Re: [RFC mm v4 1/2] page_pool: check if nmdesc->pp is !NULL to confirm its usage as pp for net_iov

From: Byungchul Park

Date: Mon Oct 27 2025 - 22:24:34 EST


On Mon, Oct 27, 2025 at 06:25:38PM -0700, Mina Almasry wrote:
> On Thu, Oct 23, 2025 at 12:44 AM Byungchul Park <byungchul@xxxxxx> wrote:
> >
> > ->pp_magic field in struct page is current used to identify if a page
> > belongs to a page pool. However, ->pp_magic will be removed and page
> > type bit in struct page e.g. PGTY_netpp should be used for that purpose.
> >
> > As a preparation, the check for net_iov, that is not page-backed, should
> > avoid using ->pp_magic since net_iov doens't have to do with page type.
> > Instead, nmdesc->pp can be used if a net_iov or its nmdesc belongs to a
> > page pool, by making sure nmdesc->pp is NULL otherwise.
> >
> > For page-backed netmem, just leave unchanged as is, while for net_iov,
> > make sure nmdesc->pp is initialized to NULL and use nmdesc->pp for the
> > check.
> >
> > Signed-off-by: Byungchul Park <byungchul@xxxxxx>
> > ---
> > net/core/devmem.c | 1 +
> > net/core/netmem_priv.h | 8 ++++++++
> > net/core/page_pool.c | 16 ++++++++++++++--
> > 3 files changed, 23 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/core/devmem.c b/net/core/devmem.c
> > index d9de31a6cc7f..f81b700f1fd1 100644
> > --- a/net/core/devmem.c
> > +++ b/net/core/devmem.c
> > @@ -291,6 +291,7 @@ net_devmem_bind_dmabuf(struct net_device *dev,
> > niov = &owner->area.niovs[i];
> > niov->type = NET_IOV_DMABUF;
> > niov->owner = &owner->area;
> > + niov->desc.pp = NULL;
>
> Don't you also need to = NULL the niov allocations in io_uring zcrx,
> or is that already done? Maybe mention in commit message.

Yes, that's been already done by kvmalloc_array(__GFP_ZERO). I want to
leave a comment explaining that on io_uring side like:

diff --git a/io_uring/zcrx.c b/io_uring/zcrx.c
index e5ff49f3425e..f771bb3e756d 100644
--- a/io_uring/zcrx.c
+++ b/io_uring/zcrx.c
@@ -444,6 +444,10 @@ static int io_zcrx_create_area(struct io_zcrx_ifq *ifq,
area->freelist[i] = i;
atomic_set(&area->user_refs[i], 0);
niov->type = NET_IOV_IOURING;
+
+ /* niov->pp is already initialized to NULL by
+ * kvmalloc_array(__GFP_ZERO).
+ */
}

area->free_count = nr_iovs;

However, I dropped it as Pavel requested:

https://lore.kernel.org/lkml/8d833a3f-ae18-4ea6-9092-ddaa48290a63@xxxxxxxxx/

I will mention it in commit message then.

> Other than that, looks correct,
>
> Reviewed-by: Mina Almasry <almasrymina@xxxxxxxxxx>

Thanks.

Byungchul