[SunRescue] 4/260 speed clocking (adjustable?)

BSD Bob bsdbob at weedcon1.cropsci.ncsu.edu
Fri Dec 3 14:58:02 CST 1999

> > I have an old 4/260 crate that I am trying to get OpenBSD up on.  It runs
> > SunOS 4.1 fine, but, in loading up Net or OpenBSD, it gets a lot of scsi
> > dma timeout errors.
> Before you drag out the jumpers, check and rechecy your scsi cabling,
most particularly that blasted internal ribbon cable that seems to go bad
when you look at it.

I hesitate to call it scsi cabling, since it has these problems on every
sun vme crate I have using the older controllers and hardware.

It is non-destructive to the OS, so it points to being merely a timing
hardware problem for one end of the horsie needing to wait for the other
end.  That is a code thing and not a hardware thing.

> The behavior you are describing (workie fine under SunOS, scsi errors out
the wazoo under Net|Open-BSD) sounds suspiciously like some stuff we saw in
the (very) early days of the Linux Sparc port.  Flakey hardware was the answer. 

It could be hardware, but it doesn't feel like it.  The same code works fine
as the processors and hardware speed up.   It begins to have problems on
the older VME hardware that is possibly pushing the speed limit.

In the old days, we had fixes for that kind of thing by hard-coding a
few assembler nops here and there, to pad the timing out.

> SunOS was handling the errors with retrys silently and logging no errors.
Dave Miller had purged silent error handling out of his code in pursuit of
performance.  Both OSs were seeing the same lowlevel errors, Linux just
refused to go on.

But, it is not an error.  It is a handshake problem.  The system continues
to work fine, after it has had time to spit out the ``error'' message and
cycle through a bit reclocking a cycle.

I suspicion Sun may have used such tricks to silently to handle minor
timing problems.  Another cycle or two and things have settled down.
Nothing inherently wrong in that.

That is why I was wanting to try a slower clock speed, so that if it
really was only a timing thing, slowing down should reduce the number
of ``error'' messages.  Buying a wait cycle would do the same thing.

> Hope this helps.....

I would like to see some sun4 Linux vme scsi code.   Can you point me

> AL

Thanks, Al....


More information about the rescue mailing list