version 1.7

Jussi Kivilinna jussi.kivilinna at iki.fi
Sun Mar 20 13:54:47 CET 2016


On 18.03.2016 11:21, Werner Koch wrote:
> On Fri, 12 Feb 2016 14:15, ametzler at bebt.de said:
> 
>> Out of the yet tried builds only one failed, PPC64:
> 
>> selftest for CFB failed - see syslog for details
>> pass 0, algo 4, mode 1, gcry_cipher_setkey failed: Selftest failed
> 
>> This is also present in 1.6 where we work around it with
>> http://sources.debian.net/src/libgcrypt20/1.6.5-2/debian/patches/20_fedora_libgcrypt-1.6.3-aliasing.patch/
> 
> which adds 
> 
>   __attribute__ ((optimize("no-strict-aliasing")))
> 
> to the inline functions of bufhelp.h.
> 
> 
> @jussi: I recall that you fixed some alignment problems - might that
> problem be related?
> 

Problem here was wrong assumption that unaligned memory accesses
can be done on PPC64. This is now solved by disabling unaligned
accesses on PowerPC:
http://git.gnupg.org/cgi-bin/gitweb.cgi?p=libgcrypt.git;a=commit;h=1da793d089b65ac8c1ead65dacb6b8699f5b6e69

The earlier alignment problems, if I remember right, were clang
undefined behaviour sanitizer warnings and fixed with:
http://git.gnupg.org/cgi-bin/gitweb.cgi?p=libgcrypt.git;a=commit;h=92fa5f16d69707e302c0f85b2e5e80af8dc037f1

-Jussi

> 
> Shalom-Salam,
> 
>    Werner
> 
> 




More information about the Gcrypt-devel mailing list