Bad packet length error

[Pete Flugstad]

Hari wrote:
> I just tried with WORDS_BIGENDIAN defined and it helped solve this issue.
> Apparently this is what happened:
> I am using a cross-compiler for building openssh and I could not run
> configure on the cross-compiler environment. 
> I have initially run the
> configure on an x86 box and used the thus created config.h and Makefile for
> building in the cross-compiler env. At some point when I moved to power-pc,
> I failed to modify/obtain new config.h and Makefile.

Yup, same thing happened to me.  The only thing I found that
mattered was the WORDS_BIGENDIAN issue.  I had more issues
with OpenSSL for PPC as I'd left some RC4 defines turned on
(they were in X86 code) I should not have (PPC860 doesn't
like RC4_INDEX).

> How do we handle this for different platforms when using a cross-compiler ?
> Right now, I am doing an #ifdef _<PLATFORM>_ in config.h

The proper solution is to use the newer code - it appears to be
endian independant.  That eliminates this issue entirely.

So I would expect that with 3.6 you can run the configure script
on the X86 host and the generated config.h, etc will probably be
OK for the PPC system (since the endian issues will have been taken
care of), assuming similar/same OS.   You may need to edit config.h
for LynxOS to more accurately reflect what features it has defined
vs what Linux has defined, but I'd guess it'd be pretty close.

Pete Flugstad
Icon Labs

PS: Note that the 3.6.1p2 configure script still seems to
     setup WORDS_BIGENDIAN, even though nothing appears to
     use that #define or any other endian releated define
     that I can find.