Re: [PATCH v12 18/45] KVM: guest_memfd: Handle lru_add fbatch refcounts during conversion safety check

[Hugh Dickins <hughd@google.com>]

On Sun, 30 Aug 2026, Ackerley Tng via B4 Relay wrote:

> From: Ackerley Tng <ackerleytng@google.com>
> 
> A guest_memfd folio has no outstanding references if guest_memfd holds the
> only references on it. Any other references on the folio may indicate
> another user, and guest_memfd cannot convert it to private if there may be
> an existing host user.
> 
> A folio will have outstanding references if it is present in a per-CPU
> lru_add fbatch. guest_memfd does not actually participate in LRU, but
> freshly-allocated folios are still added to the lru_add fbatch for batch
> LRU statistics processing.
> 
> A folio may also have extra refcounts if it is on the mlock fbatch.
> 
> These two known "usages" of the folio are handled by calling
> lru_cache_drain_for_folio, which drains both the lru_add and mlock
> fbatches. After draining, if the refcount is still elevated, then there are
> truly outstanding references.
> 
> If the page may be dma pinned, DMA is using it and hence there are
> outstanding references. folio_maybe_dma_pinned() can have false positives,
> but that's only with a significant number of refcounts, at which point
> draining LRU is not going to move the needle - it can still be concluded
> that the folio has outstanding references.
> 
> If the page is still mapped after guest_memfd tried to unmap it earlier in
> the conversion process, it also has outstanding references.
> 
> Return true and exit early to avoid unnecessary draining in these 2 cases.
> 
> Provide a drain status to only drain once ever while processing a batch of
> folios.
> 
> Acked-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
> Suggested-by: David Hildenbrand <david@kernel.org>
> Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev>
> Reviewed-by: Binbin Wu <binbin.wu@linux.intel.com>
> Signed-off-by: Ackerley Tng <ackerleytng@google.com>
> ---
>  mm/folio.c             |  2 ++
>  virt/kvm/guest_memfd.c | 30 ++++++++++++++++++++++--------
>  2 files changed, 24 insertions(+), 8 deletions(-)
> 
> diff --git a/mm/folio.c b/mm/folio.c
> index c02dcea9c03c2..50a6dbe55998e 100644
> --- a/mm/folio.c
> +++ b/mm/folio.c
> @@ -33,6 +33,7 @@
>  #include <linux/page_idle.h>
>  #include <linux/local_lock.h>
>  #include <linux/buffer_head.h>
> +#include <linux/kvm_types.h>
>  
>  #include "internal.h"
>  #include "page_alloc.h"
> @@ -926,6 +927,7 @@ void lru_cache_drain_for_folio(const struct folio *folio,
>  			*drained = LRU_CACHE_DRAINED_ALL;
>  	}
>  }
> +EXPORT_SYMBOL_FOR_KVM(lru_cache_drain_for_folio);
>  
>  atomic_t lru_disable_count = ATOMIC_INIT(0);
>  

I don't mind about the virt/kvm/guest_memfd.c part of it, but I'm finding
a KVM patchset modifying mm/folio.c there hard to deal with: and notice
Sean also suggesting to separate this part out.

As you know, I've worked up a patchset "mm/fbatch: drain lru_add_drain()
and _all()" which finally removes the problem lru_cache_drain_for_folio()
works around. In the initial version posted a week ago, there was no
lru_cache_drain_for_folio() in the tree. Now 7.3-rc1 has it, so I
intended a replacement 13/25 in my series, giving you just an empty
inline lru_cache_drain_for_folio() stub (and enum lru_cache_drained)
in linux/swap.h.

But that won't work for you, if you're adding an EXPORT_SYMBOL_FOR_KVM()
in mm/folio.c, and of course conflicts with my removals (in context both
above and below your EXPORT line). It's easy for me to remove what's in
mm/gup.c and mm/folio.c, but I cannot remove what is not yet there.

I've wasted hours on this, hoping not to trouble either of you; but
seeing now that I shall have to rebase anyway (an unrelated mlock fix),
I'm electing to take the only clean way out: I'm going to submit this
mm/folio.c part of your patch to Andrew tonight (with a shorter Cc list!),
in the hope that it can be accelerated into 7.3-rc2 (or at least get an
mm-stable stable base-commit id) which we can both work off independently.

Whether that's acceptable to Ackerley and to Andrew, I don't know
(just as we don't know when either of our patchsets will go further),
but let me try.

Thanks,
Hugh