From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [patch 3/4] KVM: introduce irq_lock, use it to protect ioapic Date: Tue, 26 May 2009 14:33:33 +0300 Message-ID: <4A1BD38D.6040207@redhat.com> References: <20090520184841.954066003@localhost.localdomain> <20090520185134.748295218@localhost.localdomain> <4A19553F.1090903@redhat.com> <20090525111114.GC2789@amt.cnet> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org, Gregory Haskins To: Marcelo Tosatti Return-path: Received: from mx2.redhat.com ([66.187.237.31]:42456 "EHLO mx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751525AbZEZLdf (ORCPT ); Tue, 26 May 2009 07:33:35 -0400 In-Reply-To: <20090525111114.GC2789@amt.cnet> Sender: kvm-owner@vger.kernel.org List-ID: Marcelo Tosatti wrote: > On Sun, May 24, 2009 at 05:10:07PM +0300, Avi Kivity wrote: > >> Marcelo Tosatti wrote: >> >>> Subject says it all. >>> >>> >> I hate those changelogs. I guess Subject never reviews code. >> >> You might put some evidence that we're suffering from contention here >> (I'm very willing to believe it, but hard evidence is better). >> > > Don't have any evidence yet, the main purpose of the split is to fix the > deadlock. > You might have said that. > But, with the data we have so far, slots_lock and kvm->lock will cause > significant cache bouncing on EPT hardware (with measurements on shadow > mmu_lock is #1 offender). > Well, mmu_lock is probably contended, which is much worse than just cache bouncing. -- error compiling committee.c: too many arguments to function