From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sheng Yang Subject: Re: [PATCH 1/1] KVM: VMX: Fix base_ptes for EPT Date: Wed, 19 Nov 2008 18:32:32 +0800 Message-ID: <200811191832.32698.sheng@linux.intel.com> References: <1227083462-12313-1-git-send-email-sheng@linux.intel.com> <4923E99C.1030506@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org To: Avi Kivity Return-path: Received: from mga05.intel.com ([192.55.52.89]:19261 "EHLO fmsmga101.fm.intel.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751939AbYKSKgL (ORCPT ); Wed, 19 Nov 2008 05:36:11 -0500 In-Reply-To: <4923E99C.1030506@redhat.com> Content-Disposition: inline Sender: kvm-owner@vger.kernel.org List-ID: On Wednesday 19 November 2008 18:25:32 Avi Kivity wrote: > Sheng Yang wrote: > > After support MTRR for EPT, memory type WB shouldn't take as default in > > base_ptes. (Something wrong during the merging...) > > Yes, my fault. > > But don't we still have a big hole here? The guest can set RAM to > various attributes, even without device assignment, and create > inconsistencies with the qemu and host kernel mapping. The current reply from hardware guys is: the only situation that cause MCE is to access MMIO with cacheable memory type, and at least for EPT, they say using different memory type for same memory region don't have problem as far as we know... But for now, the hardware guys suggest to use host MTRR as the base for EPT memory type rather than using guest MTRR, at least it would protect us from MCE. And guest can operate it's PAT then promote UC to WC for performance and use UC for DMA. We are still in discussion to get more details now. For now, I may try to using host MTRR for EPT as suggested. -- regards Yang, Sheng