Subject: Re: openssh unaligned access in swap_bytes (on alpha)
To: Simon Burge <firstname.lastname@example.org>
From: Jason R Thorpe <email@example.com>
Date: 10/22/2000 11:07:28
On Mon, Oct 23, 2000 at 03:46:07AM +1100, Simon Burge wrote:
> So the arguments obviously aren't 8 (or even 4) byte aligned, contrary
> to the comment before swap_bytes() in cipher.c.
> SA-110 (233MHz) 1.6488 1.8607 1.6500
> I'm tempted to leave it at 4 byte loops. The code is clearer and
> overall it doesn't seem to make that much off a difference either way.
> I've tested this change actually in ssh on alpha and sparc (just to
> check that I didn't botch any endian-type stuff).
> I also think that such a change should also be pulled up to 1.5...
Agreed -- without the fix, SSH will simply not function on e.g.
ARM systems (unaligned accesses produce unexpected results rather
than faults on the ARM).
-- Jason R. Thorpe <firstname.lastname@example.org>