pc-bios/s390-ccw/bootmap.c

   1 /*
   2  * QEMU S390 bootmap interpreter
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or (at
   7  * your option) any later version. See the COPYING file in the top-level
   8  * directory.
   9  */
  10
  11 #include "libc.h"
  12 #include "s390-ccw.h"
  13 #include "bootmap.h"
  14 #include "virtio.h"
  15 #include "bswap.h"
  16
  17 #ifdef DEBUG
  18 /* #define DEBUG_FALLBACK */
  19 #endif
  20
  21 #ifdef DEBUG_FALLBACK
  22 #define dputs(txt) \
  23     do { sclp_print("zipl: " txt); } while (0)
  24 #else
  25 #define dputs(fmt, ...) \
  26     do { } while (0)
  27 #endif
  28
  29 /* Scratch space */
  30 static uint8_t sec[MAX_SECTOR_SIZE*4] __attribute__((__aligned__(PAGE_SIZE)));
  31
  32 const uint8_t el_torito_magic[] = "EL TORITO SPECIFICATION"
  33                                   "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
  34
  35 /*
  36  * Match two CCWs located after PSW and eight filler bytes.
  37  * From libmagic and arch/s390/kernel/head.S.
  38  */
  39 const uint8_t linux_s390_magic[] = "\x02\x00\x00\x18\x60\x00\x00\x50\x02\x00"
  40                                    "\x00\x68\x60\x00\x00\x50\x40\x40\x40\x40"
  41                                    "\x40\x40\x40\x40";
  42
  43 static inline bool is_iso_vd_valid(IsoVolDesc *vd)
  44 {
  45     const uint8_t vol_desc_magic[] = "CD001";
  46
  47     return !memcmp(&vd->ident[0], vol_desc_magic, 5) &&
  48            vd->version == 0x1 &&
  49            vd->type <= VOL_DESC_TYPE_PARTITION;
  50 }
  51
  52 /***********************************************************************
  53  * IPL an ECKD DASD (CDL or LDL/CMS format)
  54  */
  55
  56 static unsigned char _bprs[8*1024]; /* guessed "max" ECKD sector size */
  57 static const int max_bprs_entries = sizeof(_bprs) / sizeof(ExtEckdBlockPtr);
  58 static uint8_t _s2[MAX_SECTOR_SIZE * 3] __attribute__((__aligned__(PAGE_SIZE)));
  59 static void *s2_prev_blk = _s2;
  60 static void *s2_cur_blk = _s2 + MAX_SECTOR_SIZE;
  61 static void *s2_next_blk = _s2 + MAX_SECTOR_SIZE * 2;
  62
  63 static inline void verify_boot_info(BootInfo *bip)
  64 {
  65     IPL_assert(magic_match(bip->magic, ZIPL_MAGIC), "No zIPL sig in BootInfo");
  66     IPL_assert(bip->version == BOOT_INFO_VERSION, "Wrong zIPL version");
  67     IPL_assert(bip->bp_type == BOOT_INFO_BP_TYPE_IPL, "DASD is not for IPL");
  68     IPL_assert(bip->dev_type == BOOT_INFO_DEV_TYPE_ECKD, "DASD is not ECKD");
  69     IPL_assert(bip->flags == BOOT_INFO_FLAGS_ARCH, "Not for this arch");
  70     IPL_assert(block_size_ok(bip->bp.ipl.bm_ptr.eckd.bptr.size),
  71                "Bad block size in zIPL section of the 1st record.");
  72 }
  73
  74 static block_number_t eckd_block_num(EckdCHS *chs)
  75 {
  76     const uint64_t sectors = virtio_get_sectors();
  77     const uint64_t heads = virtio_get_heads();
  78     const uint64_t cylinder = chs->cylinder
  79                             + ((chs->head & 0xfff0) << 12);
  80     const uint64_t head = chs->head & 0x000f;
  81     const block_number_t block = sectors * heads * cylinder
  82                                + sectors * head
  83                                + chs->sector
  84                                - 1; /* block nr starts with zero */
  85     return block;
  86 }
  87
  88 static bool eckd_valid_address(BootMapPointer *p)
  89 {
  90     const uint64_t head = p->eckd.chs.head & 0x000f;
  91
  92     if (head >= virtio_get_heads()
  93         ||  p->eckd.chs.sector > virtio_get_sectors()
  94         ||  p->eckd.chs.sector <= 0) {
  95         return false;
  96     }
  97
  98     if (!virtio_guessed_disk_nature() &&
  99         eckd_block_num(&p->eckd.chs) >= virtio_get_blocks()) {
 100         return false;
 101     }
 102
 103     return true;
 104 }
 105
 106 static block_number_t load_eckd_segments(block_number_t blk, uint64_t *address)
 107 {
 108     block_number_t block_nr;
 109     int j, rc;
 110     BootMapPointer *bprs = (void *)_bprs;
 111     bool more_data;
 112
 113     memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
 114     read_block(blk, bprs, "BPRS read failed");
 115
 116     do {
 117         more_data = false;
 118         for (j = 0;; j++) {
 119             block_nr = eckd_block_num(&bprs[j].xeckd.bptr.chs);
 120             if (is_null_block_number(block_nr)) { /* end of chunk */
 121                 break;
 122             }
 123
 124             /* we need the updated blockno for the next indirect entry
 125              * in the chain, but don't want to advance address
 126              */
 127             if (j == (max_bprs_entries - 1)) {
 128                 break;
 129             }
 130
 131             IPL_assert(block_size_ok(bprs[j].xeckd.bptr.size),
 132                        "bad chunk block size");
 133             IPL_assert(eckd_valid_address(&bprs[j]), "bad chunk ECKD addr");
 134
 135             if ((bprs[j].xeckd.bptr.count == 0) && unused_space(&(bprs[j+1]),
 136                 sizeof(EckdBlockPtr))) {
 137                 /* This is a "continue" pointer.
 138                  * This ptr should be the last one in the current
 139                  * script section.
 140                  * I.e. the next ptr must point to the unused memory area
 141                  */
 142                 memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
 143                 read_block(block_nr, bprs, "BPRS continuation read failed");
 144                 more_data = true;
 145                 break;
 146             }
 147
 148             /* Load (count+1) blocks of code at (block_nr)
 149              * to memory (address).
 150              */
 151             rc = virtio_read_many(block_nr, (void *)(*address),
 152                                   bprs[j].xeckd.bptr.count+1);
 153             IPL_assert(rc == 0, "code chunk read failed");
 154
 155             *address += (bprs[j].xeckd.bptr.count+1) * virtio_get_block_size();
 156         }
 157     } while (more_data);
 158     return block_nr;
 159 }
 160
 161 static bool find_zipl_boot_menu_banner(int *offset)
 162 {
 163     int i;
 164
 165     /* Menu banner starts with "zIPL" */
 166     for (i = 0; i < virtio_get_block_size() - 4; i++) {
 167         if (magic_match(s2_cur_blk + i, ZIPL_MAGIC_EBCDIC)) {
 168             *offset = i;
 169             return true;
 170         }
 171     }
 172
 173     return false;
 174 }
 175
 176 static int eckd_get_boot_menu_index(block_number_t s1b_block_nr)
 177 {
 178     block_number_t cur_block_nr;
 179     block_number_t prev_block_nr = 0;
 180     block_number_t next_block_nr = 0;
 181     EckdStage1b *s1b = (void *)sec;
 182     int banner_offset;
 183     int i;
 184
 185     /* Get Stage1b data */
 186     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 187     read_block(s1b_block_nr, s1b, "Cannot read stage1b boot loader");
 188
 189     memset(_s2, FREE_SPACE_FILLER, sizeof(_s2));
 190
 191     /* Get Stage2 data */
 192     for (i = 0; i < STAGE2_BLK_CNT_MAX; i++) {
 193         cur_block_nr = eckd_block_num(&s1b->seek[i].chs);
 194
 195         if (!cur_block_nr) {
 196             break;
 197         }
 198
 199         read_block(cur_block_nr, s2_cur_blk, "Cannot read stage2 boot loader");
 200
 201         if (find_zipl_boot_menu_banner(&banner_offset)) {
 202             /*
 203              * Load the adjacent blocks to account for the
 204              * possibility of menu data spanning multiple blocks.
 205              */
 206             if (prev_block_nr) {
 207                 read_block(prev_block_nr, s2_prev_blk,
 208                            "Cannot read stage2 boot loader");
 209             }
 210
 211             if (i + 1 < STAGE2_BLK_CNT_MAX) {
 212                 next_block_nr = eckd_block_num(&s1b->seek[i + 1].chs);
 213             }
 214
 215             if (next_block_nr) {
 216                 read_block(next_block_nr, s2_next_blk,
 217                            "Cannot read stage2 boot loader");
 218             }
 219
 220             return menu_get_zipl_boot_index(s2_cur_blk + banner_offset);
 221         }
 222
 223         prev_block_nr = cur_block_nr;
 224     }
 225
 226     sclp_print("No zipl boot menu data found. Booting default entry.");
 227     return 0;
 228 }
 229
 230 static void run_eckd_boot_script(block_number_t bmt_block_nr,
 231                                  block_number_t s1b_block_nr)
 232 {
 233     int i;
 234     unsigned int loadparm = get_loadparm_index();
 235     block_number_t block_nr;
 236     uint64_t address;
 237     BootMapTable *bmt = (void *)sec;
 238     BootMapScript *bms = (void *)sec;
 239
 240     if (menu_is_enabled_zipl()) {
 241         loadparm = eckd_get_boot_menu_index(s1b_block_nr);
 242     }
 243
 244     debug_print_int("loadparm", loadparm);
 245     IPL_assert(loadparm < MAX_BOOT_ENTRIES, "loadparm value greater than"
 246                " maximum number of boot entries allowed");
 247
 248     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 249     read_block(bmt_block_nr, sec, "Cannot read Boot Map Table");
 250
 251     block_nr = eckd_block_num(&bmt->entry[loadparm].xeckd.bptr.chs);
 252     IPL_assert(block_nr != -1, "Cannot find Boot Map Table Entry");
 253
 254     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 255     read_block(block_nr, sec, "Cannot read Boot Map Script");
 256
 257     for (i = 0; bms->entry[i].type == BOOT_SCRIPT_LOAD ||
 258                 bms->entry[i].type == BOOT_SCRIPT_SIGNATURE; i++) {
 259
 260         /* We don't support secure boot yet, so we skip signature entries */
 261         if (bms->entry[i].type == BOOT_SCRIPT_SIGNATURE) {
 262             continue;
 263         }
 264
 265         address = bms->entry[i].address.load_address;
 266         block_nr = eckd_block_num(&bms->entry[i].blkptr.xeckd.bptr.chs);
 267
 268         do {
 269             block_nr = load_eckd_segments(block_nr, &address);
 270         } while (block_nr != -1);
 271     }
 272
 273     IPL_assert(bms->entry[i].type == BOOT_SCRIPT_EXEC,
 274                "Unknown script entry type");
 275     jump_to_IPL_code(bms->entry[i].address.load_address); /* no return */
 276 }
 277
 278 static void ipl_eckd_cdl(void)
 279 {
 280     XEckdMbr *mbr;
 281     EckdCdlIpl2 *ipl2 = (void *)sec;
 282     IplVolumeLabel *vlbl = (void *)sec;
 283     block_number_t bmt_block_nr, s1b_block_nr;
 284
 285     /* we have just read the block #0 and recognized it as "IPL1" */
 286     sclp_print("CDL\n");
 287
 288     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 289     read_block(1, ipl2, "Cannot read IPL2 record at block 1");
 290
 291     mbr = &ipl2->mbr;
 292     IPL_assert(magic_match(mbr, ZIPL_MAGIC), "No zIPL section in IPL2 record.");
 293     IPL_assert(block_size_ok(mbr->blockptr.xeckd.bptr.size),
 294                "Bad block size in zIPL section of IPL2 record.");
 295     IPL_assert(mbr->dev_type == DEV_TYPE_ECKD,
 296                "Non-ECKD device type in zIPL section of IPL2 record.");
 297
 298     /* save pointer to Boot Map Table */
 299     bmt_block_nr = eckd_block_num(&mbr->blockptr.xeckd.bptr.chs);
 300
 301     /* save pointer to Stage1b Data */
 302     s1b_block_nr = eckd_block_num(&ipl2->stage1.seek[0].chs);
 303
 304     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 305     read_block(2, vlbl, "Cannot read Volume Label at block 2");
 306     IPL_assert(magic_match(vlbl->key, VOL1_MAGIC),
 307                "Invalid magic of volume label block");
 308     IPL_assert(magic_match(vlbl->f.key, VOL1_MAGIC),
 309                "Invalid magic of volser block");
 310     print_volser(vlbl->f.volser);
 311
 312     run_eckd_boot_script(bmt_block_nr, s1b_block_nr);
 313     /* no return */
 314 }
 315
 316 static void print_eckd_ldl_msg(ECKD_IPL_mode_t mode)
 317 {
 318     LDL_VTOC *vlbl = (void *)sec; /* already read, 3rd block */
 319     char msg[4] = { '?', '.', '\n', '\0' };
 320
 321     sclp_print((mode == ECKD_CMS) ? "CMS" : "LDL");
 322     sclp_print(" version ");
 323     switch (vlbl->LDL_version) {
 324     case LDL1_VERSION:
 325         msg[0] = '1';
 326         break;
 327     case LDL2_VERSION:
 328         msg[0] = '2';
 329         break;
 330     default:
 331         msg[0] = vlbl->LDL_version;
 332         msg[0] &= 0x0f; /* convert EBCDIC   */
 333         msg[0] |= 0x30; /* to ASCII (digit) */
 334         msg[1] = '?';
 335         break;
 336     }
 337     sclp_print(msg);
 338     print_volser(vlbl->volser);
 339 }
 340
 341 static void ipl_eckd_ldl(ECKD_IPL_mode_t mode)
 342 {
 343     block_number_t bmt_block_nr, s1b_block_nr;
 344     EckdLdlIpl1 *ipl1 = (void *)sec;
 345
 346     if (mode != ECKD_LDL_UNLABELED) {
 347         print_eckd_ldl_msg(mode);
 348     }
 349
 350     /* DO NOT read BootMap pointer (only one, xECKD) at block #2 */
 351
 352     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 353     read_block(0, sec, "Cannot read block 0 to grab boot info.");
 354     if (mode == ECKD_LDL_UNLABELED) {
 355         if (!magic_match(ipl1->bip.magic, ZIPL_MAGIC)) {
 356             return; /* not applicable layout */
 357         }
 358         sclp_print("unlabeled LDL.\n");
 359     }
 360     verify_boot_info(&ipl1->bip);
 361
 362     /* save pointer to Boot Map Table */
 363     bmt_block_nr = eckd_block_num(&ipl1->bip.bp.ipl.bm_ptr.eckd.bptr.chs);
 364
 365     /* save pointer to Stage1b Data */
 366     s1b_block_nr = eckd_block_num(&ipl1->stage1.seek[0].chs);
 367
 368     run_eckd_boot_script(bmt_block_nr, s1b_block_nr);
 369     /* no return */
 370 }
 371
 372 static void print_eckd_msg(void)
 373 {
 374     char msg[] = "Using ECKD scheme (block size *****), ";
 375     char *p = &msg[34], *q = &msg[30];
 376     int n = virtio_get_block_size();
 377
 378     /* Fill in the block size and show up the message */
 379     if (n > 0 && n <= 99999) {
 380         while (n) {
 381             *p-- = '0' + (n % 10);
 382             n /= 10;
 383         }
 384         while (p >= q) {
 385             *p-- = ' ';
 386         }
 387     }
 388     sclp_print(msg);
 389 }
 390
 391 static void ipl_eckd(void)
 392 {
 393     XEckdMbr *mbr = (void *)sec;
 394     LDL_VTOC *vlbl = (void *)sec;
 395
 396     print_eckd_msg();
 397
 398     /* Grab the MBR again */
 399     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 400     read_block(0, mbr, "Cannot read block 0 on DASD");
 401
 402     if (magic_match(mbr->magic, IPL1_MAGIC)) {
 403         ipl_eckd_cdl(); /* no return */
 404     }
 405
 406     /* LDL/CMS? */
 407     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 408     read_block(2, vlbl, "Cannot read block 2");
 409
 410     if (magic_match(vlbl->magic, CMS1_MAGIC)) {
 411         ipl_eckd_ldl(ECKD_CMS); /* no return */
 412     }
 413     if (magic_match(vlbl->magic, LNX1_MAGIC)) {
 414         ipl_eckd_ldl(ECKD_LDL); /* no return */
 415     }
 416
 417     ipl_eckd_ldl(ECKD_LDL_UNLABELED); /* it still may return */
 418     /*
 419      * Ok, it is not a LDL by any means.
 420      * It still might be a CDL with zero record keys for IPL1 and IPL2
 421      */
 422     ipl_eckd_cdl();
 423 }
 424
 425 /***********************************************************************
 426  * IPL a SCSI disk
 427  */
 428
 429 static void zipl_load_segment(ComponentEntry *entry)
 430 {
 431     const int max_entries = (MAX_SECTOR_SIZE / sizeof(ScsiBlockPtr));
 432     ScsiBlockPtr *bprs = (void *)sec;
 433     const int bprs_size = sizeof(sec);
 434     block_number_t blockno;
 435     uint64_t address;
 436     int i;
 437     char err_msg[] = "zIPL failed to read BPRS at 0xZZZZZZZZZZZZZZZZ";
 438     char *blk_no = &err_msg[30]; /* where to print blockno in (those ZZs) */
 439
 440     blockno = entry->data.blockno;
 441     address = entry->load_address;
 442
 443     debug_print_int("loading segment at block", blockno);
 444     debug_print_int("addr", address);
 445
 446     do {
 447         memset(bprs, FREE_SPACE_FILLER, bprs_size);
 448         fill_hex_val(blk_no, &blockno, sizeof(blockno));
 449         read_block(blockno, bprs, err_msg);
 450
 451         for (i = 0;; i++) {
 452             uint64_t *cur_desc = (void *)&bprs[i];
 453
 454             blockno = bprs[i].blockno;
 455             if (!blockno) {
 456                 break;
 457             }
 458
 459             /* we need the updated blockno for the next indirect entry in the
 460                chain, but don't want to advance address */
 461             if (i == (max_entries - 1)) {
 462                 break;
 463             }
 464
 465             if (bprs[i].blockct == 0 && unused_space(&bprs[i + 1],
 466                 sizeof(ScsiBlockPtr))) {
 467                 /* This is a "continue" pointer.
 468                  * This ptr is the last one in the current script section.
 469                  * I.e. the next ptr must point to the unused memory area.
 470                  * The blockno is not zero, so the upper loop must continue
 471                  * reading next section of BPRS.
 472                  */
 473                 break;
 474             }
 475             address = virtio_load_direct(cur_desc[0], cur_desc[1], 0,
 476                                          (void *)address);
 477             IPL_assert(address != -1, "zIPL load segment failed");
 478         }
 479     } while (blockno);
 480 }
 481
 482 /* Run a zipl program */
 483 static void zipl_run(ScsiBlockPtr *pte)
 484 {
 485     ComponentHeader *header;
 486     ComponentEntry *entry;
 487     uint8_t tmp_sec[MAX_SECTOR_SIZE];
 488
 489     read_block(pte->blockno, tmp_sec, "Cannot read header");
 490     header = (ComponentHeader *)tmp_sec;
 491
 492     IPL_assert(magic_match(tmp_sec, ZIPL_MAGIC), "No zIPL magic in header");
 493     IPL_assert(header->type == ZIPL_COMP_HEADER_IPL, "Bad header type");
 494
 495     dputs("start loading images\n");
 496
 497     /* Load image(s) into RAM */
 498     entry = (ComponentEntry *)(&header[1]);
 499     while (entry->component_type == ZIPL_COMP_ENTRY_LOAD ||
 500            entry->component_type == ZIPL_COMP_ENTRY_SIGNATURE) {
 501
 502         /* We don't support secure boot yet, so we skip signature entries */
 503         if (entry->component_type == ZIPL_COMP_ENTRY_SIGNATURE) {
 504             entry++;
 505             continue;
 506         }
 507
 508         zipl_load_segment(entry);
 509
 510         entry++;
 511
 512         IPL_assert((uint8_t *)(&entry[1]) <= (tmp_sec + MAX_SECTOR_SIZE),
 513                    "Wrong entry value");
 514     }
 515
 516     IPL_assert(entry->component_type == ZIPL_COMP_ENTRY_EXEC, "No EXEC entry");
 517
 518     /* should not return */
 519     jump_to_IPL_code(entry->load_address);
 520 }
 521
 522 static void ipl_scsi(void)
 523 {
 524     ScsiMbr *mbr = (void *)sec;
 525     int program_table_entries = 0;
 526     BootMapTable *prog_table = (void *)sec;
 527     unsigned int loadparm = get_loadparm_index();
 528     bool valid_entries[MAX_BOOT_ENTRIES] = {false};
 529     size_t i;
 530
 531     /* Grab the MBR */
 532     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 533     read_block(0, mbr, "Cannot read block 0");
 534
 535     if (!magic_match(mbr->magic, ZIPL_MAGIC)) {
 536         return;
 537     }
 538
 539     sclp_print("Using SCSI scheme.\n");
 540     debug_print_int("MBR Version", mbr->version_id);
 541     IPL_check(mbr->version_id == 1,
 542               "Unknown MBR layout version, assuming version 1");
 543     debug_print_int("program table", mbr->pt.blockno);
 544     IPL_assert(mbr->pt.blockno, "No Program Table");
 545
 546     /* Parse the program table */
 547     read_block(mbr->pt.blockno, sec, "Error reading Program Table");
 548     IPL_assert(magic_match(sec, ZIPL_MAGIC), "No zIPL magic in PT");
 549
 550     for (i = 0; i < MAX_BOOT_ENTRIES; i++) {
 551         if (prog_table->entry[i].scsi.blockno) {
 552             valid_entries[i] = true;
 553             program_table_entries++;
 554         }
 555     }
 556
 557     debug_print_int("program table entries", program_table_entries);
 558     IPL_assert(program_table_entries != 0, "Empty Program Table");
 559
 560     if (menu_is_enabled_enum()) {
 561         loadparm = menu_get_enum_boot_index(valid_entries);
 562     }
 563
 564     debug_print_int("loadparm", loadparm);
 565     IPL_assert(loadparm < MAX_BOOT_ENTRIES, "loadparm value greater than"
 566                " maximum number of boot entries allowed");
 567
 568     zipl_run(&prog_table->entry[loadparm].scsi); /* no return */
 569 }
 570
 571 /***********************************************************************
 572  * IPL El Torito ISO9660 image or DVD
 573  */
 574
 575 static bool is_iso_bc_entry_compatible(IsoBcSection *s)
 576 {
 577     uint8_t *magic_sec = (uint8_t *)(sec + ISO_SECTOR_SIZE);
 578
 579     if (s->unused || !s->sector_count) {
 580         return false;
 581     }
 582     read_iso_sector(bswap32(s->load_rba), magic_sec,
 583                     "Failed to read image sector 0");
 584
 585     /* Checking bytes 8 - 32 for S390 Linux magic */
 586     return !memcmp(magic_sec + 8, linux_s390_magic, 24);
 587 }
 588
 589 /* Location of the current sector of the directory */
 590 static uint32_t sec_loc[ISO9660_MAX_DIR_DEPTH];
 591 /* Offset in the current sector of the directory */
 592 static uint32_t sec_offset[ISO9660_MAX_DIR_DEPTH];
 593 /* Remained directory space in bytes */
 594 static uint32_t dir_rem[ISO9660_MAX_DIR_DEPTH];
 595
 596 static inline uint32_t iso_get_file_size(uint32_t load_rba)
 597 {
 598     IsoVolDesc *vd = (IsoVolDesc *)sec;
 599     IsoDirHdr *cur_record = &vd->vd.primary.rootdir;
 600     uint8_t *temp = sec + ISO_SECTOR_SIZE;
 601     int level = 0;
 602
 603     read_iso_sector(ISO_PRIMARY_VD_SECTOR, sec,
 604                     "Failed to read ISO primary descriptor");
 605     sec_loc[0] = iso_733_to_u32(cur_record->ext_loc);
 606     dir_rem[0] = 0;
 607     sec_offset[0] = 0;
 608
 609     while (level >= 0) {
 610         IPL_assert(sec_offset[level] <= ISO_SECTOR_SIZE,
 611                    "Directory tree structure violation");
 612
 613         cur_record = (IsoDirHdr *)(temp + sec_offset[level]);
 614
 615         if (sec_offset[level] == 0) {
 616             read_iso_sector(sec_loc[level], temp,
 617                             "Failed to read ISO directory");
 618             if (dir_rem[level] == 0) {
 619                 /* Skip self and parent records */
 620                 dir_rem[level] = iso_733_to_u32(cur_record->data_len) -
 621                                  cur_record->dr_len;
 622                 sec_offset[level] += cur_record->dr_len;
 623
 624                 cur_record = (IsoDirHdr *)(temp + sec_offset[level]);
 625                 dir_rem[level] -= cur_record->dr_len;
 626                 sec_offset[level] += cur_record->dr_len;
 627                 continue;
 628             }
 629         }
 630
 631         if (!cur_record->dr_len || sec_offset[level] == ISO_SECTOR_SIZE) {
 632             /* Zero-padding and/or the end of current sector */
 633             dir_rem[level] -= ISO_SECTOR_SIZE - sec_offset[level];
 634             sec_offset[level] = 0;
 635             sec_loc[level]++;
 636         } else {
 637             /* The directory record is valid */
 638             if (load_rba == iso_733_to_u32(cur_record->ext_loc)) {
 639                 return iso_733_to_u32(cur_record->data_len);
 640             }
 641
 642             dir_rem[level] -= cur_record->dr_len;
 643             sec_offset[level] += cur_record->dr_len;
 644
 645             if (cur_record->file_flags & 0x2) {
 646                 /* Subdirectory */
 647                 if (level == ISO9660_MAX_DIR_DEPTH - 1) {
 648                     sclp_print("ISO-9660 directory depth limit exceeded\n");
 649                 } else {
 650                     level++;
 651                     sec_loc[level] = iso_733_to_u32(cur_record->ext_loc);
 652                     sec_offset[level] = 0;
 653                     dir_rem[level] = 0;
 654                     continue;
 655                 }
 656             }
 657         }
 658
 659         if (dir_rem[level] == 0) {
 660             /* Nothing remaining */
 661             level--;
 662             read_iso_sector(sec_loc[level], temp,
 663                             "Failed to read ISO directory");
 664         }
 665     }
 666
 667     return 0;
 668 }
 669
 670 static void load_iso_bc_entry(IsoBcSection *load)
 671 {
 672     IsoBcSection s = *load;
 673     /*
 674      * According to spec, extent for each file
 675      * is padded and ISO_SECTOR_SIZE bytes aligned
 676      */
 677     uint32_t blks_to_load = bswap16(s.sector_count) >> ET_SECTOR_SHIFT;
 678     uint32_t real_size = iso_get_file_size(bswap32(s.load_rba));
 679
 680     if (real_size) {
 681         /* Round up blocks to load */
 682         blks_to_load = (real_size + ISO_SECTOR_SIZE - 1) / ISO_SECTOR_SIZE;
 683         sclp_print("ISO boot image size verified\n");
 684     } else {
 685         sclp_print("ISO boot image size could not be verified\n");
 686     }
 687
 688     read_iso_boot_image(bswap32(s.load_rba),
 689                         (void *)((uint64_t)bswap16(s.load_segment)),
 690                         blks_to_load);
 691
 692     jump_to_low_kernel();
 693 }
 694
 695 static uint32_t find_iso_bc(void)
 696 {
 697     IsoVolDesc *vd = (IsoVolDesc *)sec;
 698     uint32_t block_num = ISO_PRIMARY_VD_SECTOR;
 699
 700     if (virtio_read_many(block_num++, sec, 1)) {
 701         /* If primary vd cannot be read, there is no boot catalog */
 702         return 0;
 703     }
 704
 705     while (is_iso_vd_valid(vd) && vd->type != VOL_DESC_TERMINATOR) {
 706         if (vd->type == VOL_DESC_TYPE_BOOT) {
 707             IsoVdElTorito *et = &vd->vd.boot;
 708
 709             if (!memcmp(&et->el_torito[0], el_torito_magic, 32)) {
 710                 return bswap32(et->bc_offset);
 711             }
 712         }
 713         read_iso_sector(block_num++, sec,
 714                         "Failed to read ISO volume descriptor");
 715     }
 716
 717     return 0;
 718 }
 719
 720 static IsoBcSection *find_iso_bc_entry(void)
 721 {
 722     IsoBcEntry *e = (IsoBcEntry *)sec;
 723     uint32_t offset = find_iso_bc();
 724     int i;
 725     unsigned int loadparm = get_loadparm_index();
 726
 727     if (!offset) {
 728         return NULL;
 729     }
 730
 731     read_iso_sector(offset, sec, "Failed to read El Torito boot catalog");
 732
 733     if (!is_iso_bc_valid(e)) {
 734         /* The validation entry is mandatory */
 735         panic("No valid boot catalog found!\n");
 736         return NULL;
 737     }
 738
 739     /*
 740      * Each entry has 32 bytes size, so one sector cannot contain > 64 entries.
 741      * We consider only boot catalogs with no more than 64 entries.
 742      */
 743     for (i = 1; i < ISO_BC_ENTRY_PER_SECTOR; i++) {
 744         if (e[i].id == ISO_BC_BOOTABLE_SECTION) {
 745             if (is_iso_bc_entry_compatible(&e[i].body.sect)) {
 746                 if (loadparm <= 1) {
 747                     /* found, default, or unspecified */
 748                     return &e[i].body.sect;
 749                 }
 750                 loadparm--;
 751             }
 752         }
 753     }
 754
 755     panic("No suitable boot entry found on ISO-9660 media!\n");
 756
 757     return NULL;
 758 }
 759
 760 static void ipl_iso_el_torito(void)
 761 {
 762     IsoBcSection *s = find_iso_bc_entry();
 763
 764     if (s) {
 765         load_iso_bc_entry(s);
 766         /* no return */
 767     }
 768 }
 769
 770 /***********************************************************************
 771  * Bus specific IPL sequences
 772  */
 773
 774 static void zipl_load_vblk(void)
 775 {
 776     if (virtio_guessed_disk_nature()) {
 777         virtio_assume_iso9660();
 778     }
 779     ipl_iso_el_torito();
 780
 781     if (virtio_guessed_disk_nature()) {
 782         sclp_print("Using guessed DASD geometry.\n");
 783         virtio_assume_eckd();
 784     }
 785     ipl_eckd();
 786 }
 787
 788 static void zipl_load_vscsi(void)
 789 {
 790     if (virtio_get_block_size() == VIRTIO_ISO_BLOCK_SIZE) {
 791         /* Is it an ISO image in non-CD drive? */
 792         ipl_iso_el_torito();
 793     }
 794
 795     sclp_print("Using guessed DASD geometry.\n");
 796     virtio_assume_eckd();
 797     ipl_eckd();
 798 }
 799
 800 /***********************************************************************
 801  * IPL starts here
 802  */
 803
 804 void zipl_load(void)
 805 {
 806     VDev *vdev = virtio_get_device();
 807
 808     if (vdev->is_cdrom) {
 809         ipl_iso_el_torito();
 810         panic("\n! Cannot IPL this ISO image !\n");
 811     }
 812
 813     if (virtio_get_device_type() == VIRTIO_ID_NET) {
 814         jump_to_IPL_code(vdev->netboot_start_addr);
 815     }
 816
 817     ipl_scsi();
 818
 819     switch (virtio_get_device_type()) {
 820     case VIRTIO_ID_BLOCK:
 821         zipl_load_vblk();
 822         break;
 823     case VIRTIO_ID_SCSI:
 824         zipl_load_vscsi();
 825         break;
 826     default:
 827         panic("\n! Unknown IPL device type !\n");
 828     }
 829
 830     panic("\n* this can never happen *\n");
 831 }