On Nov 14, 2006, at 7:24 AM, Izumi Tsutsui wrote:

> Hmm.
>
> Should some upper layer provide such function?

Yah, that might be useful.

> Well, but for now most of our drivers which pad ETHER_ALIGN on RX
> manually allocate a new mbuf (with a cluster if needed) and copy
> entire each packet by memcpy(9), don't they? ;-p

I think you mean which do NOT pad with ETHER_ALIGN on Rx.

> The vge(4)'s method looks a bit faster than that.
> (though maybe I should have more precise benchmarks)

Yes, that may be true, but it still requires bringing the entire  
packet into the data cache which would otherwise not be necessary (if  
you're using checksum offloading).

-- thorpej