Re: [PATCH 0/7] of: overlay: Add support for export-symbols node feature

[Ayush Singh]

On 10/12/24 16:25, Herve Codina wrote:
> Hi Ayush,
>
> On Tue, 10 Dec 2024 15:26:44 +0530
> Ayush Singh <ayush@xxxxxxxxxxxxxxx> wrote:
>
> > On 10/12/24 15:11, Herve Codina wrote:
> > > Hi Ayush,
> > >
> > > On Tue, 10 Dec 2024 14:52:22 +0530
> > > Ayush Singh <ayush@xxxxxxxxxxxxxxx> wrote:
> > >
> > > ...
> > > > What is the reason for not using symbols directly as described here [3]?
> > > >
> > > > I do like this approach since it does not pollute the global symbols.
> > > > Just want to know if there are any other reasons for it.
> > > >   
> > >
> > > Modifying the __symbols__ node at runtime (adding / removing properties in
> > > it) exposes memory leaks if __symbols__ already exist in the live DT.
> > > This __symbols__ node exist if the dtb was compiled with '-@' or if you
> > > chain the overlay (i.e. __symbols__ node created by the first overlay).
> >
> > Yeah, that is a problem, specially in a setup which might involve
> > hot-plugging.
> >
> > > I think also that some conflicts can appears. What happens if you want to
> > > add a new label but this label is already present for some other purpose?
> >
> > I do not think that actually is a problem. As described in the original
> > patch [0], the symbol and connector overlay is supposed to be applied as
> > a group (overwriting any conflicting symbols in the process).
> >
> > The reason why this is not a problem is that `__symbols__` are only used
> > to resolve the phandles (overlays do not support path references yet),
> > but do not really have a purpose in the livetree (at least far as I
> > know, but I can be wrong).
> >
> > > Best regards,
> > > Hervé
> >
> > [0]: https://lore.kernel.org/lkml/20240702164403.29067-1-afd@xxxxxx/
>
>
> Also, in your first overlay (adding symbols in __sympbols__ node), you have
> something like:
>     GROVE_PIN1_MUX_I2C_SCL = "/bus@f0000/pinctrl@f4000/grove-i2c-pins";
>
> If I understood correctly, other overlays will have GROVE_PIN1_MUX_I2C_SCL
> as unresolved symbols and will use GROVE_PIN1_MUX_I2C_SCL to reference the
> grove-i2c-pins node.
> This unresolved symbol from the overlay is resolved thanks to the __symbols__
> table where you added GROVE_PIN1_MUX_I2C_SCL (first overlay operation).
>
> In order to work, you need to have a phandle property set in the
> grove-i2c-pins node.
>
> This is done by dtc when you compile the dtb containing the grove-i2c-pins
> node (i.e. k3-am625-beagleplay.dts)
>
> The phandle property will be set only if:
> - a label for grove-i2c-pins already exist and -@ option is used
> or
> - a label for grove-i2c-pins already exist and it is referenced as a phandle
>    in the dts (k3-am625-beagleplay.dts).
>
> Otherwise, dtc will not create the phandle property and without this
> property, the symbol resolution will not be correct.
>
> Best regards,
> Hervé

Hello Hervé

Thanks for the clarification. things have changed a bit since the last 
message and it seems like trying to add path reference support to 
overlays is not the best way forward [0]. So I would love to help move 
this approach forward.

I do have a question regarding this approach, so here I go:

Can the `export-symbols` node be added to devicetree spec and be 
resolved by the devicetree compiler (and fdtoverlay) instead of being 
runtime resolution.

To get some context, I would like to share the addon-board overlays 
between ZephyrRTOS and Linux kernel. I would be happy to try adding 
support to dtc compiler for it. I am also tagging David Gibson (dtc 
maintainer) in this discussion since he also had some ideas regarding 
the feasibility and pitfalls of adding it to devicetree compiler (and spec).


[0]: 
https://lore.kernel.org/devicetree-compiler/6b2dba90-3c52-4933-88f3-b47f96dc7710@xxxxxxxxxxxxxxx/T/#m900b5ca13cfc28396d4d46d9c3130a7070fa8c90

Best regards,
Ayush Singh