Subject: Re: NVIDIA nForce ethernet driver commited
To: Jonathan A. Kollasch <email@example.com>
From: Chuck Silvers <firstname.lastname@example.org>
Date: 03/17/2006 07:45:46
On Wed, Mar 15, 2006 at 03:01:33PM -0600, Jonathan A. Kollasch wrote:
> On Sun, Mar 12, 2006 at 03:17:56PM -0800, Chuck Silvers wrote:
> > hi folks,
> > I ported OpenBSD's "nfe" driver for the NVIDIA ethernet and added it
> > to -current just now. please try it out and report any problems
> > (or successes, for that matter).
> On a nForce 4 board, running:
> NetBSD dolomite.xxx.xxx.xxx 3.99.17 NetBSD 3.99.17 (GENERIC_ACPI) #0: Wed Mar 15 04:38:49 CST 2006 email@example.com:/mnt/wd1a/jakllsch/dolomite-current-20060315/src/sys/arch/amd64/compile/obj/GENERIC_ACPI amd64
> and a 3.0 userland.
> nfe0 at pci0 dev 10 function 0: ioapic0 pin 5 (irq 5), address 00:04:61:9d:xx:xx
> ciphy0 at nfe0 phy 1: Cicada CS8201 10/100/1000TX PHY, rev. 3
> pcictl shows device ID 0x0057.
that's the same one I have. mine has a makphy though.
> I can get full (a 9.6MB/s scp) receive performance, yet transmit is bad,
> it starts at about 3 MB/s then drops to about 24 KB/s. Sometimes it
> seems to lock up the DSS-8+ HW ver G1 switch; the 10Mbit hub doesn't seem
> to lock up. But it's useable enough to connect to home and write this
this is probably caused by the same problem as the remaining mbuf leak.
I don't know what that is yet, though.
try lowering the MTU. for me, setting it to 243 or lower gets good tx
performance, but 244 or higher causes the tx throughput to fall off a cliff.
> However this is great progress for NetBSD. Truly amazing.
> Jonathan Kollasch