From mboxrd@z Thu Jan 1 00:00:00 1970 From: Gleb Natapov Subject: Re: [PATCH RFC dontapply 5/5] kvm_para: guest side for eoi avoidance Date: Tue, 8 May 2012 19:32:14 +0300 Message-ID: <20120508163214.GG8988@redhat.com> References: <5e1399d9e743668a87f31d8f2b83cad862c60728.1334833141.git.mst@redhat.com> <4FA93B3F.3000906@redhat.com> <20120508152843.GE8988@redhat.com> <4FA93FA3.20306@zytor.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Paolo Bonzini , "Michael S. Tsirkin" , x86@kernel.org, kvm@vger.kernel.org, Ingo Molnar , Avi Kivity , Marcelo Tosatti , Linus Torvalds , linux-kernel@vger.kernel.org To: "H. Peter Anvin" Return-path: Content-Disposition: inline In-Reply-To: <4FA93FA3.20306@zytor.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On Tue, May 08, 2012 at 08:45:39AM -0700, H. Peter Anvin wrote: > On 05/08/2012 08:28 AM, Gleb Natapov wrote: > > On Tue, May 08, 2012 at 05:26:55PM +0200, Paolo Bonzini wrote: > >> Il 23/04/2012 16:04, Michael S. Tsirkin ha scritto: > >>> +/* Our own copy of __test_and_clear_bit to make sure > >>> + * it is done with a single instruction */ > >> > >> Is this for microoptimization or correctness? If the latter, it does > >> not ensure anything without a "lock" prefix. > >> > > It can't race with other vcpus, only with vmexit on the same vcpu. > > > > That doesn't answer the question very well... I really don't understand > the point of having a private copy here. __test_and_clear_bit() is not guarantied to be local CPU atomic an that is what we need here. > > I really, really don't want a bunch of private interfaces around. It > would be a lot better to define a test_and_{set,clear}_bit_local() in > which is defined to be local CPU atomic. > Yes, this will be definitely better. -- Gleb.