[Qemu-devel] KVM developer call minutes (Jan 19)

[Chris Wright]

Minutes (please reply w/ corrections or follow-ups):

how is stable branch working (both qemu and kvm)?
- qemu 0.12 cherry picking pretty aggressively
  - could use more community involvement
  - any distro patches for qemu/kvm packages?
    - doesn't sound like it

vhost-net command line syntax
- Anthony not a huge fan of current syntax, but will take it
- long term, 0.13, like to revisit all network syntax
  - feature driven UI instead of interface driven UI (user shouldn't
    have to know about tun/tap).
- hard to properly express proper raw socket interface
  - users confused by subtle differences between raw and tun/

SR-IOV network device status
- require new enough host kernel to allocate VFs
- once allocated, can be used as PCI networking device in the host
  - shared device w/ bridge and taps
  - assigned device, assigned to guest
- need vhost-net above to get VF dedicated to guest via something like
  raw socket
- not capable to manage the embedded bridge, host OS needs these interfaces

todo collection/status updates
- can use this call for some quick status updates on feature development
- could also use this to highlight areas where feature is stuck or the
  todo list could use some community help.
- may prove useful as a running todo list...

mmio bug fix
- Avi is adding SSE support (kills 8 byte assumption, can be 16 bytes)

vmchannel integration
- qemu merge status
  - buffering should be done by guest not qemu (to avoid using host memory)
    - can't inform guest about the size of buffer (can't be infinite)
    - makes transport unreliable
  - guest driver should not succeed on write until consumed by host
    - guest app writes buffer
    - put data into ring queue
    - move entries to used queue, now complete write (notify guest)
    - guest kernel can force guest app to block
- VNC clipboard copy/paste
  - working w/ Amit's scripts, but needs proper GNOME integration (d-bus
    pointers appreciated)
- virtio console just doing one-byte at a time
- qemu char device will either block when full or drop bytes
  - desparately needs to be re-written

SPICE code overview tomorrow (Wed, Jan 20, 15:00 UTC)
- Send email to Dor Laor <address@hidden> if interested

QMP status update
- working on feature negotiation, RFC coming tomorrow
- still working on switching libvirt to using QMP handlers
  - few series outstanding on libvirt list
- Markus working on self-description for protocol