From owner-freebsd-hackers  Mon Nov  4 15:21:43 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id PAA11369
          for hackers-outgoing; Mon, 4 Nov 1996 15:21:43 -0800 (PST)
Received: from plains.nodak.edu (tinguely@plains.NoDak.edu [134.129.111.64])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id PAA11348
          for <freebsd-hackers@freebsd.org>; Mon, 4 Nov 1996 15:21:37 -0800 (PST)
Received: (from tinguely@localhost) by plains.nodak.edu (8.7.6/8.7.3) id RAA10608 for freebsd-hackers@freebsd.org; Mon, 4 Nov 1996 17:21:31 -0600 (CST)
Date: Mon, 4 Nov 1996 17:21:31 -0600 (CST)
From: Mark Tinguely <tinguely@plains.nodak.edu>
Message-Id: <199611042321.RAA10608@plains.nodak.edu>
To: freebsd-hackers@freebsd.org
Subject: Virtual/phyical alignments.
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk


I allocated a contiguous block that needs to be 8k aligned. I needed
to calculate the current offset to keep to use with a hardware head/tail
combination. I have been using:

		/* rsqh == virtual address of head pointer */
	idt->rsqh = ((u_long) rsq) & 0x1ffc;

This has been working for several weeks and then today, after a compile
that added more paranoid checks, everything started failing. It took me at
least an hour to discover that (rsq & 0x1000) == 0x1000 not 0 as I would have
assumed. The physical address is 8k aligned but the virtual address is not.
A substitution of :

	idt->rsqh = vtophys((u_long) rsq) & 0x1ffc;

I know it is too difficult to assure alignments of physical/virtual are
the same in the kernel (we demanding people always have these "it would be
nice if" thoughts). Yes, I am glad it happened during developement instead of
hate mail and in the very least let this be a warning to others...
don't assume virtual alignment when allocating for physical alignment.

--mark "in the belltower with a high powered rifle :)".