From owner-freebsd-current  Tue May 28 12:30:00 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id MAA23299
          for current-outgoing; Tue, 28 May 1996 12:30:00 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id MAA23289
          for <current@FreeBSD.ORG>; Tue, 28 May 1996 12:29:52 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id FAA31591; Wed, 29 May 1996 05:25:44 +1000
Date: Wed, 29 May 1996 05:25:44 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199605281925.FAA31591@godzilla.zeta.org.au>
To: bde@zeta.org.au, terry@lambert.org
Subject: Re: memset/memcopy vs bzero/bcopy
Cc: current@FreeBSD.ORG, scrappy@ki.net
Sender: owner-current@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>If we have a processor where we can turn on unaligned access exceptions,
>and we turn them on, how much of the kernel code starts failing?  I would

None, because alignment checking is only done in ring 3 :-).  On
Pentiums you can count misaligned data references in counter 11.

>like to see this flag on by default for P5 or better systems...  the
>better to encourage writing code that can run on RISC systems.

The library certainly isn't well aligned.  The following example dumps
core in printf:

int ef;

main()
{
    asm("pushfl; orl $0x40000,(%esp); popfl");
    asm("pushfl; popl _ef");
    printf("%x", ef);
    asm("pushfl; andl $~0x40000,(%esp); popfl");
}

It works if 0 or 1 is printed instead of `ef' but fails if 0x40000 is
printed instead of `ef'

I know of only one compiler bug in this area.  The builtin memcpy()
doesn't bother aligning things.  The builtin memcpy() is used for
structs that aren't necessarily aligned.

Bruce