Re: [PATCH v8 8/9] admin: command list discovery

[Cornelia Huck <cohuck@redhat.com>]

On Tue, Nov 22 2022, "Michael S. Tsirkin" <mst@redhat.com> wrote:

> On Tue, Nov 22, 2022 at 04:25:23PM +0100, Cornelia Huck wrote:
>> On Sun, Nov 20 2022, "Michael S. Tsirkin" <mst@redhat.com> wrote:
> Using 0x instead of h suffix? Sure.
>
> We should probably document the syntax in introduction.tex
> though this seems low priority.

Nod.
>> > +When \field{status} is VIRTIO_ADMIN_STATUS_EINVAL,
>> > +the following table describes possible \field{status_qialifier} values:
>> > +\begin{tabular}{|l|l|l|}
>> > +\hline
>> > +Status & Name & Description \\
>> > +\hline \hline
>> > +00h   & VIRTIO_ADMIN_STATUS_Q_INVALID_COMMAND   & command error: no additional information  \\
>> 
>> Either 0x00, or decimal (which one is better?)
>
> I think I prefer 0x here. And maybe I will add status values in both hex
> and decimal (I used decimal to be consistent with linux headers but
> fundamentally what we use most of the time is hex).

Ok.

>
>> > +\hline
>> > +01h   & VIRTIO_ADMIN_STATUS_Q_INVALID_OPCODE    & unsupported or invalid \field{opcode}  \\
>> > +\hline
>> > +02h   & VIRTIO_ADMIN_STATUS_Q_INVALID_FIELD    & unsupported or invalid field within \field{command_specific_data}  \\
>> > +\hline
>> > +03h   & VIRTIO_ADMIN_STATUS_Q_INVALID_GROUP    & unsupported or invalid \field{group_type} was set  \\
>> 
>> s/was set//
>> 
>> > +\hline
>> > +04h   & VIRTIO_ADMIN_STATUS_Q_INVALID_MEM    & unsupported or invalid \field{group_member_id} was set  \\
>> 
>> s/was set//
>> 
>> > +\hline
>> > +other   & -    & group administration command error  \\
>> 
>> Again the question whether this is something that can be defined per
>> group type.
>
> probably - above ones are generic, let's see if we need specific ones.
> if yes will be easy to add.

I think we want to distinguish between "reserved" (not defined yet, may
get a meaning later on) and "group type specific" (a group type may use
it, don't reuse for generic stuff). If we need group type specific
errors (and don't want a free-for-all), we could go with eg.

0x05 & VIRTIO_ADMIN_STATUS_Q_GROUP_ERR_0 & group type specific error \\

or so? Do we see any need for that yet?

>> > +The driver then enables use of some of the opcodes by sending to
>> > +the device the command VIRTIO_ADMIN_CMD_LIST_USE with a subset
>> > +of the list returned by VIRTIO_ADMIN_CMD_LIST_QUERY that is
>> > +both understood and used by the driver.
>> > +
>> > +If the device supports the command list used by the driver, the
>> > +device completes the command with status VIRTIO_ADMIN_STATUS_OK.
>> > +If the device does not support the command list, the device
>> > +completes the command with status
>> > +VIRTIO_ADMIN_STATUS_INVALID_FIELD.
>> > +
>> > +Note: drivers are assumed not to set bits in device_admin_cmds
>> > +if they are not familiar with how the command opcode
>> > +is used, since devices could have dependencies between
>> > +command opcodes.
>> > +
>> > +It is assumed that all members in a group support and are used
>> > +with the same list of commands.
>> 
>> Can the driver later change its mind, i.e. use VIRTIO_ADMIN_CMD_LIST_USE
>> a second time with a different set of commands? If yes, can it add
>> commands, or only withdraw them?
>
> I think it's ok to allow changing them arbitrarily at any time.
> If driver wants to "lock down" the list then all it has to do
> is send a list with VIRTIO_ADMIN_CMD_LIST_USE cleared.
>
> It seemed clear along the lines of since it's not prohibited it's
> allowed but since the question arose I will add a conformance clause for
> this.

So if the driver sends LIST_USE with an additional command, and the
device rejects that list (because of dependencies or whatever), is the
list of commands supposed to stay whatever it was prior to that?

>
>
>> What happens if a driver tries to send a command that it had not
>> included in the list?
>
>
> This is covered in conformance statements in the next patch.
>
> +
> +Device MUST validate commands against the list used by
> +driver and MUST fail any commands not in the list with
> +\field{status} set to VIRTIO_ADMIN_STATUS_EINVAL
> +and \field{status_qualifier} set to
> +VIRTIO_ADMIN_STATUS_Q_INVALID_OPCODE.

Ok.