Skip to contentSkip to conversation

memcg: accounting for allocations called with disabled BH

PATCH In review Vasily Averin 4 participants 12 messages Mar 9, 2021 – Mar 10, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/9 memcg: accounting for allocations called with disabled BH

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memcontrol.c +1 -1
index 845eec0..568f2cb 100644
@@ -1076,7 +1076,7 @@ static __always_inline bool memcg_kmem_bypass(void)
 		return false;
 
 	/* Memcg to charge can't be determined. */
-	if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
+	if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
 		return true;
 
 	return false;
Signature · 1 line
1.8.3.1
#1 lore raw
Quoted 3 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Is there any existing user in the tree? Or is this more of a preparatory patch for a later one which will need it? In other words, is this a bug fix or a preparatory work.

Quoted 20 lines
---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memcontrol.c +1 -1
index 845eec0..568f2cb 100644
@@ -1076,7 +1076,7 @@ static __always_inline bool memcg_kmem_bypass(void)
 		return false;
 
 	/* Memcg to charge can't be determined. */
-	if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
+	if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
 		return true;
 
 	return false;
Signature · 1 line
1.8.3.1
Signature · 2 lines
Michal Hocko
SUSE Labs
#2 lore raw
Quoted 8 lines · Michal Hocko
Quoted 3 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Is there any existing user in the tree? Or is this more of a preparatory patch for a later one which will need it? In other words, is this a bug fix or a preparatory work.

struct fib6_node objects are allocated by this way net/ipv6/route.c::__ip6_ins_rt()

...        write_lock_bh(&table->tb6_lock);
        err = fib6_add(&table->tb6_root, rt, info, mxc);
        write_unlock_bh(&table->tb6_lock);

I spend some time to understand why properly entries from properly configured cache was not accounted to container's memcg.

Thank you,
Vasily Averin

#3 lore raw
Quoted 10 lines · Vasily Averin
Quoted 8 lines · Michal Hocko
Quoted 3 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Is there any existing user in the tree? Or is this more of a preparatory patch for a later one which will need it? In other words, is this a bug fix or a preparatory work.

Quoted 8 lines

struct fib6_node objects are allocated by this way net/ipv6/route.c::__ip6_ins_rt()

...        write_lock_bh(&table->tb6_lock);
        err = fib6_add(&table->tb6_root, rt, info, mxc);
        write_unlock_bh(&table->tb6_lock);

I spend some time to understand why properly entries from properly configured cache was not accounted to container's memcg.

OK, that is a valuable information. If there are no other users currently then I would recommend squashing this patch into the one which introduces accounting for fib6_node cache (patch 2, IIUC).

Signature · 2 lines
Michal Hocko
SUSE Labs
#4 lore raw
In reply to Vasily Averin
Quoted 6 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

In that file in_interrupt() is used at other places too. Should we change those too?

Quoted 20 lines
---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memcontrol.c +1 -2
index 845eec0..568f2cb 100644
@@ -1076,7 +1076,7 @@ static __always_inline bool memcg_kmem_bypass(void)
                return false;
 
        /* Memcg to charge can't be determined. */
-       if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
+       if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
                return true;
 
        return false;
--

1.8.3.1

#5 lore raw
Quoted 10 lines · Shakeel Butt
Quoted 6 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

In that file in_interrupt() is used at other places too. Should we change those too?

Yes, it seems so. Let me prepare a fix (it seems like most of them were introduced by me).

Thanks!

#6 lore raw
Quoted 7 lines · Roman Gushchin
Quoted 6 lines · Shakeel Butt
Quoted 5 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

In that file in_interrupt() is used at other places too. Should we change those too?

Yes, it seems so. Let me prepare a fix (it seems like most of them were introduced by me).

No objects from my side, just keep me informed and add me in "Reported-by:" in your patches.

Thank you,
Vasily Averin

#7 lore raw
In reply to Roman Gushchin
Quoted 7 lines · Roman Gushchin
Quoted 6 lines · Shakeel Butt
Quoted 5 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

In that file in_interrupt() is used at other places too. Should we change those too?

Yes, it seems so. Let me prepare a fix (it seems like most of them were introduced by me).

Does this affect any existing in-tree users?

Signature · 2 lines
Michal Hocko
SUSE Labs
#8 lore raw
Quoted 10 lines · Michal Hocko
Quoted 9 lines · Roman Gushchin
Quoted 8 lines · Shakeel Butt
Quoted 6 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

In that file in_interrupt() is used at other places too. Should we change those too?

Yes, it seems so. Let me prepare a fix (it seems like most of them were introduced by me).

Does this affect any existing in-tree users?

I'll double check this, but I doubt. We should fix it anyway, but I don't think we need any stable backports.

#9 lore raw
In reply to Shakeel Butt
Quoted 10 lines · Shakeel Butt
Quoted 6 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

In that file in_interrupt() is used at other places too. Should we change those too?

Yes, you're right, in_interrupt() is used incorrectly in other places too, but
1) these cases are not so critical as this one,
2) and are not related to current patch set

They can be replaced later without urgency
(unless I missed something imporant).

thank you,
Vasily Averin

Quoted 20 lines
Quoted 20 lines
---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memcontrol.c +1 -2
index 845eec0..568f2cb 100644
@@ -1076,7 +1076,7 @@ static __always_inline bool memcg_kmem_bypass(void)
                return false;
 
        /* Memcg to charge can't be determined. */
-       if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
+       if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
                return true;
 
        return false;
--

1.8.3.1

Acked-by on v1 1/9
#10 lore raw
In reply to Vasily Averin
Quoted 5 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Good catch!

It looks like the bug was there for years: in_interrupt() was there since the commit 7ae1e1d0f8ac ("memcg: kmem controller infrastructure") from 2012! So I guess there is no point for a stable fix, but it's definitely nice to have it fixed.

for this patch and the rest of the series.

Thank you!

Quoted 20 lines
---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memcontrol.c +1 -1
index 845eec0..568f2cb 100644
@@ -1076,7 +1076,7 @@ static __always_inline bool memcg_kmem_bypass(void)
 		return false;
 
 	/* Memcg to charge can't be determined. */
-	if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
+	if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
 		return true;
 
 	return false;
Signature · 1 line
1.8.3.1
#11 lore raw
Quoted 5 lines · Roman Gushchin
Quoted 4 lines · Vasily Averin

in_interrupt() check in memcg_kmem_bypass() is incorrect because it does not allow to account memory allocation called from task context with disabled BH, i.e. inside spin_lock_bh()/spin_unlock_bh() sections

Quoted 8 lines
Quoted 1 line

Good catch!

It looks like the bug was there for years: in_interrupt() was there since the commit 7ae1e1d0f8ac ("memcg: kmem controller infrastructure") from 2012! So I guess there is no point for a stable fix, but it's definitely nice to have it fixed.

I did not noticed that this problem affect existing allocation,
I just found that it blocked accounting of "struct fib6_node" in __ip6_ins_rt() added in my current patch set. So I do not think it should went to stable@.

Thank you,
Vasily Averin