Hello, Rulinhuang! > > Hi Rezki, we have submitted patch v2 to avoid the partial > initialization issue of vm's members and separated insert_vmap_area() > from alloc_vmap_area() so that setup_vmalloc_vm() has no need to > require lock protection. We have also verified the improvement of > 6.3% at 160 vcpu on intel icelake platform with this patch. > Thank you for your patience. > Please work on the mm-unstable and try to measure it on the latest tip. -- Uladzislau Rezki