From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 7/7] KVM: MMU: cache guest page number to guest frame number Date: Tue, 22 Feb 2011 16:32:25 +0200 Message-ID: <4D63C8F9.1020708@redhat.com> References: <4D636EF8.60800@cn.fujitsu.com> <4D6370E8.6080008@cn.fujitsu.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Cc: Marcelo Tosatti , LKML , KVM To: Xiao Guangrong Return-path: In-Reply-To: <4D6370E8.6080008@cn.fujitsu.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On 02/22/2011 10:16 AM, Xiao Guangrong wrote: > Cache guest page number to guest frame number to avoid walk guest page table > frequently, the 'vtlb' idea is from Xen. > > Note: > we can't use vtlb in ept guests since the guest tlb invalid operation is not > intercept(reload CR3, invlpg), also can't used in L2 nnpt guest for the same > reason, but we can used it to cache L1's npt page table. > I'm not so hot about introducing a new mechanism strictly for older hosts... EPT exists in three generations of Intel processors now (Sandy Bridge, Westmere, and Nehalem), and NPT is significantly older. I'd rather make the guest page table walk faster. I believe that things like get_user() can be significantly faster than copy_from_user() on older hosts, and a four-level page table walk shouldn't be that slow. -- error compiling committee.c: too many arguments to function