[PATCH RFC v1 2/5] KVM: guest_memfd: Only access KVM memory attributes when appropriate

[Michael Roth <michael.roth@amd.com>]

When a memslot is configured with KVM_MEMSLOT_SUPPORTS_GMEM_SHARED, the
KVM MMU will not rely on KVM's memory attribute tracking to determine
whether a page is shared/private, but will instead call into guest_memfd
to obtain this information.

In the case of kvm_gmem_populate(), KVM's memory attributes are used to
determine the max order for pages that will be used for the guest's
initial memory payload, but this information will not be valid if
KVM_MEMSLOT_SUPPORTS_GMEM_SHARED is set, so update the handling to
account for this. Just hard-code the order to 0 for now since there
isn't yet hugepage support in guest_memfd.

Signed-off-by: Michael Roth <michael.roth@amd.com>
---
 virt/kvm/guest_memfd.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
index cc93c502b5d8..b77cdccd340e 100644
--- a/virt/kvm/guest_memfd.c
+++ b/virt/kvm/guest_memfd.c
@@ -1429,12 +1429,16 @@ long kvm_gmem_populate(struct kvm *kvm, gfn_t start_gfn, void __user *src, long
 			(npages - i) < (1 << max_order));
 
 		ret = -EINVAL;
-		while (!kvm_range_has_memory_attributes(kvm, gfn, gfn + (1 << max_order),
-							KVM_MEMORY_ATTRIBUTE_PRIVATE,
-							KVM_MEMORY_ATTRIBUTE_PRIVATE)) {
-			if (!max_order)
-				goto put_folio_and_exit;
-			max_order--;
+		if (!kvm_gmem_memslot_supports_shared(slot)) {
+			while (!kvm_range_has_memory_attributes(kvm, gfn, gfn + (1 << max_order),
+								KVM_MEMORY_ATTRIBUTE_PRIVATE,
+								KVM_MEMORY_ATTRIBUTE_PRIVATE)) {
+				if (!max_order)
+					goto put_folio_and_exit;
+				max_order--;
+			}
+		} else {
+			max_order = 0;
 		}
 
 		p = src ? src + i * PAGE_SIZE : NULL;
-- 
2.25.1