From: "H. Peter Anvin" Subject: Re: [Xen-devel] [PATCH] x86/amd: fix crash as Xen Dom0 on AMD Trinity systems Date: Wed, 30 May 2012 11:20:17 -0700 Message-ID: <4FC664E1.9050504@zytor.com> References: <4FC63DAF0200007800086DC5@nat28.tlf.novell.com> <4FC62888.9010407@amd.com> <20120530144851.GA12184@jshin-Toonie> <20120530145005.GI3207@phenom.dumpdata.com> <20120530150334.GA13349@jshin-Toonie> <20120530171754.GA5115@phenom.dumpdata.com> <20120530173247.GC15635@x1.osrc.amd.com> <4FC65D34.1060803@zytor.com> <20120530175150.GE15438@x1.osrc.amd.com> <4FC66037.6020506@zytor.com> <20120530181722.GF15438@x1.osrc.amd.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20120530181722.GF15438@x1.osrc.amd.com> Sender: linux-kernel-owner@vger.kernel.org To: Borislav Petkov , Konrad Rzeszutek Wilk , Jacob Shin , Andre Przywara , jeremy@goop.org, xen-devel@lists.xensource.com, linux-kernel@vger.kernel.org, Jan Beulich , mingo@elte.hu, tglx@linutronix.de List-Id: xen-devel@lists.xenproject.org On 05/30/2012 11:17 AM, Borislav Petkov wrote: > On Wed, May 30, 2012 at 11:00:23AM -0700, H. Peter Anvin wrote: >> OK. But there is still the general problem, no? > > With this patch xen crashes go away because they paravirt > native_{read,write}_msr_safe. > > The other place where we use the amd_safe variants is an obscure K8, > revC and earlier fix for _some_ BIOSen and this hasn't bitten us yet > so I'm assuming people haven't run xen on such boxes yet. Does it need > fixing? Probably, if we really really have to. > > Now, someone probably needs to paravirt the *safe_regs variants in case > something else decides to use them. I don't know what to do here, do I > want more paravirt code in there? No. I guess if this is done carefully > and cleanly, then it should be ok but it can't be done like that - it > needs to adhere to the current pv_cpu_ops thing which is already there. > I thought I was being told that Xen would trap and emulate the rdmsr/wrmsr instructions. I guess they don't want to do that for the handful of performance-sensitive MSRs there are, but those don't use the *_regs variants. -hpa