Date:   Thu, 11 Nov 2021 18:06:57 +0000
From:   David Matlack <dmatlack@google.com>
To:     Ben Gardon <bgardon@google.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>,
        Peter Xu <peterx@redhat.com>,
        Sean Christopherson <seanjc@google.com>,
        Peter Shier <pshier@google.com>,
        Mingwei Zhang <mizhang@google.com>,
        Yulei Zhang <yulei.kernel@gmail.com>,
        Wanpeng Li <kernellwp@gmail.com>,
        Xiao Guangrong <xiaoguangrong.eric@gmail.com>,
        Kai Huang <kai.huang@intel.com>,
        Keqian Zhu <zhukeqian1@huawei.com>,
        David Hildenbrand <david@redhat.com>
Subject: Re: [RFC 02/19] KVM: x86/mmu: Batch TLB flushes for a single zap
Message-ID: <YY1bweJeOLaBtbA0@google.com>
References: <20211110223010.1392399-1-bgardon@google.com>
 <20211110223010.1392399-3-bgardon@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20211110223010.1392399-3-bgardon@google.com>
Precedence: bulk

On Wed, Nov 10, 2021 at 02:29:53PM -0800, Ben Gardon wrote:
> When recursively handling a removed TDP page table, the TDP MMU will
> flush the TLBs and queue an RCU callback to free the PT. If the original
> change zapped a non-leaf SPTE at PG_LEVEL_1G or above, that change will
> result in many unnecessary TLB flushes when one would suffice. Queue all
> the PTs which need to be freed on a list and wait to queue RCU callbacks
> to free them until after all the recursive callbacks are done.
> 
> 
> Signed-off-by: Ben Gardon <bgardon@google.com>
> ---
>  arch/x86/kvm/mmu/tdp_mmu.c | 88 ++++++++++++++++++++++++++++++--------
>  1 file changed, 70 insertions(+), 18 deletions(-)
> 
> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index 866c2b191e1e..5b31d046df78 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -220,7 +220,8 @@ hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
>  
>  static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
>  				u64 old_spte, u64 new_spte, int level,
> -				bool shared);
> +				bool shared,
> +				struct list_head *disconnected_sps);
>  
>  static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level)
>  {
> @@ -302,6 +303,11 @@ static void tdp_mmu_unlink_page(struct kvm *kvm, struct kvm_mmu_page *sp,
>   * @shared: This operation may not be running under the exclusive use
>   *	    of the MMU lock and the operation must synchronize with other
>   *	    threads that might be modifying SPTEs.
> + * @disconnected_sps: If null, the TLBs will be flushed and the disconnected
> + *		      TDP MMU page will be queued to be freed after an RCU
> + *		      callback. If non-null the page will be added to the list
> + *		      and flushing the TLBs and queueing an RCU callback to
> + *		      free the page will be the caller's responsibility.
>   *
>   * Given a page table that has been removed from the TDP paging structure,
>   * iterates through the page table to clear SPTEs and free child page tables.
> @@ -312,7 +318,8 @@ static void tdp_mmu_unlink_page(struct kvm *kvm, struct kvm_mmu_page *sp,
>   * early rcu_dereferences in the function.
>   */
>  static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt,
> -					bool shared)
> +					bool shared,

>  {
>  	struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(pt));
>  	int level = sp->role.level;
> @@ -371,13 +378,16 @@ static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt,
>  		}
>  		handle_changed_spte(kvm, kvm_mmu_page_as_id(sp), gfn,
>  				    old_child_spte, REMOVED_SPTE, level,
> -				    shared);
> +				    shared, disconnected_sps);
>  	}
>  
> -	kvm_flush_remote_tlbs_with_address(kvm, base_gfn,
> -					   KVM_PAGES_PER_HPAGE(level + 1));
> -
> -	call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback);
> +	if (disconnected_sps) {
> +		list_add_tail(&sp->link, disconnected_sps);
> +	} else {
> +		kvm_flush_remote_tlbs_with_address(kvm, base_gfn,
> +						   KVM_PAGES_PER_HPAGE(level + 1));
> +		call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback);
> +	}
>  }
>  
>  /**
> @@ -391,13 +401,21 @@ static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt,
>   * @shared: This operation may not be running under the exclusive use of
>   *	    the MMU lock and the operation must synchronize with other
>   *	    threads that might be modifying SPTEs.
> + * @disconnected_sps: Only used if a page of page table memory has been
> + *		      removed from the paging structure by this change.
> + *		      If null, the TLBs will be flushed and the disconnected
> + *		      TDP MMU page will be queued to be freed after an RCU
> + *		      callback. If non-null the page will be added to the list
> + *		      and flushing the TLBs and queueing an RCU callback to
> + *		      free the page will be the caller's responsibility.
>   *
>   * Handle bookkeeping that might result from the modification of a SPTE.
>   * This function must be called for all TDP SPTE modifications.
>   */
>  static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
>  				  u64 old_spte, u64 new_spte, int level,
> -				  bool shared)
> +				  bool shared,
> +				  struct list_head *disconnected_sps)
>  {
>  	bool was_present = is_shadow_present_pte(old_spte);
>  	bool is_present = is_shadow_present_pte(new_spte);
> @@ -475,22 +493,39 @@ static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
>  	 */
>  	if (was_present && !was_leaf && (pfn_changed || !is_present))
>  		handle_removed_tdp_mmu_page(kvm,
> -				spte_to_child_pt(old_spte, level), shared);
> +				spte_to_child_pt(old_spte, level), shared,
> +				disconnected_sps);
>  }
>  
>  static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
>  				u64 old_spte, u64 new_spte, int level,
> -				bool shared)
> +				bool shared, struct list_head *disconnected_sps)
>  {
>  	__handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level,
> -			      shared);
> +			      shared, disconnected_sps);
>  	handle_changed_spte_acc_track(old_spte, new_spte, level);
>  	handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte,
>  				      new_spte, level);
>  }
>  
>  /*
> - * tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically
> + * The TLBs must be flushed between the pages linked from disconnected_sps
> + * being removed from the paging structure and this function being called.
> + */
> +static void handle_disconnected_sps(struct kvm *kvm,
> +				    struct list_head *disconnected_sps)

handle_disconnected_sps() does a very specific task so I think we could
go with a more specific function name to make the code more readable.

How about free_sps_rcu() or call_rcu_free_sps()?

> +{
> +	struct kvm_mmu_page *sp;
> +	struct kvm_mmu_page *next;
> +
> +	list_for_each_entry_safe(sp, next, disconnected_sps, link) {
> +		list_del(&sp->link);
> +		call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback);
> +	}
> +}
> +
> +/*
> + * __tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically
>   * and handle the associated bookkeeping.  Do not mark the page dirty
>   * in KVM's dirty bitmaps.
>   *
> @@ -500,9 +535,10 @@ static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
>   * Returns: true if the SPTE was set, false if it was not. If false is returned,
>   *	    this function will have no side-effects.
>   */
> -static inline bool tdp_mmu_set_spte_atomic(struct kvm *kvm,
> -					   struct tdp_iter *iter,
> -					   u64 new_spte)
> +static inline bool __tdp_mmu_set_spte_atomic(struct kvm *kvm,
> +					     struct tdp_iter *iter,
> +					     u64 new_spte,
> +					     struct list_head *disconnected_sps)
>  {
>  	lockdep_assert_held_read(&kvm->mmu_lock);
>  
> @@ -522,22 +558,32 @@ static inline bool tdp_mmu_set_spte_atomic(struct kvm *kvm,
>  		return false;
>  
>  	__handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
> -			      new_spte, iter->level, true);
> +			      new_spte, iter->level, true, disconnected_sps);
>  	handle_changed_spte_acc_track(iter->old_spte, new_spte, iter->level);
>  
>  	return true;
>  }
>  
> +static inline bool tdp_mmu_set_spte_atomic(struct kvm *kvm,
> +					   struct tdp_iter *iter,
> +					   u64 new_spte)
> +{
> +	return __tdp_mmu_set_spte_atomic(kvm, iter, new_spte, NULL);

Why not leverage disconnected_sps here as well? Then you can remove the
NULL case (and comments) from handle_removed_tdp_mmu_page.

> +}
> +
>  static inline bool tdp_mmu_zap_spte_atomic(struct kvm *kvm,
>  					   struct tdp_iter *iter)
>  {
> +	LIST_HEAD(disconnected_sps);
> +
>  	/*
>  	 * Freeze the SPTE by setting it to a special,
>  	 * non-present value. This will stop other threads from
>  	 * immediately installing a present entry in its place
>  	 * before the TLBs are flushed.
>  	 */
> -	if (!tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE))
> +	if (!__tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE,
> +				       &disconnected_sps))
>  		return false;
>  
>  	kvm_flush_remote_tlbs_with_address(kvm, iter->gfn,
> @@ -553,6 +599,8 @@ static inline bool tdp_mmu_zap_spte_atomic(struct kvm *kvm,
>  	 */
>  	WRITE_ONCE(*rcu_dereference(iter->sptep), 0);
>  
> +	handle_disconnected_sps(kvm, &disconnected_sps);
> +
>  	return true;
>  }
>  
> @@ -577,6 +625,8 @@ static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
>  				      u64 new_spte, bool record_acc_track,
>  				      bool record_dirty_log)
>  {
> +	LIST_HEAD(disconnected_sps);
> +
>  	lockdep_assert_held_write(&kvm->mmu_lock);
>  
>  	/*
> @@ -591,7 +641,7 @@ static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
>  	WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte);
>  
>  	__handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
> -			      new_spte, iter->level, false);
> +			      new_spte, iter->level, false, &disconnected_sps);
>  	if (record_acc_track)
>  		handle_changed_spte_acc_track(iter->old_spte, new_spte,
>  					      iter->level);
> @@ -599,6 +649,8 @@ static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
>  		handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn,
>  					      iter->old_spte, new_spte,
>  					      iter->level);
> +
> +	handle_disconnected_sps(kvm, &disconnected_sps);

Where is the TLB flush for these disconnected_sps?

>  }
>  
>  static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
> -- 
> 2.34.0.rc0.344.g81b53c2807-goog
>