Re: [PATCH 5/7] random: replace non-blocking pool with a Chacha20-based CRNG

[Theodore Ts'o]

On Mon, Jun 20, 2016 at 05:49:17PM +0200, Stephan Mueller wrote:
> 
> Is speed everything we should care about? What about:
> 
> - offloading of crypto operation from the CPU

In practice CPU offland is not helpful, and in fact, in most cases is
harmful, when one is only encrypting a tiny amount of data.  That's
because the cost of setup and teardown, not to mention key scheduling,
dominate.  This is less of the case in the case of the SIMD / AVX
optimizations --- but that's because these are CPU instructions, and
there really isn't any CPU offloading going on.

> - potentially additional security features a hardware cipher may provide like 
> cache coloring attack resistance?

Um.... have you even taken a *look* at how ChaCha20 is implemented?
*What* cache coloring attack is possible at all, period?

Hint: where are the lookup tables?  Where are the data-dependent
memory accesses in the ChaCha20 core?

						- Ted