On Tue, May 03, 2022 at 09:13:20AM +0800, Tong Tiangen wrote:
> 在 2022/4/29 14:18, Anshuman Khandual 写道:
> > On 4/26/22 13:40, Tong Tiangen wrote:
> > > @@ -884,7 +928,11 @@ static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
> > > static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
> > > unsigned long address, pmd_t *pmdp)
> > > {
> > > - return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
> > > + pmd_t pmd = pte_pmd(__pte(xchg_relaxed(&pte_val(*(pte_t *)pmdp), 0)));
> >
> > Why there is any change here ? Why not just the following instead, like what you did
> > in ptep_get_and_clear() above. The page table entry return value here just needs to
> > be preserved for subsequent page table check helpers.
> >
> > pmd_t pmd = pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
>
> The purpose of what I do here is to avoid doing page_table_check_pte_clear()
> in ptep_get_and_clear(). there is no functional change here.
I'm fine with using xchg here but I'd go for pmd_val directly, no need
for conversion to pte_t:
pmd_t pmd = __pmd(xchg_relaxed(&pmd_val(*pmdp), 0));
--
Catalin
