Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 28 Mar 2023 10:39:12 +0800
From:   Xu Yilun <yilun.xu@intel.com>
To:     isaku.yamahata@intel.com
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        isaku.yamahata@gmail.com, Paolo Bonzini <pbonzini@redhat.com>,
        erdemaktas@google.com, Sean Christopherson <seanjc@google.com>,
        Sagi Shahar <sagis@google.com>,
        David Matlack <dmatlack@google.com>,
        Kai Huang <kai.huang@intel.com>,
        Zhi Wang <zhi.wang.linux@gmail.com>,
        Chao Gao <chao.gao@intel.com>
Subject: Re: [PATCH v13 039/113] KVM: x86/mmu: Assume guest MMIOs are shared
Message-ID: <ZCJTUIJsaJDNjcPP@yilunxu-OptiPlex-7050>
References: <cover.1678643051.git.isaku.yamahata@intel.com>
 <80912a430d4642acf85c454e97c8320f96e4b737.1678643052.git.isaku.yamahata@intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <80912a430d4642acf85c454e97c8320f96e4b737.1678643052.git.isaku.yamahata@intel.com>
Precedence: bulk

On 2023-03-12 at 10:56:03 -0700, isaku.yamahata@intel.com wrote:
> From: Chao Gao <chao.gao@intel.com>
> 
> Current TD guest doesn't invoke MAP_GPA to convert MMIO range to shared
> before accessing it. It implies that current TD guest assumes MMIOs are
> shared.
> 
> When TD tries to access assigned device's MMIO as shared, an EPT violation

Seems the patch is dealing with emulated MMIO, not assigned device's MMIO.

> is raised first. kvm_mem_is_private() checks the page shared or private
> attribute against the access type (shared bit in GPA). Then since no
> MAP_GPA is called for the MMIO, KVM thinks the MMIO is private and refuses
> shared access and doesn't set up shared EPT. Then KVM returns to TD and TD
> just retries and this causes an infinite loop.
> 
> Instead of requiring guest to invoke MAP_GPA for MMIOs, assume guest MMIOs
> are shared in KVM as well and don't expect explicit calls of MAP_GAP for
> guest MMIOs (i.e., GPAs either have no kvm_memory_slot or are backed by
> host MMIOs). So, allow shared access to guest MMIOs and move the page type
> check after the corresponding pfn is available.

Didn't see the movement.

Seems the commit message needs update.

> 
> Signed-off-by: Chao Gao <chao.gao@intel.com>
> ---
>  arch/x86/kvm/mmu/mmu.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 5883ab95ff07..ce8a896a3cfa 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -4314,7 +4314,12 @@ static int __kvm_faultin_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault
>  			return RET_PF_EMULATE;
>  	}
>  
> -	if (fault->is_private != kvm_mem_is_private(vcpu->kvm, fault->gfn))
> +	/*
> +	 * !fault->slot means MMIO.  Don't require explicit GPA conversion for
> +	 * MMIO because MMIO is assigned at the boot time.
> +	 */
> +	if (fault->slot &&

This only exempts emulated MMIO, how about the passthrough device's MMIO?

Thanks,
Yilun

> +	    fault->is_private != kvm_mem_is_private(vcpu->kvm, fault->gfn))
>  		return kvm_do_memory_fault_exit(vcpu, fault);
>  
>  	if (fault->is_private)
> -- 
> 2.25.1
>