Re: [v5 01/15] mm/zone_device: support large zone device private folios

[Balbir Singh]

On 9/18/25 22:27, Chris Mason wrote:
> 
> [ apologies for resending this, debugging sendmail ]
> 
> Hi Balbir,
> 
> On Mon,  8 Sep 2025 10:04:34 +1000 Balbir Singh <balbirs@xxxxxxxxxx> wrote:
>> Add routines to support allocation of large order zone device folios
>> and helper functions for zone device folios, to check if a folio is
>> device private and helpers for setting zone device data.
>>
>> When large folios are used, the existing page_free() callback in
>> pgmap is called when the folio is freed, this is true for both
>> PAGE_SIZE and higher order pages.
>>
>> Zone device private large folios do not support deferred split and
>> scan like normal THP folios.
> 
> [ ... ]
> 
>> diff --git a/mm/memremap.c b/mm/memremap.c
>> index 46cb1b0b6f72..66f9186b5500 100644
>> --- a/mm/memremap.c
>> +++ b/mm/memremap.c
>> @@ -453,11 +452,15 @@ void free_zone_device_folio(struct folio *folio)
>>  
>>  	switch (pgmap->type) {
>>  	case MEMORY_DEVICE_PRIVATE:
>> +		percpu_ref_put_many(&folio->pgmap->ref, nr);
> 
> Here we're dropping nr refs
> 
>> +		pgmap->ops->page_free(&folio->page);
>> +		folio->page.mapping = NULL;
>> +		break;
>>  	case MEMORY_DEVICE_COHERENT:
>>  		if (WARN_ON_ONCE(!pgmap->ops || !pgmap->ops->page_free))
>>  			break;
>> -		pgmap->ops->page_free(folio_page(folio, 0));
>> -		put_dev_pagemap(pgmap);
>> +		pgmap->ops->page_free(&folio->page);
>> +		percpu_ref_put(&folio->pgmap->ref);
> 
> Here we're dropping one ref?
> 
>>  		break;
>>  
>>  	case MEMORY_DEVICE_GENERIC:
>> @@ -480,14 +483,23 @@ void free_zone_device_folio(struct folio *folio)
>>  	}
>>  }
>>  
>> -void zone_device_page_init(struct page *page)
>> +void zone_device_folio_init(struct folio *folio, unsigned int order)
>>  {
>> +	struct page *page = folio_page(folio, 0);
>> +
>> +	VM_WARN_ON_ONCE(order > MAX_ORDER_NR_PAGES);
>> +
>>  	/*
>>  	 * Drivers shouldn't be allocating pages after calling
>>  	 * memunmap_pages().
>>  	 */
>> -	WARN_ON_ONCE(!percpu_ref_tryget_live(&page_pgmap(page)->ref));
>> -	set_page_count(page, 1);
>> +	WARN_ON_ONCE(!percpu_ref_tryget_many(&page_pgmap(page)->ref, 1 << order));
> 
> Here we always bump by 1 << order
> 
> I hesitate to send this one because I don't know the code at all, but the
> AI review prompts keep flagging this apparent refcount mismatch, and it looks
> real to me.
> 
> Are the differences in refcount handling inside free_zone_device_folio()
> intentional?
> 

Thanks for the review! I've posted a v6, but in general

1. Large folios for supported for device-private pages, hence the nr (from folio_order)
2. The user of the folios specifies the order (for mTHP), a folio gets get created
   with that order
3. What you are seeing with the percpu_ref_put for two different cases
   MEMORY_DEVICE_COHERENT and MEMORY_DEVICE_PRIVATE

The reason I point you to v6, is that the diff is a little bit different there

Balbir