Re: [RESEND RFC 3/3] nvmem: Add 'nvmem-composite' driver

[Srinivas Kandagatla]

On 02/03/16 18:33, Andrey Smirnov wrote:
> On Wed, Mar 2, 2016 at 5:59 AM, Srinivas Kandagatla
> <srinivas.kandagatla@xxxxxxxxxx> wrote:
> > On 01/03/16 16:59, Andrey Smirnov wrote:
> > > Add 'nvmem-composite' driver which allows to combine multiple chunks of
> > > various NVMEM cells into a single continuous NVMEM device.
> >
> >
> > My plan on this feature was add support inside the nvmem_cell_get itself,
> > this makes the nvmem bindings more inline with bindings like pinctrl.
> > Also I still want to keep nvmem simple as it can.
>
> That's perfectly fine with me, I can change my patch to do that. My
> thinking on implementing it as a standalone module was that this seems
> to be a very niche functionality which not a lot of people would use,
> so the code would be a part of your kernel only if you directly use
> this feature.
>
> > DT would look something like this.
> >
> > nvmem-provider-a {
> >          cell_a {
> >                  reg = <0 2>;
> >          };
> > };
> >
> > nvmem-provider-b {
> >          cell_b: cell_c {
> >                  reg = <0 1>;
> >          };
> > };
> >
> > nvmem-provider-c {
> >          cell_c: cell_c {
> >                  reg = <3 2>;
> >          }
> > };
> >
> > a-node {
> >          nvmem-cells = <&cell_a &cell_b &cell_c>
> >          nvmem-cell-names = "some-data";
> > };
>
> It's not very clear to me, possibly to my DT ignorance, how this would
> handle the case of multiple variables. Say we have
>
> a-node {
>           nvmem-cells = <&cell_a &cell_b>, ????;
>           nvmem-cell-names = "some-data", "more-data";
> };
Should have replied you long back :-)
> and I want "more-data" to reference only one phandle, how would this be handled?
yes this would fail.

The device tree compiler would concatenate all the cells and we have no 
means to know where did "more-data" starts.

sounds like composite driver is the way forward.


--srini
> Another minor nitpick about his is that if one's goal is to just
> byteswap already defined nvmem_cell:
>    - N, where N is the length of the variable, new nvmem cells would
> have to be defined
>    - All of them would have to use absolute address within the nvmem
> provided, instead of referencing a byte relative to the position of
> pre-defined cell
>
> Thanks,
> Andrey