On October 1, 2026, a report said AMD’s five-patch v16 RMPOPT series had been queued in the Linux tip tree’s x86/sev branch ahead of the Linux 7.4 merge window. The series proposes reducing Reverse Map Table (RMP) check overhead on systems using SEV-SNP, AMD’s technology for confidential virtual machines. The report forecast submission during the merge window later in October and a stable Linux 7.4 release around late December.
What RMPOPT proposes to change in SEV-SNP
RMP checks could be skipped for a 1 GB memory region only after verifying that every RMP entry in that region is owned by the hypervisor, rather than assigned to an SEV-SNP guest. That ownership check is the gate: the proposal targets regions that do not contain SNP guest memory.
The aim is to reduce RMP-check overhead for hypervisors and non-SNP guests on SEV-SNP systems. The optimization would run asynchronously, rather than as immediate work on the guest-termination path.
The v16 proposal’s memory scope and asynchronous work
The v16 series proposes coverage of up to 2 TB of system RAM. The AMD patch author deferred coverage beyond 2 TB to a follow-on series.
The proposal schedules optimization work after a 10-second delay so that terminations of multiple SNP guests can be batched. After a guest shuts down and its pages return to shared memory, the proposal would allow RMP optimization to be re-enabled.
The series also proposes disabling CPU hotplug while SNP is active, keeping the CPU set stable for RMPOPT setup and asynchronous scans.
