Re: [RFC PATCH 2/2] dt-bindings: add device tree binding for silex multipk

[Greg KH <gregkh@linuxfoundation.org>]

On Thu, Mar 13, 2025 at 02:55:07PM +0530, Gupta, Nipun wrote:
> On 12-03-2025 16:21, Krzysztof Kozlowski wrote:
> > On 12/03/2025 10:54, Nipun Gupta wrote:
> > > +maintainers:
> > > +  - Nipun Gupta <nipun.gupta@amd.com>
> > > +  - Praveen Jain <praveen.jain@amd.com>
> > > +
> > > +description: |
> > > +  Silex Multipk device handles the Asymmetric crypto operations. The
> > > +  driver provides interface to user-space to directly interact with the
> > > +  Silex MultiPK device.
> > 
> > Why this isn't in crypto?
> 
> It is mentioned in patch RFC 1/2 - because Crypto AF_ALG does not support
> offloading asymmetric operations from user-space (which was attempted to be
> added earlier in Linux at: https://lore.kernel.org/all/146672253157.23101.15291203749122389409.stgit@tstruk-mobl1.ra.intel.com/)

That was 8 years ago.  We now have the accel framework, why not use
that?  Or work on getting this framework properly submitted as it looks
like you now have 2 different users for it.

> > Device name part is weirdly generic. How is this device exactly called?
> > Where is it used? Where is datasheet?
> 
> The device is on AMD versal series and is named "Multi PKI" device. I
> will update to use compatible as xlnx,multipk (AMD versal series link: https://www.amd.com/en/products/adaptive-socs-and-fpgas/versal/premium-series.html).
> Seems also renaming files to xlnx_multipk.c etc would be better. Will
> update.
> 
> The device is used for PKI offload for OpenSSL based applications.
> Unfortunately data sheet is not available in public domain.

Expecting an undocumented crypto accelerator to be used by anyone feels
like a very odd ask.  Would you want to trust that?

thanks,

greg k-h