Re: nfe problem on 8.0-BETA2

On Thu, Aug 20, 2009 at 02:07:39PM -0400, Travis Mikalson wrote:
> Pyun YongHyeon wrote:
> > On Thu, Aug 13, 2009 at 07:56:05PM -0700, Ryan Rogers wrote:
> >> Pyun YongHyeon wrote:
> >>> On Wed, Jul 22, 2009 at 03:26:43PM -0700, Ryan Rogers wrote:
> >>>> Weongyo Jeong wrote:
> >>>>> On Wed, Jul 22, 2009 at 05:26:07AM -0500, Sam Fourman Jr. wrote:
> >>>>>>> svn co svn://svn.freebsd.org/base/head
> >>>>>>> svn merge -c -193289
> >>>>>>>
> >>>>>>> Thereafter, in the root of the repo:
> >>>>>>> svn up
> >>>>>>> svn merge
> >>>>>> Confirmed, running the above set of commands fixed my 88E1116 nfe 
> >>>>>> problem
> >>>>>> everything works as it should now. thanks guys for the help.
> >>>>>>
> >>>>>> Sam Fourman Jr.
> >>>>>>
> >>>>>> ps. svn was WAY faster than csup, I think I am going to use svn all
> >>>>>> the time now.
> >>>>> I know that yongari_at_ knows what the problem is and the solution but he
> >>>>> could not access the internet right now due to relocation to USA.
> >>>>>
> >>>>> It looks he is busy with looking for house and etc...
> >>>>>
> >>>>> regards,
> >>>>> Weongyo Jeong
> >>>>>
> >>>>>
> >>>>>
> >>>> Well, the good news is that r193289 was the only thing keeping my nics 
> >>> >from working.  I was able to update to current as of today and revert 
> >>>> that patch, and everything is working nicely now.  So I'll just keep my 
> >>>> eye on the commit logs and patch that up by hand whenever I need to 
> >>>> update until yongari is able to get situated.
> >>>>
> >>> Would you try attached patch and let me know how it goes on your
> >>> box?
> >> The patch works great, thanks!
> > 
> > Thanks for testing.
> > Unfortunately the change requires more testing on various
> > controllers. 88E1116 PHY is commonly found on Yukon Ultra or newer
> > Yukon controllers as well as NVIDIA controllers. Since we are in
> > 8.0-BETA stage changing common PHY code at this time looks
> > dangerous.
> 
> Before I start a bug report, I'm curious, do you think it is possible
> for this issue to also affect the general stability in operation of the
> nfe driver? I mean if nfe comes up and works fine for many days and then
> suddenly stops passing traffic?
> 
> In both 8.0 and 7.2-STABLE some time after the beginning of June, two
> separate systems with two different nfe NICs started showing exactly the
> same issue. The 7.2-STABLE system was built from June 16th source and
> has a Marvell 88E1116. The 8.0-BETA2 system has a Marvell 88E1111 and is
> built from Aug 13th source. Both systems did not ever have this issue
> with code from before June (7.2-RELEASE included).
> 

Sorry, I don't know. Both Marvell, vendor of PHY hardware and NIVIA
, vendor of ethernet controller didn't release documentation to
open source developers, so some part of code might be wrong on some
controllers. I think nfe(4) didn't have code changes during that
time window. e1000phy(4) received a couple changes to support LED
for Yukon FE+, Yukon Ultra and Yukon Extrme. But I don't think it
may cause problems on nfe(4) as the change only affects LED
blinking rate. You can verify it by backing out r193291.

> Both will stop passing traffic across their nfe0 NIC randomly and
> require an ifconfig nfe0 down; ifconfig nfe0 up to continue. In the
> meantime before the interface is reset, ping says sendto no buffer space
> available and named starts spitting out a lot of "error sending
> response: not enough free resources" from the DNS queries it was in the
> middle of responding to.

By chance are you using jumbo frame? Before digging into source
code it would be even better to know which controllers you have.
Just sending dmesg(1) output related with nfe(4)/e1000phy(4) would
be enough to know that.