bouyer@antioche.eu.org wrote:

> > Both patches (padding mbuf, or padding by zero'ed dmamem) are attached.
> 
> I find the second better. There shouldn't be much differences in performances,
> but avoiding a mbuf allocation if possible is always better :)

The latter has been committed.

BTW, it seems a bit faster then the former on my Pentium 90 PC
(because struct re_txq is cache aligned?) but I guess there is
few difference on modern machines...
---
Izumi Tsutsui.