On Thu, Mar 15, 2012 at 10:05:04AM +1100, Russell Coker wrote: > On Thu, 15 Mar 2012, Kurt Roeckx <[email protected]> wrote: > > > # eu-findtextrel /usr/lib/i386-linux-gnu/i686/cmov/libcrypto.so.1.0.0 > > > either the file containing the function 'OPENSSL_DIR_end' or the file > > > containing the function 'OPENSSL_cleanse' is not compiled with > > > -fpic/-fPIC either the file containing the function 'OPENSSL_DIR_end' or > > > the file containing the function 'OPENSSL_cleanse' is not compiled with > > > -fpic/-fPIC either the file containing the function 'OPENSSL_DIR_end' or > > > the file containing the function 'OPENSSL_cleanse' is not compiled with > > > -fpic/-fPIC > > > > Which is weird since nothing seems to have changed related to it. > > root@unstable32:/usr/src/openssl-1.0.0h# eu-findtextrel > ./i686/cmov/libcrypto.so > /usr/src/openssl-1.0.0h/crypto/x86cpuid.s not compiled with -fpic/-fPIC > root@unstable32:/usr/src/openssl-1.0.0h# eu-findtextrel ./i586/libcrypto.so > /usr/src/openssl-1.0.0h/crypto/x86cpuid.s not compiled with -fpic/-fPIC > > From the above test on a build tree it seems that the x86cpuid.s file is the > problem, but the detail is apparently stripped when the .deb is produced.
If you install libssl1.0.0-dbg and rename /usr/lib/debug/usr/lib/i386-linux-gnu/libcrypto.so.1.0.0 to /usr/lib/debug/usr/lib/i386-linux-gnu/libcrypto.so.1.0.0.debug It will say the same. This look like a regression in elfutils. Anyway, the problem is the usage of OPENSSL_ia32cap_P in x86cpuid as far as I can see. > > I do know what this all means, text relocations really aren't > > acceptable. > > That is the first time I've received such a response from a DD. In the past > I've seen little indication that maintainers of such packages know what it's > about and a great aversion to fixing things. This is one of the reasons policy says to build with -fPIC. Kurt -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
