Re: [PATCH] VIRTIO_F_USED_EVENT_AUTO_DISABLE: add new used buffer notification suppression mechanism

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

On Fri, Jul 05, 2024 at 04:42:52AM +0000, Parav Pandit wrote:
> Hi Lege Wang,
> 
> 
> > From: Lege Wang <lege.wang@jaguarmicro.com>
> > Sent: Friday, July 5, 2024 9:05 AM
> > 
> > hi
> > 
> > Thanks for your comments in previous discussions.
> > I'd like to confirm that do you understand this proposal's idea now. If yes, I
> > may start to prepare a more formal V2.
> > 
> 
> Notification enablement offset adjacent to current notification offset seems a better approach than adding the new capability for following reasons.
> 
> 1. pci capability area (non-extended cap) is full to its capacity. New addition can only work in theory.
> An obvious alternative is to add the extended cap and add there.
> However, it is not a good idea either as the PCI spec strongly recommends to not keep adding vendor specific stuff in the PCI capability section.

Can you point me to the relevant section in the spec?

> (even though an option is kept for the vendor capability).

Interesting. EP guys are also asking about putting
the capabilities in a BAR. So we might want an option
to move all capabilities there.

> 2. New area adjacent to current offset can further is simple enough to utilize.
> There is no good reason to complicate by another extended cap.
> If there is one, lets discuss it first.

We might want to pass more data inline in the notification
going forward.
Also kick and suppression can be handled by different entities,
they would need to be in different pages.
Basically, don't combine unrelated things in one place.

-- 
MST