PUCK Notes - 2024.01.03 - Post-copy for guest_memfd()
Sean Christopherson
reviewerRecording and slides of the post-copy for guest_memfd discussion:
Key takeaways:
- There is no known use case for tying "data invalid" to the guest_memfd inode, i.e. letting each "view" (struct file / struct kvm / VM) control its own behavior is acceptable, and arguably desirable as it provides userspace more flexibility.
- If the "data invalid" attribute is tied to a given view, then routing the ioctl() through KVM provides a superset of functionality compared to making a guest_memfd specific ioctl(), e.g. KVM can apply the attribute to any gfn regardless of backing store.
- Pursuing a fully generic file-based solution is undesirable as there are a large number of questions that need answers, and there is no known use case beyond KVM to drive those discussions.
Feel free to chime in with anything important that I missed.