[PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[Ritesh Harjani (IBM)]

Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
built with CONFIG_DEBUG_VM triggers this warning on every munmap:

    WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
    NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
    LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
    Call Trace:
     zap_huge_pud+0x170/0x2d0 (unreliable)
     __zap_vma_range+0x1604/0x1e20
     unmap_vmas+0xa0/0x1f0
     unmap_region+0xb0/0x130
     vms_complete_munmap_vmas+0x120/0x310
     do_vmi_align_munmap+0x1ec/0x240
     __vm_munmap+0x148/0x220
     sys_munmap+0x28/0x40

To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
byte so that the whole 1G is mapped by a single PUD and then munmap()
it.

PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
the page size for which tlb flush is needed. zap_huge_pmd(),
zap_pte_range(), change_huge_pud() etc all do this already. However,
since commit:
  dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")

a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
tlb->page_size is still whatever the previous user left behind, which is
0 for the fresh gather set up by munmap() and so the check fires.

Fix this in the same way like how other similar paths do.

Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
Reported-by: Pavithra <pavrampu@linux.ibm.com>
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
---
[v1]->v2:
=========
- Split the 2 part patch series. Patch-1 is now superseded by a different patch
  from Muchun which he already posted [2].
- No difference in this patch from v1.

[v1]: https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
[2]: https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/

 mm/huge_memory.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index bd44d58e8684..7e589a2fb02d 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -3040,6 +3040,8 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	spinlock_t *ptl;
 	pud_t orig_pud;

+	tlb_change_page_size(tlb, HPAGE_PUD_SIZE);
+
 	ptl = __pud_trans_huge_lock(pud, vma);
 	if (!ptl)
 		return 0;
--
2.39.5

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[Zi Yan]

On 29 Oct 2026, at 20:07, Ritesh Harjani (IBM) wrote:

> Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
>
>     WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
>     NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>     LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>     Call Trace:
>      zap_huge_pud+0x170/0x2d0 (unreliable)
>      __zap_vma_range+0x1604/0x1e20
>      unmap_vmas+0xa0/0x1f0
>      unmap_region+0xb0/0x130
>      vms_complete_munmap_vmas+0x120/0x310
>      do_vmi_align_munmap+0x1ec/0x240
>      __vm_munmap+0x148/0x220
>      sys_munmap+0x28/0x40
>
> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
> byte so that the whole 1G is mapped by a single PUD and then munmap()
> it.
>
> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
> the page size for which tlb flush is needed. zap_huge_pmd(),
> zap_pte_range(), change_huge_pud() etc all do this already. However,
> since commit:
>   dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
>
> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
> tlb->page_size is still whatever the previous user left behind, which is
> 0 for the fresh gather set up by munmap() and so the check fires.
>
> Fix this in the same way like how other similar paths do.
>
> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
> Reported-by: Pavithra <pavrampu@linux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> ---
> [v1]->v2:
> =========
> - Split the 2 part patch series. Patch-1 is now superseded by a different patch
>   from Muchun which he already posted [2].
> - No difference in this patch from v1.
>
> [v1]: https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
> [2]: https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/
>
>  mm/huge_memory.c | 2 ++
>  1 file changed, 2 insertions(+)
>

Makes sense. Thanks.

Acked-by: Zi Yan <ziy@nvidia.com>


Best Regards,
Yan, Zi

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[Muchun Song]

> On Oct 30, 2026, at 08:07, Ritesh Harjani (IBM) <ritesh.list@gmail.com> wrote:
> 
> Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
> 
>    WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
>    NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>    LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>    Call Trace:
>     zap_huge_pud+0x170/0x2d0 (unreliable)
>     __zap_vma_range+0x1604/0x1e20
>     unmap_vmas+0xa0/0x1f0
>     unmap_region+0xb0/0x130
>     vms_complete_munmap_vmas+0x120/0x310
>     do_vmi_align_munmap+0x1ec/0x240
>     __vm_munmap+0x148/0x220
>     sys_munmap+0x28/0x40
> 
> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
> byte so that the whole 1G is mapped by a single PUD and then munmap()
> it.
> 
> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
> the page size for which tlb flush is needed. zap_huge_pmd(),
> zap_pte_range(), change_huge_pud() etc all do this already. However,
> since commit:
>  dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
> 
> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
> tlb->page_size is still whatever the previous user left behind, which is
> 0 for the fresh gather set up by munmap() and so the check fires.
> 
> Fix this in the same way like how other similar paths do.
> 
> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
> Reported-by: Pavithra <pavrampu@linux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>

Acked-by: Muchun Song <muchun.song@linux.dev>

Thanks.

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[pavrampu]

On 2026-10-30 05:37, Ritesh Harjani (IBM) wrote:
> Unmapping a device-dax mapping with 1G alignment on powerpc radix 
> kernel
> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
> 
>     WARNING: mm/mmu_gather.c:176 at 
> __tlb_remove_folio_pages_size+0x38/0x230
>     NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>     LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>     Call Trace:
>      zap_huge_pud+0x170/0x2d0 (unreliable)
>      __zap_vma_range+0x1604/0x1e20
>      unmap_vmas+0xa0/0x1f0
>      unmap_region+0xb0/0x130
>      vms_complete_munmap_vmas+0x120/0x310
>      do_vmi_align_munmap+0x1ec/0x240
>      __vm_munmap+0x148/0x220
>      sys_munmap+0x28/0x40
> 
> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
> byte so that the whole 1G is mapped by a single PUD and then munmap()
> it.
> 
> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
> the page size for which tlb flush is needed. zap_huge_pmd(),
> zap_pte_range(), change_huge_pud() etc all do this already. However,
> since commit:
>   dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
> 
> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
> tlb->page_size is still whatever the previous user left behind, which 
> is
> 0 for the fresh gather set up by munmap() and so the check fires.
> 
> Fix this in the same way like how other similar paths do.
> 
> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
> Reported-by: Pavithra <pavrampu@linux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> ---
> [v1]->v2:
> =========
> - Split the 2 part patch series. Patch-1 is now superseded by a 
> different patch
>   from Muchun which he already posted [2].
> - No difference in this patch from v1.
> 
> [v1]:
> https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
> [2]:
> https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/
> 
>  mm/huge_memory.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index bd44d58e8684..7e589a2fb02d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3040,6 +3040,8 @@ int zap_huge_pud(struct mmu_gather *tlb, struct
> vm_area_struct *vma,
>  	spinlock_t *ptl;
>  	pud_t orig_pud;
> 
> +	tlb_change_page_size(tlb, HPAGE_PUD_SIZE);
> +
>  	ptl = __pud_trans_huge_lock(pud, vma);
>  	if (!ptl)
>  		return 0;
> --
> 2.39.5

Tested on Power10 (ppc64le), kernel v7.3

Tested-by: Pavithra Prakash <pavrampu@linux.ibm.com>

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[David Hildenbrand (Arm)]

On 10/30/26 01:07, Ritesh Harjani (IBM) wrote:
> Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
> 
>     WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
>     NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>     LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>     Call Trace:
>      zap_huge_pud+0x170/0x2d0 (unreliable)
>      __zap_vma_range+0x1604/0x1e20
>      unmap_vmas+0xa0/0x1f0
>      unmap_region+0xb0/0x130
>      vms_complete_munmap_vmas+0x120/0x310
>      do_vmi_align_munmap+0x1ec/0x240
>      __vm_munmap+0x148/0x220
>      sys_munmap+0x28/0x40
> 
> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
> byte so that the whole 1G is mapped by a single PUD and then munmap()
> it.
> 
> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
> the page size for which tlb flush is needed. zap_huge_pmd(),
> zap_pte_range(), change_huge_pud() etc all do this already. However,
> since commit:
>   dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
> 
> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
> tlb->page_size is still whatever the previous user left behind, which is
> 0 for the fresh gather set up by munmap() and so the check fires.
> 
> Fix this in the same way like how other similar paths do.
> 
> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
> Reported-by: Pavithra <pavrampu@linux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> ---
> [v1]->v2:
> =========
> - Split the 2 part patch series. Patch-1 is now superseded by a different patch
>   from Muchun which he already posted [2].
> - No difference in this patch from v1.
> 
> [v1]: https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
> [2]: https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/
> 
>  mm/huge_memory.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index bd44d58e8684..7e589a2fb02d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3040,6 +3040,8 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
>  	spinlock_t *ptl;
>  	pud_t orig_pud;
> 
> +	tlb_change_page_size(tlb, HPAGE_PUD_SIZE);
> +
>  	ptl = __pud_trans_huge_lock(pud, vma);
>  	if (!ptl)
>  		return 0;
> --
> 2.39.5
> 


Matches zap_huge_pmd().

I will wait a bit  and then queue this shortly.

This sounds like stable material?

-- 
Cheers,

David

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[Ritesh Harjani]

"David Hildenbrand (Arm)" <david@kernel.org> writes:

> On 10/30/26 01:07, Ritesh Harjani (IBM) wrote:
>> Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
>> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
>> 
>>     WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
>>     NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>>     LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>>     Call Trace:
>>      zap_huge_pud+0x170/0x2d0 (unreliable)
>>      __zap_vma_range+0x1604/0x1e20
>>      unmap_vmas+0xa0/0x1f0
>>      unmap_region+0xb0/0x130
>>      vms_complete_munmap_vmas+0x120/0x310
>>      do_vmi_align_munmap+0x1ec/0x240
>>      __vm_munmap+0x148/0x220
>>      sys_munmap+0x28/0x40
>> 
>> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
>> byte so that the whole 1G is mapped by a single PUD and then munmap()
>> it.
>> 
>> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
>> the page size for which tlb flush is needed. zap_huge_pmd(),
>> zap_pte_range(), change_huge_pud() etc all do this already. However,
>> since commit:
>>   dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
>> 
>> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
>> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
>> tlb->page_size is still whatever the previous user left behind, which is
>> 0 for the fresh gather set up by munmap() and so the check fires.
>> 
>> Fix this in the same way like how other similar paths do.
>> 
>> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
>> Reported-by: Pavithra <pavrampu@linux.ibm.com>
>> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
>> ---
>> [v1]->v2:
>> =========
>> - Split the 2 part patch series. Patch-1 is now superseded by a different patch
>>   from Muchun which he already posted [2].
>> - No difference in this patch from v1.
>> 
>> [v1]: https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
>> [2]: https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/
>> 
>>  mm/huge_memory.c | 2 ++
>>  1 file changed, 2 insertions(+)
>> 
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index bd44d58e8684..7e589a2fb02d 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -3040,6 +3040,8 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
>>  	spinlock_t *ptl;
>>  	pud_t orig_pud;
>> 
>> +	tlb_change_page_size(tlb, HPAGE_PUD_SIZE);
>> +
>>  	ptl = __pud_trans_huge_lock(pud, vma);
>>  	if (!ptl)
>>  		return 0;
>> --
>> 2.39.5
>> 
>
>
> Matches zap_huge_pmd().
>
> I will wait a bit  and then queue this shortly.
>

Thanks David, however, looks like Andrew might have queued this up in
his mm tree [1], right?

[1]: https://lore.kernel.org/all/20261030021905.2DCA71F000C1@smtp.kernel.org/

> This sounds like stable material?
>

Sorry, my bad. Yes, we should have this Cc: stable@vger.kernel.org.

Kindly let me know if I should update the tag and send v3, or can it be
taken care? Thanks!

-ritesh

Re: [PATCHv2] mm/huge_memory: set the mmu_gather page size in zap_huge_pud()

[David Hildenbrand (Arm)]

On 10/30/26 11:15, Ritesh Harjani (IBM) wrote:
> "David Hildenbrand (Arm)" <david@kernel.org> writes:
> 
>> On 10/30/26 01:07, Ritesh Harjani (IBM) wrote:
>>> Unmapping a device-dax mapping with 1G alignment on powerpc radix kernel
>>> built with CONFIG_DEBUG_VM triggers this warning on every munmap:
>>>
>>>     WARNING: mm/mmu_gather.c:176 at __tlb_remove_folio_pages_size+0x38/0x230
>>>     NIP [c00000000063d508] __tlb_remove_folio_pages_size+0x38/0x230
>>>     LR [c0000000006fb688] zap_huge_pud+0x1b8/0x2d0
>>>     Call Trace:
>>>      zap_huge_pud+0x170/0x2d0 (unreliable)
>>>      __zap_vma_range+0x1604/0x1e20
>>>      unmap_vmas+0xa0/0x1f0
>>>      unmap_region+0xb0/0x130
>>>      vms_complete_munmap_vmas+0x120/0x310
>>>      do_vmi_align_munmap+0x1ec/0x240
>>>      __vm_munmap+0x148/0x220
>>>      sys_munmap+0x28/0x40
>>>
>>> To trigger this - it is enough to mmap() /dev/dax0.0, touch the first
>>> byte so that the whole 1G is mapped by a single PUD and then munmap()
>>> it.
>>>
>>> PowerPC needs MMU_GATHER_PAGE_SIZE in which mmu_gather needs to record
>>> the page size for which tlb flush is needed. zap_huge_pmd(),
>>> zap_pte_range(), change_huge_pud() etc all do this already. However,
>>> since commit:
>>>   dbe54153296d ("mm/huge_memory: add vmf_insert_folio_pud()")
>>>
>>> a device-dax PUD maps a real folio and zap_huge_pud() now hands it to
>>> tlb_remove_page_size(tlb, page, HPAGE_PUD_SIZE). At that point
>>> tlb->page_size is still whatever the previous user left behind, which is
>>> 0 for the fresh gather set up by munmap() and so the check fires.
>>>
>>> Fix this in the same way like how other similar paths do.
>>>
>>> Fixes: dbe54153296d0 ("mm/huge_memory: add vmf_insert_folio_pud()")
>>> Reported-by: Pavithra <pavrampu@linux.ibm.com>
>>> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
>>> ---
>>> [v1]->v2:
>>> =========
>>> - Split the 2 part patch series. Patch-1 is now superseded by a different patch
>>>   from Muchun which he already posted [2].
>>> - No difference in this patch from v1.
>>>
>>> [v1]: https://lore.kernel.org/all/97f85592bb78a273be8519e6897e7f9b9bd31469.1793070429.git.ritesh.list@gmail.com/
>>> [2]: https://lore.kernel.org/all/20261029063601.1961415-1-songmuchun@bytedance.com/
>>>
>>>  mm/huge_memory.c | 2 ++
>>>  1 file changed, 2 insertions(+)
>>>
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index bd44d58e8684..7e589a2fb02d 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -3040,6 +3040,8 @@ int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,
>>>  	spinlock_t *ptl;
>>>  	pud_t orig_pud;
>>>
>>> +	tlb_change_page_size(tlb, HPAGE_PUD_SIZE);
>>> +
>>>  	ptl = __pud_trans_huge_lock(pud, vma);
>>>  	if (!ptl)
>>>  		return 0;
>>> --
>>> 2.39.5
>>>
>>
>>
>> Matches zap_huge_pmd().
>>
>> I will wait a bit  and then queue this shortly.
>>
> 
> Thanks David, however, looks like Andrew might have queued this up in
> his mm tree [1], right?
> 
> [1]: https://lore.kernel.org/all/20261030021905.2DCA71F000C1@smtp.kernel.org/

No worries, one of us can drop it before sending it upstream.

I'm about to send out the MAINTAINTERS update in a bit, so right now it's still
confusing.

(but also there is no harm in having it in two trees temporarily)

> 
>> This sounds like stable material?
>>
> 
> Sorry, my bad. Yes, we should have this Cc: stable@vger.kernel.org.
> 
> Kindly let me know if I should update the tag and send v3, or can it be
> taken care? Thanks!

I would just apply that when picking, no need to resend.

Thanks!

-- 
Cheers,

David