Re: [PATCH] mm, page_pool: introduce a new page type for page pool in page type

[Byungchul Park <byungchul@sk.com>]

On Thu, Jul 24, 2025 at 02:23:05PM -0700, Mina Almasry wrote:
> On Tue, Jul 22, 2025 at 9:46 PM Byungchul Park <byungchul@sk.com> wrote:
> >
> > On Tue, Jul 22, 2025 at 03:17:15PM -0700, Mina Almasry wrote:
> > > On Sun, Jul 20, 2025 at 10:49 PM Byungchul Park <byungchul@sk.com> wrote:
> > > >
> > > > Hi,
> > > >
> > > > I focused on converting the existing APIs accessing ->pp_magic field to
> > > > page type APIs.  However, yes.  Additional works would better be
> > > > considered on top like:
> > > >
> > > >    1. Adjust how to store and retrieve dma index.  Maybe network guys
> > > >       can work better on top.
> > > >
> > > >    2. Move the sanity check for page pool in mm/page_alloc.c to on free.
> > > >
> > > >    Byungchul
> > > >
> > > > ---8<---
> > > > From 7d207a1b3e9f4ff2a72f5b54b09e3ed0c4aaaca3 Mon Sep 17 00:00:00 2001
> > > > From: Byungchul Park <byungchul@sk.com>
> > > > Date: Mon, 21 Jul 2025 14:05:20 +0900
> > > > Subject: [PATCH] mm, page_pool: introduce a new page type for page pool in page type
> > > >
> > > > ->pp_magic field in struct page is current used to identify if a page
> > > > belongs to a page pool.  However, page type e.i. PGTY_netpp can be used
> > > > for that purpose.
> > > >
> > > > Use the page type APIs e.g. PageNetpp(), __SetPageNetpp(), and
> > > > __ClearPageNetpp() instead, and remove the existing APIs accessing
> > > > ->pp_magic e.g. page_pool_page_is_pp(), netmem_or_pp_magic(), and
> > > > netmem_clear_pp_magic() since they are totally replaced.
> > > >
> > > > This work was inspired by the following link by Pavel:
> > > >
> > > > [1] https://lore.kernel.org/all/582f41c0-2742-4400-9c81-0d46bf4e8314@gmail.com/
> > > >
> > > > Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
> > > > Signed-off-by: Byungchul Park <byungchul@sk.com>
> > > > ---
> > > >  .../net/ethernet/mellanox/mlx5/core/en/xdp.c  |  2 +-
> > > >  include/linux/mm.h                            | 28 ++-----------------
> > > >  include/linux/page-flags.h                    |  6 ++++
> > > >  include/net/netmem.h                          |  2 +-
> > > >  mm/page_alloc.c                               |  4 +--
> > > >  net/core/netmem_priv.h                        | 16 ++---------
> > > >  net/core/page_pool.c                          | 10 +++++--
> > > >  7 files changed, 24 insertions(+), 44 deletions(-)
> > > >
> > > > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
> > > > index 5d51600935a6..def274f5c1ca 100644
> > > > --- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
> > > > +++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
> > > > @@ -707,7 +707,7 @@ static void mlx5e_free_xdpsq_desc(struct mlx5e_xdpsq *sq,
> > > >                                 xdpi = mlx5e_xdpi_fifo_pop(xdpi_fifo);
> > > >                                 page = xdpi.page.page;
> > > >
> > > > -                               /* No need to check page_pool_page_is_pp() as we
> > > > +                               /* No need to check PageNetpp() as we
> > > >                                  * know this is a page_pool page.
> > > >                                  */
> > > >                                 page_pool_recycle_direct(pp_page_to_nmdesc(page)->pp,
> > > > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > > > index ae50c1641bed..736061749535 100644
> > > > --- a/include/linux/mm.h
> > > > +++ b/include/linux/mm.h
> > > > @@ -4135,10 +4135,9 @@ int arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status);
> > > >   * DMA mapping IDs for page_pool
> > > >   *
> > > >   * When DMA-mapping a page, page_pool allocates an ID (from an xarray) and
> > > > - * stashes it in the upper bits of page->pp_magic. We always want to be able to
> > > > - * unambiguously identify page pool pages (using page_pool_page_is_pp()). Non-PP
> > > > - * pages can have arbitrary kernel pointers stored in the same field as pp_magic
> > > > - * (since it overlaps with page->lru.next), so we must ensure that we cannot
> > > > + * stashes it in the upper bits of page->pp_magic. Non-PP pages can have
> > > > + * arbitrary kernel pointers stored in the same field as pp_magic (since
> > > > + * it overlaps with page->lru.next), so we must ensure that we cannot
> > > >   * mistake a valid kernel pointer with any of the values we write into this
> > > >   * field.
> > > >   *
> > > > @@ -4168,25 +4167,4 @@ int arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status);
> > > >
> > > >  #define PP_DMA_INDEX_MASK GENMASK(PP_DMA_INDEX_BITS + PP_DMA_INDEX_SHIFT - 1, \
> > > >                                   PP_DMA_INDEX_SHIFT)
> > > > -
> > > > -/* Mask used for checking in page_pool_page_is_pp() below. page->pp_magic is
> > > > - * OR'ed with PP_SIGNATURE after the allocation in order to preserve bit 0 for
> > > > - * the head page of compound page and bit 1 for pfmemalloc page, as well as the
> > > > - * bits used for the DMA index. page_is_pfmemalloc() is checked in
> > > > - * __page_pool_put_page() to avoid recycling the pfmemalloc page.
> > > > - */
> > > > -#define PP_MAGIC_MASK ~(PP_DMA_INDEX_MASK | 0x3UL)
> > > > -
> > > > -#ifdef CONFIG_PAGE_POOL
> > > > -static inline bool page_pool_page_is_pp(const struct page *page)
> > > > -{
> > > > -       return (page->pp_magic & PP_MAGIC_MASK) == PP_SIGNATURE;
> > > > -}
> > > > -#else
> > > > -static inline bool page_pool_page_is_pp(const struct page *page)
> > > > -{
> > > > -       return false;
> > > > -}
> > > > -#endif
> > > > -
> > > >  #endif /* _LINUX_MM_H */
> > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > index 4fe5ee67535b..906ba7c9e372 100644
> > > > --- a/include/linux/page-flags.h
> > > > +++ b/include/linux/page-flags.h
> > > > @@ -957,6 +957,7 @@ enum pagetype {
> > > >         PGTY_zsmalloc           = 0xf6,
> > > >         PGTY_unaccepted         = 0xf7,
> > > >         PGTY_large_kmalloc      = 0xf8,
> > > > +       PGTY_netpp              = 0xf9,
> > > >
> > > >         PGTY_mapcount_underflow = 0xff
> > > >  };
> > > > @@ -1101,6 +1102,11 @@ PAGE_TYPE_OPS(Zsmalloc, zsmalloc, zsmalloc)
> > > >  PAGE_TYPE_OPS(Unaccepted, unaccepted, unaccepted)
> > > >  FOLIO_TYPE_OPS(large_kmalloc, large_kmalloc)
> > > >
> > > > +/*
> > > > + * Marks page_pool allocated pages.
> > > > + */
> > > > +PAGE_TYPE_OPS(Netpp, netpp, netpp)
> > > > +
> > > >  /**
> > > >   * PageHuge - Determine if the page belongs to hugetlbfs
> > > >   * @page: The page to test.
> > > > diff --git a/include/net/netmem.h b/include/net/netmem.h
> > > > index f7dacc9e75fd..3667334e16e7 100644
> > > > --- a/include/net/netmem.h
> > > > +++ b/include/net/netmem.h
> > > > @@ -298,7 +298,7 @@ static inline struct net_iov *__netmem_clear_lsb(netmem_ref netmem)
> > > >   */
> > > >  #define pp_page_to_nmdesc(p)                                           \
> > > >  ({                                                                     \
> > > > -       DEBUG_NET_WARN_ON_ONCE(!page_pool_page_is_pp(p));               \
> > > > +       DEBUG_NET_WARN_ON_ONCE(!PageNetpp(p));                          \
> > > >         __pp_page_to_nmdesc(p);                                         \
> > > >  })
> > > >
> > > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > > index 2ef3c07266b3..71c7666e48a9 100644
> > > > --- a/mm/page_alloc.c
> > > > +++ b/mm/page_alloc.c
> > > > @@ -898,7 +898,7 @@ static inline bool page_expected_state(struct page *page,
> > > >  #ifdef CONFIG_MEMCG
> > > >                         page->memcg_data |
> > > >  #endif
> > > > -                       page_pool_page_is_pp(page) |
> > > > +                       PageNetpp(page) |
> > > >                         (page->flags & check_flags)))
> > > >                 return false;
> > > >
> > > > @@ -925,7 +925,7 @@ static const char *page_bad_reason(struct page *page, unsigned long flags)
> > > >         if (unlikely(page->memcg_data))
> > > >                 bad_reason = "page still charged to cgroup";
> > > >  #endif
> > > > -       if (unlikely(page_pool_page_is_pp(page)))
> > > > +       if (unlikely(PageNetpp(page)))
> > > >                 bad_reason = "page_pool leak";
> > > >         return bad_reason;
> > > >  }
> > > > diff --git a/net/core/netmem_priv.h b/net/core/netmem_priv.h
> > > > index cd95394399b4..39a97703d9ed 100644
> > > > --- a/net/core/netmem_priv.h
> > > > +++ b/net/core/netmem_priv.h
> > > > @@ -8,21 +8,11 @@ static inline unsigned long netmem_get_pp_magic(netmem_ref netmem)
> > > >         return __netmem_clear_lsb(netmem)->pp_magic & ~PP_DMA_INDEX_MASK;
> > > >  }
> > > >
> > > > -static inline void netmem_or_pp_magic(netmem_ref netmem, unsigned long pp_magic)
> > > > -{
> > > > -       __netmem_clear_lsb(netmem)->pp_magic |= pp_magic;
> > > > -}
> > > > -
> > > > -static inline void netmem_clear_pp_magic(netmem_ref netmem)
> > > > -{
> > > > -       WARN_ON_ONCE(__netmem_clear_lsb(netmem)->pp_magic & PP_DMA_INDEX_MASK);
> > > > -
> > > > -       __netmem_clear_lsb(netmem)->pp_magic = 0;
> > > > -}
> > > > -
> > > >  static inline bool netmem_is_pp(netmem_ref netmem)
> > > >  {
> > > > -       return (netmem_get_pp_magic(netmem) & PP_MAGIC_MASK) == PP_SIGNATURE;
> > > > +       if (netmem_is_net_iov(netmem))
> > > > +               return true;
> > >
> > > As Pavel alludes, this is dubious, and at least it's difficult to
> > > reason about it.
> > >
> > > There could be net_iovs that are not attached to pp, and should not be
> > > treated as pp memory. These are in the devmem (and future net_iov) tx
> > > paths.
> > >
> > > We need a way to tell if a net_iov is pp or not. A couple of options:
> > >
> > > 1. We could have it such that if net_iov->pp is set, then the
> > > netmem_is_pp == true, otherwise false.
> > > 2. We could implement a page-flags equivalent for net_iov.
> > >
> > > Option #1 is simpler and is my preferred. To do that properly, you need to:
> > >
> > > 1. Make sure everywhere net_iovs are allocated that pp=NULL in the
> > > non-pp case and pp=non NULL in the pp case. those callsites are
> > > net_devmem_bind_dmabuf (devmem rx & tx path), io_zcrx_create_area
> > > (io_uring rx path).
> > >
> > > 2. Change netmem_is_pp to check net_iov->pp in the net_iov case.
> >
> > Good idea, but I'm not sure if I could work on it without consuming your
> > additional review efforts.  Can anyone add net_iov_is_pp() helper?
> >
> 
> Things did indeed get busy for me with work work the past week and I
> still need to look at your merged netmem desc series, but I'm happy to
> review whenever I can.
> 
> > Or use the page type, Netpp, as an additional way to identify if it's a
> > pp page for system memory, keeping the current way using ->pp_magic.
> > So the page type, Netpp, is used for system memory, and ->pp_magic is
> > used for net_iov.  The clean up for ->pp_magic can be done if needed.
> >
> 
> IMO I would like to avoid deviations like this, especially since
> ->pp_magic is in the netmem_desc struct that is now shared between
> page and net_iov. I'd rather both use pp_magic or both not, but that
> may just be me.

Totally Agree.  Lemme try your option #1 then.

	Byungchul
> 
> 
> --
> Thanks,
> Mina