[Openswan Users] [Solved] Openswan 'errs' on PIII, but works fine on newer CPUs

[Neal Murphy]

Posting in case anyone else encounters similar trouble.

Re: below (I never sent it). Problem solved. For the moment. And it wasn't 
really an Openswan problem after all.

I was specifying "-m32 -march=i486". But pkgs were *still* picking up the 
host's CPU. I finally (effectively) added
  --build=i486-pc-linux-gnu --host=i486-pc-linux-gnu
to almost all of the pkgs' configure options. After dealing with old working 
code with new breakages and a bumping glibc to 2.18 to fix a bug, I now have 
an i486 version of Smoothwall 3.1 that boots, installs and runs on my PhenomII 
in a 486 KVM and the Y2K PIII; IPSEC VPN comes up right smartly. I'll find out 
Tuesday evening if it runs onae VIA CPU and if IPSEC works (working with 
someone else).

Next I'll build for i586 since most 'old' systems today should be at *least* 
Pentium Overdrive (i.e., there shouldn't be any i486 systems still running 
that even *could* run Smoothwall). And verify that x86_64 still builds.

In sum, I can now build Openswan (and the rest of the system) for i486 and 
have the tools actually produce i486 code. It should also work for i586 and 
i686.

N

>OK. Now I've run into another problem I cannot solve.
>
>I've built openswan 2.6.39 (and 2.6.38, same results) and Linux 3.4.71
>(all of Smoothwall 3.1, really) for i486; have to support the 'oldest'
>systems in use. I am nearly certain the system is being built correctly
>for -march=i486; some pkgs made it a challenge.
>
>The problem is that the tunnel will not come up when the system is running
>on a year 2000 PIII. The same system running in a KVM on a 'modern' CPU
>brings the VPN up as expected. The symptom was that I could see the MAIN_x1 
>packets leaving and entering (via tcpdump), but pluto misinterpreted and
>'dropped' all such incoming packets. Some code somewhere just isn't being
>made right.