Re: [PATCH] mm/fadvise: avoid remote LRU drain for mapped folio failures

[Fujunjie]

On 5/11/2026 1:47 PM, David Hildenbrand (Arm) wrote:
>> -long mapping_evict_folio(struct address_space *mapping, struct folio *folio)
>> +static long __mapping_evict_folio(struct address_space *mapping,
>> +				  struct folio *folio, bool *lru_refs)
>>  {
>> +	if (lru_refs)
>> +		*lru_refs = false;
>>  	/* The page may have been truncated before it was locked */
>>  	if (!mapping)
>>  		return 0;
>> @@ -331,14 +323,38 @@ long mapping_evict_folio(struct address_space *mapping, struct folio *folio)
>>  		return 0;
>>  	/* The refcount will be elevated if any page in the folio is mapped */
>>  	if (folio_ref_count(folio) >
>> -			folio_nr_pages(folio) + folio_has_private(folio) + 1)
>> +			folio_nr_pages(folio) + folio_has_private(folio) + 1) {
>> +		/*
>> +		 * A remote LRU drain can only help with extra references on
>> +		 * otherwise evictable folios.  Mapped folios also have an
>> +		 * elevated refcount, but draining LRU caches cannot unmap them.
>> +		 */
>> +		if (lru_refs && !folio_mapped(folio))
>> +			*lru_refs = true;
> 
> How can you conclude that it's a LRU refs? Could also just be something trivial
> like a remaining GUP reference, or am I wrong?
> 

You're right. The elevated refcount does not prove that the extra
reference is from a remote LRU batch; it could be GUP or another transient
reference.

The intent was to use this as an approximate filter: avoid the remote
drain for failures where it clearly cannot help, such as mapped folios,
rather than to prove the exact source of the extra reference.

"lru_refs" is too strong or misleading as a name. For v2 I will take
Matthew's suggestion into account as well: avoid adding
__mapping_evict_folio(), do the folio_mapped() test in
mapping_try_invalidate(), and think more carefully about where the
fallback boundary should be. I will also use naming that describes the
decision more accurately rather than implying that the extra reference
has been proven to be an LRU-batch reference.