Re: [Qemu-devel] [PATCH 1/2] vfio/mdev: add version field as mandatory attribute for mdev device

[Cornelia Huck <cohuck@redhat.com>]

On Tue, 23 Apr 2019 01:41:57 -0400
Yan Zhao <yan.y.zhao@intel.com> wrote:

> On Tue, Apr 23, 2019 at 09:21:00AM +0800, Alex Williamson wrote:
> > On Mon, 22 Apr 2019 21:01:52 -0400
> > Yan Zhao <yan.y.zhao@intel.com> wrote:
> >   
> > > On Mon, Apr 22, 2019 at 10:39:50PM +0800, Alex Williamson wrote:  
> > > > On Fri, 19 Apr 2019 04:35:04 -0400
> > > > Yan Zhao <yan.y.zhao@intel.com> wrote:
> > > >     
> > > > > device version attribute in mdev sysfs is used by user space software
> > > > > (e.g. libvirt) to query device compatibility for live migration of VFIO
> > > > > mdev devices. This attribute is mandatory if a mdev device supports live
> > > > > migration.    
> > > > 
> > > > The Subject: doesn't quite match what's being proposed here.
> > > >      
> > > > > It consists of two parts: common part and vendor proprietary part.
> > > > > common part: 32 bit. lower 16 bits is vendor id and higher 16 bits
> > > > >              identifies device type. e.g., for pci device, it is
> > > > >              "pci vendor id" | (VFIO_DEVICE_FLAGS_PCI << 16).    
> > > > 
> > > > What purpose does this serve?  If it's intended as some sort of
> > > > namespace feature, shouldn't we first assume that we can only support
> > > > migration to devices of the same type?  Therefore each type would
> > > > already have its own namespace.  Also that would make the trailing bit
> > > > of the version string listed below in the example redundant.  A vendor
> > > > is still welcome to include this in their version string if they wish,
> > > > but I think the string should be entirely vendor defined.
> > > >    
> > > hi Alex,
> > > This common part is a kind of namespace.
> > > Because if version string is entirely defined by vendors, I'm worried about
> > > if there is a case that one vendor's version string happens to deceive and
> > > interfere with another vendor's version checking?
> > > e.g.
> > > vendor A has a version string like: vendor id + device id + mdev type
> > > vendor B has a version string like: device id + vendor id + mdev type
> > > but vendor A's vendor id is 0x8086, device id is 0x1217
> > > vendor B's vendor id is 0x1217, device id is 0x8086.
> > > 
> > > In this corner case, the two vendors may regard the two device is
> > > migratable but actually they are not.
> > > 
> > > That's the reason for this common part that serve as a kind of namespace
> > > that all vendors will comply with to avoid overlap.  
> > 
> > If we assume that migration can only occur between matching mdev types,
> > this is redundant, each type already has their own namespace.
> >  
> hi Alex,
> do you mean user space software like libvirt needs to first check whether
> mdev type is matching and then check whether version is matching?
> 
> if user space software only checks version for migration, it means vendor
> driver has to include mdev type in their vendor proprietary part string,
> right?

Can't userspace simply check for the driver in use and only then check
the version attribute?

> Another thing is that could there be any future mdev parent driver that
> applies to all mdev devices, just like vfio-pci? like Yi's vfio-pci-mdev
> driver (https://lkml.org/lkml/2019/3/13/114)?

Hm, I think that the vfio-pci-mdev driver then needs to expose
information regarding compatibility (and not the core).