In free path, we don't check number of cpu_partial, so one slab can
be linked in cpu partial list even if cpu_partial is 0. To prevent this,
we should check number of cpu_partial in put_cpu_partial().
Signed-off-by: Joonsoo Kim <[email protected]>
diff --git a/mm/slub.c b/mm/slub.c
index 57707f0..7033b4f 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1955,6 +1955,9 @@ static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
int pages;
int pobjects;
+ if (!s->cpu_partial)
+ return;
+
do {
pages = 0;
pobjects = 0;
--
1.7.9.5
On Wed, Jun 19, 2013 at 04:00:32PM +0800, Wanpeng Li wrote:
> On Wed, Jun 19, 2013 at 03:33:55PM +0900, Joonsoo Kim wrote:
> >In free path, we don't check number of cpu_partial, so one slab can
> >be linked in cpu partial list even if cpu_partial is 0. To prevent this,
> >we should check number of cpu_partial in put_cpu_partial().
> >
>
> How about skip get_partial entirely? put_cpu_partial is called
> in two paths, one is during refill cpu partial lists in alloc
> slow path, the other is in free slow path. And cpu_partial is 0
> just in debug mode.
>
> - alloc slow path, there is unnecessary to call get_partial
> since cpu partial lists won't be used in debug mode.
> - free slow patch, new.inuse won't be true in debug mode
> which lead to put_cpu_partial won't be called.
>
In debug mode, put_cpu_partial() can't be called already on both path.
But, if we assign 0 to cpu_partial via sysfs, put_cpu_partial() will be called
on free slow path. On alloc slow path, it can't be called, because following
test in get_partial_node() is always failed.
available > s->cpu_partial / 2
Thanks.
> Regards,
> Wanpeng Li
>
> >Signed-off-by: Joonsoo Kim <[email protected]>
> >
> >diff --git a/mm/slub.c b/mm/slub.c
> >index 57707f0..7033b4f 100644
> >--- a/mm/slub.c
> >+++ b/mm/slub.c
> >@@ -1955,6 +1955,9 @@ static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
> > int pages;
> > int pobjects;
> >
> >+ if (!s->cpu_partial)
> >+ return;
> >+
> > do {
> > pages = 0;
> > pobjects = 0;
> >--
> >1.7.9.5
> >
> >--
> >To unsubscribe, send a message with 'unsubscribe linux-mm' in
> >the body to [email protected]. For more info on Linux MM,
> >see: http://www.linux-mm.org/ .
> >Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
On Wed, 19 Jun 2013, Joonsoo Kim wrote:
> In free path, we don't check number of cpu_partial, so one slab can
> be linked in cpu partial list even if cpu_partial is 0. To prevent this,
> we should check number of cpu_partial in put_cpu_partial().
Acked-by: Christoph Lameeter <[email protected]>
On Thu, Jun 20, 2013 at 08:26:03AM +0800, Wanpeng Li wrote:
> On Wed, Jun 19, 2013 at 05:52:50PM +0900, Joonsoo Kim wrote:
> >On Wed, Jun 19, 2013 at 04:00:32PM +0800, Wanpeng Li wrote:
> >> On Wed, Jun 19, 2013 at 03:33:55PM +0900, Joonsoo Kim wrote:
> >> >In free path, we don't check number of cpu_partial, so one slab can
> >> >be linked in cpu partial list even if cpu_partial is 0. To prevent this,
> >> >we should check number of cpu_partial in put_cpu_partial().
> >> >
> >>
> >> How about skip get_partial entirely? put_cpu_partial is called
> >> in two paths, one is during refill cpu partial lists in alloc
> >> slow path, the other is in free slow path. And cpu_partial is 0
> >> just in debug mode.
> >>
> >> - alloc slow path, there is unnecessary to call get_partial
> >> since cpu partial lists won't be used in debug mode.
> >> - free slow patch, new.inuse won't be true in debug mode
> >> which lead to put_cpu_partial won't be called.
> >>
> >
> >In debug mode, put_cpu_partial() can't be called already on both path.
> >But, if we assign 0 to cpu_partial via sysfs, put_cpu_partial() will be called
> >on free slow path. On alloc slow path, it can't be called, because following
> >test in get_partial_node() is always failed.
> >
> >available > s->cpu_partial / 2
>
> Is it always true? We can freeze slab from partial list, and
> s->cpu_partial is 0.
Do you mean node partial list?
At first, acquire_slab() is called for a cpu slab(not cpu partial list)
in get_partial_node(), and then, check above test. In this time, available
is always higher than 0, so, if we assign 0 to s->cpu_partial, we break
the loop and we don't try to get a slab for cpu partial list.
Thanks.
>
> Regards,
> Wanpeng Li
>
> >
> >Thanks.
> >
> >> Regards,
> >> Wanpeng Li
> >>
> >> >Signed-off-by: Joonsoo Kim <[email protected]>
> >> >
> >> >diff --git a/mm/slub.c b/mm/slub.c
> >> >index 57707f0..7033b4f 100644
> >> >--- a/mm/slub.c
> >> >+++ b/mm/slub.c
> >> >@@ -1955,6 +1955,9 @@ static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
> >> > int pages;
> >> > int pobjects;
> >> >
> >> >+ if (!s->cpu_partial)
> >> >+ return;
> >> >+
> >> > do {
> >> > pages = 0;
> >> > pobjects = 0;
> >> >--
> >> >1.7.9.5
> >> >
> >> >--
> >> >To unsubscribe, send a message with 'unsubscribe linux-mm' in
> >> >the body to [email protected]. For more info on Linux MM,
> >> >see: http://www.linux-mm.org/ .
> >> >Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
> >>
> >> --
> >> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> >> the body to [email protected]. For more info on Linux MM,
> >> see: http://www.linux-mm.org/ .
> >> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
On Wed, Jun 19, 2013 at 9:33 AM, Joonsoo Kim <[email protected]> wrote:
> In free path, we don't check number of cpu_partial, so one slab can
> be linked in cpu partial list even if cpu_partial is 0. To prevent this,
> we should check number of cpu_partial in put_cpu_partial().
>
> Signed-off-by: Joonsoo Kim <[email protected]>
Applied, thanks a lot!