RE: [PATCH v10 052/108] KVM: x86/tdp_mmu: Ignore unsupported mmu operation on private GFNs

From: Wang, Wei W
Date: Mon Dec 05 2022 - 09:24:14 EST


On Sunday, October 30, 2022 2:23 PM, Yamahata, Isaku wrote:
> From: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
>
> Some KVM MMU operations (dirty page logging, page migration, aging page)
> aren't supported for private GFNs (yet) with the first generation of TDX.
> Silently return on unsupported TDX KVM MMU operations.
>
> Signed-off-by: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> ---
> arch/x86/kvm/mmu/mmu.c | 3 ++
> arch/x86/kvm/mmu/tdp_mmu.c | 73
> +++++++++++++++++++++++++++++++++++---
> arch/x86/kvm/x86.c | 3 ++
> 3 files changed, 74 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c index
> 02e7b5cf3231..efc3b3f2dd12 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -6588,6 +6588,9 @@ static bool kvm_mmu_zap_collapsible_spte(struct
> kvm *kvm,
> for_each_rmap_spte(rmap_head, &iter, sptep) {
> sp = sptep_to_sp(sptep);
>
> + /* Private page dirty logging is not supported yet. */
> + KVM_BUG_ON(is_private_sptep(sptep), kvm);
> +
> /*
> * We cannot do huge page mapping for indirect shadow pages,
> * which are found on the last rmap (level = 1) when not using diff --git
> a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c index
> 0e053b96444a..4b207ce83ffe 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -1469,7 +1469,8 @@ typedef bool (*tdp_handler_t)(struct kvm *kvm,
> struct tdp_iter *iter,
>
> static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm,
> struct kvm_gfn_range *range,
> - tdp_handler_t handler)
> + tdp_handler_t handler,
> + bool only_shared)
> {
> struct kvm_mmu_page *root;
> struct tdp_iter iter;
> @@ -1480,9 +1481,23 @@ static __always_inline bool
> kvm_tdp_mmu_handle_gfn(struct kvm *kvm,
> * into this helper allow blocking; it'd be dead, wasteful code.
> */
> for_each_tdp_mmu_root(kvm, root, range->slot->as_id) {
> + gfn_t start;
> + gfn_t end;
> +
> + if (only_shared && is_private_sp(root))
> + continue;
> +
> rcu_read_lock();
>
> - tdp_root_for_each_leaf_pte(iter, root, range->start, range->end)
> + /*
> + * For TDX shared mapping, set GFN shared bit to the range,
> + * so the handler() doesn't need to set it, to avoid duplicated
> + * code in multiple handler()s.
> + */
> + start = kvm_gfn_for_root(kvm, root, range->start);
> + end = kvm_gfn_for_root(kvm, root, range->end);
> +
> + tdp_root_for_each_leaf_pte(iter, root, start, end)
> ret |= handler(kvm, &iter, range);
>
> rcu_read_unlock();
> @@ -1526,7 +1541,12 @@ static bool age_gfn_range(struct kvm *kvm, struct
> tdp_iter *iter,
>
> bool kvm_tdp_mmu_age_gfn_range(struct kvm *kvm, struct kvm_gfn_range
> *range) {
> - return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range);
> + /*
> + * First TDX generation doesn't support clearing A bit for private
> + * mapping, since there's no secure EPT API to support it. However
> + * it's a legitimate request for TDX guest.
> + */
> + return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range, true);
> }
>
> static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter, @@ -1537,7
> +1557,8 @@ static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter,
>
> bool kvm_tdp_mmu_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range
> *range) {
> - return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn);
> + /* The first TDX generation doesn't support A bit. */
> + return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn, true);
> }
>
> static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter, @@ -1582,8
> +1603,11 @@ bool kvm_tdp_mmu_set_spte_gfn(struct kvm *kvm, struct
> kvm_gfn_range *range)
> * No need to handle the remote TLB flush under RCU protection, the
> * target SPTE _must_ be a leaf SPTE, i.e. cannot result in freeing a
> * shadow page. See the WARN on pfn_changed in
> __handle_changed_spte().
> + *
> + * .change_pte() callback should not happen for private page, because
> + * for now TDX private pages are pinned during VM's life time.
> */
> - return kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn);
> + return kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn, true);
> }
>
> /*
> @@ -1637,6 +1661,14 @@ bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm,
>
> lockdep_assert_held_read(&kvm->mmu_lock);
>
> + /*
> + * Because first TDX generation doesn't support write protecting private
> + * mappings and kvm_arch_dirty_log_supported(kvm) = false, it's a bug
> + * to reach here for guest TD.
> + */
> + if (WARN_ON_ONCE(!kvm_arch_dirty_log_supported(kvm)))
> + return false;
> +
> for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
> spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn,
> slot->base_gfn + slot->npages, min_level); @@ -1902,6
> +1934,14 @@ bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm,
>
> lockdep_assert_held_read(&kvm->mmu_lock);
>
> + /*
> + * First TDX generation doesn't support clearing dirty bit,
> + * since there's no secure EPT API to support it. It is a
> + * bug to reach here for TDX guest.
> + */
> + if (WARN_ON_ONCE(!kvm_arch_dirty_log_supported(kvm)))
> + return false;
> +

It might not be a good choice to intercept everywhere in kvm_mmu just as tdx
doesn't support it. I'm thinking maybe we could do the check in tdx.c, which is
much simpler. For example:

@@ -2592,6 +2605,12 @@ static void tdx_handle_changed_private_spte(struct kvm *kvm,
lockdep_assert_held(&kvm->mmu_lock);

if (change->new.is_present) {
+ /* Only flags change. This isn't supported currently. */
+ KVM_BUG_ON(change->old.is_present, kvm);

Then we can have kvm_arch_dirty_log_supported completely removed.