Re: [PATCH v8 2/8] mfd: simple-mfd-i2c: specify max_register

[Lee Jones <lee@kernel.org>]

On Wed, 23 Jul 2025, Alex Elder wrote:

> On 7/23/25 4:51 AM, Lee Jones wrote:
> > On Thu, 10 Jul 2025, Alex Elder wrote:
> > 
> > > All devices supported by simple MFD use the same 8-bit register 8-bit
> > > value regmap configuration.  There is an option available for a device
> > > to specify a custom configuration, but no existing device uses it.
> > > 
> > > Rather than specify a "full" regmap configuration to change only
> > > the max_register value, Lee Jones suggested allowing max_register
> > > to be specified in the simple_mfd_data structure.  If regmap_config
> > > and max_register are both supplied, the max_register field is ignored.
> > > 
> > > Signed-off-by: Alex Elder <elder@riscstar.com>
> > > Suggested-by: Lee Jones <lee@kernel.org>
> > > ---
> > > v8: - Use regmap_config_8r_8v, modifying it if max_register supplied
> > > 
> > >   drivers/mfd/simple-mfd-i2c.c | 8 ++++++--
> > >   drivers/mfd/simple-mfd-i2c.h | 3 ++-
> > >   2 files changed, 8 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/mfd/simple-mfd-i2c.c b/drivers/mfd/simple-mfd-i2c.c
> > > index 22159913bea03..5138aa72140b5 100644
> > > --- a/drivers/mfd/simple-mfd-i2c.c
> > > +++ b/drivers/mfd/simple-mfd-i2c.c
> > > @@ -24,15 +24,16 @@
> > >   #include "simple-mfd-i2c.h"
> > > -static const struct regmap_config regmap_config_8r_8v = {
> > > +static struct regmap_config regmap_config_8r_8v = {
> > >   	.reg_bits = 8,
> > >   	.val_bits = 8,
> > > +	/* .max_register can be specified in simple_mfd_data */
> > 
> > Drop this comment please.
> > 
> > >   };
> > >   static int simple_mfd_i2c_probe(struct i2c_client *i2c)
> > >   {
> > >   	const struct simple_mfd_data *simple_mfd_data;
> > > -	const struct regmap_config *regmap_config;
> > > +	struct regmap_config *regmap_config;
> > >   	struct regmap *regmap;
> > >   	int ret;
> > > @@ -43,8 +44,11 @@ static int simple_mfd_i2c_probe(struct i2c_client *i2c)
> > >   		regmap_config = &regmap_config_8r_8v;
> > >   	else
> > >   		regmap_config = simple_mfd_data->regmap_config;
> > > +	if (simple_mfd_data && !simple_mfd_data->regmap_config)
> > > +		regmap_config->max_register = simple_mfd_data->max_register;
> > 
> > If max_register is set in simple_mfd_data, it should take precedence.
> 
> I don't really agree with that.  If simple_mfd_data->regmap_config
> is provided, why not use the max_register field already available
> there?

Why would a user add a max_register override to simple_mfd_data if they
didn't want to use it?

> This is why I said above that I think this feature doesn't add
> much value.  It provides a second way to specify something, but
> in the end it complicates the code more than it's worth.
> 
> The only time this new simple_mfd_data->max_register field seems
> to make sense is if it were the only thing provided (without
> simple_mfd_data->regmap_config being supplied).  In that case,
> I see the benefit--a null simple_mfd_data->regmap_config means
> use regmap_config_8r_8v, and overlay it with the max_register
> value.  The new max_register field avoids defining another huge
> but mostly empty regmap_config structure.

This is your use-case, right?

> Anyway, back to your original point:  I said in v7 "If both
> are specified, the max_register value is ignored" and I think
> that's the simplest.  Specify one or the other--if you want
> to define things in regmap_config, then that's where you add
> your max_register.  If you like regmap_config_8r_8v but want
> to define a max_register value, just provide max_register.
> 
> If you insist, I'll do what you say but before I sent another
> version I wanted to explain my reasoning.

I hear you and I get what you're saying.

I see no use-case where a user would provide both regmap_config AND
max_register either.  However, I see max_register in simple_mfd_data as
an override, so I would like it to take precedence please.

> > if (simple_mfd_data && simple_mfd_data->max_register)
> > 	regmap_config->max_register = simple_mfd_data->max_register;
> > 
> > >   	regmap = devm_regmap_init_i2c(i2c, regmap_config);
> > > +	regmap_config->max_register = 0;
> > 
> > Does max_register definitely have persistence over subsequent calls?
> 
> It is a global variable.  Isn't that how they work?  When
> it was read-only there was no concern about that, nor about
> any possible concurrent access (though I don't think multiple
> probes can be using this code at once).
> 
> We could allocate a new one each time instead.
> 
> I think what I offered in v5 was acceptable.  If you're
> willing to accept that I will be happy to keep discussing
> (and implementing) the max_register feature.

Yes, I'm inclined to agree.

Make the call and I will respect your decision.

-- 
Lee Jones [李琼斯]