Re: [libvirt] [PATCH 0/3] start of patches to support VFIO device assignment

This work isn't finished, but since there's a freeze coming up in a few days, I thought I should send these patches that *are* finished to get them out of the way before the last minute rush. Patch 1 is trivial. Patch 2 is longer, but completely mechanical (NOTE: there may be some uses of that struct living in code that I lack the environment to compile for, so test builds by people with odd platforms would be appreciated!) Patch 3 is standard parser/formatter/RNG/docs for the new XML. I haven't added any test cases, because lacking backend functionality I could only test xml-to-xml, which would require adding yet another test xml file, and I'd rather wait until the backend commandline-building code is in place, then simply add the new element to a few existing test XML files. In email awhile ago, danpb had suggested <driver name='vfio|qemu'/> I've used <driver name='kvm'/> because, unlike vhost-net where <driver name='qemu'/> indicates that the driver used is in the usermode qemu process (and *not* in the kernel), in this case <driver name='kvm'/> indicates that the driver used is in the kernel (and I've seen it referenced in several places as being "done by KVM", but never as "done by QEMU" (makes sense, since qemu is all in userland) Laine Stump (3): qemu: detect vfio-pci device assignment support conf: put hostdev pci address in a struct conf: formatter/parser/RNG/docs for hostdev <driver name='kvm|vfio'/>