From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 0/1] x2apic implementation for kvm Date: Mon, 25 May 2009 12:50:44 +0300 Message-ID: <4A1A69F4.2010904@redhat.com> References: <1242927475-6140-1-git-send-email-gleb@redhat.com> <200905251448.04183.sheng@linux.intel.com> <4A1A5FD5.5090406@redhat.com> <200905251719.16009.sheng@linux.intel.com> <4A1A635A.1000000@redhat.com> <9832F13BD22FB94A829F798DA4A8280501B99466D1@pdsmsx503.ccr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Sheng Yang , Gleb Natapov , "kvm@vger.kernel.org" To: "Dong, Eddie" Return-path: Received: from mx2.redhat.com ([66.187.237.31]:34513 "EHLO mx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753643AbZEYJup (ORCPT ); Mon, 25 May 2009 05:50:45 -0400 In-Reply-To: <9832F13BD22FB94A829F798DA4A8280501B99466D1@pdsmsx503.ccr.corp.intel.com> Sender: kvm-owner@vger.kernel.org List-ID: Dong, Eddie wrote: >> x2apic has the following benefit: >> >> - msr exits are faster than mmio (no page table walk, emulation) >> - no need to read back ICR to look at the busy bit >> - one ICR write instead of two >> - potential to support large guests once we add interrupt remapping >> - shared code with the Hyper-V paravirt interface >> >> > Is there any plan to implement an PV irqchip such as Xenirqchip for KVM? No. PV irqchips (and PV in general) have the following drawbacks: - need to define and maintain an ABI - only works on newer Linux guests - obsoleted when hardware improves - increase code size and maintenance effort - have problems during transitions (boot, kexec) - don't integrate well with device assignment - require effort outside the kvm codebase If a significant performance benefit can be demonstrated, I'll consider it, but until then my preference is full virtualization augmented by optional, targeted pv assists (like the TPR patching). -- error compiling committee.c: too many arguments to function