PiBox » Xeon

The init scripts are updated, but this problem remains. I don't know what's going on yet, possibly that the devices are not mounted yet so firstboot's attempt to make symlinks fails.

This may be due to the following line in firstboot:

for part in $(ls -1 /dev/disk/by-label)

The subshell expansion might not be working as expected. I've changed it to the following.

for part in /dev/disk/by-label/*

Interestingly, shellcheck recommended this change but I ignored it.

rootfs rebuild in progress in xeon1 sandbox. Will test later.

That didn't work either, because there are no directories under /dev/disk during firstboot. Here is the firstboot log showing that.

Jan 18 08:50:37 (none) user.debug kernel: ### Running genModel
Jan 18 08:50:37 (none) user.debug kernel: ### Appending board type to pibox-version.
Jan 18 08:50:37 (none) user.debug kernel: ### Disks directories:
Jan 18 08:50:37 (none) user.debug kernel: ### 
Jan 18 08:50:37 (none) user.debug kernel: ### Disks by label:
Jan 18 08:50:37 (none) user.debug kernel: ### 
Jan 18 08:50:37 (none) user.debug kernel: ### Running symlink updates
Jan 18 08:50:37 (none) user.debug kernel: ### Symlink processing for *...
Jan 18 08:50:37 (none) user.debug kernel: ### Saving logs

So I need to either fix it so firstboot finds those directories or rerun the symlink creation on second boot. No modules are loaded on second boot so I don't know what's missing on firstboot. It might be problem with udev vs mdev processing since I use the latter for event processing with blockhandler.sh, but that seems unlikely.

I might be wrong about udev vs mdev

After reading this discussion I found that I can find the matching device name by the label, as in the following.

# blkid -L BOOT  
/dev/mmcblk1p1
# blkid -L rootfs
/dev/mmcblk1p2
# blkid -L data
/dev/mmcblk1p3

So this means I can use a different mechanism on Xeon than is used in PiBox systems to create these symlinks. I've tested this on Xeon after 2nd boot but will still need to verify it works on firstboot.