2022-03-17 06:08:00

by Marc Zyngier

[permalink] [raw]
Subject: Re: [PATCH 3/3] clocksource: arch_timer: Add arm,cortex-a7/15-timer in of_match list

On Wed, 16 Mar 2022 17:41:08 +0000,
Kuldeep Singh <[email protected]> wrote:
>
> On Wed, Mar 16, 2022 at 05:30:26PM +0100, Krzysztof Kozlowski wrote:
> > On 16/03/2022 10:54, Kuldeep Singh wrote:
> > > Few platforms such as Renesas RZ/N1D, Calxeda, Alpine etc. are using
> > > arm,cortex-a15-timer and arm,cortex-a7-timer entries in conjugation with
> > > arm,armv7-timer which are not currently defined in driver file. Add
> > > these entries in arch_timer_of_match list to bring them in use.
> > >
> >
> > This looks wrong (also Marc pointed this out) and rationale is not
> > sufficient. Why do you need these compatibles in the driver?
>
> Hi Krzysztof and Marc,
>
> I find myself in trouble whenever dealing with compatible entries and
> had 2 options when I stumble this issue.
> 1. Remove unused compatible

That'd be silly.

> 2. Add required compatible to binding and driver

To the binding, yes. But to the driver?

> My past experience and advise from other developer says not to remove an
> existing compatible. And also I found "arm,cortex-a15-timer" in binding
> which was again not documented and was present in DT. This prompted me
> to go for second option and make necessary additions in binding and
> driver following current entries.

The "arm,cortex-a15-timer" compatible is documentation, and only
that. If, one day, we find a bug in this implementation, we could work
around it in the driver thanks to the separate compatible (although in
this case, we'd have much better way of doing that).

> As per your perspective, current configuration isn't apt which means
> "arm,cortex-a15-timer" is a stub and is wrongly present in binding.

That's not what I said. This compatible string is perfectly fine, and
accurately describe the HW. The driver doesn't need to know about the
fine details of the implementation, and is perfectly happy with the
current state of things.

Think of it as an instance of a class. The driver doesn't need to know
the instance, only that it is a certain class.

> I also observed many other DTs have compatibles which are not present in
> driver. What is an ideal idealogy behind such cases?

I think I've made myself clear above.

Thanks,

M.

--
Without deviation from the norm, progress is not possible.


2022-03-17 08:35:34

by Kuldeep Singh

[permalink] [raw]
Subject: Re: [PATCH 3/3] clocksource: arch_timer: Add arm,cortex-a7/15-timer in of_match list

On Wed, Mar 16, 2022 at 06:43:15PM +0000, Marc Zyngier wrote:
> On Wed, 16 Mar 2022 17:41:08 +0000,
> Kuldeep Singh <[email protected]> wrote:
> >
> > On Wed, Mar 16, 2022 at 05:30:26PM +0100, Krzysztof Kozlowski wrote:
> > > On 16/03/2022 10:54, Kuldeep Singh wrote:
> > > > Few platforms such as Renesas RZ/N1D, Calxeda, Alpine etc. are using
> > > > arm,cortex-a15-timer and arm,cortex-a7-timer entries in conjugation with
> > > > arm,armv7-timer which are not currently defined in driver file. Add
> > > > these entries in arch_timer_of_match list to bring them in use.
> > > >
> > >
> > > This looks wrong (also Marc pointed this out) and rationale is not
> > > sufficient. Why do you need these compatibles in the driver?
> >
> > Hi Krzysztof and Marc,
> >
> > I find myself in trouble whenever dealing with compatible entries and
> > had 2 options when I stumble this issue.
> > 1. Remove unused compatible
>
> That'd be silly.
>
> > 2. Add required compatible to binding and driver
>
> To the binding, yes. But to the driver?
>
> > My past experience and advise from other developer says not to remove an
> > existing compatible. And also I found "arm,cortex-a15-timer" in binding
> > which was again not documented and was present in DT. This prompted me
> > to go for second option and make necessary additions in binding and
> > driver following current entries.
>
> The "arm,cortex-a15-timer" compatible is documentation, and only
> that. If, one day, we find a bug in this implementation, we could work
> around it in the driver thanks to the separate compatible (although in
> this case, we'd have much better way of doing that).
>
> > As per your perspective, current configuration isn't apt which means
> > "arm,cortex-a15-timer" is a stub and is wrongly present in binding.
>
> That's not what I said. This compatible string is perfectly fine, and
> accurately describe the HW. The driver doesn't need to know about the
> fine details of the implementation, and is perfectly happy with the
> current state of things.
>
> Think of it as an instance of a class. The driver doesn't need to know
> the instance, only that it is a certain class.
>

Thanks Marc for sharing knowledge. This was indeed helpful.
To sum up from what I understood, bindings and DTs should always be in
sync and driver file may not need to define all compatible entries as
long as purpose is served.

This means no driver change will be required to address
"arm,cortex-a7-timer". To which I have a question to Krzysztof.

Hi Krzysztof,

As per your comments on 2/3 patch, that it's DT which is not aligned
with binding w.r.t arm,cortex-a7-timer.

What makes "arm,cortex-a7-timer" an invalid entry from binding
perspective when we have a similar entry "arm,cortex-a15-timer" already
present?

I think we should share some common grounds here and keep both of them
in bindings or remove them altogether. I prefer first option, What's
your say?
Or please let me know in case there's better way to address this.

- Kuldeep

> > I also observed many other DTs have compatibles which are not present in
> > driver. What is an ideal idealogy behind such cases?
>
> I think I've made myself clear above.
>
> Thanks,
>
> M.
>
> --
> Without deviation from the norm, progress is not possible.