diff mbox series

arm64: mm: use ptep_clear() instead of pte_clear() in clear_flush()

Message ID 20230810093241.1181142-1-qi.zheng@linux.dev
State New
Headers
Series arm64: mm: use ptep_clear() instead of pte_clear() in clear_flush() |

Commit Message

Qi Zheng Aug. 10, 2023, 9:32 a.m. UTC 
  From: Qi Zheng <zhengqi.arch@bytedance.com>

In clear_flush(), the original pte may be a present entry, so we should
use ptep_clear() to let page_table_check track the pte clearing operation,
otherwise it may cause false positive in subsequent set_pte_at().

Fixes: 42b2547137f5 ("arm64/mm: enable ARCH_SUPPORTS_PAGE_TABLE_CHECK")
Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
---
 arch/arm64/mm/hugetlbpage.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Qi Zheng Aug. 11, 2023, 2:38 a.m. UTC | #1 
I wrote wrong Kefeng's email address before, correct it now.


On 2023/8/10 17:32, Qi Zheng wrote:
> From: Qi Zheng <zhengqi.arch@bytedance.com>
> 
> In clear_flush(), the original pte may be a present entry, so we should
> use ptep_clear() to let page_table_check track the pte clearing operation,
> otherwise it may cause false positive in subsequent set_pte_at().
> 
> Fixes: 42b2547137f5 ("arm64/mm: enable ARCH_SUPPORTS_PAGE_TABLE_CHECK")
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
>   arch/arm64/mm/hugetlbpage.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
> index 21716c940682..9c52718ea750 100644
> --- a/arch/arm64/mm/hugetlbpage.c
> +++ b/arch/arm64/mm/hugetlbpage.c
> @@ -236,7 +236,7 @@ static void clear_flush(struct mm_struct *mm,
>   	unsigned long i, saddr = addr;
>   
>   	for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
> -		pte_clear(mm, addr, ptep);
> +		ptep_clear(mm, addr, ptep);
>   
>   	flush_tlb_range(&vma, saddr, addr);
>   }
Will Deacon Aug. 11, 2023, 11:03 a.m. UTC | #2 
On Thu, Aug 10, 2023 at 09:32:41AM +0000, Qi Zheng wrote:
> From: Qi Zheng <zhengqi.arch@bytedance.com>
> 
> In clear_flush(), the original pte may be a present entry, so we should
> use ptep_clear() to let page_table_check track the pte clearing operation,
> otherwise it may cause false positive in subsequent set_pte_at().

Isn't this true for most users of pte_clear()? There are some in the core
code, so could they trigger the false positive as well?

Will
Will Deacon Aug. 11, 2023, 11:21 a.m. UTC | #3 
On Fri, Aug 11, 2023 at 07:16:20PM +0800, Qi Zheng wrote:
>    Will Deacon <[1]will@kernel.org>于2023年8月11日 周五19:03写道：
> 
>      On Thu, Aug 10, 2023 at 09:32:41AM +0000, Qi Zheng wrote:
>      > From: Qi Zheng <[2]zhengqi.arch@bytedance.com>
>      >
>      > In clear_flush(), the original pte may be a present entry, so we
>      should
>      > use ptep_clear() to let page_table_check track the pte clearing
>      operation,
>      > otherwise it may cause false positive in subsequent set_pte_at().
> 
>      Isn't this true for most users of pte_clear()? There are some in the
>      core
>      code, so could they trigger the false positive as well?
> 
>    No, the PTE entry in other places where pte_clear() is used is non-present
>    PTE. 
>    The page_table_check does not does track the pte operation in this case,
>    so it will not cause false positives.

Are you sure? For example, the call from flush_all_zero_pkmaps() in
highmem.c really looks like it's clearing a valid entry. Not that arm64
cares about highmem, but still.

Will
Will Deacon Aug. 22, 2023, 9:58 a.m. UTC | #4 
On Mon, Aug 21, 2023 at 01:21:19PM -0700, Andrew Morton wrote:
> On Fri, 11 Aug 2023 19:28:41 +0800 Qi Zheng <zhengqi.arch@bytedance.com> wrote:
> 
> > Will Deacon <will@kernel.org>于2023年8月11日 周五19:21写道：
> > 
> > > On Fri, Aug 11, 2023 at 07:16:20PM +0800, Qi Zheng wrote:
> > > >    Will Deacon <[1]will@kernel.org>于2023年8月11日 周五19:03写道：
> > > >
> > > >      On Thu, Aug 10, 2023 at 09:32:41AM +0000, Qi Zheng wrote:
> > > >      > From: Qi Zheng <[2]zhengqi.arch@bytedance.com>
> > > >      >
> > > >      > In clear_flush(), the original pte may be a present entry, so we
> > > >      should
> > > >      > use ptep_clear() to let page_table_check track the pte clearing
> > > >      operation,
> > > >      > otherwise it may cause false positive in subsequent set_pte_at().
> > > >
> > > >      Isn't this true for most users of pte_clear()? There are some in the
> > > >      core
> > > >      code, so could they trigger the false positive as well?
> > > >
> > > >    No, the PTE entry in other places where pte_clear() is used is
> > > non-present
> > > >    PTE.
> > > >    The page_table_check does not does track the pte operation in this
> > > case,
> > > >    so it will not cause false positives.
> > >
> > > Are you sure? For example, the call from flush_all_zero_pkmaps() in
> > > highmem.c really looks like it's clearing a valid entry. Not that arm64
> > > cares about highmem, but still.
> > 
> > 
> > Ah, this is init_mm, not user mm, page_table_check does not care about this
> > case.
> 
> It's unclear where we stand with this patch.  An ack or a nack, please?

Sorry Andrew, I saw you'd queued it so I marked it as "done" on my list. I
think it's fine:

Acked-by: Will Deacon <will@kernel.org>

Will
diff mbox series

Patch

diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
index 21716c940682..9c52718ea750 100644
--- a/arch/arm64/mm/hugetlbpage.c
+++ b/arch/arm64/mm/hugetlbpage.c
@@ -236,7 +236,7 @@  static void clear_flush(struct mm_struct *mm,
 	unsigned long i, saddr = addr;
 
 	for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
-		pte_clear(mm, addr, ptep);
+		ptep_clear(mm, addr, ptep);
 
 	flush_tlb_range(&vma, saddr, addr);
 }