[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[XenARM] RE: Project Planning



Hi Xing, 

> -----Original Message-----
> From: Xing Zhao [mailto:xingzhao@xxxxxxxxxxxxx]
> Sent: mardi, 5. mai 2009 05:57
> To: ROSSIER Daniel
> Cc: GERBER Patrick; CURCHOD Gilles
> Subject: RE: Project Planning
> 
> Hi, Daniel,
> 
> Yes. I want to help. But I am still not sure about the general idea of
> this
> "embedded image approach". I am kind of slow on this one.
> 
> I think the high-level design choices that we have are
> 
> => linux as the resource manager
>     => xen runs as a linux application, that is, a hosted vmm;

No, not really. The hypervisor is the first component which starts at the 
beginning. The Linux domain will start right after
the domain creation (now it is still miniOS which is considered, not Linux). 
The hypervisor and the domains will have their own 
page tables, which guarantees perfect isolation between domains. Nevertheless, 
the code of XEN, Linux and miniOS is currently located (mixed) at the same 
place, namely 0xc0000000. All the XEN-related data will be placed at 
0xff000000, leaving free the virtual address space between "_end" and 
0xff000000 (including I/Os and DMA mapping) required by Linux and miniOS. 

I will send a pdf which shows how the virtual memory address space looks like.

> => xen as the resource manager
>     => no hw assistance / with linux modifications
>         => binary modification
>         => source modification, that is, paravirtualization
>     => with hw assistance / no linux modifications
> 
> If we go by the paravirtualization scenario but use a standard kernel
> instead, then after the kernel starts, it's going to set up its page
> table,
> (re)configure and (re)enable mmu, schedule all the tasks, handle all
> exceptions and interrupts, and so on. I don't see how Xen can regain
> control
> of the machine, let alone managing all the resources. Please let me
> know if
> I missed anything here.
> 

You're right. But we will not consider full virtualization since it does not 
make sense for the time being in the embedded world (only a few recent 
ARM-cores support instructions for virtualization mechanisms), and this will 
probably introduce too much overhead.

So, we need to adapt the initialization code of Linux in order to be start it 
as guest OS on top of XEN. This will be done after miniOS correctly boots.

> Best regards,
> 
> Xing
> 
> 

Daniel
ps: I also push your mails to the mailing list in order to keep the community 
informed.



_______________________________________________
Xen-arm mailing list
Xen-arm@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/mailman/listinfo/xen-arm


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.