Hi William,
On 15/08/23 21:49, William R Sowerbutts wrote:
We need to fix pata_falcon for Q40, and with data on William's disk in
little endian order, but data on other users' disks in big endian order, we
need to come up with a way to support both for now.
I don't mind if you break my disk -- I can always byte-swap it if required,
or patch my kernel to use my preferred byte ordering.
Valid points have been made about needing to keep existing disks working.
For the kernel driver to switch byte order after some specific version seems
unexpected.
On the other hand, having disks compatible with other systems is important,
and as a naive new user the byte-swapped disk format was quite unexpected to
me.
It seems there are reasonable arguments for both sides.
I had a cursory look through other drivers and could not find any that make
byte ordering a user-selectable option. Is there any precedent for this? It
would be interesting to learn how they exposed the option to the user.
As Geert explained, it was agreed by the IDE maintainers that any data
byte order fixup needed would be handled in the block layer, and
existing byte swapping options were removed as a result (as far as I
recall, that was only ever necessary for Falcon and Q40).
I'm not clear if pata_falcon on Atari records data on disk in normal or
swapped byte order. Can I take a drive from my Atari Falcon Linux machine
and connect it to a PC without needing to byte swap it?
No, such a disk can only be read on a PC when running ARAnyM (where the
m68k Linux kernel then handles the necessary fixup). Data are in the
wrong byte order when hooking up a Falcon IDE disk to a PC.
I want to float some potential solutions for discussion;
* Add a kernel configuration option to choose between legacy and compatible
byte ordering on Q40 at compile time (affecting all disks)
That's easiest ...
* Add a pata_falcon driver option to choose which of the connected disks
should use legacy or compatible byte ordering at run time
I'm trying to do that at present. I don't expect that to fly with the
IDE or block maintainers though (for afore mentioned reasons).
* Extend pata_falcon to examine the connected disk's contents, looking for
some marker that reliably indicates a legacy byte order disk is connected.
Default to compatible byte ordering if this mark is not found. Maybe it
would be cleaner for some optional module atop the driver to do this.
Your only hope for this is to get such autodetection into the Atari
partition table code. It has no place in a device driver. As far as the
device driver is concerned, a disk is just a bag of bytes, no structure.
You'd then have to communicate the result back to user space. If you
have enough RAM to first boot into an initrd that can then sort out what
byte order to use for the root filesystem, this may work. I've only got
14 MB, so won't be able to test such a scheme.
* Assuming pata_legacy can be made to work well: Have pata_falcon retain
legacy byte ordering, while pata_legacy drives the same hardware in
compatible byte ordering mode. User chooses the driver to choose byte
ordering.
That could be a way out, but this applies to both disk either way. You
can't have one driver handle drive 1, the other drive 2.
* Switch pata_falcon to compatible byte order, document for any legacy disk
users how they can either do a one-time migration of their data to
compatible format, or setup the block layer to do byte swapping
on-the-fly.
'compatible byte order' would have to be the byte order used by the
native OS here. Meaning we pretty much do nothing, just use my latest
RFC patch.
As Geert also pointed out, the block layer mechanism to effect byte
swapping an entire disk or any partition appears to have been lost. This
has to be revived before we can think about compatibility schemes.
Cheers,
Michael
Thanks
Will
_________________________________________________________________________
William R Sowerbutts will@xxxxxxxxxxxxxx
"Carpe post meridiem" http://sowerbutts.com
main(){char*s=">#=0> ^#X@#@^7=",c=0,m;for(;c<15;c++)for
(m=-1;m<7;putchar(m++/6&c%3/2?10:s[c]-31&1<<m?42:32));}
