[PATCH] Adding shared code support for ia32 and amd64 -- x86 sub-branch

[M. Warner Losh]

In message: <20100216.135906.812628220523048669.imp at bsdimp.com>
            "M. Warner Losh" <imp at bsdimp.com> writes:
: In message: <3bbf2fe11002151610l41526f55r5e60b5e46ce42b64 at mail.gmail.com>
:             Attilio Rao <attilio at FreeBSD.org> writes:
: : There are few things to discuss. One, that I had not necessity to dig
: : about still, is about how to organize headers (include/). Maybe some
: : replication ala pc98 may be good.
: 
: Most of the pc98 replication these days is in the form of "#include
: <i386/foo.h>" and we install i386/include into /usr/include/i386 as
: well as pc98/include into /usr/include/machine.
: 
: I'm working in a similar vein.  For a while, one has had to specify
: TARGET_BIG_ENDIAN in order to build big endian MIPS or ARM binaries.
: This is lame.  I've eliminated it and have made it possible to build
: mipsel, mipseb, arm or armeb binaries by setting TARGET_ARCH to one of
: these, while still sharing all the mips code in one directory and all
: the arm code in another.

You can see the results of this work in
	http://people.freebsd.org/~imp/tbemd.diff

the patch is a little rough, and there's an issue with make universe
for mips kernels still that I need to track down[*], but otherwise it
looks good.

Warner

[*] Well, and a bigger, meta question: if we're building all the
kernels for a given MACHINE setting, and that MACHINE setting supports
both endians, how do we cope (short of putting back in the big endian
hacks I'm trying to eliminate).