Re: [PATCH v3 16/16] ARM: marvell/dt: add crypto node to kirkwood dtsi

[Boris Brezillon <boris.brezillon@free-electrons.com>]

On Tue, 26 May 2015 09:06:29 +0000
Jason Cooper <jason@lakedaemon.net> wrote:

> On Mon, May 25, 2015 at 08:43:02PM +0200, Boris Brezillon wrote:
> > Jason, Gregory,
> > 
> > On Mon, 25 May 2015 16:46:51 +0000
> > Jason Cooper <jason@lakedaemon.net> wrote:
> > 
> > > On Mon, May 25, 2015 at 05:39:13PM +0200, Gregory CLEMENT wrote:
> > > > Hi Boris, Arnaud,
> > > > 
> > > > On 22/05/2015 15:34, Boris Brezillon wrote:
> > > > > From: Arnaud Ebalard <arno@natisbad.org>
> > > > > 
> > > > > Add crypto related nodes to kirkwood.dtsi.
> > > > 
> > > > Here you use a new compatible string but with an old binding
> > > > to let the user chose between the old and the new driver. Am I right?
> > 
> > That was not the intention, but you're right, that's exactly what's
> > happening here.
> > 
> > > 
> > > I thought we had settled on the user choosing by module load/ which driver is
> > > compiled in?  The DT should be describing the hardware, not which driver the
> > > user chooses to use.
> > 
> > Right, but I didn't want to add new compatible strings to the old
> > driver in the first place, neither I wanted to support the new way of
> > defining/referencing the crypto SRAMs.
> 
> 
> > ITOH, if we want to benefit from the TDMA optimization on Kirkwood SoCs,
> > we have to add a new compatible (unlike Orion SoCs, Kirkwood ones embed
> > a TDMA engine).
> 
> Ah, there's the HW difference I must have missed in my previous thousand-foot
> overview scans :-/
> 
> So "marvell,orion-crypto" matches IP blocks without the TDMA engine,
> "marvell,kirkwood-crypto" matches IP blocks *with* the TDMA engine.
> 
> > This leaves the following solutions:
> >  - avoid changing the compatible in existing orion and kirkwood dtsi
> >    files
> 
> no, in light of the above HW difference, it makes sense to change these.
> 
> >  - adding kirkwood compatible string support to the existing CESA
> >    driver (and I think supporting the new approach to retrieve SRAM
> >    memory region would make sense too)
> 
> Or, old driver matches "marvell,orion-crypto", and the new driver matches
> either compatible string.  If dt has "marvell,kirkwood-crypto" then new driver
> enables TDMA with the provided properties.
> 
> We then update the dtsi for all but orion to "marvell,kirkwood-crypto".
> 
> This may be what you are already doing.  If so, please ignore my rambling. ;-)

Yes, that's what I'm doing :-). But this means we're forcing kirkwood
users to switch to the new driver, which is not really what you
suggested in the first place.


-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com