--7AUc2qLy4jB3hD7Z
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Thu, Apr 13, 2006 at 06:47:23PM +0000, Edward B. DREGER wrote:
> RK> Date: Wed, 12 Apr 2006 13:11:08 -0700
> RK> From: Rahul Kulkarni
>=20
> RK> Any known issues or experiences where a byte or 2 bytes get
> RK> corrupted randomly at different addresses but always on a 4k page
> RK> boundary.
>=20
> I'm no PPC expert, but this almost sounds like something to do with=20
> misaligned data, alignment interrupts, and the MMU.
>=20
>=20
> RK> This happens in a different user process and the address where the
>=20
> Which further makes me wonder about the MMU and alignment trap.

Another possibility is too fast access to memory. To be precise:
when a cpu cache burst is back to back with DMA access from/to the
disk controler, the timing between the two is too "tight".=20

This might be because the memory (if of the intelligent variant) doesn't
live up to its specifications, or because you're using too small access
times if set manually somewhere.=20

I'd try to relax timing, if easily settable, and check whether the=20
problem goes away.

Regards,
	-is

--7AUc2qLy4jB3hD7Z
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (NetBSD)

iD8DBQFEPrIQN4tiz3B8hB0RAiWxAJ9GY52iFxZBRZAGsKmBdxY3mhMQ0wCdH9Rk
IhgVccGtwZPWiKXk2OSOUQM=
=BqUj
-----END PGP SIGNATURE-----

--7AUc2qLy4jB3hD7Z--