Re: [PATCH] mm/rmap: fix BUG at rmap_walk

[Wanpeng Li <liwanp@linux.vnet.ibm.com>]

Hi Andrew,
On Wed, Dec 18, 2013 at 04:28:58PM -0800, Andrew Morton wrote:
>On Thu, 19 Dec 2013 08:16:35 +0800 Wanpeng Li <liwanp@linux.vnet.ibm.com> wrote:
>
>> page_get_anon_vma() called in page_referenced_anon() will lock and 
>> increase the refcount of anon_vma, page won't be locked for anonymous 
>> page. This patch fix it by skip check anonymous page locked.
>> 
>> [  588.698828] kernel BUG at mm/rmap.c:1663!
>
>Why is all this suddenly happening.  Did we change something, or did a
>new test get added to trinity?
>

They are introduced by Joonsoo's rmap_walk.

>> --- a/mm/rmap.c
>> +++ b/mm/rmap.c
>> @@ -1660,7 +1660,8 @@ done:
>>  
>>  int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
>>  {
>> -	VM_BUG_ON(!PageLocked(page));
>> +	if (!PageAnon(page) || PageKsm(page))
>> +		VM_BUG_ON(!PageLocked(page));
>>  
>>  	if (unlikely(PageKsm(page)))
>>  		return rmap_walk_ksm(page, rwc);
>
>Is there any reason why rmap_walk_ksm() and rmap_walk_file() *need*
>PageLocked() whereas rmap_walk_anon() does not?  If so, let's implement
>it like this:

All callsites of rmap_walk() 

try_to_unmap() page should be lockecd (checked in rmap_walk()) 
try_to_munlock() pages should be locked (checked in try_to_munlock())
page_referenced() pages should be locked except anonymous page (checked in rmap_walk())
page_mkclean() pages should be locked (checked in page_mkclean())
remove_migration_ptes() pages should be locked (checked in rmap_walk())

We can move PageLocked(page) check to the callsites instead of in
rmap_walk() since anonymous page is not locked in page_referenced().

Regards,
Wanpeng Li 

>
>
>--- a/mm/rmap.c~a
>+++ a/mm/rmap.c
>@@ -1716,6 +1716,10 @@ static int rmap_walk_file(struct page *p
> 	struct vm_area_struct *vma;
> 	int ret = SWAP_AGAIN;
>
>+	/*
>+	 * page must be locked because <reason goes here>
>+	 */
>+	VM_BUG_ON(!PageLocked(page));
> 	if (!mapping)
> 		return ret;
> 	mutex_lock(&mapping->i_mmap_mutex);
>@@ -1737,8 +1741,6 @@ static int rmap_walk_file(struct page *p
> int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
> 		struct vm_area_struct *, unsigned long, void *), void *arg)
> {
>-	VM_BUG_ON(!PageLocked(page));
>-
> 	if (unlikely(PageKsm(page)))
> 		return rmap_walk_ksm(page, rmap_one, arg);
> 	else if (PageAnon(page))
>--- a/mm/ksm.c~a
>+++ a/mm/ksm.c
>@@ -2006,6 +2006,9 @@ int rmap_walk_ksm(struct page *page, int
> 	int search_new_forks = 0;
>
> 	VM_BUG_ON(!PageKsm(page));
>+	/*
>+	 * page must be locked because <reason goes here>
>+	 */
> 	VM_BUG_ON(!PageLocked(page));
>
> 	stable_node = page_stable_node(page);
>
>
>Or if there is no reason why the page must be locked for
>rmap_walk_ksm() and rmap_walk_file(), let's just remove rmap_walk()'s
>VM_BUG_ON()?  And rmap_walk_ksm()'s as well - it's duplicative anyway.
>
>--
>To unsubscribe, send a message with 'unsubscribe linux-mm' in
>the body to majordomo@kvack.org.  For more info on Linux MM,
>see: http://www.linux-mm.org/ .
>Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>