Re: qe0: xmit logic died, resetting...

[Johnny Billquist <bqt%update.uu.se@localhost>]

On 2020-08-19 08:23, Anders Magnusson wrote:
> Den 2020-08-18 kl. 16:48, skrev Silas:
> > On Tue, Aug 18, 2020 at 02:32:31PM +0200, Johnny Billquist wrote:
> > > On 2020-08-18 14:22, Mouse wrote:
> > > > if_qe.c sets an if_watchdog function which, if there is anything in
> > > > sc_inq when it's called, prints "qe0: xmit logic died, resetting..."
> > > > (or qe1, as appropriate) and calls qeinit().
> > > >
> > > > The reason this is interesting is that I'm seeing that message on my
> > > > uV2 simulator, and I'm quite sure the simulated DEQNA doesn't have
> > > > whatever real-hardware issue that code was designed to work around.
> > > >
> > > > This leads me to suspect a bug somewhere else that's leading to packets
> > > > sitting on sc_inq but not getting pushed to the hardware rather than a
> > > > hardware bug after all.
> > > >
> > > > Thoughts?  Is anyone else seeing that on a simulated DEQNA?
> > >
> > > Not sure, but I think I've seen it on my 4000/90 as well (well, ze0 
> > > instead of qe0), which would suggest that the code gets triggered for 
> > > a lot more than the original hardware bug of the DEQNA.
> >
> > Hi.  I've also been seeing this when trying to run NetBSD on my VAX 
> > 4000/90 too, using ze0:
> >
> > ze0: xmit logic died, resetting...
> >
> > This happened when using a diskless environment that booted fine, but 
> > started to fail with this error message on the first heavy read/write 
> > operations on NFS...
> >
> > That is what stopped me from playing with my beautiful VAXstation 
> > 4000/90 that is laying around since 7.0 days...
> >
> > There are some old messages in port-vax about that, probably an 
> > unrelated bug?
> The DEQNA is likely to stop transmitting once in a while (at least the 
> earlier firmwares), especially on busy shared networks IIRC.
> But the SGEC in the /90 should be quite stable.  I have one myself and 
> never gets any error, so I haven't reflected over it.
>
> Most likely it is something missed in the driver that causes it to stop 
> (or do not turn off the timer by some reason).

Like with Silas, I see it on a machine with lots of nfs traffic. While 
I'm not running diskless, I do have /usr/src mounted over nfs. When 
stressing the system, it can happen a lot. But unlike Silas, for me this 
isn't a big problem, so I never bothered with it. The system continues 
operating, and the network is also still good. It's just annoying 
messages, and I think the network gets a temporary hiccup at that point 
in time. But it recovers.

  Johnny

--
Johnny Billquist                  || "I'm on a bus
                                  ||  on a psychedelic trip
email: bqt%softjar.se@localhost             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol