[Cryptech Tech] EIM Baseline Project

[Rob Austein]

At Sun, 01 Feb 2015 01:27:44 +0300, Шатов Павел wrote:
>
> The other important thing is that you need to mmap() EIM region into
> our address space. I can't tell what overhead is caused by this
> call, because I'm not even close to being a linux kernel expert.

I don't really know either (most of my experience with memory mapping
dates back to PDP-10s), but any time I see writing four bytes mutate
into multiple operating system calls I get worried. :)

> The idea to save last mapped page address and not call mmap() again
> if access is requested to already mapped page was actually in the
> original code that I downloaded from the repository. I do know that
> one can't mmap() smaller than one page at a time, but I belive that
> you can mmap() a larger region if you need. You will not need to
> remap it that often then. I actually haven't tried it, because my
> demo adder only has 4 registers.

Yeah, the total memory footprint of all of the cores is probably not
going to be all that large, so most likely we can just mmap() enough
pages to cover all of it during process initialization and have done.

Thanks!