Re: [PATCH] MIPS: update tlb even if pte entry has no change

From: Mike Rapoport
Date: Thu May 14 2020 - 05:34:21 EST


On Thu, May 14, 2020 at 10:17:57AM +0800, Bibo Mao wrote:
> From: bibo mao <maobibo@xxxxxxxxxxx>
>
> If there are two threads reading the same memory and tlb miss happens,
> one thread fills pte entry, the other reads new pte value during page fault
> handling. PTE value may be updated before page faul, so the process need
> need update tlb still.
>
> Also this patch define flush_tlb_fix_spurious_fault as empty, since it not
> necessary to flush the page for all CPUs

I'm not familiar enough with MIPS TLB workings, but it seems that adding
a MIPS version of ptep_set_access_flags() and relaxing
flush_tlb_fix_spurious_fault() are two separate patches.

> Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
> ---
> arch/mips/include/asm/pgtable.h | 20 ++++++++++++++++++++
> 1 file changed, 20 insertions(+)
>
> diff --git a/arch/mips/include/asm/pgtable.h b/arch/mips/include/asm/pgtable.h
> index aab0ec1..d0a4940 100644
> --- a/arch/mips/include/asm/pgtable.h
> +++ b/arch/mips/include/asm/pgtable.h
> @@ -635,6 +635,26 @@ static inline pmd_t pmd_mknotpresent(pmd_t pmd)
> return pmd;
> }
>
> +#define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
> +
> +#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
> +int ptep_set_access_flags(struct vm_area_struct *vma,
> + unsigned long address, pte_t *ptep,
> + pte_t entry, int dirty)
> +{
> + int changed = !pte_same(*ptep, entry);
> +
> + if (changed)
> + set_pte_at(vma->vm_mm, address, ptep, entry);
> + else
> + /* update tlb with latest pte entry still, tlb entry is old
> + * since there is page fault
> + */
> + update_mmu_cache(vma, address, ptep);
> +
> + return changed;
> +}
> +
> /*
> * The generic version pmdp_huge_get_and_clear uses a version of pmd_clear() with a
> * different prototype.
> --
> 1.8.3.1
>

--
Sincerely yours,
Mike.