Re: disklabel endian issues

[Lloyd Parkes <lloyd%must-have-coffee.gen.nz@localhost>]

On 30/06/2009, at 11:47 PM, Izumi Tsutsui wrote:

> (1) When byteswap should be done
>
> For crossbuild toolchain, we can deteremine both
> target's endian and host's endian at compile time.

That's a good point.

> So, struct disklabel in target's endian is required only in buffers
> to read from/write to a target regular file specified by `-F' option,
> i.e. byteswap ops should be done only on copying disklabel between
> the buffer and struct disklabel for the host to calculate parameters.
>
> Is this right?  If so, there is no hard thing there.

Right. The actual byte swapping shouldn't be too hard.

> (2) How byteswap should be done
>
> First, the struct disklabel contains some data in char arrays,
> and those members must not be swapped in the byteswap functions.

All the members that need swapping are uint16 and uint32, so that it  
really is just a lot of typing.

> Second, we have to consider how d_checksum member should be handled.
> d_checksum is calculated by XOR in uint16_t, but as noted above
> not all even/odd bytes are swapped, so d_checksum must be
> (re)calculated after all other non-char members are properly swapped,
> and no more swap is needed for a recalculated sum.

I remembered the checksum this morning. Since it's only an XOR, we  
might not need to recalculate the checksum. I'm pretty sure XOR is  
associative and commutative.

> I hope src/sys/arch/sh3/sh3/disklabel.c might help.

I'll check it out.


> martin%duskware.de@localhost wrote:
> > And where to find/store the label ;-}

The disklabel program is built as a target specific binary, so I'm  
assuming that someone else has taken care of this. I also assumed that  
someone else had taken care of the byte swapping, so ...

> > (But isn't this the wrong list?)
>
> It depends?

There isn't a tech-tools for hosted tools issues, and the toolchain is  
the largest hosted tool.

Cheers,
Lloyd