Am 01.08.2012 22:13, schrieb Marti Raudsepp:
On Wed, Aug 1, 2012 at 10:47 PM, Thomas Bächler <thomas@archlinux.org> wrote:
This might be unavoidable.
Try lvmwait=/dev/sdXY on the command line, where sdXY is the partition that holds the LVM PV.
So you're saying that Linux 3.5 didn't introduce any bugs, but simply boots up so fast that the disk isn't initialized yet by the time the initrd tries to activate LVM?
Yes, this has happened before.
Couldn't this be fixed in initrd to try again every second for 5 seconds or so, if the root device hasn't appeared? (Or maybe monitoring udev events, although that sounds much more complicated)
We do this already, for every device but for LVM and RAID. For RAID, there is the auto-assembly with the mdadm_udev hook. For LVM2, I wrote something similar once, but it had problems. If you want the initramfs to wait for the devices, use the lvmwait parameter and you should be fine.