svn commit: r208995 - stable/7/sys/dev/bge

[Pyun YongHyeon]

On Thu, Jul 01, 2010 at 06:28:35PM -0700, Garrett Cooper wrote:
> On Thu, Jul 1, 2010 at 2:52 PM, Pyun YongHyeon <pyunyh at gmail.com> wrote:
> > On Tue, Jun 29, 2010 at 10:13:45AM -0400, Ken Smith wrote:
> >> On Tue, 2010-06-29 at 09:48 +0200, Anders Nordby wrote:
> >> > Is it possible to get this fix into FreeBSD 7.3-RELEASE?
> >>
> >> We typically rely on developers to decide if fixes might warrant
> >> consideration as an Errata Notice and send email to re@ suggesting
> >> it. ?It's hard for us to judge on our own - factors include things
> >> like the individual developers' confidence level in the fix, etc.
> >>
> >
> > Even though I didn't receive positive feedback from jumbo frame
> > users I'm pretty sure the change set will fix regression introduced
> > in r199011. As you might know, the bug was not detected for a very
> > long time(about 6 months) so I think it's better to wait 2 or 3
> > weeks to get more confidence on the patch.
> 
> If you have jumbo packet functional tests, I'll gladly run it on some
> r200 machines we have in the lab with bge(4) cards sometime in the
> next couple of weeks. If not, I'll have to produce the tests which
> will take longer. I'll see whether or not some of the test code in
> http://wiki.freebsd.org/VictorBilouro/TCP-IP_regression_test_suite
> addresses this.
> 

To reproduce the issue you should be able to generate very high
rate of PPS and make kernel resource shortage condition. It would
be hard to trigger the issue with self-clocking protocols like TCP.
Alternatively, patching driver to add intentional delays and
reducing number of available mbufs may help to trigger the issue.

> >> If someone(s) were to send mail to re@ suggesting this be an
> >> Errata Notice given your description here we would definitely
> >> consider it seriously. ?As far as the general process goes that's
> >> the first step.
> 
> Thanks!
> -Garrett