On Thu, Jul 30 2026, Ackerley Tng wrote: > Tarun Sahu <[email protected]> writes: > >> Extract 'struct gmem_file', 'struct gmem_inode', and GMEM_I() from >> virt/kvm/guest_memfd.c into a new internal header virt/kvm/guest_memfd.h. >> Also split __kvm_gmem_create() to expose a non-static >> __kvm_gmem_create_file() helper that returns a 'struct file *' instead of >> an fd. >> > > This is done basically to support a new virt/kvm/guest_memfd_luo.c > file.
My $0.02 on this: I think it is useful to split out the live update code into a separate file. It makes for easier reviews, and puts a clear boundary between core and live update. We have done it for memfd/shmem for example (see mm/memfd_luo.c) and for other live update series in flight like PCI or IOMMU. > > Would like to know what Sean thinks of this! I'd like to know for future > guest_memfd work too. > >> These internal definitions and helpers allow upcoming guest_memfd Live >> Update Orchestrator (LUO) preservation code to access guest_memfd >> internals and reconstruct guest_memfd file instances from preserved state >> without installing them into a file descriptor table up front. >> >> Signed-off-by: Tarun Sahu <[email protected]> [...] -- Regards, Pratyush Yadav
