Xen devel - Apr 2004 - Role of Xen/Xenolinux in New IO networking

I know I have asked this before, but I am having trouble understanding
the role of each component in the new 1.3 world:
 
In 1.3 networking if Xen has no devices, and domain 0 owns and operates
the NIC - does Xen have any role at all in networking for the guest
OS''s?
 
Which network device driver will a guest OS use?
 
Do all the vnetif rules defined in Domain 0 now get interpreted by
Domain 0 vnet driver? Or still by Xen?
 
How do the actual packets (that will be routed to a guest) flow from the
physical NIC - (received by a domain 0 driver, routed/filtered by ???,
passed over a evtchn to a guest driver?)
 
Barry Silverman

> I know I have asked this before, but I am having trouble understanding
> the role of each component in the new 1.3 world:
>  
> In 1.3 networking if Xen has no devices, and domain 0 owns and operates
> the NIC - does Xen have any role at all in networking for the guest
> OS''s?
No -- Xen will have no direct involvement in I/O accesses, except
that:
 1. It will still own the ''emergency console'' (ie. serial
line) so
 that higher-level software can be debugged/post-mortemed.
 2. It may make special scheduling and CPU-allocation decisions for
 domains containign device drivers (eg. minimise wakeup latency when
 an interrupt is received).
 3. Communication between virtual device drivers and
 physical-device-driver domains will be via shared memory and event
 channels. These mechanisms need support within Xen, but they are more
 generally applicable than just device access.
> Which network device driver will a guest OS use?
Ordinary guests will still use a virtual network driver. However,
rather than talking to Xen to send/receive packets, teh driver will
have a communications channel to a device-driver domain -- this
channel comprises a shared-memory area and an event channel for async
notifications. 

This is the bit that is currently missing -- and the new world isn''t
much use without it!
> Do all the vnetif rules defined in Domain 0 now get interpreted by
> Domain 0 vnet driver? Or still by Xen?
The hideous VFR rules mechanism will be going away. Instead,
firewall/switching/routing rules will be written to the guest OS that
is runnign the physical device driver. If this guest OS is Linux, for
example, then you will get access to all the usual Linux mechanisms
--- much more fully-featured and robust than what currently exists in
Xen. 
> How do the actual packets (that will be routed to a guest) flow from the
> physical NIC - (received by a domain 0 driver, routed/filtered by ???,
> passed over a evtchn to a guest driver?)
We plan that the guiest OS conatinign the real driver will have a
network interface for each guest that is connected via it. It also
obviously has an interface for each real physical NIC that it
controls. Packets can then be routed between these interfaces just
like in a normal software router. We will add whatever fast paths are
necessary to avoid packet copying.

 -- Keir


-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

> Thanks Keir, 
> It is much clearer now. Do you plan to drop the code into the unstable
> tree a bit-at-a-time or one big change when it is done?
A bit at a time, as far as is possible. Maintaining forked trees is
too much hassle. Hopefully the two architectures can coexist, with the
appropriate one being selected at compile time (eg,. by building Xen
with or without ''nodev=y'').
> I have enjoyed watching the evolution...
> 
> If I can help in debugging, or testing, I would be glad to volunteer.
> 
> Am I right in saying, that at the moment, I can''t launch a non-0
guest?
> (I am using the demo CD for non-0 domains, and have been booting each
> from the CD. Domain 0 grants read-access to the physical CD to all the
> guests. 
> 
> The CD seems to be no longer accessible from a non-0 guest. [might be a
> configuration problem, though...]). 
If you did not build Xen with ''nodev=y'', and you did not
specifically
build real device drivers into Xenolinux, then everything should work
as it always has done. The CD should be accessible from other domains,
provided that appropriate VBD parameters are specified when the domain
is created.
> Is there any other structure I can use today to boot a non-0 guest with
> today''s 1.3?
You can still build Xen and Xenolinux as you normally would. Out of
the box all the new IO world is disabled. You have to explicitly
select it at build time by compiling Xen with ''nodev=y make''
and
configuring Xenolinux to allow physical device access (it''s a kernel
config parameter).

Once the new IO stuff is complete, the old architceture will be
removed and devices-in-domains will become the default build
configuration. 

 -- Keir


-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel