diff mbox series

[STABLE,v6.1.y] mm/migrate: set swap entry values of THP tail pages properly.

Message ID 20240305161313.90954-1-zi.yan@sent.com (mailing list archive)
State New
Headers show
Series [STABLE,v6.1.y] mm/migrate: set swap entry values of THP tail pages properly. | expand

Commit Message

Zi Yan March 5, 2024, 4:13 p.m. UTC
From: Zi Yan <ziy@nvidia.com>

The tail pages in a THP can have swap entry information stored in their
private field. When migrating to a new page, all tail pages of the new
page need to update ->private to avoid future data corruption.

Signed-off-by: Zi Yan <ziy@nvidia.com>
---
 mm/migrate.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Comments

Zi Yan March 5, 2024, 4:17 p.m. UTC | #1
On 5 Mar 2024, at 11:13, Zi Yan wrote:

> From: Zi Yan <ziy@nvidia.com>
>
> The tail pages in a THP can have swap entry information stored in their
> private field. When migrating to a new page, all tail pages of the new
> page need to update ->private to avoid future data corruption.

Corresponding swapcache entries need to be updated as well.
e71769ae5260 ("mm: enable thp migration for shmem thp") fixed it already.

Fixes: 616b8371539a ("mm: thp: enable thp migration in generic path")


>
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>  mm/migrate.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/mm/migrate.c b/mm/migrate.c
> index c93dd6a31c31..c5968021fde0 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -423,8 +423,12 @@ int folio_migrate_mapping(struct address_space *mapping,
>  	if (folio_test_swapbacked(folio)) {
>  		__folio_set_swapbacked(newfolio);
>  		if (folio_test_swapcache(folio)) {
> +			int i;
> +
>  			folio_set_swapcache(newfolio);
> -			newfolio->private = folio_get_private(folio);
> +			for (i = 0; i < nr; i++)
> +				set_page_private(folio_page(newfolio, i),
> +					page_private(folio_page(folio, i)));
>  		}
>  		entries = nr;
>  	} else {
> -- 
> 2.43.0


--
Best Regards,
Yan, Zi
David Hildenbrand March 5, 2024, 4:22 p.m. UTC | #2
On 05.03.24 17:13, Zi Yan wrote:
> From: Zi Yan <ziy@nvidia.com>
> 
> The tail pages in a THP can have swap entry information stored in their
> private field. When migrating to a new page, all tail pages of the new
> page need to update ->private to avoid future data corruption.
> 
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>   mm/migrate.c | 6 +++++-
>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/migrate.c b/mm/migrate.c
> index c93dd6a31c31..c5968021fde0 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -423,8 +423,12 @@ int folio_migrate_mapping(struct address_space *mapping,
>   	if (folio_test_swapbacked(folio)) {
>   		__folio_set_swapbacked(newfolio);
>   		if (folio_test_swapcache(folio)) {
> +			int i;
> +
>   			folio_set_swapcache(newfolio);
> -			newfolio->private = folio_get_private(folio);
> +			for (i = 0; i < nr; i++)
> +				set_page_private(folio_page(newfolio, i),
> +					page_private(folio_page(folio, i)));
>   		}
>   		entries = nr;
>   	} else {

Hopefully Charan can run the reproducer against this.

Acked-by: David Hildenbrand <david@redhat.com>
Charan Teja Kalla March 5, 2024, 4:22 p.m. UTC | #3
Thanks David for various inputs on this patch!!.

On 3/5/2024 9:47 PM, Zi Yan wrote:
> On 5 Mar 2024, at 11:13, Zi Yan wrote:
> 
>> From: Zi Yan <ziy@nvidia.com>
>>
>> The tail pages in a THP can have swap entry information stored in their
>> private field. When migrating to a new page, all tail pages of the new
>> page need to update ->private to avoid future data corruption.
> 
> Corresponding swapcache entries need to be updated as well.
> e71769ae5260 ("mm: enable thp migration for shmem thp") fixed it already.
> 
> Fixes: 616b8371539a ("mm: thp: enable thp migration in generic path")
> 

Thanks Zi Yan, for posting this patch. I think below tag too applicable?

Closes:
https://lore.kernel.org/linux-mm/1707814102-22682-1-git-send-email-quic_charante@quicinc.com/

> 
>>
>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>> ---
>>  mm/migrate.c | 6 +++++-
>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index c93dd6a31c31..c5968021fde0 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -423,8 +423,12 @@ int folio_migrate_mapping(struct address_space *mapping,
>>  	if (folio_test_swapbacked(folio)) {
>>  		__folio_set_swapbacked(newfolio);
>>  		if (folio_test_swapcache(folio)) {
>> +			int i;
>> +
>>  			folio_set_swapcache(newfolio);
>> -			newfolio->private = folio_get_private(folio);
>> +			for (i = 0; i < nr; i++)
>> +				set_page_private(folio_page(newfolio, i),
>> +					page_private(folio_page(folio, i)));
>>  		}
>>  		entries = nr;
>>  	} else {
>> -- 
>> 2.43.0
> 
> 
> --
> Best Regards,
> Yan, Zi
Zi Yan March 5, 2024, 4:26 p.m. UTC | #4
On 5 Mar 2024, at 11:22, Charan Teja Kalla wrote:

> Thanks David for various inputs on this patch!!.
>
> On 3/5/2024 9:47 PM, Zi Yan wrote:
>> On 5 Mar 2024, at 11:13, Zi Yan wrote:
>>
>>> From: Zi Yan <ziy@nvidia.com>
>>>
>>> The tail pages in a THP can have swap entry information stored in their
>>> private field. When migrating to a new page, all tail pages of the new
>>> page need to update ->private to avoid future data corruption.
>>
>> Corresponding swapcache entries need to be updated as well.
>> e71769ae5260 ("mm: enable thp migration for shmem thp") fixed it already.
>>
>> Fixes: 616b8371539a ("mm: thp: enable thp migration in generic path")
>>
>
> Thanks Zi Yan, for posting this patch. I think below tag too applicable?
>
> Closes:
> https://lore.kernel.org/linux-mm/1707814102-22682-1-git-send-email-quic_charante@quicinc.com/

Right. Let me add it to other stable fixes I am going to send.


Hi Greg,

Could you add the information above (text, Fixes, and Closes) to this patch?
Or do you want me to resend?

Thanks.

>
>>
>>>
>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>> ---
>>>  mm/migrate.c | 6 +++++-
>>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/mm/migrate.c b/mm/migrate.c
>>> index c93dd6a31c31..c5968021fde0 100644
>>> --- a/mm/migrate.c
>>> +++ b/mm/migrate.c
>>> @@ -423,8 +423,12 @@ int folio_migrate_mapping(struct address_space *mapping,
>>>  	if (folio_test_swapbacked(folio)) {
>>>  		__folio_set_swapbacked(newfolio);
>>>  		if (folio_test_swapcache(folio)) {
>>> +			int i;
>>> +
>>>  			folio_set_swapcache(newfolio);
>>> -			newfolio->private = folio_get_private(folio);
>>> +			for (i = 0; i < nr; i++)
>>> +				set_page_private(folio_page(newfolio, i),
>>> +					page_private(folio_page(folio, i)));
>>>  		}
>>>  		entries = nr;
>>>  	} else {
>>> -- 
>>> 2.43.0
>>
>>
>> --
>> Best Regards,
>> Yan, Zi


--
Best Regards,
Yan, Zi
Greg KH March 5, 2024, 10:04 p.m. UTC | #5
On Tue, Mar 05, 2024 at 11:13:13AM -0500, Zi Yan wrote:
> From: Zi Yan <ziy@nvidia.com>
> 
> The tail pages in a THP can have swap entry information stored in their
> private field. When migrating to a new page, all tail pages of the new
> page need to update ->private to avoid future data corruption.
> 
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>  mm/migrate.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)

What is the git commit id of this change in Linus's tree?

thanks,

greg k-h
David Hildenbrand March 5, 2024, 10:09 p.m. UTC | #6
On 05.03.24 23:04, Greg KH wrote:
> On Tue, Mar 05, 2024 at 11:13:13AM -0500, Zi Yan wrote:
>> From: Zi Yan <ziy@nvidia.com>
>>
>> The tail pages in a THP can have swap entry information stored in their
>> private field. When migrating to a new page, all tail pages of the new
>> page need to update ->private to avoid future data corruption.
>>
>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>> ---
>>   mm/migrate.c | 6 +++++-
>>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> What is the git commit id of this change in Linus's tree?

Unfortunately, we had to do stable-only versions, because the backport
of the "accidental" fix that removes the per-subpage "private" 
information would be non-trivial, especially for pre-folio-converison times.

The accidental fix is

07e09c483cbef2a252f75d95670755a0607288f5
Greg KH March 5, 2024, 10:32 p.m. UTC | #7
On Tue, Mar 05, 2024 at 11:09:17PM +0100, David Hildenbrand wrote:
> On 05.03.24 23:04, Greg KH wrote:
> > On Tue, Mar 05, 2024 at 11:13:13AM -0500, Zi Yan wrote:
> > > From: Zi Yan <ziy@nvidia.com>
> > > 
> > > The tail pages in a THP can have swap entry information stored in their
> > > private field. When migrating to a new page, all tail pages of the new
> > > page need to update ->private to avoid future data corruption.
> > > 
> > > Signed-off-by: Zi Yan <ziy@nvidia.com>
> > > ---
> > >   mm/migrate.c | 6 +++++-
> > >   1 file changed, 5 insertions(+), 1 deletion(-)
> > 
> > What is the git commit id of this change in Linus's tree?
> 
> Unfortunately, we had to do stable-only versions, because the backport
> of the "accidental" fix that removes the per-subpage "private" information
> would be non-trivial, especially for pre-folio-converison times.
> 
> The accidental fix is
> 
> 07e09c483cbef2a252f75d95670755a0607288f5

None of that is obvious at all here, we need loads of documentation in
the changelog text that says all of that please.

thanks,

greg k-h
Zi Yan March 5, 2024, 11:13 p.m. UTC | #8
On 5 Mar 2024, at 17:32, Greg KH wrote:

> On Tue, Mar 05, 2024 at 11:09:17PM +0100, David Hildenbrand wrote:
>> On 05.03.24 23:04, Greg KH wrote:
>>> On Tue, Mar 05, 2024 at 11:13:13AM -0500, Zi Yan wrote:
>>>> From: Zi Yan <ziy@nvidia.com>
>>>>
>>>> The tail pages in a THP can have swap entry information stored in their
>>>> private field. When migrating to a new page, all tail pages of the new
>>>> page need to update ->private to avoid future data corruption.
>>>>
>>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>>> ---
>>>>   mm/migrate.c | 6 +++++-
>>>>   1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> What is the git commit id of this change in Linus's tree?
>>
>> Unfortunately, we had to do stable-only versions, because the backport
>> of the "accidental" fix that removes the per-subpage "private" information
>> would be non-trivial, especially for pre-folio-converison times.
>>
>> The accidental fix is
>>
>> 07e09c483cbef2a252f75d95670755a0607288f5
>
> None of that is obvious at all here, we need loads of documentation in
> the changelog text that says all of that please.

How about?

Before 07e09c483cbe ("mm/huge_memory: work on folio->swap instead of
page->private when splitting folio"), when a THP is added into swapcache,
each of its subpages has its own swapcache entry and need ->private pointing
to the right swapcache entry. THP added to swapcache function is added in
38d8b4e6bdc87 ("mm, THP, swap: delay splitting THP during swap out").
When THP migration was added in 616b8371539a6 ("mm: thp: enable thp migration in generic path"), it did not take care of swapcached THP's subpages,
neither updated subpage's ->private nor replaced subpage pointer in
the swapcache. Later, e71769ae5260 ("mm: enable thp migration for shmem thp")
fixed swapcache update part. Now this patch fixes the subpage ->private
update part.



--
Best Regards,
Yan, Zi
Greg KH March 6, 2024, 1:51 p.m. UTC | #9
On Tue, Mar 05, 2024 at 06:13:39PM -0500, Zi Yan wrote:
> On 5 Mar 2024, at 17:32, Greg KH wrote:
> 
> > On Tue, Mar 05, 2024 at 11:09:17PM +0100, David Hildenbrand wrote:
> >> On 05.03.24 23:04, Greg KH wrote:
> >>> On Tue, Mar 05, 2024 at 11:13:13AM -0500, Zi Yan wrote:
> >>>> From: Zi Yan <ziy@nvidia.com>
> >>>>
> >>>> The tail pages in a THP can have swap entry information stored in their
> >>>> private field. When migrating to a new page, all tail pages of the new
> >>>> page need to update ->private to avoid future data corruption.
> >>>>
> >>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
> >>>> ---
> >>>>   mm/migrate.c | 6 +++++-
> >>>>   1 file changed, 5 insertions(+), 1 deletion(-)
> >>>
> >>> What is the git commit id of this change in Linus's tree?
> >>
> >> Unfortunately, we had to do stable-only versions, because the backport
> >> of the "accidental" fix that removes the per-subpage "private" information
> >> would be non-trivial, especially for pre-folio-converison times.
> >>
> >> The accidental fix is
> >>
> >> 07e09c483cbef2a252f75d95670755a0607288f5
> >
> > None of that is obvious at all here, we need loads of documentation in
> > the changelog text that says all of that please.
> 
> How about?
> 
> Before 07e09c483cbe ("mm/huge_memory: work on folio->swap instead of
> page->private when splitting folio"), when a THP is added into swapcache,
> each of its subpages has its own swapcache entry and need ->private pointing
> to the right swapcache entry. THP added to swapcache function is added in
> 38d8b4e6bdc87 ("mm, THP, swap: delay splitting THP during swap out").
> When THP migration was added in 616b8371539a6 ("mm: thp: enable thp migration in generic path"), it did not take care of swapcached THP's subpages,
> neither updated subpage's ->private nor replaced subpage pointer in
> the swapcache. Later, e71769ae5260 ("mm: enable thp migration for shmem thp")
> fixed swapcache update part. Now this patch fixes the subpage ->private
> update part.

That's better than what is there now :)

So yes, please resend all of these with the new text and then we can
queue them up.

thanks,

greg k-h
diff mbox series

Patch

diff --git a/mm/migrate.c b/mm/migrate.c
index c93dd6a31c31..c5968021fde0 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -423,8 +423,12 @@  int folio_migrate_mapping(struct address_space *mapping,
 	if (folio_test_swapbacked(folio)) {
 		__folio_set_swapbacked(newfolio);
 		if (folio_test_swapcache(folio)) {
+			int i;
+
 			folio_set_swapcache(newfolio);
-			newfolio->private = folio_get_private(folio);
+			for (i = 0; i < nr; i++)
+				set_page_private(folio_page(newfolio, i),
+					page_private(folio_page(folio, i)));
 		}
 		entries = nr;
 	} else {