1 /* 2 * Copyright (c) 2007, Neocleus Corporation. 3 * Copyright (c) 2007, Intel Corporation. 4 * 5 * This work is licensed under the terms of the GNU GPL, version 2. See 6 * the COPYING file in the top-level directory. 7 * 8 * Alex Novik <alex@neocleus.com> 9 * Allen Kay <allen.m.kay@intel.com> 10 * Guy Zana <guy@neocleus.com> 11 * 12 * This file implements direct PCI assignment to a HVM guest 13 */ 14 15 /* 16 * Interrupt Disable policy: 17 * 18 * INTx interrupt: 19 * Initialize(register_real_device) 20 * Map INTx(xc_physdev_map_pirq): 21 * <fail> 22 * - Set real Interrupt Disable bit to '1'. 23 * - Set machine_irq and assigned_device->machine_irq to '0'. 24 * * Don't bind INTx. 25 * 26 * Bind INTx(xc_domain_bind_pt_pci_irq): 27 * <fail> 28 * - Set real Interrupt Disable bit to '1'. 29 * - Unmap INTx. 30 * - Decrement xen_pt_mapped_machine_irq[machine_irq] 31 * - Set assigned_device->machine_irq to '0'. 32 * 33 * Write to Interrupt Disable bit by guest software(xen_pt_cmd_reg_write) 34 * Write '0' 35 * - Set real bit to '0' if assigned_device->machine_irq isn't '0'. 36 * 37 * Write '1' 38 * - Set real bit to '1'. 39 * 40 * MSI interrupt: 41 * Initialize MSI register(xen_pt_msi_setup, xen_pt_msi_update) 42 * Bind MSI(xc_domain_update_msi_irq) 43 * <fail> 44 * - Unmap MSI. 45 * - Set dev->msi->pirq to '-1'. 46 * 47 * MSI-X interrupt: 48 * Initialize MSI-X register(xen_pt_msix_update_one) 49 * Bind MSI-X(xc_domain_update_msi_irq) 50 * <fail> 51 * - Unmap MSI-X. 52 * - Set entry->pirq to '-1'. 53 */ 54 55 #include <sys/ioctl.h> 56 57 #include "hw/pci/pci.h" 58 #include "hw/xen/xen.h" 59 #include "hw/xen/xen_backend.h" 60 #include "xen_pt.h" 61 #include "qemu/range.h" 62 #include "exec/address-spaces.h" 63 64 #define XEN_PT_NR_IRQS (256) 65 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0}; 66 67 void xen_pt_log(const PCIDevice *d, const char *f, ...) 68 { 69 va_list ap; 70 71 va_start(ap, f); 72 if (d) { 73 fprintf(stderr, "[%02x:%02x.%d] ", pci_bus_num(d->bus), 74 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn)); 75 } 76 vfprintf(stderr, f, ap); 77 va_end(ap); 78 } 79 80 /* Config Space */ 81 82 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len) 83 { 84 /* check offset range */ 85 if (addr >= 0xFF) { 86 XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. " 87 "(addr: 0x%02x, len: %d)\n", addr, len); 88 return -1; 89 } 90 91 /* check read size */ 92 if ((len != 1) && (len != 2) && (len != 4)) { 93 XEN_PT_ERR(d, "Failed to access register with invalid access length. " 94 "(addr: 0x%02x, len: %d)\n", addr, len); 95 return -1; 96 } 97 98 /* check offset alignment */ 99 if (addr & (len - 1)) { 100 XEN_PT_ERR(d, "Failed to access register with invalid access size " 101 "alignment. (addr: 0x%02x, len: %d)\n", addr, len); 102 return -1; 103 } 104 105 return 0; 106 } 107 108 int xen_pt_bar_offset_to_index(uint32_t offset) 109 { 110 int index = 0; 111 112 /* check Exp ROM BAR */ 113 if (offset == PCI_ROM_ADDRESS) { 114 return PCI_ROM_SLOT; 115 } 116 117 /* calculate BAR index */ 118 index = (offset - PCI_BASE_ADDRESS_0) >> 2; 119 if (index >= PCI_NUM_REGIONS) { 120 return -1; 121 } 122 123 return index; 124 } 125 126 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len) 127 { 128 XenPCIPassthroughState *s = DO_UPCAST(XenPCIPassthroughState, dev, d); 129 uint32_t val = 0; 130 XenPTRegGroup *reg_grp_entry = NULL; 131 XenPTReg *reg_entry = NULL; 132 int rc = 0; 133 int emul_len = 0; 134 uint32_t find_addr = addr; 135 136 if (xen_pt_pci_config_access_check(d, addr, len)) { 137 goto exit; 138 } 139 140 /* find register group entry */ 141 reg_grp_entry = xen_pt_find_reg_grp(s, addr); 142 if (reg_grp_entry) { 143 /* check 0-Hardwired register group */ 144 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) { 145 /* no need to emulate, just return 0 */ 146 val = 0; 147 goto exit; 148 } 149 } 150 151 /* read I/O device register value */ 152 rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len); 153 if (rc < 0) { 154 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc); 155 memset(&val, 0xff, len); 156 } 157 158 /* just return the I/O device register value for 159 * passthrough type register group */ 160 if (reg_grp_entry == NULL) { 161 goto exit; 162 } 163 164 /* adjust the read value to appropriate CFC-CFF window */ 165 val <<= (addr & 3) << 3; 166 emul_len = len; 167 168 /* loop around the guest requested size */ 169 while (emul_len > 0) { 170 /* find register entry to be emulated */ 171 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr); 172 if (reg_entry) { 173 XenPTRegInfo *reg = reg_entry->reg; 174 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset; 175 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3); 176 uint8_t *ptr_val = NULL; 177 178 valid_mask <<= (find_addr - real_offset) << 3; 179 ptr_val = (uint8_t *)&val + (real_offset & 3); 180 181 /* do emulation based on register size */ 182 switch (reg->size) { 183 case 1: 184 if (reg->u.b.read) { 185 rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask); 186 } 187 break; 188 case 2: 189 if (reg->u.w.read) { 190 rc = reg->u.w.read(s, reg_entry, 191 (uint16_t *)ptr_val, valid_mask); 192 } 193 break; 194 case 4: 195 if (reg->u.dw.read) { 196 rc = reg->u.dw.read(s, reg_entry, 197 (uint32_t *)ptr_val, valid_mask); 198 } 199 break; 200 } 201 202 if (rc < 0) { 203 xen_shutdown_fatal_error("Internal error: Invalid read " 204 "emulation. (%s, rc: %d)\n", 205 __func__, rc); 206 return 0; 207 } 208 209 /* calculate next address to find */ 210 emul_len -= reg->size; 211 if (emul_len > 0) { 212 find_addr = real_offset + reg->size; 213 } 214 } else { 215 /* nothing to do with passthrough type register, 216 * continue to find next byte */ 217 emul_len--; 218 find_addr++; 219 } 220 } 221 222 /* need to shift back before returning them to pci bus emulator */ 223 val >>= ((addr & 3) << 3); 224 225 exit: 226 XEN_PT_LOG_CONFIG(d, addr, val, len); 227 return val; 228 } 229 230 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr, 231 uint32_t val, int len) 232 { 233 XenPCIPassthroughState *s = DO_UPCAST(XenPCIPassthroughState, dev, d); 234 int index = 0; 235 XenPTRegGroup *reg_grp_entry = NULL; 236 int rc = 0; 237 uint32_t read_val = 0; 238 int emul_len = 0; 239 XenPTReg *reg_entry = NULL; 240 uint32_t find_addr = addr; 241 XenPTRegInfo *reg = NULL; 242 243 if (xen_pt_pci_config_access_check(d, addr, len)) { 244 return; 245 } 246 247 XEN_PT_LOG_CONFIG(d, addr, val, len); 248 249 /* check unused BAR register */ 250 index = xen_pt_bar_offset_to_index(addr); 251 if ((index >= 0) && (val > 0 && val < XEN_PT_BAR_ALLF) && 252 (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) { 253 XEN_PT_WARN(d, "Guest attempt to set address to unused Base Address " 254 "Register. (addr: 0x%02x, len: %d)\n", addr, len); 255 } 256 257 /* find register group entry */ 258 reg_grp_entry = xen_pt_find_reg_grp(s, addr); 259 if (reg_grp_entry) { 260 /* check 0-Hardwired register group */ 261 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) { 262 /* ignore silently */ 263 XEN_PT_WARN(d, "Access to 0-Hardwired register. " 264 "(addr: 0x%02x, len: %d)\n", addr, len); 265 return; 266 } 267 } 268 269 rc = xen_host_pci_get_block(&s->real_device, addr, 270 (uint8_t *)&read_val, len); 271 if (rc < 0) { 272 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc); 273 memset(&read_val, 0xff, len); 274 } 275 276 /* pass directly to the real device for passthrough type register group */ 277 if (reg_grp_entry == NULL) { 278 goto out; 279 } 280 281 memory_region_transaction_begin(); 282 pci_default_write_config(d, addr, val, len); 283 284 /* adjust the read and write value to appropriate CFC-CFF window */ 285 read_val <<= (addr & 3) << 3; 286 val <<= (addr & 3) << 3; 287 emul_len = len; 288 289 /* loop around the guest requested size */ 290 while (emul_len > 0) { 291 /* find register entry to be emulated */ 292 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr); 293 if (reg_entry) { 294 reg = reg_entry->reg; 295 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset; 296 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3); 297 uint8_t *ptr_val = NULL; 298 299 valid_mask <<= (find_addr - real_offset) << 3; 300 ptr_val = (uint8_t *)&val + (real_offset & 3); 301 302 /* do emulation based on register size */ 303 switch (reg->size) { 304 case 1: 305 if (reg->u.b.write) { 306 rc = reg->u.b.write(s, reg_entry, ptr_val, 307 read_val >> ((real_offset & 3) << 3), 308 valid_mask); 309 } 310 break; 311 case 2: 312 if (reg->u.w.write) { 313 rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val, 314 (read_val >> ((real_offset & 3) << 3)), 315 valid_mask); 316 } 317 break; 318 case 4: 319 if (reg->u.dw.write) { 320 rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val, 321 (read_val >> ((real_offset & 3) << 3)), 322 valid_mask); 323 } 324 break; 325 } 326 327 if (rc < 0) { 328 xen_shutdown_fatal_error("Internal error: Invalid write" 329 " emulation. (%s, rc: %d)\n", 330 __func__, rc); 331 return; 332 } 333 334 /* calculate next address to find */ 335 emul_len -= reg->size; 336 if (emul_len > 0) { 337 find_addr = real_offset + reg->size; 338 } 339 } else { 340 /* nothing to do with passthrough type register, 341 * continue to find next byte */ 342 emul_len--; 343 find_addr++; 344 } 345 } 346 347 /* need to shift back before passing them to xen_host_pci_device */ 348 val >>= (addr & 3) << 3; 349 350 memory_region_transaction_commit(); 351 352 out: 353 if (!(reg && reg->no_wb)) { 354 /* unknown regs are passed through */ 355 rc = xen_host_pci_set_block(&s->real_device, addr, 356 (uint8_t *)&val, len); 357 358 if (rc < 0) { 359 XEN_PT_ERR(d, "pci_write_block failed. return value: %d.\n", rc); 360 } 361 } 362 } 363 364 /* register regions */ 365 366 static uint64_t xen_pt_bar_read(void *o, hwaddr addr, 367 unsigned size) 368 { 369 PCIDevice *d = o; 370 /* if this function is called, that probably means that there is a 371 * misconfiguration of the IOMMU. */ 372 XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n", 373 addr); 374 return 0; 375 } 376 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val, 377 unsigned size) 378 { 379 PCIDevice *d = o; 380 /* Same comment as xen_pt_bar_read function */ 381 XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n", 382 addr); 383 } 384 385 static const MemoryRegionOps ops = { 386 .endianness = DEVICE_NATIVE_ENDIAN, 387 .read = xen_pt_bar_read, 388 .write = xen_pt_bar_write, 389 }; 390 391 static int xen_pt_register_regions(XenPCIPassthroughState *s) 392 { 393 int i = 0; 394 XenHostPCIDevice *d = &s->real_device; 395 396 /* Register PIO/MMIO BARs */ 397 for (i = 0; i < PCI_ROM_SLOT; i++) { 398 XenHostPCIIORegion *r = &d->io_regions[i]; 399 uint8_t type; 400 401 if (r->base_addr == 0 || r->size == 0) { 402 continue; 403 } 404 405 s->bases[i].access.u = r->base_addr; 406 407 if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) { 408 type = PCI_BASE_ADDRESS_SPACE_IO; 409 } else { 410 type = PCI_BASE_ADDRESS_SPACE_MEMORY; 411 if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) { 412 type |= PCI_BASE_ADDRESS_MEM_PREFETCH; 413 } 414 if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) { 415 type |= PCI_BASE_ADDRESS_MEM_TYPE_64; 416 } 417 } 418 419 memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev, 420 "xen-pci-pt-bar", r->size); 421 pci_register_bar(&s->dev, i, type, &s->bar[i]); 422 423 XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%lx"PRIx64 424 " base_addr=0x%lx"PRIx64" type: %#x)\n", 425 i, r->size, r->base_addr, type); 426 } 427 428 /* Register expansion ROM address */ 429 if (d->rom.base_addr && d->rom.size) { 430 uint32_t bar_data = 0; 431 432 /* Re-set BAR reported by OS, otherwise ROM can't be read. */ 433 if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) { 434 return 0; 435 } 436 if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) { 437 bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK; 438 xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data); 439 } 440 441 s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr; 442 443 memory_region_init_rom_device(&s->rom, OBJECT(s), NULL, NULL, 444 "xen-pci-pt-rom", d->rom.size); 445 pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH, 446 &s->rom); 447 448 XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64 449 " base_addr=0x%08"PRIx64")\n", 450 d->rom.size, d->rom.base_addr); 451 } 452 453 return 0; 454 } 455 456 static void xen_pt_unregister_regions(XenPCIPassthroughState *s) 457 { 458 XenHostPCIDevice *d = &s->real_device; 459 int i; 460 461 for (i = 0; i < PCI_NUM_REGIONS - 1; i++) { 462 XenHostPCIIORegion *r = &d->io_regions[i]; 463 464 if (r->base_addr == 0 || r->size == 0) { 465 continue; 466 } 467 468 memory_region_destroy(&s->bar[i]); 469 } 470 if (d->rom.base_addr && d->rom.size) { 471 memory_region_destroy(&s->rom); 472 } 473 } 474 475 /* region mapping */ 476 477 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr) 478 { 479 int i = 0; 480 481 for (i = 0; i < PCI_NUM_REGIONS - 1; i++) { 482 if (mr == &s->bar[i]) { 483 return i; 484 } 485 } 486 if (mr == &s->rom) { 487 return PCI_ROM_SLOT; 488 } 489 return -1; 490 } 491 492 /* 493 * This function checks if an io_region overlaps an io_region from another 494 * device. The io_region to check is provided with (addr, size and type) 495 * A callback can be provided and will be called for every region that is 496 * overlapped. 497 * The return value indicates if the region is overlappsed */ 498 struct CheckBarArgs { 499 XenPCIPassthroughState *s; 500 pcibus_t addr; 501 pcibus_t size; 502 uint8_t type; 503 bool rc; 504 }; 505 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque) 506 { 507 struct CheckBarArgs *arg = opaque; 508 XenPCIPassthroughState *s = arg->s; 509 uint8_t type = arg->type; 510 int i; 511 512 if (d->devfn == s->dev.devfn) { 513 return; 514 } 515 516 /* xxx: This ignores bridges. */ 517 for (i = 0; i < PCI_NUM_REGIONS; i++) { 518 const PCIIORegion *r = &d->io_regions[i]; 519 520 if (!r->size) { 521 continue; 522 } 523 if ((type & PCI_BASE_ADDRESS_SPACE_IO) 524 != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) { 525 continue; 526 } 527 528 if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) { 529 XEN_PT_WARN(&s->dev, 530 "Overlapped to device [%02x:%02x.%d] Region: %i" 531 " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n", 532 pci_bus_num(bus), PCI_SLOT(d->devfn), 533 PCI_FUNC(d->devfn), i, r->addr, r->size); 534 arg->rc = true; 535 } 536 } 537 } 538 539 static void xen_pt_region_update(XenPCIPassthroughState *s, 540 MemoryRegionSection *sec, bool adding) 541 { 542 PCIDevice *d = &s->dev; 543 MemoryRegion *mr = sec->mr; 544 int bar = -1; 545 int rc; 546 int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING; 547 struct CheckBarArgs args = { 548 .s = s, 549 .addr = sec->offset_within_address_space, 550 .size = int128_get64(sec->size), 551 .rc = false, 552 }; 553 554 bar = xen_pt_bar_from_region(s, mr); 555 if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) { 556 return; 557 } 558 559 if (s->msix && &s->msix->mmio == mr) { 560 if (adding) { 561 s->msix->mmio_base_addr = sec->offset_within_address_space; 562 rc = xen_pt_msix_update_remap(s, s->msix->bar_index); 563 } 564 return; 565 } 566 567 args.type = d->io_regions[bar].type; 568 pci_for_each_device(d->bus, pci_bus_num(d->bus), 569 xen_pt_check_bar_overlap, &args); 570 if (args.rc) { 571 XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS 572 ", len: %#"FMT_PCIBUS") is overlapped.\n", 573 bar, sec->offset_within_address_space, 574 int128_get64(sec->size)); 575 } 576 577 if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) { 578 uint32_t guest_port = sec->offset_within_address_space; 579 uint32_t machine_port = s->bases[bar].access.pio_base; 580 uint32_t size = int128_get64(sec->size); 581 rc = xc_domain_ioport_mapping(xen_xc, xen_domid, 582 guest_port, machine_port, size, 583 op); 584 if (rc) { 585 XEN_PT_ERR(d, "%s ioport mapping failed! (rc: %i)\n", 586 adding ? "create new" : "remove old", rc); 587 } 588 } else { 589 pcibus_t guest_addr = sec->offset_within_address_space; 590 pcibus_t machine_addr = s->bases[bar].access.maddr 591 + sec->offset_within_region; 592 pcibus_t size = int128_get64(sec->size); 593 rc = xc_domain_memory_mapping(xen_xc, xen_domid, 594 XEN_PFN(guest_addr + XC_PAGE_SIZE - 1), 595 XEN_PFN(machine_addr + XC_PAGE_SIZE - 1), 596 XEN_PFN(size + XC_PAGE_SIZE - 1), 597 op); 598 if (rc) { 599 XEN_PT_ERR(d, "%s mem mapping failed! (rc: %i)\n", 600 adding ? "create new" : "remove old", rc); 601 } 602 } 603 } 604 605 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec) 606 { 607 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState, 608 memory_listener); 609 610 memory_region_ref(sec->mr); 611 xen_pt_region_update(s, sec, true); 612 } 613 614 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec) 615 { 616 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState, 617 memory_listener); 618 619 xen_pt_region_update(s, sec, false); 620 memory_region_unref(sec->mr); 621 } 622 623 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec) 624 { 625 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState, 626 io_listener); 627 628 memory_region_ref(sec->mr); 629 xen_pt_region_update(s, sec, true); 630 } 631 632 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec) 633 { 634 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState, 635 io_listener); 636 637 xen_pt_region_update(s, sec, false); 638 memory_region_unref(sec->mr); 639 } 640 641 static const MemoryListener xen_pt_memory_listener = { 642 .region_add = xen_pt_region_add, 643 .region_del = xen_pt_region_del, 644 .priority = 10, 645 }; 646 647 static const MemoryListener xen_pt_io_listener = { 648 .region_add = xen_pt_io_region_add, 649 .region_del = xen_pt_io_region_del, 650 .priority = 10, 651 }; 652 653 /* init */ 654 655 static int xen_pt_initfn(PCIDevice *d) 656 { 657 XenPCIPassthroughState *s = DO_UPCAST(XenPCIPassthroughState, dev, d); 658 int rc = 0; 659 uint8_t machine_irq = 0; 660 int pirq = XEN_PT_UNASSIGNED_PIRQ; 661 662 /* register real device */ 663 XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d" 664 " to devfn %#x\n", 665 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function, 666 s->dev.devfn); 667 668 rc = xen_host_pci_device_get(&s->real_device, 669 s->hostaddr.domain, s->hostaddr.bus, 670 s->hostaddr.slot, s->hostaddr.function); 671 if (rc) { 672 XEN_PT_ERR(d, "Failed to \"open\" the real pci device. rc: %i\n", rc); 673 return -1; 674 } 675 676 s->is_virtfn = s->real_device.is_virtfn; 677 if (s->is_virtfn) { 678 XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n", 679 s->real_device.domain, s->real_device.bus, 680 s->real_device.dev, s->real_device.func); 681 } 682 683 /* Initialize virtualized PCI configuration (Extended 256 Bytes) */ 684 if (xen_host_pci_get_block(&s->real_device, 0, d->config, 685 PCI_CONFIG_SPACE_SIZE) == -1) { 686 xen_host_pci_device_put(&s->real_device); 687 return -1; 688 } 689 690 s->memory_listener = xen_pt_memory_listener; 691 s->io_listener = xen_pt_io_listener; 692 693 /* Handle real device's MMIO/PIO BARs */ 694 xen_pt_register_regions(s); 695 696 /* reinitialize each config register to be emulated */ 697 if (xen_pt_config_init(s)) { 698 XEN_PT_ERR(d, "PCI Config space initialisation failed.\n"); 699 xen_host_pci_device_put(&s->real_device); 700 return -1; 701 } 702 703 /* Bind interrupt */ 704 if (!s->dev.config[PCI_INTERRUPT_PIN]) { 705 XEN_PT_LOG(d, "no pin interrupt\n"); 706 goto out; 707 } 708 709 machine_irq = s->real_device.irq; 710 rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq); 711 712 if (rc < 0) { 713 XEN_PT_ERR(d, "Mapping machine irq %u to pirq %i failed, (rc: %d)\n", 714 machine_irq, pirq, rc); 715 716 /* Disable PCI intx assertion (turn on bit10 of devctl) */ 717 xen_host_pci_set_word(&s->real_device, 718 PCI_COMMAND, 719 pci_get_word(s->dev.config + PCI_COMMAND) 720 | PCI_COMMAND_INTX_DISABLE); 721 machine_irq = 0; 722 s->machine_irq = 0; 723 } else { 724 machine_irq = pirq; 725 s->machine_irq = pirq; 726 xen_pt_mapped_machine_irq[machine_irq]++; 727 } 728 729 /* bind machine_irq to device */ 730 if (machine_irq != 0) { 731 uint8_t e_intx = xen_pt_pci_intx(s); 732 733 rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq, 734 pci_bus_num(d->bus), 735 PCI_SLOT(d->devfn), 736 e_intx); 737 if (rc < 0) { 738 XEN_PT_ERR(d, "Binding of interrupt %i failed! (rc: %d)\n", 739 e_intx, rc); 740 741 /* Disable PCI intx assertion (turn on bit10 of devctl) */ 742 xen_host_pci_set_word(&s->real_device, PCI_COMMAND, 743 *(uint16_t *)(&s->dev.config[PCI_COMMAND]) 744 | PCI_COMMAND_INTX_DISABLE); 745 xen_pt_mapped_machine_irq[machine_irq]--; 746 747 if (xen_pt_mapped_machine_irq[machine_irq] == 0) { 748 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) { 749 XEN_PT_ERR(d, "Unmapping of machine interrupt %i failed!" 750 " (rc: %d)\n", machine_irq, rc); 751 } 752 } 753 s->machine_irq = 0; 754 } 755 } 756 757 out: 758 memory_listener_register(&s->memory_listener, &address_space_memory); 759 memory_listener_register(&s->io_listener, &address_space_io); 760 XEN_PT_LOG(d, 761 "Real physical device %02x:%02x.%d registered successfully!\n", 762 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function); 763 764 return 0; 765 } 766 767 static void xen_pt_unregister_device(PCIDevice *d) 768 { 769 XenPCIPassthroughState *s = DO_UPCAST(XenPCIPassthroughState, dev, d); 770 uint8_t machine_irq = s->machine_irq; 771 uint8_t intx = xen_pt_pci_intx(s); 772 int rc; 773 774 if (machine_irq) { 775 rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq, 776 PT_IRQ_TYPE_PCI, 777 pci_bus_num(d->bus), 778 PCI_SLOT(s->dev.devfn), 779 intx, 780 0 /* isa_irq */); 781 if (rc < 0) { 782 XEN_PT_ERR(d, "unbinding of interrupt INT%c failed." 783 " (machine irq: %i, rc: %d)" 784 " But bravely continuing on..\n", 785 'a' + intx, machine_irq, rc); 786 } 787 } 788 789 if (s->msi) { 790 xen_pt_msi_disable(s); 791 } 792 if (s->msix) { 793 xen_pt_msix_disable(s); 794 } 795 796 if (machine_irq) { 797 xen_pt_mapped_machine_irq[machine_irq]--; 798 799 if (xen_pt_mapped_machine_irq[machine_irq] == 0) { 800 rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq); 801 802 if (rc < 0) { 803 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (rc: %d)" 804 " But bravely continuing on..\n", 805 machine_irq, rc); 806 } 807 } 808 } 809 810 /* delete all emulated config registers */ 811 xen_pt_config_delete(s); 812 813 xen_pt_unregister_regions(s); 814 memory_listener_unregister(&s->memory_listener); 815 memory_listener_unregister(&s->io_listener); 816 817 xen_host_pci_device_put(&s->real_device); 818 } 819 820 static Property xen_pci_passthrough_properties[] = { 821 DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr), 822 DEFINE_PROP_END_OF_LIST(), 823 }; 824 825 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data) 826 { 827 DeviceClass *dc = DEVICE_CLASS(klass); 828 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); 829 830 k->init = xen_pt_initfn; 831 k->exit = xen_pt_unregister_device; 832 k->config_read = xen_pt_pci_read_config; 833 k->config_write = xen_pt_pci_write_config; 834 set_bit(DEVICE_CATEGORY_MISC, dc->categories); 835 dc->desc = "Assign an host PCI device with Xen"; 836 dc->props = xen_pci_passthrough_properties; 837 }; 838 839 static const TypeInfo xen_pci_passthrough_info = { 840 .name = "xen-pci-passthrough", 841 .parent = TYPE_PCI_DEVICE, 842 .instance_size = sizeof(XenPCIPassthroughState), 843 .class_init = xen_pci_passthrough_class_init, 844 }; 845 846 static void xen_pci_passthrough_register_types(void) 847 { 848 type_register_static(&xen_pci_passthrough_info); 849 } 850 851 type_init(xen_pci_passthrough_register_types) 852