On Wed, Jul 27, 2022 at 02:56:40PM +0530, Nikunj A. Dadhania wrote: > On 7/26/2022 8:02 PM, Chao Peng wrote: > > On Mon, Jul 25, 2022 at 07:16:24PM +0530, Nikunj A. Dadhania wrote: > >> On 7/20/2022 8:29 PM, Chao Peng wrote: > >>> On Thu, Jul 14, 2022 at 01:03:46AM +0000, Sean Christopherson wrote: > >>> ... > >>>> > >>>> Option D). track shared regions in an Xarray, update kvm_arch_memory_slot.lpage_info > >>>> on insertion/removal to (dis)allow hugepages as needed. > >>>> > >>>> + efficient on KVM page fault (no new lookups) > >>>> + zero memory overhead (assuming KVM has to eat the cost of the Xarray anyways) > >>>> + straightforward to implement > >>>> + can (and should) be merged as part of the UPM series > >>>> > >>>> I believe xa_for_each_range() can be used to see if a given 2mb/1gb range is > >>>> completely covered (fully shared) or not covered at all (fully private), but I'm > >>>> not 100% certain that xa_for_each_range() works the way I think it does. > >>> > >>> Hi Sean, > >>> > >>> Below is the implementation to support 2M as you mentioned as option D. > >>> It's based on UPM v7 xarray code: https://lkml.org/lkml/2022/7/6/259 > >>> > >>> Everything sounds good, the only trick bit is inc/dec disallow_lpage. If > >>> we still treat it as a count, it will be a challenge to make the inc/dec > >>> balanced. So in this patch I stole a bit for the purpose, looks ugly. > >>> > >>> Any feedback is welcome. > >>> > >>> Thanks, > >>> Chao > >>> > >>> ----------------------------------------------------------------------- > >>> From: Chao Peng <chao.p.peng@xxxxxxxxxxxxxxx> > >>> Date: Wed, 20 Jul 2022 11:37:18 +0800 > >>> Subject: [PATCH] KVM: Add large page support for private memory > >>> > >>> Update lpage_info when handling KVM_MEMORY_ENCRYPT_{UN,}REG_REGION. > >>> > >>> Reserve a bit in disallow_lpage to indicate a large page has > >>> private/share pages mixed. > >>> > >>> Signed-off-by: Chao Peng <chao.p.peng@xxxxxxxxxxxxxxx> > >>> --- > >> > >> > >>> +static void update_mem_lpage_info(struct kvm *kvm, > >>> + struct kvm_memory_slot *slot, > >>> + unsigned int attr, > >>> + gfn_t start, gfn_t end) > >>> +{ > >>> + unsigned long lpage_start, lpage_end; > >>> + unsigned long gfn, pages, mask; > >>> + int level; > >>> + > >>> + for (level = PG_LEVEL_2M; level <= KVM_MAX_HUGEPAGE_LEVEL; level++) { > >>> + pages = KVM_PAGES_PER_HPAGE(level); > >>> + mask = ~(pages - 1); > >>> + lpage_start = start & mask; > >>> + lpage_end = end & mask; > >>> + > >>> + /* > >>> + * We only need to scan the head and tail page, for middle pages > >>> + * we know they are not mixed. > >>> + */ > >>> + update_mixed(lpage_info_slot(lpage_start, slot, level), > >>> + mem_attr_is_mixed(kvm, attr, lpage_start, > >>> + lpage_start + pages)); > >>> + > >>> + if (lpage_start == lpage_end) > >>> + return; > >>> + > >>> + for (gfn = lpage_start + pages; gfn < lpage_end; gfn += pages) { > >>> + update_mixed(lpage_info_slot(gfn, slot, level), false); > >>> + } > >> > >> Boundary check missing here for the case when gfn reaches lpage_end. > >> > >> if (gfn == lpage_end) > >> return; > > > > In this case, it's actually the tail page that I want to scan for with > > below code. > > What if you do not have the tail lpage? > > For example: memslot base_gfn = 0x1000 and npages is 0x800, so memslot range > is 0x1000 to 0x17ff. > > Assume a case when this function is called with start = 1000 and end = 1800. > For 2M, page mask is 0x1ff. start and end both are 2M aligned. > > First update_mixed takes care of 0x1000-0x1200 > Loop update_mixed: goes over from 0x1200 - 0x1800, there are no pages left > for last update_mixed to process. Oops, good catch. I would fix it differently by playing with lpage_end: lpage_end = (end - 1) & mask; Thanks, Chao > > > > > It's also possible I misunderstand something here. > > > > Chao > >> > >>> + > >>> + update_mixed(lpage_info_slot(lpage_end, slot, level), > >>> + mem_attr_is_mixed(kvm, attr, lpage_end, > >>> + lpage_end + pages)); > > lpage_info_slot some times causes a crash, as I noticed that > lpage_info_slot() returns out-of-bound index. > > Regards > Nikunj >
