Re: [Xen-devel] RE: information leaking between domains via granted pages?

[Keir Fraser <keir.fraser@xxxxxxxxxxxxx>]

On 4/3/08 08:53, "James Harper" <james.harper@xxxxxxxxxxxxxxxx> wrote:

>> The flipping path is not high performance these days as we expect only
>> very old domUs to use it. All skbuffs destined for a flipping
> interface
>> are laundered through netbk_copy_skb(). Note its judicious use of
>> __GFP_ZERO.
> 
> Hmmm... I thought that flipping would be a high performance path?
> 
> So if we use rx-copy, does this mean that data is received from a
> physical interface (or another domain) via DMA, and then actually copied
> into the buffers for DomU? Do we consider the performance implications
> of this too low to be worrying about?

Yes, it's a little surprising, but experiments showed that, even with lareg
datagrams, the performance of copying versus flipping was pretty comparable.
With normal Ethernet MTU copying was significantly faster. Also copying
gives us more flexibility wrt where we place the data in the domU page. So
flipping is now deprecated.

 -- Keir



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