No need to check all of the conditions in __handle_changed_spte(). Aging
a gfn range implies resetting access bit or marking spte for access
tracking.

Use atomic operation to only reset those bits. This avoids checking many
conditions in __handle_changed_spte() API. Also, clean up code by
removing dead code and API parameters.

Signed-off-by: Vipin Sharma <vipinsh@google.com>
---
 arch/x86/kvm/mmu/tdp_mmu.c | 36 +++++++++++++++++++-----------------
 1 file changed, 19 insertions(+), 17 deletions(-)

diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index c895560244de..5d6e77554797 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -758,13 +758,6 @@ static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
 	_tdp_mmu_set_spte(kvm, iter, new_spte, true);
 }
 
-static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm,
-						 struct tdp_iter *iter,
-						 u64 new_spte)
-{
-	_tdp_mmu_set_spte(kvm, iter, new_spte, false);
-}
-
 #define tdp_root_for_each_pte(_iter, _root, _start, _end) \
 	for_each_tdp_pte(_iter, _root, _start, _end)
 
@@ -1251,32 +1244,41 @@ static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm,
 /*
  * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero
  * if any of the GFNs in the range have been accessed.
+ *
+ * No need to mark corresponding PFN as accessed as this call is coming from
+ * the clear_young() or clear_flush_young() notifier, which uses the return
+ * value to determine if the page has been accessed.
  */
 static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter,
 			  struct kvm_gfn_range *range)
 {
-	u64 new_spte = 0;
+	u64 new_spte;
 
 	/* If we have a non-accessed entry we don't need to change the pte. */
 	if (!is_accessed_spte(iter->old_spte))
 		return false;
 
-	new_spte = iter->old_spte;
-
-	if (spte_ad_enabled(new_spte)) {
-		new_spte &= ~shadow_accessed_mask;
+	if (spte_ad_enabled(iter->old_spte)) {
+		iter->old_spte  = tdp_mmu_clear_spte_bits(iter->sptep,
+							  iter->old_spte,
+							  shadow_accessed_mask,
+							  iter->level);
+		new_spte = iter->old_spte & ~shadow_accessed_mask;
 	} else {
+		new_spte = mark_spte_for_access_track(iter->old_spte);
+		iter->old_spte = kvm_tdp_mmu_write_spte(iter->sptep,
+							iter->old_spte, new_spte,
+							iter->level);
 		/*
 		 * Capture the dirty status of the page, so that it doesn't get
 		 * lost when the SPTE is marked for access tracking.
 		 */
-		if (is_writable_pte(new_spte))
-			kvm_set_pfn_dirty(spte_to_pfn(new_spte));
-
-		new_spte = mark_spte_for_access_track(new_spte);
+		if (is_writable_pte(iter->old_spte))
+			kvm_set_pfn_dirty(spte_to_pfn(iter->old_spte));
 	}
 
-	tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte);
+	trace_kvm_tdp_mmu_spte_changed(iter->as_id, iter->gfn, iter->level,
+				       iter->old_spte, new_spte);
 
 	return true;
 }
-- 
2.39.1.581.gbfd45094c4-goog

On Fri, Feb 10, 2023, Vipin Sharma wrote:
>  	} else {
> +		new_spte = mark_spte_for_access_track(iter->old_spte);
> +		iter->old_spte = kvm_tdp_mmu_write_spte(iter->sptep,
> +							iter->old_spte, new_spte,
> +							iter->level);
>  		/*
>  		 * Capture the dirty status of the page, so that it doesn't get
>  		 * lost when the SPTE is marked for access tracking.
>  		 */
> -		if (is_writable_pte(new_spte))
> -			kvm_set_pfn_dirty(spte_to_pfn(new_spte));
> -
> -		new_spte = mark_spte_for_access_track(new_spte);
> +		if (is_writable_pte(iter->old_spte))
> +			kvm_set_pfn_dirty(spte_to_pfn(iter->old_spte));

Moving this block below kvm_tdp_mmu_write_spte() is an unrelated change.  Much to
my chagrin, I discovered that past me gave you this code.  I still think the change
is correct, but I dropped it for now, mostly because the legacy/shadow MMU has the
same pattern (marks the PFN dirty before setting the SPTE).

I think this might actually be a bug fix, e.g. if the XCHG races with a fast page
fault fix and drops the Writable bit, the CPU could insert writable entry into the
TLB without KVM invoking kvm_set_pfn_dirty().  But I'm not 100% confident that I'm
not missing something, and _if_ there's a bug then mmu_spte_age() needs the same
fix, so for now, I dropped it.

On Fri, Feb 10, 2023 at 05:46:23PM -0800, Vipin Sharma wrote:
> No need to check all of the conditions in __handle_changed_spte(). Aging
> a gfn range implies resetting access bit or marking spte for access
> tracking.

nit: State what the patch does first.

> 
> Use atomic operation to only reset those bits. This avoids checking many
> conditions in __handle_changed_spte() API. Also, clean up code by
> removing dead code and API parameters.
> 
> Signed-off-by: Vipin Sharma <vipinsh@google.com>

nits aside,

Reviewed-by: David Matlack <dmatlack@google.com>

> ---
>  arch/x86/kvm/mmu/tdp_mmu.c | 36 +++++++++++++++++++-----------------
>  1 file changed, 19 insertions(+), 17 deletions(-)
> 
> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index c895560244de..5d6e77554797 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -758,13 +758,6 @@ static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
>  	_tdp_mmu_set_spte(kvm, iter, new_spte, true);
>  }
>  
> -static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm,
> -						 struct tdp_iter *iter,
> -						 u64 new_spte)
> -{
> -	_tdp_mmu_set_spte(kvm, iter, new_spte, false);
> -}
> -
>  #define tdp_root_for_each_pte(_iter, _root, _start, _end) \
>  	for_each_tdp_pte(_iter, _root, _start, _end)
>  
> @@ -1251,32 +1244,41 @@ static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm,
>  /*
>   * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero
>   * if any of the GFNs in the range have been accessed.
> + *
> + * No need to mark corresponding PFN as accessed as this call is coming from
> + * the clear_young() or clear_flush_young() notifier, which uses the return
> + * value to determine if the page has been accessed.
>   */
>  static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter,
>  			  struct kvm_gfn_range *range)
>  {
> -	u64 new_spte = 0;
> +	u64 new_spte;
>  
>  	/* If we have a non-accessed entry we don't need to change the pte. */
>  	if (!is_accessed_spte(iter->old_spte))
>  		return false;
>  
> -	new_spte = iter->old_spte;
> -
> -	if (spte_ad_enabled(new_spte)) {
> -		new_spte &= ~shadow_accessed_mask;
> +	if (spte_ad_enabled(iter->old_spte)) {
> +		iter->old_spte  = tdp_mmu_clear_spte_bits(iter->sptep,

nit: Extra space before =