Re: [PATCH v2 15/17] kvm: x86: Skip shadow page resync on CR3 switch when indicated by guest

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Jun 12, 2018 at 03:52:42PM -0700, Junaid Shahid wrote:
> When the guest indicates that the TLB doesn't need to be flushed in a
> CR3 switch, we can also skip resyncing the shadow page tables since an
> out-of-sync shadow page table is equivalent to an out-of-sync TLB.
> 
> Signed-off-by: Junaid Shahid <junaids@xxxxxxxxxx>
> ---
>  arch/x86/kvm/mmu.c | 38 +++++++++++++++++++++++++++++++++-----
>  arch/x86/kvm/vmx.c | 11 ++++++-----
>  arch/x86/kvm/x86.c |  6 +++---
>  3 files changed, 42 insertions(+), 13 deletions(-)
> 
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index 4464565eec96..5b26c191052c 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -4087,9 +4087,19 @@ static bool fast_cr3_switch(struct kvm_vcpu *vcpu, gpa_t new_cr3,
>  			 */
>  
>  			kvm_make_request(KVM_REQ_LOAD_CR3, vcpu);
> -			kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
> -			if (!skip_tlb_flush)
> +			if (!skip_tlb_flush) {
> +				kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
>  				kvm_x86_ops->tlb_flush(vcpu, true);
> +			}
> +
> +			/*
> +			 * The last MMIO access's GVA and GPA are cached in the
> +			 * VCPU. When switching to a new CR3, that GVA->GPA
> +			 * mapping may no longer be valid. So clear any cached
> +			 * MMIO info even when we don't need to sync the shadow
> +			 * page tables.
> +			 */
> +			vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY);
>  
>  			__clear_sp_write_flooding_count(
>  				page_header(mmu->root_hpa));
> @@ -5205,6 +5215,21 @@ void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
>  	struct kvm_mmu *mmu = &vcpu->arch.mmu;
>  
>  	mmu->invlpg(vcpu, gva, mmu->root_hpa);
> +
> +	/*
> +	 * INVLPG is required to invalidate any global mappings for the VA,
> +	 * irrespective of PCID. Since it would take us roughly similar amount
> +	 * of work to determine whether the prev_root mapping of the VA is
> +	 * marked global, or to just sync it blindly, so we might as well just
> +	 * always sync it.
> +	 *
> +	 * Mappings not reachable via the current cr3 or the prev_root.cr3 will
> +	 * be synced when switching to that cr3, so nothing needs to be done
> +	 * here for them.
> +	 */
> +	if (VALID_PAGE(mmu->prev_root.hpa))
> +		mmu->invlpg(vcpu, gva, mmu->prev_root.hpa);
> +
>  	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>  	++vcpu->stat.invlpg;
>  }
> @@ -5220,14 +5245,17 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
>  	}
>  
>  	if (VALID_PAGE(mmu->prev_root.hpa) &&
> -	    pcid == kvm_get_pcid(vcpu, mmu->prev_root.cr3))
> +	    pcid == kvm_get_pcid(vcpu, mmu->prev_root.cr3)) {
> +		mmu->invlpg(vcpu, gva, mmu->prev_root.hpa);
>  		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
> +	}
>  
>  	++vcpu->stat.invlpg;
>  
>  	/*
> -	 * Mappings not reachable via the current cr3 will be synced when
> -	 * switching to that cr3, so nothing needs to be synced here for them.
> +	 * Mappings not reachable via the current cr3 or the prev_root.cr3 will
> +	 * be synced when switching to that cr3, so nothing needs to be done
> +	 * here for them.
>  	 */
>  }
>  EXPORT_SYMBOL_GPL(kvm_mmu_invpcid_gva);
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index ac4e78c2da38..98b2a9ca3ba9 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -8682,14 +8682,15 @@ static int handle_invpcid(struct kvm_vcpu *vcpu)
>  			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>  		}
>  
> -		if (kvm_get_pcid(vcpu, vcpu->arch.mmu.prev_root.cr3)
> +		if (kvm_get_pcid(vcpu, vcpu->arch.mmu.prev_root.cr3) 

Trailing whitespace is being addded.

>  		    == operand.pcid)
> -			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
> +			kvm_mmu_free_roots(vcpu, KVM_MMU_ROOT_PREVIOUS);
> +

Extra newline isn't needed.

>  
>  		/*
> -		 * If the current cr3 does not use the given PCID, then nothing
> -		 * needs to be synced here because a resync will happen anyway
> -		 * before switching to any other CR3.
> +		 * If neither the current cr3 nor the prev_root.cr3 use the
> +		 * given PCID, then nothing needs to be done here because a
> +		 * resync will happen anyway before switching to any other CR3.
>  		 */
>  
>  		skip_emulated_instruction(vcpu);
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 151a8f61c7f1..6a179f54b42c 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -858,10 +858,10 @@ int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
>  #endif
>  
>  	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
> -		kvm_mmu_sync_roots(vcpu);
> -
> -		if (!skip_tlb_flush)
> +		if (!skip_tlb_flush) {
> +			kvm_mmu_sync_roots(vcpu);
>  			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
> +		}
>  		return 0;
>  	}
>  
> -- 
> 2.18.0.rc1.242.g61856ae69a-goog
> 



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux