RE: [virtio-comment] Re: [PATCH v2] virtio-net: support the virtqueue coalescing moderation

[Parav Pandit <parav@nvidia.com>]

> From: Alvaro Karsz <alvaro.karsz@solid-run.com>
> Sent: Saturday, February 11, 2023 10:47 AM
> 
> You write "both are configured", "try to configure both", "disable one and
> switch to other" "mixed mode" etc..
> 
> But no modes/levels are mentioned in the patch, not once.

using RX_SET and VQ_SET simultaneously implies "both, mixed mode, configured both".
Spec cannot be silent about it.
Did I miss the requirements section that describes the behavior of above two commands in combination?

> 
> I totally agree that *if* we define 2 modes, we should have strict rules on how
> to switch between modes and how to operate with both modes.
>
Even without any explicit mode, we need to define behavior of two commands invoked one after the other.
 
> But if we decide to have modes, your suggestion is not enough IMO, we should
> explicitly define the modes.
> 
> In my POV, we should have no modes at all, everything is per queue, and
> VIRTIO_NET_CTRL_NOTF_COAL_RX_SET is exactly the same as iterating
> through the RX queues and issuing a VIRTIO_NET_CTRL_NOTF_COAL_VQ_SET
> command.
> 
> Every queue will use the last coalescing parameters it received from the driver,
> regardless to the command used to deliver the parameters.
>
This to be part of the spec. I donât see it or I missed it?
 
> No modes at all.
> 
> I really don't see what we gain from having different operation modes.
> 
> I thought that we agreed not to have operation modes in v1.
We agreed to not have explicit mode, at condition to define the clear behavior on what should happen when both commands are used one after other.