[PATCH 0/4] mm: mremap: fix move page tables

[PATCH 0/4] mm: mremap: fix move page tables

[Kefeng Wang]

The first three patches use correct flush tlb functions when move page
tables, and patch 4 is a small optimization for hugepage on arm64.

Kefeng Wang (4):
  mm: hugetlb: use flush_hugetlb_tlb_range() in
    move_hugetlb_page_tables()
  mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()
  mm: mremap: use flush_pud_tlb_range in move_normal_pud()
  arm64: tlb: set huge page size to stride for hugepage

 arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
 mm/hugetlb.c                      |  4 ++--
 mm/mremap.c                       |  4 ++--
 3 files changed, 15 insertions(+), 14 deletions(-)

-- 
2.41.0

[PATCH 1/4] mm: hugetlb: use flush_hugetlb_tlb_range() in move_hugetlb_page_tables()

[Kefeng Wang]

Archs may need to do special things when flushing hugepage tlb,
so use the more applicable flush_hugetlb_tlb_range() instead of
flush_tlb_range().

Fixes: 550a7d60bd5e ("mm, hugepages: add mremap() support for hugepage backed vma")
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/hugetlb.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 64a3239b6407..ac876bfba340 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -5281,9 +5281,9 @@ int move_hugetlb_page_tables(struct vm_area_struct *vma,
 	}
 
 	if (shared_pmd)
-		flush_tlb_range(vma, range.start, range.end);
+		flush_hugetlb_tlb_range(vma, range.start, range.end);
 	else
-		flush_tlb_range(vma, old_end - len, old_end);
+		flush_hugetlb_tlb_range(vma, old_end - len, old_end);
 	mmu_notifier_invalidate_range_end(&range);
 	i_mmap_unlock_write(mapping);
 	hugetlb_vma_unlock_write(vma);
-- 
2.41.0

[PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()

[Kefeng Wang]

Archs may need to do special things when flushing thp tlb,
so use the more applicable flush_pud_tlb_range() instead of
flush_tlb_range().

Fixes: 2c91bd4a4e2e ("mm: speed up mremap by 20x on large regions")
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mremap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/mremap.c b/mm/mremap.c
index 11e06e4ab33b..1883205fa22b 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -284,7 +284,7 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	VM_BUG_ON(!pmd_none(*new_pmd));
 
 	pmd_populate(mm, new_pmd, pmd_pgtable(pmd));
-	flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
+	flush_pmd_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
 	if (new_ptl != old_ptl)
 		spin_unlock(new_ptl);
 	spin_unlock(old_ptl);
-- 
2.41.0

[PATCH 3/4] mm: mremap: use flush_pud_tlb_range in move_normal_pud()

[Kefeng Wang]

Archs may need to do special things when flushing thp tlb,
so use the more applicable flush_pud_tlb_range() instead of
flush_tlb_range().

Fixes: c49dd3401802 ("mm: speedup mremap on 1GB or larger regions")
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mremap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/mremap.c b/mm/mremap.c
index 1883205fa22b..25114e56901f 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -333,7 +333,7 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
 	VM_BUG_ON(!pud_none(*new_pud));
 
 	pud_populate(mm, new_pud, pud_pgtable(pud));
-	flush_tlb_range(vma, old_addr, old_addr + PUD_SIZE);
+	flush_pud_tlb_range(vma, old_addr, old_addr + PUD_SIZE);
 	if (new_ptl != old_ptl)
 		spin_unlock(new_ptl);
 	spin_unlock(old_ptl);
-- 
2.41.0

[PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Kefeng Wang]

It is better to use huge_page_size() for hugepage(HugeTLB) instead of
PAGE_SIZE for stride, which has been done in flush_pmd/pud_tlb_range(),
it could reduce the loop in __flush_tlb_range().

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
 1 file changed, 11 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
index 412a3b9a3c25..25e35e6f8093 100644
--- a/arch/arm64/include/asm/tlbflush.h
+++ b/arch/arm64/include/asm/tlbflush.h
@@ -360,16 +360,17 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
 	dsb(ish);
 }
 
-static inline void flush_tlb_range(struct vm_area_struct *vma,
-				   unsigned long start, unsigned long end)
-{
-	/*
-	 * We cannot use leaf-only invalidation here, since we may be invalidating
-	 * table entries as part of collapsing hugepages or moving page tables.
-	 * Set the tlb_level to 0 because we can not get enough information here.
-	 */
-	__flush_tlb_range(vma, start, end, PAGE_SIZE, false, 0);
-}
+/*
+ * We cannot use leaf-only invalidation here, since we may be invalidating
+ * table entries as part of collapsing hugepages or moving page tables.
+ * Set the tlb_level to 0 because we can not get enough information here.
+ */
+#define flush_tlb_range(vma, start, end)				\
+	__flush_tlb_range(vma, start, end,				\
+				((vma)->vm_flags & VM_HUGETLB)		\
+				? huge_page_size(hstate_vma(vma))	\
+				: PAGE_SIZE, false, 0)
+
 
 static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 {
-- 
2.41.0

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Barry Song]

On Mon, Jul 31, 2023 at 4:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> It is better to use huge_page_size() for hugepage(HugeTLB) instead of
> PAGE_SIZE for stride, which has been done in flush_pmd/pud_tlb_range(),
> it could reduce the loop in __flush_tlb_range().
>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
>  arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
>  1 file changed, 11 insertions(+), 10 deletions(-)
>
> diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
> index 412a3b9a3c25..25e35e6f8093 100644
> --- a/arch/arm64/include/asm/tlbflush.h
> +++ b/arch/arm64/include/asm/tlbflush.h
> @@ -360,16 +360,17 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
>         dsb(ish);
>  }
>
> -static inline void flush_tlb_range(struct vm_area_struct *vma,
> -                                  unsigned long start, unsigned long end)
> -{
> -       /*
> -        * We cannot use leaf-only invalidation here, since we may be invalidating
> -        * table entries as part of collapsing hugepages or moving page tables.
> -        * Set the tlb_level to 0 because we can not get enough information here.
> -        */
> -       __flush_tlb_range(vma, start, end, PAGE_SIZE, false, 0);
> -}
> +/*
> + * We cannot use leaf-only invalidation here, since we may be invalidating
> + * table entries as part of collapsing hugepages or moving page tables.
> + * Set the tlb_level to 0 because we can not get enough information here.
> + */
> +#define flush_tlb_range(vma, start, end)                               \
> +       __flush_tlb_range(vma, start, end,                              \
> +                               ((vma)->vm_flags & VM_HUGETLB)          \
> +                               ? huge_page_size(hstate_vma(vma))       \
> +                               : PAGE_SIZE, false, 0)
> +

seems like a good idea.

I wonder if a better implementation will be MMU_GATHER_PAGE_SIZE,  in this case,
we are going to support stride for other large folios as well, such as thp.

>
>  static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
>  {
> --
> 2.41.0
>

Thanks
Barry

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Barry Song]

On Mon, Jul 31, 2023 at 4:33 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Mon, Jul 31, 2023 at 4:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> >
> > It is better to use huge_page_size() for hugepage(HugeTLB) instead of
> > PAGE_SIZE for stride, which has been done in flush_pmd/pud_tlb_range(),
> > it could reduce the loop in __flush_tlb_range().
> >
> > Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> > ---
> >  arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
> >  1 file changed, 11 insertions(+), 10 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
> > index 412a3b9a3c25..25e35e6f8093 100644
> > --- a/arch/arm64/include/asm/tlbflush.h
> > +++ b/arch/arm64/include/asm/tlbflush.h
> > @@ -360,16 +360,17 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
> >         dsb(ish);
> >  }
> >
> > -static inline void flush_tlb_range(struct vm_area_struct *vma,
> > -                                  unsigned long start, unsigned long end)
> > -{
> > -       /*
> > -        * We cannot use leaf-only invalidation here, since we may be invalidating
> > -        * table entries as part of collapsing hugepages or moving page tables.
> > -        * Set the tlb_level to 0 because we can not get enough information here.
> > -        */
> > -       __flush_tlb_range(vma, start, end, PAGE_SIZE, false, 0);
> > -}
> > +/*
> > + * We cannot use leaf-only invalidation here, since we may be invalidating
> > + * table entries as part of collapsing hugepages or moving page tables.
> > + * Set the tlb_level to 0 because we can not get enough information here.
> > + */
> > +#define flush_tlb_range(vma, start, end)                               \
> > +       __flush_tlb_range(vma, start, end,                              \
> > +                               ((vma)->vm_flags & VM_HUGETLB)          \
> > +                               ? huge_page_size(hstate_vma(vma))       \
> > +                               : PAGE_SIZE, false, 0)
> > +
>
> seems like a good idea.
>
> I wonder if a better implementation will be MMU_GATHER_PAGE_SIZE,  in this case,
> we are going to support stride for other large folios as well, such as thp.
>

BTW, in most cases we have already had right stride:

arch/arm64/include/asm/tlb.h has already this to get stride:

static inline void tlb_flush(struct mmu_gather *tlb)
{
   struct vm_area_struct vma = TLB_FLUSH_VMA(tlb->mm, 0);
   bool last_level = !tlb->freed_tables;
   unsigned long stride = tlb_get_unmap_size(tlb);
   int tlb_level = tlb_get_level(tlb);

   /*
    * If we're tearing down the address space then we only care about
    * invalidating the walk-cache, since the ASID allocator won't
    * reallocate our ASID without invalidating the entire TLB.
  */
  if (tlb->fullmm) {
      if (!last_level)
           flush_tlb_mm(tlb->mm);
       return;
  }

   __flush_tlb_range(&vma, tlb->start, tlb->end, stride,
    last_level, tlb_level);
}

> >
> >  static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
> >  {
> > --
> > 2.41.0
> >
>
> Thanks
> Barry

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Kefeng Wang]

On 2023/7/31 16:43, Barry Song wrote:
> On Mon, Jul 31, 2023 at 4:33 PM Barry Song <21cnbao@gmail.com> wrote:
>>
>> On Mon, Jul 31, 2023 at 4:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>>
>>> It is better to use huge_page_size() for hugepage(HugeTLB) instead of
>>> PAGE_SIZE for stride, which has been done in flush_pmd/pud_tlb_range(),
>>> it could reduce the loop in __flush_tlb_range().
>>>
>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>>> ---
>>>   arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
>>>   1 file changed, 11 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
>>> index 412a3b9a3c25..25e35e6f8093 100644
>>> --- a/arch/arm64/include/asm/tlbflush.h
>>> +++ b/arch/arm64/include/asm/tlbflush.h
>>> @@ -360,16 +360,17 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
>>>          dsb(ish);
>>>   }
>>>
>>> -static inline void flush_tlb_range(struct vm_area_struct *vma,
>>> -                                  unsigned long start, unsigned long end)
>>> -{
>>> -       /*
>>> -        * We cannot use leaf-only invalidation here, since we may be invalidating
>>> -        * table entries as part of collapsing hugepages or moving page tables.
>>> -        * Set the tlb_level to 0 because we can not get enough information here.
>>> -        */
>>> -       __flush_tlb_range(vma, start, end, PAGE_SIZE, false, 0);
>>> -}
>>> +/*
>>> + * We cannot use leaf-only invalidation here, since we may be invalidating
>>> + * table entries as part of collapsing hugepages or moving page tables.
>>> + * Set the tlb_level to 0 because we can not get enough information here.
>>> + */
>>> +#define flush_tlb_range(vma, start, end)                               \
>>> +       __flush_tlb_range(vma, start, end,                              \
>>> +                               ((vma)->vm_flags & VM_HUGETLB)          \
>>> +                               ? huge_page_size(hstate_vma(vma))       \
>>> +                               : PAGE_SIZE, false, 0)
>>> +
>>
>> seems like a good idea.
>>
>> I wonder if a better implementation will be MMU_GATHER_PAGE_SIZE,  in this case,
>> we are going to support stride for other large folios as well, such as thp.
>>
> 
> BTW, in most cases we have already had right stride:
> 
> arch/arm64/include/asm/tlb.h has already this to get stride:

MMU_GATHER_PAGE_SIZE works for tlb_flush, but flush_tlb_range()
directly called without mmu_gather, see above 3 patches is to
use correct flush_[hugetlb/pmd/pud]_tlb_range(also there are
some other places, like get_clear_contig_flush/clear_flush on arm64),
so enable MMU_GATHER_PAGE_SIZE for arm64 is independent thing, right?

> 
> static inline void tlb_flush(struct mmu_gather *tlb)
> {
>     struct vm_area_struct vma = TLB_FLUSH_VMA(tlb->mm, 0);
>     bool last_level = !tlb->freed_tables;
>     unsigned long stride = tlb_get_unmap_size(tlb);
>     int tlb_level = tlb_get_level(tlb);
> 
>     /*
>      * If we're tearing down the address space then we only care about
>      * invalidating the walk-cache, since the ASID allocator won't
>      * reallocate our ASID without invalidating the entire TLB.
>    */
>    if (tlb->fullmm) {
>        if (!last_level)
>             flush_tlb_mm(tlb->mm);
>         return;
>    }
> 
>     __flush_tlb_range(&vma, tlb->start, tlb->end, stride,
>      last_level, tlb_level);
> }
> 
>>>
>>>   static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
>>>   {
>>> --
>>> 2.41.0
>>>
>>
>> Thanks
>> Barry

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Barry Song]

On Mon, Jul 31, 2023 at 5:29 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
>
>
> On 2023/7/31 16:43, Barry Song wrote:
> > On Mon, Jul 31, 2023 at 4:33 PM Barry Song <21cnbao@gmail.com> wrote:
> >>
> >> On Mon, Jul 31, 2023 at 4:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> >>>
> >>> It is better to use huge_page_size() for hugepage(HugeTLB) instead of
> >>> PAGE_SIZE for stride, which has been done in flush_pmd/pud_tlb_range(),
> >>> it could reduce the loop in __flush_tlb_range().
> >>>
> >>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >>> ---
> >>>   arch/arm64/include/asm/tlbflush.h | 21 +++++++++++----------
> >>>   1 file changed, 11 insertions(+), 10 deletions(-)
> >>>
> >>> diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
> >>> index 412a3b9a3c25..25e35e6f8093 100644
> >>> --- a/arch/arm64/include/asm/tlbflush.h
> >>> +++ b/arch/arm64/include/asm/tlbflush.h
> >>> @@ -360,16 +360,17 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
> >>>          dsb(ish);
> >>>   }
> >>>
> >>> -static inline void flush_tlb_range(struct vm_area_struct *vma,
> >>> -                                  unsigned long start, unsigned long end)
> >>> -{
> >>> -       /*
> >>> -        * We cannot use leaf-only invalidation here, since we may be invalidating
> >>> -        * table entries as part of collapsing hugepages or moving page tables.
> >>> -        * Set the tlb_level to 0 because we can not get enough information here.
> >>> -        */
> >>> -       __flush_tlb_range(vma, start, end, PAGE_SIZE, false, 0);
> >>> -}
> >>> +/*
> >>> + * We cannot use leaf-only invalidation here, since we may be invalidating
> >>> + * table entries as part of collapsing hugepages or moving page tables.
> >>> + * Set the tlb_level to 0 because we can not get enough information here.
> >>> + */
> >>> +#define flush_tlb_range(vma, start, end)                               \
> >>> +       __flush_tlb_range(vma, start, end,                              \
> >>> +                               ((vma)->vm_flags & VM_HUGETLB)          \
> >>> +                               ? huge_page_size(hstate_vma(vma))       \
> >>> +                               : PAGE_SIZE, false, 0)
> >>> +
> >>
> >> seems like a good idea.
> >>
> >> I wonder if a better implementation will be MMU_GATHER_PAGE_SIZE,  in this case,
> >> we are going to support stride for other large folios as well, such as thp.
> >>
> >
> > BTW, in most cases we have already had right stride:
> >
> > arch/arm64/include/asm/tlb.h has already this to get stride:
>
> MMU_GATHER_PAGE_SIZE works for tlb_flush, but flush_tlb_range()
> directly called without mmu_gather, see above 3 patches is to
> use correct flush_[hugetlb/pmd/pud]_tlb_range(also there are
> some other places, like get_clear_contig_flush/clear_flush on arm64),
> so enable MMU_GATHER_PAGE_SIZE for arm64 is independent thing, right?
>

You are right. I was thinking of those zap_pte/pmd_range cases especially
for those vmas where large folios engage. but it is not very relevant.
In that case, one vma might have mixed different folio sizes.
your patch, for sure, will benefit hugetlb with arm64 contiguous bits.

Thanks
Barry

Re: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()

[Catalin Marinas]

On Mon, Jul 31, 2023 at 03:48:27PM +0800, Kefeng Wang wrote:
> Archs may need to do special things when flushing thp tlb,
> so use the more applicable flush_pud_tlb_range() instead of
> flush_tlb_range().
> 
> Fixes: 2c91bd4a4e2e ("mm: speed up mremap by 20x on large regions")
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
>  mm/mremap.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/mremap.c b/mm/mremap.c
> index 11e06e4ab33b..1883205fa22b 100644
> --- a/mm/mremap.c
> +++ b/mm/mremap.c
> @@ -284,7 +284,7 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
>  	VM_BUG_ON(!pmd_none(*new_pmd));
>  
>  	pmd_populate(mm, new_pmd, pmd_pgtable(pmd));
> -	flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
> +	flush_pmd_tlb_range(vma, old_addr, old_addr + PMD_SIZE);

I don't think that's correct for arm64. The assumption in the
flush_p*d_tlb_range() was that they are called only for block mappings
at that p*d level (and we use FEAT_TTL on arm64 indicating that the leaf
level is level 2 for pmd, 1 for pud). IIUC move_normal_pmd() is only
called for table pmds which would have a leaf level of 3 (the pte).

Same for the next patch doing the equivalent for the pud.

-- 
Catalin

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Catalin Marinas]

On Mon, Jul 31, 2023 at 03:48:29PM +0800, Kefeng Wang wrote:
> +/*
> + * We cannot use leaf-only invalidation here, since we may be invalidating
> + * table entries as part of collapsing hugepages or moving page tables.
> + * Set the tlb_level to 0 because we can not get enough information here.
> + */
> +#define flush_tlb_range(vma, start, end)				\
> +	__flush_tlb_range(vma, start, end,				\
> +				((vma)->vm_flags & VM_HUGETLB)		\
> +				? huge_page_size(hstate_vma(vma))	\
> +				: PAGE_SIZE, false, 0)

This won't work if we use the contiguous PTE to get 64K hugetlb pages on
a 4K base page configuration. The 16 base pages in the range would have
to be invalidated individually (the contig PTE bit is just a hint, the
hardware may or may not take it into account).

-- 
Catalin

Re: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()

[Kefeng Wang]

On 2023/7/31 19:05, Catalin Marinas wrote:
> On Mon, Jul 31, 2023 at 03:48:27PM +0800, Kefeng Wang wrote:
>> Archs may need to do special things when flushing thp tlb,
>> so use the more applicable flush_pud_tlb_range() instead of
>> flush_tlb_range().
>>
>> Fixes: 2c91bd4a4e2e ("mm: speed up mremap by 20x on large regions")
>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>> ---
>>   mm/mremap.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/mm/mremap.c b/mm/mremap.c
>> index 11e06e4ab33b..1883205fa22b 100644
>> --- a/mm/mremap.c
>> +++ b/mm/mremap.c
>> @@ -284,7 +284,7 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
>>   	VM_BUG_ON(!pmd_none(*new_pmd));
>>   
>>   	pmd_populate(mm, new_pmd, pmd_pgtable(pmd));
>> -	flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
>> +	flush_pmd_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
> 
> I don't think that's correct for arm64. The assumption in the
> flush_p*d_tlb_range() was that they are called only for block mappings
> at that p*d level (and we use FEAT_TTL on arm64 indicating that the leaf
> level is level 2 for pmd, 1 for pud). IIUC move_normal_pmd() is only
> called for table pmds which would have a leaf level of 3 (the pte).

oops, yes, this is for NORMAL_PMD case, not HPAGE_PMD, please ignore 
patch 2/3.

> 
> Same for the next patch doing the equivalent for the pud.
> 

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Kefeng Wang]

On 2023/7/31 19:11, Catalin Marinas wrote:
> On Mon, Jul 31, 2023 at 03:48:29PM +0800, Kefeng Wang wrote:
>> +/*
>> + * We cannot use leaf-only invalidation here, since we may be invalidating
>> + * table entries as part of collapsing hugepages or moving page tables.
>> + * Set the tlb_level to 0 because we can not get enough information here.
>> + */
>> +#define flush_tlb_range(vma, start, end)				\
>> +	__flush_tlb_range(vma, start, end,				\
>> +				((vma)->vm_flags & VM_HUGETLB)		\
>> +				? huge_page_size(hstate_vma(vma))	\
>> +				: PAGE_SIZE, false, 0)
> 
> This won't work if we use the contiguous PTE to get 64K hugetlb pages on
> a 4K base page configuration. The 16 base pages in the range would have
> to be invalidated individually (the contig PTE bit is just a hint, the
> hardware may or may not take it into account).

Got it, the contig huge page is depended on hardware implementation,
but for normal hugepage(2M/1G), we could use this, right?
> 

Re: [PATCH 4/4] arm64: tlb: set huge page size to stride for hugepage

[Catalin Marinas]

On Mon, Jul 31, 2023 at 07:27:14PM +0800, Kefeng Wang wrote:
> On 2023/7/31 19:11, Catalin Marinas wrote:
> > On Mon, Jul 31, 2023 at 03:48:29PM +0800, Kefeng Wang wrote:
> > > +/*
> > > + * We cannot use leaf-only invalidation here, since we may be invalidating
> > > + * table entries as part of collapsing hugepages or moving page tables.
> > > + * Set the tlb_level to 0 because we can not get enough information here.
> > > + */
> > > +#define flush_tlb_range(vma, start, end)				\
> > > +	__flush_tlb_range(vma, start, end,				\
> > > +				((vma)->vm_flags & VM_HUGETLB)		\
> > > +				? huge_page_size(hstate_vma(vma))	\
> > > +				: PAGE_SIZE, false, 0)
> > 
> > This won't work if we use the contiguous PTE to get 64K hugetlb pages on
> > a 4K base page configuration. The 16 base pages in the range would have
> > to be invalidated individually (the contig PTE bit is just a hint, the
> > hardware may or may not take it into account).
> 
> Got it, the contig huge page is depended on hardware implementation,
> but for normal hugepage(2M/1G), we could use this, right?

Right. Only the pmd/pud cases.

-- 
Catalin

Re: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()

[kernel test robot]

Hi Kefeng,

kernel test robot noticed the following build errors:

[auto build test ERROR on arm64/for-next/core]
[also build test ERROR on arm-perf/for-next/perf linus/master v6.5-rc4 next-20230731]
[cannot apply to akpm-mm/mm-everything]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Kefeng-Wang/mm-hugetlb-use-flush_hugetlb_tlb_range-in-move_hugetlb_page_tables/20230731-154016
base:   https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git for-next/core
patch link:    https://lore.kernel.org/r/20230731074829.79309-3-wangkefeng.wang%40huawei.com
patch subject: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()
config: x86_64-defconfig (https://download.01.org/0day-ci/archive/20230731/202307312137.ormxuS5g-lkp@intel.com/config)
compiler: gcc-12 (Debian 12.2.0-14) 12.2.0
reproduce: (https://download.01.org/0day-ci/archive/20230731/202307312137.ormxuS5g-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202307312137.ormxuS5g-lkp@intel.com/

All errors (new ones prefixed by >>):

   In file included from <command-line>:
   In function 'move_normal_pmd',
       inlined from 'move_pgt_entry' at mm/mremap.c:463:11,
       inlined from 'move_page_tables' at mm/mremap.c:565:8:
>> include/linux/compiler_types.h:397:45: error: call to '__compiletime_assert_338' declared with attribute error: BUILD_BUG failed
     397 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
         |                                             ^
   include/linux/compiler_types.h:378:25: note: in definition of macro '__compiletime_assert'
     378 |                         prefix ## suffix();                             \
         |                         ^~~~~~
   include/linux/compiler_types.h:397:9: note: in expansion of macro '_compiletime_assert'
     397 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
         |         ^~~~~~~~~~~~~~~~~~~
   include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert'
      39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
         |                                     ^~~~~~~~~~~~~~~~~~
   include/linux/build_bug.h:59:21: note: in expansion of macro 'BUILD_BUG_ON_MSG'
      59 | #define BUILD_BUG() BUILD_BUG_ON_MSG(1, "BUILD_BUG failed")
         |                     ^~~~~~~~~~~~~~~~
   include/linux/pgtable.h:1415:49: note: in expansion of macro 'BUILD_BUG'
    1415 | #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
         |                                                 ^~~~~~~~~
   mm/mremap.c:287:9: note: in expansion of macro 'flush_pmd_tlb_range'
     287 |         flush_pmd_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
         |         ^~~~~~~~~~~~~~~~~~~


vim +/__compiletime_assert_338 +397 include/linux/compiler_types.h

eb5c2d4b45e3d2 Will Deacon 2020-07-21  383  
eb5c2d4b45e3d2 Will Deacon 2020-07-21  384  #define _compiletime_assert(condition, msg, prefix, suffix) \
eb5c2d4b45e3d2 Will Deacon 2020-07-21  385  	__compiletime_assert(condition, msg, prefix, suffix)
eb5c2d4b45e3d2 Will Deacon 2020-07-21  386  
eb5c2d4b45e3d2 Will Deacon 2020-07-21  387  /**
eb5c2d4b45e3d2 Will Deacon 2020-07-21  388   * compiletime_assert - break build and emit msg if condition is false
eb5c2d4b45e3d2 Will Deacon 2020-07-21  389   * @condition: a compile-time constant condition to check
eb5c2d4b45e3d2 Will Deacon 2020-07-21  390   * @msg:       a message to emit if condition is false
eb5c2d4b45e3d2 Will Deacon 2020-07-21  391   *
eb5c2d4b45e3d2 Will Deacon 2020-07-21  392   * In tradition of POSIX assert, this macro will break the build if the
eb5c2d4b45e3d2 Will Deacon 2020-07-21  393   * supplied condition is *false*, emitting the supplied error message if the
eb5c2d4b45e3d2 Will Deacon 2020-07-21  394   * compiler has support to do so.
eb5c2d4b45e3d2 Will Deacon 2020-07-21  395   */
eb5c2d4b45e3d2 Will Deacon 2020-07-21  396  #define compiletime_assert(condition, msg) \
eb5c2d4b45e3d2 Will Deacon 2020-07-21 @397  	_compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
eb5c2d4b45e3d2 Will Deacon 2020-07-21  398  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 3/4] mm: mremap: use flush_pud_tlb_range in move_normal_pud()

[kernel test robot]

Hi Kefeng,

kernel test robot noticed the following build errors:

[auto build test ERROR on arm64/for-next/core]
[also build test ERROR on arm-perf/for-next/perf linus/master v6.5-rc4 next-20230731]
[cannot apply to akpm-mm/mm-everything]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Kefeng-Wang/mm-hugetlb-use-flush_hugetlb_tlb_range-in-move_hugetlb_page_tables/20230731-154016
base:   https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git for-next/core
patch link:    https://lore.kernel.org/r/20230731074829.79309-4-wangkefeng.wang%40huawei.com
patch subject: [PATCH 3/4] mm: mremap: use flush_pud_tlb_range in move_normal_pud()
config: riscv-allmodconfig (https://download.01.org/0day-ci/archive/20230801/202308010022.uY01vAew-lkp@intel.com/config)
compiler: riscv64-linux-gcc (GCC) 12.3.0
reproduce: (https://download.01.org/0day-ci/archive/20230801/202308010022.uY01vAew-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202308010022.uY01vAew-lkp@intel.com/

All errors (new ones prefixed by >>):

   mm/mremap.c: In function 'move_normal_pud':
>> mm/mremap.c:336:9: error: implicit declaration of function 'flush_pud_tlb_range'; did you mean 'flush_pmd_tlb_range'? [-Werror=implicit-function-declaration]
     336 |         flush_pud_tlb_range(vma, old_addr, old_addr + PUD_SIZE);
         |         ^~~~~~~~~~~~~~~~~~~
         |         flush_pmd_tlb_range
   cc1: some warnings being treated as errors


vim +336 mm/mremap.c

   302	
   303	#if CONFIG_PGTABLE_LEVELS > 2 && defined(CONFIG_HAVE_MOVE_PUD)
   304	static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
   305			  unsigned long new_addr, pud_t *old_pud, pud_t *new_pud)
   306	{
   307		spinlock_t *old_ptl, *new_ptl;
   308		struct mm_struct *mm = vma->vm_mm;
   309		pud_t pud;
   310	
   311		if (!arch_supports_page_table_move())
   312			return false;
   313		/*
   314		 * The destination pud shouldn't be established, free_pgtables()
   315		 * should have released it.
   316		 */
   317		if (WARN_ON_ONCE(!pud_none(*new_pud)))
   318			return false;
   319	
   320		/*
   321		 * We don't have to worry about the ordering of src and dst
   322		 * ptlocks because exclusive mmap_lock prevents deadlock.
   323		 */
   324		old_ptl = pud_lock(vma->vm_mm, old_pud);
   325		new_ptl = pud_lockptr(mm, new_pud);
   326		if (new_ptl != old_ptl)
   327			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
   328	
   329		/* Clear the pud */
   330		pud = *old_pud;
   331		pud_clear(old_pud);
   332	
   333		VM_BUG_ON(!pud_none(*new_pud));
   334	
   335		pud_populate(mm, new_pud, pud_pgtable(pud));
 > 336		flush_pud_tlb_range(vma, old_addr, old_addr + PUD_SIZE);
   337		if (new_ptl != old_ptl)
   338			spin_unlock(new_ptl);
   339		spin_unlock(old_ptl);
   340	
   341		return true;
   342	}
   343	#else
   344	static inline bool move_normal_pud(struct vm_area_struct *vma,
   345			unsigned long old_addr, unsigned long new_addr, pud_t *old_pud,
   346			pud_t *new_pud)
   347	{
   348		return false;
   349	}
   350	#endif
   351	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()

[kernel test robot]

Hi Kefeng,

kernel test robot noticed the following build errors:

[auto build test ERROR on arm64/for-next/core]
[also build test ERROR on arm-perf/for-next/perf linus/master v6.5-rc4]
[cannot apply to akpm-mm/mm-everything]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Kefeng-Wang/mm-hugetlb-use-flush_hugetlb_tlb_range-in-move_hugetlb_page_tables/20230731-154016
base:   https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git for-next/core
patch link:    https://lore.kernel.org/r/20230731074829.79309-3-wangkefeng.wang%40huawei.com
patch subject: [PATCH 2/4] mm: mremap: use flush_pmd_tlb_range() in move_normal_pmd()
config: x86_64-randconfig-x003-20230731 (https://download.01.org/0day-ci/archive/20230801/202308010553.KxefZFdO-lkp@intel.com/config)
compiler: clang version 16.0.4 (https://github.com/llvm/llvm-project.git ae42196bc493ffe877a7e3dff8be32035dea4d07)
reproduce: (https://download.01.org/0day-ci/archive/20230801/202308010553.KxefZFdO-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202308010553.KxefZFdO-lkp@intel.com/

All errors (new ones prefixed by >>):

>> ld.lld: error: call to __compiletime_assert_860 marked "dontcall-error": BUILD_BUG failed

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 1/4] mm: hugetlb: use flush_hugetlb_tlb_range() in move_hugetlb_page_tables()

[Mike Kravetz]

On 07/31/23 15:48, Kefeng Wang wrote:
> Archs may need to do special things when flushing hugepage tlb,
> so use the more applicable flush_hugetlb_tlb_range() instead of
> flush_tlb_range().
> 
> Fixes: 550a7d60bd5e ("mm, hugepages: add mremap() support for hugepage backed vma")
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Thanks!

Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>

Although, I missed this in 550a7d60bd5e :(

Looks like only powerpc provides an arch specific flush_hugetlb_tlb_range
today.
-- 
Mike Kravetz

> ---
>  mm/hugetlb.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 64a3239b6407..ac876bfba340 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -5281,9 +5281,9 @@ int move_hugetlb_page_tables(struct vm_area_struct *vma,
>  	}
>  
>  	if (shared_pmd)
> -		flush_tlb_range(vma, range.start, range.end);
> +		flush_hugetlb_tlb_range(vma, range.start, range.end);
>  	else
> -		flush_tlb_range(vma, old_end - len, old_end);
> +		flush_hugetlb_tlb_range(vma, old_end - len, old_end);
>  	mmu_notifier_invalidate_range_end(&range);
>  	i_mmap_unlock_write(mapping);
>  	hugetlb_vma_unlock_write(vma);
> -- 
> 2.41.0
> 

Re: [PATCH 1/4] mm: hugetlb: use flush_hugetlb_tlb_range() in move_hugetlb_page_tables()

[Muchun Song]

> On Jul 31, 2023, at 15:48, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> 
> Archs may need to do special things when flushing hugepage tlb,
> so use the more applicable flush_hugetlb_tlb_range() instead of
> flush_tlb_range().
> 
> Fixes: 550a7d60bd5e ("mm, hugepages: add mremap() support for hugepage backed vma")
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Acked-by: Muchun Song <songmuchun@bytedance.com>

Re: [PATCH 1/4] mm: hugetlb: use flush_hugetlb_tlb_range() in move_hugetlb_page_tables()

[Mina Almasry]

On Mon, Jul 31, 2023 at 4:40 PM Mike Kravetz <mike.kravetz@oracle.com> wrote:
>
> On 07/31/23 15:48, Kefeng Wang wrote:
> > Archs may need to do special things when flushing hugepage tlb,
> > so use the more applicable flush_hugetlb_tlb_range() instead of
> > flush_tlb_range().
> >
> > Fixes: 550a7d60bd5e ("mm, hugepages: add mremap() support for hugepage backed vma")
> > Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>
> Thanks!
>
> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
>

Sorry for jumping in late, but given the concerns raised around HGM
and the deviation between hugetlb and the rest of MM, does it make
sense to try to make an incremental effort towards avoiding hugetlb
specialization?

In the context of this patch, I would prefer that the arch upgrade
flush_tlb_range() to handle hugetlb correctly, instead of adding more
hugetlb specific deviations, ala flush_hugetlb_tlb_range. While it's
at it, maybe replace flush_hugetlb_tlb_range() in the code with
flush_tlb_range().

Although, I don't have the expertise to judge if upgrading
flush_tlb_range() to handle hugetlb is easy or feasible at all.

> Although, I missed this in 550a7d60bd5e :(
>
> Looks like only powerpc provides an arch specific flush_hugetlb_tlb_range
> today.
> --
> Mike Kravetz
>
> > ---
> >  mm/hugetlb.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 64a3239b6407..ac876bfba340 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -5281,9 +5281,9 @@ int move_hugetlb_page_tables(struct vm_area_struct *vma,
> >       }
> >
> >       if (shared_pmd)
> > -             flush_tlb_range(vma, range.start, range.end);
> > +             flush_hugetlb_tlb_range(vma, range.start, range.end);
> >       else
> > -             flush_tlb_range(vma, old_end - len, old_end);
> > +             flush_hugetlb_tlb_range(vma, old_end - len, old_end);
> >       mmu_notifier_invalidate_range_end(&range);
> >       i_mmap_unlock_write(mapping);
> >       hugetlb_vma_unlock_write(vma);
> > --
> > 2.41.0
> >



-- 
Thanks,
Mina