Hi Stephen,


On Sat, May 8, 2021 at 10:59 AM Stephen Rothwell [off-list ref] wrote:

Hi Masahiro,

On Thu,  6 May 2021 02:45:15 +0900 Masahiro Yamada [off-list ref] wrote:

quoted

<linux/kconfig.h> is included from all the kernel-space source files,
including C, assembly, linker scripts. It is intended to contain a
minimal set of macros to evaluate CONFIG options.

IF_ENABLED() is an intruder here because (x ? y : z) is C code, which
should not be included from assembly files or linker scripts.

Except it doesn't matter unless IF_ENABLED() is used by one of those.

Yes. This is rather coding policy of <linux/kconfig.h>

quoted

Also, <linux/kconfig.h> is no longer self-contained because NULL is
defined in <linux/stddef.h>.

Again, it doesn't matter unless IF_ENABLED() is used.

If you want to use IF_ENABLED(), you  must understand that
it is internally using NULL and include <linux/stddef.h>.

It is the reason why we should make headers self-contained.

quoted

Move IF_ENABLED() out to <linux/kernel.h> as PTR_IF(). PTF_IF()
takes the general boolean expression instead of a CONFIG option
so that it fits better in <linux/kernel.h>.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
---

Changes in v2:
  - Keep PTF_IF macro in pinctrl-ingenic.c

 drivers/pinctrl/pinctrl-ingenic.c | 2 ++
 include/linux/kconfig.h           | 6 ------
 include/linux/kernel.h            | 2 ++
 3 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/pinctrl/pinctrl-ingenic.c b/drivers/pinctrl/pinctrl-ingenic.c
index 651a36b9dcc0..0ee69f8e20b2 100644
--- a/drivers/pinctrl/pinctrl-ingenic.c
+++ b/drivers/pinctrl/pinctrl-ingenic.c

@@ -3854,6 +3854,8 @@ static int __init ingenic_pinctrl_probe(struct platform_device *pdev)
      return 0;
 }

+#define IF_ENABLED(cfg, ptr) PTR_IF(IS_ENABLED(cfg), (ptr))
+
 static const struct of_device_id ingenic_pinctrl_of_match[] = {
      {
              .compatible = "ingenic,jz4730-pinctrl",

You also need to include linux/kernel.h in
drivers/pinctrl/pinctrl-ingenic.c (for completeness).

<linux/kernel.h> is too widely used, so some other headers
would have already included it eventually.
But, correct, it is better to include it explicitly for completeness.

Also, I don't understand why the use of IF_ENABLED doesn't produce
"defined but not used" warnings (if the function "ptr" is not marked as
__maybe_unused) ...

The function is optimized out by the compiler, not by the pre-processor.
So, -Wunused-but-set-variable does not complain about it.

Also, if there is only one user of IF_ENABLED (and therefore PTR_IF),
why not just put it in that file and save me rebuilding the world again
every day because kernel.h is changed (again).  I guess that is going
to happen just because kconfig.h is being changed and that is also
included by everything :-(

I understand, but this is a one-time fix.

<linux/kconfig.h> is not changed so often.

I believe this will be useful for SET_SYSTEM_SLEEP_PM_OPS cleanups.
So, the macro must be placed in a public header.

Also, is anyone else ever going to use PTR_IF() without having to also
use IS_ENABLED()?

Yes or no. I am not sure about the future of this.

I prefer this in a more general expression,
so IS_REACHABLE or the combination of two CONFIG options
will work.

  PTR_IF(IS_REACHABLE(CONFIG_FOO),  ...)

  PTR_IF( IS_ENABLED(CONFIG_FOO) && IS_ENABLED(CONFIG_BAR), ...)

So, in case it is not obvious, I consider this patch unnecessary churn
(as was probably the patch that introduced IF_ENABLED in the first
place).

As an aside, this should not have been added to the kbuild tree in
linux-next until after -rc1 was released ...

Rather, I was considering sending a pull req before -rc1.

Otherwise, other drivers may start to use it.

--
Cheers,
Stephen Rothwell

--
Best Regards
Masahiro Yamada