ceph/src/spdk/dpdk/drivers/bus/pci/linux/pci.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <string.h>
   6 #include <dirent.h>
   7
   8 #include <rte_log.h>
   9 #include <rte_bus.h>
  10 #include <rte_pci.h>
  11 #include <rte_bus_pci.h>
  12 #include <rte_eal_memconfig.h>
  13 #include <rte_malloc.h>
  14 #include <rte_devargs.h>
  15 #include <rte_memcpy.h>
  16 #include <rte_vfio.h>
  17
  18 #include "eal_filesystem.h"
  19
  20 #include "private.h"
  21 #include "pci_init.h"
  22
  23 /**
  24  * @file
  25  * PCI probing under linux
  26  *
  27  * This code is used to simulate a PCI probe by parsing information in sysfs.
  28  * When a registered device matches a driver, it is then initialized with
  29  * IGB_UIO driver (or doesn't initialize, if the device wasn't bound to it).
  30  */
  31
  32 extern struct rte_pci_bus rte_pci_bus;
  33
  34 static int
  35 pci_get_kernel_driver_by_path(const char *filename, char *dri_name,
  36                               size_t len)
  37 {
  38         int count;
  39         char path[PATH_MAX];
  40         char *name;
  41
  42         if (!filename || !dri_name)
  43                 return -1;
  44
  45         count = readlink(filename, path, PATH_MAX);
  46         if (count >= PATH_MAX)
  47                 return -1;
  48
  49         /* For device does not have a driver */
  50         if (count < 0)
  51                 return 1;
  52
  53         path[count] = '\0';
  54
  55         name = strrchr(path, '/');
  56         if (name) {
  57                 strlcpy(dri_name, name + 1, len);
  58                 return 0;
  59         }
  60
  61         return -1;
  62 }
  63
  64 /* Map pci device */
  65 int
  66 rte_pci_map_device(struct rte_pci_device *dev)
  67 {
  68         int ret = -1;
  69
  70         /* try mapping the NIC resources using VFIO if it exists */
  71         switch (dev->kdrv) {
  72         case RTE_KDRV_VFIO:
  73 #ifdef VFIO_PRESENT
  74                 if (pci_vfio_is_enabled())
  75                         ret = pci_vfio_map_resource(dev);
  76 #endif
  77                 break;
  78         case RTE_KDRV_IGB_UIO:
  79         case RTE_KDRV_UIO_GENERIC:
  80                 if (rte_eal_using_phys_addrs()) {
  81                         /* map resources for devices that use uio */
  82                         ret = pci_uio_map_resource(dev);
  83                 }
  84                 break;
  85         default:
  86                 RTE_LOG(DEBUG, EAL,
  87                         "  Not managed by a supported kernel driver, skipped\n");
  88                 ret = 1;
  89                 break;
  90         }
  91
  92         return ret;
  93 }
  94
  95 /* Unmap pci device */
  96 void
  97 rte_pci_unmap_device(struct rte_pci_device *dev)
  98 {
  99         /* try unmapping the NIC resources using VFIO if it exists */
 100         switch (dev->kdrv) {
 101         case RTE_KDRV_VFIO:
 102 #ifdef VFIO_PRESENT
 103                 if (pci_vfio_is_enabled())
 104                         pci_vfio_unmap_resource(dev);
 105 #endif
 106                 break;
 107         case RTE_KDRV_IGB_UIO:
 108         case RTE_KDRV_UIO_GENERIC:
 109                 /* unmap resources for devices that use uio */
 110                 pci_uio_unmap_resource(dev);
 111                 break;
 112         default:
 113                 RTE_LOG(DEBUG, EAL,
 114                         "  Not managed by a supported kernel driver, skipped\n");
 115                 break;
 116         }
 117 }
 118
 119 static int
 120 find_max_end_va(const struct rte_memseg_list *msl, void *arg)
 121 {
 122         size_t sz = msl->memseg_arr.len * msl->page_sz;
 123         void *end_va = RTE_PTR_ADD(msl->base_va, sz);
 124         void **max_va = arg;
 125
 126         if (*max_va < end_va)
 127                 *max_va = end_va;
 128         return 0;
 129 }
 130
 131 void *
 132 pci_find_max_end_va(void)
 133 {
 134         void *va = NULL;
 135
 136         rte_memseg_list_walk(find_max_end_va, &va);
 137         return va;
 138 }
 139
 140
 141 /* parse one line of the "resource" sysfs file (note that the 'line'
 142  * string is modified)
 143  */
 144 int
 145 pci_parse_one_sysfs_resource(char *line, size_t len, uint64_t *phys_addr,
 146         uint64_t *end_addr, uint64_t *flags)
 147 {
 148         union pci_resource_info {
 149                 struct {
 150                         char *phys_addr;
 151                         char *end_addr;
 152                         char *flags;
 153                 };
 154                 char *ptrs[PCI_RESOURCE_FMT_NVAL];
 155         } res_info;
 156
 157         if (rte_strsplit(line, len, res_info.ptrs, 3, ' ') != 3) {
 158                 RTE_LOG(ERR, EAL,
 159                         "%s(): bad resource format\n", __func__);
 160                 return -1;
 161         }
 162         errno = 0;
 163         *phys_addr = strtoull(res_info.phys_addr, NULL, 16);
 164         *end_addr = strtoull(res_info.end_addr, NULL, 16);
 165         *flags = strtoull(res_info.flags, NULL, 16);
 166         if (errno != 0) {
 167                 RTE_LOG(ERR, EAL,
 168                         "%s(): bad resource format\n", __func__);
 169                 return -1;
 170         }
 171
 172         return 0;
 173 }
 174
 175 /* parse the "resource" sysfs file */
 176 static int
 177 pci_parse_sysfs_resource(const char *filename, struct rte_pci_device *dev)
 178 {
 179         FILE *f;
 180         char buf[BUFSIZ];
 181         int i;
 182         uint64_t phys_addr, end_addr, flags;
 183
 184         f = fopen(filename, "r");
 185         if (f == NULL) {
 186                 RTE_LOG(ERR, EAL, "Cannot open sysfs resource\n");
 187                 return -1;
 188         }
 189
 190         for (i = 0; i<PCI_MAX_RESOURCE; i++) {
 191
 192                 if (fgets(buf, sizeof(buf), f) == NULL) {
 193                         RTE_LOG(ERR, EAL,
 194                                 "%s(): cannot read resource\n", __func__);
 195                         goto error;
 196                 }
 197                 if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
 198                                 &end_addr, &flags) < 0)
 199                         goto error;
 200
 201                 if (flags & IORESOURCE_MEM) {
 202                         dev->mem_resource[i].phys_addr = phys_addr;
 203                         dev->mem_resource[i].len = end_addr - phys_addr + 1;
 204                         /* not mapped for now */
 205                         dev->mem_resource[i].addr = NULL;
 206                 }
 207         }
 208         fclose(f);
 209         return 0;
 210
 211 error:
 212         fclose(f);
 213         return -1;
 214 }
 215
 216 /* Scan one pci sysfs entry, and fill the devices list from it. */
 217 static int
 218 pci_scan_one(const char *dirname, const struct rte_pci_addr *addr)
 219 {
 220         char filename[PATH_MAX];
 221         unsigned long tmp;
 222         struct rte_pci_device *dev;
 223         char driver[PATH_MAX];
 224         int ret;
 225
 226         dev = malloc(sizeof(*dev));
 227         if (dev == NULL)
 228                 return -1;
 229
 230         memset(dev, 0, sizeof(*dev));
 231         dev->addr = *addr;
 232
 233         /* get vendor id */
 234         snprintf(filename, sizeof(filename), "%s/vendor", dirname);
 235         if (eal_parse_sysfs_value(filename, &tmp) < 0) {
 236                 free(dev);
 237                 return -1;
 238         }
 239         dev->id.vendor_id = (uint16_t)tmp;
 240
 241         /* get device id */
 242         snprintf(filename, sizeof(filename), "%s/device", dirname);
 243         if (eal_parse_sysfs_value(filename, &tmp) < 0) {
 244                 free(dev);
 245                 return -1;
 246         }
 247         dev->id.device_id = (uint16_t)tmp;
 248
 249         /* get subsystem_vendor id */
 250         snprintf(filename, sizeof(filename), "%s/subsystem_vendor",
 251                  dirname);
 252         if (eal_parse_sysfs_value(filename, &tmp) < 0) {
 253                 free(dev);
 254                 return -1;
 255         }
 256         dev->id.subsystem_vendor_id = (uint16_t)tmp;
 257
 258         /* get subsystem_device id */
 259         snprintf(filename, sizeof(filename), "%s/subsystem_device",
 260                  dirname);
 261         if (eal_parse_sysfs_value(filename, &tmp) < 0) {
 262                 free(dev);
 263                 return -1;
 264         }
 265         dev->id.subsystem_device_id = (uint16_t)tmp;
 266
 267         /* get class_id */
 268         snprintf(filename, sizeof(filename), "%s/class",
 269                  dirname);
 270         if (eal_parse_sysfs_value(filename, &tmp) < 0) {
 271                 free(dev);
 272                 return -1;
 273         }
 274         /* the least 24 bits are valid: class, subclass, program interface */
 275         dev->id.class_id = (uint32_t)tmp & RTE_CLASS_ANY_ID;
 276
 277         /* get max_vfs */
 278         dev->max_vfs = 0;
 279         snprintf(filename, sizeof(filename), "%s/max_vfs", dirname);
 280         if (!access(filename, F_OK) &&
 281             eal_parse_sysfs_value(filename, &tmp) == 0)
 282                 dev->max_vfs = (uint16_t)tmp;
 283         else {
 284                 /* for non igb_uio driver, need kernel version >= 3.8 */
 285                 snprintf(filename, sizeof(filename),
 286                          "%s/sriov_numvfs", dirname);
 287                 if (!access(filename, F_OK) &&
 288                     eal_parse_sysfs_value(filename, &tmp) == 0)
 289                         dev->max_vfs = (uint16_t)tmp;
 290         }
 291
 292         /* get numa node, default to 0 if not present */
 293         snprintf(filename, sizeof(filename), "%s/numa_node",
 294                  dirname);
 295
 296         if (access(filename, F_OK) != -1) {
 297                 if (eal_parse_sysfs_value(filename, &tmp) == 0)
 298                         dev->device.numa_node = tmp;
 299                 else
 300                         dev->device.numa_node = -1;
 301         } else {
 302                 dev->device.numa_node = 0;
 303         }
 304
 305         pci_name_set(dev);
 306
 307         /* parse resources */
 308         snprintf(filename, sizeof(filename), "%s/resource", dirname);
 309         if (pci_parse_sysfs_resource(filename, dev) < 0) {
 310                 RTE_LOG(ERR, EAL, "%s(): cannot parse resource\n", __func__);
 311                 free(dev);
 312                 return -1;
 313         }
 314
 315         /* parse driver */
 316         snprintf(filename, sizeof(filename), "%s/driver", dirname);
 317         ret = pci_get_kernel_driver_by_path(filename, driver, sizeof(driver));
 318         if (ret < 0) {
 319                 RTE_LOG(ERR, EAL, "Fail to get kernel driver\n");
 320                 free(dev);
 321                 return -1;
 322         }
 323
 324         if (!ret) {
 325                 if (!strcmp(driver, "vfio-pci"))
 326                         dev->kdrv = RTE_KDRV_VFIO;
 327                 else if (!strcmp(driver, "igb_uio"))
 328                         dev->kdrv = RTE_KDRV_IGB_UIO;
 329                 else if (!strcmp(driver, "uio_pci_generic"))
 330                         dev->kdrv = RTE_KDRV_UIO_GENERIC;
 331                 else
 332                         dev->kdrv = RTE_KDRV_UNKNOWN;
 333         } else
 334                 dev->kdrv = RTE_KDRV_NONE;
 335
 336         /* device is valid, add in list (sorted) */
 337         if (TAILQ_EMPTY(&rte_pci_bus.device_list)) {
 338                 rte_pci_add_device(dev);
 339         } else {
 340                 struct rte_pci_device *dev2;
 341                 int ret;
 342
 343                 TAILQ_FOREACH(dev2, &rte_pci_bus.device_list, next) {
 344                         ret = rte_pci_addr_cmp(&dev->addr, &dev2->addr);
 345                         if (ret > 0)
 346                                 continue;
 347
 348                         if (ret < 0) {
 349                                 rte_pci_insert_device(dev2, dev);
 350                         } else { /* already registered */
 351                                 if (dev2->driver == NULL) {
 352                                         dev2->kdrv = dev->kdrv;
 353                                         dev2->max_vfs = dev->max_vfs;
 354                                         pci_name_set(dev2);
 355                                         memmove(dev2->mem_resource,
 356                                                 dev->mem_resource,
 357                                                 sizeof(dev->mem_resource));
 358                                 } else {
 359                                         /**
 360                                          * If device is plugged and driver is
 361                                          * probed already, we don't need to do
 362                                          * anything here. (This happens when we
 363                                          * call rte_eal_hotplug_add)
 364                                          */
 365                                         if (dev2->kdrv != dev->kdrv ||
 366                                                 dev2->max_vfs != dev->max_vfs)
 367                                                 /*
 368                                                  * This should not happens.
 369                                                  * But it is still possible if
 370                                                  * we unbind a device from
 371                                                  * vfio or uio before hotplug
 372                                                  * remove and rebind it with
 373                                                  * a different configure.
 374                                                  * So we just print out the
 375                                                  * error as an alarm.
 376                                                  */
 377                                                 RTE_LOG(ERR, EAL, "Unexpected device scan at %s!\n",
 378                                                         filename);
 379                                 }
 380                                 free(dev);
 381                         }
 382                         return 0;
 383                 }
 384
 385                 rte_pci_add_device(dev);
 386         }
 387
 388         return 0;
 389 }
 390
 391 int
 392 pci_update_device(const struct rte_pci_addr *addr)
 393 {
 394         char filename[PATH_MAX];
 395
 396         snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
 397                  rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
 398                  addr->function);
 399
 400         return pci_scan_one(filename, addr);
 401 }
 402
 403 /*
 404  * split up a pci address into its constituent parts.
 405  */
 406 static int
 407 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
 408 {
 409         /* first split on ':' */
 410         union splitaddr {
 411                 struct {
 412                         char *domain;
 413                         char *bus;
 414                         char *devid;
 415                         char *function;
 416                 };
 417                 char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
 418         } splitaddr;
 419
 420         char *buf_copy = strndup(buf, bufsize);
 421         if (buf_copy == NULL)
 422                 return -1;
 423
 424         if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
 425                         != PCI_FMT_NVAL - 1)
 426                 goto error;
 427         /* final split is on '.' between devid and function */
 428         splitaddr.function = strchr(splitaddr.devid,'.');
 429         if (splitaddr.function == NULL)
 430                 goto error;
 431         *splitaddr.function++ = '\0';
 432
 433         /* now convert to int values */
 434         errno = 0;
 435         addr->domain = strtoul(splitaddr.domain, NULL, 16);
 436         addr->bus = strtoul(splitaddr.bus, NULL, 16);
 437         addr->devid = strtoul(splitaddr.devid, NULL, 16);
 438         addr->function = strtoul(splitaddr.function, NULL, 10);
 439         if (errno != 0)
 440                 goto error;
 441
 442         free(buf_copy); /* free the copy made with strdup */
 443         return 0;
 444 error:
 445         free(buf_copy);
 446         return -1;
 447 }
 448
 449 /*
 450  * Scan the content of the PCI bus, and the devices in the devices
 451  * list
 452  */
 453 int
 454 rte_pci_scan(void)
 455 {
 456         struct dirent *e;
 457         DIR *dir;
 458         char dirname[PATH_MAX];
 459         struct rte_pci_addr addr;
 460
 461         /* for debug purposes, PCI can be disabled */
 462         if (!rte_eal_has_pci())
 463                 return 0;
 464
 465 #ifdef VFIO_PRESENT
 466         if (!pci_vfio_is_enabled())
 467                 RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
 468 #endif
 469
 470         dir = opendir(rte_pci_get_sysfs_path());
 471         if (dir == NULL) {
 472                 RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
 473                         __func__, strerror(errno));
 474                 return -1;
 475         }
 476
 477         while ((e = readdir(dir)) != NULL) {
 478                 if (e->d_name[0] == '.')
 479                         continue;
 480
 481                 if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
 482                         continue;
 483
 484                 snprintf(dirname, sizeof(dirname), "%s/%s",
 485                                 rte_pci_get_sysfs_path(), e->d_name);
 486
 487                 if (pci_scan_one(dirname, &addr) < 0)
 488                         goto error;
 489         }
 490         closedir(dir);
 491         return 0;
 492
 493 error:
 494         closedir(dir);
 495         return -1;
 496 }
 497
 498 /*
 499  * Is pci device bound to any kdrv
 500  */
 501 static inline int
 502 pci_one_device_is_bound(void)
 503 {
 504         struct rte_pci_device *dev = NULL;
 505         int ret = 0;
 506
 507         FOREACH_DEVICE_ON_PCIBUS(dev) {
 508                 if (dev->kdrv == RTE_KDRV_UNKNOWN ||
 509                     dev->kdrv == RTE_KDRV_NONE) {
 510                         continue;
 511                 } else {
 512                         ret = 1;
 513                         break;
 514                 }
 515         }
 516         return ret;
 517 }
 518
 519 /*
 520  * Any one of the device bound to uio
 521  */
 522 static inline int
 523 pci_one_device_bound_uio(void)
 524 {
 525         struct rte_pci_device *dev = NULL;
 526         struct rte_devargs *devargs;
 527         int need_check;
 528
 529         FOREACH_DEVICE_ON_PCIBUS(dev) {
 530                 devargs = dev->device.devargs;
 531
 532                 need_check = 0;
 533                 switch (rte_pci_bus.bus.conf.scan_mode) {
 534                 case RTE_BUS_SCAN_WHITELIST:
 535                         if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
 536                                 need_check = 1;
 537                         break;
 538                 case RTE_BUS_SCAN_UNDEFINED:
 539                 case RTE_BUS_SCAN_BLACKLIST:
 540                         if (devargs == NULL ||
 541                             devargs->policy != RTE_DEV_BLACKLISTED)
 542                                 need_check = 1;
 543                         break;
 544                 }
 545
 546                 if (!need_check)
 547                         continue;
 548
 549                 if (dev->kdrv == RTE_KDRV_IGB_UIO ||
 550                    dev->kdrv == RTE_KDRV_UIO_GENERIC) {
 551                         return 1;
 552                 }
 553         }
 554         return 0;
 555 }
 556
 557 /*
 558  * Any one of the device has iova as va
 559  */
 560 static inline int
 561 pci_one_device_has_iova_va(void)
 562 {
 563         struct rte_pci_device *dev = NULL;
 564         struct rte_pci_driver *drv = NULL;
 565
 566         FOREACH_DRIVER_ON_PCIBUS(drv) {
 567                 if (drv && drv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) {
 568                         FOREACH_DEVICE_ON_PCIBUS(dev) {
 569                                 if (dev->kdrv == RTE_KDRV_VFIO &&
 570                                     rte_pci_match(drv, dev))
 571                                         return 1;
 572                         }
 573                 }
 574         }
 575         return 0;
 576 }
 577
 578 #if defined(RTE_ARCH_X86)
 579 static bool
 580 pci_one_device_iommu_support_va(struct rte_pci_device *dev)
 581 {
 582 #define VTD_CAP_MGAW_SHIFT      16
 583 #define VTD_CAP_MGAW_MASK       (0x3fULL << VTD_CAP_MGAW_SHIFT)
 584 #define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
 585         struct rte_pci_addr *addr = &dev->addr;
 586         char filename[PATH_MAX];
 587         FILE *fp;
 588         uint64_t mgaw, vtd_cap_reg = 0;
 589
 590         snprintf(filename, sizeof(filename),
 591                  "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
 592                  rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
 593                  addr->function);
 594         if (access(filename, F_OK) == -1) {
 595                 /* We don't have an Intel IOMMU, assume VA supported*/
 596                 return true;
 597         }
 598
 599         /* We have an intel IOMMU */
 600         fp = fopen(filename, "r");
 601         if (fp == NULL) {
 602                 RTE_LOG(ERR, EAL, "%s(): can't open %s\n", __func__, filename);
 603                 return false;
 604         }
 605
 606         if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
 607                 RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
 608                 fclose(fp);
 609                 return false;
 610         }
 611
 612         fclose(fp);
 613
 614         mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
 615         if (mgaw < X86_VA_WIDTH)
 616                 return false;
 617
 618         return true;
 619 }
 620 #elif defined(RTE_ARCH_PPC_64)
 621 static bool
 622 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
 623 {
 624         return false;
 625 }
 626 #else
 627 static bool
 628 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
 629 {
 630         return true;
 631 }
 632 #endif
 633
 634 /*
 635  * All devices IOMMUs support VA as IOVA
 636  */
 637 static bool
 638 pci_devices_iommu_support_va(void)
 639 {
 640         struct rte_pci_device *dev = NULL;
 641         struct rte_pci_driver *drv = NULL;
 642
 643         FOREACH_DRIVER_ON_PCIBUS(drv) {
 644                 FOREACH_DEVICE_ON_PCIBUS(dev) {
 645                         if (!rte_pci_match(drv, dev))
 646                                 continue;
 647                         if (!pci_one_device_iommu_support_va(dev))
 648                                 return false;
 649                 }
 650         }
 651         return true;
 652 }
 653
 654 /*
 655  * Get iommu class of PCI devices on the bus.
 656  */
 657 enum rte_iova_mode
 658 rte_pci_get_iommu_class(void)
 659 {
 660         bool is_bound;
 661         bool is_vfio_noiommu_enabled = true;
 662         bool has_iova_va;
 663         bool is_bound_uio;
 664         bool iommu_no_va;
 665
 666         is_bound = pci_one_device_is_bound();
 667         if (!is_bound)
 668                 return RTE_IOVA_DC;
 669
 670         has_iova_va = pci_one_device_has_iova_va();
 671         is_bound_uio = pci_one_device_bound_uio();
 672         iommu_no_va = !pci_devices_iommu_support_va();
 673 #ifdef VFIO_PRESENT
 674         is_vfio_noiommu_enabled = rte_vfio_noiommu_is_enabled() == true ?
 675                                         true : false;
 676 #endif
 677
 678         if (has_iova_va && !is_bound_uio && !is_vfio_noiommu_enabled &&
 679                         !iommu_no_va)
 680                 return RTE_IOVA_VA;
 681
 682         if (has_iova_va) {
 683                 RTE_LOG(WARNING, EAL, "Some devices want iova as va but pa will be used because.. ");
 684                 if (is_vfio_noiommu_enabled)
 685                         RTE_LOG(WARNING, EAL, "vfio-noiommu mode configured\n");
 686                 if (is_bound_uio)
 687                         RTE_LOG(WARNING, EAL, "few device bound to UIO\n");
 688                 if (iommu_no_va)
 689                         RTE_LOG(WARNING, EAL, "IOMMU does not support IOVA as VA\n");
 690         }
 691
 692         return RTE_IOVA_PA;
 693 }
 694
 695 /* Read PCI config space. */
 696 int rte_pci_read_config(const struct rte_pci_device *device,
 697                 void *buf, size_t len, off_t offset)
 698 {
 699         const struct rte_intr_handle *intr_handle = &device->intr_handle;
 700
 701         switch (intr_handle->type) {
 702         case RTE_INTR_HANDLE_UIO:
 703         case RTE_INTR_HANDLE_UIO_INTX:
 704                 return pci_uio_read_config(intr_handle, buf, len, offset);
 705
 706 #ifdef VFIO_PRESENT
 707         case RTE_INTR_HANDLE_VFIO_MSIX:
 708         case RTE_INTR_HANDLE_VFIO_MSI:
 709         case RTE_INTR_HANDLE_VFIO_LEGACY:
 710                 return pci_vfio_read_config(intr_handle, buf, len, offset);
 711 #endif
 712         default:
 713                 RTE_LOG(ERR, EAL,
 714                         "Unknown handle type of fd %d\n",
 715                                         intr_handle->fd);
 716                 return -1;
 717         }
 718 }
 719
 720 /* Write PCI config space. */
 721 int rte_pci_write_config(const struct rte_pci_device *device,
 722                 const void *buf, size_t len, off_t offset)
 723 {
 724         const struct rte_intr_handle *intr_handle = &device->intr_handle;
 725
 726         switch (intr_handle->type) {
 727         case RTE_INTR_HANDLE_UIO:
 728         case RTE_INTR_HANDLE_UIO_INTX:
 729                 return pci_uio_write_config(intr_handle, buf, len, offset);
 730
 731 #ifdef VFIO_PRESENT
 732         case RTE_INTR_HANDLE_VFIO_MSIX:
 733         case RTE_INTR_HANDLE_VFIO_MSI:
 734         case RTE_INTR_HANDLE_VFIO_LEGACY:
 735                 return pci_vfio_write_config(intr_handle, buf, len, offset);
 736 #endif
 737         default:
 738                 RTE_LOG(ERR, EAL,
 739                         "Unknown handle type of fd %d\n",
 740                                         intr_handle->fd);
 741                 return -1;
 742         }
 743 }
 744
 745 #if defined(RTE_ARCH_X86)
 746 static int
 747 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
 748                 struct rte_pci_ioport *p)
 749 {
 750         uint16_t start, end;
 751         FILE *fp;
 752         char *line = NULL;
 753         char pci_id[16];
 754         int found = 0;
 755         size_t linesz;
 756
 757         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
 758                  dev->addr.domain, dev->addr.bus,
 759                  dev->addr.devid, dev->addr.function);
 760
 761         fp = fopen("/proc/ioports", "r");
 762         if (fp == NULL) {
 763                 RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
 764                 return -1;
 765         }
 766
 767         while (getdelim(&line, &linesz, '\n', fp) > 0) {
 768                 char *ptr = line;
 769                 char *left;
 770                 int n;
 771
 772                 n = strcspn(ptr, ":");
 773                 ptr[n] = 0;
 774                 left = &ptr[n + 1];
 775
 776                 while (*left && isspace(*left))
 777                         left++;
 778
 779                 if (!strncmp(left, pci_id, strlen(pci_id))) {
 780                         found = 1;
 781
 782                         while (*ptr && isspace(*ptr))
 783                                 ptr++;
 784
 785                         sscanf(ptr, "%04hx-%04hx", &start, &end);
 786
 787                         break;
 788                 }
 789         }
 790
 791         free(line);
 792         fclose(fp);
 793
 794         if (!found)
 795                 return -1;
 796
 797         p->base = start;
 798         RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
 799
 800         return 0;
 801 }
 802 #endif
 803
 804 int
 805 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
 806                 struct rte_pci_ioport *p)
 807 {
 808         int ret = -1;
 809
 810         switch (dev->kdrv) {
 811 #ifdef VFIO_PRESENT
 812         case RTE_KDRV_VFIO:
 813                 if (pci_vfio_is_enabled())
 814                         ret = pci_vfio_ioport_map(dev, bar, p);
 815                 break;
 816 #endif
 817         case RTE_KDRV_IGB_UIO:
 818                 ret = pci_uio_ioport_map(dev, bar, p);
 819                 break;
 820         case RTE_KDRV_UIO_GENERIC:
 821 #if defined(RTE_ARCH_X86)
 822                 ret = pci_ioport_map(dev, bar, p);
 823 #else
 824                 ret = pci_uio_ioport_map(dev, bar, p);
 825 #endif
 826                 break;
 827         case RTE_KDRV_NONE:
 828 #if defined(RTE_ARCH_X86)
 829                 ret = pci_ioport_map(dev, bar, p);
 830 #endif
 831                 break;
 832         default:
 833                 break;
 834         }
 835
 836         if (!ret)
 837                 p->dev = dev;
 838
 839         return ret;
 840 }
 841
 842 void
 843 rte_pci_ioport_read(struct rte_pci_ioport *p,
 844                 void *data, size_t len, off_t offset)
 845 {
 846         switch (p->dev->kdrv) {
 847 #ifdef VFIO_PRESENT
 848         case RTE_KDRV_VFIO:
 849                 pci_vfio_ioport_read(p, data, len, offset);
 850                 break;
 851 #endif
 852         case RTE_KDRV_IGB_UIO:
 853                 pci_uio_ioport_read(p, data, len, offset);
 854                 break;
 855         case RTE_KDRV_UIO_GENERIC:
 856                 pci_uio_ioport_read(p, data, len, offset);
 857                 break;
 858         case RTE_KDRV_NONE:
 859 #if defined(RTE_ARCH_X86)
 860                 pci_uio_ioport_read(p, data, len, offset);
 861 #endif
 862                 break;
 863         default:
 864                 break;
 865         }
 866 }
 867
 868 void
 869 rte_pci_ioport_write(struct rte_pci_ioport *p,
 870                 const void *data, size_t len, off_t offset)
 871 {
 872         switch (p->dev->kdrv) {
 873 #ifdef VFIO_PRESENT
 874         case RTE_KDRV_VFIO:
 875                 pci_vfio_ioport_write(p, data, len, offset);
 876                 break;
 877 #endif
 878         case RTE_KDRV_IGB_UIO:
 879                 pci_uio_ioport_write(p, data, len, offset);
 880                 break;
 881         case RTE_KDRV_UIO_GENERIC:
 882                 pci_uio_ioport_write(p, data, len, offset);
 883                 break;
 884         case RTE_KDRV_NONE:
 885 #if defined(RTE_ARCH_X86)
 886                 pci_uio_ioport_write(p, data, len, offset);
 887 #endif
 888                 break;
 889         default:
 890                 break;
 891         }
 892 }
 893
 894 int
 895 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
 896 {
 897         int ret = -1;
 898
 899         switch (p->dev->kdrv) {
 900 #ifdef VFIO_PRESENT
 901         case RTE_KDRV_VFIO:
 902                 if (pci_vfio_is_enabled())
 903                         ret = pci_vfio_ioport_unmap(p);
 904                 break;
 905 #endif
 906         case RTE_KDRV_IGB_UIO:
 907                 ret = pci_uio_ioport_unmap(p);
 908                 break;
 909         case RTE_KDRV_UIO_GENERIC:
 910 #if defined(RTE_ARCH_X86)
 911                 ret = 0;
 912 #else
 913                 ret = pci_uio_ioport_unmap(p);
 914 #endif
 915                 break;
 916         case RTE_KDRV_NONE:
 917 #if defined(RTE_ARCH_X86)
 918                 ret = 0;
 919 #endif
 920                 break;
 921         default:
 922                 break;
 923         }
 924
 925         return ret;
 926 }