Re: [2.6 PATCH] Exporting mmu_cr4_features again for i386 & x86_64

Re: [2.6 PATCH] Exporting mmu_cr4_features again for i386 & x86_64

[Andi Kleen]

On Thursday 29 June 2006 01:01, Milena Milenkovic wrote:
> Hi all,
> 
> The mmu_cr4_features variable was "unexported" in 2.6.12 kernel
> (A patch submitted by Adrian Bunk, [2.6 patch] unexport mmu_cr4_features).

Normal policy is to unexport symbols that are not used in the core 
kernel.

However I think there is a deprecation policy with prior 
warning that might not have been followed here.

> However, this variable is needed if a loadable kernel module wants to 
> enable
> access to performance counters from user space.

Ok I guess it's a reasonable request (although we will likely
eventually add a standard API for this and then it might 
be removed again) 


-Andi

Re: [2.6 PATCH] Exporting mmu_cr4_features again for i386 & x86_64

[Arjan van de Ven]

> > However, this variable is needed if a loadable kernel module wants to 
> > enable
> > access to performance counters from user space.
> 
> Ok I guess it's a reasonable request (although we will likely
> eventually add a standard API for this and then it might 
> be removed again) 

I kind of disagree that being a reasonable request; there are several
tools (perfmon/oprofile) that do this already without the export, and if
you want another one, adding a very simple API to do just this to the
kernel should not at all be a big deal, just add that to the patch when
submitting the module for inclusion....

Re: [2.6 PATCH] Exporting mmu_cr4_features again for i386 & x86_64

[Andrew Morton]

On Thu, 29 Jun 2006 12:55:36 +0200
Andi Kleen <ak@suse.de> wrote:

> On Thursday 29 June 2006 01:01, Milena Milenkovic wrote:
> > Hi all,
> > 
> > The mmu_cr4_features variable was "unexported" in 2.6.12 kernel
> > (A patch submitted by Adrian Bunk, [2.6 patch] unexport mmu_cr4_features).
> 
> Normal policy is to unexport symbols that are not used in the core 
> kernel.
> 
> However I think there is a deprecation policy with prior 
> warning that might not have been followed here.

We have EXPORT_UNUSED_SYMBOL() and EXPORT_UNUSED_SYMBOL_GPL() as of this
morning.  We can convert unused exports to use those and users will get a
nasty warning at modprobe-time when they use the exported symbols.