'wrong pmg' error - SUMMARY (l

[Carlo Tiana]

In article <8904222139.AA11276 at eagle.cvs.rochester.edu> of this group,
I asked whether anyone knew how to deal with the fatal error that leads
a Sun 3/160 running 4.0 to crash with the error message:

vmunix: panic: wrong pmg

I was lost, as indicated by the line:
>What/who/where is 'pmg'? What's happening? What shall I do?

Well, I have got some useful pointers and am now a little less lost. In
case any of you out there are interested (am I *really* the only one with
this problem??), here is a summary of the replies I got:
[as an aside, the replies strictly came in increasing order of specificity
- ie every reply narrowed it down a little more than the previous one - it
was great: the next thing that was due to come in the sequence was a new
CPU board 8^)] [it did not come, yet]

From: Guy Harris <guy at auspex.com>
>What/who/where is 'pmg'?
	It's a "page map group".... while I was at Sun I heard lots and lots
	of references to "pmeg"s, which were "page map entry groups"; I don't
	know if "pmg"s and "pmeg"s are the same thing or not....
[he found out later - see below]


From: Laurence R. Brothers <quasar at ctt.bellcore.com>
	I got this panic (on an old sun 3/75) -- sun couldn't figure it out.
	Our repairman never heard of it. His technical support back home never
	heard of it. In a triumph of techincal brilliance he replaced the cpu
	board and the problem never reappeared....


From: Greg Wageman <"PSI%SINET1::PSI%SJSCA2::SNJSN1::SJ.ATE.SLB.COM
::GREG%sdr.slb.com"@RELAY.CS.NET>
[gotta love the above address 8^) Probably better: <greg at sj.ate.slb.com>]
	The "pmg" refers to the "page map group".  The kernel apparently
	keeps a shadow copy of a value it programs into the MMU.  When the
	kernel discovers that the hardware and software disagree on this
	value, it panics with the error we've seen.

	On the 3/75 with the problem, the message log gave the values
	expected and read.  In each case, there was a difference of 1 bit
	in the hardware, always the same bit.  We assumed it was a hardware
	problem, and replacing the CPU board has cured it.

	By the way, the reason you didn't see it under 3.x is that 4.x uses
	the page maps much more heavily, since files as well as virtual
	memory are handled via the page map.


From: Guy Harris <guy at auspex.com>
	I asked the guy who did most of the SunOS 4.0 VM code, and he said
	yes, a pmg and a pmeg are the same thing, and they are what I said
	they are. He also said that you probably have a down-rev CPU board;
	check with Sun, because there are probably ECOs that should be applied
	to it. (Basically, the OS will read items from the MMU, and find
	that the entry that is there doesn't match the entry that should be
	there - it has the wrong pmg/pmeg value.  Prior to 4.0, the OS was
	less likely to read stuff back from the MMU.)


From: David Barto <barto at net1.ucsd.edu>
	What you have is an out of rev board.  Get the patch CPU board
	fix [its number is] is 2550.


Thanks to all. I am getting the CPU board fixed.
Carlo.