407 408 switch(size) { 409 case 1: 410 src8 = (uint8_t *)((void *)entry + entry_offset); 411 data = *src8; 412 break; 413 case 2: 414 src16 = (uint16_t *)((void *)entry + entry_offset); 415 data = *src16; 416 break; 417 case 4: 418 src32 = (uint32_t *)((void *)entry + entry_offset); 419 data = *src32; 420 break; 421 case 8: 422 src64 = (uint64_t *)((void *)entry + entry_offset); 423 data = *src64; 424 break; 425 default: 426 return (-1); 427 } 428 429 return (data); 430 } 431 432 static void 433 passthru_msix_table_write(struct vmctx *ctx, int vcpu, 434 struct passthru_softc *sc, uint64_t offset, int size, uint64_t data) 435 { 436 struct pci_devinst *pi; 437 struct msix_table_entry *entry; 438 uint8_t *dest8; 439 uint16_t *dest16; 440 uint32_t *dest32; 441 uint64_t *dest64; 442 size_t entry_offset; 443 uint32_t vector_control; 444 int index; 445 446 pi = sc->psc_pi; 447 if (offset >= pi->pi_msix.pba_offset && 448 offset < pi->pi_msix.pba_offset + pi->pi_msix.pba_size) { 449 switch(size) { 450 case 1: 451 dest8 = (uint8_t *)(pi->pi_msix.pba_page + offset - 452 pi->pi_msix.pba_page_offset); 453 *dest8 = data; 454 break; 455 case 2: 456 dest16 = (uint16_t *)(pi->pi_msix.pba_page + offset - 457 pi->pi_msix.pba_page_offset); 458 *dest16 = data; 459 break; 460 case 4: 461 dest32 = (uint32_t *)(pi->pi_msix.pba_page + offset - 462 pi->pi_msix.pba_page_offset); 463 *dest32 = data; 464 break; 465 case 8: 466 dest64 = (uint64_t *)(pi->pi_msix.pba_page + offset - 467 pi->pi_msix.pba_page_offset); 468 *dest64 = data; 469 break; 470 default: 471 break; 472 } 473 return; 474 } 475 476 if (offset < pi->pi_msix.table_offset) 477 return; 478 479 offset -= pi->pi_msix.table_offset; 480 index = offset / MSIX_TABLE_ENTRY_SIZE; 481 if (index >= pi->pi_msix.table_count) 482 return; 483 484 entry = &pi->pi_msix.table[index]; 485 entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; 486 487 /* Only 4 byte naturally-aligned writes are supported */ 488 assert(size == 4); 489 assert(entry_offset % 4 == 0); 490 491 vector_control = entry->vector_control; 492 dest32 = (uint32_t *)((void *)entry + entry_offset); 493 *dest32 = data; 494 /* If MSI-X hasn't been enabled, do nothing */ 495 if (pi->pi_msix.enabled) { 496 /* If the entry is masked, don't set it up */ 497 if ((entry->vector_control & PCIM_MSIX_VCTRL_MASK) == 0 || 498 (vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { 499 (void) vm_setup_pptdev_msix(ctx, vcpu, sc->pptfd, 500 index, entry->addr, entry->msg_data, 501 entry->vector_control); 502 } 503 } 504 } 505 506 static int 507 init_msix_table(struct vmctx *ctx, struct passthru_softc *sc, uint64_t base) 508 { 509 int idx; 510 size_t remaining __unused; 511 uint32_t table_size, table_offset; 512 uint32_t pba_size, pba_offset; 513 vm_paddr_t start __unused; 514 struct pci_devinst *pi = sc->psc_pi; 515 516 assert(pci_msix_table_bar(pi) >= 0 && pci_msix_pba_bar(pi) >= 0); 517 518 /* 519 * If the MSI-X table BAR maps memory intended for 520 * other uses, it is at least assured that the table 521 * either resides in its own page within the region, 522 * or it resides in a page shared with only the PBA. 523 */ 524 table_offset = rounddown2(pi->pi_msix.table_offset, 4096); 525 526 table_size = pi->pi_msix.table_offset - table_offset; 527 table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE; 528 table_size = roundup2(table_size, 4096); 529 530 idx = pi->pi_msix.table_bar; 531 start = pi->pi_bar[idx].addr; 532 remaining = pi->pi_bar[idx].size; 533 534 if (pi->pi_msix.pba_bar == pi->pi_msix.table_bar) { 535 pba_offset = pi->pi_msix.pba_offset; 536 pba_size = pi->pi_msix.pba_size; 537 if (pba_offset >= table_offset + table_size || 538 table_offset >= pba_offset + pba_size) { 539 /* 540 * If the PBA does not share a page with the MSI-x 541 * tables, no PBA emulation is required. 542 */ 543 pi->pi_msix.pba_page = NULL; 544 pi->pi_msix.pba_page_offset = 0; 545 } else { 546 /* 547 * The PBA overlaps with either the first or last 548 * page of the MSI-X table region. Map the 549 * appropriate page. 550 */ 551 if (pba_offset <= table_offset) 552 pi->pi_msix.pba_page_offset = table_offset; 553 else 554 pi->pi_msix.pba_page_offset = table_offset + 555 table_size - 4096; 556 pi->pi_msix.pba_page = mmap(NULL, 4096, PROT_READ | 557 PROT_WRITE, MAP_SHARED, sc->pptfd, 558 pi->pi_msix.pba_page_offset); 559 if (pi->pi_msix.pba_page == MAP_FAILED) { 560 warn("Failed to map PBA page for MSI-X on %d", 561 sc->pptfd); 562 return (-1); 563 } 564 } 565 } 566 567 return (0); 568 } 569 570 static int 571 cfginitbar(struct vmctx *ctx, struct passthru_softc *sc) 572 { 573 struct pci_devinst *pi = sc->psc_pi; 574 uint_t i; 575 576 /* 577 * Initialize BAR registers 578 */ 579 for (i = 0; i <= PCI_BARMAX; i++) { 580 enum pcibar_type bartype; 581 uint64_t base, size; 582 int error; 583 584 if (passthru_get_bar(sc, i, &bartype, &base, &size) != 0) { 585 continue; 586 } 587 588 if (bartype != PCIBAR_IO) { 589 if (((base | size) & PAGE_MASK) != 0) { 590 warnx("passthru device %d BAR %d: " 591 "base %#lx or size %#lx not page aligned\n", 592 sc->pptfd, i, base, size); 593 return (-1); 594 } 595 } 596 597 /* Cache information about the "real" BAR */ 598 sc->psc_bar[i].type = bartype; 599 sc->psc_bar[i].size = size; 600 sc->psc_bar[i].addr = base; 601 602 /* Allocate the BAR in the guest I/O or MMIO space */ 603 error = pci_emul_alloc_bar(pi, i, bartype, size); 604 if (error) 605 return (-1); 606 607 /* The MSI-X table needs special handling */ 608 if (i == pci_msix_table_bar(pi)) { 609 error = init_msix_table(ctx, sc, base); 610 if (error) 611 return (-1); 612 } 613 614 /* 615 * 64-bit BAR takes up two slots so skip the next one. 616 */ 617 if (bartype == PCIBAR_MEM64) { 618 i++; 619 assert(i <= PCI_BARMAX); 620 sc->psc_bar[i].type = PCIBAR_MEMHI64; 621 } 622 } 623 return (0); 624 } 625 626 static int 627 cfginit(struct vmctx *ctx, struct passthru_softc *sc) 628 { 629 struct pci_devinst *pi = sc->psc_pi; 630 631 if (cfginitmsi(sc) != 0) { 632 warnx("failed to initialize MSI for PCI %d", sc->pptfd); 633 return (-1); 634 } 635 636 if (cfginitbar(ctx, sc) != 0) { 637 warnx("failed to initialize BARs for PCI %d", sc->pptfd); 638 return (-1); 639 } 640 641 pci_set_cfgdata16(pi, PCIR_COMMAND, read_config(sc, PCIR_COMMAND, 2)); 642 643 return (0); 644 } 645 646 static int 647 passthru_legacy_config(nvlist_t *nvl, const char *opts) 648 { 649 if (opts == NULL) 650 return (0); 651 652 if (strncmp(opts, "/dev/ppt", 8) == 0) 653 set_config_value_node(nvl, "path", opts); 654 655 return (0); 656 } 657 658 static int 659 passthru_init(struct vmctx *ctx, struct pci_devinst *pi, nvlist_t *nvl) 660 { 661 int error, memflags, pptfd; 662 struct passthru_softc *sc; 663 const char *path; 664 665 pptfd = -1; 666 sc = NULL; 667 error = 1; 668 669 memflags = vm_get_memflags(ctx); 670 if (!(memflags & VM_MEM_F_WIRED)) { 671 warnx("passthru requires guest memory to be wired"); 672 goto done; 673 } 674 675 path = get_config_value_node(nvl, "path"); 676 if (path == NULL || passthru_dev_open(path, &pptfd) != 0) { 677 warnx("invalid passthru options"); 678 goto done; 679 } 680 681 if (vm_assign_pptdev(ctx, pptfd) != 0) { 682 warnx("PCI device at %d is not using the ppt driver", pptfd); 683 goto done; 684 } 685 686 sc = calloc(1, sizeof(struct passthru_softc)); 687 688 pi->pi_arg = sc; 689 sc->psc_pi = pi; 690 sc->pptfd = pptfd; 691 692 if ((error = vm_get_pptdev_limits(ctx, pptfd, &sc->msi_limit, 693 &sc->msix_limit)) != 0) 694 goto done; 695 696 /* initialize config space */ 697 if ((error = cfginit(ctx, sc)) != 0) 698 goto done; 699 700 error = 0; /* success */ 701 done: 702 if (error) { 703 free(sc); 704 if (pptfd != -1) 705 vm_unassign_pptdev(ctx, pptfd); 706 } 707 return (error); 708 } 709 710 static int 711 bar_access(int coff) 712 { 713 if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) 714 return (1); 715 else 716 return (0); 717 } 718 719 static int 720 msicap_access(struct passthru_softc *sc, int coff) 721 { 722 int caplen; 723 724 if (sc->psc_msi.capoff == 0) 725 return (0); 726 727 caplen = msi_caplen(sc->psc_msi.msgctrl); 728 729 if (coff >= sc->psc_msi.capoff && coff < sc->psc_msi.capoff + caplen) 730 return (1); 731 else 732 return (0); 733 } 734 735 static int 736 msixcap_access(struct passthru_softc *sc, int coff) 737 { 738 if (sc->psc_msix.capoff == 0) 739 return (0); 740 741 return (coff >= sc->psc_msix.capoff && 742 coff < sc->psc_msix.capoff + MSIX_CAPLEN); 743 } 744 745 static int 746 passthru_cfgread(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, 747 int coff, int bytes, uint32_t *rv) 748 { 749 struct passthru_softc *sc; 750 751 sc = pi->pi_arg; 752 753 /* 754 * PCI BARs and MSI capability is emulated. 755 */ 756 if (bar_access(coff) || msicap_access(sc, coff)) 757 return (-1); 758 759 /* 760 * MSI-X is also emulated since a limit on interrupts may be imposed by 761 * the OS, altering the perceived register state. 762 */ 763 if (msixcap_access(sc, coff)) 764 return (-1); 765 766 #ifdef LEGACY_SUPPORT 767 /* 768 * Emulate PCIR_CAP_PTR if this device does not support MSI capability 769 * natively. 770 */ 771 if (sc->psc_msi.emulated) { 772 if (coff >= PCIR_CAP_PTR && coff < PCIR_CAP_PTR + 4) 773 return (-1); 774 } 775 #endif 776 777 /* 778 * Emulate the command register. If a single read reads both the 779 * command and status registers, read the status register from the 780 * device's config space. 781 */ 782 if (coff == PCIR_COMMAND) { 783 if (bytes <= 2) 784 return (-1); 785 *rv = pci_get_cfgdata16(pi, PCIR_COMMAND) << 16 | 786 read_config(sc, PCIR_STATUS, 2); 787 return (0); 788 } 789 790 /* Everything else just read from the device's config space */ 791 *rv = read_config(sc, coff, bytes); 792 793 return (0); 794 } 795 796 static int 797 passthru_cfgwrite(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, 798 int coff, int bytes, uint32_t val) 799 { 800 int error, msix_table_entries, i; 801 struct passthru_softc *sc; 802 uint16_t cmd_old; 803 804 sc = pi->pi_arg; 805 806 /* 807 * PCI BARs are emulated 808 */ 809 if (bar_access(coff)) 810 return (-1); 811 812 /* 813 * MSI capability is emulated 814 */ 815 if (msicap_access(sc, coff)) { 816 pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msi.capoff, 817 PCIY_MSI); 818 error = vm_setup_pptdev_msi(ctx, vcpu, sc->pptfd, 819 pi->pi_msi.addr, pi->pi_msi.msg_data, pi->pi_msi.maxmsgnum); 820 if (error != 0) 821 err(1, "vm_setup_pptdev_msi"); 822 return (0); 823 } 824 825 if (msixcap_access(sc, coff)) { 826 pci_emul_capwrite(pi, coff, bytes, val, sc->psc_msix.capoff, 827 PCIY_MSIX); 828 if (pi->pi_msix.enabled) { 829 msix_table_entries = pi->pi_msix.table_count; 830 for (i = 0; i < msix_table_entries; i++) { 831 error = vm_setup_pptdev_msix(ctx, vcpu, 832 sc->pptfd, i, 833 pi->pi_msix.table[i].addr, 834 pi->pi_msix.table[i].msg_data, 835 pi->pi_msix.table[i].vector_control); 836 837 if (error) 838 err(1, "vm_setup_pptdev_msix"); 839 } 840 } else { 841 error = vm_disable_pptdev_msix(ctx, sc->pptfd); 842 if (error) 843 err(1, "vm_disable_pptdev_msix"); 844 } 845 return (0); 846 } 847 848 #ifdef LEGACY_SUPPORT 849 /* 850 * If this device does not support MSI natively then we cannot let 851 * the guest disable legacy interrupts from the device. It is the 852 * legacy interrupt that is triggering the virtual MSI to the guest. 853 */ 854 if (sc->psc_msi.emulated && pci_msi_enabled(pi)) { 855 if (coff == PCIR_COMMAND && bytes == 2) 856 val &= ~PCIM_CMD_INTxDIS; 857 } 858 #endif 859 860 write_config(sc, coff, bytes, val); 861 if (coff == PCIR_COMMAND) { 862 cmd_old = pci_get_cfgdata16(pi, PCIR_COMMAND); 863 if (bytes == 1) 864 pci_set_cfgdata8(pi, PCIR_COMMAND, val); 865 else if (bytes == 2) 866 pci_set_cfgdata16(pi, PCIR_COMMAND, val); 867 pci_emul_cmd_changed(pi, cmd_old); 868 } 869 870 return (0); 871 } 872 873 static void 874 passthru_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, 875 uint64_t offset, int size, uint64_t value) 876 { 877 struct passthru_softc *sc = pi->pi_arg; 878 879 if (baridx == pci_msix_table_bar(pi)) { 880 passthru_msix_table_write(ctx, vcpu, sc, offset, size, value); 881 } else { 882 struct ppt_bar_io pbi; 883 884 assert(pi->pi_bar[baridx].type == PCIBAR_IO); 885 886 pbi.pbi_bar = baridx; 887 pbi.pbi_width = size; 888 pbi.pbi_off = offset; 889 pbi.pbi_data = value; 890 (void) ioctl(sc->pptfd, PPT_BAR_WRITE, &pbi); 891 } 892 } 893 894 static uint64_t 895 passthru_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, 896 uint64_t offset, int size) 897 { 898 struct passthru_softc *sc = pi->pi_arg; 899 uint64_t val; 900 901 if (baridx == pci_msix_table_bar(pi)) { 902 val = passthru_msix_table_read(sc, offset, size); 903 } else { 904 struct ppt_bar_io pbi; 905 906 assert(pi->pi_bar[baridx].type == PCIBAR_IO); 907 908 pbi.pbi_bar = baridx; 909 pbi.pbi_width = size; 910 pbi.pbi_off = offset; 911 if (ioctl(sc->pptfd, PPT_BAR_READ, &pbi) == 0) { 912 val = pbi.pbi_data; 913 } else { 914 val = 0; 915 } 916 } 917 918 return (val); 919 } 920 921 static void 922 passthru_msix_addr(struct vmctx *ctx, struct pci_devinst *pi, int baridx, 923 int enabled, uint64_t address) 924 { 925 struct passthru_softc *sc; 926 size_t remaining; 927 uint32_t table_size, table_offset; 928 929 sc = pi->pi_arg; 930 table_offset = rounddown2(pi->pi_msix.table_offset, 4096); 931 if (table_offset > 0) { 932 if (!enabled) { 933 if (vm_unmap_pptdev_mmio(ctx, sc->pptfd, address, 934 table_offset) != 0) 935 warnx("pci_passthru: unmap_pptdev_mmio failed"); 936 } else { 937 if (vm_map_pptdev_mmio(ctx, sc->pptfd, address, 938 table_offset, sc->psc_bar[baridx].addr) != 0) 939 warnx("pci_passthru: map_pptdev_mmio failed"); 940 } 941 } 942 table_size = pi->pi_msix.table_offset - table_offset; 943 table_size += pi->pi_msix.table_count * MSIX_TABLE_ENTRY_SIZE; 944 table_size = roundup2(table_size, 4096); 945 remaining = pi->pi_bar[baridx].size - table_offset - table_size; 946 if (remaining > 0) { 947 address += table_offset + table_size; 948 if (!enabled) { 949 if (vm_unmap_pptdev_mmio(ctx, sc->pptfd, address, 950 remaining) != 0) 951 warnx("pci_passthru: unmap_pptdev_mmio failed"); 952 } else { 953 if (vm_map_pptdev_mmio(ctx, sc->pptfd, address, 954 remaining, sc->psc_bar[baridx].addr + 955 table_offset + table_size) != 0) 956 warnx("pci_passthru: map_pptdev_mmio failed"); 957 } 958 } 959 } 960 961 static void 962 passthru_mmio_addr(struct vmctx *ctx, struct pci_devinst *pi, int baridx, 963 int enabled, uint64_t address) 964 { 965 struct passthru_softc *sc; 966 967 sc = pi->pi_arg; 968 if (!enabled) { 969 if (vm_unmap_pptdev_mmio(ctx, sc->pptfd, address, 970 sc->psc_bar[baridx].size) != 0) 971 warnx("pci_passthru: unmap_pptdev_mmio failed"); 972 } else { 973 if (vm_map_pptdev_mmio(ctx, sc->pptfd, address, 974 sc->psc_bar[baridx].size, sc->psc_bar[baridx].addr) != 0) 975 warnx("pci_passthru: map_pptdev_mmio failed"); 976 } 977 } 978 979 static void 980 passthru_addr(struct vmctx *ctx, struct pci_devinst *pi, int baridx, 981 int enabled, uint64_t address) 982 { 983 984 if (pi->pi_bar[baridx].type == PCIBAR_IO) 985 return; 986 if (baridx == pci_msix_table_bar(pi)) 987 passthru_msix_addr(ctx, pi, baridx, enabled, address); 988 else 989 passthru_mmio_addr(ctx, pi, baridx, enabled, address); 990 } 991 992 struct pci_devemu passthru = { 993 .pe_emu = "passthru", 994 .pe_init = passthru_init, 995 .pe_legacy_config = passthru_legacy_config, 996 .pe_cfgwrite = passthru_cfgwrite, 997 .pe_cfgread = passthru_cfgread, 998 .pe_barwrite = passthru_write, 999 .pe_barread = passthru_read, 1000 .pe_baraddr = passthru_addr, 1001 }; 1002 PCI_EMUL_SET(passthru); 1003