Matt Thomas <matt@lkg.dec.com> writes:

[Toru Nishimura wrote]

>>Although there is little possibility to have big-endian NetBSD/mips
>>in near future, it's surely wise to make NetBSD/mips endian neutral.

>This issue is similar to support native 64LP mode on R4000 (or better)
>and having an 32LP mode for applications.  You'll need multiple syscall
>tables (similar to emulations) and syscall converters.  

>I'd like to see both be possible.

I'm confused. Does "bi-endian" mean that /sys/arch mips supports
either big- or little-endian kernels, configured statically; or does
it mean support for running big-endian binaries on a machine with a
little-endian kernel(or vice-versa?)

The former is relatively straightforward.  The second is *much* more
complicated, especially where ``untyped'' data crosses syscall
boundaries (e.g., ioctl()).  I'm unsure whether simple syscall
converters can always DTRT there.

>BTW, did you know that ULTRIX actually run big-endian at the start
>of MIPS port?

Are you sure that was really ULTRIX, and not a mangled mipsCo OS?
Was it driving little-endian bus hardware?