1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2014 Intel Corporation. 3 * Copyright 2013-2014 6WIND S.A. 4 */ 5 6 #include <string.h> 7 #include <inttypes.h> 8 #include <stdint.h> 9 #include <stdbool.h> 10 #include <stdlib.h> 11 #include <stdio.h> 12 #include <sys/queue.h> 13 #include <rte_errno.h> 14 #include <rte_interrupts.h> 15 #include <rte_log.h> 16 #include <rte_bus.h> 17 #include <rte_pci.h> 18 #include <rte_bus_pci.h> 19 #include <rte_lcore.h> 20 #include <rte_per_lcore.h> 21 #include <rte_memory.h> 22 #include <rte_eal.h> 23 #include <rte_eal_paging.h> 24 #include <rte_string_fns.h> 25 #include <rte_common.h> 26 #include <rte_devargs.h> 27 #include <rte_vfio.h> 28 29 #include "private.h" 30 31 32 #define SYSFS_PCI_DEVICES "/sys/bus/pci/devices" 33 34 const char *rte_pci_get_sysfs_path(void) 35 { 36 const char *path = NULL; 37 38 #ifdef RTE_EXEC_ENV_LINUX 39 path = getenv("SYSFS_PCI_DEVICES"); 40 if (path == NULL) 41 return SYSFS_PCI_DEVICES; 42 #endif 43 44 return path; 45 } 46 47 static struct rte_devargs * 48 pci_devargs_lookup(const struct rte_pci_addr *pci_addr) 49 { 50 struct rte_devargs *devargs; 51 struct rte_pci_addr addr; 52 53 RTE_EAL_DEVARGS_FOREACH("pci", devargs) { 54 devargs->bus->parse(devargs->name, &addr); 55 if (!rte_pci_addr_cmp(pci_addr, &addr)) 56 return devargs; 57 } 58 return NULL; 59 } 60 61 void 62 pci_name_set(struct rte_pci_device *dev) 63 { 64 struct rte_devargs *devargs; 65 66 /* Each device has its internal, canonical name set. */ 67 rte_pci_device_name(&dev->addr, 68 dev->name, sizeof(dev->name)); 69 devargs = pci_devargs_lookup(&dev->addr); 70 dev->device.devargs = devargs; 71 72 /* When using a blocklist, only blocked devices will have 73 * an rte_devargs. Allowed devices won't have one. 74 */ 75 if (devargs != NULL) 76 /* If an rte_devargs exists, the generic rte_device uses the 77 * given name as its name. 78 */ 79 dev->device.name = dev->device.devargs->name; 80 else 81 /* Otherwise, it uses the internal, canonical form. */ 82 dev->device.name = dev->name; 83 } 84 85 /* map a particular resource from a file */ 86 void * 87 pci_map_resource(void *requested_addr, int fd, off_t offset, size_t size, 88 int additional_flags) 89 { 90 void *mapaddr; 91 92 /* Map the PCI memory resource of device */ 93 mapaddr = rte_mem_map(requested_addr, size, 94 RTE_PROT_READ | RTE_PROT_WRITE, 95 RTE_MAP_SHARED | additional_flags, fd, offset); 96 if (mapaddr == NULL) { 97 RTE_LOG(ERR, EAL, 98 "%s(): cannot map resource(%d, %p, 0x%zx, 0x%llx): %s (%p)\n", 99 __func__, fd, requested_addr, size, 100 (unsigned long long)offset, 101 rte_strerror(rte_errno), mapaddr); 102 } else 103 RTE_LOG(DEBUG, EAL, " PCI memory mapped at %p\n", mapaddr); 104 105 return mapaddr; 106 } 107 108 /* unmap a particular resource */ 109 void 110 pci_unmap_resource(void *requested_addr, size_t size) 111 { 112 if (requested_addr == NULL) 113 return; 114 115 /* Unmap the PCI memory resource of device */ 116 if (rte_mem_unmap(requested_addr, size)) { 117 RTE_LOG(ERR, EAL, "%s(): cannot mem unmap(%p, %#zx): %s\n", 118 __func__, requested_addr, size, 119 rte_strerror(rte_errno)); 120 } else 121 RTE_LOG(DEBUG, EAL, " PCI memory unmapped at %p\n", 122 requested_addr); 123 } 124 /* 125 * Match the PCI Driver and Device using the ID Table 126 */ 127 int 128 rte_pci_match(const struct rte_pci_driver *pci_drv, 129 const struct rte_pci_device *pci_dev) 130 { 131 const struct rte_pci_id *id_table; 132 133 for (id_table = pci_drv->id_table; id_table->vendor_id != 0; 134 id_table++) { 135 /* check if device's identifiers match the driver's ones */ 136 if (id_table->vendor_id != pci_dev->id.vendor_id && 137 id_table->vendor_id != RTE_PCI_ANY_ID) 138 continue; 139 if (id_table->device_id != pci_dev->id.device_id && 140 id_table->device_id != RTE_PCI_ANY_ID) 141 continue; 142 if (id_table->subsystem_vendor_id != 143 pci_dev->id.subsystem_vendor_id && 144 id_table->subsystem_vendor_id != RTE_PCI_ANY_ID) 145 continue; 146 if (id_table->subsystem_device_id != 147 pci_dev->id.subsystem_device_id && 148 id_table->subsystem_device_id != RTE_PCI_ANY_ID) 149 continue; 150 if (id_table->class_id != pci_dev->id.class_id && 151 id_table->class_id != RTE_CLASS_ANY_ID) 152 continue; 153 154 return 1; 155 } 156 157 return 0; 158 } 159 160 /* 161 * If vendor/device ID match, call the probe() function of the 162 * driver. 163 */ 164 static int 165 rte_pci_probe_one_driver(struct rte_pci_driver *dr, 166 struct rte_pci_device *dev) 167 { 168 int ret; 169 bool already_probed; 170 struct rte_pci_addr *loc; 171 172 if ((dr == NULL) || (dev == NULL)) 173 return -EINVAL; 174 175 loc = &dev->addr; 176 177 /* The device is not blocked; Check if driver supports it */ 178 if (!rte_pci_match(dr, dev)) 179 /* Match of device and driver failed */ 180 return 1; 181 182 RTE_LOG(DEBUG, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n", 183 loc->domain, loc->bus, loc->devid, loc->function, 184 dev->device.numa_node); 185 186 /* no initialization when marked as blocked, return without error */ 187 if (dev->device.devargs != NULL && 188 dev->device.devargs->policy == RTE_DEV_BLOCKED) { 189 RTE_LOG(INFO, EAL, " Device is blocked, not initializing\n"); 190 return 1; 191 } 192 193 if (dev->device.numa_node < 0) { 194 if (rte_socket_count() > 1) 195 RTE_LOG(INFO, EAL, "Device %s is not NUMA-aware, defaulting socket to 0\n", 196 dev->name); 197 dev->device.numa_node = 0; 198 } 199 200 already_probed = rte_dev_is_probed(&dev->device); 201 if (already_probed && !(dr->drv_flags & RTE_PCI_DRV_PROBE_AGAIN)) { 202 RTE_LOG(DEBUG, EAL, "Device %s is already probed\n", 203 dev->device.name); 204 return -EEXIST; 205 } 206 207 RTE_LOG(DEBUG, EAL, " probe driver: %x:%x %s\n", dev->id.vendor_id, 208 dev->id.device_id, dr->driver.name); 209 210 /* 211 * reference driver structure 212 * This needs to be before rte_pci_map_device(), as it enables to use 213 * driver flags for adjusting configuration. 214 */ 215 if (!already_probed) { 216 enum rte_iova_mode dev_iova_mode; 217 enum rte_iova_mode iova_mode; 218 219 dev_iova_mode = pci_device_iova_mode(dr, dev); 220 iova_mode = rte_eal_iova_mode(); 221 if (dev_iova_mode != RTE_IOVA_DC && 222 dev_iova_mode != iova_mode) { 223 RTE_LOG(ERR, EAL, " Expecting '%s' IOVA mode but current mode is '%s', not initializing\n", 224 dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA", 225 iova_mode == RTE_IOVA_PA ? "PA" : "VA"); 226 return -EINVAL; 227 } 228 229 dev->driver = dr; 230 } 231 232 if (!already_probed && (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)) { 233 /* map resources for devices that use igb_uio */ 234 ret = rte_pci_map_device(dev); 235 if (ret != 0) { 236 dev->driver = NULL; 237 return ret; 238 } 239 } 240 241 RTE_LOG(INFO, EAL, "Probe PCI driver: %s (%x:%x) device: "PCI_PRI_FMT" (socket %i)\n", 242 dr->driver.name, dev->id.vendor_id, dev->id.device_id, 243 loc->domain, loc->bus, loc->devid, loc->function, 244 dev->device.numa_node); 245 /* call the driver probe() function */ 246 ret = dr->probe(dr, dev); 247 if (already_probed) 248 return ret; /* no rollback if already succeeded earlier */ 249 if (ret) { 250 dev->driver = NULL; 251 if ((dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) && 252 /* Don't unmap if device is unsupported and 253 * driver needs mapped resources. 254 */ 255 !(ret > 0 && 256 (dr->drv_flags & RTE_PCI_DRV_KEEP_MAPPED_RES))) 257 rte_pci_unmap_device(dev); 258 } else { 259 dev->device.driver = &dr->driver; 260 } 261 262 return ret; 263 } 264 265 /* 266 * If vendor/device ID match, call the remove() function of the 267 * driver. 268 */ 269 static int 270 rte_pci_detach_dev(struct rte_pci_device *dev) 271 { 272 struct rte_pci_addr *loc; 273 struct rte_pci_driver *dr; 274 int ret = 0; 275 276 if (dev == NULL) 277 return -EINVAL; 278 279 dr = dev->driver; 280 loc = &dev->addr; 281 282 RTE_LOG(DEBUG, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n", 283 loc->domain, loc->bus, loc->devid, 284 loc->function, dev->device.numa_node); 285 286 RTE_LOG(DEBUG, EAL, " remove driver: %x:%x %s\n", dev->id.vendor_id, 287 dev->id.device_id, dr->driver.name); 288 289 if (dr->remove) { 290 ret = dr->remove(dev); 291 if (ret < 0) 292 return ret; 293 } 294 295 /* clear driver structure */ 296 dev->driver = NULL; 297 dev->device.driver = NULL; 298 299 if (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) 300 /* unmap resources for devices that use igb_uio */ 301 rte_pci_unmap_device(dev); 302 303 return 0; 304 } 305 306 /* 307 * If vendor/device ID match, call the probe() function of all 308 * registered driver for the given device. Return < 0 if initialization 309 * failed, return 1 if no driver is found for this device. 310 */ 311 static int 312 pci_probe_all_drivers(struct rte_pci_device *dev) 313 { 314 struct rte_pci_driver *dr = NULL; 315 int rc = 0; 316 317 if (dev == NULL) 318 return -EINVAL; 319 320 FOREACH_DRIVER_ON_PCIBUS(dr) { 321 rc = rte_pci_probe_one_driver(dr, dev); 322 if (rc < 0) 323 /* negative value is an error */ 324 return rc; 325 if (rc > 0) 326 /* positive value means driver doesn't support it */ 327 continue; 328 return 0; 329 } 330 return 1; 331 } 332 333 /* 334 * Scan the content of the PCI bus, and call the probe() function for 335 * all registered drivers that have a matching entry in its id_table 336 * for discovered devices. 337 */ 338 static int 339 pci_probe(void) 340 { 341 struct rte_pci_device *dev = NULL; 342 size_t probed = 0, failed = 0; 343 int ret = 0; 344 345 FOREACH_DEVICE_ON_PCIBUS(dev) { 346 probed++; 347 348 ret = pci_probe_all_drivers(dev); 349 if (ret < 0) { 350 if (ret != -EEXIST) { 351 RTE_LOG(ERR, EAL, "Requested device " 352 PCI_PRI_FMT " cannot be used\n", 353 dev->addr.domain, dev->addr.bus, 354 dev->addr.devid, dev->addr.function); 355 rte_errno = errno; 356 failed++; 357 } 358 ret = 0; 359 } 360 } 361 362 return (probed && probed == failed) ? -1 : 0; 363 } 364 365 /* dump one device */ 366 static int 367 pci_dump_one_device(FILE *f, struct rte_pci_device *dev) 368 { 369 int i; 370 371 fprintf(f, PCI_PRI_FMT, dev->addr.domain, dev->addr.bus, 372 dev->addr.devid, dev->addr.function); 373 fprintf(f, " - vendor:%x device:%x\n", dev->id.vendor_id, 374 dev->id.device_id); 375 376 for (i = 0; i != sizeof(dev->mem_resource) / 377 sizeof(dev->mem_resource[0]); i++) { 378 fprintf(f, " %16.16"PRIx64" %16.16"PRIx64"\n", 379 dev->mem_resource[i].phys_addr, 380 dev->mem_resource[i].len); 381 } 382 return 0; 383 } 384 385 /* dump devices on the bus */ 386 void 387 rte_pci_dump(FILE *f) 388 { 389 struct rte_pci_device *dev = NULL; 390 391 FOREACH_DEVICE_ON_PCIBUS(dev) { 392 pci_dump_one_device(f, dev); 393 } 394 } 395 396 static int 397 pci_parse(const char *name, void *addr) 398 { 399 struct rte_pci_addr *out = addr; 400 struct rte_pci_addr pci_addr; 401 bool parse; 402 403 parse = (rte_pci_addr_parse(name, &pci_addr) == 0); 404 if (parse && addr != NULL) 405 *out = pci_addr; 406 return parse == false; 407 } 408 409 /* register a driver */ 410 void 411 rte_pci_register(struct rte_pci_driver *driver) 412 { 413 TAILQ_INSERT_TAIL(&rte_pci_bus.driver_list, driver, next); 414 driver->bus = &rte_pci_bus; 415 } 416 417 /* unregister a driver */ 418 void 419 rte_pci_unregister(struct rte_pci_driver *driver) 420 { 421 TAILQ_REMOVE(&rte_pci_bus.driver_list, driver, next); 422 driver->bus = NULL; 423 } 424 425 /* Add a device to PCI bus */ 426 void 427 rte_pci_add_device(struct rte_pci_device *pci_dev) 428 { 429 TAILQ_INSERT_TAIL(&rte_pci_bus.device_list, pci_dev, next); 430 } 431 432 /* Insert a device into a predefined position in PCI bus */ 433 void 434 rte_pci_insert_device(struct rte_pci_device *exist_pci_dev, 435 struct rte_pci_device *new_pci_dev) 436 { 437 TAILQ_INSERT_BEFORE(exist_pci_dev, new_pci_dev, next); 438 } 439 440 /* Remove a device from PCI bus */ 441 static void 442 rte_pci_remove_device(struct rte_pci_device *pci_dev) 443 { 444 TAILQ_REMOVE(&rte_pci_bus.device_list, pci_dev, next); 445 } 446 447 static struct rte_device * 448 pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp, 449 const void *data) 450 { 451 const struct rte_pci_device *pstart; 452 struct rte_pci_device *pdev; 453 454 if (start != NULL) { 455 pstart = RTE_DEV_TO_PCI_CONST(start); 456 pdev = TAILQ_NEXT(pstart, next); 457 } else { 458 pdev = TAILQ_FIRST(&rte_pci_bus.device_list); 459 } 460 while (pdev != NULL) { 461 if (cmp(&pdev->device, data) == 0) 462 return &pdev->device; 463 pdev = TAILQ_NEXT(pdev, next); 464 } 465 return NULL; 466 } 467 468 /* 469 * find the device which encounter the failure, by iterate over all device on 470 * PCI bus to check if the memory failure address is located in the range 471 * of the BARs of the device. 472 */ 473 static struct rte_pci_device * 474 pci_find_device_by_addr(const void *failure_addr) 475 { 476 struct rte_pci_device *pdev = NULL; 477 uint64_t check_point, start, end, len; 478 int i; 479 480 check_point = (uint64_t)(uintptr_t)failure_addr; 481 482 FOREACH_DEVICE_ON_PCIBUS(pdev) { 483 for (i = 0; i != RTE_DIM(pdev->mem_resource); i++) { 484 start = (uint64_t)(uintptr_t)pdev->mem_resource[i].addr; 485 len = pdev->mem_resource[i].len; 486 end = start + len; 487 if (check_point >= start && check_point < end) { 488 RTE_LOG(DEBUG, EAL, "Failure address %16.16" 489 PRIx64" belongs to device %s!\n", 490 check_point, pdev->device.name); 491 return pdev; 492 } 493 } 494 } 495 return NULL; 496 } 497 498 static int 499 pci_hot_unplug_handler(struct rte_device *dev) 500 { 501 struct rte_pci_device *pdev = NULL; 502 int ret = 0; 503 504 pdev = RTE_DEV_TO_PCI(dev); 505 if (!pdev) 506 return -1; 507 508 switch (pdev->kdrv) { 509 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE 510 case RTE_PCI_KDRV_VFIO: 511 /* 512 * vfio kernel module guaranty the pci device would not be 513 * deleted until the user space release the resource, so no 514 * need to remap BARs resource here, just directly notify 515 * the req event to the user space to handle it. 516 */ 517 rte_dev_event_callback_process(dev->name, 518 RTE_DEV_EVENT_REMOVE); 519 break; 520 #endif 521 case RTE_PCI_KDRV_IGB_UIO: 522 case RTE_PCI_KDRV_UIO_GENERIC: 523 case RTE_PCI_KDRV_NIC_UIO: 524 /* BARs resource is invalid, remap it to be safe. */ 525 ret = pci_uio_remap_resource(pdev); 526 break; 527 default: 528 RTE_LOG(DEBUG, EAL, 529 "Not managed by a supported kernel driver, skipped\n"); 530 ret = -1; 531 break; 532 } 533 534 return ret; 535 } 536 537 static int 538 pci_sigbus_handler(const void *failure_addr) 539 { 540 struct rte_pci_device *pdev = NULL; 541 int ret = 0; 542 543 pdev = pci_find_device_by_addr(failure_addr); 544 if (!pdev) { 545 /* It is a generic sigbus error, no bus would handle it. */ 546 ret = 1; 547 } else { 548 /* The sigbus error is caused of hot-unplug. */ 549 ret = pci_hot_unplug_handler(&pdev->device); 550 if (ret) { 551 RTE_LOG(ERR, EAL, 552 "Failed to handle hot-unplug for device %s", 553 pdev->name); 554 ret = -1; 555 } 556 } 557 return ret; 558 } 559 560 static int 561 pci_plug(struct rte_device *dev) 562 { 563 return pci_probe_all_drivers(RTE_DEV_TO_PCI(dev)); 564 } 565 566 static int 567 pci_unplug(struct rte_device *dev) 568 { 569 struct rte_pci_device *pdev; 570 int ret; 571 572 pdev = RTE_DEV_TO_PCI(dev); 573 ret = rte_pci_detach_dev(pdev); 574 if (ret == 0) { 575 rte_pci_remove_device(pdev); 576 rte_devargs_remove(dev->devargs); 577 free(pdev); 578 } 579 return ret; 580 } 581 582 static int 583 pci_dma_map(struct rte_device *dev, void *addr, uint64_t iova, size_t len) 584 { 585 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev); 586 587 if (!pdev || !pdev->driver) { 588 rte_errno = EINVAL; 589 return -1; 590 } 591 if (pdev->driver->dma_map) 592 return pdev->driver->dma_map(pdev, addr, iova, len); 593 /** 594 * In case driver don't provides any specific mapping 595 * try fallback to VFIO. 596 */ 597 if (pdev->kdrv == RTE_PCI_KDRV_VFIO) 598 return rte_vfio_container_dma_map 599 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr, 600 iova, len); 601 rte_errno = ENOTSUP; 602 return -1; 603 } 604 605 static int 606 pci_dma_unmap(struct rte_device *dev, void *addr, uint64_t iova, size_t len) 607 { 608 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev); 609 610 if (!pdev || !pdev->driver) { 611 rte_errno = EINVAL; 612 return -1; 613 } 614 if (pdev->driver->dma_unmap) 615 return pdev->driver->dma_unmap(pdev, addr, iova, len); 616 /** 617 * In case driver don't provides any specific mapping 618 * try fallback to VFIO. 619 */ 620 if (pdev->kdrv == RTE_PCI_KDRV_VFIO) 621 return rte_vfio_container_dma_unmap 622 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr, 623 iova, len); 624 rte_errno = ENOTSUP; 625 return -1; 626 } 627 628 bool 629 rte_pci_ignore_device(const struct rte_pci_addr *pci_addr) 630 { 631 struct rte_devargs *devargs = pci_devargs_lookup(pci_addr); 632 633 switch (rte_pci_bus.bus.conf.scan_mode) { 634 case RTE_BUS_SCAN_ALLOWLIST: 635 if (devargs && devargs->policy == RTE_DEV_ALLOWED) 636 return false; 637 break; 638 case RTE_BUS_SCAN_UNDEFINED: 639 case RTE_BUS_SCAN_BLOCKLIST: 640 if (devargs == NULL || devargs->policy != RTE_DEV_BLOCKED) 641 return false; 642 break; 643 } 644 return true; 645 } 646 647 enum rte_iova_mode 648 rte_pci_get_iommu_class(void) 649 { 650 enum rte_iova_mode iova_mode = RTE_IOVA_DC; 651 const struct rte_pci_device *dev; 652 const struct rte_pci_driver *drv; 653 bool devices_want_va = false; 654 bool devices_want_pa = false; 655 int iommu_no_va = -1; 656 657 FOREACH_DEVICE_ON_PCIBUS(dev) { 658 /* 659 * We can check this only once, because the IOMMU hardware is 660 * the same for all of them. 661 */ 662 if (iommu_no_va == -1) 663 iommu_no_va = pci_device_iommu_support_va(dev) 664 ? 0 : 1; 665 666 if (dev->kdrv == RTE_PCI_KDRV_UNKNOWN || 667 dev->kdrv == RTE_PCI_KDRV_NONE) 668 continue; 669 FOREACH_DRIVER_ON_PCIBUS(drv) { 670 enum rte_iova_mode dev_iova_mode; 671 672 if (!rte_pci_match(drv, dev)) 673 continue; 674 675 dev_iova_mode = pci_device_iova_mode(drv, dev); 676 RTE_LOG(DEBUG, EAL, "PCI driver %s for device " 677 PCI_PRI_FMT " wants IOVA as '%s'\n", 678 drv->driver.name, 679 dev->addr.domain, dev->addr.bus, 680 dev->addr.devid, dev->addr.function, 681 dev_iova_mode == RTE_IOVA_DC ? "DC" : 682 (dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA")); 683 if (dev_iova_mode == RTE_IOVA_PA) 684 devices_want_pa = true; 685 else if (dev_iova_mode == RTE_IOVA_VA) 686 devices_want_va = true; 687 } 688 } 689 if (iommu_no_va == 1) { 690 iova_mode = RTE_IOVA_PA; 691 if (devices_want_va) { 692 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but IOMMU does not support 'VA'.\n"); 693 RTE_LOG(WARNING, EAL, "The devices that want 'VA' won't initialize.\n"); 694 } 695 } else if (devices_want_va && !devices_want_pa) { 696 iova_mode = RTE_IOVA_VA; 697 } else if (devices_want_pa && !devices_want_va) { 698 iova_mode = RTE_IOVA_PA; 699 } else { 700 iova_mode = RTE_IOVA_DC; 701 if (devices_want_va) { 702 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but forcing 'DC' because other devices want 'PA'.\n"); 703 RTE_LOG(WARNING, EAL, "Depending on the final decision by the EAL, not all devices may be able to initialize.\n"); 704 } 705 } 706 return iova_mode; 707 } 708 709 off_t 710 rte_pci_find_ext_capability(struct rte_pci_device *dev, uint32_t cap) 711 { 712 off_t offset = RTE_PCI_CFG_SPACE_SIZE; 713 uint32_t header; 714 int ttl; 715 716 /* minimum 8 bytes per capability */ 717 ttl = (RTE_PCI_CFG_SPACE_EXP_SIZE - RTE_PCI_CFG_SPACE_SIZE) / 8; 718 719 if (rte_pci_read_config(dev, &header, 4, offset) < 0) { 720 RTE_LOG(ERR, EAL, "error in reading extended capabilities\n"); 721 return -1; 722 } 723 724 /* 725 * If we have no capabilities, this is indicated by cap ID, 726 * cap version and next pointer all being 0. 727 */ 728 if (header == 0) 729 return 0; 730 731 while (ttl != 0) { 732 if (RTE_PCI_EXT_CAP_ID(header) == cap) 733 return offset; 734 735 offset = RTE_PCI_EXT_CAP_NEXT(header); 736 737 if (offset < RTE_PCI_CFG_SPACE_SIZE) 738 break; 739 740 if (rte_pci_read_config(dev, &header, 4, offset) < 0) { 741 RTE_LOG(ERR, EAL, 742 "error in reading extended capabilities\n"); 743 return -1; 744 } 745 746 ttl--; 747 } 748 749 return 0; 750 } 751 752 int 753 rte_pci_set_bus_master(struct rte_pci_device *dev, bool enable) 754 { 755 uint16_t old_cmd, cmd; 756 757 if (rte_pci_read_config(dev, &old_cmd, sizeof(old_cmd), 758 RTE_PCI_COMMAND) < 0) { 759 RTE_LOG(ERR, EAL, "error in reading PCI command register\n"); 760 return -1; 761 } 762 763 if (enable) 764 cmd = old_cmd | RTE_PCI_COMMAND_MASTER; 765 else 766 cmd = old_cmd & ~RTE_PCI_COMMAND_MASTER; 767 768 if (cmd == old_cmd) 769 return 0; 770 771 if (rte_pci_write_config(dev, &cmd, sizeof(cmd), 772 RTE_PCI_COMMAND) < 0) { 773 RTE_LOG(ERR, EAL, "error in writing PCI command register\n"); 774 return -1; 775 } 776 777 return 0; 778 } 779 780 struct rte_pci_bus rte_pci_bus = { 781 .bus = { 782 .scan = rte_pci_scan, 783 .probe = pci_probe, 784 .find_device = pci_find_device, 785 .plug = pci_plug, 786 .unplug = pci_unplug, 787 .parse = pci_parse, 788 .devargs_parse = rte_pci_devargs_parse, 789 .dma_map = pci_dma_map, 790 .dma_unmap = pci_dma_unmap, 791 .get_iommu_class = rte_pci_get_iommu_class, 792 .dev_iterate = rte_pci_dev_iterate, 793 .hot_unplug_handler = pci_hot_unplug_handler, 794 .sigbus_handler = pci_sigbus_handler, 795 }, 796 .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list), 797 .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list), 798 }; 799 800 RTE_REGISTER_BUS(pci, rte_pci_bus.bus); 801