cvs commit: src/sys/net if.c

[Scott Long]

On Fri, 24 Sep 2004, Brian Fundakowski Feldman wrote:
> >
> > This is mostly solved, and Julian is just tracking down some edge cases
> > that people have reported.  I'm not sure if we are going to turn on
> > PREEMPTION for 5.3, though.
>
> Every time it "appears" to be fixed, I can't get any kind of stability
> with SMP, PREEMPTION, and SCHED_ULE; I'm certainly willing to try it
> each time it looks to be fixed.

Would you please copy me on the reports that you've sent to Julian so I
can track them also?

>
> >
> > > * Reports of UFS2 "large disk" problems
> >
> > I guess you haven't been paying attention to all of the good work that
> > Pawel is doing here.  I also have fixes in my queue for growfs.
>
> Just based on observation of "what of these things look stalled," not
> meant as any slight to pjd.

Well, just because it doesn't happen on your IRC channel doesn't mean that
it's not happening ;-)

>
> > > "Things not on the lists that I would love to be fixed":
> > > * IPv6 route lookup recursion I've fixed already/still getting review
> > > * ugen detachment crash I've fixed already/still getting review
> > > * BPF panics due to bpfdetach() on live interface
> > > * cbb child interrupt handler deregistration race+panic
> > > * SoftUpdates panics
> > > * anything from http://www.holm.cc/stress/log/
> > >
> >
> > Of the things are you are working on right now, how many do you think
> > you'll have fixed in the next week?  I'm definitely not trying to rush
> > you, just curious on what you have planned.
>
> From this list:
>
> The IPv6 routing error/INVARIANTS-crash should be in shortly.
> The ugen detachment requires imp's testing; since he backed
> out my previous suboptimal workaround, and then unbacked it out, it is
> back to not being a crash again but could (interruptibly) hang an
> application using the ugen device.  The bpfdetach() is something that
> should be fixed after rwatson's latest bpf.c delta, but needs
> testing.  The cbb problem problem is fixable by adding locking or by
> adding ithread pinning, but I am unable to do anything about it.

Sounds good, thanks!

Scott