From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alex Williamson Subject: Re: VFIO based vGPU(was Re: [Announcement] 2015-Q3 release of XenGT - a Mediated ...) Date: Tue, 26 Jan 2016 14:43:33 -0700 Message-ID: <1453844613.18049.1.camel@redhat.com> References: <569C5071.6080004@intel.com> <1453092476.32741.67.camel@redhat.com> <569CA8AD.6070200@intel.com> <1453143919.32741.169.camel@redhat.com> <569F4C86.2070501@intel.com> <56A6083E.10703@intel.com> <1453757426.32741.614.camel@redhat.com> <56A72313.9030009@intel.com> <56A77D2D.40109@gmail.com> <1453826249.26652.54.camel@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Gerd Hoffmann , Paolo Bonzini , "Lv, Zhiyuan" , "Ruan, Shuai" , "kvm@vger.kernel.org" , qemu-devel , "igvt-g@lists.01.org" , Neo Jia To: "Tian, Kevin" , Yang Zhang , "Song, Jike" Return-path: Received: from mx1.redhat.com ([209.132.183.28]:41342 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751266AbcAZVne (ORCPT ); Tue, 26 Jan 2016 16:43:34 -0500 In-Reply-To: Sender: kvm-owner@vger.kernel.org List-ID: On Tue, 2016-01-26 at 21:21 +0000, Tian, Kevin wrote: > > From: Alex Williamson [mailto:alex.williamson@redhat.com] > > Sent: Wednesday, January 27, 2016 12:37 AM > >=C2=A0 > > On Tue, 2016-01-26 at 22:05 +0800, Yang Zhang wrote: > > > On 2016/1/26 15:41, Jike Song wrote: > > > > On 01/26/2016 05:30 AM, Alex Williamson wrote: > > > > > [cc +Neo @Nvidia] > > > > >=C2=A0 > > > > > Hi Jike, > > > > >=C2=A0 > > > > > On Mon, 2016-01-25 at 19:34 +0800, Jike Song wrote: > > > > > > On 01/20/2016 05:05 PM, Tian, Kevin wrote: > > > > > > > I would expect we can spell out next level tasks toward a= bove > > > > > > > direction, upon which Alex can easily judge whether there= are > > > > > > > some common VFIO framework changes that he can help :-) > > > > > >=C2=A0 > > > > > > Hi Alex, > > > > > >=C2=A0 > > > > > > Here is a draft task list after a short discussion w/ Kevin= , > > > > > > would you please have a look? > > > > > >=C2=A0 > > > > > >=C2=A0 Bus Driver > > > > > >=C2=A0 > > > > > >=C2=A0 { in i915/vgt/xxx.c } > > > > > >=C2=A0 > > > > > >=C2=A0 - define a subset of vfio_pci interfaces > > > > > >=C2=A0 - selective pass-through (say aperture) > > > > > >=C2=A0 - trap MMIO: interface w/ QEMU > > > > >=C2=A0 > > > > > What's included in the subset?=C2=A0=C2=A0Certainly the bus r= eset ioctls really > > > > > don't apply, but you'll need to support the full device inter= face, > > > > > right?=C2=A0=C2=A0That includes the region info ioctl and acc= ess through the vfio > > > > > device file descriptor as well as the interrupt info and setu= p ioctls. > > > > >=C2=A0 > > > >=C2=A0 > > > > [All interfaces I thought are via ioctl:)=C2=A0=C2=A0For other = stuff like file > > > > descriptor we'll definitely keep it.] > > > >=C2=A0 > > > > The list of ioctl commands provided by vfio_pci: > > > >=C2=A0 > > > >=C2=A0 - VFIO_DEVICE_GET_PCI_HOT_RESET_INFO > > > >=C2=A0 - VFIO_DEVICE_PCI_HOT_RESET > > > >=C2=A0 > > > > As you said, above 2 don't apply. But for this: > > > >=C2=A0 > > > >=C2=A0 - VFIO_DEVICE_RESET > > > >=C2=A0 > > > > In my opinion it should be kept, no matter what will be provide= d in > > > > the bus driver. > > > >=C2=A0 > > > >=C2=A0 - VFIO_PCI_ROM_REGION_INDEX > > > >=C2=A0 - VFIO_PCI_VGA_REGION_INDEX > > > >=C2=A0 > > > > I suppose above 2 don't apply neither? For a vgpu we don't prov= ide a > > > > ROM BAR or VGA region. > > > >=C2=A0 > > > >=C2=A0 - VFIO_DEVICE_GET_INFO > > > >=C2=A0 - VFIO_DEVICE_GET_REGION_INFO > > > >=C2=A0 - VFIO_DEVICE_GET_IRQ_INFO > > > >=C2=A0 - VFIO_DEVICE_SET_IRQS > > > >=C2=A0 > > > > Above 4 are needed of course. > > > >=C2=A0 > > > > We will need to extend: > > > >=C2=A0 > > > >=C2=A0 - VFIO_DEVICE_GET_REGION_INFO > > > >=C2=A0 > > > >=C2=A0 > > > > a) adding a flag: DONT_MAP. For example, the MMIO of vgpu > > > > should be trapped instead of being mmap-ed. > > >=C2=A0 > > > I may not in the context, but i am curious how to handle the DONT= _MAP in > > > vfio driver? Since there are no real MMIO maps into the region an= d i > > > suppose the access to the region should be handled by vgpu in i91= 5 > > > driver, but currently most of the mmio accesses are handled by Qe= mu. > >=C2=A0 > > VFIO supports the following region attributes: > >=C2=A0 > > #define VFIO_REGION_INFO_FLAG_READ=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0(1 << 0) /* Region supports read */ > > #define VFIO_REGION_INFO_FLAG_WRITE=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0(1= << 1) /* Region supports write */ > > #define VFIO_REGION_INFO_FLAG_MMAP=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0(1 << 2) /* Region supports mmap */ > >=C2=A0 > > If MMAP is not set, then the QEMU driver will do pread and/or pwrit= e to > > the specified offsets of the device file descriptor, depending on w= hat > > accesses are supported. =C2=A0This is all reported through the REGI= ON_INFO > > ioctl for a given index. =C2=A0If mmap is supported, the VM will ha= ve direct > > access to the area, without faulting to KVM other than to populate = the > > mapping. =C2=A0Without mmap support, a VM MMIO access traps into KV= M, which > > returns out to QEMU to service the request, which then finds the > > MemoryRegion serviced through vfio, which will then perform a > > pread/pwrite through to the kernel vfio bus driver to handle the > > access. =C2=A0Thanks, > >=C2=A0 >=C2=A0 > Today KVMGT (not using VFIO yet) registers I/O emulation callbacks to= =C2=A0 > KVM, so VM MMIO access will be forwarded to KVMGT directly for=C2=A0 > emulation in kernel. If we reuse above R/W flags, the whole emulation= =C2=A0 > path would be unnecessarily long with obvious performance impact. We > either need a new flag here to indicate in-kernel emulation (bias fro= m > passthrough support), or just hide the region alternatively (let KVMG= T > to handle I/O emulation itself like today). That sounds like a future optimization TBH.=C2=A0=C2=A0There's very str= ict layering between vfio and kvm.=C2=A0=C2=A0Physical device assignment co= uld make use of it as well, avoiding a round trip through userspace when an ioread/write would do.=C2=A0=C2=A0Userspace also needs to orchestrate t= hose kinds of accelerators, there might be cases where userspace wants to see thos= e transactions for debugging or manipulating the device.=C2=A0=C2=A0We ca= n't simply take shortcuts to provide such direct access.=C2=A0=C2=A0Thanks, Alex