
On 12.10.2017 14:53, Chen-Yu Tsai wrote:
On Thu, Oct 12, 2017 at 8:24 PM, Felix Brack fb@ltec.ch wrote:
On 12.10.2017 04:46, Chen-Yu Tsai wrote:
On Mon, Oct 9, 2017 at 6:04 PM, Felix Brack fb@ltec.ch wrote:
This patch extends pmic_bind_children prefix matching. In addition to the node name the property regulator-name is used while trying to match prefixes. This allows assigning different drivers to regulator nodes named regulator@1 and regulator@10 for example.
No. See the regulator bindings:
Optional properties:
- regulator-name: A string used as a descriptive name for regulator outputs
The actual regulator.txt states:
Optional properties:
- regulator-name: a string, required by the regulator uclass
This was the reason for choosing the regulator-name property.
Mine was from the Linux Kernel. Are there two sets of bindings? Shouldn't there be just one?
Mine was from U-Boot as this is the U-Boot mailing list and things do not seem to be fully synchronized between LINUX and U-Boot.
This can vary from board to board. The name should match the power rail name of the board (which may not be the same as the regulator chip's output name).
Exactly. I totally agree but as stated in an earlier post: I did not define the names for the regulators and modifying them is almost certainly not the right way to go. Let me explain this briefly. The regulator names I'm trying to match are those from tps65910.dtsi, an include file. The exact same file is part of the LINUX kernel. Therefore I resigned suggesting the modification of the node names.
First, it is an include file. Board files are free to override the name. We did that for sunxi at first, have the .dtsi file provide some default names, then have board .dts files override them with names that make more sense. Later on we just dropped the default names altogether.
I am definitely confused now, maybe we are using same terms for different things. When I'm talking about a 'name' I mean the 'node name' according to DT specification (as for instance returned by ofnode_get_name). I'm not talking about a property or a node label. The following code defines a node name 'regulator@2' ore more precisely a node named 'regulator' having unit-address 2:
vdd1_reg: regulator@2 { reg = <2>; regulator-compatible = "vdd1"; };
This is found in tps65910.dtsi include file. You say "board files are free to override the name". Hence I could include the tps65910.dtsi into my board dts file and kind of rename node 'regulator@2' by let's say 'buck_vdd1@2'? The only way of "overriding" I can think of is by not including the dtsi file and redefining all nodes.
The same pattern is also seen in tps65910.dtsi and am3517-craneboard.dts. And also am335x-evm.dts, which the tps65910.dtsi was tested with.
Looking at the am3517-craneboard.dts file I don't see how node names are getting overwritten. What gets set, changed or overwritten is the node property regulator-name.
Now I couldn't find the binding document for tps65910, but looking at tps65217 instead, it says:
- regulators: list of regulators provided by this controller, must be named after their hardware counterparts: dcdc[1-3] and ldo[1-4]
And indeed that is what's used in the example.
Yes, exactly and correct. Again, this tps65217.txt does only exist in LINUX and not in U-Boot.
So clearly the dts files aren't following the binding.
And what about the dtsi files? Are they correct in defining node names as 'regulator@[unit-address]'?
If you have multiple regulator nodes which need to be differentiated, you need to use the deprecated "regulator-compatible" property, or just use the standard compatible property.
Good point. I would not use a deprecated property but the compatible property seems reasonable to me. So you agree that the patch's concept could be retained while substituting the node-name property by the compatible property?
If you're going to modify the binding and/or device tree files, please do it in both places. Ideally the platform should have one canonical source for them.
Linux's standard regulator DT matching code only matches against node names and the deprecated "regulator-compatible" property. Not even the standard "compatible" is used, though I see a few PMICs using that. So even if you do get them working this way in U-boot, it's still not going to work in Linux, and you might get other comments when pushing your changes.
I guess if LINUX would not match against the regulator-compatible property but only against the node name things would not work properly. Again looking the file tps65217.dtsi shows that every regulator node (named regulator@0 to regulator@6) defines the regulator-compatible property. Only matching against this property therefore makes things working.
Felix