On Fri, Mar 01, 2019 at 01:09:51PM -0800, Andrew Morton wrote:
> On Thu, 28 Feb 2019 19:49:03 -0500 Qian Cai <cai@xxxxxx> wrote:
>
> > The commit a00cc7d9dd93 ("mm, x86: add support for PUD-sized transparent
> > hugepages") introduced pudp_huge_get_and_clear_full() but no one uses
> > its return code, so just make it void.
> >
> > mm/huge_memory.c: In function 'zap_huge_pud':
> > mm/huge_memory.c:1982:8: warning: variable 'orig_pud' set but not used
> > [-Wunused-but-set-variable]
> > pud_t orig_pud;
> > ^~~~~~~~
> >
> > ...
> >
> > --- a/include/asm-generic/pgtable.h
> > +++ b/include/asm-generic/pgtable.h
> > @@ -167,11 +167,11 @@ static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
> > #endif
> >
> > #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
> > -static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
> > - unsigned long address, pud_t *pudp,
> > - int full)
> > +static inline void pudp_huge_get_and_clear_full(struct mm_struct *mm,
> > + unsigned long address,
> > + pud_t *pudp, int full)
> > {
> > - return pudp_huge_get_and_clear(mm, address, pudp);
> > + pudp_huge_get_and_clear(mm, address, pudp);
> > }
>
> Not sure this is a good change. Future callers might want that return
> value.
>
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -1979,7 +1979,6 @@ spinlock_t *__pud_trans_huge_lock(pud_t *pud, struct vm_area_struct *vma)
> > int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
> > pud_t *pud, unsigned long addr)
> > {
> > - pud_t orig_pud;
> > spinlock_t *ptl;
> >
> > ptl = __pud_trans_huge_lock(pud, vma);
> > @@ -1991,8 +1990,7 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
> > * pgtable_trans_huge_withdraw after finishing pudp related
> > * operations.
> > */
> > - orig_pud = pudp_huge_get_and_clear_full(tlb->mm, addr, pud,
> > - tlb->fullmm);
> > + pudp_huge_get_and_clear_full(tlb->mm, addr, pud, tlb->fullmm);
>
> In fact this code perhaps should be passing orig_pud into
> pudp_huge_get_and_clear_full(). That could depend on what future
> per-arch implementations of pudp_huge_get_and_clear_full() choose to
> do.
>
> Anyway, I'll await Matthew's feedback.
I'm not sure it's wise to diverge from pmdp_huge_get_and_clear_full()
which does return the orig_pud. I agree we don't currently use it,
so maybe we should just change zap_huge_pud() to not assign the return
value from pudp_huge_get_and_clear_full()?
