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 /* Description: Read text value of dev_path/entry field
 243  * Parameters:
 244  *      dev_path - sysfs path to the device
 245  *      entry - entry to be read
 246  *      buf - buffer for read value
 247  *      len - size of buf
 248  *      verbose - error logging level
 249  */
 250 int devpath_to_char(const char *dev_path, const char *entry, char *buf, int len,
 251                     int verbose)
 252 {
 253         char path[PATH_MAX];
 254
 255         snprintf(path, sizeof(path), "%s/%s", dev_path, entry);
 256         if (load_sys(path, buf, len)) {
 257                 if (verbose)
 258                         pr_err("Cannot read %s, aborting\n", path);
 259                 return 1;
 260         }
 261
 262         return 0;
 263 }
 264
 265 struct sys_dev *find_intel_devices(void)
 266 {
 267         struct sys_dev *ahci, *isci, *nvme;
 268
 269         if (valid_time > time(0) - 10)
 270                 return intel_devices;
 271
 272         if (intel_devices)
 273                 free_sys_dev(&intel_devices);
 274
 275         isci = find_driver_devices("pci", "isci");
 276         ahci = find_driver_devices("pci", "ahci");
 277         /* Searching for NVMe will return list of NVMe and VMD controllers */
 278         nvme = find_driver_devices("pci", "nvme");
 279
 280         if (!isci && !ahci) {
 281                 ahci = nvme;
 282         } else if (!ahci) {
 283                 ahci = isci;
 284                 struct sys_dev *elem = ahci;
 285                 while (elem->next)
 286                         elem = elem->next;
 287                 elem->next = nvme;
 288         } else {
 289                 struct sys_dev *elem = ahci;
 290                 while (elem->next)
 291                         elem = elem->next;
 292                 elem->next = isci;
 293                 while (elem->next)
 294                         elem = elem->next;
 295                 elem->next = nvme;
 296         }
 297         intel_devices = ahci;
 298         valid_time = time(0);
 299         return intel_devices;
 300 }
 301
 302 /*
 303  * PCI Expansion ROM Data Structure Format */
 304 struct pciExpDataStructFormat {
 305         __u8  ver[4];
 306         __u16 vendorID;
 307         __u16 deviceID;
 308         __u16 devListOffset;
 309         __u16 pciDataStructLen;
 310         __u8 pciDataStructRev;
 311 } __attribute__ ((packed));
 312
 313 struct orom_entry *orom_entries;
 314
 315 const struct orom_entry *get_orom_entry_by_device_id(__u16 dev_id)
 316 {
 317         struct orom_entry *entry;
 318         struct devid_list *devid;
 319
 320         for (entry = orom_entries; entry; entry = entry->next) {
 321                 for (devid = entry->devid_list; devid; devid = devid->next) {
 322                         if (devid->devid == dev_id)
 323                                 return entry;
 324                 }
 325         }
 326
 327         return NULL;
 328 }
 329
 330 const struct imsm_orom *get_orom_by_device_id(__u16 dev_id)
 331 {
 332         const struct orom_entry *entry = get_orom_entry_by_device_id(dev_id);
 333
 334         if (entry)
 335                 return &entry->orom;
 336
 337         return NULL;
 338 }
 339
 340 static struct orom_entry *add_orom(const struct imsm_orom *orom)
 341 {
 342         struct orom_entry *list;
 343         struct orom_entry *prev = NULL;
 344
 345         for (list = orom_entries; list; prev = list, list = list->next)
 346                 ;
 347
 348         list = xmalloc(sizeof(struct orom_entry));
 349         list->orom = *orom;
 350         list->devid_list = NULL;
 351         list->next = NULL;
 352
 353         if (prev == NULL)
 354                 orom_entries = list;
 355         else
 356                 prev->next = list;
 357
 358         return list;
 359 }
 360
 361 static void add_orom_device_id(struct orom_entry *entry, __u16 dev_id)
 362 {
 363         struct devid_list *list;
 364         struct devid_list *prev = NULL;
 365
 366         for (list = entry->devid_list; list; prev = list, list = list->next) {
 367                 if (list->devid == dev_id)
 368                         return;
 369         }
 370         list = xmalloc(sizeof(struct devid_list));
 371         list->devid = dev_id;
 372         list->next = NULL;
 373
 374         if (prev == NULL)
 375                 entry->devid_list = list;
 376         else
 377                 prev->next = list;
 378 }
 379
 380 static int scan(const void *start, const void *end, const void *data)
 381 {
 382         int offset;
 383         const struct imsm_orom *imsm_mem = NULL;
 384         int len = (end - start);
 385         struct pciExpDataStructFormat *ptr= (struct pciExpDataStructFormat *)data;
 386
 387         if (data + 0x18 > end) {
 388                 dprintf("cannot find pciExpDataStruct \n");
 389                 return 0;
 390         }
 391
 392         dprintf("ptr->vendorID: %lx __le16_to_cpu(ptr->deviceID): %lx \n",
 393                 (ulong) __le16_to_cpu(ptr->vendorID),
 394                 (ulong) __le16_to_cpu(ptr->deviceID));
 395
 396         if (__le16_to_cpu(ptr->vendorID) != 0x8086)
 397                 return 0;
 398
 399         if (get_orom_by_device_id(ptr->deviceID))
 400                 return 0;
 401
 402         for (offset = 0; offset < len; offset += 4) {
 403                 const void *mem = start + offset;
 404
 405                 if ((memcmp(mem, IMSM_OROM_SIGNATURE, 4) == 0)) {
 406                         imsm_mem = mem;
 407                         break;
 408                 }
 409         }
 410
 411         if (!imsm_mem)
 412                 return 0;
 413
 414         struct orom_entry *orom = add_orom(imsm_mem);
 415
 416         /* only PciDataStructure with revision 3 and above supports devices list. */
 417         if (ptr->pciDataStructRev >= 3 && ptr->devListOffset) {
 418                 const __u16 *dev_list = (void *)ptr + ptr->devListOffset;
 419                 int i;
 420
 421                 for (i = 0; dev_list[i] != 0; i++)
 422                         add_orom_device_id(orom, dev_list[i]);
 423         } else {
 424                 add_orom_device_id(orom, __le16_to_cpu(ptr->deviceID));
 425         }
 426
 427         return 0;
 428 }
 429
 430 const struct imsm_orom *imsm_platform_test(struct sys_dev *hba)
 431 {
 432         struct imsm_orom orom = {
 433                 .signature = IMSM_OROM_SIGNATURE,
 434                 .rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 435                                         IMSM_OROM_RLC_RAID10 | IMSM_OROM_RLC_RAID5,
 436                 .sss = IMSM_OROM_SSS_4kB | IMSM_OROM_SSS_8kB |
 437                                         IMSM_OROM_SSS_16kB | IMSM_OROM_SSS_32kB |
 438                                         IMSM_OROM_SSS_64kB | IMSM_OROM_SSS_128kB |
 439                                         IMSM_OROM_SSS_256kB | IMSM_OROM_SSS_512kB |
 440                                         IMSM_OROM_SSS_1MB | IMSM_OROM_SSS_2MB,
 441                 .dpa = IMSM_OROM_DISKS_PER_ARRAY,
 442                 .tds = IMSM_OROM_TOTAL_DISKS,
 443                 .vpa = IMSM_OROM_VOLUMES_PER_ARRAY,
 444                 .vphba = IMSM_OROM_VOLUMES_PER_HBA
 445         };
 446         orom.attr = orom.rlc | IMSM_OROM_ATTR_ChecksumVerify;
 447
 448         if (check_env("IMSM_TEST_OROM_NORAID5")) {
 449                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 450                                 IMSM_OROM_RLC_RAID10;
 451         }
 452         if (check_env("IMSM_TEST_AHCI_EFI_NORAID5") && (hba->type == SYS_DEV_SAS)) {
 453                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 454                                 IMSM_OROM_RLC_RAID10;
 455         }
 456         if (check_env("IMSM_TEST_SCU_EFI_NORAID5") && (hba->type == SYS_DEV_SATA)) {
 457                 orom.rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 458                                 IMSM_OROM_RLC_RAID10;
 459         }
 460
 461         struct orom_entry *ret = add_orom(&orom);
 462
 463         add_orom_device_id(ret, hba->dev_id);
 464
 465         return &ret->orom;
 466 }
 467
 468 static const struct imsm_orom *find_imsm_hba_orom(struct sys_dev *hba)
 469 {
 470         unsigned long align;
 471
 472         if (check_env("IMSM_TEST_OROM"))
 473                 return imsm_platform_test(hba);
 474
 475         /* return empty OROM capabilities in EFI test mode */
 476         if (check_env("IMSM_TEST_AHCI_EFI") || check_env("IMSM_TEST_SCU_EFI"))
 477                 return NULL;
 478
 479         find_intel_devices();
 480
 481         if (intel_devices == NULL)
 482                 return NULL;
 483
 484         /* scan option-rom memory looking for an imsm signature */
 485         if (check_env("IMSM_SAFE_OROM_SCAN"))
 486                 align = 2048;
 487         else
 488                 align = 512;
 489         if (probe_roms_init(align) != 0)
 490                 return NULL;
 491         probe_roms();
 492         /* ignore return value - True is returned if both adapater roms are found */
 493         scan_adapter_roms(scan);
 494         probe_roms_exit();
 495
 496         return get_orom_by_device_id(hba->dev_id);
 497 }
 498
 499 #define GUID_STR_MAX    37  /* according to GUID format:
 500                              * xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" */
 501
 502 #define EFI_GUID(a, b, c, d0, d1, d2, d3, d4, d5, d6, d7) \
 503 ((struct efi_guid) \
 504 {{ (a) & 0xff, ((a) >> 8) & 0xff, ((a) >> 16) & 0xff, ((a) >> 24) & 0xff, \
 505   (b) & 0xff, ((b) >> 8) & 0xff, \
 506   (c) & 0xff, ((c) >> 8) & 0xff, \
 507   (d0), (d1), (d2), (d3), (d4), (d5), (d6), (d7) }})
 508
 509 #define SYS_EFI_VAR_PATH "/sys/firmware/efi/vars"
 510 #define SYS_EFIVARS_PATH "/sys/firmware/efi/efivars"
 511 #define SCU_PROP "RstScuV"
 512 #define AHCI_PROP "RstSataV"
 513 #define AHCI_SSATA_PROP "RstsSatV"
 514 #define AHCI_TSATA_PROP "RsttSatV"
 515 #define VMD_PROP "RstUefiV"
 516
 517 #define VENDOR_GUID \
 518         EFI_GUID(0x193dfefa, 0xa445, 0x4302, 0x99, 0xd8, 0xef, 0x3a, 0xad, 0x1a, 0x04, 0xc6)
 519
 520 #define PCI_CLASS_RAID_CNTRL 0x010400
 521
 522 static int read_efi_var(void *buffer, ssize_t buf_size,
 523                         const char *variable_name, struct efi_guid guid)
 524 {
 525         char path[PATH_MAX];
 526         char buf[GUID_STR_MAX];
 527         int fd;
 528         ssize_t n;
 529
 530         snprintf(path, PATH_MAX, "%s/%s-%s", SYS_EFIVARS_PATH, variable_name, guid_str(buf, guid));
 531
 532         fd = open(path, O_RDONLY);
 533         if (fd < 0)
 534                 return 1;
 535
 536         /* read the variable attributes and ignore it */
 537         n = read(fd, buf, sizeof(__u32));
 538         if (n < 0) {
 539                 close(fd);
 540                 return 1;
 541         }
 542
 543         /* read the variable data */
 544         n = read(fd, buffer, buf_size);
 545         close(fd);
 546         if (n < buf_size)
 547                 return 1;
 548
 549         return 0;
 550 }
 551
 552 static int read_efi_variable(void *buffer, ssize_t buf_size,
 553                              const char *variable_name, struct efi_guid guid)
 554 {
 555         char path[PATH_MAX];
 556         char buf[GUID_STR_MAX];
 557         int dfd;
 558         ssize_t n, var_data_len;
 559
 560         /* Try to read the variable using the new efivarfs interface first.
 561          * If that fails, fall back to the old sysfs-efivars interface. */
 562         if (!read_efi_var(buffer, buf_size, variable_name, guid))
 563                 return 0;
 564
 565         snprintf(path, PATH_MAX, "%s/%s-%s/size", SYS_EFI_VAR_PATH, variable_name, guid_str(buf, guid));
 566
 567         dprintf("EFI VAR: path=%s\n", path);
 568         /* get size of variable data */
 569         dfd = open(path, O_RDONLY);
 570         if (dfd < 0)
 571                 return 1;
 572
 573         n = read(dfd, &buf, sizeof(buf));
 574         close(dfd);
 575         if (n < 0)
 576                 return 1;
 577         buf[n] = '\0';
 578
 579         errno = 0;
 580         var_data_len = strtoul(buf, NULL, 16);
 581         if ((errno == ERANGE && (var_data_len == LONG_MAX)) ||
 582             (errno != 0 && var_data_len == 0))
 583                 return 1;
 584
 585         /* get data */
 586         snprintf(path, PATH_MAX, "%s/%s-%s/data", SYS_EFI_VAR_PATH, variable_name, guid_str(buf, guid));
 587
 588         dprintf("EFI VAR: path=%s\n", path);
 589         dfd = open(path, O_RDONLY);
 590         if (dfd < 0)
 591                 return 1;
 592
 593         n = read(dfd, buffer, buf_size);
 594         close(dfd);
 595         if (n != var_data_len || n < buf_size) {
 596                 return 1;
 597         }
 598
 599         return 0;
 600 }
 601
 602 const struct imsm_orom *find_imsm_efi(struct sys_dev *hba)
 603 {
 604         struct imsm_orom orom;
 605         struct orom_entry *ret;
 606         static const char * const sata_efivars[] = {AHCI_PROP, AHCI_SSATA_PROP,
 607                                                     AHCI_TSATA_PROP};
 608         unsigned long i;
 609
 610         if (check_env("IMSM_TEST_AHCI_EFI") || check_env("IMSM_TEST_SCU_EFI"))
 611                 return imsm_platform_test(hba);
 612
 613         /* OROM test is set, return that there is no EFI capabilities */
 614         if (check_env("IMSM_TEST_OROM"))
 615                 return NULL;
 616
 617         switch (hba->type) {
 618         case SYS_DEV_SAS:
 619                 if (!read_efi_variable(&orom, sizeof(orom), SCU_PROP,
 620                                        VENDOR_GUID))
 621                         break;
 622
 623                 return NULL;
 624         case SYS_DEV_SATA:
 625                 if (hba->class != PCI_CLASS_RAID_CNTRL)
 626                         return NULL;
 627
 628                 for (i = 0; i < ARRAY_SIZE(sata_efivars); i++) {
 629                         if (!read_efi_variable(&orom, sizeof(orom),
 630                                                 sata_efivars[i], VENDOR_GUID))
 631                                 break;
 632
 633                 }
 634                 if (i == ARRAY_SIZE(sata_efivars))
 635                         return NULL;
 636
 637                 break;
 638         case SYS_DEV_VMD:
 639                 if (!read_efi_variable(&orom, sizeof(orom), VMD_PROP,
 640                                        VENDOR_GUID))
 641                         break;
 642                 return NULL;
 643         default:
 644                 return NULL;
 645         }
 646
 647         ret = add_orom(&orom);
 648         add_orom_device_id(ret, hba->dev_id);
 649         ret->type = hba->type;
 650
 651         return &ret->orom;
 652 }
 653
 654 const struct imsm_orom *find_imsm_nvme(struct sys_dev *hba)
 655 {
 656         static struct orom_entry *nvme_orom;
 657
 658         if (hba->type != SYS_DEV_NVME)
 659                 return NULL;
 660
 661         if (!nvme_orom) {
 662                 struct imsm_orom nvme_orom_compat = {
 663                         .signature = IMSM_NVME_OROM_COMPAT_SIGNATURE,
 664                         .rlc = IMSM_OROM_RLC_RAID0 | IMSM_OROM_RLC_RAID1 |
 665                                                 IMSM_OROM_RLC_RAID10 | IMSM_OROM_RLC_RAID5,
 666                         .sss = IMSM_OROM_SSS_4kB | IMSM_OROM_SSS_8kB |
 667                                                 IMSM_OROM_SSS_16kB | IMSM_OROM_SSS_32kB |
 668                                                 IMSM_OROM_SSS_64kB | IMSM_OROM_SSS_128kB,
 669                         .dpa = IMSM_OROM_DISKS_PER_ARRAY_NVME,
 670                         .tds = IMSM_OROM_TOTAL_DISKS_NVME,
 671                         .vpa = IMSM_OROM_VOLUMES_PER_ARRAY,
 672                         .vphba = IMSM_OROM_TOTAL_DISKS_NVME / 2 * IMSM_OROM_VOLUMES_PER_ARRAY,
 673                         .attr = IMSM_OROM_ATTR_2TB | IMSM_OROM_ATTR_2TB_DISK,
 674                         .driver_features = IMSM_OROM_CAPABILITIES_EnterpriseSystem
 675                 };
 676                 nvme_orom = add_orom(&nvme_orom_compat);
 677         }
 678         add_orom_device_id(nvme_orom, hba->dev_id);
 679         nvme_orom->type = SYS_DEV_NVME;
 680         return &nvme_orom->orom;
 681 }
 682
 683 const struct imsm_orom *find_imsm_capability(struct sys_dev *hba)
 684 {
 685         const struct imsm_orom *cap = get_orom_by_device_id(hba->dev_id);
 686
 687         if (cap)
 688                 return cap;
 689
 690         if (hba->type == SYS_DEV_NVME)
 691                 return find_imsm_nvme(hba);
 692         if ((cap = find_imsm_efi(hba)) != NULL)
 693                 return cap;
 694         if ((cap = find_imsm_hba_orom(hba)) != NULL)
 695                 return cap;
 696
 697         return NULL;
 698 }
 699
 700 /* Check whether the nvme device is represented by nvme subsytem,
 701  * if yes virtual path should be changed to hardware device path,
 702  * to allow IMSM capabilities detection.
 703  * Returns:
 704  *      hardware path to device - if the device is represented via
 705  *              nvme virtual subsytem
 706  *      NULL - if the device is not represented via nvme virtual subsytem
 707  */
 708 char *get_nvme_multipath_dev_hw_path(const char *dev_path)
 709 {
 710         DIR *dir;
 711         struct dirent *ent;
 712         char *rp = NULL;
 713
 714         if (strncmp(dev_path, NVME_SUBSYS_PATH, strlen(NVME_SUBSYS_PATH)) != 0)
 715                 return NULL;
 716
 717         dir = opendir(dev_path);
 718         if (!dir)
 719                 return NULL;
 720
 721         for (ent = readdir(dir); ent; ent = readdir(dir)) {
 722                 char buf[strlen(dev_path) + strlen(ent->d_name) + 1];
 723
 724                 /* Check if dir is a controller, ignore namespaces*/
 725                 if (!(strncmp(ent->d_name, "nvme", 4) == 0) ||
 726                     (strrchr(ent->d_name, 'n') != &ent->d_name[0]))
 727                         continue;
 728
 729                 sprintf(buf, "%s/%s", dev_path, ent->d_name);
 730                 rp = realpath(buf, NULL);
 731                 break;
 732         }
 733
 734         closedir(dir);
 735         return rp;
 736 }
 737
 738 /* Description: Return part or whole realpath for the dev
 739  * Parameters:
 740  *      dev - the device to be quered
 741  *      dev_level - level of "/device" entries. It allows to caller to access
 742  *                  virtual or physical devices which are on "path" to quered
 743  *                  one.
 744  *      buf - optional, must be PATH_MAX size. If set, then will be used.
 745  */
 746 char *devt_to_devpath(dev_t dev, int dev_level, char *buf)
 747 {
 748         char device[PATH_MAX];
 749         char *hw_path;
 750         int i;
 751         unsigned long device_free_len = sizeof(device) - 1;
 752         char dev_str[] = "/device";
 753         unsigned long dev_str_len = strlen(dev_str);
 754
 755         snprintf(device, sizeof(device), "/sys/dev/block/%d:%d", major(dev),
 756                  minor(dev));
 757
 758         /* If caller wants block device, return path to it even if it is exposed
 759          * via virtual layer.
 760          */
 761         if (dev_level == 0)
 762                 return realpath(device, buf);
 763
 764         device_free_len -= strlen(device);
 765         for (i = 0; i < dev_level; i++) {
 766                 if (device_free_len < dev_str_len)
 767                         return NULL;
 768
 769                 strncat(device, dev_str, device_free_len);
 770
 771                 /* Resolve nvme-subsystem abstraction if needed
 772                  */
 773                 device_free_len -= dev_str_len;
 774                 if (i == 0) {
 775                         char rp[PATH_MAX];
 776
 777                         if (!realpath(device, rp))
 778                                 return NULL;
 779                         hw_path = get_nvme_multipath_dev_hw_path(rp);
 780                         if (hw_path) {
 781                                 strcpy(device, hw_path);
 782                                 device_free_len = sizeof(device) -
 783                                                   strlen(device) - 1;
 784                                 free(hw_path);
 785                         }
 786                 }
 787         }
 788
 789         return realpath(device, buf);
 790 }
 791
 792 char *diskfd_to_devpath(int fd, int dev_level, char *buf)
 793 {
 794         /* return the device path for a disk, return NULL on error or fd
 795          * refers to a partition
 796          */
 797         struct stat st;
 798
 799         if (fstat(fd, &st) != 0)
 800                 return NULL;
 801         if (!S_ISBLK(st.st_mode))
 802                 return NULL;
 803
 804         return devt_to_devpath(st.st_rdev, dev_level, buf);
 805 }
 806
 807 int path_attached_to_hba(const char *disk_path, const char *hba_path)
 808 {
 809         int rc;
 810
 811         if (check_env("IMSM_TEST_AHCI_DEV") ||
 812             check_env("IMSM_TEST_SCU_DEV")) {
 813                 return 1;
 814         }
 815
 816         if (!disk_path || !hba_path)
 817                 return 0;
 818         dprintf("hba: %s - disk: %s\n", hba_path, disk_path);
 819         if (strncmp(disk_path, hba_path, strlen(hba_path)) == 0)
 820                 rc = 1;
 821         else
 822                 rc = 0;
 823
 824         return rc;
 825 }
 826
 827 int devt_attached_to_hba(dev_t dev, const char *hba_path)
 828 {
 829         char *disk_path = devt_to_devpath(dev, 1, NULL);
 830         int rc = path_attached_to_hba(disk_path, hba_path);
 831
 832         if (disk_path)
 833                 free(disk_path);
 834
 835         return rc;
 836 }
 837
 838 int disk_attached_to_hba(int fd, const char *hba_path)
 839 {
 840         char *disk_path = diskfd_to_devpath(fd, 1, NULL);
 841         int rc = path_attached_to_hba(disk_path, hba_path);
 842
 843         if (disk_path)
 844                 free(disk_path);
 845
 846         return rc;
 847 }
 848
 849 char *vmd_domain_to_controller(struct sys_dev *hba, char *buf)
 850 {
 851         struct dirent *ent;
 852         DIR *dir;
 853         char path[PATH_MAX];
 854
 855         if (!hba)
 856                 return NULL;
 857
 858         if (hba->type != SYS_DEV_VMD)
 859                 return NULL;
 860
 861         dir = opendir("/sys/bus/pci/drivers/vmd");
 862         if (!dir)
 863                 return NULL;
 864
 865         for (ent = readdir(dir); ent; ent = readdir(dir)) {
 866                 sprintf(path, "/sys/bus/pci/drivers/vmd/%s/domain/device",
 867                         ent->d_name);
 868
 869                 if (!realpath(path, buf))
 870                         continue;
 871
 872                 if (strncmp(buf, hba->path, strlen(buf)) == 0) {
 873                         sprintf(path, "/sys/bus/pci/drivers/vmd/%s", ent->d_name);
 874                         closedir(dir);
 875                         return realpath(path, buf);
 876                 }
 877         }
 878
 879         closedir(dir);
 880         return NULL;
 881 }
 882
 883 /* Scan over all controller's namespaces and compare nsid value to verify if
 884  * current one is supported. The routine doesn't check IMSM capabilities for
 885  * namespace. Only one nvme namespace is supported by IMSM.
 886  * Paramteres:
 887  *      fd - open descriptor to the nvme namespace
 888  *      verbose - error logging level
 889  * Returns:
 890  *      1 - if namespace is supported
 891  *      0 - otherwise
 892  */
 893 int imsm_is_nvme_namespace_supported(int fd, int verbose)
 894 {
 895         DIR *dir = NULL;
 896         struct dirent *ent;
 897         char cntrl_path[PATH_MAX];
 898         char ns_path[PATH_MAX];
 899         unsigned long long lowest_nsid = ULLONG_MAX;
 900         unsigned long long this_nsid;
 901         int rv = 0;
 902
 903
 904         if (!diskfd_to_devpath(fd, 1, cntrl_path) ||
 905             !diskfd_to_devpath(fd, 0, ns_path)) {
 906                 if (verbose)
 907                         pr_err("Cannot get device paths\n");
 908                 goto abort;
 909         }
 910
 911
 912         if (devpath_to_ll(ns_path, "nsid", &this_nsid)) {
 913                 if (verbose)
 914                         pr_err("Cannot read nsid value for %s",
 915                                basename(ns_path));
 916                 goto abort;
 917         }
 918
 919         dir = opendir(cntrl_path);
 920         if (!dir)
 921                 goto abort;
 922
 923         /* The lowest nvme namespace is supported */
 924         for (ent = readdir(dir); ent; ent = readdir(dir)) {
 925                 unsigned long long curr_nsid;
 926                 char curr_ns_path[PATH_MAX + 256];
 927
 928                 if (!strstr(ent->d_name, "nvme"))
 929                         continue;
 930
 931                 snprintf(curr_ns_path, sizeof(curr_ns_path), "%s/%s",
 932                          cntrl_path, ent->d_name);
 933
 934                 if (devpath_to_ll(curr_ns_path, "nsid", &curr_nsid))
 935                         goto abort;
 936
 937                 if (lowest_nsid > curr_nsid)
 938                         lowest_nsid = curr_nsid;
 939         }
 940
 941         if (this_nsid == lowest_nsid)
 942                 rv = 1;
 943         else if (verbose)
 944                 pr_err("IMSM is supported on the lowest NVMe namespace\n");
 945
 946 abort:
 947         if (dir)
 948                 closedir(dir);
 949
 950         return rv;
 951 }
 952
 953 /* Verify if multipath is supported by NVMe controller
 954  * Returns:
 955  *      0 - not supported
 956  *      1 - supported
 957  */
 958 int is_multipath_nvme(int disk_fd)
 959 {
 960         char ns_path[PATH_MAX];
 961
 962         if (!diskfd_to_devpath(disk_fd, 0, ns_path))
 963                 return 0;
 964
 965         if (strncmp(ns_path, NVME_SUBSYS_PATH, strlen(NVME_SUBSYS_PATH)) == 0)
 966                 return 1;
 967
 968         return 0;
 969 }