ceph/src/seastar/dpdk/drivers/bus/pci/linux/pci_uio.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <string.h>
   6 #include <unistd.h>
   7 #include <fcntl.h>
   8 #include <dirent.h>
   9 #include <inttypes.h>
  10 #include <sys/stat.h>
  11 #include <sys/mman.h>
  12 #include <sys/sysmacros.h>
  13 #include <linux/pci_regs.h>
  14
  15 #if defined(RTE_ARCH_X86)
  16 #include <sys/io.h>
  17 #endif
  18
  19 #include <rte_string_fns.h>
  20 #include <rte_log.h>
  21 #include <rte_pci.h>
  22 #include <rte_bus_pci.h>
  23 #include <rte_eal_memconfig.h>
  24 #include <rte_common.h>
  25 #include <rte_malloc.h>
  26
  27 #include "eal_filesystem.h"
  28 #include "pci_init.h"
  29
  30 void *pci_map_addr = NULL;
  31
  32 #define OFF_MAX              ((uint64_t)(off_t)-1)
  33
  34 int
  35 pci_uio_read_config(const struct rte_intr_handle *intr_handle,
  36                     void *buf, size_t len, off_t offset)
  37 {
  38         return pread(intr_handle->uio_cfg_fd, buf, len, offset);
  39 }
  40
  41 int
  42 pci_uio_write_config(const struct rte_intr_handle *intr_handle,
  43                      const void *buf, size_t len, off_t offset)
  44 {
  45         return pwrite(intr_handle->uio_cfg_fd, buf, len, offset);
  46 }
  47
  48 static int
  49 pci_uio_set_bus_master(int dev_fd)
  50 {
  51         uint16_t reg;
  52         int ret;
  53
  54         ret = pread(dev_fd, &reg, sizeof(reg), PCI_COMMAND);
  55         if (ret != sizeof(reg)) {
  56                 RTE_LOG(ERR, EAL,
  57                         "Cannot read command from PCI config space!\n");
  58                 return -1;
  59         }
  60
  61         /* return if bus mastering is already on */
  62         if (reg & PCI_COMMAND_MASTER)
  63                 return 0;
  64
  65         reg |= PCI_COMMAND_MASTER;
  66
  67         ret = pwrite(dev_fd, &reg, sizeof(reg), PCI_COMMAND);
  68         if (ret != sizeof(reg)) {
  69                 RTE_LOG(ERR, EAL,
  70                         "Cannot write command to PCI config space!\n");
  71                 return -1;
  72         }
  73
  74         return 0;
  75 }
  76
  77 static int
  78 pci_mknod_uio_dev(const char *sysfs_uio_path, unsigned uio_num)
  79 {
  80         FILE *f;
  81         char filename[PATH_MAX];
  82         int ret;
  83         unsigned major, minor;
  84         dev_t dev;
  85
  86         /* get the name of the sysfs file that contains the major and minor
  87          * of the uio device and read its content */
  88         snprintf(filename, sizeof(filename), "%s/dev", sysfs_uio_path);
  89
  90         f = fopen(filename, "r");
  91         if (f == NULL) {
  92                 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs to get major:minor\n",
  93                         __func__);
  94                 return -1;
  95         }
  96
  97         ret = fscanf(f, "%u:%u", &major, &minor);
  98         if (ret != 2) {
  99                 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs to get major:minor\n",
 100                         __func__);
 101                 fclose(f);
 102                 return -1;
 103         }
 104         fclose(f);
 105
 106         /* create the char device "mknod /dev/uioX c major minor" */
 107         snprintf(filename, sizeof(filename), "/dev/uio%u", uio_num);
 108         dev = makedev(major, minor);
 109         ret = mknod(filename, S_IFCHR | S_IRUSR | S_IWUSR, dev);
 110         if (ret != 0) {
 111                 RTE_LOG(ERR, EAL, "%s(): mknod() failed %s\n",
 112                         __func__, strerror(errno));
 113                 return -1;
 114         }
 115
 116         return ret;
 117 }
 118
 119 /*
 120  * Return the uioX char device used for a pci device. On success, return
 121  * the UIO number and fill dstbuf string with the path of the device in
 122  * sysfs. On error, return a negative value. In this case dstbuf is
 123  * invalid.
 124  */
 125 static int
 126 pci_get_uio_dev(struct rte_pci_device *dev, char *dstbuf,
 127                            unsigned int buflen, int create)
 128 {
 129         struct rte_pci_addr *loc = &dev->addr;
 130         int uio_num = -1;
 131         struct dirent *e;
 132         DIR *dir;
 133         char dirname[PATH_MAX];
 134
 135         /* depending on kernel version, uio can be located in uio/uioX
 136          * or uio:uioX */
 137
 138         snprintf(dirname, sizeof(dirname),
 139                         "%s/" PCI_PRI_FMT "/uio", rte_pci_get_sysfs_path(),
 140                         loc->domain, loc->bus, loc->devid, loc->function);
 141
 142         dir = opendir(dirname);
 143         if (dir == NULL) {
 144                 /* retry with the parent directory */
 145                 snprintf(dirname, sizeof(dirname),
 146                                 "%s/" PCI_PRI_FMT, rte_pci_get_sysfs_path(),
 147                                 loc->domain, loc->bus, loc->devid, loc->function);
 148                 dir = opendir(dirname);
 149
 150                 if (dir == NULL) {
 151                         RTE_LOG(ERR, EAL, "Cannot opendir %s\n", dirname);
 152                         return -1;
 153                 }
 154         }
 155
 156         /* take the first file starting with "uio" */
 157         while ((e = readdir(dir)) != NULL) {
 158                 /* format could be uio%d ...*/
 159                 int shortprefix_len = sizeof("uio") - 1;
 160                 /* ... or uio:uio%d */
 161                 int longprefix_len = sizeof("uio:uio") - 1;
 162                 char *endptr;
 163
 164                 if (strncmp(e->d_name, "uio", 3) != 0)
 165                         continue;
 166
 167                 /* first try uio%d */
 168                 errno = 0;
 169                 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 170                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 171                         snprintf(dstbuf, buflen, "%s/uio%u", dirname, uio_num);
 172                         break;
 173                 }
 174
 175                 /* then try uio:uio%d */
 176                 errno = 0;
 177                 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 178                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 179                         snprintf(dstbuf, buflen, "%s/uio:uio%u", dirname, uio_num);
 180                         break;
 181                 }
 182         }
 183         closedir(dir);
 184
 185         /* No uio resource found */
 186         if (e == NULL)
 187                 return -1;
 188
 189         /* create uio device if we've been asked to */
 190         if (rte_eal_create_uio_dev() && create &&
 191                         pci_mknod_uio_dev(dstbuf, uio_num) < 0)
 192                 RTE_LOG(WARNING, EAL, "Cannot create /dev/uio%u\n", uio_num);
 193
 194         return uio_num;
 195 }
 196
 197 void
 198 pci_uio_free_resource(struct rte_pci_device *dev,
 199                 struct mapped_pci_resource *uio_res)
 200 {
 201         rte_free(uio_res);
 202
 203         if (dev->intr_handle.uio_cfg_fd >= 0) {
 204                 close(dev->intr_handle.uio_cfg_fd);
 205                 dev->intr_handle.uio_cfg_fd = -1;
 206         }
 207         if (dev->intr_handle.fd >= 0) {
 208                 close(dev->intr_handle.fd);
 209                 dev->intr_handle.fd = -1;
 210                 dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
 211         }
 212 }
 213
 214 int
 215 pci_uio_alloc_resource(struct rte_pci_device *dev,
 216                 struct mapped_pci_resource **uio_res)
 217 {
 218         char dirname[PATH_MAX];
 219         char cfgname[PATH_MAX];
 220         char devname[PATH_MAX]; /* contains the /dev/uioX */
 221         int uio_num;
 222         struct rte_pci_addr *loc;
 223
 224         loc = &dev->addr;
 225
 226         /* find uio resource */
 227         uio_num = pci_get_uio_dev(dev, dirname, sizeof(dirname), 1);
 228         if (uio_num < 0) {
 229                 RTE_LOG(WARNING, EAL, "  "PCI_PRI_FMT" not managed by UIO driver, "
 230                                 "skipping\n", loc->domain, loc->bus, loc->devid, loc->function);
 231                 return 1;
 232         }
 233         snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 234
 235         /* save fd if in primary process */
 236         dev->intr_handle.fd = open(devname, O_RDWR);
 237         if (dev->intr_handle.fd < 0) {
 238                 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
 239                         devname, strerror(errno));
 240                 goto error;
 241         }
 242
 243         snprintf(cfgname, sizeof(cfgname),
 244                         "/sys/class/uio/uio%u/device/config", uio_num);
 245         dev->intr_handle.uio_cfg_fd = open(cfgname, O_RDWR);
 246         if (dev->intr_handle.uio_cfg_fd < 0) {
 247                 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
 248                         cfgname, strerror(errno));
 249                 goto error;
 250         }
 251
 252         if (dev->kdrv == RTE_KDRV_IGB_UIO)
 253                 dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
 254         else {
 255                 dev->intr_handle.type = RTE_INTR_HANDLE_UIO_INTX;
 256
 257                 /* set bus master that is not done by uio_pci_generic */
 258                 if (pci_uio_set_bus_master(dev->intr_handle.uio_cfg_fd)) {
 259                         RTE_LOG(ERR, EAL, "Cannot set up bus mastering!\n");
 260                         goto error;
 261                 }
 262         }
 263
 264         /* allocate the mapping details for secondary processes*/
 265         *uio_res = rte_zmalloc("UIO_RES", sizeof(**uio_res), 0);
 266         if (*uio_res == NULL) {
 267                 RTE_LOG(ERR, EAL,
 268                         "%s(): cannot store uio mmap details\n", __func__);
 269                 goto error;
 270         }
 271
 272         strlcpy((*uio_res)->path, devname, sizeof((*uio_res)->path));
 273         memcpy(&(*uio_res)->pci_addr, &dev->addr, sizeof((*uio_res)->pci_addr));
 274
 275         return 0;
 276
 277 error:
 278         pci_uio_free_resource(dev, *uio_res);
 279         return -1;
 280 }
 281
 282 int
 283 pci_uio_map_resource_by_index(struct rte_pci_device *dev, int res_idx,
 284                 struct mapped_pci_resource *uio_res, int map_idx)
 285 {
 286         int fd = -1;
 287         char devname[PATH_MAX];
 288         void *mapaddr;
 289         struct rte_pci_addr *loc;
 290         struct pci_map *maps;
 291         int wc_activate = 0;
 292
 293         if (dev->driver != NULL)
 294                 wc_activate = dev->driver->drv_flags & RTE_PCI_DRV_WC_ACTIVATE;
 295
 296         loc = &dev->addr;
 297         maps = uio_res->maps;
 298
 299         /* allocate memory to keep path */
 300         maps[map_idx].path = rte_malloc(NULL, sizeof(devname), 0);
 301         if (maps[map_idx].path == NULL) {
 302                 RTE_LOG(ERR, EAL, "Cannot allocate memory for path: %s\n",
 303                                 strerror(errno));
 304                 return -1;
 305         }
 306
 307         /*
 308          * open resource file, to mmap it
 309          */
 310         if (wc_activate) {
 311                 /* update devname for mmap  */
 312                 snprintf(devname, sizeof(devname),
 313                         "%s/" PCI_PRI_FMT "/resource%d_wc",
 314                         rte_pci_get_sysfs_path(),
 315                         loc->domain, loc->bus, loc->devid,
 316                         loc->function, res_idx);
 317
 318                 if (access(devname, R_OK|W_OK) != -1) {
 319                         fd = open(devname, O_RDWR);
 320                         if (fd < 0)
 321                                 RTE_LOG(INFO, EAL, "%s cannot be mapped. "
 322                                         "Fall-back to non prefetchable mode.\n",
 323                                         devname);
 324                 }
 325         }
 326
 327         if (!wc_activate || fd < 0) {
 328                 snprintf(devname, sizeof(devname),
 329                         "%s/" PCI_PRI_FMT "/resource%d",
 330                         rte_pci_get_sysfs_path(),
 331                         loc->domain, loc->bus, loc->devid,
 332                         loc->function, res_idx);
 333
 334                 /* then try to map resource file */
 335                 fd = open(devname, O_RDWR);
 336                 if (fd < 0) {
 337                         RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
 338                                 devname, strerror(errno));
 339                         goto error;
 340                 }
 341         }
 342
 343         /* try mapping somewhere close to the end of hugepages */
 344         if (pci_map_addr == NULL)
 345                 pci_map_addr = pci_find_max_end_va();
 346
 347         mapaddr = pci_map_resource(pci_map_addr, fd, 0,
 348                         (size_t)dev->mem_resource[res_idx].len, 0);
 349         close(fd);
 350         if (mapaddr == MAP_FAILED)
 351                 goto error;
 352
 353         pci_map_addr = RTE_PTR_ADD(mapaddr,
 354                         (size_t)dev->mem_resource[res_idx].len);
 355
 356         maps[map_idx].phaddr = dev->mem_resource[res_idx].phys_addr;
 357         maps[map_idx].size = dev->mem_resource[res_idx].len;
 358         maps[map_idx].addr = mapaddr;
 359         maps[map_idx].offset = 0;
 360         strcpy(maps[map_idx].path, devname);
 361         dev->mem_resource[res_idx].addr = mapaddr;
 362
 363         return 0;
 364
 365 error:
 366         rte_free(maps[map_idx].path);
 367         return -1;
 368 }
 369
 370 #if defined(RTE_ARCH_X86)
 371 int
 372 pci_uio_ioport_map(struct rte_pci_device *dev, int bar,
 373                    struct rte_pci_ioport *p)
 374 {
 375         char dirname[PATH_MAX];
 376         char filename[PATH_MAX];
 377         int uio_num;
 378         unsigned long start;
 379
 380         uio_num = pci_get_uio_dev(dev, dirname, sizeof(dirname), 0);
 381         if (uio_num < 0)
 382                 return -1;
 383
 384         /* get portio start */
 385         snprintf(filename, sizeof(filename),
 386                  "%s/portio/port%d/start", dirname, bar);
 387         if (eal_parse_sysfs_value(filename, &start) < 0) {
 388                 RTE_LOG(ERR, EAL, "%s(): cannot parse portio start\n",
 389                         __func__);
 390                 return -1;
 391         }
 392         /* ensure we don't get anything funny here, read/write will cast to
 393          * uin16_t */
 394         if (start > UINT16_MAX)
 395                 return -1;
 396
 397         /* FIXME only for primary process ? */
 398         if (dev->intr_handle.type == RTE_INTR_HANDLE_UNKNOWN) {
 399
 400                 snprintf(filename, sizeof(filename), "/dev/uio%u", uio_num);
 401                 dev->intr_handle.fd = open(filename, O_RDWR);
 402                 if (dev->intr_handle.fd < 0) {
 403                         RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
 404                                 filename, strerror(errno));
 405                         return -1;
 406                 }
 407                 dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
 408         }
 409
 410         RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%lx\n", start);
 411
 412         p->base = start;
 413         p->len = 0;
 414         return 0;
 415 }
 416 #else
 417 int
 418 pci_uio_ioport_map(struct rte_pci_device *dev, int bar,
 419                    struct rte_pci_ioport *p)
 420 {
 421         FILE *f;
 422         char buf[BUFSIZ];
 423         char filename[PATH_MAX];
 424         uint64_t phys_addr, end_addr, flags;
 425         int fd, i;
 426         void *addr;
 427
 428         /* open and read addresses of the corresponding resource in sysfs */
 429         snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT "/resource",
 430                 rte_pci_get_sysfs_path(), dev->addr.domain, dev->addr.bus,
 431                 dev->addr.devid, dev->addr.function);
 432         f = fopen(filename, "r");
 433         if (f == NULL) {
 434                 RTE_LOG(ERR, EAL, "Cannot open sysfs resource: %s\n",
 435                         strerror(errno));
 436                 return -1;
 437         }
 438         for (i = 0; i < bar + 1; i++) {
 439                 if (fgets(buf, sizeof(buf), f) == NULL) {
 440                         RTE_LOG(ERR, EAL, "Cannot read sysfs resource\n");
 441                         goto error;
 442                 }
 443         }
 444         if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
 445                         &end_addr, &flags) < 0)
 446                 goto error;
 447         if ((flags & IORESOURCE_IO) == 0) {
 448                 RTE_LOG(ERR, EAL, "BAR %d is not an IO resource\n", bar);
 449                 goto error;
 450         }
 451         snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT "/resource%d",
 452                 rte_pci_get_sysfs_path(), dev->addr.domain, dev->addr.bus,
 453                 dev->addr.devid, dev->addr.function, bar);
 454
 455         /* mmap the pci resource */
 456         fd = open(filename, O_RDWR);
 457         if (fd < 0) {
 458                 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n", filename,
 459                         strerror(errno));
 460                 goto error;
 461         }
 462         addr = mmap(NULL, end_addr + 1, PROT_READ | PROT_WRITE,
 463                 MAP_SHARED, fd, 0);
 464         close(fd);
 465         if (addr == MAP_FAILED) {
 466                 RTE_LOG(ERR, EAL, "Cannot mmap IO port resource: %s\n",
 467                         strerror(errno));
 468                 goto error;
 469         }
 470
 471         /* strangely, the base address is mmap addr + phys_addr */
 472         p->base = (uintptr_t)addr + phys_addr;
 473         p->len = end_addr + 1;
 474         RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%"PRIx64"\n", p->base);
 475         fclose(f);
 476
 477         return 0;
 478
 479 error:
 480         fclose(f);
 481         return -1;
 482 }
 483 #endif
 484
 485 void
 486 pci_uio_ioport_read(struct rte_pci_ioport *p,
 487                     void *data, size_t len, off_t offset)
 488 {
 489         uint8_t *d;
 490         int size;
 491         uintptr_t reg = p->base + offset;
 492
 493         for (d = data; len > 0; d += size, reg += size, len -= size) {
 494                 if (len >= 4) {
 495                         size = 4;
 496 #if defined(RTE_ARCH_X86)
 497                         *(uint32_t *)d = inl(reg);
 498 #else
 499                         *(uint32_t *)d = *(volatile uint32_t *)reg;
 500 #endif
 501                 } else if (len >= 2) {
 502                         size = 2;
 503 #if defined(RTE_ARCH_X86)
 504                         *(uint16_t *)d = inw(reg);
 505 #else
 506                         *(uint16_t *)d = *(volatile uint16_t *)reg;
 507 #endif
 508                 } else {
 509                         size = 1;
 510 #if defined(RTE_ARCH_X86)
 511                         *d = inb(reg);
 512 #else
 513                         *d = *(volatile uint8_t *)reg;
 514 #endif
 515                 }
 516         }
 517 }
 518
 519 void
 520 pci_uio_ioport_write(struct rte_pci_ioport *p,
 521                      const void *data, size_t len, off_t offset)
 522 {
 523         const uint8_t *s;
 524         int size;
 525         uintptr_t reg = p->base + offset;
 526
 527         for (s = data; len > 0; s += size, reg += size, len -= size) {
 528                 if (len >= 4) {
 529                         size = 4;
 530 #if defined(RTE_ARCH_X86)
 531                         outl_p(*(const uint32_t *)s, reg);
 532 #else
 533                         *(volatile uint32_t *)reg = *(const uint32_t *)s;
 534 #endif
 535                 } else if (len >= 2) {
 536                         size = 2;
 537 #if defined(RTE_ARCH_X86)
 538                         outw_p(*(const uint16_t *)s, reg);
 539 #else
 540                         *(volatile uint16_t *)reg = *(const uint16_t *)s;
 541 #endif
 542                 } else {
 543                         size = 1;
 544 #if defined(RTE_ARCH_X86)
 545                         outb_p(*s, reg);
 546 #else
 547                         *(volatile uint8_t *)reg = *s;
 548 #endif
 549                 }
 550         }
 551 }
 552
 553 int
 554 pci_uio_ioport_unmap(struct rte_pci_ioport *p)
 555 {
 556 #if defined(RTE_ARCH_X86)
 557         RTE_SET_USED(p);
 558         /* FIXME close intr fd ? */
 559         return 0;
 560 #else
 561         return munmap((void *)(uintptr_t)p->base, p->len);
 562 #endif
 563 }