This is a very long term goal, we may need more than one tickets to track the whole project if to let it land.  It may also involve not only QEMU but kernel changes. However let's use this ticket to start tracking this effort.

This ticket tracks the possibility of supporting guest-memfd 1G huge pages to be live migrated using postcopy.

QEMU used to support either 1G huge pages (based on hugetlbfs), or postcopy on generic types of memory.  Postcopy can functionally work with 1G huge pages, but it suffers from a major performance issue so it's almost not usable.

It looks like guest-memfd will at some point start to support 1G huge pages.  There is a chance we could start to support postcopy on 1G huge pages using guest-memfd.

When enabled, the VM can benefit from both the high TLB hit rate due to 1G huge pages, meanwhile allows live migration to always converge with the help of postcopy, being able to remotely request pages.

One major use case can be SAP HANA.  SAP HANA requires 1G huge pages in the VMs as a baseline.  It's also one of the known workload that is easily heavy enough so live migration may not converge when the database is running.  With 1G guest-memfd postcopy support, it'll both satisfy the SAP demand on 1G page utilization, meanwhile provide a way to guarantee live migration convergence.