2 * QEMU S390 bootmap interpreter
4 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
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
16 /* #define DEBUG_FALLBACK */
21 do { sclp_print("zipl: " txt); } while (0)
23 #define dputs(fmt, ...) \
28 static uint8_t sec
[MAX_SECTOR_SIZE
*4] __attribute__((__aligned__(PAGE_SIZE
)));
30 typedef struct ResetInfo
{
33 uint32_t ipl_continue
;
36 static ResetInfo save
;
38 static void jump_to_IPL_2(void)
40 ResetInfo
*current
= 0;
42 void (*ipl
)(void) = (void *) (uint64_t) current
->ipl_continue
;
44 ipl(); /* should not return */
47 static void jump_to_IPL_code(uint64_t address
)
49 /* store the subsystem information _after_ the bootmap was loaded */
50 write_subsystem_identification();
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.
56 ResetInfo
*current
= 0;
59 current
->ipl_addr
= (uint32_t) (uint64_t) &jump_to_IPL_2
;
60 current
->ipl_continue
= address
& 0x7fffffff;
62 debug_print_int("set IPL addr to", current
->ipl_continue
);
64 /* Ensure the guest output starts fresh */
69 * We use the load normal reset to keep r15 unchanged. jump_to_IPL_2
70 * can then use r15 as its stack pointer.
72 asm volatile("lghi 1,1\n\t"
75 virtio_panic("\n! IPL returns !\n");
78 /***********************************************************************
79 * IPL an ECKD DASD (CDL or LDL/CMS format)
82 static unsigned char _bprs
[8*1024]; /* guessed "max" ECKD sector size */
83 static const int max_bprs_entries
= sizeof(_bprs
) / sizeof(ExtEckdBlockPtr
);
85 static inline void verify_boot_info(BootInfo
*bip
)
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.");
96 static block_number_t
eckd_block_num(BootMapPointer
*p
)
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
106 - 1; /* block nr starts with zero */
110 static bool eckd_valid_address(BootMapPointer
*p
)
112 const uint64_t head
= p
->eckd
.head
& 0x000f;
114 if (head
>= virtio_get_heads()
115 || p
->eckd
.sector
> virtio_get_sectors()
116 || p
->eckd
.sector
<= 0) {
120 if (!virtio_guessed_disk_nature() &&
121 eckd_block_num(p
) >= virtio_get_blocks()) {
128 static block_number_t
load_eckd_segments(block_number_t blk
, uint64_t *address
)
130 block_number_t block_nr
;
132 BootMapPointer
*bprs
= (void *)_bprs
;
135 memset(_bprs
, FREE_SPACE_FILLER
, sizeof(_bprs
));
136 read_block(blk
, bprs
, "BPRS read failed");
141 block_nr
= eckd_block_num((void *)&(bprs
[j
].xeckd
));
142 if (is_null_block_number(block_nr
)) { /* end of chunk */
146 /* we need the updated blockno for the next indirect entry
147 * in the chain, but don't want to advance address
149 if (j
== (max_bprs_entries
- 1)) {
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");
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
162 * I.e. the next ptr must point to the unused memory area
164 memset(_bprs
, FREE_SPACE_FILLER
, sizeof(_bprs
));
165 read_block(block_nr
, bprs
, "BPRS continuation read failed");
170 /* Load (count+1) blocks of code at (block_nr)
171 * to memory (address).
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");
177 *address
+= (bprs
[j
].xeckd
.bptr
.count
+1) * virtio_get_block_size();
183 static void run_eckd_boot_script(block_number_t mbr_block_nr
)
186 block_number_t block_nr
;
188 ScsiMbr
*scsi_mbr
= (void *)sec
;
189 BootMapScript
*bms
= (void *)sec
;
191 memset(sec
, FREE_SPACE_FILLER
, sizeof(sec
));
192 read_block(mbr_block_nr
, sec
, "Cannot read MBR");
194 block_nr
= eckd_block_num((void *)&(scsi_mbr
->blockptr
));
196 memset(sec
, FREE_SPACE_FILLER
, sizeof(sec
));
197 read_block(block_nr
, sec
, "Cannot read Boot Map Script");
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
));
204 block_nr
= load_eckd_segments(block_nr
, &address
);
205 } while (block_nr
!= -1);
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 */
213 static void ipl_eckd_cdl(void)
216 Ipl2
*ipl2
= (void *)sec
;
217 IplVolumeLabel
*vlbl
= (void *)sec
;
218 block_number_t block_nr
;
220 /* we have just read the block #0 and recognized it as "IPL1" */
223 memset(sec
, FREE_SPACE_FILLER
, sizeof(sec
));
224 read_block(1, ipl2
, "Cannot read IPL2 record at block 1");
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.");
233 /* save pointer to Boot Script */
234 block_nr
= eckd_block_num((void *)&(mbr
->blockptr
));
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
);
244 run_eckd_boot_script(block_nr
);
248 static void print_eckd_ldl_msg(ECKD_IPL_mode_t mode
)
250 LDL_VTOC
*vlbl
= (void *)sec
; /* already read, 3rd block */
251 char msg
[4] = { '?', '.', '\n', '\0' };
253 sclp_print((mode
== ECKD_CMS
) ? "CMS" : "LDL");
254 sclp_print(" version ");
255 switch (vlbl
->LDL_version
) {
263 msg
[0] = vlbl
->LDL_version
;
264 msg
[0] &= 0x0f; /* convert EBCDIC */
265 msg
[0] |= 0x30; /* to ASCII (digit) */
270 print_volser(vlbl
->volser
);
273 static void ipl_eckd_ldl(ECKD_IPL_mode_t mode
)
275 block_number_t block_nr
;
276 BootInfo
*bip
= (void *)(sec
+ 0x70); /* BootInfo is MBR for LDL */
278 if (mode
!= ECKD_LDL_UNLABELED
) {
279 print_eckd_ldl_msg(mode
);
282 /* DO NOT read BootMap pointer (only one, xECKD) at block #2 */
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 */
290 sclp_print("unlabeled LDL.\n");
292 verify_boot_info(bip
);
294 block_nr
= eckd_block_num((void *)&(bip
->bp
.ipl
.bm_ptr
.eckd
.bptr
));
295 run_eckd_boot_script(block_nr
);
299 static void print_eckd_msg(void)
301 char msg
[] = "Using ECKD scheme (block size *****), ";
302 char *p
= &msg
[34], *q
= &msg
[30];
303 int n
= virtio_get_block_size();
305 /* Fill in the block size and show up the message */
306 if (n
> 0 && n
<= 99999) {
308 *p
-- = '0' + (n
% 10);
318 /***********************************************************************
322 static void zipl_load_segment(ComponentEntry
*entry
)
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
;
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) */
333 blockno
= entry
->data
.blockno
;
334 address
= entry
->load_address
;
336 debug_print_int("loading segment at block", blockno
);
337 debug_print_int("addr", address
);
340 memset(bprs
, FREE_SPACE_FILLER
, bprs_size
);
341 fill_hex_val(blk_no
, &blockno
, sizeof(blockno
));
342 read_block(blockno
, bprs
, err_msg
);
345 uint64_t *cur_desc
= (void *)&bprs
[i
];
347 blockno
= bprs
[i
].blockno
;
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)) {
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.
368 address
= virtio_load_direct(cur_desc
[0], cur_desc
[1], 0,
370 IPL_assert(address
!= -1, "zIPL load segment failed");
375 /* Run a zipl program */
376 static void zipl_run(ScsiBlockPtr
*pte
)
378 ComponentHeader
*header
;
379 ComponentEntry
*entry
;
380 uint8_t tmp_sec
[MAX_SECTOR_SIZE
];
382 read_block(pte
->blockno
, tmp_sec
, "Cannot read header");
383 header
= (ComponentHeader
*)tmp_sec
;
385 IPL_assert(magic_match(tmp_sec
, ZIPL_MAGIC
), "No zIPL magic");
386 IPL_assert(header
->type
== ZIPL_COMP_HEADER_IPL
, "Bad header type");
388 dputs("start loading images\n");
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
);
397 IPL_assert((uint8_t *)(&entry
[1]) <= (tmp_sec
+ MAX_SECTOR_SIZE
),
398 "Wrong entry value");
401 IPL_assert(entry
->component_type
== ZIPL_COMP_ENTRY_EXEC
, "No EXEC entry");
403 /* should not return */
404 jump_to_IPL_code(entry
->load_address
);
407 static void ipl_scsi(void)
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
;
415 /* The 0-th block (MBR) was already read into sec[] */
417 sclp_print("Using SCSI scheme.\n");
418 debug_print_int("program table", mbr
->blockptr
.blockno
);
420 /* Parse the program table */
421 read_block(mbr
->blockptr
.blockno
, sec
,
422 "Error reading Program Table");
424 IPL_assert(magic_match(sec
, ZIPL_MAGIC
), "No zIPL magic");
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
) {
433 program_table_entries
++;
436 debug_print_int("program table entries", program_table_entries
);
438 IPL_assert(program_table_entries
!= 0, "Empty Program Table");
440 /* Run the default entry */
442 prog_table_entry
= (ScsiBlockPtr
*)(sec
+ pte_len
);
444 zipl_run(prog_table_entry
); /* no return */
447 /***********************************************************************
448 * IPL El Torito ISO9660 image or DVD
451 static bool is_iso_bc_entry_compatible(IsoBcSection
*s
)
453 uint8_t *magic_sec
= (uint8_t *)(sec
+ ISO_SECTOR_SIZE
);
455 if (s
->unused
|| !s
->sector_count
) {
458 read_iso_sector(bswap32(s
->load_rba
), magic_sec
,
459 "Failed to read image sector 0");
461 /* Checking bytes 8 - 32 for S390 Linux magic */
462 return !_memcmp(magic_sec
+ 8, linux_s390_magic
, 24);
465 /* Location of the current sector of the directory */
466 static uint32_t sec_loc
[ISO9660_MAX_DIR_DEPTH
];
467 /* Offset in the current sector of the directory */
468 static uint32_t sec_offset
[ISO9660_MAX_DIR_DEPTH
];
469 /* Remained directory space in bytes */
470 static uint32_t dir_rem
[ISO9660_MAX_DIR_DEPTH
];
472 static inline uint32_t iso_get_file_size(uint32_t load_rba
)
474 IsoVolDesc
*vd
= (IsoVolDesc
*)sec
;
475 IsoDirHdr
*cur_record
= &vd
->vd
.primary
.rootdir
;
476 uint8_t *temp
= sec
+ ISO_SECTOR_SIZE
;
479 read_iso_sector(ISO_PRIMARY_VD_SECTOR
, sec
,
480 "Failed to read ISO primary descriptor");
481 sec_loc
[0] = iso_733_to_u32(cur_record
->ext_loc
);
486 IPL_assert(sec_offset
[level
] <= ISO_SECTOR_SIZE
,
487 "Directory tree structure violation");
489 cur_record
= (IsoDirHdr
*)(temp
+ sec_offset
[level
]);
491 if (sec_offset
[level
] == 0) {
492 read_iso_sector(sec_loc
[level
], temp
,
493 "Failed to read ISO directory");
494 if (dir_rem
[level
] == 0) {
495 /* Skip self and parent records */
496 dir_rem
[level
] = iso_733_to_u32(cur_record
->data_len
) -
498 sec_offset
[level
] += cur_record
->dr_len
;
500 cur_record
= (IsoDirHdr
*)(temp
+ sec_offset
[level
]);
501 dir_rem
[level
] -= cur_record
->dr_len
;
502 sec_offset
[level
] += cur_record
->dr_len
;
507 if (!cur_record
->dr_len
|| sec_offset
[level
] == ISO_SECTOR_SIZE
) {
508 /* Zero-padding and/or the end of current sector */
509 dir_rem
[level
] -= ISO_SECTOR_SIZE
- sec_offset
[level
];
510 sec_offset
[level
] = 0;
513 /* The directory record is valid */
514 if (load_rba
== iso_733_to_u32(cur_record
->ext_loc
)) {
515 return iso_733_to_u32(cur_record
->data_len
);
518 dir_rem
[level
] -= cur_record
->dr_len
;
519 sec_offset
[level
] += cur_record
->dr_len
;
521 if (cur_record
->file_flags
& 0x2) {
523 if (level
== ISO9660_MAX_DIR_DEPTH
- 1) {
524 sclp_print("ISO-9660 directory depth limit exceeded\n");
527 sec_loc
[level
] = iso_733_to_u32(cur_record
->ext_loc
);
528 sec_offset
[level
] = 0;
535 if (dir_rem
[level
] == 0) {
536 /* Nothing remaining */
538 read_iso_sector(sec_loc
[level
], temp
,
539 "Failed to read ISO directory");
546 static void load_iso_bc_entry(IsoBcSection
*load
)
548 IsoBcSection s
= *load
;
550 * According to spec, extent for each file
551 * is padded and ISO_SECTOR_SIZE bytes aligned
553 uint32_t blks_to_load
= bswap16(s
.sector_count
) >> ET_SECTOR_SHIFT
;
554 uint32_t real_size
= iso_get_file_size(bswap32(s
.load_rba
));
557 /* Round up blocks to load */
558 blks_to_load
= (real_size
+ ISO_SECTOR_SIZE
- 1) / ISO_SECTOR_SIZE
;
559 sclp_print("ISO boot image size verified\n");
561 sclp_print("ISO boot image size could not be verified\n");
564 read_iso_boot_image(bswap32(s
.load_rba
),
565 (void *)((uint64_t)bswap16(s
.load_segment
)),
568 /* Trying to get PSW at zero address */
569 if (*((uint64_t *)0) & IPL_PSW_MASK
) {
570 jump_to_IPL_code((*((uint64_t *)0)) & 0x7fffffff);
573 /* Try default linux start address */
574 jump_to_IPL_code(KERN_IMAGE_START
);
577 static uint32_t find_iso_bc(void)
579 IsoVolDesc
*vd
= (IsoVolDesc
*)sec
;
580 uint32_t block_num
= ISO_PRIMARY_VD_SECTOR
;
582 if (virtio_read_many(block_num
++, sec
, 1)) {
583 /* If primary vd cannot be read, there is no boot catalog */
587 while (is_iso_vd_valid(vd
) && vd
->type
!= VOL_DESC_TERMINATOR
) {
588 if (vd
->type
== VOL_DESC_TYPE_BOOT
) {
589 IsoVdElTorito
*et
= &vd
->vd
.boot
;
591 if (!_memcmp(&et
->el_torito
[0], el_torito_magic
, 32)) {
592 return bswap32(et
->bc_offset
);
595 read_iso_sector(block_num
++, sec
,
596 "Failed to read ISO volume descriptor");
602 static IsoBcSection
*find_iso_bc_entry(void)
604 IsoBcEntry
*e
= (IsoBcEntry
*)sec
;
605 uint32_t offset
= find_iso_bc();
612 read_iso_sector(offset
, sec
, "Failed to read El Torito boot catalog");
614 if (!is_iso_bc_valid(e
)) {
615 /* The validation entry is mandatory */
616 virtio_panic("No valid boot catalog found!\n");
621 * Each entry has 32 bytes size, so one sector cannot contain > 64 entries.
622 * We consider only boot catalogs with no more than 64 entries.
624 for (i
= 1; i
< ISO_BC_ENTRY_PER_SECTOR
; i
++) {
625 if (e
[i
].id
== ISO_BC_BOOTABLE_SECTION
) {
626 if (is_iso_bc_entry_compatible(&e
[i
].body
.sect
)) {
627 return &e
[i
].body
.sect
;
632 virtio_panic("No suitable boot entry found on ISO-9660 media!\n");
637 static void ipl_iso_el_torito(void)
639 IsoBcSection
*s
= find_iso_bc_entry();
642 load_iso_bc_entry(s
);
647 /***********************************************************************
653 ScsiMbr
*mbr
= (void *)sec
;
654 LDL_VTOC
*vlbl
= (void *)sec
;
657 memset(sec
, FREE_SPACE_FILLER
, sizeof(sec
));
658 read_block(0, mbr
, "Cannot read block 0");
660 dputs("checking magic\n");
662 if (magic_match(mbr
->magic
, ZIPL_MAGIC
)) {
663 ipl_scsi(); /* no return */
666 /* Check if we can boot as ISO media */
667 if (virtio_guessed_disk_nature()) {
668 virtio_assume_iso9660();
672 /* We have failed to follow the SCSI scheme, so */
673 if (virtio_guessed_disk_nature()) {
674 sclp_print("Using guessed DASD geometry.\n");
675 virtio_assume_eckd();
678 if (magic_match(mbr
->magic
, IPL1_MAGIC
)) {
679 ipl_eckd_cdl(); /* no return */
683 memset(sec
, FREE_SPACE_FILLER
, sizeof(sec
));
684 read_block(2, vlbl
, "Cannot read block 2");
686 if (magic_match(vlbl
->magic
, CMS1_MAGIC
)) {
687 ipl_eckd_ldl(ECKD_CMS
); /* no return */
689 if (magic_match(vlbl
->magic
, LNX1_MAGIC
)) {
690 ipl_eckd_ldl(ECKD_LDL
); /* no return */
693 ipl_eckd_ldl(ECKD_LDL_UNLABELED
); /* it still may return */
695 * Ok, it is not a LDL by any means.
696 * It still might be a CDL with zero record keys for IPL1 and IPL2
700 virtio_panic("\n* this can never happen *\n");