On 27/04/2019 16:14, John Paul Adrian Glaubitz wrote: > Hi! > > On 4/27/19 5:02 PM, Mark Cave-Ayland wrote: >> Adrian: based upon this I think what you need to do is: >> >> - For consistency rename /boot/grub/sparc64.elf to /boot/grub/core.img to >> match >> the descriptions of the grub components in the documentation > > Ok. But still generate it with grub-mkimage? > > + grub-mkimage -O sparc64-ieee1275-cdcore -p '()/boot/grub' \ > + -o $(TEMP_CD_INFO_DIR)/boot/grub/sparc64.elf \ > + $(GRUB_MODULES) $(GRUB_MODULES_CDROM)
Yes, I believe that grub-mkimage generates the grub core image (which by convention is called core.img). >> - Confirm that cdboot.img is the a.out executable generated by boot.S and >> diskboot.S >> (use dd and hexdump to check for the 0x01 0x03 0x01 0x07 signature in the >> first 4 >> bytes) > > glaubitz@kyoto:/usr/lib/grub/sparc64-ieee1275$ xxd cdboot.img |head -n2 > 00000000: 0103 0107 0000 01e0 0000 0000 0000 0000 ................ > 00000010: 0000 0000 0000 4000 0000 0000 0000 0000 ......@......... > glaubitz@kyoto:/usr/lib/grub/sparc64-ieee1275$ > > Yes. Excellent! >> - In your second patch re-enable the -G/-B options but with -G set to >> cdboot.img: >> add_mkisofs_opt $CDDIR/../$N.mkisofs_opts "-G cdboot.img -B ..." > > I set $CDDIR to the source directory which contains core.img? How does the > initial > bootloader know that it has to boot core.img? This is the part which Thomas explained: the offset of core.img *as embedded within the ISO filesystem image* and its size are embedded at byte offsets 552 and 560 respectively from the start of the image. From what I can see boot.S/diskboot.S seek to sector 1, read in these values, seek to the start offset and then read the relevant number of bytes into memory before finally transferring control. >> Once this is done and the ISO image is generated, the basic setup can be >> checked as >> follows: >> >> - Use fdisk or similar tool to confirm that the Sun disk label exists at the >> start of the .iso with all 8 partitions starting at sector 0 with an end >> sector >> representing the contents of the whole CDROM image >> >> - Use dd and hexdump to confirm that cdboot.img appears 512 bytes into the >> .iso >> by checking for the 0x01 0x03 0x01 0x07 signature >> >> - Next check the offset and size of /boot/grub/core.img embedded in sector 1 >> using >> dd and hexdump. As per Thomas' message you should find the offset in bytes >> of >> the start of /boot/grub/core.img at offset 552 from the start of the .iso >> (64-bit >> big endian) and its size at offset 560 from the start of the .iso (32-bit >> big >> endian) >> >> - If everything is correct then again using dd and hexdump you should be >> able to >> see the ELF signature for core.img from the offset discovered above. >> >> Assuming all of this looks correct, then I believe that you should end up >> with a >> bootable CDROM image using grub... > > Nice. I'll give it a shot tonight and report back. Actually one thought: are you sure that your grub-mkimage generates an ELF binary? Given that boot.S/diskboot.S don't contain an ELF loader this makes me think that core.img is actually a plain non-relocatable binary. In which case, of course, the final ELF header check above is incorrect. ATB, Mark.
