[PATCH RFC 3/3] kvm: Add additional check to determine if a page is refcounted

[Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>]

The function kvm_is_refcounted_page is used primarily to determine if KVM
is allowed to take a reference on the page. It was using the PG_reserved
flag to determine this previously, however in the case of DAX the page has
the PG_reserved flag set, but supports pinning by taking a reference on
the page. As such I have updated the check to add a special case for
ZONE_DEVICE pages that have the new support_refcount_pinning flag set.

Signed-off-by: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
---
 virt/kvm/kvm_main.c |   16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 5e666df5666d..2e7e9fbb67bf 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -148,8 +148,20 @@ __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
 
 bool kvm_is_refcounted_pfn(kvm_pfn_t pfn)
 {
-	if (pfn_valid(pfn))
-		return !PageReserved(pfn_to_page(pfn));
+	if (pfn_valid(pfn)) {
+		struct page *page = pfn_to_page(pfn);
+
+		/*
+		 * The reference count for MMIO pages are not updated.
+		 * Previously this was being tested for with just the
+		 * PageReserved check, however now ZONE_DEVICE pages may
+		 * also allow for the refcount to be updated for the sake
+		 * of pinning the pages so use the additional check provided
+		 * to determine if the reference count on the page can be
+		 * used to pin it.
+		 */
+		return !PageReserved(page) || is_device_pinnable_page(page);
+	}
 
 	return false;
 }