From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: [RFC] para virt interface of perf to support kvm guest os statistics collection in guest os Date: Wed, 9 Jun 2010 12:27:43 +0200 Message-ID: <20100609102743.GC1150@elte.hu> References: <1276054214.2096.383.camel@ymzhang.sh.intel.com> <4C0F51DD.3080200@redhat.com> <1276075280.2096.427.camel@ymzhang.sh.intel.com> <4C0F61D3.9000402@redhat.com> <1276078122.2046.1227.camel@twins> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Avi Kivity , "Zhang, Yanmin" , LKML , kvm@vger.kernel.org, Fr??d??ric Weisbecker , Arnaldo Carvalho de Melo , Cyrill Gorcunov , Lin Ming , Sheng Yang , Marcelo Tosatti , oerg Roedel , Jes Sorensen , Gleb Natapov , Zachary Amsden , zhiteng.huang@intel.com, tim.c.chen@intel.com To: Peter Zijlstra Return-path: Content-Disposition: inline In-Reply-To: <1276078122.2046.1227.camel@twins> Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org * Peter Zijlstra wrote: > On Wed, 2010-06-09 at 12:41 +0300, Avi Kivity wrote: > > > > >> Disabling the watchdog is unfortunate. Why is it necessary? > > >> > > > perf always uses NMI, so we disable the nmi_watchdog when a perf_event is > > > set up in case they might have impact. > > > > > > > Ok. Is that the case for the hardware pmus as well? If so it might be > > done in common code. > > The x86 hardware pmu implementation disables the lapic watchdog too, but > recent kernels come with a watchdog implementation on top of perf, the old > lapic one will be depricated. Note, that code is in -tip, queued for v2.6.36. Ingo