linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
* [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
@ 2014-06-30 22:48 Hugh Dickins
  2014-06-30 23:02 ` Andrew Morton
  2014-07-02 22:37 ` Johannes Weiner
  0 siblings, 2 replies; 6+ messages in thread
From: Hugh Dickins @ 2014-06-30 22:48 UTC (permalink / raw)
  To: Johannes Weiner; +Cc: Andrew Morton, Michal Hocko, linux-mm, linux-kernel

Under shmem swapping and swapoff load, I sometimes hit the
VM_BUG_ON_PAGE(!page->mapping) in mem_cgroup_commit_charge() at
mm/memcontrol.c:6502!  Each time it has been a call from shmem_unuse().

Yes, there are some cases (most commonly when the page being unswapped
is in a file being unlinked and evicted at that time) when the charge
should not be committed.  In the old scheme, the page got uncharged
again on release; but in the new scheme, it hits that BUG beforehand.

It's a useful BUG, so adapt shmem_unuse() to allow for it.  Which needs
more info from shmem_unuse_inode(): so abuse -EAGAIN internally to
replace the previous !found state (-ENOENT would be a more natural
code, but that's exactly what you get when the swap has been evicted).

Signed-off-by: Hugh Dickins <hughd@google.com>
---

 mm/shmem.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

--- 3.16-rc2-mm1/mm/shmem.c	2014-06-25 18:43:59.868588121 -0700
+++ linux/mm/shmem.c	2014-06-30 15:05:50.736335600 -0700
@@ -611,7 +611,7 @@ static int shmem_unuse_inode(struct shme
 	radswap = swp_to_radix_entry(swap);
 	index = radix_tree_locate_item(&mapping->page_tree, radswap);
 	if (index == -1)
-		return 0;
+		return -EAGAIN;
 
 	/*
 	 * Move _head_ to start search for next from here.
@@ -670,7 +670,6 @@ static int shmem_unuse_inode(struct shme
 			spin_unlock(&info->lock);
 			swap_free(swap);
 		}
-		error = 1;	/* not an error, but entry was found */
 	}
 	return error;
 }
@@ -683,7 +682,6 @@ int shmem_unuse(swp_entry_t swap, struct
 	struct list_head *this, *next;
 	struct shmem_inode_info *info;
 	struct mem_cgroup *memcg;
-	int found = 0;
 	int error = 0;
 
 	/*
@@ -702,22 +700,24 @@ int shmem_unuse(swp_entry_t swap, struct
 	if (error)
 		goto out;
 	/* No radix_tree_preload: swap entry keeps a place for page in tree */
+	error = -EAGAIN;
 
 	mutex_lock(&shmem_swaplist_mutex);
 	list_for_each_safe(this, next, &shmem_swaplist) {
 		info = list_entry(this, struct shmem_inode_info, swaplist);
 		if (info->swapped)
-			found = shmem_unuse_inode(info, swap, &page);
+			error = shmem_unuse_inode(info, swap, &page);
 		else
 			list_del_init(&info->swaplist);
 		cond_resched();
-		if (found)
+		if (error != -EAGAIN)
 			break;
 	}
 	mutex_unlock(&shmem_swaplist_mutex);
 
-	if (found < 0) {
-		error = found;
+	if (error) {
+		if (error != -ENOMEM)
+			error = 0;
 		mem_cgroup_cancel_charge(page, memcg);
 	} else
 		mem_cgroup_commit_charge(page, memcg, true);

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
  2014-06-30 22:48 [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse Hugh Dickins
@ 2014-06-30 23:02 ` Andrew Morton
  2014-07-01  0:10   ` Hugh Dickins
  2014-07-02 22:37 ` Johannes Weiner
  1 sibling, 1 reply; 6+ messages in thread
From: Andrew Morton @ 2014-06-30 23:02 UTC (permalink / raw)
  To: Hugh Dickins; +Cc: Johannes Weiner, Michal Hocko, linux-mm, linux-kernel

On Mon, 30 Jun 2014 15:48:39 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:

> Under shmem swapping and swapoff load, I sometimes hit the
> VM_BUG_ON_PAGE(!page->mapping) in mem_cgroup_commit_charge() at
> mm/memcontrol.c:6502!  Each time it has been a call from shmem_unuse().
> 
> Yes, there are some cases (most commonly when the page being unswapped
> is in a file being unlinked and evicted at that time) when the charge
> should not be committed.  In the old scheme, the page got uncharged
> again on release; but in the new scheme, it hits that BUG beforehand.
> 
> It's a useful BUG, so adapt shmem_unuse() to allow for it.  Which needs
> more info from shmem_unuse_inode(): so abuse -EAGAIN internally to
> replace the previous !found state (-ENOENT would be a more natural
> code, but that's exactly what you get when the swap has been evicted).
> 
> ...
>
> --- 3.16-rc2-mm1/mm/shmem.c	2014-06-25 18:43:59.868588121 -0700
> +++ linux/mm/shmem.c	2014-06-30 15:05:50.736335600 -0700
> @@ -611,7 +611,7 @@ static int shmem_unuse_inode(struct shme
>  	radswap = swp_to_radix_entry(swap);
>  	index = radix_tree_locate_item(&mapping->page_tree, radswap);
>  	if (index == -1)
> -		return 0;
> +		return -EAGAIN;

Maybe it's time to document the shmem_unuse_inode() return values.

>  	/*
>  	 * Move _head_ to start search for next from here.
> @@ -670,7 +670,6 @@ static int shmem_unuse_inode(struct shme
>  			spin_unlock(&info->lock);
>  			swap_free(swap);
>  		}
> -		error = 1;	/* not an error, but entry was found */
>  	}
>  	return error;
>  }
> @@ -683,7 +682,6 @@ int shmem_unuse(swp_entry_t swap, struct
>  	struct list_head *this, *next;
>  	struct shmem_inode_info *info;
>  	struct mem_cgroup *memcg;
> -	int found = 0;
>  	int error = 0;
>  
>  	/*
> @@ -702,22 +700,24 @@ int shmem_unuse(swp_entry_t swap, struct
>  	if (error)
>  		goto out;
>  	/* No radix_tree_preload: swap entry keeps a place for page in tree */
> +	error = -EAGAIN;
>  
>  	mutex_lock(&shmem_swaplist_mutex);
>  	list_for_each_safe(this, next, &shmem_swaplist) {
>  		info = list_entry(this, struct shmem_inode_info, swaplist);
>  		if (info->swapped)
> -			found = shmem_unuse_inode(info, swap, &page);
> +			error = shmem_unuse_inode(info, swap, &page);
>  		else
>  			list_del_init(&info->swaplist);
>  		cond_resched();
> -		if (found)
> +		if (error != -EAGAIN)
>  			break;
>  	}
>  	mutex_unlock(&shmem_swaplist_mutex);
>  
> -	if (found < 0) {
> -		error = found;
> +	if (error) {
> +		if (error != -ENOMEM)
> +			error = 0;
>  		mem_cgroup_cancel_charge(page, memcg);
>  	} else
>  		mem_cgroup_commit_charge(page, memcg, true);

If I'm reading this correctly, shmem_unuse() can now return -EAGAIN and
that can get all the way back to userspace.  `man 2 swapoff' doesn't
know this...

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
  2014-06-30 23:02 ` Andrew Morton
@ 2014-07-01  0:10   ` Hugh Dickins
  2014-07-01  0:34     ` Andrew Morton
  0 siblings, 1 reply; 6+ messages in thread
From: Hugh Dickins @ 2014-07-01  0:10 UTC (permalink / raw)
  To: Andrew Morton
  Cc: Hugh Dickins, Johannes Weiner, Michal Hocko, linux-mm,
	linux-kernel

On Mon, 30 Jun 2014, Andrew Morton wrote:
> On Mon, 30 Jun 2014 15:48:39 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:
> > -		return 0;
> > +		return -EAGAIN;
> 
> Maybe it's time to document the shmem_unuse_inode() return values.

Oh dear.  I had hoped they would look after themselves.  This one is a
private matter between shmem_unuse_inode and its one caller, just below.

> > +	if (error) {
> > +		if (error != -ENOMEM)
> > +			error = 0;
> >  		mem_cgroup_cancel_charge(page, memcg);
> >  	} else
> >  		mem_cgroup_commit_charge(page, memcg, true);
> 
> If I'm reading this correctly, shmem_unuse() can now return -EAGAIN and
> that can get all the way back to userspace.  `man 2 swapoff' doesn't
> know this...

if (error) {
	if (error != -ENOMEM)
		error = 0;
...
	return error;

So the only values returned from shmem_unuse_inode() to its caller
try_to_unuse() are 0 and -ENOMEM.  Those may get passed back to the
user, but -EAGAIN was just an internal shmem.c detail.

Hugh

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
  2014-07-01  0:10   ` Hugh Dickins
@ 2014-07-01  0:34     ` Andrew Morton
  2014-07-01  1:06       ` Hugh Dickins
  0 siblings, 1 reply; 6+ messages in thread
From: Andrew Morton @ 2014-07-01  0:34 UTC (permalink / raw)
  To: Hugh Dickins; +Cc: Johannes Weiner, Michal Hocko, linux-mm, linux-kernel

On Mon, 30 Jun 2014 17:10:54 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:

> On Mon, 30 Jun 2014, Andrew Morton wrote:
> > On Mon, 30 Jun 2014 15:48:39 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:
> > > -		return 0;
> > > +		return -EAGAIN;
> > 
> > Maybe it's time to document the shmem_unuse_inode() return values.
> 
> Oh dear.  I had hoped they would look after themselves.  This one is a
> private matter between shmem_unuse_inode and its one caller, just below.

Well, readers of shmem_unuse_inode() won't know that unless we tell them.


> > > +	if (error) {
> > > +		if (error != -ENOMEM)
> > > +			error = 0;
> > >  		mem_cgroup_cancel_charge(page, memcg);
> > >  	} else
> > >  		mem_cgroup_commit_charge(page, memcg, true);
> > 
> > If I'm reading this correctly, shmem_unuse() can now return -EAGAIN and
> > that can get all the way back to userspace.  `man 2 swapoff' doesn't
> > know this...
> 
> if (error) {
> 	if (error != -ENOMEM)
> 		error = 0;
> ...
> 	return error;
> 
> So the only values returned from shmem_unuse_inode() to its caller
> try_to_unuse() are 0 and -ENOMEM.  Those may get passed back to the
> user, but -EAGAIN was just an internal shmem.c detail.

OK.

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
  2014-07-01  0:34     ` Andrew Morton
@ 2014-07-01  1:06       ` Hugh Dickins
  0 siblings, 0 replies; 6+ messages in thread
From: Hugh Dickins @ 2014-07-01  1:06 UTC (permalink / raw)
  To: Andrew Morton
  Cc: Hugh Dickins, Johannes Weiner, Michal Hocko, linux-mm,
	linux-kernel

On Mon, 30 Jun 2014, Andrew Morton wrote:
> On Mon, 30 Jun 2014 17:10:54 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:
> > On Mon, 30 Jun 2014, Andrew Morton wrote:
> > > On Mon, 30 Jun 2014 15:48:39 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote:
> > > > -		return 0;
> > > > +		return -EAGAIN;
> > > 
> > > Maybe it's time to document the shmem_unuse_inode() return values.
> > 
> > Oh dear.  I had hoped they would look after themselves.  This one is a
> > private matter between shmem_unuse_inode and its one caller, just below.
> 
> Well, readers of shmem_unuse_inode() won't know that unless we tell them.

Add comments on the private use of -EAGAIN.

Signed-off-by: Hugh Dickins <hughd@google.com>
---

 mm/shmem.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

--- 3.16-rc2-mm1+/mm/shmem.c	2014-06-30 15:05:50.736335600 -0700
+++ linux/mm/shmem.c	2014-06-30 18:00:02.820584009 -0700
@@ -611,7 +611,7 @@ static int shmem_unuse_inode(struct shme
 	radswap = swp_to_radix_entry(swap);
 	index = radix_tree_locate_item(&mapping->page_tree, radswap);
 	if (index == -1)
-		return -EAGAIN;
+		return -EAGAIN;	/* tell shmem_unuse we found nothing */
 
 	/*
 	 * Move _head_ to start search for next from here.
@@ -712,6 +712,7 @@ int shmem_unuse(swp_entry_t swap, struct
 		cond_resched();
 		if (error != -EAGAIN)
 			break;
+		/* found nothing in this: move on to search the next */
 	}
 	mutex_unlock(&shmem_swaplist_mutex);
 

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse
  2014-06-30 22:48 [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse Hugh Dickins
  2014-06-30 23:02 ` Andrew Morton
@ 2014-07-02 22:37 ` Johannes Weiner
  1 sibling, 0 replies; 6+ messages in thread
From: Johannes Weiner @ 2014-07-02 22:37 UTC (permalink / raw)
  To: Hugh Dickins; +Cc: Andrew Morton, Michal Hocko, linux-mm, linux-kernel

On Mon, Jun 30, 2014 at 03:48:39PM -0700, Hugh Dickins wrote:
> Under shmem swapping and swapoff load, I sometimes hit the
> VM_BUG_ON_PAGE(!page->mapping) in mem_cgroup_commit_charge() at
> mm/memcontrol.c:6502!  Each time it has been a call from shmem_unuse().
> 
> Yes, there are some cases (most commonly when the page being unswapped
> is in a file being unlinked and evicted at that time) when the charge
> should not be committed.  In the old scheme, the page got uncharged
> again on release; but in the new scheme, it hits that BUG beforehand.
> 
> It's a useful BUG, so adapt shmem_unuse() to allow for it.  Which needs
> more info from shmem_unuse_inode(): so abuse -EAGAIN internally to
> replace the previous !found state (-ENOENT would be a more natural
> code, but that's exactly what you get when the swap has been evicted).
> 
> Signed-off-by: Hugh Dickins <hughd@google.com>

Acked-by: Johannes Weiner <hannes@cmpxchg.org>

Thanks, Hugh!

--
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>

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2014-07-02 22:37 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-06-30 22:48 [PATCH mmotm/next] mm: memcontrol: rewrite charge API: fix shmem_unuse Hugh Dickins
2014-06-30 23:02 ` Andrew Morton
2014-07-01  0:10   ` Hugh Dickins
2014-07-01  0:34     ` Andrew Morton
2014-07-01  1:06       ` Hugh Dickins
2014-07-02 22:37 ` Johannes Weiner

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).