Jeremy Fitzhardinge wrote: > Zachary Amsden wrote: > >> Actually, vmalloc mapping replication is not quite straightforward - >> you have to replicate it under locks and track all pgds using pgd_list >> if you support large pages. >> > > But that machinery already exists. It would continue to work under PAE > wouldn't it? > No, it doesn't exist under PAE, since PAE uses a common pgd[3] therefore the init_mm pgd never needs changing for large page / split page PD transitions. The code is there for non-PAE, but compiled out of the PAE version. Zach