Re: [PATCH] (0/4) Entropy accounting fixes

[D. Hugh Redelmeier]

| From: Alan Cox <[e-mail]>
| Date: 19 Aug 2002 13:29:10 +0100

| On Mon, 2002-08-19 at 11:47, Marco Colombo wrote:
| > BTW, I know you wrote the amd768-rng driver, I wonder if you have any
| > indication of how good these rng are. What is the typical output bits/
| > random bits ratio in normal applications?
| 
| It seems random. People have subjected both the intel and AMD one to
| statistical test sets. I'm not a cryptographer or a statistician so I
| can't answer usefully

[Note: I am not a cryptographer.]

The Intel (and, I assume, the AMD) hardware random generator cannot be
audited.  There is no way to tell if it is a RNG.  These days, you
don't need to be paranoid to lack trust in such devices.

Governments could easily pressure Intel or AMD in a number of ways.
Perhaps other forces could corrupt the RNG implementation.

I've heard that Intel's RNG "whitens" its output in a way that hides
failure and makes analysis difficult.  This cannot be turned off.
This doesn't add to my confidence.

Adding the putative RNG's output to the pool is a Good Thing.
Depending on it may not be.

Hugh Redelmeier
[e-mail]  voice: [phone]