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 "s390-ccw.h"
  12 #include "bootmap.h"
  13 #include "virtio.h"
  14
  15 #ifdef DEBUG
  16 /* #define DEBUG_FALLBACK */
  17 #endif
  18
  19 #ifdef DEBUG_FALLBACK
  20 #define dputs(txt) \
  21     do { sclp_print("zipl: " txt); } while (0)
  22 #else
  23 #define dputs(fmt, ...) \
  24     do { } while (0)
  25 #endif
  26
  27 /* Scratch space */
  28 static uint8_t sec[MAX_SECTOR_SIZE*4] __attribute__((__aligned__(PAGE_SIZE)));
  29
  30 typedef struct ResetInfo {
  31     uint32_t ipl_mask;
  32     uint32_t ipl_addr;
  33     uint32_t ipl_continue;
  34 } ResetInfo;
  35
  36 static ResetInfo save;
  37
  38 static void jump_to_IPL_2(void)
  39 {
  40     ResetInfo *current = 0;
  41
  42     void (*ipl)(void) = (void *) (uint64_t) current->ipl_continue;
  43     *current = save;
  44     ipl(); /* should not return */
  45 }
  46
  47 static void jump_to_IPL_code(uint64_t address)
  48 {
  49     /* store the subsystem information _after_ the bootmap was loaded */
  50     write_subsystem_identification();
  51     /*
  52      * The IPL PSW is at address 0. We also must not overwrite the
  53      * content of non-BIOS memory after we loaded the guest, so we
  54      * save the original content and restore it in jump_to_IPL_2.
  55      */
  56     ResetInfo *current = 0;
  57
  58     save = *current;
  59     current->ipl_addr = (uint32_t) (uint64_t) &jump_to_IPL_2;
  60     current->ipl_continue = address & 0x7fffffff;
  61
  62     debug_print_int("set IPL addr to", current->ipl_continue);
  63
  64     /* Ensure the guest output starts fresh */
  65     sclp_print("\n");
  66
  67     /*
  68      * HACK ALERT.
  69      * We use the load normal reset to keep r15 unchanged. jump_to_IPL_2
  70      * can then use r15 as its stack pointer.
  71      */
  72     asm volatile("lghi 1,1\n\t"
  73                  "diag 1,1,0x308\n\t"
  74                  : : : "1", "memory");
  75     virtio_panic("\n! IPL returns !\n");
  76 }
  77
  78 /***********************************************************************
  79  * IPL an ECKD DASD (CDL or LDL/CMS format)
  80  */
  81
  82 static unsigned char _bprs[8*1024]; /* guessed "max" ECKD sector size */
  83 static const int max_bprs_entries = sizeof(_bprs) / sizeof(ExtEckdBlockPtr);
  84
  85 static inline void verify_boot_info(BootInfo *bip)
  86 {
  87     IPL_assert(magic_match(bip->magic, ZIPL_MAGIC), "No zIPL magic");
  88     IPL_assert(bip->version == BOOT_INFO_VERSION, "Wrong zIPL version");
  89     IPL_assert(bip->bp_type == BOOT_INFO_BP_TYPE_IPL, "DASD is not for IPL");
  90     IPL_assert(bip->dev_type == BOOT_INFO_DEV_TYPE_ECKD, "DASD is not ECKD");
  91     IPL_assert(bip->flags == BOOT_INFO_FLAGS_ARCH, "Not for this arch");
  92     IPL_assert(block_size_ok(bip->bp.ipl.bm_ptr.eckd.bptr.size),
  93                "Bad block size in zIPL section of the 1st record.");
  94 }
  95
  96 static block_number_t eckd_block_num(BootMapPointer *p)
  97 {
  98     const uint64_t sectors = virtio_get_sectors();
  99     const uint64_t heads = virtio_get_heads();
 100     const uint64_t cylinder = p->eckd.cylinder
 101                             + ((p->eckd.head & 0xfff0) << 12);
 102     const uint64_t head = p->eckd.head & 0x000f;
 103     const block_number_t block = sectors * heads * cylinder
 104                                + sectors * head
 105                                + p->eckd.sector
 106                                - 1; /* block nr starts with zero */
 107     return block;
 108 }
 109
 110 static bool eckd_valid_address(BootMapPointer *p)
 111 {
 112     const uint64_t head = p->eckd.head & 0x000f;
 113
 114     if (head >= virtio_get_heads()
 115         ||  p->eckd.sector > virtio_get_sectors()
 116         ||  p->eckd.sector <= 0) {
 117         return false;
 118     }
 119
 120     if (!virtio_guessed_disk_nature() &&
 121         eckd_block_num(p) >= virtio_get_blocks()) {
 122         return false;
 123     }
 124
 125     return true;
 126 }
 127
 128 static block_number_t load_eckd_segments(block_number_t blk, uint64_t *address)
 129 {
 130     block_number_t block_nr;
 131     int j, rc;
 132     BootMapPointer *bprs = (void *)_bprs;
 133     bool more_data;
 134
 135     memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
 136     read_block(blk, bprs, "BPRS read failed");
 137
 138     do {
 139         more_data = false;
 140         for (j = 0;; j++) {
 141             block_nr = eckd_block_num((void *)&(bprs[j].xeckd));
 142             if (is_null_block_number(block_nr)) { /* end of chunk */
 143                 break;
 144             }
 145
 146             /* we need the updated blockno for the next indirect entry
 147              * in the chain, but don't want to advance address
 148              */
 149             if (j == (max_bprs_entries - 1)) {
 150                 break;
 151             }
 152
 153             IPL_assert(block_size_ok(bprs[j].xeckd.bptr.size),
 154                        "bad chunk block size");
 155             IPL_assert(eckd_valid_address(&bprs[j]), "bad chunk ECKD addr");
 156
 157             if ((bprs[j].xeckd.bptr.count == 0) && unused_space(&(bprs[j+1]),
 158                 sizeof(EckdBlockPtr))) {
 159                 /* This is a "continue" pointer.
 160                  * This ptr should be the last one in the current
 161                  * script section.
 162                  * I.e. the next ptr must point to the unused memory area
 163                  */
 164                 memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
 165                 read_block(block_nr, bprs, "BPRS continuation read failed");
 166                 more_data = true;
 167                 break;
 168             }
 169
 170             /* Load (count+1) blocks of code at (block_nr)
 171              * to memory (address).
 172              */
 173             rc = virtio_read_many(block_nr, (void *)(*address),
 174                                   bprs[j].xeckd.bptr.count+1);
 175             IPL_assert(rc == 0, "code chunk read failed");
 176
 177             *address += (bprs[j].xeckd.bptr.count+1) * virtio_get_block_size();
 178         }
 179     } while (more_data);
 180     return block_nr;
 181 }
 182
 183 static void run_eckd_boot_script(block_number_t mbr_block_nr)
 184 {
 185     int i;
 186     block_number_t block_nr;
 187     uint64_t address;
 188     ScsiMbr *scsi_mbr = (void *)sec;
 189     BootMapScript *bms = (void *)sec;
 190
 191     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 192     read_block(mbr_block_nr, sec, "Cannot read MBR");
 193
 194     block_nr = eckd_block_num((void *)&(scsi_mbr->blockptr));
 195
 196     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 197     read_block(block_nr, sec, "Cannot read Boot Map Script");
 198
 199     for (i = 0; bms->entry[i].type == BOOT_SCRIPT_LOAD; i++) {
 200         address = bms->entry[i].address.load_address;
 201         block_nr = eckd_block_num(&(bms->entry[i].blkptr));
 202
 203         do {
 204             block_nr = load_eckd_segments(block_nr, &address);
 205         } while (block_nr != -1);
 206     }
 207
 208     IPL_assert(bms->entry[i].type == BOOT_SCRIPT_EXEC,
 209                "Unknown script entry type");
 210     jump_to_IPL_code(bms->entry[i].address.load_address); /* no return */
 211 }
 212
 213 static void ipl_eckd_cdl(void)
 214 {
 215     XEckdMbr *mbr;
 216     Ipl2 *ipl2 = (void *)sec;
 217     IplVolumeLabel *vlbl = (void *)sec;
 218     block_number_t block_nr;
 219
 220     /* we have just read the block #0 and recognized it as "IPL1" */
 221     sclp_print("CDL\n");
 222
 223     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 224     read_block(1, ipl2, "Cannot read IPL2 record at block 1");
 225
 226     mbr = &ipl2->u.x.mbr;
 227     IPL_assert(magic_match(mbr, ZIPL_MAGIC), "No zIPL section in IPL2 record.");
 228     IPL_assert(block_size_ok(mbr->blockptr.xeckd.bptr.size),
 229                "Bad block size in zIPL section of IPL2 record.");
 230     IPL_assert(mbr->dev_type == DEV_TYPE_ECKD,
 231                "Non-ECKD device type in zIPL section of IPL2 record.");
 232
 233     /* save pointer to Boot Script */
 234     block_nr = eckd_block_num((void *)&(mbr->blockptr));
 235
 236     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 237     read_block(2, vlbl, "Cannot read Volume Label at block 2");
 238     IPL_assert(magic_match(vlbl->key, VOL1_MAGIC),
 239                "Invalid magic of volume label block");
 240     IPL_assert(magic_match(vlbl->f.key, VOL1_MAGIC),
 241                "Invalid magic of volser block");
 242     print_volser(vlbl->f.volser);
 243
 244     run_eckd_boot_script(block_nr);
 245     /* no return */
 246 }
 247
 248 static void print_eckd_ldl_msg(ECKD_IPL_mode_t mode)
 249 {
 250     LDL_VTOC *vlbl = (void *)sec; /* already read, 3rd block */
 251     char msg[4] = { '?', '.', '\n', '\0' };
 252
 253     sclp_print((mode == ECKD_CMS) ? "CMS" : "LDL");
 254     sclp_print(" version ");
 255     switch (vlbl->LDL_version) {
 256     case LDL1_VERSION:
 257         msg[0] = '1';
 258         break;
 259     case LDL2_VERSION:
 260         msg[0] = '2';
 261         break;
 262     default:
 263         msg[0] = vlbl->LDL_version;
 264         msg[0] &= 0x0f; /* convert EBCDIC   */
 265         msg[0] |= 0x30; /* to ASCII (digit) */
 266         msg[1] = '?';
 267         break;
 268     }
 269     sclp_print(msg);
 270     print_volser(vlbl->volser);
 271 }
 272
 273 static void ipl_eckd_ldl(ECKD_IPL_mode_t mode)
 274 {
 275     block_number_t block_nr;
 276     BootInfo *bip = (void *)(sec + 0x70); /* BootInfo is MBR for LDL */
 277
 278     if (mode != ECKD_LDL_UNLABELED) {
 279         print_eckd_ldl_msg(mode);
 280     }
 281
 282     /* DO NOT read BootMap pointer (only one, xECKD) at block #2 */
 283
 284     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 285     read_block(0, sec, "Cannot read block 0 to grab boot info.");
 286     if (mode == ECKD_LDL_UNLABELED) {
 287         if (!magic_match(bip->magic, ZIPL_MAGIC)) {
 288             return; /* not applicable layout */
 289         }
 290         sclp_print("unlabeled LDL.\n");
 291     }
 292     verify_boot_info(bip);
 293
 294     block_nr = eckd_block_num((void *)&(bip->bp.ipl.bm_ptr.eckd.bptr));
 295     run_eckd_boot_script(block_nr);
 296     /* no return */
 297 }
 298
 299 static void print_eckd_msg(void)
 300 {
 301     char msg[] = "Using ECKD scheme (block size *****), ";
 302     char *p = &msg[34], *q = &msg[30];
 303     int n = virtio_get_block_size();
 304
 305     /* Fill in the block size and show up the message */
 306     if (n > 0 && n <= 99999) {
 307         while (n) {
 308             *p-- = '0' + (n % 10);
 309             n /= 10;
 310         }
 311         while (p >= q) {
 312             *p-- = ' ';
 313         }
 314     }
 315     sclp_print(msg);
 316 }
 317
 318 /***********************************************************************
 319  * IPL a SCSI disk
 320  */
 321
 322 static void zipl_load_segment(ComponentEntry *entry)
 323 {
 324     const int max_entries = (MAX_SECTOR_SIZE / sizeof(ScsiBlockPtr));
 325     ScsiBlockPtr *bprs = (void *)sec;
 326     const int bprs_size = sizeof(sec);
 327     block_number_t blockno;
 328     uint64_t address;
 329     int i;
 330     char err_msg[] = "zIPL failed to read BPRS at 0xZZZZZZZZZZZZZZZZ";
 331     char *blk_no = &err_msg[30]; /* where to print blockno in (those ZZs) */
 332
 333     blockno = entry->data.blockno;
 334     address = entry->load_address;
 335
 336     debug_print_int("loading segment at block", blockno);
 337     debug_print_int("addr", address);
 338
 339     do {
 340         memset(bprs, FREE_SPACE_FILLER, bprs_size);
 341         fill_hex_val(blk_no, &blockno, sizeof(blockno));
 342         read_block(blockno, bprs, err_msg);
 343
 344         for (i = 0;; i++) {
 345             uint64_t *cur_desc = (void *)&bprs[i];
 346
 347             blockno = bprs[i].blockno;
 348             if (!blockno) {
 349                 break;
 350             }
 351
 352             /* we need the updated blockno for the next indirect entry in the
 353                chain, but don't want to advance address */
 354             if (i == (max_entries - 1)) {
 355                 break;
 356             }
 357
 358             if (bprs[i].blockct == 0 && unused_space(&bprs[i + 1],
 359                 sizeof(ScsiBlockPtr))) {
 360                 /* This is a "continue" pointer.
 361                  * This ptr is the last one in the current script section.
 362                  * I.e. the next ptr must point to the unused memory area.
 363                  * The blockno is not zero, so the upper loop must continue
 364                  * reading next section of BPRS.
 365                  */
 366                 break;
 367             }
 368             address = virtio_load_direct(cur_desc[0], cur_desc[1], 0,
 369                                          (void *)address);
 370             IPL_assert(address != -1, "zIPL load segment failed");
 371         }
 372     } while (blockno);
 373 }
 374
 375 /* Run a zipl program */
 376 static void zipl_run(ScsiBlockPtr *pte)
 377 {
 378     ComponentHeader *header;
 379     ComponentEntry *entry;
 380     uint8_t tmp_sec[MAX_SECTOR_SIZE];
 381
 382     read_block(pte->blockno, tmp_sec, "Cannot read header");
 383     header = (ComponentHeader *)tmp_sec;
 384
 385     IPL_assert(magic_match(tmp_sec, ZIPL_MAGIC), "No zIPL magic");
 386     IPL_assert(header->type == ZIPL_COMP_HEADER_IPL, "Bad header type");
 387
 388     dputs("start loading images\n");
 389
 390     /* Load image(s) into RAM */
 391     entry = (ComponentEntry *)(&header[1]);
 392     while (entry->component_type == ZIPL_COMP_ENTRY_LOAD) {
 393         zipl_load_segment(entry);
 394
 395         entry++;
 396
 397         IPL_assert((uint8_t *)(&entry[1]) <= (tmp_sec + MAX_SECTOR_SIZE),
 398                    "Wrong entry value");
 399     }
 400
 401     IPL_assert(entry->component_type == ZIPL_COMP_ENTRY_EXEC, "No EXEC entry");
 402
 403     /* should not return */
 404     jump_to_IPL_code(entry->load_address);
 405 }
 406
 407 static void ipl_scsi(void)
 408 {
 409     ScsiMbr *mbr = (void *)sec;
 410     uint8_t *ns, *ns_end;
 411     int program_table_entries = 0;
 412     const int pte_len = sizeof(ScsiBlockPtr);
 413     ScsiBlockPtr *prog_table_entry;
 414
 415     /* The 0-th block (MBR) was already read into sec[] */
 416
 417     sclp_print("Using SCSI scheme.\n");
 418     debug_print_int("program table", mbr->blockptr.blockno);
 419
 420     /* Parse the program table */
 421     read_block(mbr->blockptr.blockno, sec,
 422                "Error reading Program Table");
 423
 424     IPL_assert(magic_match(sec, ZIPL_MAGIC), "No zIPL magic");
 425
 426     ns_end = sec + virtio_get_block_size();
 427     for (ns = (sec + pte_len); (ns + pte_len) < ns_end; ns++) {
 428         prog_table_entry = (ScsiBlockPtr *)ns;
 429         if (!prog_table_entry->blockno) {
 430             break;
 431         }
 432
 433         program_table_entries++;
 434     }
 435
 436     debug_print_int("program table entries", program_table_entries);
 437
 438     IPL_assert(program_table_entries != 0, "Empty Program Table");
 439
 440     /* Run the default entry */
 441
 442     prog_table_entry = (ScsiBlockPtr *)(sec + pte_len);
 443
 444     zipl_run(prog_table_entry); /* no return */
 445 }
 446
 447 /***********************************************************************
 448  * IPL starts here
 449  */
 450
 451 void zipl_load(void)
 452 {
 453     ScsiMbr *mbr = (void *)sec;
 454     LDL_VTOC *vlbl = (void *)sec;
 455
 456     /* Grab the MBR */
 457     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 458     read_block(0, mbr, "Cannot read block 0");
 459
 460     dputs("checking magic\n");
 461
 462     if (magic_match(mbr->magic, ZIPL_MAGIC)) {
 463         ipl_scsi(); /* no return */
 464     }
 465
 466     /* We have failed to follow the SCSI scheme, so */
 467     if (virtio_guessed_disk_nature()) {
 468         sclp_print("Using guessed DASD geometry.\n");
 469         virtio_assume_eckd();
 470     }
 471     print_eckd_msg();
 472     if (magic_match(mbr->magic, IPL1_MAGIC)) {
 473         ipl_eckd_cdl(); /* no return */
 474     }
 475
 476     /* LDL/CMS? */
 477     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
 478     read_block(2, vlbl, "Cannot read block 2");
 479
 480     if (magic_match(vlbl->magic, CMS1_MAGIC)) {
 481         ipl_eckd_ldl(ECKD_CMS); /* no return */
 482     }
 483     if (magic_match(vlbl->magic, LNX1_MAGIC)) {
 484         ipl_eckd_ldl(ECKD_LDL); /* no return */
 485     }
 486
 487     ipl_eckd_ldl(ECKD_LDL_UNLABELED); /* it still may return */
 488     /*
 489      * Ok, it is not a LDL by any means.
 490      * It still might be a CDL with zero record keys for IPL1 and IPL2
 491      */
 492     ipl_eckd_cdl();
 493
 494     virtio_panic("\n* this can never happen *\n");
 495 }