On Tue, Jan 24, 2023 at 10:00 AM Ben Gardon <bgardon@xxxxxxxxxx> wrote: > > On Tue, Jan 24, 2023 at 8:32 AM Ricardo Koller <ricarkol@xxxxxxxxxx> wrote: > > > > On Tue, Jan 24, 2023 at 12:56:11AM +0000, Oliver Upton wrote: > > > On Mon, Jan 23, 2023 at 04:51:16PM -0800, Ben Gardon wrote: > > > > On Thu, Jan 12, 2023 at 7:50 PM Ricardo Koller <ricarkol@xxxxxxxxxx> wrote: > > > > > > > > > > Add a flag to kvm_pgtable_visit_ctx, KVM_PGTABLE_WALK_REMOVED, to > > > > > indicate that the walk is on a removed table not accesible to the HW > > > > > page-table walker. Then use it to avoid doing break-before-make or > > > > > performing CMOs (Cache Maintenance Operations) when mapping a removed > > > > > > > > Nit: Should this say unmapping? Or are we actually going to use this > > > > to map memory ? > > > > > > I think the *_REMOVED term feels weird as it relates to constructing a > > > page table. It'd be better if we instead added flags to describe the > > > operations we intend to elide (i.e. CMOs and TLBIs). > > > > What about KVM_PGTABLE_WALK_ELIDE_BBM and KVM_PGTABLE_WALK_ELIDE_CMO? > > I like this, but please don't use elide in the code. I'm all for > vocabulary, but that's not a common enough word to expect everyone to > know. Perhaps just SKIP? No problem, SKIP should be fine. > > > > > > That way the > > > implementation is generic enough that we can repurpose it for other use > > > cases. > > > > Aha, good point. I actually have a use case for it (FEAT_BBM). > > > > > > > > -- > > > Thanks, > > > Oliver
