Re: [PATCH 6/7] iio: adc: ad485x: add ad485x driver

[Jonathan Cameron <jic23@kernel.org>]

On Fri, 4 Oct 2024 14:07:37 +0000
"Miclaus, Antoniu" <Antoniu.Miclaus@analog.com> wrote:

> > On 10/3/24 5:14 AM, Miclaus, Antoniu wrote:  
> > >> On 10/1/24 8:51 AM, Miclaus, Antoniu wrote:  
> > >>>>> Regarding the bulk writes/reads, the msb/mid/lsb registers need to be
> > >>>>> read/write in a specific order and the addresses are not incremental,  
> > >>>>
> > >>>> We have _noinc() variants of regmap accessors.  
> > >>> [Miclaus, Antoniu]
> > >>> I think _noinc() functions read from the same register address so it doesn't
> > >>> apply.
> > >>> I am reading values from multiple register addresses that are not reg_addr,
> > >>> reg_addr+1, reg_addr+2.  
> > >>
> > >> I'm confused by the statement that the registers are not incremental.
> > >>
> > >> For example, this patch defines...
> > >>
> > >> +#define AD485X_REG_CHX_OFFSET_LSB(ch)
> > >> 	AD485X_REG_CHX_OFFSET(ch)
> > >> +#define AD485X_REG_CHX_OFFSET_MID(ch)
> > >> 	(AD485X_REG_CHX_OFFSET_LSB(ch) + 0x01)
> > >> +#define AD485X_REG_CHX_OFFSET_MSB(ch)
> > >> 	(AD485X_REG_CHX_OFFSET_MID(ch) + 0x01)
> > >>
> > >> This looks exactly like reg_addr, reg_addr+1, reg_addr+2 to me.  
> > > Yes you are right. Although I tested with hardware and it seems that the  
> > registers  
> > > are not properly written when using bulk operations. My guess is that  
> > holding CS low during  
> > > the entire transaction might be a possible issue. Any suggestions are  
> > appreciated.
> > 
> > Is ADDR_DIR in SPI_CONFIG_A set to the correct value to match how
> > the regmap is configured for bulk writes?
> > 
> > I had to set this bit for AD4695 which has a similar register map
> > (although on that one I used two regmaps, an 8-bit one and a 16-bit
> > one, instead of doing bulk operations on the 8-bit one).
> >   
> Thanks for the input! I tried your suggested approach: set the ADDR_DIR
> to 1 during probe. Unfortunately, this did not fix the issue. I am still not able
> to perform bulk writes properly to the device.
> 
> For now I will keep the only working version in v2, since there will be
> most probably  other iterations of the this patch series 😊.
I'd definitely like to know what is going on here if you can dig into it.
But if we really get stuck then the code at least needs a comment saying
it is necessary and we aren't sure why.  If we know why and can't change
it then the comment should give that reasoning.

Jonathan

> 
> > >  
> > >>>  
> > >>>>> so I am not sure how the bulk functions fit. On this matter, we will need
> > >>>>> unsigned int (not u8) to store the values read via regmap_read, and in  
> > this  
> > >>>>> case we will need extra casts and assignments to use get_unaligned.  
> > >>>>
> > >>>> --
> > >>>> With Best Regards,
> > >>>> Andy Shevchenko
> > >>>>  
> > >>>  
> > >  
>