platform-intel.c

   1 /*
   2  * Intel(R) Matrix Storage Manager hardware and firmware support routines
   3  *
   4  * Copyright (C) 2008 Intel Corporation
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms and conditions of the GNU General Public License,
   8  * version 2, as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19 #include "mdadm.h"
  20 #include "platform-intel.h"
  21 #include "probe_roms.h"
  22 #include <stdio.h>
  23 #include <stdlib.h>
  24 #include <string.h>
  25 #include <unistd.h>
  26 #include <dirent.h>
  27 #include <fcntl.h>
  28 #include <sys/mman.h>
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <limits.h>
  32
  33 #define NVME_SUBSYS_PATH "/sys/devices/virtual/nvme-subsystem/"
  34
  35 static int devpath_to_ll(const char *dev_path, const char *entry,
  36                          unsigned long long *val);
  37
  38 static void free_sys_dev(struct sys_dev **list)
  39 {
  40         while (*list) {
  41                 struct sys_dev *next = (*list)->next;
  42
  43                 if ((*list)->path)
  44                         free((*list)->path);
  45                 free(*list);
  46                 *list = next;
  47         }
  48 }
  49
  50 struct sys_dev *find_driver_devices(const char *bus, const char *driver)
  51 {
  52         /* search sysfs for devices driven by 'driver' */
  53         char path[PATH_MAX];
  54         char link[PATH_MAX];
  55         char *c, *p;
  56         DIR *driver_dir;
  57         struct dirent *de;
  58         struct sys_dev *head = NULL;
  59         struct sys_dev *list = NULL;
  60         struct sys_dev *vmd = NULL;
  61         enum sys_dev_type type;
  62         unsigned long long dev_id;
  63         unsigned long long class;
  64
  65         if (strcmp(driver, "isci") == 0)
  66                 type = SYS_DEV_SAS;
  67         else if (strcmp(driver, "ahci") == 0)
  68                 type = SYS_DEV_SATA;
  69         else if (strcmp(driver, "nvme") == 0) {
  70                 /* if looking for nvme devs, first look for vmd */
  71                 vmd = find_driver_devices("pci", "vmd");
  72                 type = SYS_DEV_NVME;
  73         } else if (strcmp(driver, "vmd") == 0)
  74                 type = SYS_DEV_VMD;
  75         else
  76                 type = SYS_DEV_UNKNOWN;
  77
  78         sprintf(path, "/sys/bus/%s/drivers/%s", bus, driver);
  79         driver_dir = opendir(path);
  80         if (!driver_dir) {
  81                 if (vmd)
  82                         free_sys_dev(&vmd);
  83                 return NULL;
  84         }
  85         for (de = readdir(driver_dir); de; de = readdir(driver_dir)) {
  86                 int n;
  87                 int skip = 0;
  88
  89                 /* is 'de' a device? check that the 'subsystem' link exists and
  90                  * that its target matches 'bus'
  91                  */
  92                 sprintf(path, "/sys/bus/%s/drivers/%s/%s/subsystem",
  93                         bus, driver, de->d_name);
  94                 n = readlink(path, link, sizeof(link));
  95                 if (n < 0 || n >= (int)sizeof(link))
  96                         continue;
  97                 link[n] = '\0';
  98                 c = strrchr(link, '/');
  99                 if (!c)
 100                         continue;
 101                 if (strncmp(bus, c+1, strlen(bus)) != 0)
 102                         continue;
 103
 104                 sprintf(path, "/sys/bus/%s/drivers/%s/%s",
 105                         bus, driver, de->d_name);
 106
 107                 /* if searching for nvme - skip vmd connected one */
 108                 if (type == SYS_DEV_NVME) {
 109                         struct sys_dev *dev;
 110                         char *rp = realpath(path, NULL);
 111                         for (dev = vmd; dev; dev = dev->next) {
 112                                 if ((strncmp(dev->path, rp, strlen(dev->path)) == 0))
 113                                         skip = 1;
 114                         }
 115                         free(rp);
 116                 }
 117
 118                 /* if it's not Intel device or mark as VMD connected - skip it. */
 119                 if (devpath_to_vendor(path) != 0x8086 || skip == 1)
 120                         continue;
 121
 122                 if (devpath_to_ll(path, "device", &dev_id) != 0)
 123                         continue;
 124
 125                 if (devpath_to_ll(path, "class", &class) != 0)
 126                         continue;
 127
 128                 /*
 129                  * Each VMD device (domain) adds separate PCI bus, it is better
 130                  * to store path as a path to that bus (easier further
 131                  * determination which NVMe dev is connected to this particular
 132                  * VMD domain).
 133                  */
 134                 if (type == SYS_DEV_VMD) {
 135                         sprintf(path, "/sys/bus/%s/drivers/%s/%s/domain/device",
 136                                 bus, driver, de->d_name);
 137                 }
 138                 p = realpath(path, NULL);
 139                 if (p == NULL) {
 140                         pr_err("Unable to get real path for '%s'\n", path);
 141                         continue;
 142                 }
 143
 144                 /* start / add list entry */
 145                 if (!head) {
 146                         head = xmalloc(sizeof(*head));
 147                         list = head;
 148                 } else {
 149                         list->next = xmalloc(sizeof(*head));
 150                         list = list->next;
 151                 }
 152
 153                 if (!list) {
 154                         free_sys_dev(&head);
 155                         break;
 156                 }
 157
 158                 list->dev_id = (__u16) dev_id;
 159                 list->class = (__u32) class;
 160                 list->type = type;
 161                 list->next = NULL;
 162                 list->path = p;
 163
 164                 if ((list->pci_id = strrchr(list->path, '/')) != NULL)
 165                         list->pci_id++;
 166         }
 167         closedir(driver_dir);
 168
 169         if (vmd) {
 170                 if (list)
 171                         list->next = vmd;
 172                 else
 173                         head = vmd;
 174         }
 175
 176         return head;
 177 }
 178
 179 static struct sys_dev *intel_devices=NULL;
 180 static time_t valid_time = 0;
 181
 182 struct sys_dev *device_by_id(__u16 device_id)
 183 {
 184         struct sys_dev *iter;
 185
 186         for (iter = intel_devices; iter != NULL; iter = iter->next)
 187                 if (iter->dev_id == device_id)
 188                         return iter;
 189         return NULL;
 190 }
 191
 192 struct sys_dev *device_by_id_and_path(__u16 device_id, const char *path)
 193 {
 194         struct sys_dev *iter;
 195
 196         for (iter = intel_devices; iter != NULL; iter = iter->next)
 197                 if ((iter->dev_id == device_id) && strstr(iter->path, path))
 198                         return iter;
 199         return NULL;
 200 }
 201
 202 static int devpath_to_ll(const char *dev_path, const char *entry, unsigned long long *val)
 203 {
 204         char path[strlen(dev_path) + strlen(entry) + 2];
 205         int fd;
 206         int n;
 207
 208         sprintf(path, "%s/%s", dev_path, entry);
 209
 210         fd = open(path, O_RDONLY);
 211         if (fd < 0)
 212                 return -1;
 213         n = sysfs_fd_get_ll(fd, val);
 214         close(fd);
 215         return n;
 216 }
 217
 218 __u16 devpath_to_vendor(const char *dev_path)
 219 {
 220         char path[strlen(dev_path) + strlen("/vendor") + 1];
 221         char vendor[7];
 222         int fd;
 223         __u16 id = 0xffff;
 224         int n;
 225
 226         sprintf(path, "%s/vendor", dev_path);
 227
 228         fd = open(path, O_RDONLY);
 229         if (fd < 0)
 230                 return 0xffff;
 231
 232         n = read(fd, vendor, sizeof(vendor));
 233         if (n == sizeof(vendor)) {
 234                 vendor[n - 1] = '\0';
 235                 id = strtoul(vendor, NULL, 16);
 236         }
 237         close(fd);
 238
 239         return id;
 240 }
 241
 242 struct sys_dev *find_intel_devices(void)
 243 {
 244         struct sys_dev *ahci, *isci, *nvme;
 245
 246         if (valid_time > time(0) - 10)
 247                 return intel_devices;
 248
 249         if (intel_devices)
 250                 free_sys_dev(&intel_devices);
 251
 252         isci = find_driver_devices("pci", "isci");
 253         ahci = find_driver_devices("pci", "ahci");
 254         /* Searching for NVMe will return list of NVMe and VMD controllers */
 255         nvme = find_driver_devices("pci", "nvme");
 256
 257         if (!isci && !ahci) {
 258                 ahci = nvme;
 259         } else if (!ahci) {
 260                 ahci = isci;
 261                 struct sys_dev *elem = ahci;
 262                 while (elem->next)
 263                         elem = elem->next;
 264                 elem->next = nvme;
 265         } else {
 266                 struct sys_dev *elem = ahci;
 267                 while (elem->next)
 268                         elem = elem->next;
 269                 elem->next = isci;
 270                 while (elem->next)
 271                         elem = elem->next;
 272                 elem->next = nvme;
 273         }
 274         intel_devices = ahci;
 275         valid_time = time(0);
 276         return intel_devices;
 277 }
 278
 279 /*
 280  * PCI Expansion ROM Data Structure Format */
 281 struct pciExpDataStructFormat {
 282         __u8  ver[4];
 283         __u16 vendorID;
 284         __u16 deviceID;
 285         __u16 devListOffset;
 286         __u16 pciDataStructLen;
 287         __u8 pciDataStructRev;
 288 } __attribute__ ((packed));
 289
 290 struct orom_entry *orom_entries;
 291
 292 const struct orom_entry *get_orom_entry_by_device_id(__u16 dev_id)
 293 {
 294         struct orom_entry *entry;
 295         struct devid_list *devid;
 296
 297         for (entry = orom_entries; entry; entry = entry->next) {
 298                 for (devid = entry->devid_list; devid; devid = devid->next) {
 299                         if (devid->devid == dev_id)
 300                                 return entry;
 301                 }
 302         }
 303
 304         return NULL;
 305 }
 306
 307 const struct imsm_orom *get_orom_by_device_id(__u16 dev_id)
 308 {
 309         const struct orom_entry *entry = get_orom_entry_by_device_id(dev_id);
 310
 311         if (entry)
 312                 return &entry->orom;
 313
 314         return NULL;
 315 }
 316
 317 static struct orom_entry *add_orom(const struct imsm_orom *orom)
 318 {
 319         struct orom_entry *list;
 320         struct orom_entry *prev = NULL;
 321
 322         for (list = orom_entries; list; prev = list, list = list->next)
 323                 ;
 324
 325         list = xmalloc(sizeof(struct orom_entry));
 326         list->orom = *orom;
 327         list->devid_list = NULL;
 328         list->next = NULL;
 329
 330         if (prev == NULL)
 331                 orom_entries = list;
 332         else
 333                 prev->next = list;
 334
 335         return list;
 336 }
 337
 338 static void add_orom_device_id(struct orom_entry *entry, __u16 dev_id)
 339 {
 340         struct devid_list *list;
 341         struct devid_list *prev = NULL;
 342
 343         for (list = entry->devid_list; list; prev = list, list = list->next) {
 344                 if (list->devid == dev_id)
 345                         return;
 346         }
 347         list = xmalloc(sizeof(struct devid_list));
 348         list->devid = dev_id;
 349         list->next = NULL;
 350
 351         if (prev == NULL)
 352                 entry->devid_list = list;
 353         else
 354                 prev->next = list;
 355 }
 356
 357 static int scan(const void *start, const void *end, const void *data)
 358 {
 359         int offset;
 360         const struct imsm_orom *imsm_mem = NULL;
 361         int len = (end - start);
 362         struct pciExpDataStructFormat *ptr= (struct pciExpDataStructFormat *)data;
 363
 364         if (data + 0x18 > end) {
 365                 dprintf("cannot find pciExpDataStruct \n");
 366                 return 0;
 367         }
 368
 369         dprintf("ptr->vendorID: %lx __le16_to_cpu(ptr->deviceID): %lx \n",
 370                 (ulong) __le16_to_cpu(ptr->vendorID),
 371                 (ulong) __le16_to_cpu(ptr->deviceID));
 372
 373         if (__le16_to_cpu(ptr->vendorID) != 0x8086)
 374                 return 0;
 375
 376         if (get_orom_by_device_id(ptr->deviceID))
 377                 return 0;
 378
 379         for (offset = 0; offset < len; offset += 4) {
 380                 const void *mem = start + offset;
 381
 382                 if ((memcmp(mem, IMSM_OROM_SIGNATURE, 4) == 0)) {
 383                         imsm_mem = mem;
 384                         break;
 385                 }
 386         }
 387
 388         if (!imsm_mem)
 389                 return 0;
 390
 391         struct orom_entry *orom = add_orom(imsm_mem);
 392
 393         /* only PciDataStructure with revision 3 and above supports devices list. */
 394         if (ptr->pciDataStructRev >= 3 && ptr->devListOffset) {
 395                 const __u16 *dev_list = (void *)ptr + ptr->devListOffset;
 396                 int i;
 397
 398                 for (i = 0; dev_list[i] != 0; i++)
 399                         add_orom_device_id(orom, dev_list[i]);
 400         } else {
 401                 add_orom_device_id(orom, __le16_to_cpu(ptr->deviceID));
 402         }
 403
 404         return 0;
 405 }
 406
 407 const struct imsm_orom *imsm_platform_test(struct sys_dev *hba)
 408 {
 409         struct imsm_orom orom = {
 410                 .signature = IMSM_OROM_SIGNATURE,
 411                 .rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 412                                         IMSM_OROM_RLC_RAID10 | IMSM_OROM_RLC_RAID5,
 413                 .sss = IMSM_OROM_SSS_4kB | IMSM_OROM_SSS_8kB |
 414                                         IMSM_OROM_SSS_16kB | IMSM_OROM_SSS_32kB |
 415                                         IMSM_OROM_SSS_64kB | IMSM_OROM_SSS_128kB |
 416                                         IMSM_OROM_SSS_256kB | IMSM_OROM_SSS_512kB |
 417                                         IMSM_OROM_SSS_1MB | IMSM_OROM_SSS_2MB,
 418                 .dpa = IMSM_OROM_DISKS_PER_ARRAY,
 419                 .tds = IMSM_OROM_TOTAL_DISKS,
 420                 .vpa = IMSM_OROM_VOLUMES_PER_ARRAY,
 421                 .vphba = IMSM_OROM_VOLUMES_PER_HBA
 422         };
 423         orom.attr = orom.rlc | IMSM_OROM_ATTR_ChecksumVerify;
 424
 425         if (check_env("IMSM_TEST_OROM_NORAID5")) {
 426                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 427                                 IMSM_OROM_RLC_RAID10;
 428         }
 429         if (check_env("IMSM_TEST_AHCI_EFI_NORAID5") && (hba->type == SYS_DEV_SAS)) {
 430                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 431                                 IMSM_OROM_RLC_RAID10;
 432         }
 433         if (check_env("IMSM_TEST_SCU_EFI_NORAID5") && (hba->type == SYS_DEV_SATA)) {
 434                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 435                                 IMSM_OROM_RLC_RAID10;
 436         }
 437
 438         struct orom_entry *ret = add_orom(&orom);
 439
 440         add_orom_device_id(ret, hba->dev_id);
 441
 442         return &ret->orom;
 443 }
 444
 445 static const struct imsm_orom *find_imsm_hba_orom(struct sys_dev *hba)
 446 {
 447         unsigned long align;
 448
 449         if (check_env("IMSM_TEST_OROM"))
 450                 return imsm_platform_test(hba);
 451
 452         /* return empty OROM capabilities in EFI test mode */
 453         if (check_env("IMSM_TEST_AHCI_EFI") || check_env("IMSM_TEST_SCU_EFI"))
 454                 return NULL;
 455
 456         find_intel_devices();
 457
 458         if (intel_devices == NULL)
 459                 return NULL;
 460
 461         /* scan option-rom memory looking for an imsm signature */
 462         if (check_env("IMSM_SAFE_OROM_SCAN"))
 463                 align = 2048;
 464         else
 465                 align = 512;
 466         if (probe_roms_init(align) != 0)
 467                 return NULL;
 468         probe_roms();
 469         /* ignore return value - True is returned if both adapater roms are found */
 470         scan_adapter_roms(scan);
 471         probe_roms_exit();
 472
 473         return get_orom_by_device_id(hba->dev_id);
 474 }
 475
 476 #define GUID_STR_MAX    37  /* according to GUID format:
 477                              * xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" */
 478
 479 #define EFI_GUID(a, b, c, d0, d1, d2, d3, d4, d5, d6, d7) \
 480 ((struct efi_guid) \
 481 {{ (a) & 0xff, ((a) >> 8) & 0xff, ((a) >> 16) & 0xff, ((a) >> 24) & 0xff, \
 482   (b) & 0xff, ((b) >> 8) & 0xff, \
 483   (c) & 0xff, ((c) >> 8) & 0xff, \
 484   (d0), (d1), (d2), (d3), (d4), (d5), (d6), (d7) }})
 485
 486 #define SYS_EFI_VAR_PATH "/sys/firmware/efi/vars"
 487 #define SYS_EFIVARS_PATH "/sys/firmware/efi/efivars"
 488 #define SCU_PROP "RstScuV"
 489 #define AHCI_PROP "RstSataV"
 490 #define AHCI_SSATA_PROP "RstsSatV"
 491 #define AHCI_CSATA_PROP "RstCSatV"
 492 #define VMD_PROP "RstUefiV"
 493
 494 #define VENDOR_GUID \
 495         EFI_GUID(0x193dfefa, 0xa445, 0x4302, 0x99, 0xd8, 0xef, 0x3a, 0xad, 0x1a, 0x04, 0xc6)
 496
 497 #define PCI_CLASS_RAID_CNTRL 0x010400
 498
 499 static int read_efi_var(void *buffer, ssize_t buf_size, char *variable_name, struct efi_guid guid)
 500 {
 501         char path[PATH_MAX];
 502         char buf[GUID_STR_MAX];
 503         int fd;
 504         ssize_t n;
 505
 506         snprintf(path, PATH_MAX, "%s/%s-%s", SYS_EFIVARS_PATH, variable_name, guid_str(buf, guid));
 507
 508         fd = open(path, O_RDONLY);
 509         if (fd < 0)
 510                 return 1;
 511
 512         /* read the variable attributes and ignore it */
 513         n = read(fd, buf, sizeof(__u32));
 514         if (n < 0) {
 515                 close(fd);
 516                 return 1;
 517         }
 518
 519         /* read the variable data */
 520         n = read(fd, buffer, buf_size);
 521         close(fd);
 522         if (n < buf_size)
 523                 return 1;
 524
 525         return 0;
 526 }
 527
 528 static int read_efi_variable(void *buffer, ssize_t buf_size, char *variable_name, struct efi_guid guid)
 529 {
 530         char path[PATH_MAX];
 531         char buf[GUID_STR_MAX];
 532         int dfd;
 533         ssize_t n, var_data_len;
 534
 535         /* Try to read the variable using the new efivarfs interface first.
 536          * If that fails, fall back to the old sysfs-efivars interface. */
 537         if (!read_efi_var(buffer, buf_size, variable_name, guid))
 538                 return 0;
 539
 540         snprintf(path, PATH_MAX, "%s/%s-%s/size", SYS_EFI_VAR_PATH, variable_name, guid_str(buf, guid));
 541
 542         dprintf("EFI VAR: path=%s\n", path);
 543         /* get size of variable data */
 544         dfd = open(path, O_RDONLY);
 545         if (dfd < 0)
 546                 return 1;
 547
 548         n = read(dfd, &buf, sizeof(buf));
 549         close(dfd);
 550         if (n < 0)
 551                 return 1;
 552         buf[n] = '\0';
 553
 554         errno = 0;
 555         var_data_len = strtoul(buf, NULL, 16);
 556         if ((errno == ERANGE && (var_data_len == LONG_MAX)) ||
 557             (errno != 0 && var_data_len == 0))
 558                 return 1;
 559
 560         /* get data */
 561         snprintf(path, PATH_MAX, "%s/%s-%s/data", SYS_EFI_VAR_PATH, variable_name, guid_str(buf, guid));
 562
 563         dprintf("EFI VAR: path=%s\n", path);
 564         dfd = open(path, O_RDONLY);
 565         if (dfd < 0)
 566                 return 1;
 567
 568         n = read(dfd, buffer, buf_size);
 569         close(dfd);
 570         if (n != var_data_len || n < buf_size) {
 571                 return 1;
 572         }
 573
 574         return 0;
 575 }
 576
 577 const struct imsm_orom *find_imsm_efi(struct sys_dev *hba)
 578 {
 579         struct imsm_orom orom;
 580         struct orom_entry *ret;
 581         int err;
 582
 583         if (check_env("IMSM_TEST_AHCI_EFI") || check_env("IMSM_TEST_SCU_EFI"))
 584                 return imsm_platform_test(hba);
 585
 586         /* OROM test is set, return that there is no EFI capabilities */
 587         if (check_env("IMSM_TEST_OROM"))
 588                 return NULL;
 589
 590         if (hba->type == SYS_DEV_SATA && hba->class != PCI_CLASS_RAID_CNTRL)
 591                 return NULL;
 592
 593         err = read_efi_variable(&orom, sizeof(orom), hba->type == SYS_DEV_SAS ? SCU_PROP : AHCI_PROP, VENDOR_GUID);
 594
 595         /* try to read variable for second AHCI controller */
 596         if (err && hba->type == SYS_DEV_SATA)
 597                 err = read_efi_variable(&orom, sizeof(orom), AHCI_SSATA_PROP, VENDOR_GUID);
 598
 599         /* try to read variable for combined AHCI controllers */
 600         if (err && hba->type == SYS_DEV_SATA) {
 601                 static struct orom_entry *csata;
 602
 603                 err = read_efi_variable(&orom, sizeof(orom), AHCI_CSATA_PROP, VENDOR_GUID);
 604                 if (!err) {
 605                         if (!csata)
 606                                 csata = add_orom(&orom);
 607                         add_orom_device_id(csata, hba->dev_id);
 608                         csata->type = hba->type;
 609                         return &csata->orom;
 610                 }
 611         }
 612
 613         if (hba->type == SYS_DEV_VMD) {
 614                 err = read_efi_variable(&orom, sizeof(orom), VMD_PROP, VENDOR_GUID);
 615         }
 616
 617         if (err)
 618                 return NULL;
 619
 620         ret = add_orom(&orom);
 621         add_orom_device_id(ret, hba->dev_id);
 622         ret->type = hba->type;
 623
 624         return &ret->orom;
 625 }
 626
 627 const struct imsm_orom *find_imsm_nvme(struct sys_dev *hba)
 628 {
 629         static struct orom_entry *nvme_orom;
 630
 631         if (hba->type != SYS_DEV_NVME)
 632                 return NULL;
 633
 634         if (!nvme_orom) {
 635                 struct imsm_orom nvme_orom_compat = {
 636                         .signature = IMSM_NVME_OROM_COMPAT_SIGNATURE,
 637                         .rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 638                                                 IMSM_OROM_RLC_RAID10 | IMSM_OROM_RLC_RAID5,
 639                         .sss = IMSM_OROM_SSS_4kB | IMSM_OROM_SSS_8kB |
 640                                                 IMSM_OROM_SSS_16kB | IMSM_OROM_SSS_32kB |
 641                                                 IMSM_OROM_SSS_64kB | IMSM_OROM_SSS_128kB,
 642                         .dpa = IMSM_OROM_DISKS_PER_ARRAY_NVME,
 643                         .tds = IMSM_OROM_TOTAL_DISKS_NVME,
 644                         .vpa = IMSM_OROM_VOLUMES_PER_ARRAY,
 645                         .vphba = IMSM_OROM_TOTAL_DISKS_NVME / 2 * IMSM_OROM_VOLUMES_PER_ARRAY,
 646                         .attr = IMSM_OROM_ATTR_2TB | IMSM_OROM_ATTR_2TB_DISK,
 647                         .driver_features = IMSM_OROM_CAPABILITIES_EnterpriseSystem
 648                 };
 649                 nvme_orom = add_orom(&nvme_orom_compat);
 650         }
 651         add_orom_device_id(nvme_orom, hba->dev_id);
 652         nvme_orom->type = SYS_DEV_NVME;
 653         return &nvme_orom->orom;
 654 }
 655
 656 const struct imsm_orom *find_imsm_capability(struct sys_dev *hba)
 657 {
 658         const struct imsm_orom *cap = get_orom_by_device_id(hba->dev_id);
 659
 660         if (cap)
 661                 return cap;
 662
 663         if (hba->type == SYS_DEV_NVME)
 664                 return find_imsm_nvme(hba);
 665         if ((cap = find_imsm_efi(hba)) != NULL)
 666                 return cap;
 667         if ((cap = find_imsm_hba_orom(hba)) != NULL)
 668                 return cap;
 669
 670         return NULL;
 671 }
 672
 673 /* Check whether the nvme device is represented by nvme subsytem,
 674  * if yes virtual path should be changed to hardware device path,
 675  * to allow IMSM capabilities detection.
 676  * Returns:
 677  *      hardware path to device - if the device is represented via
 678  *              nvme virtual subsytem
 679  *      NULL - if the device is not represented via nvme virtual subsytem
 680  */
 681 char *get_nvme_multipath_dev_hw_path(const char *dev_path)
 682 {
 683         DIR *dir;
 684         struct dirent *ent;
 685         char *rp = NULL;
 686
 687         if (strncmp(dev_path, NVME_SUBSYS_PATH, strlen(NVME_SUBSYS_PATH)) != 0)
 688                 return NULL;
 689
 690         dir = opendir(dev_path);
 691         if (!dir)
 692                 return NULL;
 693
 694         for (ent = readdir(dir); ent; ent = readdir(dir)) {
 695                 char buf[strlen(dev_path) + strlen(ent->d_name) + 1];
 696
 697                 /* Check if dir is a controller, ignore namespaces*/
 698                 if (!(strncmp(ent->d_name, "nvme", 4) == 0) ||
 699                     (strrchr(ent->d_name, 'n') != &ent->d_name[0]))
 700                         continue;
 701
 702                 sprintf(buf, "%s/%s", dev_path, ent->d_name);
 703                 rp = realpath(buf, NULL);
 704                 break;
 705         }
 706
 707         closedir(dir);
 708         return rp;
 709 }
 710
 711 char *devt_to_devpath(dev_t dev)
 712 {
 713         char device[46];
 714         char *rp;
 715         char *buf;
 716
 717         sprintf(device, "/sys/dev/block/%d:%d/device", major(dev), minor(dev));
 718
 719         rp = realpath(device, NULL);
 720         if (!rp)
 721                 return NULL;
 722
 723         buf = get_nvme_multipath_dev_hw_path(rp);
 724         if (buf) {
 725                 free(rp);
 726                 return buf;
 727         }
 728
 729         return rp;
 730 }
 731
 732 char *diskfd_to_devpath(int fd)
 733 {
 734         /* return the device path for a disk, return NULL on error or fd
 735          * refers to a partition
 736          */
 737         struct stat st;
 738
 739         if (fstat(fd, &st) != 0)
 740                 return NULL;
 741         if (!S_ISBLK(st.st_mode))
 742                 return NULL;
 743
 744         return devt_to_devpath(st.st_rdev);
 745 }
 746
 747 int path_attached_to_hba(const char *disk_path, const char *hba_path)
 748 {
 749         int rc;
 750
 751         if (check_env("IMSM_TEST_AHCI_DEV") ||
 752             check_env("IMSM_TEST_SCU_DEV")) {
 753                 return 1;
 754         }
 755
 756         if (!disk_path || !hba_path)
 757                 return 0;
 758         dprintf("hba: %s - disk: %s\n", hba_path, disk_path);
 759         if (strncmp(disk_path, hba_path, strlen(hba_path)) == 0)
 760                 rc = 1;
 761         else
 762                 rc = 0;
 763
 764         return rc;
 765 }
 766
 767 int devt_attached_to_hba(dev_t dev, const char *hba_path)
 768 {
 769         char *disk_path = devt_to_devpath(dev);
 770         int rc = path_attached_to_hba(disk_path, hba_path);
 771
 772         if (disk_path)
 773                 free(disk_path);
 774
 775         return rc;
 776 }
 777
 778 int disk_attached_to_hba(int fd, const char *hba_path)
 779 {
 780         char *disk_path = diskfd_to_devpath(fd);
 781         int rc = path_attached_to_hba(disk_path, hba_path);
 782
 783         if (disk_path)
 784                 free(disk_path);
 785
 786         return rc;
 787 }
 788
 789 char *vmd_domain_to_controller(struct sys_dev *hba, char *buf)
 790 {
 791         struct dirent *ent;
 792         DIR *dir;
 793         char path[PATH_MAX];
 794
 795         if (!hba)
 796                 return NULL;
 797
 798         if (hba->type != SYS_DEV_VMD)
 799                 return NULL;
 800
 801         dir = opendir("/sys/bus/pci/drivers/vmd");
 802         if (!dir)
 803                 return NULL;
 804
 805         for (ent = readdir(dir); ent; ent = readdir(dir)) {
 806                 sprintf(path, "/sys/bus/pci/drivers/vmd/%s/domain/device",
 807                         ent->d_name);
 808
 809                 if (!realpath(path, buf))
 810                         continue;
 811
 812                 if (strncmp(buf, hba->path, strlen(buf)) == 0) {
 813                         sprintf(path, "/sys/bus/pci/drivers/vmd/%s", ent->d_name);
 814                         closedir(dir);
 815                         return realpath(path, buf);
 816                 }
 817         }
 818
 819         closedir(dir);
 820         return NULL;
 821 }
 822 /* Verify that NVMe drive is supported by IMSM
 823  * Returns:
 824  *      0 - not supported
 825  *      1 - supported
 826  */
 827 int imsm_is_nvme_supported(int disk_fd, int verbose)
 828 {
 829         char nsid_path[PATH_MAX];
 830         char buf[PATH_MAX];
 831         struct stat stb;
 832
 833         if (disk_fd < 0)
 834                 return 0;
 835
 836         if (fstat(disk_fd, &stb))
 837                 return 0;
 838
 839         snprintf(nsid_path, PATH_MAX-1, "/sys/dev/block/%d:%d/nsid",
 840                  major(stb.st_rdev), minor(stb.st_rdev));
 841
 842         if (load_sys(nsid_path, buf, sizeof(buf))) {
 843                 pr_err("Cannot read %s, rejecting drive\n", nsid_path);
 844                 return 0;
 845         }
 846         if (strtoll(buf, NULL, 10) != 1) {
 847                 if (verbose)
 848                         pr_err("Only first namespace is supported by IMSM, aborting\n");
 849                 return 0;
 850         }
 851         return 1;
 852 }
 853
 854 /* Verify if multipath is supported by NVMe controller
 855  * Returns:
 856  *      0 - not supported
 857  *      1 - supported
 858  */
 859 int is_multipath_nvme(int disk_fd)
 860 {
 861         char path_buf[PATH_MAX];
 862         char ns_path[PATH_MAX];
 863         char *kname = fd2kname(disk_fd);
 864
 865         if (!kname)
 866                 return 0;
 867         sprintf(path_buf, "/sys/block/%s", kname);
 868
 869         if (!realpath(path_buf, ns_path))
 870                 return 0;
 871
 872         if (strncmp(ns_path, NVME_SUBSYS_PATH, strlen(NVME_SUBSYS_PATH)) == 0)
 873                 return 1;
 874
 875         return 0;
 876 }