I had the following problem:
> We have just got a Fujitsu 2382 with a Xylogics 753 controller
> for our Sun 4/260 which has already a Fujitsu 2392 on a
> Xylogics 7053 controller installed.
>
> The controller which has been already there sits in the tenth
> slot of the backplane. Both subsystems work well with their respective
> controller inserted into this slot, but not when inserted into another
> slot.
>
> Putting one controller into the ninth slot and the other into the
> tenth lets the system crash during boot with a BAD TRAP, kernel read
> error, just when the controllers are accessed.
>
> Leaving the Xylogics 7053 in the tenth and adding the Xylogics 753
> into the eleventh slot lets the system boot up without problems,
> both controllers and disks are recognized:
>
> Mar 14 18:46:09 az3 vmunix: xdc0 at vme16d32 0xee80 vec 0x44
> Mar 14 18:46:09 az3 vmunix: xd0 at xdc0 slave 0
> Mar 14 18:46:09 az3 vmunix: xd0: <Fujitsu-M2392K cyl 1914 alt 2 hd 21 sec 83>
> Mar 14 18:46:09 az3 vmunix: xdc1 at vme16d32 0xee90 vec 0x45
> Mar 14 18:46:09 az3 vmunix: xd4 at xdc1 slave 0
> Mar 14 18:46:09 az3 vmunix: xd4: <Fujitsu-M2382K cyl 743 alt 2 hd 27 sec 80>
>
> But calling format or accessing xd4 with newfs results immediatly in
> lots of "spurious interrupt" messages.
>
> This lets me think, that I have to do something with the backplane.
> Are there any jumpers which must be set correctly?
Yes indeed (see below). My new disk-subsystem works fine, now.
Thanks to all who replied:
Jeff Nieusma <nieusma@cs.Colorado.EDU>
curt@ecn.purdue.edu (Curt Freeland)
daryl@dash.mitre.org (Daryl Crandall)
tom@yac.llnl.gov (tom slezak)
kevin%synergy@Sun.COM (Kevin Sheehan)
Sandy Napel <snapel@irus.rri.uwo.ca>
Steve Rumsby <steve@maths.warwick.ac.uk>
lentz@ody1.uchicago.edu (Gordon Lentz)
slf@beta.lanl.gov (Stuart L Flicker)
curt@ecn.purdue.edu (Curt Freeland) sent the following instructions:
> There are backplane jumpers you must remove to get the controller to work in
> the slot. You need to remove the BG3 and IACK jumpers from the backplane for
> the slot(s) you put the disk controller(s) into. For slot 11 they will be
> labeled P1103 and P1104.
>
> You should be able to figure it out pretty quickly from looking at
> slot 10's jumpers.
Jeff Nieusma <nieusma@cs.Colorado.EDU> gave an explanation why this is
necessary:
> yes, make sure you set the jumpers on the FRONT of the backplane...
>
> The bottom two jumpers pass the bus_request and bus_grant signals. So,
> if they are still in and you have another card above the card, here is
> a bad situation:
>
> 1. the 9th card requests the bus
> 2. the 10th card requests the bus
> 3. the CPU asserts the bus_grant line
> 4. the 9th card sees the bus_grant line and assumes it has the bus
> 5. the jumper passes the bus_grant signal
> 6. the 10th card sees the bus_grant line and assumes it has the bus
> 7. both cards try to use the bus...
>
> It might actually work without changing jumpers if you put the new card
> in the last slot
I tried this, but it didn't work. The disk was accessible, but after
some time it hang (during newfs, fsck).
Steve Rumsby <steve@maths.warwick.ac.uk> mentioned some problems he
had doing something similiar:
> I have a 3/280. For a while I had a Sun supplied 451 and a couple of 753s.
> Then the 451 was upgraded to a 7053 and the 753s stopped working. In fact
> the machine would crash if they were plugged in. The solution was to take
> out the 7053 and put it at the end of the bus (ie. highest numbered slot)
> and then everything worked fine. I've since had a further 753 put in the
> slot left by that and everything is still OK.
tom slezak <tom@yac.llnl.gov> sent some info which is related to this:
> you definitely want the faster 7053 before the 753 on the bus, and don't
> get too greedy setting a high throttle count on the 7053 or the 753 will
> never get any DMA slots.
Since I didn't understand this, I asked Tom to explain and he sent me
the following reply:
> On some controllers, you can set a "throttle count" to allow the controller
> to hog the DMA for multiple-word xfers. This is generally a bad idea in
> a mix of old and new (ie, fast and slow) controllers since the hog will
> cause the other controller to take too many latency hits.
>
> As your other responder noted, some people find the faster controller needs
> to be at the end of the bus, others need it earlier! A lot depends on your
> load and the delicate interplay on the bus between the controllers.
My 7053 now sits in slot 10 and the 753 in slot 11 and up to now
everything is working fine in this configuration.
Once again thanks to all who replied,
Franz Haberhauer
P.S.: Has anyone got performance data on the 753 vs. the 7053?
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:12 CDT