Re: [PATCH] if_sk(4) rx/tx "hangs"

On Fri, Nov 12, 2004 at 05:39:40PM -0800, Peter Wemm wrote:
> On Friday 12 November 2004 05:04 pm, Daniel O'Connor wrote:
> > On Sat, 13 Nov 2004 11:17, Bjoern A. Zeeb wrote:
> > > Hi,
> > >
> > > the patch at [1] gave me 8-10MB/s depending on direction ftping 1G
> > > /dev/urandom data with the onboard NIC of my ASUS K8V SE Deluxe.
> > >
> > > This is the most I can get out of my 100Mbit/s consumer equipment
> > > and local ftp machine I guess.
> > >
> > > Any feedback appreciated.
> >
> > Try something less CPU intensive like /dev/zero :)
> 
> I think he was saying he created a 1GB file from the contents 
> of /dev/urandom and then timed the transfer of this over 100Mbit/sec 
> link.
> 
> Anyway, the point was that it worked!  Could the problem with the K8V SE 
> really be as simple as we've been hardcoding 128K of ram for a device 
> that only has 64K?
> 
Looks familiar.  I was in the lge(4) driver code recently, trying to
add polling(4) support but gave up.  The driver unconditionally uses
64/128 depth RX/TX lists while the chip only provides FIFOs for 31.
I haven't had much time to play with this NIC to determine if the
problem is real, but I suspect so.


Cheers,
-- 
Ruslan Ermilov
ru_at_FreeBSD.org
FreeBSD committer