On Mon, Jun 27, 2022 at 02:53:41PM -0700, isaku.yamahata@xxxxxxxxx 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/tdp_mmu.c | 74 +++++++++++++++++++++++++++++++++++---
> arch/x86/kvm/x86.c | 3 ++
> 2 files changed, 72 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index 12f75e60a254..fef6246086a8 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -387,6 +387,8 @@ static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn,
>
> if ((!is_writable_pte(old_spte) || pfn_changed) &&
> is_writable_pte(new_spte)) {
> + /* For memory slot operations, use GFN without aliasing */
> + gfn = gfn & ~kvm_gfn_shared_mask(kvm);
This should be part of enabling, please consider to squash it into patch 46.
> slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn);
> mark_page_dirty_in_slot(kvm, slot, gfn);
> }
> @@ -1398,7 +1400,8 @@ 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;
> @@ -1409,9 +1412,23 @@ 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();
> @@ -1455,7 +1472,12 @@ 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,
> @@ -1466,7 +1488,7 @@ 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);
> + return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn, false);
The "false" here means we will do young testing for even private
pages, but we don't have actual A bit state in iter->old_spte for
them, so may here should be "true" ?
> }
>
> static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter,
> @@ -1511,8 +1533,11 @@ 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.
> */
Worth to catch this by WARN_ON() ? Depends on you.
> - return kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn);
> + return kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn, true);
> }
>
> /*
> @@ -1566,6 +1591,14 @@ 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(!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);
> @@ -1830,6 +1863,14 @@ 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(!kvm_arch_dirty_log_supported(kvm)))
> + return false;
> +
> for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
> spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn,
> slot->base_gfn + slot->npages);
> @@ -1896,6 +1937,13 @@ void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm,
> struct kvm_mmu_page *root;
>
> lockdep_assert_held_write(&kvm->mmu_lock);
> + /*
> + * First TDX generation doesn't support clearing dirty bit,
> + * since there's no secure EPT API to support it. For now silently
> + * ignore KVM_CLEAR_DIRTY_LOG.
> + */
> + if (!kvm_arch_dirty_log_supported(kvm))
> + return;
> for_each_tdp_mmu_root(kvm, root, slot->as_id)
> clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot);
> }
> @@ -1975,6 +2023,13 @@ void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm,
>
> lockdep_assert_held_read(&kvm->mmu_lock);
>
> + /*
> + * This should only be reachable when diryt-log is supported. It's a
> + * bug to reach here.
> + */
> + if (WARN_ON(!kvm_arch_dirty_log_supported(kvm)))
> + return;
> +
> for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
> zap_collapsible_spte_range(kvm, root, slot);
> }
> @@ -2028,6 +2083,15 @@ bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
> bool spte_set = false;
>
> lockdep_assert_held_write(&kvm->mmu_lock);
> +
> + /*
> + * First TDX generation doesn't support write protecting private
> + * mappings, silently ignore the request. KVM_GET_DIRTY_LOG etc
> + * can reach here, no warning.
> + */
> + if (!kvm_arch_dirty_log_supported(kvm))
> + return false;
> +
> for_each_tdp_mmu_root(kvm, root, slot->as_id)
> spte_set |= write_protect_gfn(kvm, root, gfn, min_level);
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index dcd1f5e2ba05..8f57dfb2a8c9 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -12243,6 +12243,9 @@ static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
> u32 new_flags = new ? new->flags : 0;
> bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
>
> + if (!kvm_arch_dirty_log_supported(kvm) && log_dirty_pages)
> + return;
> +
> /*
> * Update CPU dirty logging if dirty logging is being toggled. This
> * applies to all operations.
> --
> 2.25.1
>
