Re: [PATCH 2/3] i2c: designware: Introduce per-variant register offset and bit-layout tables

[Aniket Limaye]

On 25/09/26 13:41, Andy Shevchenko wrote:
> On Mon, Sep 21, 2026 at 01:11:48PM +0200, Mika Westerberg wrote:
> > On Sat, Sep 19, 2026 at 02:36:07PM +0530, Aniket Limaye wrote:
> > > Every DW_IC_* register offset and CON-register bit position is currently
> > > baked in as a compile-time constant, which only works while there is a
> > > single register layout. Introduce a logical register-ID enum (enum
> > > dw_i2c_reg_idx) plus a per-variant offset table (dev->regs[]) and a
> > > per-variant CON-register bit-layout descriptor (dev->con_bits), selected
> > > at probe time via the new i2c_dw_select_variant().
> > >
> > > Replace every direct DW_IC_* offset/bit-position reference with a lookup
> > > through dev->regs[]/dev->con_bits. Also fold the read-to-clear
> > > interrupt-acknowledgment pattern into a new i2c_dw_ack_intr() helper,
> > > driven by a per-variant dev->intr_clr[] table.
> > >
> > > Only one variant exists at this point (DW_apb_i2c), so this is a
> > > mechanical, behavior-preserving change: the values in
> > > dw_i2c_reg_offsets[] and dw_i2c_con_bits match the DW_IC_* macros
> > > exactly. It lays the groundwork for adding a second register layout
> > > (DWC_i2c) without duplicating the whole driver.
> ...
>
> > > -		regmap_write(dev->map, DW_IC_TX_TL, dev->tx_fifo_depth / 2);
> > > -		regmap_write(dev->map, DW_IC_RX_TL, 0);
> > > -		regmap_write(dev->map, DW_IC_CON, dev->master_cfg);
> > > +		regmap_write(dev->map, dev->regs[DW_REG_IDX_TX_TL], dev->tx_fifo_depth / 2);
> > > +		regmap_write(dev->map, dev->regs[DW_REG_IDX_RX_TL], 0);
> > > +		regmap_write(dev->map, dev->regs[DW_REG_IDX_CON], dev->master_cfg);
> > Instead of all this. Can't you do this inside the regmap so that here and
> > elsewhere in the driver we continue to do:
> >
> > 	regmap_write(dev->map, DW_IC_RX_TL, 0);
> >
> > but internally, depending on the hardware it then maps this into the
> > corresponding register offset.
> Exactly what I was going to say when I hit "reply".
> These series is definitely NAKed (in terms of the approach taken).
Hello Mika, Andy,

I have posted a v3 for the series [0] with a different approach, 
preserving existing call sites.
Note that v3 version makes changes to all regmaps (native and inherited) 
to now handle enums instead of actual offsets.

[0]: 
https://lore.kernel.org/all/20260925-tda54-upstream-i2c-v3-0-544d74e992ff@xxxxxx/

Thanks for your reviews!
Aniket