The document describes an issue where VMs were soft locking up on a host machine after numad was enabled and several VMs were running on the same host. Investigation found that the KSM page merging list was extremely long, with over 2 million entries merged into a single page, consuming around 9.2GB of memory. This was causing tasks like ksmd, khugepaged, and qemu-system-x86 to wait on locks and become unresponsive. Addressing this likely requires limits on the size of the KSM page merging list to prevent a single page from dominating memory usage.