Re: [PATCH v3 1/2] virtio-net: Describe dev cfg fields read only

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

On Wed, Feb 22, 2023 at 01:07:07PM +0100, Cornelia Huck wrote:
> On Wed, Feb 22 2023, "Michael S. Tsirkin" <mst@redhat.com> wrote:
> 
> > On Wed, Feb 22, 2023 at 10:01:24AM +0100, Cornelia Huck wrote:
> >> On Tue, Feb 21 2023, "Michael S. Tsirkin" <mst@redhat.com> wrote:
> >> 
> >> > On Tue, Feb 21, 2023 at 05:59:52PM +0000, Parav Pandit wrote:
> >> >> 
> >> >> > From: Michael S. Tsirkin <mst@redhat.com>
> >> >> > Sent: Tuesday, February 21, 2023 12:52 PM
> >> >> > 
> >> >> > On Tue, Feb 21, 2023 at 05:50:09PM +0000, Parav Pandit wrote:
> >> >> > > Hence, it should be mentioned as read-only fields, so when the driver writes
> >> >> > something to read-only fields, it can be considered as undefined behavior on
> >> >> > such fields.
> >> >> > >
> >> >> > 
> >> >> > In the description not in the normative statements. normative sections just tell
> >> >> > driver what it must and must not do, in the standard RFC terms.
> >> >> > 
> >> >> Got it.
> >> >> I will shift them as read-only in the description section.
> >> >> And normative in the device and driver section.
> >> >> Device section:
> >> >> Any writes to config space fields is ignored by the device, because these are read-only fields for the driver.
> >> >
> >> > writes is plural so "are ignored"
> >> >
> >> > but more importantly use rfc terms in normative sections.
> >> 
> >> I don't think you need to talk about "read-only" in the normative
> >> sections (that belongs to the descriptive sections.) I'd use
> >> 
> >> "The device MUST ignore any writes to config space fields by the
> >> driver."
> >
> > Hmm. Is this something we previously required for read only fields?
> 
> So, better make it SHOULD?
> 
> (The only alternative to ignoring I see is breaking the device, and I
> think ignoring is preferable.)

We can just skip adding a new requirement completely - we'll never get
there with a compliant driver.  This is what we do e.g. for MMIO.
Why not?
This has an advantage as this allows backing config with regular RAM.
Also I feel that since it always said "read only for driver" then
this implies a restriction on driver not the device.

> >
> >
> >> >
> >> >> 
> >> >> Driver section:
> >> >> Driver must not write to read-only fields.
> >> 
> >> "The driver MUST NOT write to any config space field."