RE: [Xen-devel] problem with netfront.c

["Ling, Xiaofeng" <xiaofeng.ling@xxxxxxxxx>]

Ian Pratt <mailto:m+Ian.Pratt@xxxxxxxxxxxx> wrote:
>> Yes, but does it still need hypervisor to modify the phys_to_machine
>> table? I just hope to save some page faults or tlb flush vmexit to
>> get high performance and also have less modification to the
>> original netfront.c. So still only one multicall needed each
>> time when gets a net event.
> 
> The netfront/backdriver driver needs to be switched over to using the
> grant tables interface.
> 
> Chris Clark has already checked in the grant tables implementation,
> but no-one has got around to updating netfront/back to use it. There
> are patches to switch blkfront/back over to use grant tables, which
> will be checked in soon, but since the blk driver use the foreign
> access rather than the page transfer mechanism its probably not too
> much help. 
> 
> Using grant tables, the front end doesn't need to know about machine
> addresses, and the whole thing ends up rather cleaner, particulary for
> domains running with virtualized VMs.
Yes, there do have security problem to use machine address in netfront.
I'm originally  thinking that we can pass virtual address to backend through 
the ring buffer and let the backend to set the mapping for the frontend. 
I've already done a blkfront module driver in unmodified linux with an event 
channel channel  and ctrlif module driver. I'll wait for the new blkfront with 
grant table. For netfront, if no one has worked on the grant table update,
I can do it together.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel