Re: linux-ipsec: Re: FreeS/WAN allergic to egcs64 (SPARC)
Apparently "David S. Miller" <davem@redhat.com> wrote:
> It looks like conmakehash.c is also including stdio.h --- though
> it wasn't tripping over the __restrict problem like this.
> conmakehash is build with the user application compiler,
> ie. HOSTCC
> Later,
> David S. Miller
> davem@redhat.com
David,
I'd tried building the .../ipsec/libdes with gcc and with
gcc272 before building the rest of the kernel. I haven't
yet figured out how to hack KLIPS libdes to use HOSTCC
instead of egcs-sparc-64 and I have no confidence that
this would work.
I don't suppose its possible to compile a 32-bit kernel and
run it on an UltraSPARC. Is it? (I've tried a couple of
ways --- and it doesn't seem to work).
Incidentally I also tried unpacking a fresh copy of 2.2.14
and applying the patch-int-2.1.2.14.1 (international crypto
patch) to that. It seems to do a bit better though it has
basically the same problem with some of the modules there.
(I've unpacked several different copies and started
fresh a couple of different times trying different options
--- just to be sure that I wasn't bumping into some artifact of
my earlier failures).
I can give you and others (as needed) access to this
system via ssh, if you're willing to take a look at it.
I'm willing to switch to Red Hat (SPARC) if the tool chain
included with that can cope with this.
I've been looking at pipsecd --- which seems to install O.K.
(though I have to get the right version of the userlink.o
kernel modules build and installed to actually test it).
Unfortunately I haven't seen any response from anyone on
the linux-ipsec team. I know that the message got to their
list, since I see it in their archives, as linked from their
own web site. I see that another question about SPARC issues
has been ignored as well.
Do they believe in a PC centric world out there? Seems like
an odd bias to me; particularly considering the background
of their principle sponsor.
Of course the FreeS/WAN patches are completely separate from
the international crypto patch. I don't suppose it would make
sense to integrate the two in some way. Maybe we'd have one
implementation of the relevant DES, IDEA, 3DES, MD5 and SHA-1
algorithms. Maybe FreeS/WAN could then be more easily extended
to use Twofish, or whatever the AES winner turns out to be.
Anyway, David, Jakub, and Anton,
Is there any chance that I might get this working if I
were to get the latest egcs out of CVS (and/or upgrade to
a new glibc?). Would any of you like access to this box?
(I can to a scratch new install if you like. This was a
Debian Potato using Ben Collins 60Mb ISO image and an
apt-get update && apt-get upgrade. I'll see if I can
dig up a Red Hat if that would help.
Is it just that SPARC is a "third class citizen" and that
I'm expecting too much?
--
Jim Dennis jdennis@linuxcare.com
Linuxcare: Linux Corporate Support Team: http://www.linuxcare.com
Reply to: