[v1] mm: move vma_(kernel|mmu)_pagesize() out of hugetlb.c

[PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by David Hildenbrand (Arm) 1 month ago

In the past, only hugetlb had special "vma_kernel_pagesize()"
requirements, so it provided its own implementation.

In commit 05ea88608d4e ("mm, hugetlbfs: introduce ->pagesize() to
vm_operations_struct") we generalized that approach by providing a
vm_ops->pagesize() callback to be used by device-dax.

Once device-dax started using that callback in commit c1d53b92b95c
("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
it was missed that CONFIG_DEV_DAX does not depend on hugetlb support.

So building a kernel with CONFIG_DEV_DAX but without CONFIG_HUGETLBFS
would not pick up that value.

Fix it by moving vma_kernel_pagesize() to mm.h, providing only a single
implementation. While at it, improve the kerneldoc a bit.

Ideally, we'd move vma_mmu_pagesize() as well to the header. However,
its __weak symbol might be overwritten by a PPC variant in hugetlb code.
So let's leave it in there for now, as it really only matters for some
hugetlb oddities.

This was found by code inspection.

Fixes: c1d53b92b95c ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>
---
 include/linux/hugetlb.h |  7 -------
 include/linux/mm.h      | 20 ++++++++++++++++++++
 mm/hugetlb.c            | 17 -----------------
 3 files changed, 20 insertions(+), 24 deletions(-)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 65910437be1c..44c1848a2c21 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -777,8 +777,6 @@ static inline unsigned long huge_page_size(const struct hstate *h)
 	return (unsigned long)PAGE_SIZE << h->order;
 }
 
-extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
-
 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
 
 static inline unsigned long huge_page_mask(struct hstate *h)
@@ -1177,11 +1175,6 @@ static inline unsigned long huge_page_mask(struct hstate *h)
 	return PAGE_MASK;
 }
 
-static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
-{
-	return PAGE_SIZE;
-}
-
 static inline unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
 {
 	return PAGE_SIZE;
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 44e04a42fe77..227809790f1a 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1307,6 +1307,26 @@ static inline bool vma_is_shared_maywrite(const struct vm_area_struct *vma)
 	return is_shared_maywrite(&vma->flags);
 }
 
+/**
+ * vma_kernel_pagesize - Default page size granularity for this VMA.
+ * @vma: The user mapping.
+ *
+ * The kernel page size specifies in which granularity VMA modifications
+ * can be performed. Folios in this VMA will be aligned to, and at least
+ * the size of the number of bytes returned by this function.
+ *
+ * The default kernel page size is not affected by Transparent Huge Pages
+ * being in effect.
+ *
+ * Return: The default page size granularity for this VMA.
+ */
+static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
+{
+	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
+		return vma->vm_ops->pagesize(vma);
+	return PAGE_SIZE;
+}
+
 static inline
 struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max)
 {
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 1d41fa3dd43e..66eadfa9e958 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1017,23 +1017,6 @@ static pgoff_t vma_hugecache_offset(struct hstate *h,
 			(vma->vm_pgoff >> huge_page_order(h));
 }
 
-/**
- * vma_kernel_pagesize - Page size granularity for this VMA.
- * @vma: The user mapping.
- *
- * Folios in this VMA will be aligned to, and at least the size of the
- * number of bytes returned by this function.
- *
- * Return: The default size of the folios allocated when backing a VMA.
- */
-unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
-{
-	if (vma->vm_ops && vma->vm_ops->pagesize)
-		return vma->vm_ops->pagesize(vma);
-	return PAGE_SIZE;
-}
-EXPORT_SYMBOL_GPL(vma_kernel_pagesize);
-
 /*
  * Return the page size being used by the MMU to back a VMA. In the majority
  * of cases, the page size used by the kernel matches the MMU size. On
-- 
2.43.0

Re: [PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by Lorenzo Stoakes (Oracle) 1 month ago

On Fri, Mar 06, 2026 at 11:15:57AM +0100, David Hildenbrand (Arm) wrote:
> In the past, only hugetlb had special "vma_kernel_pagesize()"
> requirements, so it provided its own implementation.
>
> In commit 05ea88608d4e ("mm, hugetlbfs: introduce ->pagesize() to
> vm_operations_struct") we generalized that approach by providing a
> vm_ops->pagesize() callback to be used by device-dax.
>
> Once device-dax started using that callback in commit c1d53b92b95c
> ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> it was missed that CONFIG_DEV_DAX does not depend on hugetlb support.
>
> So building a kernel with CONFIG_DEV_DAX but without CONFIG_HUGETLBFS
> would not pick up that value.
>
> Fix it by moving vma_kernel_pagesize() to mm.h, providing only a single
> implementation. While at it, improve the kerneldoc a bit.
>
> Ideally, we'd move vma_mmu_pagesize() as well to the header. However,
> its __weak symbol might be overwritten by a PPC variant in hugetlb code.
> So let's leave it in there for now, as it really only matters for some
> hugetlb oddities.
>
> This was found by code inspection.
>
> Fixes: c1d53b92b95c ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> Cc: Dan Williams <dan.j.williams@intel.com>
> Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>

LGTM, but you need to fix up VMA tests, I attach a patch below to do this. Will
this resolved:

Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>

> ---
>  include/linux/hugetlb.h |  7 -------
>  include/linux/mm.h      | 20 ++++++++++++++++++++
>  mm/hugetlb.c            | 17 -----------------
>  3 files changed, 20 insertions(+), 24 deletions(-)
>
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index 65910437be1c..44c1848a2c21 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -777,8 +777,6 @@ static inline unsigned long huge_page_size(const struct hstate *h)
>  	return (unsigned long)PAGE_SIZE << h->order;
>  }
>
> -extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
> -
>  extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
>
>  static inline unsigned long huge_page_mask(struct hstate *h)
> @@ -1177,11 +1175,6 @@ static inline unsigned long huge_page_mask(struct hstate *h)
>  	return PAGE_MASK;
>  }
>
> -static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
> -{
> -	return PAGE_SIZE;
> -}
> -
>  static inline unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
>  {
>  	return PAGE_SIZE;
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 44e04a42fe77..227809790f1a 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -1307,6 +1307,26 @@ static inline bool vma_is_shared_maywrite(const struct vm_area_struct *vma)
>  	return is_shared_maywrite(&vma->flags);
>  }
>
> +/**
> + * vma_kernel_pagesize - Default page size granularity for this VMA.
> + * @vma: The user mapping.
> + *
> + * The kernel page size specifies in which granularity VMA modifications
> + * can be performed. Folios in this VMA will be aligned to, and at least
> + * the size of the number of bytes returned by this function.
> + *
> + * The default kernel page size is not affected by Transparent Huge Pages
> + * being in effect.
> + *
> + * Return: The default page size granularity for this VMA.
> + */
> +static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
> +{
> +	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
> +		return vma->vm_ops->pagesize(vma);
> +	return PAGE_SIZE;
> +}
> +
>  static inline
>  struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max)
>  {
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 1d41fa3dd43e..66eadfa9e958 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1017,23 +1017,6 @@ static pgoff_t vma_hugecache_offset(struct hstate *h,
>  			(vma->vm_pgoff >> huge_page_order(h));
>  }
>
> -/**
> - * vma_kernel_pagesize - Page size granularity for this VMA.
> - * @vma: The user mapping.
> - *
> - * Folios in this VMA will be aligned to, and at least the size of the
> - * number of bytes returned by this function.
> - *
> - * Return: The default size of the folios allocated when backing a VMA.
> - */
> -unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
> -{
> -	if (vma->vm_ops && vma->vm_ops->pagesize)
> -		return vma->vm_ops->pagesize(vma);
> -	return PAGE_SIZE;
> -}
> -EXPORT_SYMBOL_GPL(vma_kernel_pagesize);
> -
>  /*
>   * Return the page size being used by the MMU to back a VMA. In the majority
>   * of cases, the page size used by the kernel matches the MMU size. On
> --
> 2.43.0
>

----8<----
This breaks the VMA tests when patch 2/4 removes the references in other
headers. So this patch should also update them, I enclose a simple fix for
convenience:

From bec84895cbdbe28e3495c4d90e097074598419e5 Mon Sep 17 00:00:00 2001
From: "Lorenzo Stoakes (Oracle)" <ljs@kernel.org>
Date: Fri, 6 Mar 2026 11:05:12 +0000
Subject: [PATCH] fix

---
 tools/testing/vma/include/dup.h | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/tools/testing/vma/include/dup.h b/tools/testing/vma/include/dup.h
index 3078ff1487d3..65b1030a7fdf 100644
--- a/tools/testing/vma/include/dup.h
+++ b/tools/testing/vma/include/dup.h
@@ -1318,3 +1318,10 @@ static inline void vma_set_file(struct vm_area_struct *vma, struct file *file)
 	swap(vma->vm_file, file);
 	fput(file);
 }
+
+static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
+{
+	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
+		return vma->vm_ops->pagesize(vma);
+	return PAGE_SIZE;
+}
--
2.53.0

Re: [PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by David Hildenbrand (Arm) 1 month ago

On 3/6/26 12:07, Lorenzo Stoakes (Oracle) wrote:
> On Fri, Mar 06, 2026 at 11:15:57AM +0100, David Hildenbrand (Arm) wrote:
>> In the past, only hugetlb had special "vma_kernel_pagesize()"
>> requirements, so it provided its own implementation.
>>
>> In commit 05ea88608d4e ("mm, hugetlbfs: introduce ->pagesize() to
>> vm_operations_struct") we generalized that approach by providing a
>> vm_ops->pagesize() callback to be used by device-dax.
>>
>> Once device-dax started using that callback in commit c1d53b92b95c
>> ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
>> it was missed that CONFIG_DEV_DAX does not depend on hugetlb support.
>>
>> So building a kernel with CONFIG_DEV_DAX but without CONFIG_HUGETLBFS
>> would not pick up that value.
>>
>> Fix it by moving vma_kernel_pagesize() to mm.h, providing only a single
>> implementation. While at it, improve the kerneldoc a bit.
>>
>> Ideally, we'd move vma_mmu_pagesize() as well to the header. However,
>> its __weak symbol might be overwritten by a PPC variant in hugetlb code.
>> So let's leave it in there for now, as it really only matters for some
>> hugetlb oddities.
>>
>> This was found by code inspection.
>>
>> Fixes: c1d53b92b95c ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
>> Cc: Dan Williams <dan.j.williams@intel.com>
>> Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>
> 
> LGTM, but you need to fix up VMA tests, I attach a patch below to do this. Will
> this resolved:

Thanks!

I assume that should go into patch #2 instead?

> 
> Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>


[...]

> ---
>  tools/testing/vma/include/dup.h | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/tools/testing/vma/include/dup.h b/tools/testing/vma/include/dup.h
> index 3078ff1487d3..65b1030a7fdf 100644
> --- a/tools/testing/vma/include/dup.h
> +++ b/tools/testing/vma/include/dup.h
> @@ -1318,3 +1318,10 @@ static inline void vma_set_file(struct vm_area_struct *vma, struct file *file)
>  	swap(vma->vm_file, file);
>  	fput(file);
>  }
> +
> +static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
> +{
> +	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
> +		return vma->vm_ops->pagesize(vma);
> +	return PAGE_SIZE;

Should we just KIS and use PAGE_SIZE for the test?

-- 
Cheers,

David

Re: [PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by Lorenzo Stoakes (Oracle) 1 month ago

On Fri, Mar 06, 2026 at 02:25:51PM +0100, David Hildenbrand (Arm) wrote:
> On 3/6/26 12:07, Lorenzo Stoakes (Oracle) wrote:
> > On Fri, Mar 06, 2026 at 11:15:57AM +0100, David Hildenbrand (Arm) wrote:
> >> In the past, only hugetlb had special "vma_kernel_pagesize()"
> >> requirements, so it provided its own implementation.
> >>
> >> In commit 05ea88608d4e ("mm, hugetlbfs: introduce ->pagesize() to
> >> vm_operations_struct") we generalized that approach by providing a
> >> vm_ops->pagesize() callback to be used by device-dax.
> >>
> >> Once device-dax started using that callback in commit c1d53b92b95c
> >> ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> >> it was missed that CONFIG_DEV_DAX does not depend on hugetlb support.
> >>
> >> So building a kernel with CONFIG_DEV_DAX but without CONFIG_HUGETLBFS
> >> would not pick up that value.
> >>
> >> Fix it by moving vma_kernel_pagesize() to mm.h, providing only a single
> >> implementation. While at it, improve the kerneldoc a bit.
> >>
> >> Ideally, we'd move vma_mmu_pagesize() as well to the header. However,
> >> its __weak symbol might be overwritten by a PPC variant in hugetlb code.
> >> So let's leave it in there for now, as it really only matters for some
> >> hugetlb oddities.
> >>
> >> This was found by code inspection.
> >>
> >> Fixes: c1d53b92b95c ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> >> Cc: Dan Williams <dan.j.williams@intel.com>
> >> Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>
> >
> > LGTM, but you need to fix up VMA tests, I attach a patch below to do this. Will
> > this resolved:
>
> Thanks!
>
> I assume that should go into patch #2 instead?

(Sorry missed this on first reply)

It doesn't matter too much from compilation point of view but thought it made
more sense as this is where you pull vma_kernel_pagesize() in? But I'm fine with
either!

Cheers, Lorenzo

Re: [PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by Lorenzo Stoakes (Oracle) 1 month ago

On Fri, Mar 06, 2026 at 02:25:51PM +0100, David Hildenbrand (Arm) wrote:
> On 3/6/26 12:07, Lorenzo Stoakes (Oracle) wrote:
> > On Fri, Mar 06, 2026 at 11:15:57AM +0100, David Hildenbrand (Arm) wrote:
> >> In the past, only hugetlb had special "vma_kernel_pagesize()"
> >> requirements, so it provided its own implementation.
> >>
> >> In commit 05ea88608d4e ("mm, hugetlbfs: introduce ->pagesize() to
> >> vm_operations_struct") we generalized that approach by providing a
> >> vm_ops->pagesize() callback to be used by device-dax.
> >>
> >> Once device-dax started using that callback in commit c1d53b92b95c
> >> ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> >> it was missed that CONFIG_DEV_DAX does not depend on hugetlb support.
> >>
> >> So building a kernel with CONFIG_DEV_DAX but without CONFIG_HUGETLBFS
> >> would not pick up that value.
> >>
> >> Fix it by moving vma_kernel_pagesize() to mm.h, providing only a single
> >> implementation. While at it, improve the kerneldoc a bit.
> >>
> >> Ideally, we'd move vma_mmu_pagesize() as well to the header. However,
> >> its __weak symbol might be overwritten by a PPC variant in hugetlb code.
> >> So let's leave it in there for now, as it really only matters for some
> >> hugetlb oddities.
> >>
> >> This was found by code inspection.
> >>
> >> Fixes: c1d53b92b95c ("device-dax: implement ->pagesize() for smaps to report MMUPageSize")
> >> Cc: Dan Williams <dan.j.williams@intel.com>
> >> Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>
> >
> > LGTM, but you need to fix up VMA tests, I attach a patch below to do this. Will
> > this resolved:
>
> Thanks!
>
> I assume that should go into patch #2 instead?
>
> >
> > Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
>
>
> [...]
>
> > ---
> >  tools/testing/vma/include/dup.h | 7 +++++++
> >  1 file changed, 7 insertions(+)
> >
> > diff --git a/tools/testing/vma/include/dup.h b/tools/testing/vma/include/dup.h
> > index 3078ff1487d3..65b1030a7fdf 100644
> > --- a/tools/testing/vma/include/dup.h
> > +++ b/tools/testing/vma/include/dup.h
> > @@ -1318,3 +1318,10 @@ static inline void vma_set_file(struct vm_area_struct *vma, struct file *file)
> >  	swap(vma->vm_file, file);
> >  	fput(file);
> >  }
> > +
> > +static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
> > +{
> > +	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
> > +		return vma->vm_ops->pagesize(vma);
> > +	return PAGE_SIZE;
>
> Should we just KIS and use PAGE_SIZE for the test?

Yeah that's fine, but then should go in tools/testing/vma/include/custom.h :>)

I tidied things up there to make it easier to understand WTH is going on with
the headers used by VMA tests.

>
> --
> Cheers,
>
> David

Thanks, Lorenzo

Re: [PATCH v1 1/4] mm: move vma_kernel_pagesize() from hugetlb to mm.h

Posted by David Hildenbrand (Arm) 1 month ago

On 3/9/26 14:41, Lorenzo Stoakes (Oracle) wrote:
> On Fri, Mar 06, 2026 at 02:25:51PM +0100, David Hildenbrand (Arm) wrote:
>> On 3/6/26 12:07, Lorenzo Stoakes (Oracle) wrote:
>>>
>>> LGTM, but you need to fix up VMA tests, I attach a patch below to do this. Will
>>> this resolved:
>>
>> Thanks!
>>
>> I assume that should go into patch #2 instead?
>>
>>>
>>> Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
>>
>>
>> [...]
>>
>>> ---
>>>  tools/testing/vma/include/dup.h | 7 +++++++
>>>  1 file changed, 7 insertions(+)
>>>
>>> diff --git a/tools/testing/vma/include/dup.h b/tools/testing/vma/include/dup.h
>>> index 3078ff1487d3..65b1030a7fdf 100644
>>> --- a/tools/testing/vma/include/dup.h
>>> +++ b/tools/testing/vma/include/dup.h
>>> @@ -1318,3 +1318,10 @@ static inline void vma_set_file(struct vm_area_struct *vma, struct file *file)
>>>  	swap(vma->vm_file, file);
>>>  	fput(file);
>>>  }
>>> +
>>> +static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
>>> +{
>>> +	if (unlikely(vma->vm_ops && vma->vm_ops->pagesize))
>>> +		return vma->vm_ops->pagesize(vma);
>>> +	return PAGE_SIZE;
>>
>> Should we just KIS and use PAGE_SIZE for the test?
> 
> Yeah that's fine, but then should go in tools/testing/vma/include/custom.h :>)

I'll place it in there, then, thanks.

-- 
Cheers,

David