Hans-Christian Egtvedt wrote: > Matt Johnston wrote: >> On Tue, Aug 08, 2006 at 10:59:15AM +0200, Hans-Christian Egtvedt wrote:
<cut link to test application> >> I've put what I see as dump2.gz, it should be the same as >> you see on the workstation (at least for the calculations). >> >> For debugging, I'd compare the two traces (avr versus x86) >> using vimdiff or something similar. Find the first point >> where they begin to differ in their numeric output (the >> big-numbers printed etc), and then add more debugging >> statements before and after that to narrow down the point >> where they start to differ. (Note that some of the >> hexadecimal output will differ with endianness, those should >> be marked). > > I looked in your dump2-ppc vs. dump2, and found that in my x86 vs avr32 > I have something faulty with the lines below: > > X86 > mp_int mp_exptmod_fast X: > 57177274837176085710416002504255433114092271661685833986820045309615224297446777341551735779854568082929739086299104916069027366371956577873538725199184070295968371188748133537718214011550432576681564852689543774176794595038388421094953024485861852872341010938808415634567578860387138830446329229275810102561 > > AVR32 > mp_int mp_exptmod_fast X: > 26144629297143774191971619660891134373249478019090098695819374263386738314451768455401121680453697629379694697503955208976597983567979834942305834318644583895981375065787344042040457278746854017585464708420108154208341890971865764261601653839167444443692604372044120220262714803272834772806762057598126328370 > >> Have a look at libtommath/bn_mp_exptmod_fast.c for how I've >> added the debugging output there - it requires the extra >> #include up the top, and order of includes does matter. > > I'll have a look inside here to see what I find :) <snip logs> And then I have found the problem, and the solution. The avr32-linux-gcc compiler had a wrong implementation of the ror and rol functions. And the uClibc library had a bug in memcmp function. After these were corrected dropbear and dbclient is working nicely :) Thank you for the help with debugging. If there are any AVR32-users who need these patches, drop me an email. -- With kind regards, Hans-Christian Egtvedt Applications Engineer - AVR Applications Lab
