netdev - Re: [RFC net-next v1 5/6] net: page_pool: convert refcounting helpers to nmdesc

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAHS8izMfqgzA75Wo9fkeLkHdCa512vqr+5iQ0u1zHKZX9uGoNQ@mail.gmail.com>
Date: Wed, 13 Aug 2025 09:55:18 -0700
From: Mina Almasry <almasrymina@...gle.com>
To: Pavel Begunkov <asml.silence@...il.com>
Cc: netdev@...r.kernel.org, Jakub Kicinski <kuba@...nel.org>, 
	Eric Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>, davem@...emloft.net, 
	sdf@...ichev.me, dw@...idwei.uk, Jesper Dangaard Brouer <hawk@...nel.org>, 
	Ilias Apalodimas <ilias.apalodimas@...aro.org>, Byungchul Park <byungchul@...com>
Subject: Re: [RFC net-next v1 5/6] net: page_pool: convert refcounting helpers
 to nmdesc

On Wed, Aug 13, 2025 at 2:10 AM Pavel Begunkov <asml.silence@...il.com> wrote:
>
> On 8/13/25 01:14, Mina Almasry wrote:
> > On Mon, Aug 11, 2025 at 9:28 AM Pavel Begunkov <asml.silence@...il.com> wrote:
> ...>> -static inline long page_pool_unref_netmem(netmem_ref netmem, long nr)
> >> +static inline long page_pool_unref_nmdesc(struct netmem_desc *desc, long nr)
> >>   {
> >> -       atomic_long_t *pp_ref_count = netmem_get_pp_ref_count_ref(netmem);
> >> +       atomic_long_t *pp_ref_count = &desc->pp_ref_count;
> >
> > nit: I think we can also kill the pp_ref_count local var and use
> > desc->pp_ref_count directly.
>
> I stopped there to save the churn, I'd rather have it on top and outside
> of cross tree branches. But I agree in general, and there is more that
> we can do as well.
>
> ...>>   static inline bool page_pool_unref_and_test(netmem_ref netmem)
> >> diff --git a/net/core/devmem.c b/net/core/devmem.c
> >> index 24c591ab38ae..e084dad11506 100644
> >> --- a/net/core/devmem.c
> >> +++ b/net/core/devmem.c
> >> @@ -440,14 +440,9 @@ void mp_dmabuf_devmem_destroy(struct page_pool *pool)
> >>
> >>   bool mp_dmabuf_devmem_release_page(struct page_pool *pool, netmem_ref netmem)
> >>   {
> >> -       long refcount = atomic_long_read(netmem_get_pp_ref_count_ref(netmem));
> >> -
> >>          if (WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
> >>                  return false;
> >>
> >> -       if (WARN_ON_ONCE(refcount != 1))
> >> -               return false;
> >> -
> >
> > Rest of the patch looks good to me, but this comes across as a
> > completely unrelated clean up/change or something? Lets keep the
> > WARN_ON_ONCE?
> I was killing netmem_get_pp_ref_count_ref(), which is why it's here.
> It checks an assumption that's guaranteed by page pools and shared
> with non-mp pools, so not like devmem needs it, and it'd not catch
> any recycling problems either. Regardless, I can leave the warning.
>

Ack. I also agree the WARN_ON_ONCE is not necessary, even the one
above it for the net_iov check is not necessary.

But since devmem was the first memory provider I'm paranoid that I got
something wrong in the general memory provider infra or in the devmem
implementation specifically; I think some paranoid WARN_ON_ONCEs are
justified, maybe. I'd rather debug the warning firing rather than a
very subtle refcounting issue or provider mixup or something at a
later point. I'm still surprised there aren't many bug reports about
any memory providers. They probably aren't used much in production
yet.

I think after 2025 or 2026 LTS it's probably time to clean up these
unnecessary WARN_ONs, but until then, let's keep them in if you don't
mind.

-- 
Thanks,
Mina