On 02/10/2012 01:41 PM, Stefan Hajnoczi wrote:
At this early stage we still have the option of designing the device
to be little- or big-endian. I'm in favour of a fixed endianness even
though existing virtio devices tend to use guest-endian.
What do you think?
Interesting idea, but I'd say no for three reasons:
1) The virtqueue would remain in guest-endianness.
2) Even if the result is better, it is still different from every other
device, so right now it would be a one-off exception (and it would
likely remain so; it's not like there are new virtio devices every week :)).
3) I'm afraid of breaking something else, for example pSeries virtio is
doing complicated things that I don't really understand to split the I/O
BAR in two parts, the first in PCI endianness and the second in native
endianness.
Paolo
