[PATCH] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[PATCH] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Farhad Alemi]

cpuset_update_tasks_nodemask() rebinds a task's own mempolicy to the
cpuset's effective, online mems (newmems, from guarantee_online_mems()),
but rebinds that task's VMA mempolicies to the *configured* mask instead:

	cpuset_change_task_nodemask(task, &newmems);
	...
	mpol_rebind_mm(mm, &cs->mems_allowed);

On the default (v2) hierarchy a cpuset that has never had cpuset.mems
written keeps mems_allowed empty while effective_mems is inherited
non-empty from the parent, and tasks may be attached to it (the
empty-mems attach check is v1-only).  A subsequent rebind -- e.g. from a
CPU hotplug event walking the cpuset -- then calls mpol_rebind_mm() with
an empty mask.  For a VMA policy created with MPOL_F_RELATIVE_NODES this
reaches mpol_relative_nodemask() ->
nodes_fold(..., nodes_weight(cs->mems_allowed) == 0) -> bitmap_fold(),
whose set_bit(oldbit % sz, dst) divides by zero:

  Oops: divide error: 0000 [#1] SMP KASAN NOPTI
  RIP: 0010:bitmap_fold+0x5e/0xb0
   mpol_rebind_nodemask
   mpol_rebind_mm
   cpuset_update_tasks_nodemask
   cpuset_handle_hotplug
   sched_cpu_deactivate
   cpuhp_thread_fun

cs->mems_allowed is the only nodemask in this function that is not the
effective set: the task-policy rebind, the page-migration target and
cs->old_mems_allowed all use newmems.  The sibling cpuset_attach() path
already rebinds VMA policies against the effective mems
(cpuset_attach_nodemask_to = cs->effective_mems) and explicitly notes
that mems_allowed can be empty under hotplug.  Rebind the VMA policies to
newmems too: it is guaranteed non-empty by guarantee_online_mems(), which
fixes the divide-by-zero, and it makes the VMA policies consistent with
the task policy and with the nodes the task is actually allowed to use.

Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
Suggested-by: Gregory Price <gourry@gourry.net>
Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
Cc: stable@vger.kernel.org
---
 kernel/cgroup/cpuset.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)

 		migrate = is_memory_migrate(cs);

-		mpol_rebind_mm(mm, &cs->mems_allowed);
+		mpol_rebind_mm(mm, &newmems);
 		if (migrate)
 			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
 		else
-- 
2.43.0

Re: [PATCH] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Andrew Morton]

On Tue, 9 Jun 2026 19:57:41 -0400 Farhad Alemi <farhad.alemi@berkeley.edu> wrote:

> cpuset_update_tasks_nodemask() rebinds a task's own mempolicy to the
> cpuset's effective, online mems (newmems, from guarantee_online_mems()),
> but rebinds that task's VMA mempolicies to the *configured* mask instead:

Hard to understand.  Was "rebinds" supposed to be "is supposed to
rebind"?

> 	cpuset_change_task_nodemask(task, &newmems);
> 	...
> 	mpol_rebind_mm(mm, &cs->mems_allowed);
> 
> On the default (v2) hierarchy a cpuset that has never had cpuset.mems
> written keeps mems_allowed empty while effective_mems is inherited
> non-empty from the parent, and tasks may be attached to it (the
> empty-mems attach check is v1-only).  A subsequent rebind -- e.g. from a
> CPU hotplug event walking the cpuset -- then calls mpol_rebind_mm() with
> an empty mask.  For a VMA policy created with MPOL_F_RELATIVE_NODES this
> reaches mpol_relative_nodemask() ->
> nodes_fold(..., nodes_weight(cs->mems_allowed) == 0) -> bitmap_fold(),
> whose set_bit(oldbit % sz, dst) divides by zero:
> 
>   Oops: divide error: 0000 [#1] SMP KASAN NOPTI
>   RIP: 0010:bitmap_fold+0x5e/0xb0
>    mpol_rebind_nodemask
>    mpol_rebind_mm
>    cpuset_update_tasks_nodemask
>    cpuset_handle_hotplug
>    sched_cpu_deactivate
>    cpuhp_thread_fun
> 
> cs->mems_allowed is the only nodemask in this function that is not the
> effective set: the task-policy rebind, the page-migration target and
> cs->old_mems_allowed all use newmems.  The sibling cpuset_attach() path
> already rebinds VMA policies against the effective mems
> (cpuset_attach_nodemask_to = cs->effective_mems) and explicitly notes
> that mems_allowed can be empty under hotplug.  Rebind the VMA policies to
> newmems too: it is guaranteed non-empty by guarantee_online_mems(), which
> fixes the divide-by-zero, and it makes the VMA policies consistent with
> the task policy and with the nodes the task is actually allowed to use.

How is this bug triggered?

Re: [PATCH] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Gregory Price]

On Tue, Jun 09, 2026 at 07:57:41PM -0400, Farhad Alemi wrote:
> cpuset_update_tasks_nodemask() rebinds a task's own mempolicy to the
> cpuset's effective, online mems (newmems, from guarantee_online_mems()),
> but rebinds that task's VMA mempolicies to the *configured* mask instead:
> 
> 	cpuset_change_task_nodemask(task, &newmems);
> 	...
> 	mpol_rebind_mm(mm, &cs->mems_allowed);
> 
> On the default (v2) hierarchy a cpuset that has never had cpuset.mems
> written keeps mems_allowed empty while effective_mems is inherited
> non-empty from the parent, and tasks may be attached to it (the
> empty-mems attach check is v1-only).  A subsequent rebind -- e.g. from a
> CPU hotplug event walking the cpuset -- then calls mpol_rebind_mm() with
> an empty mask.  For a VMA policy created with MPOL_F_RELATIVE_NODES this
> reaches mpol_relative_nodemask() ->
> nodes_fold(..., nodes_weight(cs->mems_allowed) == 0) -> bitmap_fold(),
> whose set_bit(oldbit % sz, dst) divides by zero:
> 
>   Oops: divide error: 0000 [#1] SMP KASAN NOPTI
>   RIP: 0010:bitmap_fold+0x5e/0xb0
>    mpol_rebind_nodemask
>    mpol_rebind_mm
>    cpuset_update_tasks_nodemask
>    cpuset_handle_hotplug
>    sched_cpu_deactivate
>    cpuhp_thread_fun
> 
> cs->mems_allowed is the only nodemask in this function that is not the
> effective set: the task-policy rebind, the page-migration target and
> cs->old_mems_allowed all use newmems.  The sibling cpuset_attach() path
> already rebinds VMA policies against the effective mems
> (cpuset_attach_nodemask_to = cs->effective_mems) and explicitly notes
> that mems_allowed can be empty under hotplug.  Rebind the VMA policies to
> newmems too: it is guaranteed non-empty by guarantee_online_mems(), which
> fixes the divide-by-zero, and it makes the VMA policies consistent with
> the task policy and with the nodes the task is actually allowed to use.
>

I think you can make this a bit more concise:

  Creating a child cpuset where cpuset.mems is never set leads to
  a div/0 when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds
  in response to a CPU hotplug event.

  Reproduction steps:
     1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
     2) Move the task into the child cpuset
     3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
     4) unplug and hotplug a cpu
          echo 0 > /sys/devices/system/cpu/cpu1/oneline
          echo 1 > /sys/devices/system/cpu/cpu1/oneline
     5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
        call to __nodes_fold()

  The cpuset code passes (cs->mems_allowed) which is not guaranteed to
  have nodes to the rebind routine.  Use mems_effective - the value
  returned by guarantee_online_mems() - instead, which is guaranteed to
  have a non-empty nodemask..

Maybe add a link to your reproducer and the original [BUG]

Link: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/
Link: https://lore.kernel.org/all/CA+0ovCiEz6SP_sn3kN4Tb+_oC=eHMXy_Ffj=usV3wREdQrUtww@mail.gmail.com/

Does this need a Closes tag?

> Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
> Suggested-by: Gregory Price <gourry@gourry.net>
> Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
> Cc: stable@vger.kernel.org
> ---
>  kernel/cgroup/cpuset.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
> 
>  		migrate = is_memory_migrate(cs);
> 
> -		mpol_rebind_mm(mm, &cs->mems_allowed);
> +		mpol_rebind_mm(mm, &newmems);
>  		if (migrate)
>  			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
>  		else
> -- 
> 2.43.0

Re: [PATCH] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/9/26 7:57 PM, Farhad Alemi wrote:
> cpuset_update_tasks_nodemask() rebinds a task's own mempolicy to the
> cpuset's effective, online mems (newmems, from guarantee_online_mems()),
> but rebinds that task's VMA mempolicies to the *configured* mask instead:
>
> 	cpuset_change_task_nodemask(task, &newmems);
> 	...
> 	mpol_rebind_mm(mm, &cs->mems_allowed);
>
> On the default (v2) hierarchy a cpuset that has never had cpuset.mems
> written keeps mems_allowed empty while effective_mems is inherited
> non-empty from the parent, and tasks may be attached to it (the
> empty-mems attach check is v1-only).  A subsequent rebind -- e.g. from a
> CPU hotplug event walking the cpuset -- then calls mpol_rebind_mm() with
> an empty mask.  For a VMA policy created with MPOL_F_RELATIVE_NODES this
> reaches mpol_relative_nodemask() ->
> nodes_fold(..., nodes_weight(cs->mems_allowed) == 0) -> bitmap_fold(),
> whose set_bit(oldbit % sz, dst) divides by zero:
>
>    Oops: divide error: 0000 [#1] SMP KASAN NOPTI
>    RIP: 0010:bitmap_fold+0x5e/0xb0
>     mpol_rebind_nodemask
>     mpol_rebind_mm
>     cpuset_update_tasks_nodemask
>     cpuset_handle_hotplug
>     sched_cpu_deactivate
>     cpuhp_thread_fun
>
> cs->mems_allowed is the only nodemask in this function that is not the
> effective set: the task-policy rebind, the page-migration target and
> cs->old_mems_allowed all use newmems.  The sibling cpuset_attach() path
> already rebinds VMA policies against the effective mems
> (cpuset_attach_nodemask_to = cs->effective_mems) and explicitly notes
> that mems_allowed can be empty under hotplug.  Rebind the VMA policies to
> newmems too: it is guaranteed non-empty by guarantee_online_mems(), which
> fixes the divide-by-zero, and it makes the VMA policies consistent with
> the task policy and with the nodes the task is actually allowed to use.
>
> Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
> Suggested-by: Gregory Price <gourry@gourry.net>
> Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
> Cc: stable@vger.kernel.org
> ---
>   kernel/cgroup/cpuset.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>
>   		migrate = is_memory_migrate(cs);
>
> -		mpol_rebind_mm(mm, &cs->mems_allowed);
> +		mpol_rebind_mm(mm, &newmems);
>   		if (migrate)
>   			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
>   		else

Could you change it to &cs->effecitve_mems instead? For v2, 
effective_mems will never be empty.

In fact, this is part of the following patch

https://lore.kernel.org/lkml/20260604150229.414135-2-longman@redhat.com/

Given that this bug can crash the kernel, it should be separated out as 
a separate patch.

Cheers,
Longman

[PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Farhad Alemi]

Creating a child cpuset where cpuset.mems is never set leads to a div/0
when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
CPU hotplug event.

Reproduction steps:
 1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
 2) Move the task into the child cpuset
 3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
 4) unplug and hotplug a cpu
      echo 0 > /sys/devices/system/cpu/cpu1/online
      echo 1 > /sys/devices/system/cpu/cpu1/online
 5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
    call to __nodes_fold()

The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
nodes to the rebind routine.  Use cs->effective_mems instead, which is
guaranteed to have a non-empty nodemask.

Link: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/
Link: https://lore.kernel.org/all/CA+0ovCiEz6SP_sn3kN4Tb+_oC=eHMXy_Ffj=usV3wREdQrUtww@mail.gmail.com/
Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
Suggested-by: Gregory Price <gourry@gourry.net>
Suggested-by: Waiman Long <longman@redhat.com>
Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
Cc: stable@vger.kernel.org
---
v2: rebind to cs->effective_mems instead of newmems (Waiman Long);
    condense the changelog.

 kernel/cgroup/cpuset.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)

 		migrate = is_memory_migrate(cs);

-		mpol_rebind_mm(mm, &cs->mems_allowed);
+		mpol_rebind_mm(mm, &cs->effective_mems);
 		if (migrate)
 			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
 		else
-- 
2.43.0

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/14/26 15:25, Farhad Alemi wrote:

Hi, thanks for your patch!

For the future, please don't submit new revisions as reply to previous submissions.

> Creating a child cpuset where cpuset.mems is never set leads to a div/0
> when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
> CPU hotplug event.
> 
> Reproduction steps:
>  1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
>  2) Move the task into the child cpuset
>  3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
>  4) unplug and hotplug a cpu
>       echo 0 > /sys/devices/system/cpu/cpu1/online
>       echo 1 > /sys/devices/system/cpu/cpu1/online
>  5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
>     call to __nodes_fold()
> 
> The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
> nodes to the rebind routine.  Use cs->effective_mems instead, which is
> guaranteed to have a non-empty nodemask.

Probably worth mentioning here that this makes the linked reproducer happy.

> 
> Link: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/

This should be a

Closes:
https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/

> Link: https://lore.kernel.org/all/CA+0ovCiEz6SP_sn3kN4Tb+_oC=eHMXy_Ffj=usV3wREdQrUtww@mail.gmail.com/
> Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
> Suggested-by: Gregory Price <gourry@gourry.net>
> Suggested-by: Waiman Long <longman@redhat.com>
> Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
> Cc: stable@vger.kernel.org
> ---
> v2: rebind to cs->effective_mems instead of newmems (Waiman Long);
>     condense the changelog.
> 
>  kernel/cgroup/cpuset.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
> 
>  		migrate = is_memory_migrate(cs);
> 
> -		mpol_rebind_mm(mm, &cs->mems_allowed);
> +		mpol_rebind_mm(mm, &cs->effective_mems);

God this is confusing.

So, we obtain newmems from guarantee_online_mems(), which guarantees that
newmems is non-empty.

In cpuset_change_task_nodemask(), we set tsk->mems_allowed to newmems, and call
mpol_rebind_task(tsk, newmems).

So at least tsk->mems_allowed should be non-empty.

Then we call mpol_rebind_mm(mm, &cs->mems_allowed);


Naturally I wonder: Why are we not using "task->mems_allowed" (maybe cs vs. tsk
was the original bug?), which is effectively just newmems?

guarantee_online_mems() computes newmems as "cs->effective_mems &
node_states[N_MEMORY]", but walks up to the parent if it would be empty.

-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Gregory Price]

On Mon, Jun 15, 2026 at 10:08:51AM +0200, David Hildenbrand (Arm) wrote:
> On 6/14/26 15:25, Farhad Alemi wrote:
> > 
> > diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> > --- a/kernel/cgroup/cpuset.c
> > +++ b/kernel/cgroup/cpuset.c
> > @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
> > 
> >  		migrate = is_memory_migrate(cs);
> > 
> > -		mpol_rebind_mm(mm, &cs->mems_allowed);
> > +		mpol_rebind_mm(mm, &cs->effective_mems);
> 
> God this is confusing.
>

All interactions between mempolicy and cpuset are horrible and
confusing.  Much like Lorenzo's anon_vma work, I have to keep
notes on how this whole thing doesn't just spew SIGBUS constantly.

The short answer is: mempolicy is advisory and cpuset is strictly
followed - in a dispute cpuset wins... except for file backed memory,
then everyon loses and nothing is consistent.

> Naturally I wonder: Why are we not using "task->mems_allowed" (maybe cs vs. tsk
> was the original bug?), which is effectively just newmems?
>

Short answer: task->mems_allowed is protected by the task lock and we
don't hold the task lock for a foreign task (not-current) over mm
operations.

Long answer: Reasons and "Stop looking at the spaghetti, it's going to
break"

~Gregory

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/15/26 11:38, Gregory Price wrote:
> On Mon, Jun 15, 2026 at 10:08:51AM +0200, David Hildenbrand (Arm) wrote:
>> On 6/14/26 15:25, Farhad Alemi wrote:
>>>
>>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>>> --- a/kernel/cgroup/cpuset.c
>>> +++ b/kernel/cgroup/cpuset.c
>>> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>>>
>>>  		migrate = is_memory_migrate(cs);
>>>
>>> -		mpol_rebind_mm(mm, &cs->mems_allowed);
>>> +		mpol_rebind_mm(mm, &cs->effective_mems);
>>
>> God this is confusing.
>>
> 
> All interactions between mempolicy and cpuset are horrible and
> confusing.  Much like Lorenzo's anon_vma work, I have to keep
> notes on how this whole thing doesn't just spew SIGBUS constantly.
> 
> The short answer is: mempolicy is advisory and cpuset is strictly
> followed - in a dispute cpuset wins... except for file backed memory,
> then everyon loses and nothing is consistent.
> 
>> Naturally I wonder: Why are we not using "task->mems_allowed" (maybe cs vs. tsk
>> was the original bug?), which is effectively just newmems?
>>
> 
> Short answer: task->mems_allowed is protected by the task lock and we
> don't hold the task lock for a foreign task (not-current) over mm
> operations.

Well, we can just use newmems, which cannot change? Again, that is based on
cs->effective_mems but is guaranteed to return something non-empty.

AI was not able to convince me (neither was I able to convince AI) that there is
not some obscure cgroup v1 scenario where the current fix would also be wrong.

With newmems it's clear that it is guaranteed to not be empty.

-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Gregory Price]

On Mon, Jun 15, 2026 at 01:08:16PM +0200, David Hildenbrand (Arm) wrote:
> With newmems it's clear that it is guaranteed to not be empty.

I hadn't noticed he switched the patch from newmems -> effective_mems.

This needs to be changed back to newmems, otherwise we're depending on
a derivative value set somewhere else in the code being correct instead
of using what we *know* is correct *at the moment we need to use it*.

So yes, go back to using newmems.

~Gregory

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/15/26 13:19, Gregory Price wrote:
> On Mon, Jun 15, 2026 at 01:08:16PM +0200, David Hildenbrand (Arm) wrote:
>> With newmems it's clear that it is guaranteed to not be empty.
> 
> I hadn't noticed he switched the patch from newmems -> effective_mems.
> 
> This needs to be changed back to newmems, otherwise we're depending on
> a derivative value set somewhere else in the code being correct instead
> of using what we *know* is correct *at the moment we need to use it*.
> 
> So yes, go back to using newmems.

Right, that's what v1 did looking at this now. Waiman requested the change, but
I don't think we want that.

So for v1:

Acked-by: David Hildenbrand (Arm) <david@kernel.org>

-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/15/26 5:38 AM, Gregory Price wrote:
> On Mon, Jun 15, 2026 at 10:08:51AM +0200, David Hildenbrand (Arm) wrote:
>> On 6/14/26 15:25, Farhad Alemi wrote:
>>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>>> --- a/kernel/cgroup/cpuset.c
>>> +++ b/kernel/cgroup/cpuset.c
>>> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>>>
>>>   		migrate = is_memory_migrate(cs);
>>>
>>> -		mpol_rebind_mm(mm, &cs->mems_allowed);
>>> +		mpol_rebind_mm(mm, &cs->effective_mems);
>> God this is confusing.
>>
> All interactions between mempolicy and cpuset are horrible and
> confusing.  Much like Lorenzo's anon_vma work, I have to keep
> notes on how this whole thing doesn't just spew SIGBUS constantly.
>
> The short answer is: mempolicy is advisory and cpuset is strictly
> followed - in a dispute cpuset wins... except for file backed memory,
> then everyon loses and nothing is consistent.

That is what I believe why mpol_rebind_mm() a bit differently from the 
others and it is historically done this way a long time ago before 
cgroup v2.

For cgroup v1, mems_allowed can't be empty or you can't put any task 
into the cpuset. Also effective_mems is the same as mems_allowed. cgroup 
v2 is quite different in how it handles memory nodes and CPUs. Users can 
isn't forced to set mems_allowed and cpus_allowed as effective_mems and 
effective_cpus will inherit parent version if mems_allowed and 
cpus_allowed are not set. IOW, effective_mems will never be empty. Yes, 
it is a bug with the introduction of cpuset v2 that we should have 
replaced mems_allowed by effective_mems at that time. With v2, 
effective_mems should contain only online nodes. The only exception is 
during the short transition period when a memory node hotunplug 
operation is in progress when a write to cpuset.mems is happening at the 
same time. With v1, it is theoretically possible that none of the nodes 
in mems_allowed is online.

The reason why I am suggesting to use cs->effective_mems to keep the old 
cgroup v1 behavior. If the consensus is to use the output of 
guarantee_online_mems() for mpol_rebind_mm(), I will not be against that 
but it will be a slight change in user-visible behavior.

Cheers, Longman

>> Naturally I wonder: Why are we not using "task->mems_allowed" (maybe cs vs. tsk
>> was the original bug?), which is effectively just newmems?
>>
> Short answer: task->mems_allowed is protected by the task lock and we
> don't hold the task lock for a foreign task (not-current) over mm
> operations.
>
> Long answer: Reasons and "Stop looking at the spaghetti, it's going to
> break"
>
> ~Gregory
>

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/15/26 10:26 PM, Waiman Long wrote:
>
> On 6/15/26 5:38 AM, Gregory Price wrote:
>> On Mon, Jun 15, 2026 at 10:08:51AM +0200, David Hildenbrand (Arm) wrote:
>>> On 6/14/26 15:25, Farhad Alemi wrote:
>>>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>>>> --- a/kernel/cgroup/cpuset.c
>>>> +++ b/kernel/cgroup/cpuset.c
>>>> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct 
>>>> cpuset *cs)
>>>>
>>>>           migrate = is_memory_migrate(cs);
>>>>
>>>> -        mpol_rebind_mm(mm, &cs->mems_allowed);
>>>> +        mpol_rebind_mm(mm, &cs->effective_mems);
>>> God this is confusing.
>>>
>> All interactions between mempolicy and cpuset are horrible and
>> confusing.  Much like Lorenzo's anon_vma work, I have to keep
>> notes on how this whole thing doesn't just spew SIGBUS constantly.
>>
>> The short answer is: mempolicy is advisory and cpuset is strictly
>> followed - in a dispute cpuset wins... except for file backed memory,
>> then everyon loses and nothing is consistent.
>
> That is what I believe why mpol_rebind_mm() a bit differently from the 
> others and it is historically done this way a long time ago before 
> cgroup v2.
>
> For cgroup v1, mems_allowed can't be empty or you can't put any task 
> into the cpuset. Also effective_mems is the same as mems_allowed. 
> cgroup v2 is quite different in how it handles memory nodes and CPUs. 
> Users can isn't forced to set mems_allowed and cpus_allowed as 
> effective_mems and effective_cpus will inherit parent version if 
> mems_allowed and cpus_allowed are not set. IOW, effective_mems will 
> never be empty. Yes, it is a bug with the introduction of cpuset v2 
> that we should have replaced mems_allowed by effective_mems at that 
> time. With v2, effective_mems should contain only online nodes. The 
> only exception is during the short transition period when a memory 
> node hotunplug operation is in progress when a write to cpuset.mems is 
> happening at the same time. With v1, it is theoretically possible that 
> none of the nodes in mems_allowed is online.
>
> The reason why I am suggesting to use cs->effective_mems to keep the 
> old cgroup v1 behavior. If the consensus is to use the output of 
> guarantee_online_mems() for mpol_rebind_mm(), I will not be against 
> that but it will be a slight change in user-visible behavior. 

BTW, I still prefer the v2 patch. If it is decided we should use the 
guarantee_online_mems() value instead, it will have to be a separate 
patch with changes in the relevant documentation like 
Documentation/admin-guide/cgroup-v1/cpuset.rst.

Cheers,
Longman

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/14/26 9:25 AM, Farhad Alemi wrote:
> Creating a child cpuset where cpuset.mems is never set leads to a div/0
> when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
> CPU hotplug event.
>
> Reproduction steps:
>   1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
>   2) Move the task into the child cpuset
>   3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
>   4) unplug and hotplug a cpu
>        echo 0 > /sys/devices/system/cpu/cpu1/online
>        echo 1 > /sys/devices/system/cpu/cpu1/online
>   5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
>      call to __nodes_fold()
>
> The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
> nodes to the rebind routine.  Use cs->effective_mems instead, which is
> guaranteed to have a non-empty nodemask.
>
> Link: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/
> Link: https://lore.kernel.org/all/CA+0ovCiEz6SP_sn3kN4Tb+_oC=eHMXy_Ffj=usV3wREdQrUtww@mail.gmail.com/
> Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
> Suggested-by: Gregory Price <gourry@gourry.net>
> Suggested-by: Waiman Long <longman@redhat.com>
> Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
> Cc: stable@vger.kernel.org
> ---
> v2: rebind to cs->effective_mems instead of newmems (Waiman Long);
>      condense the changelog.
>
>   kernel/cgroup/cpuset.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2649,7 +2649,7 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>
>   		migrate = is_memory_migrate(cs);
>
> -		mpol_rebind_mm(mm, &cs->mems_allowed);
> +		mpol_rebind_mm(mm, &cs->effective_mems);
>   		if (migrate)
>   			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
>   		else
Acked-by: Waiman Long <longman@redhat.com>

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/16/26 05:43, Waiman Long wrote:
> On 6/15/26 10:26 PM, Waiman Long wrote:
>>
>> On 6/15/26 5:38 AM, Gregory Price wrote:
>>> All interactions between mempolicy and cpuset are horrible and
>>> confusing.  Much like Lorenzo's anon_vma work, I have to keep
>>> notes on how this whole thing doesn't just spew SIGBUS constantly.
>>>
>>> The short answer is: mempolicy is advisory and cpuset is strictly
>>> followed - in a dispute cpuset wins... except for file backed memory,
>>> then everyon loses and nothing is consistent.
>>
>> That is what I believe why mpol_rebind_mm() a bit differently from the others
>> and it is historically done this way a long time ago before cgroup v2.
>>
>> For cgroup v1, mems_allowed can't be empty or you can't put any task into the
>> cpuset. Also effective_mems is the same as mems_allowed. cgroup v2 is quite
>> different in how it handles memory nodes and CPUs. Users can isn't forced to
>> set mems_allowed and cpus_allowed as effective_mems and effective_cpus will
>> inherit parent version if mems_allowed and cpus_allowed are not set. IOW,
>> effective_mems will never be empty. Yes, it is a bug with the introduction of
>> cpuset v2 that we should have replaced mems_allowed by effective_mems at that
>> time. With v2, effective_mems should contain only online nodes. The only
>> exception is during the short transition period when a memory node hotunplug
>> operation is in progress when a write to cpuset.mems is happening at the same
>> time. With v1, it is theoretically possible that none of the nodes in
>> mems_allowed is online.
>>
>> The reason why I am suggesting to use cs->effective_mems to keep the old
>> cgroup v1 behavior. If the consensus is to use the output of
>> guarantee_online_mems() for mpol_rebind_mm(), I will not be against that but
>> it will be a slight change in user-visible behavior. 
> 
> BTW, I still prefer the v2 patch. If it is decided we should use the
> guarantee_online_mems() value instead, it will have to be a separate patch with
> changes in the relevant documentation like Documentation/admin-guide/cgroup-v1/
> cpuset.rst.

newmems is "obviously" correct, so I really don't see why we should add
something that needs half a page of text to explain why it is fine -- if newmems
just does the trick?

Please enlighten me.

-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Gregory Price]

On Tue, Jun 16, 2026 at 08:59:07AM +0200, David Hildenbrand (Arm) wrote:
> On 6/16/26 05:43, Waiman Long wrote:
> > On 6/15/26 10:26 PM, Waiman Long wrote:
> >>
> >>
> >> The reason why I am suggesting to use cs->effective_mems to keep the old
> >> cgroup v1 behavior. If the consensus is to use the output of
> >> guarantee_online_mems() for mpol_rebind_mm(), I will not be against that but
> >> it will be a slight change in user-visible behavior.

I'm not grok'ing what is user-visible here.

The two values should effectively be equivalent because we're
using this value to constrain mpol's during a hotplug event.

If the values differed, you would be saying there's a race condition
that could affect correctness of the rebind (which can't happen,
because this whole thing is done under the hotplug lock btw).

Can you help me understand?

~Gregory

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/16/26 2:59 AM, David Hildenbrand (Arm) wrote:
> On 6/16/26 05:43, Waiman Long wrote:
>> On 6/15/26 10:26 PM, Waiman Long wrote:
>>> On 6/15/26 5:38 AM, Gregory Price wrote:
>>>> All interactions between mempolicy and cpuset are horrible and
>>>> confusing.  Much like Lorenzo's anon_vma work, I have to keep
>>>> notes on how this whole thing doesn't just spew SIGBUS constantly.
>>>>
>>>> The short answer is: mempolicy is advisory and cpuset is strictly
>>>> followed - in a dispute cpuset wins... except for file backed memory,
>>>> then everyon loses and nothing is consistent.
>>> That is what I believe why mpol_rebind_mm() a bit differently from the others
>>> and it is historically done this way a long time ago before cgroup v2.
>>>
>>> For cgroup v1, mems_allowed can't be empty or you can't put any task into the
>>> cpuset. Also effective_mems is the same as mems_allowed. cgroup v2 is quite
>>> different in how it handles memory nodes and CPUs. Users can isn't forced to
>>> set mems_allowed and cpus_allowed as effective_mems and effective_cpus will
>>> inherit parent version if mems_allowed and cpus_allowed are not set. IOW,
>>> effective_mems will never be empty. Yes, it is a bug with the introduction of
>>> cpuset v2 that we should have replaced mems_allowed by effective_mems at that
>>> time. With v2, effective_mems should contain only online nodes. The only
>>> exception is during the short transition period when a memory node hotunplug
>>> operation is in progress when a write to cpuset.mems is happening at the same
>>> time. With v1, it is theoretically possible that none of the nodes in
>>> mems_allowed is online.
>>>
>>> The reason why I am suggesting to use cs->effective_mems to keep the old
>>> cgroup v1 behavior. If the consensus is to use the output of
>>> guarantee_online_mems() for mpol_rebind_mm(), I will not be against that but
>>> it will be a slight change in user-visible behavior.
>> BTW, I still prefer the v2 patch. If it is decided we should use the
>> guarantee_online_mems() value instead, it will have to be a separate patch with
>> changes in the relevant documentation like Documentation/admin-guide/cgroup-v1/
>> cpuset.rst.
> newmems is "obviously" correct, so I really don't see why we should add
> something that needs half a page of text to explain why it is fine -- if newmems
> just does the trick?
>
> Please enlighten me.

Yes, taking newmems is a reasonable choice and there are pros and cons 
with each options. My focus is more on not changing how v1 cpuset 
behaves as it is well defined in the v1 cpusets.rst file:

     Requests by a task, using the sched_setaffinity(2) system call to
     include CPUs in its CPU affinity mask, and using the mbind(2) and
     set_mempolicy(2) system calls to include Memory Nodes in its memory
     policy, are both filtered through that task's cpuset, filtering out any
     CPUs or Memory Nodes not in that cpuset.  The scheduler will not
     schedule a task on a CPU that is not allowed in its cpus_allowed
     vector, and the kernel page allocator will not allocate a page on a
     node that is not allowed in the requesting task's mems_allowed vector.

v2, OTOH, is more vague as to what setting cpuset.mems will mean and we 
generally follow what v1 is doing, but we have more leeway of what we 
can do.

Using newmems will make the above text not totally correct. At least the 
offline memory nodes will be filtered out which will not be utilized by 
the task when the offline node becomes online. That is why I am saying 
that we will have to correct the documentation if we want to make this 
change.

Cheers,
Longman

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/16/26 9:44 AM, Gregory Price wrote:
> On Tue, Jun 16, 2026 at 08:59:07AM +0200, David Hildenbrand (Arm) wrote:
>> On 6/16/26 05:43, Waiman Long wrote:
>>> On 6/15/26 10:26 PM, Waiman Long wrote:
>>>>
>>>> The reason why I am suggesting to use cs->effective_mems to keep the old
>>>> cgroup v1 behavior. If the consensus is to use the output of
>>>> guarantee_online_mems() for mpol_rebind_mm(), I will not be against that but
>>>> it will be a slight change in user-visible behavior.
> I'm not grok'ing what is user-visible here.
>
> The two values should effectively be equivalent because we're
> using this value to constrain mpol's during a hotplug event.
>
> If the values differed, you would be saying there's a race condition
> that could affect correctness of the rebind (which can't happen,
> because this whole thing is done under the hotplug lock btw).
>
> Can you help me understand?

The effective_mems and guarantee_online_mems() should be more or less 
equivalent for v2, but not for v1. In my response to David, I gave an 
example of how the behavior will differ.

Cheers,
Longman

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/16/26 17:23, Waiman Long wrote:
> On 6/16/26 2:59 AM, David Hildenbrand (Arm) wrote:
>> On 6/16/26 05:43, Waiman Long wrote:
>>> BTW, I still prefer the v2 patch. If it is decided we should use the
>>> guarantee_online_mems() value instead, it will have to be a separate patch with
>>> changes in the relevant documentation like Documentation/admin-guide/cgroup-v1/
>>> cpuset.rst.
>> newmems is "obviously" correct, so I really don't see why we should add
>> something that needs half a page of text to explain why it is fine -- if newmems
>> just does the trick?
>>
>> Please enlighten me.
> 
> Yes, taking newmems is a reasonable choice and there are pros and cons with each
> options. My focus is more on not changing how v1 cpuset behaves as it is well
> defined in the v1 cpusets.rst file:
> 
>     Requests by a task, using the sched_setaffinity(2) system call to
>     include CPUs in its CPU affinity mask, and using the mbind(2) and
>     set_mempolicy(2) system calls to include Memory Nodes in its memory
>     policy, are both filtered through that task's cpuset, filtering out any
>     CPUs or Memory Nodes not in that cpuset.  The scheduler will not
>     schedule a task on a CPU that is not allowed in its cpus_allowed
>     vector, and the kernel page allocator will not allocate a page on a
>     node that is not allowed in the requesting task's mems_allowed vector.
> 
> v2, OTOH, is more vague as to what setting cpuset.mems will mean and we
> generally follow what v1 is doing, but we have more leeway of what we can do.
> 
> Using newmems will make the above text not totally correct. At least the offline
> memory nodes will be filtered out which will not be utilized by the task when
> the offline node becomes online. That is why I am saying that we will have to
> correct the documentation if we want to make this change.

So IIUC:

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 1335e437098e..cdfc615f35a5 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2645,7 +2645,13 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
 
                migrate = is_memory_migrate(cs);
 
-               mpol_rebind_mm(mm, &cs->mems_allowed);
+               /*
+                * For v1 we can have empty effective_mems, but we cannot
+                * attach any tasks (see cpuset_can_attach_check()). For v2,
+                * it's guaranteed to not be empty.
+                */
+               VM_WARN_ON_ONCE(nodes_empty(cs->effective_mems));
+               mpol_rebind_mm(mm, &cs->effective_mems);
                if (migrate)
                        cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
                else


-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Waiman Long]

On 6/18/26 4:41 AM, David Hildenbrand (Arm) wrote:
> On 6/16/26 17:23, Waiman Long wrote:
>> On 6/16/26 2:59 AM, David Hildenbrand (Arm) wrote:
>>> On 6/16/26 05:43, Waiman Long wrote:
>>>> BTW, I still prefer the v2 patch. If it is decided we should use the
>>>> guarantee_online_mems() value instead, it will have to be a separate patch with
>>>> changes in the relevant documentation like Documentation/admin-guide/cgroup-v1/
>>>> cpuset.rst.
>>> newmems is "obviously" correct, so I really don't see why we should add
>>> something that needs half a page of text to explain why it is fine -- if newmems
>>> just does the trick?
>>>
>>> Please enlighten me.
>> Yes, taking newmems is a reasonable choice and there are pros and cons with each
>> options. My focus is more on not changing how v1 cpuset behaves as it is well
>> defined in the v1 cpusets.rst file:
>>
>>      Requests by a task, using the sched_setaffinity(2) system call to
>>      include CPUs in its CPU affinity mask, and using the mbind(2) and
>>      set_mempolicy(2) system calls to include Memory Nodes in its memory
>>      policy, are both filtered through that task's cpuset, filtering out any
>>      CPUs or Memory Nodes not in that cpuset.  The scheduler will not
>>      schedule a task on a CPU that is not allowed in its cpus_allowed
>>      vector, and the kernel page allocator will not allocate a page on a
>>      node that is not allowed in the requesting task's mems_allowed vector.
>>
>> v2, OTOH, is more vague as to what setting cpuset.mems will mean and we
>> generally follow what v1 is doing, but we have more leeway of what we can do.
>>
>> Using newmems will make the above text not totally correct. At least the offline
>> memory nodes will be filtered out which will not be utilized by the task when
>> the offline node becomes online. That is why I am saying that we will have to
>> correct the documentation if we want to make this change.
> So IIUC:
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index 1335e437098e..cdfc615f35a5 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2645,7 +2645,13 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>   
>                  migrate = is_memory_migrate(cs);
>   
> -               mpol_rebind_mm(mm, &cs->mems_allowed);
> +               /*
> +                * For v1 we can have empty effective_mems, but we cannot
> +                * attach any tasks (see cpuset_can_attach_check()). For v2,
> +                * it's guaranteed to not be empty.
> +                */
> +               VM_WARN_ON_ONCE(nodes_empty(cs->effective_mems));
> +               mpol_rebind_mm(mm, &cs->effective_mems);
>                  if (migrate)
>                          cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
>                  else

That is true, but I don't think we need a VM_WARN_ON_ONCE() here.

Cheers,
Longman

>

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/21/26 05:24, Waiman Long wrote:
> On 6/18/26 4:41 AM, David Hildenbrand (Arm) wrote:
>> On 6/16/26 17:23, Waiman Long wrote:
>>> Yes, taking newmems is a reasonable choice and there are pros and cons with each
>>> options. My focus is more on not changing how v1 cpuset behaves as it is well
>>> defined in the v1 cpusets.rst file:
>>>
>>>      Requests by a task, using the sched_setaffinity(2) system call to
>>>      include CPUs in its CPU affinity mask, and using the mbind(2) and
>>>      set_mempolicy(2) system calls to include Memory Nodes in its memory
>>>      policy, are both filtered through that task's cpuset, filtering out any
>>>      CPUs or Memory Nodes not in that cpuset.  The scheduler will not
>>>      schedule a task on a CPU that is not allowed in its cpus_allowed
>>>      vector, and the kernel page allocator will not allocate a page on a
>>>      node that is not allowed in the requesting task's mems_allowed vector.
>>>
>>> v2, OTOH, is more vague as to what setting cpuset.mems will mean and we
>>> generally follow what v1 is doing, but we have more leeway of what we can do.
>>>
>>> Using newmems will make the above text not totally correct. At least the offline
>>> memory nodes will be filtered out which will not be utilized by the task when
>>> the offline node becomes online. That is why I am saying that we will have to
>>> correct the documentation if we want to make this change.
>> So IIUC:
>>
>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>> index 1335e437098e..cdfc615f35a5 100644
>> --- a/kernel/cgroup/cpuset.c
>> +++ b/kernel/cgroup/cpuset.c
>> @@ -2645,7 +2645,13 @@ void cpuset_update_tasks_nodemask(struct cpuset *cs)
>>                    migrate = is_memory_migrate(cs);
>>   -               mpol_rebind_mm(mm, &cs->mems_allowed);
>> +               /*
>> +                * For v1 we can have empty effective_mems, but we cannot
>> +                * attach any tasks (see cpuset_can_attach_check()). For v2,
>> +                * it's guaranteed to not be empty.
>> +                */
>> +               VM_WARN_ON_ONCE(nodes_empty(cs->effective_mems));
>> +               mpol_rebind_mm(mm, &cs->effective_mems);
>>                  if (migrate)
>>                          cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
>>                  else
> 
> That is true, but I don't think we need a VM_WARN_ON_ONCE() here.

I'd prefer if we catch such stuff in the future more easily than running into
late divide-by-zero. Maybe we should check in mpol_rebind_mm() instead.

-- 
Cheers,

David

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[Andrew Morton]

On Sun, 14 Jun 2026 06:25:55 -0700 Farhad Alemi <farhad.alemi@berkeley.edu> wrote:

> Creating a child cpuset where cpuset.mems is never set leads to a div/0
> when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
> CPU hotplug event.
> 
> Reproduction steps:
>  1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
>  2) Move the task into the child cpuset
>  3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
>  4) unplug and hotplug a cpu
>       echo 0 > /sys/devices/system/cpu/cpu1/online
>       echo 1 > /sys/devices/system/cpu/cpu1/online
>  5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
>     call to __nodes_fold()

Oops.

> The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
> nodes to the rebind routine.  Use cs->effective_mems instead, which is
> guaranteed to have a non-empty nodemask.

Well gee, what happened with this patch.

I apologize for misfiling a cc:stable bugfix into my post-rc1 backlog
pile, but I got there in the end.

I guess this is an MM patch, even though it's against
kernel/cgroup/cpuset.c.

Nobody cc'ed Tejun.  Fixed.

David acked v1 but is being coy about the v2 patch?

Sashiko AI review suggests that there's a similar bug in
sys_set_mempolicy():

	https://sashiko.dev/#/patchset/CA+0ovCgfHJHv5d1mzapWWvF-LhjppzDX8NPPLvCPZxPKg8RiYw@mail.gmail.com


Anyway, I'll queue the v2 patch as an mm.git hotfix, but not with a lot
of confidence at this time.  Can people please refocus on this and help
recommend a way forward?


From: Farhad Alemi <farhad.alemi@berkeley.edu>
Subject: cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed
Date: Sun, 14 Jun 2026 06:25:55 -0700

Creating a child cpuset where cpuset.mems is never set leads to a div/0
when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
CPU hotplug event.

Reproduction steps:
 1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
 2) Move the task into the child cpuset
 3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
 4) unplug and hotplug a cpu
      echo 0 > /sys/devices/system/cpu/cpu1/online
      echo 1 > /sys/devices/system/cpu/cpu1/online
 5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
    call to __nodes_fold()

The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
nodes to the rebind routine.  Use cs->effective_mems instead, which is
guaranteed to have a non-empty nodemask.

Link: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/
Link: https://lore.kernel.org/all/CA+0ovCiEz6SP_sn3kN4Tb+_oC=eHMXy_Ffj=usV3wREdQrUtww@mail.gmail.com/
Link: https://lore.kernel.org/CA+0ovCgfHJHv5d1mzapWWvF-LhjppzDX8NPPLvCPZxPKg8RiYw@mail.gmail.com
Fixes: ae1c802382f7 ("cpuset: apply cs->effective_{cpus,mems}")
Signed-off-by: Farhad Alemi <farhad.alemi@berkeley.edu>
Suggested-by: Gregory Price <gourry@gourry.net>
Suggested-by: Waiman Long <longman@redhat.com>
Closes: https://lore.kernel.org/linux-mm/CA+0ovCgxbZkXa+OU8w3s84R3KNPNxxRfmsNR-udh+afQBbGNmw@mail.gmail.com/
Acked-by: Waiman Long <longman@redhat.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Gregory Price <gourry@gourry.net>
Cc: "Huang, Ying" <ying.huang@linux.alibaba.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 kernel/cgroup/cpuset.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/kernel/cgroup/cpuset.c~cgroup-cpuset-rebind-mm-mempolicy-to-effective_mems-not-mems_allowed
+++ a/kernel/cgroup/cpuset.c
@@ -2653,7 +2653,7 @@ void cpuset_update_tasks_nodemask(struct
 
 		migrate = is_memory_migrate(cs);
 
-		mpol_rebind_mm(mm, &cs->mems_allowed);
+		mpol_rebind_mm(mm, &cs->effective_mems);
 		if (migrate)
 			cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems);
 		else
_

Re: [PATCH v2] cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed

[David Hildenbrand (Arm)]

On 6/28/26 08:15, Andrew Morton wrote:
> On Sun, 14 Jun 2026 06:25:55 -0700 Farhad Alemi <farhad.alemi@berkeley.edu> wrote:
> 
>> Creating a child cpuset where cpuset.mems is never set leads to a div/0
>> when a VMA mempolicy with MPOL_F_RELATIVE_NODES rebinds in response to a
>> CPU hotplug event.
>>
>> Reproduction steps:
>>  1) Create a cgroup w/ cpuset controls (do not set cpuset.mems)
>>  2) Move the task into the child cpuset
>>  3) Create a VMA mempolicy for that task with MPOL_F_RELATIVE_NODES
>>  4) unplug and hotplug a cpu
>>       echo 0 > /sys/devices/system/cpu/cpu1/online
>>       echo 1 > /sys/devices/system/cpu/cpu1/online
>>  5) mempolicy rebind does a div/0 in mpol_relative_nodemask on the
>>     call to __nodes_fold()
> 
> Oops.
> 
>> The cpuset code passes (cs->mems_allowed) which is not guaranteed to have
>> nodes to the rebind routine.  Use cs->effective_mems instead, which is
>> guaranteed to have a non-empty nodemask.
> 
> Well gee, what happened with this patch.
> 
> I apologize for misfiling a cc:stable bugfix into my post-rc1 backlog
> pile, but I got there in the end.
> 
> I guess this is an MM patch, even though it's against
> kernel/cgroup/cpuset.c.
> 
> Nobody cc'ed Tejun.  Fixed.
> 
> David acked v1 but is being coy about the v2 patch?

Yes, after the discussion, I think we should add comment similar to what I proposed.
-- 
Cheers,

David