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, 22 Feb 2022 14:11:10 +0300
From:   "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
To:     Dingji Li <dj_lee@sjtu.edu.cn>
Cc:     tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
        dave.hansen@intel.com, luto@kernel.org, peterz@infradead.org,
        sathyanarayanan.kuppuswamy@linux.intel.com, aarcange@redhat.com,
        ak@linux.intel.com, dan.j.williams@intel.com, david@redhat.com,
        hpa@zytor.com, jgross@suse.com, jmattson@google.com,
        joro@8bytes.org, jpoimboe@redhat.com, knsathya@kernel.org,
        pbonzini@redhat.com, sdeep@vmware.com, seanjc@google.com,
        tony.luck@intel.com, vkuznets@redhat.com, wanpengli@tencent.com,
        x86@kernel.org, linux-kernel@vger.kernel.org,
        Sean Christopherson <sean.j.christopherson@intel.com>
Subject: Re: [PATCHv3 08/32] x86/traps: Add #VE support for TDX guest
Message-ID: <20220222111110.qe3bjqq6huomqqmi@black.fi.intel.com>
References: <20220218161718.67148-1-kirill.shutemov@linux.intel.com>
 <20220218161718.67148-9-kirill.shutemov@linux.intel.com>
 <6A6FFE5E-16C3-4054-837D-77D28A490C85@sjtu.edu.cn>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <6A6FFE5E-16C3-4054-837D-77D28A490C85@sjtu.edu.cn>
Precedence: bulk

On Tue, Feb 22, 2022 at 03:19:47PM +0800, Dingji Li wrote:
> Hi all,
> 
> I hope it is appropriate to ask these questions here:
> 
> I'm wondering if there are any performance comparisons available between
> TDX guests and VMX guests. The #VE processing adds non-trivial overhead
> to various VM exits, but how does it affect the performance of
> real-world applications? Existing patches have listed alternative
> methods to avoid the #VE in the first place, but there are trade-offs
> (e.g., bloated code, reduced generality). Besides, how much does the
> time spent in the TDX module affect VM exits / applications? (I guess
> the TDX module has a low overhead when compared to the #VE processing,
> but there is no public data.) Maybe some performance data can help make
> better trade-offs?

This is basic enabling of TDX guest support. The goal is to make TDX guest
functional. Yes, #VE handling adds non-trivial overhead and we have plan
to migrate it: there are patches in the queue that help to avoid bulk of
#VE, like replacing #VE-based MMIO with direct hypercalls. TDX will still
have performance penalty over plain VMX no matter what, but we aim to
minimize it.

I don't have any performance numbers to share at the moment.

-- 
 Kirill A. Shutemov