On 11/20/20 17:16, Krzysztof Kozlowski wrote:
> of_match_device() already handles properly !CONFIG_OF case, so passing
> the argument via of_match_ptr() is not needed.
>
> Signed-off-by: Krzysztof Kozlowski <[email protected]>

Reviewed-by: Sylwester Nawrocki <[email protected]>

2020-11-20 16:59:42

On Fri, Nov 20, 2020 at 08:42:45PM +0100, Krzysztof Kozlowski wrote:
> On Fri, Nov 20, 2020 at 04:56:34PM +0000, Mark Brown wrote:

> > It would be better to fix these by annotating the table as potentially
> > unused, if nothing else it means if someone wants to add ACPI support
> > (or it just works on their ACPI system with the plain old I2C ID) then
> > they don't need to revert this change.

> The point is after this patch - removal of of_match_ptr() - they will
> already support the ACPI matching through the PRP0001.

> Keeping of_match_ptr() and maybe_unused will prevent any ACPI re-usage
> unless explicit ACPI table is added

Surely if that's the desired outcome the fix is to change the definition
of of_match_ptr() such that it leaves the reference with CONFIG_ACPI,
perhaps hidden behind a config option for PRP0001? That seems better
than going through the entire tree like this.

Attachments:

(No filename) (919.00 B)
signature.asc (499.00 B)
Download all attachments

2020-11-22 11:02:16

by Krzysztof Kozlowski

[permalink] [raw]

Subject: Re: [PATCH 01/38] ASoC: ak5558: drop of_match_ptr from of_device_id table

On Fri, Nov 20, 2020 at 08:04:29PM +0000, Mark Brown wrote:
> On Fri, Nov 20, 2020 at 08:42:45PM +0100, Krzysztof Kozlowski wrote:
> > On Fri, Nov 20, 2020 at 04:56:34PM +0000, Mark Brown wrote:
>
> > > It would be better to fix these by annotating the table as potentially
> > > unused, if nothing else it means if someone wants to add ACPI support
> > > (or it just works on their ACPI system with the plain old I2C ID) then
> > > they don't need to revert this change.
>
> > The point is after this patch - removal of of_match_ptr() - they will
> > already support the ACPI matching through the PRP0001.
>
> > Keeping of_match_ptr() and maybe_unused will prevent any ACPI re-usage
> > unless explicit ACPI table is added
>
> Surely if that's the desired outcome the fix is to change the definition
> of of_match_ptr() such that it leaves the reference with CONFIG_ACPI,
> perhaps hidden behind a config option for PRP0001? That seems better
> than going through the entire tree like this.

That could be indeed an easier way to achieve this.

+Cc Andy, Rafael,

I saw you were doing similar way as I did here [1] for the 698fffc2705c
("rtc: ds1307: Drop of_match_ptr and CONFIG_OF protections") with the
same reasoning as mine ("These prevent use of this driver with ACPI via
PRP0001.").

Do you have thoughts on Mark's proposal above (to change the
of_match_ptr())?

[1] https://lore.kernel.org/lkml/[email protected]/

Best regards,
Krzysztof

2020-11-23 10:51:07

by Andy Shevchenko

[permalink] [raw]

Subject: Re: [PATCH 01/38] ASoC: ak5558: drop of_match_ptr from of_device_id table

On Sun, Nov 22, 2020 at 11:59:20AM +0100, Krzysztof Kozlowski wrote:
> On Fri, Nov 20, 2020 at 08:04:29PM +0000, Mark Brown wrote:
> > On Fri, Nov 20, 2020 at 08:42:45PM +0100, Krzysztof Kozlowski wrote:
> > > On Fri, Nov 20, 2020 at 04:56:34PM +0000, Mark Brown wrote:
> >
> > > > It would be better to fix these by annotating the table as potentially
> > > > unused, if nothing else it means if someone wants to add ACPI support
> > > > (or it just works on their ACPI system with the plain old I2C ID) then
> > > > they don't need to revert this change.
> >
> > > The point is after this patch - removal of of_match_ptr() - they will
> > > already support the ACPI matching through the PRP0001.
> >
> > > Keeping of_match_ptr() and maybe_unused will prevent any ACPI re-usage
> > > unless explicit ACPI table is added
> >
> > Surely if that's the desired outcome the fix is to change the definition
> > of of_match_ptr() such that it leaves the reference with CONFIG_ACPI,
> > perhaps hidden behind a config option for PRP0001? That seems better
> > than going through the entire tree like this.
>
> That could be indeed an easier way to achieve this.

...easier and wrong in my opinion. Not all drivers need that.
What the point to touch it in the driver which is OF-only?
(For IP which will quite unlikely to be present in ACPI world)
Or if the device will get the correct ACPI ID?

> +Cc Andy, Rafael,

I guess Rafael can correct me or others.

> I saw you were doing similar way as I did here [1] for the 698fffc2705c
> ("rtc: ds1307: Drop of_match_ptr and CONFIG_OF protections") with the
> same reasoning as mine ("These prevent use of this driver with ACPI via
> PRP0001.").

The above is a device which can be connected to any system, including
ACPI-based one. The patch has been cooked to have some means to make
it usable on such systems (because previous patch removes wrong ACPI IDs).

> Do you have thoughts on Mark's proposal above (to change the
> of_match_ptr())?
>
> [1] https://lore.kernel.org/lkml/[email protected]/

--
With Best Regards,
Andy Shevchenko

2020-11-23 12:42:37

On Mon, Nov 23, 2020 at 01:41:29PM +0100, Krzysztof Kozlowski wrote:
> On Mon, Nov 23, 2020 at 12:37:31PM +0000, Mark Brown wrote:

> > That feels like something that should be done with Kconfig dependencies
> > like a direct OF dependency (possibly a !PRP0001 dependency?) for the
> > driver or possibly with having a variant of_match_ptr() for things that
> > really don't want to support PRP0001. Just removing all the use of
> > of_match_ptr() is both noisy and confusing in that it looks like it's
> > creating issues to fix, it makes it hard to understand when and why one
> > should use the macro.

> For the OF-only drivers (without other ID table), there is no point to
> use the macro. Driver can bind only with DT, so what is the point of
> of_match_ptr? To skip the OF table when building without OF? Driver
> won't be usable at all in such case. So maybe for compile testing?
> There is no need to remove OF table for simple build tests.

If nothing else it means you don't have to check if the driver is OF
only or not. I can see not bothering to add it but actively going round
removing some instances of it doesn't seem great, and it seems like
people will constantly be adding new uses on the basis that it's just
such an obviously correct thing to do.

Attachments:

(No filename) (1.26 kB)
signature.asc (499.00 B)
Download all attachments

2020-11-23 15:02:16

by Krzysztof Kozlowski

[permalink] [raw]

Subject: Re: [PATCH 01/38] ASoC: ak5558: drop of_match_ptr from of_device_id table

On Mon, Nov 23, 2020 at 01:50:06PM +0000, Mark Brown wrote:
> On Mon, Nov 23, 2020 at 01:41:29PM +0100, Krzysztof Kozlowski wrote:
> > On Mon, Nov 23, 2020 at 12:37:31PM +0000, Mark Brown wrote:
>
> > > That feels like something that should be done with Kconfig dependencies
> > > like a direct OF dependency (possibly a !PRP0001 dependency?) for the
> > > driver or possibly with having a variant of_match_ptr() for things that
> > > really don't want to support PRP0001. Just removing all the use of
> > > of_match_ptr() is both noisy and confusing in that it looks like it's
> > > creating issues to fix, it makes it hard to understand when and why one
> > > should use the macro.
>
> > For the OF-only drivers (without other ID table), there is no point to
> > use the macro. Driver can bind only with DT, so what is the point of
> > of_match_ptr? To skip the OF table when building without OF? Driver
> > won't be usable at all in such case. So maybe for compile testing?
> > There is no need to remove OF table for simple build tests.
>
> If nothing else it means you don't have to check if the driver is OF
> only or not. I can see not bothering to add it but actively going round
> removing some instances of it doesn't seem great, and it seems like
> people will constantly be adding new uses on the basis that it's just
> such an obviously correct thing to do.

If my patch was not changing anything, I would agree that it might be
just a churn. But the patch fixes a real warning.

The other way of fixing warning is the one you proposed at beginning -
adding maybe_unused. Here we go to the second reason:

Having these of_match_ptr() for OF-only drivers is not the correct way
but rather something which is copied from existing drivers into new
ones. This is another reason for removing them - people will stop
copying this code all over again.

Best regards,
Krzysztof

2020-11-23 16:49:24

On Mon, Nov 23, 2020 at 12:37:31PM +0000, Mark Brown wrote:
> On Mon, Nov 23, 2020 at 12:48:32PM +0200, Andy Shevchenko wrote:
> > On Sun, Nov 22, 2020 at 11:59:20AM +0100, Krzysztof Kozlowski wrote:
> > > On Fri, Nov 20, 2020 at 08:04:29PM +0000, Mark Brown wrote:
>
> > > > Surely if that's the desired outcome the fix is to change the definition
> > > > of of_match_ptr() such that it leaves the reference with CONFIG_ACPI,
> > > > perhaps hidden behind a config option for PRP0001? That seems better
> > > > than going through the entire tree like this.
>
> > > That could be indeed an easier way to achieve this.
>
> > ...easier and wrong in my opinion. Not all drivers need that.
> > What the point to touch it in the driver which is OF-only?
> > (For IP which will quite unlikely to be present in ACPI world)
> > Or if the device will get the correct ACPI ID?
>
> That feels like something that should be done with Kconfig dependencies
> like a direct OF dependency (possibly a !PRP0001 dependency?) for the
> driver or possibly with having a variant of_match_ptr() for things that
> really don't want to support PRP0001. Just removing all the use of
> of_match_ptr() is both noisy and confusing in that it looks like it's
> creating issues to fix, it makes it hard to understand when and why one
> should use the macro.

My personal opinion is that in 99% using that macro (as well as ACPI_PTR() one)
is kinda mistake. We save dozen of bytes here and there by adding macro,
necessary ifdeferry, and dropping user to know that the driver might serve for
other device IDs as provided by OF / ACPI.

But I'm not the one who makes a decision here and I could see some want to have
a possibility to reduce their kernel memory footprint as much as possible.

--
With Best Regards,
Andy Shevchenko