home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cs.utexas.edu!swrinde!mips!odin!fido!zola!zuni!anchor!olson
- From: olson@anchor.esd.sgi.com (Dave Olson)
- Newsgroups: comp.sys.sgi
- Subject: Re: VME differences between 4D/20 & 4D/35?
- Message-ID: <ntsommo@zuni.esd.sgi.com>
- Date: 29 Jul 92 22:46:05 GMT
- References: <1992Jul29.151301.25998@eagle.lerc.nasa.gov>
- Sender: news@zuni.esd.sgi.com (Net News)
- Distribution: na
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 37
-
- In <1992Jul29.151301.25998@eagle.lerc.nasa.gov> fsset@bach.lerc.nasa.gov (Scott Townsend) writes:
- | I have a machine here that was recently upgraded from a 4D/20 running 3.3.(2?)
- | to a 4D/35 running 4.0.1. It has a Bit3 VME-VME adaptor in it used to
- | connect to an external VME bus.
- |
- | Prior to the upgrade things had been running fine. I'm now experiencing
- | fairly odd behaviour which might be related to the VME implementation
- | in the 4D/35. "Odd behaviour" includes things like write(2) returning -1
- | _and_ setting errno to -1 (fixed by using memcpy from VME to local buffer
- | and then writing the local buffer to the socket), SIGBUS on valid VME
- | addresses in an almost repeatable fashion, etc.
-
- Any VME drivers you had running under 3.3 need to be rebuilt to run under 4.0.
- If all the boards are just memory mapped, with no VME driver at all, there
- shouldn't be a problem. The IP6 and IP12 VME addresses are all in the
- same ranges.
-
- | Is there anyone out there who could tell me what (if anything) has changed
- | in the VME implementation, or could point me to documentation?
- |
- | NOTE: there's no kernel driver code in question here, the VME adaptor merely
- | supplies a window into the remote VME bus via a suitable mmap() call.
- | Ordinary user-level code then uses the window to do reads & writes
- | of remote memory. The remote system does NOT access the Iris or pass
- | any interrupts to it.
-
- But does it do things on the VME bus that might affect the 35? That is,
- does it use the VME interface for anything other than the slave accesses
- from the 35? If so, there could be some problems. The actual hardware
- VME interface in the 35 is completely redone from the 20, and it may be
- more sensitive to strange VME bus states than the 20 (or less, it really
- hasn't come up as an issue before for slave devices that are mmap'ed).
- VME PIO is slightly faster on the 35 than on the 20/25.
- --
- Let no one tell me that silence gives consent, | Dave Olson
- because whoever is silent dissents. | Silicon Graphics, Inc.
- Maria Isabel Barreno | olson@sgi.com
-