[PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Baolin Wang]

Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page cache
if needed"), ARM64 can support 512MB PMD-sized THP when the base page size is
64KB, which is larger than the maximum supported page cache size MAX_PAGECACHE_ORDER.
This is not expected. To fix this issue, use THP_ORDERS_ALL_FILE_DEFAULT for
shmem to filter allowable huge orders.

Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/shmem.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/shmem.c b/mm/shmem.c
index 2faa9daaf54b..a4332a97558c 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1630,10 +1630,10 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
 	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
 	unsigned long vm_flags = vma->vm_flags;
 	/*
-	 * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
+	 * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
 	 * are enabled for this vma.
 	 */
-	unsigned long orders = BIT(PMD_ORDER + 1) - 1;
+	unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
 	loff_t i_size;
 	int order;
 
-- 
2.39.3

[PATCH 2/2] mm: shmem: fix incorrect aligned index when checking conflicts

[Baolin Wang]

In the shmem_suitable_orders() function, xa_find() is used to check for
conflicts in the pagecache to select suitable huge orders. However, when
checking each huge order in every loop, the aligned index is calculated
from the previous iteration, which may cause suitable huge orders to be
missed.

We should use the original index each time in the loop to calculate a
new aligned index for checking conflicts to avoid this issue.

Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/shmem.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/mm/shmem.c b/mm/shmem.c
index a4332a97558c..6e9836b1bd1d 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1686,6 +1686,7 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
 					   unsigned long orders)
 {
 	struct vm_area_struct *vma = vmf->vma;
+	pgoff_t aligned_index;
 	unsigned long pages;
 	int order;
 
@@ -1697,9 +1698,9 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
 	order = highest_order(orders);
 	while (orders) {
 		pages = 1UL << order;
-		index = round_down(index, pages);
-		if (!xa_find(&mapping->i_pages, &index,
-			     index + pages - 1, XA_PRESENT))
+		aligned_index = round_down(index, pages);
+		if (!xa_find(&mapping->i_pages, &aligned_index,
+			     aligned_index + pages - 1, XA_PRESENT))
 			break;
 		order = next_order(&orders, order);
 	}
-- 
2.39.3

Re: [PATCH 2/2] mm: shmem: fix incorrect aligned index when checking conflicts

[David Hildenbrand]

On 31.07.24 07:46, Baolin Wang wrote:
> In the shmem_suitable_orders() function, xa_find() is used to check for
> conflicts in the pagecache to select suitable huge orders. However, when
> checking each huge order in every loop, the aligned index is calculated
> from the previous iteration, which may cause suitable huge orders to be
> missed.
> 
> We should use the original index each time in the loop to calculate a
> new aligned index for checking conflicts to avoid this issue.
> 
> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
>   mm/shmem.c | 7 ++++---
>   1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index a4332a97558c..6e9836b1bd1d 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1686,6 +1686,7 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
>   					   unsigned long orders)
>   {
>   	struct vm_area_struct *vma = vmf->vma;
> +	pgoff_t aligned_index;
>   	unsigned long pages;
>   	int order;
>   
> @@ -1697,9 +1698,9 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
>   	order = highest_order(orders);
>   	while (orders) {
>   		pages = 1UL << order;
> -		index = round_down(index, pages);
> -		if (!xa_find(&mapping->i_pages, &index,
> -			     index + pages - 1, XA_PRESENT))
> +		aligned_index = round_down(index, pages);
> +		if (!xa_find(&mapping->i_pages, &aligned_index,
> +			     aligned_index + pages - 1, XA_PRESENT))
>   			break;
>   		order = next_order(&orders, order);
>   	}

Acked-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Barry Song]

On Wed, Jul 31, 2024 at 1:46 PM Baolin Wang
<baolin.wang@linux.alibaba.com> wrote:
>
> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page cache
> if needed"), ARM64 can support 512MB PMD-sized THP when the base page size is
> 64KB, which is larger than the maximum supported page cache size MAX_PAGECACHE_ORDER.
> This is not expected. To fix this issue, use THP_ORDERS_ALL_FILE_DEFAULT for
> shmem to filter allowable huge orders.
>
> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>

Reviewed-by: Barry Song <baohua@kernel.org>

> ---
>  mm/shmem.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 2faa9daaf54b..a4332a97558c 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1630,10 +1630,10 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
>         unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
>         unsigned long vm_flags = vma->vm_flags;
>         /*
> -        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
> +        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
>          * are enabled for this vma.

Nit:
THP_ORDERS_ALL_FILE_DEFAULT should be self-explanatory enough.
I feel we don't need this comment?

>          */
> -       unsigned long orders = BIT(PMD_ORDER + 1) - 1;
> +       unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
>         loff_t i_size;
>         int order;
>
> --
> 2.39.3
>

Thanks
Barry

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Baolin Wang]

On 2024/7/31 14:18, Barry Song wrote:
> On Wed, Jul 31, 2024 at 1:46 PM Baolin Wang
> <baolin.wang@linux.alibaba.com> wrote:
>>
>> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page cache
>> if needed"), ARM64 can support 512MB PMD-sized THP when the base page size is
>> 64KB, which is larger than the maximum supported page cache size MAX_PAGECACHE_ORDER.
>> This is not expected. To fix this issue, use THP_ORDERS_ALL_FILE_DEFAULT for
>> shmem to filter allowable huge orders.
>>
>> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> 
> Reviewed-by: Barry Song <baohua@kernel.org>

Thanks for reviewing.

> 
>> ---
>>   mm/shmem.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/shmem.c b/mm/shmem.c
>> index 2faa9daaf54b..a4332a97558c 100644
>> --- a/mm/shmem.c
>> +++ b/mm/shmem.c
>> @@ -1630,10 +1630,10 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
>>          unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
>>          unsigned long vm_flags = vma->vm_flags;
>>          /*
>> -        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
>> +        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
>>           * are enabled for this vma.
> 
> Nit:
> THP_ORDERS_ALL_FILE_DEFAULT should be self-explanatory enough.
> I feel we don't need this comment?

Sure.

Andrew, please help to squash the following changes into this patch. Thanks.

diff --git a/mm/shmem.c b/mm/shmem.c
index 6e9836b1bd1d..432faec21547 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1629,10 +1629,6 @@ unsigned long shmem_allowable_huge_orders(struct 
inode *inode,
         unsigned long mask = READ_ONCE(huge_shmem_orders_always);
         unsigned long within_size_orders = 
READ_ONCE(huge_shmem_orders_within_size);
         unsigned long vm_flags = vma->vm_flags;
-       /*
-        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
-        * are enabled for this vma.
-        */
         unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
         loff_t i_size;
         int order;

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Kefeng Wang]

On 2024/7/31 16:56, Baolin Wang wrote:
> 
> 
> On 2024/7/31 14:18, Barry Song wrote:
>> On Wed, Jul 31, 2024 at 1:46 PM Baolin Wang
>> <baolin.wang@linux.alibaba.com> wrote:
>>>
>>> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page 
>>> cache
>>> if needed"), ARM64 can support 512MB PMD-sized THP when the base page 
>>> size is
>>> 64KB, which is larger than the maximum supported page cache size 
>>> MAX_PAGECACHE_ORDER.
>>> This is not expected. To fix this issue, use 
>>> THP_ORDERS_ALL_FILE_DEFAULT for
>>> shmem to filter allowable huge orders.
>>>
>>> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
>>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>>
>> Reviewed-by: Barry Song <baohua@kernel.org>
> 
> Thanks for reviewing.
> 
>>
>>> ---
>>>   mm/shmem.c | 4 ++--
>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/mm/shmem.c b/mm/shmem.c
>>> index 2faa9daaf54b..a4332a97558c 100644
>>> --- a/mm/shmem.c
>>> +++ b/mm/shmem.c
>>> @@ -1630,10 +1630,10 @@ unsigned long 
>>> shmem_allowable_huge_orders(struct inode *inode,
>>>          unsigned long within_size_orders = 
>>> READ_ONCE(huge_shmem_orders_within_size);
>>>          unsigned long vm_flags = vma->vm_flags;
>>>          /*
>>> -        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
>>> +        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 
>>> that
>>>           * are enabled for this vma.
>>
>> Nit:
>> THP_ORDERS_ALL_FILE_DEFAULT should be self-explanatory enough.
>> I feel we don't need this comment?
> 
> Sure.
> 
> Andrew, please help to squash the following changes into this patch. 
> Thanks.

Maybe drop unsigned long orders too?

diff --git a/mm/shmem.c b/mm/shmem.c
index 6af95f595d6f..8485eb6f2ec4 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1638,11 +1638,6 @@ unsigned long shmem_allowable_huge_orders(struct 
inode *inode,
         unsigned long mask = READ_ONCE(huge_shmem_orders_always);
         unsigned long within_size_orders = 
READ_ONCE(huge_shmem_orders_within_size);
         unsigned long vm_flags = vma ? vma->vm_flags : 0;
-       /*
-        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
-        * are enabled for this vma.
-        */
-       unsigned long orders = BIT(PMD_ORDER + 1) - 1;
         bool global_huge;
         loff_t i_size;
         int order;
@@ -1698,7 +1693,7 @@ unsigned long shmem_allowable_huge_orders(struct 
inode *inode,
         if (global_huge)
                 mask |= READ_ONCE(huge_shmem_orders_inherit);

-       return orders & mask;
+       return THP_ORDERS_ALL_FILE_DEFAULT & mask;
  }

> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 6e9836b1bd1d..432faec21547 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1629,10 +1629,6 @@ unsigned long shmem_allowable_huge_orders(struct 
> inode *inode,
>          unsigned long mask = READ_ONCE(huge_shmem_orders_always);
>          unsigned long within_size_orders = 
> READ_ONCE(huge_shmem_orders_within_size);
>          unsigned long vm_flags = vma->vm_flags;
> -       /*
> -        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
> -        * are enabled for this vma.
> -        */
>          unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
>          loff_t i_size;
>          int order;
> 

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Baolin Wang]

On 2024/7/31 17:59, Kefeng Wang wrote:
> 
> 
> On 2024/7/31 16:56, Baolin Wang wrote:
>>
>>
>> On 2024/7/31 14:18, Barry Song wrote:
>>> On Wed, Jul 31, 2024 at 1:46 PM Baolin Wang
>>> <baolin.wang@linux.alibaba.com> wrote:
>>>>
>>>> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size 
>>>> page cache
>>>> if needed"), ARM64 can support 512MB PMD-sized THP when the base 
>>>> page size is
>>>> 64KB, which is larger than the maximum supported page cache size 
>>>> MAX_PAGECACHE_ORDER.
>>>> This is not expected. To fix this issue, use 
>>>> THP_ORDERS_ALL_FILE_DEFAULT for
>>>> shmem to filter allowable huge orders.
>>>>
>>>> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
>>>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>>>
>>> Reviewed-by: Barry Song <baohua@kernel.org>
>>
>> Thanks for reviewing.
>>
>>>
>>>> ---
>>>>   mm/shmem.c | 4 ++--
>>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/mm/shmem.c b/mm/shmem.c
>>>> index 2faa9daaf54b..a4332a97558c 100644
>>>> --- a/mm/shmem.c
>>>> +++ b/mm/shmem.c
>>>> @@ -1630,10 +1630,10 @@ unsigned long 
>>>> shmem_allowable_huge_orders(struct inode *inode,
>>>>          unsigned long within_size_orders = 
>>>> READ_ONCE(huge_shmem_orders_within_size);
>>>>          unsigned long vm_flags = vma->vm_flags;
>>>>          /*
>>>> -        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
>>>> +        * Check all the (large) orders below MAX_PAGECACHE_ORDER + 
>>>> 1 that
>>>>           * are enabled for this vma.
>>>
>>> Nit:
>>> THP_ORDERS_ALL_FILE_DEFAULT should be self-explanatory enough.
>>> I feel we don't need this comment?
>>
>> Sure.
>>
>> Andrew, please help to squash the following changes into this patch. 
>> Thanks.
> 
> Maybe drop unsigned long orders too?
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 6af95f595d6f..8485eb6f2ec4 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1638,11 +1638,6 @@ unsigned long shmem_allowable_huge_orders(struct 
> inode *inode,
>          unsigned long mask = READ_ONCE(huge_shmem_orders_always);
>          unsigned long within_size_orders = 
> READ_ONCE(huge_shmem_orders_within_size);
>          unsigned long vm_flags = vma ? vma->vm_flags : 0;
> -       /*
> -        * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
> -        * are enabled for this vma.
> -        */
> -       unsigned long orders = BIT(PMD_ORDER + 1) - 1;
>          bool global_huge;
>          loff_t i_size;
>          int order;
> @@ -1698,7 +1693,7 @@ unsigned long shmem_allowable_huge_orders(struct 
> inode *inode,
>          if (global_huge)
>                  mask |= READ_ONCE(huge_shmem_orders_inherit);
> 
> -       return orders & mask;
> +       return THP_ORDERS_ALL_FILE_DEFAULT & mask;
>   }

Yes. Good point. Thanks.
(Hope Andrew can help to squash these changes :))

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Andrew Morton]

On Wed, 31 Jul 2024 18:22:17 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:

> (Hope Andrew can help to squash these changes :))

I'm seeing some rejects against, amongst other things, your own "Some
cleanups for shmem" series.

So... v2, please?

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Baolin Wang]

On 2024/8/1 04:48, Andrew Morton wrote:
> On Wed, 31 Jul 2024 18:22:17 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:
> 
>> (Hope Andrew can help to squash these changes :))
> 
> I'm seeing some rejects against, amongst other things, your own "Some
> cleanups for shmem" series.
> 
> So... v2, please?

These two bugfix patches are based on the mm-hotfixes-unstable branch 
and need to be merged into 6.11-rcX, so they should be queued first.

For the 'Some cleanups for shmem' series, I can send a new V4 version to 
you after resolving conflicts with the shmem bugfix patches. Sorry for 
the inconvenience.

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Andrew Morton]

On Thu, 1 Aug 2024 08:06:59 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:

> 
> 
> On 2024/8/1 04:48, Andrew Morton wrote:
> > On Wed, 31 Jul 2024 18:22:17 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:
> > 
> >> (Hope Andrew can help to squash these changes :))
> > 
> > I'm seeing some rejects against, amongst other things, your own "Some
> > cleanups for shmem" series.
> > 
> > So... v2, please?
> 
> These two bugfix patches are based on the mm-hotfixes-unstable branch 
> and need to be merged into 6.11-rcX, so they should be queued first.

OK.

> For the 'Some cleanups for shmem' series, I can send a new V4 version to 
> you after resolving conflicts with the shmem bugfix patches. Sorry for 
> the inconvenience.

I fixed things up.

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Baolin Wang]

On 2024/8/2 03:55, Andrew Morton wrote:
> On Thu, 1 Aug 2024 08:06:59 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:
> 
>>
>>
>> On 2024/8/1 04:48, Andrew Morton wrote:
>>> On Wed, 31 Jul 2024 18:22:17 +0800 Baolin Wang <baolin.wang@linux.alibaba.com> wrote:
>>>
>>>> (Hope Andrew can help to squash these changes :))
>>>
>>> I'm seeing some rejects against, amongst other things, your own "Some
>>> cleanups for shmem" series.
>>>
>>> So... v2, please?
>>
>> These two bugfix patches are based on the mm-hotfixes-unstable branch
>> and need to be merged into 6.11-rcX, so they should be queued first.
> 
> OK.
> 
>> For the 'Some cleanups for shmem' series, I can send a new V4 version to
>> you after resolving conflicts with the shmem bugfix patches. Sorry for
>> the inconvenience.
> 
> I fixed things up.

Thank you for your help, Andrew. I have checked the patches after 
resolving the conflicts, and all look good.

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[David Hildenbrand]

On 31.07.24 07:46, Baolin Wang wrote:
> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page cache
> if needed"), ARM64 can support 512MB PMD-sized THP when the base page size is
> 64KB, which is larger than the maximum supported page cache size MAX_PAGECACHE_ORDER.
> This is not expected. To fix this issue, use THP_ORDERS_ALL_FILE_DEFAULT for
> shmem to filter allowable huge orders.
> 
> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
>   mm/shmem.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 2faa9daaf54b..a4332a97558c 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1630,10 +1630,10 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
>   	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
>   	unsigned long vm_flags = vma->vm_flags;
>   	/*
> -	 * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
> +	 * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
>   	 * are enabled for this vma.
>   	 */
> -	unsigned long orders = BIT(PMD_ORDER + 1) - 1;
> +	unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
>   	loff_t i_size;
>   	int order;
>   

Acked-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb

Re: [PATCH 1/2] mm: shmem: avoid allocating huge pages larger than MAX_PAGECACHE_ORDER for shmem

[Zi Yan]

[-- Attachment #1: Type: text/plain, Size: 1348 bytes --]

On 31 Jul 2024, at 1:46, Baolin Wang wrote:

> Similar to commit d659b715e94ac ("mm/huge_memory: avoid PMD-size page cache
> if needed"), ARM64 can support 512MB PMD-sized THP when the base page size is
> 64KB, which is larger than the maximum supported page cache size MAX_PAGECACHE_ORDER.
> This is not expected. To fix this issue, use THP_ORDERS_ALL_FILE_DEFAULT for
> shmem to filter allowable huge orders.
>
> Fixes: e7a2ab7b3bb5 ("mm: shmem: add mTHP support for anonymous shmem")
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
>  mm/shmem.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 2faa9daaf54b..a4332a97558c 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1630,10 +1630,10 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
>  	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
>  	unsigned long vm_flags = vma->vm_flags;
>  	/*
> -	 * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
> +	 * Check all the (large) orders below MAX_PAGECACHE_ORDER + 1 that
>  	 * are enabled for this vma.
>  	 */
> -	unsigned long orders = BIT(PMD_ORDER + 1) - 1;
> +	unsigned long orders = THP_ORDERS_ALL_FILE_DEFAULT;
>  	loff_t i_size;
>  	int order;

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

Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]