Re: [PATCH] mm/thp: expose deferred split folio memory usage in meminfo

[Ye Liu <ye.liu@linux.dev>]

在 2026/7/17 16:08, Lorenzo Stoakes (ARM) 写道:
> +cc Johannes
> 
> On Fri, Jul 17, 2026 at 02:30:22PM +0800, Ye Liu wrote:
>> From: Ye Liu <liuye@kylinos.cn>
>>
>> Folios on the deferred split list hold physical memory that is
>> invisible in meminfo. When a THP becomes partially mapped, the
>> unmapped pages are removed from AnonPages but remain physically
>> allocated until the shrinker splits the folio. This creates a
>> memory accounting gap where used memory cannot be attributed to
>> any meminfo field.
> 
> Is this really that much of an issue? You're not giving any use cases here.
> 
> What real-world use case motivated this?
> 
I have indeed encountered this situation in a customer environment,                           
but the environment is complex, and I cannot clearly explain why this memory                  
black hole occurs in that scenario.                                                           
However, I will provide an example to reproduce it below.                                     
Use case: The system has 100GB of memory. A user-space program requests 80GB of memory.       
For each 2MB block, only one page is retained, and the mapping of all other pages is unmapped,
while the process continues to run.                                                           
You'll find that in meminfo, MemFree: only 12GB remains,                                      
but the process actually uses very little memory.                                             
This creates a memory black hole situation.                                                   
Below is the test program I used, from AI.                                                    


>>
>> Add NR_DEFERRED_SPLIT_PAGES to track the total memory consumed by
>> folios currently on the deferred_split_lru, updated via
>> mod_node_page_state() at all enqueue/dequeue points. The new field
>> DeferredSplitPages is visible in /proc/meminfo, /proc/vmstat, and
>> per-node /sys/devices/system/node/node*/meminfo.
> 
> Again you're not justifying anything, why do you need a per-node breakdown?
> 
Initially, the only plan was to add it to /proc/meminfo;
the others were more of a transitional design.          

>>
>> Signed-off-by: Ye Liu <liuye@kylinos.cn>
> 
> You're updating these stats in a number of places and you've given zero
> explanataion or evidence that you're accounting this correctly.
> 

The counter tracks the current total pages of partially_mapped folios on deferred_split_lru.
The six hook points:                                                                        
- Add: deferred_split_folio() when partially_mapped → the only first-enqueue path           
- Add: deferred_split_scan() requeue → net effect is zero paired with isolate's del         
- Del: deferred_split_isolate() → temporary removal for shrinker processing                 
- Del: __folio_unqueue_deferred_split() → permanent removal when folio is freed             
- Del: split path → permanent removal when folio is split into sub-folios                   

> This is very subtle code and you're a very new contributor to this so I
> wouldn't want us to take someting like this even if it was justified
> without a very strong argument.

True – I’m new to the kernel code, so I’ll definitely benefit from checking in 
with you guys more regularly.                                                  

> 
> You'd probably want to do some refactoring stuff first also to combine any
> such updates with moves.
> 
Indeed, it would be better to integrate it with folio updates.
>> ---
>>  drivers/base/node.c    |  4 +++-
>>  fs/proc/meminfo.c      |  2 ++
>>  include/linux/mmzone.h |  1 +
>>  mm/huge_memory.c       | 24 +++++++++++++++++++++---
>>  mm/vmstat.c            |  1 +
>>  5 files changed, 28 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/base/node.c b/drivers/base/node.c
>> index 3da91929ad4e..69463438300e 100644
>> --- a/drivers/base/node.c
>> +++ b/drivers/base/node.c
>> @@ -519,6 +519,7 @@ static ssize_t node_read_meminfo(struct device *dev,
>>  			     "Node %d ShmemPmdMapped: %8lu kB\n"
>>  			     "Node %d FileHugePages:  %8lu kB\n"
>>  			     "Node %d FilePmdMapped:  %8lu kB\n"
>> +			     "Node %d DeferredSplitPages: %8lu kB\n"
>>  #endif
>>  #ifdef CONFIG_UNACCEPTED_MEMORY
>>  			     "Node %d Unaccepted:     %8lu kB\n"
>> @@ -553,7 +554,8 @@ static ssize_t node_read_meminfo(struct device *dev,
>>  			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
>>  			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
>>  			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
>> -			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
>> +			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED)),
>> +			     nid, K(node_page_state(pgdat, NR_DEFERRED_SPLIT_PAGES))
>>  #endif
>>  #ifdef CONFIG_UNACCEPTED_MEMORY
>>  			     ,
>> diff --git a/fs/proc/meminfo.c b/fs/proc/meminfo.c
>> index b2813ff13cb2..76dbf51ad1e0 100644
>> --- a/fs/proc/meminfo.c
>> +++ b/fs/proc/meminfo.c
>> @@ -149,6 +149,8 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
>>  		    global_node_page_state(NR_FILE_THPS));
>>  	show_val_kb(m, "FilePmdMapped:  ",
>>  		    global_node_page_state(NR_FILE_PMDMAPPED));
>> +	show_val_kb(m, "DeferredSplitPages: ",
>> +		    global_node_page_state(NR_DEFERRED_SPLIT_PAGES));
>>  #endif
>>
>>  #ifdef CONFIG_CMA
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index 0507193b3ae3..1e75adb73d00 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -267,6 +267,7 @@ enum node_stat_item {
>>  	NR_FILE_THPS,
>>  	NR_FILE_PMDMAPPED,
>>  	NR_ANON_THPS,
>> +	NR_DEFERRED_SPLIT_PAGES, /* THP/mTHP pages pending deferred split */
>>  	NR_VMSCAN_WRITE,
>>  	NR_VMSCAN_IMMEDIATE,	/* Prioritise for reclaim when writeback ends */
>>  	NR_DIRTIED,		/* page dirtyings since bootup */
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index c642bec967fa..1094b844675c 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -77,6 +77,14 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
>>  					 struct shrink_control *sc);
>>  static bool split_underused_thp = true;
>>
>> +#define deferred_split_pages_add(folio) \
>> +	mod_node_page_state(NODE_DATA(folio_nid(folio)), \
>> +			    NR_DEFERRED_SPLIT_PAGES, folio_nr_pages(folio))
>> +
>> +#define deferred_split_pages_del(folio) \
>> +	mod_node_page_state(NODE_DATA(folio_nid(folio)), \
>> +			    NR_DEFERRED_SPLIT_PAGES, -folio_nr_pages(folio))
> 
> Why are these macros?
> 
> I really dislike the naming too, these read like they're actually
> adding/deleting pages.
> 
Rename the helpers to avoid the "add/del pages" confusion           
(e.g. mod_deferred_split_stat() or just inline mod_node_page_state).

>> +
>>  static atomic_t huge_zero_refcount;
>>  struct folio *huge_zero_folio __read_mostly;
>>  unsigned long huge_zero_pfn __read_mostly = ~0UL;
>> @@ -3913,8 +3921,10 @@ static int __folio_freeze_and_split_unmapped(struct folio *folio, unsigned int n
>>  		struct lruvec *lruvec;
>>
>>  		if (dequeue_deferred) {
>> -			__list_lru_del(&deferred_split_lru, lru,
>> -				       &folio->_deferred_list, folio_nid(folio));
>> +			if (__list_lru_del(&deferred_split_lru, lru,
>> +					  &folio->_deferred_list, folio_nid(folio)) &&
>> +			    folio_test_partially_mapped(folio))
>> +				deferred_split_pages_del(folio);
>>  			if (folio_test_partially_mapped(folio)) {
>>  				folio_clear_partially_mapped(folio);
>>  				mod_mthp_stat(old_order,
>> @@ -4415,6 +4425,7 @@ bool __folio_unqueue_deferred_split(struct folio *folio)
>>  	lru = list_lru_lock_irqsave(&deferred_split_lru, nid, &memcg, &flags);
>>  	if (__list_lru_del(&deferred_split_lru, lru, &folio->_deferred_list, nid)) {
>>  		if (folio_test_partially_mapped(folio)) {
>> +			deferred_split_pages_del(folio);
>>  			folio_clear_partially_mapped(folio);
>>  			mod_mthp_stat(folio_order(folio),
>>  				      MTHP_STAT_NR_ANON_PARTIALLY_MAPPED, -1);
>> @@ -4473,6 +4484,8 @@ void deferred_split_folio(struct folio *folio, bool partially_mapped)
>>  		VM_WARN_ON_FOLIO(folio_test_partially_mapped(folio), folio);
>>  	}
>>  	__list_lru_add(&deferred_split_lru, lru, &folio->_deferred_list, nid, memcg);
>> +	if (partially_mapped)
>> +		deferred_split_pages_add(folio);
>>  	list_lru_unlock_irqrestore(lru, &flags);
>>  	rcu_read_unlock();
>>  }
>> @@ -4524,8 +4537,11 @@ static enum lru_status deferred_split_isolate(struct list_head *item,
>>  {
>>  	struct folio *folio = container_of(item, struct folio, _deferred_list);
>>  	struct list_head *freeable = cb_arg;
>> +	bool partially_mapped = folio_test_partially_mapped(folio);
>>
>>  	if (folio_try_get(folio)) {
>> +		if (partially_mapped)
>> +			deferred_split_pages_del(folio);
>>  		list_lru_isolate_move(lru, item, freeable);
>>  		return LRU_REMOVED;
>>  	}
>> @@ -4535,7 +4551,8 @@ static enum lru_status deferred_split_isolate(struct list_head *item,
>>  	 * isolate: folio_unqueue_deferred_split() checks list_empty()
>>  	 * locklessly, so once removed the folio can be freed any time.
>>  	 */
>> -	if (folio_test_partially_mapped(folio)) {
>> +	if (partially_mapped) {
>> +		deferred_split_pages_del(folio);
>>  		folio_clear_partially_mapped(folio);
>>  		mod_mthp_stat(folio_order(folio),
>>  			      MTHP_STAT_NR_ANON_PARTIALLY_MAPPED, -1);
>> @@ -4596,6 +4613,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
>>  					 &folio->_deferred_list,
>>  					 folio_nid(folio),
>>  					 folio_memcg(folio));
>> +			deferred_split_pages_add(folio);
>>  			rcu_read_unlock();
>>  		}
>>  		folio_put(folio);
>> diff --git a/mm/vmstat.c b/mm/vmstat.c
>> index 3b5cb1031f72..ba71cc3e4c30 100644
>> --- a/mm/vmstat.c
>> +++ b/mm/vmstat.c
>> @@ -1248,6 +1248,7 @@ const char * const vmstat_text[] = {
>>  	[I(NR_FILE_THPS)]			= "nr_file_hugepages",
>>  	[I(NR_FILE_PMDMAPPED)]			= "nr_file_pmdmapped",
>>  	[I(NR_ANON_THPS)]			= "nr_anon_transparent_hugepages",
>> +	[I(NR_DEFERRED_SPLIT_PAGES)]		= "nr_deferred_split_pages",
>>  	[I(NR_VMSCAN_WRITE)]			= "nr_vmscan_write",
>>  	[I(NR_VMSCAN_IMMEDIATE)]		= "nr_vmscan_immediate_reclaim",
>>  	[I(NR_DIRTIED)]				= "nr_dirtied",
>> --
>> 2.43.0
>>
> 
> Cheers, Lorenzo

-- 
Thanks,
Ye Liu