And indeed, setting max_loop to 32, upgrading again to TW's lvm2-2.02.141-64.2.x86_64, ensuring "use_lvmetad = 1" is still set, rebuilding initrd, followed by a reboot now succeeds... While my last question remains valid "shouldn't the dm-udev logic be able to deal with such slow scan results", I mean, shouldn't /dev/loop devices be excluded by a *host*'s lvm2 filter in the first place? Stuff like /dev/*/by-* is already excluded, I'll add /dev/loop*, too, and see how far I get with that.