Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <9b9b7f7a-cfc6-1e56-aeb2-d3a7e445fe14@redhat.com>
Date:   Mon, 7 Feb 2022 15:35:23 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.5.0
Subject: Re: [PATCH 06/23] KVM: MMU: load new PGD once nested two-dimensional
 paging is initialized
Content-Language: en-US
From:   Paolo Bonzini <pbonzini@redhat.com>
To:     David Matlack <dmatlack@google.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        seanjc@google.com, vkuznets@redhat.com
References: <20220204115718.14934-1-pbonzini@redhat.com>
 <20220204115718.14934-7-pbonzini@redhat.com> <Yf178LYEY4pFJcLc@google.com>
 <180d7f0f-8c58-2a52-02a7-bd014d81d7a3@redhat.com>
In-Reply-To: <180d7f0f-8c58-2a52-02a7-bd014d81d7a3@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2/7/22 14:50, Paolo Bonzini wrote:
>>>
>> Those checks are just for performance correct (to skip iterating through
>> the list of roots)?
>>
>> Either way, it's probably worth including a Fixes tag below.
> 
> There's no bug because __kvm_mmu_new_pgd is passed a correct new_role. 
> But it's unnecessarily complex as shown by patches 7 and 9.

root_level is not part of the role, but we're still safe because 
changing it requires to flip CR0.PG to 0 and back to 1.  This calls 
kvm_mmu_reset_context and unloads all roots, so it's not possible to 
have a cached PGD with the wrong root level.  On the first call the 
root_level might be incorrect, but then again there's no cached PGD.

(Also, shadow_root_level is either always or never >= PT64_ROOT_4LEVEL, 
so it's okay unless the guest_mmu was never set up and therefore no 
cached PGD exists).

Paolo