diff mbox series

[v2] mm: memcontrol: fix kernel stack account

Message ID 20210303093956.72318-1-songmuchun@bytedance.com (mailing list archive)
State New
Headers show
Series [v2] mm: memcontrol: fix kernel stack account | expand

Commit Message

Muchun Song March 3, 2021, 9:39 a.m. UTC
For simplification 991e7673859e ("mm: memcontrol: account kernel stack
per node") has changed the per zone vmalloc backed stack pages
accounting to per node. By doing that we have lost a certain precision
because those pages might live in different NUMA nodes. In the end
NR_KERNEL_STACK_KB exported to the userspace might be over estimated on
some nodes while underestimated on others.

This doesn't impose any real problem to correctnes of the kernel
behavior as the counter is not used for any internal processing but it
can cause some confusion to the userspace.

Address the problem by accounting each vmalloc backing page to its own
node.

Fixes: 991e7673859e ("mm: memcontrol: account kernel stack per node")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
---
Changelog in v2:
 - Rework commit log suggested by Michal.

 Thanks to Michal and Shakeel for review.

 kernel/fork.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

Comments

Michal Hocko March 3, 2021, 10:25 a.m. UTC | #1
On Wed 03-03-21 17:39:56, Muchun Song wrote:
> For simplification 991e7673859e ("mm: memcontrol: account kernel stack
> per node") has changed the per zone vmalloc backed stack pages
> accounting to per node. By doing that we have lost a certain precision
> because those pages might live in different NUMA nodes. In the end
> NR_KERNEL_STACK_KB exported to the userspace might be over estimated on
> some nodes while underestimated on others.
> 
> This doesn't impose any real problem to correctnes of the kernel
> behavior as the counter is not used for any internal processing but it
> can cause some confusion to the userspace.

You have skipped over one part of the changelog I have proposed and that
is to provide an actual data.

> Address the problem by accounting each vmalloc backing page to its own
> node.
> 
> Fixes: 991e7673859e ("mm: memcontrol: account kernel stack per node")

Fixes tag might make somebody assume this is worth backporting but I
highly doubt so.

> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Reviewed-by: Shakeel Butt <shakeelb@google.com>

Anyway
Acked-by: Michal Hocko <mhocko@suse.com>

as the patch is correct with one comment below

> ---
> Changelog in v2:
>  - Rework commit log suggested by Michal.
> 
>  Thanks to Michal and Shakeel for review.
> 
>  kernel/fork.c | 15 ++++++++++-----
>  1 file changed, 10 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/fork.c b/kernel/fork.c
> index d66cd1014211..6e2201feb524 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -379,14 +379,19 @@ static void account_kernel_stack(struct task_struct *tsk, int account)
>  	void *stack = task_stack_page(tsk);
>  	struct vm_struct *vm = task_stack_vm_area(tsk);
>  
> +	if (vm) {
> +		int i;
>  
> -	/* All stack pages are in the same node. */
> -	if (vm)
> -		mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
> -				      account * (THREAD_SIZE / 1024));
> -	else
> +		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);

I do not think we need this BUG_ON. What kind of purpose does it serve?

> +
> +		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
> +			mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
> +					      account * (PAGE_SIZE / 1024));
> +	} else {
> +		/* All stack pages are in the same node. */
>  		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
>  				      account * (THREAD_SIZE / 1024));
> +	}
>  }
>  
>  static int memcg_charge_kernel_stack(struct task_struct *tsk)
> -- 
> 2.11.0
Muchun Song March 3, 2021, 1:27 p.m. UTC | #2
On Wed, Mar 3, 2021 at 6:25 PM Michal Hocko <mhocko@suse.com> wrote:
>
> On Wed 03-03-21 17:39:56, Muchun Song wrote:
> > For simplification 991e7673859e ("mm: memcontrol: account kernel stack
> > per node") has changed the per zone vmalloc backed stack pages
> > accounting to per node. By doing that we have lost a certain precision
> > because those pages might live in different NUMA nodes. In the end
> > NR_KERNEL_STACK_KB exported to the userspace might be over estimated on
> > some nodes while underestimated on others.
> >
> > This doesn't impose any real problem to correctnes of the kernel
> > behavior as the counter is not used for any internal processing but it
> > can cause some confusion to the userspace.
>
> You have skipped over one part of the changelog I have proposed and that
> is to provide an actual data.

Because this is a problem I found by looking at the code, not a real world
problem. I do not have any actual data. :-(

>
> > Address the problem by accounting each vmalloc backing page to its own
> > node.
> >
> > Fixes: 991e7673859e ("mm: memcontrol: account kernel stack per node")
>
> Fixes tag might make somebody assume this is worth backporting but I
> highly doubt so.

OK. I can remove the Fixes tag.

>
> > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > Reviewed-by: Shakeel Butt <shakeelb@google.com>
>
> Anyway
> Acked-by: Michal Hocko <mhocko@suse.com>

Thanks for your review.

>
> as the patch is correct with one comment below
>
> > ---
> > Changelog in v2:
> >  - Rework commit log suggested by Michal.
> >
> >  Thanks to Michal and Shakeel for review.
> >
> >  kernel/fork.c | 15 ++++++++++-----
> >  1 file changed, 10 insertions(+), 5 deletions(-)
> >
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index d66cd1014211..6e2201feb524 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -379,14 +379,19 @@ static void account_kernel_stack(struct task_struct *tsk, int account)
> >       void *stack = task_stack_page(tsk);
> >       struct vm_struct *vm = task_stack_vm_area(tsk);
> >
> > +     if (vm) {
> > +             int i;
> >
> > -     /* All stack pages are in the same node. */
> > -     if (vm)
> > -             mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
> > -                                   account * (THREAD_SIZE / 1024));
> > -     else
> > +             BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
>
> I do not think we need this BUG_ON. What kind of purpose does it serve?

vm->nr_pages should be always equal to THREAD_SIZE / PAGE_SIZE
if the system is not corrupted. It makes sense to remove the BUG_ON.
I will remove it in the next version. Thanks.

>
> > +
> > +             for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
> > +                     mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
> > +                                           account * (PAGE_SIZE / 1024));
> > +     } else {
> > +             /* All stack pages are in the same node. */
> >               mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
> >                                     account * (THREAD_SIZE / 1024));
> > +     }
> >  }
> >
> >  static int memcg_charge_kernel_stack(struct task_struct *tsk)
> > --
> > 2.11.0
>
> --
> Michal Hocko
> SUSE Labs
Michal Hocko March 3, 2021, 2:02 p.m. UTC | #3
On Wed 03-03-21 21:27:24, Muchun Song wrote:
> On Wed, Mar 3, 2021 at 6:25 PM Michal Hocko <mhocko@suse.com> wrote:
> >
> > On Wed 03-03-21 17:39:56, Muchun Song wrote:
> > > For simplification 991e7673859e ("mm: memcontrol: account kernel stack
> > > per node") has changed the per zone vmalloc backed stack pages
> > > accounting to per node. By doing that we have lost a certain precision
> > > because those pages might live in different NUMA nodes. In the end
> > > NR_KERNEL_STACK_KB exported to the userspace might be over estimated on
> > > some nodes while underestimated on others.
> > >
> > > This doesn't impose any real problem to correctnes of the kernel
> > > behavior as the counter is not used for any internal processing but it
> > > can cause some confusion to the userspace.
> >
> > You have skipped over one part of the changelog I have proposed and that
> > is to provide an actual data.
> 
> Because this is a problem I found by looking at the code, not a real world
> problem. I do not have any actual data. :-(

As I've mentioned several times already, this is all fine but it should
be made explicit in the changelog. There might be people spending their
time to evaluate this code to find out whether this is something that
somebody depend on.

[...]
> > > -     /* All stack pages are in the same node. */
> > > -     if (vm)
> > > -             mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
> > > -                                   account * (THREAD_SIZE / 1024));
> > > -     else
> > > +             BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
> >
> > I do not think we need this BUG_ON. What kind of purpose does it serve?
> 
> vm->nr_pages should be always equal to THREAD_SIZE / PAGE_SIZE
> if the system is not corrupted.

BUG_ON is not an annotation for "this shouldn't happen". Even if the
system was corrupted and nr_pages wouldn't match then this is not a
reason to crash the kernel right away.

In general there should be a very _strong_ reason to add a BUG_ON.
Shakeel Butt March 3, 2021, 2:42 p.m. UTC | #4
On Wed, Mar 3, 2021 at 6:02 AM Michal Hocko <mhocko@suse.com> wrote:
>
[...]
> > > > +             BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
> > >
> > > I do not think we need this BUG_ON. What kind of purpose does it serve?
> >
> > vm->nr_pages should be always equal to THREAD_SIZE / PAGE_SIZE
> > if the system is not corrupted.
>
> BUG_ON is not an annotation for "this shouldn't happen". Even if the
> system was corrupted and nr_pages wouldn't match then this is not a
> reason to crash the kernel right away.
>
> In general there should be a very _strong_ reason to add a BUG_ON.
>

I agree with Michal. We should remove this BUG_ON or at least convert
it into VM_BUG_ON.
Johannes Weiner March 3, 2021, 3:14 p.m. UTC | #5
On Wed, Mar 03, 2021 at 05:39:56PM +0800, Muchun Song wrote:
> For simplification 991e7673859e ("mm: memcontrol: account kernel stack
> per node") has changed the per zone vmalloc backed stack pages
> accounting to per node. By doing that we have lost a certain precision
> because those pages might live in different NUMA nodes. In the end
> NR_KERNEL_STACK_KB exported to the userspace might be over estimated on
> some nodes while underestimated on others.
> 
> This doesn't impose any real problem to correctnes of the kernel
> behavior as the counter is not used for any internal processing but it
> can cause some confusion to the userspace.
> 
> Address the problem by accounting each vmalloc backing page to its own
> node.
> 
> Fixes: 991e7673859e ("mm: memcontrol: account kernel stack per node")
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Reviewed-by: Shakeel Butt <shakeelb@google.com>

With changes proposed by Shakeel and Michal, this looks good to me.

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

I guess the BUG_ON() was inspired by memcg_charge_kernel_stack() - not
really your fault for following that example. But yeah, please drop it
from your patch. Thanks!
diff mbox series

Patch

diff --git a/kernel/fork.c b/kernel/fork.c
index d66cd1014211..6e2201feb524 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -379,14 +379,19 @@  static void account_kernel_stack(struct task_struct *tsk, int account)
 	void *stack = task_stack_page(tsk);
 	struct vm_struct *vm = task_stack_vm_area(tsk);
 
+	if (vm) {
+		int i;
 
-	/* All stack pages are in the same node. */
-	if (vm)
-		mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
-				      account * (THREAD_SIZE / 1024));
-	else
+		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
+
+		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
+			mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
+					      account * (PAGE_SIZE / 1024));
+	} else {
+		/* All stack pages are in the same node. */
 		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 				      account * (THREAD_SIZE / 1024));
+	}
 }
 
 static int memcg_charge_kernel_stack(struct task_struct *tsk)