On Tue, Aug 20, 2013 at 06:19:49AM +0100, Shawn Guo wrote:
On Mon, Aug 19, 2013 at 10:54:33AM +0100, Mark Rutland wrote:
I guess it's better to drop the 'imx6q-spdif' here?
That depends:
If the two IP blocks are identical, only the "imx35-spdif" name is necessary, and we can forget about "fsl,imx6q-spdif".
If "fsl,imx6q-spdif" is a strict superset of "fsl,imx35-spdif", having both names documented and in a compatible list for a "fsl,imx6q-spdif" device makes sense.
Practically, I found it's very useful to have "fsl,<soc>-<ip>" in the device compatible property in <soc>.dtsi, even when device driver does not match it right now. For this example, I still prefer to have the following line for spdif device in imx6q.dtsi.
compatible = "fsl,imx6q-spdif", "fsl,imx35-spdif";
This is a good thing, but I was not referring to the driver. Nicolin presumably has enough information about the hardware to know if the new block is fully compatible, even if the driver doesn't yet poke the hardware in such a way as to make that obvious.
The reason for that is we usually do not see all the differences of an IP block from one SoC to another when we firstly define the bindings for the device by looking at hardware reference manual. Some programming model differences are only identified when we're actually programming. That said, if some day we find there is difference between imx6q-spdif and imx35-spdif to be handled when we add something new to the driver, we only need to add "fsl,imx6q-spdif" as a new compatible into device driver and bindings document. The existing device tree would need no update to work with the new kernel driver.
That's a valid concern, but I think that bindings should be documented from the start (even if unused by the driver), otherwise we're in an equally bad position later if we discover a device isn't backwards compatible as we have no guarantee the new (previously undocumented) string was picked up in all dts.
Thanks, Mark.