On 6/18/2025 8:42 PM, Gregory Price wrote:

On Wed, Jun 18, 2025 at 11:29:32AM +0000, Shivank Garg wrote:

quoted

KVM guest_memfd wants to implement support for NUMA policies just like
shmem already does using the shared policy infrastructure. As
guest_memfd currently resides in KVM module code, we have to export the
relevant symbols.

In the future, guest_memfd might be moved to core-mm, at which point the
symbols no longer would have to be exported. When/if that happens is
still unclear.

Acked-by: David Hildenbrand <redacted>
Acked-by: Vlastimil Babka <redacted>
Signed-off-by: Shivank Garg <redacted>
---
 mm/mempolicy.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 3b1dfd08338b..d98243cdf090 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c

@@ -354,6 +354,7 @@ struct mempolicy *get_task_policy(struct task_struct *p)
 
 	return &default_policy;
 }
+EXPORT_SYMBOL_GPL(get_task_policy);
 
 static const struct mempolicy_operations {
 	int (*create)(struct mempolicy *pol, const nodemask_t *nodes);

@@ -487,6 +488,7 @@ void __mpol_put(struct mempolicy *pol)
 		return;
 	kmem_cache_free(policy_cache, pol);
 }
+EXPORT_SYMBOL_GPL(__mpol_put);
 

I'm concerned that get_task_policy doesn't actually increment the policy
refcount - and mpol_cond_put only decrements the refcount for shared
policies (vma policies) - while __mpol_put decrements it unconditionally.

If you look at how get_task_policy is used internally to mempolicy,
you'll find that it either completes the operation in the context of the
task lock (allocation time) or it calls mpol_get afterwards.

I agree. But the semantics of my usage isn't new. shmem use this in same way.

I think the alloc_frozen_pages_noprof(), alloc_pages_bulk_mempolicy_noprof()
calls get_task_policy without task_lock or calling mpol_get.

Exporting this as-is creates a triping hazard, if only because get/put
naming implies reference counting.

Since KVM is the only user, we could consider newly added EXPORT_SYMBOL_GPL_FOR_MODULES(..., "kvm")
to avoid wider exposure.
Does this solve your concern?
Or should we rename these functions.
What should be the preferred approach?

Thanks,
Shivank