Re: [PATCH v2] export: fix string handling of namespace in EXPORT_SYMBOL_NS

[Greg Kroah-Hartman]

On Thu, Apr 28, 2022 at 12:11:13AM +0900, Masahiro Yamada wrote:
> On Wed, Apr 27, 2022 at 11:50 PM Greg Kroah-Hartman
> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, Apr 27, 2022 at 11:29:19PM +0900, Masahiro Yamada wrote:
> > > On Wed, Apr 27, 2022 at 6:06 PM Greg Kroah-Hartman
> > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > Commit c3a6cf19e695 ("export: avoid code duplication in
> > > > include/linux/export.h") broke the ability for a defined string to be
> > > > used as a namespace value.
> > >
> > > In hindsight, this was a bad idea.
> > >
> > >
> > > EXPORT_SYMBOL_NS_GPL(dw_spi_resume_host, "SPI_DW_CORE")
> > >
> > >    is much much better than:
> > >
> > > EXPORT_SYMBOL_NS_GPL(dw_spi_resume_host, SPI_DW_CORE)
> >
> > I agree, but it's really not that big of a deal.  We could change it if
> > you want.
> >
> > > ccflags-y += -DDEFAULT_SYMBOL_NAMESPACE=USB_STORAGE
> > >
> > > is also a bad idea.
> >
> > That's not such a bad idea as it lets you set a namespace for a
> > directory and below easily.  What would you want to use instead?
> 
> I like "explicit" better than "implicit".
> 
> With  EXPORT_SYMBOL_GPL(usb_stor_resume);
> you need to check Makefile to know whether it is global
> or is in a specific namespace.
> 
> EXPORT_SYMBOL_NS_GPL(usb_stor_resume, "USB_STORAGE");
> will clarify the namespace at a glance.

I agree, but currently I use the makefile change in a big out-of-tree
project to do "all exported symbols in this directory and lower are now
in this namespace".  That's a huge help and makes for a 2 line change
vs. a hundreds-of-lines change.

> > > When you look at EXPORT_SYMBOL_GPL() in C files, you will not be
> > > aware of the presence of the namespace.
> >
> > It's easy to tell when things do not link properly :)
> 
> Right, modpost will tell you that once you compile it.

And that's the best thing, 'make nsdeps' will then even generate a patch
to fix things up.

> > > Anyway, it is presumably too late to fix it.
> >
> > Not really, the number of in-kernel users are still small and can be
> > changed if you like.  External users can update when they hit the change
> > as well, not a big deal.
> >
> > Other than using a string for the namespace definition, what would you
> > like to see done differently?
> 
> Nothing else.
> 
> (but I am not a big fan of the module namespace itself.)

I think more subsystems need to start using them as it instantly shows
where some drivers are doing things that maybe they shouldn't be doing.
It was insightful when the dma-buf code moved to a module namespace as
it found places the maintainers were not even aware of.

Anyway, it's on my long-term todo list...

thanks,

greg k-h