The LLVM/Clang compiler has merged support for AMD’s RMPCHKD, RMPREAD, and RMPOPT instructions, designed for Zen 5 and newer processors. These instructions aim to optimize performance on servers utilizing AMD’s Secure Encrypted Virtualization – Scalable Nested Paging (SEV-SNP) technology.
The v13 patches enabling RMPOPT were released last week. RMPOPT is designed to minimize performance overhead associated with Reverse Map Table (RMP) checks on the hypervisor, and for non-SNP guests, by allowing these checks to be skipped when 1GB regions of memory are confirmed to not contain SEV-SNP guest memory. RMPREAD is the instruction used for reading Reverse Map Table entries, while RMPCHKD checks the dirty status bit on guest private pages within the Reverse Map Table.
Initially, RMPOPT was anticipated to debut with Zen 6 processors, but it is already supported on Zen 5 EPYC processors. The software enablement is described as belated. The code is being considered for inclusion in the Linux v7.4 merge window in late October. The merged pull request to the LLVM toolchain Git code was labeled “support further Zen5 and Zen6 instructions.”
Read the original coverage
💬 Comments
📜 Comment Policy