[PATCH] mm/z3fold.c: Reinitialize zhdr structs after migration

[PATCH] mm/z3fold.c: Reinitialize zhdr structs after migration

[Henry Burns]

z3fold_page_migration() calls memcpy(new_zhdr, zhdr, PAGE_SIZE).
However, zhdr contains fields that can't be directly coppied over (ex:
list_head, a circular linked list). We only need to initialize the
linked lists in new_zhdr, as z3fold_isolate_page() already ensures
that these lists are empty.

Additionally it is possible that zhdr->work has been placed in a
workqueue. In this case we shouldn't migrate the page, as zhdr->work
references zhdr as opposed to new_zhdr.

Fixes: bba4c5f96ce4 ("mm/z3fold.c: support page migration")
Signed-off-by: Henry Burns <henryburns@google.com>
---
 mm/z3fold.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/mm/z3fold.c b/mm/z3fold.c
index 42ef9955117c..9da471bcab93 100644
--- a/mm/z3fold.c
+++ b/mm/z3fold.c
@@ -1352,12 +1352,22 @@ static int z3fold_page_migrate(struct address_space *mapping, struct page *newpa
 		z3fold_page_unlock(zhdr);
 		return -EBUSY;
 	}
+	if (work_pending(&zhdr->work)) {
+		z3fold_page_unlock(zhdr);
+		return -EAGAIN;
+	}
 	new_zhdr = page_address(newpage);
 	memcpy(new_zhdr, zhdr, PAGE_SIZE);
 	newpage->private = page->private;
 	page->private = 0;
 	z3fold_page_unlock(zhdr);
 	spin_lock_init(&new_zhdr->page_lock);
+	INIT_WORK(&new_zhdr->work, compact_page_work);
+	/*
+	 * z3fold_page_isolate() ensures that this list is empty, so we only
+	 * have to reinitialize it.
+	 */
+	INIT_LIST_HEAD(&new_zhdr->buddy);
 	new_mapping = page_mapping(page);
 	__ClearPageMovable(page);
 	ClearPagePrivate(page);
-- 
2.22.0.510.g264f2c817a-goog

Re: [PATCH] mm/z3fold.c: Reinitialize zhdr structs after migration

[Shakeel Butt]

On Mon, Jul 15, 2019 at 9:47 AM Henry Burns <henryburns@google.com> wrote:
>
> z3fold_page_migration() calls memcpy(new_zhdr, zhdr, PAGE_SIZE).
> However, zhdr contains fields that can't be directly coppied over (ex:
> list_head, a circular linked list). We only need to initialize the
> linked lists in new_zhdr, as z3fold_isolate_page() already ensures
> that these lists are empty.
>
> Additionally it is possible that zhdr->work has been placed in a
> workqueue. In this case we shouldn't migrate the page, as zhdr->work
> references zhdr as opposed to new_zhdr.
>
> Fixes: bba4c5f96ce4 ("mm/z3fold.c: support page migration")
> Signed-off-by: Henry Burns <henryburns@google.com>

Reviewed-by: Shakeel Butt <shakeelb@google.com>

> ---
>  mm/z3fold.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/mm/z3fold.c b/mm/z3fold.c
> index 42ef9955117c..9da471bcab93 100644
> --- a/mm/z3fold.c
> +++ b/mm/z3fold.c
> @@ -1352,12 +1352,22 @@ static int z3fold_page_migrate(struct address_space *mapping, struct page *newpa
>                 z3fold_page_unlock(zhdr);
>                 return -EBUSY;
>         }
> +       if (work_pending(&zhdr->work)) {
> +               z3fold_page_unlock(zhdr);
> +               return -EAGAIN;
> +       }
>         new_zhdr = page_address(newpage);
>         memcpy(new_zhdr, zhdr, PAGE_SIZE);
>         newpage->private = page->private;
>         page->private = 0;
>         z3fold_page_unlock(zhdr);
>         spin_lock_init(&new_zhdr->page_lock);
> +       INIT_WORK(&new_zhdr->work, compact_page_work);
> +       /*
> +        * z3fold_page_isolate() ensures that this list is empty, so we only
> +        * have to reinitialize it.
> +        */
> +       INIT_LIST_HEAD(&new_zhdr->buddy);
>         new_mapping = page_mapping(page);
>         __ClearPageMovable(page);
>         ClearPagePrivate(page);
> --
> 2.22.0.510.g264f2c817a-goog
>

Re: [PATCH] mm/z3fold.c: Reinitialize zhdr structs after migration

[Henry Burns]

>>
>> z3fold_page_migration() calls memcpy(new_zhdr, zhdr, PAGE_SIZE).
>> However, zhdr contains fields that can't be directly coppied over (ex:
>> list_head, a circular linked list). We only need to initialize the
>> linked lists in new_zhdr, as z3fold_isolate_page() already ensures
>> that these lists are empty.
>>
>> Additionally it is possible that zhdr->work has been placed in a
>> workqueue. In this case we shouldn't migrate the page, as zhdr->work
>> references zhdr as opposed to new_zhdr.
>>
>> Fixes: bba4c5f96ce4 ("mm/z3fold.c: support page migration")
>> Signed-off-by: Henry Burns <henryburns@google.com>
>> ---
>>  mm/z3fold.c | 10 ++++++++++
>>  1 file changed, 10 insertions(+)
>>
>> diff --git a/mm/z3fold.c b/mm/z3fold.c
>> index 42ef9955117c..9da471bcab93 100644
>> --- a/mm/z3fold.c
>> +++ b/mm/z3fold.c
>> @@ -1352,12 +1352,22 @@ static int z3fold_page_migrate(struct address_space *mapping, struct page *newpa
>>                 z3fold_page_unlock(zhdr);
>>                 return -EBUSY;
>>         }
>> +       if (work_pending(&zhdr->work)) {
>> +               z3fold_page_unlock(zhdr);
>> +               return -EAGAIN;
>> +       }
>>         new_zhdr = page_address(newpage);
>>         memcpy(new_zhdr, zhdr, PAGE_SIZE);
>>         newpage->private = page->private;
>>         page->private = 0;
>>         z3fold_page_unlock(zhdr);
>>         spin_lock_init(&new_zhdr->page_lock);
>> +       INIT_WORK(&new_zhdr->work, compact_page_work);
>> +       /*
>> +        * z3fold_page_isolate() ensures that this list is empty, so we only
>> +        * have to reinitialize it.
>> +        */
>
>
> On the nitpicking side, we seem to have ensured that directly in migrate :) Looks OK to me otherwise.
Ok, I see it happens in the call to do_compact_page(). Got it, new
patch coming out now.

>
> ~Vitaly
>
>> +       INIT_LIST_HEAD(&new_zhdr->buddy);
>>         new_mapping = page_mapping(page);
>>         __ClearPageMovable(page);
>>         ClearPagePrivate(page);
>> --
>> 2.22.0.510.g264f2c817a-goog
>>

Re: [PATCH] mm/z3fold.c: Reinitialize zhdr structs after migration

[Henry Burns]

Sorry Vitaly, I had the wrong impression from your email that
INIT_LIST_HEAD() was being ensured directly in migrate and got
confused. (I thought you were saying it happened through the call to
do_compact_page() queued).

That being said, I don't see where in migrate new_zhdr->buddy is being
checked. We do check for new_zhdr.work with
 if (work_pending(&zhdr->work)) {
...
}

Is that what you were referring to?