Re: [PATCH] Loongarch:Make pte/pmd_modify can set _PAGE_MODIFIED

[Tianyang Zhang]

在 2025/11/6 下午3:07, Bibo Mao 写道:
> On 2025/11/4 下午3:30, Tianyang Zhang wrote:
> > In the current pte_modify operation, _PAGE_DIRTY might be cleared. Since
> > the hardware-page-walk does not have a predefined _PAGE_MODIFIED flag,
> > this could lead to loss of valid data in certain scenarios.
> >
> > The new modification involves checking whether the original PTE has the
> > _PAGE_DIRTY flag. If it exists, the _PAGE_MODIFIED bit is set, ensuring
> > that the pte_dirty interface can return accurate information.
> >
> > Co-developed-by: Liupu Wang <wangliupu@xxxxxxxxxxx>
> > Signed-off-by: Liupu Wang <wangliupu@xxxxxxxxxxx>
> > Signed-off-by: Tianyang Zhang <zhangtianyang@xxxxxxxxxxx>
> > ---
> >   arch/loongarch/include/asm/pgtable.h | 17 +++++++++++++----
> >   1 file changed, 13 insertions(+), 4 deletions(-)
> >
> > diff --git a/arch/loongarch/include/asm/pgtable.h 
> > b/arch/loongarch/include/asm/pgtable.h
> > index bd128696e96d..106abfa5183b 100644
> > --- a/arch/loongarch/include/asm/pgtable.h
> > +++ b/arch/loongarch/include/asm/pgtable.h
> > @@ -424,8 +424,13 @@ static inline unsigned long 
> > pte_accessible(struct mm_struct *mm, pte_t a)
> >     static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
> >   {
> > -    return __pte((pte_val(pte) & _PAGE_CHG_MASK) |
> > -             (pgprot_val(newprot) & ~_PAGE_CHG_MASK));
> > +    unsigned long val = (pte_val(pte) & _PAGE_CHG_MASK) |
> > +             (pgprot_val(newprot) & ~_PAGE_CHG_MASK);
> > +
> > +    if (pte_val(pte) & _PAGE_DIRTY)
> > +        val |= _PAGE_MODIFIED;
> Since ptep_get_and_clear() is not atomic operation on LoongArch like 
> other architectures, considering this scenery with HW PTW enabled:
>     CPU 0:                         CPU1
>                              old_pte = ptep_modify_prot_start();
>                                   old_pte = ptep_get(ptep);
>
>     write(buf);
> *HW will set _PAGE_DIRTY bit*
>                                   pte_clear(mm, address, ptep);
> ^^^^^^^^^^ For CPU1, bit _PAGE_DIRTY is no set in old_pte, _PAGE_DIRTY 
> will be lost also. ^^^^^^^^^^^
>                              pte = pte_modify(old_pte,)
>                              ptep_modify_prot_commit(.., pte)

There does appear to be an issue here. It seems we should use 
`__HAVE_ARCH_PTEP_GET_AND_CLEAR` and implement ptep_get_and_clear via 
`atomic_xchg`.

However, I believe this change should be submitted in a new patch.

Thanks

Tianyang

> Regards
> Bibo Mao
> > +
> > +    return __pte(val);
> >   }
> >     extern void __update_tlb(struct vm_area_struct *vma,
> > @@ -547,9 +552,13 @@ static inline struct page *pmd_page(pmd_t pmd)
> >     static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
> >   {
> > -    pmd_val(pmd) = (pmd_val(pmd) & _HPAGE_CHG_MASK) |
> > +    unsigned long val = (pmd_val(pmd) & _HPAGE_CHG_MASK) |
> >                   (pgprot_val(newprot) & ~_HPAGE_CHG_MASK);
> > -    return pmd;
> > +
> > +    if (pmd_val(pmd) & _PAGE_DIRTY)
> > +        val |= _PAGE_MODIFIED;
> > +
> > +    return __pmd(val);
> >   }
> >     static inline pmd_t pmd_mkinvalid(pmd_t pmd)