Subject: Re: Some observations about DMA segment count
To: None <tech-net@netbsd.org>
From: Jason R Thorpe <thorpej@zembu.com>
List: tech-net
Date: 07/26/2001 12:02:09
On Tue, Jul 24, 2001 at 08:32:41PM -0700, Jason R Thorpe wrote:
> Thoughts on what to do about this? There should certainly be room in front
> of the packet for the link header, and we should try to avoid allocating
> the extra mbuf.
So, FWIW, I fixed this. See M_READONLY() in sys/mbuf.h, and how it is
used by M_LEADINGSPACE() and M_TRAILINGSPACE().
Between this and another change (which I will commit shortly) to use
pool_cache_get() for mbuf/cluster allocation, I've gained another 2000KB/sec
on my Gig-E benchmarks with the Sundance TC9021 chip.
...time to do another kernel profile, I suppose :-)
--
-- Jason R. Thorpe <thorpej@zembu.com>