[Ksummit-2008-discuss] proposal for discussion..

Evgeniy Polyakov johnpol at 2ka.mipt.ru
Mon Aug 25 12:34:50 PDT 2008


Hi Marcel.

On Mon, Aug 25, 2008 at 07:02:16PM +0200, Marcel Holtmann (holtmann at linux.intel.com) wrote:
> I do have a basic prototype running. It needs some extra polishing  
> before I can publish it. It is implemented as AF_DBUS. For  
> compatibility we just add another transport to the D-Bus library to  
> talk to AF_DBUS instead of dbus-daemon. In the end we hope that dbus- 
> daemon on Linux systems goes away and we have full zero-copy for  
> messages between processes.
> 
> Let me repeat one thing. Nobody wants to put HAL into the kernel :)
> 
> If the kernel has direct access to the D-Bus message bus, we could use  
> that to export APIs from the kernel. It is similar to netlink, but  
> with the advantage that D-Bus has bindings for C, C++, Mono, Python,  
> Perl etc. already and these will all keep working without any breakage.

Just a small note (not to break this healthy discussion about protocol
design errors, which do not allow to extend it, after protocol was
released to wild nature of the userspace horrors, when kernel needs
upgrade): very likely using own socket family for the
kernel/userspace protocol is a bit excessive. I would even say that it
is an overkill, since netlink was specially designed for exactly this
kind of tasks. Its attribute-based architecture allows infinite
extension with ability to deprecate some arguments and introduce new
ones without backward dependency breakage.

Anyway you need some abstraction layer, since right now no one can call
socket(AF_DBUS), so no one will really notice that something was changed
to use netlink.

-- 
	Evgeniy Polyakov


More information about the Ksummit-2008-discuss mailing list