Intel em receive hang and possible pr #72970

[Jack Vogel]

On 9/1/06, Rob Watt <rob at hudson-trading.com> wrote:
> >> On Thu, 31 Aug 2006, Joe Holden wrote:
>
> >>> Sounds like its at least possible this is your problem, worth setting up a
> >>> system to test with I would say.
>
> > There's also another possibility these days -- we support errata fixes
> going
> > into release branches, as we do with security fixes.  These changes must be
> > low-impact (not affect API, ABI, etc), but are useful for critical stability
> > fixes or to correct widely experienced problems.  If there is a small tweak
> > which could make a big difference, and has been well QA'd, then that approach
> > can be taken.  However, we're currently accepting errata fixes only for the
> > most recent release, so expanding the scope to include older releases
> > (i.e., 6.0 in addition to 6.1) then that would require some discussion
> > and careful thinking.  So far, we've been very conservative in accepting
> > errata fixes
> > on the basis that we want it to be a trickle, not a waterfall, for a
> > great many good reasons :-).
>
> > Robert N M Watson
> > Computer Laboratory
> > University of Cambridge
>
>
> Thanks for all of the quick feedback.
>
> Stability and consistency are very important to us. We are not going to
> use STABLE on one machine and a RELEASE on another, and we are not going
> to roll out some random version of stable to all of our machines (there
> are many of them). When troubleshooting we need to be able to have a
> good standard reference to talk to people about and to compare from
> machine to machine. This is possible when using a full release, but much
> less so when dealing with the head of a branch. We are willing to apply
> small patches provided they are well tested (we are willing to do the
> testing).
>
> I am happy to test against 6.1-RELEASE. We are considering upgrading
> all of our servers to 6.1, so I have started testing 6.1 anyways. It is
> not clear to me from the cvs log whether rev 1.129 in branch MAIN
> mentioned by Mikhail will work with 6.1 (I suspect it will not). The
> 1.65.2.18 rev against RELENG_6 seems to include the changes in 1.129. I
> am going to start testing that tonight. How extensively have the changes
> in that version been tested? Will that version play nicely with 6.1?

This was my earlier point Rob, that delta of the driver WILL NOT
build against 6.1 RELEASE, it has taskqueue changes in it that
are not in the release, as well as a few other small gotchas.
To get that fix you will need to wait for 6.2 if you want it all rolled
into a determined whole.

Jack