Re: [PATCH 1/3] pinctrl: mcp23s08: Fixup mcp23x17 regmap_config

[Thomas Preston]

Hey Andy, Linus,
Thanks for looking at this.

On 28/08/2020 11:09, Andy Shevchenko wrote:
> On Fri, Aug 14, 2020 at 1:35 PM Thomas Preston
> <thomas.preston@xxxxxxxxxxxxxxx> wrote:
> > - Fix a typo where mcp23x17 configs are referred to as mcp23x16.
>
> I'm not sure it's correct. MPC23016 is an existing I²C IO expander.

The MCP23016 device is not mentioned anywhere else in this driver. The 
only place this string is used is in `struct regmap_config 
mcp23x17_regmap` (another device). It seems to me that this is a typo 
but I might be wrong.

~/w/linux$ git grep -h compatible drivers/pinctrl/pinctrl-mcp23s08*
                .compatible = "microchip,mcp23008",
                .compatible = "microchip,mcp23017",
                .compatible = "microchip,mcp23018",
                .compatible = "mcp,mcp23008",
                .compatible = "mcp,mcp23017",
                .compatible = "microchip,mcp23s08",
                .compatible = "microchip,mcp23s17",
                .compatible = "microchip,mcp23s18",
                .compatible = "mcp,mcp23s08",
                .compatible = "mcp,mcp23s17",

Also I don't have an MC23016, so I can't test configuration for it.

> > - Fix precious range to include INTCAP{A,B}, which clear on read.
> > - Fix precious range to include GPIOB, which clears on read.
> > - Fix volatile range to include GPIOB, to fix debugfs registers
> >    reporting different values than `gpioget gpiochip2 {0..15}`.
>
> I'm wondering if you read all the datasheets before doing these changes.
> MPC2308
> MPC23016
> MPC23017
> ...

I did not! I was only changing configuration for MCP23017 devices.
What have I missed?

For reference, I think you are referring to [0], [1], [2]. I'm familiar 
with the last one.

> > -static const struct regmap_range mcp23x16_volatile_range = {
> > +static const struct regmap_range mcp23x17_volatile_range = {
> >          .range_min = MCP_INTF << 1,
> > -       .range_max = MCP_GPIO << 1,
> > +       .range_max = (MCP_GPIO << 1) + 1,
>
> This looks weird. Usually we do a mask or a bit based mask, like (1 << x) - 1.

I don't think these are masks, they're addresses.

I believe the author has doubled the register indexing using a 1 bit 
shift, because the MCP23017 device is configured with sequential 
addresses (IOCON.BANK = 0). On page 12 of the datasheet [2] this looks like:

0x00 IODIRA, MCP_IODIR << 1
0x01 IODIRB
0x02 IPOLA,  MCP_IPOL << 1
0x03 IPOLB
...
0x12 GPIOA,  MCP_GPIO << 1
0x13 GPIOB

This means you can read 16 bits from MCP_GPIO << 1 and get the register 
values for both banks, or even use this for .range_min.

However, this trick doesn't work for .range_max:

	.range_max = MCP_GPIO << 1; /* 0x12 */

But I think it needs to be 0x13 to include GPIOB. Now that I'm looking 
into it, how does `mcp23x17_regmap.val_bits = 16` affect this? Perhaps 
`MCP_GPIO << 1` is fine after all.

I will whip up a v2 and test this. I'll split the changes across patches 
and fix the typo last patch - in case you don't agree with me.

Many thanks,
Thomas

[0] MCP23008 https://ww1.microchip.com/downloads/en/DeviceDoc/21919e.pdf
[1] MCP23016 http://ww1.microchip.com/downloads/en/devicedoc/20090c.pdf
[2] MCP23017 https://ww1.microchip.com/downloads/en/DeviceDoc/20001952C.pdf