When large images, like SDK images, are booted as live images (hddimg) they often run out of memory, the OOM killer kicks in and kills random things, resulting in a variety of odd boot failures. These images should be installed via something like mkefidisk or wic first and not booted as live images. An early detection and consistent sort of warning at boot time for live images would be helpful and possibly reduce the number of times this "bug" gets reported against various BSPs.
We suspect some of the issues may have been as a result of the issue in http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=5c71337cd79b03322f39e56c22c3e7634a2d9abd which is now resolved (as an error). That has perhaps removed some of this kind of issue. Moving to unassigned as I doubt Brendan will get to this soon. Checking the system memory size in the live image would be good to ensure we have enough space to run as a live image.
I've opted to take a look at this issue as my introduction to Yocto bugs. I'm going to replicate the issue through QEMU before looking at creating some runtime scripts that will see if we have sufficient memory to run the live image. It'll probably just write something to dmesg as a warning if memory is low enough to start triggering the OOM_killer
In addition to ensuring that the image is created properly, we should add a check to the initramfs to ensure that the live image is smaller than available ram at runtime.
Merged into master: http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=8a2dfa16917aca821caf1a1d9b53faa3b025656c