Re: [PATCH] hwpoison: fix uninitialized warning

[Wu Fengguang <fengguang.wu@intel.com>]

On Wed, Sep 16, 2009 at 08:51:06AM +0800, Hugh Dickins wrote:
> On Wed, 16 Sep 2009, Wu Fengguang wrote:
> > On Wed, Sep 16, 2009 at 05:19:07AM +0800, Hugh Dickins wrote:
> > > Fix mmotm build warning, presumably also in linux-next:
> > > mm/memory.c: In function `do_swap_page':
> > > mm/memory.c:2498: warning: `pte' may be used uninitialized in this function
> > > 
> > > Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
> > > ---
> > > I've only noticed this warning on one machine, the powerpc: certainly it
> > > needs CONFIG_MIGRATION or CONFIG_MEMORY_FAILURE to see it, but I thought
> > > I had one of those set on other machines - just musing in case it's being
> > > masked elsewhere by some other bug...
> 
> > The lines was introduced in this patch:
> > 
> >         entry = pte_to_swp_entry(orig_pte);
> > -       if (is_migration_entry(entry)) {
> > -               migration_entry_wait(mm, pmd, address);
> > +       if (unlikely(non_swap_entry(entry))) {
> > +               if (is_migration_entry(entry)) {
> > +                       migration_entry_wait(mm, pmd, address);
> > +               } else if (is_hwpoison_entry(entry)) {
> > +                       ret = VM_FAULT_HWPOISON;
> > +               } else {
> > +                       print_bad_pte(vma, address, pte, NULL);
> > +                       ret = VM_FAULT_OOM;
> > +               }
> >                 goto out;
> >         }
> > 
> > Given that currently there are only two types of non swap entries:
> > migration/hwpoison, the last 'else' block is in fact dead code..
> 
> Ah, yes, I think it is dead code on x86 (32 and 64), where the
> swp_entry_t is well packed.  But not dead code on ppc64, which has
> 
> #define __swp_type(entry)	(((entry).val >> 1) & 0x3f)
> 
> which is allowing swap types up to 63, when in fact the highest
> we use is 31: that leaves space for 32 more non_swap_entry types.
> 
> So the compiler was absolutely right to complain about the
> uninitialized variable on ppc64, but not on x86.  It's a little
> surprising that ppc64 allows 64 swap types, but nothing wrong.

Ah I know. It seems that gcc is smart enough to remove that dead code
and hence the warning message in x86 :)

Thanks,
Fengguang