On Mon, 6 Jul 2020 14:25:20 +0100 Stefan Hajnoczi wrote: > On Thu, May 28, 2020 at 05:35:55PM +0200, Cornelia Huck wrote: > > On Wed, 27 May 2020 11:29:21 +0100 > > Stefan Hajnoczi wrote: > > > > > Multi-queue devices achieve the best performance when each vCPU has a > > > dedicated queue. This ensures that virtqueue used notifications are > > > handled on the same vCPU that submitted virtqueue buffers. When another > > > vCPU handles the the notification an IPI will be necessary to wake the > > > submission vCPU and this incurs a performance overhead. > > > > > > Provide a helper function that virtio-pci devices will use in later > > > patches to automatically select the optimal number of queues. > > > > > > Signed-off-by: Stefan Hajnoczi > > > --- > > > hw/virtio/virtio-pci.h | 9 +++++++++ > > > hw/virtio/virtio-pci.c | 7 +++++++ > > > 2 files changed, 16 insertions(+) > > > > That looks like a good idea, since the policy can be easily tweaked in > > one place later. > > > > For ccw, I don't see a good way to arrive at an optimal number of > > queues. Is there something we should do for mmio? If yes, should this > > be a callback in VirtioBusClass? > > I looked at this but virtio-pci devices need to do num_queues -> > num_vectors -> .realize() in that order. It's hard to introduce a > meaningful VirtioBusClass method. (The problem is that some devices > automatically calculate the number of PCI MSI-X vectors based on the > number of queues, but that needs to happen before .realize() and > involves PCI-specific qdev properties.) > > Trying to go through a common interface for all transports doesn't > simplify things here. That makes sense, thanks for checking.