Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site rochester.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!wjh12!harvard!seismo!rochester!bukys
From: bukys@rochester.UUCP (Liudvikas Bukys)
Newsgroups: net.unix-wizards
Subject: obscure problem with 4.xBSD vaxuba/uba.c
Message-ID: <3474@rochester.UUCP>
Date: Thu, 15-Nov-84 19:47:20 EST
Article-I.D.: rocheste.3474
Posted: Thu Nov 15 19:47:20 1984
Date-Received: Sat, 17-Nov-84 04:37:58 EST
Distribution: net
Organization: U. of Rochester, CS Dept.
Lines: 41

This is probably not a real problem with any machines out there,
but I thought I'd point it out as a gift to posterity.

The ubamem() routine reserves pieces of the unibus address space for
devices with on-board memory (like 3Com Ethernet controllers).  One of
the things it does is twiddle the UBA configuration register to disable
some map registers...

#if VAX780
		/*
		 * On a 780, set up the map register disable
		 * field in the configuration register.  Beware
		 * of callers that request memory ``out of order''.
		 */
		if (cpu == VAX_780) {
			int cr = uh->uh_uba->uba_cr;

			i = (addr + npg * 512 + 8191) / 8192;
			if (i > (cr >> 26))
				uh->uh_uba->uba_cr |= i << 26;
		}
#endif

I believe this will disable a few too many map registers.
Consider two calls to ubamem, one which ends up with i==3,
another with i==4.  The result will be 7 banks of registers
disabled, not 4, which is what I think the result should be.
In my opinion, the correct code in that assignment statement
would be

				uh->uh_uba->uba_cr = (i<<26)|(cr&0x3ffffff);
				/* (notice the absence of '|=') */

I can't vouch for this, as I don't have *any* devices like this,
and it probably doesn't matter unless you stuff your unibus full
of them anyway.  I hope it helps some poor future hacker grepping
through unix-wizards archives, though.

Liudvikas Bukys
rochester!bukys (uucp) via allegra, decvax, seismo
bukys@rochester (arpa)