Re: [virtio-dev] RFC: Doorbell suppression, packed-ring mode and hardware offload

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Feb 04, 2019 at 01:36:28PM +0800, Stefan Hajnoczi wrote:
> On Fri, Feb 01, 2019 at 09:43:02AM -0800, Rob Miller wrote:
> > Agreed that this is needed.
> > 
> > I would also like to suggest splitting the F_IN_ORDER into
> > F_RX_IN_ORDER and F_TX_IN_ORDER to support hw LRO implementations,
> > which can be more of a scatter/gather than tx. This would allow
> > batchmode for tx at least in packed rings.
> > 
> > Finally, i would suggest a means to specify a given rings ring mode
> > and packed leans more towards TX, whilst split can be either really
> > depending upon LRO, jumbo, rx buff size, ect.. just like F_IN_ORDER,
> > we can have RX & TX, split out.
> 
> Device types beside virtio-net might also want per-virtqueue F_IN_ORDER
> and other features,

Do you know that they will? Now that we already have implementations,
how about a prototype patch showing the performance gains?

> so let's find a way to make it independent of
> virtio-net concepts like rx/tx.

Add another field along with VQ PA. It's not rocket science.

> The per-virtqueue in-order flag could live in struct
> pvirtq_event_suppress (VIRTIO 1.1 2.7.14 Event Suppression Structure
> Format).

That's for dynamic things though.

> Or a new structure could be used for per-virtqueue configuration.  Doing
> this is a little tricky if you want to select split vs packed ring
> layout on a per-virtqueue basis, since this structure is part of the
> split/packed ring layout.  It would probably be necessary to add vring
> mode selection to the transport (PCI, MMIO, CCW) instead.
> 
> Stefan

Yes.

-- 
MST