MAD24 316 Why TLBI matters on ARM server scalability issues we found and solutions

TLBI is the TLB invalidation which is needed for updating mappings and attributes of the page table, on ARM the TLBI will broadcast to every CPU core in the inner shareable domain, and we found that TLBI affects the scalability a lot on some workload (such as virtual machine) on ARM64 server with many CPU cores. In this presentation, we introduce some scalability mitigations for TLBI such as TLBI by range, batched TLBI in the Linux mainline kernel. Then present the scalability issue we encountered on our ARM server when running VMs, we limit the TLBI in VM cpu context domain then we got about 50% performance boost on a 96 core ARM server.