Re: [PATCH 2/6] nvmem: Add backwards compatibility support for older EEPROM drivers.

[Andrew Lunn <andrew@lunn.ch>]

On Fri, Dec 11, 2015 at 02:03:25PM +0100, Wolfram Sang wrote:
> On Tue, Dec 08, 2015 at 03:05:07PM +0100, Andrew Lunn wrote:
> > Older drivers made an 'eeprom' file available in the /sys device
> > directory. Have the NVMEM core provide this to retain backwards
> > compatibility.
> > 
> > Signed-off-by: Andrew Lunn <andrew@lunn.ch>
> > ---
> >  drivers/nvmem/Kconfig          |  7 ++++
> >  drivers/nvmem/core.c           | 75 +++++++++++++++++++++++++++++++++++++++---
> >  include/linux/nvmem-provider.h | 10 ++++++
> >  3 files changed, 88 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/nvmem/Kconfig b/drivers/nvmem/Kconfig
> > index bc4ea585b42e..b4e79ba7d502 100644
> > --- a/drivers/nvmem/Kconfig
> > +++ b/drivers/nvmem/Kconfig
> > @@ -13,6 +13,13 @@ menuconfig NVMEM
> >  	  If unsure, say no.
> >  
> >  if NVMEM
> > +config NVMEM_COMPAT
> > +       bool "Enable /sys compatibility with old eeprom drivers"
> > +       help
> > +	 Older EEPROM drivers, such as AT24, AT25, provide access to
> > +	 the eeprom via a file called "eeprom" in /sys under the
> > +	 device node. Enabling this option makes the NVMEM core
> > +	 provide this file to retain backwards compatibility
> 
> I don't like this being a Kconfig option TBH. In most cases, when I read
> "retain backwards compatibility" in Kconfig help texts, I keep the
> option activated because I don't know the details when exactly it is
> safe to disable it. Plus, we have too many Kconfig symbols already.
> 
> I suggest to add this flag to nvmem_config and let the old eeprom
> drivers always set this flag because they need to provide this file for
> some more time, if not forever. New drivers using the nvmem_layer will
> probably not want to set this.

I'm happy to do this, if the NVMEM core maintainers agree.

> BTW how does this NVMEM framework relate to the memory_accessor
> framework. Can it be used to replace it? I think we should keep the
> number of eeprom interfaces at a sane level, preferably 1 ;)

The memory_accessor framework only seems to work with old style
platform data devices, where you can register the callback function to
be used during probe. Because it cannot be used in a DT system, there
are very few users of it, only boards in arch/arm/mach-davinci. They
use it to get their MAC address out of the EEPROM. There are no users
of the AT25 equivalent, which is why i removed it. So this API is
dying on its own.

The NVMEM framework has a similar API for accessing the whole EEPROM,
and a much finer grained API for accessing cells within the EEPROM,
for example saying bytes 16-22 is the MAC address cell.

However, the NVMEM APIs are DT only, so are not a replacement for
memory_accessor. We need to keep memory_accessor until Davinci gets
converted to DT.

> Also adding Pantelis to CC who also submitted at24 NVMEM support a while
> ago.

Thanks for pointing this out, i was not aware of that patch.

       Thanks
	    Andrew