Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 9 Apr 2004 13:31:45 -0700
From:      Mike Hunter <mhunter@berkeley.edu>
To:        freebsd-net@freebsd.org
Subject:   Re: Broadcom BCM5788 jumbo frames? (FBSD 5.2.1)
Message-ID:  <20040409203145.GC8722@ack.Berkeley.EDU>
In-Reply-To: <20040406180020.GA21071@ack.Berkeley.EDU>
References:  <20040406180020.GA21071@ack.Berkeley.EDU>

next in thread | previous in thread | raw e-mail | index | archive | help
On Apr 06, "To freebsd-net@freebsd.org" wrote:

> Hi,
>
> I have a Broadcom BCM5788 card that I'd like to use with jumboframes, but
> it seems that the card (or the driver) only allows a max of 1500 MTU.
>
> ifconfig bge0 mtu 9000
> ifconfig: ioctl (set mtu): Invalid argument
>
> 1500 and less works.
>
> I had this problem with my BCM5707M, and lo and behold if you look in
> if_bge.c it says it can't do it, but I looked and I can't find a similar
> comment for the BCM5788.
>
> Any thoughts?  The system in question is 5.2.1-RELEASE-p3.  The bge card
> is builtin to the motherboard (shuttle sb75g2 on 64 bit pci.)
>
> While I have everybody's attention, what's a reasonable goal for
> throughput between two of these cards?  I broke through the 33mhz pci
> barrier and am now getting iperf results of:
>
> [  6]  0.0-10.0 sec   535 MBytes   449 Mbits/sec
>
> ( Obviously, without jumbo frames.)
>
> I've increased the tcp send and recv space....Any other suggestions?

I received this off-list:

There's nothing to think about: Broadcom NICs based on the 5705 ASIC
do not support jumbo frames. You did not bother to provide the dmesg
output from your system, so I can't see the ASIC rev from the probe
message, but I'm pretty sure it's in the 5705 family. Compare the
output against the ASIC revision listing in if_bgereg.h.

There's no driver jiggling involved: the hardware just won't do what
you're trying to do.

As far as expected performance goes, if you actually have a 64-bit
NIC (or a 32-bit/66Mhz NIC), you should be able to reach 900Mbps
speeds with TCP, although your chances of achieving this are better
if you have checksum offload capability.

Mike



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040409203145.GC8722>