Comment # 10 on bug 950857 from Andrei Borzenkov

(In reply to Marcus Meissner from comment #9)
> The systemd-journald.socket also will start the .service as soon as someone
> accesses the syslog socket. This might just have been haveged itself with it
> syslog() call, deadlocking it nicely. :/
> 

That's true. But it also true for haveged stdout/stderr which are connected to
journald socket. IOW unit that must be started before journald cannot have its
output connected to journald. I wonder if this could be a reason for failure to
start journald some people reported (I have observed it myself).