when mdadm is too fast for the kernel...

... you have to put options scsi_mod scan=sync somewhere in /etc/modprobe.d/ and regenerate the initrd. (thanks to Michal Ludvig in http://lists.debian.org/debian-boot/2010/11/msg00369.html) Just happened to me on my Sun Netra T1-200 after the upgrade from Lenny to Squeeze, which threw me back into busybox of the initrd when it could not find the root-fs (on raid) and where mdadm -A /dev/md0 worked just fine to confuse me.