Question about compatible fallback and documentation

Question about compatible fallback and documentation

[Clément Péron]

Hi,

I have to bind an already existing IP by a vendor in a new SoC called
"SOC3" device-tree.

In the 1st gen of "SOC1" the IP is introduced :
soc1.dtsi :
compatible = "vendor,ip-soc1";

Then a 2nd gen of the IP is introduced in "SOC2" with new registers.
But the driver of the 1st gen is still working fine and no update of
the existing driver has been introduced because not required.
soc2.dtsi :
compatible = "vendor,ip-soc2", "vendor,ip-soc1";

Finally in "SOC3" and regardind the user manual we think that the IP
introduced is the same as "SOC2".
Should the compatible in soc3.dtsi be A or B?
A) compatible = "vendor,ip-soc2", "vendor,ip-soc1";
or
B) compatible = "vendor,ip-soc3", "vendor,ip-soc2", "vendor,ip-soc1";

I propose the solution B) because we don't know what could happens
maybe the IP could need a quirks only for "SOC3". And device tree
shouldn't move for the user only the driver.

Last question does we have to document all the compatible use in
DTS(i) files in the Documentation ? or only the compatible used by the
drivers ?

Thanks for answers,
Clement

Re: Question about compatible fallback and documentation

[Rob Herring]

On Sun, Apr 14, 2019 at 06:18:04PM +0200, Cl�ment P�ron wrote:
> Hi,
> 
> I have to bind an already existing IP by a vendor in a new SoC called
> "SOC3" device-tree.
> 
> In the 1st gen of "SOC1" the IP is introduced :
> soc1.dtsi :
> compatible = "vendor,ip-soc1";
> 
> Then a 2nd gen of the IP is introduced in "SOC2" with new registers.
> But the driver of the 1st gen is still working fine and no update of
> the existing driver has been introduced because not required.
> soc2.dtsi :
> compatible = "vendor,ip-soc2", "vendor,ip-soc1";
> 
> Finally in "SOC3" and regardind the user manual we think that the IP
> introduced is the same as "SOC2".
> Should the compatible in soc3.dtsi be A or B?
> A) compatible = "vendor,ip-soc2", "vendor,ip-soc1";
> or
> B) compatible = "vendor,ip-soc3", "vendor,ip-soc2", "vendor,ip-soc1";
> 
> I propose the solution B) because we don't know what could happens
> maybe the IP could need a quirks only for "SOC3". And device tree
> shouldn't move for the user only the driver.

B is correct.

Or you could list soc3 and soc2 given you do know there are additional 
features. That would require a driver update, but likely the new SoC 
requires some OS changes. Maybe someday SoC design will be disciplined 
enough that new SoCs are fully backwards compatible.

> Last question does we have to document all the compatible use in
> DTS(i) files in the Documentation ? or only the compatible used by the
> drivers ?

What is used in DTS files.

Rob

Re: Question about compatible fallback and documentation

[Clément Péron]

Hi,

On Tue, 30 Apr 2019 at 00:05, Rob Herring <robh@kernel.org> wrote:
>
> On Sun, Apr 14, 2019 at 06:18:04PM +0200, Clément Péron wrote:
> > Hi,
> >
> > I have to bind an already existing IP by a vendor in a new SoC called
> > "SOC3" device-tree.
> >
> > In the 1st gen of "SOC1" the IP is introduced :
> > soc1.dtsi :
> > compatible = "vendor,ip-soc1";
> >
> > Then a 2nd gen of the IP is introduced in "SOC2" with new registers.
> > But the driver of the 1st gen is still working fine and no update of
> > the existing driver has been introduced because not required.
> > soc2.dtsi :
> > compatible = "vendor,ip-soc2", "vendor,ip-soc1";
> >
> > Finally in "SOC3" and regardind the user manual we think that the IP
> > introduced is the same as "SOC2".
> > Should the compatible in soc3.dtsi be A or B?
> > A) compatible = "vendor,ip-soc2", "vendor,ip-soc1";
> > or
> > B) compatible = "vendor,ip-soc3", "vendor,ip-soc2", "vendor,ip-soc1";
> >
> > I propose the solution B) because we don't know what could happens
> > maybe the IP could need a quirks only for "SOC3". And device tree
> > shouldn't move for the user only the driver.
>
> B is correct.
>
> Or you could list soc3 and soc2 given you do know there are additional
> features. That would require a driver update, but likely the new SoC
> requires some OS changes. Maybe someday SoC design will be disciplined
> enough that new SoCs are fully backwards compatible.

Thanks for the answer,
Clement

>
> > Last question does we have to document all the compatible use in
> > DTS(i) files in the Documentation ? or only the compatible used by the
> > drivers ?
>
> What is used in DTS files.
>
> Rob