[syslinux] Support for 4kb disks

[Frediano Ziglio]

Thanks Peter for advices.

I managed now to load the second step (loading ldlinux.sys on ram) and
it seems to get the configuration file (hitting tab show me the two
entries I have in extlinux.conf). However kernel does not load.

Not that easy to debug, do you have some advice? Currently I enabled
debugging of BIOS (SeaBIOS) and I look for calls. Calling printf in
core/fs/diskio.c seems to not work.

In many places rdwr_sectors get called assuming that sectors are 512
(for instance to read BPB or boot sector) so I changed to assume sectors
calling rdwr_sectors are 512 however there are other places (like
getoneblk) where this assumption is wrong. Also in iso9660.c you shift
from iso blocks (2k) to disk one (4k) loosing precision. I would
personally write a rdwr_bytes to couple with this problems leading
rdwr_sectors as a low level functions.

Another problem I'm facing is ADV "sectors". There is a string
assumption that ADV sectors are 512 bytes long. Currently I align
initial ldlinux and ADV sectors to 4k but I didn't tested if all works
correctly (I think not...). Writing into a single sectors have problems
if write fails (you corrupt both), aligning to sector size is a waste of
space for systems like btrfs (they would occupy 8kb out of 64kb).

Frediano


On Mon, 2012-07-30 at 07:17 -0700, H. Peter Anvin wrote:
> Partition tables are invalid when moving disk sizes anyway.  As far as removing legacy int13 support that is a non-option, although it probably is not necessary for those to work with larger sector sizes.
> 
> Frediano Ziglio <frediano.ziglio at citrix.com> wrote:
> 
> >On Wed, 2012-07-18 at 18:59 -0700, H. Peter Anvin wrote:
> >> On 07/17/2012 01:37 AM, Frediano Ziglio wrote:
> >> >
> >> > It's a physical machine with a beta firmware on it. I wrote a small
> >> > program to read/write two sectors and I can confirm that each
> >sector is
> >> > 4kb.
> >> >
> >> 
> >> Support for non-512-byte logical sectors is sparse since such support
> >in 
> >> actual systems is scant to nonexistent.  Most of the pieces are there
> >
> >> but as you have noticed there are bits still missing.
> >> 
> >> 	-hpa
> >> 
> >
> >I finally got a repro using Qemu and patched SeaBIOS (just changed the
> >check in src/blockcmd.c for DISK_SECTOR_SIZE adding 4096).
> >
> >After some though I think there are 2 way to make it work:
> >1- patch extlinux installer in order to detect and use sector > 512
> >2- make extlinux code assume that sectors specified are 512 but handle
> >case where BIOS use bigger sizes
> >
> >1 make simpler extlinux code (probably code is already there) but it
> >could be a problem if a user wants to switch the disk from 512 to 4096
> >(this can happen changing jumpers on disk or firmware on some disk
> >controller).
> >
> >2 however require more code on extlinux (it's hard to make it works
> >only
> >using 512 bytes).
> >
> >I'd vote for 2 and I think there are some options to avoid the 512
> >bytes
> >problems:
> >- as sectors are bigger mbr code load more code, ext2/3/4 let us 1024
> >bytes so we can use entire 1kb, just check on first 512 bytes we can
> >use
> >entire kb
> >- remove old floppy and no int13x code and use only int13x.
> >
> >In this case I would vote for first.
> >
> >Any suggestion?
> >
> >Regards
> >  Frediano
>