[openssl-dev] [RFC 0/4] Kernel TLS socket API

[Kurt Roeckx]

On Thu, Jun 08, 2017 at 06:26:28PM +0000, Ilya Lesokhin wrote:
> Hi Kurt,
> I think this it's better to have this discussion in the kernel mailing list.
> But basically, we were debating this issue ourselves. 
> Previously we had another field in the attach API which could be {SW only, HW only and auto}.
> However, we thought that most application wouldn't know what to do with it.
> So, we thought we could simplify the API, and make the HW/SW decision a global administrative configuration. 
> 
> The downside is that the ability to do HW offload depends on the routing, so the software
> Can't be configured to use KTLS only if HW offload is available.
> 
> Do you think we should restore the old {SW only, HW only and auto} API?

I currently have no idea what the best way is. But I would like to
avoid a regression in performance by default.

So I guess part of the question is if you think this kind of
hardware would become common, or that it's only going to be used
in specialised settings?


Kurt