On my freshly built 3.14-rc1 kernel, I get the following new warning splat from
slub:
[ 69.008845] ------------[ cut here ]------------
[ 69.008861] WARNING: CPU: 0 PID: 1578 at mm/slub.c:1007
deactivate_slab+0x4bb/0x520()
[ 69.008863] Modules linked in: rfcomm nfs fscache af_packet lockd sunrpc bnep
rtl8723be arc4 rtl8723_common b43 rtl_pci rtlwifi mac80211 cfg80211 btusb bl
uetooth powernow_k8 kvm_amd kvm snd_hda_intel snd_hda_codec bcma snd_hwdep r852
sdhci_pci sdhci snd_pcm sr_mod snd_seq ssb cdrom forcedeth sm_common pcspkr s
nd_timer serio_raw snd_seq_device snd nand mtd mmc_core rfkill ata_generic
pata_amd nand_ids nand_bch bch nand_ecc 6lowpan_iphc soundcore btcoexist r592 ac v
ideo memstick button battery sg dm_mod autofs4 thermal processor thermal_sys hwmon
[ 69.008934] CPU: 0 PID: 1578 Comm: akonadi_newmail Not tainted 3.14.0-rc1-wl+ #60
[ 69.008936] Hardware name: Hewlett-Packard HP Pavilion dv2700 Notebook
PC/30D6, BIOS F.27 11/27/2008
[ 69.008939] 0000000000000009 ffff880085ab1cd0 ffffffff815ee4a0 0000000000000000
[ 69.008945] ffff880085ab1d08 ffffffff8104e768 ffffea0001ee69c0 ffff8800bb401800
[ 69.008950] 0000000000000000 ffff8800bb400c80 0000000000000002 ffff880085ab1d18
[ 69.008956] Call Trace:
[ 69.008965] [<ffffffff815ee4a0>] dump_stack+0x4d/0x6f
[ 69.008971] [<ffffffff8104e768>] warn_slowpath_common+0x78/0xa0
[ 69.008976] [<ffffffff8104e845>] warn_slowpath_null+0x15/0x20
[ 69.008980] [<ffffffff8118581b>] deactivate_slab+0x4bb/0x520
[ 69.008985] [<ffffffff81183429>] ? new_slab+0x1f9/0x300
[ 69.008989] [<ffffffff815ebe57>] __slab_alloc+0x34d/0x4f5
[ 69.008994] [<ffffffff81185a6b>] ? kmem_cache_alloc+0x18b/0x1c0
[ 69.008999] [<ffffffff810770d1>] ? prepare_creds+0x21/0x1a0
[ 69.009003] [<ffffffff810770d1>] ? prepare_creds+0x21/0x1a0
[ 69.009007] [<ffffffff81185a6b>] kmem_cache_alloc+0x18b/0x1c0
[ 69.009011] [<ffffffff810770d1>] prepare_creds+0x21/0x1a0
[ 69.009016] [<ffffffff8119f955>] SyS_access+0x35/0x1f0
[ 69.009019] [<ffffffff815fe8a2>] system_call_fastpath+0x16/0x1b
[ 69.009019] ---[ end trace c63f75644bfb030a ]---
There is a similar warning for my other CPU as well. The warning comes from
"lockdep_assert_held(&n->list_lock);".
If needed, I can bisect this issue.
Larry
Commit c65c1877bd68 ("slub: use lockdep_assert_held") incorrectly required
that add_full() and remove_full() hold n->list_lock. The lock is only
taken when kmem_cache_debug(s), since that's the only time it actually
does anything.
Require that the lock only be taken under such a condition.
Reported-by: Larry Finger <[email protected]>
Signed-off-by: David Rientjes <[email protected]>
---
mm/slub.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/mm/slub.c b/mm/slub.c
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1004,21 +1004,19 @@ static inline void slab_free_hook(struct kmem_cache *s, void *x)
static void add_full(struct kmem_cache *s,
struct kmem_cache_node *n, struct page *page)
{
- lockdep_assert_held(&n->list_lock);
-
if (!(s->flags & SLAB_STORE_USER))
return;
+ lockdep_assert_held(&n->list_lock);
list_add(&page->lru, &n->full);
}
static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
{
- lockdep_assert_held(&n->list_lock);
-
if (!(s->flags & SLAB_STORE_USER))
return;
+ lockdep_assert_held(&n->list_lock);
list_del(&page->lru);
}
On Mon, 3 Feb 2014, David Rientjes wrote:
> Commit c65c1877bd68 ("slub: use lockdep_assert_held") incorrectly required
> that add_full() and remove_full() hold n->list_lock. The lock is only
> taken when kmem_cache_debug(s), since that's the only time it actually
> does anything.
Acked-by: Christoph Lameter <[email protected]>
On 02/03/2014 02:39 PM, David Rientjes wrote:
> Commit c65c1877bd68 ("slub: use lockdep_assert_held") incorrectly required
> that add_full() and remove_full() hold n->list_lock. The lock is only
> taken when kmem_cache_debug(s), since that's the only time it actually
> does anything.
>
> Require that the lock only be taken under such a condition.
>
> Reported-by: Larry Finger <[email protected]>
> Signed-off-by: David Rientjes <[email protected]>
You may add a "Tested-by: Larry Finger <[email protected]>". The patch
cleans up the splat on my system. Thanks for the quick response.
Larry
On Mon, Feb 03, 2014 at 05:08:11PM -0600, Larry Finger wrote:
> On 02/03/2014 02:39 PM, David Rientjes wrote:
> >Commit c65c1877bd68 ("slub: use lockdep_assert_held") incorrectly required
> >that add_full() and remove_full() hold n->list_lock. The lock is only
> >taken when kmem_cache_debug(s), since that's the only time it actually
> >does anything.
> >
> >Require that the lock only be taken under such a condition.
> >
> >Reported-by: Larry Finger <[email protected]>
> >Signed-off-by: David Rientjes <[email protected]>
>
> You may add a "Tested-by: Larry Finger <[email protected]>".
> The patch cleans up the splat on my system. Thanks for the quick
> response.
Please feel free to add mine as well:
Tested-by: Paul E. McKenney <[email protected]>
And also feel free to ignore my patch as well. ;-)
Thanx, Paul
Commit c65c1877bd68 ("slub: use lockdep_assert_held") incorrectly required
that add_full() and remove_full() hold n->list_lock. The lock is only
taken when kmem_cache_debug(s), since that's the only time it actually
does anything.
Require that the lock only be taken under such a condition.
Reported-by: Larry Finger <[email protected]>
Tested-by: Larry Finger <[email protected]>
Tested-by: Paul E. McKenney <[email protected]>
Acked-by: Christoph Lameter <[email protected]>
Signed-off-by: David Rientjes <[email protected]>
---
mm/slub.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/mm/slub.c b/mm/slub.c
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1004,21 +1004,19 @@ static inline void slab_free_hook(struct kmem_cache *s, void *x)
static void add_full(struct kmem_cache *s,
struct kmem_cache_node *n, struct page *page)
{
- lockdep_assert_held(&n->list_lock);
-
if (!(s->flags & SLAB_STORE_USER))
return;
+ lockdep_assert_held(&n->list_lock);
list_add(&page->lru, &n->full);
}
static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
{
- lockdep_assert_held(&n->list_lock);
-
if (!(s->flags & SLAB_STORE_USER))
return;
+ lockdep_assert_held(&n->list_lock);
list_del(&page->lru);
}