Re: [PATCH 3/5] mm: export zap_page_range_single and list_lru_add/del

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Thu, Feb 05, 2026 at 10:51:28AM +0000, Alice Ryhl wrote:
> These are the functions needed by Binder's shrinker.
>
> Binder uses zap_page_range_single in the shrinker path to remove an
> unused page from the mmap'd region. Note that pages are only removed
> from the mmap'd region lazily when shrinker asks for it.
>
> Binder uses list_lru_add/del to keep track of the shrinker lru list, and
> it can't use _obj because the list head is not stored inline in the page
> actually being lru freed, so page_to_nid(virt_to_page(item)) on the list
> head computes the nid of the wrong page.
>
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>
> ---
>  mm/list_lru.c | 2 ++
>  mm/memory.c   | 1 +
>  2 files changed, 3 insertions(+)
>
> diff --git a/mm/list_lru.c b/mm/list_lru.c
> index ec48b5dadf519a5296ac14cda035c067f9e448f8..bf95d73c9815548a19db6345f856cee9baad22e3 100644
> --- a/mm/list_lru.c
> +++ b/mm/list_lru.c
> @@ -179,6 +179,7 @@ bool list_lru_add(struct list_lru *lru, struct list_head *item, int nid,
>  	unlock_list_lru(l, false);
>  	return false;
>  }
> +EXPORT_SYMBOL_GPL(list_lru_add);
>
>  bool list_lru_add_obj(struct list_lru *lru, struct list_head *item)
>  {
> @@ -216,6 +217,7 @@ bool list_lru_del(struct list_lru *lru, struct list_head *item, int nid,
>  	unlock_list_lru(l, false);
>  	return false;
>  }
> +EXPORT_SYMBOL_GPL(list_lru_del);

Same point as before about exporting symbols, but given the _obj variants are
exported already this one is more valid.

>
>  bool list_lru_del_obj(struct list_lru *lru, struct list_head *item)
>  {
> diff --git a/mm/memory.c b/mm/memory.c
> index da360a6eb8a48e29293430d0c577fb4b6ec58099..64083ace239a2caf58e1645dd5d91a41d61492c4 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -2168,6 +2168,7 @@ void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
>  	zap_page_range_single_batched(&tlb, vma, address, size, details);
>  	tlb_finish_mmu(&tlb);
>  }
> +EXPORT_SYMBOL(zap_page_range_single);

Sorry but I don't want this exported at all.

This is an internal implementation detail which allows fine-grained control of
behaviour via struct zap_details (which binder doesn't use, of course :)

We either need a wrapper that eliminates this parameter (but then we're adding a
wrapper to this behaviour that is literally for one driver that is _temporarily_
being modularised which is weak justifiction), or use of a function that invokes
it that is currently exported.

Again the general policy with exports is that we really don't want to provide
them at all if we can help it, and if we do, only when it's really justified.

Drivers by nature generally abuse any interfaces provided, we reduce the surface
area of bugs in the kernel by minimising what's available (even via header for
in-tree...)

>
>  /**
>   * zap_vma_ptes - remove ptes mapping the vma
>
> --
> 2.53.0.rc2.204.g2597b5adb4-goog
>

Cheers, Lorenzo