On Wed, Jul 01, 2026 at 01:48:24PM -0500, David Lechner wrote: > Note that a few subsystems, including spi want the subject > to be `spi: dt-bindings:` rather than the other way around. > > See https://www.kernel.org/doc/html/latest/devicetree/bindings/submitting-patches.html > > > On 7/1/26 1:29 PM, Jonathan Cameron wrote: > > On Wed, 1 Jul 2026 12:04:37 +0100 > > Conor Dooley wrote: > > > >> On Wed, Jul 01, 2026 at 08:40:39AM +0200, Janani Sunil wrote: > >>> Some SPI devices support sharing a single chip select across multiple > >>> physical chips by encoding a device address in the SPI frame itself. > >>> Add a generic spi,device-addr property to document this per-peripheral > >>> address. This property belongs in channel or sub-device nodes of > >>> peripherals that use this addressing scheme. > >>> > >>> Signed-off-by: Janani Sunil > >>> --- > >>> Documentation/devicetree/bindings/spi/spi-peripheral-props.yaml | 5 +++++ > >>> 1 file changed, 5 insertions(+) > >>> > >>> diff --git a/Documentation/devicetree/bindings/spi/spi-peripheral-props.yaml b/Documentation/devicetree/bindings/spi/spi-peripheral-props.yaml > >>> index 880a9f624566..3774e8018355 100644 > >>> --- a/Documentation/devicetree/bindings/spi/spi-peripheral-props.yaml > >>> +++ b/Documentation/devicetree/bindings/spi/spi-peripheral-props.yaml > >>> @@ -142,6 +142,11 @@ properties: > >>> minItems: 2 > >>> maxItems: 4 > >>> > >>> + spi,device-addr: > >> > >> To match other generic spi properties, s/,/-/. > >> > >> However, you don't actually use this as a spi peripheral's property in > >> your device binding, so you've got your wires crossed here somewhere. > > > > If we are going to make this generic (which I'm not against) I think > > it should also work for the case of multiple independent devices. > > So it can also be a top level device node spi property. > > > > That kind of makes me wonder if we are better off having it always > > in the top level node, but allowing multiple values to represent > > sub devices under this. That would leave figuring out mappings of which > > channels are on which device to the driver. The driver must know the > > mapping afterall. For the example something like > > > > > > #include > > spi { > > #address-cells = <1>; > > #size-cells = <0>; > > dac@0 { > > compatible = "adi,ad5529r-16"; > > reg = <0>; > > spi-max-frequency = <25000000>; > > > > spi-device-addreses = <0 3> > > ... > > > > #address-cells = <1>; > > #size-cells = <0>; > > > > channel@0 { > > reg = <0>; > > adi,output-range-microvolt = <0 5000000>; > > }; > > > > channel@16 { #on second device using dev addr 3 > > reg = <16>; > > adi,output-range-microvolt = <(-10000000) 10000000>; > > }; > > channel@18 { #3rd channel on device using dev addr 3 > > reg = <18>; > > adi,output-range-microvolt = <0 40000000>; > > }; > > }; > > }; > > > > Where devices are truely independent then you would have separate device > > nodes each with one entry in spi-device-addresses > > > > I'm a bit dubious about putting this in the spi namespace though given > > it is not part of any standard specification. Do we have any precedence > > for that sort of thing? > > It seems like most SPI controllers/devices don't really follow any > standards, so I think there is plenty of precedence for a property > like this. It would be nice to see one or two more examples of SPI > peripherals with this feature though other than the one chip in > this series. Otherwise, I wouldn't try to make it a standard property. There are two microchip devices with a property like "microchip,hw-addr" that I pointed out on ?v3? (whichever thread had the long discussion) that do the exact same thing as this device. I did ask that the submitter convert those to the new property as evidence that this is actually generic, but that request was not implemented. If done at the device level, this should be trivial, as the microchip devices don't actually have support for multiple devices on the same cs in the drivers, they just parse the property to correctly support a single device. > I'm also in favor of making it an array and letting the device-specific > bindings decide what multiple devices with different addresses on the > same CS line means. I kinda wanted it to be in the channels, but I think, on reflection, that putting it at the device level is probably fine. Pretending that two devices are one extended device is always going to have some ickyness about presentation, and putting it in the channels means defining it again and again and again for each type of device that wants it. > But... if we are leaving it up to devices to deal with the property rather > than the core SPI code, maybe it shouldn't be a standard SPI property. > Although, I suppose the core SPI code could parse the property and just > pass that information in the struct spi_device to let the device driver > do what it wants with it.