Am 2022-11-28 07:59, schrieb Rafał Miłecki:
> From: Rafał Miłecki <[email protected]>
>
> Pass whole NVMEM cell struct and length pointer as arguments to
> callback
> functions.
>
> This allows:
>
> 1. Cells content to be modified based on more info
> Some cells (identified by their names) contain specific data that
> needs further processing. This can be e.g. MAC address stored in an
> ASCII format. NVMEM consumers expect MAC to be read in a binary
> form.
> More complex cells may be additionally described in DT. This change
> allows also accessing relevant DT nodes and reading extra info.
>
> 2. Adjusting data length
> If cell processing results in reformatting it, it's required to
> adjust length. This again applies e.g. to the MAC format change from
> ASCII to the byte-based.
>
> Later on we may consider more cleanups & features like:
> 1. Dropping "const char *id" and just using NVMEM cell name
> 2. Adding extra argument for cells providing multiple values
>
> Signed-off-by: Rafał Miłecki <[email protected]>
> ---
> This solution conflicts with 1 part of Michael's work:
> [PATCH v2 00/20] nvmem: core: introduce NVMEM layouts
> https://lore.kernel.org/linux-arm-kernel/[email protected]/
>
> Instead of:
> 1. Adding NVMEM cell-level post_process callback
> 2. Adding callback (.fixup_cell_info()) for setting callbacks
> 3. Dropping NVMEM device-level post_process callback
> I decided to refactor existing callback.
>
> Michael's work on adding #nvmem-cell-cells should be possible to easily
> rebase on top of those changes.
As yours should be easily added on top of my series. I've showed that
providing a global post process hook is bad because that way you need
to have *all* cells of your device read-only.
-michael
Hi Rafal,
[email protected] wrote on Mon, 28 Nov 2022 08:35:24 +0100:
> Am 2022-11-28 07:59, schrieb Rafał Miłecki:
> > From: Rafał Miłecki <[email protected]>
> >
> > Pass whole NVMEM cell struct and length pointer as arguments to > callback
> > functions.
> >
> > This allows:
> >
> > 1. Cells content to be modified based on more info
> > Some cells (identified by their names) contain specific data that
> > needs further processing. This can be e.g. MAC address stored in an
> > ASCII format. NVMEM consumers expect MAC to be read in a binary > form.
> > More complex cells may be additionally described in DT. This change
> > allows also accessing relevant DT nodes and reading extra info.
> >
> > 2. Adjusting data length
> > If cell processing results in reformatting it, it's required to
> > adjust length. This again applies e.g. to the MAC format change from
> > ASCII to the byte-based.
Michael's series brings read_post_process, isn't what you need here?
> >
> > Later on we may consider more cleanups & features like:
> > 1. Dropping "const char *id" and just using NVMEM cell name
> > 2. Adding extra argument for cells providing multiple values
> >
> > Signed-off-by: Rafał Miłecki <[email protected]>
> > ---
> > This solution conflicts with 1 part of Michael's work:
> > [PATCH v2 00/20] nvmem: core: introduce NVMEM layouts
> > https://lore.kernel.org/linux-arm-kernel/[email protected]/
> >
> > Instead of:
> > 1. Adding NVMEM cell-level post_process callback
> > 2. Adding callback (.fixup_cell_info()) for setting callbacks
> > 3. Dropping NVMEM device-level post_process callback
> > I decided to refactor existing callback.
> >
> > Michael's work on adding #nvmem-cell-cells should be possible to easily
> > rebase on top of those changes.
Yeah, I guess since Michael's series has been out for 2 years and we
finally agreed on the bindings plus some implementation points, I would
expect it to be merged very soon (I don't know if Srinivas still plans
to take it for this release or for the next?) unless someone speaks up
against it.
> As yours should be easily added on top of my series. I've showed that
> providing a global post process hook is bad because that way you need
> to have *all* cells of your device read-only.
>
> -michael
Thanks,
Miquèl