Re: [Qemu-devel] [RFC] Streamlining endian handling in TCG

[Richard Henderson]

On 08/28/2013 10:28 AM, Peter Maydell wrote:
> Well, what do ldl_p or ldl_phys or any of the other functions
> without an le/be qualifier mean any more if "is this CPU in
> big endian mode?" now requires you to have a CPUState
> to ask? I guess we can tackle that separately from getting
> the assumptions on endianness out of the tcg backends
> though.

ldl_p is the host native endian unaligned load primitive.

ldl_phys is the one that'll need to be adjusted or eliminated.

>From a quick glance, this doesn't look too bad.  Most of the uses are in the
target-* directories already, and mostly for emulating page tables.  Those
should be trivial to adjust as needed.

The troublesome offender appears to be virtio.  I know some discussion of
cleaning that up has happened in the context of LE PowerPC, but I didn't really
follow what went on there.


r~