Hacker News new | ask | show | jobs
by notacoward 4226 days ago
It might, but I wouldn't say probably. It might also mean more resources devoted to ensuring that PolarSSL runs well on the ARM architectures/chips that are (or will be) out there anyway. That's probably more interesting to most people, except for those making dedicated network hardware.
1 comments
xgbi 4226 days ago
I don't see the problem, PolarSSL runs on ARM very well!
link
notacoward 4226 days ago
Who said there was a problem? I'm sure PolarSSL runs very well on ARM already. However, it's also extremely likely that it could run even better if the PolarSSL developers were more fully plugged into what the people who work on the ARM crypto hardware know. It's amazing what one can do if one knows exact details of what's going on within each functional unit, between them on the internal coherency bus, etc. ARM probably saw an opportunity, not a problem.
link
bravo22 4225 days ago
I don't think it is that. ARM doesn't make the crypto. The SoC makers put their own crypto cores so it wouldn't help ARM that way.

I think just like they bought Keil (a dev tools maker) this is a strategy play to make it easier for end devs to add SSL or other crypto to their products. One shop solution.

link
notacoward 4225 days ago
Actually ARM does make crypto. It's part of ARMv8, licensable as an option for at least the Cortex-A53.

http://infocenter.arm.com/help/topic/com.arm.doc.ddi0500e/DD... (section 2.1.4)

There are undoubtedly other bits as well, as part of their "trusted computing" blahblah. Even if that weren't the case, knowing more about the internals of current and upcoming ARM IP could help optimize even an all-software implementation of PolarSSL. You could be right that it's mostly about "one stop shopping" but that doesn't mean there won't be other benefits.

link
bravo22 4225 days ago
You are correct.

However, Trusted Computing crypto is different than the crypto accelerators you find as peripherals in an SoC.

I should have been more clear that I meant crypto peripherals. There are crypto instruction extensions but don't require a separate library implementation -- just asm code optimization in something like openSSL.

PolarSSL is also targeted (mainly) towards much lower power than A53 or Cortex-A. It is targeted towards Cortex-M where you are dealing with KB of data and often don't have an MMU. You can't just port OpenSSL to those platforms and run it at will, hence their optimized libraries.

link