Re: [PATCH] module: Allow DEFAULT_SYMBOL_NAMESPACE be set after export.h included

[David Laight <david.laight.linux@gmail.com>]

On Mon, 30 Dec 2024 10:42:55 +0100
Uwe Kleine-König <u.kleine-koenig@baylibre.com> wrote:

> On Sun, Dec 29, 2024 at 12:59:36AM +0000, David Laight wrote:
> > On Sat, 28 Dec 2024 15:29:24 -0800
> > Linus Torvalds <torvalds@linux-foundation.org> wrote:
> >   
> > > On Sat, 28 Dec 2024 at 10:43, David Laight <david.laight.linux@gmail.com> wrote:  
> > > >
> > > > Instead just default DEFAULT_SYMBOL_NAMESPACE to "" and remove the
> > > > extra _EXPORT_SYMBOL() wrapper.
> > > >
> > > > This lets DEFAULT_SYMBOL_NAMESPACE be defined after export.h is included.    
> > > 
> > > Grr. This is horribly ugly.  
> > 
> > I thought it was a neater 'ugly' than the current definitions in export.h
> >   
> > > I think the i2c code should just be fixed to use the proper "define
> > > namespace early".  
> > 
> > The i2c changes were needed because I found the code wouldn't compile.
> > It is pretty easy mistake to make and will happen again.  
> 
> There is
> https://lore.kernel.org/linux-i2c/20241203173640.1648939-2-u.kleine-koenig@baylibre.com
> that moves the DEFAULT_SYMBOL_NAMESPACE above the #include block for the
> i2c driver. Though it seems I missed ...master.c. (I'll address that.)
> 
> > and does - I missed drivers/pwm/pwm-lpss.c drivers/hwmon/nct6775-core.c
> > and drivers/pwm/pwm-lpss.c  
> 
> drivers/pwm/pwm-lpss.c is addressed by
> https://lore.kernel.org/linux-pwm/cover.1733245406.git.ukleinek@kernel.org
> 
> the hwmon driver is addressed by
> https://lore.kernel.org/linux-hwmon/20241203173149.1648456-2-u.kleine-koenig@baylibre.com
> (and applied in next)
> 
> There is also drivers/gpio/gpio-idio-16.c (which I guess you intended to
> list instead of the pwm driver twice), which I sent a patch for at
> https://lore.kernel.org/linux-gpio/20241203172631.1647792-2-u.kleine-koenig@baylibre.com
> (also already applied in next).


With all those applied it is probably worth applying my change to export.h
(which is all I really wanted to do - until the build failures.)

diff --git a/include/linux/export.h b/include/linux/export.h
index 2633df4d31e6..6cea1c3982cd 100644
--- a/include/linux/export.h
+++ b/include/linux/export.h
@@ -59,14 +59,12 @@

 #endif

-#ifdef DEFAULT_SYMBOL_NAMESPACE
-#define _EXPORT_SYMBOL(sym, license)   __EXPORT_SYMBOL(sym, license, DEFAULT_SYMBOL_NAMESPACE)
-#else
-#define _EXPORT_SYMBOL(sym, license)   __EXPORT_SYMBOL(sym, license, "")
+#ifndef DEFAULT_SYMBOL_NAMESPACE
+#define DEFAULT_SYMBOL_NAMESPACE ""
 #endif

-#define EXPORT_SYMBOL(sym)             _EXPORT_SYMBOL(sym, "")
-#define EXPORT_SYMBOL_GPL(sym)         _EXPORT_SYMBOL(sym, "GPL")
+#define EXPORT_SYMBOL(sym)             __EXPORT_SYMBOL(sym, "", DEFAULT_SYMBOL_NAMESPACE)
+#define EXPORT_SYMBOL_GPL(sym)         __EXPORT_SYMBOL(sym, "GPL", DEFAULT_SYMBOL_NAMESPACE)
 #define EXPORT_SYMBOL_NS(sym, ns)      __EXPORT_SYMBOL(sym, "", ns)
 #define EXPORT_SYMBOL_NS_GPL(sym, ns)  __EXPORT_SYMBOL(sym, "GPL", ns)

It would be 'nice' to get that into 6.13 (along with the other changes that
remove __stringify()) - but it is getting late in the rc cycle now.

Whether it is better to define DEFAULT_SYMBOL_NAMESPACE at the top of the
file or after the includes is another matter.
If the file is a module then it really makes sense to put the definition
with all the other module-related definitions.
The fact that it needs a #undef is annoying, but not the end of the world.

	David