From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konrad Rzeszutek Wilk Subject: Re: [PATCH RFC] x86/hvm: unify HVM and PVH hypercall tables. Date: Fri, 16 May 2014 08:45:11 -0400 Message-ID: <20140516124511.GD2888@phenom.dumpdata.com> References: <201405151432.s4FEWZZd010054@aserz7022.oracle.com> <20140515163557.70c821e8@mantra.us.oracle.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Return-path: Content-Disposition: inline In-Reply-To: <20140515163557.70c821e8@mantra.us.oracle.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Mukesh Rathor Cc: Keir Fraser , George Dunlap , Tim Deegan , Jan Beulich , xen-devel@lists.xen.org List-Id: xen-devel@lists.xenproject.org On Thu, May 15, 2014 at 04:35:57PM -0700, Mukesh Rathor wrote: > On Thu, 15 May 2014 10:32:24 -0400 > Konrad Rzeszutek Wilk wrote: > = > > = > > On May 15, 2014 6:30 AM, Tim Deegan wrote: > > > > > > At 14:39 -0400 on 08 May (1399556383), Konrad Rzeszutek Wilk wrote: = > > > > On Thu, May 08, 2014 at 04:31:30PM +0100, Tim Deegan wrote: = > > > > > Stage one of many in merging PVH and HVM code in the > > > > > hypervisor. = > > > > > = > > > > > This exposes a few new hypercalls to HVM guests, all of which > > > > > were already available to PVH ones: = > > > > > = > > > > >=A0 - XENMEM_memory_map / XENMEM_machine_memory_map / > > > > >XENMEM_machphys_mapping: These are basically harmless, if a bit > > > > >useless to plain HVM. = > > > > > = > > > > >=A0 - VCPUOP_send_nmi / VCPUOP_initialise / VCPUOP[_is]_up / > > > > >VCPUOP_down This will eventually let HVM guests bring up APs the > > > > >way PVH ones do. For now, the VCPUOP_initialise paths are still > > > > >gated on is_pvh. = > > > > = > > > > I had a similar patch to enable this under HVM and found out that = > > > > if the guest issues VCPUOP_send_nmi we get in Linux: = > > > > = > > > > [=A0=A0=A0 3.611742] Corrupted low memory at c000fffc (fffc phys) = =3D > > > > 00029b00 [=A0=A0=A0 2.386785] Corrupted low memory at ffff88000000f= ff8 > > > > (fff8 phys) =3D 2990000000000 = > > > > = > > > > http://mid.gmane.org/20140422183443.GA6817@phenom.dumpdata.com = > > > > > > Right, thanks.=A0 Do you think that's likely to be a hypervisor bug, > > > or just a "don't do that then"? > > = > > It is a bug. But I don't know where and had not had a chance to > > investigate this further. > > = > > My feeling is that it is APIC emulation but I might be quite off. > > > > > > AFAICT PVH domains need this as they have no other way of sending = > > > NMIs. = > > > > > = > > Perhaps. The vAPIC that Boris had been looking at could make this > > work via the APIC path. > = > But, VCPUOP_send_nmi works for PVH right, and only HVM has problem? I did not test PVH - just tried with HVM. It might be that the PVH won't have the issue. > = > mukesh > =