Re: [PATCH 09/30] x86_64: ifdef out struct thread_struct::ip

[Alexey Dobriyan <adobriyan@gmail.com>]

On Fri, Apr 10, 2009 at 11:19:31AM +0200, Ingo Molnar wrote:
> 
> * Matt Helsley <matthltc@us.ibm.com> wrote:
> 
> > On Fri, Apr 10, 2009 at 06:35:22AM +0400, Alexey Dobriyan wrote:
> > > struct thread_struct::ip isn't used on x86_64, struct pt_regs::ip is used
> > > instead. 
> > > 
> > > kgdb should be reading 0, but I can't check it.
> > > 
> > > Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
> > > ---
> > > 
> > >  arch/x86/include/asm/processor.h |    2 ++
> > >  arch/x86/kernel/kgdb.c           |    2 +-
> > >  2 files changed, 3 insertions(+), 1 deletion(-)
> > > 
> > > --- a/arch/x86/include/asm/processor.h
> > > +++ b/arch/x86/include/asm/processor.h
> > > @@ -421,7 +421,9 @@ struct thread_struct {
> > >  	unsigned short		fsindex;
> > >  	unsigned short		gsindex;
> > >  #endif
> > > +#ifdef CONFIG_X86_32
> > >  	unsigned long		ip;
> > > +#endif
> > >  #ifdef CONFIG_X86_64
> > >  	unsigned long		fs;
> > >  #endif
> > 
> > Do these make struct thread_struct behave better in cachelines 
> > (smaller, less aliasing)? Can we really fit more in the slab du 
> > jour?
> > 
> > Otherwise it seems like we're littering these structs with #ifdefs 
> > and not really saving anything. [...]
> 
> Removing fields always saves memory (even if it does not show up 
> currently due to allocators cache-aligning sizes).
> 
> But the #ifdef ugliness is a real worry.

You should have thought about it when i386/x86_64 unification was
introduced.

pagefault code was full of ifdefs (it's less now), and this is trivial
ifdef in a header.

> > [...] If these #ifdefs don't save any 
> > space why not just put in a comment:
> > 
> > >  	unsigned long		ip; /* Used only on i386 */
> 
> Yes.
> 
> > Or maybe even:
> > 
> > 	union {
> > 	  	unsigned long		ip; /* Used only on i386 */
> > 	  	unsigned long		fs; /* Used only on x86_64 */
> > 	};
> 
> Maybe. If this ever gets misunderstood somewhere in platform code we 
> will get ugly failure modes and zero compiler help.