Received-SPF: pass (google.com: domain of linux-kernel+bounces-88463-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20240229213018.work.556-kees@kernel.org> <20240229225910.79e224cf@kernel.org>
 <CANn89iKeJGvhY0K=kLhhR39NVbaizP2UBk0Vk0r_XCe2XMBZHg@mail.gmail.com> <9050bdec-b34a-4133-8ba5-021dfd4b1c75@intel.com>
In-Reply-To: <9050bdec-b34a-4133-8ba5-021dfd4b1c75@intel.com>
From: Eric Dumazet <edumazet@google.com>
Date: Fri, 1 Mar 2024 14:25:37 +0100
Message-ID: <CANn89iLSLPn5PyXNQcA2HO0e5jGYvHKhTh_9_EMmfbTJKZPfbg@mail.gmail.com>
Subject: Re: [PATCH] netdev: Use flexible array for trailing private bytes
To: Alexander Lobakin <aleksander.lobakin@intel.com>
Cc: Jakub Kicinski <kuba@kernel.org>, Kees Cook <keescook@chromium.org>, 
	"David S. Miller" <davem@davemloft.net>, Paolo Abeni <pabeni@redhat.com>, 
	Andy Shevchenko <andriy.shevchenko@linux.intel.com>, 
	"Gustavo A. R. Silva" <gustavoars@kernel.org>, netdev@vger.kernel.org, 
	linux-hardening@vger.kernel.org, Simon Horman <horms@kernel.org>, 
	Jiri Pirko <jiri@resnulli.us>, Daniel Borkmann <daniel@iogearbox.net>, Coco Li <lixiaoyan@google.com>, 
	Amritha Nambiar <amritha.nambiar@intel.com>, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Fri, Mar 1, 2024 at 1:59=E2=80=AFPM Alexander Lobakin
<aleksander.lobakin@intel.com> wrote:
>
> From: Eric Dumazet <edumazet@google.com>
> Date: Fri, 1 Mar 2024 09:03:55 +0100
>
> > On Fri, Mar 1, 2024 at 7:59=E2=80=AFAM Jakub Kicinski <kuba@kernel.org>=
 wrote:
> >>
> >> On Thu, 29 Feb 2024 13:30:22 -0800 Kees Cook wrote:
>
> Re WARN_ONCE() in netdev_priv(): netdev_priv() is VERY hot, I'm not sure
> we want to add checks there. Maybe under CONFIG_DEBUG_NET?
>
> >
> >>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> >>> index 118c40258d07..b476809d0bae 100644
> >>> --- a/include/linux/netdevice.h
> >>> +++ b/include/linux/netdevice.h
> >>> @@ -1815,6 +1815,8 @@ enum netdev_stat_type {
> >>>       NETDEV_PCPU_STAT_DSTATS, /* struct pcpu_dstats */
> >>>  };
> >>>
> >>> +#define      NETDEV_ALIGN            32
> >>
> >> Unless someone knows what this is for it should go.
> >> Align priv to cacheline size.
> >
> > +2
> >
>
> Maybe
>
> > #define NETDEV_ALIGN    L1_CACHE_BYTES
>
> #define NETDEV_ALIGN    max(SMP_CACHE_BYTES, 32)

Why would we care if some arches have a very small SMP_CACHE_BYTES ?
Bet it !

IMO nothing in networking mandates this minimal 32 byte alignment.

>
> ?
>
> (or even max(1 << INTERNODE_CACHE_SHIFT, 32))

I do not think so.

INTERNODE_CACHE_SHIFT is a bit extreme on allyesconfig on x86 :/
(with CONFIG_X86_VSMP=3Dy)


>
> >
> > or a general replacement of NETDEV_ALIGN....
> >
> >
>
> + I'd align both struct net_device AND its private space to
> %NETDEV_ALIGN and remove this weird PTR_ALIGN. {k,v}malloc ensures
> natural alignment of allocations for at least a couple years already
> (IOW if struct net_device is aligned to 64, {k,v}malloc will *always*
> return a 64-byte aligned address).

I think that with SLAB or SLOB in the past with some DEBUG options
there was no such guarantee.

But this is probably no longer the case, and heavy DEBUG options these
days (KASAN, KFENCE...)
do not expect fast networking anyway.