From mboxrd@z Thu Jan 1 00:00:00 1970 From: Wanpeng Li Subject: Re: [RFC PATCH 1/4] KVM: x86: Split the APIC from the rest of IRQCHIP. Date: Mon, 18 May 2015 10:11:57 +0800 Message-ID: <20150518021157.GA15596@kernel> References: <1431481652-27268-1-git-send-email-srutherford@google.com> <555303CC.5030507@redhat.com> <20150513221001.GA23867@google.com> <20150515050302.GA7255@kernel> <20150515181023.GA10465@google.com> Reply-To: Wanpeng Li Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andrew Honig , "Wu, Feng" , Paolo Bonzini , "kvm@vger.kernel.org" To: Steve Rutherford Return-path: Received: from mga02.intel.com ([134.134.136.20]:12616 "EHLO mga02.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750770AbbERC3q (ORCPT ); Sun, 17 May 2015 22:29:46 -0400 Content-Disposition: inline In-Reply-To: <20150515181023.GA10465@google.com> Sender: kvm-owner@vger.kernel.org List-ID: On Fri, May 15, 2015 at 11:10:23AM -0700, Steve Rutherford wrote: >On Fri, May 15, 2015 at 01:03:02PM +0800, Wanpeng Li wrote: >> On Thu, May 14, 2015 at 12:29:21PM -0700, Andrew Honig wrote: >> >> >> >> BTW, what is the purpose of this series. If I understand it correctly, you only want to >> >> use the in-kernel lapic and leave the others (pic, ioapic) in userspace, what is the >> >> benefit of it? >> > >> >The purpose is to achieve the security benefit of removing some of the >> >interrupt handling into userspace, without incurring a significant >> >performance penalty. If you move the entire IRQCHIP into userspace, >> >we've seen perf impacts from 15-200% depending on the workload. With >> >this patch series, we're seeing perf penalty <1% on our tests (TCP_RR >> >> Why keep pic and ioapic in kernel space not get obvious benefit, what's >> the bottleneck? > >It's the other way around. The PIC and IOAPIC are going up to userspace, and the APICs are staying in the kernel. Yeah, this is what you have done in your patchset. I mean do you observe that why keep pic and ioapic in kernel space not get obvious benefit than move pic and ioapic to userspace. Regards, Wanpeng Li > >> >> Regards, >> Wanpeng Li >> >> >latency, TCP throughput, and Disk I/O). See > >(https://lwn.net/Articles/619332/) >> > >> >> >> >> Thanks, >> >> Feng >> >> >> >>> >> >-- >> >To unsubscribe from this list: send the line "unsubscribe kvm" in >> >the body of a message to majordomo@vger.kernel.org >> >More majordomo info at http://vger.kernel.org/majordomo-info.html