[PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Jean Delvare]

Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
can be enabled on all architectures. Therefore depending on
COMPILE_TEST as an alternative is no longer needed.

Signed-off-by: Jean Delvare <jdelvare@suse.de>
---
 drivers/net/can/rockchip/Kconfig |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- linux-6.12-rc4.orig/drivers/net/can/rockchip/Kconfig
+++ linux-6.12-rc4/drivers/net/can/rockchip/Kconfig
@@ -2,7 +2,7 @@
 
 config CAN_ROCKCHIP_CANFD
 	tristate "Rockchip CAN-FD controller"
-	depends on OF || COMPILE_TEST
+	depends on OF
 	select CAN_RX_OFFLOAD
 	help
 	  Say Y here if you want to use CAN-FD controller found on


-- 
Jean Delvare
SUSE L3 Support

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Vincent MAILHOL]

Hi Jean,

+cc: Miguel Ojeda <ojeda@kernel.org>
+cc: Masahiro Yamada <masahiroy@kernel.org>
+cc: Andrew Morton <akpm@linux-foundation.org>

On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
> Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> can be enabled on all architectures. Therefore depending on
> COMPILE_TEST as an alternative is no longer needed.

I understand the motivation behind this patch, but for me, as a
maintainer, it becomes more work when I want to do a compile test.
Before I would have needed to only select COMPILE_TEST but now, I
would need to remember to also select OF for that driver to appear in
the menuconfig.

Well, I am not strongly against this simplification, but, wouldn't it
be good to make COMPILE_TEST automatically select OF then? Looking at
the description of COMPILE_TEST, I read:

 If you are a developer and want to build everything available, say Y here.

So having COMPILE_TEST automatically select OF looks sane to me as it
goes in the direction of "building everything". If this makes sense, I
can send a patch for this. Thoughts?

> Signed-off-by: Jean Delvare <jdelvare@suse.de>
> ---
>  drivers/net/can/rockchip/Kconfig |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> --- linux-6.12-rc4.orig/drivers/net/can/rockchip/Kconfig
> +++ linux-6.12-rc4/drivers/net/can/rockchip/Kconfig
> @@ -2,7 +2,7 @@
>
>  config CAN_ROCKCHIP_CANFD
>         tristate "Rockchip CAN-FD controller"
> -       depends on OF || COMPILE_TEST
> +       depends on OF
>         select CAN_RX_OFFLOAD
>         help
>           Say Y here if you want to use CAN-FD controller found on
>
>
> --
> Jean Delvare
> SUSE L3 Support
>

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Jean Delvare]

Hi Vincent,

On Tue, 22 Oct 2024 22:22:05 +0900, Vincent MAILHOL wrote:
> On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
> > Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> > can be enabled on all architectures. Therefore depending on
> > COMPILE_TEST as an alternative is no longer needed.  
> 
> I understand the motivation behind this patch, but for me, as a
> maintainer, it becomes more work when I want to do a compile test.
> Before I would have needed to only select COMPILE_TEST but now, I
> would need to remember to also select OF for that driver to appear in
> the menuconfig.

Not sure what your typical process looks like, but if you need to
test-build drivers on a regular basis, why don't you have
CONFIG_COMPILE_TEST always enabled? Switching back and forth must be a
hassle.

About CONFIG_OF, well you probably want it enabled regardless. For one
thing, there are at least 940 drivers in the kernel tree which depend on
OF, so you are already missing a lot of test build coverage if you have
it disabled.

For another, if test-building a driver which would normally depend on
OF, but with OF disabled (as you do for rockchip_canfd at the moment,
as I understand it), then you may not be building the full driver. Part
of the code could be stripped out at build time because the compiler
notices that it is unreachable. Apparently this isn't the case of the
rockchip_canfd driver though.

> Well, I am not strongly against this simplification, but, wouldn't it
> be good to make COMPILE_TEST automatically select OF then? Looking at
> the description of COMPILE_TEST, I read:
> 
>  If you are a developer and want to build everything available, say Y here.

As a side note, I don't think this is a proper description of how to
use this option. If you "want to build everything available" then what
you want to do is "make allmodconfig", not manually enabling all
drivers in "make menuconfig". The purpose of manually selecting
COMPILE_TEST would presumably be to test-build a specific driver or set
of drivers.

> So having COMPILE_TEST automatically select OF looks sane to me as it
> goes in the direction of "building everything". If this makes sense, I
> can send a patch for this. Thoughts?

My initial reaction was that this would be an artificial dependency,
which seems wrong.

However, considering that the purpose of COMPILE_TEST is essentially to
let the user build drivers from other architectures, and that enabling
OF on an architecture which doesn't have it enabled by default
essentially lets you select a whole lot of drivers from other
architectures... I must admit that both options indeed share a
purpose.

But I still have a concern. Some drivers can be built with or without
OF (they support both OF and non-OF devices). If someone wants to
test-build such a driver on a different architecture (and therefore
must enable COMPILE_TEST) and wants to ensure that it builds fine with
and without OF, then if COMPILE_TEST selects OF as you suggested, they
would no longer be able to test the CONFIG_OF=n case.

So I'm afraid we can't actually do that.

-- 
Jean Delvare
SUSE L3 Support

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Vincent MAILHOL]

Hi Jean,

On Wed. 23 Oct. 2024, 02:04, Jean Delvare <jdelvare@suse.de> wrote:
> Hi Vincent,
>
> On Tue, 22 Oct 2024 22:22:05 +0900, Vincent MAILHOL wrote:
> > On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
> > > Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> > > can be enabled on all architectures. Therefore depending on
> > > COMPILE_TEST as an alternative is no longer needed.
> >
> > I understand the motivation behind this patch, but for me, as a
> > maintainer, it becomes more work when I want to do a compile test.
> > Before I would have needed to only select COMPILE_TEST but now, I
> > would need to remember to also select OF for that driver to appear in
> > the menuconfig.
>
> Not sure what your typical process looks like, but if you need to
> test-build drivers on a regular basis, why don't you have
> CONFIG_COMPILE_TEST always enabled? Switching back and forth must be a
> hassle.

The point is that I do not need to test build all drivers on a regular
basis. The CAN subsystem is a small tree with not so many activities.
And this OF thing is a pitfall (in which I have fallen into it in the
past…)

> About CONFIG_OF, well you probably want it enabled regardless. For one
> thing, there are at least 940 drivers in the kernel tree which depend on
> OF, so you are already missing a lot of test build coverage if you have
> it disabled.

The CAN subtree for which I care has only four drivers with OF, only
two of which are not automatically selected by COMPILE_TEST:

  $ git grep "depends.* OF" drivers/net/can/
  drivers/net/can/Kconfig: depends on OF || COLDFIRE || COMPILE_TEST
  drivers/net/can/Kconfig: depends on OF && HAS_DMA && HAS_IOMEM
  drivers/net/can/ctucanfd/Kconfig: depends on HAS_IOMEM && OF
  drivers/net/can/rockchip/Kconfig: depends on OF || COMPILE_TEST

> For another, if test-building a driver which would normally depend on
> OF, but with OF disabled (as you do for rockchip_canfd at the moment,
> as I understand it), then you may not be building the full driver. Part
> of the code could be stripped out at build time because the compiler
> notices that it is unreachable. Apparently this isn't the case of the
> rockchip_canfd driver though.
>
> > Well, I am not strongly against this simplification, but, wouldn't it
> > be good to make COMPILE_TEST automatically select OF then? Looking at
> > the description of COMPILE_TEST, I read:
> >
> >  If you are a developer and want to build everything available, say Y here.
>
> As a side note, I don't think this is a proper description of how to
> use this option. If you "want to build everything available" then what
> you want to do is "make allmodconfig", not manually enabling all
> drivers in "make menuconfig". The purpose of manually selecting
> COMPILE_TEST would presumably be to test-build a specific driver or set
> of drivers.

Ack. It would be helpful to rephrase that description.

> > So having COMPILE_TEST automatically select OF looks sane to me as it
> > goes in the direction of "building everything". If this makes sense, I
> > can send a patch for this. Thoughts?
>
> My initial reaction was that this would be an artificial dependency,
> which seems wrong.
>
> However, considering that the purpose of COMPILE_TEST is essentially to
> let the user build drivers from other architectures, and that enabling
> OF on an architecture which doesn't have it enabled by default
> essentially lets you select a whole lot of drivers from other
> architectures... I must admit that both options indeed share a
> purpose.
>
> But I still have a concern. Some drivers can be built with or without
> OF (they support both OF and non-OF devices). If someone wants to
> test-build such a driver on a different architecture (and therefore
> must enable COMPILE_TEST) and wants to ensure that it builds fine with
> and without OF, then if COMPILE_TEST selects OF as you suggested, they
> would no longer be able to test the CONFIG_OF=n case.
>
> So I'm afraid we can't actually do that.

Ack, this concern seems valid.

Thanks for your time and your good explanations.
I guess it is now my duty to not forget anymore to also enable OF when
doing compile tests. With this said:

Reviewed-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>


Yours sincerely,
Vincent Mailhol

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Geert Uytterhoeven]

 	Hi Jean, Vincent,

On Tue, 22 Oct 2024, Vincent MAILHOL wrote:
> On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
>> Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
>> can be enabled on all architectures. Therefore depending on
>> COMPILE_TEST as an alternative is no longer needed.
>
> I understand the motivation behind this patch, but for me, as a
> maintainer, it becomes more work when I want to do a compile test.
> Before I would have needed to only select COMPILE_TEST but now, I
> would need to remember to also select OF for that driver to appear in
> the menuconfig.

IMHO these are two different things: to get a working driver, you need
to enable OF; to do (may be limited, i.e. may not give a working driver)
compile-testing, you need to enable COMPILE_TEST.

So I think commit 51e102ec23b25e6c ("can: rockchip_canfd: Drop obsolete
dependency on COMPILE_TEST") should be reverted.

> Well, I am not strongly against this simplification, but, wouldn't it
> be good to make COMPILE_TEST automatically select OF then? Looking at
> the description of COMPILE_TEST, I read:
>
> If you are a developer and want to build everything available, say Y here.
>
> So having COMPILE_TEST automatically select OF looks sane to me as it
> goes in the direction of "building everything". If this makes sense, I
> can send a patch for this. Thoughts?

Please don't do that! Merely enabling COMPILE_TEST should not enable
any additional code in the kernel.

>> --- linux-6.12-rc4.orig/drivers/net/can/rockchip/Kconfig
>> +++ linux-6.12-rc4/drivers/net/can/rockchip/Kconfig
>> @@ -2,7 +2,7 @@
>>
>>  config CAN_ROCKCHIP_CANFD
>>         tristate "Rockchip CAN-FD controller"
>> -       depends on OF || COMPILE_TEST
>> +       depends on OF
>>         select CAN_RX_OFFLOAD
>>         help
>>           Say Y here if you want to use CAN-FD controller found on
>>

Gr{oetje,eeting}s,

 						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
 							    -- Linus Torvalds

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Jean Delvare]

Hi Geert,

On Tue, 12 Nov 2024 12:15:06 +0100 (CET), Geert Uytterhoeven wrote:
>  	Hi Jean, Vincent,
> 
> On Tue, 22 Oct 2024, Vincent MAILHOL wrote:
> > On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:  
> >> Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> >> can be enabled on all architectures. Therefore depending on
> >> COMPILE_TEST as an alternative is no longer needed.  
> >
> > I understand the motivation behind this patch, but for me, as a
> > maintainer, it becomes more work when I want to do a compile test.
> > Before I would have needed to only select COMPILE_TEST but now, I
> > would need to remember to also select OF for that driver to appear in
> > the menuconfig.  
> 
> IMHO these are two different things: to get a working driver, you need
> to enable OF;

True.

>(...) to do (may be limited, i.e. may not give a working driver)
> compile-testing, you need to enable COMPILE_TEST.

No, you don't *need* it. Enabling COMPILE_TEST is (or was) one way to
do compile-testing, but it was not the only way. Which is the reason
why it was dropped.

Your reasoning would hold only if building a limited, maybe not-working
driver, was a purpose in itself. I personally can't see any value in
doing this.

-- 
Jean Delvare
SUSE L3 Support

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Geert Uytterhoeven]

Hi Jean,

On Thu, Nov 21, 2024 at 2:50 PM Jean Delvare <jdelvare@suse.de> wrote:
> On Tue, 12 Nov 2024 12:15:06 +0100 (CET), Geert Uytterhoeven wrote:
> > On Tue, 22 Oct 2024, Vincent MAILHOL wrote:
> > > On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
> > >> Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> > >> can be enabled on all architectures. Therefore depending on
> > >> COMPILE_TEST as an alternative is no longer needed.
> > >
> > > I understand the motivation behind this patch, but for me, as a
> > > maintainer, it becomes more work when I want to do a compile test.
> > > Before I would have needed to only select COMPILE_TEST but now, I
> > > would need to remember to also select OF for that driver to appear in
> > > the menuconfig.
> >
> > IMHO these are two different things: to get a working driver, you need
> > to enable OF;
>
> True.
>
> >(...) to do (may be limited, i.e. may not give a working driver)
> > compile-testing, you need to enable COMPILE_TEST.
>
> No, you don't *need* it. Enabling COMPILE_TEST is (or was) one way to
> do compile-testing, but it was not the only way. Which is the reason
> why it was dropped.

You could still do it the other way, by enabling CONFIG_OF.

> Your reasoning would hold only if building a limited, maybe not-working
> driver, was a purpose in itself. I personally can't see any value in
> doing this.

It may help detecting more configuration issues using randconfig.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Geert Uytterhoeven]

On Fri, Nov 22, 2024 at 3:33 PM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> On Thu, Nov 21, 2024 at 2:50 PM Jean Delvare <jdelvare@suse.de> wrote:
> > On Tue, 12 Nov 2024 12:15:06 +0100 (CET), Geert Uytterhoeven wrote:
> > > On Tue, 22 Oct 2024, Vincent MAILHOL wrote:
> > > > On Tue. 22 Oct. 2024 at 20:06, Jean Delvare <jdelvare@suse.de> wrote:
> > > >> Since commit 0166dc11be91 ("of: make CONFIG_OF user selectable"), OF
> > > >> can be enabled on all architectures. Therefore depending on
> > > >> COMPILE_TEST as an alternative is no longer needed.
> > > >
> > > > I understand the motivation behind this patch, but for me, as a
> > > > maintainer, it becomes more work when I want to do a compile test.
> > > > Before I would have needed to only select COMPILE_TEST but now, I
> > > > would need to remember to also select OF for that driver to appear in
> > > > the menuconfig.
> > >
> > > IMHO these are two different things: to get a working driver, you need
> > > to enable OF;
> >
> > True.
> >
> > >(...) to do (may be limited, i.e. may not give a working driver)
> > > compile-testing, you need to enable COMPILE_TEST.
> >
> > No, you don't *need* it. Enabling COMPILE_TEST is (or was) one way to
> > do compile-testing, but it was not the only way. Which is the reason
> > why it was dropped.
>
> You could still do it the other way, by enabling CONFIG_OF.

And you still need to enable COMPILE_TEST, to get around the
ARCH_ROCKCHIP dependency...

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Jean Delvare]

Hi Geert,

On Fri, 22 Nov 2024 15:33:21 +0100, Geert Uytterhoeven wrote:
> On Thu, Nov 21, 2024 at 2:50 PM Jean Delvare <jdelvare@suse.de> wrote:
> > Your reasoning would hold only if building a limited, maybe
> > not-working driver, was a purpose in itself. I personally can't see
> > any value in doing this.  
> 
> It may help detecting more configuration issues using randconfig.

Maybe I would agree if build time, energy and engineering time were
free and unlimited resources.

In our world though, the time spent by a test build farm to build a
known-crippled driver would be better used to start the next test build
earlier. This makes for better quality coverage. Or the energy used to
do that could also be saved altogether.

And this suboptimal use of build time and energy is only the best case.
The worst case is if the randconfig case causes a "false positive" build
error or warning. This will generate a report, which will be sent to
mailing lists, read by human beings and investigated, wasting
engineering time.

I'm not making this up, this is in fact exactly what happened and led
me to investigate a first "depends on OF || COMPILE_TEST" case a few
years ago. I found out that the build warning would only possibly
trigger if COMPILE_TEST was selected and OF was not. It would never
trigger with any configuration options combination not including
COMPILE_TEST. Fixing it would require adding #ifdefs and using macros.
But why should we make the code more complex, less readable, to prevent
a warning which can never happen in a non-test scenario? And more
importantly, why should we allocate engineering time to look into such
issues, when there are so many more useful tasks to work on?

-- 
Jean Delvare
SUSE L3 Support