[PATCH v5 2/6] dt-bindings: auxdisplay: Add Titan Micro Electronics TM1628
Heiner Kallweit
hkallweit1 at gmail.com
Wed Mar 23 13:33:20 PDT 2022
On 21.03.2022 09:34, Krzysztof Kozlowski wrote:
> On 18/03/2022 21:50, Robin Murphy wrote:
>> On 2022-02-25 21:13, Heiner Kallweit wrote:
>>> Add a YAML schema binding for TM1628 auxdisplay
>>> (7/11-segment LED) controller.
>>>
>>> This patch is partially based on previous work from
>>> Andreas Färber <afaerber at suse.de>.
>>>
>>> Co-developed-by: Andreas Färber <afaerber at suse.de>
>>> Signed-off-by: Andreas Färber <afaerber at suse.de>
>>> Signed-off-by: Heiner Kallweit <hkallweit1 at gmail.com>
>>> ---
>>> v5:
>>> - add vendor prefix to driver-specific properties
>>> ---
>>> .../bindings/auxdisplay/titanmec,tm1628.yaml | 92 +++++++++++++++++++
>>> 1 file changed, 92 insertions(+)
>>> create mode 100644 Documentation/devicetree/bindings/auxdisplay/titanmec,tm1628.yaml
>>>
>>> diff --git a/Documentation/devicetree/bindings/auxdisplay/titanmec,tm1628.yaml b/Documentation/devicetree/bindings/auxdisplay/titanmec,tm1628.yaml
>>> new file mode 100644
>>> index 000000000..2a1ef692c
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/auxdisplay/titanmec,tm1628.yaml
>>> @@ -0,0 +1,92 @@
>>> +# SPDX-License-Identifier: (GPL-2.0-or-later OR BSD-2-Clause)
>>> +%YAML 1.2
>>> +---
>>> +$id: http://devicetree.org/schemas/auxdisplay/titanmec,tm1628.yaml#
>>> +$schema: http://devicetree.org/meta-schemas/core.yaml#
>>> +
>>> +title: Titan Micro Electronics TM1628 LED controller
>>> +
>>> +maintainers:
>>> + - Andreas Färber <afaerber at suse.de>
>>> + - Heiner Kallweit <hkallweit1 at gmail.com>
>>> +
>>> +allOf:
>>> + - $ref: /schemas/spi/spi-peripheral-props.yaml#
>>> +
>>> +properties:
>>> + compatible:
>>> + const: titanmec,tm1628
>>> +
>>> + reg:
>>> + maxItems: 1
>>> +
>>> + titanmec,grid:
>>> + description:
>>> + Mapping of display digit position to grid number.
>>> + This implicitly defines the display size.
>>> + $ref: /schemas/types.yaml#/definitions/uint8-array
>>> + minItems: 1
>>> + maxItems: 7
>>> +
>>> + titanmec,segment-mapping:
>>> + description:
>>> + Mapping of 7 segment display segments A-G to bit numbers 1-12.
>>> + $ref: /schemas/types.yaml#/definitions/uint8-array
>>> + minItems: 7
>>> + maxItems: 7
>>> +
>>> + "#address-cells":
>>> + const: 2
>>> +
>>> + "#size-cells":
>>> + const: 0
>>> +
>>> +required:
>>> + - compatible
>>> + - reg
>>
>> Would it be fair to say that "spi-lsb-first" and "spi-3wire" are also
>> required? The chips aren't configurable so won't exactly be usable any
>> other way. Furthermore I believe the transmission format actually works
>> out equivalent to SPI mode 3, so should warrant "spi-cpha" and
>> "spi-cpol" as well.
>>
>>> +
>>> +patternProperties:
>>> + "^.*@[1-7],([1-9]|1[0-6])$":
>>> + type: object
>>> + $ref: /schemas/leds/common.yaml#
>>> + unevaluatedProperties: false
>>> + description: |
>>> + Properties for a single LED.
>>> +
>>> + properties:
>>> + reg:
>>> + description: |
>>> + 1-based grid number, followed by 1-based segment bit number.
>>> + maxItems: 1
>>> +
>>> + required:
>>> + - reg
>>
>> I'm concerned that this leaves us no room to support the additional
>> keypad functionality in future. Having now double-checked a datasheet,
>> the inputs are also a two-dimensional mux (sharing the segment lines),
>> so the device effectively has two distinct but numerically-overlapping
>> child address spaces - one addressed by (grid,segment) and the other by
>> (segment,key).
>>
>> Rob, Krysztof, any thoughts on the best DT idiom to leave accommodation
>> for that? I'm thinking either require an intermediate node to contain
>> each notional address space, or perhaps add another leading address cell
>> to select between them? I don't believe any of these things have further
>> functionality beyond that.
>
> I think intermediate nodes - leds, keys - are more appropriate, because
> it is self-describing. Additional address space number would require
> decoding this "0" or "1" into LED/key. For complex devices - like PMICs
> with regulators, RTC and clocks - we already have such patterns.
>
Then it's just the question who can implement such an intermediate node
based on what has been done so far.
> Best regards,
> Best regards,
> Krzysztof
More information about the linux-amlogic
mailing list