Message-ID: <25249e7d-4fd9-e1c1-8efb-31750740ec27@redhat.com>
Date:   Mon, 6 Mar 2023 22:47:18 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Subject: Re: [PATCH 0/2] KVM: x86: Propagate AMD-specific IBRS bits to guests
Content-Language: en-US
To:     Borislav Petkov <bp@alien8.de>,
        Sean Christopherson <seanjc@google.com>
Cc:     Takahiro Itazuri <itazur@amazon.com>, dave.hansen@linux.intel.com,
        kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        mingo@redhat.com, tglx@linutronix.de, x86@kernel.org,
        zulinx86@gmail.com
References: <Y/5oBKi6vjZe83ac@zn.tnic>
 <20230228222416.61484-1-itazur@amazon.com> <Y/6FIeJ5KCOfKEPN@zn.tnic>
 <ZAZYKe4L8jhMG4An@google.com>
 <20230306214419.GDZAZes941k+4NPgDL@fat_crate.local>
From:   Paolo Bonzini <pbonzini@redhat.com>
In-Reply-To: <20230306214419.GDZAZes941k+4NPgDL@fat_crate.local>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 3/6/23 22:44, Borislav Petkov wrote:
>> I thought that the consensus was that adding unused-by-the-kernel flags to
>> cpufeatures.h is ok so long as the feature is hidden from /proc/cpuinfo and the
>> kernel already dedicates a word to the CPUID leaf?
> I guess we should finally write it down in Documentation/x86/cpuinfo.rst
> 
> And in case there's no dedicated word, it should be resorted to KVM-only
> feature flags.
> 
> In any case, I'd like for baremetal CPUID stuff to be decoupled from
> KVM's machinery as far as possible as both have different goals wrt
> feature flags.

It's very rare that KVM can provide a CPUID feature if the kernel has 
masked it, so if the kernel needs to know about a feature word than KVM 
most likely needs to know what kind of massaging the kernel has done.

Paolo