Re: [RFC] Introduce __ARCH_WANT_SYS_SYSFS

[Sam Ravnborg <sam-uyr5N9Q2VtJg9hUCZPvPmw@public.gmane.org>]

On Wed, Apr 23, 2008 at 03:36:23PM +0100, Will Newton wrote:
> On Tue, Apr 22, 2008 at 4:38 PM, Kyle McMartin <kyle-pfcGkIkfWfAsA/PxXw9srA@public.gmane.org> wrote:
> > On Tue, Apr 22, 2008 at 04:34:03PM +0100, Will Newton wrote:
> >  > On Tue, Apr 22, 2008 at 4:24 PM, Kyle McMartin <kyle-pfcGkIkfWfAsA/PxXw9srA@public.gmane.org> wrote:
> >  > > On Tue, Apr 22, 2008 at 04:16:17PM +0100, Will Newton wrote:
> >  > >  > It can be done with Kconfig. Whether it should be or not depends on
> >  > >  > your point of view, hence RFC. Currently __ARCH_WANT macros is the way
> >  > >  > syscalls are enabled and disabled across architectures. If there's
> >  > >  > consensus that it should be done via Kconfig that could certainly be
> >  > >  > implmented, but that's a different patch.
> >  > >  >
> >  > >
> >  > >  It's currently done in unistd.h for hysterical raisins. All new
> >  > >  conditional syscalls have been done in Kconfig.
> >  >
> >  > Do you have a syscall in mind that does this in the correct way?
> >  >
> >
> >  epoll is the most immediate example at hand, since it touches
> >  fs/compat.c similarly to how you would be touching fs/filesystem.c, and
> >  touches kernel/sys_ni.c and init/Kconfig in much the same way.
> 
> I implemented the approach you suggested - Kconfig symbol and
> cond_syscall definition. I think I actually like the previous approach
> better:
> 
> 1. The arch Kconfig files are quite non-uniform compared to unistd.h
> so the definitions wind up at different places in the file which is a
> bit messy.
> 2. Changes to Kconfig may cause churn in defconfigs perhaps?
> 3. There is more churn in arch Kconfig than unistd.h so getting a
> cross arch patch applied is likely to be more difficult.
> 4. The patch is about 4 times as many lines.
> 
> What do you think?

If we go the Kconfig route we should use the
HAVE_ semantic as expressed by the patch below.
It is preferable to defining a config symbol for each
arch.
This also address your point 1) and 4)
2) is not an issue since the symbols are not visible.

	Sam

Note: cut'n'pasted...

diff --git a/arch/Kconfig b/arch/Kconfig
index 694c9af..759bd5b 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -36,3 +36,9 @@ config HAVE_KPROBES

 config HAVE_KRETPROBES
        def_bool n
+
+# syscall symbols.
+# archs shall select the SYMBOL if they
+# implment this syscall
+config HAVE_SYS_SYSFS
+       defbool n
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 87a693c..487310f 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -24,6 +24,8 @@ config X86
        select HAVE_KRETPROBES
        select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64)
        select HAVE_ARCH_KGDB
+       # kconfig selectable syscalls
+       select HAVE_SYS_SYSFS