[Cryptech Tech] EIM Baseline Project

[Rob Austein]

At Sat, 31 Jan 2015 09:19:04 +0100, Joachim Strömbergson wrote:
> 
> Paul and Rob could you look at whats in the sw/ and the doc/ and comment
> on both the code and what is suggested in the documentation.

C code looks very clean.  There are a few constructs Pavel used that I
would not ordinarily use in portable network code, but this code is
(very) specific to a particular platform and compiler, so what Pavel
did is fine.

Unfortunate that we can't do bulk memcpy(), particularly since modern
compilers tend to be clever about inlining memcpy() calls, but it is
what it is.  Reading between the lines, I assume that we can write
(eg) a complete block as a sequence of eim_write_32() without having
to poll for an acknowledgement after every 32-bit word, we just can't
do that in a single memcmp().

I'm a little worried about overhead from _eim_remap_mem() if we end up
spreading things out on a large number of pages (thus requiring
frequent remapping), but let's get the basic stuff right before
worrying about that kind of performance issue.  If we can get to the
point where overhead from _eim_remap_mem() is our worst bottleneck,
I'll be very happy. :)

I will wait for somebody to hook some hash cores into the revised EIM
framework before trying to revise the HAL code.