I was now able to reproduce it with the jessie schroot and I've done a trace etc: http://labs.seravo.fi/~otto/mariadb-repo/logs/powerpc/ I am chatting with upstream devs to get their analysis what might have broke in 10.0.20.