[PATCH v6 1/5] mm/zone_device: Reinitialize large zone device private folios

[Vlastimil Babka]

On 1/16/26 17:07, Vlastimil Babka wrote:
> On 1/16/26 12:10, Francois Dugast wrote:
>> From: Matthew Brost <matthew.brost at intel.com>
>> diff --git a/mm/memremap.c b/mm/memremap.c
>> index 63c6ab4fdf08..ac7be07e3361 100644
>> --- a/mm/memremap.c
>> +++ b/mm/memremap.c
>> @@ -477,10 +477,43 @@ void free_zone_device_folio(struct folio *folio)
>>  	}
>>  }
>>  
>> -void zone_device_page_init(struct page *page, unsigned int order)
>> +void zone_device_page_init(struct page *page, struct dev_pagemap *pgmap,
>> +			   unsigned int order)
>>  {
>> +	struct page *new_page = page;
>> +	unsigned int i;
>> +
>>  	VM_WARN_ON_ONCE(order > MAX_ORDER_NR_PAGES);
>>  
>> +	for (i = 0; i < (1UL << order); ++i, ++new_page) {
>> +		struct folio *new_folio = (struct folio *)new_page;
>> +
>> +		/*
>> +		 * new_page could have been part of previous higher order folio
>> +		 * which encodes the order, in page + 1, in the flags bits. We
>> +		 * blindly clear bits which could have set my order field here,
>> +		 * including page head.
>> +		 */
>> +		new_page->flags.f &= ~0xffUL;	/* Clear possible order, page head */
>> +
>> +#ifdef NR_PAGES_IN_LARGE_FOLIO
>> +		/*
>> +		 * This pointer math looks odd, but new_page could have been
>> +		 * part of a previous higher order folio, which sets _nr_pages
>> +		 * in page + 1 (new_page). Therefore, we use pointer casting to
>> +		 * correctly locate the _nr_pages bits within new_page which
>> +		 * could have modified by previous higher order folio.
>> +		 */
>> +		((struct folio *)(new_page - 1))->_nr_pages = 0;
>> +#endif
>> +
>> +		new_folio->mapping = NULL;
>> +		new_folio->pgmap = pgmap;	/* Also clear compound head */
>> +		new_folio->share = 0;   /* fsdax only, unused for device private */
>> +		VM_WARN_ON_FOLIO(folio_ref_count(new_folio), new_folio);
>> +		VM_WARN_ON_FOLIO(!folio_is_zone_device(new_folio), new_folio);
>> +	}
>> +
>>  	/*
>>  	 * Drivers shouldn't be allocating pages after calling
>>  	 * memunmap_pages().
> 
> Can't say I'm a fan of this. It probably works now (so I'm not nacking) but
> seems rather fragile. It seems likely to me somebody will try to change some
> implementation detail in the page allocator and not notice it breaks this,
> for example. I hope we can eventually get to something more robust.

For doing this as a hotfix for 6.19, assuming we'll refactor later:

Acked-by: Vlastimil Babka <vbabka at suse.cz>