Date: Tue, 16 Jan 2018 18:16:51 +0100
From: Radim =?utf-8?B?S3LEjW3DocWZ?= <rkrcmar@redhat.com>
To: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: kvm@vger.kernel.org, x86@kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>,
        "K. Y. Srinivasan" <kys@microsoft.com>,
        Haiyang Zhang <haiyangz@microsoft.com>,
        Stephen Hemminger <sthemmin@microsoft.com>,
        "Michael Kelley (EOSG)" <Michael.H.Kelley@microsoft.com>,
        Mohammed Gamal <mmorsy@redhat.com>,
        Cathy Avery <cavery@redhat.com>, Bandan Das <bsd@redhat.com>,
        linux-kernel@vger.kernel.org
Subject: Re: [RFC 0/6] Enlightened VMCS support for KVM on Hyper-V
Message-ID: <20180116171650.GB1824@flask>
References: <20180115173105.31845-1-vkuznets@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180115173105.31845-1-vkuznets@redhat.com>
Sender: linux-kernel-owner@vger.kernel.org

2018-01-15 18:30+0100, Vitaly Kuznetsov:
> Early RFC. I'll refer to this patchset in my DevConf/FOSDEM
> presentations.
> 
> When running nested KVM on Hyper-V it's possible to use so called
> 'Enlightened VMCS' and do normal memory reads/writes instead of
> doing VMWRITE/VMREAD instructions. Tests show that this speeds up
> tight CPUID loop almost 3 times:
> 
> Before:
> ./cpuid_tight
> 20459
> 
> After:
> ./cpuid_tight
> 7698

Nice!

> checkpatch.pl errors/warnings and possible 32bit brokenness are known
> things.
>     
> Main RFC questions I have are:
> - Do we want to have this per L2 VM or per L1 host?

IIUC, eVMCS replaces VMCS when enabled, hence doing it for all VMs would
be simplest -- we wouldn't need to setup VMCS nor reconfigure Hyper-V on
the fly.  (I'm thinking we could have a union in loaded_vmcs for
actually used type of VMCS.)

> - How can we achieve zero overhead for non-Hyper-V deployments? Use static
>   keys? But this will only work if we decide to do eVMCS per host.

Static keys seem like a good choice.

> - Can we do better than a big switch in evmcs_read()/evmcs_write()? And
>   probably don't use 'case' defines which checkpatch.pl hates.

I'd go for a separate mapping from Intel VMCS into its MS eVMCS and
dirty bit, something like vmcs_field_to_offset_table.

Thanks.