[PATCH v3 1/4] crypto: Add Allwinner Security System crypto accelerator

[marex@denx.de (Marek Vasut)]

On Sunday, June 22, 2014 at 02:33:35 PM, Russell King - ARM Linux wrote:
> On Sun, Jun 22, 2014 at 02:23:15PM +0200, Marek Vasut wrote:
> > On Sunday, June 22, 2014 at 01:58:08 PM, Corentin LABBE wrote:
> > [...]
> > 
> > > >> + * This program is free software; you can redistribute it and/or
> > > >> modify + * it under the terms of the GNU General Public License as
> > > >> published by + * the Free Software Foundation version 2 of the
> > > >> License
> > > > 
> > > > The license text seems incomplete.
> > > > [...]
> > > 
> > > I will replace it with a simplier line "Licensed under the GPL-2."
> > 
> > I'd suggest you to use the SPDX license identifiers then, but that's not
> > something the kernel crowd agreed upon yet IIRC.Therefore , just make the
> > text complete please.
> 
> Marek,
> 
> The full text of the first paragraph (in COPYING) is:
> 
>     This program is free software; you can redistribute it and/or modify
>     it under the terms of the GNU General Public License as published by
>     the Free Software Foundation; either version 2 of the License, or
>     (at your option) any later version.
> 
> which is deemed to be entirely sufficient; quoting the full text is
> discouraged, especially when it includes the FSF address.  It is also
> acceptable to restrict it to version 2 only, in which case something
> like this can be used:
> 
>    This program is free software; you can redistribute it and/or modify
>    it under the terms of the GNU General Public License version 2 as
>    published by the Free Software Foundation.
> 
> Of course, it's up to the author to make up their own mind at the end
> of the day.

Ah, I fully agree. Thank you for clarifying this.

> > Oh, ok, good question -- dear list, shall one use sg_page()+kmap or
> > sg_virt()?
> 
> sg_page() + kmap() is preferred, because sg_virt() fails with highmem.
> Using sg_virt() means you restrict the driver to non-highmem memory,
> and if the kernel wants to place the data into a highmem page, it will
> have to use bounce buffers (so it's inefficient).

Ah, thanks for explaining this!

> What's even better is to use the scatterlist iterator, which will handle
> this for you.  See the sg_miter_*() functions.

The scatterwalk ( include/crypto/scatterwalk.h ) should work pretty much the 
same then, right ? But with the added benefit that it's crypto-friendly such 
that it handles both the input and output scatterlists of the crypto 
transformation etc.

Please correct me if I'm mistaken.

Best regards,
Marek Vasut