[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1801152.Iu9ZVUxv1v@tauon.chronox.de>
Date: Wed, 16 Aug 2017 14:51:42 +0200
From: Stephan Mueller <smueller@...onox.de>
To: Theodore Ts'o <tytso@....edu>
Cc: LKML <linux-kernel@...r.kernel.org>, linux-crypto@...r.kernel.org,
david.fontaine@...gemini.com, olivier.vivolo@...nge.com
Subject: Re: random.c: LFSR polynomials are not irreducible/primitive
Am Dienstag, 15. August 2017, 17:12:24 CEST schrieb Theodore Ts'o:
Hi Theodore,
>
> Stephan, if you have any comments on the proposal made by David
> Fontaine and Olivier Vivolo, I'd appreciate hearing them!
I think I have some news: The magma code I used for GF(2^32) testing was not
correct.
The corrected magma code is attached (thanks to Dr. Peter Birkner, BSI, who
helped me here).
That magma code shows:
- the current polynomials for Q(X) = α**3 (P(X) − 1) + 1 are irreducible but
not primitive over GF(2^32)
- the polynomials suggested in https://eprint.iacr.org/2017/726.pdf Q(X) =
α**4 (P(X) − 1) + 1 are both, irreducible and primitive over GF(2^32)
The use of GF(2^32) is important, because we apply the LFSR to a 32 bit word.
Hence, we have 2^32 permutations the LFSR should evenly cover.
Bottom line, I would recommend that random.c is patched to take the
polynomials suggested in https://eprint.iacr.org/2017/726.pdf.
If it is of any help, the attached magma code could be preserved somewhere
useful (in random.c?)
Ciao
Stephan
Download attachment "LFSR_polynomials eprint 251.mag" of type "application/x-wine-extension-mag" (2655 bytes)
Powered by blists - more mailing lists