As for this, we also fall back onto the need to write a good virtual ne2k or
similar virtual firmware device. Maybe we could run such an item in host
space? (As below?)
guest OS host OS
\ /
guest device driver<--->virtual firmware device stub monitor
\ /
nexus<--->virtual device handler code
I base this speculation on the fact that many peripheral devices have there
own on-board custom processors these days--hiding the nitty-gritty from the
host machine and its physical domain.
Drew Northup, N1XIM
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf
> Of James Mastros
> Sent: Monday, December 18, 2000 12:01 PM
> To: [EMAIL PROTECTED]
> Subject: Re: [plex86] brain fart regarding networking
>
>
> On Mon, Dec 18, 2000 at 10:28:15AM -0800, Colin Davidson wrote:
> > Looks like I didn't express myself clearly. Funny, my thoughts were
> > perfectly clear to me (:-). I saw a driver in the host talking
> directly to a
> > driver in the guest, with no physical LAN card involved. The
> host would have
> > and own the physical LAN card(s). The Guest would think that
> the driver that
> > talked to the Host was driving an ethernet card and would treat it
> > accordingly. The Host would think the the driver that talked to
> the Guest
> > was driving a physical LAN card and would treat it accordingly. The
> > Pseudo-device on the Host side would need an IP (could get
> complicated if
> > another machine is doing DHCP resolution, although the
> Host/Guest linkage
> > could always use 192.168 pair of IP addresses with the host
> configured as a
> > router).
> >
> > If you see what I mean, perhaps you can describe it better. If
> not, perhaps
> > some questions would help clarify. Let's talk some more.
> I think I see what you mean now...
>
> In the host, we have a net device, lets say, virt0 with IP 10.86.0.1. In
> the guest, we have a net device, virt1 with IP 10.86.0.2. Now, for both
> sides, the external API thinks it's a normal network card.
> However, the way
> it acatualy communicates looks somthing like this:
>
> HOST-------------------- MONITOR -----------------GUEST
> API Raw packets API
> net stack -> virt0 -> plex86 -> virt1 -> net stack
>
> Now, between virt0 and virt1 sits the monitor. Instead of talking between
> them as if there was wire there, we remap the area where virt0's to-send
> buffer into guest space, and have the guest take an interupt.
> Then when the
> guest is done with it, the guest calls an interupt and we unmap
> it so virt0
> can send another buffer-full of data. (Of course, we could copy
> instead of
> remaping, but since we're playing MMU games anyway, we might as well.)
>
> If this is what you're thinking, it sounds pretty good, except, of course,
> that it relies on writing drivers for both sides. You have to do
> NAT on the
> host side if you want to get to the outside world and don't have
> a IP to do
> it, but if you don't, it's pretty simple routing.
>
> -=- James Mastros
> --
> midendian: She never sleeps.
> mousetrout: But I do. I just regret it after I wake up.
> AIM: theorbtwo homepage: http://www.rtweb.net/theorb/
> ICBM: 40:04:15.100 N, 76:18:53.165 W
>
