Re: [RFC mm v5 1/2] page_pool: check nmdesc->pp to see its usage as page pool for net_iov not page-backed

[Byungchul Park <byungchul@sk.com>]

On Sat, Nov 08, 2025 at 11:24:58AM +0900, Byungchul Park wrote:
> On Fri, Nov 07, 2025 at 05:41:29PM -0800, Jakub Kicinski wrote:
> > On Fri, 7 Nov 2025 13:47:08 +0900 Byungchul Park wrote:
> > > The offset of page_type in struct page cannot be used in struct net_iov
> > > for the same purpose, since the offset in struct net_iov is for storing
> > > (struct net_iov_area *)owner.
> > 
> > owner does not have to be at a fixed offset. Can we not move owner
> > to _pp_mapping_pad ? Or reorder it with type, enum net_iov_type
> > only has 2 values we can smoosh it with page_type easily.
> 
> I'm still confused.  I think you probably understand what this work is
> for.  (I've explained several times with related links.)  Or am I
         ^
Please don't mind.  It's not a blame.

	Byungchul

> missing something from your questions?
> 
> I've answered your question directly since you asked, but the point is
> that, struct net_iov will no longer overlay on struct page.
> 
> Instead, struct netmem_desc will be responsible for keeping the pp
> fields while struct page will lay down the resonsibility, once the pp
> fields will be removed from struct page like:
> 
> <before> (the current form is:)
> 
>    struct page {
>  	memdesc_flags_t flags;
>  	union {
>  		...
>  		struct {
>  			unsigned long pp_magic;
>  			struct page_pool *pp;
>  			unsigned long _pp_mapping_pad;
>  			unsigned long dma_addr;
>  			atomic_long_t pp_ref_count;
>  		};
>  		...
>  	};
>  	unsigned int page_type;
>  	...
>    };
>  
>    struct net_iov {
>  	union {
>  		struct netmem_desc desc;
>  		struct
>  		{
>  			unsigned long _flags;
>  			unsigned long pp_magic;
>  			struct page_pool *pp;
>  			unsigned long _pp_mapping_pad;
>  			unsigned long dma_addr;
>  			atomic_long_t pp_ref_count;
>  		};
>  	};
>  	struct net_iov_area *owner;
>  	enum net_iov_type type;
>    };
> 
> <after> (the final form should be, just around the corner:)
> 
>    struct page {
>  	memdesc_flags_t flags;
>  	union {
>  		...
> 		/* pp fields are gone. */
>  		...
>  	};
>  	unsigned int page_type;
>  	...
>    };
>  
>    struct net_iov {
>  	struct netmem_desc desc;
>  	struct net_iov_area *owner;
>  	enum net_iov_type type;
>    };
> 
> After that, struct page and struct net_iov(or struct netmem_desc) will
> not share any fields with each other, probably they will be connected
> e.i. through some ways between struct page and netmem_desc tho.
> 
> 	Byungchul
>  
> > > Yeah, you can tell 'why don't we add the field, page_type, to struct
> > > net_iov (or struct netmem_desc)' so as to be like:
> > >
> > >   struct net_iov {
> > >       union {
> > >               struct netmem_desc desc;
> > >               struct
> > >               {
> > >                       unsigned long _flags;
> > >                       unsigned long pp_magic;
> > >                       struct page_pool *pp;
> > >                       unsigned long _pp_mapping_pad;
> > >                       unsigned long dma_addr;
> > >                       atomic_long_t pp_ref_count;
> > > +                     unsigned int page_type; // add this field newly
> > >               };
> > >       };
> > >       struct net_iov_area *owner; // the same offet of page_type
> > >       enum net_iov_type type;
> > >   };
> > >
> > > I think we can make it anyway but it makes less sense to add page_type
> > > to struct net_iov, only for PGTY_netpp.
> > >
> > > It'd be better to use netmem_desc->pp for that purpose, IMO.