Re: [PATCH RFC 15/17] KVM: guest_memfd: implement userfaultfd missing mode

[Nikita Kalyazin <kalyazin@amazon.com>]

+ David's new email


On 04/03/2026 17:12, Nikita Kalyazin wrote:
> 
> 
> On 27/01/2026 19:29, Mike Rapoport wrote:
>> From: Nikita Kalyazin <kalyazin@amazon.com>
>>
>> userfaultfd missing mode allows populating guest memory with the content
>> supplied by userspace on demand.
>>
>> Extend guest_memfd implementation of vm_uffd_ops to support MISSING
>> mode.
>>
>> Signed-off-by: Nikita Kalyazin <kalyazin@amazon.com>
>> Co-developed-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> ---
>>   virt/kvm/guest_memfd.c | 60 +++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 59 insertions(+), 1 deletion(-)
>>
>> diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
>> index 087e7632bf70..14cca057fc0e 100644
>> --- a/virt/kvm/guest_memfd.c
>> +++ b/virt/kvm/guest_memfd.c
>> @@ -431,6 +431,14 @@ static vm_fault_t 
>> kvm_gmem_fault_user_mapping(struct vm_fault *vmf)
>>                          ret = VM_FAULT_UFFD_MINOR;
>>                          goto out_folio;
>>                  }
>> +
>> +               /*
>> +                * Check if userfaultfd is registered in missing mode. 
>> If so,
>> +                * check if a folio exists in the page cache. If not, 
>> return
>> +                * VM_FAULT_UFFD_MISSING to trigger the userfaultfd 
>> handler.
>> +                */
>> +               if (userfaultfd_missing(vmf->vma) && 
>> IS_ERR_OR_NULL(folio))
>> +                       return VM_FAULT_UFFD_MISSING;
>>          }
>>
>>          /* folio not in the pagecache, try to allocate */
>> @@ -507,9 +515,59 @@ static bool kvm_gmem_can_userfault(struct 
>> vm_area_struct *vma, vm_flags_t vm_fla
>>          return true;
>>   }
>>
>> +static struct folio *kvm_gmem_folio_alloc(struct vm_area_struct *vma,
>> +                                         unsigned long addr)
>> +{
>> +       struct inode *inode = file_inode(vma->vm_file);
>> +       pgoff_t pgoff = linear_page_index(vma, addr);
>> +       struct mempolicy *mpol;
>> +       struct folio *folio;
>> +       gfp_t gfp;
>> +
>> +       if (unlikely(pgoff >= (i_size_read(inode) >> PAGE_SHIFT)))
>> +               return NULL;
>> +
>> +       gfp = mapping_gfp_mask(inode->i_mapping);
>> +       mpol = mpol_shared_policy_lookup(&GMEM_I(inode)->policy, pgoff);
>> +       mpol = mpol ?: get_task_policy(current);
>> +       folio = folio_alloc_mpol(gfp, 0, mpol, pgoff, numa_node_id());
> 
> It looks like folio_alloc_mpol_noprof() and filemap_remove_folio() are 
> not actually exported to modules.  Would it be ok to export them similar 
> to how it was done in f634f10809ec ("mm/mempolicy: Export memory policy 
> symbols")?
> 
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 3e4579e4b8bb..041c7719e524 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -261,6 +261,7 @@ void filemap_remove_folio(struct folio *folio)
> 
>          filemap_free_folio(mapping, folio);
>   }
> +EXPORT_SYMBOL_FOR_MODULES(filemap_remove_folio, "kvm");
> 
>   /*
>    * page_cache_delete_batch - delete several folios from page cache
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 3d797d47a040..1dbbbb28a36e 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -2442,6 +2442,7 @@ struct folio *folio_alloc_mpol_noprof(gfp_t gfp, 
> unsigned int order,
>          set_page_refcounted(page);
>          return page_rmappable_folio(page);
>   }
> +EXPORT_SYMBOL_FOR_MODULES(folio_alloc_mpol_noprof, "kvm");
> 
>   /**
>    * vma_alloc_folio - Allocate a folio for a VMA.
> 
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/ 
> commit/?id=f634f10809ec3d51d9529dfb0f99bc7cec1b5eff
> 
>> +       mpol_cond_put(mpol);
>> +
>> +       return folio;
>> +}
>> +
>> +static int kvm_gmem_filemap_add(struct folio *folio,
>> +                               struct vm_area_struct *vma,
>> +                               unsigned long addr)
>> +{
>> +       struct inode *inode = file_inode(vma->vm_file);
>> +       struct address_space *mapping = inode->i_mapping;
>> +       pgoff_t pgoff = linear_page_index(vma, addr);
>> +       int err;
>> +
>> +       __folio_set_locked(folio);
>> +       err = filemap_add_folio(mapping, folio, pgoff, GFP_KERNEL);
>> +       if (err) {
>> +               folio_unlock(folio);
>> +               return err;
>> +       }
>> +
>> +       return 0;
>> +}
>> +
>> +static void kvm_gmem_filemap_remove(struct folio *folio,
>> +                                   struct vm_area_struct *vma)
>> +{
>> +       filemap_remove_folio(folio);
>> +       folio_unlock(folio);
>> +}
>> +
>>   static const struct vm_uffd_ops kvm_gmem_uffd_ops = {
>> -       .can_userfault = kvm_gmem_can_userfault,
>> +       .can_userfault     = kvm_gmem_can_userfault,
>>          .get_folio_noalloc = kvm_gmem_get_folio_noalloc,
>> +       .alloc_folio       = kvm_gmem_folio_alloc,
>> +       .filemap_add       = kvm_gmem_filemap_add,
>> +       .filemap_remove    = kvm_gmem_filemap_remove,
>>   };
>>   #endif /* CONFIG_USERFAULTFD */
>>
>> -- 
>> 2.51.0
>>
>