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 Thu, Nov 06, 2025 at 06:08:10PM -0800, Jakub Kicinski wrote:
> On Fri, 7 Nov 2025 10:59:02 +0900 Byungchul Park wrote:
> > > > page-backed, the identification cannot be based on the page_type.
> > > > Instead, nmdesc->pp can be used to see if it belongs to a page pool, by
> > > > making sure nmdesc->pp is NULL otherwise.
> > >
> > > Please explain why. Isn't the type just a value in a field?
> > > Which net_iov could also set accordingly.. ?
> >
> > page_type field is in 'struct page', so 'struct page' can check the type.
> >
> > However, the field is not in 'struct net_iov', so 'struct net_iov' that
> > is not backed by page, cannot use the type checking to see if it's page
> > pool'ed instance.
> >
> > I'm afraid I didn't get your questions.  I will try to explain again
> > properly if you give me more detail and example about your questions or
> > requirement.
> 
> net_iov has members in the same place as page. page_type is just
> a field right now.

The current layout 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; // the same offet of page_type
	enum net_iov_type type;
  };

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.

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.

Thoughts?

	Byungchul

> static __always_inline int Page##uname(const struct page *page)         \
> {                                                                       \
>         return data_race(page->page_type >> 24) == PGTY_##lname;        \
> }                                                                       \
> 
> The whole thing works right now by overlaying one struct on top of
> another, and shared members being in the same places.
> 
> Is this clear enough?