Re: [PATCH 5/7] KVM: X86: Don't unsync pagetables when speculative

[Maxim Levitsky <mlevitsk@redhat.com>]

On Tue, 2021-08-24 at 15:55 +0800, Lai Jiangshan wrote:
> From: Lai Jiangshan <laijs@linux.alibaba.com>
> 
> We'd better only unsync the pagetable when there just was a really
> write fault on a level-1 pagetable.
> 
> Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
> ---
>  arch/x86/kvm/mmu/mmu.c          | 6 +++++-
>  arch/x86/kvm/mmu/mmu_internal.h | 3 ++-
>  arch/x86/kvm/mmu/spte.c         | 2 +-
>  3 files changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index a165eb8713bc..e5932af6f11c 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -2600,7 +2600,8 @@ static void kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
>   * were marked unsync (or if there is no shadow page), -EPERM if the SPTE must
>   * be write-protected.
>   */
> -int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync)
> +int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync,
> +			    bool speculative)
>  {
>  	struct kvm_mmu_page *sp;
>  	bool locked = false;
> @@ -2626,6 +2627,9 @@ int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync)
>  		if (sp->unsync)
>  			continue;
>  
> +		if (speculative)
> +			return -EEXIST;

Woudn't it be better to ensure that callers set can_unsync = false when speculating?

Also if I understand correctly this is not fixing a bug, but an optimization?

Best regards,
	Maxim Levitsky


> +
>  		/*
>  		 * TDP MMU page faults require an additional spinlock as they
>  		 * run with mmu_lock held for read, not write, and the unsync
> diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
> index 658d8d228d43..f5d8be787993 100644
> --- a/arch/x86/kvm/mmu/mmu_internal.h
> +++ b/arch/x86/kvm/mmu/mmu_internal.h
> @@ -116,7 +116,8 @@ static inline bool kvm_vcpu_ad_need_write_protect(struct kvm_vcpu *vcpu)
>  	       kvm_x86_ops.cpu_dirty_log_size;
>  }
>  
> -int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync);
> +int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync,
> +			    bool speculative);
>  
>  void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
>  void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
> diff --git a/arch/x86/kvm/mmu/spte.c b/arch/x86/kvm/mmu/spte.c
> index 3e97cdb13eb7..b68a580f3510 100644
> --- a/arch/x86/kvm/mmu/spte.c
> +++ b/arch/x86/kvm/mmu/spte.c
> @@ -159,7 +159,7 @@ int make_spte(struct kvm_vcpu *vcpu, unsigned int pte_access, int level,
>  		 * e.g. it's write-tracked (upper-level SPs) or has one or more
>  		 * shadow pages and unsync'ing pages is not allowed.
>  		 */
> -		if (mmu_try_to_unsync_pages(vcpu, gfn, can_unsync)) {
> +		if (mmu_try_to_unsync_pages(vcpu, gfn, can_unsync, speculative)) {
>  			pgprintk("%s: found shadow page for %llx, marking ro\n",
>  				 __func__, gfn);
>  			ret |= SET_SPTE_WRITE_PROTECTED_PT;