Hello Jari,
thanks for your quick reply.
Sorry, but I guess it's all me too blame. Should have done some more
research before posting here ;-)
Indeed my box is running Debian Etch 32bit ... which in the first place
explains my problems compiling with 64bit optimizations I guess ...
My quick guess was, that as long as I have CPU support for the K8 in the
kernel, that I am able to use those assembler optimizations, which was
obviously wrong.
Anyways, I will get a AMD X2 soon and then setup a new Debian AMD64
system from scratch. With that I should be able to use your AMD64 code,
right?
thanks again
Thorsten
Jari Ruusu wrote:
Thorsten Peter wrote:
I have been trying to build latest loop-aes from source today on my
AMD64 (Athlon 64 3000+) machine today and ran into these errors:
[snip]
AS [M] /usr/src/work/loop-AES-v3.2b/tmp-d-kbuild/aes-amd64.o
/usr/src/work/loop-AES-v3.2b/tmp-d-kbuild/aes-amd64.S: Assembler messages:
/usr/src/work/loop-AES-v3.2b/tmp-d-kbuild/aes-amd64.S:201: Error: bad
register name `%rsi)'
To me it looks like your compiler/assembler/linker tool chain is creating 32
bit binaries by default. Correct?
And you compiled your kernel overriding ARCH= and/or CROSS_COMPILE= kernel
Makefile defaults. Correct?
On 2.6 kernels, loop-AES Makefile calls kernel build system to
compile/assemble/link kernel modules. It needs exact same ARCH= and/or
CROSS_COMPILE= defaults overridden on make command line as the ones that you
used to compile your kernel.
If above guess is not correct, then please provide more info about your
build environment:
1) Fully 64 bit distro, or 32 bit distro with 64 bit kernel?
2) What distro?
3) gcc and binutils versions?
4) Output of "id" command.
-
Linux-crypto: cryptography in and on the Linux system
Archive: http://mail.nl.linux.org/linux-crypto/
