Still have BCE driver issues (dell pe 1950) and NFS

[Scott Long]

Josh Paetzel wrote:
> On Tuesday 07 November 2006 10:10, Scott Long wrote:
>> Josh Paetzel wrote:
>>> On Tuesday 07 November 2006 03:30, Fredrik Widlund wrote:
>>>> Hi,
>>>>
>>>> Will a fix/this fix be part of the 6.2 Release? We will be
>>>> relying heavily on fbsd6.2 and pe1950 and are worried about the
>>>> BCE stability?
>>>>
>>>> Kind regards,
>>>> Fredrik Widlund
>>>>
>>>> Scott Long wrote:
>>>>> Olivier Mueller wrote:
>>>>>> Le 7 nov. 06 à 01:15, Scott Long a écrit :
>>>>>>> Olivier Mueller wrote:
>>>>>>>> NFS Server: dell poweredge 1950, with the 1.2.2.6 version of
>>>>>>>> if_bce.c: bce0: <Broadcom NetXtreme II BCM5708 1000Base-T
>>>>>>>> (B1), v0.9.6> mem - Start a directory listing on it:
>>>>>>>> immediate (network) crash of the NFS
>>>>>>>>   server.   (reproduced 3 times)
>>>>>>> Do the following, then retry your test:
>>>>>>> ifconfig bce0 -txcsum
>>>>>> Oh, this way it looks much better, thanks.  Directory listing
>>>>>> was fine, and copying files during 2-3 minutes over NFS worked
>>>>>> as well. More tests will follow tomorrow.
>>>>>>
>>>>>> Next step? :-)  Should I put that command somewhere in my init
>>>>>> scripts, or even directly in rc.conf, or wait for a new
>>>>>> if_bce0.c version?  I am available for any other
>>>>>> PE1950-related tests if this may help.
>>>>>>
>>>>>> Regards,
>>>>>> Olivier
>>>>> Change /sys/dev/bce/if_bcereg.h so that BCE_IF_HWASSIST is
>>>>> defined to 0. Then recompile.
>>>>>
>>>>> Scott
>>> I know I've brought this up before, but I have a PE1950 with a
>>> pair of bce nics that get pounded on 24/7.  I've been using 6.1-R
>>> with the 0.9.6 version of the bce driver for a couple of months
>>> now.  The driver is available here: 
>>> http://yogurt.org/FreeBSD/if_bce.c  I've emailed the author of
>>> the driver and I've at least mentioned it to Scott once but I
>>> really can't understand why we don't just import this driver into
>>> the tree.
>> What you just posted is exactly what committed to CVS for 7-CURRENT
>> and 6-STABLE, and what was proven to break down under moderate to
>> heavy UDP traffic.  I don't doubt that your servers have a load
>> that doesn't trigger the problem, but if you're curious I can send
>> you a couple of very simple test cases that will cause your driver
>> to panic and your network interface to wedge.
>>
>> Scott
> 
> My bad.  Thanks for clearing this up. (In my case pretty much all of 
> the traffic is TCP which I guess would explain why it's working for 
> me)
> 

Actually, I think that I was a little unclear.  The link that you posted
is what was in FreeBSD prior to me fixing all of the problems.  What is
in FreeBSD now fixes just about every problem for every person.  There
are still a few reports of mystery problems, but they are in very rare
and specialized cases.  In all, the driver works now.

Scott