Re: [PATCH v6 3/7] mfd: Add support for the Lantiq PEF2256 framer

[Lee Jones <lee@kernel.org>]

On Fri, 21 Apr 2023, Krzysztof Kozlowski wrote:

> On 21/04/2023 09:45, Lee Jones wrote:
> > On Fri, 21 Apr 2023, Herve Codina wrote:
> > 
> >> Hi Lee, Krzysztof,
> >>
> >> On Thu, 20 Apr 2023 14:47:03 +0100
> >> Lee Jones <lee@kernel.org> wrote:
> >>
> >>> On Thu, 20 Apr 2023, Herve Codina wrote:
> >>>
> >>>> On Thu, 20 Apr 2023 13:39:46 +0100
> >>>> Lee Jones <lee@kernel.org> wrote:
> >>>>   
> >>>>> On Mon, 17 Apr 2023, Herve Codina wrote:
> >>>>>   
> >>>>>> The Lantiq PEF2256 is a framer and line interface component designed to
> >>>>>> fulfill all required interfacing between an analog E1/T1/J1 line and the
> >>>>>> digital PCM system highway/H.100 bus.
> >>>>>>
> >>>>>> Signed-off-by: Herve Codina <herve.codina@bootlin.com>
> >>>>>> ---
> >>>>>>  drivers/mfd/Kconfig         |  16 +
> >>>>>>  drivers/mfd/Makefile        |   1 +
> >>>>>>  drivers/mfd/pef2256-regs.h  | 250 ++++++++++
> >>>>>>  drivers/mfd/pef2256.c       | 950 ++++++++++++++++++++++++++++++++++++    
> >>>>>
> >>>>> 95% of this driver needs to be moved somewhere else.
> >>>>>
> >>>>> What is a Framer?  Perhaps sound/ is a good candidate?  
> >>>>
> >>>> The pef2256 framer is a device that transfers data to/from a TDM (time-slots
> >>>> data) from/to quite old telecommunication lines (E1 in my case).
> >>>> Several subsystem can set/get data to/from the TDM. Each device using their
> >>>> own time-slots set.
> >>>>
> >>>> On my use-case, I have some audio consumer and a not yet upstreamed HDLC
> >>>> consumer. Both of them uses the framer to know the E1 link state.
> >>>> The framer needs to be initialized 'globally' and not by a specific consumer
> >>>> as several consumers can use the framer.  
> >>>
> >>> I can't think of a good place for this.
> >>>
> >>> If all else fails, it's drivers/misc
> >>>
> >>>>>>  include/linux/mfd/pef2256.h |  52 ++
> >>>>>>  5 files changed, 1269 insertions(+)
> >>>>>>  create mode 100644 drivers/mfd/pef2256-regs.h
> >>>>>>  create mode 100644 drivers/mfd/pef2256.c
> >>>>>>  create mode 100644 include/linux/mfd/pef2256.h    
> >>>>>
> >>>>> [...]
> >>>>>   
> >>>>>> +static int pef2256_add_audio_devices(struct pef2256 *pef2256)
> >>>>>> +{
> >>>>>> +	const char *compatible = "lantiq,pef2256-codec";
> >>>>>> +	struct mfd_cell *audio_devs;
> >>>>>> +	struct device_node *np;
> >>>>>> +	unsigned int count = 0;
> >>>>>> +	unsigned int i;
> >>>>>> +	int ret;
> >>>>>> +
> >>>>>> +	for_each_available_child_of_node(pef2256->dev->of_node, np) {
> >>>>>> +		if (of_device_is_compatible(np, compatible))
> >>>>>> +			count++;
> >>>>>> +	}    
> >>>>>
> >>>>> Converting Device Tree nodes into MFD cells to register with the
> >>>>> Platform Device API is not a reasonable use-case of MFD.
> >>>>>
> >>>>> Have the CODEC driver match on "lantiq,pef2256-codec" and let it
> >>>>> instantiate itself.  
> >>>>
> >>>> As the framer is going to used by several subsystem, I cannot instantiate
> >>>> it in the specific ASoC subsystem.
> >>>>   
> >>>>>
> >>>>> Your first version using of_platform_populate() was closer to the mark.  
> >>>>
> >>>> The issue was that I need MFD cells for the pinctrl part.  
> >>>
> >>> Why can't it be represented in DT?
> >>
> >> The pinctrl part has no specific compatible string.
> >> Not sure that a compatible string for pinctrl can be accepted
> >> as there is only one pinctrl subnode and no specific reg for this
> >> subnode.
> >>
> >> The DT looks like this:
> >>     framer@2000000 {
> >>       compatible = "lantiq,pef2256";
> >>       reg = <0x2000000 0x100>;
> >>       ...
> >>       pinctrl {
> >>         pef2256_rpa_sypr: rpa-pins {
> >>           pins = "RPA";
> >>           function = "SYPR";
> >>         };
> >>       };
> >>
> >>       pef2256_codec0: codec-0 {
> >>         compatible = "lantiq,pef2256-codec";
> >>         #sound-dai-cells = <0>;
> >>         sound-name-prefix = "PEF2256_0";
> >>       };
> >>     };
> >>
> >> Krzysztof, is it acceptable to have a compatible string in the pinctrl node ?
> > 
> > Why wouldn't it be?
> > 
> > $ git grep ".compatible" -- drivers/pinctrl/
> > 
> >> In this case, it will looks like this:
> >>     framer@2000000 {
> >>       compatible = "lantiq,pef2256";
> >>       reg = <0x2000000 0x100>;
> >>       ...
> >>       pinctrl {
> >>         compatible = "lantiq,pef2256-pinctrl";
> 
> If you do not have any resources, there is no point in having separate
> compatible for separate device node.

That's a new rule.  Is that documented somewhere?  I'm sure we already
have device nodes for devices whom only operate on shared resources. 

> Anyway this discussions should not be about DT. How Linux drivers are
> implementing DT is not really a guide how to write DT. Since these
> series were brought there were some DT decisions made based how you want
> to write the driver. No, please don't. I also do not see any problems in
> handling more-or-less complex driver structures without poking the DT.
> We have already many such device families.

-- 
Lee Jones [李琼斯]