Re: [Qemu-devel] [PATCH 3/3] hw/qxl: support client monitor configuration via device

[Alon Levy <alevy@redhat.com>]

> Hi,
> 
> Sorry for top posting, but trying to summarize this thread here.
> 
> I must say I like Gerd's approach, as it unifies code paths mostly,
> instead of having yet another interface where we do 2 way
> capabilities
> negotiation, with all the extra test matrix entries that would entice
> for full testing, we keep things simple.

So you are suggesting to send the message to both parties, and ignore it in the guest agent if it sees a qxl device. That's the only way this works, since otherwise you need a handshake between spice and qxl:

server
(1) receive VDAgentMonitorsConfig config
(2) call qxl->client_monitors_config
(3) wait for qxl->client_monitors_config_not_acked <-- after timeout? when do we decide interrupt wasn't cleared due to guest not supporting it, or due to not enough time having passed?
(4a) if timeout, send VDAgentMonitorsConfig to agent
(4b) else done

> 
> So we would have:
> 1) monitor config in rom space
> 2) QXL_INTERRUPT_CLIENT_MONITORS_CONFIG to tell the guest it is
> updated
> 3) Some way to avoid a new monitor config arriving and the guest
> being
>     busy reading the previous race.
> 4) The server will always update the monitor config in rom space
> 5) If the guest has not requested
> QXL_INTERRUPT_CLIENT_MONITORS_CONFIG
>     and there is an agent the server will send the monitor info to
>     the
>     agent
> 
> Note an alternative to the handshake suggested is simply adding a crc
> to the monitor config block. If that fails we hit the the (rare) race
> and
> the guest re-reads it.
> 
> Regards,
> 
> Hans
> 
>