[PATCH v6 2/2] mfd: atmel-flexcom: add a driver for Atmel Flexible Serial Communication Unit

[Lee Jones]

On Thu, 23 Jul 2015, Cyrille Pitchen wrote:

> Hi all,
> 
> Le 23/07/2015 14:50, Boris Brezillon a écrit :
> > On Thu, 23 Jul 2015 10:13:11 +0100
> > Lee Jones <lee.jones at linaro.org> wrote:
> > 
> >> On Thu, 23 Jul 2015, Boris Brezillon wrote:
> >>
> >>> Hi Lee,
> >>>
> >>> On Thu, 23 Jul 2015 08:32:17 +0100
> >>> Lee Jones <lee.jones at linaro.org> wrote:
> >>>
> >>>> On Wed, 22 Jul 2015, Cyrille Pitchen wrote:
> >>>>> +	for_each_child_of_node(np, child) {
> >>>>> +		const char *compatible;
> >>>>> +		int cplen;
> >>>>> +
> >>>>> +		if (!of_device_is_available(child))
> >>>>> +			continue;
> >>>>> +
> >>>>> +		compatible = of_get_property(child, "compatible", &cplen);
> >>>>> +		if (!compatible || strlen(compatible) > cplen)
> >>>>> +			continue;
> >>>>> +
> >>>>> +		if (strstr(compatible, "-usart")) {
> >>>>> +			opmode = FLEX_MR_OPMODE_USART;
> >>>>> +			break;
> >>>>> +		}
> >>>>> +
> >>>>> +		if (strstr(compatible, "-spi")) {
> >>>>> +			opmode = FLEX_MR_OPMODE_SPI;
> >>>>> +			break;
> >>>>> +		}
> >>>>> +
> >>>>> +		if (strstr(compatible, "-i2c")) {
> >>>>> +			opmode = FLEX_MR_OPMODE_TWI;
> >>>>> +			break;
> >>>>> +		}
> >>>>> +	}
> >>>>
> >>>> From what I understand Flexcom is a wrapper which can sit above any
> >>>> number of SPI, I2C and/or UART devices.  Devices which you don't
> >>>> really have any control over (source code wise).  So wouldn't it be
> >>>> better to match on the details you do have control over i.e. the node
> >>>> name, rather than the compatible string?
> >>>>
> >>>> I would personally match on of_find_node_by_name() to future-proof
> >>>> your implementation.
> >>>
> >>> Actually, I think using compatible strings is more future-proof than
> >>> using the node names, because nothing in the DT bindings doc enforce the
> >>> node name, and usually what we use to attach a node to a specific
> >>> driver is the compatible string (this one is specified in the bindings
> >>> doc).
> >>
> >> I know what you're saying, but what if someone uses the Flexcom driver
> >> to wrap a different type of SPI driver where (for instance) the
> >> compatible string used is "<name>-<newtype>".  Then we'd have to keep
> >> adding more lines here to accommodate.
> >>
> >> Whereas if we used the child node name which only pertains to _this_
> >> driver, we would then have full control and know that (unless it
> >> Flexcom is used for a completely different type of serial controller)
> >> we wouldn't have to keep expanding the code to accommodate.
> > 
> > You're right about the complexity implied by the compat string
> > maintenance, but I still think using node names to detect the mode is
> > a bad idea.
> > 
> > Let's take another example making both solution unsuitable: what if
> > the flexcom-v2 exposes 2 devices of the same type, they will both have
> > the same name and the same compatible string, and we'll have no way to
> > detect the appropriate mode. That's why I think none of our suggestion
> > is future-proof.
> > 
> >>
> >>> Regarding the implementation itself, I would match the child node with
> >>> an of_device_id table rather than trying to find a specific substring
> >>> in the compatible string, but I think that's only a matter of taste.
> >>
> >> You mean duplicate each of the supported device's compatible strings
> >> in this driver, then fetch the attributed of_match_device()->data
> >> value?
> >>
> > 
> > Yes, and that's definitely not a good idea, but I think Cyrille has
> > found a better approach (I'll let him explain).
> 
> Indeed, what about taking advantage of the "ranges" property?
> 
> For the Flexcom:
> #address-cells = <2>;
> #size-cells = <1>;
> ranges = <1 0 0xf8034200 0x200    /* opmode 1: USART */
>           2 0 0xf8034400 0x200    /* opmode 2: SPI */
>           3 0 0xf8034600 0x200>;  /* opmode 3: I2C */
> 
> Then for the single available child (for instance the SPI controller):
> reg = <2 0 0x200>;
> 
> So the Operating Mode to be set into the Flexcom Mode Register is read from
> the very first u32 of the "reg" property of the child.
> 
> No need to introduce any new DT property and the mapping remains easy to
> maintain to follow hardware upgrades.

When we do things like this we normally do:

    reg <base base_size>, <mode mode_size>;
    reg-names "base", "mode";

Then use:

     platform_get_resource_byname()

.... to fetch them.

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog