This set of patches implements the QEMU end of the MMIO virtio transport (as specified by Appendix X of the latest virtio spec from here http://ozlabs.org/~rusty/virtio-spec/virtio.pdf and implemented by patches which I think are going into Linux 3.2).
I submit this only as an RFC because although it works (tested virtio-blk anyway) it has some issues: * extra vring alignment field not saved/restored (because virtio layer isn't using VMState and doesn't allow the virtio base layer to specify a version for its data so back-compat would be tricky) * you have to specify which kind of virtio device you want in the board model. In particular this means that for virtio-blk the user has to say "-drive if=none,file=whatever.img,id=myimg -global virtio-blk-mmio.drive=myimg" or the virtio-blk layer will refuse to start, which is not at all userfriendly Anthony's suggestion was that we should have a qdev bus between the transport layer and the -blk/-net/-etc parts. Then the board could instantiate N virtio-mmio transport instances, and the user uses -device to create a virtio-blk device and assign it to the relevant slot. (PCI virtio would also split similarly so you'd end up with pci-controller -> n lots of [ virtio-pci-transport + virtio-blk]) plus some syntactic sugar for back-compatibility I guess.) However I'm not currently sure when I'll have time to do that refactoring, so I'm just sending this patchset out as an RFC for now. Peter Maydell (4): virtio: Add support for guest setting of queue size virtio: Support transports which can specify the vring alignment Add MMIO based virtio transport hw/vexpress.c: Add MMIO virtio transport Makefile.objs | 1 + hw/vexpress.c | 3 + hw/virtio-mmio.c | 426 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ hw/virtio.c | 20 +++- hw/virtio.h | 2 + 5 files changed, 450 insertions(+), 2 deletions(-) create mode 100644 hw/virtio-mmio.c
