This seems to be caused by the call to "timeval_inject_offset_valid()" introduced in kernel/time/ntp.c (in "ntp_validate_timex()") by 2ed8b5bac95a96e2334a6b9bcbe99eabaf83f931. timeval_inject_offset_valid only handles timevals with microseconds, but systemd-timesyncd uses the ADJ_NANO mode which stores nanosecond values in tv_usec. upstream 4.5 already has a fix to handle this situation by copying the tv content to a (temporary) timespec and validating that iff ADJ_NANO is used.
This seems to be caused by the call to "timeval_ inject_ offset_ valid() " introduced in kernel/time/ntp.c (in "ntp_validate_ timex() ") by 2ed8b5bac95a96e 2334a6b9bcbe99e abaf83f931. timeval_ inject_ offset_ valid only handles timevals with microseconds, but systemd-timesyncd uses the ADJ_NANO mode which stores nanosecond values in tv_usec. upstream 4.5 already has a fix to handle this situation by copying the tv content to a (temporary) timespec and validating that iff ADJ_NANO is used.