From: Avi K. <av...@qu...> - 2008-01-25 06:35:54
|
Christoph Lameter wrote: > On Thu, 24 Jan 2008, Andrea Arcangeli wrote: > > >> I think you should consider if you can also build a rmap per-MM like >> KVM does and index it by the virtual address like KVM does. >> > > Yes we have that. > > If we have that then we do not need the mmu_notifier. > We could call it with a page parameter and then walk the KVM or XPmem > reverse map to directly find all the ptes we need to clear. There is no > need then to add a new field to the mm_struct. > The reason the new field is needed is because the Linux mm does not understand the secondary pte format and zapping protocol. Locating the secondary ptes is just a part of the problem. -- Any sufficiently difficult bug is indistinguishable from a feature. |