netdev - Re: [RFC net-next] netmem: replace __netmem_clear_lsb() with netmem_to

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: <20250729011024.GD56089@system.software.com>
Date: Tue, 29 Jul 2025 10:10:24 +0900
From: Byungchul Park <byungchul@...com>
To: Mina Almasry <almasrymina@...gle.com>
Cc: netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
	davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
	pabeni@...hat.com, horms@...nel.org, hawk@...nel.org,
	toke@...hat.com, asml.silence@...il.com, kernel_team@...ynix.com
Subject: Re: [RFC net-next] netmem: replace __netmem_clear_lsb() with
 netmem_to_nmdesc()

On Mon, Jul 28, 2025 at 10:44:31AM -0700, Mina Almasry wrote:
> On Sun, Jul 27, 2025 at 9:21 PM Byungchul Park <byungchul@...com> wrote:
> >
> > Now that we have struct netmem_desc, it'd better access the pp fields
> > via struct netmem_desc rather than struct net_iov.
> >
> > Introduce netmem_to_nmdesc() for safely converting netmem_ref to
> > netmem_desc regardless of the type underneath e.i. netmem_desc, net_iov.
> >
> > While at it, remove __netmem_clear_lsb() and make netmem_to_nmdesc()
> > used instead.
> >
> > Signed-off-by: Byungchul Park <byungchul@...com>
> 
> Thank you for working on paying this tech debt!

I thought it was appropriate to organize the code I modified to some
extent.

> > ---
> >  include/net/netmem.h   | 33 ++++++++++++++++-----------------
> >  net/core/netmem_priv.h | 16 ++++++++--------
> >  2 files changed, 24 insertions(+), 25 deletions(-)
> >
> > diff --git a/include/net/netmem.h b/include/net/netmem.h
> > index f7dacc9e75fd..33ae444a9745 100644
> > --- a/include/net/netmem.h
> > +++ b/include/net/netmem.h
> > @@ -265,24 +265,23 @@ static inline struct netmem_desc *__netmem_to_nmdesc(netmem_ref netmem)
> >         return (__force struct netmem_desc *)netmem;
> >  }
> >
> > -/* __netmem_clear_lsb - convert netmem_ref to struct net_iov * for access to
> > - * common fields.
> > - * @netmem: netmem reference to extract as net_iov.
> > +/* netmem_to_nmdesc - convert netmem_ref to struct netmem_desc * for
> > + * access to common fields.
> > + * @netmem: netmem reference to get netmem_desc.
> >   *
> > - * All the sub types of netmem_ref (page, net_iov) have the same pp, pp_magic,
> > - * dma_addr, and pp_ref_count fields at the same offsets. Thus, we can access
> > - * these fields without a type check to make sure that the underlying mem is
> > - * net_iov or page.
> > + * All the sub types of netmem_ref (netmem_desc, net_iov) have the same
> > + * pp, pp_magic, dma_addr, and pp_ref_count fields via netmem_desc.
> >   *
> > - * The resulting value of this function can only be used to access the fields
> > - * that are NET_IOV_ASSERT_OFFSET'd. Accessing any other fields will result in
> > - * undefined behavior.
> > - *
> 
> I think instead of removing this warning, we want to add an
> NET_IOV_ASSERT_OFFSET that asserts that net_iov->netmem_desc and
> page->netmem_desc are in the same offset, and then add a note here
> that this works because we assert that the netmem_desc offset in both
> net_iov and page are the same.

It doesn't have to have the same offset.  Why do you want it?  Is it for
some optimizaiton?  Or I think it's unnecessary constraint.

> > - * Return: the netmem_ref cast to net_iov* regardless of its underlying type.
> > + * Return: the pointer to struct netmem_desc * regardless of its
> > + * underlying type.
> >   */
> > -static inline struct net_iov *__netmem_clear_lsb(netmem_ref netmem)
> > +static inline struct netmem_desc *netmem_to_nmdesc(netmem_ref netmem)
> >  {
> > -       return (struct net_iov *)((__force unsigned long)netmem & ~NET_IOV);
> > +       if (netmem_is_net_iov(netmem))
> > +               return &((struct net_iov *)((__force unsigned long)netmem &
> > +                                           ~NET_IOV))->desc;
> > +
> > +       return __netmem_to_nmdesc(netmem);
> 
> The if statement generates overhead. I'd rather avoid it. We can
> implement netmem_to_nmdesc like this, no?
> 
> netmem_to_nmdesc(netmem_ref netmem)
> {
>   return (struct netmem_desc)((__force unsigned long)netmem & ~NET_IOV);
> }

I see.  You want this kind of optimization.  I will do this way if you
want.

> Because netmem_desc is the first element in both net_iov and page for
> the moment. (yes I know that will change eventually, but we don't have
> to incur overhead of an extra if statement until netmem_desc is
> removed from page, right?)

Okay.

	Byungchul
> 
> 
> --
> Thanks,
> Mina